IMF-Supported Programs: Recent Staff Research; Ashoka Mody ... · Wolfgang Mayer Professor of...

IMF-Supported Programs: Recent Staff Research

IMF-Supported P

rograms:

Recent S

taff Research

IMF

© 2006 International Monetary Fund

Production: IMF Multimedia Services DivisionInterior Design: Alicia Etchebarne-BourdinCover Design: Martine Rossignol-Winner

Cataloging-in-Publication Data

[IMF-supported programs: recent staff research/edited by Ashoka Mody and Alessan-dro Rebucci]—Washington, D.C.: International Monetary Fund, 2006]

p. cm.

Includes bibliographical referencesISBN 1-58906-361-91. International Monetary Fund—Research. I. Mody, Ashoka. II. Rebucci, Alessan-

dro. III. International Monetary Fund.HG3881.5.I58I5 2006

Price: US$37.50

Please send orders to:International Monetary Fund, Publication Services

700 19th Street, N.W., Washington, DC 20431, U.S.A.Tel.: (202) 623-7430 Telefax: (202) 623-7201

E-mail: [email protected]: http://www.imf.org

recycled paper

The following symbols have been used throughout this paper:

. . . to indicate that data are not available;

— to indicate that the figure is zero or less than half the final digit shown, orthat the item does not exist;

– between years or months (e.g., 1997–98 or January–June) to indicate theyears or months covered, including the beginning and ending years ormonths;

/ between years (e.g., 1997/98) to indicate a fiscal (financial) year.

“n.a.” means not applicable.

“Billion” means a thousand million.

Minor discrepancies between constituent figures and totals are due to rounding.

The views expressed in this publication are those of the authors and do notnecessarily represent the views of the International Monetary Fund. The term“country,” as used in this publication, does not in all cases refer to a territorialentity that is a state as understood by international law and practice; the termalso covers some territorial entities that are not states, but for which statisticaldata are maintained and provided internationally on a separate andindependent basis.

Foreword v

Contributors vii

Overview ix

PART I. DESIGN

1 Programs Objectives and Program SuccessAtish Ghosh, Timothy Lane, Alun Thomas, and Juan Zalduendo 3

2 IMF Programs and Growth: Is Optimism Defensible?Reza Baqir, Rodney Ramcharan, and Ratna Sahay 17

3 Do IMF-Supported Programs Boost Private Capital Inflows? The Role of Program Size and Policy AdjustmentRoberto Benelli 35

4 Macroeconomic Adjustment in IMF-Supported Programs: Projections and RealityRuben Atoyan, Patrick Conway, Marcelo Selowsky, and Tsidi Tsikata 52

5 Fiscal Policy, Expenditure Composition, and Growth in Low-Income CountriesSanjeev Gupta, Benedict Clements, Emanuele Baldacci, and Carlos Mulas-Granados 84

PART II. IMPLEMENTATION

6 Efficiency and Legitimacy: Trade-Offs in IMF GovernanceCarlo Cottarelli 103

7 IMF Conditionality and Country Ownership of Adjustment ProgramsMohsin S. Khan and Sunil Sharma 119

8 Different Strokes? Common and Uncommon Responses to Financial CrisesJames M. Boughton 131

9 Institutions, Program Implementation, and Macroeconomic PerformanceSaleh M. Nsouli, Ruben Atoyan, and Alex Mourmouras 140

Contents

iii

CONTENTS

10 What Determines the Implementation of IMF-Supported Programs?Anna Ivanova, Wolfgang Mayer, Alex Mourmouras, and George Anayiotos 160

PART III. EFFECTIVENESS

11 International Bailouts, Moral Hazard, and ConditionalityOlivier Jeanne and Jeromin Zettelmeyer 189

12 Bedfellows, Hostages, or Perfect Strangers? Global Capital Marketsand the Catalytic Effect of IMF Crisis LendingCarlo Cottarelli and Curzio Giannini 202

13 Keeping Capital Flowing: The Role of the IMFMichael Bordo, Ashoka Mody, and Nienke Oomes 228

14 Long-Term Fiscal Developments and IMF Conditionality: Is There a Link?Ales Bulír and Soojin Moon 245

15 Evaluating IMF-Supported Programs in the 1990s: The Importance of Taking Explicit Account of StoppagesChuling Chen and Alun Thomas 260

iv

A long tradition of research has sought to evaluate the effectiveness of IMF-supported programs. The IMF itself has devoted much institutional energy to assessing its performance, both to learn from the past and to deal with new chal-lenges. In doing so, the staff of the IMF has contributed significantly to analyzingthe IMF’s ability to achieve its stated goals of fostering external viability andgrowth. IMF staff members have had the advantage of access to program detailsoften not as easily available to external researchers. Staff contributions have alsoincluded methodological advances. Importantly, in conducting this research, wehave not shied away from unpleasant conclusions.

This volume includes recent research that moves decisively beyond the typi-cal characterization of programs. Though researchers have sometimes made thebasic distinctions between the different types of IMF program—Stand-ByArrangements, and arrangements under the Extended Fund Facility and thePoverty Reduction and Growth Facility (as well as their predecessor facilities)—the nuances of programs are more extensive. In particular, the circumstancesunder which the program is designed, the limitations in program implementa-tion, and the specific national and international economic conditions when theprogram is in effect all influence the ultimate outcome. The key achievement ofthe research reported here is to deal with this complexity while also providingsimple insights. Of course, all research is work in progress, and these insightsneed to be verified by further work, but they do offer useful working hypothesesfor our operations.

Looking ahead, I see two obvious areas for further work. First, as the essays inthis volume demonstrate, the IMF’s advice and financing are conditioned by avariety of political economy considerations. International political economy hasa bearing on program selection, conditionality, and implementation. Similarly,domestic political economy shapes attitudes toward the IMF and, hence, theability of the authorities to enter into constructive IMF-supported programs. A richer political economy analysis could make for more effective engagement.

Second, the IMF’s governance structure also has implications for program de-sign and implementation, and, hence, for the IMF’s economic effectiveness.Though, once again, the task is a difficult one, careful analysis could provideuseful input to the ongoing debates.

Raghuram G. RajanEconomic Counsellor and Director, Research Department

International Monetary Fund

Foreword

v

George Anayiotos Senior Economist, Finance Department

Ruben Atoyan Economist, Western Hemisphere Department

Emanuele Baldacci Lead Researcher, Italian Statistical Office (ISTAT)

Reza Baqir Resident Representative, Philippines

Roberto Benelli Economist, Western Hemisphere Department

Michael Bordo Professor of Economics, Rutgers University

James M. Boughton IMF Historian

Ales Bulír Senior Economist, Policy Development and Review Department

Chuling Chen Analyst, Modeling and Research Department, (U.S.) Federal Home Loan Mortgage Corporation

Benedict Clements Advisor, Western Hemisphere Department

Patrick Conway Professor of Economics, University of North Carolinaat Chapel Hill

Carlo Cottarelli Deputy Director, Policy Development and ReviewDepartment

Atish Ghosh Division Chief, Policy Development and ReviewDepartment

Curzio Giannini2 Deputy Head, International Relations Department, Banca d’Italia

Sanjeev Gupta Assistant Director, African Department

Anna Ivanova Economist, Fiscal Affairs Department

Olivier Jeanne Deputy Division Chief, Research Department

Mohsin S. Khan Director, Middle East and Central Asia Department

Contributors1

vii

1 All contributors are staff members of the International Monetary Fund unless otherwise indicated. 2 Deceased.

CONTRIBUTORS

Timothy Lane Assistant Director, Western Hemisphere Department

Wolfgang Mayer Professor of Economics, University of Cincinnati

Ashoka Mody Assistant Director, European Department

Soojin Moon Candidate Member, American Institute of Certified Public Accountants (AICPA)

Alex Mourmouras Deputy Division Chief, IMF Institute

Carlos Mulas-Granados Vice-Chairman, Prime Minister’s Economic Bureau (Madrid)

Saleh M. Nsouli Director, IMF Offices in Europe

Nienke Oomes Economist, European Department

Rodney Ramcharan Economist, Research Department

Alessandro Rebucci Economist, Research Department

Ratna Sahay Assistant Director, Western Hemisphere Department

Marcelo Selowsky Assistant Director, Independent Evaluation Office

Sunil Sharma Director, IMF-Singapore Regional Training Institute

Alun Thomas Senior Economist, Policy Development and ReviewDepartment

Tsidi Tsikata Deputy Division Chief, African Department

Juan Zalduendo Senior Economist, Policy Development and ReviewDepartment

Jeromin Zettelmeyer Assistant to the Director, Western HemisphereDepartment

viii

IMF lending programs, though varying in objectives and duration, are often associatedwith a sharp and sustained redirection of the course of economic policy. An IMF-supported program is typically initiated when a country faces the need for exter-nal adjustment. The IMF provides financing and the country puts in place a pro-gram of policies to redress actual or potential external imbalances; also, whereappropriate, economic reforms to raise long-term growth are introduced or accel-erated. Continued lending is predicated on progress in implementing the pro-gram, which in turn is assessed on the basis of preset conditions (performancecriteria) to be met by specified dates in the context of periodic reviews.

Does the policy redirection supported by IMF lending achieve its goals? The effec-tiveness of program lending in achieving its goals has long been the subject of research. The evaluation is amenable to analyses that assess the variations ineconomic indicators before and after the start of a program.2 And although evensophisticated before-and-after analyses have their pitfalls, researchers have con-tinued to exercise their ingenuity to identify, or isolate, the effects of an IMFprogram.3

The IMF’s own staff has been active in—and, often, at the forefront of—assessingthe effectiveness of program lending. This book describes the recent evolution ofstaff research on program lending, highlighting both methodological contribu-tions and substantive findings. In the mid- and late 1980s, the overarching ques-tion posed was whether an IMF-supported program helped improve macroeco-nomic performance—as measured by growth, inflation, and current accountoutcomes—with the latter a proxy for the extent and speed of economic adjust-ment. (Haque and Khan (1998) summarize this research.) Recent IMF researchhas focused on more nuanced questions and made more pointed inquiries intothe factors contributing to program success.

An advance in the research—and an organizing device for this book as well as a dis-tinguishing feature of many of its chapters—has been to recognize that program effec-tiveness needs to be linked to the quality of program design and implementation, as wellas to the degree of a country’s external vulnerability. Analyses of effects before andafter the adoption of a program implicitly assume either that the quality of policydesign and the implementation of those policies do not vary across programs orthat they do not matter. Since they obviously do matter, the links among design,implementation, and effectiveness are increasingly being explored. These links

Overview

Ashoka Mody and Alessandro Rebucci1

ix

1 The authors are grateful to Timothy Lane, Anne Krueger, and Raghuram Rajan for their valuablecomments in the preparation of this overview and to Paul Gleason for overseeing the copyeditingand production of this publication.

2 In contrast, the effects of surveillance (the analysis of economic conditions, policies, andprospects in member countries and the global economy) are felt in a diffuse manner, contemporane-ous with many other economic developments, over an extended period and are thus not easy to iden-tify. See IMF (2005).

3 The appropriate techniques have, not surprisingly, remained controversial. A program must bejudged against a benchmark, or the outcome that would have occurred if the program had not beenput in place. This counterfactual experiment is not observable. Moreover, the benchmark evolveswith the changing world economy and the country’s economic conditions.

are complex, and, hence, researchers have focused on various elements of a largeset of possibilities, guided, in part, by their inherent interest in the links but alsoconstrained by their limited ability to quantify measures of design and imple-mentation and, hence, subject them to statistical analysis. At the same time, in-terest has persisted in carrying out direct examinations of program effectivenesswithout focusing on design and/or implementation but with a clearer recogni-tion that country conditions influence both the incentives for policy reform andthe confidence of international capital markets that is necessary to reinforce do-mestic reform efforts.

Thus, the questions motivating recent research are framed as follows:

• How—and how much—does success depend upon the design of the program? • To the extent that differences in program outcomes reflect differences in the

quality of their actual implementation, why does implementation vary and howis it related, in particular, to program “ownership,” or the commitment by coun-try authorities to the policy reform package?

• Do differing economic conditions, such as the levels of external debt and re-serves, influence the willingness and ability of countries, private markets, andthe IMF to coordinate to achieve success?

• If a program brings an improvement in a country’s economy, does that reflectthe IMF’s policy advice, its lending, its monitoring of the country’s policies, orits “seal of approval” effect?

Though the questions are posed in these ambitious terms, the answers are necessarilymore piecemeal and reflect, among other limitations, the particular ways in which theterms “design,” “implementation,” and “country conditions” are operationalized. Thequality of design, sometimes construed as the composition of the policy advice(e.g., relative emphasis on fiscal or monetary discipline), is often inferred bycomparing actual outcomes with those projected at the inception of the pro-gram. But the interpretation of these comparisons is not always straightforward.Similarly, assessments of implementation rely on such measures as the adherenceof a program to its original conditions or its timetable or to the drawing down ofthe originally anticipated loan amount. The implicit assumption is that if theperiodic reviews were delayed, the program was canceled, or a smaller-than-anticipated amount was borrowed, the program was not fully implemented. Thismay sometimes be true, but failure to meet conditions on the agreed schedulemay also reflect corrections made as new information becomes available; andfailing to draw the full amount may also reflect an improved economy withscaled-back requirements for external financing. Moreover, carrying out pre-scribed actions during the course of the program may be of little use if these ac-tions are discontinued or reversed as soon as the program is over, necessitatingthe nontrivial task of assessing the sustainability of the policies put in placethrough the program. Finally, it is not straightforward to parsimoniously definethe international and country context within which a program is formulated andimplemented. Yet recent theoretical advances and empirical research suggestthat such a context is relevant for program success. For example, applying theo-ries proposing that success is most likely to be achieved when a country is in the“intermediate” state of external vulnerability—neither safe from a crisis nor im-minently facing one—entails making judgments about, for example, levels ofdebt and reserves.

The rest of this overview reports on the contributions to this book, discussing them inthe context of the broader questions they raise and pointing to the scope for further re-search. The next section discusses program design. This is followed by a sectionreviewing research on the determinants and the value of program implementa-tion, also raising the issue of whether and how implementation is influenced by

x

OVERVIEW

Ashoka Mody and Alessandro Rebucci

the nature of IMF governance. Next, the discussion of program effectiveness focuses—in line with recent research—on the IMF’s role in capital accountcrises and in “catalyzing” private capital flows. A final section summarizes keyfindings and possible research directions.

Program Design

A challenge for good design—and a recurring theme in this research—is the need toadapt design to changing global and country economic conditions while preserving uni-formity of treatment (as emphasized, in particular, by Boughton in Chapter 8). Tothis end, an eclectic approach to program design has evolved. Ghosh and others(Chapter 1) state that “a variety of different analytical tools are used to analyzeand predict developments in particular sectors while the financial programmingframework is used to ensure that projections for different sectors are consistentwith key balance-sheet identities.” They further note that although the IMF hasmaintained the overarching objective of “achieving (or maintaining) externalviability,” the variation in program objectives has resulted in different types ofprograms:

• classic stabilization and adjustment programs (whose primary objective is to cor-rect a current account imbalance and restore official reserves to a safe level);

• capital account crisis programs (whose primary objective is to restore confidencein international capital markets to staunch a sudden loss of private external financing); and

• reform programs (whose priority is to support structural reforms designed to con-tribute to economic growth and stability over the long term to, for instance,meet the early needs of former transition economies and, more recently, of low-income developing countries).

Are IMF-supported programs well designed? The answer to this question dependson the definition of “well designed.” One measure of good design is the realismof—or, at least, the absence of systematic bias in—macroeconomic projectionsand performance targets subject to conditionality (Baqir, Ramcharan, andSahay, Chapter 2). Although such a representation of good design has some va-lidity, it needs to be interpreted with caution. First, if outcomes fall short of pro-jections, it could be that ambitious goals were set to achieve the best possibleoutcomes; more cautious goals may well be met, or even exceeded, but may beinsufficient to achieve a sustainable transition to a higher growth path. Second,the evidence is that the goals are more overoptimistic as the horizon gets longer;however, the early stage, when a program has to be jump-started, may be the mostimportant. An alternative approach to assessing program design is by examiningthe composition of policy adjustment. Baqir, Ramcharan, and Sahay (Chapter 2)and Gupta and others (Chapter 5) examine, respectively, the relative importanceand the composition of fiscal measures. Given the range of policy measures un-dertaken, composition can be assessed in many ways, and in future research thechoices made could be guided by appropriate theoretical considerations.

Divergence in program projections and outcomes is particularly evident for medium-to-long-run growth projections. Consistent with the analysis of Musso and Phillips(2002), Ghosh and others (Chapter 1) do not find a significant bias in short-rungrowth projections. Beyond a horizon of about one year, however, the overpre-diction of growth increases as the horizon lengthens, and this is so regardless of thetype of program. Baqir, Ramcharan, and Sahay (Chapter 2) also conclude thatoverprediction of growth remains even when account is taken of performancebenchmarks (or, in their terminology, “intermediate targets”) that are not met.

xi

OVERVIEW

In other words, the overprediction is not just a reflection of proposed ambitiousmeasures that are eventually not realized. They conclude that, therefore, the biasin predictions reflects either unwarranted optimism or an inadequate analyticalframework. They reach similar conclusions about inflation. There is no consen-sus on the sources of these biases or on their implications, though.

Any bias in current account projections appears to depend on the type of program.Ghosh and others (Chapter 1) find that classic IMF stabilization and adjustmentprograms and capital account crisis programs systematically tend to underpredictexternal adjustment. This occurs, in part, because growth—and, hence, im-ports—do not pick up as rapidly as expected, leading to a greater-than-expectedimprovement in the trade balance. In contrast, reform programs tend to overpre-dict external adjustment—leading to a greater-than-projected buildup of exter-nal debt—particularly for low-income countries.

The differences between projections and outcomes appear to have different causes.These may include insufficient information on the economic conditions prevail-ing at the time of program design, shortcomings in the framework applied forprogram projections, and the need for projections that are acceptable to both theauthorities and the IMF’s Executive Board. Formally, in fact, program projec-tions are proposed by the authorities under programs formulated by them. Chapter 4 by Atoyan and others, which focuses on fiscal and external projec-tions, concludes that inaccuracies in the preliminary statistical information baseare the most serious source of projection errors. The authors note, however, thatthis is a common source of projection errors in policymaking that has been welldocumented in, for example, the formulation of U.S. monetary policy decisions.They further point to a “learning process” that takes place as new data becomeavailable and the size of the divergence declines. In contrast, in Chapter 3,Benelli suggests that the limited pool of IMF lending resources may induce a biastoward excessive optimism in the balance of payments projections of programs;although this does not explain why countries undertaking Stand-By Arrange-ments tend, on the contrary, to underestimate the balance of payments adjust-ment, it points to the possibility that, in practice, balance of payments projec-tions are tailored to the financing provided rather than vice versa.

The contribution of the alleged analytical inconsistency to projection errors is, as yet,speculative. Although the underlying frameworks for program design meet thediscipline of accounting identities, Baqir, Ramcharan, and Sahay (Chapter 2)conjecture that the design may not always be theoretically consistent. Behav-ioral or analytical inconsistency could arise because the toolkits currently usedin program design—financial programming, the balance-sheet approach, vulner-ability assessments, and debt-sustainability analyses—are not model-based, mu-tually consistent theoretical frameworks. In the face of a theoretical inconsis-tency, it would not be surprising to see policy outturns deviating from targets.Moreover, because inconsistently designed programs are likely to be more difficultto implement, they may also be less effective. Atoyan and others (Chapter 4) findthat the projected timing and speed of adjustment in fiscal and external balancesdiffers from those actually achieved and attribute this to inadequacies in the the-oretical underpinnings, a plausible but not established conclusion. Thus, a fruit-ful area for future research would be to establish the extent to which behavioralconsistency is lacking and assess how that affects program implementation and,ultimately, effectiveness.

Research could aid in the further integration of analytical frameworks, adapted todiffering country circumstances. In Chapter 1, for instance, Ghosh and others notethat the financial-programming framework is not easily adapted to designing re-form programs or programs in capital-account-driven crises, because it takesgrowth and foreign financing as exogenous. Batista and Zalduendo (2004) arguethat adopting an empirical, cross-country growth framework of reference may

xii


enhance the accuracy of growth projections under reform programs. Someprogress along these lines has been achieved on both capital account issues andofficial financing with the development of debt-sustainability analysis and a balance-sheet approach to vulnerability, but there is scope for further advance inthis direction. For instance, policy packages supported by reform programs andthe associated conditionality now take into account the need to avoid buildupsof external debt and debt-service obligations beyond specific thresholds in low-income countries. For capital account crisis programs, the typical policy packagetakes into account the linkages among different sectors of the economy in termsof both flows and stock positions, thus providing a more adequate cushionagainst relative price changes that may occur during adjustment to the crisis.Boughton (Chapter 8) also highlights, in the context of financial crises, the im-portance to program design of international investor psychology and domesticstructural conditions.

In contrast to the comparison of projections and outcomes, the research on the com-position of policy adjustment is more limited but points to the value of fiscal adjustment.In Chapter 2, Baqir, Ramcharan, and Sahay compare program objectives andperformance targets with actual outcomes in a large set of recent programs. Theyconclude that, relative to the targets set, more ambitious fiscal contractions areassociated with better growth performance while more ambitious monetary con-tractions are associated with worse outcomes. Chapter 5 by Gupta and othersfinds not only that strong fiscal consolidations (in absolute terms) are associatedwith higher economic growth in both the short and long terms but also that thecomposition of fiscal adjustment matters: fiscal adjustments achieved by curtail-ing current expenditures are, in general, more sustained and more conducive togrowth. When public investment is also protected, the positive effect of fiscaladjustment on growth is further accentuated. Furthermore, Chapter 14 by Bulírand Moon makes the point that the quality of fiscal adjustment is better underIMF programs (than in countries where there is no program) in the sense that itis directed more toward expenditure reductions than increases in revenue.

Program Implementation

Recent research on program implementation has focused on domestic political economyconstraints. As noted, in empirical analyses, implementation has been measuredin terms of program interruptions and completions. This, of course, is not neces-sarily the same thing as implementation of a policy package. Within the limits ofthe measures used, however, both the determinants of implementation and itsinfluence on program outcomes have been examined. The findings suggest that astronger political and institutional environment is conducive to better programimplementation. Compared with programs that falter, better-implemented pro-grams lead, in turn, to superior macroeconomic outcomes.

The implementation of IMF-supported programs depends to a significant extent onthe domestic political and institutional environment. In Chapter 10, Ivanova andothers quantify the factors that determine successful implementation of IMF-supported programs. They find that program implementation is weakened bystrong special interests in the parliament, lack of political cohesion, political in-stability, ethno-linguistic divisions, and inefficient bureaucracies. In contrast,they find that IMF effort—proxied by the number of IMF staff members in-volved and the design of conditionality—does not significantly affect the proba-bility of successful implementation of IMF-supported programs. Despite the au-thors’ efforts to disentangle the direction of causation, the possibility remainsthat dealing with more difficult cases requires more staff and management re-sources.

xiii

OVERVIEW

Implementation has real consequences: the closer a program’s implementation is tospecifications originally agreed with the IMF’s Executive Board, the more effective itappears to be. Chapter 9 by Nsouli, Atoyan, and Mourmouras finds that betterprogram implementation is associated with better macroeconomic outcomes, in-cluding lower inflation, a stronger external position, and faster economic growth(though the results on growth are not statistically significant). Using a some-what different approach (comparing program and non-program periods for thesame country rather than comparing across countries), Chapter 15 by Chen andThomas arrives at similar conclusions. It finds that program interruptions are as-sociated subsequently with higher inflation, higher budget deficits, and lowergrowth than periods without a program. Completed programs are, instead, asso-ciated with marginally increased growth three years after the termination of theprogram.

The extent to which program conditions are implemented depends on the extent ofdomestic “ownership”—but this may itself depend on the content and form of condi-tionality. In Chapter 7, Khan and Sharma concede that ownership is an elusiveconcept but argue that it may be enhanced by giving the authorities greater flex-ibility in deciding how to achieve agreed outcomes. To this end, they favor “outcomes-based” conditionality, which involves “conditioning disbursementson the achievement of results rather than on the implementation of policies ex-pected to eventually attain program objectives.” They note that much of pro-gram conditionality is already outcomes-based—particularly with regard tomacroeconomic conditions such as a floor on net international reserves. Theysuggest other possible outcomes that could form the basis for conditionality: thetrade balance, the current account, investment, and growth.

Although outcomes-based conditionality has attractive features, it is not straightfor-ward to operationalize. Country authorities are not in full control of outcomes.Thus, despite their best efforts, outcomes may diverge from those intended be-cause of unforeseen developments. In contrast, the authorities have greater con-trol over policies. Khan and Sharma note that policy- and outcomes-based con-ditionality will likely be combined in practice, with the precise combinationreflecting “country circumstances” and “preferences.” A variation of outcomes-based conditionality is the targeting of financial assistance to countries “that arelikely to make the best use of it,” as discussed in Chapter 11 by Jeanne andZettelmeyer. These authors argue for ex ante conditionality, established before acountry faces a crisis, rather than the ex post conditionality that currently fol-lows a crisis in the context of an IMF-supported program. Ex ante conditionalityspecifies policy and outcome benchmarks that a country should meet to be eligi-ble for financial support in the event of a liquidity crisis. As with outcomes-based conditionality, specification of ex ante conditionality has remained con-troversial. Experimenting with alternative forms of conditionality may bedesirable, however, to improve its effectiveness. (In Chapter 14, for instance,Bulír and Moon suggest that structural conditionality has not had a significantinfluence on the achievement of fiscal targets.)

The governance constraints faced by the IMF may hamper program design and im-plementation. In Chapter 6, Cottarelli argues that designing appropriate gover-nance structures for an institution such as the IMF is difficult, because of under-lying trade-offs between legitimacy and effectiveness. Attempts to enhance theIMF’s legitimacy by placing constraints on its decision-making process may re-duce the organization’s operational flexibility and, hence, lower its effectiveness.He illustrates these trade-offs in three contexts. First, increased control of theIMF’s functioning by national political authorities may restrict the possiblerange of technical decisions. Echoing the suggestions reported earlier for changesto the IMF’s own approach to conditionality, he proposes that the IMF itself bemonitored ex post—that is, on the basis of outcomes and results—rather than

xiv


subjected to efforts to influence its technical processes. Second, Cottarelli isconcerned that the legitimate need for transparency may conflict with the IMF’soperational goals. Third, and finally, while recognizing that uniformity of treat-ment across countries grants legitimacy to IMF programs, he cautions that it maycreate inefficiencies in their technical design and processing. In particular, he argues that too much emphasis has been given to formalistic procedural unifor-mity of treatment rather than to establishing “substantive evenhandedness.”

Further research, with a greater empirical bent, on the implications of the IMF’s gov-ernance structure is needed. Is the distribution of power among different membersof the Executive Board only an issue of member equality and representation, ordoes it also affect the overall institution’s effectiveness? Do the governancestructure of the IMF and domestic political economy interact? Does the gover-nance structure exacerbate the political economy constraints that country au-thorities face by influencing their behavior in program negotiation and imple-mentation?

Program Effectiveness

Pioneering research done at the IMF on program effectiveness examined the macroeco-nomic consequences of IMF-supported programs. These early efforts asked the over-arching question whether an IMF-supported program helped improve macroeco-nomic performance as measured by growth, inflation, and current accountoutcomes. Despite the difficulties posed by the presence of “treatment effects”and the lack of differentiation across programs, this early literature reached threeconclusions that have proved durable. (See Haque and Khan (1998) for a sur-vey.) First, and not surprisingly, an IMF program helps to improve the country’sexternal balance. Second, inflation falls following the start of a program, but theeffect is not statistically significant. Third, growth falls initially, but recovers inthe medium term, though possibly not to its pre-program level, under classicaland capital account crisis programs. Dicks-Mireaux, Mecagni, and Schadler(2000) find, though, that the growth effects have been stronger in low-incomecountries. Academics have largely continued to view IMF programs as zero-oneevents and pursued the analysis of their impact on overall economic perfor-mance. Examples of more recent academic research in this vein include Prze-worski and Vreeland (2000), Hutchinson (2001), and Barro and Lee (2002).

Recent IMF research has tended to identify conditions under which IMF programsmay succeed and to infer from those findings the features of the IMF that contribute tosuccess. Research by the IMF’s staff has moved beyond the earlier literature intwo important respects. First, researchers have sought to examine the conditionsunder which IMF programs are effective. In other words, rather than treatingIMF programs as undifferentiated (and represented, therefore, in empirical workas a single dummy variable for the presence or absence of an IMF program), a greater effort has been made to identify the influence of initial country condi-tions. In this spirit, greater effort has also been made to distinguish differenttypes of programs and, as noted above, to assess the costs of inadequate programimplementation. Second, these differentiations have allowed a more forceful ex-amination of the IMF’s comparative advantages and, therefore, of its informa-tional and lending roles.

In turn, the conditions for IMF effectiveness have been sought in the context of capi-tal account crises and catalysis of capital flows. Outcomes under programs support-ing adjustment to capital account crises have been mixed. The usual experiencehas been that the relatively large shock faced by the country is reflected, duringthe first year of the program, in a continuing decline in output, employment,and consumption, but this is followed by a sharp, positive reversal in confidence

xv

OVERVIEW

and private capital flows (see Chapter 1 by Ghosh and others). Jeanne andZettelmeyer conclude, in Chapter 10, that official lending in the context of the1990s financial crises did not create moral hazard at the expense of global tax-payers. Rather, domestic taxpayers, who ultimately had to repay the IMF andother official creditors, bore the costs of the crises. To limit the risk of govern-ments borrowing irresponsibly from international markets—and placing them-selves subsequently in the position of requiring official financing—Jeanne andZettelmeyer suggest that the appropriate response is not to raise the interest ratescharged by the IMF, but rather to condition large-scale crisis lending on soundpre-crisis policies and institutions.

IMF research has also focused on a more medium-term response of capital flows toIMF-supported programs, in addition to the implications for financing at times of capi-tal account crises. In Chapter 12, Cottarelli and Giannini find that the aggregateevidence on the response of capital flows to IMF-supported programs, which isoften referred to as the catalytic effect of IMF financing, is weak. They arguethat the absence of such an effect is all the more remarkable, since empiricalstudies often fail to control for actual policy change, thereby biasing their resultsin favor of such an effect. This is one area, however, in which country differenti-ation turns out to be crucial.

The evidence suggests IMF-supported programs are most catalytic when the initialconditions are not too negative or the external financing need is only potential and notactual. Bordo and others (Chapter 13) reach this conclusion (as does an earlierstudy by Mody and Saravia (2003)). They investigate the IMF’s role in main-taining emerging market economies’ access to international capital markets andfind that both macroeconomic aggregates and capital flows improve followingthe adoption of an IMF-supported program, although they may initially deterio-rate somewhat. They also find that IMF programs are most successful in improv-ing capital flows to countries in a state of vulnerability—as distinct from ex-treme distress. In such countries, IMF programs are also associated withimprovements in external fundamentals. Consistent with these findings, the cat-alytic effect of IMF lending also appears salient in the context of precautionaryprograms—that is, those undertaken for crisis prevention.

IMF research has also identified specific features of IMF-supported programs thatmay make a catalytic effect possible. Building on the extensive taxonomy proposedby Cottarelli and Giannini, Chapter 13 by Bordo and others concludes that it isimplausible to assert that the IMF’s signaling role (the IMF’s conferring of its“seal of approval” on a country’s reform policies) is crucial to the catalyticprocess, since that assumes the IMF has information that others do not. Instead,they suggest, the IMF’s monitoring role allows countries on the reform path tosignal commitment. They also find only ambiguous support for the hypothesisthat more lending is more catalytic.

Looking ahead, assessing program effectiveness remains a challenge, but progress ispossible. The challenge arises from the absence of counterfactual experiments,pervasive endogeneity issues, measurement errors in the data, and potentialomitted-variable biases. Establishing causal links between IMF programs andoutcomes requires considerable finesse. The creative use of alternative datasources and methodologies may help overcome some of the methodological diffi-culties. Case studies, nonparametric and Bayesian estimation methods (such asthose used in Rossi and Rebucci (2005)), and simulations of alternative programscenarios could be attempted. A data-collection effort to extend the sample ofprograms analyzed by IMF researchers to include programs approved in the1970s and 1980s would be very valuable. Finally, the data used by the IMF staff ’sresearch are generally not publicly available, limiting their use by the academiccommunity. Finding ways to provide access to program data without violatingcountry confidentiality could enhance the credibility of the IMF staff ’s analyses

xvi


and encourage further research in this area. Perhaps more importantly, there isscope for further investigating program modalities and the channels throughwhich they succeed or fail. This would help in the design of more effective pro-grams in the future. Joint analysis of program design, implementation, and effec-tiveness, in particular, is likely to be a promising direction for future research inlight of linkages highlighted by the research reviewed in this book.

Conclusions

The contribution of recent IMF research has been the more precise identification of theconditions necessary for the success of IMF programs. The research has moved fruit-fully from the question “Do programs work or not?” to “Under what conditionsdo programs work?” The challenge has been to give structure to this more nu-anced question. The contributions to this book point to features of program de-sign, the quality of program implementation, and the country conditions likelyto contribute to program success. Much remains to be done, however. Not leastis the challenge to further characterize the relevant design, implementation, andcountry conditions in the context of different types of programs and countries.

On program design, the research both points to specific recommendations and callsfor future research. Among the specific recommendations, the value of fiscal ad-justment is underscored, as is the effort to increase the accuracy of program pro-jections. In turn, the quality of program projection is identified with more accu-rate and comprehensive information, especially on initial conditions. Authorscall for more refined theoretical analytical frameworks to deal, for example, withcapital account crises and the determinants of long-term growth.

Variations in program-implementation experiences tend to reflect differences in do-mestic institutions and political constraints. These findings are consistent with thecall for greater country ownership of IMF-supported programs. Future researchcould usefully analyze more carefully the determinants and impact of programconditionality. The IMF’s governance structure—in the specific sense of how op-erational control is exercised by national authorities, the degree of transparency,and the manner in which uniformity of treatment across member countries isachieved—may also be an important influence on program design and imple-mentation.

Recent research on program effectiveness has focused on the IMF’s role in the con-text of capital account crises and in catalyzing capital flows, distinguishing among dif-ferent country economic conditions in evaluating success. In the resolution of capitalaccount crises, the finding is that private creditors and the IMF were largely re-paid; hence, to protect domestic taxpayers who bore much of the costs of thecrises, the suggestion is that IMF lending be conditioned on the quality of pre-crisis policies and institutions. In helping countries maintain medium-term ac-cess to international capital markets, the IMF may be most effective when amember country is vulnerable to, but not yet in, a crisis. The research points tothe value of a country committing itself to an altered policy course and usingIMF monitoring as a signal of its commitment to reform. Further research willbenefit from a more precise characterization and analysis of the varieties of IMFprograms. New empirical approaches that enhance analysts’ ability to attributespecific outcomes to IMF programs would also be useful.

ReferencesBarro, Robert J., and Jong-Wha Lee, 2002, “IMF Programs: Who Is Chosen and What

Are the Effects?” NBER Working Paper No. 8951 (Cambridge, Massachusetts: National Bureau of Economic Research).

xvii

OVERVIEW

Batista, Catia, and Juan Zalduendo, 2004, “Can the IMF’s Medium-Term Growth Projec-tions Be Improved?” IMF Working Paper 04/203 (Washington: International Mone-tary Fund).

Dicks-Mireaux, Louis, Mauro Mecagni, and Susan Schadler, 2000, “Evaluating the Effectof IMF Lending to Low-Income Countries,” Journal of Development Economics,Vol. 61, No. 2, pp. 495–526.

Haque, Nadeem Ul, and Mohsin S. Khan, 1998, “Do IMF-Supported Programs Work? ASurvey of the Cross-Country Empirical Evidence,” IMF Working Paper 98/169(Washington: International Monetary Fund).

Hutchison, Michael M., 2001, “A Cure Worse than the Disease? Currency Crises and theOutput Costs of IMF-Supported Stabilization Programs,” NBER Working PaperNo. 8305 (Cambridge, Massachusetts: National Bureau of Economic Research).

International Monetary Fund (IMF), 2005, “The IMF’s Multilateral Surveillance: DraftIssues Paper for an Evaluation by the Independent Evaluation Office,” available onthe Web at http://www.imf.org/external/np/ieo/index.htm.

Mody, Ashoka, and Diego Saravia, 2003 “Catalyzing Capital Flows: Do IMF ProgramsWork as Commitment Devices?” IMF Working Paper 03/10 (Washington: Interna-tional Monetary Fund). Forthcoming in Economic Journal.

Musso, Alberto, and Steven Phillips, 2002, “Comparing Projections and Outcomes ofIMF-Supported Programs,” Staff Papers, International Monetary Fund, Vol. 49,No. 1, pp. 22–48.

Przeworski, Adam, and James Raymond Vreeland, 2000, “The Effect of IMF Programs onEconomic Growth,” Journal of Development Economics, Vol. 62, No. 2, pp. 385–421.

Rossi, Marco, and Alessandro Rebucci, 2005, “Measuring Disinflation Credibility UnderIMF-Supported Programs: A Bayesian Approach with an Application to Turkey”(unpublished manuscript; Washington: International Monetary Fund). An earlierversion of this paper was issued as IMF Working Paper 04/208.

xviii

Part I. Design

1

This paper examines how IMF-supported programs haveevolved over the past decade, and to what extent na-tional authorities have been successful in achieving theirprogram goals.2

Introduction

Are IMF-supported programs a success or a failure?This is a key question in evaluating the role of theIMF, and a large body of literature has set about an-swering it.3 Yet defining success, let alone measuringit, is far from straightforward. In part, this is becausethe circumstances facing member countries haveevolved over time, as has the nature of problemsthat national authorities expect to address with IMFsupport. These changing circumstances have beenaccompanied by a broad range of facilities for IMFfinancing, which are often associated with differentobjectives: in addition to the Stand-By Arrange-ment, introduced in 1952 to address short-term bal-ance of payments needs, financing is providedthrough the Extended Fund Facility (EFF), estab-lished in 1974 to address longer-term payments im-balances rooted in structural problems, and thePoverty Reduction and Growth Facility (PRGF), in-troduced in 1999, with its explicit eponymous goals.4

The paper is structured as follows. The secondsection presents the classic type of IMF-supportedprogram and discusses how IMF-supported programshave evolved from that classic type. The third sec-tion examines experience under IMF-supported pro-grams, including external adjustment and macro-

economic performance. The fourth section studiesperformance, while the fifth section draws someconclusions.

Types of IMF-Supported Programs

A useful starting point in discussing program designis the classic type of IMF-supported program. In thistype of program, a country turns to the IMF for fi-nancing because it faces a loss of international re-serves stemming from an external current accountimbalance, often in the context of poor macroeco-nomic performance—such as high inflation or lowgrowth. In this setting, the primary objective is tocorrect the current account imbalance and restorereserves to a safe level. The IMF provides temporaryfinancing while this process is taking place, both toease the adjustment burden and minimize the out-put loss, even though some temporary contractionin economic activity is unavoidable.5 Indeed,growth typically dips during the program periodowing to restrictive monetary and fiscal policies in-tended to bring aggregate demand into alignmentwith national income and the drying up of externalfinancing that leads to lower investment. As confi-dence is restored, capital flows resume and eco-nomic activity picks up, enabling the country to fi-nance its now sustainable deficits.

Although this classic adjustment paradigm re-mains surprisingly relevant, the past decade has alsoseen an important evolution in the kinds of circum-stances in which the IMF has been providing fi-nancing. These newer programs can be classifiedinto two broad categories. The first are the so-calledcapital account crisis programs, where the salient ini-tial condition is a sudden loss of private external fi-nancing, which is in the nature of a stock adjust-ment—in fact, there need not be a sizable initialcurrent account imbalance when these crisesemerge—with pervasive consequences for economicperformance.6 Typically, this loss of financing puts

3

Program Objectives and Program SuccessAtish Ghosh, Timothy Lane, Alun Thomas, and Juan Zalduendo1

1 This paper draws on analytical work undertaken as part of the2004 Program Design project, which served as background mate-rial for the review of the 2002 IMF Conditionality Guidelines.

2 The IMF-supported programs examined cover the period1995–2000. Although this period includes only the early years ofthe PRGF, during the latter years of ESAF arrangements, thesehad already been redirected toward growth and poverty-reductionobjectives.

3 This literature has been reviewed by Haque and Khan(1998); some recent contributions include Barro and Lee (2002);Dicks-Mireaux, Mecagni, and Schadler (2000); and Przeworskiand Vreeland (2000).

4 The PRGF replaced the Enhanced Structural Adjustment Fa-cility (ESAF) and the Structural Adjustment Facility (SAF),which were designed to support structural reforms in low-incomecountries.

5 The IMF also requires that the resources it provides be usedin a manner consistent with IMF purposes, which precludes theimposition of trade and payments restrictions for balance of pay-ments purposes during the program period.

6 See Ghosh and others (2002).

CHAPTER

the exchange rate under pressure, leading to theabandonment of the initial exchange rate pegregime (which characterizes these countries) and toa large currency depreciation. In the presence ofcurrency mismatches in domestic public and privatesector balance sheets, the depreciation leads to asharp contraction in economic activity, whichbrings about a large current account adjustment de-spite the fact that the initial current imbalanceswere small. In fact, the policy package in these pro-grams is not intended to bring about current ac-count adjustment, as this is already being forced bythe markets. Instead, the proximate objective is torestore confidence with a view to attenuating theloss of financing.

The second group can broadly be characterized asreform programs, in which the country’s main prior-ity is to undertake a set of structural reforms de-signed to contribute to economic growth and stabil-ity over the longer term. Again, there need not bean external imbalance in the first place, but main-taining a sustainable external position is a con-straint on policy choice.

Within the broad category of reform programs,one can identify some important and diverse group-ings. Many transition economies fit the description(whether or not they have access to private financialmarkets): beyond their initial stabilization phase,their main priorities are undertaking structural re-forms aimed at transition to the market economy,maintaining macroeconomic stability, and keeping aviable external position while they are doing so.

IMF-supported programs in low-income countriesalso have the broad features of reform programs and,though they differ from transition economies inmany respects, share a common logic of program de-sign: while the need to maintain external viability isgenerally an important constraint, the primary ob-jective is not short-run adjustment of the currentaccount but rather structural transformation of theeconomy to create a sound basis for growth. Low-income countries also have some important distinc-tive characteristics—in particular, the very long-term nature of the reform agenda and the prospectthat they will be supported by concessional lendingand grant aid over the foreseeable future.

A final type of program that fits the reform profiledeals with non-crisis emerging market countries. Thesecountries do not face acute balance of paymentspressure, in part because they continue to enjoy ac-cess to private financial markets, but they seek sup-port from the IMF to help maintain stability whilethey are undertaking other policy initiatives de-signed for longer-term growth and stability—whichmay include disinflation and various aspects ofstructural reform.

In sum, these programs are designed primarily tosupport sound policies and enhance the credibilityof the authorities’ policy programs, rather than toclose an immediate balance of payments gap. In sev-

eral emerging market countries, the purpose of suchprograms is to elicit lower inflationary expectationsand interest rates to put domestic debt dynamics ona more sustainable footing, while maintaining accessto external markets.

While the differing nature of these programsimply different objectives, all IMF-supported pro-grams share the goal of achieving (or maintaining)external viability for at least two reasons. First, a lossof external viability would eventually force anabrupt adjustment, in turn jeopardizing any otherprogram goals. Second, both economic logic and theIMF’s Article of Agreements dictate that IMF re-sources be used to meet a country’s balance of pay-ments needs. Since the use of IMF resources adds tothe country’s external obligations, which will subse-quently need to be repaid, IMF support essentiallyinvolves shifting the time profile of the country’s ex-ternal adjustment to minimize the associated eco-nomic and social disruption, including by allowingtime for the adjustment to come through a supplyresponse rather than demand management alone.

As IMF-supported programs have been designedto respond to different initial conditions than origi-nally envisaged, with a different set of objectives inmind, the analytical framework used to design themhas also evolved. In the past, the connections be-tween policies and objectives were often analyzedon the basis of projections using the financial pro-gramming framework based on the monetary ap-proach to the balance of payments—a special caseof the Mundell-Fleming model.7 In recent years, thisframework has given way to a more eclectic ap-proach, in which a variety of analytical tools are usedto analyze and predict developments in particularsectors while the financial programming frameworkis used to ensure that projections for different sectorsare consistent with key balance-sheet identities.

This eclectic approach has been viable in mostcases, given that projections are often revised after aquarter or two and over that period are typically rea-sonably accurate (Table 1, Figure 1, and Figure 2).The IMF’s analytical approach has been less success-ful in capital account crises—since neither IMF staffnor anyone else have been able to construct an ac-curate short-run model of the rapid and often mas-sive stock adjustments that take place in a crisis set-ting. The balance-sheet approach has beendeveloped to analyze these adjustments and to in-form a sharper assessment of crisis vulnerability; this

1 � PROGRAM OBJECTIVES AND PROGRAM SUCCESS

4

7 Specifically, the financial programming model assumes thatreal economic activity is determined on the supply side andwhere money demand and capital flows are interest inelastic.These assumptions are not unrealistic for many countries withlimited capital mobility, repressed financial mobility, and con-strained supply sides—which characterized many countries withIMF-supported programs in the past—but have become less rele-vant, particularly for emerging market countries. See IMF, Re-search Department (1987) on theoretical aspects of program de-sign and Polak (1991).

approach needs to be developed further to provide aclearer basis for the policy response in a crisis. A sec-ond area in which more work is clearly needed is theanalysis of long-term growth and debt dynamics:while the IMF’s debt-sustainability template pro-vides a basis for prudent assessments of the debt dy-namics, it has to contend with a continuing ten-dency toward over-optimism in medium-termgrowth projections. Deriving realistic growth pro-jections poses a significant challenge, especiallygiven that growth remains far from well understoodby the economics profession at large.8 These analyt-ical issues should be borne in mind in consideringthe experience with IMF-supported programs in theremainder of the paper.9

What Happens in IMF-Supported Programs?

With these basic types in mind, we may now considerthe macroeconomic outcomes of IMF-supported pro-grams in the 1990s. Here we are simply trying to

characterize what happened during the course of the programs and thereafter—rather than trying toestablish what would have happened in the IMF’sabsence.10

We may first examine the experience of Stand-ByArrangements. Programs supported by thesearrangements look, on average, surprisingly close tothe classic type (Figure 3). At the outset of a typicalprogram, the country faces substantial external im-balances, and these imbalances are reduced duringthe course of the program. Monetary and fiscal poli-cies are tightened to help promote the external ad-justment. Economic activity dips below trend duringthe program and subsequently recovers.

Programs adopted in the context of capital ac-count crises display similar but more extreme pat-terns (Figure 4). In these cases, large current ac-count adjustment takes place in the face of capitaloutflows. This adjustment is associated with a majorslump in economic activity, followed by a sharp re-covery. The major difference between capital ac-count crisis programs and the classic adjustment par-adigm lies in the orientation of policies: in capital

8 See, for instance, Easterly (2001).9 A fuller discussion of the analytical frameworks for program

design is presented in the set of papers on “The Design of IMF-Supported Programs,” which is available on the IMF website(http://www.imf.org/external/np/pdr/2004/eng/design.htm).

5

Ghosh, Lane, Thomas, and Zalduendo

10 In contrast, the existing literature on the macroeconomicoutcomes of IMF-supported programs typically seeks to distin-guish the role of the IMF from some counterfactual—but, on theother hand, does not focus on the fact that different types of pro-grams may have different objectives.

Table 1

Statistical Characteristics of Program Projection Errors1

Average Period Period t Period t+1 t+1: t+3

Number of Mean Mean MeanObservations error RMSE error RMSE error RMSE

Real GDP Growth

PRGF-supported programs 56 –0.4 2.2 –1.2*** 3.1 –1.3*** 2.8GRA-supported programs

Transition countries2 27 0.1 3.7 –0.7 4.4 –0.5 2.9Non-transition countries2 35 –0.3 2.7 –0.7 4.4 –2.4*** 4.3CACs3 9 –9.3*** 10.7 –0.9 5.0 –0.4 2.3

Current Account Balance

PRGF-supported programs 48 –1.5** 4.9 –1.9*** 4.8 –2.2*** 4.1GRA-supported programs

Transition countries2 24 0.4 2.5 –0.4 2.3 –0.8 2.6Non-transition countries2 28 2.4*** 4.6 2.7*** 5.2 1.6*** 3.2CACs3 8 5.6** 7.2 6.5** 8.7 … …

Sources: IMF, Monitoring of Fund Arrangements (MONA) and World Economic Outlook (WEO) databases; and IMF staff estimates.Notes: *** significant at 1 percent level; ** significant at 5 percent level; and * significant at 10 percent level. PRGF denotes the Poverty Re-

duction and Growth Facility. RMSE denotes root-mean-squared error. GRA denotes the IMF’s General Resources Account.1 Data have been transformed so that they map into the interval (–100, 100). Errors are defined as actual minus projections. Table was con-

structed using a dataset of countries with available information for years t, t+1, t+2, and t+3.2 Excludes capital account crises.3 CAC stands for capital account crisis.

6


–15

–10

–5

0

5

10

–15 –10 –5 5 10 15

–15

–10

–5

0

5

10

–15 –10 –5 5 10 15

–15

–10

–5

0

5

10

15

–15 –10 –5 5

–15

–10

–5

0

5

10

15

–15 –10 –5 5

Real GDP Growth (in percent)

Period t Average Period t +1: t+3

Period t Average Period t+1: t +3

Current Account Balance (percent of GDP)

45degree

line

45degree

line

45degree

line

45degree

line

Figure 1

Growth and Current Account Balances: Projections and Actuals(x-axis projections; y-axis actuals; GRA programs only)

Sources: IMF, World Economic Outlook (WEO) and Monitoring of Fund Arrangements (MONA) databases; and IMF staff estimates.Notes: Data were mapped into the interval (–100, 100). GRA denotes the IMF's General Resources Account. Capital account crisis countries

are depicted by triangles.

7


Real GDP Growth (in percent)

Period t Average Period t +1: t +3

Period t Average Period t+1: t +3

Current Account Balance (percentage of GDP)

–10

–5

0

5

10

–5 5 10

45degree

line

45degree

line 45degree

line

45degree

line

–10

–5

0

5

10

–5 5 10

–30

–25

–20

–15

–10

–5

0

5

10

–20 –15 –10 –5 5

–30

–25

–20

–15

–10

–5

0

5

10

–20 –15 –10 –5 5

Figure 2

Growth and Current Account Balances: Projections and Actuals(x-axis projections; y-axis actuals; PRGF programs only)

Sources: IMF, World Economic Outlook (WEO) and Monitoring of Fund Arrangements (MONA) databases; and IMF staff estimates.Notes: Data were mapped into the interval (–100, 100). PRGF denotes the Poverty Reduction and Growth Facility.

8


–1.0

0.0

1.0

2.0

3.0

4.0

5.0

6.0

t–3 t–2 t–1 t t+1 t+2 t+3 t–3 t–2 t–1 t t+1 t+2 t+3

t–3 t–2 t–1 t t+1 t+2 t+3

t–3 t–2 t–1 t t+1 t+2 t+3t–3 t–2 t–1 t t+1 t+2 t+3

t–3 t–2 t–1 t t+1 t+2 t+3

6.0

8.0

10.0

12.0

14.0

16.0

Realexchange rate

index

Reserves inmonths of

imports(right scale)

95

100

105

110

115

120

3.0

3.5

4.0

4.5

5.0

Currentaccount

Investment(right scale)

Savings (rightscale)

–4.0

–3.0

–2.0

–1.0

0.0

1.0

15

16

17

18

19

20

Overallbalance

Primarybalance

–6

–4

–2

0

2

4

6

8

10

Broad money growth

Real interest rate (right

scale)

0

5

10

15

20

25

0

2

4

6

8

10

12

Real GDP (growth, in percent per year) Inflation (in percent per year)

Reserves and Real Exchange Rate (index t=100) Savings, Investment, and Current Account (percent of GDP)

Central Government Balances (percent of GDP) Real Interest Rate and Broad Money Growth (percent)

Figure 3

Macroeconomic Performance Under GRA-Supported Programs, 1995–2003(Excluding transition economies)

Sources: IMF, World Economic Outlook (WEO) database; and IMF staff estimates.Notes: GRA denotes the IMF's General Resources Account. Standard error bands for real GDP growth, inflation, and government balances are

shown by the dotted lines.

9


–8

–5

–2

1

4

7

10

t–3 t–2 t–1 t t+1 t+2 t+3

t–3 t–2 t–1 t t+1 t+2 t+3

t–3 t–2 t–1 t t+1 t+2 t+3 t–3 t–2 t–1 t t+1 t+2 t+3

t–3 t–2 t–1 t t+1 t+2 t+3

t–3 t–2 t–1 t t+1 t+2 t+3

0

5

10

15

20

25

30

Real exchange rate

Reserves inmonths of

imports (right scale)

80

100

120

140

160

3

4

5

6

7

8

Current account

Savings(right scale)


–6

–4

–2

0

2

4

16

17

18

19

20

21

22

Overallbalance

Primarybalance

–8

–6

–4

–2

0

2

4

Broad money growth

Real interest rate (right

scale)

10

20

30

40

50

0

2

4

6

8

10


Reserves and Real Exchange Rate (index t=100) Savings, Investment, and Current Account (percent of GDP)


Figure 4

Macroeconomic Performance Under Capital Account Crisis Countries, 1995–20031

(Excluding transition economies)

Sources: IMF, World Economic Outlook (WEO) database; and IMF staff estimates.1 Standard error bands for real GDP growth, inflation, and government balances are shown by the dotted lines.

account crises, fiscal policy is loosened to supporteconomic activity (except, of course, in those crisesthat are driven by public sector imbalances), sincethe external adjustment is being forced on the coun-try through capital outflows anyway. Monetary pol-icy is tightened, although this is done mainly to at-tract capital rather than to foster adjustmentthrough demand restraint.

Macroeconomic developments in low-incomecountry programs supported by IMF concessional fa-cilities display quite a different pattern (Figure 5). Inthose cases, while external deficits were quite large,little current account adjustment took place. How-ever, in this set of programs, growth increased dur-ing the course of the program. On the negative side,these countries tended to experience an increase inexternal debt, corresponding to the continuing ex-acerbation of low-income countries’ debt problemsduring this period.

Performance

External Adjustment

How well did programs succeed? Since IMF finan-cial support is intended to bring about orderly exter-nal adjustment, and a failure to maintain externalviability is likely to jeopardize any other programgoals, a good starting point is the record on externaladjustment. But even here, establishing what shouldconstitute a “successful” program is not straightfor-ward, since it is not only the overall amount but alsothe means and the time path of external adjustmentthat are important.

The importance of the time path of adjustmentcan be illustrated with reference to two extremecases: Argentina (1995) and the Republic of Korea(1997). In the aftermath of the “tequila” crisis, in1995 Argentina faced large bank deposit with-drawals, putting severe pressure on the balance ofpayments and calling into question the viability ofthe currency board arrangement. In the event, theauthorities were able to stabilize capital outflows, adevaluation was avoided, and the government waseven tapping the capital markets by year’s end. Theprogram was thus very successful in dealing with theimmediate balance of payments problem. Yet, in ret-rospect, it is also clear that the failure to tackle theunderlying weaknesses of the public finances re-sulted in a mounting public and external debt prob-lem, culminating in the 2002 crisis.11 In contrast,Korea’s 1997 program met with very little initialsuccess in stemming capital outflows or preventing acollapse of the exchange rate and of economic ac-tivity. Over the longer term, however, by enhancing

the credibility of policies and instituting structuralreforms, the IMF-supported program succeeded inrestoring confidence and bringing about a return ofprivate capital together with a replenishment of for-eign exchange reserves. Neither extreme is optimal:Korea achieved a sharp reduction in its externaldebt, but at a cost of a wrenching external adjust-ment and contraction of output. In Argentina, theimpact of adjustment was avoided in the shortrun—but at the cost of a highly disruptive crisis inthe longer run.

This suggests that any measure of successful exter-nal adjustment must weigh the benefits of attenuat-ing adjustment in the short run against the longer-run considerations of maintaining external viability.Medium-term debt sustainability provides one suchmeasure. The basic principle is that if a country is sol-vent, it should be able to obtain financing ratherthan having to adjust its current account in responseto a temporary shock. Therefore, unless the country isconstrained in the financing it is able to obtain, thecurrent account balance should adjust by as much asis required to maintain solvency—except inasmuchas it has a high level of external debt or low level ofreserves, in which case a larger surplus (or smallerdeficit) would be appropriate to reduce vulnerabilityto future balance of payments problems.

Programmed and actual current account balances(relative to the debt-stabilizing balance) for middle-income countries (those supported by Stand-By andExtended Arrangements) are plotted against theinitial external debt ratio in Figure 6. Three pointsare apparent. First, consistent with reducing vulner-ability, there is indeed a positive relationship be-tween the initial level of external debt and the tar-geted improvement in the current account balance(relative to the debt-stabilizing balance). Second,actual adjustment was generally greater than theprogrammed adjustment—on average by about1 percent of GDP. Third, in a large number of cases,adjustment was greater than would be required tostabilize the debt ratio. While it is difficult to estab-lish exact thresholds at which debt should be con-sidered high, most studies suggest a range of about40–60 percent of GDP for developing and emergingmarket countries (Figure 6). Segmenting the figureaccording to whether the country undertook moreadjustment than necessary to stabilize the debt ratioand whether the initial debt ratio exceeded 40–60percent of GDP shows that in about one-quarter ofthe cases, the current account balance was higherthan necessary to stabilize the debt ratio eventhough debt was below 40 percent of GDP (in thesecond section in Figure 6). In a further 21 percentof cases—including some notable capital accountcrises such as Korea (1997) and Mexico (1995)—current account balances exceeded the debt-stabilizing balance although the initial debt ratiowas within the 40–60 percent of GDP range (in the second section in Figure 6). Finally, in about

10


11 Daseking and others (2005).

11


0.0

1.0

2.0

3.0

4.0

5.0

t–3 t–2 t–1 t t+1 t+2 t+3

0.0

5.0

10.0

15.0

20.0

25.0

30.0

t–3 t–2 t–1 t t+1 t+2 t+3

Realexchange

rate

Reserves inmonths

of imports(right scale)

90

95

100

105

t–3 t–2 t–1 t t+1 t+2 t+3

2

3

4

5

Currentaccount


Savings (right scale)

–10

–9

–8

–7

t–3 t–2 t–1 t t+1 t+2 t+3

8

10

12

14

16

18

Overallbalance

Primarybalance

–6

–5

–4

–3

–2

–1

0

t–3 t–2 t–1 t t+1 t+2 t+3

Broad money growth

Real interestrate (right

scale)

5

10

15

20

25

t–3 t–2 t–1 t t+1 t+2 t+3

2

4

6

8

10

12


Reserves and Real Exchange Rate (index t =100) Savings, Investment, and Current Account (percent of GDP)


Figure 5

Macroeconomic Performance Under Enhanced Structural Adjustment and PovertyReduction and Growth Facility Programs, 1995–20031

Sources: IMF, World Economic Outlook (WEO) database; and IMF staff estimates.1 Standard error bands for real GDP growth, inflation, and government balances are shown by the dotted lines.

12


MEX 95

HTIUKR 95

EST 95

RUS 95LVA 95

MKD

PNG

SLV 95GAB

CRI PAN 95

URY 96

HUNRUS 96

ARG 96

UKR 96

LVA 96PER 96

VEN

BGR 96

EST 96

EGY

SLV 97

HRV

BGR 97

ROM

URY 97THA

UKR 97

LVA 97KOR

PAN 97

EST 97

ARG 98

CPV

PHLZWE

UKR 98

SLV 98BGR 98

BRA URY 98

JOR 98

PER 99

MEX 99

RUS 99

ZWE

LVA 99

KAZ

COL

TUR

ARG 00

ECU

EST 00

GAB

LTUMKD 00

NGA

PAK

PAN 00

PNG

URY 00

y = –1.06 + 0.06x(R2 = 0.06)

–10

–5

0

5

10

15

20

25

30

External debt

Proj

ecte

d m

inus

deb

t-st

abili

zing

cur

rent

acc

ount

URY 00

PNG

PAN 00

PAK

NGA

MKD 00LTU

GAB

EST 00

ECU

ARG 00

TUR

COL

KAZ

LVA 99

ZWE

RUS 99

MEX 99

PER 99

JOR 98

URY 98BRA BGR 98SLV 98

UKR 98

ZWE

PHL

CPV

ARG 98

EST 97

PAN 97

KOR

LVA 97UKR 97

THA

URY 97

ROM

BGR 97

HRV

SLV 97

EGY

EST 96

BGR 96

VEN

PER 96LVA 96

UKR 96

ARG 96 RUS 96

HUN

URY 96PAN 95

CRI

GAB

SLV

PNG

DZA

MKD

LVA 95

RUS 95

EST 95

UKR 95

HTI

MEX 95

y = –1.83 + 0.11x (R2 = 0.10)

–10

–5

0

5

10

15

20

25

30

External debt

Actu

al m

inus

deb

t-st

abili

zing

cur

rent

acc

ount

III(31%)

VI(13%)

IV(5%)

I(20%)

II(20%)

III(31%)

IV(5%)

V(8%)

VI(16%)

II(21%)

I(23%)

V(7%)

0 20 30 40 50 60 70 80 90 100

0 20 30 40 50 60 70 80 90 100

Figure 6

Projected, Actual, and Debt-Stabilizing Current Account Balances in GRA-Supported Programs(In percent of GDP)

Sources: International Monetary Fund, Monitoring of Fund Arrangements (MONA) and World Economic Outlook (WEO) databases; and IMFstaff estimates.

Notes: Country abbreviations are based on three-letter country codes used in IMF automated database systems. GRA denotes the IMF's General Resources Account. Capital account crisis countries are depicted by triangles.

30 percent of cases, current account balances ex-ceeded the debt-stabilizing balance—but the countrywas starting from a high initial level of indebtedness.

To some extent, countries may have run largercurrent account balances than necessary to stabilizethe debt ratio in order to build up reserves to reducevulnerability to future crises. While this is part of theexplanation, it does not account for it fully, since theexcess current account balance (for countries in thefirst section) was, on average, almost 3 percent ofGDP—against a programmed increase in reserves of11⁄2 percent of GDP. For countries in the second sec-tion, the difference is even more dramatic—8.7 per-cent of GDP against a programmed increase in re-serves of 11⁄2 percent of GDP, although a number ofthese countries are oil exporters that enjoyed a posi-tive terms of trade boost from higher oil prices. Assuch, capital outflows are likely to have forced someof these countries to adjust their current accountbalances by more than would be indicated by con-siderations of debt sustainability.

IMF-supported programs in low-income countriespresent a sharply contrasting picture of external ad-

justment. First, the positive relationship betweenprogrammed current account adjustment and thecountry’s initial external debt characteristic of themiddle-income countries is not evident for the low-income countries—indeed, the relationship isslightly negative (Figure 7). Second, actual currentaccount adjustment typically fell short of even thisplanned adjustment—often because of delays in dis-bursements of external grants. Facing such a delay,however, not only did these countries not adjust theircurrent account deficits but, on average, they bor-rowed more than the corresponding shortfall. Whilethese results may be consistent with the primary pur-poses of these programs to promote growth and re-duce poverty, they also imply a growing debt burdenin the absence of debt relief—corresponding to thepersistent debt problems of low-income countries.

Macroeconomic Performance

Beyond external adjustment, national authoritiesoften have a number of other objectives in theireconomic programs for which they are seeking IMF

13


Table 2

Macroeconomic Performance of Countries with IMF-Supported Programs1

Number of Three Years Program Three Years Observations Before Program Period2 After Program

PRGF-Eligible Countries3

Inflation3

1980–91 169 68 26 731992–2002 62 28 16** 8*

Real GDP growth3

1980–91 169 2.1 2.6 2.41992–2002 62 2.4 4.1* 3.4**

Standard deviation of growth1980–91 169 3.9 3.4 3.21992–2002 62 3.0 3.0 2.4

Non-PRGF-Eligible Countries2

Inflation3

1980–91 104 42 72 551992–2002 51 80 39 27**

Real GDP growth3

1980–91 104 2.7 2.0 3.11992–2002 51 3.6 2.7 3.4

Standard deviation of growth1980–91 104 3.7 n.a. 3.51992–2002 51 3.0 n.a. 3.0

Sources: IMF, Monitoring of Fund Arrangements (MONA) and World Economic Outlook (WEO) databases; and IMF staff estimates.1 Average annual growth rates over 3-year periods unless otherwise specified.2 PRGF-eligible countries: program period is three years and includes the year in which the program begins. Non-PRGF-eligible countries:

program period is one year—the year the program begins.3 Statistical significance of the average rate relative to the pre-program average rate; * at 1 percent; ** at 10 percent; n.a. stands for “not ap-

plicable.” PRGF denotes Poverty Reduction and Growth Facility.

14


GNB

MRT

GHATCD

MWI

ZMBMLI

KEN

NERBFA

BEN

ETH

HTI

TZA

MDG

GIN

MNG

CMR

PAK

UGA

CIV

NIC

SEN

ALB

RWA

GMB

GUY

CAF

BOL

HND

GHA MRT

BFADJI

KHM

ZMB

MOZMLI

TZA

TCD

BEN

KEN

NER

MWI

CMR

ARM

GEO

AZE

KGZ

TJK

YEM

GNB

MDA

y = –2.80 –0.06x(R

2 = 0.30)

–40

–30

–20

–10

0

10

20

0 50 100 150 200 250 300 350 400

External debt

Proj

ecte

d m

inus

deb

t-st

abili

zing

cur

rent

acc

ount

GNB

MRT

GHATCD

MWI

ZMB

MLI

KEN

NERBFA

COG

BEN

ETH

HTI

TZA

MDG

GIN

MNGCMR

PAK

UGA

CIV

NIC

SEN

ALB

RWA

GMB

GUY

CAF

BOL HND

GHAMRT

BFADJI

KHM

ZMB

MOZMLI

TZA

TCDBEN

KEN

NER

MWI

CMR

ARM

GEOAZE

KGZ

TJK

YEM

GNB

MDA

y = 0.41– 0.05x (R2 = 0.32)

–50

–40

–30

–20

–10

0

10

20

0 50 100 150 200 250 300 350 400

External debt

Actu

al m

inus

deb

t-st

abili

zing

cur

rent

acc

ount

Figure 7

Projected, Actual, and Debt-Stabilizing Current Account Balances in Poverty Reduction andGrowth Facility (PRGF)-Supported Programs

Sources: IMF, Monitoring of Fund Arrangements (MONA) and World Economic Outlook (WEO) databases; and IMF staff estimates.Notes: Country abbreviations are based on three-letter country codes used in IMF automated database systems. Non-HIPCs (highly indebted

poor countries) are depicted by triangles.

support. These typically include lowering inflationand achieving macroeconomic stability, raisinggrowth, and promoting poverty reduction. Withoutattributing outcomes to the IMF’s support, Table 2reports macroeconomic performance under coun-tries’ programs. Among the middle-income coun-tries, during the 1990s, annual inflation fell from anaverage of 80 percent to less than 30 percent for thethree years following the program. This contrastswith the experience of the 1980s, when inflation ac-tually rose during the program period—and re-mained higher in the years following the programthan it had been previously. Consistent with the clas-sic adjustment paradigm, growth recovers to its pre-program rates, but there is not a marked improve-ment in the growth performance of the country.

Among low-income countries, in contrast, notonly did inflation come down significantly, growthimproved as well. Indeed, during the program pe-riod, annual growth was about 11⁄2 percentage pointshigher, and in the three years following the program,about 1 percentage point per year higher, than thepre-program rate. Applying standard cross-countrygrowth regressions suggests that the coincidence ofbetter inflation performance and higher growth isnot coincidental—improved growth in low-incomecountries can be explained by better macroeco-nomic policies (as captured by inflation and smallerfiscal deficits) as well as a more benign external anddomestic environment (Table 3). At the same time,inasmuch as external resource flows have helpedthese countries contain their deficits (after grants)

15


Table 3

Explaining Growth in PRGF Countries(1992–2002, average annual growth rates)

Three Years Three Years Coefficient Preceding Three-Year FollowingEstimates Program Program Program

Real GDP per capita growth –0.56 1.34 0.68

Change in per capita growth 1.90 –0.66

Contributing factorsG-7 real GDP growth 0.6828*** 0.50 0.10Initial conditions

Of whichInitial income level –0.0595*** –0.27 –0.37Fertility rates –0.0090 0.25 0.24

Macro policiesOf whichInflation –0.0376*** 0.02 0.22Fiscal balance 0.1360*** 0.29 –0.01

Structural reforms 0.0285** 0.08 0.01Shocks (internal and external)

Of whichDomestic shocks –0.0409*** 0.50 –0.06Terms of trade 0.0247 0.08 –0.02

Constant 0.0052* 0.52 0.52

Unexplained –0.08 –1.29

Number of observations 162 46 46Number of countries 46 31 31

R-squared adj. 0.23F-statistic 26.21***Standard error of regression 0.039

Sources: IMF, Monitoring of Fund Arrangements (MONA) and World Economic Outlook (WEO) databases; and IMF staff estimates.Notes: Asterisks indicate statistically significant coefficient estimates; *** at 1 percent, ** at 5 percent, and* at 10 percent. PRGF denotes

Poverty Reduction and Growth Facility. G7 denotes the Group of Seven.

and foreign borrowing has limited their recourse toinflationary finance, it remains an open questionwhether this improved performance can be main-tained without an unsustainable buildup of debt orgreater grant financing.

Conclusions

This paper contrasts the classic framework of pro-gram design and the growing diversity of countrycircumstances and program objectives for whichIMF-supported programs are now being designed. Inview of this diversity, the concept of a financial pro-gram has changed. In the classic type of program,IMF financing enables a country to undertakeneeded external adjustment in a more gradual andorderly manner than would be possible in the ab-sence of such support. The program achieves thecountry’s objectives and safeguards IMF resources byensuring that the financial flows implied by thepolicies envisaged will enable the country to repaythe IMF on the arranged schedule.

Although this classic adjustment paradigm re-mains—perhaps surprisingly—relevant, the pastcouple of decades have also seen the emergence ofprograms with different characteristics. Foremostamong these are capital account crises, where capi-tal outflows force external adjustment on the coun-try, and IMF-supported programs in low-incomecountries, where promoting growth and reducingpoverty are the key objectives.

These differing objectives and characteristics ofIMF-supported programs are reflected in outcomes.Defining program success requires a metric for judg-ing the appropriate degree of adjustment; medium-term debt sustainability may provide such a metric.Among middle-income countries, especially but notexclusively in capital account crises, capital out-flows (or, more generally, a lack of sufficient exter-nal financing) sometimes forced larger external ad-justments than should be required for debtsustainability considerations. This experience raisesimportant questions of whether larger official fi-nancing packages or action to “bail in” private cred-itors, or both, could result in better outcomes.

In low-income countries, the experience wasquite different: adjustment typically fell short bothof the programmed current account adjustment andthe adjustment required to stabilize external debt ra-tios. In large part, this reflects the goals of these pro-grams—promoting growth and reducing poverty.However, to the extent that the growth came at theprice of mounting debt, it nevertheless raises ques-

tions about the longer-term viability of this ap-proach. This experience argues for an increase in thegrant element of external financing for low-incomecountries.

ReferencesBarro, Robert, and Jong-Wha Lee, 2002, “IMF Programs:

Who Is Chosen and What Are the Effects?” NBERWorking Paper No. 8951 (Cambridge, Massachu-setts: National Bureau of Economic Research).

Daseking, Christina, Atish Ghosh, Timothy Lane, andAlun Thomas, 2005, Lessons from the Crisis in Ar-gentina, IMF Occasional Paper No. 236 (Washington:International Monetary Fund).

Dicks-Mireaux, Louis, Mauro Mecagni, and SusanSchadler, 2000, “Evaluating the Effect of IMF Lend-ing to Low-Income Countries,” Journal of Develop-ment Economics, Vol. 61 (April), pp. 495–526.

Easterly, William, 2001, The Elusive Quest for Growth:Economic Adventures and Misadventures in the Tropics(Cambridge, Massachusetts: MIT Press).

Ghosh, A., T. Lane, M. Schulze-Ghattas, A. Bulír, J.Hamann, and A. Mourmouras, 2002, IMF-SupportedPrograms in Capital Account Crises, IMF OccasionalPaper No. 210 (Washington: International MonetaryFund).

Haque, Nadeem Ul, and Mohsin S. Khan, 1998, “DoIMF-Supported Programs Work—A Survey of theCross-Country Empirical Evidence,” IMF WorkingPaper 98/169 (Washington: International MonetaryFund).

International Monetary Fund, Research Department,1987, Theoretical Aspects of the Design of Fund-Supported Adjustment Programs: A Study, IMF Occa-sional Paper No. 55 (Washington: InternationalMonetary Fund).

International Monetary Fund, 2004a, “The Design ofFund-Supported Programs—Overview” (unpublished;Washington: International Monetary Fund).

______, 2004b, “Fund-Supported Programs—Objectivesand Outcomes” (unpublished; Washington: Interna-tional Monetary Fund).

______, 2004c, “Policy Formulation, Analytical Frame-works, and Program Design” (unpublished; Washing-ton: International Monetary Fund).

______, 2004d, “Macroeconomic and Structural Policiesin Fund-Supported Programs—Review of Experi-ence” (unpublished; Washington: InternationalMonetary Fund).

Polak, Jacques, 1991, “Changing Nature of IMF Condi-tionality,” OECD Technical Paper No. 41 (Paris: Or-ganization for Economic Cooperation and Develop-ment), pp. 7–82.

Przeworski, J.A., and J. Vreeland, 2000, “The Effect ofIMF Programs on Economic Growth,” Journal of De-velopment Economics, Vol. 62, pp. 215–76.

16


2

IMF-supported programs focus on key objectives (suchas growth, inflation, and the external current account)and on intermediate policy targets (such as monetaryand fiscal policies) needed to achieve these objectives. Inthis paper, we use a new, large dataset, with informationon 94 programs between 1989 and 2002, to compareprogrammed objectives and policy targets to actual out-comes. We report two broad sets of results. First, wefind that outcomes typically fell short of expectations ongrowth and inflation but were broadly in line with theprogrammed external current account objectives. Simi-larly, programmed intermediate policy targets were gen-erally more ambitious than the policy outcomes. Second,and focusing on growth, we examine the relationship be-tween objectives and policy targets, and find differencesin the way ambitious monetary and fiscal targets affectedthe achievement of the growth objective. On the onehand, more ambitious fiscal targets, even when theywere missed, led to better growth performance. On theother hand, more ambitious monetary targets tended tobe associated with lower growth performance.

Introduction

IMF-supported programs are often described bythose on the left as creating hardships on the popu-lation because they are said to be “too tight”(Stiglitz, 2002). Those on the right frequently dis-parage the objectives that were set in the programsbut were not achieved. These criticisms refer to theintermediate targets set in IMF-supported programsin the areas of monetary and fiscal policy, as well asto the macroeconomic outcomes—such as inflation,employment, and growth. Are both groups correct?Is there any validity to these criticisms? Or, are thebenchmarks by which IMF programs judged simplymisplaced?

Defenders of IMF-supported programs wouldargue that the programmed objectives and targetsshould not be viewed as forecasts. The objectives areset high so that countries can aspire to achievethem. Similarly, targets are set tight to ensure thatpolicy slippages are kept to a minimum. If targets aremissed either because of negative exogenous shocksor because the programs were set too tight, mecha-nisms in IMF policies and procedures exist to pro-vide waivers for missing these targets. As a matter offact, ample evidence exists on the waivers given inIMF-supported programs to ensure that IMF loandisbursements are not interrupted unless a majorpolicy slippage occurs. This raises the question ofwhether tight policy targets and ambitious objec-tives are deliberate. Also, if they are deliberate, dothey help countries achieve better outcomes thanthey could otherwise?

In an earlier paper based on a much smaller sam-ple size (Baqir, Ramcharan, and Sahay, 2003), wefound that (a) IMF-supported programs were, in-deed, optimistic—in particular, programmed objec-tives on inflation and growth were often not fullyachieved; and (b) meeting the fiscal target was asso-ciated with meeting the growth target. Given thesmall sample of 29 countries in that paper, however,we were unable to report conclusive results and, inparticular, to explore systematically the relationshipbetween objectives and policy targets.

In this paper, we expand the dataset used in Baqir,Ramcharan, and Sahay (2003) to 94 countries andconfirm our previous findings on the optimism ongrowth projections in IMF-supported programs. Wethen compare the programmed and actual values ofintermediate policy targets and objectives sepa-rately, and uncover systematic patterns. We also ex-plore the relationship between the intermediatepolicy targets and the objectives to understand whythere are persistent shortfalls in achieving some ob-jectives. On the latter, we focus on a recurrent find-ing in reviews of IMF-supported programs—the rel-atively poor performance on meeting the growthobjective—by looking at the main intermediate pol-icy targets in monetary and fiscal policy to explorethese questions.

This paper is organized as follows. The next sec-tion discusses the IMF’s financial programming

17

IMF Programs and Growth:Is Optimism Defensible?

Reza Baqir, Rodney Ramcharan, and Ratna Sahay1

1 The paper was presented at the IMF’s Annual Research Con-ference held in Washington on November 4–5, 2004, the proceed-ings of which are to be published in IMF Staff Papers. The authorsare grateful to the conference discussant, Michael Clemens, for hiscomments; Russell Kincaid, Alessandro Rebucci, and CarlosVégh for helpful discussions; Manzoor Gill for excellent researchassistance; and Maansi Sahay Seth for her editorial suggestions.

CHAPTER

framework. The third section describes the data.The fourth section systematically compares pro-grammed objectives and policies with their actualoutcomes. We examine the frequency with whichprogram objectives are met simultaneously. We alsolook at the extent of adjustments that are pro-grammed in different types of IMF-supported pro-grams (Stand-By Arrangements and arrangementsunder the Poverty Reduction and Growth Facility)to see whether the adjustments differ across thesegroups. In the fifth section, we examine the rela-tionship between objectives and fiscal and monetarypolicy targets, respectively. The sixth section con-cludes the paper.

The IMF’s Financial Programming Framework

The relationship between intermediate policy tar-gets (such as the fiscal balance and monetary aggre-gates) and macroeconomic outcomes (such as infla-tion and growth) in IMF-supported programs isderived from the monetary approach to the balanceof payments. In turn, this approach produces aframework known as financial programming, whichuses a series of macroeconomic accounting identi-ties to link economic growth, inflation, the moneysupply, the external current account, the budgetdeficit, and other macroeconomic variables.2

The intermediate policy targets derived withinthe financial programming framework, such as do-mestic credit and the fiscal balance, are designed tobe consistent with the set of macroeconomic objec-tives—such as growth, current account adjustment,and inflation—chosen to help resolve the country’seconomic difficulties.3 In other words, countriesthat meet the intermediate policy targets shouldconditionally expect to achieve the macroeconomicoutcomes that underlie these targets.

To illustrate the financial programming approach,consider the classical money equation:

MV = PY

where M is money supply, V is velocity, P is the ag-gregate price level in the economy, and Y is the ag-gregate output. Typically, objectives are first estab-lished for inflation and growth, yielding P and Y.Next—and importantly—an assumption on veloc-ity is made to arrive at the level of money supply

consistent with program objectives. Money creationin excess of this amount would be inflationary. Inpractice, velocity is often chosen either by examin-ing its historical pattern and making some assump-tion about how it is likely to be affected by particu-lar factors in the near future and/or by estimatingmoney demand functions. With money supply pro-grammed, and given an external target on the netforeign assets of the country, the banking system’sbalance sheet yields the maximum tolerable level ofnet domestic assets:

∆NDA = ∆M − ∆NFA.

Given the balance of payments objective underly-ing the IMF-supported program, the assumption onvelocity therefore directly affects the scope forcredit creation in the economy. Programminghigher velocity reflects an assumption that moneydemand will be low. In the event that money de-mand is higher than expected, tight money woulddrive up interest rates and constrain real activity inthe economy, thereby affecting the growth outcome.

Net domestic assets can, in turn, be decomposedinto net credit to the private sector (CPS), netcredit to the government (NCG), and other itemsnet (OIN):

∆CPS + ∆NCG + ∆OIN = ∆M − ∆NFA.

This equation gives the other set of relationships between fiscal policy and real activity. Once veloc-ity has been set and the external objective chosen, ahigher government deficit financed by the bankingsystem would crowd out credit to the private sector.And to the extent that private sector credit facili-tates investment, such crowding out would affectreal output.4 We use these relationships to examine,in the empirical section that follows, how assump-tions on velocity and programmed fiscal adjust-ments affect growth outcomes.

Data

The data for this paper have been assembled froman internal IMF database on IMF-supported pro-grams. In the sampling methodology, a unit of obser-vation is defined as a program country-year: a calen-dar year in which disbursements were made to aparticular country. Before disbursements are made, adocument known as a staff report is issued and dis-cussed at a meeting of the Executive Board, the bodythat decides IMF policy and approves IMF-supportedprograms. As their name suggests, staff reports contain the IMF staff ’s assessment of a country’seconomic situation and policies. They include the

2 � IMF PROGRAMS AND GROWTH: IS OPTIMISM DEFENSIBLE?

18

2 Underlying these identities are several behavioral relation-ships. Depending on data availability, IMF country desk econo-mists estimate relationships—the typical ones include money demand functions, export and import functions, and investmentand saving functions.

3 Additional performance criteria are often set on structural re-forms. These are not derived directly from the financial program-ming framework but are meant to be consistent with, and sup-port, the policy targets.

4 The trade-off with private sector credit would be correspond-ingly less if the deficit were financed from nonbank or externalfinancing.

program’s intermediate policy targets and theirmacroeconomic counterparts that are meant to cor-rect the particular problem(s) that prompted thecountry to seek IMF assistance. After each such Ex-ecutive Board meeting, the data in the staff reporton the key macroeconomic indicators are recordedin the database.

Typically there are several Board meetings on acountry’s program in a given year. The staff report is-sued for each successive meeting contains an up-dated set of historical and programmed/projecteddata on key macroeconomic indicators. As such,there are several vintages of the programmed valuesfor any variable of interest. We make use of the in-formation in the evolving forecasts/programs byrecording the programmed values for a variable xt inyears t, t –1, t–2, and t–3 from the most recent staffreport in that particular year.

Data on outcomes are generally not released untilafter the end of the year. We therefore define thewithin-year horizon as the forecast made for xt inyear t. Similarly, a one-year horizon is defined as thevalue programmed for xt in year t–1. For most em-pirical work, we focus on up to two-year horizons,since the number of observations declines sharply asthe horizon length increases. We measure the actualas the most recent historical observation availableon a particular variable for the entire set of staff re-ports for a country. For example, we record the actual fiscal balance for 1995 as that contained in astaff report dated 1998 if that particular report is the most recent available in the database for thatcountry.

Conceivably, we could expand our data on actualoutcomes by combining these data with other popu-lar databases, such as the IMF’s Government Fi-nance Statistics (GFS) or International FinancialStatistics (IFS). However, aside from growth and in-flation, which are generally measured in the sameway across databases, nearly all other variables of in-terest in the areas of monetary, fiscal, and externalpolicies can potentially be measured in differentways across databases. This is particularly true forfiscal policy targets—indeed, staff report data on fis-cal measures are often somewhat different fromthose reported in GFS. Hence, to avoid contaminat-ing our data, we focus only on actual outcomes asrecorded in the staff reports.

To facilitate our analysis by type of program, wedivide all programs into three groups—the Stand-ByArrangements (SBAs), a subset of SBAs that we call“high-profile” SBAs, and arrangements under thePoverty Reduction and Growth Facility (PRGFs).Borrowings under the SBAs are typically for shorterperiods and carry higher rates of charge than thoseunder the PRGF. The high-profile SBAs are distin-guished from other SBAs by the greater amounts ofaccess they provide to the IMF’s resources—they arealso typically covered prominently by the media.We defined “large access” as all programs in thedatabase with access exceeding two billion Special

Drawing Rights (SDRs).5 The list of large-accesscountries in our sample consists of Argentina,Brazil, Indonesia, the Republic of Korea, Mexico,the Russian Federation, Thailand, Turkey, andUruguay.

The universe of our data consists of 94 countriesfor the years 1989–2002. The number of observa-tions varies by country for each variable. Table 1shows the distribution of available observations onactuals for key variables we use in the empiricalwork. On average, we have about 7–8 observationsper country, which allows us to capture significantvariation, both across countries and within coun-tries, over time. We exploit both dimensions of thisvariation in the empirical work discussed later inthis paper. The corresponding number of observa-tions available for forecasts is considerably smaller.For example, a one-year growth forecast is availablefor 495 country-years, compared with 776 country-years for actuals.

Objectives and Targets: Programmed Versus Actual

To evaluate IMF-supported programs, it is of centralinterest to know whether both the objectives andthe policy targets were met. If the objectives werenot met (in either direction), it could suggest thatprograms were either not sufficiently ambitious ortoo ambitious. If the policy targets were not met (ineither direction), it suggests either that policy effortsby the borrowing countries were insufficient or thatthe government exceeded its targets. If the policytargets were met but objectives were not (and viceversa), it may imply that the IMF program designwas faulty or that the targets and objectives were in-consistent.6

Table 2 and Figures 1–3 summarize the pro-grammed and actual outcomes for the main eco-nomic objectives in IMF-supported programs—theIMF’s Articles of Agreement suggest that the mostimportant goals include inflation, growth, and ex-ternal current account balance (see Baqir, Ramcha-ran, and Sahay, 2003 for a detailed discussion). Thetables compare the programmed outcomes with theactual ones. For each of the three objectives, therows indicate values for all programs, PRGFs, SBAs,and high-profile SBAs.

19

Reza Baqir, Rodney Ramcharan, and Ratna Sahay

5 The SDR is an international reserve asset created by the IMFin 1969 to supplement the existing official reserves of membercountries. SDRs are allocated to member countries in proportionto their IMF quotas. The SDR also serves as the unit of accountof the IMF and some other international organizations. Its valueis based on a basket of key international currencies. The SDRequaled roughly US$1.55 in December 2004.

6 Of course, these inferences can be drawn only after takinginto account exogenous shocks that could not have been antici-pated when the program was designed and the targets and objec-tives were set. We assume that shocks are randomly distributedacross the programs.

20


Table 1

Country List and Number of Observations for Key Variables

Number of Observations for Actuals on

Real GDP Current account Fiscal Broad Country ID Country Name growth Inflation balance balance money

ALB Albania 10 9 10 10 10ALG Algeria 7 7 7 7 7ARG Argentina 12 12 5 12 8ARM Armenia 11 11 6 10 6AZE Azerbaijan 10 10 9 10 10BEL Belarus 3 3 3 3 3BEN Benin 13 13 9 13 13BOL Bolivia 9 9 9 9 0BOS Bosnia and Herzegovina 6 1 0 5 5BRA Brazil 6 6 1 6 2BUL Bulgaria 12 12 7 11 10BUR Burkina Faso 12 12 10 12 11CAM Cameroon 12 11 12 12 11CAP Cape Verde 5 3 4 3 3CEN Central African Republic 9 9 9 9 9CHA Chad 11 11 11 10 10CMB Cambodia 11 11 9 10 8COL Colombia 6 6 3 6 2CON Congo, Republic of 8 8 5 8 8COS Costa Rica 6 6 5 5 5COT Côte d'Ivoire 7 5 6 6 5CRO Croatia 10 10 4 9 7CZE Czech Republic 4 4 4 2 3DJI Djibouti 7 7 6 3 7DOM Dominican Republic 3 3 3 3 3ECU Ecuador 7 7 5 7 0EGY Egypt 7 7 7 7 7ELS El Salvador 8 8 8 8 8EQU Equatorial Guinea 3 3 3 3 3EST Estonia 10 10 8 9 9ETH Ethiopia 11 9 11 11 11GAB Gabon 9 9 9 9 8GAM Gambia, The 4 4 3 2 4GEO Georgia 10 7 5 10 9GHA Ghana 10 10 10 10 10GUB Guinea-Bissau 5 5 5 5 5GUI Guinea 8 8 5 8 8GUY Guyana 11 11 11 11 11HAI Haiti 4 4 4 4 4HON Honduras 11 11 11 11 11HUN Hungary 7 7 7 7 7IND Indonesia 7 7 4 7 2JAM Jamaica 7 7 7 7 7JOR Jordan 11 11 11 11 11KAZ Kazakhstan 8 8 4 8 7KEN Kenya 9 9 9 9 9KOR Korea, Republic of 6 6 6 6 4KYR Kyrgyz Republic 13 12 8 12 7

(Table continues on next page)

21


Table 1 (concluded)

Country List and Number of Observations for Key Variables

Number of Observations for Actuals on

Real GDP Current account Fiscal Broad Country ID Country Name growth Inflation balance balance money

LAO Lao People’s Democratic Rep. 11 11 10 11 8LAT Latvia 11 11 10 11 11LES Lesotho 9 7 7 9 8LIT Lithuania 12 12 3 11 9MAC Macedonia, FYR of 8 8 8 8 8MAD Madagascar 10 10 10 10 10MAL Mali 14 11 13 13 13MAU Mauritania 13 11 9 11 11MEX Mexico 8 8 8 8 8MLW Malawi 9 9 9 9 9MOL Moldova 10 10 7 10 8MON Mongolia 11 10 11 11 11MOZ Mozambique 9 9 9 8 8NEP Nepal 4 4 4 4 4NGR Nigeria 3 3 3 3 3NIC Nicaragua 8 8 6 7 7NIG Niger 12 12 11 10 11PAK Pakistan 13 10 11 12 12PAN Panama 8 8 8 8 8PAP Papua New Guinea 8 8 8 8 8PER Peru 10 10 6 10 9PHI Philippines 9 9 9 9 9POL Poland 5 5 5 5 5ROM Romania 10 10 8 10 8RUS Russian Federation 7 7 7 7 7RWA Rwanda 6 6 5 6 4SAO São Tomé and Príncipe 3 3 3 3 3SEN Senegal 11 11 11 11 11SIE Sierra Leone 6 6 6 6 6SLO Slovak Republic 5 5 4 4 4SRI Sri Lanka 4 4 4 4 4TAJ Tajikistan 6 6 6 6 4TAN Tanzania 8 8 6 7 5THA Thailand 6 6 6 3 5TOG Togo 6 6 6 6 6TUR Turkey 11 11 9 8 7UGA Uganda 9 9 9 9 9UKR Ukraine 9 9 7 8 7URU Uruguay 10 10 9 10 7UZB Uzbekistan 3 3 3 3 3VEN Venezuela, República

Bolivariana de 3 3 3 3 3VIE Vietnam 10 10 7 10 9YEM Yemen 8 8 8 7 7YUG Yugoslavia 4 4 4 2 3ZAM Zambia 10 10 5 10 9ZIM Zimbabwe 10 10 10 10 8Total 776 748 649 735 665Average number of observations per country 8.3 8.0 6.9 7.8 7.1

22


Table 2

Objectives in IMF Programs: Program Versus Actual

Difference Program Horizon (program minus actual)

Two One Within- Two One Within-years year year Actual years year year

Real GDP growth (in percent)

All program years 5.2 4.6 3.5 1.8 3.4 2.8 1.7

PRGFs 5.7 5.3 4.7 3.3 2.4 2.0 1.4

SBAs 4.5 3.8 2.0 0.3 4.2 3.5 1.7

High-profile SBAs 4.1 2.9 1.3 1.1 3.0 1.8 0.2

CPI inflation (percent, end of period)

All program years 5.0 6.0 8.0 10.3 –5.3 –4.3 –2.3

PRGFs 4.3 5.0 7.0 8.4 –4.1 –3.4 –1.4

SBAs 6.0 7.0 9.1 13.2 –7.2 –6.2 –4.1

High-profile SBAs 6.0 6.3 6.6 8.9 –2.9 –2.6 –2.3

Current account balance (percentage of GDP)

All program years –8.6 –9.1 –9.4 –9.4 0.8 0.3 0.0

PRGFs –11.4 –12.4 –13.2 –13.9 2.5 1.5 0.7

SBAs –4.1 –4.7 –4.6 –4.5 0.4 –0.2 –0.1

High-profile SBAs –2.1 –1.3 –1.3 –1.0 –1.1 –0.3 –0.3

Sources: IMF; Authors' calculations.Notes: Table reports means by group except for inflation, for which medians due to outliers are reported. All observations are used for each

sample. The same general pattern is preserved if sample size is kept constant across columns. The last three columns report the difference be-tween the program columns and the actual columns. PRGFs denotes arrangements under the Poverty Reduction and Growth Facility. SBAs de-notes Stand-By Arrangements.

1.9

1.3

0.9

0.2

–1.0

–0.5

0.0

0.5

1.0

1.5

2.0

2.5

3.0

0–1 1–2 2–3 3–4

Years-ahead projection

Perc

ent

Figure 1

Projection Errors, by Program Horizon: Growth(Error = projected minus actual, mean, and 95 percent confidence interval)

23


–0.8

–0.6

–0.4

–0.2

0.0

0.2

0–1

1–2

2–3

3–4


Log

diffe

renc

eFigure 2

Projection Errors, by Program Horizon: Inflation(Mean and 95 percent confidence interval)

–2

–1

0

1

2

3

0–1

1–2

2–3

3–4


Perc

enta

ge o

f GD

P

Figure 3

Projection Errors, by Program Horizon: Current Account(Mean and 95 percent confidence interval, percentage of GDP)

Objectives

Table 2 indicates that for all types and subsets ofprograms, programmed real GDP growth was consis-tently higher than actual outcomes. Moreover, pro-grammed growth was progressively higher, thelonger was the horizon of the forecasting period(Figure 1). When we compare the forecast errors inabsolute terms, we see that the errors were higher inSBAs than in PRGF programs. It is notable, how-ever, that the errors in the high-profile SBAs werelower than in the SBAs and even lower than thosein the PRGF programs. This suggests that growthprojections are more optimistic in SBAs than inPRGF programs, with one caveat: the projections inthe high-profile SBAs were more realistic than inother SBAs and PRGFs, although the direction ofthe bias was the same in all types of program.

In the second set of rows in Table 2, the pro-grammed and actual inflation rates are compared.As in our results on real GDP growth, programmedinflation is lower than the actual outcomes in alltypes of program. And as in our results on growthforecasts, the errors decrease as the horizon of theforecasting period becomes smaller (Figure 2). Com-paring across programs, the inflation objectives aremore optimistic in the SBAs than in the PRGFs.Within SBAs, the high-profile ones had more realis-tic programmed inflation, although differences be-tween actuals and program objectives were less forthe PRGFs. Again, the direction of the bias was thesame across programs, which points to optimism to-ward achieving inflation objectives.

The results on the current account objectives arequalitatively different from those obtained on thegrowth and inflation objectives. Although the fore-casting error falls with the length of the forecasting

horizon, as in the previous cases, there is no bias, onaverage, in all programs. There are some differencesacross the types of program. In PRGF programs, onthe one hand, the programmed current account bal-ance is somewhat optimistic relative to the realizedvalues; on the other hand, in the SBAs, the realizedvalues were higher than the programmed ones. Thehigh-profile SBAs performed best, since this grouphad the smallest bias compared with other SBAsand PRGFs.

We also explored the unconditional probability ofmeeting all three objectives at the same time (Figure4). The figure shows that when all programs are con-sidered, the probability of achieving all three objec-tives at the same time is about 10 percent. As is to beexpected, this probability rises as the horizon of theforecast shortens, but only marginally. Figure 4 alsoindicates that the probability of meeting the currentaccount objective is the highest, followed by the in-flation and growth objectives, respectively. Thisshould not be surprising, since the core function ofIMF-supported programs is stabilization and restora-tion of balance of payments viability.

In summary, all three objectives—growth, infla-tion, and the current account—are unlikely to bemet at the same time. Second, the inflation andgrowth objectives consistently reflect optimism inthe formulation of IMF-supported programs, whilethe current account balance is met more frequently.Optimism about inflation and growth is highest inSBAs, followed by PRGFs and high-profile SBAs,respectively. Third, the extent to which the targetsfor the current account balance are exceeded isgreatest in high-profile SBAs, followed by otherSBAs and PRGFs, respectively. These results indi-cate that when they are judged by the values of theprogrammed objectives, the high-profile SBAs ap-

24


0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0–1 1–2 2–3Program horizon (years)

Growth Inflation Current account All three

Figure 4

Unconditional Probability of Meeting Program Objectives

pear to have performed best, since either the bias issmaller than for other programs or the targets areexceeded. A question that arises is whether the IMFdoes a better job of designing programs in high-profile cases or simply sets them more realistically insuch instances because, almost by definition, exter-nal scrutiny is greater.

Fiscal Policy Targets

Table 3 compares the fiscal policy targets set in pro-grams with those realized. From top to bottom, thefirst two sets of rows relate to measures of fiscal bal-

ance, the next two to revenues, and the last two toexpenditures.

The table indicates that both the fiscal-balanceand primary-balance targets (shown in first two setsof rows) are missed consistently in all types of pro-gram; and, as expected, the forecast errors shrink asthe forecast horizon declines. Three results are note-worthy. First, the targets in SBAs were missed bysmaller margins than in PRGFs and in the one- andtwo-year horizons. Second, the targets in SBAs andPRGFs were missed by larger margins than in high-profile SBAs. Third, and finally, the bias in theoverall fiscal balance is in the opposite direction in

25


Table 3

Fiscal Policy Targets in IMF Programs: Program Versus Actual(Percentage of GDP)



Fiscal balance, broadest coverage All program years –2.5 –3.0 –3.5 –4.7 2.2 1.7 1.2PRGFs –3.1 –3.7 –4.3 –5.6 2.5 1.9 1.3SBAs –1.3 –2.0 –2.5 –3.8 2.5 1.8 1.3

High-profile SBAs –1.9 –3.0 –3.8 –3.3 1.4 0.3 –0.5

Primary balance (excluding grants)All program years –2.1 –2.5 –2.9 –3.8 1.7 1.3 0.9PRGFs –3.5 –4.2 –5.2 –6.1 2.6 1.9 0.9SBAs 1.8 1.0 0.8 –0.7 2.5 1.7 1.5

High-profile SBAs 0.7 –0.4 0.0 –0.5 1.2 0.1 0.5

Revenues (excluding grants)All program years 20.1 20.6 21.0 21.4 –1.3 –0.8 –0.4PRGFs 17.7 17.8 17.6 17.8 –0.1 0.0 –0.2SBAs 26.7 26.7 27.1 27.3 –0.6 –0.6 –0.2

High-profile SBAs 22.6 21.5 20.4 21.7 0.9 –0.2 –1.3

Revenues (including grants)All program years 22.8 23.5 23.9 24.2 –1.4 –0.7 –0.3PRGFs 20.7 21.2 21.3 21.3 –0.6 –0.1 0.0SBAs 27.0 27.3 27.6 27.9 –0.9 –0.6 –0.3


Total expenditures All program years 25.2 26.3 27.0 28.2 –3.0 –1.9 –1.2PRGFs 23.8 24.4 24.7 25.9 –2.1 –1.5 –1.2SBAs 28.2 29.3 30.1 31.3 –3.1 –2.0 –1.2

High-profile SBAs 23.2 24.3 24.1 23.4 –0.2 0.9 0.7

Primary expenditures All program years 22.8 23.5 23.9 25.3 –2.5 –1.8 –1.4PRGFs 21.8 22.0 22.2 23.1 –1.3 –1.1 –0.9SBAs 25.1 25.8 26.4 28.0 –2.9 –2.2 –1.6


Sources: IMF; Authors' calculations.Notes: Table entries report means by group. All available observations are used for each sample. The same general pattern is preserved if

sample size is left constant across columns. PRGFs denotes arrangements under the Poverty Reduction and Growth Facility. SBAs denotes Stand-By Arrangements.

high-profile SBAs, compared with PRGFs and otherSBAs for the within-year forecast horizon. That is tosay, the actual outcomes on overall fiscal balance inhigh-profile SBAs were better than the ones pro-grammed the previous year.

Regarding revenue targets and performance, thepattern is unexpected and striking. The actual rev-enue outcomes—whether measured with or withoutgrants—are consistently better than the pro-grammed targets for all programs and across almostall time horizons. This pattern is unexpected be-cause we have seen that the growth outcomes werefar worse than programmed, which should lead us tobelieve that the revenue performance would beworse than programmed. The second notable fea-ture is that contrary to our expectations, errors inforecasting do not necessarily fall over time whenrevenues are measured without grants. It almostseems as if programs were made tighter over timewhen the targets came close to being reached earlyin their implementation.

The pattern of expenditure (programmed and ac-tual values) is similar to that of the fiscal balance.Actual expenditures were higher than the pro-grammed ones across all types of program. Also, asexpected, forecast errors generally became smallerwith the shortening of the forecast horizon. Theonly puzzling result is for high-profile SBAs: the pro-grammed total expenditures were higher than theactuals, though this result did not hold when pri-mary expenditures were considered. It appears thatthe interest costs were overestimated for the high-profile SBAs—the interest rate spreads turned outto be smaller than expected, perhaps owing to betterperformance, as we saw earlier, or to the credibilityof the IMF programs themselves that IMF staffmembers did not fully take into account when theprograms were designed.

In summary, the fiscal targets appear to have beenmet more often in the high-profile SBA programs,although, in general, more fiscal targets wereachieved in PRGFs than in SBAs. Although it isgenerally true that the forecasting errors improvedas the horizon shortened, this result did not neces-sarily hold for revenue projections, which did notchange very much with the forecast horizon.

Monetary Policy Targets

Table 4 compares the programmed monetary poli-cy targets with the actual outcomes under IMF-supported programs. To analyze adjustments underprograms and to facilitate comparisons across coun-tries, we look at the first differences (rather than theactual levels) of broad money, net domestic assets,and net foreign assets. In addition, the absolute val-ues of velocity are compared across program types.

Several broad patterns emerge in comparing theprogrammed and actual values of the monetary pol-icy targets. First, targets for broad money and do-mestic asset growth were generally missed in all

types of program. Second, targets for foreign assetswere met with greater precision, which is consistentwith our earlier finding that external current ac-count objectives are generally met in IMF-supportedprograms. Third, the errors in forecasting monetarytargets were similar across PRGFs and SBAs, buthigher for high-profile SBAs.

Interpreting the results on the income velocity ofmoney is not a trivial task. We find that programmedvelocity, relative to the realized values, is highest forPRGFs, followed by all SBAs and high-profile SBAs,respectively. In fact, for the high-profile SBAs, theforecasting error (programmed minus actual value)was negative. One interpretation of this result isthat IMF-supported programs underestimated thepickup in the demand for money in PRGFs andmost SBAs but overestimated the increase in thedemand for money in the high-profile SBAs. An-other interpretation is that the monetary programswere looser for the high-profile SBAs, comparedwith the other two types of program.

Were Objectives Less Optimistic and FiscalTargets Less Tight for High-Profile SBAs?

One stylized fact that emerges from the previoussubsections is that fiscal outcomes were closer to tar-gets in high-profile SBAs than in other types of pro-gram. This could indicate that either program tar-gets were not ambitious—so that it was easier toattain them—or that programs were designed better,so that outcomes were close to expectations. In thissubsection, we examine evidence for the first ofthese two possible interpretations. Table 5 showsprogrammed fiscal adjustment by type of programand by type of the fiscal measure. Here, instead ofcomparing actuals to program values, as we did be-fore, we summarize programmed fiscal effort (mea-sured as the fiscal measure programmed for next yearminus this year’s actual outcome). The results arestriking and systematic: first, the adjustment plannedin all SBAs is always more than in high-profileSBAs. The adjustments programmed for high-pro-file SBAs, however, are not only always less than forother SBAs but also less than for the PRGFs. In fact,virtually all fiscal targets are relaxed in the within-year horizon in the high-profile SBAs.

Program Objectives and Intermediate Policies

IMF-supported programs are designed to set policiesthat are consistent with achieving certain objec-tives. As part of this exercise, the IMF’s staff pro-duces a “program scenario,” which quantifies the ob-jectives (growth, inflation, and others) and theintermediate policies (fiscal balance, monetary ex-pansion, and others) consistent with these objec-tives. Our approach to examining the link betweenintermediate policy targets and objectives is to ask

26


27


Table 4

Monetary Policy Targets in IMF Programs: Program Versus Actual(Percentage of GDP)



Broad money


PRGFs 20.1 20.2 19.6 21.9 –1.8 –1.7 –2.3

SBAs 38.4 37.4 32.1 34.3 4.1 3.1 –2.2


Increase in broad money


PRGFs 2.8 2.7 2.6 4.0 –1.2 –1.3 –1.4

SBAs 6.1 7.2 6.7 8.2 –2.1 –1.0 –1.5


Increase in net domestic assets


PRGFs 1.4 1.4 1.3 1.6 –0.2 –0.2 –0.3

SBAs 3.3 4.2 3.7 4.8 –1.5 –0.6 –1.1


Increase in net foreign assets


PRGFs 1.3 1.4 1.6 1.9 –0.6 –0.5 –0.3

SBAs 1.7 2.2 2.0 1.9 –0.2 0.3 0.1


Velocity

All program years 4.4 4.3 4.3 3.9 0.5 0.4 0.4

PRGFs 5.0 5.0 5.1 4.6 0.4 0.4 0.5

SBAs 2.6 2.7 3.1 2.9 –0.3 –0.2 0.2


Sources: IMF; authors' calculations.Notes: Table reports medians by group. The median is a better indicator of the central tendency for monetary variables owing to several out-

liers in the monetary series. All observations are used for each sample. The same general pattern is preserved if sample size is kept constantacross columns. The last three columns report the difference between the program columns and the actual columns. PRGFs denotes arrange-ments under the Poverty Reduction and Growth Facility. SBAs denotes Stand-By Arrangements.

whether achieving the intermediate policy targetshelps to achieve program objectives. To address thisquestion, we focus on the deviation of the outcomesfrom the programmed values (which we will refer toas “projection errors” for lack of a better term).7 Forexample, the question posed is “does growth fall fur-ther short of its programmed value when thegrowth-consistent policy falls further short of itsprogrammed value?” If there is no such relationship,or the relationship is in the opposite direction, itwould cast serious doubt on the validity of theframework underlying program design. Conversely,the empirical relationship may turn out to be in theexpected direction yet growth outcomes may stillfall systematically short of programmed values evenafter controlling for the extent to which policy targets are achieved. That would suggest that thereare other elements missing in the programmingframework and/or that the optimism in settinggrowth targets is greater than could be justified bypolicy shortfalls.

We examine the relationship between the growthobjective and two types of macro policies: fiscal andmonetary.

Fiscal Policy

We start our investigation by recapitulating the sta-tistics presented earlier on the systematic shortfallin growth outcomes compared with the programmed

values. The equation shown in Table 6’s second col-umn regresses the projection error in growth on aconstant, reflecting the normal approach to examin-ing the extent of bias in a projection. Projection er-rors are defined as programmed values minus actualvalues. Such errors can be presented at differenttime horizons. For the sake of brevity, we presentthe results with the one-year horizon.8 Thus, the fig-ure in the first column indicates that, on average,actual growth is about 0.9 percentage points lessthan what was programmed a year earlier.9

In the second specification, we regress the projec-tion error in growth on the projection error in theoverall fiscal balance:10

(1)

where, for any variable x (g and f are growth and the fiscal balance, respectively) for countryi, et–s(xt) denotes the projection error based on a projection made s periods ahead and defined aset–s(xt)0

t–sxt – xt. In our notation, t–sxt denotes thes-period-ahead forecast, and xt simply denotes theoutcome for x in period t.

There are two points worth noting in the regres-sion results. First, the coefficient of the projection

28


Table 5

Programmed Fiscal Adjustments, by Program Type(Percentage of GDP)

Programmed Change in Fiscal Measure

SBAs

All PRGFs All High-profile

Fiscal balance, broadest coverage 0.54 0.55 0.53 –0.84

Primary fiscal balance excluding grants,broadest coverage 0.55 0.41 0.80 0.46

Revenue 0.53 0.67 0.33 0.02

Revenue excluding grants 0.36 0.36 0.36 –0.95

Expenditure 0.10 0.29 –0.19 –0.18

Primary expenditure 0.07 0.27 –0.21 –1.06

Notes: Table entries report the fiscal measure programmed for one year ahead less this year's actual. PRGFs denotes arrangements under thePoverty Reduction and Growth Facility. SBAs denotes Stand-By Arrangements.

7 As discussed previously, it is not quite right to think of theprogram numbers as projections in the sense that this term isgenerally used. Program numbers are best understood as the IMFstaff’s projections of outcomes conditional on the member coun-try’s achieving certain policy targets and adequate implementa-tion of other elements of the program.

8 While a within-year horizon may be too short for a meaning-ful test of program design, a two-year horizon may be too long, inthat ensuing events can seriously weaken the assumptions onwhich targets were based. Thus, in general, we focus on the one-year horizon, although we conducted robustness checks for otherlengths of horizon. The results for different horizon lengths weregenerally consistent.

9 The slight variations from the summary statistics presentedearlier were due to small differences in the sample sizes.

10 We use the broadest available measure of the fiscal balancethroughout.

e g e ft s it t s it it− −= + ⋅ +( ) ( )α β ε

error on the fiscal balance is consistent with the fi-nancial programming framework. That frameworkimplies that with other factors remaining the same,a smaller fiscal deficit creates more room for privatesector credit while respecting overall conditions formoney growth. To the extent that private sectorcredit is conducive to financing investment andgrowth, this is expected to allow a greater expansionof output. The coefficient suggests that a 1 percent-age point improvement in the extent to which thefiscal target is met is associated with a 1⁄4 of 1 per-centage point improvement in the extent to whichthe growth target is met.

The second notable point is that the growth ob-jective is not met, on average, even after controllingfor the extent to which the intermediate policy tar-get is met. This is indicated by the continued statis-

tically significant coefficient on the constantterm—the conventional measure of bias. When theprogrammed fiscal balance is exactly equal to theactual fiscal balance, actual growth performance re-mains systematically less than programmed, thoughthe magnitude of the shortfall is somewhat less thanthe unconditional bias when we do not control forthe extent to which policy targets are met. System-atically being optimistic in setting growth objectivescan have serious consequences for other aspects ofprogram design, particularly for debt dynamics (Hel-bling, Mody, and Sahay, 2003). Taken together,these two points suggest that although programs getthe direction of the framework right, their growth as-sumptions are more optimistic than can be justified.

In the third data column of Table 6, we allow forcountry-specific heterogeneity by including a com-

29


Table 6

Regressions for Projection Errors in Growth and Fiscal Targets

Dependent Variable

Proj. error Proj. error Proj. error Actual Programmed Proj. errorin growth in growth in growth growth growth in growth

Fiscal measure = Broad fiscal balanceCountry fixed effects No No Yes Yes Yes YesConstant 0.890*** 0.736*** –4.717 4.995 2.734 –2.538

(0.000) (0.002) (0.233) (0.268) (0.242) (0.519)Proj. error in fiscal measure 0.251*** 0.471***

(0.000) (0.000)Actual fiscal measure 0.559*** –0.512***

(0.000) (0.000)Programmed fiscal measure 0.106** 0.431***

(0.018) (0.000)Wald test (p-value) 0.59No. of observations 313 287 287 735 445 287R-squared 0.000 0.057 0.309 0.398 0.417 0.310

Fiscal measure = Broad primary fiscal balance, excluding grantsCountry fixed effects No No Yes Yes Yes YesConstant 0.890*** 0.599** 4.439 –10.465** 7.892*** 2.849

(0.000) (0.023) (0.207) (0.045) (0.006) (0.458)Proj. error in fiscal measure 0.298*** 0.276***

(0.000) (0.009)Actual fiscal measure 0.502*** –0.345***

(0.000) (0.007)Programmed fiscal measure 0.112** 0.210*

(0.023) (0.090)Wald test (p-value) 0.33No. of observations 313 207 207 584 361 207R-squared 0.000 0.061 0.430 0.453 0.444 0.434

Notes: Projection error is defined as the programmed value minus the realized value. This table presents results for programmed values atthe one-year horizon (see text). “Growth” refers to growth of real GDP in percentage points. Fiscal measures are in percentage of GDP. Paren-theses report p-values for the estimated coefficients. An asterisk (*) denotes significance at 10 percent, ** at 5 percent, and *** at 1 percent. TheWald test corresponds to the null hypothesis that the sum of the coefficients on the actual and programmed fiscal measure in the last specifica-tion equals zero.

plete set of country fixed effects in the equation.The coefficient on the projection error on the fiscalbalance strengthens, suggesting that programs use-fully use country-specific information in programdesign. In terms of bias, in this specification there isone estimated constant per country. The joint testfor all country-specific constants being equal to zerois not rejected, suggesting that one constant couldhave been estimated.11

A potential issue of interpretation in the previousspecification is that a relationship estimated in theform of projection errors may be suppressing usefulinformation in the respective relationships betweenactual growth and actual fiscal balance, and be-tween programmed growth and programmed fiscalbalance. The next two specifications in Table 6 es-sentially unravel this relationship. We first regressactual growth on actual fiscal balance and then dothe same for the programmed values:

(2)

In each case, we get a significant relationship, al-though the magnitude is somewhat stronger for therelationship estimated in actuals. We formally testfor whether actuals and programmed values can bepooled in the next column, where we regress theprojection error in growth on both the actual andprogrammed levels of the fiscal balance:

(3)

If β1=β 2 =β and the errors are uncorrelated, wewould simply get (1).12 Table 6 shows the proximitybetween the estimated coefficients on β1 and β2. AWald test for β1=β2 is not rejected, vindicating ouroriginal approach.

The measure of fiscal balance we have used so faris the overall balance. There are two potential prob-lems with it. First, to the extent that some revenueconsists of fully funded grants—for instance, from of-ficial donors—an expansion of the deficit may notcrowd out private sector credit and may not ad-versely affect growth. Hence, a more appropriatemeasure of fiscal balance in the context of the pro-gram framework may be one that excludes grants.Second, it may be more appropriate to look at theprimary fiscal balance to more appropriately measurefiscal effort by a country. The bottom panel of Table6 repeats the above set of specifications for the pri-mary fiscal balance excluding grants. We get the

same pattern, with very similarly sized estimated co-efficients, and again the Wald test is not rejected.13

Implicit in the preceding discussion is the as-sumption that an improvement in the fiscal balanceleads to an improvement in growth. In reality,growth outcomes may well affect the realized fiscalbalance. In particular, such endogeneity could arisein two forms. First, buoyancy in revenues may yieldprocyclical movements in the revenue-to-GDPratio. Second, government spending may react toexternal shocks to stabilize output. Externally driven slowdowns in growth may cause the govern-ment to increase public outlays. Similarly, in goodtimes, the government may let the private sectortake the lead and roll back its own spending. We ad-dress each of these potential problems in turn.

As a first step toward reducing potential bias inthe previously estimated equations, we start by firstdifferencing our data. Hence, we look at how thechange in growth is correlated with the change inthe fiscal balance. Although this automatically getsrid of country fixed effects, it allows us to addition-ally control for country-specific trends. Some coun-tries may be on a “good path” with rising growth andfiscal balances. Using first differences and a com-plete set of country fixed effects allows us to controlfor such differences among countries. The first tworows of Table 7 show that the previously estimatedrelationships in levels survive when estimated infirst differences, with and without country fixed ef-fects. For example, an improvement of 1 percent ofGDP in the fiscal balance is associated with an 0.5percentage point increase in growth. The next tworows of Table 7 show that this relationship is notvalidated from the revenue side. There is no rela-tionship between changes in the revenue ratio (in-cluding or excluding grants) and changes in growth.Thus, buoyancy is probably not contaminating ourresults. The last two rows show that the relationshipbetween the fiscal balance and growth emanatesfrom the expenditure side. A 1 percentage pointincrease in expenditure is associated with about an 0.3 percentage point reduction in growth.

To test whether expenditure, and hence our fiscal-balance measures, may be reacting to output shocksowing to countercyclical fiscal policy, we present re-sults from instrumental variable regressions in Table8.14 In this specification, we identify the actualchange in the fiscal balance by using the pro-grammed change in the fiscal balance and exportgrowth. Since adjustment programmed one year in

30


11 The test may be compromised owing to the limited number ofobservations per country, however; in this specification, there are,on average, only 3–4 observations for each country. Since time-invariant, country-specific heterogeneity can be an importantsource of bias—which could contaminate our results—we include acomplete set of fixed effects in all subsequent specifications.

12 We address issues of endogeneity later in this subsection.

13 We repeated these regressions for all possible permutationsof the fiscal measures along the following dimensions: level ofcoverage (central government versus broadest available), treat-ment of grants (excluded from versus included in revenues), andinterest expenditure (excluded from versus included in fiscal bal-ance). We found the same general pattern of results reported previously.

14 Kaminsky, Reinhart, and Végh (2004) find that fiscal policyis, in fact, procyclical for non-industrial countries.

g f

g fit i it it

t s it i t s it i

= + ⋅ +

= + ⋅ +− −

α β ε

α β ε1 1 1

2 2 2 tt

( ) ( )

(t s it it i i t s it

it

g g f

f− −− = − + ⋅

− ⋅ +

α α β

β ε1 2 2

1 2 iit it−ε1

).

31


Table 7

Regressions for Growth and Fiscal Targets, First Differences

Dependent Variable = First Difference of Growth Rate

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)

Country fixed effects No Yes No Yes No Yes No Yes No Yes No Yes

Fiscal balance, broadest coverage 0.526*** 0.532***(first difference) (0.000) (0.000)

Primary fiscal balance excluding grants, broadest coverage 0.451*** 0.458***(first difference) (0.000) (0.000)

Revenue 0.046 0.047(first difference) (0.658) (0.700)

Revenue, excluding grants 0.103 0.106(first difference) (0.425) (0.499)

Expenditure –0.319*** -0.312***(first difference) (0.000) (0.000)

Primary expenditure –0.280*** –0.263***(first difference) (0.000) (0.002)

Constant 0.217 0.713 0.180 1.758 0.485 -0.328 0.383 2.306 0.451 1.343 0.488 1.157(0.425) (0.916) (0.534) (0.786) (0.108) (0.960) (0.232) (0.720) (0.121) (0.832) (0.106) (0.855)

No. of observations 609 609 459 459 407 407 349 349 414 414 385 385

R-squared 0.086 0.201 0.088 0.166 0.000 0.066 0.002 0.091 0.047 0.118 0.038 0.126

Notes: The table reports results from regressions of the change in the growth rate on the change in the fiscal measure listed in the first col-umn. “Growth” refers to growth of real GDP in percentage points. Fiscal measures are in percent of GDP. Parentheses report p-values for the es-timated coefficients. An asterisk (*) denotes significance at 10 percent, ** at 5 percent, and *** at 1 percent.

Table 8Instrumental Variable Regressions for Growth and Fiscal Targets, First Differences

Dependent Variable =First Difference of Growth Rate

(1) (2) (3) (4)

Country fixed effects No Yes No Yes

Fiscal balance, broadest coverage 1.274*** 1.188***(first difference) (0.000) (0.000)

Primary fiscal balance excluding grants,broadest coverage 0.399*** 0.418**(first difference) (0.008) (0.016)

Constant 0.261 2.345 0.735* 2.440(0.541) (0.723) (0.072) (0.688)

Test of overidentifying restrictions (p-value)Sargan test 0.40 0.28 0.62 0.39Basmann's test 0.40 0.20 0.62 0.48

No. of observations 268 268 199 199

R-squared 0.141 0.060 0.272

Notes: The table reports the results from instrumental variable regressions of the change in the growth rate on the change in the fiscal bal-ance measure. The change in the fiscal balance is instrumented with the change in the fiscal balance programmed 1–2 years ago and with ex-port growth. The test of overidentifying restrictions is the test of the joint hypothesis that the instruments are valid and correctly excluded fromthe estimated equation. A rejection of the test casts doubt on the validity of the instruments. "Growth" refers to growth of real GDP in percent-age points. Fiscal measures are in percentage of GDP. Parentheses report p-values for the estimated coefficients. An asterisk (*) denotes signifi-cance at 10 percent, ** at 5 percent, and *** at 1 percent.

32


advance is predetermined relative to the actual real-ization of the shock in period t, we think it is a goodinstrument for identifying the exogenous variationin the actual change in the fiscal balance. In addi-tion, export growth may capture external shocks towhich fiscal policy may react. We run this specifica-tion both with and without country fixed effects. Ineach case, we find that the improvement in the fis-cal balance, as identified, likely increases growth.We also test whether we should instead have thesevariables directly in the regression as right-hand-side variables by running a test of overidentifying re-strictions. In each case, the test is not rejected, cor-roborating our approach.

Monetary Policy

We now turn to examining the relationship be-tween growth and monetary policy in the context ofIMF-supported programs. The approach we follow issimilar to the one we followed for fiscal policy. Thekey relationship examined is between growth andvelocity. An assumption on velocity is one of thefirst and integral assumptions made as part of pro-gram design. After the growth and inflation objec-tives have been set, an implicit assumption regard-ing money demand is made by projecting a specificvelocity. Alternatively, a money demand function is

estimated and an estimate for velocity is then de-rived. Setting the amount of monetary expansionunder the program is key, since it establishes theoverall “tightness” of the program. As discussed inthe previous section on financial programming, afterthe monetary growth and the net foreign asset(NFA) targets have been set, the maximum tolera-ble expansion in net domestic assets is determinedas a residual. Programming higher velocity wouldsystematically lead to tighter monetary objectives,which, in turn, with other things held constant,would constrain total credit to the economy and,hence, output.15

Table 9 shows the results of the specifications weran. One problem we encountered was the signifi-cant large volatility in the monetary aggregates typi-cally observed in the early years in the transitioncountries, when many systemic changes and struc-tural transformations took place. Under such cir-cumstances, money demand was virtually impossibleto predict. To be on the safe side, we therefore ex-cluded all transition countries from the regressions

Table 9

Regressions for Growth and Velocity

Dependent Variable = Programmed Less Actual GDP Growth

(1) (2) (3) (4) (5) (6)

Country fixed effects No No Yes Yes

Programmed velocity 0.398** 0.635*** 0.643***less actual velocity (0.014) (0.003) (0.002)

Lagged programmed velocity 0.438*less actual velocity (0.061)

Programmed velocity 0.645***(0.003)

Actual velocity –0.603**(0.013)

Fiscal balance projection error 0.144***(broadest available measure) (0.010)

Constant 0.138 0.185 1.033 4.770** –2.072 –1.498(0.333) (0.168) (0.648) (0.023) (0.413) (0.510)

No. of observations 332 279 279 176 279 275

R-squared 0.000 0.021 0.259 0.294 0.259 0.287

Notes: Projection error is defined as the programmed value minus the realized value. The table presents results for programmed values atthe within-year horizon (see text). "Growth" refers to growth of real GDP in percentage points. Fiscal balance is in percent of GDP. Parenthesesreport p-values for the estimated coefficients. An asterisk (*) denotes significance at 10 percent, ** at 5 percent, and *** at 1 percent.

15 As an alternative, one could also focus on the projection er-rors in net domestic assets. We found considerable instability inthe measures of net domestic assets in our database, however. Inpart this is due to cases of very high inflation in the sample dur-ing which the relationships among monetary aggregates becomeparticularly unstable.

in this subsection. Since this exclusion reduces oursample size, we use the within-year horizons in thissection to maximize available observations. The firstcolumn regresses the projection error in growth on aconstant. The second regression adds the projectionerror in velocity:

(4)

where v denotes velocity. The positive estimated co-efficient suggests that programming higher velocitydrives actual growth performance below the pro-grammed value. The next specification adds a com-plete set of country fixed effects. Controlling forcountry-specific heterogeneity strengthens the rela-tionship between the projection errors in velocityand growth. To reduce the scope for contemporane-ous correlation between velocity and growth, thenext specification lags the projection error in veloc-ity. Although the number of observations drops, thecoefficient is still significant at 10 percent. The nextspecification removes constraints on the coefficientson actual and programmed velocity and shows thatthe two coefficients are close in magnitude and op-posite in sign, as hypothesized. A Wald test forβ1=β2 is not rejected, indicating that the regressioncould be run in terms of projection errors.

The last specification in Table 9 regresses the pro-jection error in growth on both the projection errorin velocity and the projection error in the broad fis-cal balance. These results suggest that even aftercontrolling for the projection error in the fiscal bal-ance, higher-than-actual programmed velocity de-presses growth; and conversely, even after controllingfor the tightness of the monetary program, a higherfiscal surplus is associated with greater growth.

Conclusion

In this paper, we have attempted to analyze severalaspects of IMF program design. We have docu-mented systematically the relationship between pro-grammed values and outcomes for key program ob-jectives and the intermediate policies designed toachieve them. We find that IMF-supported pro-grams achieve the objectives set for external currentaccount adjustment more frequently than those setfor inflation and growth. All three objectives aremet simultaneously in about 10 percent of the pro-grams. Likewise, the programmed values on inter-mediate policy targets on the fiscal and monetaryvariables were generally more ambitious than thoseactually achieved in the programs.

Second, we have explored the relationship be-tween errors in growth objectives, on the one hand,and errors in fiscal and monetary policy targets, onthe other hand. The evidence suggests that an im-provement in the fiscal balance is associated withbetter growth outcomes, and that programming

more ambitious fiscal targets helps to achieve highergrowth. Fiscal targets are more often missed thanmet, however. Recognizing the difficulty in meetingfiscal targets, programs may tend to overcompensateby being tougher on the monetary policy side. Pro-gramming tight velocity may protect the countryagainst missing the fiscal objective but does so at thecost of dampening growth.16

Third, we find systematic biases in growth and in-flation projections even after controlling for policyimplementation.17 To the extent that ambitious ob-jectives are used to spur authorities into action, thismay not, in itself, be a problem. To the extent thatthe bias is more than what could be justified ongrounds of inadequate policy implementation, how-ever, there is cause for concern. One example of thecosts of getting growth projections wrong is in thecontext of debt dynamics where IMF-supported pro-grams may predict much lower debt-to-GDP ratiosthan are actually achieved.

One question we were not able to address iswhether, in a constrained world where fiscal targetsare likely to be missed, overcompensating by havingtighter monetary programs is the best strategy fordesigning programs to achieve more ambitious ob-jectives. Although a tighter monetary program islikely to entail output costs, it may be necessary topromote fiscal discipline, ensure inflation stability,and restore external current account balance (twoother key objectives that we do not explore ingreater depth in this paper).

Returning to the broader questions that we beganwith in this paper, we note that it is indeed the casethat IMF-supported programs are ambitious with re-spect to their objectives and intermediate policy tar-gets. In that sense, both those on the right and thoseon the left are correct: most program objectives arerarely fully achieved, and fiscal and monetary policytargets are ambitious. On the more interesting ques-tion of whether such ambition is defensible, thispaper has attempted to substantiate that it is justifi-able for the fiscal targets, because it helps achievehigher growth objectives than would otherwise bepossible. There is also evidence, however, thatgrowth objectives are ambitious to an extent thatexceeds what can be explained by the need to ensure consistency with ambitious intermediate pol-icy targets. The latter can, and have tended to, haveunwarranted side effects: when growth is pro-grammed too high, IMF-supported programs end upprojecting lower debt-to-GDP ratios than those ac-tually realized.

33


16 Tighter monetary programs may be designed to bring downinflation, which may necessarily entail output costs. In thispaper, we do not explore the relationship between intermediatepolicy targets and the inflation objective.

17 Our results contrast with those of Musso and Phillips (2002),which does not find statistical bias in growth projections underIMF-supported programs. We note, however, that their sample,consisting of 54 countries, was much smaller than ours.

e g e vt s it t s it it− −= + ⋅ +( ) ( )α β ε

34


BibliographyBagci, P., and W. Perraudin, 1997, “Do IMF Programs

Work?” Global Economic Institutions Working Paper.Baqir, R., R. Ramcharan, and R. Sahay, 2003, “IMF Pro-

gram Design” (unpublished: Washington: Interna-tional Monetary Fund).

Barro, Robert J., and Jong-Wha Lee, 2002, “IMF Pro-grams: Who Is Chosen and What Are the Effects?”NBER Working Paper No. 8951 (Cambridge, Massa-chusetts: National Bureau of Economic Research),pp. 1–37.

Bird, Graham, 2002, “The Credibility and Signaling ofIMF Programs,” Journal of Policy Modelling, Vol. 24,pp. 799–811.

Bulír, A., and S. Moon, 2004, “Is Fiscal Adjustment MoreDurable When the IMF Is Involved?” ComparativeEconomic Studies, Vol. 46, pp. 373–99. This paper alsoappears, in somewhat different form, as Chapter 14 inthis volume.

Conway, P., 1994, “IMF Lending Programs: Participationand Impact,” Journal of Development Economics,Vol. 45 (December), pp. 365–91.

Dicks-Mireaux, Louis, Mauro Mecagni, and SusanSchadler, 2000, “Evaluating the Effect of IMF Lend-ing to Low Income Countries,” Journal of Develop-ment Economics, Vol. 61 (April), pp. 495–525.

Ghosh, A., T. Lane, M. Schulze-Ghattas, A. Bulír, J.Hamann, and A. Mourmouras, 2002, IMF-SupportedPrograms in Capital Account Crises, IMF OccasionalPaper No. 210 (Washington: International MonetaryFund).

Goldstein, Morris, and P. Montiel, 1996, “EvaluatingFund Stabilization Programs with Multicountry Data:Some Methodological Pitfalls,” Staff Papers, Interna-tional Monetary Fund, Vol. 33 (June), pp. 304–44.

Haque, Nadeem Ul, and Mohsin S. Khan, 1998, “DoIMF-Supported Programs Work? A Survey of Cross-Country Empirical Evidence,” IMF Working Paper98/169 (Washington: International Monetary Fund).

Helbling, Thomas, Ashoka Mody, and Ratna Sahay, 2004,“Debt Accumulation in the CIS-7 Countries: BadLuck, Bad Policies, or Bad Advice?” in The Low-Income Countries of the Commonwealth of IndependentStates: Progress and Challenges in Transition, ed. by

Clinton R. Shiells and Sarosh Sattar (Washington:International Monetary Fund and World Bank).

Kaminsky, Graciela L., Carmen M. Reinhart, and CarlosA. Végh, 2004, “When It Rains, It Pours: ProcyclicalCapital Flows and Macroeconomic Policies,” inNBER Macroeconomic Annual, ed. by Kenneth Rogoff and Mark Gertler (Cambridge, Massachusetts:National Bureau of Economic Research).

Khan, Mohsin S., 1990, “ The Macroeconomic Effects ofIMF-Supported Adjustment Programs,” Staff Papers,International Monetary Fund, Vol. 37 (June),pp. 195–231.

———, and M. Knight, 1981, “Stabilization Programs inDeveloping Countries: A Formal Framework,” StaffPapers, International Monetary Fund, Vol. 28(March), pp. 1–53.

———, 1985, IMF-Supported Programs and EconomicGrowth, IMF Occasional Paper No. 41 (Washington:International Monetary Fund).

Killick, T., M. Malik, and M. Manuel, 1995, “What CanWe Know About the Effects of IMF Programmes?”World Economy, Vol. 15 (September), pp. 575–97.

Mussa, Michael, and Miguel Savastano, 1999, “The IMFApproach to Economic Stabilization,” IMF WorkingPaper 99/104 (Washington: International MonetaryFund).

Musso, Alberto, and Steven Phillips, 2002, “ComparingProjections and Outcomes of IMF-Supported Pro-grams.” Staff Papers, International Monetary Fund,Vol. 49, No. 1, pp. 22–48.

Przeworski, Adam, and J.R. Vreeland, 2000, “The Effectof IMF Programs on Economic Growth,” Journal ofDevelopment Economics, Vol. 62, pp. 385–421.

Reichmann, T., and R. Stillson, 1978, “Experience withPrograms of Balance Payments Adjustment: Stand-By Arrangements in the Higher Credit Tranches,1963–1972,” Staff Papers, International MonetaryFund, Vol. 25 (June), pp. 119–39.

Santaella, J., 1995, “Four Decades of Fund Arrangements:Macroeconomics Stylized Facts Before the Adjust-ment Programs,” IMF Working Paper 95/74 (Wash-ington: International Monetary Fund).

Stiglitz, J., 2002, Globalization and Its Discontents (NewYork: Norton).

3

I analyze empirically whether program size (the size of fi-nancial assistance) and policy adjustment matter for thesuccess of IMF-supported programs. I define a programas successful in boosting private capital flows if the initialprogram projections for net private capital flows are metor exceeded. I find that success is negatively associatedwith the size of financial assistance, especially in coun-tries with market access, and that projection biases aris-ing from binding constraints on the amount of IMF lend-ing may account for this association. Moreover, policyadjustment seems to have a causal positive effect on thelikelihood of program success.

Introduction

What determines the success of IMF-supported pro-grams? A simple approach is to ask whether the fi-nancing that a country can raise from privatesources meets or exceeds the target set by the pro-gram. Among the many criteria, this one is particu-larly relevant, because restoring capital market ac-cess is necessary for program viability whenprograms can provide only a fraction of a country’soverall external financing need and, as a result,have to rely on private sources to help cover the re-mainder. Therefore, in this paper, I will define a pro-gram to be successful if net private capital flows donot fall short of their projections.

The purpose of this study is to analyze what deter-mines program success. At a stylized level, a programconsists of financial assistance—whose magnitude isoften referred to as the “size of the program”—and pol-icy adjustment; thus, it is worth asking whether thesetwo components contribute to success. Theory doesnot permit one to make clear predictions onwhether larger programs are more likely to succeed.On the one hand, a larger financial commitmentmay, for example, improve the IMF’s incentives forbetter monitoring and program design, thereby gen-

erating a positive response from private investors(see Rodrik, 1996). On the other hand, IMF pro-grams may encourage capital outflows by providingmuch-needed foreign currency reserves. Likewise,how policy adjustment affects capital flows is un-clear in principle. Tight macroeconomic policies areusually considered desirable to restore investor con-fidence and encourage the return of private capital.However, critics of IMF programs during the Asiancrisis in 1997 have seriously questioned that view(see Furman and Stiglitz, 1998).

Besides being straightforward, this approach hasfour advantages. First, by using program projectionsfor net private capital flows as a benchmark for ac-tual flows, I avoid the problem of defining a “coun-terfactual” scenario (what would have happenedwithout a program) against which to define programsuccess. Since programs contain projections for cap-ital flows, each program itself provides the relevantbenchmark for actual flows, and no counterfactualneeds to be constructed.

Second, this approach allows me to deal directlywith the implications of the requirement, mandatedby the IMF’s Articles of Agreement, that financialassistance be provided only in the presence of aproven balance of payments need. As a result of thisrequirement, programs typically aim at improvingthe current account balance; in fact, one of the mostrobust and less controversial findings in the litera-ture on the effects of IMF programs is that the cur-rent account balance improves under IMF programs(see Haque and Khan, 1998). In turn, improving thecurrent account tends to lower the country’s de-mand for external sources of finance. Thus, one hasto be careful not to confuse the reduction in netcapital inflows caused by the external adjustmentwith the failure to generate private flows. Compar-ing net capital flows against a program-dependentbenchmark eliminates any potential confusion.2

35

Do IMF-Supported Programs Boost Private Capital Inflows? The Role of Program Size and Policy Adjustment

Roberto Benelli1

1 I am very grateful to Eduardo Borenzstein, Peter Clark, CarloCottarelli, Olivier Jeanne, Gian Maria Milesi-Ferretti, AshokaMody, Alex Mourmouras, Alessandro Prati, Rodney Ramcharan,Roberto Rigobon, Jeronimo Zettelmeyer, and especially PaoloMauro and Alessandro Rebucci for helpful discussions and com-ments. Any errors that remain are mine.

2 The current account could improve more than projected be-cause of an unpredictable exogenous shock such as, for example,an improvement in terms of trade, thus reducing the need forcapital flows. Although in this case my approach signals a “fail-ure,” there will be no consequence for the unbiasedness of theeconometric estimates of the effect of programs. This is discussedin more detail in the next section.

CHAPTER

Third, this approach allows me to measure thestrength (or weakness) of the “catalytic” effects ofIMF lending. Although programs, particularly dur-ing the capital account crises of the 1990s, may havebeen designed to catalyze private capitals—that is,spur private capital inflows—the empirical evidenceof catalytic effects is weak (see Cottarelli and Gian-nini, 2002). A direct way to assess catalytic effects isto ask whether the programs generate the net capi-tal flows that have been projected by the IMF at thestage of program design. In this light, a shortfall inactual flows relative to these projections shows thatthe factors that should have generated catalytic ef-fects have not played the expected role.

Fourth, this approach allows me to estimate howpolicy adjustment affects program success by isolat-ing the causal effect of policy adjustment from its en-dogenous response to external shocks. The endogene-ity problem arises because, for example, tightmonetary and fiscal policies may not be sustainableif an exogenously determined, unexpected worsen-ing in capital inflows causes a deep recession. Ad-dressing this endogeneity problem requires valid in-struments for the policy adjustment that takes placeunder the program. I use the projected policy adjust-ment—the change in inflation and the change infiscal balance—as an instrument for the actual pol-icy adjustment. The projected policy adjustmentshould meet the two requirements of a valid instru-ment: the projections should be correlated withwhat actually happened and not be correlated withthe capital account shocks that occur during theprogram.3

The quality of the macroeconomic data for thesample considered here is not always high. This im-plies that clear and robust empirical regularities areunlikely to emerge and that all results have to be in-terpreted with caution. Nevertheless, the empiricalevidence presented in this paper seems to provide aconsistent picture of the effects of IMF programs.The first main finding is that program success doesnot seem to be a purely a random event; in particu-lar, success is negatively associated with the size offinancial assistance, especially in countries with ac-cess to private capital markets. Although this find-ing could mean that larger programs cause capitaloutflows, this causal interpretation is somewhat am-biguous, because I find some evidence that biases inprojections for net private capital flows may accountfor the negative association between success andprogram size. These biases might be caused by thefact that the IMF’s limited resources constrain theamount that can be lent in a program, especially

when this amount is large (in relative terms). Whenthese constraints are binding or close to being bind-ing, the IMF’s staff is more likely to feel pressure togenerate relatively more optimistic projections tocover any residual financing gap that cannot be cov-ered by additional program lending. Since this opti-mism is likely to be larger in larger programs, itcould explain why shortfalls in net private capitalflows are more likely to occur in larger programs.

The second main finding is that policy adjust-ment, especially in monetary policy, contributes toprogram success. Although it is not possible to esti-mate the causal effect precisely, it is, as expected,considerably smaller than the ordinary-least-squares(OLS) estimate—that is, the estimate that is com-puted without correcting for policy endogeneity. Thisfinding therefore suggests that, in countries with mar-ket access, exogenous capital account shocks are veryimportant determinants of what policy adjustmentthe domestic authorities can undertake.

The remainder of this section reviews some re-lated literature. The following section presents themain data sources and the empirical framework usedin the paper. The third section, “Are the Shortfallsin Net Private Capital Flows Random Errors?” pro-vides some basic stylized facts about the paper’smain object of interest—the shortfalls in net privatecapital flows—and a simple test of projection effi-ciency. The fourth and fifth sections—“Does Pro-gram Size Matter?” and “How Do Capital Flows Re-spond to Policy Adjustment?” respectively—studywhether financial assistance and policy adjustmentmatter for program success. The sixth, and final, sec-tion provides conclusions.

The empirical literature on the effects of IMF pro-grams is large and growing; here I will review only afew papers on the effects of programs on capitalflows.4 Several papers have studied the catalytic ef-fects of IMF programs—that is, the validity of thehypothesis that a program can restore the confi-dence of international investors and thus spur pri-vate capital inflows. After thoroughly reviewing theliterature, Cottarelli and Giannini (2002) and Birdand Rowlands (2002a and 2002b) conclude that theevidence of catalytic effects is weak. Ghosh and oth-ers (2002) reach similar conclusions in the contextof the capital account crises that occurred duringthe 1990s.

Most of the literature does not take into accountthe potential bias that can result from the fact thatadjusting the current account tends to lower the de-mand for net private capital inflows. For example,

3 � DO IMF-SUPPORTED PROGRAMS BOOST PRIVATE CAPITAL INFLOWS?

36

3 One problem with this empirical strategy is that the instru-mental variable estimate of the causal effect of policy adjustmentis inconsistent if the program projections are biased. However,the magnitude of this bias can be quantified and turns out to befairly small.

4 A recent example of the more general literature on the ef-fects of IMF programs is Barro and Lee (2002). See Haque andKhan (1998) and Krueger (1998) for surveys of earlier studies onthe effects of IMF programs, and Ramcharan (2003) for a recentsurvey. There is also an abundant literature on the moral hazardimplications of IMF programs; see Dell’Ariccia, Schnabel, andZettelmeyer (2002) and the references therein.

Rodrik (1996) studies the effect of past net bilateraland multilateral transfers on net private capitalflows by regressing the latter on the former and findsthat the effect of IMF lending is either not signifi-cantly different from zero or negative; Bird andRowlands (1997) reach similar conclusions using ananalogous approach. Mody and Saravia (2003) lookat gross capital inflows (specifically, new bond is-sues) to avoid confusing the program-induced exter-nal adjustment with the lack of catalytic effects, be-cause, if programs have a positive effect on investorconfidence, one should observe more issues at betterterms. It is not entirely clear, however, how achange in the net demand for external funds trans-lates into a change in gross flows. Similar conceptualdifficulties affect those papers that focus on specificasset classes or investment decisions, such as debtrestructuring (Marchesi, 2003) or foreign direct in-vestment (Edwards, 2000); moreover, it is difficultto gauge the macroeconomic relevance of findingsbased on individual asset classes.5 Since programprojections for net private capital flows have, inprinciple, to take the current account adjustmentinto account, my definition of program success isnot subject to this problem.

Little empirical research has been devoted to theeffect of program size. An exception is the paper byMody and Saravia (2003), which finds that largerprograms help raise the probability of issuing newbonds at a lower spread. My paper sheds light on theeffect of size by studying whether size has a system-atic relationship with the shortfall in net privatecapital flows.

Most of the literature does not take into accountthe fact that policy adjustment is endogenous—inspite of playing a central role in the programs, policyadjustment is often not taken into account at all.One approach that does attempt to control formacroeconomic policies is the Generalized Evalua-tion Estimator (GEE) discussed in Haque and Khan(1998) and recently applied by, among others,Dicks-Mireaux, Mecagni, and Schadler (2000) andBulír and Moon (2003). This approach attempts toestimate the economic policies that would havebeen in place in the absence of the program; thesepolicy counterfactuals are then introduced as a con-trol variable in the equation for the variable of in-terest, such as GDP growth. Because it uses policycounterfactuals rather than the actual policies fol-lowed under the program, the GEE approach doesnot have much to say about the effect of actual ad-justment—in contrast to this paper. The approachof this paper is close to the one followed by Mussoand Phillips (2001), which analyzes the effect of ac-tual program implementation on programs’ projec-

tion errors. However, these authors do not considerthe fact that program implementation is likely to beendogenous. To control for policy endogeneity, I usethe projected adjustment as an instrument for actualadjustment; a similar approach has been used byBerg and others (1999).6

Empirical Framework

In this section, I present the main source of the dataused in the paper and then discuss the empiricalframework that underpins the analysis of the follow-ing sections.

Data Sources

The main source is the Monitoring of IMF Arrange-ments (MONA) database, maintained by the IMF’sPolicy Development and Review Department. Istudy the Stand-By Arrangements and extendedarrangements that have been included in MONAfrom its inception in 1992 through 2001. These arethe arrangements used by the IMF for its ordinary,nonconcessional lending activity; in the following, Irefer to them as “programs.” The sample contains136 programs, including 105 Stand-By Arrange-ments and 31 extended arrangements.7

I construct a residual measure of net private capitalflows by subtracting from the overall balance thecurrent account balance, official transfers, and offi-cial multilateral and bilateral borrowing. This mea-sure of net private capital flows corresponds to thesum of the financial account balance and net errorsand omissions in the Fifth Edition of the Balance ofPayments Manual (IMF, 1993).8 Using the informa-tion available at the start and conclusion of the pro-gram (or last available review), I construct measuresof the initial program projections for and actual re-alizations of net private capital flows as percentagesof gross domestic product, also provided by MONA.

The timing is delicate. For example, I could com-pare the projections and the realizations during thecalendar year the program is approved, which I referto as year T. However, this comparison would con-fuse the preprogram developments with the responseto the program. Therefore I choose to compare the

5 This bias is likely to be less severe in those papers that ana-lyze how interest rate spreads respond to programs, such asEichengreen and Mody (2001) and Mody and Saravia (2003).

37

Roberto Benelli

6 Finally, there are a few related studies on the consequences oftight monetary policy during the Asian crisis (see, for example, Ba-surto and Ghosh, 2000) or more generally on the effect of mone-tary policy on speculative attacks (see, for example, Kraay, 2003).

7 MONA includes only three programs started in 1992. Theactual number of programs used in the empirical analysis belowwill vary depending on data availability.

8 My definition of net private capital flows corresponds to thesum of the financial account and net errors and omissions exceptfor any transfer and borrowing from multilateral and bilateral en-tities that may be included in the financial account. This defini-tion implicitly assumes that net errors and omissions are financialaccount transactions.

projections and outcomes in year T+1,9 ignoringthe very short-term response of net private capitalflows.

The basic unit of observation is an individual pro-gram. Treating each program as a distinct episoderaises thorny conceptual difficulties. First of all, anycountry may have multiple programs over the sam-ple period. Furthermore, many substantial changes—for example, in program length or size—can inter-vene during a program.

Finally, I use the IMF’s World Economic Outlook(WEO) database for series on purchasing-power-parity (PPP)-adjusted gross domestic product andthe stock of external liabilities.

Why Use the Shortfalls in Net Private Capital Flows?

Suppose that the net private capital flows (denotedby k) to a country over a certain period followingthe adoption of a program are governed by the fol-lowing equation:

(1)

where α is a constant term; k0 is a term that capturesthe effect on capital flows of initial conditions; P is avector of program variables, e.g., policy adjustmentand size of financial assistance; ε is an unobservedshock; and β is a vector of coefficients on programvariables. In this framework, the signs and magni-tudes of the coefficients in β determine whether andto what extent program variables help generate cap-ital flows. Besides program variables, net privatecapital inflows respond to exogenous shocks, cap-tured by the term ε. Examples of exogenous shocksare changes in the terms of trade, unexpected risesin export production, or changes in interest rates inadvanced countries. As these shocks are, by defini-tion, uncorrelated with the right-hand-side variablesin the regression, there will be no bias in the esti-mated coefficients.

Many problems in empirically evaluating the ef-fects of IMF programs stem from the fact that theeconometrician cannot observe precisely the pro-gram’s initial conditions, i.e., the term k0 in equa-tion (1). For example, countries are likely to startprograms when they face serious economic difficul-ties, i.e., when k0 is “unusually” low. Since the pro-gram characteristics, such as the size of financial as-sistance and policy adjustment, are likely to benegatively correlated with k0—because the largerthe initial economic difficulties, the larger the fi-nancial assistance and policy adjustment—ignoringk0 when estimating equation (1) produces inconsis-tent estimates of β.

The approach based on explicitly modeling the se-lection of countries into programs (see, e.g., Edwards,2000; and Przeworski and Vreeland, 2002) aims atinserting an omitted variable into an outcome equa-tion such as (1). Intuitively, this approach comparesprogram countries with countries without programsto estimate the probability of starting a program.This information is then used to construct a variable(the inverse Mill’s ratio), which plays the role of k0in equation (1), that corrects for the omitted vari-able bias owing to self-selection. Having correctedfor the self-selection problem, the econometriciancan consistently estimate β and thus infer the effectsof program variables on private capital flows.

In addition to the self-selection problem, k0 isalso related to the size of external adjustment thatthe country is to undertake under the program. Forexample, suppose that the program requires that acountry considerably reduce its current accountdeficit. This adjustment would also reduce the de-mand for net private capital flows, in turn implyinga smaller value for k0. Since the magnitude of theexpected external adjustment is likely to be posi-tively correlated with program variables, the omis-sion of k0 would bias the estimate of β.

Rather than constructing k0, I use the projectionsfor net private capital flows made by the IMF at thebeginning of the program. Using projections to con-trol for the initial conditions has two main advan-tages. First, I avoid the complications of explicitlymodeling the selection into IMF programs. Second, Iindirectly take into account the size and nature of theshocks that lead to the program; since these shocksshould in principle be taken into account by IMFstaff “on the ground” when formulating their projec-tions, they are included in the estimated equation.The main assumption underlying this approach isthat the projections are formulated within a uniformand consistent framework across programs.10

Finally, to highlight that the shortfall in net pri-vate capital flows relative to their projections is mymeasure of program success, I rearrange equation (1)to work with the shortfall

–k – k0 as my dependent

variable.

Are the Shortfalls in Net Private Capital Flows Random Errors?

This section documents some basic facts about theprojections for net private capital flows and their

38


k k P= + + +α β ε0 ,

9 Because of data constraints—my dataset is annual—I cannotcontrol for the exact timing of program approval.

10 It is also worth noting that the exchange rate regime doesnot affect the notion of program success. The exchange rateregime is likely to affect what happens when actual flows differfrom their projections but not whether a country receives asmuch capital as projected at the start of the program. For exam-ple, if net private capital flows exceed their projections, foreignreserve accumulation will occur in a fixed exchange rate regimeand the exchange rate will appreciate in a pure float.

shortfalls and then provides a simple test of projec-tion efficiency.

Some Summary Statistics

Figure 1 plots the distribution of the main object ofinterest, the shortfall in net private capital flows inyear T+1 of the program, defined throughout thepaper as the difference between projected and actualnet private capital flows in percent of GDP. Al-though the distribution of the shortfalls appears to befairly symmetric, it is centered below zero, i.e., theprojections tend to be above the outcomes. The fig-ure also shows outliers, large positive and large nega-tive shortfalls corresponding to big failures and suc-cesses, respectively.

An important characteristic is whether a programis precautionary—that is, whether the domestic au-thorities state their intention not to draw the re-sources available under the program. There is empir-ical evidence that precautionary programs may beparticularly suitable for conveying positive signals toprivate investors (see Mody and Saravia, 2003). Tominimize the likelihood that the precautionary na-ture of a program is endogenous, my definition of“precautionary” includes only those programs thatare precautionary on approval and not those that

“turn precautionary” at a later stage.11 This is be-cause turning a program into a precautionary one isunlikely to be an exogenous policy decision. For ex-ample, unexpected large capital inflows owing to ex-ogenous external reasons could ease the external fi-nancial constraint faced by a country, allowing it toturn its program into a precautionary one.

I identify the following stylized facts from Panel Aof Table 1:

• Precautionary programs project larger net pri-vate capital inflows than do nonprecautionaryprograms, in terms of both mean and median.This is likely to depend on the fact that precau-tionary programs start in more tranquil times.

• In spite of being higher, the projections in pre-cautionary programs tend to be “conserva-tive”—that is, both the mean and medianshortfall are negative—the mean shortfall isclose to being statistically negative at the 10

39

Roberto Benelli

0

5

10

15

20

25

s < –4 –4 < s < –3 –3 < s < –2 –2 < s < –1 –1< s < 0 0 < s < 1 1 < s < 2 2 < s < 3 3 < s < 4 s > 4

Shortfalls in Net Private Capital Flows (In percent of GDP)

Freq

uenc

yFigure 1

Distribution of Shortfalls in Year T+1 for All Programs

11 Two types of programs are precautionary on approval. Thefirst type consists of those programs that are explicitly negotiatedas such. The second type consists of a smaller subset of programsthat are approved as nonprecautionary programs but in whichthe domestic authorities choose not to make the first drawing. Ido not consider those programs (13 in the sample) that start asnonprecautionary but turn precautionary at a later stage.

percent confidence level. By contrast, the non-precautionary programs tend to be “right,” i.e.,both the mean and median shortfalls are closeto zero.

• Nonprecautionary programs tend to be moreheterogeneous than precautionary programs,both in terms of initial projections and subse-quent shortfalls.

Does a country’s access to private capital marketsaffect how capital flows respond to a program? I mea-sure a country’s market access using the IMF’s WEOclassification of developing, nontransition countriesby main source of external financing. This classifica-tion defines three groups of countries depending onthe main source of external financing: countries rely-ing on official financing, on private financing, andon diversified financing.12 I interpret this classifica-tion as a measure of capital account openness be-cause countries with more open capital accounts aremore likely to borrow from private sources.13

Panel B shows that relying on private sources ofexternal financing does not necessarily lead to largershortfalls; instead, it is the countries with diversified

financing that tend to have the largest shortfalls.Although, as expected, the projected net privatecapital flows are larger in countries with private fi-nancing (in terms of both mean and median), theshortfalls tend to be larger in countries with diversi-fied financing; only for precautionary programs arethe median (but not the mean) shortfalls larger incountries with private financing. The average short-falls are not, however, significantly different fromzero, possibly because of the very small number ofobservations in each subgroup.

To include transition economies, I construct anew dummy variable for market access by takinginto account whether a transition economy is in-cluded in the JP Morgan Emerging Markets BondIndex (EMBI) Global. Specifically, the dummy vari-able takes the value of 1 if a transition economy isincluded in this index and if a nontransition econ-omy (classified by WEO) relies on private or diversi-fied financing.14 Panel C shows that the shortfallstend to be larger in those countries with market ac-cess (in terms of both mean and median).

A Simple Test of Projection Efficiency

Comparing projections of private capital flows withactual flows is interesting per se because of its impli-cations for projection efficiency. Although statisti-cal tests of the efficiency of forecasts are common in

40


Table 1

Summary Statistics of Projections and Shortfalls in Net Private Capital Flows, by Type of Program and External Source of Financing

Nonprecautionary Programs Precautionary Programs

Projected NKF1 Shortfalls in NKF1 Projected NKF1 Shortfalls in NKF1

Mean Median Mean Median S.D. Mean = 02 N Mean Median Mean Median S.D. Mean = 02 N

A: All –0.42 0.04 0.02 –0.17 5.82 0.48 84 3.09 2.44 –1.11 –0.59 5.57 0.12 35

B: Source of external finance3

Private –0.76 1.57 0.69 –0.18 4.79 0.28 16 2.9 2.32 0.51 0.61 1.1 0.11 8Diversified –2.9 –1.41 4.45 0.65 11.42 0.15 8 2.54 2.21 2.58 0.23 6.1 0.17 6Public –2.91 –2.16 0.51 0 4.98 0.31 23 –1.02 –0.32 –5.17 –0.69 9.14 0.07 8

C: Market access4

Yes –1.04 0.07 1.47 0 6.58 0.1 33 2.79 2.32 1.19 0.49 3.82 0.11 16No 0.05 0.04 –0.92 –0.49 5.12 0.1 51 3.35 2.74 –3.06 –1.21 6.14 0.02 19

1 NKF denotes net private capital flows. S.D. denotes standard deviation. N denotes number of programs.2 P-value for the test of the hypothesis that the mean shortfalls is zero against the alternative hypothesis that the mean is positive (if mean

shortfall is positive) or negative (if mean shortfall is negative).

Classification

3 The main source of external financing is from IMF, World Economic Outlook, various issues. (See text for discussion.)4 Market access is a dummy variable that takes the value one if the country’s main source of external financing is not official (for the devel-

oping, nontransition economies) or if the country is included in the JP Morgan’s EMBI Global Plus Index (for the transition economies).

14 According to this criterion, 33 nonprecautionary programs(39 percent of the total) and 16 precautionary programs (45 per-cent of the total) take place in countries with market access.

12 A net debtor country is allocated to either of the first twosubgroups if official sources (including official grants) or privatesources (including direct and portfolio investment) account forat least two-thirds of its total external financing in the four yearsbefore the country is classified. Countries that do not meet thesetwo criteria are classified as relying on diversified financing.

13 This notion is somewhat similar to other de facto capital ac-count openness measures that are based on the actual behavior ofcapital flows; see Edison and others (2002) for a survey.

a variety of contexts,15 they may not be appropriatein the context of programs. Their main limitation isthat they assume that the projections are uncondi-tional optimal forecasts (in a statistical sense) of thevariable being projected. Yet, program projectionsare conditional on the implementation of the policyadjustment negotiated between the domestic au-thorities and the IMF; moreover, they may them-selves result from negotiations (see Musso andPhillips, 2001; and Mussa and Savastano, 1999).

Figure 2 plots actual net private capital flows inpercentage of GDP against their projections; the fig-ure confirms the large variation in projections andoutcomes pointed out above. However, no majorbias is apparent, since actual flows are roughly dis-tributed along the 45-degree line. (The slope of theleast-square line is 0.91, with an intercept of 0.28.)

A formal test of projection efficiency can be basedon the intuitive idea that projections should be“right on average.” In other words, if projectionsrely on all the available information, it should notbe possible to predict systematically the projectionerror. For example, if the projections were on aver-

age optimistic, lowering them would eliminate theerror. This implies that the projection error shouldnot be correlated with the projection itself. In theregression of actual net private capital flows on pro-jections,

(2)

efficiency requires that the intercept be 0 and theslope 1. In column 1 of Table 2, based on the sampleof all programs, the projections for net private capitalflows meet the minimal efficiency requirement (thisis also documented by Musso and Phillips, 2001): theslope of the equation is very close to 1. Althoughthe intercept is positive, it is not statistically signifi-cant at the usual confidence levels; the joint hy-pothesis that the constant is 0 and the slope is 1cannot be rejected at the usual confidence levels.

Columns 2 and 3 carry out the test separately forprecautionary and nonprecautionary programs. Incolumn 2, the test of efficiency does not reject thehypothesis that projections are efficient for non-precautionary programs. On the other hand, column3 shows that projections do not meet the efficiencyrequirement for precautionary programs: the slopecoefficient is well below 1 (and not statistically sig-nificant from 0) and the constant is very large andstatistically significant from 0 at the 10 percent significance level; the test statistic of the null joint

41

Roberto Benelli

15 See Loungani (2000) and Zitzewitz (2001) for applicationsto consensus forecasts and equity earning forecasts, respectively.Musso and Phillips (2001) reviews various concepts of forecastefficiency.

–40

–30

–20

–10

0

10

20

30

40

–20 –15 –10 –5 0 5 10 15 20 25

Projected Net Private Capital Flows in Percent of GDP

Actu

al N

et P

rivat

e N

et C

apita

l Flo

ws

in P

erce

nt o

f GD

PFigure 2

Projected and Actual Net Private Capital Flows for All Programs

k k u= + +α β ,

hypothesis that the slope is 1 and the constant is 0 isclose to the rejection region. (The p-value is 0.155.)However, this result may be driven by three observa-tions corresponding to two large negative shortfallsand one large shortfall: if these observations weredropped, projections would pass the efficiency test.

For future reference, it is worth noting that thesmall slope coefficient found for precautionary pro-grams implies that the shortfalls tend to grow withthe magnitude of the projections; this observationwill be useful below to interpret the findings on pro-gram access.16

These tests, together with the summary statisticspreviously shown, suggest that the shortfalls in netprivate capital flows may not be truly random errors.In particular, it is worth asking whether programvariables can explain them, in the following sense:

• Are the shortfalls systematically associated withprogram size? This question is addressed in thefourth section.

• Can macroeconomic adjustment explain theshortfalls? Does the association between short-falls and policy adjustment, if present, arisefrom the causal effect of policy adjustment oncapital flows or is it due to reverse causationfrom capital flows to policy adjustment? Thisissue is studied in the fifth section.

Does Program Size Matter?

At a theoretical level, it is unclear whether programsize affects private capital flows. On the one hand,

larger programs could help attract private investorsbecause, for example, larger programs could inducebetter program design and policy monitoring by im-proving IMF incentives (Rodrik, 1996) or becausethey could signal IMF confidence in a country’s poli-cies.17 Moreover, by providing foreign currency re-serves that can be used to fend off a speculative at-tack, a larger IMF program could lower both capitaloutflows and the likelihood of crises (see Morris andShin, 1998; and Corsetti, Guimarães, and Roubini,2003). The effect of program size could be nonlinearbecause programs may need to be as large as thecountry’s external liabilities to restore investor confi-dence (see Chang and Velasco, 2000)18 or becausethe government’s optimal adjustment effort couldvary in a nonmonotonic fashion with program char-acteristics (see Morris and Shin, 2003; and Corsetti,Guimarães, and Roubini, 2003). On the other hand,larger programs could simply facilitate the flight ofdomestic and foreign investors by providing much-needed foreign currency resources. This paper ad-dresses this ambiguity by studying the associationbetween shortfalls in net private capital flows andprogram size.

A country quota, the country share in the IMFcapital, constrains the amount a country can borrowunder a program—program “access” in the IMF ter-minology. Currently, annual and cumulative limitsconstrain access to 100 percent and 300 percent ofquota, respectively. However, the IMF’s ExecutiveBoard can waive these access limits in case of excep-tional circumstances; starting from the end of 1997,

42


17 Cottarelli and Giannini (2002) review the channels throughwhich adopting an IMF program could affect the response of in-ternational investors.

18 Partial bailouts may be ineffective, or even precipitate crises(see Zettelmeyer, 2000).

Table 2

Test of Projection Efficiency

(1) (2) (3)All Nonprecautionary Precautionary

Dependent variable: NKF in year T+1

Projection 0.918*** 0.972*** 0.312(0.103) (0.113) (0.376)

Constant 0.280 –0.032 3.180*(0.536) (0.641) (1.614)

R-squared 0.428 0.470 0.048F-test for unbiased projection1 0.44 0.03 1.97Probability > F 0.644 0.970 0.155Number of observations 119 84 35

Notes: Robust standard errors appear in parentheses. NKF stands for net private capital flows. An asterisk (*) denotes significance at 10 per-cent, (**) at 5 percent, and (***) at 1 percent.

1 Test of the joint hypothesis that slope is 1 and constant is 0.

16 The large point estimate of the constant implies that precau-tionary programs with low projections for net private capital in-flows tend to exhibit large negative shortfalls.

exceptional access has been granted using a dedi-cated lending facility, the Supplemental Reserve Fa-cility.19

The first data issue is to measure access itself. Imeasure program access as the original program accessper program year—that is, access approved at thestart of the program divided by the original programmaturity (in years). This definition has two advan-tages. First, it ensures that access is comparable tothe variable of interest, shortfalls in net private cap-ital flows in year T+1. Second, it ensures that I donot treat programs with longer maturity as largerprograms. This definition also implies that I do nottake into account that access can change during theprogram—in fact, augmentations of program accessare frequent, although reductions are less frequent.Since I do not include augmentations or reductionsas part of program access, I minimize the likelihoodthat my measure of access is endogenous.20

The second data issue arises because, to carry outa cross-program analysis, I need to scale program ac-cess. Quotas, the official IMF scaling factor for pro-gram access, may not reflect accurately the “true”country size, implying that access as a percentage ofquota may not accurately reflect program size.21

Therefore I consider other scaling factors for pro-

gram access. The first plausible alternative is to use acountry’s GDP. In particular, I use the average PPP-adjusted GDP between year T–3 and year T to pre-vent nominal exchange rate movements and outputcollapses before the program from affecting my mea-sure of program size—although I find similar resultsif I use GDP in U.S. dollars at market exchangerates. I also scale access by the stock of external lia-bilities at the end of year T–1 because some of thetheoretical papers discussed above predict that thestock of external liabilities is the relevant measureto gauge program size.

Although quotas may not accurately reflect pro-gram size, they may still matter as a scaling factor be-cause they determine how the IMF operates as a mul-tilateral policy institution. On the one hand, a largeprogram in percentage of quota may signal IMF con-fidence in a country’s economic policies. On theother hand, large access in percentage of quota maysignal that the lending constraints are binding orclose to being binding. In practice, these lendingconstraints are likely to be smaller than the officialannual and cumulative access limits mentioned.

Table 3 shows that, uniformly across the threescaling factors, precautionary programs tend to grantlower access (in terms of mean and median) and tobe more homogeneous (as measured by the standarddeviation) than nonprecautionary programs.22 Fur-thermore, mean access is larger than median accessacross the three scaling factors, reflecting the pres-ence of a few very large access programs. A recent,thorough review of access in IMF programs is pro-vided by IMF (2003a, 2003b).

43

Roberto Benelli

19 See the IMF (2003a, 2003b) review on access limits. Thepurpose of the Supplemental Reserve Facility is to lend to coun-tries that face capital account pressures owing to financial conta-gion.

20 This problem arises if, as plausible, changes in access takeplace in response to how private capital flows react to programcharacteristics. A similar problem would arise if actual disburse-ments or outstanding credit were used (as done by Barro and Lee,2002).

21 For example, although the programs in the Republic ofKorea (1997), Brazil (1998), Thailand (1997), and Indonesia(1997 and 1998) stand out as large programs in terms of percent-age of quota, they are comparable to many other programs in per-centage of GDP.

Table 3

Summary Statistics of Program Access


Access Access Access Access (percent of Access Access (percent of

(percent of (percent of external (percent of (percent of external quota) GDP)1 debt)2 quota) GDP)1 debt)2

Mean 59.8 0.46 5.89 40.03 0.33 2.40Median 43.3 0.40 2.17 27.7 0.27 1.43Standard deviation 79.84 0.29 9.86 48.94 0.20 3.24Minimum 11.1 0.10 0.39 15.6 0.12 0.53Maximum 646.2 1.95 58.67 320 1.07 15.73Number of observations 97 95 94 39 39 3

1 Average of purchasing-power-parity (PPP)-adjusted GDP in years T–3 through T. 2 Total external debt, excluding programs with average initial access exceeding 100 percent of external debt.

22 Table 3 excludes two nonprecautionary programs in transi-tion economies for which access in percentage of external debtexceeds 100 percent owing to the very low initial external debt.Including them causes the standard deviation of access in per-centage of external debt to jump to over 80 (from below 10) andthe mean to jump to about 15 (from about 5).

The main finding from the regression analysis inTable 4 is that program access is positively and sig-nificantly associated with the shortfalls in net pri-vate capital flows only when it is scaled by quotas.The coefficient on access in percentage of quota ispositive and significant for both nonprecautionaryand precautionary programs (see columns 1 and 8,respectively). By contrast, the other two measures of program access—access in percentage of PPP-adjusted average GDP and access in percentage ofexternal debt—are not statistically significant; forboth types of programs, access in percentage of PPP-adjusted GDP is positively associated with the short-falls (columns 2 and 9) while access in percentage ofexternal debt is negatively associated with them

(columns 4 and 10). It is also worth pointing out thatthe coefficient on access in percentage of quota is anorder of magnitude larger for precautionary programs;it is also quantitatively nonnegligible: an increase inaccess of 10 percentage points of quota is associatedwith a larger shortfall of about 1.5 percent of GDP.23

Why is access in percentage of quota significantlyassociated with the shortfalls in private flows whilethe other access measures are not? An explanation

44


Table 4

Shortfalls and Program Access

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)


Dependent variable: Shortfall in NKF in year T+1

Access (percent 0.017** 0.019** 0.019 0.073* –0.013 0.143** –0.016 0.006of quota) (0.008) (0.009) (0.018) (0.042) (0.028) (0.067) (0.090) (0.088)

Access (percent 2.391 1.721of average PPP- (1.566) (3.517)adjusted GDP)

Access (percent –0.023 –0.116of external debt) (0.041) (0.108)

Access (percent –0.001 0.021*of quota × (0.002) (0.012)projected NKF)

Exceptional access 22.469**(10.082)

Access (percent of –0.049*quota × exceptional (0.026)access)

Squared access –0.001***(percent of quota) 0.000

Access (percent of 0.029 0.111*quota × market (0.022) (0.058)access)

Constant –1.034 –0.990 0.269 –1.148* –1.478 –3.346** –0.244 –5.656** –1.642 –0.842 –2.884* –3.177*(0.630) (0.908) (0.753) (0.670) (1.063) (1.623) (0.825) (2.210) (1.556) (1.130) (1.678) (1.877)

R-squared 0.065 0.016 0.001 0.071 0.289 0.137 0.08 0.151 0.002 0.005 0.301 0.225

Number of observations 84 81 83 84 84 84 84 35 35 35 35 35

Notes: Robust standard errors appear in parentheses. NKF denotes net capital flows. PPP denotes purchasing power parity. An asterisk (*)denotes significance at 10 percent, (**) at 5 percent, and (***) at 1 percent. Column 4 is based on a sample that excludes programs with accessexceeding 100 percent of external debt.

23 If I drop the three outliers that imply that the projections inprecautionary programs are not efficient (see the third section,“Are the Shortfalls in Net Private Capital Flows Random Er-rors?”), the coefficient on access in percentage of quota is onlythree times as large for precautionary programs.

could be that the positive association betweenshortfalls and access in percentage of quotas is dueto institutional constraints on IMF lending. If lend-ing constraints—expressed in percentage of quo-tas—limit the amount that can be lent in a program,IMF staff may be forced to generate optimistic pro-jections for private capital inflows to “close” the fi-nancing gap that would otherwise occur. Since thepressure to generate more optimistic projections islikely to be stronger when access is large relative toquota, larger programs are likely to be associatedwith larger projection biases (in the direction of op-timism) and therefore with larger shortfalls.

As regards precautionary programs, projection biases may play a role to generate the positive asso-ciation between shortfalls and program access.Column 11 provides some evidence on this by intro-ducing an interaction term between program accessand projections for net private capital flows. I wouldexpect that, if optimistic projections substitute forprogram access at higher levels of access, the posi-tive association between shortfalls and program ac-cess should become stronger as the projections in-crease (because, at higher access, projections needto be more optimistic). The interaction term turnsout to be positive and statistically significant androughly accounts for the magnitude of the coeffi-cient on access in column 8,24 while the coefficienton access becomes statistically insignificant. As re-gards nonprecautionary programs, the interactionterm is fairly small and statistically insignificant (seecolumn 4), implying that the association betweenprogram access and shortfalls may be genuinelystructural, i.e., that larger programs could causelarger shortfalls.

Do exceptional access programs account for thepositive association between access and shortfalls?25

Column 5 introduces a dummy variable for excep-tional access and an interaction term between thisdummy and program access. Interestingly, the coeffi-cient on access in percentage of quota remains veryclose to the estimate in column 1, but is not signifi-cant at the usual confidence levels. The dummyvariable for exceptional access is large and statisti-cally significant (at the 5 percent level)—not a sur-prising finding given that IMF programs during thecapital account crises of the 1990s witnessed verylarge and unexpected net capital outflows (seeGhosh and others, 2002)—and the interaction term

is negative and statistically significant (at the 5 per-cent level). Thus, within the exceptional accessprograms, larger programs were associated withsmaller shortfalls. Although the very small numberof cases does not allow general conclusions on thesystematic effect of exceptional access, the largestprograms do not seem to account for the positive as-sociation between program access and shortfallsdocumented earlier. Column 6 bolsters this conclu-sion: the quadratic term does not pick up any impor-tant nonlinearities.

Does access to private capital markets matter forthe association between program access and short-falls in net private capital flows? Columns 7 and 12introduce an interaction term between program ac-cess and the market access dummy described in theprevious section. Relative to columns 1 and 8,where the coefficient on program access does notvary with market access, the coefficient on access issmaller and statistically insignificant for both non-precautionary (column 7) and precautionary pro-grams (column 12). Instead, program access has alarge coefficient when interacted with the marketaccess dummy:26 the interaction term accounts en-tirely for the effect of program access. In otherwords, the positive association between shortfallsand program access is present only in countries withmarket access.27

Why does program access matter only in coun-tries with market access? An explanation could bethat restricting the sample to countries with marketaccess simply “cleans” the data by removing coun-tries that rely mainly on official financing and thusdo not provide information on how private capitalflows respond to programs. Yet, countries withoutmarket access may still have significant interactionswith private capital markets. For example, capitalflight can take place even in countries that officiallyhave low access to private capital markets. Thisfinding could then mean that market access booststhe capital outflows financed by IMF-provided for-eign currency reserves. This conclusion is not com-pletely uncontroversial, since projection biasesmight be more likely in countries with market ac-cess because, in these countries, the constraints onIMF lending might be more likely to bind. Overall,

45

Roberto Benelli

24 The marginal effect of access is 0.034 when evaluated at themedian program access, climbing to 0.112 when evaluated at themedian plus one standard deviation of access. However, this re-sult is not robust to outliers: the interaction term is negative butstatistically insignificant if outliers are dropped.

25 There are six nonprecautionary programs in the sample thatgranted “exceptional access” on approval: Mexico in 1995; theRepublic of Korea, Thailand, and Indonesia in 1997; and Brazil in1998 and in 2001. Cases in which exceptional access was grantedby augmenting existing programs are not considered here.

26 The interaction term is significant at the 10 percent levelonly for the precautionary programs, and is very close to beingsignificant at the 10 percent level for the nonprecautionary pro-grams.

27 Shortfalls and program access are positively associated onlyin countries with market access especially when I measure accessin percentage of external debt: the coefficient on access in per-centage of external debt estimated on countries with market ac-cess is positive, large, and statistically significant, and is evenlarger when I exclude exceptional access cases (the R-squared co-efficient climbs to over 40 percent). There is virtually no associa-tion between shortfalls and program access in the sample ofcountries without market access. These estimates are availablefrom the author upon request.

however, it seems difficult to explain the positive as-sociation between shortfalls and program access entirely as a consequence of projection bi-ases,28 i.e., this association provides some evidencethat market access tends to lower the likelihood ofsuccess.

I can summarize the findings on program access asfollows:

• I have found some evidence of a positive associ-ation between shortfalls in net private capitalflows and program access.

• This evidence is consistent with the hypothesisthat larger programs cause private capital out-flows. However, the causal interpretation of thisassociation is ambiguous because of potentialprojection biases. In particular, I have foundsome evidence that lending constraints mayforce IMF staff to produce more optimistic pro-jections at higher levels of access, which, inturn, implies that larger shortfalls are morelikely at higher levels of access.

• Exceptional access cases or failure to accountfor nonlinearities do not seem to account forthese findings. Moreover, this association isstronger in countries with market access, i.e.,market access seems to lower the probability ofsuccess (but the effect of market access is likelyto be nonmonotonic).

How Do Capital Flows Respond to Policy Adjustment?

How private capital flows respond to policy adjust-ment has received considerable attention. In thissection, I assume that domestic authorities can in-fluence capital flows by adjusting their macroeco-nomic policies—I do not consider here the othermajor dimension of policy adjustment in programs,structural adjustment. The econometric problem inestimating the causal effect of policy adjustment isthat policy adjustment is endogenous, i.e., it is likelyto depend on the developments in the balance ofpayments, and hence on capital flows. For example,tight monetary and fiscal policies may not be sus-tainable if a negative capital account shock causes adeep recession. The following simple system de-scribes the feedback between policy adjustment andshortfalls in net private capital flows:

(3)

(4)

with E(ε)=E(η)=E(εη)=0. Since my purpose is toidentify the portion of the shortfalls in net privatecapital flows owing to policy adjustment, equation(3) decomposes the shortfall in net private capitalflows (

–k– k) as a sum of the effect of actual policy

adjustment (p) and an exogenous shock (ε). Thisshock is shorthand for all the factors that affect cap-ital flows for reasons beyond the authorities’ control,e.g., interest rates in industrialized countries orterms of trade changes; for simplicity, I call it a capi-tal account shock. In equation (4), actual policy ad-justment is the sum of the initial projection (

_p); an

exogenous policy shock (η), e.g., a shock to thepreferences of the domestic authorities; and a termcapturing the feedback of capital account shocks onpolicy adjustment (λε). For example, if λ < 0, then ashock ε > 0 raises the shortfall and simultaneouslyreduces policy adjustment. Whenever λ ≠ 0, theOLS estimate of the policy parameter β is inconsis-tent because policy adjustment is endogenous.

The main identifying assumption in equation (3)is that the projected policy adjustment

_p does

not appear on the right-hand side. This exclusion re-striction is based on the assumption that the IMF’sstaff takes into account all the relevant informationon what is known to affect capital flows when it for-mulates its projection

–k, including, in principle, the

relationship between the projected policy adjust-ment

_p and the projected capital flows

–k.29 Under

this exclusion restriction and assuming that theshocks ε and η are orthogonal to the projection

_p,

the projection _p is an instrument for the actual pol-

icy adjustment p in equation (3). However, this in-strument may be invalid if the projections

_p and–

k have a common bias, e.g., owing to optimism inthe projections; I discuss this possibility at the endof the section.

Focusing on the relationship between policy ad-justment and shortfalls in net private capital flowsonly, as in equation (3), is not necessarily restric-tive. This is because the shortfalls are defined rela-tive to the initial projections for private capitalflows; these should, in principle, take into accountthe economic conditions that led to the program.For brevity, in what follows I focus on nonprecau-tionary programs.

Measuring Policy Adjustment

The first empirical problem is to measure policy ad-justment in a way that is comparable across countries.

46


k k p− = +β ε

p p= + +η λε,

29 Working with an equation in terms of shortfalls such as (3)rather than with an equation that includes the projected flows asa right-hand-side determinant for actual flows is not restrictivebecause the coefficient on projected flows turns out to be veryclose to one.

28 In particular, it is hard to believe that the statistically andquantitatively strong positive association between shortfalls andprogram access in percentage of external debt in countries withmarket access (see previous footnote), in which access in per-centage of external debt accounts for up to 40 percent of thevariation in shortfalls, is purely due to projection biases.

Measuring the stance of economic policies is a well-known problem in the literature on cross-countrygrowth regressions, where the economic policystance is studied as a possible determinant of eco-nomic growth (see, e.g., Fischer, 1993), and in theliterature on the effects of monetary policy on spec-ulative attacks, where the stance of monetary policyaffects the likelihood of a currency attack (see, e.g.,Kraay, 2003).

To measure the monetary policy stance, I wouldideally use a policy instrument that is directly con-trolled by the monetary authorities, e.g., the dis-count rate charged by the central bank on the liq-uidity provided to the commercial banks. A simplealternative is to measure the monetary policy stancewith a policy outcome, e.g., inflation. The mainproblem with using inflation is that it can be seen asa more general indicator of macroeconomic policy(see Fischer, 1993).

I use inflation to measure monetary policy. Sinceprograms typically require an adjustment in eco-nomic policy, I focus on the change in the end-of-period inflation rate between years T and T+ 1,even though some policy adjustment may start be-fore the program is formally approved as “prior ac-tions” (see Knight and Santaella, 1997; and Mussaand Savastano, 1999). Given the lags in monetarypolicy, the inflation rate in year T is unlikely to fullyreflect the program-induced adjustment in mone-tary policy, especially for the programs approved inthe later months of the year. However, the inflationrate in year T+1 should reflect the policy adjust-ment owing to the program, and, as a result, thechange in inflation between years T and T+ 1should capture the monetary policy adjustment.

I use two alternative proxies for adjustment in fis-cal policy: the change in overall balance and in pri-mary balance between year T and year T+ 1. Theoverall fiscal balance captures the weight of thepublic sector in the economy and, in particular, thedemand for foreign capital originating from the pub-lic sector. However, using the overall balance hasthe shortcoming that interest payments are likely todepend on the supply of capital flows, implying thatthe overall balance may be endogenous. The pri-mary balance does not suffer from this shortcoming;moreover, it may better capture the policy effort ofdomestic authorities. I use both measures of fiscaladjustment in this paper.

Interpreting the association between shortfalls innet capital flows and fiscal adjustment is somewhatproblematic because, holding private savings con-stant, fiscal consolidation raises national savingsand correspondingly lowers the need for external fi-nancing. If the shortfall rises more than one for onewith the improvement in fiscal balance, fiscal ad-justment has “perverse” effects on capital flows be-cause the shortfall exceeds the reduction in the de-mand for external financing owing to the fiscaladjustment. By contrast, if the increase in the short-

fall is less than one for one, fiscal adjustment hascatalytic effects.30

Association Between Policy Adjustment and Capital Flows

When I do not take into account that policy adjust-ment is endogenous, I find a positive association be-tween the shortfalls in net private capital flows andthe change in inflation, i.e., a tightening in mone-tary policy is associated with a lower shortfall. Col-umn 1 of Table 5 shows that this association is sta-tistically significant at the 5 percent confidencelevel but quantitatively very small: reducing infla-tion by 10 percentage points is associated with asmaller shortfall of only 0.02 percent of GDP.

To assess whether large changes in inflation ac-count for this association, I estimate it by allowingthe coefficient on inflation change to vary with themagnitude of inflation change. There are reasons toexpect that large inflation changes may be neutralbecause high inflation economies tend to developsophisticated indexation practices; in the context ofeconomic growth, for example, Fischer (1993) findsno relationship between inflation and growth wheninflation is high. Column 2 estimates a piecewiselinear function with breaks at –70 and 70.31 I findthat the coefficient on intermediate inflationchanges is 0.083, which is larger than the coeffi-cients on large inflation increases and decreases,with both coefficients very close to zero. Althoughstatistically insignificant, the coefficient on inter-mediate inflation change is economically meaning-ful: reducing inflation by 10 percentage points is as-sociated with a smaller shortfall of almost 1 percentof GDP. Column 3 shows a similar point estimatewhen I restrict the sample to programs with inter-mediate inflation changes.

If restoring investor confidence is critical for howcapital flows respond to a program, as is often arguedin the literature on catalytic effects of IMF pro-grams, I would expect that the association betweenshortfalls and policy adjustment is stronger in coun-tries with market access. Column 4 shows that thisassociation is quantitatively stronger for these coun-tries: the point estimate (with a p-value of 0.12) im-plies that reducing inflation by 10 percentage pointsis associated with a smaller shortfall of about1.6 percent of GDP. To gauge its order of magnitude,it is worth recalling that average annual access fornonprecautionary programs is 1.6 percent of GDP.Thus, the reduction in the shortfall owing to a 10percentage point fall in inflation is as large as aver-age program access!

47

Roberto Benelli

30 This problem seems less severe with regard to monetary ad-justment because, in contrast to fiscal policy, this effect does notarise directly from the balance of payments identity.

31 I found virtually identical estimates using higher thresholdsbut no statistically significant results using smaller thresholds.

Concerning fiscal policy, the association betweenfiscal adjustment and shortfalls in net private capitalflows is negative for both measures of fiscal adjust-ment, indicating that a larger fiscal adjustment is as-sociated with a smaller shortfall, but statistically in-significant and quantitatively small (see columns 5and 6 of Table 5). The point estimates imply that afiscal adjustment of 1 percent of GDP is associatedwith a lower shortfall of between 0.2 and 0.4 per-cent of GDP. This association is statistically andquantitatively stronger in countries with market ac-cess: in columns 7 and 8, the coefficients on overalland primary adjustment are much larger (in absolutevalue) and close to being significant (their p-valuesare about 0.2); the R-squared coefficients are alsolarger. These estimates imply that improving fiscalbalance by 1 percent of GDP is associated with asmaller shortfall of about 1 percent of GDP.

Causal Effect of Policy Adjustment

Although the previous findings show that policy ad-justment is associated with lower shortfalls in netprivate capital flows, it is not clear which way thecausation runs, i.e., the association between policyadjustment and shortfalls could be the effect of ex-

ogenous capital account shocks on policy adjust-ment. To control for policy endogeneity, I instru-ment for the actual policy adjustment using the pro-jected adjustment.

Starting from monetary policy, in equation (3) Iinstrument the actual inflation change using theprojected inflation change. Table 6 reports the two-stage least-square (2SLS) estimates of equation (3).The bottom panel of Table 6 reports the first-stageregressions of actual inflation change on its projec-tion. In the countries with market access (column2), the projections are statistically significant at the1 percent level in predicting actual inflationchanges and explain a large portion of its variation.The projections are less accurate in the broader sub-sample that includes the countries without marketaccess (column 1). Overall, there seems to be a suffi-ciently strong association between projections andactual inflation changes, as shown by the largeR-squared coefficients. I discuss at the end of thesection whether the second requirement of a validinstrument—that it be uncorrelated with the errorterm in equation (3)—is likely to be met.

Turning to the two-stage least-square estimates inthe top panel, the Hausman test points out a statisti-cally significant difference between the OLS and

48


Table 5

Association Between Shortfalls and Policy Adjustment for Nonprecautionary Programs

(1) (2) (3) (4) (5) (6) (7) (8)

Adjustment in Inflation Adjustment in Fiscal Balance

Excluding Excluding large

large inflationinflation changes— Market Market

All All changes1 market access All All access access

Dependent variable: Shortfall in NKF in year T+1

Inflation change1 0.002** 0.091 0.159(0.001) (0.084) (0.101)

Moderate inflation change2 0.083(0.085)

Large inflation decrease2 0.002**(0.001)

Large inflation increase2 –0.001 (0.002)

Overall adjustment –0.357 –0.927(0.408) (0.760)

Primary adjustment –0.206 –1.032(0.343) (0.894)

Constant 0.144 0.915 1.277 3.089* 0.329 0.178 1.446 0.806(0.678) (0.219) (1.363) (1.661) (0.767) (0.733) (1.049) (0.888)

R-squared 0.01 0.067 0.078 0.284 0.031 0.011 0.167 0.139Number of observations 83 83 73 29 82 78 33 30

Notes: NKF denotes net capital flows. An asterisk (*) denotes significance at 10 percent, (**) at 5 percent, and (***) at 1 percent.1 Large inflation changes are changes in inflation between years T and T+1 above 70 percentage points in absolute value.2 Moderate inflation changes, large inflation decreases, and large inflation increases are changes in inflation between –70 and +70 percent-

age points, below –70 percentage points, and above +70 percentage points, respectively.

2SLS estimates for only countries with market ac-cess, providing evidence that reverse causality is po-tentially important, at least in countries with marketaccess. As expected, the 2SLS estimates of the effectof inflation change are smaller than the previousOLS estimates because the latter pick up the effect ofexogenous capital account shocks on policy adjust-ment as well. In column 1, based on all programs, thecausal effect of inflation change is quantitativelysmall and not statistically significant. In countrieswith market access (column 2), the point estimate islarger, 0.048 (with a p-value of 0.26), but only one-third of the OLS estimate. Although reverse causal-ity explains a large portion of the association be-tween inflation changes and shortfalls, the causaleffect of policy adjustment is not quantitatively ir-relevant: reducing inflation by 10 percentage pointslowers the shortfall by about 0.5 percent of GDP.

Turning to fiscal adjustment, the first-stage regres-sions show that the association between actual fiscaladjustment and its projection is not as strong as formonetary adjustment. In three out of four casesshown in Table 6, the Hausman test rejects the nullhypothesis that the OLS estimates are not statisti-cally different from the 2SLS estimates, indicatingthat the endogeneity bias in the OLS estimates ispotentially important. This result is not surprisinggiven that all the 2SLS point estimates of the coefficient on fiscal adjustment in the top panel ofTable 6 have a different sign relative to the OLS es-timates in Table 5. However, these estimates are notsignificant at the usual confidence levels—but incolumn 2 the estimate is close to being significant atthe 10 percent level, and the p-value is 0.11. Onlyfor primary adjustment in countries with market ac-cess (column 4) is the coefficient larger than 1,

49

Roberto Benelli

Table 6

Causal Effect of Policy Adjustment for Nonprecautionary Programs

(1) (2) (3) (4) (5) (6)

Adjustment in Inflation Adjustment in Fiscal Balance

Market Market MarketAll access All All access access

Two-stage least squaresDependent variable: Shortfall in NKF in year T+1

Inflation change1 0.030 0.048(0.041) (0.041)

Overall adjustment 1.002 0.619(0.792) (1.706)

Primary balance 0.686 1.778(0.431) (1.714)

Constant 0.846 2.357 –0.120 0.119 1.918 3.309(1.097) (1.563) (0.970) (0.951) (1.500) (2.970)

Hausman test statistic (H)2 1.09 5.00** 4.00** 11.65*** 1.03 3.69*Probability > H 0.300 0.025 0.045 0.001 0.311 0.055Number of observations 66 27 68 59 28 22

First-stage regressionDependent variable: Actual policy adjustment

Projected inflation change 0.507*** 0.915***(0.089) (0.142)

Projected overall adjustment 0.684*** 0.676(0.189) (0.532)

Projected primary adjustment 0.867*** 0.767*(0.182) (0.411)

Constant –3.109 3.406 –0.125 –0.156 –0.667 –1.098*(2.284) (3.393) (0.365) (0.377) (0.722) (0.549)

R-squared 0.334 0.624 0.166 0.285 0.058 0.149Number of observations 66 27 68 59 28 22

Notes: Robust standard errors in parentheses. An asterisk (*) denotes significance at 10 percent, (**) at 5 percent, and (***) at 1 percent.1 Excluding large inflation changes (larger than 70 percentage points in absolute value).2 Hausman test for adjustment in inflation is not based on robust covariance matrix, because when the robust covariance matrix is used, the

finite-sample covariance matrix of the estimators’ difference is not positive definite.

which, according to the previous interpretation, in-dicates “perverse” effects of fiscal adjustment.

To summarize, on the one hand, the associationbetween shortfalls in net private capital flows andinflation change is not merely the result of reversecausation from capital flows to policy adjustment—it supports the hypothesis that, in countries withmarket access, tight monetary policy helps stimulateprivate capital inflows (as claimed by Rogoff, 2003).On the other hand, there is no evidence that fiscaladjustment has either systematic catalytic or per-verse effects. The negative association between fis-cal adjustment and shortfalls in net private capitalflows found in Table 5 thus appears to be due to thereverse causation of capital account shocks on pol-icy adjustment. Given the poor statistical precisionof the estimates, though, caution should be used ininterpreting these findings.32

Concluding Remarks

In this paper, I have defined a program to be success-ful if the program projections for net private capitalflows are met or exceeded. Starting from the premisethat a program consists of financial assistance andpolicy adjustment, I have focused on the size of fi-nancial assistance (the “program size”) and policyadjustment as determinants of program success. Thisapproach has several advantages: it avoids thethorny issue of defining counterfactual scenarios; itallows me to control for the fact that programs differin the mix of policy adjustment and financing theyprescribe; and it allows me to estimate the causal ef-fect of policy adjustment. Although data qualityhampers the statistical precision of some of the find-ings, the empirical evidence seems to depict a con-sistent picture of the effects of IMF programs on pri-vate capital flows.

The first main finding is that program success isnot purely a random event. In particular, success isnegatively associated with the size of financial assis-tance, especially in those countries with access toprivate capital markets. This empirical association isdifficult to interpret unambiguously: it could meanthat larger programs facilitate capital outflows, but itcould also mean that the IMF often hits a bindinglending constraint on the size of its lending. In fact,I have found some evidence that the negative asso-ciation between success and program size could arisefrom projection biases. These biases, in turn, seemto arise from the fact that the pressure on the IMF’sstaff to generate optimistic projections in response

to binding lending constraints is likely to bestronger in larger programs.

The second main finding is that policy adjust-ment—more in monetary policy than in fiscal pol-icy—seems to contribute to program success, espe-cially in countries with market access. Although the causal effect of policy adjustment is not esti-mated precisely, the estimates are smaller when con-trolling for policy endogeneity. This suggests thatexogenous capital account shocks are important de-terminants of what policy adjustment the domesticauthorities undertake, especially in countries withmarket access.

ReferencesBarro, Robert J., and Jong-Wha Lee, 2002, “IMF Pro-

grams: Who Is Chosen and What Are the Effects?”NBER Working Paper No. 8951 (Cambridge, Massa-chusetts: National Bureau of Economic Research).

Basurto, Gabriela, and Atish R. Ghosh, 2000, “The Inter-est Rate-Exchange Rate Nexus in the Asian CrisisCountries,” IMF Working Paper 00/19 (Washington:International Monetary Fund).

Benelli, Roberto, 2003, “Do IMF-Supported ProgramsBoost Private Capital Inflows? The Role of ProgramSize and Policy Adjustment,” IMF Working Paper03/231 (Washington: International Monetary Fund).

Berg, Andrew, Eduardo Borensztein, Ratna Sahay, andJeromin Zettelmeyer, 1999, “The Evolution of Out-put in Transition Economies: Explaining the Differ-ences,” IMF Working Paper 99/73 (Washington: In-ternational Monetary Fund).

Bird, Graham, and Dane Rowlands, 1997, “The CatalyticEffects of Lending by the International Financial Insti-tutions,” World Economy, Vol. 20, No. 2, pp. 955–75.

———, 2002a, “IMF Lending: How Is It Affected by Eco-nomic, Political and Institutional Factors?” Policy Re-form, Vol. 4, pp. 243–70.

———, 2002b, “Do IMF Programmes Have a CatalyticEffect on Other International Capital Flows?” OxfordDevelopment Studies, Vol. 30, No. 3, pp. 229–49.

Bulír, Ales, and Soojin Moon, 2003, “Do IMF-SupportedPrograms Help Make Fiscal Adjustment MoreDurable?” IMF Working Paper 03/38 (Washington:International Monetary Fund). This paper also ap-pears, in somewhat different form, as Chapter 14 inthis volume.

Chang, Roberto, and Andrés Velasco, 2000, “LiquidityCrises in Emerging Markets: Theory and Policy,”NBER Macroeconomics Annual 1999, ed. by Ben S.Bernanke and Julio Rotemberg (Cambridge, Massa-chusetts: MIT Press).

Corsetti, Giancarlo, Bernardo Guimarães, and NourielRoubini, 2003, “The Trade-Off Between an Interna-tional Lender of Last Resort to Deal with LiquidityCrises and Moral Hazard Distortions: A Model of theIMF’s Catalytic Finance Effect,” NBER WorkingPaper No. 10125 (Cambridge, Massachusetts: Na-tional Bureau of Economic Research).

Cottarelli, Carlo, and Curzio Giannini, 2002, “Bedfel-lows, Hostages, or Perfect Strangers? Global CapitalMarkets and the Catalytic Effect of IMF Crisis Lend-ing,” IMF Working Paper 02/193 (Washington:

50


32 Estimating the causal effects of policy adjustment relies onthe assumption that the projected policy adjustment is a valid in-strument for the actual adjustment. In an appendix of the IMFWorking Paper version of this chapter (Benelli, 2003), I showthat this problem is not likely to be quantitatively important.

International Monetary Fund). This paper also ap-pears, in somewhat different form, as Chapter 12 inthis volume.

Dell’Ariccia, Giovanni, Isabel Schnabel, and JerominZettelmeyer, 2002, “Moral Hazard and InternationalCrisis Lending: A Test,” IMF Working Paper 02/181(Washington: International Monetary Fund).


Edison, Hali, Michael Klein, Luca A. Ricci, and TorstenM. Sløk, 2002, “Capital Account Liberalization andEconomic Performance: Survey and Synthesis,” IMFWorking Paper 02/120 (Washington: InternationalMonetary Fund).

Edwards, Martin S., 2000, “Reevaluating the Catalytic Ef-fects of IMF Programs,” paper presented at the An-nual Meeting of the American Political Science As-sociation, Washington, September.

Eichengreen, Barry, and Ashoka Mody, 2001, “Bail-Ins,Bailouts, and Borrowing Costs,” Staff Papers, Interna-tional Monetary Fund, Vol. 47, Special Issue,pp. 155–87.

Fischer, Stanley, 1993, “The Role of Macroeconomic Factors in Growth,” Journal of Monetary Economics,Vol. 32, pp. 485–512.

Furman, Jason, and Joseph E. Stiglitz, 1998, “EconomicCrises: Evidence and Insight from East Asia,” Brook-ing Papers on Economic Activity: 2, Brookings Institu-tion, pp. 1–135.

Ghosh, Atish, Timothy Lane, Marianne Schulze-Ghattas,Ales Bulír, Javier Hamann, and Alex Mourmouras,2002, IMF-Supported Programs in Capital AccountCrises, IMF Occasional Paper No. 210 (Washington:International Monetary Fund).

Haque, Nadeem Ul, and Mohsin S. Khan, 1998, “DoIMF-Supported Programs Work? A Survey of theCross-Country Empirical Evidence,” IMF WorkingPaper 98/169 (Washington: International MonetaryFund).

International Monetary Fund (IMF), 1993, Balance ofPayments Manual, Fifth Edition (Washington: Inter-national Monetary Fund).

———, 2003a, “Review of Access Policy Under theCredit Tranches and the Extended Fund Facility,”available on the Web at http://www.imf.org/external/np/tre/access/2003/011403a.htm.

———, 2003b, “Access Policy in Capital AccountCrises—Modifications to the Supplemental ReserveFacility (SRF) and Follow-Up Issues Related to Ex-ceptional Access Policy,” available on the Web athttp://www.imf.org/external/np/tre/access/2003/pdf/011403.pdf.

Knight, Malcolm, and Julio A. Santaella, 1997, “Eco-nomic Determinants of IMF Financial Arrange-

ments,” Journal of Development Economics, Vol. 54,pp. 405–36.

Kraay, Aart, 2003, “Do High Interest Rates Defend Cur-rencies During Speculative Attacks?” Journal of Inter-national Economics, Vol. 59, pp. 297–321.

Krueger, Anne O., 1998, “Whither the World Bank andthe IMF?” Journal of Economic Literature, Vol. 36 (De-cember), pp. 1983–2020.

Loungani, Prakash, 2000, “How Accurate Are PrivateSector Forecasts—Cross-Country Evidence fromConsensus Forecasts of Output Growth,” IMF Work-ing Paper 00/77 (Washington: International Mone-tary Fund).

Marchesi, Silvia, 2003, “Adoption of an IMF Programmeand Debt Rescheduling: An Empirical Analysis,”Journal of Development Economics, Vol. 70, pp. 403–23.

Mody, Ashoka, and Diego Saravia, 2003, “CatalyzingCapital Flows: Do IMF Programs Work as Commit-ment Devices?” IMF Working Paper 03/100 (Wash-ington: International Monetary Fund).

Morris, Stephen, and Hyun Song Shin, 1998, “UniqueEquilibrium in a Model of Self-Fulfilling CurrencyAttacks,” American Economic Review, Vol. 88, No. 3,pp. 587–97.

———, 2003, “Catalytic Finance: When Does It Work?”(unpublished; New Haven, Connecticut: Yale Uni-versity).


Musso, Alberto, and Steven Phillips, 2001, “ComparingProjections and Outcomes of IMF-Supported Pro-grams,” IMF Working Paper 01/45 (Washington: In-ternational Monetary Fund).

Przeworski, Adam, and James Raymond Vreeland, 2002,“The Effect of IMF Programs on Economic Growth,”Journal of Development Economics, Vol. 62, No. 2,pp. 385–421.

Ramcharan, Rodney, 2003, “Assessing IMF Program Ef-fectiveness” IMF Research Bulletin, Vol. 4 (June).

Rodrik, Dani, 1996, “Why Is There Multilateral Lend-ing?” in Annual World Bank Conference on Develop-ment Economics 1995, ed. by M. Bruno and B.Pleskovic (Washington: World Bank).

Rogoff, Kenneth, 2003, “The IMF Strikes Back,” ForeignPolicy, Vol. 134 (January/February).

Zettelmeyer, Jeromin, 2000, “Can Official Crisis LendingBe Counterproductive in the Short Run?” EconomicNotes, Vol. 29, No. 1, pp. 13–29.

Zitzewitz, Eric, 2001, “Measuring Herding and Exaggera-tion by Equity Analysts and Other Opinion Sellers”(unpublished; Palo Alto, California: Stanford Uni-versity).

51

Roberto Benelli

52

4

This paper examines the accuracy of IMF projections in175 programs approved in the period 1993–2001, fo-cusing specifically on ratios of the fiscal surplus to GDPand external current account surplus to GDP. Four po-tential reasons for the divergence of projections from ac-tual values are identified: (a) mismeasured data on ini-tial conditions; (b) differences between the “model”underlying the IMF projections and the “model” sug-gested by the data on outturns; (c) differences betweenreforms and measures underlying the projections andthose actually undertaken; and (d) random errors in theactual data. Our analysis suggests that while all are im-portant, incomplete information on initial conditions isthe largest contributor to projection inaccuracy. We alsoinvestigate the role of revisions over time in projectionerror, and find that they improve projections for fiscalaccount data, while the current account continues to in-dicate a great deal of variability in the revision process.

Introduction

In this paper, we examine the accuracy of IMF pro-jections associated with 175 IMF-supported pro-grams approved in the period 1993–2001. For eachprogram, the IMF staff prepares a projection of thecountry’s future performance. This projection isbased on the country’s initial situation and upon thepredicted impact of reforms agreed on in the con-text of the IMF program.2 We focus on the projec-

tions of macroeconomic aggregates—specifically, onthe ratios of fiscal surplus to GDP and of current ac-count surplus to GDP—during the years immedi-ately following the approval of the IMF program.We will compare these projections to the actualdata for the same years.

Our comparison is statistical. We begin with de-scriptive statistics for the two macroeconomic aggregates, and demonstrate that the projection de-viates substantially from the observed value. Wethen use a simple vector autoregressive model of thedetermination of these two aggregates to decom-pose the deviation into components. We find thatthe “model” revealed by IMF staff ’s projections dif-fers significantly from the model evident in histori-cal data. We also find, however, that a substantialamount of the deviation in projections fromhistorical values is due to the incomplete informa-tion on which the IMF staff bases projections. We provide a complete decomposition of these ef-fects. We also investigate the degree to which revi-sions to the projection eliminate these deviationsowing to incomplete information. We find that re-visions tend to approximate more closely the historical data, but that substantial differences re-main between the revised projections and the his-torical data.

The data we analyze come from two distinctsources. The projections (also called “envisaged”outcomes) are drawn from the Monitoring of IMFArrangements (MONA) database.3 The data onhistorical outcomes are drawn from the World Economic Outlook (WEO) database of the IMF

Macroeconomic Adjustment in IMF-Supported Programs: Projections and Reality1

Ruben Atoyan, Patrick Conway, Marcelo Selowsky, and Tsidi Tsikata

1 A version of this paper was presented at the Conference onthe Role of the World Bank and the IMF in the Global Economy,held at Yale University during April 25–27, 2003. An abbrevi-ated version is to be published by Routledge in a volume of con-ference proceedings.

2 We will hold to a specific definition of “projections” in thispaper. We do not consider projections to be identical to “fore-casts.” We define a forecast to be the best prediction possible ofwhat is to occur at a given time in the future. A projection in thiscontext is a prediction based on the participating country under-taking and completing all structural and policy reforms agreed toin the letter of intent approved between the participating gov-ernment and the IMF. The two could diverge if the best predic-tion includes only partial implementation of policy and struc-tural reform.

3 When an IMF program is approved, the IMF staff uses thebest statistics available at that time for current and past macro-economic data to create projections for the evolution of thosevariables over the following years. These projections representthe “original program” projections for that IMF program. Pro-gram performance is reviewed periodically over time, and at eachreview the IMF staff creates a new set of projections for themacroeconomic data reflecting the best available information ofthat time. We will use the “first review” projections for each pro-gram in a later section.

CHAPTER

53

Atoyan, Conway, Selowsky, and Tsikata

as reported in June 2002. Given the difference insources, some data manipulation is necessary to ensure comparability.4 The data are redefined ineach case to be relative to the initial program year:it is denoted the “year T” of the program.5 We willexamine four projection “horizons” in this study. Foreach projection horizon, we will compare the IMFstaff projection with the historical outcome. Theyear prior to “year T” is denoted T–1. The horizon-T data will be projections of macroeconomic out-comes in period T based on information available inT–1: in other words, a one-year-ahead projection.The horizon-T+1 data are projections of macroeco-nomic outcomes in period T+1 based on informa-tion available in T–1, and are as such two-year-ahead projections. The horizon-T+2 and hori-zon-T+3 projections are defined analogously. Thenumber of observations available differs for each pro-jection horizon owing to (a) missing projectiondata, or (b) projection horizons that extend beyondthe end of the available historical data. The num-bers of observations available for comparisons are asfollows for horizons T through T+3, respectively:175, 147, 115, and 79.

We will focus on two macroeconomic aggregates.The historical fiscal surplus as a share of GDP forcountry j in year t will be denoted yjt. The historicalcurrent account surplus as a share of GDP will bedenoted cjt. The projections of these two variableswill be denoted yjt and cjt, respectively. Other variables will be introduced as necessary and definedat that time. It will be useful for exposition to de-scribe projections of these ratios as the change observed in the ratio between period T–1 (just be-fore the program began) and the end of the timehorizon. We use the notation ∆yjk and ∆cjk to repre-sent the change in the projection ratio between period T–1 and the end of horizon k: for example,∆ cjT= cjT– cjT–1. Historical data from the WEO aredifferenced analogously.

Each program is treated as an independent obser-vation in what follows. However, it is important tonote that the database includes multiple programsfor many participating countries. These programsmay overlap for a given country, in the sense thatthe initial year (year T) for one program may coin-cide with a projection year (e.g., year T+2) for aprevious program in that country.

What Does the Record Show?

For an initial pass, we compare the historical outcomes for the countries participating in IMF-supported programs with the outcomes projected byIMF staff when the programs were originally ap-proved.6 When we compare the mean of ∆yjk and∆ cjk for various projection horizons k with the meanof the actual ∆yjk and ∆cjk, we find that projectionsdiffer substantially from those actually observed.Figure 1a illustrates the pattern of mean changes inprojected and historical fiscal ratios.7 The two meanchanges are nearly coincident for horizon T, whilefor longer horizons the historical and envisagedchanges diverge sharply. The mean projectedchange in the fiscal ratio rises with the length of thehorizon; at horizon T+3, the projected change inthe fiscal ratio is 3.5 percentage points. The changeactually observed over those time horizons was quitedifferent: 0.68 percentage point for horizon T+1 andup to 1.12 percentage points for horizon T+3.

Figure 1b illustrates the pattern for changes inprojected and actual current account ratios. Themean projected change in the current account ratiois negative for horizon T and horizon T+1. Thechange becomes positive and grows for longer projection horizons. The historical change in thecurrent account ratio for participating countries fol-lowed a different dynamic: improvement for hori-zon T, followed by deterioration in longer horizons.Negative changes in mean current account ratiocontinued three and four years after adoption of theIMF program.

0

0.5

1.0

1.5

2.0

2.5

3.0

3.5

4.0

T T+1 T+2 T+3

Cumulative change as percentage of GDP

Projected

Historical

Figure 1a

Mean Historical and Projected Changes inFiscal Ratios

4 For example, the projections are reported on an annual basisbut the year is not invariably a calendar year. For some programs,the fiscal year was used as the basis of data collection and for pro-jections. In those instances, the historical data are convertedinto fiscal-year equivalents through weighted-average conversionof the calendar-year data.

5 The “year T” of each program is defined by IMF staff to bethat fiscal year (as defined by the country) in which the programis approved. Programs are typically not approved at the begin-ning of year T, but rather at some point within the year.

6 In this section, we use the projections from the “original pro-gram.”

7 The data on which Figures 1 and 2 are based are reported inTable 1.

While these mean differences are suggestive, theycover up the great variability in projections and realizations for both ratios. Figures 2a through 2h il-lustrate the historical and envisaged changes ineach variable for each projection horizon. The diag-onal line indicates values at which projected changeis just equal to historical change. The dispersion invalues around the diagonal lines is quite striking.8For both fiscal and current account ratios, there isevidence of projected changes exceeding historicalchanges. This is especially striking for the fiscal ra-tios over time, as the proportion of observations tothe right of the diagonal line rises with the projec-tion horizon. There is also evidence of historicalvalues more extreme than projected, especially forthe changes in the current account ratio.

The correlations between projected and historicalchanges of the two ratios over the various time hori-zons are as follows: There is a good, though not perfect, correlation between projected and historical changes for hori-zon T. For longer projection horizons the correla-tion is lower. The horizon T+2 correlations exhibitthe lowest values, with horizon T+3 correlations ris-ing again to equal those of horizon T+1.

It is not surprising that the projections are inexactat any projection horizon. Nor is it surprising thatthe shortest horizon exhibits the closest fit to theactual, since longer-horizon projections requiredpredictions on intermediate-year outcomes that al-most surely will be inexact. It will be useful, how-ever, to decompose the projection error into parts—can we learn from the record to identify the sourceof the projected imprecision?

Decomposing Projection Error

Begin with gT, a macroeconomic variable observedat time T. Define sT as the vector of policy forcingvariables observed at time T. Denote the projectionof ∆gT to be

∆ gT = f(XT–1, ∆ sT), (1)

with XT–1 a matrix representing that informationavailable to the forecaster at time T–1 and sT thematrix of projected policy outcomes consistent withthe government’s letter of intent.9 The actual evolu-tion of the variable g T can be represented by the expression

∆gT = φ(ζT–1, ∆sT), (2)

with ζT–1 the matrix of forcing variables at timeT–1 (including a random error in time T), sT thematrix of observed policy outcomes, and φ the truereduced-form model. Projection error can then berepresented by the difference (∆ gT – ∆gT).10

(∆ gT – ∆gT)= φ(ζT–1, ∆sT) – f(XT–1, ∆sT). (3)

There are four potential sources for this projectionerror. First, the projection model f(.) may not beidentical with the true model φ(.). Second, the his-torical policy adjustment (∆sT) may differ from theprojected policy adjustment (∆ sT). Third, the infor-mation set XT–1 available for the projections maynot include the same information as the forcing vec-tor ζT–1 for the true process. Finally, there is randomerror in realizations of the macroeconomic variable.

Consider a simple example. There is a single pro-jection of change in a variable gT. The forcing ma-trix is simply the lagged variables gT–1 and gT–2.11

The policy matrix is represented by the single in-strument sT.

54

4 � MACROECONOMIC ADJUSTMENT IN IMF-SUPPORTED PROGRAMS

Horizons T T+1 T+2 T+3

Fiscal ratio 0.61 0.56 0.31 0.56(∆yjk, ∆yjk)

Current account 0.54 0.38 0.32 0.38ratio (∆cjk, ∆cjk)

8 Table 1 reports the standard deviations of the projected andactual changes in the database; these are in all cases and for allprojection horizons at least as large as the mean values.

–1.0

–0.5

0

0.5

1.0

1.5

T T+1 T+2 T+3

Cumulative change as percentage of GDP

Projected

Historical

Figure 1b

Mean Historical and Projected Changes inCurrent Account Ratio

9 By contrast, we consider the forecast of ∆gT to be defined as∆g eT = f(XT–1;∆seT), with ∆seT representing the forecaster’s bestprediction as of period T–1 of the policy vector to be observed inperiod T.

10 Hendry (1997) provides an excellent summary of the possiblesources of projection (in his case forecasting) error when the pro-jection model is potentially different from the actual model. Thisexample can be thought of as a special case of his formulation.

11 gT–2 enters the expression through the term ∆gT–1.

55


Figure 2a. Fiscal Ratios in Horizon T

–15

–10

–5

0

5

10

15

–10 –5 0 5 10 15

Projection

His

toric

al

–20

–15

–10

–5

0

5

10

15

–20 –15 –10 –5 0 5 10 15

–25

–20

–15

–10

–5

0

5

10

15

20

–10 –5 0 5 10 15

Figure 2b. Current Account Ratios in Horizon T

Projection

His

toric

al

Projection

His

toric

al

Figure 2c. Fiscal Ratios in Horizon T+1 Figure 2d. Current Account Ratios in Horizon T+1

–40

–30

–20

–10

0

10

20

30

–25 –20 –15 –10 –5 0 5 10 15

Projection

His

toric

al

–16

–12

–8

–40

4

8

12

16

–4 4 8 12 16

Projection

His

toric

al

Figure 2e. Fiscal Ratios in Horizon T+2

–20

–15

–10

–5

0

5

10

15

20

–5 5 10 15 20

Projection

His

toric

al

Figure 2g. Fiscal Ratios in Horizon T+3

–25

–20

–15

–1

–5

5

10

15

20

–25 –20 –15 –10 –5 0 5 10 15

Figure 2h. Current Account Ratios in horizon T+3

Projection

His

toric

al

Figure 2f. Current Account Ratios in Horizon T+2

–50

–40

–30

–20

–10

0

10

20

30

–25 –20 –15 –10 –5 0 5 10 15

Projection

His

toric

al

Figures 2a–2h

Fiscal Ratios and Current Account Ratios

56


Table 1

Projecting the Change in Macroeconomic Aggregates

Horizon T

Simple Statistics

Variable N Mean Std dev Sum Minimum Maximum

∆ yjT 175 1.08651 2.88011 190.14000 –7.60000 12.50000

∆yjT 175 0.87778 3.25935 153.61161 –11.33896 12.72751

∆cjT 175 –0.22187 3.47920 –38.82699 –13.89236 11.66200

∆cjT 175 0.72340 4.77454 126.59449 –17.68986 14.49604

Correlations: ∆ yjT ∆yjT ∆cjT ∆cjT

∆ yjT 1.00000

∆yjT 0.60489 1.00000

∆cjT 0.24256 0.12334 1.00000

∆cjT 0.19968 0.30303 0.53486 1.00000

Horizon T+1

Simple Statistics


∆ yjT+1 147 1.62408 3.22486 238.74000 –5.60000 12.90000

∆yjT+1 147 0.67722 3.93298 99.55073 –19.42935 13.69233

∆cjT+1 147 –0.37867 4.98040 –55.66390 –22.23187 12.01531

∆cjT+1 147 –0.03233 7.07135 –4.75294 –35.61176 25.90529

Correlations: ∆ yjT+1 ∆yjT+1 ∆cjT+1 ∆cjT+1

∆ yjT+1 1.00000

∆yjT+1 0.56182 1.00000

∆cjT+1 0.13572 –0.02165 1.00000

∆cjT+1 0.12453 0.04358 0.38254 1.00000

(Continued on next page)

57


Table 1 (concluded)

Horizon T+2

Simple Statistics


∆ yjT+2 115 2.59478 3.67922 298.40000 –3.30000 15.60000

∆yjT+2 115 0.81807 4.85553 94.07814 –16.72117 11.88877

∆cjT+2 115 0.64742 4.64897 74.45280 –22.04280 11.68855

∆cjT+2 115 –0.49056 7.32857 –56.41476 –38.14743 21.78397


∆ yjT+2 1.00000

∆yjT+2 0.31046 1.00000

∆cjT+2 0.11603 –0.21840 1.00000

∆cjT+2 –0.03683 –0.11606 0.32365 1.00000

Horizon T+3

Simple Statistics


∆ yjT+3 79 3.51000 4.27596 277.29000 –2.70000 19.50000

∆yjT+3 79 1.11918 4.85320 88.41557 –17.48994 13.35470

∆cjT+3 79 1.28198 4.91608 101.27681 –19.89594 14.73079

∆cjT+3 79 –1.37587 12.09842 –108.69398 –81.569321 21.75981


∆ yjT+3 1.00000

∆yjT+3 0.55890 1.00000

∆cjT+3 0.14194 –0.12499 1.00000

∆cjT+3 –0.02829 0.01113 0.38530 1.00000

Notes: Std dev denotes standard deviation. N denotes the number of observations.

58


Equations (1) and (2) can then be rewritten in thefollowing form:

∆ gT = a1∆ gT–1+ a2(gT–1+ ηT–1)+b1∆ (1e)

∆gT = α1∆gT–1+ α2gT–1+ β1∆sT+ εT. (2e)

The coefficients (α1, α2, β1) represent the truemodel while (a1, a2, b1) are coefficients from themodel used for projections. In the projection rule,the forecaster perceives gT–1 = (g T–1 + ηT–1) withηT–1 a random error. This imprecision may occur be-cause the information set available to the forecasteris less precise than the set available after later revi-sions. The variable εT represents the stochastic na-ture of realizations of the actual variable.

(∆ gT – ∆gT)=[(a1– α1)∆gT–1+(a2– α2)gT –1+(b1– β1)∆sT]+b1(∆sT – ∆sT) +[a2ηT–1+ a1(∆ gT–1– ∆gT–1)] + eT (3e)

The projection error thus illustrates the four com-ponents mentioned previously. First, there is thepossibility that the forecaster’s model differs fromthat evident in the historical data; this will lead tothe deviations summarized within the first set ofsquare brackets. Second, there could be a diver-gence between the projected policy adjustment andthe actual policy adjustment. Third, there is the po-tential that projection error is due to mismeasure-ment of initial conditions or to past errors in fore-casts of variable growth. Fourth, the error maysimply be due to the stochastic nature of the vari-able being projected.

In the sections that follow, we decompose theprojection error into these four parts for the fiscal balance/GDP ratio and the current account balance/GDP ratio in countries with IMF-supportedprograms. First, we create a reduced-form modelthat represents well the evolution of the actual data. We estimate the model implicit in the pro-jected data, and compare the coefficients from thisprojection model with those from the actual data. Second, we examine the envisaged and historicaldata for evidence that revisions in the data led to the discrepancies. Third, we perform a decompo-sition exercise to determine the percentages of deviations of projection from historical values thatcan be attributed to differences in models, differ-ences in initial conditions, differences in policy re-sponse, or simply random variation in the historicaldata.

Fiscal and Current Accounts

We begin with the macro identity

yjt ≡ cjt– pjt (4)

holding for all countries j and time periods t. yjt isthe fiscal surplus as a share of GDP, cjt is the currentaccount surplus as a share of GDP, and pjt is privatesavings as a share of GDP.

We posit as well that there is a “normal” level ofprivate saving specific to each country and to eachtime period. This normal level pn

jt can be representedby a country-specific component, a component thatis common to all countries for a given time period,and a positive relationship between foreign savingopportunities and private saving.

pnjt = α j+ βt+ δ cjt . (5)

Combining (4) and (5), and defining ejt= (pjt – pnjt) as

the excess private saving in any period yields

yjt= –α j – βt+(1– δ) cjt – ejt. (6)

The variables yjt and cjt can be represented by a vec-tor autoregression. With appropriate substitution,this vector autoregression can be rewritten in error-correction form as12

∆yjt= ao – a12∆yjt–1– b12∆cjt–1+(a11+ a12–1)yjt–1+(b11+ b12)cjt–1+ εyjt (7a)

∆cjt= bo– a22∆yjt–1– b22∆cjt–1+(a21+ a22)yjt–1+(b12+ b22–1)cjt–1+ εc. (7b)

There is, in general, no way to assign contempo-raneous causality in (7a) and (7b). If it were possibleto assert that the current account ratio is exogenously

12 We will refer to the “error-correction form” as one that in-cludes both lagged differences and lagged levels of the two vari-ables as explanatory variables for the current differenced vari-ables. This can be derived from a general AR specification of thetwo variables; the AR(2) specification is used here for ease of il-lustration. The form presented in the text can be derived fromthe following AR(2) set of equations:

yjt= ao+a11yjt–1+ a12yjt–2+b11cjt–1+b12cjt–2+εyjt

cjt=bo+ a21yjt–1+ a22yjt–2+ b21cjt–1+b22cjt–2+εc.

Specification tests are used to choose the lag length appropri-ate to the empirical work. In a world in which yjt and cjt are non-stationary but are cointegrated on a country-by-country basis,further simplification is possible. If yjt and cjt are nonstationary inthe current dataset, equation (6) represents a cointegrating rela-tionship. The “error-correction” variable ejt can then be insertedin the equations (7) in place of the terms in yjt–1 and cjt–1 and willhave the coefficient associated with yjt–1 in (7). It is impossible toverify a nonstationary relationship in this dataset, given that wehave only scattered observations from each country’s time series.We do investigate that possibility in the second and fourthcolumns of Tables 2 and 3, with support for that interpretation ofthe error-correction term in the ∆yjt equation. Hamilton (1994,Ch. 19) provides a clear derivation of this error-correction formfrom the underlying autoregression.

determined, for example, then the contemporaneouschange ∆cjt could be a separate regressor in the ∆yjtequation to account for that contemporaneous cor-relation.

The econometric effects modeled here can be di-vided into three groups. The first group, representedby the terms in ∆cjt–1 and ∆yjt–1, capture the autore-gressive structure of the system. The second group,represented by the terms in yjt–1 and cjt–1, capture theadjustment of these variables in response to devia-tions from the “normal” relationship described in(6). The third group represents random errors. Al-though the direction of contemporaneous causalitycannot be verified, there is a version of dynamiccausality that can be checked. The coefficients ofyjt–1 and cjt–1 represent the degree to which the cur-rent account and fiscal ratios respond to deviationsfrom the norm.

The system of equations in (7) will hold for all t,and thus should be in evidence at time T when theIMF-supported program is introduced. The systemhas excluded policy interventions from the deriva-tion for simplicity, but it is straightforward, thoughmessy, to introduce them. One way to do so will bethrough definition of a policy response function, bywhich ∆sjT is itself a function of cjT–1 and yjT–1. Thesecond will be to incorporate the policy variables asexogenous forcing variables. The approach we usewill incorporate parts of each.

Estimation Using Historical Data

The results of the coefficient estimates from equa-tions (7) for all programs in the sample at horizon Tusing historical data are summarized in Table 2a.Specification testing revealed that lagged first-difference terms with lag length greater than twodid not contribute significantly to the regression.13

The contemporaneous causality imposed on themodel is that changes in the fiscal account arecaused by changes in the current account, and notvice versa.14 The error-correction term (ejT–1) wasderived from the regression in levels (i.e., not firstdifferenced) reported in Annex I.

For the ratio of fiscal balance to GDP, the estima-tion results suggest the following insights (see thefirst two columns of Table 2a):

• There is significant positive contemporaneouscorrelation between the two variables, and thenormalization chosen here assigns causation to

∆cjT. For a 1 percent increase in the current ac-count ratio, there is a 0.28 percent increase inthe fiscal ratio.

• The current first-difference responds positivelyand significantly to shocks in the own ratio inprevious periods. For a unit shock to ∆yjT–1,there are other things equal a 0.25 increase in∆yjT. For a unit shock to ∆yjT–2, the transmittedshock is positive and significant at 0.16. Past pos-itive current account shocks have small negativeeffects on ∆yjT with the two-period lagged effectsignificant at the 90 percent level of confidence.

• The coefficient on yjT–1 is significantly differentfrom 0, but not from –1. It implies that for anaverage country, a deviation from its “normal”fiscal account ratio will lead to an adjustment inthe next period that erases 82 percent of thatdeviation.

For the ratio of current account to GDP, the esti-mation explains a lower percentage of the variation(as indicated by the R2 statistic of 0.56). The secondset of columns reports coefficients and standard errorsfor that specification, and indicates the following:

• The lagged first-difference terms have no signif-icant effect on the current first difference.

• The coefficient on cjT–1 of –0.40 is significantlydifferent from both 0 and –1. It indicates that40 percent of any deviations of the current ac-count ratio from its normal value is made up inthe following period.

The last four columns of Table 2a report the re-sults of error-correction regressions in which yjT–1and cjT–1 are replaced by ejT–1 from equation (6), asimplied by a cointegrating relationship between thetwo variables. As is evident in comparing the firstset and third set of results, the cointegrating rela-tionship captures nearly all of the explanatorypower in the ∆yjT regression. The cointegrating rela-tionship is less effective in the ∆cjT equation, how-ever, as indicated by the R2 statistic.15

These results are specific to the data for horizon T.Results for horizon T+1 are presented in Table 2b.The construction of these data differs somewhat, inthat the endogenous variable is a two-period fore-cast; we chose to use two-period lags on the right-hand side of the equation for comparability. Forhorizon T+1, the contemporaneous effect of thecurrent account ratio on the fiscal ratio is halved—

59


13 Statistical confidence in this paper will be measured at the90 percent, 95 percent, and 99 percent levels. In the text, statis-tical significance will indicate a degree of confidence greaterthan 95 percent unless otherwise indicated.

14 This assumption will be justified, for example, if the partici-pating country is constrained in its international borrowing, sothat the ratio of current account surplus to GDP is set by foreignlenders.

15 While imposition of the cointegration condition throughthe error-correction variable is effective for the fiscal ratio, ourcomparison of projections with historical data will be based onthe system without this condition imposed. As Clements andHendry (1995) demonstrate, the imposition of the cointegrationcondition in estimation when cointegration exists improves fore-cast accuracy, most notably for small (i.e., N= 50) samples. Forlarger samples, the improvements in forecast accuracy are small.

60


this is perhaps due to the doubling of the length of thetime horizon. The autoregressive structure of the fis-cal ratio, significant in horizon T, is no longer signifi-cant for horizon T+1. By contrast, the lagged “level”effects have larger coefficients. This effect in thecurrent account ratio equation is significantly larger,as well, with the coefficient (–0.833) more thandouble the comparable term for horizon T (–0.40).

Estimation Using Projected Data

If we interpret the estimated model of the precedingsection to be the “true” model (2), we posit that themodel used in forming projections for IMF programsshould have a similar form. We can use similareconometric techniques to those of the previoussection to derive the economic model implied by

Table 2a

Regression Results, Historical Current and Fiscal Account Ratios, Horizon T

∆yjT ∆cjT ∆yjT ∆cjT

Coefficient S.E. Coefficient S.E. Coefficient S.E. Coefficient S.E.

∆cjT 0.28** (0.06) 0.25** (0.05)∆yjT–1 0.25** (0.10) –0.08 (0.19) 0.23** (0.10) –0.04 (0.20)∆cjT–1 –0.05 (0.05) –0.04 (0.10) –0.02 (0.05) –0.23** (0.09)∆yjT–2 0.16** (0.08) –0.01 (0.16) 0.14* (0.08) 0.13 (0.16)∆cjT–2 –0.07* (0.04) –0.02 (0.08) –0.05 (0.04) –0.17** (0.07)

yjT–1 –0.82** (0.11) –0.09 (0.21)cjT–1 0.16** (0.07) –0.40** (0.12)

ejT–1 –0.81** (0.11) –0.17 (0.22)

N 176 176 176 176R2 0.78 0.56 0.78 0.50

Notes: Full sample, Horizon T. Standard errors (S.E.) appear in parentheses.* Indicates significance at the 90 percent level of confidence.** Indicates significance at the 95 percent confidence level.A complete set of time and country dummies was included in the regressions, but their coefficients have been suppressed for brevity.

Table 2b

Regression Results, Historical Current and Fiscal Account Ratios, Horizon T+1

∆yjT+1 ∆cjT+1 ∆yjT+1 ∆cjT+1


∆cjT+1 0.14** (0.06) 0.12** (0.05)∆yjT–1 0.09 (0.11) –0.15 (0.24) 0.08 (0.10) 0.01 (0.26)∆cjT–1 –0.04 (0.07) –0.12 (0.16) –0.01 (0.06) –0.55*** (0.12)

yjT–1 –1.13*** (0.17) 0.47 (0.37)cjT–1 0.23** (0.10) –0.83*** (0.21)

ejT–1 –1.10*** (0.16) 0.27 (0.41)

N 147 147 147 147R2 0.83 0.72 0.83 0.65

Notes: Variable definition (this table only for all variables gj):

∆gjT+1= gjT+1– gjT–1

∆gjT–1= gjT–1– gjT–3

Full sample, Horizon T+1. Standard errors (S.E.) appear in parentheses. ** Indicates significance at the 95 percent confidence level.*** Indicates significance at the 99 percent confidence level.A complete set of time and country dummies was included in the regressions, but their coefficients have been suppressed for brevity.

61


the projections. We report the results of this estima-tion exercise in Table 3a for projection horizon T.

The results from estimating the projection modelfor the fiscal ratio suggest the following (see the firstset of columns in Table 3a):

• There is significant contemporaneous correla-tion between the projected fiscal and currentaccount ratios. For a 1 percentage point in-crease in the current account ratio, there is evi-dence of a 0.15 percentage point increase inthe fiscal ratio. This is roughly half of the re-sponse found in the actual data. By implication,the IMF staff model will project a 0.85 percent-age point increase in the ratio of private netsavings to GDP in response to such a currentaccount shock, while the historical data indi-cate a 0.72 percentage point increase in the pri-vate saving ratio in response to such a shock.

• A 1 percentage point increase in last period’sfiscal ratio will trigger an 0.15 percentage pointdecrease in this period’s ratio. This suggests thatthe projection is relying on fiscal policy correc-tion to overcome any inertia in fiscal stanceover time and to offset past excesses with cur-rent austerity.16 This response also is less thanwas observed in the historical data.

• There is evidence of an error-correction effectin the data. The coefficients on the lagged ra-

tios have the correct signs, and that associatedwith yjT–1 is significantly different from zero.The coefficient –0.44 indicates that the projec-tion is designed to make up 44 percent of anydeviation of fiscal ratio from the country’s “nor-mal” ratio within a single year. This adjustmentis also roughly half of the adjustment observedin the historical data.

The results from estimating the projection modelfor the current account ratio are reported in the sec-ond set of columns in Table 3a:

• There is no significant evidence of an autore-gressive structure in ∆cjt, just as was true in thehistorical analysis.

• Past shocks to the fiscal ratio have a significantlagged effect on the current account ratio, a fea-ture unobserved in the actual data.

• There is a significant error-correction effect asevidenced by the coefficient on ∆cjt–1. The coef-ficient –0.33 indicates that the projection isconstructed to make up about 1⁄3 of any devia-tion of the current account ratio from its nor-mal value within a single year. The coefficienton yjT–1 is insignificantly different from zero.These features are quite similar to those ob-served in the historical data.

When the envisaged data are examined with thecointegrating relationship imposed, the evidence isonce again stronger for the fiscal ratio. In that re-gression (reported in the third set of columns), thecointegrating variable (emjT–1) has explanatory

Table 3a

Regression Results, Envisaged Current and Fiscal Account Ratios, Horizon T

∆yjT ∆ cjT ∆yjT ∆ cjT


∆ cjT 0.15* (0.09) 0.20** (0.08)∆ yjT–1 –0.15* (0.09) –0.15 (0.12) –0.14 (0.09) –0.15 (0.13)∆ cjT–1 –0.09 (0.06) –0.0004 (0.08) –0.09 (0.06) –0.06 (0.09)∆ yjT–2 –0.03 (0.08) 0.16* (0.10) –0.05 (0.08) 0.11 (0.11)∆ cjT–2 –0.06 (0.04) 0.04 (0.06) –0.08* (0.04) –0.04 (0.06)

yjT–1 –0.44** (0.09) –0.02 (0.11)cjT–1 0.08 (0.07) –0.33** (0.08)

emjT–1 –0.44** (0.09) –0.04 (0.12)

N 165 165 165 165R2 0.85 0.76 0.85 0.69

Notes: Full sample, Horizon T. Standard errors (S.E.) appear in parentheses.* Indicates significance at the 90 percent confidence level.** Indicates significance at the 95 percent confidence level.A complete set of time and country dummies was included in the regressions, but their coefficients have been suppressed for brevity.

16 We would observe this negative coefficient, for example, ifwe had a model that required the government to balance its bud-get over each two-year period. There could be excess spending inodd years, but it would be offset by spending cuts in even years.

62


power nearly equal to the lagged cjT–1 and yjT–1 re-ported in the first set of columns. In the equation forthe current account ratio, the results are muchweaker.

For time horizon T+1 (Table 3b), the projection“model” is quite similar to that of horizon T. Thecontemporaneous and lagged “level” effects are al-most identical for the fiscal ratio, as is the lagged“level” effect for the current account ratio. The au-toregressive terms differ somewhat, but the differ-ences are not statistically significant. The similarityof error-correction effects is quite striking, as it sug-gests that the projected adjustment from imbalanceoccurs totally in horizon T—there is no further ad-justment in horizon T+1. This is quite differentfrom the historical record, where adjustment con-tinues in fairly equal increments from horizon T tohorizon T+1.

Divergence Between Projected and Actual Policy

We note from the preceding discussion that there issubstantial evidence of difference between the coef-ficients in Tables 2a and 3a, and between Tables 2band 3b. We interpret these differences as evidencethat the “model” used in IMF projections and the“model” generating the historical data are signifi-cantly different. However, as the earlier discussiondemonstrated, model differences are only one sourceof projection errors. In this subsection, we use theframework of equation (3e) to decompose the ob-

served projection error for horizon T into compo-nents.

As the earlier discussion indicated, the projectionerror can conceptually be decomposed into fourparts: differences in models, differences in policy re-sponse, mismeasurement of initial conditions attime of projection, and random errors.17 Projectionerror is measured directly as the projection of thevariable for horizon T minus the realization of thevariable. Errors in initial conditions are measured asthe difference between projected and historical ob-servations of the level of the variable in period T–1.Two policy variables are considered indicators of theimportance of policy reform conditions in the error:the difference between projected and historical de-preciation of the real exchange rate (∆êjT – ∆ejT),and the difference between projected and historicalchange in government consumption expenditures asa share of GDP (∆wjT – ∆wjT).18 We hypothesize

Table 3b

Regression Results, Envisaged Current and Fiscal Account Ratios, Horizon T+1

∆yjT+1 ∆ cjT+1 ∆yjT+1 ∆ cjT+1


∆ cjT+1 0.158* (0.08) 0.179** (0.08)∆ yjT–1 –0.175* (0.10) 0.020 (0.16) –0.165* (0.09) 0.104 (0.16)∆ cjT–1 0.049 (0.08) –0.208* (0.12) 0.034 (0.07) –0.364*** (0.12)

yjT–1 –0.462*** (0.11) –0.029 (0.18)cjT–1 0.048 (0.09) –0.370*** (0.14)

emjT–1 –0.474*** (0.11) –0.128 (0.19)

N 129 129 129 129R2 0.86 0.72 0.86 0.68

Notes: Variable definitions (this table only for all variables gj):

∆gjT+1 = gjT+1 – gjT–1

∆gjT–1 = gjT–1 – gjT–3

Full sample, Horizon T+1. Standard errors (S.E.) appear in parentheses. * Indicates significance at the 90 percent confidence level.** Indicates significance at the 95 percent confidence level.*** Indicates significance at the 99 percent confidence level.A complete set of time and country dummies was included in the regressions, but their coefficients have been suppressed for brevity.

17 For now we treat each program as if it were approved at thebeginning of year T, so that the projected effects of the programon macroeconomic adjustment have a full year to take hold. Infact, programs are approved at different times within year T.Thus, the timing of approval within the year may explain part ofthe projection error. We explore this in Annex II.

18 The variable for government consumption expenditures isavailable in consistent format in both historical and envisageddata. The variable on real depreciation is constructed in bothcases as nominal depreciation minus CPI inflation for the horizonin question. These variables are explicit in the historical data. Inthe envisaged data, the nominal exchange rate is derived as theratio between GDP in home currency and GDP in U.S. dollars.

that the former should have a significant effect onthe current account, while the latter should be a sig-nificant component of the fiscal surplus.

Estimation of (3e) using the error-correctionframework presented in equations (8) is compli-cated by the simultaneity of the macroeconomicbalances and the policy variables over which condi-tions are defined. As (3e) indicates, (∆êjT – ∆ejT),(∆wjT – ∆wjT), ∆ejT and ∆wjT will all be included asregressors in the estimation framework, but all ofthese are potentially simultaneously determinedwith the macro balances. We address this by esti-mating the equations with both ordinary leastsquares (OLS) and two-stage least squares (2SLS),with the 2SLS results presumed to be free of simul-taneity bias.19 For each equation, as implied by (6),year-specific dummy variables are included to con-trol for year-to-year differences in capital availabil-ity on world markets; we also include significantcountry-specific dummy variables to control for ab-normally large cross-country differences in macrobalances. Those results are reported in Table 4. Thetop panel reports the results of regressions in the cur-rent account ratio and the fiscal ratio. There are twocolumns: the first with OLS estimates, on a slightlylarger sample, and the second with 2SLS estimateson a consistent-size sample of 162 observationsacross all variables. The bottom panel reports the re-gressions that served as the “first stage” of the 2SLS.The first column reports OLS results over the largestsample for which data were available for that regres-sion, while the second column reports OLS resultsover the consistent 2SLS sample of 162 observations.

We interpret the results as follows. Take as exam-ple the coefficient on cjT–1 in the two regressions.Given our derivation in equation (3e), this coeffi-cient should represent the difference between theprojection coefficient and the actual coefficient.When we compare the results of Tables 2 and 3, wefind this to be the case. Consider the 2SLS results.In the fiscal ratio regression of Table 4, the coeffi-cient of –0.11 is quite similar to the difference(0.08–0.16) of the coefficients reported in Tables 2and 3. For the current account ratio, the coefficientof 0.04 is also very similar to the difference(–0.33–(–0.40)) of the coefficients reported in Ta-bles 2a and 3a. A positive coefficient in this regres-sion indicates that the projection incorporated amore positive response to that variable than wasfound in the actual data.

We separate the discussion into the various typesof errors.

Differences in modeling. If the projections used adifferent model from that evident in the actual data,we expect to find significant coefficients on the

variables cjT–1, yjT–1, ∆cjT–1, ∆yjT–1, ∆ejT, and ∆wjT inthe top panel. Our discussion of Tables 2a and 3a in-dicated that we anticipated greater evidence of dif-fering models in the fiscal projections than in thecurrent account projections. This point is partiallysupported by results reported in Table 4. Considerthe OLS results. In the fiscal ratio estimation, thereare significant coefficients on cjT–1 (–0.11), ∆ejT(0.01), ∆yjT–1 (–0.08), and ∆wjT (0.10). If we con-sider the last case for illustration: a positive ∆wjTshould reduce the fiscal balance. The coefficient(0.10) indicates that the IMF projections incorpo-rated less pass-through of increased government ex-penditures into the reduced fiscal ratio than was ac-tually observed, leaving a positive projection error.However, the 2SLS results suggest that differencesin modeling are less apparent than is suggested bythe OLS estimates, since only the coefficient on cjT–1(–0.10) remains significantly different from zero.

For the current account ratio, there is no signifi-cant evidence of differences in modeling. All coeffi-cients on these variables are both small and insignif-icantly different from zero.

Mismeasurement of initial conditions. Anothersource of projection error will be the difference be-tween the initial conditions known to IMF staff andthe actual initial conditions available in historicaldata. For these differences to be a significant sourceof projection error, the coefficients on the variables(cjT–1– cjT–1) and (yjT–1– yjT–1) must be significantlydifferent from zero.

In the fiscal ratio regression, the difference in ini-tial fiscal ratios (yjT–1– yjT–1) is a significant contrib-utor to projection error. The 2SLS coefficient(–0.30) indicates that when IMF staff had access toartificially high estimates of the previous period’s fis-cal ratio, they adjusted downward the projected nec-essary policy adjustment necessary. This responsewas a rational one, given the error-correction natureof the fiscal ratio, but was based on an incorrectstarting point.

In the current account ratio regressions, the dif-ferences in initial conditions are the only significantdeterminants of projection error. With coefficients(–0.31) for (cjT–1– cjT–1) and (–0.25) of (yjT–1 – yjT–1)in the 2SLS version, the regressions suggest that theprojections were in error largely because of incom-plete information about the true value of the cur-rent account ratio in the preceding period.

Differences in policy response. If the projections in-cluded a policy response at variance with that actu-ally observed, the coefficients on (∆wjT – ∆wjT) and(∆êjT – ∆ejT) will be significant in the two regressions. In both the 2SLS and OLS results, thereis little evidence of this. In the fiscal 2SLS regression, there is a significant coefficient (–0.47)on (∆wjT – ∆wjT). This indicates that when the IMFprojected smaller expenditure increases than actu-ally occurred, the projection error on the fiscal ratiowas, on average, positive—as expected.

63


19 Both sets of results are reported because the systems approachto estimation reduces the number of observations usable in estima-tion. The OLS results thus provide a more comprehensive analy-sis, although one potentially tainted by simultaneity bias.

64


The regressions in the bottom panel hold someclues as to why the projections differed from histori-cal values. As is evident in the (∆wjT – ∆wjT) regres-sion, previous forecast errors were significant determinants of this policy projection error, as was abias toward more positive projections as the previ-ous period’s fiscal ratio rose. The policy projectionerrors in the real exchange rate depreciation(∆êjT – ∆ejT) had no significant contribution to ei-ther regression in either specification.

Random errors. As the R2 statistics indicate for thetwo regressions, the preceding three sources of pro-

jection error explain only 59 percent (for the cur-rent account ratio) and 71 percent (for the fiscalratio) of total projection error. The remaindershould be considered random shocks.

An Empirical Decomposition of Projection Error

In previous subsections, we identified several poten-tial sources of projection errors. The magnitude andsignificance of the regression coefficients reportedin Table 4 shed some light on the relative importance

Table 4

Estimation of Projection Error Equations

∆cjT – ∆cjT ∆yjT – ∆yT

OLS 2SLS OLS 2SLS

cjT–1 0.01 0.04 cjT–1 –0.11** –0.10**cjT–1– cjT–1 –0.41** –0.31** cjT–1– cjT–1 –0.06 –0.07yjT–1 –0.10 –0.02 yjT–1 0.01 0.04yjT–1– yjT–1 –0.19 –0.25* yjT–1–yjT–1 –0.34** –0.30**∆ejT –0.0003 –0.002 ∆ejT 0.01** 0.01∆êjT – ∆ejT –0.005 –0.001 ∆êjT –∆ejT 0.007 0.009

∆wjT 0.10** 0.02 ∆wjT –∆wjT –0.47** –0.47**∆cjT–1 0.03 0.03 ∆yjT–1 –0.08* –0.08

N 172 162 167 162R2 0.59 0.59 0.74 0.71

∆êjT – ∆ejT ∆wjT – ∆wjT

OLS OLS OLS OLS

∆êjT–1– ∆ejT–1 0.14** –0.03 wjT–1– wjT–1 –0.02 –0.03 ∆ejT–1 –0.03** –0.05** wjT–1 0.15** 0.14**

∆wjT–1 –0.06 –0.06wjT–1–∆wjT–1 –0.28** –0.23**

N 166 162 166 162R2 0.68 0.68 0.52 0.53

∆ejT ∆wjT

OLS OLS OLS OLS

∆ejT–1 0.01** 0.03** wjT–1 –0.39** –0.38**cjT–1 –0.10 0.04 ∆wjT–1 –0.13** –0.16**yjT–1 0.37 0.28 yjT–1 0.27** 0.29**N 174 162 173 162R2 0.65 0.74 0.61 0.60

Notes: The two-stage-least-squares (2SLS) procedure used the estimating equations in the two lower panels for ∆eT, ∆êT – ∆eT, ∆wT, and∆wT –∆wT, and estimated those equations simultaneously with the two reported in the upper panel. The equations in the two lower panels areall ordinary least squares (OLS), since they did not include endogenous regressors. The coefficients differ because of the number of observationsincluded: those with 165 were estimated in the simultaneous equation system, while those with other numbers of observations were estimatedas single equations.

* Indicates significance at the 90 percent confidence level. ** Indicates significance at the 95 percent confidence level.Standard errors and other regression statistics are available from the authors on request.

of each of the sources. To investigate this issue inmore detail and to get a better insight into the rela-tive contributions of each of the sources to the re-sulting projection errors, we implement the follow-ing exercise. Setting variables used in the 2SLSregressions of Table 4 to their mean values and usingestimated coefficients, we compute the contributionof each of the model variables to the projection er-rors for current account and fiscal balance ratios,(∆ cjT – ∆cjT) and (∆yjT – ∆yjT), respectively. Usingmeans of projection errors as anchors, we can drawsome conclusions about the relative contributions ofdifferences in modeling, differences in initial condi-tions, and differences in policy response to the pro-jection errors. Tables 5a and 5b summarize the resultsof the described experiment for current account andfiscal balance projection errors, respectively.

In the case of current account ratios, the most sig-nificant source of projection error comes from themeasurement of the initial conditions. This compo-nent is responsible for 44.55 percent of the total pro-jection error, while differences in modeling and dif-ferences in policy response generate forecast errorswith magnitudes of only 16.63 percent and 0.85 per-cent, respectively. The positive signs of percentagecontributions of all three sources suggest that thesesources of errors tend to bias the current accountmean projection error toward negative values.

However, when the components of the forecastingerror for fiscal-balance ratios are considered, the twomajor sources of the errors are differences in model-ing (166.17 percent) and mismeasurement of initialconditions (–52.78 percent). It appears that themodel used in projections tends to make the projec-tion error more positive while the measurementerror in the initial conditions pulls the projectionerror in the negative direction, as occurred in the caseof the current account projection errors. Differencesin policy response are responsible for approximately16 percent of the total mean projection error.

Projection errors of both variables are greatly in-fluenced by the year- and country-specific factorscaptured by the corresponding dummy variables.

It is evident, in examining the data, that there issubstantial mismeasurement in the fiscal and cur-rent account ratios when the initial values in thetwo databases are compared. Simple statistics for theactual and projection ratios are as follows (based on175 observations):

The difference in mean between historical and pro-jected data for the current account ratio is quite

striking. The value of cjT–1 should be known (i.e.,historical) at the time of the projection. Differencesof this magnitude are an indication that there hasbeen substantial revision in the macroeconomic ag-gregates over time.20 The difference in mean for thefiscal ratio is not so pronounced. The standard devi-ations are large, and these differ substantially be-tween actual and projection databases. There ismore variability in the actual current account ratiosthan in those projected; by contrast, there is morevariability in the projected fiscal ratios than there isin the actual ratios.

Figures 3 and 4 present the scatter-plots of actualand projected ratios. The 45 degree line representsthose combinations for which projected coincideswith actual. As the figures show, there is tremen-dous measurement error even in these initial condi-tions. There is also a strong positive correlation ofprojection with actual: for (cT–1, cT–1) it is 0.84,while for (yT–1, yT–1) it is 0.86. There is not the per-fect match that would exist in theory, but the matchis quite strong.

Examining the Role of Revisions

New information is made available to IMF staff on acontinuous basis throughout the duration of theIMF program. The staff periodically revisits its ini-tial projections in the context of a program review,and updates them to reflect the information morerecently received. We should then observe that theIMF projections converge to the actual performanceas revisions are made over the duration of a multi-period IMF program: imprecision in initial condi-tions will be eliminated, projected policy reform canbe revised in light of observed behavior, and inaccu-racy in the forecasting model can be reduced. More-over, one can expect that for multiperiod programs,the IMF staff ’s major efforts in the design of theoriginal programs would be concentrated on the im-provement of short-horizon projections while lessemphasis is placed on long-horizon projections sinceinitial projections can be fine-tuned in the contextof later reviews.

Assuming that the new information is efficientlyincorporated, we expect to observe that the IMFprojections converge to the historical performanceas revisions are considered. Therefore, any assess-ment of the quality of the IMF projections will beincomplete without examining the evolution of pro-jections. We address this issue by comparing theprojections of the original programs (OPs) with

65


20 We have been careful in constructing the dataset, but wemust admit as well the possibility that the definition of currentaccount used in the historical data may differ in some instancesfrom the definition used in the envisaged data. While we see noreason for this difference to be systematic, it may well representsome of the observed difference in mean values.

StandardHorizons Mean Deviation Minimum Maximum

cjT–1 –6.24 7.89 –39.82 11.08cjT–1 –5.09 5.99 –39.92 10.41yjT–1 –4.36 4.11 –20.48 5.61yjT–1 –4.22 4.42 –22.60 4.00

66


Table 5a

Forecast Error Components: Current Account Ratios

PercentVariable Coeff. Mean Effect Effect Total Percent Effect by Type

cjT–1 0.04 –6.64 –0.25 25.46yjT–1 –0.02 –4.39 0.08 –8.23 Differences in modeling: 16.63 percent∆ejT 0.00 –3.14 0.01 –0.60

(cjT–1– cjT–1) –0.31 1.26 –0.40 40.38Mismeasurement of initial conditions: 44.55 percent

( yjT–1– yjT–1) –0.25 0.16 –0.04 4.17

(∆êjT –∆ejT) 0.00 8.46 –0.01 0.85 Differences in policy response: 0.85 percent

t93 0.97 0.10 0.10 –10.43t94 0.37 0.15 0.06 –5.78t95 1.14 0.14 0.16 –16.54t96 1.99 0.11 0.22 –22.60t97 1.84 0.11 0.20 –20.83 Year-specific: –125.53 percentt98 3.35 0.10 0.35 –35.93t99 1.31 0.11 0.15 –14.90t00 0.50 0.10 0.05 –5.37t01 –1.21 0.06 –0.07 6.85

Country dummies Country-specific: 163.50 percent

∆cjT –∆cjT –0.98

Total: –0.98 100 100 percent

Number of observations 162

Table 5b

Forecast Error Components: Fiscal Balance Ratios

PercentVariable Coeff. Mean Effect Effect Total Percent Effect by Type

cjT–1 –0.10 –6.64 0.68 265.67∆cjT–1 0.03 –0.57 –0.01 –5.63yjT–1 0.04 –4.39 –0.19 –74.68 Differences in modeling: 166.17 percent∆yjT–1 –0.08 0.10 –0.01 –3.03∆ejT 0.01 –3.14 –0.04 –14.43∆wjT 0.02 –0.28 0.00 –1.73

(cjT–1 – cjT–1) –0.07 1.26 –0.09 –33.98Mismeasurement of initial conditions: –52.78 percent

(yjT–1 – yjT–1) –0.30 0.16 –0.05 –18.80

∆êjT – ∆ejT 0.01 8.46 0.07 28.22Differences in policy response: 15.99 percent

∆wjT – ∆wjT –0.47 0.07 –0.03 –12.24

t93 –0.12 0.10 –0.01 –4.77t94 0.42 0.15 0.06 25.34t95 1.19 0.14 0.17 66.65t96 –0.11 0.11 –0.01 –4.96 Year-specific variables: 182.08 percent t97 0.08 0.11 0.01 3.38t98 0.35 0.10 0.04 14.53t99 1.30 0.11 0.14 56.64t00 0.42 0.10 0.04 17.40t01 0.36 0.06 0.02 7.87

Country dummies –1.36 0.01 –0.02 –6.60 Country-specific variables: –211.46 percent

∆yjT –∆yT 0.25

Total: 0.25 100 100 percent

Number of observations: 162

67


–25

–20

–15

–10

–5

0

5

10

–25 –20 –15 –10 –5 0 5 10

Projection

His

toric

al

Figure 3

Initial Conditions for Fiscal Account Ratio

–50

–40

–30

–20

–10

10

20

–40 –35 –30 –25 –20 –15 -5 0 5 10 15

Projection

His

toric

al

–10

Figure 4

Initial Conditions for Current Account Ratio

68


those reported in the first reviews (FRs), which takeplace during the first program year.21 Some basicqualitative information on the evolution of the out-come projections can be illustrated by Figures 5aand 5b, where we compare historical and envisagedmean changes in the fiscal and current account ra-

tios. These plots are based on the data summarizedin Table 6.22 An obvious observation is that for thevast majority of projection horizons, the first-reviewprojections, ∆yjk

FR and ∆ c jkFR, are closer to the actual

outcomes, ∆yjk and ∆cjk, than the original programprojections, ∆y jk

OP and ∆ c jkOP. The only exception is

the change in the fiscal ratio for the horizon T.

21 Actual timing and number of reviews vary from program toprogram. In general, Stand-By Arrangements and extendedarrangements (SBAs and EFFs) have more frequent reviews thanarrangements under the structural adjustment facilities (SAF,ESAF, and PRGF). Further, the completion of reviews is oftenheld up by difficulties in complying with conditionality. Forthese reasons, we plan in future research to address the relation-ship between review timing and projection error.

22 Figures 5a and 5b include similar information to that of Fig-ures 1a and 1b. They differ, however, in the number of observa-tions used in creating the mean values. For example, Figure 1auses 175 observations for horizon T to calculate the mean histor-ical and envisaged change in the fiscal ratio, while Figure 5a uses120 observations for which both original program and first re-view observations of fiscal ratio are available.

0

0.5

1

1.5

2

2.5

3

3.5

T T+1 T+2 T+3

Cum

ulat

ive

chan

ge a

s pe

rcen

tage

of G

DP

Original Program ActualFirst Review

Figure 5a

Means of Projected and Historical Changes in Fiscal Ratios

–2.5

–2

–1.5

–1

–0.5

0

0.5

1

TT+1

T+2

+ T+3

Cum

ulat

ive

chan

ge a

s pe

rcen

tage

of G

DP

Original Program ActualFirst Review

Figure 5b

Means of Projected and Historical Changes in Current Account Ratios

69


Table 6

Projecting the Change in Macroeconomic Aggregates (Original Program (OP) Versus First Review (FR))

Variable N Mean Std Dev Sum Minimum Maximum

Horizon T∆c jT

FR 120 0.18624 3.67482 22.34863 –11.25961 14.09314

∆c jTOP 120 –0.26392 3.25906 –31.66988 –13.89236 9.14219

∆cjT 120 0.64113 4.33661 76.93582 –16.63708 14.29500

∆y jTFR 120 1.21250 2.68248 145.50000 –5.50000 9.40000

∆y jTOP 120 0.92700 2.82961 111.24000 –7.60000 11.00000

∆yjt 120 0.97491 2.95418 116.98875 –6.27027 12.72751

Horizon T+1∆cjT+1

FR 95 –0.74583 5.81160 –70.85403 –26.17494 11.90127

∆c jT+1OP 95 –0.81342 5.31906 –77.27518 –22.23187 11.90477

∆cjT+1 95 –0.39908 7.35763 –37.91262 –35.61176 17.19095

∆yjT+1FR 95 1.27189 3.35196 120.83000 –5.60000 13.30000

∆yjT+1OP 95 1.40253 3.18396 133.24000 –5.60000 12.90000

∆yjT+1 95 1.05813 3.75009 100.52207 –6.76704 13.69233

Horizon T+2∆cjT+2

FR 74 –0.31151 5.18034 –23.05211 –17.31635 11.10499

∆c jT+2OP 74 –0.22570 4.91495 –16.70172 –22.04280 10.81110

∆cjT+2 74 –0.37456 8.12439 –27.71773 –38.14743 21.78397

∆yjT+2FR 74 1.98784 3.49077 147.10000 –4.70000 13.50000

∆yjT+2OP 74 2.27838 3.15904 168.60000 –3.30000 13.20000

∆yjT+2 74 1.51000 4.07910 111.73996 –14.88070 11.88877

Horizon T+3∆cjT+3

FR 50 0.24462 5.16137 12.23123 –14.73737 12.91443

∆c jT+3OP 50 0.24960 4.89254 12.48019 –19.89594 7.35647

∆cjT+3 50 –2.07739 14.24848 –103.86963 –81.56932 21.40533

∆yjT+3FR 50 2.96400 4.02060 148.20000 –3.90000 13.10000

∆yjT+3OP 50 3.11580 3.52157 155.79000 –1.90000 13.00000

∆yjT+3 50 1.48644 4.21497 74.32213 –12.33862 11.64537

Notes: N denotes number of observations; Std Dev denotes standard deviation.

Assessment of only mean changes might be mis-leading, since, as we showed in the previous sec-tions, there is a great deal of variability in envisagedand historical data. Some additional insights on theevolution of the projections can be obtained by ex-amining developments in correlations between en-visaged and actual changes. Table 7 reports thosecorrelations for fiscal and current account ratios forboth first reviews and original programs. As wefound in the regressions of the previous section, en-visaged changes in fiscal ratios exhibit higher corre-lation with the actual changes than do comparable

changes in envisaged and actual current account ra-tios. This observation is true for projections drawnfrom the original program as well as those from thefirst reviews. Inaccuracy of the current account pro-jections seems to worsen significantly with thelength of the projection horizon. Also, there is astrong pattern showing that the projection perfor-mance of the first reviews, measured by the correla-tion coefficient, improves relative to the projectionsof the original programs for all variables and all pro-jection horizons. The gain in forecasting power isparticularly noticeable over short horizons—and

70


decreases as the length of the projection horizon extends.

Bias, Efficiency, and Accuracy of Revisions

Musso and Phillips (2001) suggested an interestingapproach to evaluating projections. They analyzeprojections on the basis of three major characteris-tics: bias, efficiency, and accuracy. In this paper, wefollow their approach and document some of thefacts along these three dimensions to compare therelative performance of the projections of the origi-nal programs and their first revisions.

Bias. By bias, we refer to the divergence of the dis-tribution of projection errors from the zero-meannormal distribution. Table 8 presents statistics characterizing the distribution of (∆ cT – ∆cT) and(∆yT – ∆yT) for the original programs as well as fortheir first revisions.23 Several observations can bemade from the information presented in Table 8:

• The null hypothesis of the true mean of the dis-tribution being zero is rejected more frequentlyfor the original program projection errors thanfor those from the first reviews. It is especiallynoticeable for the fiscal balance ratios.

• Standard deviations are considerably smaller forthe first-review projection errors than those forthe original programs. The difference is greaterfor short horizons and becomes very small oreven reverses for longer horizons.

• For the horizon T, positive skew of the distribu-tion of the projection errors for both variablessuggests that projection errors are more likely tobe far above the mean than they are to be farbelow the mean. This result can be observed forboth groups of projections. However, for longerhorizons, the skew tends to be negative, reflect-ing the opposite trend.

• For both variables and for most of the horizonlengths, the distribution of errors has more mass

in the tails than a Gaussian distribution withthe same variance. The only exceptions are pro-jection error distributions for horizon T changesin the current account (FR), and horizon T+1changes in the fiscal balance (both OP and FR).

• For the OP projection errors, most of the testsfind statistically significant evidence that thedistributions exhibit lack of normality. The onlyexception is the T+1 horizon for fiscal balances.For the FR projection errors, the results aremixed. Some of the goodness-of-fit tests for nor-mal distribution cannot reject the null hypothe-sis of normality.24

Efficiency. We test the efficiency of the FR andOP projections by regressing the value of the histor-ical change on a constant term and the value of pro-jected change as illustrated in equations (8a) and(8b) for macroeconomic variable gT, with vT and uTas random errors. We perform the estimation forchanges in variables as well as for the levels.

∆gT= c0+ c1∆gT+vT (8a)

gT= d0+ d1 gT+ uT. (8b)

This type of efficiency test is referred to as theweak criterion since it uses a limited information set(Musso and Phillips, 2001). We would concludethat the projection was an efficient estimate of thehistorical datum if the intercept were insignificantlydifferent from zero and the slope were insignifi-cantly different from unity. Tables 9 and 10 reportresults of the estimation in changes and in levels, re-spectively.

There is a striking relationship between historicaland projected changes found in the data: in eachcase, for the original program (except horizon T+3changes in current account ratios), the hypothesis ofweak efficiency is strongly rejected by the data. The

Table 7

Correlations Between Projected and Actual Outcomes for Changes in MacroeconomicAggregates (Original Program (OP) Versus First Review (FR))

Horizons T T+1 T+2 T+3

Fiscal ratio (∆yjk, ∆yjkFR) 0.69635 0.76157 0.70624 0.60535

Fiscal ratio (∆yjk, ∆yjkOP) 0.60742 0.69037 0.65761 0.57322

Correlation improvement (ρyFR – ρy

OP) 0.08893 0.0712 0.04863 0.03213

Current account ratio (∆cjk, ∆cjkFR) 0.69175 0.46345 0.33955 0.35714

Current account ratio (∆cjk, ∆cjkOP) 0.50390 0.34193 0.30747 0.35449

Correlation improvement (ρcFR – ρc

OP) 0.18785 0.12152 0.03208 0.00265

23 Projection errors are calculated as the differences betweenenvisaged values and actual realizations.

24 We used Kolmogorov-Smirnov, Cramer-von Mises, Anderson-Darling, and chi-square tests to check normality. The results arereported in Table AIII.3.

71


Table 8

Program Projection Errors

Horizon T T+1 T+2 T+3

Projection Errors in Changes of Fiscal Balance Ratios to GDP

Mean OP 0.048 –0.492** –1.334* –1.809*

FR –0.238 –0.252 –0.543 –1.478*

Median OP –0.014 –0.492 –0.786 –1.541

FR –0.232 –0.341 –0.548 –1.728

Standard deviation OP 2.565 2.852 4.070 3.673

FR 2.211 2.471 2.952 3.663

Skewness OP 0.028 –0.009 –2.406 –1.582

FR 0.827 0.312 –1.731 –1.591

Kurtosis OP 4.32 1.636 10.100 4.747

FR 6.142 1.057 9.812 6.467

Normality test OP Rejected Mixed (3/4) Rejected Rejected

FR Rejected Mixed (3/4) Mixed (1/4) Mixed (2/4)

Projection Errors in Changes of Current Account Ratios to GDP

Mean OP 0.905* 0.258 –0.595 –2.176

FR 0.455 0.262 –0.222 –2.220

Median OP 0.583 0.669 –0.614 –0.666

FR 0.281 0.410 –0.228 –0.994

Standard deviation OP 3.897 7.260 7.766 12.492

FR 3.204 6.824 7.832 12.726

Skewness OP 1.222 –2.630 –2.555 –4.612

FR 0.241 –3.405 –3.072 –4.702

Kurtosis OP 5.442 17.961 17.032 29.405

FR 1.825 24.079 18.618 29.359

Normality test OP Rejected Rejected Rejected Rejected

FR Rejected Mixed (1/4) Rejected Rejected

Notes: * Significantly different from zero at the 90 percent confidence level (based on student’s t-test).** Significantly different from zero at the 95 percent confidence level (based on student’s t-test).Mixed (X/4): X out of four tests cannot reject normality of the error terms at the 95 percent confidence level.OP denotes original program; FR denotes first review.

72


Table 9

Test of “Weak” Efficiency (in changes)

Original Program First Review

Coeff. S.E. t–statistic Coeff. S.E. t–statistic

Fiscal balance ratiosHorizon: T

Intercept (H0: Intercept=0) 0.387* 0.227 1.709 0.045 0.214 0.210

Slope (H0: Slope =1) 0.634*** 0.076 –4.792 0.767*** 0.073 –3.192

R2 0.369 0.485

Horizon: T+1

Intercept (H0: Intercept=0) –0.084 0.296 –0.284 –0.056 0.265 –0.211

Slope (H0: Slope =1) 0.731*** 0.082 –3.281 0.848** 0.075 –2.027

R2 0.439 0.577

Horizon: T+2


Slope (H0: Slope =1) 0.542*** 0.114 –4.018 0.814** 0.096 –1.938

R2 0.211 0.492

Horizon: T+3


Slope (H0: Slope =1) 0.704** 0.12 –2.467 0.635*** 0.12 –3.042

R2 0.377 0.366

Current account ratiosHorizon: T

Intercept (H0: Intercept=0) 0.818** 0.345 2.371 0.489* 0.288 1.698

Slope (H0: Slope =1) 0.671*** 0.106 –3.104 0.816** 0.079 –2.329

R2 0.254 0.479

Horizon: T+1


Slope (H0: Slope =1) 0.467*** 0.130 –4.100 0.583*** 0.113 –3.690

R2 0.114 0.215

Horizon: T+2


Slope (H0: Slope =1) 0.507*** 0.169 –2.917 0.550*** 0.169 –2.663

R2 0.097 0.120

Horizon: T+3


Slope (H0: Slope =1) 1.016 0.353 0.045 1.004 0.350 0.011

R2 0.127 0.134

Notes: * The null hypothesis (H0) can be rejected at the 90 percent confidence level.** The null hypothesis can be rejected at the 95 percent confidence level.*** The null hypothesis can be rejected at the 99 percent confidence level.Coeff. denotes coefficient; S.E. denotes standard error.

73


rejection is in all cases based on an estimate of c1 ord1 that is significantly less than unity. When the FRresults are examined, weak efficiency is once againrejected. However, when compared with the OP re-sults, the coefficient estimates of c1 and d1 are closerto the hypothesized value of unity.25

Accuracy. We test relative accuracy of the OP andFR projections by comparing them with a random-walk benchmark projection. That is, we investigatewhether the IMF projections of the year-T values ofthe variables do better than if the projections weresimply set equal to the T–1 values. We draw ourconclusions from Theil’s U statistic and report re-sults in Table 11.26 Larger values of the U statisticindicate a poor projection performance. The bench-mark random-walk projections for OP are based onthe T–1 value of the variable as it is documented inOP, while the benchmark random-walk projectionfor FR uses the initial conditions from the reviseddata of FR.

For the fiscal balance, both OP and FR projec-tions perform better than the random walk. How-ever, only the FR projection outperforms the ran-dom walk for the current account; the OPprojection for this variable is slightly worse thanthat of its random-walk counterpart. Overall, theFR projections exhibit lower values of the U statis-tic, reflecting their more accurate projections.27

An Empirical Decomposition

The preceding results suggest that the IMF staffmodifies its projections to incorporate new informa-tion, and that the revised projections have betterforecasting power when compared with the projec-tions of the original programs. It is possible to de-compose the difference in the OP and FR projec-tions using a methodology similar to that ofequation (3e) and Table 4. The details of this analy-sis are reported in Annexes I–III. The salient find-ings for our purpose are as follows:

• There is a substantial difference in initial condi-tions used in the two projections, and these dif-ferences contribute significantly to the im-provement of FR over OP.

• There is also evidence that the model used inthe FR projections differs significantly from thatused in the OP projections.

Table 10

Test of “Weak” Efficiency (in levels)

Original Program First Review

Coeff. S.E. t-statistic Coeff. S.E. t-statistic

Fiscal balance ratiosYear: T

Intercept (H0: Intercept=0) –1.870*** 0.345 –5.420 –1.483*** 0.297 –4.993

Slope (H0: Slope =1) 0.493*** 0.076 –6.671 0.685*** 0.071 –4.437

R2 0.263 0.437

Current account ratiosYear: T

Intercept (H0: Intercept=0) –1.529** 0.694 –2.203 –1.121** 0.529 –2.119

Slope (H0: Slope =1) 0.706*** 0.101 –2.911 0.912 0.081 –1.086

R2 0.289 0.512

Notes: ** The null hypothesis (H0) can be rejected at the 95 percent confidence level.*** The null hypothesis can be rejected at the 99 percent confidence level.Coeff. denotes coefficient; S.E. denotes standard error.

25 There is a difficulty in this type of estimation that is not ad-dressed by Musso and Phillips (2001). Since the right-hand-sidevariable is only an estimate of the true OP projection, the regres-sion may be characterized by error-in-variables. This will causethe slope coefficient to be biased downward. We investigated thispossibility using an instrumental-variable technique. The result-ing slope coefficients were, in most cases, closer to unity and in-significantly different from unity for the fiscal ratio, thus exhibit-ing weak efficiency. They were farther from unity for the currentaccount ratio, thus sustaining the conclusion of inefficiency forthat variable.

26 The Theil’s U Statistic:

UN g g

N g

jt jtjt

jtjt

=−∑

∑

1

1

2

2

( ˆ ).

27 The pattern of errors in OP and FR projections are similar tothose observed by Howrey (1984) in his study of inventory invest-ment. He found in that case that there was evidence of substantialrevision to initial data in inventory investment over the period1954–80. He also found, however, that knowledge of the revisionreduced only marginally the variance of projection error.

74


There is evidently “learning by doing” in theseprojections at the modeling stage as well as at thestage of data collection.

Conclusions

Envisaged and historical observations on the fiscaland current account ratios in countries participatingin 175 IMF programs between 1993 and 2001 devi-ated strongly from one another. Our statisticalanalysis suggests that the causes can be separatedinto four components.

First, the IMF staff was apparently working withquite different information about the initial condi-tions of the program countries than is currently ac-cepted as historical. This difference leads to substan-tial divergence even if the IMF staff used the modelrevealed by the historical data. This result is consis-tent with the conclusions of Orphanides (2001);and Callan, Ghysels, and Swanson (2002) on themaking of U.S. monetary policy.

Second, the IMF staff did appear to have a differ-ent model in mind when making its projections. Itsmodel was characterized by gradual fiscal accountadjustment, both in response to contemporaneouscurrent account shocks and to long-run imbalances,while the model revealed by historical data wascharacterized by more rapid adjustment to bothtypes of imbalances. Further, its envisaged responsewas concentrated in horizon T, while the historicalresponse to shocks was roughly equally proportionedacross horizons T and T+1.

Third, there is a difference between projected andhistorical implementation of policy adjustment.Given the level of aggregation of the policy vari-ables investigated (total government consumptionexpenditures, real exchange rate depreciation) wecannot conclude that the difference is due to a fail-

ure to meet the conditions of the program; the dif-ferences could also be due to shocks that worsenedthe performance of these aggregates even when con-ditions were fulfilled. This is a question that can,and should, be investigated further.

Fourth, there is ample evidence that, like othermacroeconomic projections, IMF projections arequite inaccurate. The evidence on “accuracy” re-ported here is instructive—while the projectionsoutperform a random walk most of the time, theyare not much better. The Meese and Rogoff (1983)results remind us of the difficulty in projecting ex-change rates in time series. The project describedhere indicates the inaccuracy of simple models in apanel (i.e., time-series and cross-section of coun-tries) format.

Our results on revisions indicate that the IMFstaff learns from past projection errors—and fromnew information. However, even that learningleaves large gaps to fill. The largest margin for im-provement may well be in “just-in-time” data col-lection, so that the errors owing to incomplete in-formation, especially from initial conditions, can beeliminated.

Annex I. Creating the Error-Correction Residuals

In the following tables (AI.1–AI.3), we use theIMF’s World Economic Outlook (WEO) datasetcovering those programs with time horizon T. Thereare 175 observations in general, although somewhatmore when considered in levels.

Creating the error correction residual ejt

Dependent variable: yjt (WEO)

Table 11

Test of Accuracy (in levels)

Theil’s U Statistic

Number of Fiscal Current Projection Model Observations balance ratios account ratios

Original program 121 0.695 0.696

Benchmark for OP (random walk) 121 0.788 0.639

First review 120 0.571 0.568

Benchmark for FR (random walk) 120 0.760 0.635

Notes: OP denotes original program; FR denotes first review.

75


Table AI.1

Analysis of Variance

Source DF Sum of Squares Mean Square F Value Pr > F

Model 86 5,518.14566 64.16448 6.58 < .0001Error 96 935.61705 9.74601Uncorrected total 182 6,453.76271

Root MSE 3.12186 R–square 0.8550Dependent mean –4.33059 Adjusted R–square 0.7252Coefficient of variance –72.08859

Parameter Estimates

Variable DF Parameter Estimate Standard Error t Value Pr > | t |

Cjt 1 0.09996 0.06203 1.61 0.1103t93 1 –7.41751 1.72365 –4.30 < .0001t94 1 –4.83851 1.91288 –2.53 0.0131t95 1 –6.31586 1.84898 –3.42 0.0009t96 1 –5.37486 1.92894 –2.79 0.0064t97 1 –3.98082 1.88383 –2.11 0.0372t98 1 –3.63622 1.95216 –1.86 0.0656t99 1 –4.64533 1.95383 –2.38 0.0194t00 1 –5.26644 1.97374 –2.67 0.0090t01 1 –5.92937 1.83106 –3.24 0.0017

Notes: DF denotes degrees of freedom; MSE denotes mean square error.

This is the formulation used to create the errorcorrection variable (eT–1= yt – predicted value) forWEO data. A complete set of country dummies wasused as well, but is suppressed here.

The following regression results report the coeffi-cients used in creating the error-correction variablefor envisaged data:

Table AI.2

Analysis of Variance Dependent variable: yjt (envisaged)

Source DF Sum of Squares Mean Square F Value Pr > F

Model 95 6,449.94187 67.89412 5.29 <0.0001Error 97 1,244.26623 12.82749Uncorrected total 192 7,694.20810

Root MSE 3.58155 R–square 0.8383Dependent mean –4.47401 Adj. R–square 0.6799Coefficient of variation –80.05230

Notes: DF denotes degrees of freedom; MSE denotes mean square error.

76


Annex II. Does the Timing of Approval of IMF-Supported Programs

Matter to These Results?

Projection errors, especially for the initial programyear (year T), may reasonably be hypothesized to de-pend on the point in time during the year when aprogram was approved. We investigated this hy-pothesis in two ways. First, we calculated Pearsoncorrelations of the approval month with the size of the projection error for horizons T and T+1(Table AII.1). Second, we regressed the projectionerror on dummy variables indicating the quarter ofyear T in which approval occurred (Table AII.2).

The Pearson correlations provide no evidence ofa significant approval-time effect in either variable.For the fiscal ratio, there is no evidence of a signifi-cant approval-time effect for either OP or FR pro-jection errors. For the current account ratio, a num-ber of coefficients are positive and significant.However, they do not grow uniformly over the sam-

ple; the largest deviations from the mean occur forprograms approved in the second and third quartersof “year T.”

We did the same exercise for the deviation in ini-tial conditions (Table AII.3); in that case, the hy-pothesis is that programs approved later in year Twill have more accurate information on the initialconditions, so that deviations will be lessened.There is no evidence of a significant effect in thePearson correlations. There is some evidence of thisin the regression results, however (Table AII.4). Forboth OP and FR versions of the fiscal ratio and theOP version of the current account ratio, the devia-tion in initial conditions is significantly larger, onaverage, for programs approved in the first quarter ofyear T than for those approved later in year T. Thereis thus a downward bias in the fiscal ratios used asinitial conditions in projections created in the firstquarter of year T relative to the historical data, mostlikely because the IMF staff did not have access tothe later revisions when creating its projections.

Table AI.3

Parameter Estimates

Variable DF Parameter Estimate Standard Error t Value Pr > | t |

Cjt 1 0.31664 0.07861 4.03 0.0001t93 1 –6.84332 1.81926 –3.76 0.0003t94 1 –4.68806 1.95701 –2.40 0.0185t95 1 –5.69861 1.90593 –2.99 0.0035t96 1 –3.85602 1.93132 –2.00 0.0487t97 1 –3.34252 1.93759 –1.73 0.0877t98 1 –2.74118 1.85499 –1.48 0.1427t99 1 –4.38718 2.07410 –2.12 0.0370t00 1 –3.95966 2.08914 –1.90 0.0610t01 1 –5.05367 2.00406 –2.52 0.0133

Notes: DF denotes degrees of freedom. A complete set of country dummies was used as well, but it has been suppressed here for brevity.

Table AII.1

Pearson Correlations for Projection Errors

Fiscal Balance: Fiscal Balance: Current Account: Current Account:Original Program First Review Original Program First Review

(approval month in T) (approval month in T) (approval month in T) (approval month in T)

Horizon T 0.01905 –0.02677 0.00687 0.067110.8364 0.7716 0.9406 0.4665

120 120 120 120

Horizon T+1 –0.05439 –0.15750 0.14289 0.149160.5853 0.1234 0.1499 0.1406

103 97 103 99

Notes: Each cell in this table includes three statistics: the top entry is the Pearson correlation coefficient; the middle entry is Prob > | r | underthe null hypothesis of zero correlation; and the bottom entry is the is the number of observations.

77


Table AII.2

Regressions on Quarterly Dummies (Horizon T ) for Projection Errors

Fiscal balance (OP) Fiscal balance (FR) Current account (OP) Current account (FR)

Quarter 1 0.06 –0.11 0.44 –0.15(0.46) (0.40) (0.69) (0.57)

Quarter 2 –0.27 –0.47 1.08* 0.94*(0.39) (0.34) (0.60) (0.49)

Quarter 3 0.26 –0.16 1.76** 0.26(0.51) (0.44) (0.77) (0.63)

Quarter 4 0.44 –0.04 0.13 0.63(0.59) (0.51) (0.90) (0.74)

R2 0.01 0.02 0.07 0.04Number of observations 120 120 120 120

Notes: OP denotes original program; FR denotes first review. Standard errors appear in parentheses.* Indicates significance at the 90 percent confidence level.** Indicates significance at the 95 percent confidence level.

Table AII.3

Pearson Correlations for Discrepancies in Initial Conditions (Actual – Projection)

Fiscal Balance: Fiscal Balance: Current Account: Current Account:Original Program First Review Original Program First Review

(approval month in T) (approval month in T) (approval month in T) (approval month in T)

All horizons 0.13744 0.10628 0.09385 –0.116110.1328 0.2440 0.3076 0.2028

121 122 120 122

Notes: Each cell in this table includes three statistics: the top entry is the Pearson correlation coefficient; the middle entry is Prob > | r | underthe null hypothesis of zero correlation; and the bottom entry is the is the number of observations.

Table AII.4

Regressions on Quarterly Dummies for Discrepancies in Initial Conditions

Fiscal Balance (OP) Fiscal Balance (FR) Current Account (OP) Current Account (FR)

Quarter 1 –0.80** –0.73** –1.75** 1.56(0.39) (0.35) (0.73) (1.11)

Quarter 2 –0.37 –0.40 –0.95 –0.82(0.34) (0.31) (0.63) (0.96)

Quarter 3 0.27 0.05 –0.75 0.62(0.44) (0.39) (0.82) (1.23)

Quarter 4 –0.18 –0.21 –0.34 –2.17(0.51) (0.46) (0.95) (1.44)

R2 0.05 0.05 0.07 0.04Number of observations 120 120 120 120

Notes: OP denotes original program; FR denotes first review. Standard errors appear in parentheses.** Indicates significance at the 95 percent confidence level.

78


If there is a value to this information, it should alsobe evident in the initial conditions as reported in FRrelative to OP for each program. In Table AII.5, wecompare the initial conditions, with deviations mea-sured as FR values minus OP values. A similar regres-sion on approval times within year T yields little evi-dence of a systematic bias, with only the currentaccount ratio showing any deviation of significance.The estimated coefficients are suggestive, though,rising from negative values for Q1 approval to ever-increasing values for subsequent quarters.

Annex III. What New Information Do Revisions Incorporate?

The results in the text suggest that IMF staff modi-fies its projections to incorporate new informationand that the revised projections have better fore-casting power relative to the original program. How-ever, it is not yet clear whether this is a reflection ofadjusting projections for new values of the initialconditions that contain less measurement error, or asign of using new information to modify the entirescope of the model used in projection. We choose toaddress this issue by estimating regressions of thefollowing general form:

(A3a)

(A3b)

The form is the same as that advanced in the pre-vious section. The difference in projections can bestated in somewhat different form in equation (A3c).When we subtract (A3a) from (A3b), we note fourdifferent reasons why the two projections will not bethe same: an updating of information on past events(within the first set of square brackets in (A3c)), in-creased information on policy implementation, achange in the “model” used in projection (within

the second set of square brackets in (A3c)), and pro-jection errors.

(A3c)

Here, we regress projected changes in the macro-economic variable as projected in the first review ofthe program (∆g jT

FR) on the projected change of thevariable as it was originally planned at the outset ofthe program (∆ g jT

OP) and on the terms reflecting improvement of the information on the initial condi-tions (∆g jT–1

FR – ∆g jT–1OP ) and (g jT–1

FR – g jT–1OP ). We also in-

corporate a term representing differences in projectedchanges in the policy variable, (∆sjT

FR – ∆ sjTOP), to cap-

ture changes in the implementation of conditions as-sociated with the programs. Finally, all the termswithin the second set of square brackets are includedto study whether the forecasting model has changed.

We predict that the value of the coefficient on∆g jT

OP will be unity, as would be the case, for example,if the first review simply caused a mean-preservingcontraction in the distribution of random errors.Values of ã1 and ã2 differing significantly from zerowill indicate that the revision observed in FR reflectsthe improved information about the initial condi-tions governing the economic success of the pro-gram. Figure AIII.1 illustrates the interpretation ofthis model. With ã1 and ã2 significantly differentfrom zero and the coefficient on ∆g jT

OP being unity,the revision should trigger the “old model, new ini-tial conditions” scenario pictured there. However, ifthe new information available during implementa-tion of the program called for correction of the entire

Table AII.5

Regressions on Quarterly Dummies (Horizon T ) for Differences in Initial ConditionsBetween First Review (FR) and Original Program (OP)

Fiscal Balance (FR – OP) Current Account (FR – OP)

Quarter 1 –0.072 –0.176(0.171) (0.367)

Quarter 2 0.023 0.074(0.148) (0.317)

Quarter 3 0.215 0.323(0.190) (0.408)

Quarter 4 0.032 0.904*(0.222) (0.477)

R2 0.013 0.039Number of observations 120 120

Notes: * Indicates significance at the 90 percent confidence level. Standard errors appear in parentheses.

∆ ∆ ∆ˆ ˆ ˆ ˆg a g a g b s vjTOP

jTOP

jTOP jT

OPj= + + +− −1 1 2 1 1 TTOP

∆ ∆ ∆ˆ ˆ ˆ ˆg a g a g b sjTFR

jTFR

jTFR jT

FR= + +− −� � �1 1 2 1 1 ++ vjT

FR

∆ = ∆

+ ∆ −∆ +− −

ˆ ˆ

( ˆ ˆ )

g g

a g g a

jTFR

jTOP

jTFR

jTOP� �1 1 1 22 1 1

1

(ˆ ˆ )

( ˆ ˆ

g g

b s s

jTFR

jTOP

jTFR

− −−

+ ∆ −∆�jjTOP

jTOP

jTOPa a g a a g

)

( ) ˆ ( )ˆ

(+

− ∆ + −

+− −� �1 1 1 2 2 1

��b b s

v v

jTOP

jTFR

jTOP

1 1− ∆

+ −

) ˆ

( ).

79


projection model, the coefficients on ∆g jT–1OP , g jT–1

OP ,and ∆ sjT

OP would be significantly different from zeroand the estimates would follow the “new model,new initial conditions” scenario in Figure AIII.1.

Table AIII.1 summarizes the results of the modelestimation for the ratio of fiscal balances to GDP forall programs in the sample at horizon T. Changes infiscal ratios as they are projected in the first reviewsof the programs are regressed not only on terms rep-resenting the error-correction structure of fiscal ra-tios but also on the similar terms corresponding tothe current account ratios. A complete set of timeand country dummy variables was also included inthe regressions. The following insights can be ob-tained from the first column of Table AIII.1:

• The value of the coefficient on ∆yjTOP is 0.986,

which is not statistically different from unity.This could be interpreted as if the correction ofthe projection reported in the first review of theprogram is just a modification of the projectionowing to the more accurate initial conditions.The updated information set available at themoment of revision is incorporated into thesame projection model that was used to createOP projections. This result is consistent withthe fact that none of the terms included to cap-ture projection model modification is signifi-cantly different from zero.

• The coefficient on (∆yjT–1FR – ∆yjT–1

OP ) is negativeand significant. One of the potential explana-tions of this fact can be outlined as follows. Sup-pose that reduction of the measurement errorresults in an improvement in the fiscal balancein the years preceding the program relative towhat it had been originally thought to be when

the program was designed. That would meanthat (∆yjT–1

FR – ∆yjT–1OP ) is a positive number. Given

our finding, this would result in a reduction ofthe projected change in the fiscal ratio projectedin the first review of the program. Moreover,the value of the coefficient, –0.973, is not sig-nificantly different from –1, which suggests thatthis is a one-to-one relationship. This findingmakes intuitive economic sense, because if thegovernment’s budget deficit is not as bad as wasoriginally thought, the required correction ofthe fiscal balance is also less demanding.

• Specification testing reveals that changes inlagged first-difference terms with lag lengthgreater than 1 do not contribute significantly tothe regression. At the same time, none of thecurrent account ratio terms is significantly dif-ferent from 0, which suggests that improvementin the data quality of the current account has lit-tle effect on the projections of the fiscal ratios.

• The coefficient on the difference in the policyvariable, (∆ s jT

FR – ∆ s jTOP), is negative and signifi-

cant, implying that differences in policy be-tween OP and FR are also responsible for theamendments of the original projections. More-over, the negative sign of this coefficient sug-gests that a greater observed real depreciationresults in less positive forecasts of changes in fis-cal balance ratios.28

• Finally, testing jointly that both lagged levelterms are not significantly different from zero

T–1 T time

gtFR

gt–1FR

gtOP

gt–1OP

∆gtOP

∆gtFR

Old model, new initialconditions

New model, newinitial conditions

Figure A.III.1

Incorporation of New Information in Projections

28 Although it would be more reasonable to use total govern-ment expenditure as a policy variable in the regression for fiscalbalances, the number of observations available for the first re-views limits the use of this variable as a proxy for a policy variable.

80


allows us to conclude that revisions to initialconditions do not contribute systematically tothe changes observed in FR relative to OP.

The second column in Table AIII.2 reports results ofthe estimation of a similar model when the laggedlevel terms are excluded from the regression.

• The coefficient on ∆yjTOP is still insignificantly

different from unity and the hypothesis that theIMF staff does not modify the projection modelas the new information arrives is strongly sup-ported by the data.

• At the same time, the coefficient on ∆êjTOP is

significantly different from zero at the 90 per-cent confidence level, providing some supportfor the hypothesis that the scope of the project-ing model was amended.

• The coefficient on (∆yjT–1FR – ∆yjT–1

OP ) is still nega-tive, although much smaller in absolute value.

• The policy variable coefficient is still signifi-cantly different from zero.

Similarly, Table AIII.2 presents outcomes of themodel estimation for the current account ratios forall programs in the sample at horizon T. Once again,we regress changes in current account ratios fromthe first reviews of the programs on terms represent-ing the error-correction structure of current accountratios and on the similar terms corresponding to thefiscal ratios, as well as on the policy variable and theset of time and country dummies. The first columnof the table represents the case in which the error-correction terms are included in the regression:

• The value of the coefficient on the originallyprojected change in the current account, ∆cjT

OP,is 0.411 and the null hypothesis of the truevalue of this coefficient being unity is rejectedat the 99 percent confidence level. Unlike our

Table AIII.1

Regression Results, Fiscal Account Ratios (first review versus original program)

∆yjTFR ∆yjT

FR

Coefficient Standard error Coefficient Standard error

∆ yjTOP 0.986*** 0.141 0.970*** 0.130

(∆cjTFR – ∆cjT

OP) 0.040 0.137 0.047 0.124

(∆yjT–1FR – ∆yjT–1

OP ) –0.973*** 0.357 –0.756*** 0.271

(∆cjT–1FR – ∆cjT–1

OP ) 0.314 0.237 0.210 0.224


OP ) –0.350 0.406 –0.203 0.316


OP ) –0.246 0.221 –0.206 0.192

( yjT–1FR – yjT–1

OP ) 0.211 0.553

(cjT–1FR – cjT–1

OP ) –0.308 0.209

∆yjT–1OP 0.075 0.087 0.035 0.076

yjT–1OP –0.043 0.133 –0.066 0.114

êjTOP –0.009 0.008 –0.012* 0.007

(∆êjTFR – ∆ê jT

OP ) –0.045*** 0.016 –0.041*** 0.016

Number of observations 91 91

R2 0.988 0.986

Adjusted R2 0.926 0.925

Notes: Full sample, Horizon T. * Significantly different from zero at the 90 percent confidence level. *** Significantly different from zero at the 99 percent confidence level. A complete set of time and country dummies was included in the regressions, but their coefficients have been suppressed for brevity.

81


result for the case of fiscal ratio projections, theprojected change in the current account ratio inthe revision of the program appears to be de-rived under a different model relative to thechange in the current account projected at thebeginning of the program.

• Modification of the projection model is alsostrongly supported by the fact that the coeffi-cient on ∆c jT–1

OP is significant at the 99 percentconfidence level.

Excluding the lagged level terms from the regressiongives us a slightly better understanding of the rela-tionship between the considered variables.

• The coefficient on the originally projectedchange in the current account, ∆cjT

OP, is still sig-nificantly different from unity at the 99 percentconfidence level and takes the value of 0.439.Thus, we still find strong support for distin-guishing between the original program and first-review projection models.

• However, one of the terms representing chan-ges in the initial conditions for fiscal ratio,(∆yjT–2

FR – ∆yjT–2OP ), is significantly different from

zero at the 90 percent confidence level, with acoefficient value of –1.279. This suggests thatthe projection of the current account ratio is sig-nificantly affected by the changes in the initialconditions of the fiscal balance ratios. Moreover,the sign of the estimated coefficients indicatesthat an improvement in the initial conditions ofthe fiscal balance relative to what it was origi-nally assumed to be when the program was de-signed induces a reduction in the projectedchange in the current account for some givenvalues of the other variables. This result is sup-ported by our previous finding that the coeffi-cient on (∆yjT–1

FR – ∆yjT–1OP ) in the regression of fis-

cal balances reported in Table AIII.1 is negative.To illustrate this, suppose that reduction of themeasurement error results in the improvementof the fiscal balance initial conditions relativeto what it had been originally thought to be

Table AIII.2

Regression Results, Current Account Ratios (first review versus original program)

∆cjTFR ∆cjT

FR

Coefficient Standard error Coefficient Standard error

∆cjTOP 0.411*,+++ 0.234 0.439***,+++ 0.207


OP ) –0.194 1.037 –0.211 0.666


OP ) –0.297 0.691 –0.171 0.547


OP ) –1.099 0.939 –1.279* 0.728


OP ) 0.037 0.474 0.084 0.428

( yjT–1FR – yjT–1

OP ) 0.185 1.250

(cjT–1FR – cjT–1

OP ) 0.178 0.586

∆cjT–1OP –0.085 0.201 –0.066 0.180

cjT–1OP –0.366*** 0.142 –0.382*** 0.126

êjTOP 0.013 0.012 0.013 0.011

(∆êjTFR – ∆ê jT

OP ) –0.018 0.043 –0.025 0.036

Number of observations 91 91

R2 0.939 0.938

Adjusted R2 0.651 0.688

Notes: Full sample, Horizon T. * Significantly different from zero at the 90 percent confidence level. *** Significantly different from zero at the 99 percent confidence level. +++ Significantly different from unity at the 99 percent confidence level. A complete set of time and country dummies was included in the regressions, but their coefficients have been suppressed for brevity.

82


when the program was designed, implying that(∆yjT–1

FR – ∆yjT–1OP ) is positive. Since the coefficient

on this term is negative, this would result in thereduction of the projected change in the fiscalratio projected in the first review of the pro-gram, ∆yjT

FR. Then the macro identity written inthe first-difference form, ∆yjT

FR = ∆ cjTFR – ∆ pjT

FR,suggests that for any given value of private sav-ing, ∆ pjT

FR, the projected change in the currentaccount, ∆ cjT

FR, also reduces. This decrease inthe current account ratio as a result of improve-ment in the initial conditions for fiscal ratios iscaptured in our model by the negative sign ofthe coefficient on the corresponding terms.

Our analysis shows that the correction in the ini-tial conditions has a strong influence on the magni-tude of the projections for both fiscal and currentaccount ratios. Therefore, it appears to be logical tolook at the magnitude of those corrections and theirdistribution. Figures AIII.2 and AIII.3 illustrate thedistribution of the corrections in the levels of fiscalbalance ratio to GDP and the distribution of thecorrections in the levels of current account ratio toGDP, respectively, for the year T–1. These correc-tions are large, varying between –5.3 percent and4.6 percent of GDP for fiscal ratios and between–9.3 percent and 8.1 percent of GDP for current account ratios. The mass of the distributions is

–6.0

50

40

30

20

10

0

–4.5 –3.0 –1.5 0.0

m_gm 1_FRmOP

Normal (Mu = 0.0398; Sigma = 0.9561)

Perc

ent

1.5 3.0 4.5 6.0

Minimum –5.300Median 0.000Maximum 4.600Mean 0.040Standard deviation 0.956Skewness –0.337Kurtosis 14.135

Figure A.III.2

Distribution of Differences in T–1 Levels Between FR and OP for Fiscal Ratios

–12

40

35

30

25

20

15

10

5

0

–8 –4 0m_bm 1_FRmOP

Curve: Normal (Mu = 0.1926; Sigma = 2.082)

Perc

ent

4 8 12

Minimum –9.319Median 0.000Maximum 8.081Mean 0.193Standard deviation 2.082Skewness –0.154Kurtosis 8.349

Figure A.III.3

Distribution of Differences in T–1 Levels Between FR and OP for Current Account Ratio

Notes: FR denotes first review; OP denotes original program.

Notes: FR denotes first review; OP denotes original program.

83


concentrated around 0. The negative skew in bothcases shows that the corrections of the initial condi-tions are more likely to be far below the mean thanthey are to be far above the mean. Also, both distri-butions have kurtosis that exceeds 3, which impliesthat they have more mass in the tails than a Gaussian distribution with the same variance.Table AIII.3 reports results of the goodness-of-fittests for the normal distribution. All of the testsstrongly reject the null hypothesis of the initial con-dition corrections having a Gaussian distribution.

ReferencesCallan, Myles, Eric Ghysels, and Norman R. Swanson,

2002, “Monetary Policy Rules with Model and DataUncertainty,” Southern Economic Journal, Vol. 69,pp. 239–65.

Clements, Michael P., and David Hendry, 1995, “Fore-casting in Cointegrated Systems,” Journal of AppliedEconometrics, Vol. 10, pp. 127–46.

Hamilton, James, 1994, Time Series Analysis (Princeton,New Jersey: Princeton University Press).

Hendry, David, 1997, “The Econometrics of Macroeco-nomic Forecasting,” Economic Journal, Vol. 107,pp. 1330–57.

Howrey, E.P., 1984, “Data Revision, Reconstruction andPrediction: An Application to Inventory Invest-ment,” Review of Economics and Statistics, Vol. 66,No. 3, pp. 386–93.

Meese, Richard, and Kenneth Rogoff, 1983, “EmpiricalExchange Rate Models of the Seventies: Do They FitOut of Sample?” Journal of International Economics,Vol. 14, pp. 3–24.

Musso, Alberto, and Steven Phillips, 2001, “ComparingProjections and Outcomes of IMF-Supported Pro-grams,” IMF Working Paper 01/45 (Washington: In-ternational Monetary Fund).

Orphanides, Athanasios, 2001, “Monetary Policy Rules,Macroeconomic Stability and Inflation: A View fromthe Trenches,” Working Paper (Washington: Boardof Governors of the Federal Reserve System).

Table AIII.3

Goodness-of-Fit Tests for Normal Distribution for Corrections in Initial Conditions

Variable Test Statistic P Value (H0: normal)

Kolmogorov-Smirnov 0.3179 < 0.010

Cramer-von Mises 3.9281 < 0.005(yjt–1

FR – yjt–1OP)

Anderson-Darling 18.6995 < 0.005

chi-square 17,877.8796 < 0.001

Kolmogorov-Smirnov 0.24224 < 0.010

Cramer-von Mises 2.73701 < 0.005(cjt–1

FR – cjt–1OP )

Anderson-Darling 13.70785 < 0.005

chi-square 9,843.90054 < 0.001

Note: H0 denotes the null hypothesis.

84

5

This paper assesses the effects of fiscal consolidation andexpenditure composition on economic growth in a sam-ple of 39 low-income countries during the 1990s. Thepaper finds that strong budgetary positions are generallyassociated with higher economic growth in both the shortand long terms. The composition of public outlays alsomatters: countries where spending is concentrated onwages tend to have lower growth, while those that allo-cate higher shares to capital and nonwage goods and ser-vices enjoy faster output expansion. Finally, initial fiscalconditions also have a bearing on the nexus between fis-cal deficits and growth.

Introduction

A large body of empirical research supports the no-tion that healthy budgetary balances are, over thelong run, good for growth (Easterly, Rodriguez, andSchmidt-Hebbel, 1994). The effect of fiscal consoli-dation on growth in the short run, however, remainsopen to question, as a number of studies—largely forindustrial countries—have drawn the conclusionthat under some circumstances, fiscal contractionscan stimulate growth.2 A central theme in theseworks is that the composition of fiscal adjustmentplays a key role in determining whether fiscal con-tractions lead to higher growth and are also sustain-able over time. These studies show that improvingfiscal positions through the rationalization of the gov-ernment wage bill and public transfers, rather thanincreasing revenues and cutting public investment,can foster higher growth, even in the short run.

The purpose of this paper is to assess whether fis-cal consolidation and improvements in the compo-

sition of public expenditure have positive repercus-sions for growth in low-income countries. Whilesome aspects of this issue have been assessed inother studies,3 an in-depth econometric evalua-tion—drawing on a wide sample of low-incomecountries—has yet to be undertaken. For example,in the group of 36 different empirical studies thatKneller, Bleaney, and Gemmell (1998) identify asthe core of the empirical research on the effects offiscal policy on growth, only three studies (includingLandau, 1986; and Easterly, Rodriguez, andSchmidt-Hebbel, 1994) were based on developingcountries, and none of those was based on low-income countries alone.

A number of important related issues have not yetbeen fully examined in the literature. None of thesestudies, for example, have addressed whetherdeficits that are financed from abroad have a differ-ent impact on growth than those financed from do-mestic sources. In addition, the important issue ofwhether the macroeconomic effects of fiscal policydiffer in low-deficit countries—as opposed to thosethat have yet to achieve a modicum of macroeco-nomic stability—has yet to be assessed for a widesample of countries.4

This paper attempts to fill some of these gaps andaims to provide some empirical evidence of the ef-fects of fiscal adjustment and expenditure composi-tion on economic growth. More specifically, thepaper addresses the following two questions:

• What is the impact of the fiscal stance, expen-diture composition, and the nature of budget fi-nancing on economic growth in low-incomecountries?

• Are these effects independent of initial fiscalconditions?

This paper does not restrict its analysis to episodesof fiscal adjustment, as has been done in studies for

Fiscal Policy, Expenditure Composition, and Growth in Low-Income CountriesSanjeev Gupta, Benedict Clements, Emanuele Baldacci, and Carlos Mulas-Granados1

1 The authors wish to thank Shamit Chakravarti and ErwinTiongson for their help in preparing this paper. Emanuele Baldacci was an economist, and Carlos Mulas-Granados was a sum-mer intern, in the IMF’s Fiscal Affairs Department when this paperwas prepared. This article was previously published in the April2005 issue of the Journal of International Money and Finance andappears here with permission of Elsevier, the journal’s publisher.

2 See, for example, McDermott and Wescott (1996); Alesinaand Perotti (1996); Alesina, Perotti, and Tavares (1998); Alesinaand Ardagna (1998); Buti and Sapir (1998); Alesina and others(1999); and Von Hagen and Strauch (2001).

3 See Mackenzie, Orsmond, and Gerson (1997); Abed andothers (1998); and Kneller, Bleaney, and Gemmell (1999).

4 See Adam and Bevan (2000) for a study based on 17 low-income countries.

CHAPTER

85

Gupta, Clements, Baldacci, and Mulas-Granados

industrial countries. Instead, it assesses the effects ofboth fiscal expansions and fiscal consolidations ongrowth in 39 low-income countries with IMF-supported programs in the 1990s.5 These programs,on average, have targeted relatively small reductionsin budget deficits.6 Furthermore, the elimination ofbudget imbalances has not been the sole aim ofthese IMF-supported programs, which also sought,among other things, to improve the composition ofpublic expenditure and revenues. As such, an exclu-sive focus on episodes of fiscal adjustment—definedas periods of sharp deficit reduction—would be ofonly limited interest in examining the impact of fis-cal policy on growth in low-income countries.

The results of this study confirm that there is astrong link between public expenditure reform andgrowth, as fiscal adjustments achieved through cur-tailing current expenditures are, in general, moreconducive to growth. Fiscal consolidations tend tohave the most positive effects on growth when theylead to a reduction in the government’s domesticborrowing requirement. When public investment isalso protected, the positive effect of fiscal adjust-ment on growth is further accentuated. The fiscalconsolidation-growth nexus is also influenced by acountry’s initial fiscal conditions—in particular,whether the country has reached a certain degree ofmacroeconomic stability or not.

The rest of this paper is structured as follows: thesecond section surveys the literature on the effectsof fiscal policy and budget composition on eco-nomic growth; the third section describes the dataused in the empirical subsection; and the fourth sec-tion presents some baseline econometric results ofthe effects of fiscal policy and expenditure composi-tion on economic growth. Particular attention isgiven to examining the robustness of the results and whether results differ for low-deficit (“post-stabilization”) countries. The fifth section concludesthe paper and elaborates on some policy implica-tions of the results.

Literature Review

The effects of fiscal policy on economic growthhave been the subject of long debate. With respectto short-term effects, a large body of empirical re-

search, primarily for industrial countries, has beendevoted to understanding under which conditionsfiscal multipliers can be small (and even negative)(Alesina and Perotti, 1996; Alesina and Ardagna,1998; and Perotti, 1999). Perotti (1999), for exam-ple, shows that consolidations tend to be expansion-ary when debt is high or growing rapidly, whileAlesina and Perotti (1995) and Alesina andArdagna (1998) find that in addition to the size andpersistence of the fiscal impulse, budget compositionmatters in explaining different private sector re-sponses to fiscal policy (and hence the effect ongrowth). Fiscal adjustments that rely primarily oncuts in transfers and the wage bill tend to last longerand can be expansionary, while those that rely pri-marily on tax increases and cuts in public invest-ment tend to be contractionary and unsustainable(Von Hagen and Strauch, 2001).

The potential effects of fiscal policy on long-termgrowth has also generated substantial attention(Tanzi and Zee, 1996). Most recently, the burgeon-ing work in the field of endogenous growth suggeststhat fiscal policy can either promote or retard eco-nomic growth, as investment in physical and humancapital—both of which can be affected by taxationand government expenditures—can affect steady-state growth rates (Chamley, 1986; Barro, 1990 and1991; Barro and Sala-ì-Martin, 1995; and Mendoza,Milesi-Ferretti, and Asea, 1997).

In both strands of the literature, the effect of fis-cal policy on growth can be nonlinear. This mayoccur, for example, because the private sector’s re-sponse to fiscal policy may be nonlinear, implying acomplex relationship between the size and composi-tion of public spending and revenues and growth.Giavazzi, Jappelli, and Pagano (2000), for example,find that in industrial and developing countries, thenonlinear effects of fiscal policy on national savingstend to be associated with large and persistent in-creases in the primary deficit.

There are good reasons to believe that for some(but not all) low-income countries, fiscal contrac-tions may also be expansionary. As in the industrialcountries, expansionary contractions are more likelyto be observed in countries that have not yetachieved a degree of macroeconomic stability.7 Forthese countries, the overriding imperative of reiningin inflation and achieving low budget deficits aresuch that increases in public spending—even if potentially productive—may not have a salutary ef-fect on growth. By contrast, countries in a “post-stabilization” phase can exercise more choice overexpenditure priorities, including by allocating re-sources to important structural reforms, such as thedecompression of the civil service payscale. In these

7 For an empirical analysis of the impact of initial conditionson the effectiveness of fiscal policy during recessions in industrialand middle-income countries, see Baldacci and others (2001).

5 This includes countries that have obtained concessionalloans from the IMF since 1999 under the Poverty Reduction andGrowth Facility (PRGF), which replaced the Enhanced Struc-tural Adjustment Facility (ESAF). One of the basic tenets of thePRGF is that a stable macroeconomic position is critical for promoting growth and reducing poverty. For further informationon the characteristics of the PRGF, see http://www.imf.org/external/np/exr/facts/prgf.htm.

6 For example, for ESAF-supported programs over the 1986–95period, the deficit was targeted, on average, to decrease by about1 percentage point of GDP relative to the preprogram year(Abed and others, 1998).

86

5 � FISCAL POLICY, EXPENDITURE COMPOSITION, AND GROWTH IN LOW-INCOME COUNTRIES

countries, higher public spending—even if it results inhigher deficits—could expand, rather than contract,economic activity. In sum, the relationship betweenthe fiscal policy stance and growth will differ acrosscountries, depending on their initial fiscal condi-tions. This also has important implications for theeconometric specifications used to link fiscal policyand growth (see the subsequent discussion).

Another important issue to be considered in theanalysis is the nexus between the composition of fis-cal deficit financing and growth. Many studies havefound that fiscal consolidations can have an indirectimpact on private investment (and thus growth) byaffecting the level of aggregate demand and mone-tary variables. Deficits largely financed by domesticsources may also lead to inflationary pressures. Highlevels of inflation have been found to reduce growthand can lead to macroeconomic and financial insta-bility (Fischer, 1983; Sarel, 1996; and Khan andSenhadji, 2001).

In sum, the theoretical framework underlying theempirical analysis carried out in this paper assumesthat fiscal policy can affect the steady-state andshort-run growth rate through its effects on privatesector behavior and on human and physical capitalformation. It also acknowledges that initial and ac-companying macroeconomic and fiscal conditionsare important.

Statistical Data and Descriptive Analysis

Data

In this paper, three aspects of a country’s fiscal pol-icy are examined in relation to their impact ongrowth: the fiscal policy stance, as measured by thelevel and changes in the general government bud-getary balance; the financing of budgetary deficits;and expenditure composition. Data for these vari-ables were constructed on the basis of the IMF’sWorld Economic Outlook (WEO) database, as wellas a database for 39 IMF member countries supportedby Enhanced Structural Adjustment Facility (ESAF)and Poverty Reduction and Growth Facility (PRGF)arrangements during the period 1990–2000.8

The fiscal policy stance is measured by the gen-eral government budget balance on a cash basis.

8 The countries are Albania, Armenia, Benin, Bolivia, BurkinaFaso, Cambodia, Cameroon, the Central African Republic,Chad, Djibouti, Ethiopia, The Gambia, Ghana, Georgia,Guinea, Guinea-Bissau, Guyana, Honduras, Kenya, the KyrgyzRepublic, the Lao People’s Democratic Republic, Lesotho, theformer Yugoslav Republic of Macedonia, Madagascar, Malawi,Mali, Mauritania, Moldova, Mozambique, Nicaragua, Niger,Rwanda, São Tomé and Príncipe, Senegal, Tajikistan, Tanzania,Vietnam, the Republic of Yemen, and Zambia.

Table 1

Descriptive Statistics (as percentages of GDP, unless otherwise specified)

Variable Observations Mean Standard Deviation

Budget balance 429 –6.30 7.9Tax revenue 425 15.00 7.5Nontax revenue 423 2.50 2.2Grants 426 4.20 4.6Current spending 425 19.70 9.5Capital spending 425 9.00 7.2

Domestic financing 372 1.70 4.9External financing 372 4.60 6.1Per capita real GDP growth 429 –0.50 8.3

Change in:Budget balance 390 0.40 5.8Tax revenue 386 0.02 3.2Nontax revenue 384 –0.06 1.2Grants 386 0.03 2.6Current spending 386 –0.50 4.8Capital spending 386 0.05 3.4Domestic financing 333 –0.20 4.8External financing 333 –0.10 5.2Per capita GDP growth 390 0.50 10.1

Source: Authors’ calculations.Note: Sample averages using data from 1990 until 2000.

87


This is defined as total revenues and grants minustotal expenditures and net lending.9 A positivechange in the budget balance can be interpreted as aconsolidation, and a negative change as an expan-sion. As reported in Table 1, the average budgetdeficit for the sample is 6.3 percent of GDP. Deficitswere generally reduced during the period, with anaverage annual improvement of approximately 1⁄2 ofone percentage point of GDP.

The deficit can be financed from either domesticor external sources. Domestic financing includesboth bank and nonbank financing, with the lattermeasure including privatization receipts. For thecountries included in the sample, external financingpredominated, while domestic financing averagedless than 2 percent of GDP.

Fiscal deficits are also used to identify “post-stabilization” countries, which are defined as thosethat had an average budget deficit (after grants)below 2.5 percent of GDP during the 1990–2000 period.10 Based on this criterion, only seven coun-tries can be considered post-stabilizers (Benin, TheGambia, Lesotho, the former Yugoslav Republic ofMacedonia, Mauritania, Senegal, and Tanzania).

Macroeconomic indicators have also been ex-tracted from the WEO database. Following earlierstudies, growth is measured on a real per capitabasis.11 Other variables used in the regression analy-sis to control for initial and accompanying condi-tions include the labor force (as a percentage of thetotal population), terms of trade, and private invest-ment. These variables are used to control for the ef-fects of private sector and external sector activity ongrowth. We also control for the level of initial pri-mary and secondary enrollment as indicators ofhuman capital endowment in each country. Dataare taken from World Development Indicators ofthe World Bank.

Fiscal Policy and Growth: Bivariate Analysis

Simple correlations reported in Table 2 show a sig-nificant association between fiscal adjustment, ex-penditure composition, and growth consistent withprevious findings in the literature on industrialcountries. For example, stronger budget balances are

9 The difference between revenues and expenditures can bedifferent from the cash deficit for countries that measure expen-ditures on a commitment basis.

10 This roughly corresponds to the low-deficit country groupidentified in the ESAF review (Abed and others, 1998).

Table 2

Bivariate Correlations (variables expressed as percentages of GDP, unless otherwise specified)

Variables Per Capita Real GDP Growth Number of Observations

Budget balance 0.23*** 429Tax revenue –0.03 425Nontax revenue 0.03 423Grants 0.05 425Current spending –0.24*** 425Capital spending 0.16*** 425

Domestic financing –0.25*** 372External financing –0.07 372

Change in:Budget balance 0.20*** 390Tax revenue 0.09** 386Nontax revenue 0.08* 386Grants 0.11** 384Current spending –0.16*** 386Capital spending 0.12*** 386Domestic financing –0.16*** 333External financing –0.01 333

Source: Authors’ calculations.Notes: Bilateral correlations using annual data from 1990 through 2000.* significant at 10 percent; ** significant at 5 percent; *** significant at 1 percent.

11 Growth of per capita GDP is used most frequently in the em-pirical literature assessing the effects of fiscal policy on growth, asthis controls for differences among countries in the populationgrowth rate. See, for example, Aschauer (1989); Barro (1990,1991); Easterly and Rebelo (1993); Devarajan, Swaroop, and Zou(1996); Easterly, Loayza, and Montiel (1997); and Kneller,Bleaney, and Gemmell (1999, 2000).

88


strongly and positively associated with per capitagrowth. The composition of public expenditure alsomatters for growth; higher capital outlays are associ-ated with more buoyant growth, while higher cur-rent expenditures and domestic financing of thedeficit are associated with less favorable economicperformance.

These results hold for the short-run correlationsas well. Annual changes in the budget balance arepositively correlated with changes in per capitagrowth. Correlation coefficients12 are also signifi-cant for the various measures of public expendi-ture (including capital outlays) and for domestic fi-nancing.

These preliminary findings are consistent withthe empirical results obtained by Easterly and Rebelo(1993) and Kneller, Bleaney, and Gemmell (1999,2000), who found that balanced budgets and invest-ment in transport and communications are consis-tently correlated with growth in a sample of low-income countries.

Econometric Analysis

The Econometric Models

The relationship between expenditure composition,fiscal adjustment, and growth can be estimated byregressing the annual rate of real per capita GDPgrowth on a set of regressors, including fiscal vari-ables and other control variables. Three specifica-tions of the relationship are used here. In Model A,fiscal variables are measured as a share of GDP,without a variable included on the fiscal balance;this allows us to capture the effects of particular ex-penditure items (e.g., wages) not only on the com-position of expenditure but also on the deficit. InModel B, we measure fiscal variables in relation tototal expenditures or total revenues, so as to assessdirectly the impact of expenditure or revenue com-position on growth, while at the same time includ-ing a variable for the budget balance. In Model C,we address how the nature of the deficit financingaffects growth by substituting the budget balancevariable with variables for domestic and external fi-nancing of the deficit.

Each of the three models is formulated as follows:

Budget components (revenue and expenditure) mea-sured as a share of GDP (Model A):

(1)

where git is the growth rate of real per capita GDP;Yilt is a vector of nonfiscal independent variables

(initial level of GDP per capita, private investmentratio, terms of trade, labor force, initial level of pri-mary and secondary enrollment rates); and XGDPihtis a vector of independent fiscal variables aimed atcapturing the effect of the composition of the bud-get. These variables are measured in percentage ofGDP and include public sector wages and salaries,expenditures on other goods and services, transfersand subsidies, interest payments on governmentdebt, capital expenditures, tax revenues, nontaxrevenues, and grants. In order to avoid perfectcollinearity among regressors, the budget balance isnot included.13,14

Fiscal balance as share of GDP and expenditure compo-sition by economic category (Model B):

(2)

where gi,t and Yilt are defined as before and XBALEXPiht is a vector of independent fiscal vari-ables aimed at capturing the effect of the budget bal-ance and the composition of expenditures. The bud-get balance is measured as a percentage of GDP,while all expenditure items are measured as shares oftotal public expenditures. The expenditure cate-gories include public wages and salaries, publictransfers and subsidies, interest payments on govern-ment debt, public expenditures on other goods andservices, and public capital expenditures.

Source of deficit financing expressed as a share of GDPand expenditure composition by economic category(Model C):

(3)

where gi,t and Yilt are defined as before and XFINEXPiht is a vector of independent fiscal vari-ables aimed at capturing the effect of the deficit fi-nancing (both domestic and external financing as percentages of GDP), and the composition of expenditures as shares of total public expenditures.This specification is the same as the previous one,

g Y XGDP ui t l ilt h iht ith

q

l

k

, = + + +==

∑∑α β β11

12 Correlation coefficients are calculated using the Spearmanrank correlation formula to avoid the effect of outliers.

13 Theoretical models have generally incorporated the govern-ment budget constraint, which implies that a change in revenuesor spending of a given magnitude has to be matched by offsettingchanges elsewhere. This has not, however, been the approachtaken in the empirical literature. In many cases, applied studiesestimate the effect of selected expenditures and revenues ongrowth, which implicitly assumes that the effect of the excludeditems on growth is neutral. We avoid this by including all budgetitems in the specification. In this respect, we follow Kneller,Bleaney, and Gemmell (1999), who emphasize the need to in-clude all fiscal policy variables in the equations to avoid omittedvariables bias.

14 For example, adjustment based on selective increases in im-port tariff rates would most likely have a more adverse effect ongrowth than would raising revenues from a broad-based value-added tax (VAT).

g Y XBALEXP ui t l ilt h ihl ith

q

l

k

, = + + +==

∑∑α β β11

g Y XFINEXP ui t l ilt h iht ith

q

l

k

, = + + +==

∑∑α β β11

89


but it replaces the budget balance with its financingsources (expressed as ratios to GDP).

The baseline regressions use a fixed-effects esti-mator. The results are then tested for robustness byrunning a Generalized Method of Moments (GMM)estimator to address potential problems with endo-geneity and serial correlation arising from the dy-namic specification mentioned previously. A pooledmean-group (PMG) estimator is also used to capturethe effects of both short-run and long-run dynamicsand to relax the assumption of homogeneity ofshort-run coefficients. The relative merits of thesemethods are discussed in the respective subsections.

Baseline Regressions

The models above are estimated in levels and in firstdifferences (changes) in order to capture both long-and short-run effects of fiscal policy on growth. Analternative formulation of this model, involving anested specification in which both short-run andlong-run effects are estimated simultaneously, isfound in Bassanini, Scarpetta, and Hemmings(2001). This model could not be fully estimated inthe present context owing to the relatively shortlength of our sample. The results from a revised ver-sion of the Bassanini, Scarpetta, and Hemmingsmodel are discussed in the next subsection.

An important problem that is encountered inpanel data estimation is the presence of unobservedcountry-specific effects (Easterly, Loayza, and Montiel, 1997).15 Excluding unobservable country-specific effects could lead to serious biases in theeconometric estimates, notably when these effectsare correlated with the other covariates. To addressthis, we used a least-squares-dummy-variable (LSDV)estimator that allows the intercept in the regressionto be country-specific for the estimation of ModelsA, B, and C.16

Results from the baseline regressions (see Table 3)are consistent with the empirical literature andshow that, on average, fiscal adjustments have notbeen harmful for growth—in the long as well as theshort term. According to these results, a 1 percentimprovement in the fiscal balance has a positive andsignificant impact in the long term on the rate ofGDP growth, raising it by 1⁄2 of one percentage point(Model B). An even larger coefficient is estimatedfor the short-term effect of a change in the fiscalbalance on growth. The composition of deficit fi-nancing also matters. Domestic financing of thebudget tends to be more harmful for growth than ex-

ternal financing (Model C): in the long term, an in-crease in domestic financing by 1 percent reduces theper capita growth rate by 3⁄4 of a percentage point.The estimated coefficient for the short-term rela-tionship is even larger.

Expenditure composition is also critical forgrowth. In Model A, a one percentage point of GDPincrease in spending on wages and salaries reducesgrowth by 1⁄2 of a percentage point, while a one percentage point increase in the ratio of capital out-lays to GDP increases growth by more than 1⁄2 of apercentage point. Expenditures on other goods andservices are also found to increase the growth rate,but only in the short term. Interest payments have astatistically insignificant impact on growth. Finally,in the models that assess the impact of expenditurecomposition directly (Models B and C), the coeffi-cients for spending on wages are significant, butonly in the long run. The share of capital expendi-tures in total outlays is positively related to growthunder all model specifications, except for the long-run coefficients estimated in Model C. The resultssuggest that a 1 percent increase in the allocation ofpublic spending to capital outlays can raise thegrowth rate by 0.1 percentage point of GDP in thelong term and by almost 0.25 percentage point ofGDP in the short term. The share of public outlaysdevoted to the interest bill is also negatively corre-lated with growth. A 1 percent increase in the ratioof interest to total public spending tends to reducegrowth by 1⁄4 of one percentage point in the long runand by more than 1⁄3 of one percentage point in theshort run.

Sensitivity Analysis

In order to assess the sensitivity of the econometricresults presented above, this subsection reports themain results of the robustness analysis.

Reverse causality is not found to affect signifi-cantly the parameter estimates. A common issue inthe literature on fiscal policy and growth is thelikely presence of endogeneity or reverse causality. Itcould be the case that economic growth itself influ-ences fiscal variables. For example, when economicgrowth slows down, the ratio of government spend-ing to GDP is likely to increase if the nominal levelof expenditure is fixed, or if the revenue effort issensitive to cyclical developments. Moreover, somedegree of reverse causality could also be present inthe relationship between growth and investment.17

17 A related issue is whether the model fully captures the effectof the budget balance on growth, as the inclusion of private in-vestment (as an independent variable) de facto blocks the indi-rect effects of the budget deficit on growth via its effects on pri-vate investment. Estimates that omit private investment fromthe specification, however, do not lead to significantly differentresults, including for the fiscal balance. This assessment shouldbe viewed as preliminary, however, given the need to assess thedeficit-investment relationship in a model especially specified forthat purpose.

15 Unobservable time-specific effects are less common. In fact,following Greene (2000), when such effects do exist, it would bemore efficient to include an explicit linear or nonlinear timetrend in the equation.

16 Test for serial correlation for the three models revealed nofirst-order autocorrelation for the residuals.

The number of countries included in the regression varies ac-cording to the specification. On average, about 28 countries areincluded.

90


Table 3

Budget Composition and Growth in Low-Income Countries: Fixed Effects

Model A. Model B. Model C: Budget Composition Budget Balance Budget Financing

(as percentage and Composition and Composition of GDP) of Expenditures of Expenditures

Change in Change in Change in Real per real per Real per real per Real per real per

capita GDP capita GDP capita GDP capita GDP capita GDP capita GDP growth growth growth growth growth growth

Initial GDP per capita level1 0.205 0.203 –0.816 0.475 –0.314 0.332 (0.92) (0.10) (–0.39) (0.23) (–0.14) (0.16)

Labor force 0.837*** 2.894*** 0.618** 2.799*** 0.687** 2.329*** (2.88) (5.24) (2.21) (4.68) (2.33) (3.85)

Terms of trade –0.003 0.001 –0.005 –0.362 –0.005 –0.005(–0.52) (0.18) (–0.90) (–0.04) (–1.01) (–0.55)

Private investment 0.267* 0.279 0.396*** 0.582*** 0.391** 0.75***(1.77) (1.44) (2.75) (2.81) (2.28) (3.27)

Initial primary enrollment –0.136 –0.006 –0.159 –0.016 –0.211 –0.031 (–0.99) (–0.05) (–1.21) (–0.11) (–1.54) (–0.21)

Initial secondary enrollment 0.057 –0.048 0.162 –0.051 0.150 –0.034 (0.51) (–.037) (1.51) (–0.36) (1.33) (–0.25)

Budget balance (as percentage of GDP) 0.458*** 0.551***

(4.22) (3.39) Domestic financing(as percentage of GDP) –0.797*** –1.336***

(–5.08) (–5.94)External financing (as percentage of GDP) –0.383*** –0.595***

(–2.93) (–3.22) Wages and salaries (as percentage of GDP) –0.525* –0.396

(–1.78) (–0.87) Wages and salaries (as percentage of total expenditure) –0.213** –0.229 –0.250** –0.235

(–2.23) (–1.53) (–2.38) (–1.51) Transfers and subsidies (as percentage of GDP) 0.110 –0.424

(0.42) (–1.08) Transfers and subsidies (as percentage of total expenditure) 0.054 0.033 –0.047 –0.008

(0.49) (0.19) (–0.37) (–0.05)


91


Table 3 (concluded)

Model A. Model B. Model C: Budget Composition Budget Balance Budget Financing

(as percentage and Composition and Composition of GDP) of Expenditures of Expenditures

Change in Change in Change in Real per real per Real per real per Real per real per

capita GDP capita GDP capita GDP capita GDP capita GDP capita GDP growth growth growth growth growth growth

Interest payments (as percentage of GDP) –0.293 –0.367

(–0.90) (–0.73) Interest payments (as percentage of total expenditure) –0.118 –0.370*** –0.227* –0.415***

(–1.11) (2.20) (–1.92) (–2.35) Other goods and services (as percentage of GDP) 0.420 1.722***

(1.36) (3.96) Other goods and services (as percentage of total expenditure) 0.015 0.068 0.043 0.175

(0.16) (0.45) (0.44) (1.10) Capital expenditure (as percentage of GDP) 0.567*** 0.874***

(2.96) (3.52) Capital expenditure (as percentage of total expenditure) 0.154* 0.282** 0.072 0.237*

(1.96) (2.25) (0.82) (1.81) Tax revenue (as percentage of GDP) –0.056 0.053

(–0.29) (0.17) Nontax revenue (as percentage of GDP) 0.095 1.49***

(0.81) (2.63)Grants (as percentage of GDP) 0.079 0.209

(0.33) (0.71)

Number of observations 249 220 250 221 225 197 Adjusted R-squared 0.10 0.33 0.16 0.21 0.17 0.31 F test 1.77 3.86 2.41 2.62 2.30 3.50 Probability 0.00 0.00 0.00 0.00 0.00 0.00

Notes: t-statistics appear in parentheses; * denotes significance at 10 percent, ** significance at 5 percent, and *** significance at 1 percent..1 Multiplied by 100.

92


Tabl

e 4

Fisc

al P

olic

y, B

udge

t C

ompo

sitio

n, a

nd G

row

th in

Low

-Inc

ome

Cou

ntrie

s: C

ontr

ollin

g fo

r R

ever

se C

ausa

lity,

199

0–20

00

Mod

el A

. M

odel

B.

Mod

el C

: B

udge

t Com

posi

tion

Bud

get B

alan

ce a

nd

Bud

get F

inan

cing

and

(a

s pe

rcen

tage

of G

DP)

Com

posi

tion

of E

xpen

ditu

res

Com

posi

tion

of E

xpen

ditu

res

Real

Re

al

Real

Re

al

Real

Re

al

per

capi

tape

r ca

pita

Cha

nge

Cha

nge

per

capi

tape

r ca

pita

Cha

nge

Cha

nge

per

capi

tape

r ca

pita

Cha

nge

Cha

nge

grow

th

grow

th

grow

th

grow

th

grow

th

grow

th

grow

th

grow

th

grow

th

grow

th

grow

th

grow

th

(GM

M-

(GM

M-

(GM

M-

(GM

M-

(GM

M-

(GM

M-

(GM

M-

(GM

M-

(GM

M-

(GM

M-

(GM

M-

(GM

M-

IVR

EG)

ABon

d)IV

REG

)AB

ond)

IVR

EG)

ABon

d)IV

REG

)AB

ond)

IVR

EG)

ABon

d)IV

REG

)AB

ond)

Per

capi

ta g

row

th (

t–l)

–0.1

09**

–0.3

29**

*–0

.265

***

–0.3

71**

*–0

.229

***

–0.4

24**

*(–

2.05

)(–

9.24

)(–

21.7

8)(1

6.38

)(–

6.18

)(–

16.2

1)In

itial

GD

P pe

r ca

pita

leve

l10.

396*

0.03

9–0

.041

0.03

3–0

.061

–0.0

06(1

.96)

(0.3

1)(–

0.22

)(0

.34)

(–0.

32)

(–0.

06)

Labo

r fo

rce

0.84

7*1.

784*

**2.

89**

*2.

669*

**0.

750.

936*

**2.

120*

*2.

18**

0.80

2*1.

149*

**1.

76**

1.69

***

(1.7

4)(9

.60)

(3.1

2)(1

1.30

)(1

.60)

(4.3

2)(2

.03)

(16.

72)

(1.7

4)(5

.38)

(2.0

5)(5

.90)

Term

s of

trad

e–0

.003

0.00

3***

0.00

10.

004

–0.0

05**

*–0

.002

***

–0.0

010.

003

–0.0

05**

*–0

.004

***

–0.0

04–0

.001

(1.3

4)(3

.66)

(0.4

2)(1

.47)

(–2.

94)

(–2.

64)

(–0.

29)

(1.0

1)(–

3.18

)(–

3.38

)(–

0.73

)(–

0.62

)Pr

ivat

e in

vest

men

t0.

266*

0.10

60.

279*

0.16

70.

383*

**0.

393*

**0.

466*

*0.

642*

**0.

356*

*0.

760*

**0.

602*

*0.

533*

**(1

.69)

(1.4

9)(1

.69)

(1.3

0)(2

.84)

(5.2

8)(2

.29)

(5.9

7)(2

.36)

(3.9

0)(2

.26)

(3.1

4)In

itial

prim

ary

enro

llmen

t–0

.797

–0.0

68–0

.546

0.00

6–0

.519

0.07

2(–

1.80

)(–

0.89

)(–

1.27

)(0

.09)

(–1.

24)

(0.6

7)In

itial

sec

onda

ry e

nrol

lmen

t0.

346

–0.0

210.

352

–0.0

520.

335

–0.0

72(1

.45)

(–0.

23)

(1.5

8)(–

0.77

)(1

.51)

(–0.

87)

Bud

get b

alan

ce

(as

perc

enta

ge o

f GD

P)0.

435*

**0.

536*

**0.

462*

**0.

662*

**(4

.24)

(9.5

9)(2

.21)

(17.

35)

Dom

estic

fina

ncin

g (a

s pe

rcen

tage

of G

DP)

–0.6

06**

*–0

.929

***

–0.9

77**

*–0

.991

***

(–3.

06)

(–6.

80)

(2.7

3)(–

6.79

)Ex

tern

al fi

nanc

ing

(as

perc

enta

ge o

f GD

P)–0

.415

***

–0.4

59**

*–0

.511

**–0

.605

***

(–3.

54)

(–7.

07)

(–2.

57)

(–8.

86)

Wag

es a

nd s

alar

ies

(a

s pe

rcen

tage

of G

DP)

–0.5

21–0

.385

*–0

.396

–0.5

11*

(–0.

86)

(–1.

89)

(–0.

58)

(–1.

92)

Wag

es a

nd s

alar

ies

(as

perc

enta

ge o

f to

tal e

xpen

ditu

re)

–0.2

14*

–0.3

50**

*–0

.244

*–0

.263

***

–0.2

20*

–0.2

80**

*–0

.255

*–0

.344

***

(–1.

85)

(–7.

09)

(–1.

76)

(–6.

26)

(–1.

65)

(–3.

58)

(1.8

3)(–

3.14

)Tr

ansf

ers

and

subs

idie

s (a

s pe

rcen

tage

of G

DP)

0.10

9–0

.059

–0.4

24–0

.528

***

(0.4

6)(–

0.50

)(–

0.58

)Tr

ansf

ers

and

subs

idie

s (a

s pe

rcen

tage

of

tota

l exp

endi

ture

)0.

081

–0.0

04–0

.001

–0.0

050.

039

–0.0

18–0

.070

0.23

0***

(0.9

2)(–

0.09

)(0

.00)

(–0.

09)

(0.4

0)(–

0.19

)(–

0.21

)(3

.00)

(Tab

le c

ontin

ues

on n

ext p

age)

93


Tabl

e 4

(con

clud

ed)

Mod

el A

. M

odel

B.

Mod

el C

: B

udge

t Com

posi

tion

Bud

get B

alan

ce a

nd

Bud

get F

inan

cing

and

(a

s pe

rcen

tage

of G

DP)

Com

posi

tion

of E

xpen

ditu

res

Com

posi

tion

of E

xpen

ditu

res

Real

Real

Real

Re

alRe

alRe

alpe

r ca

pita

per

capi

ta

Cha

nge

Cha

nge

per

capi

ta

per

capi

ta

Cha

nge

Cha

nge

per

capi

ta

per

capi

ta

Cha

nge

Cha

nge

grow

th

grow

th

grow

th

grow

th

grow

th

grow

th

grow

th

grow

th

grow

th

grow

th

grow

th

grow

th

(GM

M-

(GM

M-

(GM

M-

(GM

M-

(GM

M-

(GM

M-

(GM

M-

(GM

M-

(GM

M-

(GM

M-

(GM

M-

(GM

M-

IVR

EG)

ABon

d)IV

REG

)AB

ond)

IVR

EG)

ABon

d)IV

REG

)AB

ond)

IVR

EG)

ABon

d)IV

REG

)AB

ond)

Inte

rest

pay

men

ts

(as

perc

enta

ge o

f GD

P)–0

.298

–0.1

88–0

.367

0.00

5(–

0.74

)(–

1.63

)(–

0.87

)(0

.03)

Inte

rest

pay

men

ts

(as

perc

enta

ge o

f to

tal e

xpen

ditu

re)

–0.1

39–0

.248

***

–0.3

03–0

.379

***

–0.2

18–0

.194

**–0

.314

*–0

.412

***

(–1.

09)

(–4.

42)

(–1.

59)

(–6.

17)

(–1.

47)

(–2.

08)

(–1.

67)

(–5.

16)

Oth

er g

oods

and

ser

vice

s (a

s pe

rcen

tage

of G

DP)

0.40

01.

417*

**1.

72**

*0.

932*

**(0

.95)

(6.2

6)(3

.05)

(4.1

2)O

ther

goo

ds a

nd s

ervi

ces

(as

perc

enta

ge o

f to

tal e

xpen

ditu

re)

–0.0

020.

129*

*0.

048

0.03

50.

003

0.14

9**

0.10

90.

064

(–0.

03)

(2.2

8)(0

.40)

(0.8

7)(0

.05)

(2.4

4)(0

.89)

(0.7

7)C

apita

l exp

endi

ture

(a

s pe

rcen

tage

of G

DP)

0.55

20.

710*

**0.

874*

*0.

735*

**(1

.57)

(7.1

0)(2

.41)

(5.4

8)C

apita

l exp

endi

ture

(a

s pe

rcen

tage

of

tota

l exp

endi

ture

)0.

116

0.18

5***

0.19

40.

358*

**0.

071

0.18

0***

0.14

50.

293*

**(1

.35)

(5.0

3)(1

.38)

(10.

35)

(0.6

9)(3

.69)

(1.0

0)(4

.44)

Tax

reve

nue

(as

perc

enta

ge o

f GD

P)–0

.043

0.06

70.

053

0.05

9(–

0.18

)(1

.04)

(0.2

0)(0

.31)

Non

tax

reve

nue

(as

perc

enta

ge o

f GD

P)0.

104

1.51

8***

1.49

**1.

44**

* (0

.27)

(5.0

1)(2

.35)

(3.8

8)

Gra

nts

(as

perc

enta

ge o

f GD

P)0.

090

0.18

50.

209

0.32

8***

(0.2

5)(0

.97)

(0.6

1)(5

.10)

Num

ber

of o

bser

vatio

ns24

920

122

017

225

020

122

117

222

517

919

715

1Re

gres

sion

test

s—W

ald

χ2(1

2)—

3,64

4.13

—18

,158

.87

—26

,400

.36

—18

,255

.63

—7,

060.

64—

108,

082.

74

Not

es: A

bsol

ute

valu

e of

tan

d z

stat

istic

s ap

pear

in p

aren

thes

es; *

den

otes

sig

nific

ance

at 1

0 pe

rcen

t, **

sig

nific

ance

at 5

per

cent

, and

***

sig

nific

ance

at 1

per

cent

. Mod

el A

: Han

sen-

JTe

st:

(GM

M 0

-leve

ls)=

0.01

; (G

MM

0-c

hang

es)=

0.00

; ser

ial o

f ove

r 2:

(G

MM

-AB

ond-

leve

ls)=

0.41

; (G

MM

-AB

ond-

chan

ges)

=0.

25. M

odel

B: H

anse

n-J

Test

: (G

MM

0-le

vels

)=0.

62; (

GM

M 0

-cha

nges

)=1.

32;

seria

l of o

ver

2: (

GM

M-A

Bon

d-le

vels

)=0.

08; (

GM

M-A

Bon

d-ch

ange

s)=

0.21

. Mod

el C

: Han

sen-

JTe

st: (

GM

M 0

-leve

ls )

=1.

04; (

GM

M 0

-cha

nges

)=2.

32; s

eria

l of o

ver

2: (

GM

M-A

Bon

d-le

vels

)=0.

12;

(GM

M-A

Bon

d-ch

ange

s)=

0.10

.1

Mul

tiplie

d by

100

.

94


If economic growth is a determinant of any of theright-hand-side variables in our model, estimationtechniques that do not take into account this endo-geneity will yield biased and inconsistent parameterestimates. To address this concern, we estimate theprevious models using a GMM estimator,18 instru-menting for the investment rate, fiscal balanceratio, and the shares of government spending andrevenues to GDP. We use as instruments the laggedvalues of these variables, the other exogenous vari-ables in the model, and a set of instruments not in-cluded in the model.19 Results are presented in Table4 and broadly confirm the findings of the previoussubsection. Accounting for the endogeneity of fiscalbalances leads to the same positive effect of fiscalconsolidation on growth as in the baseline regres-sions. A difference in the results, however, is that thecoefficient for the share of wages and salaries becomesinsignificant in Model A, although it remains signifi-cant and negatively correlated with growth in the re-maining specifications. The short-run effect of capitaloutlays on growth is not affected by the use of theGMM estimator; however, the long-run coefficientturns insignificant. The specification in the preced-ing regression does not allow for any dynamics be-tween the dependent and independent variables.Growth relationships are dynamic, however, asgrowth in a given period is not unconnected withpast growth trends. If the true model is not static,parameter estimates based on a static fixed-effectsestimator are biased and inconsistent, even whenthe error terms are not serially correlated. Thus, weestimated Models A, B, and C using unobservedcountry-specific effects and allowed for the laggedgrowth rate to be included among the determinantsof economic growth. These models can be esti-mated using the GMM estimator proposed by Arel-lano and Bond (1991). The GMM estimate alsocontrols for endogeneity by using the lagged valuesof the levels of the endogenous and the predeter-mined variables as instruments. Both the validity ofthe instruments and the presence of serial correla-tion in the residual, which would eliminate theconsistency of the estimator, can be tested once theequation is estimated.

Introducing a dynamic specification does not leadto significantly different results from the baseline,while it improves the results compared with the sta-tic GMM estimator. GMM estimates of the dynamicmodel with country-specific effects are reported inTable 4. The results are, in general, consistent with

the static fixed-effects estimates presented in theprevious subsection. The effect of fiscal consolida-tion on growth is larger and more significant thanunder the GMM and LSDV estimates of the staticmodel. The contributions of capital outlays andgovernment spending on wages are still correctlysigned and statistically significant, and in most caseslarger in size than in the baseline and GMM regres-sions. The negative effect on growth of an increasein domestic financing is larger, while the effect ofexternal financing of the deficit is broadly un-changed. The coefficient of the lagged dependentvariable is negative and significant, as expected,20

for all models. Finally, both the Sargan test for thevalidity of instruments and the test for the serialcorrelation of residuals confirm that GMM providesconsistent estimates of the parameters.

A variety of other estimators were utilized to test therobustness of the results, including the generalized-least-squares (GLS) estimate of the random effectmodel. The results confirm the main findings of theprevious subsection.21 Results are also consistentwith these estimates when we use a robust tech-nique to control for the possible presence of outliersin the data. The method is based on an iterative al-gorithm that first runs ordinary-least-squares (OLS)estimates and calculates22 Cook’s D statistics for theresiduals, eliminating those observations for whichD>1. The second step of the algorithm is to run aregression on the new dataset, and calculate caseweights based on the inverse of the residual.23 Theresults show that the effect of outliers in our data isnot substantial.24

A further robustness test was carried out by repli-cating a modified version of the model used by Bassanini, Scarpetta, and Hemmings (2001). Thisspecification tries to capture the effect of the simul-taneous inclusion of both short-run and long-run ef-fects of fiscal variables in the growth equation usingthe PMG estimator. We were not able to fully repli-cate the nested specification used by Bassanini,Scarpetta, and Hemmings, given the short time

18 The GMM estimator used here deals with a heteroskedasticerror process. This estimator is more efficient than the tradi-tional instrumental variables estimator.

19 The instruments include total revenue, current governmentspending, and total government spending, all as ratios to GDP.All instruments were found to be valid according to the Hansen-Sargan test.

20 A negative coefficient for the lagged growth rate can be in-terpreted as the tendency of the annual growth rate to convergetoward an average long-run trend. Countries would still tend toward different, specific growth rates as a result of the error-component structure in the equation.

21 These and other results not reported in the paper are avail-able from the authors upon request.

22 Although most results are consistent with the baseline re-gression for Models A and B, in the case of Model C the coeffi-cient of the level of domestic financing is not statistically signifi-cant. For the majority of the short-run coefficients, the variablesare significant and correctly signed.

23 For a full description of this procedure, see Hamilton (1991).24 Although most results are consistent with the baseline re-

gression for Models A and B, in the case of Model C, the coeffi-cient of the level of domestic financing is not statistically signifi-cant. For the majority of the short-run coefficients, the variablesare significant and correctly signed.

95


Table 5

Budget Composition and Growth in Low-Income Countries: Nested Models with Fixed Effects1

(Dependent variable: real per capita GDP growth)

Model A. Model B. Model C: Budget Budget Balance Budget Financing

Composition and Composition and Composition(percent of GDP) of Expenditures of Expenditures

Initial GDP per capita level2 0.003 –0.001 –0.001(1.07) (–0.38) (–0.43)

Labor force 0.510* 0.443 0.413(1.77) (1.48) (1.06)

Terms of trade –0.006 –0.006 –0.007(–1.00) (–0.93) (–0.92)

Private investment –0.026 0.343* 0.452*(–0.15) (1.91) (1.80)

Initial primary enrollment –0.139 –0.130 –0.139(–0.36) (–0.91) (–0.67)

Initial secondary enrollment 0.020 0.133 0.135(0.10) (1.19) (0.94)

Budget balance (percent of GDP) 0.348**(2.54)

Domestic financing (percent of GDP) –0.521*(–1.79)

External financing (percent of GDP) –0.378(–1.63)

Wages and salaries (percent of GDP) –0.833***(–2.82)

Wages and salaries (percent of total expenditure) –0.196* –0.145 (–1.79) (–0.96)

Transfers and subsidies (percent of GDP) 0.287(0.76)

Transfers and subsidies (percent of total expenditure) 0.063 0.084(0.50) (0.50)

Interest payments (percent of GDP) –0.023(–0.07)

Interest payments (percent of total expenditure) –0.159 –0.138(–1.33) (–0.83)

Other goods and services (percent of GDP) 0.424(1.28)

Other goods and services (percent of total expenditure) 0.032 –0.016(0.27) (–0.10)

Capital expenditure (percent of GDP) 0.214(1.05)

Capital expenditure (percent of total expenditure) 0.172* 0.213*(1.96) (1.71)

Tax revenue (percent of GDP) –0.300(–1.31)

Nontax revenue (percent of GDP) –0.049(–0.12)

Grants (percent of GDP) 0.041(0.15)

Number of observations 229 230 201Adjusted R-squared 0.68 0.30 0.17

F test 3.07 2.53 1.47Probability 0.00 0.00 0.02

Notes: t-statistics appear in parentheses. * significant at 10 percent; ** significant at 5 percent; and *** significant at 1 percent.1 The regressions includes the following variables, denoted in changes, and their interaction with country dummies: total expenditure and

total revenue (Model A); deficit (Model B); and domestic and external financing (Model C).2 Multiplied by 100.

dimension of our sample. Instead, we included themost important fiscal variables in first differences25

in the level specification of the three models, andallowed their coefficients to be country-specific toaccount for differentiated short-term responses ofgrowth to fiscal policy. We estimated this modelusing a fixed-effects estimator. The results confirmthe stability of the fiscal coefficients estimated inthe baseline regressions (Table 5). The negative ef-fect of fiscal deficits on growth is confirmed by theseestimates. In Model B, the long-run coefficient ofthe fiscal balance is significantly and positivelysigned, but smaller than the corresponding coeffi-cient in the baseline regression. Similar significantand consistently signed coefficients are found forthe share of wages in total government spendingand the ratio of capital to total public outlays. Re-sults for Model A show a much larger and signifi-cant negative effect of the wage bill on growth. Theratio of capital spending to GDP, however, becomesinsignificant in this specification. Model C alsoconfirms the main findings of the baseline model. Inthis model, however, domestic financing is found tobe detrimental to growth but external financingdoes not significantly affect growth.

Finally, results do not change much when the pos-sible effects of the business cycle and time trends areremoved from the data. The possible effects of thebusiness cycle are partially eliminated by smoothingthe data using a three-year moving-average filter.Once again, the results are not sensitive to thistransformation of the original data. The reason whybusiness cycle effects may be weaker in low-incomecountries than in the industrial countries is the ab-sence of automatic stabilizers. This feature makes ithighly unlikely that business cycles affect tax collec-tion or public expenditures, and thus the overallbudget balance. Moreover, in our sample, we do notfind sufficient evidence that unobservable time ef-fects are a serious problem, as evidenced by the re-sults for regressions that include time dummies tocontrol for nonlinear time trends in the data.

Nonlinear Effects of Fiscal Policy on Growth:Pre- and Post-Stabilization Countries

The results in the previous section suggest that fiscalconsolidation is not harmful for growth in low-in-come countries. Quality fiscal adjustments based onthe reallocation of public expenditure to more pro-ductive uses, and the reduction of the budget deficit,were found to be conducive to higher growth. Of in-terest is whether these results hold for all countriesin the sample, in particular for countries that havealready achieved a modicum of macroeconomic sta-

bility (i.e., “post-stabilization” countries (Adam andBevan, 2001)).

With the purpose of assessing the effect of initialfiscal conditions on the fiscal-policy-growth nexus,we split the sample into post- and pre-stabilizationcountries. A post-stabilization country is defined asa country that maintained an average fiscal deficit(after grants) below 2.5 percent of GDP during theperiod 1990–2000.26

Results for post-stabilization countries point tothe positive effects of capital outlays and selectedcurrent expenditures on growth. Econometric re-sults for the two subgroups are reported in Table 6using LSDV.27 Interestingly, the results suggest thatfor countries with low budget deficits, additional fis-cal consolidation may not yield higher growth. Evenmore important, domestic financing is not harmfulfor growth in the short run and less harmful than ex-ternal financing in the long run in these countries,unlike the case of countries that have not yetachieved stabilization. The results should be inter-preted with caution, though, in view of the smallsample size for post-stabilization countries and thepoor performance of some of these models in termsof F tests. Nevertheless, the results support the no-tion that the relationship between budget deficitsand growth in these countries differs from that forthe sample as a whole. Results for pre-stabilizationcountries are fully consistent with the “expansion-ary contractions” thesis.

Conclusions and Policy Implications

The empirical evidence provided in this study sug-gests that in low-income countries fiscal consolida-tions were not harmful for long- or short-termgrowth during the period 1990–2000. This papersought to shed light on the relationship between fis-cal adjustment, expenditure composition, and eco-nomic growth in low-income countries. Consistentwith the previous findings in the literature on indus-trial countries, the results point to a significant

96


26 The criterion used to group the countries in the sample issimilar to the one used in a study of ESAF-supported programsfor 1986–95 (see Abed and others, 1998), where “low initialdeficit” countries were defined as those with initial deficits (be-fore grants) of 5 percent, with grants of approximately 21⁄2 percentof GDP. Post-stabilization countries included in the regressionsare The Gambia, Lesotho, the former Yugoslav Republic ofMacedonia, Mauritania, Senegal, and Tanzania. Benin is the sev-enth post-stabilization country but is excluded because data onthe other control variables are unavailable.

We also applied the same threshold to the 1996–2000 averagesin order to test the robustness of our results to alternative defini-tions of “post-stabilization.” The results confirm the findings ofthis subsection.

27 Results for Model A are reported in Table 6. Results forModels A–C were also replicated using the LSDV estimator,which confirmed these findings. These results are not included inthe paper for the sake of brevity but are available from the au-thors upon request.

25 In Model A, we use the first difference of total governmentspending and total revenues as a share of GDP. In Models B andC, we use the fiscal balance and domestic and external deficit fi-nancing, respectively.

97

Gupta, Clements, Baldacci, and Mulas-GranadosTa

ble

6

Fisc

al P

olic

y, B

udge

t C

ompo

sitio

n, a

nd G

row

th in

Low

-Inc

ome

Cou

ntrie

s (P

re-

and

Post

-Sta

biliz

atio

n C

ount

ries)

: Fi

xed

Effe

cts,

199

0–20

00

Mod

el B

. M

odel

C:

Mod

el A

.B

udge

t Bal

ance

and

B

udge

t Fin

anci

ng a

nd

Bud

get C

ompo

sitio

nC

ompo

sitio

n of

Exp

endi

ture

sC

ompo

sitio

n of

Exp

endi

ture

s

Real

per

cap

itaC

hang

e in

per

cap

itaRe

al p

er c

apita

Cha

nge

in p

er c

apita

Real

per

cap

itaC

hang

e in

per

cap

itaG

DP

grow

thG

DP

grow

thG

DP

grow

thG

DP

grow

thG

DP

grow

thG

DP

grow

th

Pre-

Post

-Pr

e-Po

st-

Pre-

Post

-Pr

e-Po

st-

Pre-

Post

-Pr

e-Po

st-

stab

iliza

tion

stab

iliza

tion

stab

iliza

tion

stab

iliza

tion

stab

iliza

tion

stab

iliza

tion

stab

iliza

tion

stab

iliza

tion

stab

iliza

tion

stab

iliza

tion

stab

iliza

tion

stab

iliza

tion

coun

trie

sco

untr

ies

coun

trie

sco

untr

ies

coun

trie

sco

untr

ies

coun

trie

sco

untr

ies

coun

trie

sco

untr

ies

coun

trie

sco

untr

ies

Initi

al G

DP

per

capi

ta le

vel1

1.46

6–1

.423

–0.0

18–0

.012

1.01

0–3

.402

0.29

0–1

.483

0.01

3–9

.726

0.42

2–0

.132

(1.5

5)(–

0.41

)(–

0.02

)(–

0.05

)(1

.21)

(–1.

18)

(0.3

1)(–

0.61

)(1

.51)

(–2.

75)

(0.4

8)(–

0.55

)La

bor

forc

e0.

833*

*0.

967

2.79

2***

0.21

80.

571*

1.68

2.56

***

0.60

50.

601*

4.59

1***

2.10

***

4.50

(2.6

2)(0

.63)

(4.7

3)(0

.11)

(1.8

7)(1

.29)

(4.0

3)(0

.29)

(1.9

1)(2

.83)

(3.3

2)(1

.52)

Term

s of

trad

e–0

.004

–0.0

040.

002

0.04

5–0

.006

0.03

3–0

.001

0.30

–0.0

070.

030

–0.0

060.

043

(–0.

63)

(–0.

09)

(0.2

3)(0

.73)

(–1.

00)

(0.8

6)(–

0.09

)(0

.50)

(–1.

13)

(0.7

9)(–

0.65

)(0

.70)

Priv

ate

inve

stm

ent

0.60

0***

–0.2

07*

0.51

8**

–0.2

85**

0.65

0***

–0.1

92*

0.99

0***

–0.3

67**

*0.

695*

**–0

.114

1.22

***

–0.3

46**

(2.8

3)(–

2.01

)(2

.01)

(–2.

34)

(3.2

9)(–

1.91

)(3

.71)

(–2.

91)

(2.9

5)(–

0.96

)(4

.12)

(–2.

48)

Initi

al p

rimar

y en

rollm

ent

0.05

2–0

.179

–0.0

09–0

.015

–0.0

10–0

.55

0.01

97–0

.031

0.01

8–1

.638

**–0

.003

–0.0

55(0

.40)

(–0.

31)

(–0.

10)

(–0.

30)

(–0.

09)

(–1.

12)

(0.2

1)(–

0.59

)(0

.15)

(–2.

73)

(–0.

04)

(–1.

03)

Initi

al s

econ

dary

enr

ollm

ent

–0.4

691.

463

–0.0

450.

050

–0.2

733.

21–0

.151

0.15

7–0

.527

9.15

4**

–0.1

160.

153

(–1.

57)

(0.4

4)(–

0.16

)(0

.23)

(–1.

03)

(1.1

8)(_

0.48

)(0

.69)

(–1.

54)

(2.7

2)(–

0.36

)(0

.69)

Bud

get b

alan

ce

(per

cent

age

of G

DP)

0.56

3***

0.11

80.

804*

**–0

.043

(4.1

6)(1

.41)

(4.1

1)(–

0.36

)D

omes

tic fi

nanc

ing

(per

cent

age

of G

DP)

–1.0

9***

–0.2

06*

–1.5

7***

0.14

1(–

5.36

)(–

1.98

)(–

6.21

)(0

.75)

Exte

rnal

fina

ncin

g (p

erce

ntag

e of

GD

P)–0

.44*

**–0

.364

***

–0.7

23**

*–0

.108

(–2.

81)

(–3.

02)

(–3.

31)

(–0.

70)

Wag

es a

nd s

alar

ies

(per

cent

age

of G

DP)

–0.3

05–0

.768

*–0

.896

*–6

.01

(–0.

86)

(–1.

91)

(–1.

74)

(–0.

82)

Wag

es a

nd s

alar

ies

(per

cent

age

of

tota

l exp

endi

ture

)–0

.233

*0.

019

–0.3

00*

–0.0

64–0

.246

*0.

033

–0.2

830.

002

(–1.

97)

(0.2

5)(–

1.68

)(–

0.52

)(–

1.98

)(0

.28)

(–1.

58)

(0.0

2)Tr

ansf

ers

and

subs

idie

s (p

erce

ntag

e of

GD

P)–0

.039

0.41

1–0

.985

**0.

862*

*(–

0.11

)(1

.28)

(–2.

08)

(2.3

7)Tr

ansf

ers

and

subs

idie

s (p

erce

ntag

e of

to

tal e

xpen

ditu

re)

–0.0

100.

338*

**–0

.110

0.30

6**

–0.1

590.

527*

**–0

.188

0.42

9**

(–0.

08)

(3.3

6)(–

0.53

)(2

.11)

(–1.

06)

(4.4

5)(–

0.84

)(2

.76)

(Tab

le c

ontin

ues

on n

ext p

age)

98


Tabl

e 6

(con

clud

ed)

Mod

el B

. M

odel

C:

Mod

el A

.B

udge

t Bal

ance

and

B

udge

t Fin

anci

ng a

nd

Bud

get C

ompo

sitio

nC

ompo

sitio

n of

Exp

endi

ture

sC

ompo

sitio

n of

Exp

endi

ture

s

Real

per

cap

itaC

hang

e pe

r ca

pita

Real

per

cap

itaC

hang

e pe

r ca

pita

Real

per

cap

itaC

hang

e pe

r ca

pita

GD

P gr

owth

GD

P gr

owth

GD

P gr

owth

GD

P gr

owth

GD

P gr

owth

GD

P gr

owth

Pre-

Post

-Pr

e-Po

st-

Pre-

Post

-Pr

e-Po

st-

Pre-

Post

-Pr

e-Po

st-

stab

iliza

tion

stab

iliza

tion

stab

iliza

tion

stab

iliza

tion

stab

iliza

tion

stab

iliza

tion

stab

iliza

tion

stab

iliza

tion

stab

iliza

tion

stab

iliza

tion

stab

iliza

tion

stab

iliza

tion

coun

trie

sco

untr

ies

coun

trie

sco

untr

ies

coun

trie

sco

untr

ies

coun

trie

sco

untr

ies

coun

trie

sco

untr

ies

coun

trie

sco

untr

ies

Inte

rest

pay

men

ts

(per

cent

age

of G

DP)

–0.5

38–0

.756

–0.5

260.

755

(–1.

44)

(–1.

02)

(–0.

97)

(0.7

4)In

tere

st p

aym

ents

(p

erce

ntag

e of

to

tal e

xpen

ditu

re)

–0.1

500.

290*

–0.4

03**

0.44

4*–0

.260

**0.

277

–0.4

54**

0.31

3(–

1.25

)(1

.79)

(–2.

20)

(1.7

2)(–

2.00

)(1

.61)

(–2.

41)

(1.0

7)O

ther

goo

ds a

nd s

ervi

ces

(per

cent

age

of G

DP)

0.66

70.

655*

*1.

932*

**0.

711*

(1.6

3)(2

.33)

(3.5

9)(2

.01)

Oth

er g

oods

and

ser

vice

s (p

erce

ntag

e of

to

tal e

xpen

ditu

re)

0.03

70.

142*

0.04

10.

197*

0.09

00.

060

0.13

50.

125

(0.3

1)(1

.80)

(0.2

2)(1

.87)

(0.7

3)(0

.67)

(0.7

2)(1

.06)

Cap

ital e

xpen

ditu

re

(per

cent

age

of G

DP)

0.68

2***

–0.0

371.

327*

**–0

.032

(3.0

0)(–

0.18

)(4

.32)

(–0.

18)

Cap

ital e

xpen

ditu

re

(per

cent

age

of

tota

l exp

endi

ture

)0.

148

0.13

9*0.

327*

*0.

171*

0.04

40.

248*

**0.

221

0.26

0**

(1.5

6)(1

.98)

(2.2

1)(1

.86)

(0.4

4)(3

.15)

(1.4

8)(2

.65)

Tax

reve

nue

(per

cent

age

of G

DP)

–0.0

57–0

.099

0.23

8–0

.312

(–0.

21)

(–0.

48)

(0.6

0)(–

1.45

)N

onta

x re

venu

e (p

erce

ntag

e of

GD

P)–0

.387

0.52

4**

1.30

6*0.

479

(–0.

58)

(2.0

8)(1

.85)

(1.1

6)G

rant

s (p

erce

ntag

e of

GD

P)–0

.100

0.48

7–0

.099

0.55

4(–

0.35

)(0

.85)

(–0.

31)

(0.9

4)N

umbe

r of

obs

erva

tions

200

4917

743

201

4917

843

184

4116

136

Adju

sted

R-s

quar

ed0.

130.

160.

390.

140.

200.

330.

280.

130.

220.

510.

380.

30F

test

1.97

1.59

4.64

1.45

2.71

2.74

3.37

1.48

2.69

4.08

4.29

2.07

Prob

abili

ty0.

000.

120.

000.

190.

000.

000.

000.

180.

000.

000.

000.

06

Not

es: T

he t-

stat

istic

s ap

pear

in p

aren

thes

es.;

* de

note

s si

gnifi

canc

e at

10

perc

ent,

** s

igni

fican

ce a

t 5 p

erce

nt, a

nd *

** s

igni

fican

ce a

t 1 p

erce

nt.

1 M

ultip

lied

by 1

00.

relationship between fiscal adjustment and percapita growth. A reduction of 1 percentage point inthe ratio of the fiscal deficit to GDP leads to an av-erage increase in per capita growth of 1⁄2 of a percent-age point both in the long term and in the shortterm. This implies that a reduction in the averagedeficit in low-income countries from about 4 per-cent of GDP to 2 percent of GDP could boost percapita growth by about 1–2 percentage points a year.

Tilting the overall composition of public expendi-ture toward more productive uses is particularly im-portant for boosting growth. Fiscal consolidationsachieved through cutting selected current expendi-tures tend to trigger higher growth rates than ad-justments based on revenue increases and cuts inmore productive spending—a result consistent withthe findings for industrial countries. According tothe results of our analysis, protecting capital expen-ditures during a fiscal adjustment leads to highergrowth. Reductions in the public sector wage billare not harmful for growth for the sample as awhole.

The composition of deficit financing is also a keyfactor affecting growth in low-income countries. Fiscal consolidations, especially those leading to asizable reduction in domestic financing of the deficitare likely to trigger higher growth rates. The empiri-cal estimates indicate that adjustments based on re-ducing domestic financing have about 11⁄2 times theeffect on growth as adjustments based on reductionsin both domestic and external financing.

The effects of fiscal policy on growth tend to benonlinear. The results above hold for countries thathave not yet achieved stable macroeconomic condi-tions. In post-stabilization countries, fiscal adjust-ments no longer have a salutary effect on growth. Inthis context, an expansion of selected current ex-penditures for these countries is compatible withhigher growth. The design of fiscal frameworks inPRGF-supported programs is consistent with theseresults, as post-stabilization countries target rela-tively larger increases in public spending and in thefiscal deficit (IMF, 2002).

Additional research is needed to disentangle thechannels through which fiscal policy affects growth.Given the reduced-form model tested here, thepaper has not examined the demand- and supply-side channels through which fiscal policy affectsgrowth, nor the role of accompanying policies (suchas monetary and external sector policies) that havebeen underscored in previous work in this field (Bal-dacci and others, 2001; and Thomas, 2001). Addi-tional research is needed in this area.

ReferencesAbed, George, and others, 1998, Fiscal Reform in Low-

Income Countries: Experience Under IMF-SupportedPrograms, IMF Occasional Paper No. 160 (Washing-ton: International Monetary Fund).

Adam, Christopher S., and David L. Bevan, 2000, “FiscalPolicy Design in Low-Income Countries,” paper pre-pared for UNU/WIDER (United Nations Univer-sity/World Institute for Development Economics Research) research project on “New Fiscal Policiesfor Poverty Reduction and Growth” (unpublished,November).

———, 2001, “Nonlinear Effects of Public Deficits onGrowth,” paper prepared for the Cornell/ISPE (Inter-national Seminar in Public Economics) Conference,New York, September.

Alesina, Alberto, and Silvia Ardagna, 1998, “Tales of Fis-cal Adjustment,” Economic Policy: A EuropeanForum, No. 27, pp. 487–546.

———, Roberto Perotti, and Fabio Schiantarelli, 2002,“Fiscal Policy, Profits, and Investment,” AmericanEconomic Review, Vol. 92, pp. 571–89.

Alesina, Alberto, and Roberto Perotti, 1995, “Fiscal Ex-pansion and Fiscal Adjustments in OECD Coun-tries,” Economic Policy, Vol. 21, pp. 205–48.

———, 1996, “Fiscal Adjustments in OECD Countries:Composition and Macroeconomic Effects,” IMFWorking Paper 96/70 (Washington: InternationalMonetary Fund).

———, and Jose Tavares, 1998, “The Political Economyof Fiscal Adjustments,” Brookings Papers on EconomicActivity: 2, Brookings Institution, pp. 192–248.

Arellano, Manuel, and Stephen Bond, 1991, “Some Testsof Specific Action for Panel Data: Monte Carlo Evi-dence and an Application to Employment Equa-tions,” Review of Economic Studies, Vol. 58, pp. 277–97.

Aschauer, David A., 1989, “Is Public Expenditure Produc-tive?” Journal of Monetary Economics, Vol. 23,pp. 177–200.

Baldacci, Emanuele, Marco Cangiano, Selma Mahfouz,and Axel Schimmelpfennig, 2001, “The Effective-ness of Fiscal Policy in Stimulating Economic Activ-ity: An Empirical Investigation,” paper presented atthe IMF Annual Research Conference (Washington:International Monetary Fund).

Barro, Robert J., 1990, “Government Spending in a Sim-ple Model of Endogenous Growth,” Journal of PoliticalEconomy, Vol. 98, No. 1, pp. 103–17.

———, 1991, “Economic Growth in a Cross Section ofCountries,” Quarterly Journal of Economics, Vol. 106,pp. 407–43.

———, and Xavier Sala-i-Martin, 1995, EconomicGrowth (New York: McGraw-Hill).

Bassanini, Andrea, Stefano Scarpetta, and Phillip Hem-mings, 2001, “Economic Growth: The Role of Poli-cies and Institutions. Panel Data Evidence fromOECD Countries,” Economics Department WorkingPapers, No. 283 (Paris: Organization for EconomicCooperation and Development).

Buti, Marco, and Andre Sapir, 1998, Economic Policy inEMU (Oxford, United Kingdom: Oxford UniversityPress).

Chamley, C., 1986, “Optimal Taxation of Capital IncomeEquilibrium with Infinite Lives,” Econometrica, Vol. 54,No. 3, pp. 607–22.

Chu, Ke-young, and others, 1995, Unproductive Public Ex-penditure: A Pragmatic Approach to Policy Analysis,IMF Pamphlet Series, No. 48 (Washington: Interna-tional Monetary Fund).

Devarajan, Shantayanan, Vinaya Swaroop, and Heng-fuZou, 1996, “The Composition of Public Expenditure

99


100


and Economic Growth,” Journal of Monetary Econom-ics, Vol. 37, pp. 313–44.

Easterly, William, Norman Loayza, and Peter Montiel,1997, “Has Latin America’s Post-Reform GrowthBeen Disappointing?” Journal of International Econom-ics, Vol. 43, pp. 287–311.

Easterly, William, and Sergio Rebelo, 1993, “Fiscal Policyand Economic Growth: An Empirical Investigation,”NBER Working Paper No. 4499 (Cambridge, Massa-chusetts: National Bureau of Economic Research).

Easterly, William, Carlos A. Rodríguez, and KlausSchmidt-Hebbel, 1994, Public Sector Deficits andMacroeconomic Performance (Washington: WorldBank).

Fischer, Stanley, 1983, “Inflation and Growth,” NBERWorking Paper No. 1235 (Cambridge, Massachu-setts: National Bureau of Economic Research).

Gerson, Philip, 1998, “The Impact of Fiscal Policy Vari-ables on Output Growth,” IMF Working Paper 98/1(Washington: International Monetary Fund).

Giavazzi, Francesco, Tullio Jappelli, and Marco Pagano,2000, “Searching for Nonlinear Effects of Fiscal Pol-icy: Evidence from Industrial and Developing Coun-tries,” NBER Working Paper No. 7460 (Cambridge,Massachusetts: National Bureau of Economic Re-search).

Greene, William, 2000, Econometric Analysis (Upper Sad-dle River, New Jersey: Prentice Hall).

Hamilton, L.C., 1991, “How Robust Is Robust Regres-sion?” Stata Technical Bulletin, Vol. 2, pp. 21–26.

International Monetary Fund, 2002, “Review of the KeyFeatures of the Poverty Reduction and Growth Facility—Staff Analyses,” available on the Web athttp://www.imf.org/external/np/prgf/2002/031502.htm.

Ize, Alain, 1991, “Measurement of Fiscal Performance inIMF-Supported Programs: Some Methodological Is-sues,” in How to Measure the Fiscal Deficit, ed. byMario I. Blejer and Adrienne Cheasty (Washington:International Monetary Fund).

Jones, Larry E., Rodolfo E. Manuelli, and Peter E. Rossi,1993, “On the Optimal Taxation of Capital Income,”NBER Working Paper No. 4525 (Cambridge, Massa-chusetts: National Bureau of Economic Research).

Khan, Mohsin S., and Abdelhak S. Senhadji, 2001,“Threshold Effects in the Relationship Between In-flation and Growth,” Staff Papers, InternationalMonetary Fund, Vol. 48, No. 1, pp. 1–21.

King, Robert G., and Sergio Rebelo, 1990, “Public Policyand Economic Growth: Developing Neoclassical

Implications,” NBER Working Paper No. 3338(Cambridge, Massachusetts: National Bureau of Eco-nomic Research).

Kneller, Richard, Michael Bleaney, and Norman Gem-mell, 1998, “Growth, Public Policy and the Govern-ment Budget Constraint: Evidence from OECDCountries,” University of Nottingham, Departmentof Economics Discussion Paper No. 98/14.

______, 1999, “Fiscal Policy and Growth: Evidence from OECD Countries,” Journal of Public Economics,Vol. 74, pp. 171–90.

______, 2000, “Testing the Endogenous Growth Model:Public Expenditure, Taxation and Growth Over theLong Run,” University of Nottingham, Departmentof Economics Discussion Paper No. 00/25.

Landau, Daniel, 1986, “Government and EconomicGrowth in the Less Developed Countries: An Empir-ical Study for 1960–80,” Economic Development andCultural Change, Vol. 35, pp. 35–75.

Mackenzie, George A., David Orsmond, and Philip Ger-son, 1997, The Composition of Fiscal Adjustment andGrowth: Lessons from Fiscal Reforms in EightEconomies, IMF Occasional Paper No. 149 (Wash-ington: International Monetary Fund).

McDermott, C. John, and Robert F. Wescott, 1996, “AnEmpirical Analysis of Fiscal Adjustments,” Staff Pa-pers, International Monetary Fund, Vol. 43 (Decem-ber), pp. 725–53.

Mendoza, E., Gian Maria Milesi-Ferretti, and P. Asea,1997, “On The Ineffectiveness of Tax Policy in Al-tering Long-Run Growth: Harberger’s Superneutral-ity Conjecture,” Journal of Public Economics, Vol. 66,pp. 99–126.

Perotti, Roberto, 1999, “Fiscal Policy in Good Times andBad,” Quarterly Journal of Economics, Vol. 114, No. 4,pp. 1399–1436.

Sarel, Michael, 1996, “Nonlinear Effects of Inflation onEconomic Growth,” Staff Papers, International Mon-etary Fund, Vol. 43 (March), pp. 199–215.

Tanzi, Vito, and Howell H. Zee, 1996, “Fiscal Policy andLong-Run Growth,” IMF Working Paper 96/119(Washington: International Monetary Fund).

Thomas, Alun, 2001, “An Exploration of the Private Sec-tor Response to Changes in Government SavingAcross OECD Countries,” IMF Working Paper 01/69(Washington: International Monetary Fund).

Von Hagen, Jürgen, and Rolf Strauch, 2001, “Fiscal Con-solidations: Quality, Economic Conditions, and Suc-cess,” Public Choice, Vol. 109, pp. 327–46.

Part II. Implementation

103

6

Designing appropriate governance structures for an international financial institution such as the IMF is difficult, because steps to enhance the legitimacy of suchan institution through constraints on its decision-makingprocess may affect its operational efficiency. Potentialtrade-offs between legitimacy and efficiency exist for anypublic institution but are arguably more severe for an in-ternational one, because delegating power to it is politi-cally controversial and, thus, likely to imply tighter con-straints. The paper also underscores that the trade-offsare not absolute, however: they depend on the specificways in which legitimacy is pursued—that is, on the spe-cific constraints that are set. Strategic reforms should,thus, aim at improving the terms of the trade-off by ex-ploring steps that are Pareto-improving in the dimensionsof legitimacy and efficiency.

Introduction

The ability of an international organization toachieve its goals ultimately rests on its legitimacy(Woods, 2000; and Van Houtven, 2002, p. 66). Le-gitimacy means that its actions must be seen as ex-pressing an accepted source of power delegated to itby sovereign countries. An institution that is not re-garded as legitimate will face key obstacles inachieving its goals and will likely be ineffective.However, establishing legitimacy typically requiresthe imposition of constraints on the operations ofsuch international organization—rules, systems ofchecks and balances, transparency requirements—and these, in principle, may come into conflict withthe efficient pursuit of the international institution’sgoals, or what will be called here its operational effi-ciency (the ability to achieve its goals without wast-ing resources). Governance structures aimed at en-hancing the institution’s legitimacy may reduce itsoperational efficiency, giving rise to the potential

trade-offs between legitimacy and operational effi-ciency.

This paper is about these trade-offs as they applyto the governance of the IMF. It should be stressedfrom the outset that these trade-offs are common toany public or political institution, but they becomemore severe for an international institution becausethe delegation of power to an international institu-tion remains a politically difficult act. This impliesthat the constraints set on international institutionsto boost their legitimacy in the eyes of sovereigncountries (and their national voters) may have to bestronger than those set on domestic institutions,thus potentially affecting more deeply their opera-tional efficiency. This potentially low efficiency hasimplications for the achievement of the institution’sstated goals. In principle, low efficiency can be off-set by the adequate provision of resources, so as to atleast preserve an institution’s effectiveness. But re-leasing resources to an international institution isalso a controversial action for sovereign states. In-ternational institutions may end up being subject toparticularly tight resource constraints. Conse-quently, limited efficiency is likely to translate intolimited effectiveness.

The paper discusses the legitimacy-efficiencytrade-offs with respect to three dimensions:

• Control of political power over the operational deci-sions of international “technocrats” (second sec-tion). Close control by national political author-ities is one way to enhance the legitimacy of aninternational institution, but it may lead to decisions that are suboptimal from a technicalperspective. The paper argues that the politicalcontrol over the IMF is pervasive at the formallevel. In practice, owing to information-processing constraints, the control is less perva-sive, although unevenly applied across countries.Yet, its costs in terms of efficiency are not trivial.Here the challenge is to find forms of controlthat reduce the disturbance to operational efficiency, based perhaps on more operationalindependence coupled with strong ex post ac-countability.

• Transparency in IMF decision making (third sec-tion). Transparency enhances legitimacy, and,

Efficiency and Legitimacy:Trade-Offs in IMF Governance

Carlo Cottarelli 1

1 I would like to thank Mark Allen, Jim Boughton, MichaelDeppler, Enrica Detragiache, Atish Ghosh, Luc Hobloue, JeroenKremers, Kate Langdon, Alessandro Leipold, Ashoka Mody,Miria Pigato, Alessandro Rebucci, Abebe Selassie, and IgnazioVisco for their helpful comments; Jeff Frieden, Anne Krueger,and J.J. Polak for useful discussions on some of the topics coveredin this paper; and Jolanta Stefanska and Michael Filippello forexcellent research assistance.

CHAPTER

104

6 � EFFICIENCY AND LEGITIMACY: TRADE-OFFS IN IMF GOVERNANCE

in many respects, can also lead to increased ef-fectiveness. However, it may also come intoconflict with operational efficiency, given theconfidential nature of the financial matters theIMF deals with. This section explores the vari-ous channels through which transparency mayinvolve costs, a necessary step in the search formore effective forms of transparency.

• Uniformity of treatment across countries (fourthsection). Uniformity of treatment also enhanceslegitimacy but may involve spending resourceson activities that have low priority for the pur-pose of achieving the IMF’s operational goals.Avoiding waste and arbitrary selection at thesame time is a difficult task.

The resource-constraint issue, which, as noted,implies that low efficiency may impair effectiveness,is discussed in the fifth section. That section also il-lustrates how the particular nature of IMF work, aswell as outdated work processes, may further con-strain the efficient management of IMF resources.The sixth section concludes the paper.

Before proceeding with the paper, two pointsmust be underlined. First, this paper is, by its nature,somewhat lopsided, as it focuses more on the needfor efficiency than on the need for legitimacy. Thisaims at provoking the discussion on an issue—thesearch for efficiency—that is often disregarded.Moreover, the paper’s focus on the existence oftrade-offs between efficiency and legitimacy doesnot mean that the author believes those trade-offsare inevitable. Indeed, their extent depends on thespecific ways through which legitimacy is pursued.The challenge is thus to find possible ways in whichthe terms of the trade-off could be improved. For ex-ample, as noted, forms of political monitoring thatare focused on ex post accountability do not directlyaffect operational decisions, contrary to direct polit-ical control. Also, other steps advocated in the pastto boost IMF legitimacy—such as more transparentprocedures for the selection of IMF management orchanges in the distribution of political controlacross various member countries—may achieve thatgoal with no undesired effects on efficiency.2 Re-forms should, thus, aim at improving the terms ofthe trade-off by exploring steps that are Pareto-im-proving in the legitimacy and efficiency dimensions.However, while examples of such steps are provided,a full discussion of possible lines of reform goes be-yond the scope of this paper—whose goal is to high-light the existence of potential trade-offs, not tofind solutions. More work, as well as expertise inareas such as delegation theory, is needed to addressthe difficult issues raised in this paper.

Second, assessing the magnitude of the trade-offsis admittedly a difficult and essentially judgmentaltask. But this does not justify neglecting their exis-tence. Indeed, one should wonder to what extentsome of the alleged mistakes in IMF work during thelast few years were due to governance structures re-flecting an inadequate appreciation of the trade-offs,rather than to specific errors in judgment.3

Appropriate Degree and Form of Political Control

Public opinion often perceives the IMF as a mono-lith, and little attention is paid to the various com-ponents of its governance structure. A bird’s-eyeview of the latter is therefore useful (see VanHoutven, 2002, Chapter 3 for a more detailed dis-cussion).

IMF Governance Structure

The IMF governance structure involves four entities:

• The direct representatives of countries’ interests(the “capitals,” as they will be referred to here-inafter). Formally, the main bodies representingthe capitals are the Board of Governors—whichincludes ministers of finance and central bankgovernors—and its International Monetary andFinancial Committee (IMFC). These bodiesprovide the ultimate source of political over-sight but meet infrequently (typically twice ayear). Meetings of the IMFC are now routinelypreceded by preparatory meetings at theDeputies’ level.

• The Board of Executive Directors (EDs)—here-inafter referred to as “the Board”—where 24EDs in Washington, D.C. represent the IMF’s184 member countries. The Board “exercises allthe powers for conducting the IMF’s business,except those that the Articles of Agreement—that is, the Fund’s statute—have reserved forthe Board of Governors” (Van Houtven, 2002,p. 14). The Board provides close political over-sight over management (see next paragraph) andstaff, meeting at IMF headquarters for at leastthree days a week. Thus, the “representatives” of

2 See, for example, Akyüz (2002), Woods (2000), Buira(2003), Caliari and Schroeder (2003), and Passacantando(2004).

3 For example, the report of the Independent Evaluation Officeon Argentina (IMF, 2004) argues that the IMF supported weakpolicies in that country for too long, focused its fiscal analysis toonarrowly, did not analyze sufficiently the long-term sustainabilityof the exchange rate regime, and did not enforce conditionality.While highlighting these shortcomings is important, it is equallyimportant to understand the underlying causes of those short-comings and, in the perspective of this paper, whether they couldbe explained by inadequate governance structures. For example,and bringing forward some of the issues raised later, did those er-rors reflect “clientelism” of IMF staff with respect to the authori-ties (see the relevant subsection), or implicit political pressures,or the inadequacy of the provision of human resources?

105

Carlo Cottarelli

IMF shareholders follow the IMF’s work, andapprove its decisions, virtually on a daily basis.This is an unusual governance feature with re-spect to, say, a private corporation. Note alsothat the EDs, while to some extent indepen-dent,4 generally operate, for key decisions, inconcert with “the capitals.”

• The Managing Director, appointed by the Board,is at the same time the IMF’s Chief ExecutiveOfficer, head of the staff, and the Chairman ofthe Board. It is common to refer to the Manag-ing Director, First Deputy Managing Director,and Deputy Managing Directors collectively as“management.”

• The staff—currently some 2,700 internationalcivil servants—represent the IMF’s “technical”voice. They execute IMF policies in the sameway that, for example, central bank employeesexecute monetary policy at the national level.

In addition to this formal structure, there are otherforums in which IMF business is discussed, such as theG-7 (Group of Seven), G-20, and G-24 meetings.

This governance structure—particularly the roleof the Board in making and monitoring IMF deci-sions—suggests a relatively limited degree of delega-tion from the political to the technical dimension.Things are more complex than they appear, though,and quite naturally differ across the range of activi-ties performed by the IMF. To explore the actual al-location of decision-making power, it is useful to review how the relationship between various com-ponents of IMF governance—and, in particular, be-tween what I will call the “political pole” (the Boardand “the capitals”) and the technical pole (manage-ment and staff)—has evolved over time.5

Some History

The issue of the appropriate degree of political over-sight was hotly debated when the IMF was estab-lished. British and U.S. views diverged on this. TheU.S. side, headed by Harry Dexter White, envisagedclose political oversight, involving a Board workingat the IMF headquarters. The British side, headedby John Maynard Keynes, was against close over-sight, and the British “plea for independence” wasclear from the very beginning. In the words ofRobert Skidelsky (Keynes’s biographer):

The British wanted the two institutions [IMF andthe World Bank] to be apolitical, deciding matterson technical grounds. To this end they wantedthem located outside Washington; and wanted theFund, in particular, to be under the unencumberedcontrol of the Managing Director and his staff,with the twelve Executive Directors and their al-ternates representing their countries and regionson a part-time basis, and at part-time salaries. TheAmericans wanted the Fund and Bank to be lo-cated in Washington: they wanted the executivedirectors to be full time…. (Skidelsky, 2003,pp. 829–30).

The British view reflected a number of concerns:(a) political considerations might overrule technicalconsiderations;6 (b) insufficient delegation mightprevent efficiency;7 and (c) the IMF might be seenas operating more under rules than under discretion(Martin, 2002, p. 16). American views eventuallyprevailed, as close political oversight was seen as theneeded counterpart for accepting the obligations ofmembership (Van Houtven, 2002, p. 65), and theIMF ended up with a Board of political appointeessitting in continuous session. Indeed, in the earlydays of the IMF’s life, EDs had a direct role not onlyin vetting IMF decisions daily but also in runningIMF business, including by actively negotiating withcountries and heading field missions (Martin, 2002,p. 20). However, it soon became clear that this gov-ernance structure was de facto impracticable, owingprimarily to informational constraints, as the mem-bership and activities of the IMF increased:“Growth of membership and turnover on the Boardmeant that it did not build up the kind of institu-tional memory that the staff gained over time”(Martin, 2002, p. 23).8

Thus, as of the early 1950s, the role of manage-ment and staff gradually increased (Horsefield,1969, p. 470–73; Gold, 1972, p. 172; and Strange,1973, p. 279).9 In the words of one of the U.S. EDs,“The result was a strong Management/staff and an

4 An Executive Director cannot be legally dismissed until hisor her term has expired. However, this formal independence doesnot apply to the “appointed EDs” (those representing the coun-tries with the five largest quotas), who can be dismissed at anytime and for any reason (see, for example, Kafka, 1996, p. 331).

5 This split between political and technical poles is a simplifica-tion, as each component of each of the poles plays a specific role.Moreover, the role of management, as interface between staff andthe Board, cannot be seen as fully technical. More generally, re-flecting their personalities and backgrounds, some members ofmanagement may be closer to the political pole than others.

6 The submission of the British delegation at the Atlantic Citymeeting of June 1944 that preceded the July 1944 Bretton Woodsconference considered that, “so far as practicable, we want to aimat a governing structure doing a technical job and developing asense of corporate responsibility to all the members, and not theneed to guard the interests of particular countries” (quoted inHorsefield (1969), p. 86). See also Boughton (2001), p. 1032.

7 At the Savannah conference of March 1946, which at thatpoint had to give operational interpretation to the Articles ofAgreement, the Canadian representatives argued that “the Boardcould not achieve the best results if it was engaged in a continu-ous study of figures and memoranda” (Horsefield, 1969, p. 132).

8 See also Kafka (1996, p. 327).9 The Board’s decisions of January 12, 1948 formalized and de-

tailed the relationship among staff, management, and the Board,including assigning to the staff the task of conducting negotiationswith country authorities. However, the Board’s rein on the staffremained initially tight, as “the composition of each staff missionwas subject to Board approval, and the Board outlined detailedinstruction for them” (de Vries and Horsefield, 1969, p. 471).

106


Executive Board that acted largely on managementrecommendations” (Southard, 1979, p.7).

Between the early 1950s and the early 1990s thebalance of power between the two poles wentthrough various fluctuations, with no clear trend.Overall, while the Board never became a rubberstamp for management/staff decisions, it retained amore significant role in setting general policies andguidelines, rather than in decisions on specificcountries (Boughton, 2001, p. 1031). The obviousexceptions were country cases with higher politicalvalue for the key shareholders—where the politicalrole of the capitals over the Board was probablymore important, however.

A more definite trend may have emerged sincethe mid-1990s, with a number of steps aimed at in-creasing the information available to the politicalpole and its capability of vetting, ex ante, the posi-tions taken by the technocratic pole:

• More information has been required to be dis-closed to the Board: for example, side letters10

have been required to be disclosed to the Boardsince September 1999. Also, as of early 2003, all“comfort letters” and similar statements madeby staff to other international financial institu-tions (IFIs), donors, and creditors on countrydevelopments must be transmitted to the Board.

• The technical pole’s discretion in shaping pro-gram conditions and modalities has been con-strained: conditionality has been streamlined,particularly as concerns structural measures.Moreover, prior actions—steps that staff re-quires to be taken by countries as a condition ofpresenting a program (or the review of a pro-gram) to the Board—have become formal con-ditions for the use of IMF resources. Finally, de-tailed provisions have been issued regarding theconditions under which large lending opera-tions can be recommended by staff (the so-called exceptional access cases, for which earlyBoard involvement is now also required).

• Requirements for ex ante scrutiny of programsby the Board have been tightened: as of mid-2003, before presenting a new program to theBoard for countries with prolonged use of IMFresources (some 80 percent of program coun-tries outstanding in mid-2004), an ex post as-sessment report justifying the need for a newprogram and outlining its main features needsto be discussed by the Board. Negotiations on anew program cannot be completed before suchdiscussion has taken place.

Moreover, within the political pole, there seemsto be a clear trend toward shifting control away from

the Board and toward the capitals of the largestcountries:

The major industrial countries, the Group ofSeven … have exhibited a growing tendency in re-cent years to act as a self-appointed steering groupor “Directoire” of the IMF. Recent reports of the fi-nance ministers to the heads of state and govern-ment at the annual summit meetings have some-times tended to deal with IMF matters in a mannerthat raises the question of whether they will leavethe Executive Directors representing the Group ofSeven countries with the necessary margin for dis-cussion and room for give-and-take that is essentialfor consensus building (Van Houtven, 2002,pp. 30–31).11

There are four reasons for these trends: (1) the in-creasing size of IMF financial support in severalheadline cases during the 1990s called for closer po-litical oversight over the use of “international tax-payers’ money;” (2) the role played by the IMF asthe number of programs increased raised the ques-tion of whether unelected technocrats had not ac-quired too much power; (3) the end-of-the-centurycrises raised questions on IMF effectiveness (dele-gating political power is always controversial, butmore so if results are mixed); and (4) with the dra-matic development in communication technologies,it has become easier for the capitals to exert theirinfluence on the Board (Kenen and others, 2004,pp. 99–100).

Key Features of IMF Governance and Its Implications for Efficiency

The key features of the IMF governance structurecan thus be summarized as follows.

• First, at the formal level, political control is per-vasive, as most IMF actions have to be approvedby the Board. This formal control has increasedin recent years.

• Second, in practice, the ability of the politicalpole to exert its influence is limited by its con-straints in processing information. The viewthat the Board does not have the resources tomonitor staff effectively is indeed quite wide-spread (De Gregorio and others, 1999, pp. 21,78–82; Harper, 1998, pp. 284–85; Caliari andSchroeder, 2004; and, perhaps more signifi-cantly, IMF, 1999, pp. 33–34). Thus, it remainsto be seen whether the recent effort to

10 Side letters are letters that program countries send to man-agement on confidential aspects of program policies (e.g., in theexchange rate area).

11 As these changes were implemented, more radical proposalsto enhance political control were being discussed. See a descrip-tion of the French reform proposal in De Gregorio and others(1999, p. 99) and of the Miyazawa proposal also in De Gregorioand others (1999, p. 95).

strengthen formal control over the techno-cratic pole will succeed in enhancing actualcontrol.12 The information processing con-straints remain severe; in this respect, the ratiobetween professional staff working for theBoard and regular professional staff has re-mained roughly constant over the last 10 years(at about 91⁄2 percent).13

• Third, although there is no firm evidence ofthis, it stands to reason that, in the most rele-vant cases as well as in the setting of key policyguidelines, the balance of power is tilted towardthe political pole.14 Within the political pole,however, power rests primarily with the capitalsof the larger shareholders. Even leaving asidearguments of political weight, larger countries(and, more generally, advanced countries) candevote more resources to monitoring the IMF,and can thus play a more significant role. And,as argued, the role of the “large” capitals has in-creased in recent years.

The preceding configuration raises issues of legiti-macy and equality of treatment across countries.15

Here, however, we focus on its implications for theIMF’s efficiency.

The first implication relates to the risk that IMFdecisions (in cases where the political stakes arehigh) reflect direct political pressures, originatingtypically in the capitals, rather than technical fac-tors. How severe is this problem? Leo Van Houtven,former IMF Secretary, while noting that “it couldnot be expected that decisions would always betaken exclusively on technical grounds,” eventuallyconcludes that “the limited occurrence of politicaldecisions in the IMF has been remarkable” (VanHoutven, 2002, pp. 43–44). Econometric work onwhether IMF decisions reflect political factors yields

mixed results.16 Surveys do indicate that politicalpressures are not unusual, with 7 percent of missionchiefs reporting that technical judgment was over-ridden by political pressures “frequently” or “al-ways,” and 48 percent reporting that political pres-sures had been experienced “occasionally” or“sometimes” (IMF, 2002, p. 64). Moreover, variousIMF watchers have reported several cases of directpolitical pressures.17 Be this as it may, the frequencyof cases of direct political interference is not all thatmatters: the perception that there have been casesof political interference affects the IMF’s credibility,and hence its effectiveness, in all cases. Further-more, the perception that the IMF is a “geopoliticalslush fund” (Willett, 2004) and that the IMF serves“the ad hoc political purposes of broad foreign pol-icy” (Calomiris, 2000, p. 86) remains quite wide-spread (see also Allegret and Dulbecco, 2004). Thecontroversy surrounding IMF decisions taken inearly 2003 to roll over the loans to Argentina is stillfresh.18

A second, and perhaps more subtle, implicationof the IMF’s formal political dependence is the al-leged “clientelism” of IMF staff vis-à-vis country au-thorities. Bordo and James (2000, p. 8) notes that“IMF staff reports on member countries are thoughtto be insufficiently critical because of the develop-ment of a sort of ‘clientelism,’ in which good rela-tions with officials and ministers develop.” The issueof “clientelism” is also raised in IMF (1999, p. 65):“A view that exists in the institution is that a reportthat is incisive but offends the authorities is damag-ing to a mission chief ’s career while one that isbland and later turns out to be lacking in some im-portant respects will be overlooked.”

While, again, the existence of clientelism shouldnot be exaggerated, some aspects of IMF culture mayencourage it. In particular, managers are assessedpartly on the basis of their ability in keeping goodrelationships with country authorities. This may af-fect, in particular, surveillance cases, as this is wherea harsh assessment given by staff may be morestrongly objected to by country authorities (particu-larly in “calm waters,” that is, when surveillance ispotentially more effective in preventing futurecrises).

107

Carlo Cottarelli

12 Reversals did occur in the past. For example, some 20 yearsago, the Board asked that at the end of a program, an assessmentof its success or failure be included in the first staff report follow-ing the conclusion of the program. This initiative—similar to thenew “ex post assessments” discussed above—eventually fell intodisuse.

13 Moreover it has been argued that, because of the increasedspeed at which decisions have to be taken in today’s world ofhigh capital mobility, the actual balance of control has shiftedtoward the staff (De Gregorio and others, 1999, p. 80), exceptwith respect to large countries such as the United States (DeGregorio and others, 1999, p. 82), which can devote larger re-sources to staff monitoring.

14 The difficulties faced by the IMF management’s initiative toestablish a new debt-restructuring mechanism for sovereigndebt—in plain English, a mechanism to deal with the cases ofcountry bankruptcies—show that when it comes to major policyissues, the technocratic pole has limited traction.

15 For example, Van Houtven (2002) argues that the shift ofpower toward the capitals and away from the Board has reducedthe role of developing countries: decisions at the Board are takenby consensus, thus partly enhancing the role of countries withlimited formal voting rights.

16 Bird (2003, pp. 248–52) lists a number of papers that do findeconometric evidence of political factors in IMF decisions, in-cluding Stiles, 1991; Thacker, 1999; and Barro and Lee (2002).However, the evidence in Bird and Rowlands (2001) is more nu-anced. After reviewing this literature, Bird (2003) concludes thatthe picture is not “completely clear.”

17 See lists in De Gregorio and others (1999) and Bordo andJames (2000, pp. 39–40); see also Stiles (1990) and FinancialTimes, October 5, 2004, p. 4, “G7 Interfered in IMF Bid to PushThrough Russia Reform.”

18 See, for example, the leader (editorial) in Financial Times, January 20, 2003, “The G7 Blinks: The IMF Has Been Forced toTake a Huge Gamble.”

Does clientelism reflect the IMF’s political gover-nance structure? This is hard to prove. But it standsto reason that the voice of country authorities is am-plified by the fact that the IMF is not formally inde-pendent from political forces, and that political rep-resentatives sit in Washington, D.C. in continuoussessions.19

The third implication relates to the ongoing shiftin decision-making power from the Board to thecapitals. As argued by Kenen and others (2004,pp. 99–101), this shift affects efficiency as decisionsare taken by officials who have more limited knowl-edge of the IMF than do EDs and have only occasional or no contact with staff. Allegret andDulbecco (2004, p. 10) take a similar view. Thisproblem is particularly severe in areas where the political pole retains a higher degree of control,namely in setting policy guidelines.

The fourth implication relates to the role of theBoard in the IMF’s daily management. From a cor-porate governance perspective, it is unusual thatshareholders’ representatives participate in dailymanagement activities. The rationale for more de-tached participation (focused, for example, on set-ting goals and monitoring results) rests on the information-processing constraints that are at thebasis of any principal-agent relationship. Given thecost of evaluating information, it is more efficientfor decisions at the daily level to be delegated to themanagement of a company or an institution. More-over, a body representing shareholders must poten-tially be large. In the IMF’s case, the large size of theBoard also reflects its political nature and the re-lated need to give voice, without excessive pooling,to all member countries.20

The report of the external evaluation of surveil-lance points candidly at the atypical role of theBoard in terms of governance: “Everything we knowabout institutional governance indicates to us that agroup of 24 is, to put it mildly, extremely large foruseful exchanges of views, discussion, and group de-cision making” (IMF, 1999, p. 75).

Van Houtven (2002, p. 23) also notes that the“decision making and management of the IMFwould be better served by a smaller Board.” Kafka(1996, p. 333) expresses a similar view.21

What impact does the Board’s atypical role haveon efficiency? As discussed, many believe that theBoard does not have enough resources to challengethe staff ’s views and that, in practice, in the major-ity of cases it performs only a formal role. However,this activity of formal control is quite time consum-ing—for the Board as well as the staff. It is time con-suming for the Board because trying to absorb (andreact to) all the information provided by staff is noteasy; about two-thirds of the Board’s time is spent oncountry matters rather than in setting policy guide-lines or in monitoring their overall implementation.And for the staff, it is time consuming because it re-sults in the production of activities that may be car-ried out largely as a formality. With the increasedweight of the political pole in the last few years, thetrend toward formal “micromanagement” has, ifanything, increased, with the Board expressingviews and imposing formal requirements on quitetechnical and detailed issues.22

Trade-Off and Right Balance

Political oversight is needed by domestic institu-tions and, to an even greater extent, by interna-tional organizations, because they must be seen bysovereign countries as exercising legitimate powers.Without legitimacy, the IMF ultimately would beineffective.23 Some forms of political control canhamper operational efficiency, however; hence atrade-off arises. The question is: what form and de-gree of delegation from the political pole to the tech-nocratic pole—in other words, the degree of “slack”in the principal-agent relationship—are most appro-priate for an institution such as the IMF? This issuehas not been studied much. Models of delegation24

have found limited application to the study of in-ternational organizations, possibly because of the

108


19 There are provisions that, in principle, protect staff from po-litical pressures, but their effective role is doubtful. In particular,Article XII, Section 4(c) of the Articles of Agreement includes aprovision obliging countries to “respect the international charac-ter of … [the staff ’s] duty and … refrain from all attempts to in-fluence any of the staff.” But this article is not well known evento IMF staff, a sign that IMF culture does not emphasize it. In asample of IMF economists I contacted, only 7 percent knew thatsuch a provision existed.

20 Indeed, the Board’s political nature has also made it difficultfor the Board to operate through committees: “The glaring absenceof meaningful committee work speaks volumes for the constraintsunder which Directors apparently operate, de facto if not legally,as country and constituency representatives” (IMF, 1999, p. 75).This “size” issue is typical of international organizations. Asnoted by Lister (1984, p. 101), “a perennial problem of interna-tional organization has been to fashion an executive organ that isboth small enough to deal expeditiously with the flow of regularbusiness and yet representative enough to act authoritatively.”

21 As IMF Secretary, Leo Van Houtven was directly in chargeof the relationships with the Board; Alexandre Kafka was an Ex-ecutive Director for more than 30 years.

22 Examples are the list of indicators staff should use to assesscountries’ vulnerabilities, templates for debt sustainability, thereporting of the effectiveness of IMF surveillance in specificcountries, the reporting of relationships with the World Bank,the reporting of statistical issues, and the background papers pre-pared for Article IV consultations.

23 Indeed, various authors have underscored that some form ofpolitical control can enhance the IMF’s effectiveness. De Gregorioand others (1999, p. 93) argues that peer pressure can be an impor-tant tool in the IMF’s hands. Cunliffe (quoted in De Gregorio andothers, p. 125) argues that “granting independence to the IMFwould result in the dissipation of support for the institution.”

24 See, for a recent survey, Bendor, Glazer, and Hammond(2001); and, regarding delegation from politicians to bureaucrats,Alesina and Tabellini (2004).

109

Carlo Cottarelli

complexity of the issues involved—for example,those related to the existence of multiple “princi-pals” (the country members), which have, in turn,various principals (each country’s stakeholders).

I have argued in the preceding text that the cur-rent approach results in (a) political influences thatare often not transparent and are exercised un-evenly across countries; and (b) excessive resourcesbeing spent in pro forma activities, reducing at thesame time the actual political oversight, the abilityto set proper guidelines, and efficiency. Could a dif-ferent approach work better? Various proposals havebeen put forward:

• Giving the IMF more operational independence,while enhancing its ex post accountability, as pro-posed, for example, by De Gregorio and others(1999), Bordo and James (2000), and Allegretand Dulbecco (2004). Note, however, that theproposal in De Gregorio and others (1999) andAllegret and Dulbecco (2004) to make the Boardindependent does not address one issue raised inthis paper, namely the difficulty a large body,such as the Board, has in managing a financialinstitution on a daily basis. This issue could per-haps be better addressed by reviving the idea of anonresident board, in charge of broad oversight,rather than of specific decisions, as advocatedoriginally by Keynes and more recently by others.

• Enhancing the protection of staff from explicit andimplicit political pressure. A critical step, recom-mended by the report of the External Evalua-tion of Surveillance, would be to alter the incentive structure by making it clear that man-agement will back up staff members who givefrank advice (IMF, 1999, p. 67).

• Reassessing the role of the Board, with the aim ofminimizing activities that result in merely formalcontrol, while increasing its role vis-à-vis the capitals(Van Houtven, 2002). One specific step proposedby Woods (2000) is to increase the amount oftime the Board could spend on setting policyguidelines by considerably reducing its responsi-bilities for country decisions.

Appropriate Degree of Transparency

Transparency is key to the legitimacy of a public in-stitution. Until recently (the mid-1990s), in deter-mining the extent of IMF transparency, the balancebetween legitimacy and (at least what was perceivedas) efficiency was biased toward the latter. It was as-sumed that the confidential nature of IMF businesswas incompatible with a high degree of operationaldisclosure. However, during the last few years, it hasbecome clear that secrecy weakens IMF legitimacy.Moreover, it has been noted that transparency isalso important for efficiency. First, it allows closermonitoring and accountability of IMF staff; and,

hence, in a principal-agent perspective, it shouldboost efficiency. Second, transparency in the IMF’sadvice (e.g., the publication of IMF staff reports)magnifies the IMF voice; if markets listen to it, in-appropriate policies are more directly penalized, en-hancing the “peer pressure” mechanism on whichIMF surveillance has traditionally relied. Indeed,the shift toward greater transparency has partly re-flected the Fund’s attempt to enhance its effective-ness in the aftermath of the Asian crisis.

While the benefits of transparency in terms of le-gitimacy and efficiency have been appropriately em-phasized over the last few years, the debate has beensomewhat one-sided. There are also potential trade-offs between transparency and efficiency, and ac-knowledging them is a necessary step in establishingappropriate transparency policies. The reader will, Ihope, forgive me if, to correct this one-sidedness,the rest of this section is somewhat biased in the op-posite direction.

Some Definitions

When I use the term “transparency policies” (or, forsimplicity’s sake, “transparency”), I refer to policiesthat require the dissemination of documents to anaudience that would not otherwise receive them.The following should be noted:

• I am referring to the dissemination ofdocuments, as one cannot assume that the con-tent of the document (that is, the information) isunaffected by the dissemination constraint.25

• Typically, IMF documents contain (a) informa-tion on a certain country or country group (e.g.,data on reserves, the decision of a governmentto devalue, information on certain financial in-stitutions, the government’s intention to intro-duce a certain measure); (b) views of staff, or ofother components of IMF governance, on a cer-tain policy issue (e.g., whether exchange ratelevels or regimes are appropriate, whether bank-ing supervision is adequate, or whether politicalrisks exist); and (c) assessments of IMF perfor-mance. In what follows, I will focus primarily on(a) and (b).26

• Discussions on transparency should alwaysidentify the original and new recipients of thedocuments. In what follows, I will focus onchanges in transparency rules that broaden the

25 Strictly speaking, using the term “transparency policies” as asynonym for publication requirements is not appropriate, sincethe latter does not necessarily involve a higher degree of dissemi-nation of information. Nevertheless, I will follow the standardconvention.

26 What follows could, however, also be applied to assessmentsof IMF performance (self-assessments or assessments by outsideexperts). The latter, however, also bring to the fore some addi-tional issues, including whether excessive emphasis on past “mis-takes” may reduce the credibility of the IMF in the future.

110


dissemination of documents from the techno-cratic pole (and the country authorities of a spe-cific country) to either the political pole or thegeneral public.

Potential Drawbacks of Transparency

I will first consider the specific mechanisms throughwhich transparency vis-à-vis the public at large canaffect IMF efficiency. I will also initially (and per-haps unrealistically) assume that the disseminationrequirement does not alter the information in thedocument to be published. In this context, trans-parency has three direct drawbacks:

• Transparency—for example, the publication ofIMF views regarding the sustainability of an ex-change rate regime under current policies—cancause a negative market reaction (say, a specula-tive attack) that could have been avoidedthrough a change in policies (IMF, 1999, p. 75).The stress here is on the fact that the negativemarket reaction could have been avoided hadcountry authorities been given sufficient timeto react to the IMF’s views. Supporters of fulltransparency often argue that if the IMF’s views can trigger a crisis, they might as well bereleased, since the crisis will occur one way oranother. This view misses the point that theIMF’s confidential advice could, one wouldhope, lead to a policy change that would avoidthe crisis. Moreover, a vulnerability assessmentis always probabilistic in nature, but most read-ers will only consider the modal projection oroutlook, which could precipitate a crisis.

• In countries with an IMF-supported program,transparency (in particular, the publication ofletters of intent) may disrupt the conditionalityprocess. A program, as well as its related condi-tionality (that is, the set of conditions that needto be met for the IMF to continue to support aprogram), is not carved in stone at its inception.Conditions are often modified as developmentsunfold (Mussa and Savastano, 1999). Thisprocess of program negotiation and renegotia-tion is more difficult to manage if markets arefully aware of the contents of letters of intent.For example, failure to take certain actions cov-ered by conditionality may trigger a crisis evenwhen the IMF would have been willing towaive them. Conversely, waiving conditionsmay be inappropriately seen as a watering downof a program. The key point is that the negotia-tion/conditionality process requires time, andtransparency, in a world of high capital mobil-ity, dramatically cuts the time available for thisprocess to work effectively and for an optimalredesign of a program. Thus, in this context,conditionality may end up acting as a coordi-nating device for speculative attacks (creatinga reverse catalytic effect).

• Transparency about a government’s intention toimplement a measure can prevent the imple-mentation of that measure if public knowledgeof the government’s intention leads to a coales-cence of vested interest against it.

In the preceding cases, the original information isprovided to the public, so, at least, the expectedbenefits usually linked to transparency (a more in-formed public) can be reaped. But one cannot as-sume that the information in the documents to bepublished will not be affected by the publication re-quirement. First, given the costs arising if the origi-nal information is disseminated, country authoritiesmay be unwilling to share all of the informationthey have with IMF staff. This would prevent theIMF from effectively performing its functions, asstaff will not be in a position to provide adequateadvice to country authorities (IMF, 1999, p. 75).And second, given the limited protection of staffvis-à-vis political pressures (see the second section),or because of the risk of causing a crisis, staff may beinclined to be less candid in documents that aregoing to be published. This implies that the correctinformation would not be provided to the public,giving rise to a pro-forma transparency.

Whether the above drawbacks apply also with re-spect to transparency vis-à-vis the Board depends onwhether the information provided to the Boardwould or would not leak to the public. This is a con-troversial issue. IMF (1999, p. 78) concludes that“any such discussions [on confidential issues] couldonly be reported to the Board in a quite general wayif their substance were expected to remain confiden-tial.” Van Houtven (2002, p. 19) also notes that“delicate issues may arise in cases when the need fordisclosure of information to Executive Directors ap-pears difficult to reconcile with the requirements ofconfidentiality of a member country.” Martin (2002,p. 40) refers to the “occasional embarrassing leaksfrom the Executive Board,” although leaks of sensi-tive material have been infrequent (and there alsohave been examples of leaks from staff). However,even without leaks, countries may be unwilling toshare with other countries specific information. Asnoted by Martin (2002), “For any borrowing coun-try, some states are likely to be political adversaries.They will then be reluctant to reveal sensitive infor-mation to the Board.”

Transparency in IMF History

How have these considerations affected IMF workover the last sixty years? For many decades after theIMF was set up, the issue of transparency vis-à-visthe public was not regarded by IMF shareholders ascritical. Transparency issues did feature prominentlyin the discussions leading to the establishment ofthe IMF, but only as far as they related to the infor-mation to be provided to the Board. Thus, at the Sa-vannah conference of March 1946, Keynes took the

111

Carlo Cottarelli

view that no country would be willing to confide inthe Managing Director if the Executive Directorswere to be made acquainted with everything thatwas going on (Horsefield, 1969, p. 133).

In any event, as discussed above, the shift to thestaff in early 1948 of the responsibility for dealingwith country authorities created a barrier betweenthe Board and member countries (Martin, 2002,p. 37). This situation prevailed during the followingdecades. The issue, however, resurfaced periodically,with the technocratic pole reiterating the following:

The confidentiality of relations between membergovernments and the Fund management and staff,so fundamental to the successful operations of theFund, could be impaired…. [W]ere a member gov-ernment to believe that any information providedto the Fund would be made available to govern-ments around the world, there would be a devastat-ing effect on the future of the Fund. (de Vries,1985, p. 994, referring to staff and managementviews in the late 1970s).

During the 1990s, however, the demand from theworld public opinion for more transparency vis-à-visthe public mounted rapidly (see, for example, DeGregorio and others, 1999, pp. 84–85) and the tech-nocratic pole, also in consideration of potential ben-efits in terms of effectiveness (see the previous dis-cussion of this issue), came to accept the need forchange (Van Houtven, 2002, p. 69). Thus, the IMFhas started publishing a number of previously un-published documents, including staff reports and let-ters of intent (IMF, 2003).

The pressure on the technocratic pole for moretransparency vis-à-vis the Board has increased in par-allel. As discussed in the previous section, this has,for example, involved the disclosure to the Board ofall side letters, as well as more frequent and informalBoard discussions of country developments.

Unresolved Issues

While the move toward transparency has merits,and is in any case unavoidable, it has been assumedthat at least some of its potential drawbacks couldbe easily minimized through appropriate policies.27

In particular, rules allow the deletion of “highlymarket-sensitive material” from published staff re-

ports. However, the deletion policy may not be per-ceived as sufficient protection by the countries in-volved. First, the request for deletion needs to be ap-proved by IMF management—consequently it is farfrom certain what would be regarded as “highly mar-ket-sensitive.” Second, the flexibility in makingchanges, even for highly market-sensitive informa-tion, is limited.28 And third, since the deletion pol-icy covers only highly market-sensitive information,it does not address other concerns the authoritiesmay have, including the risk that untimely publica-tion of policy intentions would make the policy im-plementation more difficult.29

Thus, the majority of the Board has typically fo-cused on the benefits from transparency rather thanon the existence of trade-offs. For example, the fol-lowing refers to a recent Board discussion:

Directors considered that candor and transparencywere essential dimensions of surveillance, and tooknote of efforts to improve information provided tothe Executive Board and to boost publication ofstaff reports…. Some Directors noted, however,that there may be trade-offs between transparencyand candor (IMF, 2003).30

I side with those who believe these trade-offsshould be explored more extensively.31 Ultimately,increased publication of IMF documents can en-hance the IMF’s efficiency: when the voice of theIMF is heard publicly, governments (particularly indemocratic countries) may be more willing to actunder pressure from public opinion and the markets.But acknowledging the existence of costs is an im-portant step toward finding more efficient forms oftransparency. In particular, the following questionsseem to deserve further scrutiny:

28 Deletions are possible, but modifications need to be limitedto what is required to keep the text intelligible and grammati-cally correct. Moreover, deletions cannot be used to eliminateentire sections of a report or several paragraphs. Finally, the pre-sumption is for a high degree of parsimony in distinguishing material that is clearly highly market sensitive from what is onlypolitically sensitive.

29 A side letter can be used to protect the confidentiality ofmeasures whose early disclosure would make their implementa-tion more difficult. But this approach can only be followed forprogram countries, and not for countries where the IMF is onlyengaged in surveillance.

30 The reference to a trade-off between transparency and can-dor (rather than between transparency and effectiveness) shouldnot mislead the reader. The issue is that as a result of requiringpublication of staff reports, these may turn out to lack candor andthus to be ineffective.

31 Mohammed (2000, p. 203) also points out the costs arisingfrom excessive transparency. At the time of his writing. Mo-hammed was Advisor to the Chairman of the Group of Twenty-Four (G-24), the intergovernmental group of developing countriesset up to concert their position on monetary and finance issues. Hehad earlier been Director of the IMF’s External Relations Depart-ment. Kafka (1996, p. 335) also warns about the limits of trans-parency. Williamson (2000) actually blames the IMF for exces-sive transparency when it “forced Thailand to reveal inOctober 1997 that it had mortgaged all its reserves” (p. 336).

27 Others have assumed that transparency, even in delicateareas such as exchange rate assessment, cannot be a problem inany case. The recent report of the IMF’s Independent EvaluationOffice acknowledges the inherent sensitivity of exchange rate as-sessments for countries with pegged exchange rates, as such as-sessments could “alarm the markets” (IMF, 2004, paragraph 238),but concludes that the problem can be resolved by making suchdiscussions “a routine exercise, something markets expect tooccur as a matter of procedure.” However, it is not clear why thefact that assessments of exchange rate regimes are routinely madewould diminish the impact of an IMF statement that concludedthat a certain regime is not sustainable.

• How to make sure that, in spite of publicationrequirements, countries provide the necessaryinformation. Self-assessments conducted by theIMF on the implementation of transparencypolicies are quite sanguine on the outcomes ofthe experiment. However, only time will tellhow the provision of information to the IMF bymember countries has been affected by publica-tion requirements.

• How to make sure that the publication of inter-nal IMF documents does not result in a proforma transparency, where the most importantinformation is actually not disclosed.

• To what extent it would be appropriate to delaythe publication of some documents, in contrastto the current practice of virtually immediatepublication.32 This would still allow ex post ac-countability while reducing the risk of ex antecensorship.

• What the appropriate degree of transparencyvis-à-vis the political pole (of the Board and thecapitals) is. The extent to which the informa-tion acquired by the technocratic pole, and allviews held by it, should be shared with theBoard (and “the capitals”) is not entirely clearat present.

Appropriate Degree of Uniformity of Treatment Across Countries

Evenhandedness in the treatment of country mem-bers is seen as a critical requirement for the IMF33 asit seeks to enhance its legitimacy. And yet, eco-nomic conditions do differ across countries. Thus, itstands to reason that efficiency requires selectiv-ity—applying certain procedures or approaches toonly the countries that really need them. In theory,there is no contradiction between evenhandednessand selectivity. One can, for example, use somescreening process that is applied to identify coun-tries that, for example, require more in-depth work.What ultimately matters is that countries be as-sessed on objective grounds—a uniformity of treat-ment in substance. In practice, however, formalisticuniformity of treatment is much easier to establishthan substantive evenhandedness. This may lead toa bias toward the application of the same formalprocedures to all countries, with a consequent lossin terms of resources and, hence, efficiency.

While the need for selectivity in IMF surveillancework is, in principle, accepted, difficulties have arisen

in practice. If anything, trends in recent years seem tohave been toward less selective approaches—as advo-cated by some developing countries (see, for example,Mohammed, 2000, p. 203). The key surveillanceprocess—the Article IV consultation—takes placeannually for most countries, with fewer than one-tenth of the IMF membership on a 24-monthcycle.34 Even in the latter case, staff visits often takeplace in years without formal Article IV consulta-tions. Various standard surveillance tools, many ofwhich have been introduced in recent years tostrengthen IMF surveillance (for example, those toassess public debt and external debt sustainability),have to be applied to all countries. Moreover, majornew expansions in the IMF mandate—in particularthe assessment of the application of standards andcodes—have covered all countries.

This bias against selectivity emerged clearly dur-ing the Board discussion of the so-called vulnerabil-ity assessment exercise, one of the few examples ofselective practices, that was introduced by IMF staffand management in 2001 to monitor more closelycountries with higher vulnerabilities. The press in-formation notice summarizing the Board discussionon this issue indicated the following:

The periodic vulnerability assessment exercise …provides a platform for an independent assessmentby relevant functional departments of key issues inindividual countries, offers an opportunity to ex-change views on analyses of vulnerability acrossdifferent regions, and provides inputs for bilateralsurveillance activities and program design. SeveralDirectors saw merit in applying this exercise to ad-vanced economies and not just emerging marketeconomies. Other Directors pointed out that de-velopments in advanced economies were exam-ined at high frequencies through other mecha-nisms (IMF, 2003).

More selectivity in IMF work has been advocatedin several reform proposals in recent years, but theattention has not focused on selectivity in the coun-try dimension. Following Feldstein (1998), the needfor focusing the IMF on its core areas of (macroeco-nomic) responsibility has been widely accepted, atleast in principle. Correspondingly, various authorshave underscored the need to reduce the overlap, interms of issues covered, in the mandates of the maininternational economic institutions (for a recentdiscussion, see Kenen and others, 2004, pp. 95–97).

While a pruning of IMF responsibilities is needed,one should not forget that a rigid breakdown of eco-nomic problems by their nature is not easy. In par-ticular, macroeconomic issues and structural issuesare often closely related, and excessively constrain-ing the scope of the IMF’s work would inevitably

112


32 De Gregorio and others (1999, p. 88) notes that delaying thepublication of documents would reduce the risk of negative mar-ket reactions.

33 For example, the October 2004 IMFC communiqué notedthat, “Effective and evenhanded IMF surveillance across thewhole membership is central to promoting high and sustainablegrowth in member countries and to crisis prevention” (IMFC,2004). In the IMF, the term “surveillance” refers to the monitor-ing of countries’ economic developments and policies.

34 Program countries are also on a 24-month cycle, but theyare, of course, subject to much closer monitoring as part of theprogram discussions.

113

Carlo Cottarelli

impair its effectiveness.35 Combining selectivity inthe issue dimension with selectivity in the countrydimension—dealing with certain issues only incountries where those issues are macroeconomicallyrelevant—could thus provide a more effective ap-proach. The question is how to do this without im-pairing the principle of uniformity of treatmentacross countries.

In this respect, two avenues could be explored:

• At the “micro level,” consideration should begiven to broadening the use of screeningprocesses to avoid applying essentially the sameformal procedures to all countries.

• At the “macro level,” the key issue is whetherthe responsibilities in terms of country coverageof the international economic organizations—which involves significant overlap between theOrganization for Economic Cooperation andDevelopment (OECD) and the IMF for indus-trial countries, and between the World Bankand the IMF for nonindustrial countries—should not be revised. Some have, for example,argued that the IMF should reduce its work onindustrial countries (see, for example, Polak,2004). A milder alternative would be to identifya “leading institution” in the relationship be-tween the international community and eachcountry. The issue is, once more, a difficult one,since it involves trade-offs. Leaving aside thequestion of legitimacy, it has been argued thatthe overlapping of responsibilities may fosterhealthy competition (Krueger, 1997, p. 23) and,possibly, accountability.

Resource Constraints, Efficiency, and Trade-Offs

Constraints on an institution’s efficiency would notimpair its effectiveness if an adequate amount of re-sources were made available. It is thus important toassess the tightness of the IMF’s human-resourceconstraint.

Few IMF-watchers have addressed this issue.36

This is surprising because the adequacy of human resources is critical to the effectiveness of any

decision-making process. Bordo and James (2000,p. 6), in discussing the size of the IMF, does start byproviding information on both human and financialresources, but it then tackles only the latter’s ade-quacy. Vaubel (1994, pp. 53–54) points to the rapidincrease of IMF staff since 1960 and wonderswhether it represents ”a textbook case of Parkinson’sLaw” and concludes—without reporting any sup-porting empirical evidence—that the increase wasnot explained by increased balance of payments im-balances worldwide but was more correlated withthe increase in IMF quotas. However, other crit-ics—perhaps most forcefully Jeffrey Sachs and theMeltzer Commission—have argued that the IMF isunderstaffed, given the number of program casesusually handled.37 In the same vein, IMF (1999,p. 73) also pointed out the inadequacy of its ownhuman resources, concluding that “there is no doubtthat many Fund staff are chronically overworked.”

Yet, a cursory look at the growth of IMF staff overthe last thirty years suggests no dearth of resources.From 1970 to 2002, the number of staff increased inrelation to the number of member countries, al-though much less markedly with respect to thenumber of IMF-supported programs (Table 1). How-ever, this increase needs to be interpreted in light ofrising IMF responsibilities (for example, during the1990s in international standards and codes, anti-money laundering, financial sector analysis, gover-nance, and provision of information related totransparency requirements) and the increased com-plexity of problems, particularly as a result of in-creased capital mobility.38

Moreover, the increase in IMF staff does not saymuch about the adequacy of its level. IMF teamsseem small by many standards. A typical IMF “mis-sion” team includes, besides the mission chief, four(rarely five) economists who usually have additionalassignments to carry out when at headquarters. Aresident representative assists the team in most pro-gram cases, and in key surveillance cases. Other staffare also involved in reviewing the work of missionteams on a part-time basis, with reviewers typicallyworking on tens of countries at the same time.

Is this resource endowment sufficient? Considerthe following:

• Economic policy teams in central banks and fi-nance ministries of industrial countries typically

35 For example, many of the turn-of-the-century emerging mar-ket crises deeply reflected underlying structural problems. This,of course, does not mean that the IMF should deal, at the technicallevel, with areas where it does not have enough expertise. But,the overall assessment policy design and, in the case of programcountries, the negotiation process with country authorities canhardly be broken down without an effectiveness loss.

36 At the Savannah conference of March 1946, when IMFgoals had already been crystallized in the Articles of Agreement,the views on the appropriate size of the staff ranged widely fromthe 30 professionals proposed by the British delegation to the 300proposed by the U.S. delegation (Skidelsky, 2003, p. 830). Thewidth of this range, however, partly reflects the more rule-basedapproach the British side envisaged for the IMF (see the secondsection).

37 “The Meltzer Commission noted that the IMF, with just onethousand or so professional staff, could not and should not try torun dozens of countries’ economic programs” (Sachs, 2000). TheMeltzer Commission—or, more properly, the International Fi-nancial Institution Advisory Commission—was established inNovember 1998 by the U.S. Congress to consider the future rolesof seven international financial institutions, including the IMF.

38 For example, the increase in resources used for surveillanceduring FY2001/2002 was devoted to new tasks in financial sur-veillance and standards and codes, and to multilateral surveil-lance, with no increases in the resources available for bilateralsurveillance.

114


include hundreds of economists.39 IMF teamsare expected to cooperate with country authori-ties, but, in practice, local economists may nothave adequate skills, and may have differentgoals than the IMF staff ’s, making full coopera-tion difficult.

• Each of the four mission team economists is typ-ically in charge of one sector (real, monetary, fis-cal, and balance of payments). This means, forexample, that the critical balance of paymentssector is typically the responsibility of a singleeconomist. Thus, a single person deals with mon-itoring and projecting trade flows, the serviceaccount, the capital account, external debtstocks, competitiveness, trade restrictions, issuesrelated to external debt negotiations, and so on.

• The availability of human resources assigned toprogram countries rises only modestly with thesize of the financial resources committed by theIMF, as the composition of country teams, par-ticularly in area departments (that is, thosemore directly in charge of dealing with countryauthorities), is fairly similar across countries.The relation between the amount of humanand financial resources assigned to a country isdepicted in Figure 1, which reports on the verti-cal axis the log of the ratio between human re-sources (in staff-years) and the log of millions ofSpecial Drawing Rights (SDRs) committed.The relationship is negative, since the amountof human resources does not rise proportion-ately with the size of lending. The implicit elas-ticity is quite low (0.11) for the whole IMF (top

panel) and is particularly low for area depart-ments (0.08)—indeed only one-half that ofother departments (0.16). This means that asthe size of a program increases, the amount ofresources increases more slowly for the teams incharge of a more direct contact with the author-ities, while a faster increase is observed for thestaff in charge of review work and other non-core activities.

• Program work attracts more resources than sur-veillance work, but the increase is, again, rela-tively modest. An econometric estimate relat-ing the human resources assigned to a certaincountry to measures of its size, a dummy for pro-gram countries, the amount of lending, and thecountry type (industrial or not) shows that, onaverage, the existence of a program increasesthe human-resource allocation by less than athird with respect to a surveillance-only rela-tionship (Table 2). This is not much, takinginto account the responsibilities that fall on thestaff when they are engaged in program work.

The scarcity of the IMF’s human resources is ex-acerbated by the high turnover across countries ofIMF staff. Some 60 percent of the teams of nonpro-gram countries change with each mission. Perhapsmore important, since mission teams often includeeconomists assigned only temporarily to a country,the turnover of regular desk economists (on theirmain country assignment) is also high: based on asample of area department economists, collected inMay 2004, the median time that desk economistsand mission chiefs had worked on their main assign-ments were 12 months and 13 months, respectively.

Whether high turnover is an inevitable feature ofIMF work is controversial. Some IMF watchers seethis as a requirement for an international organiza-tion, given that it allows staff to benefit from“cross-fertilization” of experiences (Feinberg andGwin, 1989, p. 26; De Gregorio and others, 1999,p. 19). High turnover may also reduce the risk ofclientelism. Others, however, have regarded this

39 The Bank of England’s staff in the Monetary Analysis and theFinancial Stability Departments includes some 220 professionals.About 200 professionals work at the U.K. Treasury on macroeco-nomic, financial, and structural reform issues. The Research andEconomics directorates of the European Central Bank (ECB) in-clude some 180 professionals. The Monetary Affairs and the Re-search and Statistics Departments of the Board of Governors ofthe (U.S.) Federal Reserve System include some 160 profession-als. The U.S. Treasury team working in the Domestic Finance andEconomic Policy Departments includes some 140 professionals.

Table 1

Number of Fund Staff Relative to IMF Member Countries and Fund-Supported Programs

Year

Staff 1970 1980 1990 2002

All staffPer country 8.7 10.1 11.9 14.7Per program 43.3 48.8 34.8 51.6

Professional staff and managersPer country ... 5.9 7.6 10.5Per program ... 28.9 22.2 36.6

Note: Three dots (...) indicate that the data were not available.

115

Carlo Cottarelli

y = –0.8894x + 0.5062

R2 = 0.9593

–4.00

–3.50

–3.00

–2.50

–2.00

–1.50

–1.00

–0.50

0.00

0.50

0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5log(F)

log(

H/F

)

y = –0.9167x + 0.3138

R2 = 0.9781

–4.00

–3.50

–3.00

–2.50

–2.00

–1.50

–1.00

–0.50

0.00

0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5

log(F)

log(

H/F

)

y = –0.8436x + 0.0084

R2 = 0.847

–4.0

–3.5

–3.0

–2.5

–2.0

–1.5

–1.0

–0.5

0.0

0.5

0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5log(F)

log(

H/F

)

All IMF departments

IMF area departments

Other IMF departments

Figure 1

Allocation of IMF Financial and Human Resources Across Program Countries

Notes: H denotes the amount of human resources, measured in staff-years, used in FY2003 and FY2004; F denotes the approved amount oflending for programs outstanding as of September 30, 2003, in millions of Special Drawing Rights (SDRs).

116


turnover as excessive (IMF, 1999, pp. 31–32; IMF,2002, pp. 88, 137), pointing out the lack of familiar-ity with country features that it generates. Other po-tential drawbacks are shortsightedness in work plan-ning (and, thus, insufficient incentives to start workinitiatives with longer-term yields) and reduced ac-countability, since successors have to deal withproblems that were left unattended (IMF, 2002,p. 68). Be this as it may—and I believe turnover isexcessive—to the extent that high turnover is re-garded as necessary, it should be taken into accountin assessing the adequacy of IMF resources.

Little work has been done by IMF staff to assessthe adequacy of the level of human resources (while,of course, the effect on the existing resource endow-ment of changes in tasks assigned to the IMF is rou-tinely assessed). One exception is Ivanova and oth-ers (2003), which tries to assess whether theprobability of success of IMF programs depends onthe size of staff teams. The results suggest that teamsize has little effect on program outcomes. This re-sult—which is puzzling, short of concluding thateconomic science is so inconsequential that thework of economists does not have value added—canhave two explanations: (a) assuming there is a mini-mum threshold for effectiveness, the size of staffteams may currently fall so short of that thresholdthat a marginal additional amount of resources doesnot help; and (b) the organization of staff teams isnot adequately modified to accommodate increasedavailability of resources.40

This discussion suggests at least the need for acomprehensive assessment of the adequacy of theIMF’s human resources. It is critical to stress that theissue has to be addressed by taking a fresh look atthe level of resources, not at increments. Of course,should resources be found insufficient, the conclu-sion should not necessarily be that they need to beincreased. Possibly, tasks need to be streamlined (seethe fourth section on selectivity) or efficiency needsto be improved (for example, by exploring ways toreduce excessive turnover).

Conclusions

This paper has highlighted a number of issues thatshould be explored further to support the ongoingeffort to strengthen IMF governance with the goalof simultaneously enhancing its legitimacy and effi-ciency. It has underscored the existence of trade-offsthat make this goal elusive: features of governancethat could enhance the IMF’s legitimacy mayweaken its operational efficiency. But synergies—or,at least, Pareto improvements—are also possible,and some of them also have been highlighted.

For example, reforms aimed at a redistribution ofpolitical oversight among countries may enhance le-gitimacy without increasing the degree of involve-ment of political control over day-to-day opera-tions. Strengthening the protection of staff againstpolitical influences would help increase candor inpublished IMF documents and, hence, lead to gen-uine, rather than pro forma, transparency. Ex postaccountability can be a substitute for close ex antemonitoring by the political pole of the technocraticpole. Reduced direct political pressure would also beconsistent with uniformity of treatment acrosscountries and alleviate the risk that selectivity inprocesses will be perceived as reflecting political discrimination.

The intention of this paper was not to propose so-lutions, but to highlight issues. Considerably more

Table 2

Factors Affecting the Allocation of IMF Human Resources Across Countries: Regression Results Based on FY 2003/04(Dependent variable: number of staff-years)

Regressors Coefficient t-Statistics

Constant 0.10 2.50Log (per capita GDP)1 0.21 8.85Log (SDR)2 –0.05 –1.03Program dummy3 0.43 4.33Industrial country dummy4 –0.30 –5.00

Notes: R2 = 0.50; standard error = 0.24; and number of observations = 177.1 At purchasing-power-parity exchange rates.2 Amount of approved lending per program, in millions of Special Drawing Rights (SDRs).3 Equal to 1 for program countries.4 Equal to 1 for industrial countries.

40 Anecdotal evidence suggests that the latter may be the case.For some time, the Russia desk benefited from a large increase instaff resources. The experiment ran into organizational difficul-ties, with many concluding that IMF teams larger than four orfive economists are ineffective. However, the work organizationof the Russian desk did not change as resources increased. Thisled to a duplication of functions (with, say, two economists cov-ering, in competition, the same sector) and related coordinationproblems. Absorbing new resources typically requires a new workorganization—for example, assigning senior economists the taskof supervising more junior ones.

117

Carlo Cottarelli

work is needed before reaching conclusions. Topicsthat deserve further work include the following:

• How the political pole could monitor the tech-nocratic pole without undue costs in terms ofefficiency, and, in this respect, (a) the role thatexplicit political considerations should play inaffecting IMF decisions; (b) the role of the Boardin day-to-day management of the IMF; and(c) steps to avoid staff’s potential “clientelism.”

• How to reconcile transparency and confiden-tiality needs, and how to avoid forms of proforma transparency that would damage both ef-ficiency and accountability.

• The possibility of improving efficiency throughmore selectivity in the country dimension—more specifically, the scope for increased pre-screening to identify countries to which moreresource-intensive procedures should be ap-plied, and for ways to reduce the overlapping ofcountry responsibilities across international or-ganizations.

• The adequacy of the amount and use of theIMF’s human resources.

ReferencesAkyüz, Yilmaz, 2002, “Towards Reform of the Interna-

tional Financial Architecture: Which Way Forward?”in Reforming the Global Financial Architecture, Issuesand Proposals, ed. by Yilmaz Akyüz (Penang: UnitedNations).

Alesina, Alberto, and Guido Tabellini, 2004, “Bureaucratsor Politicians?” NBER Working Paper No. 10241(Cambridge, Massachusetts: National Bureau of Eco-nomic Research).

Allegret, J.P.A., and P. Dulbecco, 2004, Why InternationalInstitutions Need Governance? The Case of the IMF,21st Symposium on Banking and Monetary Econom-ics, Nice, June 10–11.

Barro, Robert, and Jong-Wha Lee, 2002, “IMF Programs:Who Is Chosen and What Are the Effects?” NBERWorking Paper No. 8951 (Cambridge, Massachu-setts: National Bureau of Economic Research).

Bendor, J., A. Glazer, and T. Hammond, 2001, “Theoriesof Delegation,” in Annual Review of Political Science,Vol. 4, pp. 235–69.

Bird, Graham, 2003, “Political Economy InfluencesWithin the Life Cycle of IMF Programmes,” in TheIMF and the Future, Issues and Options Facing the Fund(New York: Routledge).

———, and D. Rowlands, 2001, “IMF Lending: How Is ItAffected by Economic, Political and InstitutionalFactors,” Journal of Policy Reform, No. 4.

Bordo, Michael D., and Harold James, 2000, “The Inter-national Monetary Fund: Its Present Role in Histori-cal Perspective,” NBER Working Paper No. 7724(Cambridge, Massachusetts: National Bureau of Eco-nomic Research).

Boughton, James M., 2001, Silent Revolution: The Interna-tional Monetary Fund, 1979–89 (Washington: Inter-national Monetary Fund).

Buira, Ariel, 2003, “The Governance of the IMF in aGlobal Economy,” in Challenges to the World Bank and

IMF, Developing Country Perspectives, ed. by ArielBuira (London: Wimbledon Publishing Company).

Caliari, Aldo, and Frank Schroeder, 2004, “Reform Pro-posals for the Governance Structures of the Interna-tional Financial Institutions” (unpublished: NewRules for Global Finance).

Calomiris, Charles W., 2000, “When Will EconomicsGuide IMF and World Bank Reforms?” in Cato Jour-nal, Vol. 20 (Spring/Summer), pp. 85–103.

De Gregorio, José, Barry Eichengreen, Takatoshi Ito, andCharles Wyplosz, 1999, “An Independent and Ac-countable IMF, Geneva Reports on the World Econ-omy 1” (London: Center for Economic Policy Re-search).

de Vries, Margaret Garritsen, 1985, The International Mon-etary Fund 1972–1978, Cooperation on Trial, Vols. Iand II: Narrative and Analysis (Washington: Interna-tional Monetary Fund).

———, and J. Keith Horsefield, The International Mone-tary Fund, 1945–1965: Twenty Years of InternationalMonetary Cooperation (Washington: InternationalMonetary Fund).

Feinberg, Richard E., and Catherine Gwin, 1989, “Re-forming the Fund,” in The International MonetaryFund in a Multipolar World: Pulling Together (Wash-ington: Overseas Development Council).

Feldstein, Martin, 1998, “Refocusing the IMF,” Foreign Af-fairs, Vol. 77 (March/April), pp. 20–33.

Gold, Joseph, 1972, Voting and Decisions in the InternationalMonetary Fund (Washington: International Mone-tary Fund).

Harper, Richard H.R., 1998, Inside the IMF, An Ethnogra-phy of Documents, Technology and Organisational Ac-tion (London: Academic Press).

Horsefield, J. Keith, 1969, The International MonetaryFund 1945–1965, Twenty Years of International Mone-tary Cooperation, Vol. I: Chronicle (Washington: In-ternational Monetary Fund).

International Monetary and Financial Committee (IMFC)of the Board of Governors of the International Mone-tary Fund, 2004, communiqué, October 2.

International Monetary Fund (IMF), 1999, External Eval-uation of IMF Surveillance, Report by a Group of Inde-pendent Experts (Washington).

———, 2002, Evaluation of Prolonged Use of IMF Re-sources, Independent Evaluation Office (Washing-ton).

______, 2003, The Fund’s Transparency Policy: Progress Re-port on Publication of Country Documents (Washington).

______, 2004, Evaluation of the Role of the Fund in Ar-gentina, 1991–2001, Independent Evaluation Office(Washington).

Ivanova, Anna, Wolfgang Mayer, Alex Mourmouras, andGeorge Anayiotos, 2003, “What Determines the Im-plementation of IMF-Supported Programs?” IMFWorking Paper 03/08 (Washington: InternationalMonetary Fund). This paper also appears, in some-what different form, as Chapter 10 in this volume.

Kafka, Alexandre, 1996, “Governance of the Fund,” inThe International Monetary and Financial System: Developing-Country Perspectives, ed. by G.K. Helleiner(New York: St. Martin’s Press).

Kenen, Peter B., Jeffrey R. Shafer, Nigel L. Wicks, andCharles Wyplosz, 2004, The International Economicand Financial Cooperation: New Issues, New Actors,New Responses (London: Center for Economic PolicyResearch).

118


Krueger, Anne O., 1997, “Whither the World Bank andthe IMF?” NBER Working Paper No. 6327 (Cam-bridge, Massachusetts: National Bureau of EconomicResearch).

Lister, Frederick K., 1984, Decision-Making Strategies forInternational Organizations: The IMF Model, Mono-graph Series in World Affairs, Vol. 20, Book 4 (Den-ver, Colorado: University of Denver, GraduateSchool of International Studies).

Martin, Lisa L., 2002, “Distribution, Information, andDelegation to International Organizations: The Caseof IMF Conditionality” (unpublished; Cambridge,Massachusetts: Harvard University).

Mohammed, Aziz Ali, 2000, “The Future Role of the IMF:A Developing Country Point of View,” in Reformingthe International Financial System, Crisis Prevention andResponse, ed. by Jan Joost Teunissen (The Hague:FONDAD).


Passacantando, Franco, 2004, “Democrazia e Governanceinternazionale: il caso delle istituzioni di Bretton Woods”(upublished; Rome: Banca d’Italia).

Polak, J.J., 2004, “If the Bretton Woods Conference Wereto Be Held Now, It Would Not Succeed,” interviewin IMF Survey, Vol. 33 (August 23).

Sachs, Jeffrey D., 2000, “Cutting the IMF and WorldBank Down to Size,” in Project Syndicate. Availableon the Web at http://www.project-syndicate.org/com-mentaries/commentary_text.php4?id=296&lang=1.

Skidelsky, Robert, 2003, John Maynard Keynes1883–1946, Economist, Philosopher, Statesman (Lon-don: Macmillan).

Southard, Frank A., 1979, The Evolution of the Interna-tional Monetary Fund (Princeton, New Jersey: Inter-national Finance Section, Department of Economics,Princeton University).

Stiles, Kendall W., 1990, “IMF Conditionality: Coercionor Compromise?” in World Development, Vol. 18,No. 7, pp. 959–74.

———, 1991, Negotiating Debt: The IMF Lending Process(Boulder, Colorado: Westview Press).

Strange, Susan, 1973, “IMF: Monetary Managers,” in TheAnatomy of Influence: Decisionmaking in InternationalOrganizations, ed. by R.W. Cox and H.K. Jacobson(New Haven, Connecticut: Yale University Press).

Thacker, S., 1999, “The High Politics of IMF Lending,”World Politics, No. 52, pp. 38–75.

Van Houtven, Leo, 2002, Governance of the IMF, DecisionMaking, Institutional Oversight, Transparency, and Ac-countability, IMF Pamphlet No. 53 (Washington: In-ternational Monetary Fund).

Vaubel, Roland, 1994, “The Political Economy of theIMF: A Public Choice Analysis,” in PerpetuatingPoverty, The World Bank, the IMF, and the DevelopingWorld, ed. by Doug Bandow and Ian Vásquez (Wash-ington: Cato Institute).

Willett, Thomas D., 2004, “IMF: Governments Need toBe Persuaded to Mend Their Ways,” letter to the Fi-nancial Times (April 27).

Williamson, John, 2000, “A Mixed Record for IMF Ad-vice,” in Reforming the International Monetary and Fi-nancial System, ed. by Peter B. Kenen and AlexanderK. Swoboda (Washington: International MonetaryFund).

Woods, Ngaire, 2000, “The Challenge of Good Gover-nance for the IMF and the World Bank Themselves,”World Development, Vol. 28, No. 5, pp. 823–41.

119

7

This article uses finance and agency theory to establishtwo key propositions about IMF conditionality andcountry ownership of IMF-supported adjustment pro-grams—first, that the conditionality attached to theseprograms is justified and, second, that country owner-ship of these programs is crucial for their success. Be-cause IMF conditionality and country ownership areboth necessary, the challenge is designing conditionalitythat maximizes ownership while providing adequate safe-guards for IMF lending. The article analyzes several re-cent proposals aimed at enhancing country ownership ofpolicies contained in IMF-supported programs. Theseproposals include encouraging countries to design theirown adjustment and reform programs, streamliningstructural conditionality, introducing flexibility in thetiming of structural policy measures (floating trancheconditionality), and applying conditionality to outcomesrather than policies (outcomes-based conditionality).

Introduction

International Monetary Fund (IMF) lending in sup-port of adjustment programs is conditional on theborrowing countries adopting certain agreed poli-cies. The conditions attached to these loans arecommonly referred to as IMF conditionality. As theliterature on the subject shows, discussions of thenature and merits of IMF conditionality have a longhistory (see, for example, Williamson, 1983; Polak,1991; Guitián, 1995; James, 1996; and Boughton,2001, Chapter 13). The issue has recently gained re-newed attention, with questions raised about whetherthe conditions the IMF imposes on borrowing coun-tries have been too intrusive and whether the designand implementation of IMF conditionality have un-dermined country ownership of adjustment programsaimed at correcting macroeconomic imbalances.

This viewpoint has gained considerable currencyas a result of the capital account crises in Mexico,

East Asia, the Russian Federation, and Brazil in the1990s. But during such crises, which require rapidresponses, there may not be enough time to securefull country support for all the policy actionsneeded. Moreover, in the midst of a serious currencyand financial crisis, there may be greater agreementbetween the country authorities and the IMF on theimmediate problems facing the country and theshort-run measures needed to address them. Condi-tionality and ownership issues are probably less rele-vant for the recent capital account crisis cases andmore relevant in standard IMF programs dealingwith current account crises in low-income coun-tries—and that is where most attention has been fo-cused. The IMF itself has engaged in a comprehen-sive analysis of conditionality and ownership issues(see IMF, 2001a).

This chapter draws on finance and agency theoryto establish two basic propositions about IMF condi-tionality and country ownership of adjustment pro-grams. First, some form of conditionality exists in allborrower-lender relationships: key to the ability toborrow is the ability to pledge income back. TheIMF must have assurances that it will be repaid, andthis requires that it place conditions on its loans.The analysis here is designed to dispel the fairlywidespread notion—articulated, for example, byDíaz Alejandro (1984)—that conditionality stemsfrom a “patron-beneficiary” relationship betweenthe IMF and borrowing countries. As he puts it,“This is the key justification for ‘conditionality’; ifyou ask for a gift, you must listen to your patron”(Díaz Alejandro, 1984, p. 7). Finance considera-tions alone provide the justification for conditional-ity being a necessary part of IMF lending. Thus, theview expressed by Killick (1997) that IMF condi-tionality should be the exception rather than therule is incorrect. Indeed, this article argues that itshould be exactly the reverse.

Second, country ownership of programs is essen-tial, because it aligns the incentives of the borrowerand the lender. For the borrowing country, programownership is critical because without a firm commit-ment from the government and other relevant con-stituencies, the difficult policy measures needed to correct economic problems are less likely to be implemented. For the IMF, country ownership in-creases the probability that programs will succeed

IMF Conditionality and Country Ownership of Adjustment Programs

Mohsin S. Khan and Sunil Sharma1

1 The authors are grateful to Shanta Devarajan, Stanley Fischer,Morris Goldstein, Laurence Harris, Harold James, Vijay Kelkar,Peter Montiel, Jean Tirole, and several IMF colleagues for helpfuldiscussions and comments. This article was previously publishedin the Fall 2003 issue of the World Bank Research Observer, and sug-gestions made by that journal’s referees helped sharpen the analysis.This article appears here with the permission of Oxford UniversityPress, the journal’s publisher.

CHAPTER

120

7 � IMF CONDITIONALITY AND COUNTRY OWNERSHIP OF ADJUSTMENT PROGRAMS

and consequently augments the protection of its re-sources provided by conditionality. Thus, thePoverty Reduction Strategy Papers (PRSPs), devel-oped jointly by the IMF and the World Bank, putsubstantial emphasis on country ownership. In sum,both IMF conditionality and country ownershiphave a clear rationale—the challenge is reconcilingthe two. This chapter considers various recent pro-posals for achieving such a reconciliation.

IMF Lending and Conditionality

This section examines the main features of IMFlending by comparing them with private loan con-tracts. It then assesses the implementation and ef-fectiveness of IMF conditionality.

Conditionality in Private Financial Contracts

Between every borrower and lender there is a funda-mental asymmetry in information. Borrowers alwaysknow more about their abilities, opportunities, andintentions than do lenders. This information asym-metry gives rise to two incentive problems: adverseselection and moral hazard.

Adverse selection arises before a transaction oc-curs and stems from the fact that information defi-ciencies make it difficult for lenders to distinguishgood risks from bad. The IMF faces a different selec-tion problem than do private lenders in that onlymembers experiencing distress approach it for fi-nancing, and all have a right to its resources. Moralhazard arises after a lender has given funds to a bor-rower.2 Having obtained the funds, it may be in theborrower’s interest to take risks that may raise re-turns—but also increase the likelihood of default. Inthe financial world, contract designs and collateral,transparency, and reporting requirements attempt tomitigate such moral hazard. Monitoring by share-holders, debt holders, market analysts, rating agen-cies, and independent company directors serves thesame purpose. Such monitoring is costly to firms,but it is done to assure investors that their claimswill be respected (Greenbaum and Thakor, 1995;and Mishkin, 1998).

Pledging collateral is expensive, but firms incurthat cost to assure lenders that borrowed funds willbe used for stated purposes and in ways that will notjeopardize their eventual repayment. Indeed, thevery existence of financial markets depends on suchassurance. To that end, corporate governance andinstitutions and practices of finance share the com-

mon rationale of increasing resources that can bepledged to outside debt and equity holders whilemaintaining appropriate incentives for managersand workers. Collateral is provided so that in theevent of default, or if the borrower does not live upto the terms of the contract, the lender can recoverits resources by taking control of the pledged assetsand selling them.

In new or emerging firms with little physical or fi-nancial collateral, relinquishing control rights toventure capitalists provides the needed assurancethat funds will be well spent, because the venturecapitalists have considerable say in decision making.The allocation of control rights to investors shouldbe designed to provide maximum assurance to in-vestors without impairing the functioning of the firmand its ability to exploit commercial opportunities.Still, in many circumstances and for many firms—es-pecially those in distress or those that have little rep-utational or financial capital—relinquishing suchcontrol rights may be costly and will almost certainlyimpose substantial limits on management discretion.

Generally, the imposition of such conditional-ity—the allocation of return streams, liquidity, andcontrol rights—is made contingent on the evolu-tion of the borrowing firm’s balance sheet and fol-lows a simple carrot-and-stick logic. Posting collat-eral that can be sold by the lender in case of defaultprovides the borrower with an incentive to preventdefault. Agency considerations dictate that thetransfer of collateral rights to the lender be madecontingent on default or on observable measures offinancial and nonfinancial performance.

Such conditionality also serves another purposewhen it comes to control rights. After an adverseshock, a borrowing firm is more likely to gamble forresurrection—that is, engage in riskier behavior—the greater is the deterioration of its financialhealth. Contingent transfer of control rights pro-tects investors in that it prevents firms from under-taking excessively risky activities to repair balancesheets. Hence it serves two purposes. Ex ante, it pro-vides incentives for preventing default. And ex post,it constrains firms’ ability to take gambles.

Even if a firm does not engage in risky strategieswhen its performance deteriorates, the contingenttransfer of control rights provides the lender withthe option of reexamining the new situation. For ex-ample, in a new enterprise the venture capitalist ob-tains full control if the firm performs poorly—whereas if the firm is profitable, the venturecapitalist may retain cash-flow rights but agree tocede control and liquidation rights to the entrepre-neur (Kaplan and Stromberg, 2000).

IMF Conditionality

IMF lending and its associated conditionality followbroadly the same principles as private financial con-tracts, though several additional dimensions makeIMF lending qualitatively different. The IMF is

2 More formally, moral hazard can be defined as actions of eco-nomic agents maximizing their utility to the detriment of othersin situations where they do not bear the full consequences oftheir actions owing to uncertainty, asymmetric information, andincomplete or restricted contracts (see Kotowitz, 1987; andStiglitz, 2000).

121

Mohsin S. Khan and Sunil Sharma

mandated by its Articles of Agreement to extendtemporary financial assistance to member countriesfacing balance of payments difficulties “under ade-quate safeguards” (Article I). Like any lender, theIMF thus needs assurances from its borrowers thatthe funds lent to them will be used for the purposesdefined by the Articles of Agreement, and in a man-ner that does not jeopardize their contractual servic-ing and repayment. Therefore many of the financepropositions relevant to private financial institu-tions also apply to the IMF.

A key aspect of IMF lending is that countries inneed of IMF loans generally do not possess interna-tionally valuable collateral. If they did, they coulduse it to borrow from private lenders and would notrequire IMF resources. In a crisis situation, though,even if a country had internationally acceptable col-lateral, it still might not be able to use it to borrowfrom private capital markets. There is an importantdifference between national income and incomethat can be pledged to foreign lenders, including theIMF. Foreign loans can be used to produce bothtradable and nontradable outputs, but foreignerstypically have no demand for a country’s nontrad-ables. In the absence of collateral, private loan con-tracts typically would include various covenants,coupled with monitoring. Formally, a covenant isdesigned to protect the lender and prohibit the bor-rower from taking actions that could reduce theprobability of repayment. Covenants can imposeclear obligations on the borrower, impose limita-tions on or prohibit certain actions, and specifywhen a borrower is considered to be in default.

IMF conditionality can be viewed as a complexcovenant written into the loan agreement. The pol-icy prescriptions in IMF-supported programs essen-tially provide safeguards that the country will beable to rectify its macroeconomic and structural im-balances and will be in a position to service andrepay the loan.3 Thus the conditionality associatedwith IMF programs can be viewed as a substitute forcollateral.

Conditionality attached to sovereign lending hasa long history. For example, in 1818, Prussia, effec-tively bankrupted by the Napoleonic wars, ap-proached Nathan Rothschild (of the House of Roths-child) for a loan and was asked to pledge “Prussianroyal domains” as collateral (Ferguson, 1998).

For many analysts, the modern model for condi-tional lending to sovereign governments in the ab-sence of collateral is the Turkish agreement of1881—known as the Decree of Mouharrem—whichwas implemented after the Turkish government de-faulted on its foreign debt in 1875. The decree cre-

ated the Council of Ottoman Debt, comprisingseven members that represented different groups ofbondholders. A large portion of the Turkish govern-ment’s revenue was placed under the council’s directcontrol and used by it to service and repay the debt(Anderson, 1966).

The League of Nations also attached strict condi-tions to its adjustment programs (or “reconstructionschemes” as they were known) for six Europeancountries in the 1920s. These conditions includedmaintenance of fiscal equilibrium and monetary dis-cipline as well as currency reform. The League de-veloped various means to enforce the programs andsafeguard the interests of foreign creditors and bond-holders, including appointing a commissioner toeach country and an adviser to each central bank toadminister and monitor the programs (Santaella,1993; and James, 2001).

Agreeing to IMF conditionality is an impositionon a country even though the two may share thesame objectives—external viability, price stability,sustained high growth, reduced systemic risk, and soon. To be sure, the covenant in the loan agreementrequired by the IMF is far more complex and has dif-ferent characteristics than covenants in simpler pri-vate financial transactions. It also may not always bepart of the explicit contract, but it is always part ofthe implicit one, regardless of the language used inthe documentation. The challenge in designing IMFconditionality is to specify the optimal covenant—the best policy conditions given the circumstancesof the country, the disbursement intervals for theloan, the type of monitoring involved, and so on—to achieve program objectives while providing suffi-cient safeguards to the IMF. Still, at a general level,IMF lending conforms to the principles governingprivate lending.

At the same time, IMF lending differs in signifi-cant ways. First, as noted, defining conditionality inIMF lending is much more complicated than in pri-vate financial transactions, where covenants may bequite straightforward. It basically involves assessingthe macroeconomic imbalances or structural defi-ciencies that led to a country’s macroeconomicproblems, and then negotiating an agreement withcountry authorities that will address those issues.

Second, it is difficult, if not impossible to estab-lish the “value” of IMF conditionality. The value ofnegotiated conditionality largely depends on the de-gree to which the authorities of the borrowing coun-try adopt the program and are willing to expend ef-fort and political credibility to implement it.

Third, unlike private lenders—for whom it maybe sufficient to deal with a firm’s management—theIMF faces what in agency theory is called “moralhazard in teams” (Holmström, 1982). This refers tosituations in which a principal’s payoff depends onthe joint efforts of two (or more) agents. Typically,the detailed negotiations for an IMF-supported program are conducted with certain governmentrepresentatives (central bank, finance ministry),

3 Additional safeguards are provided by the fact that IMFclaims are de facto senior to claims of other creditors, and thatfunds are disbursed in tranches, conditional on the implementa-tion of satisfactory policies (which are monitored) to correct theimbalances.

122


while the success of the program depends on manyother stakeholders in society—other ministries,trade unions, professional associations, civic groups,and nongovernmental organizations (NGOs).

Fourth, the IMF is by design a cooperative thatmakes loans to its sovereign members. In the eventof default, there is no court to which it can appealand no tangible collateral that can be used to coverlost resources. The enforcement mechanism for en-suring that borrowing countries live up to theirobligations essentially amounts to a combination ofmoral suasion, maintenance of the borrower’s repu-tation, peer pressure, and the threat of being shutout of international capital markets. Unlike privatefirms—where lending is generally subject to well-defined legal codes that can be enforced in courtsand where shareholders or investors can change themanagement of a company or take it over—the IMFobviously cannot replace sovereign governments,and can only refuse to provide financing to a coun-try in arrears.

Finally, relative to private lenders the IMF, givenits mandate and cooperative structure, faces what inother contexts has been called the “Samaritan’sdilemma” (Buchanan, 1975; and Lindbeck andWeibull, 1988). For example, to provide the rightincentives ex ante a private lender may imposeharsh conditions on a borrower in the event of poorperformance, but ex post (that is, after poor perfor-mance) it may not want to impose those conditionsfor a number of reasons. For instance, selling certainassets (or liquidating the company) may be worthless than keeping the assets with the company (ormaintaining the company as a going concern).

But the IMF faces a different incentive problem,because a borrowing country is always more valuableas a “going concern.” Hence country authoritiesknow that in the event of poor performance, atworst the program will be renegotiated (Drazen andFischer, 1997). This creates the wrong incentives,ex ante. Countries know that, faced with poor per-formance and a weak economy, the IMF is unlikelyto impose harsh conditionality ex post because ithas to be concerned about the welfare of borrowingcountries. And, being a cooperative institution, theIMF cannot simply walk away and “cut its losses.”Thus ex ante penalties have limited credibility be-cause they are unlikely to be enforced ex post. Thisis one reason it has been suggested that the IMFshould lend only to prequalified countries with goodpolicy environments. In other words, the IMFshould lend only to countries that have good eco-nomic track records and that can provide good “col-lateral” (see Meltzer and others, 2000).

Implementation of IMF Conditionality

Conditionality is implemented through program de-sign and monitoring that tracks whether agreedpolicies are implemented in a timely and effective

manner. Program design begins with in-depth analy-sis of the sources of a country’s macroeconomic im-balances. (See Mussa and Savastano, 1999 for a re-cent description of the IMF’s approach to economicstabilization.) The next step is to agree with the au-thorities on policy objectives and on macroeco-nomic policies and structural reforms to achievethose objectives.

Monitoring takes various forms, depending on theborrowing country’s circumstances and the IMFlending facility being used. It generally includesprior actions, performance criteria, macroeconomicand structural benchmarks, and reviews. Monitoringalso prohibits actions inconsistent with the IMF’sArticles of Agreement, such as the introduction ofnew foreign exchange restrictions. Releases of IMFfinancing are linked to compliance with monitoringarrangements.

Prior actions are required when upfront imple-mentation is critical to program success or to allaydoubts about the authorities’ commitment. Exam-ples of required actions include passing an agreedbudget, realigning the exchange rate, adoptingstructural reforms, or enacting relevant laws. Suchactions must be taken prior to the approval of a pro-gram by the IMF’s Executive Board, which then trig-gers the first disbursement.

Performance criteria normally include quantita-tive targets for specified financial aggregates (suchas bank credit, net international reserves, or fiscalbalance) and often include structural measures(such as tariff reductions, tax system revisions, orprivatization of public enterprises). Meeting thesecriteria triggers the release of subsequent tranches ofcommitted IMF resources.

Macroeconomic and structural benchmarks settargets for macroeconomic variables and structuralpolicies important for effective program implemen-tation. They do not directly affect scheduled dis-bursements.

Finally, reviews are used to assess overall progresstoward program objectives, identify any sources ofslippage (resulting from lack of policy implementa-tion, external shocks, or program design issues), andtake corrective actions. Reviews are usually speci-fied as performance criteria—precluding further dis-bursements if the review is not concluded by thescheduled review date.

A common question is whether IMF conditional-ity works. Meltzer and others (2000, p. 35) arguethat “numerous studies on the effects of IMF lendinghave failed to find any significant link between IMFinvolvement and increases in growth or income.”The IMF, not surprisingly, disagrees. The IMF’smandate is to provide short-term lending to supportbalance of payments adjustments, and it believesthat its conditional lending has generally improvedthe external accounts of borrowing countries.

What is the evidence on the effectiveness of IMFconditionality? Specifically, have IMF-supported

123


adjustment programs achieved their objectives ofimproving current account balances, increasing in-ternational reserves, lowering inflation, and raisinggrowth? This is essentially an empirical questionthat requires evaluating the effects that past pro-grams have had on the macroeconomic variables ofinterest. Such evaluations are conducted periodi-cally by the IMF’s Policy Development and ReviewDepartment and by its recently created IndependentEvaluation Office, with the results reported to theExecutive Board. In addition, a number of studiesover the past twenty years, both inside and outsidethe IMF, have examined the question using a varietyof empirical methods.

Almost all the empirical studies surveyed byHaque and Khan (2002) show that IMF-supportedprograms have improved current account balancesand the overall balance of payments in borrowingcountries. The results for inflation are less clear.Most of the studies indicate that while inflation usu-ally falls, the decline is generally statistically in-significant. For growth, output is generally depressedin the short run—that is, during the stabilizationphase—but increases as macroeconomic stability isestablished. Overall, the most recent empirical re-sults in particular indicate that, on average, IMF-supported programs and the conditionality they in-corporate have been reasonably effective inachieving their main macroeconomic objectives.

Country Ownership of Programs

The case for country ownership of IMF-supportedadjustment programs has a strong theoretical foun-dation. In the context of agency theory, principal-agent problems arise in situations where one party(the principal) relies on the other (the agent) toachieve certain objectives. Owing to informationasymmetries and lack of perfect monitoring technol-ogy, if the agent’s actions and their results cannot beeasily verified and monitored, the agent has greaterscope for pursuing its own interests rather thanthose of the principal.

But principal-agent theory also says that an agentwill do a better job for the principal if the two par-ties’ objectives are closely aligned. Thus if the real-ization of conditions hinges on cooperation and im-plementation by the agent, agent ownership of theproject is essential. As Tirole (2002) puts it, in thiscase ownership is not a goal but a necessity.

IMF lending can be cast in a principal-agentframework as well. In this case borrowing govern-ments are the agents and the IMF—the delegatedmonitor of a revolving fund—is the principal. Thisprincipal-agent relationship is complex because ofthe nature of the task and the underlying loan con-tract, the mandate and structure of the principal,and the characteristics of the agents. And this com-plexity, combined with the difficulty of specifying

all contingencies in the contract, makes countryownership of programs all the more critical for theirsuccess.

The problem is that country ownership of IMF-supported programs is an elusive concept. Implicitly,it refers to a situation in which the policy content ofthe program is similar to what the country wouldhave chosen in the absence of IMF involvement—because the country shares with the IMF both the ob-jectives of the program and an understanding of theappropriate economic model linking those objectivesto economic policies. In such a situation, the countryowns the program in the sense that it is committedto its spirit rather than to just complying with its let-ter. Thus the country will not deviate from its agree-ment with the IMF even if given the opportunity.

But because countries only borrow from the IMFwhen they face distress caused by macroeconomic orstructural imbalances, sufficient safeguards areneeded when providing access to IMF resources andto avoid moral hazard—requiring conditionalitywith some “bite.” Hence full country ownership isunlikely, and the real challenge is maximizing own-ership in the context of conditionality. As noted,program conditionality is likely to place substantiveconstraints on the authorities’ actions and use offunds. In addition, given that programs generally in-volve economic and social trade-offs, perceptionsmay be created that conditionality does not takeproper account of a country’s circumstances, includ-ing its economic priorities, political conditions, cul-ture, and traditions. And this can—and sometimesdoes—lead to differing views on objectives, programstrategies, and the pace of reform.

Ownership matters because it directly affects pro-gram implementation.4 Program agreements cannotenvisage all the possible contingencies that couldaffect a program and specify in advance actions thatthe authorities should take in response. When acountry owns a program, decisions on such actionsare likely to be made quickly and in support of theprogram, making it more likely that the program willsucceed. In addition, ownership makes it easier togenerate domestic political support for the programbecause it will likely be seen at least partly as an in-digenous product, rather than a foreign imposition.

Ownership also matters for the catalytic role thatIMF lending can play in increasing a country’s ac-cess to foreign lenders. International flows of privatecapital have become increasingly important in re-cent decades, and a critical issue for borrowing in

4 There is no direct empirical evidence on the link betweenownership and IMF-supported programs. But some evidence onthe importance of ownership for project lending is provided bythe World Bank. The Bank’s Operations Evaluation Departmentrates government commitment to each project (measuring in asense the degree of ownership) using a variety of objective andsubjective indicators. The relationship between project out-comes and government commitment turns out to be strongly pos-itive (and statistically significant).

124


international markets is lenders’ ability to exercisecontrol rights. Foreign investors also confront moralhazard in teams: the payoffs to investments dependon the behavior and efforts of private borrowers aswell as of the government of the country where theprivate borrowers reside (Tirole, 2002).

Firms investing across borders design appropriatecovenants to mitigate moral hazard among privateborrowers. Control rights not vested with the in-vestor, but that affect borrower behavior, are actu-ally shared between the borrower and the govern-ment. Returns to a foreign investment depend onthe environment created by government policies ondomestic liquidity creation and management, taxand labor laws, and other institutional factors. Andwhen differences arise between a lender and a bor-rower, or in times of distress, treatment of parties toa contract is crucially affected by public attitudesand policies toward law enforcement, bankruptcy,and corporate governance.

Government policies that reduce the amount oftradables or other internationally valuable collateralalso hurt foreign investors. Such policies includetaxing exports, failing to invest in infrastructure(which inhibits exports and tourism), depreciatingthe domestic currency when foreigners hold assetsin that currency, depleting foreign exchange re-serves, and creating incentives for currency and ma-turity mismatches that increase credit and defaultrisks.

Whether such government moral hazard is impor-tant is an empirical issue and depends on the cir-cumstances. The hazard may be limited becausegovernments lose power and credibility after a crisis,and in such cases IMF conditionality associatedwith adjustment programs places constraints on theauthorities’ policy choices (De Gregorio and others,1999). But it may be cause for concern: as in privatefirms, the threat of losing one’s job after a crisis mayprevent misbehavior—but it may also increasemoral hazard by creating incentives to gamble forresurrection. Moreover, government actions that af-fect the mix of tradables and nontradables, andhence hurt foreign investors more than domesticones, are less likely to generate adverse reactionsfrom a country’s population (Tirole, 2002).

Consequently, country ownership of policies thatreduce moral hazard related to foreign investment islikely to be important for a country’s access to inter-national capital markets. Such policies provide as-surance to foreign investors that the governmentwill not devalue their claims and, in the event ofpoor performance or adverse shocks to borrowers,will not inhibit the transfer of control rights to cred-itors. Of course, governments cannot relinquishcontrol rights as easily as firms, but that simplymeans that in their case the set of transferable rightswill be more limited.

The feasibility of achieving a particular degree ofownership, and determining when it has beenachieved, are problematic issues that vary by coun-

try. A complicating factor in assessing the degree ofownership is that most countries, especially demo-cratic ones, have multiple stakeholders.

In pluralistic societies, does ownership refer to theviews on program design and objectives held by thekey ministers and central bank officials who negoti-ate the program with the IMF? Or does it refer tothe views of the domestic bureaucracy that has toimplement the program? Or to the views of the par-liament that has to approve the necessary legisla-tion? Or to the views of civil society at large? And ifthe views of civil society carry the most weight, howare its members to be assessed and made to influenceprogram design and implementation in the face ofcompeting interests? Thus ownership is intricatelyconnected to trust in domestic institutions, effec-tiveness of political structures, and whether the gov-ernment—negotiating on behalf of its citizens—hassufficient support to speak for a fair majority.

A widespread perception exists that programownership by borrowing countries is insufficient. Inrecent years the number of program objectives hastended to increase because the IMF has taken ontasks that go beyond its traditional mandate of es-tablishing macroeconomic and financial stability.5In many programs the number of structural objec-tives has been expanded to facilitate transitions tomarket economies, integrate domestic economieswith the world economy, diversify production andexports, develop financial sectors, and promotehigh-quality growth. In the 1990s, these programgoals were explicitly specified for transitioneconomies and made prominent for sub-SaharanAfrican countries. In addition, in the Mexican(1994–95) and East Asian (1997–98) crises, the keyroles played by financial sector fragilities and corpo-rate governance shortcomings expanded the list ofgoals. The problem is that country ownership is lesslikely when programs have too many objectives—because as the number of objectives increases, it be-comes less likely that the authorities and the IMFwill agree on the full range of objectives or on howthey are to be attained.

Borrowing countries may be partly responsible forthe lack of ownership. Some countries may be soeager for initial disbursements and for the catalyticrole of IMF financing that they are willing to agreeto programs without being convinced that the asso-ciated conditionality is appropriate. Such agree-ments have a greater chance of unraveling at criticaldecision points when it becomes clear that difficultpolicy measures are unlikely to be implemented. Inprivate markets, if a lender has serious doubts abouta borrower’s intentions or is not provided sufficient

5 Feldstein (1998), for example, argues that the IMF has beentoo intrusive in its interventions and should ask two questions ofeach conditionality measure. First, is the measure needed to re-store access to international capital markets? Second, would theIMF ask the same measure of a major industrial country if thecountry had a program?

125


collateral, the optimal course of action is not tolend. Given the IMF’s cooperative structure, andthe Samaritan’s dilemma it faces, it is much moreproblematic for the organization to refuse to lend toa member in need (see Drazen and Fischer, 1997).

New Initiatives to Foster Greater Ownership

A number of proposals have recently been made toenhance country ownership of IMF-supported pro-grams. Four such initiatives are considered here: en-couraging countries to design their own programs(specifically in the context of PRSPs), streamliningstructural conditionality, introducing flexibility inthe timing of structural policy measures (floatingtranche conditionality), and applying conditionalityto outcomes rather than policies (outcomes-basedconditionality).6 The IMF is already implementingsome of these proposals (see IMF 2001a, 2001b, and2001c).

Encouraging Countries to Design Their Own Programs

The IMF could require or encourage borrowingcountries to produce “home-grown” programs. Insome cases this may be seen as forcing ownership,but program designs can be worked out coopera-tively between the country and the IMF. If countrieslack the expertise and capacity to develop their ownprograms, the IMF could provide technical assis-tance and training or encourage the authorities tohire independent technical advisers.

While there are examples of home-grown pro-grams, for several reasons this approach has gener-ally not worked very well. First, countries often useoverly optimistic assumptions when designing theirprograms. For example, countries may underesti-mate the extent of their difficulties and overesti-mate the potency of their policy instruments.

Second, if the process of formulating a program—with its associated domestic political compro-mises—hardens the authorities’ negotiating positionwith the IMF, it will likely cause significant delays inprogram negotiation.

Third, countries may prefer to have the IMF pre-pare the program because they do not have, orchoose not to have, domestic mechanisms for mak-ing decisions on difficult trade-offs. This may re-quire the IMF to force issues requiring decisions.

And fourth, from a negotiating standpoint, coun-tries may want to see the IMF’s position before offer-ing their own in a program document.

To increase program ownership by low-incomecountries, the IMF and the World Bank recentlybegan encouraging the production of Poverty Re-duction Strategy Papers.7 The papers are intendedto specify and detail a country’s policies for reducingpoverty (see Ames, Bhatt, and Plant, 2002). PRSPshave three key elements:

• they are prepared by country authorities in con-sultation with various levels of government,local communities, civil society groups, donors,and multilateral agencies;

• they diagnose the country’s poverty situation,with the aim of identifying the main obstaclesto raising incomes and welfare; and

• they set goals for poverty reduction, define theimmediate and long-term policy actions neededto achieve those goals, and design a system formonitoring progress.

Recent experience with PRSPs suggests that theycould foster program ownership in several ways. (Fora comprehensive review of the PRSP approach andits early experiences, see IMF, 2002a and 2002b.) Bycreating a forum for dialogue—both within the gov-ernment and among other stakeholders in society—the papers can identify the concerns and views of af-fected groups as well as policymakers. Thus, theyshould enable policymakers and donors to better un-derstand the different facets of poverty and the mainconcerns of poor people. In this regard, the chal-lenge for governments is to strengthen democraticinstitutions and provide a voice to all domesticconstituencies. In many developing countries, itcannot be assumed that nongovernmental organiza-tions and civil society groups have the capacity to adequately participate in such a dialogue and engage in policy design, monitoring, and imple-mentation.

PRSPs should also lead to better, more systematiccollection of data and more effective monitoring. Asa result, they can be used to manage expectationsand set realistic goals, evaluate the trade-offs andconstraints involved, and develop priorities for pol-icy action. Again, increased ownership will requireensuring that countries have the technical capacityto formulate and analyze different policy paths. Inaddition, the effectiveness of PRSPs will depend onhow much they contribute to better governance andon how policy strategies deal with corruption.

By offering a coherent strategy for attackingpoverty, PRSPs could enhance coordination of pro-gram and nonprogram aid from multiple donors andmultilateral agencies. Bulír and Hamann (2001)show that for many countries aid is highly volatile

6 Other proposals not discussed here include preselecting coun-tries eligible for IMF lending, developing policy options for coun-try authorities to choose from, and investing time and effort inselling programs at home and abroad (see Khan and Sharma,2001).

7 Heavily indebted poor countries (HIPCs) seeking debt reliefand countries eligible to borrow under the IMF’s Poverty Reduc-tion and Growth Facility are required to produce a PRSP andhave it approved by the Executive Boards of the IMF and theWorld Bank before seeking new program support.

126


(more so than fiscal revenues), mildly procyclical,and not very predictable. As Svensson (2000) ar-gues, when donors are unable to monitor reform ef-forts, aid disbursements are likely to be tied to eco-nomic performance—and hence procyclical.

Aid volatility and unpredictability can compli-cate expenditure management, especially for gov-ernments with deficient fiscal infrastructure, result-ing in adjustments borne mostly by poorer andweaker segments of society. To the extent thatPRSPs help coordinate donor efforts and reduce thecost of mobilizing and using aid, they could con-tribute to the success of IMF and World Bank pro-grams—and thus enhance country ownership.

For the IMF and the World Bank, an importantaspect of facilitating country production of povertystrategies will be coordinating advice and technicalassistance to member countries. Such coordinationwill involve ensuring that the macroeconomic andfinancial conditionality attached to IMF lending isconsistent with the sectoral and project-based con-ditionality (for structural and social policies) at-tached to World Bank lending. Such consistencywill be achieved only if the IMF and the Bank agreeon their respective conditionalities, and then usethem in programs that respect—to the extent possi-ble—the overall strategies articulated by membercountries. A collaborative approach will enablecountries to access and use effectively the IMF’sPoverty Reduction and Growth Facility and theBank’s Poverty Reduction Support Credit Facility.

Streamlining Structural Conditionality

The past two decades have seen a major increase instructural conditions in IMF-supported programs(Goldstein, 2000; and IMF, 2001c). The expansionof structural conditionality would appear to limitthe scope for domestic policy choices, reducingcountry ownership. But there is considerable valid-ity for the expansion (IMF 2001a, 2001b, and2001c). Still, many structural reforms are of a micro-economic nature and therefore likely to be more in-trusive than macroeconomic policies. Country own-ership of programs is essential for the design andimplementation of these microeconomic measuresbecause they have a different impact on various seg-ments and vested interests in society.

It is widely felt that the IMF has gone too far withstructural conditionality and overloaded programswith structural measures. Many structural reformsare not required to achieve macroeconomic stabil-ity. There is also no evidence that programs withmore structural conditions have been more success-ful. In fact, programs with more structural condi-tions seem to have the same success rate as thosewith less (IMF, 2001c).

Increased structural conditions pose two maindangers. First, they may reduce country ownershipof programs and consequently impair their effective-ness. Second, a failure to implement structural re-

forms not critical for macroeconomic stability mayundermine confidence in the overall program—pos-sibly triggering reactions in domestic and interna-tional capital markets that could make overall pro-gram objectives harder to achieve.

It would be difficult and undesirable to turn backthe clock and eliminate all structural conditionsfrom programs. But careful thought should be givento which structural reforms are essential to achiev-ing a program’s main objectives. These reforms willvary by country, but sharply pruning the list of struc-tural conditions is possible without jeopardizing aprogram’s success or the IMF’s ability to be repaid. Inother words, prioritizing or streamlining structuralconditionality does not mean weakening overallconditionality.

The IMF has acknowledged that structural condi-tionality has expanded too much, and a major effortis under way to streamline it. IMF management re-cently defined broad principles for staff to use in de-termining the appropriate scope of structural condi-tionality in programs. In summary, the principles inthe Interim Guidance Note on Streamlining Struc-tural Conditionality (IMF, 2001b, Box 3) are as fol-lows:

• Structural reforms critical to achieving a pro-gram’s macroeconomic objectives generallymust be covered by IMF conditionality.

• Structural reforms relevant—but not critical—to a program’s macroeconomic objectives andwithin the IMF’s core areas of responsibilitymay be subject to conditionality.

• For structural reforms relevant to a program’smacroeconomic objectives but neither criticalnor in the IMF’s core areas of responsibility,conditionality generally should not apply.

These principles represent an important shift by theIMF from comprehensive to parsimonious structuralconditionality. Experiences with programs negoti-ated since the issuance of these principles will showwhether this intention is being achieved.

Adopting Floating Tranche Conditionality

Performance criteria and structural benchmarks inIMF-supported programs have specific dates at-tached to them. Countries often find that rigidtimetables for major structural reforms constraintheir choices and strain their implementation ca-pacity.8 Thus programs could be designed to allowfor greater flexibility in the timing of structural re-forms, increasing the scope for country ownership.

8 In many cases, rigid program schedules have posed seriousproblems for borrowing countries. One example involves the pas-sage of laws: some legislatures have been surprised to learn that aprogram has committed them not only to a specific legislativeagenda but also to a set of deadlines under which the relevantlegislative actions must be taken.

127


One way to achieve this goal is through floatingtranche conditionality for structural reforms. Underthis approach, IMF loan disbursements would not betied to specific dates. Instead, disbursements wouldbe made available upon completion of certainagreed reforms. This approach gives countries flexi-bility in the timing of program implementation.9 Italso allows for disbursements associated with the im-plementation of one part of a program to be un-linked from those associated with another part of aprogram.

The floating tranche approach could be used todivide conditionality into two parts. One part ofIMF financing could be conditional on achievingthe usual quantitative performance criteria under apredetermined schedule, while the other part couldbe dependent on implementing certain structuralreforms at any time prior to the expiration of theprogram (provided that the macroeconomic pro-gram stays on track). In the floating tranche part,the country would have control over when it under-took reforms—and assurances that when it did, itwould receive related funding. The IMF would beprotected because it would disburse funds only whenreforms were undertaken.

The segmentation of conditionality would requiredecisions on the proportion of IMF financing sub-ject to standard fixed tranche conditionality andthat subject to floating tranche conditionality. Suchdecisions would be based on judgments about therelative importance of a program’s different parts inachieving its overall objectives.

Not all structural reforms would be subject tofloating tranche conditionality, because some are es-sential to macroeconomic improvements. For exam-ple, an independent central bank could be consid-ered necessary to promote monetary stability, and aproper tax collection system might be needed toachieve fiscal discipline. The timing of such reformscould not be left open. In other words, in decidingwhich reforms are subject to fixed or floating trancheconditionality, the interdependence between struc-tural measures and macroeconomic managementwould have to be taken into account. Final decisionswould be made on a case-by-case basis and would bethe result of negotiations and agreements betweenthe IMF and the country authorities.

There is experience with a form of floatingtranche conditionality in the context of the HigherImpact Adjustment Lending (HIAL) initiative,which was introduced by the World Bank in sub-

Saharan Africa in 1995. The tranching innovationsunder the initiative have two objectives: to givegovernments more freedom in the timing of agreedreforms, thereby increasing ownership, and to re-duce pressure on the World Bank to disburse fundswhen conditions have not been met. These objec-tives are to be achieved through multiple butsmaller tranches, increased disbursements after theimplementation of reforms, and the introduction ofindependent floating tranche arrangements.

Prior to the HIAL initiative, adjustment loanstypically were disbursed in two tranches. HIAL in-troduced floating tranches, with single tranche op-erations as an alternative in special circumstances.Floating tranches are usually targeted at policy re-forms in certain sectors, and in some cases are in ad-dition to regular tranches. Under HIAL, floatingtranches have been applied to reforms involving thefinancial and banking sectors, the parastatal andpublic sectors, privatization, and civil service re-form. A tranche is released only when the structuralcondition is met, regardless of when that happens.Of the 21 HIAL operations in 17 African countriesduring 1996–98, about two-thirds used the newtranching mechanisms.

In 1999, the World Bank’s Operations EvaluationDepartment conducted an evaluation of these 21operations (World Bank, 1999). The study foundthat the HIAL initiative was associated with posi-tive policy outcomes in terms of fiscal adjustmentand exchange and interest rate policies. Moreover,the countries involved did better than nonpartici-pating comparator countries in increasing growth,lowering inflation, improving current account bal-ances, stabilizing foreign exchange reserves, andachieving more sustainable debt paths. Even thoughHIAL programs differ in other ways from otherWorld Bank programs, and the evaluation did nottake into account the role of exogenous factors, theresults provide some support for the use of floatingtranches.

Basing Conditionality on Outcomes, Not Policies

Outcomes-based conditionality involves condition-ing disbursements on the achievement of resultsrather than on the implementation of policies expected to eventually attain program objectives.Changing from policy-based conditionality to outcomes-based conditionality—leaving the choice ofpolicies to the country authorities—has been advo-cated by, among others, Díaz Alejandro (1984, p. 7):10

I propose that the international community shouldreturn to the key rationale for conditionality, andnegotiate with countries borrowing on concessional

9 In principle, prior actions in IMF-supported programs can bethought of as a variant of floating tranche conditionality. Acountry agrees to undertake certain measures before the program(or program review) is discussed by the IMF’s Executive Board.Thus the timing of the Board meeting and the disbursement offunds depend on the prior actions having been taken. Reviewscan also be considered a form of floating tranche conditionality,since their completion (and accompanying disbursement) de-pends on agreed policy measures being taken.

10 Spraos (1986) makes a similar point but on the grounds thatthe links between outcomes and policy variables are too tenuousto make policy-based conditionality especially meaningful.

terms only regarding the balance-of-payments targets, leaving to the countries the decision as towhat policy instruments should be employed toachieve them.

Under the outcomes-based approach, IMF condi-tionality would focus on objectives rather than pol-icy instruments and actions. Performance criteria forthe disbursement of funds would be based onwhether the targets for policy objectives wereachieved by set dates. The policy objectives wouldbe negotiated with the IMF, but the policy contentof programs would largely be left up to country au-thorities.

This approach is not as radical as it might seem,because programs already define outcome variablesas performance criteria. For example, IMF-supportedprograms include a floor on net international re-serves as a performance criterion. Similarly, theadoption of an inflation-targeting framework hasmade inflation one of the key variables monitoredin the IMF program for Brazil. Presumably, as morecountries adopt the inflation-targeting approach tomonetary policy, more programs will follow theBrazilian model. Other variables that could be sub-ject to outcomes-based conditionality include thetrade balance, the current account, investment, andgrowth.

In principle, there are two major benefits to thisapproach. First, the country authorities would be re-sponsible for designing policies to achieve desiredgoals (as long as the policies are not prohibited bythe IMF’s Articles of Agreement). Hence the au-thorities would bear the risk of success or failure.This approach would enhance country ownership byrequiring that the authorities and the IMF agreeonly on program objectives—and not necessarily onthe mechanisms linking these objectives to specificpolicies.

Second, funds would be disbursed only on the at-tainment of certain goals, providing incentives forthe country and the IMF to fashion appropriatepolicies. IMF resources would be safeguarded be-cause disbursements would depend on countriesachieving the desired results. If policies did not havethe envisaged outcomes, the country and the IMFwould have to rethink the economic strategy.

Implementing this approach would undoubtedlypresent a few challenges. First, outcomes-based con-ditionality could lead to the backloading of fundsthat may be needed earlier to fill a temporary liquid-ity gap or to finance structural reforms. It would alsolead to greater uncertainty for the country authori-ties about the availability of funds, since the agreedpolicies might not lead to the anticipated results.

Second, there may be significant lags in the report-ing of data on outcomes, particularly for the real sec-tor and for trade accounts. In addition, data on out-comes may be subject to frequent revisions, makingtimely monitoring and disbursements problematic.

Third, program outcomes are influenced not justby policies under the control of country authoritiesbut also by exogenous factors beyond their control.Though true in principle, however, it is not clearthat exogenous shocks create serious problems forprogram projections. For example, a recent study ofprogram projections by Musso and Phillips (2001)shows that projections for growth, inflation, and in-ternational reserves were accurate relative to simplerandom-walk projections. But projections for cur-rent and capital accounts did not outperform projec-tions from the random-walk model.

The more difficult question in this regard is, if ex-ogenous factors force a program to go off course, towhat extent should the authorities bear the risk offailure? And should there be some assurance of IMFdisbursement if it is judged that the authoritiesmade a good effort to attain the goals? Thus, evenoutcomes-based conditionality will require siftingthe evidence to determine whether outcome targetswere missed because of exogenous factors or becausethe authorities genuinely came up short—and if it isthe former, whether there would be a case for waiversas there is under policy-based conditionality.

Outcomes-based conditionality also raises thequestion of whether and when it would be feasiblefor the IMF to disburse funds based on promises toachieve certain goals if it had not had any influenceon the policy measures to attain them. In the pri-vate sector such condition-free lending is made onlyto blue-chip clients or those with good collateral,high net worth, or both. Similarly, the IMF wouldlikely provide such loans only to countries withstrong records of economic performance and man-agement, reputations for good governance, historiesof paying debts as contracted, and to those facingsituations that are not too dire.

Otherwise, giving borrowing countries completefreedom in their choice of policy actions would notprovide adequate safeguards for IMF resources. Evenfor the best clients, contract and loan covenants arelikely to provide complete freedom for policychoices only as long as the client’s capacity to repayis not impaired by endogenous or exogenous factors.To protect its resources, the IMF, like other lenders,will specify contingencies in the loan contract if theborrower’s health deteriorates—for whatever reason.

Two points can be made in response to the prob-lems raised regarding outcomes-based conditional-ity. First, even under outcomes-based conditionality,funds will be disbursed in tranches. For example, aprogram to correct an imbalance in the balance ofpayments could take multiple steps to achieve itsgoal of attaining a comfortable level of internationalreserves. The first tranche could be disbursed basedon a promise, but subsequent tranches would be re-leased only after the country had achieved certain lev-els of international reserves. In fact, even the releaseof the first tranche could require some prior actions.Hence, by splitting monitoring and disbursement

128


into smaller components, outcomes-based criteriacan simultaneously provide sufficient safeguards andprevent excessive backloading of financing.

Second, like any creditor, the IMF would com-bine outcomes-based conditionality with a monitor-ing system so that if a borrower’s position deterio-rated sufficiently, the IMF would intervene tocontain the damage, take prompt corrective action,and try to change the borrower’s strategy. As notedearlier, there would likely be ex ante ownership butex post conditionality. That is, a program mighthave little bite initially, but stricter constraintswould be imposed if certain events occurred. Andbecause all possible contingencies cannot be speci-fied ex ante, the IMF would design outcomes-basedprograms with the option of intervening, shoulddoing so be necessary to protect its resources.

Conclusion

This article has drawn on the well-established liter-ature on agency theory and finance to argue thatconditionality must apply to all IMF lending andthat country ownership of IMF-supported programsis essential. Because ownership matters for the suc-cess of these programs, the IMF and country author-ities share an incentive to create contracts thatmaximize ownership—subject to the safeguards thatthe IMF requires for its resources.

There is a widespread perception that countriesoften lack sufficient ownership of the programs theynegotiate with the IMF. Here it is important to dis-tinguish between countries in crisis and those withemerging imbalances. Crisis countries have lessroom for maneuver, and their problems require rapidresponses. In such cases ownership may be less im-portant—and, given the exceptional and extremecircumstances, agreements may be easier to reach.But in other countries the extra time spent on nego-tiations may be a necessary price to pay for betterprogram outcomes. In fact, different IMF facilitiescould cater to countries in different circumstances,and the scope and types of conditionality could dif-fer depending on the facility used.

Ownership can be enhanced by limiting the ob-jectives of IMF-supported programs—which wouldalso allow for more focused conditionality. If a pro-gram’s objectives are to establish and maintainmacroeconomic and financial stability, the range ofstructural measures required by IMF conditionalitywould be narrower. In its capacity as policy adviser,the IMF can advise on the merits of various structuralreforms. But in the context of programs, it shouldinclude only conditions that directly support macro-economic objectives. Encouraging the domestic for-mulation of programs, selling programs to multiplecountry stakeholders, discussing different policy op-tions, and increasing information flows would allhelp increase country ownership of programs.

To date the design of conditionality has focusedon policy actions rather than outcomes. This chap-ter has argued that there is merit in shifting the em-phasis toward outcomes-based conditionality andexploring the use of floating tranches, especially forstructural reforms. Outcomes-based conditionalitywould increase country ownership by giving the authorities greater discretion and flexibility inchoosing the policy mix and the timing of structuralmeasures. This increased leeway in program imple-mentation would be tied to explicit acknowledg-ment of ownership, improvements in data and re-porting (to facilitate monitoring by externalobservers), and acceptance of responsibility for pro-gram outcomes.

For the IMF, outcomes-based conditionality wouldhave to be combined with an agreed monitoring sys-tem for programs and the establishment of rules forborrower behavior. Such rules would be applied uni-formly and enforced through peer pressure and in-ternational norms—since the IMF has no recourseto legal action.

While a good case can be made for incorporatingoutcomes-based conditionality in IMF lending, thisis not an either-or matter. Programs would presum-ably combine policy-based and outcomes-based con-ditionality. The balance would depend on the coun-try’s circumstances, preferences, and economicproblems and on the accuracy with which differentpolicy actions and outcomes can be monitored(Dixit, 2000; Drazen and Fischer, 1997). Programswith such a balance would align IMF conditionalitymore closely with country ownership—undoubtedlythe shared goal of the IMF and its member coun-tries.

ReferencesAmes, Brian, Gita Bhatt, and Mark Plant, 2002, “Taking

Stock of Poverty Reduction Efforts,” Finance & De-velopment, Vol. 39 (June), pp. 9–12.

Anderson, Matthew S., 1966, The Eastern Question:1774–1923 (New York: St. Martin’s Press).

Boughton, James M. 2001, Silent Revolution: The Interna-tional Monetary Fund 1979–1989 (Washington: Inter-national Monetary Fund).

Buchanan, James M. 1975, “The Samaritan’s Dilemma,”in Altruism, Morality and Economic Theory, ed. by Ed-mund S. Phelps (New York: Sage Foundation).

Bulír, Ales, and Javier Hamann, 2001, “How Volatile andUnpredictable Are Aid Flows, and What Are thePolicy Implications?” IMF Working Paper 01/167(Washington: International Monetary Fund).

De Gregorio, José, Barry Eichengreen, Takatoshi Ito, andCharles Wyplosz, 1999, “An Independent and Ac-countable IMF,” Geneva Reports on the World Econ-omy: 1 (Geneva: International Center for Monetaryand Banking Studies).

Díaz Alejandro, Carlos F. 1984, “IMF Conditionality:What Kind?” PIDE Tidings (January–February),pp. 7–9.

129


130


Dixit, Avinash, 2000, “IMF Programs as Incentive Mech-anisms” (Princeton, New Jersey: Princeton Univer-sity).

Drazen, Allan, and Stanley Fischer, 1997, “Conditionalityand Selectivity in Lending by International FinancialInstitutions” (Washington: International MonetaryFund).

Feldstein, Martin, 1998, “Refocusing the IMF,” Foreign Af-fairs (March–April), pp. 20–33.

Ferguson, Niall, 1998, The House of Rothschild: Money’sProphets 1798–1849 (New York: Viking).

Goldstein, Morris, 2000, “IMF Structural Conditionality:How Much Is Too Much?” (Washington: Institute forInternational Economics).

Greenbaum, Stuart I., and Anjan V. Thakor, 1995, Con-temporary Financial Intermediation (Fort Worth, Texas:Dryden Press).

Guitián, Manuel, 1995, “Conditionality Past, Present,and Future,” Staff Papers, International MonetaryFund, Vol. 42 (December), pp. 792–835.

Haque, Nadeem Ul, and Mohsin S. Khan, 2002, “DoIMF-Supported Programs Work? A Survey of theCross-Country Empirical Evidence,” in Macroeco-nomic Management: Programs and Policies, ed. by M.S.Khan, S.M. Nsouli, and C.H. Wong (Washington:International Monetary Fund).

Holmström, Bengt, 1982, “Moral Hazard in Teams,” BellJournal of Economics, Vol. 13, pp. 324–40.

International Monetary Fund (IMF), 2001a, “Condition-ality in Fund-Supported Programs—Overview”(Washington: International Monetary Fund).

———, 2001b, “Conditionality in Fund-Supported Pro-grams—Policy Issues” (Washington: InternationalMonetary Fund).

———, 2001c, “Structural Conditionality in Fund-Sup-ported Programs” (Washington: International Mone-tary Fund).

———, 2002a, “Review of the Poverty Reduction Strat-egy Paper Approach—Early Experience with InterimPRSPs and Full PRSPs” (Washington: InternationalMonetary Fund).

———, 2002b, “Review of the Poverty Reduction Strat-egy Paper Approach—Main Findings and Issues forDiscussion” (Washington: International MonetaryFund).

James, Harold, 1996, International Monetary CooperationSince Bretton Woods (Washington: InternationalMonetary Fund and New York: Oxford UniversityPress).

———, 2001, The End of Globalization: Lessons from theGreat Depression (Cambridge, Massachusetts: Har-vard University Press).

Kaplan, Steven N., and Per Stromberg, 2000, “FinancialContracting Theory Meets the Real World: An Em-pirical Analysis of Venture Capital Contracts,”NBER Working Paper No. 7660 (Cambridge, Massa-chusetts: National Bureau of Economic Research).

Khan, Mohsin S., and Sunil Sharma, 2001, “IMF Condi-tionality and Country Ownership of Programs,” IMFWorking Paper 01/142 (Washington: InternationalMonetary Fund).

Killick, Tony, 1997, “Principals, Agents and the Failingsof Conditionality,” Journal of International Develop-ment, Vol. 9, pp. 483–95.

Kotowitz, Yehuda, 1987, “Moral Hazard,” in The New Pal-grave: A Dictionary of Economics, ed. by John Eatwell,Murray Milgate, and Peter Newman (New York:Norton).

Lindbeck, Assar, and Jorgen Weibull, 1988, “Altruism andTime Consistency: The Economics of Fait Accom-pli,” Journal of Political Economy, Vol. 96, No. 6,pp. 1165–82.

Meltzer, A.H., C.F. Bergsten, C.W. Calomiris, T. Camp-bell, E.J. Feulner, W.L. Hoskins, R.L. Huber, M.H.Johnson, J.I. Levinson, J.D. Sachs, and E.E. Torres,2000, International Financial Institution Advisory Com-mission Report (Washington: U.S. Congress).

Mishkin, Frederic, 1998, The Economics of Money, Bankingand Financial Markets (New York: Addison-Wesley).


Musso, Alberto, and Steven Phillips, 2001, “ComparingProjections and Outcomes of IMF-Supported Programs,” IMF Working Paper 01/45 (Washington:International Monetary Fund).

Polak, Jacques J., 1991, “The Changing Nature of IMFConditionality,” Essays in International Finance,Vol. 184 (Princeton, New Jersey: International Fi-nance Section, Princeton University).

Santaella, Julio A., 1993, “Stabilization Programs and Ex-ternal Enforcement,” Staff Papers, InternationalMonetary Fund, Vol. 40 (September), pp. 584–621.

Spraos, John, 1986, “IMF Conditionality: Ineffectual, In-efficient, Mistargeted,” Essays in International Finance,Vol. 166 (Princeton, New Jersey: Princeton Univer-sity).

Stiglitz, Joseph E., 2000, “The Contribution of the Eco-nomics of Information to Twentieth Century Eco-nomics,” Quarterly Journal of Economics, Vol. 115,No. 4, pp. 1441–78.

Svensson, Jakob, 2000, “When Is Foreign Aid Policy Cred-ible? Aid Dependence and Conditionality,” Journal ofDevelopment Economics, Vol. 61, No. 1, pp. 61–84.

Tirole, Jean, 2002, Financial Crises, Liquidity Provision andthe International Monetary Financial System (Prince-ton, New Jersey: Princeton University Press).

Williamson, John, ed., 1983, Fund Conditionality (Wash-ington: Institute for International Economics).

World Bank, 1999, “High Impact Adjustment Lending(HIAL): Initial Evaluation,” Operations EvaluationDepartment Report No. 19797 (Washington: WorldBank).

131

8

Much of the debate on the management of financialcrises has focused on structural and psychological issuesrelated to conditions that are supposed to be necessary torestore investor confidence. Nonetheless, the paramountrequirement in the short term is for countries in crisis toadopt correct macroeconomic policies. An analysis ofconventional macroeconomic models reveals that coun-tries can afford to run expansionary policies to restoreinternal balance only if they can afford to ignore the re-quirements for external balance. This arithmetic doesnot depend on whether macroeconomic policies were in-appropriate before the crisis hit.

Introduction

The recent proliferation of literature on interna-tional financial crises has demonstrated the variega-tion of causes leading countries to economic disas-ter. Once thought to be almost universallyoriginating from fiscal and other macroeconomicpolicy errors, financial crises are now generally un-derstood to emanate just as easily from externalshocks or from internal structural imbalances. Lessattention has been paid to the implications of thisdiscovery for post-crisis economic and financialmanagement. It has become fashionable to criticizethe IMF for applying univariate solutions to multi-variate problems, for being “an overbearing organi-zation with a well-thumbed book of macroeco-nomic-policy nostrums,”1 but that line of attackleads only to other questions. For example, if acountry is initially in fiscal balance, does it followthat fiscal contraction is an inappropriate responseto an external shock? More generally, does theuniqueness of each crisis and each country’s circum-stances imply that a unique response is required?

Most of the literature on this topic has focused ontwo issues. First, it has been argued that countrieswith a strong initial macroeconomic policy stanceshould be encouraged to use their room for maneu-

ver to respond to a negative shock with expansion-ary, not contractionary, policy. Feldstein (1998), forexample, has complained that the IMF mishandledthe 1997 crisis in the Republic of Korea. The IMF,he suggested (p. 29), responded with its

traditional … prescription of budget deficit reduc-tion … and a tighter monetary policy…, which to-gether depress growth and raise unemployment.But why should Korea be required to raise taxesand cut spending to lower its 1998 budget deficitwhen its national savings rate is already one of thehighest in the world, when its 1998 budget deficitwill rise temporarily because of the policy-inducedrecession, and when the combination of higher pri-vate savings and reduced business investment arealready freeing up the resources needed to raise ex-ports and shrink the current account deficit?

Similarly, Corden (2001) argued that the “first in-stinct of the IMF was for fiscal tightening. It ignoredthe good fiscal policies of the Asian governments.Obviously this was inappropriate and was proven tobe so, given the deep slumps that developed”(p. 55). Corden acknowledged the “limits to fiscalexpansion, set by the funds available to the IMF,”but he concluded that the strategy should have beento run as expansionary a policy as possible.

A second criticism, derivative to the first, is thatmacroeconomic policies should be tailored to theneed to restore market confidence. In Feldstein’sview of the Korean crisis, “the primary need was topersuade foreign creditors to continue to lend byrolling over existing loans as they came due…. Thekey … was to persuade lenders that Korea’s lack ofadequate foreign exchange reserves was a temporaryshortage, not permanent insolvency” (1998, p. 31).Corden took a more nuanced view, noting the de-pendence of confidence on both long-term fiscaldiscipline and “the need for temporary fiscal expan-sion,” with balance between these factors being casespecific (2001, p. 48).2

Different Strokes? Common and Uncommon Responses to Financial Crises

James M. Boughton

1 “Sick Patients, Warring Doctors,” Economist (September18–24, 1999), p. 81. The opposite criticism also surfaces withsome frequency. For example, Minton-Beddoes (1995) accusedthe IMF of “ad hoc improvisations” in violation of its own rulesand mandate, as a consequence of a “quest for relevance”(pp. 128–29).

2 For earlier examples of similar criticisms of IMF crisis man-agement, see Buiter (1990) and Killick (1995). Radelet andSachs (1998) also focused on the desirability of aiming the crisisresponse at restoring market confidence.

CHAPTER

132

8 � DIFFERENT STROKES? COMMON AND UNCOMMON RESPONSES TO FINANCIAL CRISES

The central role of investor psychology in finan-cial crises is indisputable and may be illustrated bytwo examples. First, consider the infamous case ofthe closing of 16 Indonesian banks in 1997 (seeEnoch, 2000). Faced with an incipient collapse ofthe banking system because of declining confidencein the banks’ soundness, the government acceptedIMF advice to close the weakest banks but resistedclosing all of those identified as insolvent. The argu-ment that the banks remaining open were soundenough to survive was partly financial and partly po-litical, but it also reflected a psychological judgmentthat depositors would recognize and accept the sys-temic improvement and would leave their money inthe open banks. The strategy failed disastrously, notbecause the financial analysis was wrong but be-cause it misjudged the requirements for reinstatingconfidence. Instead of aborting the panic, it cat-alyzed it, as depositors feared that the 16 closingsheralded a more general shutdown.

A second example may be drawn from the Mexi-can debt crisis of August 1982 (see Boughton, 2000and 2001). Mexico’s fiscal policy was unsustainablyexpansionary, but the newly elected governmentwas promising to correct the situation when it tookoffice in December. IMF staff urged the authoritiesto act earlier and more decisively, but as long as themajor international bank creditors were prepared tomaintain their exposure, there appeared to be noreason to panic. Why did the banks approve a largesyndicated loan to Mexico at the end of July andthen refuse to roll over other jumbo loans two weekslater? That catastrophic loss of confidence was nottriggered by a spate of bad news in the first half ofAugust, but by a reassessment of the bad news thathad been accumulating for several years. Should theIMF have responded more forcefully and more pub-licly? To answer that question requires a complexjudgment about market psychology and expectations.

Evaluating the policy implications of these psy-chological factors is not straightforward. Since theoutcome depends on expectations that are extrinsicto the model, analysts are confronted with the possi-bility of multiple equilibria.3 Nonetheless, countrieshit by economic or financial crises must adjustmacroeconomic policies, economists must advisethem on how to do so, and the IMF must base itslending decisions on the soundness and appropriate-ness of those policies. What do conventional mod-els tell us about how to make such judgments? Dothey give wrong answers because they omit impor-tant economic relationships?

This paper will abstract both from second-guessing the “animal spirits” of investors and from

analyzing the structural details of IMF-supported ad-justment programs in order to look more closely atthe underlying macroeconomics.4 As noted above, akey proposition in much of the criticism of the IMFresponse to the Asian crisis is that if fiscal policy wasacceptable before the crisis, it should follow that thetraditional prescription of budget tightening is inap-propriate and will only injure the economy. The dif-ficulty with that argument is that macroeconomictheory does not generate an optimum fiscal policy.All that theory tells us is that governments mustlimit the growth of debt so as to satisfy the intertem-poral budget constraint.

As long as current and future deficits are smallenough to rule out a long-run Ponzi game, a widerange of fiscal policies might be implied by prevail-ing economic conditions. In practice, depending oncircumstances, advocates of fiscal soundness haveappealed for governments to adopt policy goals suchas (a) balancing the budget over the business cycle,(b) placing a ceiling on the ratio of debt to output,or even (c) eliminating debt or a subset of it, at leastin net terms. No generally accepted macroeconomicmodel supports any such objective as an optimizingstrategy except under restrictive assumptions. In-stead, welfare maximization places fiscal policywithin a broader structure such as stabilizing outputgrowth at a sustainable (noninflationary) rate or,more generally, optimizing a social welfare function.In any such model, shifts in exogenous conditionsthat affect national saving or investment ratesmight well require a shift in the level and timing offiscal actions.5

One way to isolate the role of macroeconomics inthe debate is to examine the macro models that arein use at the IMF to see if they rule out certain plau-sible outcomes or omit key relationships that wouldgenerate those outcomes. The next section of thispaper summarizes the properties of three theoreticalmacro models that have been developed by IMFstaff to analyze the linkages between aggregate de-mand and external payments positions. That reviewis followed by some general conclusions about theuse of macroeconomic policies in response to finan-cial crises.

Theoretical Macro Models at the IMF

Of the numerous IMF-developed policy analysismodels, three are particularly apt for illustrating the nature of the recommended policy responses to

3 A growing segment of the finance literature endogenizes ex-pectations consistently with the predictions of the model so as togenerate multiple equilibria intrinsically. For a survey, see Mas-son (2001). The basic point for the problem at hand is that ex-pectations need not be irrational to generate multiple equilibria.

4 For overviews on the actual practice of IMF financial pro-gramming, see Fischer (1997) and Mussa and Savastano (1999).For detailed inside analyses of the IMF’s handling of the Asiancrisis, see Lane and others (1999) and Boorman and others(2000).

5 For an exposition, see Blanchard and Fischer (1989, espe-cially pp. 126–35 and 583–91).

133

James M. Boughton

internal or external shocks: Jacques Polak’s classicmodel of the monetary approach to the balance ofpayments, the “merged” model developed byMohsin Kahn and others to endogenize economicgrowth, and the Mundell-Fleming model, which wasmodified recently by Timothy Lane and others tostudy the effects of capital flows in the context ofthe Asian crisis. A common feature of these threemodels, at least in their simplest form, is that theytreat capital flows as exogenous and thus can beused to analyze small open economies that dependon externally rationed credit for financing the bal-ance of payments. Although the IMF may providecredits to offset part of a capital outflow, and theIMF’s involvement may induce other capital to flowback in, countries in financial crisis typically stillface a lower net inflow than they did before.6

These models have not been used directly to gen-erate performance criteria for IMF-supported adjust-ment programs. Rather, they have been used to gen-erate key insights on how such programs are orshould be designed—on the linkages between do-mestic credit creation and the balance of payments,between output growth and the current account, be-tween capital flows and the growth-payments nexus,and so forth. The extent to which more modernmacroeconomic concepts—notably intertemporaldynamics, time consistency, expectations theory,and the role of credibility—have been incorporatedin empirical program design is difficult to deter-mine.7

The Polak Model

The basis for financial programming and conse-quently for policy conditionality at the Fund is the“Polak model,” which was first developed in the1950s to link domestic credit creation to nationalincome and international reserves. In its simplestguise, as exposited in Polak (1998), the model maybe written as follows (with some modifications fromPolak’s notation):

∆L = k∆Y (1.1)

M = mY (1.2)

∆R= X – M + K (1.3)

∆L = ∆R + ∆D, (1.4)

where all variables are expressed as nominal values:

L = stock of moneyY = GNPR = international reservesM= importsX = exportsK = capital inflow to nonbank sectorD = domestic credit of the banking system.

It is useful to think of this as a “fiscal” version ofthe model. For a country with primitive financialmarkets and given values of X and K, an increase in the fiscal deficit will be financed primarily by domestic credit expansion, D. Alternatively, equa-tion (1.4) may be replaced by a slightly less aggre-gated financial sector:

∆L = q∆H (1.4´)

∆H = ∆R + ∆Dc, (1.5)

where H = the monetary base, or “high-poweredmoney,” and Dc = domestic credit of the centralbank. This “monetary” version is a little more trans-parent, since Dc is a directly observed policy instru-ment.

Obviously, a great deal of structure—fiscal policy,inflation, exchange rate adjustment, productivityshocks—must be added to this model before it canyield useful policy prescriptions. To that end, IMF staffhave developed or adapted many country-specificmodels that build on the simplified framework andincorporate the policy instruments and other vari-ables that are needed to explain each country’s par-ticular circumstances. Dc in this general model canbe thought of as a metaphor for the full range ofmacroeconomic policy actions. Expansionary policyis represented by an increase in Dc, and policy “ad-justment” is represented by a decrease. Similarly, Ris a metaphor for external stability, which in a morecomplete model would be represented by a combi-nation of reserve and exchange rate adjustment.

The policy implications of an exogenous shift incapital flows may be illustrated by solving the modelfor internal and external balance:

Because of the simplicity of the model, not onlydo all fiscal and monetary policies have similar ef-fects; so do all external shocks. A shift in capital

6 This analysis does not examine the rationality of assumingthat capital markets allocate credit to countries exogenouslyrather than on the basis of a market-clearing response to condi-tions in the country. For an analysis of the role of internationalcapital markets in generating and spreading financial crises, seeCalvo (2000 and 2004).

7 Two key elements that are not explicit in the class of modelsreviewed here are the effects of forward-looking expectations andof the policymaker’s credibility on the timing of policy responses;see Burnside, Eichenbaum, and Rebelo (2001); and Clarida,Gali, and Gertler (1999). Those factors are at least implicit inmany IMF-supported adjustment programs, even if they are notintroduced through a formal model; see the exchange betweenSebastian Edwards and Morris Goldstein in Edwards (1989).

Y =qm + k

kY + q X + K + D

R =qm

q

c

11−

−

( )∆

∆mm k

Dmk

qm kY

kqm k

X Kc+−

++

++−∆ 1 ( ).

flows is perfectly substitutable for an export shock,regardless of whether X responds to a shift in foreigndemand or to a supply shock in the export commod-ity or industry. In any event, the preferred policy re-sponse to such a shock depends critically on whetherthe goal is to restore internal or external balance. Tostabilize domestic output in the face of an exogenousloss of capital inflows (∆K<0), Dc must be increasedcommensurately. Conversely, to prevent the capitaloutflow from depleting net international reserves orforcing depreciation, Dc must be cut.

Much of the criticism and defense of the IMF in-volves a dialogue of the deaf. Critics such as Buiter,Corden, and Feldstein accuse the IMF of urgingcountries to tighten policies in circumstances whenexpansion would be needed to stabilize and restoreoutput; the IMF defends its advice on the groundsthat an initial contraction is needed to stabilize andrestore reserves or, more generally, to stabilize theexchange rate. This debate, however, has nothing todo with the structure of the underlying macroeco-nomic model.

To reconcile these two positions requires (a) arecognition by the critics that external stability andinternational reserves are important—regardless ofthe exchange regime—and that reserves cannot bereplenished purely through official financing with-out creating moral hazard problems,8 and (b) arecognition by the IMF’s defenders that output isimportant, not only for the obvious reasons but alsobecause a sharp drop in output can bring a loss ofconfidence that will aggravate the initial adverseshock. If an objective function that weighted outputand reserve losses were added to the model, the signon the reaction of Dc to K would become ambigu-ous.9 For countries with very low initial levels of re-serves, a perceived need for exchange rate stabilityto prevent disruption of production and trade, andpoor access to new sources of external financing, thenecessity of a cut in Dc remains clear. In other cases,viable policy options involve more flexibility.

The Merged Model

The Polak model alone does not provide guidanceon the effects of adjustment programs on output.

The potential rate of output growth is left unex-plained, as is the division of nominal income growthinto its real and inflationary components. In a seriesof papers written in the late 1980s, Mohsin Khan,Peter Montiel, and Nadeem Ul Haque attempted toendogenize real growth by merging the Polak modelwith the class of growth models used at the WorldBank.10 The following is a slightly simplified versionof the model presented in Khan, Montiel, andHaque (1990). The full model includes the ex-change rate, government consumption, and lump-sum taxes as policy instruments, in addition to do-mestic credit. It thus adds some of the structure thatis needed for practical financial programming. Inthis modification, the exchange rate and govern-ment consumption are fixed, taxes are proportionalto nominal income, all lagged and fixed exogenousvariables are set equal to unity, and second-orderterms are ignored.

∆L=k∆Y (2.1)

∆M=m∆y+b∆P (2.2)

∆R=X– M+K (2.3)

∆L=∆R+∆D (2.4)

∆Y=∆y+∆P (2.5)

Y– T– C – I=∆L – Kp – ∆Dp (2.6)

G–T=Kg+∆Dg (2.7)

∆X= –c∆P (2.8)

(2.9)

C=(1– s)(Y–T) (2.10)

T= tY (2.11)

K=Kp+Kg (2.12)

D=Dp+Dg . (2.13)

The new notation is mostly standard macroeco-nomic usage. D is debt to banks, and the subscripts pand g refer to the debt of the private and public sec-tors, respectively. Kp and Kg are capital flows tothese two sectors. The coefficients b and c are re-sponses of imports and exports to relative prices, andρ is the incremental capital-output ratio (ICOR).

Although this model is much more detailed thanthe basic Polak model, the principal differencecomes from the addition of equation (2.5), which

134


8 Replacement of private with official capital bails out privateinvestors, shifts the burden of risks, and poses new threats toother countries (potentially including those in remote parts ofthe world). Even if official capital could be mobilized in unlim-ited amounts for this purpose, the quality of the new equilibriumwould be fundamentally different from the old. The solution toan exogenous capital outflow therefore must involve recreatingthe conditions—economic, political, and psychological—thatinduced capital into (or to stay in) the country before the crisis.

9 Normally, output and reserves would both be lower in thenew short-term equilibrium. The question is not whether theeconomy would have to adjust, but whether a contractionary pol-icy action would be necessary to reach that point or whether pol-icy could be eased somewhat to soften the blow from the capitaloutflow.

∆∆

yI

P=

+ρ( )1

10 For a similar but less detailed extension of the Polak model,see Chand (1989).

allows the domestic price level to absorb some of theshock that would otherwise fall on output, andequation (2.9), which allows investment to raiseoutput via the capital stock. If the dynamics in thismodel were explicit, a sustained tightening of mon-etary policy would have less severe short-run effectsbut more severe long-term effects than in the basicmodel.

The merged model can readily be solved for threeequations representing roughly the real, monetary,and external sectors. Following Khan, Montiel, andHaque (1990), these equations are labeled (BB) (forthe real, or “[World] Bank,” side), (MM), and (BP).

ρ∆y=K+ ∆D (BB)

k(∆y+∆P)=∆D +∆R (MM)

∆R+m∆y+ b´∆P=K, (BP)

where b´≡ b+ c.11

As with the Polak model, this system may besolved for the requisite demand-management policy(∆D) in response to an exogenous capital outflow(–K), first assuming that the goal is to prevent out-put from falling and then assuming that the goal isto prevent reserves from falling. The first case isquite straightforward. From (BB), with ∆y=0,∆D = –K. The central bank must raise domesticcredit so as to compensate exactly for the outflow offoreign capital. Consequently, from the solution to(MM) and (BP), ∆R = K. That is, reserves fall by thefull amount of the outflow, and the current accountis unaffected.

Stabilizing output is not feasible if reserves are in-sufficient to compensate for the outflow (R0 < –K).Taking the alternative strategy of stabilizing reserves(∆R = 0), the solution is more complex.

(2.14)

(2.15)

The signs on these expressions are ambiguous anddepend particularly on the relative magnitudes ofthe ICOR (ρ) and the income elasticity of importdemand (m). As a benchmark, consider the situa-tion if the price elasticities of demand are identicalfor imports and exports (bN= 0). Then

(2.14´)

(2.15´)

In this situation, a capital outflow results unam-biguously in a fall in output, but it does not neces-sarily follow that a contractionary demand manage-ment policy is needed to stabilize reserves. Thecapital outflow itself pulls output down and thus re-duces demand for imports. If that effect is strongenough, reserves may be stabilized without any pol-icy adjustment. If bN> 0, the need for adjustmentmay be further reduced.

The merged model, in this comparative-staticsform, is a longer-term model. In the short run, thePolak model shows unambiguously that contractionis required to stabilize reserves. The implication ofthis extension is that additional temporary financ-ing may be a sensible substitute for demand contrac-tion as a means of stabilizing reserves until the ad-justment process represented by equations (2.14)and (2.15) is completed. Nonetheless, even if re-serves are stabilized temporarily through new fi-nancing with no change in macro policies, output(or at least income, via the real exchange rate) stillmust fall eventually in order to bring about the equi-librating drop in import demand.

The Modified Mundell-Fleming Model

Fund staff working on Asian economies in the after-math of the 1997 crisis developed a variant of theMundell-Fleming model to analyze the short- andmedium-term effects of exogenous capital flows(Lane and others, 1999, Appendix 7.2). This modelis also useful for explicitly endogenizing the domes-tic interest rate and the exchange rate. The struc-ture of this model is such that an exogenous capitaloutflow leads to a depreciation that generates thecurrent account surplus (or reduced deficit) neededto offset the outflow. In a fixed-exchange-rate sys-tem, a decline in output would normally be neededto generate that surplus, but here it will forceenough currency depreciation to get the needed re-sult. Output will actually rise as a corollary effect,unless the negative wealth effect of depreciation onconsumer demand is sufficiently large.

The model is readily understood as a single reduced-form equation and an equilibrium condition:

Y = C(Y,E) + I(r) + G + X(Y,E) (3.1)0 < CY<1, CE< 0, Ir<0, XY < 0, XE> 0

X= –K, (3.2)

where X= net exports (the current account balance)and E = the exchange rate (domestic-currency price

135

James M. Boughton

11 The (BB) equation derived from the full model contains anadditional RHS term, +[s(1–t)+ t–k]y. The equation has beensimplified here by assuming k = t+ s(1– t); that is, that saving isabsorbed at the margin by money demand. This simplificationmasks the role of shifts in saving propensities or tax rates, but itdoes not affect the analytical results discussed subsequently. SeeKhan, Montiel, and Haque (1990, pp. 172–73).

∆

∆

D =k( m + b )mk + b ( k)

K

y =(b + k

ρρ

− ′′ −

′ ))mk + b ( k)

K′ −ρ

.

∆

∆

D =m

K

y =m

K

ρ−

1

1.

of foreign exchange). Other variables are in conven-tional notation. Note that this, like the Polakmodel, is a fixed-price model. Because countries areassumed to have limited resources, international re-serves are also fixed (compare equation (3.2) withequation (1.3) or (2.3)).

Now consider the model’s comparative-staticsmultiplier effects. First, the fiscal multiplier, assum-ing that monetary policy controls r and that the lat-ter is fixed, is

In the conventional Mundell-Fleming model,CE = 0, so ε = 0. Otherwise, ε > 0, and the fiscal mul-tiplier is reduced but still positive. Qualitatively, thisis a standard Keynesian effect.

Second, the effect of an exogenous capital out-flow with fixed domestic policies but a flexible ex-change rate is

This result suggests that allowing the≡ capitaloutflow to generate depreciation while holding do-mestic interest rates and government spending fixedwill lead perversely to an increase in output. In thespecial case where CE= 0, a capital outflow has thesame effect as an expansionary fiscal policy. Capitalinflows thus have a Dutch-disease effect in thismodel: they cause the exchange rate to appreciateand weaken output. This effect is, of course, inconsis-tent with the conventional wisdom that developingcountries need capital inflows to grow. As a corollary,if output was initially on target, the appropriate fiscalresponse to a capital outflow is to cut spending.

The difficulty here is in the assumption that thedomestic interest rate (r) is determined solely bymonetary policy. Capital inflows should be allowedto reduce the cost of financing investment andthereby to raise aggregate supply as well as demand.The simplest way to model this relationship is to letdomestic investment be a positive function of capi-tal inflows.12 Then

Y = C(Y,E) + I(r,K) (3.1´)+ G + X(Y,E) Ik > 0

and

Now capital inflows raise output in two ways: viaa wealth effect from currency appreciation and a fi-nancial effect on capital investment. If we assumethat those two effects are large enough to makedY/dK > 0, while the wealth effect alone is not solarge as to nullify the normal Keynesian fiscal effect,then we have the expected comparative static re-sults. Third, using this expanded version, we cancalculate the required fiscal response to an exoge-nous capital outflow, holding r constant and lettingthe exchange rate depreciate.

Let η ≡ 1+CE /χE – IK.

If η < 0 (as needed to get dY/dK > 0), thendG/dK < 0. The appropriate response to an exoge-nous outflow is to raise government spending. Thatkeeps output on target, while the exchange rate de-preciates to generate the required strengthening ofthe current account. In terms of macroeconomics,this is the real core of the criticism of the IMF’s han-dling of the Asian crisis. Rather than contractingdomestic demand, a country facing a financial crisiscan simply let the exchange rate absorb the brunt ofthe attack.

The obvious response to this criticism is that therequired depreciation might be so large as to desta-bilize the economy, and it could trigger fears of acontinuing depreciation spiral. Tightening mone-tary and fiscal policies may be a more measured andbalanced response to an outflow. Which policy ispreferable in a particular set of circumstances can-not be determined from macroeconomic theory, butinsights could be gleaned from a more detailed em-pirical analysis of the country’s productive structure(see Furman and Stiglitz, 1998). Some industries,some agricultural sectors, and some financial firmswould gain from devaluation, and some would lose.Because both the balance of interests and investorexpectations vary from case to case, so does the opti-mum exchange regime.

Fourth, consider the effect of a capital outflowwhen monetary policy keeps the exchange rate fixedrather than the interest rate. This regime may re-quire allowing international reserves (R) to vary, asin the first two models reviewed above.

136


|(dr=dK=0) Y

E Y

E

dYdG C

C

=− +

=

11 ε

εχ

χwhere .

|(dr=dG=

E E

Y

dYdK

CC

dYdK

dYdG

0)

= −+− +

<− ≤

11

0

/ χ

ε

if CE E< χ .

12 An alternative approach is to assume that the negativewealth effect from depreciation exceeds the direct effect of de-preciation on the current account (|Ce|> Xe). That would cor-rect the sign on dY/dK, but it would also reduce the magnitudesof both dY/dG and dY/dK. If neither capital flows nor fiscal policyhas a sizable effect on output, the realism and relevance of themodel seem doubtful.

|(dr=dG= )

E E K

Y

dYdK

=C I

C

dYdK

>

0

11

0

−+ −

− +/ χ

ε

if C X IE E K .− >1

Then|dY=dr=

dGdK 0

= η.

X=∆R – K. (3.2´)

In the general case, where the goal is to stabilizeoutput, the interest-rate response (with G fixed) is

and the reserve response is simply dR = dK. Thus, asin the earlier models, a capital outflow requires a cutin interest rates (or a rise in government spending),and reserves must be allowed to fall by the fullamount of the exogenous outflow.

When output stabilization is not feasible and thegoal must be to restore external balance, outputmust be allowed to fall:

In this case, however, the sign of the policy re-sponse is ambiguous:

A capital outflow will require a hike in interest ratesto stabilize reserves unless the drag on investmentreduces import demand by enough to offset the con-ventional multiplier effect. While this secondary ef-fect might be quite strong in some cases, it is lesslikely that it would work quickly enough to preventa temporary depletion of reserves. In any situationwhere the initial reserve stock is vulnerable to at-tack, the central bank is unlikely to have time toallow the adjustment of investment and import de-mands to bring about a new equilibrium.

Implications and Conclusions

What can one learn from macroeconomic theoryabout the appropriate and viable policy responses toa financial crisis? Several implications emerge fromthe above review.

First, the fundamental explanation for differencesin view on this issue is found even in the simplestmonetary model. If a country has enough resourcesto withstand the shock of a sudden capital outflow,it can afford the luxury of aiming to stabilize outputin the face of the shock. The required resourcesmight come from holding a large initial stock of for-eign exchange reserves (as explicitly modeled here),from imposing controls or taxes on outflows, or fromacquiring replacement financing from externalsources or domestic saving. In any case, output sta-bilization will call for an expansionary monetary orfiscal policy under most circumstances. If the coun-

try lacks the necessary resources, contractionary ad-justment policies will be required to stabilize exter-nal flows unless currency depreciation is a viable al-ternative. Whether depreciation would bring largeror smaller welfare losses than would a contraction ofaggregate demand is a structural issue on whichmacro models offer little guidance.

Second, although the three models reviewedabove differ substantially in their coverage ofmacroeconomic relationships, they all lead to simi-lar conclusions about the need for conventionalpolicy adjustment in cases where external balanceis essential and the availability of financing is lim-ited. Perverse situations could arise, as illustrated bythe Dutch-disease implications of the modifiedMundell-Fleming model. Those situations, however,appear to be inconsistent with prevailing condi-tions in most emerging markets. As long as capitalinflows are growth stimulating, the withdrawal ofsuch flows is likely to require a standard adjustmentresponse.

Third, the argument that a country with a stronginitial fiscal position need not tighten in response toa financial crisis is contradicted by these models.What matters for macroeconomic equilibrium is notthe fiscal balance but the overall level of nationalsaving. A withdrawal of foreign capital requires anoffsetting increase in domestic capital to restore ex-ternal balance, irrespective of the initial fiscal posi-tion. In the crisis in the Republic of Korea, the ef-fect of the initial capital outflow on net saving wasoffset by a collapse of business investment, afterwhich the IMF approved a softening of the fiscal re-quirements in the program. The problem was not aninadequate macroeconomic framework but a limitedability to predict the effects of fiscal tightening oninvestor confidence.

Fourth, the structure that is omitted from thesemodels is primarily long term in nature. No role isgiven to shifts or differences in the distribution ofwealth across production sectors or income groups.No allowance is made for deficiencies in the qualityof governance or the sustainability of the naturalenvironment or exhaustible resources. It is not hardto construct arguments for less orthodox approachesto adjustment in the presence of those types ofstructural problems. A country with a weak govern-ment, a poor political system, an inefficient eco-nomic structure, a high incidence of poverty, or avulnerable physical environment may not be able towithstand the shock of a sharp economic downturn.Even if the downturn is a temporary by-product ofpolicy corrections that are expected to bring long-term economic and social benefits, the governmentmay not have enough credibility to convince poten-tial creditors and donors to support it through thetransition. Nonetheless, if the short-run require-ments for stability are paramount in a crisis, hetero-dox solutions that require time to work will be of nopractical value.

137

James M. Boughton

drdK

=II

>K

r

− 0

dYdK Y

= − >1

0χ .

drdK

C II

I

Y Y K Y

Y r

K Y

=− − +

−<

≥

( )

(

10

χ χχ

χunless 11− −CY Yχ ).

138


Beyond these specific observations, one may drawsome more general conclusions that may help clarifythe debate on policy responses.

First, at least two methods for examining themacroeconomic requirements for stability have va-lidity and are useful under carefully specified cir-cumstances: (a) find the required policy adjustment,taking the availability of financing as given; or(b) find the level of financing required to generate adesired outcome. The IMF uses primarily the former,while the World Bank and many outside observersfocus more on the latter. The suggestion that financ-ing is readily enough available to treat it as an en-dogenous equilibrating variable in the midst of a financial crisis is not obviously valid.

Second, the core of the debate over appropriatepolicy responses has less to do with macroeconomicsthan with investor psychology and national eco-nomic and political structure. To resolve such a de-bate requires better modeling of those processes andrelationships. Would structural optimization providean escape from the need to adjust macro policies,and escape from the trade-off between currency de-preciation and financial collapse? What are the re-quirements for restoring investor confidence? Whateffects would sudden shifts in income distributionhave on production and spending decisions? Theseand related structural questions cannot be answeredsatisfactorily by reference to the current generationof economic models.

Third, macroeconomic analysis does support theargument that responses to financial crises should betailored to each country’s circumstances, but not forthe reasons most commonly advanced in the litera-ture. What matters crucially in this context is notthe initial fiscal position but the short-term con-straints on external balance. If a country in crisiscan afford to sacrifice reserve or exchange rate sta-bility or can attract replacement financing quickly,it can—and must—expand domestic policies andforgo the more usual forms of adjustment.

ReferencesBlanchard, Olivier Jean, and Stanley Fischer, 1989, Lec-

tures on Macroeconomics (Cambridge, Massachusetts:MIT Press).

Boorman, Jack, Timothy Lane, Marianne Schulze-Ghattas,Ales Bulír, Atish R. Ghosh, Javier Hamann, AlexMourmouras, and Steven Phillips, 2000, “ManagingFinancial Crises: The Experience in East Asia,”Carnegie-Rochester Conference Series on Public Policy,Vol. 53 (December), pp. 1–67.

Boughton, James M., 2000, “From Suez to Tequila: TheIMF as Crisis Manager,” Economic Journal, Vol. 110(January), pp. 273–91.

______, 2001, Silent Revolution: The International Mone-tary Fund, 1979–1989 (Washington: InternationalMonetary Fund).

Buiter, Willem H., 1990, “Allocative and StabilizationAspects of Budgetary and Financial Policy,” in

Willem H. Buiter, Principles of Budgetary and FinancialPolicy (Cambridge, Massachusetts: MIT Press).

Burnside, Craig, Martin Eichenbaum, and Sergio Rebelo,2001, “Prospective Deficits and the Asian CurrencyCrisis,” Journal of Political Economy, Vol. 109, No. 6,pp. 1155–97.

Calvo, Guillermo, 2000, “Capital Market Contagion andRecession: An Explanation of the Russian Virus,” inWanted: World Financial Stability, ed. by Eduardo Fernandez-Arias and Ricardo Hausmann (Washing-ton: Inter-American Development Bank), pp. 49–52.

_______, 2004, “Contagion in Emerging Markets: WhenWall Street Is a Carrier,” in Latin American EconomicCrises: Trade and Labour, ed. by Enrique Bour, DanielHeymann, and Fernando Navajas (Houndsmills, Bas-ingstoke, Hampshire, England and New York: Pal-grave Macmillan), pp. 81–91.

Chand, Sheetal K., 1989, “Toward a Growth-OrientedModel of Financial Programming,” World Develop-ment, Vol. 17 (April), pp. 473–90.

Clarida, Richard, Jordi Gali, and Mark Gertler, 1999,“The Science of Monetary Policy: A New KeynesianPerspective,” Journal of Economic Literature, Vol. 37(December), pp. 1661–1707.

Corden, W. Max, 2001, “The World Financial Crisis: Arethe IMF Prescriptions Right?” in The Political Econ-omy of International Financial Crisis: Interest Groups,Ideologies, and Institutions, ed. by Shale Horowitz andUk Heo (Lanham, Maryland: Rowman and Little-field), pp. 41–61.

Edwards, Sebastian, 1989, “The International MonetaryFund and the Developing Countries: A Critical Eval-uation” (with comment by Morris Goldstein, andreply), in IMF Policy Advice, Market Volatility, Com-modity Price Rules, and Other Essays, ed. by KarlBrunner and Allan H. Meltzer (Amsterdam: North-Holland), Carnegie-Rochester Conference Series onPublic Policy, Vol. 31 (Autumn).

Enoch, Charles, 2000, “Interventions in Banks DuringBanking Crises—The Experience of Indonesia,” IMFPolicy Discussion Paper PDP/00/2 (Washington: In-ternational Monetary Fund), available on the Web athttp://www.imf.org/external/pubind.htm.

Feldstein, Martin, 1998, “Refocusing the IMF,” Foreign Affairs, Vol. 77 (March/April), pp. 20–33.

Fischer, Stanley, 1997, “Applied Economics in Action:IMF Programs,” American Economic Review, Vol. 87(May), pp. 23–27.

Furman, Jason, and Joseph E. Stiglitz, 1998, “EconomicCrises: Evidence and Insights from East Asia,” Brook-ings Papers on Economic Activity: 2, Brookings Institu-tion, pp. 1–114.

Khan, Mohsin S., Peter Montiel, and Nadeem Ul Haque,1990, “Adjustment with Growth: Relating the Ana-lytical Approaches of the IMF and the World Bank,”Journal of Development Economics, Vol. 32, pp. 155–79.

Killick, Tony, 1995, IMF Programmes in Developing Coun-tries: Design and Impact (London and New York:Routledge).

Lane, Timothy, Atish R. Ghosh, Javier Hamann, StevenPhillips, Marianne Schulze-Ghattas, and TsidiTsikata, 1999, IMF-Supported Programs in Indonesia,Korea, and Thailand, IMF Occasional Paper No. 178(Washington: International Monetary Fund).

Masson, Paul, 2001, “Multiple Equilibria, Contagion, andthe Emerging Market Crises,” in Financial Crises in

139

James M. Boughton

Emerging Markets, ed. by Reuven Glick, RamonMoreno, and Mark M. Spiegel (Cambridge, UnitedKingdom: Cambridge University Press), pp. 73–98.

Minton-Beddoes, Zanny, 1995, “Why the IMF Needs Re-form,” Foreign Affairs, Vol. 74 (May/June), pp. 123–34.

Mussa, Michael, and Miguel Savastano, 1999, “The IMFApproach to Stabilization,” in NBER Macroeconom-ics Annual 1999, ed. by Ben S. Bernanke and Julio J.

Rotemberg (Cambridge, Massachusetts: National Bureau of Economic Research), pp. 79–122.

Polak, Jacques J., 1998, “The IMF Monetary Model atForty,” Economic Modelling, Vol. 15, pp. 395–410.

Radelet, Steven, and Jeffrey Sachs, 1998, “The East AsianFinancial Crisis: Diagnosis, Remedies, Prospects,”Brookings Papers on Economic Activity: 1, BrookingsInstitution, pp. 1–74.

140

9

This paper assesses empirically the links among a coun-try’s institutions and political environment, its imple-mentation of IMF-supported programs, and macroeco-nomic performance in a sample of 197 programsapproved between 1992 and 2002. We find that astronger institutional and political environment is associ-ated with better macroeconomic outcomes, especially atlonger time horizons. This direct beneficial impact of in-stitutions on macroeconomic outcomes is in addition totheir indirect impact through better program implementa-tion. We also find that program implementation exertsan independent influence on macroeconomic outcomes,especially over shorter time horizons of up to two years.Better-implemented programs are associated with lowerinflation and with initially weaker but ultimatelystronger external and fiscal outcomes, but with a statisti-cally insignificant impact on economic growth.

Introduction

The quality of a country’s institutions—broadly de-fined to include the formal and informal rules ofeconomic and political interactions—is a key deter-minant of sustainable economic progress. Weak in-stitutions are behind many development failures(IMF, 2003; Rodrik, Subramanian, and Trebbi,2002; and North, 1997). Many developing and tran-sition economies participating in IMF-supportedprograms are bedeviled by corruption, weak and un-even enforcement of property rights, and other in-stitutional failings. These institutional drawbackshave increasingly become the focus of concern forinternational financial and development institu-tions, which have made structural reform the sub-ject of their conditionality.

Despite the significance of a country’s institutionsfor its macroeconomic performance, the empiricalliterature on IMF-supported reforms lacks a system-atic quantitative assessment of their importance.Most evaluations of IMF-supported programs simply

ignore the effect on macroeconomic performance ofvariation in institutions, either across countries orover time within a country. These evaluations arealso generally inconclusive. Although inflation, thebalance of payments, and the public finances seemto improve in countries that adopt IMF-supportedprograms, the impact on economic growth is am-biguous.2

The quality of a country’s institutions also shapesthe extent to which it succeeds in implementing itsIMF-supported programs. Program interruptions andthe uneven record of implementation of some IMF-supported programs are rooted, at least in part, inweak institutions in the countries making use of IMFresources. Until recently, detailed data on programimplementation were lacking. Most studies simplycaptured countries’ decision to participate in IMF-supported programs and did not consider how varia-tion in program performance affected macroeco-nomic outcomes. This is an important omission, be-cause a large proportion of IMF-supported programsare known to suffer major interruptions (see Ivanovaand others, 2003 and Mecagni, 1999). Recently, theliterature has begun to investigate quantitatively thelinks between institutions, program implementation,and macroeconomic performance (Ivanova and oth-ers, 2003; Joyce, 2003; and Dreher, 2004).

This paper sets out to measure the effect of varia-tion in institutional quality on the macroeconomicperformance of countries implementing IMF-supported programs. Building on the recent litera-ture, it develops a statistical framework to assess em-pirically the links among a country’s institutions andpolitical environment, its implementation of IMF-supported programs, and macroeconomic perfor-mance. We first update the results on program im-plementation reported in Ivanova and others, 2003by adding the outcomes of 25 recent programs to thesample of that paper. The qualitative results regard-ing program implementation remain qualitatively

Institutions, Program Implementation,and Macroeconomic PerformanceSaleh M. Nsouli, Ruben Atoyan, and Alex Mourmouras1

1 We thank our colleagues at the IMF and the University ofNorth Carolina at Chapel Hill for useful input when the projectwas under way. We are especially indebted to Alex Dreher,Shigeru Iwata, Timothy Lane, and Gene Leon for many helpfulsuggestions on an earlier draft. We alone are responsible for anyremaining errors.

2 This is understandable, given the econometric difficulties ofconstructing counterfactuals and the long and variable lags be-tween progress in microeconomic and structural policies and im-provements in economic performance. Haque and Khan (1998)discuss the problem of the counterfactual and the stylized facts ofthe macroeconomic impacts of IMF-supported programs.

CHAPTER

141

Nsouli, Atoyan, and Mourmouras

similar: the rate of program interruption continuesto be high, exceeding 40 percent. We then assessthe effect of variation in institutions on program im-plementation and macroeconomic performanceusing four indicators of performance: inflation, eco-nomic growth, the balance of payments, and the fis-cal balance. Our measures of program implementa-tion come from the IMF’s Monitoring of IMFArrangements (MONA) database, which containsdetailed information on a large number of IMF-supported programs approved since the early 1990s.Information on borrowing countries’ institutionsand domestic politics comes from the InternationalCountry Risk Guide (ICRG).

Our empirical framework is flexible, designed totake into account both the time-series properties ofmacroeconomic variables and the endogeneity ofprogram implementation. In the data, inflation,growth, and most other macroeconomic indicatorstend to be highly serially correlated and mean-reverting. We therefore examine the impact of insti-tutions, politics, and program implementation, tak-ing into account the autoregressive structure of themain macroeconomic and institutional variables,using a methodology suggested by the literature onthe error correction mechanism. Instrumental vari-ables are used to handle the endogeneity bias thatexists because macroeconomic shocks also impactprogram performance.

Our findings are mixed. When the endogeneity ofprogram implementation is properly accounted for,we find that institutions and program implementa-tion both matter for macroeconomic performance.The response of macroeconomic variables to pro-grams is often nonmonotonic, however. For example,although better-implemented programs are associatedwith lower inflation rates, the fiscal and external cur-rent account balances typically deteriorate for twoyears after program approval before they turn around.And, as in previous work, we could not detect statisti-cally significant associations between program imple-mentation and economic growth at any time duringthe three years following program approval.

The paper is structured as follows. The secondsection sets the stage for the empirical analysis. Itdescribes the measures of program implementationand institutional development used and presents de-scriptive statistics. The third section describes theeconometric methodology and main results. Thefourth section provides a conclusion.

Measuring Program Implementation and Institutional Development

Overview

Disbursements of IMF loans are tied to prior actionsby the recipient country, the observance of perfor-

mance criteria, and the completion of program re-views—and thus to fulfillment of conditionality.Breaches of conditionality, if not followed bywaivers because the breach was judged minor ortemporary, or by required corrective action to keepthe program on track, can lead to program interrup-tions. Following Ivanova and others (2003), we usetwo complementary measures of program implemen-tation. The first measure captures the premature“cancellation” of an IMF-supported program. Thisindex takes the form of a binary variable indicatingwhether a program experienced a major and irre-versible interruption. An “irreversible” interruptionoccurs when either the last scheduled program re-view was not completed, or all scheduled reviewswere completed but the subsequent annual arrange-ment was not approved. The second implementa-tion measure is the ratio of disbursements to com-mitments. It is a continuous variable indicating theshare of available IMF credit actually drawn. Thismeasure contains information on actual programduration and the extent to which the IMF’s finan-cial commitments under the program were fulfilled.

A variety of indicators can be used to assess theinstitutional and political setting in countries par-ticipating in IMF-supported programs. Based oncountry and time coverage and the ability to capturevarious aspects of governance, we choose to focus onthe ICRG political risk indicators, which allow us toascertain the short- and medium-term impacts ofthe political and institutional environment on eco-nomic performance and program implementation.Somewhat arbitrarily, we divide the 12 ICRG com-ponents into two groups. The first group proxies forbasic institutional quality, protection of propertyrights, and contract enforcement. It includes indicesfor the Investment Profile, Corruption in Govern-ment, Law and Order, and the Quality of the Bu-reaucracy. The second group serves as a proxy forpolitical outcomes. It is captured by the followingvariables: Government Stability, SocioeconomicConditions, Internal Conflict, External Conflict,Military in Politics, Religious Tensions, Ethnic Ten-sions, and Democratic Accountability. These vari-ables provide useful information about the internaland external political factors influencing programimplementation and economic performance.3

Descriptive Statistics

Table 1 updates the results on program implementa-tion presented in Ivanova and others (2003). In oursample of 197 IMF programs approved between1992 and 2002, 41 percent of all programs (includ-ing precautionary arrangements) experienced an irreversible interruption, compared with about44 percent reported in Ivanova and others (2003).

3 See the ICRG Guide to the Rating System for details(http://www.prsonline.com).

142

9 � INSTITUTIONS, PROGRAM IMPLEMENTATION, AND MACROECONOMIC PERFORMANCE

Countries with fewer program interruptions tend tohave higher disbursement rates: the correlation co-efficient between program interruptions and the dis-bursement share is –0.7. When precautionaryarrangements are excluded, the average disburse-ment share is approximately 75 percent, comparedwith 71 percent in the sample examined by Ivanovaand others (2003). The improvement in implemen-tation reflects the fact that our sample containsmore Stand-By Arrangements, which tend to havefewer interruptions than programs supported underthe Extended Fund Facility and the Poverty Reduc-tion and Growth Facility.

Improvements in the institutional climate, as re-flected in higher ICRG indicators, are generally pos-itively correlated with better program implementa-tion, as measured by higher disbursement shares andfewer program interruptions. As in earlier studies(Dollar and Svensson, 2000; and Ivanova and oth-ers, 2003), greater government stability and astronger investment profile in the year immediatelypreceding program approval are both associatedwith fewer interruptions (Table 2). On average, therisk of program interruption is much lower in envi-ronments in which governments are friendlier to in-

ward foreign investment and are better able to carryout their programs. Improvements in the invest-ment profile over a horizon of two years after the beginning of a program lead to significantly fewerinterruptions. Lower corruption (a higher ICRGscore) and improvements in socioeconomic condi-tions in the year after program approval are also as-sociated with better implementation of conditional-ity as measured by the quantitative implementationindex.

The political and institutional climate in coun-tries in which programs are interrupted varies sys-tematically from that in countries in which pro-grams are completed. Figure 1 plots the averagechange in each of the ICRG variables during theprogram relative to the last preprogram year, distin-guishing between interrupted and uninterruptedprograms. Program interruptions are associatedwith less progress in improving the investment cli-mate and in improving the quality of the bureau-cracy, and with intensified internal conflict. Suc-cessful program implementation is associated withgreater initial influence of the military, followed bya significant reduction in subsequent years. Pro-gram interruptions tend to be accompanied by sharp

Table 1

Program Implementation by Type of Arrangement

Including ExcludingPrecautionary Arrangements Precautionary Arrangements

ESAF/ ESAF/All EFF PRGF SBA All EFF PRGF SBA

Programs having irreversible interruptions1 41.12 40.00 45.31 38.89 42.77 34.78 45.31 43.06Number of observations 197 25 64 108 159 23 64 72

Quantitative implementation index2 79.18 87.21 77.09 78.52 79.36 86.95 77.09 78.85Number of observations 182 24 62 96 151 23 62 66

Structural implementation index3 66.37 73.98 70.97 60.54 68.41 76.54 70.97 62.44Number of observations 168 24 63 81 142 22 63 57

Overall implementation index4 74.29 83.27 72.91 72.81 74.81 83.71 72.91 73.45Number of observations 166 23 62 81 141 22 62 57

Share of committed funds disbursed1 62.05 72.56 80.02 48.47 74.54 78.87 80.02 68.02Number of observations 193 25 64 104 156 23 64 69

Source: IMF, Monitoring of IMF Arrangements (MONA) database. Notes: This table updates Table 1 in Ivanova and others (2003). Multiyear arrangements are treated as one program. Each cell contains the

average percentage value of the implementation index that is based on a sample of programs approved between 1992 and 2002. SBA standsfor Stand-By Arrangement; EFF stands for Extended Fund Facility; ESAF stands for Enhanced Structural Adjustment Facility; and PRGF stands forPoverty Reduction and Growth Facility.

1 The irreversible interruption index and the share of committed funds disbursed were computed as defined in the text.2 The quantitative implementation index for a given macro performance criterion is equal to 100 percent if macro performance criterion was

met or met after modification; and it is equal to zero if macro performance criterion was not met, not met after modification, waived, or waivedafter modification. The quantitative implementation index for a program is then computed as the average of those indices across all macro per-formance criteria for this program.

3 The structural implementation index for a given structural condition is equal to 100 percent if structural condition was met or met with asmall delay for structural benchmarks; it is equal to 50 percent if the structural condition was partially met or delayed for performance criteria;and it is equal to zero if the structural condition was not met. The structural implementation index for a program is then computed as the aver-age of those indices across all structural conditions for this program.

4 The overall implementation index for a given program is the average of quantitative and structural implementation indices over all condi-tions in this program.

143


Table 2Correlation of Program Implementation Indices with ICRG Risk Ratings at Different Horizons

Panel C. Panel A. Panel B. Correlation with

Correlation with Correlation with Change in Risk Ratings T–1 Risk Ratings T+1 Risk Ratings from T–1 to T+1

Irreversible Disburse- Irreversible Disburse- Irreversible Disburse-interruption ment interruption ment interruption ment

index share1 index share1 index share1

Bureaucracy quality –0.026 –0.127 –0.016 –0.072 –0.026 0.042Corruption –0.090 0.027 –0.006 –0.028 0.119 –0.078Democratic accountability –0.045 –0.089 –0.038 –0.062 –0.014 –0.034Ethnic tensions –0.004 0.050 –0.024 0.101 0.008 0.038External conflict 0.047 –0.183* 0.025 –0.032 –0.020 0.143Government stability –0.191** –0.033 –0.067 0.011 0.076 0.008Internal conflict 0.093 –0.160 0.025 0.006 –0.098 0.142Investment profile –0.199** 0.045 –0.274*** 0.167* –0.185** 0.058Law and order –0.025 –0.167* 0.034 –0.054 0.072 0.102Military in politics –0.065 –0.059 –0.019 –0.035 0.067 –0.047Religious tensions –0.090 0.028 –0.115 0.097 –0.068 0.102Socioeconomic conditions –0.032 0.162* –0.124 0.187 –0.093 0.049

Notes: The symbols *, **, and *** denote significance at the 90, 95, and 99 percent confidence levels, respectively.1 Correlation coefficients with the share of committed funds disbursed are computed excluding precautionary arrangements.

Change in ICRG External Conflict Index for Different Horizons

–0.4

–0.3

–0.2

–0.1

0

0.1

0.2

0.3

T T+1 T+2 T+3

Horizon

Cha

nge

in E

XTC

ON

F

majint=0 majint=100

Change in ICRG Government Stability Index for Different Horizons

0

0.5

1

1.5

2

2.5

3

3.5

T T+1 T+2 T+3Horizon

Cha

nge

in G

OVS

TAB

majint=0 majint=100

Change in ICRG Internal Conflict Index for Different Horizons

–0.6

–0.4

–0.2

0

0.2

0.4

0.6

0.8

T T+1 T+2 T+3

Horizon

Cha

nge

in IN

TCO

NF

majint=0 majint=100

Change in ICRG Military in Politics Index for Different Horizons

–0.15

–0.1

–0.05

0

0.05

0.1

0.15

T T+1 T+2 T+3

Horizon

Cha

nge

in M

ILIT

ARY

majint=0 majint=100

Figure 1

Quality of Institutions in Interrupted and Completed Programs

(Figure continues on next page)

144


Change in ICRG Investment Profile Index for Different Horizons

0

0.5

1

1.5

2

2.5


Cha

nge

in IN

VPRO

F

majint=0 majint=100

Change in ICRG Corruption Index for Different Horizons

–0.4–0.35–0.3

–0.25–0.2

–0.15–0.1

–0.050

0.05

T T+1 T+2 T+3

Horizon

Cha

nge

in C

ORR

UPT

majint=0 majint=100

Change in ICRG Bureaucracy Quality Indicator for Different Horizons

–0.15

–0.1

–0.05

0

0.05

0.1

0.15

0.2

T T+1 T+2 T+3

Horizon

Cha

nge

in B

URE

AU

majint=0 majint=100

Change in ICRG Law and Order Index for Different Horizons

0

0.05

0.1

0.15

0.2

0.25


Cha

nge

in L

AW

majint=0 majint=100

Change in ICRG Democratic Accountability Index for Different Horizons

00.050.1

0.150.2

0.250.3

0.350.4

0.450.5

T T+1 T+2 T+3

Horizon

Cha

nge

in D

EMO

ACC

T

majint=0 majint=100

Change in ICRG Ethnic Tensions Index for Different Horizons

0

0.05

0.1

0.15

0.2

0.25


Cha

nge

in E

THN

IC

majint=0 majint=100

Change in ICRG Socioeconomic Conditions Index for Different Horizons

–1

–0.8

–0.6

–0.4

–0.2

0

0.2

T T+1 T+2 T+3

Horizon

Cha

nge

in S

OC

IOEC

O

majint=0 majint=100

Change in ICRG Religious Tensions Index for Different Horizons

–0.25

–0.2

–0.15

–0.1

–0.05

0

0.05

0.1

T T+1 T+2 T+3

Horizon

Cha

nge

in R

ELIG

ION

majint=0 majint=100

Figure 1 (concluded )

145


increases in the military’s involvement in the thirdand fourth years after program approval.

Countries completing IMF programs appear to bemore successful in reducing inflation than countriesthat experience program interruptions, as reflectedin mean changes in macroeconomic outcomes between period T– 1 and five different horizons(Figure 2). Uninterrupted programs are also associ-ated with sharp improvements in fiscal balances inthe first year of the program, followed by a gradualdeterioration in subsequent years. On the otherhand, countries whose programs are interrupted reg-ister very modest improvements in fiscal balancesinitially but then catch up with the others. The ex-ternal current account balance improves in coun-tries whose programs do not get interrupted. Inter-rupted programs are associated with slightimprovements in the current account in the yearimmediately following program approval, followedby steady deterioration.

Correlation analysis and comparisons of macro-economic performance in completed versus inter-rupted programs, although suggestive, mask a greatdeal of variability in the data. Each country startsfrom different initial economic, institutional, andpolitical conditions. While they are engaged inIMF-supported programs, countries are subject to a

variety of external and internal shocks that influ-ence macroeconomic outcomes and program imple-mentation. We now turn to a more rigorous econo-metric methodology to properly take into accountthis broad spectrum of country-specific effects.

Econometric Methodology and Results

Methodology

Most empirical research on the macroeconomic im-pacts of IMF-supported programs relies on paneldata. Our framework, by contrast, relies on a pooleddataset in which each program is treated as an inde-pendent observation in the context of a statisticalmodel that takes into account the short-run autore-gressive, mean-reverting nature of macroeconomicvariables. We assess the impact of program imple-mentation on growth, inflation, and the fiscal andexternal balances once country-specific institu-tional and political effects are taken into account.Similar to the “before-and-after approach,” our ap-proach compares macroeconomic outcomes underthe program with those in the last preprogram yearfor a sample of countries that chose to participate in

Change in Current Account from T–1 to Horizon

–1.5

–1

–0.5

0

0.5

1

1.5

T T+1 T+2 T+3 T+4

Horizon

Perc

enta

ge C

hang

e in

BC

A/G

DP

with interruption, N=69without interruption, N=94

Change in Fiscal Balance from T–1 to Horizon

00.20.40.60.8

11.21.41.61.8

T T+1 T+2 T+3 T+4

Horizon

Perc

enta

ge C

hang

e in

GC

B/G

DP


Change in Growth Rates from T–1 to Horizon

00.5

11.5

22.5

33.5

44.5

T T+1 T+2 T+3 T+4

Horizon

Cha

nge

in G

ROW

TH, %


Change in Inflation Rates from T–1 to Horizon

–400

–350

–300

–250

–200

–150

–100

–50

0T T+1 T+2 T+3 T+4

Horizon

Cha

nge

in IN

FL, %


Figure 2

Macroeconomic Performance in Interrupted and Completed Programs

146


such programs. We do not address issues related tosample selection bias, i.e., the systematic differencesbetween countries that agree to participate in IMF-supported programs and those that do not. Our focusis on a narrower question. Given that certain coun-tries do self-select into the “treatment” of IMF-supported programs, we ascertain the relative im-pacts of program implementation and institutionson macroeconomic outcomes.

Consider Mi,T, a macroeconomic variable ob-served at time T in country i. Since we considermacroeconomic development only in countries im-plementing IMF-supported programs, the index i isalso a unique country-program identifier. FollowingAtoyan and others (2004), the evolution of Mi,Tcan be represented by

(1)

where Xi,T–1 is a vector of noninstitutional forcingvariables at time T–1 that also includes a randomterm in time T, IMPLi is the measure of program im-plementation in country i, INSTi,T–1 is a vector ofdomestic political and institutional initial condi-tions, ∆INSTi,T is a vector of contemporaneouschanges in country i’s political and institutional environment, and f(.) is the reduced-form data-generating process.

Hypothesizing that macroeconomic variables areinfluenced by their own values in previous periods,because of institutional or psychological inertia, weassume that growth, inflation, and the current ac-count and fiscal balances follow a finite-order autoregressive process. Assuming second-order au-toregression, the reduced-form model can be conve-niently written in the following form:

(2)

where ε i,TM is a stochastic disturbance to M. Equa-

tion (2) is the “autoregressive and mean-reversionform,” as it includes both lagged differences and thelagged level as the regressors for the current first-dif-ference of the variable M. It captures the autoregres-sive structure of M via the first-difference term∆Mi,T–1. The adjustment of M in response to devia-tions from its “normal” historical value is capturedvia the mean-reversion term Mi,T–1. The coefficientγ2 is a partial adjustment coefficient. It shows whatpercentage of the deviation from the long-run equi-librium will be covered each year following the de-viation. Note that because γ2=(β1+ β2 – 1) with β1and β2 representing the autoregressive (AR) para-meters in the underlying AR(2) process, small nega-tive values of the coefficient are consistent with Mbeing highly persistent.

Equation (2) holds for all T and i, including peri-ods in which IMF-supported programs are in effect.

As these programs are designed to improve macro-economic performance, equation (2) incorporatestheir impacts into the model. Equation (2) also cap-tures the impact of institutional and political condi-tions on the macroeconomy. In implementing equation (2), we treat institutional and political de-velopments during the IMF-supported program asexogenous and mean-reverting. Including first dif-ferences and lagged levels of institutional variablesin the regressions assumes that there are long-runlevels of institutional development and that devia-tions from these levels are temporary.4 This view ofinstitutions is certainly valid in analyzing short-termprograms such as Stand-By Arrangements. It isprobably less appropriate for programs with greaterstructural orientations, such as those supportedunder the Extended Fund Facility and the PovertyReduction and Growth Facility, which aim to im-prove the supply response of the economy. The na-ture of institutional change that takes place in thecontext of IMF-supported programs is ultimately anempirical question. In the event, there is little cor-relation in the data between program implementa-tion and institutional development (see Table 2,Panel C).5

We estimate the following system of four equa-tions, one for each macroeconomic outcome variable:

(3)

Note that, in equations (3), monetary and fiscalpolicies are kept in the background. Program imple-mentation serves as a proxy for the impact of macro-economic policies on macroeconomic outcomes. Asin Ivanova and others (2003) and Dollar and Svens-son (2000), the probability of implementation of an

4 Note that excluding contemporaneous changes in institu-tions from the model addresses a potential endogeneity problemthat is present if the error term affecting macroeconomic variablesalso affects institutional developments during the program period.

5 An alternative approach would be to run regressions withonly first differences of institutional variables. The qualitative re-sults remained unchanged when we reestimated our regressions inthis manner. This makes us confident that our results are robust.

∆ ∆M f X IMPL INST INSTi T i T i i T i T, , – , – ,( , , ,= 1 1 )),

∆ ∆M M Mi T i T i T, , – , –= + +γ γ γ0 1 1 2 1

, –+ +

+

γ γ

γ3 4 1IMPL INSTi i T

55∆INSTi T i TM

, , ,+ ε

∆ ∆INFL INFL INFLi TI I

i TI

i T, , – ,= + + −γ γ γ0 1 1 2 1

, –+ +γ γ3 4 1I

iI

i TIMPL INST

, ,+ +γ ε5I

i T i TIINST

GR

∆

∆ OOWTH GROWTHi TG G

i T, , –= +γ γ0 1 1∆

, –+ +γ γ2 1 3G

i TG

iGROWTH IMPL

, –+ +γ γ4 1G

i TINST 55

0 1 1

Gi T i T

G

i TB B

i T

INST

GCB GCB

∆

∆ ∆

, ,

, ,

+

= + +−

ε

γ γ γ22 1 3

4

Bi T

Bi

B

GCB IMPL, − +

+

γ

γ IINST INST

BCA

i TB

i T i TB

i TC C

, , ,

,

− + +

= +

1 5

0 1

γ ε

γ γ

∆

∆ ∆∆BCA BCA IMPLi TC

i TC

i, ,− −+ +1 2 1 3γ γ

., , ,+ + +−γ γ ε4 1 5C

i TC

i T i TCINST INST∆

IMF-supported program is related to the underlyingpolitical and institutional factors in the borrowingcountry, to the IMF financial and human resourceeffort in the program, and to initial economic condi-tions in the country. Although the probability ofprogram implementation is unobservable, it is re-lated to the observable implementation index:

(4)

In equation (4), the θs are vectors of coefficients.INITIALi is a vector of initial conditions repre-sented by the preprogram values of real GDP percapita, inflation, the GDP growth rate, the currentaccount balance, and the fiscal balance. FUNDi is avector of program-specific variables that are impor-tant in determining program outcomes. These vari-ables are either directly under IMF control or pro-vide information about the nature of therelationship between the country and the IMF. Ourregression approach in equation (4) is similar to thatused in Ivanova and others (2003).

Since we are interested in several potentiallymean-reverting macroeconomic indicators, a vectorerror correction model (VECM) could be consid-ered. In equation (2), M would then represent a4 × 1 vector of variables (inflation, growth rate, fis-cal balance, and current account balance). We pur-sued this approach by estimating the augmentedversion of the VECM and comparing results withthe ones obtained from estimating equations (3).The results confirm the existence of a long-run rela-tionship among some of the macroeconomic vari-ables.6 On the other hand, the marginal benefit ofincorporating this information into our analysis,which focuses on the relative importance of pro-gram implementation and institutional factors formacroeconomic outcomes, seems small. If a VECMrepresentation is adopted, the testing down ap-proach on the institutional and political factorsyields identical model specification to the one wealready have. The estimated coefficients and theirsignificance levels change only marginally relativeto those obtained by considering only a variable’sown autoregressive and mean-reversion terms. Tosimplify the presentation and economize on degreesof freedom, we do not present VECM results.7 Theseare available on request from the authors.

The properties of the ordinary-least-squares(OLS) estimator in equations (3) depend on thestochastic properties of the explanatory variables,and in particular on whether or not they are distrib-uted independently of the disturbance term. In addi-tion, shocks to macroeconomic outcomes are likelyto impact program implementation, implying thatCorr(IMPLi, εi,T

M ) ≠ 0. Consequently, the OLS esti-mator is likely to be biased.8

We employ two related instrumental variable (IV)techniques to correct for potential endogeneity bias.One is the two-stage-least-squares (2SLS) proce-dure, where we first regress the program implemen-tation measure on the exogenous variables and a setof instruments that are correlated with the imple-mentation measure but are not related to the errorterms in equations (3). In the second stage, we esti-mate the system of equations (3) by OLS using thepredicted values of the implementation measures in-stead of the actual ones.

In general, it is difficult to find instruments thatare related to program implementation but do notsystematically affect economic performance. Thebest candidates are variables that describe the na-ture of the relationship between member countriesand the IMF: a country’s quota in the IMF; the cu-mulative time spent in an IMF-supported program(number of months in program mode since 1980);the amount approved in relation to the country’sIMF quota; and the dollar cost of the program start-ing six months before program approval.9

Our second IV procedure is 3SLS. This has theadvantage of incorporating information from thecross-correlations of the error terms in equations (3)and producing sharper (more efficient) parameterestimates.10 To arrive at 3SLS estimates, we use the2SLS estimates to obtain an estimate of the contem-poraneous variance-covariance matrix of the errorsin equations (3). Applying the generalized-least-squares method to the transformed single-equationrepresentation of the system yields 3SLS estimates,which are consistent and asymptotically more effi-cient than 2SLS estimates.

Model Specification

There is a broad consensus that domestic institutionsand politics are key determinants of economic per-formance in countries borrowing from international

147


IMPL INST INITIAL

FUNDi i T i

i iI

= + +

+ +−θ θ θ

θ ε0 1 1 2

3

,

MMPL .

6 The coefficient on the lagged fiscal balance term is signifi-cant in the inflation equation. The lagged first difference and thelagged level of inflation are significant in the growth equation.The lagged first difference and the lagged level of the fiscal bal-ance significantly influence the evolution of the current account.

7 We formally test a set of restrictions that turns the VECM intoan autoregressive model. With an exception of the growth equa-tion, we cannot reject the null hypothesis that the data-generatingprocess was indeed just an autoregression (the p-values are 0.15,0.02, 0.15, and 0.40 for inflation, growth, the fiscal balance, andthe current account balance, respectively).

8 The OLS estimator is still useful in model selection because itis less sensitive than the alternatives to the presence of multi-collinearity, errors in variables, or misspecification, particularlyin small samples. After relying on OLS to choose an appropriatemodel, we compare its predictions with those from the samemodel estimated by alternative means.

9 An overidentifying restrictions test could not reject the nullhypothesis of overidentified restrictions for either implementa-tion measure.

10 A shock that affects economic growth has informationalcontent for inflation, the fiscal deficit, and the external currentaccount.

148


financial institutions. There is less agreement onprecisely which aspects of the institutional and po-litical environment are especially important. Al-though all the ICRG indices could be included inthe regression analysis, this would lead to collinear-ity problems and a loss of precision. On the otherhand, omitting relevant institutional and politicalvariables would lead to biased estimates.

This dilemma dictates a parsimonious approachto model specification. We use changes in macro-economic variables over a one-year horizon follow-ing program approval as a testing horizon. This im-plicitly assumes that if a certain institution orpolitical feature is important at high frequencies, itwill also be influential over longer horizons. Thisstrategy produces results that are robust with respectto the choice of program implementation measure.

Our model specification technology is describedas a “testing down” approach. We start with an un-restricted model that includes all ICRG indices asregressors and then simplify it in light of sample evi-dence. Specifically, we estimate each of the equationsin (3) separately while systematically dropping regres-sors with low t-statistics. The adjusted R2 is used as anadditional consideration in model selection.

Our results indicate that inflation in programcountries is influenced considerably by such institu-tional factors as the prevalance of law and order, thequality of the bureaucracy, and the country’s invest-ment climate. On the political side, only variationsin ethnic tension and internal conflict appear tomatter for inflation. Economic growth is affected bythe investment profile, government stability, andinitial socioeconomic conditions.11 Corruption inthe political system, democratic accountability, eth-nic tension, external conflict, and military involve-ment in politics are important for the evolution offiscal balances. Finally, corruption, ethnic tensionand external conflict, government stability, the in-vestment climate, and military participation in thecountry’s political life have significant impacts onthe evolution of the current account.

Main Findings

What Determines Program Implementation?

Table 3 presents first-stage regressions of the imple-mentation measures on initial economic conditions,ICRG indicators during the year preceding programapproval, and our instruments. To obtain the pre-dicted values used in the second-stage regressions,we employ the complete model (columns 1 and 3).However, to overcome the collinearity problem dis-

cussed above, the discussion in the rest of this sub-section relies on estimates (columns 2 and 4 ofTable 3) that drop some of the ICRG indices thatappear to be insignificant.

When the share of committed funds disbursed isused as a measure of program implementation, noneof the variables reflecting initial economic condi-tions is significantly different from zero. This couldsuggest that programs are tailored to participatingcountries’ circumstances or that their outcomes areindependent of initial economic conditions (seeIvanova and others, 2003).

Reduced ethnic tension and greater governmentstability before program approval improve programimplementation. Coefficients on the ICRG ratingsof ethnic tensions and government stability in theyear preceding program approval are positive andsignificant. In addition to a larger proportion offunds being disbursed in countries where racial andethnic tensions are less pronounced, better programimplementation is positively correlated with thegeneral public’s perception of a government’s abilityto carry out its declared programs. Other factors re-maining the same, a one-point increase in eitherrating raises disbursements by about 8 percent.

Reductions in internal conflict and improve-ments in law and order in the year before programapproval are associated with lower disbursements.The coefficient on the initial level of the internalconflict index is negative and highly significant.The magnitude of the effect is rather large: a one-point increase in the rating would lower disburse-ments by just over 6 percent. The coefficient on theinitial level of the law and order rating is also nega-tive and significant. These results may reflect theIMF’s financial involvement in countries where ob-servance of the law is not very good initially, oftenbecause the countries are recovering from conflict.

There is some evidence that greater initial in-volvement by the military in politics is associatedwith lower disbursements of IMF financing. The co-efficient on the corresponding ICRG index is posi-tive and significant.

Countries with a history of IMF-supported pro-grams seem to have higher disbursement shares.Every additional month spent in IMF-supported pro-grams translates into 0.2 percent more funds dis-bursed. Taken literally, higher disbursement ratioscould manifest better program design and implemen-tation, and the length of IMF engagement simply re-flects the long-term nature of the needs of these low-and middle-income countries. But the reasons for—and results of—prolonged financial association be-tween member countries and the IMF are complex(see IMF, 2002 for a recent evaluation).

The size of programs, as measured by the amountof IMF financing committed in relation to a coun-try’s quota, appears to be important in determiningprogram outcomes. Countries with larger programstend to have higher disbursement shares. These

11 Contemporaneous changes in socioeconomic conditions areexcluded from the analysis. This avoids the problems associatedwith dependent variables appearing on the right-hand side of theequation.

149


packages are often provided in response to capitalaccount emergencies. They require not only more fi-nancing but also greater front-loading of assistancethan suggested by usual IMF phasing rules.

The IMF’s effort at program design and imple-mentation, as measured by staff hours and the dollarcost of staff resources, is only marginally importantin raising a program’s prospects of success. Althoughlarger quotas have an ambiguous net effect on pro-gram implementation a priori (see Box 1), the coef-ficient on the country’s IMF quota is negative, sug-gesting that the implementation of IMF-supportedprograms could be weaker in countries with largerIMF quotas.

Our findings are broadly similar when the inter-ruption index is used as the measure of program im-

plementation. Almost all the variables describingthe initial economic conditions of participatingcountries have insignificant coefficients. The onlyexception is the lagged level of a country’s growthrate, which has a marginally significant coefficient.This can be interpreted as evidence that countriesthat were growing relatively fast before program ini-tiation are less likely to have an irreversible inter-ruption of the program.

Reduced government corruption has a strikinglypositive impact on the probability of successful pro-gram implementation. The coefficient on the pre-program level of corruption is positive and signifi-cant, and its magnitude is impressive. On average, aone-point improvement in the ICRG corruptionindex, all other determinants of program success

Table 3

Determinants of Program Implementation: First-Stage Regressions

Disbursement Irreversible Interruption Dependent Variable Share1,3,4 Index2,3,4

Regression number (1) (2) (3) (4)

Intercept 0.680** 0.848*** –0.397 –0.085Initial per capita real GDP –0.012 –0.013 –0.066 –0.072INFL (T–1) 0.0002 0.0002 0.002 0.002GCB_Y (T–1) 0.116 0.209 –3.509 –3.230BCA_Y (T–1) 0.091 0.049 –0.753 –1.198GROWTH (T–1) –0.002 –0.003 0.046 0.056*

Bureaucracy quality (T–1) –0.017 –0.070Corruption (T–1) 0.006 0.322* 0.358**Democratic accountability (T–1) –0.019 –0.104Ethnic tensions (T–1) 0.073** 0.076** 0.097External conflict (T–1) 0.006 –0.034Government stability (T–1) 0.071* 0.078** 0.044Internal conflict (T–1) –0.061** –0.062*** –0.221* –0.217**Investment profile (T–1) 0.021 0.090Military in politics (T–1) 0.041 0.044* 0.101Religion tensions (T–1) 0.019 0.144 0.189Socioeconomic (T–1) –0.008 0.040Law and order (T–1) –0.066 –0.069* –0.092

Fund effort per program year 0.041 0.043* 0.036 0.050Fund quota (log) –0.034 –0.046 0.095 0.052Number of months spent in IMF programs 0.002** 0.002** –0.002 –0.0003Amount approved as a fraction of quota 0.024** 0.025** 0.088 0.094Dummy for precautionary arrangement –1.121*** –1.109***Observations 115 115 115 115Log likelihood –14.695 –15.676 –61.998 –63.251Correlation coefficient/correctly

predicted (percent) 0.807 0.803 75.66 75.66

Notes: The symbols *, **, and *** denote significance at the 90, 95, and 99 percent confidence levels, respectively.1 Results are obtained using the tobit model: y =max(X´β+ε,0).2 Results are obtained using the probit model. Parameter estimates are computed to reflect the probability of no irreversible interruption:

Pr(Interruption = 0)= F(X´β), where F is normal cumulative distribution function.3 The chi–square statistics for the estimated parameters are available from the authors upon request.4 All regressions include year dummies.

150


held constant, coincides with a 35.8 percent betterchance of having no program interruption.

As in the regressions using the disbursement shareas the measure of program implementation, the co-efficient on the initial level of internal conflict isnegative and significant, and for similar reasons.The coefficient on the preprogram level of politicalviolence is negative: an improvement in this ratingby one point is associated with a 21.7 percent higherchance of an irreversible interruption.

With the exception of the coefficient on thenumber of months spent in program mode, the vari-ables characterizing the relationship between acountry and the IMF enter the regression with theexpected signs. However, none of the coefficients issignificantly different from zero. As in Ivanova andothers (2003), this result suggests that the imple-mentation of IMF-supported programs is largely de-termined by the country’s domestic political econ-omy and institutions. Variables under IMF controlhave only a marginal impact on program outcomes.

What Determines Macroeconomic Outcomes?

This subsection summarizes the empirical linksamong macroeconomic performance, the institu-tional and political environment, and program im-plementation (Tables 4 through 11). In all the re-gressions in these tables, the dependent variable isthe change in the macroeconomic outcome between

period T–1 (the preprogram year) and the end ofthe first, second, or third years after program ap-proval (T, T+1, or T+2). Each table reports OLS,2SLS, and 3SLS estimates, using the disbursementshare or lack of program interruptions as the measureof program performance. Unless otherwise noted, inwhat follows we will refer to results obtained usingthe 3SLS procedure and the disbursement share asthe measure of program implementation.

Inflation

Inflation is highly persistent in program countries.The coefficients on lagged inflation are highly sig-nificant for all horizons (Tables 4 and 5). For the av-erage program, about three quarters of any deviationfrom “normal” inflation after a program is approvedis reversed within a year. Deviations of inflationfrom its long-run equilibrium are erased almost com-pletely in three years.

In contrast to many other studies, which were un-able to link IMF-supported programs with price sta-bility, our findings represent reasonable evidencethat better program implementation leads to lowerinflation. After correcting for endogeneity bias, thecoefficients on the disbursement share have a nega-tive sign and decline in absolute value for each of thethree years following program approval, althoughonly the result for the first year is statistically

Box 1. List of Instrumental Variables

The outcomes of IMF-supported programs are en-dogenous. Instrumental variables (IVs) help us obtainunbiased estimates of the impact of IMF-supportedprograms on the economic performance of participat-ing countries. The instruments must be correlated withprogram implementation (lack of program interrup-tions and the share of committed funds disbursed) andnot be direct determinants of the economic policy out-comes (inflation, economic growth, fiscal balance, andcurrent account). The following IVs are used in theanalysis:

IMF quota (log). A country’s quota determines themember’s voting power in the IMF. Countries withlarger quotas have more bargaining power and systemicimportance in the world economy. Greater bargainingpower could allow countries to extract more conces-sions from the IMF, leading to less conditionality andmore lenient IMF treatment. The coefficient on theIMF quota in the implementation measure regressionswould then be positive. On the other hand, the size ofthe quota also reflects a country’s systemic importancein the world economy and its access to internationalcapital markets. Governments of large countries mightbe less cooperative with IMF conditionality if the per-

ceived political costs were too high. In that case, theparameter estimate could have a negative sign.

Number of months spent in IMF programs since 1980.This variable captures the extent of a country’s finan-cial involvement with the IMF. The length of thecountry’s history under IMF-supported programs couldlead, through learning-by-doing, to better program design and higher implementation rates as governmentofficials and IMF staff gain more experience andknowledge of country-specific factors and IMF proce-dures.

Amount approved as a fraction of IMF quota. This vari-able is expected to capture the financial importance ofa particular program. Large values would be positivelycorrelated with the severity of crises and the willingnessof the authorities to implement IMF-supported reforms.

IMF effort per program year, including six months priorto program approval. This is a direct measure of the dol-lar cost of IMF programs. It is computed from the IMF’sBudget Reporting System data on hours spent by staffon program implementation and estimated averagestaff salaries by grade. More effort invested in programimplementation is expected to be positively correlatedwith program implementation.

151


Table 4

Program Implementation, Institutions, and Inflation: Regressions Using the Share of Committed Funds Disbursed

Dependent Variable ∆INFL (T–1 to horizon)

Regression number (1) (2) (3) (4) (5) (6) (7) (8) (9)

OLS 2SLS 3SLS

Horizon T T+1 T+2 T T+1 T+2 T T+1 T+2

Disbursement share 13.862 –55.659 2.778 –140.87 –90.624 –9.000 –150.828* –103.740 –12.621

RGDPPC (T–1) –8.361** –6.406* –0.277 –10.622*** –7.222* –0.643 –10.602*** –7.013* –0.474

∆INFL (T–1) –0.158 0.0003 –0.001 –0.214** –0.002 –0.001 –0.180* –0.004 –0.002

INFL (T–1) –0.731*** –0.958*** –0.976*** –0.682*** –0.950*** –0.977*** –0.739*** –0.954*** –0.975***

Bureaucracy quality (T–1) –26.488** –12.571 –4.100 –27.744** –13.581 –4.317* –27.964** –12.012 –5.537***

Change in bureaucracy quality

(T–1 to horizon) –46.700 –11.681 –2.909 –73.494* –14.381 –4.211 –74.746** –14.256 –3.954

Ethnic conflict (T–1) –15.834 –5.164 –1.969 –6.632 –4.066 –1.299 –4.275 –2.152 –1.034

Change in ethnic conflict

(T–1 to horizon) –5.246 –12.083 –3.455 –11.568 –12.110 –4.060 –12.868 –12.904 –4.606*

Internal conflict (T–1) 13.151** 7.478 1.033 6.974 6.149 0.061 6.605 7.658 0.330

Change in internal conflict

(T–1 to horizon) 17.336* 11.734 –0.425 23.794** 12.419 –0.154 24.206** 13.982* –0.461

Investment profile (T–1) –9.834 3.493 –0.809 –2.854 3.715 –0.110 –2.784 3.456 0.487

Change in investment profile

(T–1 to horizon) –17.426 –3.723 –1.623 –19.212 –7.275 –1.851 –19.558 –9.327 –1.566

Law and order (T–1) 12.738 5.862 5.700** 14.470 8.730 7.348*** 12.342 5.639 6.603***

Change in law and order

(T–1 to horizon) –87.059*** –34.277 1.430 –86.529*** –30.469 2.516 –84.579*** –22.859 4.800*

Observations 123 122 116 115 115 109 115 115 109

R2 0.586 0.596 0.982 0.626 0.592 0.983 0.625 0.590 0.983

Notes: OLS denotes ordinary least squares; 2SLS denotes two-stage least squares; and 3SLS denotes three-stage least squares. The symbols *,**, and *** denote significance at the 90, 95, and 99 percent confidence levels, respectively. Standard deviations and t-statistics for the estimatedparameters are available from the authors upon request. All regressions include intercept, precautionary arrangement, and year dummies.

significant.12 A similar pattern is observed when thelack of program interruptions is considered as themeasure of program implementation. The absenceof program interruptions is correlated with greaterprice stability in the year following program ap-proval, followed by slightly higher inflation ratesover longer horizons. However, these results areonly marginally significant.

Better institutions also lead to lower inflation incountries implementing IMF-supported programs.13

Inflation is lower, the better is the government bu-reaucracy at the start of the program and the more itimproves subsequently. The importance of the qual-ity of the bureaucracy index is highest in the firstyear of the program and declines afterward. Inflation

is also lower, the more the legal system improves andthe more the public observes the law. Interestingly, ahigher degree of law and order before the start of theprogram and improvements in this regard during theprogram are associated with slightly higher inflationat horizon T+2.

The role of political factors in inflation perfor-mance in countries under IMF-supported programsis more difficult to interpret. Lower inflation is asso-ciated with increased political violence in the firsttwo years of the program. Tight demand-side poli-cies that succeed in reducing inflation could alsotrigger public protests against austerity, as has occa-sionally been the case in countries implementingIMF-supported programs.14

Recognizing that cross-country inflation regres-sions are dominated by outliers, we also examinewhether ethnic tensions and internal conflicts arestill the primary determinants of inflation whensuch observations are excluded from the sample. We

12 The inflation dynamics reported here are similar to those inConway (1994). Killick (1995) finds reduction in inflation to besignificant. Barro and Lee (2002) reports coefficients on contem-poraneous and lagged IMF loans that are similar in sign but in-significant.

13 Our findings on the impact of institutions on macroeco-nomic performance in program countries are robust to the choiceof implementation measure.

14 This cannot be formally tested in our model, because wetreat political variables as exogenous.

152


reestimate the model on a sample that excludes allobservations with annual change in inflation greaterthan 50 percent, which cuts the sample size by ap-proximately 30 percent. The results are somewhatreassuring. In the inflation equation, ethnic ten-sions still play an important role in determining theevolution of inflation. Internal conflicts become in-significant but, by contrast, government stabilityturns out to be significant. This is not very surpris-ing since the two indices are highly correlated in oursample.

Growth

Economic growth is highly serially correlated andmean-reverting during the course of IMF-supportedprograms (Tables 6 and 7). As in the case of inflation,deviations of the growth rate from long-run equilib-rium are very short lived. Approximately 90 percentof any deviation in growth rates from the country’s“normal” growth pattern is made up within three

years. The largest adjustment, 83–84 percent, occurswithin one year after the realization of the shock.

At first glance, better program implementationappears to be associated with more rapid economicgrowth, as suggested by positive and significant esti-mated coefficients in OLS regressions of the dis-bursement share. Unfortunately, this result is not ro-bust—it appears to be driven by the endogeneity ofprogram implementation. The corresponding 2SLSand 3SLS estimates are positive at all horizons, butthe parameters are not significantly different fromzero. In addition, the impact of program performanceon economic growth is fragile to the choice of imple-mentation measure. Although fewer program inter-ruptions appear to be associated with higher growthrates, the OLS results are not significant, and the coef-ficients turn negative when IV techniques are used.

These mixed findings are consistent with those ofthe existing literature. Recovery of growth ratesfrom the initial drop (a V-shaped response of output)was reported by Conway (1994). Khan (1990) andPrzeworski and Vreeland (2000) find significantly

Table 5

Program Implementation, Institutions, and Inflation: Regressions Using Irreversible Interruption Index

Dependent Variable ∆INFL (T–1 to horizon)


OLS 2SLS 3SLS


Non–interruption dummy 1.364 –18.019 3.400 –38.409 28.532 16.575* –32.152 29.726 13.730

RGDPPC (T–1) –7.688** –6.026* –0.262 –10.512*** –6.447 –0.329 –10.182*** –6.241 –0.195

∆INFL (T–1) –0.165 –0.002 –0.001 –0.207* –0.001 0.001 –0.172 –0.003 –0.001

INFL (T–1) –0.724*** –0.957*** –0.977*** –0.706*** –0.986*** –0.991*** –0.769*** –0.992*** –0.984***

Bureaucracy quality (T–1) –25.676** –14.196 –4.103 –25.351* –13.776 –4.125* –25.928** –12.195 –5.463**

Change in bureaucracy quality

(T–1 to horizon) –38.449 –13.225 –2.957 –63.094* –13.866 –3.911 –66.295* –14.974 –4.020

Ethnic conflict (T–1) –12.467 –8.963 –1.694 –10.522 –11.185 –2.043 –7.869 –10.586 –2.102


(T–1 to horizon) –5.445 –12.690 –3.344 –7.393 –10.723 –3.092 –8.472 –12.180 –3.885

Internal conflict (T–1) 12.615** 8.928 1.082 8.375 10.639 1.211 8.677 12.822 1.279

Change in internal conflict

(T–1 to horizon) 17.151* 12.357* –0.481 19.309* 12.131 –0.501 20.607** 14.480* –0.637

Investment profile (T–1) –8.569 1.037 –0.842 –4.825 –1.951 –1.231 –6.345 –2.885 –0.315


(T–1 to horizon) –16.780 –4.559 –1.737 –17.045 –7.490 –1.706 –18.050 –9.764 –1.355

Law and order (T–1) 11.033 8.005 2.480** 18.171 9.767 6.592** 15.386 6.772 6.285***

Change in law and order

(T–1 to horizon) –85.731*** –36.979 1.589 –83.905*** –34.745 2.187 –82.277*** –28.345 4.010

Observations 126 123 116 115 115 109 115 115 109

R2 0.582 0.589 0.982 0.617 0.589 0.984 0.614 0.587 0.983

Notes: OLS denotes ordinary least squares; 2SLS denotes two-stage least squares; and 3SLS denotes three-stage least squares. The symbols*, **, and *** denote significance at the 90, 95, and 99 percent confidence levels, respectively. Standard deviations and t-statistics for the esti-mated parameters are available from the authors upon request. All regressions include intercept and year dummies.

153


Table 6

Program Implementation, Institutions, and Growth: Regressions Using the Share of Committed Funds Disbursed

Dependent Variable ∆GROWTH (T–1 to horizon)


OLS 2SLS 3SLS


Disbursement share 2.814* 2.331* 1.412 2.607 0.484 2.921 2.579 0.639 2.309

RGDPPC (T–1) 0.209 0.169 0.551*** 0.277* 0.077 0.515*** 0.266* 0.072 0.534***

∆GROWTH (T–1) –0.205** –0.203*** –0.017 –0.249*** –0.248*** –0.023 –0.210*** –0.212*** –0.033

GROWTH (T–1) –0.850*** –0.785*** –0.946*** –0.829*** –0.758*** –0.931*** –0.840*** –0.753*** –0.896***

Government stability (T–1) 0.471 0.463 0.049 0.163 0.369 0.097 –0.025 0.299 –0.076

Change in government stability

(T–1 to horizon) 1.185** 0.364 0.025 1.022* 0.339 0.062 1.117** 0.305 –0.250

Investment profile (T–1) –0.578 0.216 0.634 –0.623 –0.049 0.463 –0.568 –0.042 0.462


(T–1 to horizon) 0.963* 1.079*** 0.841*** 1.228** 0.764** 0.849** 1.029** 0.678** 0.857**

Socioeconomic (T–1) –0.536 –0.607 –0.013 –0.525 –0.467 0.019 –0.420 –0.529 –0.036

Observations 125 124 118 115 115 109 115 115 109

R2 0.687 0.744 0.660 0.704 0.744 0.641 0.701 0.743 0.638

Notes: OLS denotes ordinary least squares; 2SLS denotes two-stage least squares; and 3SLS denotes three-stage least squares. The symbols*, **, and *** denote significance at the 90, 95, and 99 percent confidence levels, respectively. Standard deviations and t-statistics for the esti-mated parameters are available from the authors upon request. All regressions include intercept, precautionary arrangement, and year dummies.

Table 7

Program Implementation, Institutions, and Growth: Regressions Using the Irreversible Interruption Index

Dependent Variable ∆GROWTH (T–1 to horizon)


OLS 2SLS 3SLS


Non-interruption dummy 0.555 0.825 0.195 –0.443 –0.388 –0.321 –0.544 –0.529 –0.268

RGDPPC (T–1) 0.144 0.177 0.566*** 0.204 0.081 0.536*** 0.190 0.070 0.525***

∆GROWTH (T–1) –0.199** –0.208*** –0.039 –0.250*** –0.242*** –0.035 –0.211** –0.208*** –0.050

GROWTH (T–1) –0.857*** –0.789*** –0.931*** –0.829*** –0.757*** –0.920*** –0.842*** –0.759*** –0.898***

Government stability (T–1) 0.354 0.483 0.095 0.094 0.369 0.091 –0.103 0.300 –0.058


(T–1 to horizon) 0.894* 0.400 0.080 0.873 0.350 0.063 0.992* 0.326 –0.210

Investment profile (T–1) –0.378 0.137 0.480 –0.335 –0.127 0.374 –0.306 –0.085 0.334


(T–1 to horizon) 1.120** 1.013*** 0.784*** 1.372** 0.700** 0.736** 1.143** 0.620** 0.738**

Socioeconomic (T–1) –0.398 –0.639* –0.078 –0.481 –0.494 –0.049 –0.370 –0.543 –0.107

Observations 128 125 118 115 115 109 115 115 109

R2 0.677 0.736 0.649 0.692 0.742 0.631 0.690 0.740 0.629


154


negative effects of IMF program participation oneconomic growth. At the same time, Killick (1995),Bagci and Perraudin (1997), and Dicks-Mireaux,Mecagni, and Schadler (2000) report positive andsignificant effects. One possibility is that the extentof program implementation does matter for eco-nomic growth, but that the leads are greater thanthree years and therefore we have been unable tocapture them. Certainly the structural reforms ofmany programs in the 1990s took a long time tocome to fruition. Many countries—including transi-tion economies—began to experience faster growthonly in the late 1990s; such a delayed responsewould not be captured in our methodology.

Not surprisingly, improvements in institutionsduring the course of program implementation are as-sociated with better growth performance. This ismost evident in the case of the investment profile,which measures the risk to foreign business opera-tions in the country, including risk of repatriation ofprofits. A one-point increase in the ICRG Invest-ment Profile rating is associated with roughly a1 percent increase in the growth rate, and this result

is robust to the length of the horizon and the choiceof estimation technique. Improvements in the abil-ity of the government to stay in office, which are in-fluenced by the cohesion of the government and bythe extent of the public’s approval of its policies, ap-pear to have a significant positive impact on growth,at least in the first year of a program. These findingsare robust to the choice of implementation measureand to omitting outliers.15

Public Finances

The fiscal balance (in relation to GDP) is persistentand mean-reverting, but less so than inflation andgrowth. Improvements in the fiscal balance persistfor two years but are then reversed (Tables 8 and 9).This pattern could be consistent with governmentsimplementing IMF-supported reforms aiming to bal-ance their budgets over a four-year horizon. The

15 We define outliers as countries growing or shrinking by morethan 10 percent a year.

Table 8

Program Implementation, Institutions, and Fiscal Balance (Ratio to GDP):Regressions Using the Share of Committed Funds Disbursed

Dependent Variable ∆GCB_Y (T–1 to horizon)


OLS 2SLS 3SLS


Disbursement share 0.024** 0.020** 0.013 –0.013 –0.039** –0.021 –0.024 –0.038* –0.019

RGDPPC (T–1) 0.002** 0.0003 –0.0006 0.002** –0.001 –0.001 0.002* –0.001 –0.001

∆GCB_Y (T–1) 0.156* –0.349*** –0.131 0.161** –0.323*** –0.096 0.100 –0.321*** –0.059

GCB_Y (T–1) –0.623*** –0.523*** –0.434*** –0.651*** –0.497*** –0.527*** –0.451*** –0.472*** –0.558***

Corruption (T–1) –0.010*** –0.005 –0.006 –0.009** –0.006 –0.006 –0.006* –0.006* –0.006

Change in corruption

(T–1 to horizon) –0.008 0.005 0.002 –0.008 0.003 0.003 –0.003 0.002 –0.001

Democratic accountability (T–1) 0.001 0.002 0.007 0.002 –0.003 –0.001 –0.002 –0.002 0.002

Change in democratic accounta-

bility (T–1 to horizon) –0.008 –0.007* 0.0007 –0.009* –0.008** –0.004 –0.008* –0.008** 0.0004

Ethnic conflict (T–1) 0.001 –0.005* –0.004 0.001 –0.002 –0.001 0.001 –0.002 –0.002


(T–1 to horizon) –0.015** 0.0006 –0.005 –0.016** –0.003 –0.006 –0.015** –0.004 –0.007

External conflict (T–1) 0.004* 0.005* 0.004 0.002 0.003 0.003 0.003 0.003 0.004

Change in external conflict

(T–1 to horizon) 0.009*** 0.004 0.005 0.006* 0.002 0.001 0.006** 0.003 0.003

Military in politics (T–1) 0.004** 0.003 0.002 0.003 0.005** 0.003 0.003 0.004* 0.003

Change in military in politics

(T–1 to horizon) 0.006 0.0007 –0.005 0.008 –0.001 –0.006 0.001 –0.001 –0.005

Observations 119 118 112 115 115 109 115 115 109

R2 0.510 0.603 0.392 0.505 0.639 0.456 0.472 0.637 0.443

Notes: OLS denotes ordinary least squares; 2SLS denotes two–stage least squares; and 3SLS denotes three-stage least squares. The symbols*, **, and *** denote significance at the 90, 95, and 99 percent confidence levels, respectively. Standard deviations and t-statistics for the esti-mated parameters are available from the authors upon request. All regressions include intercept, precautionary arrangement, and year dummies.

155


mean-reversion term is highly significant. Approxi-mately 45 percent of any deviation of the fiscal bal-ance from its long-run average is offset within a year.The speed of adjustment is much slower than for in-flation or growth.

As in the regressions explaining growth, programimplementation appears to be associated with im-provements in the public finances when simultane-ity bias issues are ignored, but these results are re-versed in the regressions using IV approaches.Regardless of the choice of implementation mea-sure, the OLS estimates are positive and significantfor the first two years, whereas the 2SLS and 3SLSestimates are negative. If anything, better programimplementation seems to be associated with largerfiscal deficits: IV estimates of the coefficient on thedisbursement share two years after program ap-proval are significant.16 The results are similarwhen the lack of program interruptions measure is

considered. They suggest that fiscal deficits incountries with completed programs are about 3 per-cent larger than in countries whose programs wereinterrupted.

This finding and our similar finding for the cur-rent account balance (see the subsequent discus-sion) likely reflect the impact of additional financialresources flowing into countries that are successfulin implementing IMF-supported reforms. Better pro-gram implementation makes more financing avail-able to countries participating in IMF-supportedprograms, which allows more gradual adjustmentand larger fiscal and external deficits.

The most important institutional factor influenc-ing fiscal outcomes is the initial level of corruption,but its effect is anomalous. Lower corruption is asso-ciated with weaker fiscal outcomes over time. We donot have a good explanation for this result.

Several aspects of the political environment playan important role in determining fiscal outcomes incountries with IMF-supported programs. First, im-provements in the government’s responsiveness toits people are associated with larger deficits. This

Table 9

Program Implementation, Institutions, and Fiscal Balance (Ratio to GDP):Regressions Using the Irreversible Interruption Index

Dependent Variable ∆GCB_Y (T–1 to horizon)


OLS 2SLS 3SLS


Non-interruption dummy 0.016*** 0.014** 0.008 –0.010 –0.023 –0.035* –0.022 –0.020 –0.032*

RGDPPC (T–1) 0.002* 0.001 –0.001 0.002* –0.001 –0.001 0.002 –0.001 –0.001

∆GCB_Y (T–1) 0.190** –0.354*** –0.141 0.168** –0.308*** –0.089 0.099 –0.310*** –0.057

GCB_Y (T–1) –0.576*** –0.534*** –0.433*** –0.632*** –0.508*** –0.522*** –0.436*** –0.484*** –0.558***

Corruption (T–1) –0.012*** –0.007* –0.007 –0.008* –0.003 –0.002 –0.004 –0.004 –0.004

Change in corruption (T–1 to horizon) –0.008 0.003 0.001 –0.007 0.003 0.004 –0.002 0.001 –0.0003

Democratic accountability (T–1) 0.002 0.003 0.007 0.002 –0.003 –0.002 –0.002 –0.002 0.002

Change in democratic accountability (T–1 to horizon) –0.006 –0.008* 0.0002 –0.008 –0.008** –0.003 –0.007* –0.008** 0.00003

Ethnic conflict (T–1) 0.002 –0.004 –0.004 0.002 –0.003 –0.002 0.001 –0.003 –0.002

Change in ethnic conflict (T–1 to horizon) –0.014** 0.001 –0.005 –0.016** –0.004 –0.006 –0.015** –0.004 –0.007

External conflict (T–1) 0.003* 0.005** 0.005 0.002 0.003 0.002 0.003 0.003 0.004

Change in external conflict (T–1 to horizon) 0.008*** 0.005* 0.005 0.007** 0.002 0.001 0.007** 0.003 0.003

Military in politics (T–1) 0.005** 0.003 0.002 0.003 0.004** 0.003 0.004* 0.004* 0.002

Change in military in politics (T–1 to horizon) 0.008 0.003 –0.004 0.008 –0.002 –0.006 0.001 –0.001 –0.005

Number of observations 122 119 112 115 115 109 115 115 109

R2 0.499 0.609 0.392 0.492 0.634 0.471 0.458 0.631 0.459


16 Schadler and others (1993) also find some evidence of nega-tive effects of IMF lending on the fiscal balance. By contrast,Conway (1994) finds significant fiscal deficit reduction.

156


could be evidence that democratic incumbents tendto postpone fiscal consolidation. Second, declines inethnic tension are contemporaneously correlatedwith improved fiscal balances. This could reflect acountry’s return to normalcy, which is associatedwith improved revenue collection and lower mili-tary spending. Third, less military involvement inpolitics in the preprogram year, as well as declines inthe risk of external conflict (ranging from trade re-strictions to full-scale warfare) are positively andsignificantly associated with lower fiscal deficits.

Current Account

Shocks to the current account are longer lived andhave larger permanent components than othermacroeconomic outcomes (Tables 10 and 11). Onlyabout 10 percent of any deviation from a country’s“normal” ratio of the current account to GDP ismade up for in one year.

Most studies find that participation in IMF-supported programs helps improve the current ac-count. Our results on the impact of program imple-mentation on the current account are more nuanced.Countries that do a better job at implementing pro-grams experience a deterioration of the current ac-count for about two years, but this is followed by asharp improvement in the trade balance for thethird year. Disbursement of 100 percent of commit-ted funds is accompanied by an 8 percent deteriora-tion of the current account in the first year (relativeto the preprogram year), followed by a numericallynoticeable but statistically insignificant 2 percentimprovement in the third year. Our mixed resultsare similar to Barro and Lee’s (2002). By contrast,Conway (1994) finds evidence of improvement inthe current account in countries participating inIMF-supported programs, but it does not correct forthe extent of program implementation.

The only institutional variables that matter forthe current account are the initial investment

Table 10

Program Implementation, Institutions, and Current Account (Ratio to GDP): Regressions Using the Share of Committed Funds Disbursed

Dependent Variable ∆BCA_Y (T–1 to horizon)


OLS 2SLS 3SLS


Disbursement share 0.008 0.022 0.060*** –0.082** –0.039 0.024 –0.080** –0.038 0.020

RGDPPC (T–1) –0.0008 –0.001 0.0006 –0.002 –0.003 –0.0001 –0.002 –0.003 –0.0001

∆BCA_Y (T–1) –0.145 –0.172* –0.118 –0.215** –0.195* –0.155 –0.160* –0.190* –0.157

BCA_Y (T–1) –0.157** –0.217** –0.371*** –0.149** –0.125 –0.394*** –0.114* –0.143* –0.435***

Corruption (T–1) –0.011 –0.005 –0.014* –0.007 –0.002 –0.016** –0.006 –0.002 –0.017**

Change in corruption (T–1 to horizon) –0.002 –0.005 0.009 0.002 –0.001 –0.002 0.002 –0.003 –0.002

Ethnic conflict (T–1) –0.008* –0.008 –0.002 –0.006 –0.007 0.001 –0.006 –0.007 0.001

Change in ethnic conflict (T–1 to horizon) –0.010 –0.002 –0.008 –0.016* –0.016* –0.006 –0.016* –0.015 –0.006

External conflict (T–1) 0.003 0.0001 –0.0001 0.001 –0.003 –0.003 0.001 –0.002 –0.003

Change in external conflict (T–1 to Horizon) 0.008 –0.0008 –0.005 0.003 –0.002 –0.010** 0.004 –0.002 –0.012**

Government stability (T–1) 0.014*** 0.013* 0.010 0.009* 0.010 0.002 0.008* 0.010 0.002

Change in government stability (T–1 to horizon) –0.004 0.005 –0.002 –0.005 0.005 –0.007 –0.003 0.004 –0.007

Investment profile (T–1) –0.014*** –0.013** –0.009* –0.009** –0.010* –0.006 –0.008** –0.011** –0.006**

Change in investment profile (T–1 to horizon) –0.016*** –0.018*** –0.007* –0.016*** –0.015*** –0.003 –0.019*** –0.016*** –0.003

Military in politics (T–1) 0.011*** 0.012*** 0.010** 0.007** 0.010** 0.008* 0.007** 0.010** 0.008*

Change in military in politics (T–1 to horizon) 0.004 0.003 –0.002 0.006 0.003 –0.001 0.005 0.004 –0.001


R2 0.392 0.379 0.443 0.460 0.407 0.451 0.460 0.406 0.449

Notes: OLS denotes ordinary least squares; 2SLS denotes two-stage least squares; and 3SLS denotes three-stage least squares. The symbols*, **, and *** denote significance at the 90, 95, and 99 percent confidence levels, respectively. Standard deviations and t-statistics for the esti-mated parameters are available from the authors upon request. All regressions include intercept, precautionary arrangement, and year dummies.

157


profile and its change during the program period.Both are highly significant and enter the regressionswith negative signs. Not surprisingly, the better agovernment’s attitude toward inward investment,the larger the current account deterioration duringthe period considered.

Of the political variables, the ones relevant forthe evolution of the external current account areexternal conflict, government stability, and militaryinvolvement in politics. The coefficient on thechange in the external conflict index is negativeand highly significant for the T+2 horizon. Im-provements in the index are associated with theelimination of embargoes and of trade restrictionsand are correlated with a worsening of the currentaccount. A one-point increase in this rating is corre-lated with a 1.2 percent deterioration of the currentaccount over three years. Governments that aremore stable in the preprogram year tend to have

better current account performance. Similar posi-tive effects on the current account appear to resultfrom less military involvement in politics before theprogram initiation. These results are robust to thechoice of program implementation measure.

Conclusion

This paper has examined the nexus among institu-tions, policy implementation, and economic perfor-mance in countries undertaking IMF-supported re-forms. We employed a short-run statistical modelthat treats institutions and politics as exogenous and mean-reverting, that takes into account theautoregressive and mean-reverting nature ofmacroeconomic outcomes, and that corrects forthe endogeneity of program implementation withrespect to macroeconomic performance.

Table 11

Program Implementation, Institutions, and Current Account (Ratio to GDP):Regressions Using Irreversible Interruption Index

Dependent Variable ∆BCA_Y (T–1 to horizon)


OLS 2SLS 3SLS


Non-interruption dummy 0.002 0.010 0.024** –0.037 –0.024 0.032 –0.032 –0.024 0.030

RGDPPC (T–1) –0.001 –0.001 0.0008 –0.002 –0.003 0.001 –0.002 –0.003 0.001

∆BCA_Y (T–1) –0.133 –0.172* –0.105 –0.235** –0.197* –0.180 –0.173* –0.192* –0.180*

BCA_Y (T–1) –0.158** –0.208** –0.366*** –0.136** –0.115 –0.365*** –0.105* –0.135* –0.399***

Corruption (T–1) –0.012* –0.007 –0.019** –0.004 0.0004 –0.019** –0.003 –0.0001 –0.020**

Change in corruption

(T–1 to horizon) 0.0002 –0.005 0.007 0.0004 –0.0004 0.0001 –0.0005 –0.002 0.0002

Ethnic conflict (T–1) –0.006 –0.005 0.002 –0.006 –0.006 0.003 –0.006* –0.006 0.003


(T–1 to horizon) –0.011 –0.002 –0.007 –0.015 –0.016* –0.004 –0.015 –0.014 –0.004

External conflict (T–1) 0.003 0.0003 0.0006 0.002 –0.002 –0.002 0.002 –0.002 –0.002

Change in external conflict

(T–1 to horizon) 0.008* 0.0004 –0.003 0.005 –0.001 –0.009** 0.006 –0.001 –0.010**

Government stability (T–1) 0.013** 0.014** 0.010 0.009* 0.010 0.001 0.008* 0.010 0.001


(T–1 to horizon) –0.003 0.006 –0.003 –0.006 0.005 –0.007 –0.003 0.005 –0.008

Investment profile (T–1) –0.011*** –0.011** –0.005 –0.009** –0.008* –0.005 –0.009*** –0.009* –0.005


(T–1 to horizon) –0.016*** –0.018*** –0.006 –0.014*** –0.014*** –0.002 –0.019*** –0.015*** –0.002

Military in politics (T–1) 0.011*** 0.012*** 0.010** 0.006* 0.010** 0.008* 0.007** 0.010** 0.008**

Change in military in politics

(T–1 to horizon) 0.004 0.005 0.0001 0.006 0.003 –0.001 0.006 0.004 –0.0003


R2 0.385 0.368 0.416 0.443 0.395 0.448 0.430 0.393 0.447


158


Our main findings are fourfold. First, the qualityof institutions and the domestic political environ-ment matter for macroeconomic outcomes in coun-tries implementing IMF-supported programs, espe-cially at longer horizons of up to three years. Thisdirect beneficial impact of institutions on themacroeconomic variables is in addition to their in-direct impact through better program implementa-tion. As expected, improvements in the govern-ment bureaucracy and better enforcement of lawand order are associated with lower inflation. How-ever, declines in internal conflict are associated withhigher inflation. Improvements in a program coun-try’s investment profile and in government stabilitylead to faster economic growth. Easing of externalconflict and lower military involvement in politicsbefore program approval are associated withstronger fiscal outcomes as military expenditure de-clines. On the other hand, reductions in ethnic ten-sion and improvements in government accountabil-ity are associated with weaker fiscal outcomes,perhaps because programs may provide for highertargeted expenditure. Greater government stabilityand reductions in the military’s involvement in pol-itics before the program starts are associated with astrengthening of the external current account.However, lower ethnic tension and improvementsin a program country’s investment profile lead to adeterioration of the current account.

Second, the institutional and political environ-ment is quantitatively important for the implemen-tation of IMF-supported programs. Rates of disburse-ment of IMF loans are higher and programinterruptions less frequent in countries where ethnictensions are low, where governments are stable andless corrupt, and where the military is less involvedin politics. In addition, more IMF loans are dis-bursed and fewer interruptions are experienced incountries in which internal conflict was intense andlaw enforcement weak before program approval. Ar-guably, this reflects the IMF’s role, as lender and pol-icy adviser, in facilitating the return to normalcy ofcountries experiencing natural or political shocks.

Third, program implementation varies systemati-cally with the duration of a country’s financial en-gagement with the IMF and the size of its quota.More funds are disbursed and fewer program interrup-tions are experienced in countries that have spentmore time in previous IMF-supported programs. Im-plementation is also better for larger programs (asmeasured by the amount of program financing ap-proved in relation to the country’s IMF quota).

And fourth, after the impact of institutions onthe macroeconomic situation is taken into account,the extent of program implementation exerts an in-dependent influence on macroeconomic outcomes,especially over shorter horizons of up to two years.Better-implemented programs are associated withlower inflation, with initially weaker but ultimatelystronger external and fiscal outcomes, and with a

statistically insignificant impact on economicgrowth. These results are to be contrasted withthose of studies that do not correct for program im-plementation; these studies conclude that programparticipation has ambiguous effects on inflation.Correcting for differences in implementation thusprovides some evidence linking successful imple-mentation of IMF-supported reforms to moreprogress in achieving price stability.

What, then, are the policy implications of thisanalysis for the IMF? The first issue is the lack ofclear-cut results linking program implementation tothe resumption of economic growth in countries im-plementing IMF-supported reforms. One possibilityis that successful program implementation has fa-vorable impacts on growth that are only felt beyondthe three-year horizon captured in our model. Thelength of lags in the operation of IMF-supportedstructural reforms should be a topic of future re-search. Further, the lack of conclusive links betweenprogram implementation and growth suggests that itmight be useful for the IMF to seek to identify struc-tural reforms that could pay off quickly in terms ofeconomic growth, both at the program design stageand at the implementation stage. At the programdesign stage, the IMF could monitor regularly pub-lished institutional and political indicators relevantto economic growth—such as the ICRG ratings ofthe level of ethnic tension, government account-ability, and the investment climate. These indica-tors would also need to be carefully monitored dur-ing program implementation to ascertain whetherIMF-supported reforms are on track toward meetingtheir growth objectives. Information on the determi-nants of the investment profile—viability of con-tracts, threat of expropriation, ease of profit repatria-tion, and payment delays—could provide high-frequency feedback concerning the extent to whichprograms are on track in implementing investment-friendly reforms.

Second, paying due attention to relevant politicaland institutional developments is critical to the successful design and implementation of IMF-supported programs. Quantitative information andanalysis could be a useful complement to informa-tion from IMF missions and resident representativesin assessing rapidly changing political environ-ments, indicating the potential for successful pro-gram implementation. A decline in political indica-tors below thresholds historically associated withinadequate program implementation could give theIMF an early warning signal, much as financial vul-nerability indicators provide useful signals of im-pending financial crisis. The IMF has on occasionresponded to heightened political uncertainty by re-quiring the major political forces in a country—thegovernment and the main opposition parties in par-liament—to endorse a program at an early stage.Systematizing these efforts, as the IMF has beendoing by increasing the emphasis on ownership,

159


could yield dividends in terms of improved programdesign, implementation, and macroeconomic per-formance. It would enable the IMF to avoid situa-tions in which, having designed and implementedfirst-best programs that failed to fully take into ac-count relevant political and institutional factors,one ends up in a third-best world when these “ideal”programs are not properly executed. In econometricterms, one would ideally want initial (T–1) institu-tional and political variables to enter implementa-tion regressions (such as those in Table 3) with in-significant coefficients. That would provideevidence that IMF-supported programs are well tai-lored to the specifics of each country’s politico-institutional climate and that the success or failureof a program is independent of initial political con-ditions. Unfortunately, we have such neutrality onlyfor the initial economic conditions. More generally,it might be useful to consider incorporating quanti-tative political and institutional indicators andanalysis in IMF surveillance work.

Third, we have treated institutions as exogenousand mean-reverting processes, yet institutional de-velopment is an important objective of IMF-supported programs with a structural orientation. Itwould be useful to assess systematically the impactof better implementation of IMF-supported pro-grams on the dynamics of institutional and politicalfactors. In such a model, the evolution of formal andinformal institutions would be endogenous to thepolitico-economic process, including participationin IMF-supported programs. To the extent that IMF-supported programs promote welfare-improving in-stitutional change, their beneficial effects are goingto be larger than suggested by models, such as ours,that treat institutions as exogenous.

ReferencesAtoyan, Ruben, Patrick Conway, Marcelo Selowsky, and

Tsidi Tsikata, 2004, “Macroeconomic Adjustment inIMF-Supported Programs: Projections and Reality,”IMF Independent Evaluation Office BP/04/2 (Wash-ington: International Monetary Fund). This paper alsoappears, in somewhat different form, as Chapter 5 inthis volume.

Bagci, Pinar, and William Perraudin, 1997, “The Impactof IMF Programmes,” Working Paper No. 24 (Lon-don: Global Economic Institutions Research Pro-gram, Center for Economic Policy Research).

Barro, Robert, and J. Lee, 2002, “IMF Programs: Who IsChosen and What Are the Effects?” NBER WorkingPaper No. 8951 (Cambridge, Massachusetts: Na-tional Bureau of Economic Research).

Conway, Patrick, 1994, “IMF Lending Programs: Partici-pation and Impact,” Journal of Development Econom-ics, Vol. 45 (December), pp. 365–91.

Dicks-Mireaux, Louis, Mauro Mecagni, and SusanSchadler, 2000, “Evaluating the Effect of IMF Lending

to Low-Income Countries,” Journal of DevelopmentEconomics, Vol. 61, pp. 495–526.

Dollar, David, and Jakob Svensson, 2000, “What Explainsthe Success or Failure of Bank-Supported Adjust-ment Programs?” Economic Journal, Vol. 110 (Octo-ber), pp. 894–917.

Dreher, Axel, 2004, “IMF and Economic Growth: The Ef-fects of Programs, Loans, and Compliance with Con-ditionality,” Working Paper, University of Konstanz(July). Available on the Web at http://econwpa.wustl.edu:8089/eps/if/papers/0404/0404004.pdf.

Haque, Nadeem Ul, and Mohsin Khan, 1998, “Do IMF-Supported Programs Work? A Survey of the Cross-Country Empirical Evidence,” IMF Working Paper98/169 (Washington: International Monetary Fund).

International Country Risk Guide (ICRG), 2003, BriefGuide to the Ratings System (East Syracuse, New York:Political Risk Services Group, Inc.).

International Monetary Fund (IMF), 2002, Evaluation ofthe Prolonged Use of IMF Resources (2 vols.). (Wash-ington: Independent Evaluation Office, Interna-tional Monetary Fund).

______, 2003, “Growth and Institutions,” Chapter III ofWorld Economic Outlook (April) (Washington: Inter-national Monetary Fund).

Ivanova, Anna, Wolfgang Mayer, Alex Mourmouras, andGeorge Anayiotos, 2003, “What Determines the Im-plementation of IMF-Supported Programs?” IMFWorking Paper 03/08 (Washington: InternationalMonetary Fund). This paper also appears, in some-what different form, as Chapter 10 in this volume.

Joyce, Joseph, 2003, “Promises Made, Promises Broken: AModel of IMF Program Implementation,” WorkingPaper No. 2003-03 (Wellesley, Massachusetts: De-partment of Economics, Wellesley College). Availableon the Web at http://www.wellesley.edu/Economics/joyce/PROM4.pdf.

Khan, Mohsin, 1990, “The Macroeconomic Effects of Fund-Supported Adjustment Programs,” Staff Papers, Inter-national Monetary Fund, Vol. 37 (June), pp. 195–231.

Killick, Tony, 1995, IMF Programmes in Developing Coun-tries: Design and Impact (London: Routledge).

Mecagni, Mauro, 1999, “The Causes of Program Interrup-tions,” in Economic Adjustment and Reform in Low-Income Countries, ed. by Hugh Bredenkamp andSusan Schadler (Washington: International Mone-tary Fund).

North, Douglass, 1997, “Introduction,” in The Frontiers ofthe New Institutional Economics, ed. by John Drobakand John Nye (New York: Academic Press), pp. 3–13.

Przeworski, Adam, and James R. Vreeland, 2000, “The Ef-fect of IMF Programs on Economic Growth,” Journalof Development Economics, Vol. 62, pp. 385–421.

Rodrik, Dani, Arvind Subramanian, and Francesco Trebbi,2002, “Institutions Rule: The Primacy of Institutionsover Geography and Integration in Economic Devel-opment,” NBER Working Paper No. 9305 (Cam-bridge, Massachusetts: National Bureau of EconomicResearch).

Schadler, Susan, Franek Rozwadowski, Siddharth Tiwari,and David Robinson, 1993, Economic Adjustment inLow-Income Countries—Experience Under the EnhancedStructural Adjustment Facility, IMF Occasional PaperNo. 106 (Washington: International Monetary Fund).

160

10

The paper provides a quantitative analysis of the factorsthat determine successful implementation of IMF-supported programs. To this end, we construct newmeasures of program implementation and compliancewith conditionality for 170 IMF programs approved be-tween 1992 and 1998. The main hypothesis tested iswhether IMF effort and the design of conditionality sig-nificantly affect the probability of successful implementa-tion of IMF-supported programs. We find that programimplementation depends primarily on the borrower’s do-mestic political economy. Strong special interests in theparliament, political instability, inefficient bureaucra-cies, lack of political cohesion, and ethno-linguistic divi-sions weaken program implementation. IMF effort, thedesign of conditionality, and initial and external condi-tions do not materially influence program prospects.

Introduction

At the heart of the International Monetary Fund’soperations are conditional lending programs thatgive borrowing countries breathing space while theycorrect their macroeconomic and structural imbal-ances. These programs provide mutual assurances.On the one hand, member countries are assured thatthey will continue receiving IMF financing if theymeet the specified conditions. On the other hand,conditionality ensures that adjustment is under-taken in ways that are conducive to national and in-ternational prosperity, providing guarantees to theIMF that it will be repaid and that the world’s finan-cial system will not suffer from disruptive systemiccrises.

In order for the effects of IMF-supported programsto be fully realized, however, the policies they envis-age must be implemented to the fullest possible ex-

tent. Many programs are in fact interrupted amidpolitical or economic turmoil, in circumstances inwhich it is not possible to agree on conditionality tounderpin new or revised programs. The implemen-tation record of IMF-supported programs has beenrather disappointing. About 44 percent of all pro-grams approved between 1992 and 1998 were notcompleted, experiencing irreversible interruptions.

In this paper, the quality of implementation ofIMF-supported programs is linked to three groups offactors: (a) the political characteristics of borrowingcountries; (b) IMF conditionality and the humanand financial effort the IMF invests in programs;and (c) internal economic conditions in countriesimplementing programs and the external conditionsaffecting them. Implementation, as we use the term,is to be contrasted with overall program success, de-fined as the achievement of its macroeconomic andstructural objectives. Previous econometric studies(Bird, 2000, provides a review) commonly assessedthe success of IMF-supported programs by examin-ing macroeconomic indicators such as budgetdeficit, international reserves, inflation, and growthbefore and after the program. However, there is noreason to expect that a program will realize itsmacroeconomic and structural objectives if imple-mentation consistently falls short of program inten-tions. Understanding the factors that affect programimplementation is thus the first step in understand-ing the determinants of overall program success.

Our analysis focuses on program implementationfor a sample of countries that made conscious deci-sions to enter into agreements with the IMF. We donot address the prior questions of what makes acountry commit to an IMF-supported program andwhether countries have a better chance of succeed-ing in their macroeconomic and structural adjust-ment by having an IMF program in place. To answerthese questions would require an assessment of eco-nomic performance of countries whose adjustmentprograms were not supported by the IMF, and com-parable information is not readily available.

The literature offers several clues indicating thatthe primary factors influencing the implementationof IMF-supported programs lie in the domestic polit-ical economy of borrowing countries. Interruptionsof programs supported by the IMF’s concessional

What Determines the Implementationof IMF-Supported Programs?Anna Ivanova, Wolfgang Mayer, Alex Mourmouras, and George Anayiotos1

1 This is a revised version of IMF Working Paper 03/08. Foruseful comments and discussion, we would like to thank the par-ticipants in the Second Annual IMF Research Conference,Washington, November 29–30, 2001. We are especially indebtedto Patrick Conway, our rapporteur at the conference; GrahamBird; Jim Boughton; Peter Clark; Charles Engel; Judy Gold;Arthur Goldberger; Mohsin Khan; Tony Killick; and SunilSharma. Tim Lane encouraged us to undertake this project andmade invaluable comments and suggestions on earlier drafts. Wealone, however, are responsible for any remaining errors.

CHAPTER

161

Ivanova, Mayer, Mourmouras, and Anayiotos

facilities—the Structural Adjustment Facility(SAF) and the Enhanced Structural Adjustment Fa-cility (ESAF)—were primarily caused by domesticpolitical economy factors, not poor program design(Mecagni, 1999). Similarly, the success of WorldBank–supported adjustment programs is attributedto favorable domestic political conditions and insti-tutions, including lack of ethnic and linguistic divi-sions, government stability, and democratic regimes(Dollar and Svensson, 2000). It is important to notethat World Bank conditionality and resources allo-cated to program design and monitoring did notseem to matter at the margin. Finally, case study ev-idence suggests that in some countries, the ambiva-lence of the top political leaders and resistance bysenior officials and special interests were key to thefailures of IMF-supported programs.2 When lack ofpolitical commitment resulted in stop-and-go pro-gram cycles, the imposition of large numbers of prioractions had limited success, pointing to the need forgreater selectivity in lending. In other countries,participatory processes that actively involved thecountry’s top leadership were instrumental in over-coming domestic divisions, building ownership forprograms, and ensuring program success.

Our analysis of the determinants of program im-plementation is made possible by the availability ofnew datasets. First, political scientists have in recentyears developed several quantitative indicators ofpolitical conditions in borrowing countries. Second,during the last decade, the IMF has improved itsmonitoring of programs and internal resource allo-cation. This allows us to develop indicators thatcapture program conditionality, the quality of its im-plementation, and the IMF’s human and financialeffort in program countries.

In ascertaining the impact on program implemen-tation of variables under IMF control, a key empiri-cal issue is the need to properly account for the en-dogenous nature of IMF decisions. A second keyissue, based on the findings of recent theoreticalwork on conditionality and program ownership, is totest for the impact on program implementation ofspecial interests resisting reforms.3 We develop anindex of the power of special interests in parliamentand examine the impact on program implementa-tion of parties representing religious, nationalistic,regional, and rural interests.

Our main results are easily summarized. On theone hand, the implementation of IMF-supportedprograms is strongly influenced by recipient coun-tries’ domestic political economy. Weak programimplementation was strongly associated with strongspecial interests in parliament, lack of political cohe-sion, inefficient bureaucracies, and ethno-linguisticdivisions. The strong association between program

implementation and political economy variables isrobust across different econometric specifications.On the other hand, initial economic conditions,IMF effort, and the breadth and depth of condition-ality do not seem to materially influence programprospects when they are properly instrumented for.Other recent studies corroborate this finding. Pro-gram implementation is not related to the numberof conditions or the number of prior actions4 (IMF,2001c; and Thomas, 2003). Structural conditional-ity does not influence medium-term fiscal develop-ments (Bulír and Moon, 2003).

The rest of the paper is organized as follows. Thesecond section describes the sample and various im-plementation measures. The third section describesthe econometric methodology and presents themain results. The fourth section is the conclusion.

Characteristics of IMF-Supported Programs

Measuring Program Implementation

We analyzed the implementation of 170 IMF-supported programs approved between 1992 and1998 (Table 1). The choice of the time period wasdetermined by the availability of information onconditionality in the MONA database5 and the dif-ficulty in assessing programs approved after 1998,some of which were still ongoing when the paper onwhich this chapter is based was prepared. Thelargest collection of programs (about 48 percent ofthe total) in the sample were Stand-By Arrange-ments (SBAs). The second-largest group of programs(38 percent) were programs under concessional fa-cilities,6 followed by programs under the ExtendedFund Facility (EFF) (15 percent).

IMF-supported programs are complex in nature,making it difficult to arrive at a single metric of pro-gram success. In general, a program is considered tobe successful if its principal macroeconomic andstructural objectives are met. Lacking a single indi-cator of success for IMF-supported programs, such asthe one produced by the World Bank’s OperationsEvaluation Department for Bank-supported pro-grams, we focus on the narrower measures of suc-cessful program implementation, which is a prereq-uisite for overall program success.

Our strategy was to construct multiple measuresof implementation for each program in our sample.These measures capture program performance from

4 Prior actions are conditions that must be implemented beforethe IMF can approve or continue disbursements of its loans.

5 The Monitoring of IMF Arrangements (MONA) database ismaintained by the IMF’s Policy Development and Review De-partment. MONA was started in 1992 and is missing 18 pro-grams approved in that year.

6 The ESAF was restructured and renamed the Poverty Reduc-tion and Growth Facility (PRGF) in 1999.

2 See IMF (2001a, 2001b), Bredenkamp and Schadler (1999),IMF (1998), and Boughton and Mourmouras (2004).

3 See Mayer and Mourmouras (2002) and Drazen (2002).

162

10 � WHAT DETERMINES THE IMPLEMENTATION OF IMF-SUPPORTED PROGRAMS?

different angles.7 Viewed from this narrower per-spective, implementation is measured by the extentto which the program was completed without unduedelays, the extent to which macroeconomic and

structural conditionality was met, and the extent towhich funds committed by the IMF were disbursed.

Our first indicator of program implementation is abinary variable measuring program interruptions.This variable captures both major and minor inter-ruptions and is motivated by Mecagni’s work. Wesay that an interruption occurred if an SBA reviewwas delayed by more than three months or not com-pleted at all, if a program review for EFF/PRGFs wasdelayed by more than six months or not completedat all, if there was an interval of more than sixmonths between two subsequent years of a multiyear

7 An alternative to this approach is to construct a comprehen-sive indicator of implementation to reflect the extent to whichprograms reach their broad objectives. Dollar and Svensson(2000) used such a definition, based on the independent (butsubjective) judgments of the World Bank’s Operations Evalua-tion Department. No such measure was available to us—the IMF’s Independent Evaluation Office (IEO) was set up onlyin 2001.

Table 1

Program Implementation by Type of Arrangement

Number of ProgramsExcluding

PrecautionaryArrangements Share of

and Share of Programs Average Average Average Average Cancelled Programs Having Macro Structural Overall Share of

and Having Irreversible Imple- Imple- Imple- Committed Type of Number of Ongoing Inter- Inter- mentation mentation mentation Funds Arrangement Programs1 Programs1 ruptions2,3 ruptions3,4 Index3,5,6 Index3,6,7 Index3,6,8 Disbursed1

(in percent)

EFF 25 13 68 40 87.0 75.4 83.3 72.1PRGF/ESAF 64 51 73.4 45.3 77.1 71.3 72.9 77.2SBA 81 41 67.9 43.2 81.0 60.8 76.0 63.7Total 170 105 70.0 43.5 80.3 67.4 75.8 71.3

Notes: EFF denotes Extended Fund Facility; PRGF denotes Poverty Reduction and Growth Facility; ESAF denoted Enhanced Structural Adjust-ment Facility; and SBA denotes Stand-By Arrangement.

1 Multiyear arrangements are treated as one program. This is a sample of programs approved between 1992 and 1998 and available fromthe IMF’s Monitoring of IMF Arrangements (MONA) database. (Our sample is missing 16 SBAs, one ESAF, and one EFF program approved in1992.) The sample of EFF programs is quite small to use for drawing reliable conclusions regarding relative performance of EFF programs com-pared with ESAF and SBA programs. The average share of disbursed funds is computed across a sample of programs that excludes arrange-ments that were either precautionary upon approval or later became precautionary, as well as canceled and ongoing programs.

2 An interruption occurs if an SBA program review was delayed by more than three months or not completed at all; if a program review forESAF/PRGF programs was delayed by more than six months or not completed at all; if there was an interval of more than six months betweentwo subsequent years of a multiyear arrangement; or if at least one of the annual arrangements was not approved (exceptions are programsthat were canceled and replaced by another program, in which case noncompleted reviews and nonapproved annual arrangements are notcounted as interruptions).

3 The macroeconomic and structural implementation indices were computed from information available in MONA. Since MONA question-naires are sent only for programs for which IMF Executive Board meetings are scheduled, implementation information is not available on manyconditions for programs with noncompleted reviews. Since these were typically interrupted programs, the macroeconomic and structural indicesoverstate program implementation. Interruption indices were constructed using additional information from country documents and other sources.

4 An irreversible interruption occurs if either (a) the last scheduled program review was not completed (all programs); or (b) all scheduled re-views were completed but the subsequent annual arrangement was not approved (ESAF/PRGF arrangements).

5 The macroeconomic implementation index for a given macroeonomic performance criterion is equal to 100 percent if the said criterion wasmet or met after modification and it is equal to zero if the said criterion was not met, not met after modification, waived, or waived after modi-fication. The macroeconomic implementation index for a program is the average of the macroeconomic implementation indices across allmacroeconomic performance criteria for this program.

6 The sample size for implementation indices was smaller (150 programs), which corresponds to the sample constructed for "Structural Con-ditionality in IMF-Supported Programs"; we simply extended the structural index used in this paper to macroeconomic and overall implementa-tion indices.

7 The structural implementation index for a given structural condition is equal to 100 percent if the structural condition was met or met witha small delay for structural benchmarks; it is equal to 50 percent if the structural condition was partially met or delayed for performance criteria;and it is equal to zero if the structural condition was not met. The structural implementation index for a program is the average of the structuralimplementation indices across all structural conditions for this program.

8 The average overall implementation index for a given program is the average of macroeconomic and structural implementation indicesover all conditions in this program.

arrangement, or if at least one of the annualarrangements was not approved.8

The second indicator is a binary variable identify-ing irreversible program interruptions. This measurecaptures programs that went off track and were notrevived subsequently (i.e., were either canceled orwere allowed to lapse because of policy slippages).More precisely, we say that an irreversible interruptionoccurred if either the last scheduled program reviewwas not completed (all programs) or all scheduledreviews were completed but the subsequent annualarrangement was not approved (ESAF/PRGFarrangements). Third, we constructed a quantitativeindicator of implementation of IMF conditionality,the overall implementation index, which representsthe average fraction of macroeconomic and struc-tural conditionality implemented. This indicator isan extension of the structural conditionality indexdeveloped in the IMF’s Policy Development and Re-view Department (PDR) during the 2000–2002 re-view of conditionality (IMF, 2001c). Finally, we alsocomputed the ratio of disbursements to commitments.The last two indicators are continuous variablesthat take values between zero and 100.

Each of our indices captures an important dimen-sion of program implementation. The macroeco-nomic and structural implementation indices pro-vide quantitative information on implementationrates by type of condition. Their main drawback isthat they overstate the degree of implementationbecause, as is well known, MONA fails to captureinformation on interrupted programs that were notsubject to further Board reviews.9 The interruptiondummies, which are based on MONA data and ad-ditional information from program documents, com-plement the macroeconomic and structural imple-mentation indices by capturing significant programstoppages. The share of disbursed funds providesuseful information on the proportion of approvedassistance actually delivered for non-precautionaryarrangements and also on the actual versus sched-uled duration of the program. The implementationindices and interruption dummies provide useful in-formation about precautionary programs, canceledprograms, and some unusual cases in which no draw-ings were made despite good results.

Descriptive Statistics

Table 1 summarizes program implementation bytype of arrangement. About 44 percent of all pro-grams experienced an irreversible interruption,while 70 percent of all programs experienced either

a major or a minor interruption. Nonetheless, ap-proximately 71 percent of committed funds weredisbursed on average (excluding precautionaryarrangements, and canceled and ongoing programs).The average implementation index for programs forwhich information is available in MONA is 76 per-cent. The macroeconomic implementation index issignificantly higher than the structural implementa-tion index (80 percent versus 67 percent). However,implementation indices most likely overstate pro-gram performance. MONA collects data only forprogram test dates subject to Board approval or re-view. Information on later stages of some programsexperiencing major interruptions is, therefore, notavailable.

The four measures of program implementation aresignificantly correlated (Table 2).10 However, thecorrelation coefficients are not very high in mostcases, reflecting the fact that the various implemen-tation measures capture quite different angles ofprogram performance. The correlation coefficientbetween the macroeconomic and structural imple-mentation index is only 0.2. This is consistent withthe recent finding by Bulír and Moon (2003) thatthe implementation of fiscal measures in IMF-supported programs was not strongly correlated withthe implementation of structural measures.

Several differences stand out between successfullyimplemented and interrupted programs (Table 3).First, countries that implemented their IMF-supported reform programs were experiencing muchhigher inflation at the start of the program thancountries whose programs were interrupted. Al-though the difference in inflation rates was not sta-tistically significant in the year in which the pro-gram was approved, inflation was significantlyhigher in countries with successfully implementedprograms one year before the program started.Countries that implemented their programs startedwith substantially smaller budget deficits (21⁄2 per-cent of GDP on average) compared with countriesin which programs were interrupted (43⁄4 percent ofGDP on average). In countries with interrupted pro-grams, terms of trade shocks were stronger, thestrength of special interests was higher, and the de-gree of political cohesion was lower. Interestingly, theeffort invested by the IMF and the extent and struc-ture of conditionality are similar in interrupted andimplemented programs.

Correlation of Implementation Measureswith Macroeconomic Performance

IMF-supported programs aim to strengthen the bor-rowing countries’ balance of payments and overall

163


8 Exceptions are programs that were canceled and replaced byother programs, in which case noncompleted reviews and nonap-proved annual arrangements are not counted as interruptions.The appendix explains in detail the definitions of program im-plementation and the political and other variables used in theeconometric work.

9 Changes in MONA submissions instituted in 2002 have cor-rected this weakness.

10 The only exception was the reversible-interruption indicator,which is not significantly correlated with the structural imple-mentation index. Since the reversible-interruption dummy cap-tures “small” policy slippages that were subsequently corrected, wedecided not to include this measure in our econometric analysis.

164


Table 2

Correlations of Implementation Indices(excluding arrangements precautionary on approval)

AverageMacro- Average Average Average

economic Structural Overall Share of Imple- Imple- Imple- Irreversible Committed

Pearson mentation mentation mentation Interruption Interruption Funds Correlation Index1,2 Index2,3 Index2,4 Index5 Index6 Disbursed

Average macroeconomic implementation index1,2 1.000

Average structural implementation index2,4 0.211 1.000

(0.01)Average overall

implementation index3,4 0.782 0.653 1.00(0.00) (0.00)

Interruption index5 –0.286 –0.050 –0.30 1.00(0.00) (0.56) (0.00)

Irreversible interruption index6 –0.263 –0.279 –0.39 0.55 1.00

(0.00) (0.00) (0.00) (0.00)Average share of

committed funds disbursed 0.211 0.346 0.38 –0.42 –0.75 1.00

(0.01) (0.00) (0.00) (0.00) (0.00)

Notes: The two-tailed significance level appears in parentheses. Figures significant at the 0.05 level are in boldface. Multiyear arrangementsare treated as one program. These programs were approved between 1992 and 1998 and are taken from the IMF’s Monitoring of IMF Arrange-ments (MONA) database.

1 The macroeconomic implementation index is equal to 100 percent if macroeconomic performance criteria were met or were met aftermodification; and it is equal to zero if macroeconomic performance criteria were not met, not met after modification, waived, or waived aftermodification.

2 The macroeconomic and structural implementation indices were computed from information available in MONA. Since MONA question-naires are sent only for programs for which IMF Executive Board meetings are scheduled, implementation information is missing on many con-ditions for programs with noncompleted reviews. Since these were typically interrupted programs, the macroeconomic and structural indicesoverstate program implementation. Interruption indices were constructed using additional information from country documents and othersources.

3 The structural implementation index is equal to 100 percent if structural criteria were met or met with a small delay for structural bench-marks; it is equal to 50 percent if structural criteria were partially met or delayed for performance criteria; and it is equal to zero if structural cri-teria were not met.

4 The average overall implementation index is the average of macroeconomic and structural implementation indices over all conditions for agiven program.

5 An interruption occurs if a Stand-By Arrangement program review was delayed by more than three months or not completed at all; if a pro-gram review for Enhanced Structural Adjustment Facility (ESAF)/Poverty Reduction and Growth Facility (PRGF) programs was delayed by morethan six months or not completed at all; if there was an interval of more than six months between two subsequent years of a multiyear arrange-ment; or if at least one of the annual arrangements was not approved. (Exceptions are programs that were cancelled and replaced by anotherprogram, in which case noncompleted reviews and nonapproved annual arrangements are not counted as interruptions.)

6 An irreversible interruption occurs if either (a) the last scheduled program review was not completed (all programs); or (b) all scheduled re-views were completed but the subsequent annual arrangement was not approved (ESAF/PRGF arrangements).

165


Table 3

Features of Successfully Implemented and Interrupted IMF Programs

Successfully t-test for Implemented Interrupted Equality of Means1

Number of Number of Average programs Average programs t-statistics p-value

Political economy characteristicsEthnic fractionalization 46 58 51 50 –0.86 0.39Political instability2 4.75 67 5.68 57 –1.16 0.25Executive index of electoral

competitiveness (in percent)3 62 86 56 66 0.69 0.49Time in power (years) 5.73 86 4.52 66 1.00 0.32Strength of special interests4 16 66 25 54 –1.74 0.04Index of political cohesion5 2.36 85 2.06 66 2.45 0.01Quality of bureaucracy6 1.72 67 1.81 57 –0.68 0.50Change of chief executive7 18.18 99 28.38 74 –1.59 0.11

Variables under IMF controlIMF effort per program year

(in millions of U.S. dollars)8 1.01 99 1.03 68 0.13 0.90Total number of conditions

per program year 40 95 38 65 0.48 0.63Share of quantitative program

criteria (PCs) waived (percent) 8.33 99 7.22 73 0.60 0.55Share of structural conditions (percent) 37 95 40 68 –0.74 0.46Loan size (agreed amount, in

millions of SDRs) 620 95 526 69 0.30 0.76

Macroeconomic characteristicsInitial GDP per capita per year

(U.S. dollars) 1,494 98 1,291 74 0.81 0.42Initial debt to the IMF (actual

holdings as a percentage of quota) 177 99 159 74 1.16 0.25Initial central government balance

(percent of GDP) –2.50 88 –4.74 68 3.70 0.00Reserve holdings

(as percent of imports)9 36.72 81 32.98 68 0.85 0.40Initial inflation (percent per year) 80 98 53 74 0.8910 0.37Initial current account balance

(percent of GDP) –5.32 98 –5.87 74 0.42 0.67Terms-of-trade shock (growth rate

during program period, in percent)11 –90 98 –15 74 –1.30 0.10

Notes: Bold figures indicate significance at the 5 percent level; bold italic figures indicate significance at the 10 percent level. PCs denote performance criteria; SDRsdenotes Special Drawing Rights.

1 The null hypothesis is stated as follows: H0: mean(implemented)– mean(interrupted) = 0. The alternative hypothesis was different for different cases. In cases inwhich the means were significantly different we report t-statistics for the relevant one-sided alternative hypothesis—for example, for the index of political cohesion wereport t-statistics for the null hypothesis as specified above versus the alternative hypothesis (that the degree of political cohesion is higher for successfully implementedcompared with interrupted programs), which in fact, cannot be rejected at the 5 percent significance level), otherwise we report t-statistics for the alternative hypothe-sis (that the difference in means is not equal to zero).

2 This index is computed based on the index of internal conflict provided by the International Country Risk Guide (ICRG) on a scale from 0 to 12. Higher values of theindex correspond to more internal political instability. We replaced the value of this variable by its maximum score (12) if there was a change of chief executive in thecourse of the IMF program.

3 Dummy variable, which equals unity if the executive index of electoral competitiveness is equal to seven, and zero otherwise. The executive index of electoral competitiveness is from the Database of Political Institutions at the World Bank. It ranges from unity to seven, with higher values corresponding to more competitiveelections.

4 Computed as the maximum share of seats in the parliament held by parties representing special interests (Political Institutions Database, World Bank). Four specialinterest groups are identified: religious, nationalistic, regional, and rural.

5 The index of political cohesion is defined as follows: in presidential systems, a high degree of political cohesion is said to exist if the same party is in control of theexecutive and legislature; in parliamentary systems, a high degree of political cohesion means a one-party majority government. See the appendix for a more detaileddefinition.

6 Bureaucracy quality (ICRG) measures the quality of a country's bureaucracy on a four-point scale. See the appendix for a more detailed definition. This variable wasinteracted with the dummy indicating that there was a change of chief executive (Political Institutions Database and CIA World Factbook for most recent years).

7 The variable, change of chief executive, is equal to 100 if the chief executive changed during the program period.8 IMF effort is the estimated dollar cost of IMF programs computed based on the IMF’s Budget Reporting System (BRS) data on hours spent by the staff on program

implementation (which includes both preparation and supervision of the program) and estimated average salaries of the staff by grade. We also made use of data provided by the IMF’s Office of Personnel Management (OPM) on the dollar costs of resident representatives.

9 Although the average inflation rate in the approval year was not significantly different for successfully implemented and interrupted programs in the year precedingthe approval year, the average inflation rate for implemented programs was significantly higher than for interrupted ones.

10 Reserves here do not include gold.11 Average growth rate of dollar export prices multiplied by the initial share of exports in GDP minus average growth rate of dollar import prices multiplied by the ini-

tial share of imports in GDP over the course of the program.

166


macroeconomic performance. This section presentsa preliminary assessment of whether program imple-mentation improves macroeconomic performance,both over the course of programs and after their ex-piration.

Figures 1–3 show how the main macroeconomicmagnitudes evolved in uninterrupted and inter-rupted programs from the year in which the programwas approved until three years after the programended. The variables plotted are the averagechanges in inflation, the ratio of reserves to imports,and real GDP growth. The eyeball test (Figure 1) in-dicates that inflation for both implemented and in-terrupted programs continued to decline after theprogram ended, but the reduction in inflation (com-pared with the approval year) was greater for imple-

mented than for interrupted programs.11 However,this difference was significant only for the end yearof the program, as indicated by solid dots on thegraph. The average level of inflation itself in theend year was also significantly lower for uninter-rupted than for interrupted programs but was notsignificantly different in later years. The high vari-ability of inflation in the data contributed to the dif-ferences in the changes in inflation being indistin-guishable between implemented and interruptedprograms in later years. Completed programs wereassociated with better performance, at least as far as the evolution of the reserve coverage of imports

0

–20

–40

–60

–80

–100

–120

0 0.5 1 1.5 2 2.5 3 3.5

Year after program end

Aver

age

chan

ge in

ann

ual i

nfla

tion

com

pare

d w

ith a

ppro

val y

ear

Interrupted

Successfully implemented

Figure 1

Inflation Dynamics for Successfully Implemented and Interrupted IMF Programs

0

2

4

6

8

10

12

14

0 0.5 1 1.5 2 2.5 3 3.5

Year after program end

Aver

age

chan

ge in

res

erve

s as

per

cent

of i

mpo

rts

com

pare

d w

ith a

ppro

val y

ear

Interrupted


Figure 2

Reserves-to-Imports Dynamics for Successfully Implemented and Interrupted IMF Programs

11 Uninterrupted programs started with significantly higher in-flation as measured one year before the approval year.

167


was concerned (Figure 2). Reserves in relation toimports experienced significantly higher growth,over the course of the program, in uninterruptedprograms than in interrupted ones. Changes in thereserve cover of imports were also significantly andpositively correlated with the share of disbursedfunds and, in one case, with the no-interruptiondummy. However, the correlation of the reserves-to-imports ratio with the overall implementation indextook the “wrong” sign (it was negative), although itwas insignificant in almost all cases.

Countries that completed their IMF-supportedprograms started with deeper recessions (more nega-tive GDP growth rates) but grew faster than countrieswhere programs were interrupted, both right afterthe programs expired and for a couple of years afterthat (Figure 3). However, these differences in growthrates were not statistically significant. Once initialGDP and inflation are controlled for, only the over-all implementation index was significantly positivelycorrelated with growth in the program’s end year.

What, then, is the association between programimplementation and macroeconomic performance?Although not especially strong in our sample, theseresults provide some evidence that countries thatcomplete their IMF-supported programs also man-age, on average, to reduce inflation, increase theirrelative reserve holdings, gain export competitive-ness, and accelerate growth more than do countrieswhere programs are interrupted. These results aregenerally consistent with those of the literature:program implementation helps countries strengthentheir current account, external reserves, and bal-

ance of payments.12 Economic growth, which is de-pressed in the short run as program reforms begin to“bite,” also improves eventually. One noteworthydifference with previous studies concerns inflationperformance. Whereas previous studies generallyhave been inconclusive regarding the impact ofIMF-supported programs on inflation, inflation per-formance improves with program implementationin our sample.13

Econometric Analysis

Model Specification

We identify three major groups of factors that mightaffect the prospects of successful implementation ofIMF-supported programs. These are political econ-omy variables, variables describing IMF behavior,and initial and external conditions.

On the political economy side, we collected datafrom various sources, namely, the Political Institu-tions Database at the World Bank (Beck and others,2001), the International Country Risk Guide (ICRG),

12 The authoritative survey of the empirical literature is Haqueand Khan (1998). See also Schadler and others (1995a, 1995b);Conway (1994, 1998); and Joyce (2002).

13 Studying fully the relationship between success in IMF pro-grams and improvement in macroeconomic performance requiresa more elaborate econometric framework. In particular, oneneeds to take into account the dynamic structure of participationin IMF programs. Conway (2000) presents such a framework.

–2

–1.5

–1

–0.5

0

0.5

1

1.5

2

2.5

3

0 1 2 3 4 5 6

Year 0 = one year before approval year 1 = approval year 2 = end year 3 = one year after end year 4 = two years after end year

Aver

age

real

GD

P gr

owth

in a

giv

en y

ear

Interrupted


Figure 3

Real GDP Growth Dynamics for Successfully Implemented and Interrupted IMF Programs

the Polity IV dataset, and the CIA World Factbook.The main hypothesis that emerges from the theoret-ical model presented in a companion paper (Mayerand Mourmouras, 2002) is that the implementationof reforms is affected by the strength of special inter-est groups in countries using IMF resources. In prac-tice, it is difficult to identify and measure thestrength of organized lobbies. To develop a suitablemeasure of the strength of special interests, we reliedon the observation that in many countries politicalparties represented in government or the legislature(or both) sometimes represent specific interests.Legislatures are crucial players in policymaking: leg-islative approval is required for successful imple-mentation of almost all key reforms.14 While manydifferent organized interest groups can and do blockreforms, special interest groups in parliament seem anatural candidate.

The Political Institutions Database (Beck andothers, 2001) identifies four groups of parties in par-liament that represent nationalistic, rural, regional,and religious special interests. Key components ofthe platforms of these parties are the creation or de-fense of a national or ethnic identity and of rural, re-gional, or religious issues. Sometimes nationalisticspecial interests have persecuted minorities (nation-alist special interests), with disastrous consequencesfor economic development. In any event, special in-terests in parliament influence government policychoices through the exercise of their political powerand, perhaps, through monetary exchanges.

An important question is whether the interests ofthe political parties representing these interestgroups run counter to the reform objectives of IMF-supported programs. While the motives of each ofthese four types of parliamentary groups are differ-ent, each is clearly committed to promoting the in-terests of only a segment of the population. As such,these parties are likely to support policies favored bythe groups they represent even if they harm aggre-gate welfare. In short, special interests in the parlia-ment serve as our proxy for special interest groups inthe theoretical model. To test whether the presenceof influential lobbies lowers the probability of suc-cessful program implementation, we use the maxi-mum share of seats in parliament held by partiesthat represent nationalistic, religious, rural, and re-gional interest groups as a measure of the strength ofspecial interests.

Regarding the remaining political economy vari-ables, we include political instability, ethnic frac-tionalization and ethnic fractionalization squared,political cohesion, and the interaction term of thequality of bureaucracy and the change of chief exec-utive. (See the appendix for more details on the def-initions of the political variables and their sources.)Program implementation might be jeopardized by

political instability, which measures the degree ofinternal conflict and the extent of drastic politicalchange, such as the installation of a new chief exec-utive. Ethnic fractionalization may lead to tensionsin society and is, therefore, a potential threat to re-forms. Political cohesion emphasizes the heteroge-neous nature of the government and the legislature.In countries with poor bureaucracies, changes ingovernment tend to be traumatic as they are oftenaccompanied by disruptions in policy formulationand day-to-day administrative functions, which canhave a negative impact on program implementa-tion. A high-quality bureaucracy has the strengthand expertise to govern without drastic changes inpolicies and, therefore, can act as a shock absorberto reduce policy deviations from program goalswhen governments change. Since the importance ofbureaucracy is more sharply felt in times of govern-ment change, we included only a term that interactsthe strength of the bureaucracy with the dummyvariable indicating a change of chief executive.15

To test how factors under IMF control affect pro-gram implementation, we included three majorgroups of factors in our regressions: measures of IMFeffort, the extent of IMF financing, and measures ofthe extent and structure of conditionality.

To test the hypothesis that more IMF support im-proves the prospects of programs, we constructedthree variables. The first is IMF effort, measured bythe dollar cost of each program. This variable isbased on (a) internal IMF data on staff hours allo-cated to Use of Fund Resources (UFR) work, whichis program related, and staff hours devoted to tech-nical assistance and support tasks in member coun-tries; (b) information on average staff salaries bygrade; and (c) the costs of running the IMF’s resi-dent representative offices in member countrieswith programs (data were provided by the IMF’s Office of Budget and Planning (OBP) and Office ofPersonnel Management (OPM)). The second vari-able is the number of IMF staff missions, and thethird is the number of mission days.16

It also has been argued that the size of IMF loansmay not be large enough to induce substantialchanges in domestic policies. To test how the extentof IMF financing influences program implementa-tion, we included loan size as a percentage of quotain our regressions.

168


14 Hence, this test is also related to the theory of veto players.See Drazen (2002) and Tsebelis (2001).

15 When we included the quality of bureaucracy itself in the re-gression, the coefficient on that term was not statistically signifi-cant.

16 For all IMF effort variables, we had to make a decision onhow to attribute the data on hours/missions available by coun-tries and months to specific programs. We used approval datesand actual end dates of programs. Recognizing that we might belosing a significant part of IMF effort invested in program prepa-ration, we also constructed alternative measures of these vari-ables, taking into account IMF effort in the country three and sixmonths before program approval. Econometric results for alter-native measures were essentially the same and are not reportedhere but are available from the authors upon request.

169


To analyze the impact of conditionality on pro-gram implementation, we employed the followingmeasures: (a) the number of conditions per programyear, which measures the extent of overall condi-tionality; (b) the share of quantitative performancecriteria waived, which measures the strength of en-forcement and associated flexibility of conditional-ity; and (c) the share of structural conditions in thetotal number of conditions, which measures theweight that programs put on structural reforms. Asan alternative to the last measure, we also includedthe number of structural conditions per programyear in the regressions to capture the extent of struc-tural conditionality. As the results were unaffected,they are not reported separately.

Variables under IMF control are endogenously de-termined. Hence, a list of appropriate instrumentalvariables (IVs) must be employed in order to gleanthe impact of IMF variables on the probability ofsuccessful implementation of IMF-supported pro-grams. These instruments must be correlated withvariables that are under IMF control, be uncorre-lated with the shocks hitting programs, and not bedirect determinants of program implementation.

The choice of instruments is described in more de-tail in Box 1.

Another key issue is the impact of initial and ex-ternal conditions and shocks on the implementa-tion of IMF-supported programs. One possibility isthat countries that start with unfavorable initialconditions or are hit by unfavorable shocks have aharder time meeting program targets. Alternatively,these countries could face stronger incentives to re-form and might be more successful in implementingIMF-supported programs. A third possibility is thatprograms are designed and negotiated optimally,taking into consideration all the relevant factors, in-cluding initial conditions and the frequency, inten-sity, and nature of economic and other shocks. Ifprograms are tailored to the circumstances of eachmember country, differences in initial or externalconditions and in exposure to shocks may not play abig role in program implementation.17 It turns out

Box 1. List of Instrumental Variables

It is difficult to find instruments for all endogenousvariables simultaneously. Out of all IMF variables, theshare of structural conditions in the total number ofconditions seems the least subject to later revisions inthe course of the program, so we treat this variable asexogenous. For the remaining IMF variables, we usethe following IVs (see Table 6 for first-stage regres-sions). F-statistics on the IV set for all endogenousvariables were significant.

The average share of bilateral aid provided to the countryby the Group of Seven (G-7) before the start of theprogram.1 This variable is positively correlated withthe loan size in relation to quota and with the share ofquantitative performance criteria (PCs) waived, al-though these correlations are not significant even atthe 10 percent significance level.

Approval year. Since the number of conditions perprogram year has been increasing over time, it is posi-tively correlated with the approval year and we can usethe latter as an IV.

Expected program duration. A program’s expected du-ration is positively correlated with the loan size in rela-tion to quota and negatively correlated with the shareof quantitative PCs waived. The longer the program,the larger is the loan and the more time the IMF has toadjust its conditionality.

IMF quota (log). The quotas of members with IMF-supported programs are significantly positively corre-lated with IMF effort per program year and with theshare of quantitative PCs waived. A higher quota is as-sociated with greater IMF effort in a program and ahigher share of quantitative PCs waived for two mainreasons. First, the quota determines the size of the IMFloan to a member and the amount “at stake” for theIMF. Second, the quota also determines the member’svoting power in the IMF.

GDP per capita (log). This variable is negatively cor-related with IMF effort per program year. Richer coun-tries require less IMF effort, get higher loans as a per-centage of quota, receive fewer waivers, and get fewerconditions per program year. (This coefficient is signif-icant at 10 percent significance level only.) This is theonly initial condition included in the IV set.2

Regional dummies. IMF effort per program year ishigher in Latin America and the Caribbean comparedwith Europe and the Middle East. Compared with theother regions, loan size in relation to quota is higher inLatin America and the Caribbean, sub-SaharanAfrica, and East Asia. The share of quantitative PCswaived is higher in East Asia (significant at the 10 per-cent significance level only).

Population (log). This variable is negatively correlatedwith the share of quantitative PCs waived and positivelycorrelated with the loan size as a percentage of quota.

1 Since G-7 members comprise 45 percent of the IMF’s vot-ing power, this variable could be related to the “weight” theIMF puts on particular borrowers. See Mayer and Mour-mouras (2002) for details.

2 This variable was not significantly correlated with suc-cessful program implementation when we included it in theoriginal regression.

17 Tailoring programs to members’ circumstances is a key prin-ciple underlying the IMF’s 2002 conditionality guidelines (IMF,2002). On flexibility in the design of IMF-supported programs,see also Mussa and Savastano (1999); and Boughton and Mour-mouras (2004).

170


that it is not possible to distinguish empiricallyamong these three possibilities. All we can say isthat the data are consistent with the notion thatinitial and external conditions do not represent amajor stumbling block for program implementation.

Variables included as initial conditions in our re-gressions were as follows: the central governmentfiscal balance in relation to GDP, the current ac-count balance in relation to GDP, the level of grossreserves at the start of the program, initial inflation,initial GDP per capita, and initial debt to the IMFin relation to a member’s IMF quota. To control forexternal conditions, we use the term “trade shock”—namely the difference between the growth rate ofdollar export prices times the share of exports inGDP and the growth rate of dollar import pricestimes the share of imports in GDP.

Econometric Methodology

Our strategy is to relate the various indicators of im-plementation, either in isolation or in a pooledsample, to various right-hand-side variables. These“explanatory” variables include observable charac-teristics of borrowing countries, such as initial con-ditions and features of their domestic political econ-omy, and variables under IMF control, as describedin the previous section.

Our choice of econometric technique was guidedby the need to make efficient use of the informationcontained in our implementation indicators and bydata availability. One complication is that one ofour indicators is a binary variable while the othertwo vary continuously, which makes it difficult tocombine all three in a single model. Limited avail-ability of political economy data is an additionalconsideration. Even though implementation mea-sures are available for 170 programs, political econ-omy variables are available for only about 60 pro-grams. Crucially, some of the political economy dataare not available for all former centrally plannedeconomies for the period under consideration.18

The limited sample also forced us to set aside prob-lems of prolonged use of IMF resources. As some ofthe countries in the complete sample had multipleprograms with the IMF, there is strong cross-sectional correlation between observations in theentire sample. But since 56 percent of our workingsample comprised countries with only one program,and only 8 percent of countries in the sample had

three or more programs, we could not apply paneldata techniques.

Owing to the small sample size, we estimate sev-eral specifications to check the robustness of ourconclusions. Our approach is to first apply the Mul-tiple Indicators and Multiple Causes (MIMIC)19

model (see Joreskog and Goldberger, 1975), whichcombines three implementation measures in oneeconometric model. We then re-estimate modelsthat feature each implementation measure sepa-rately using proper econometric techniques.Amemiya’s IV probit method is employed to esti-mate regressions where the left-hand-side variable isa binary indicator. Amemiya’s IV tobit is used in re-gressions of the share of disbursed funds and theoverall implementation index.

Formally, our model can be described as follows. If yi* is the unobservable probability of successfulprogram implementation, then

yi* = ay + γy Pi + βyFi + εyi , (1.1)

where Pi is a vector of country i political economyvariables; Fi is a vector of variables under IMF con-trol; ay, γy, and βy are vectors of coefficients; and εyiis a stochastic disturbance term. The variables con-trolled by the IMF are given by

Fi = αF + γFPi + λFZi + εFi , (1.2)

where αF, γF, and λF are vectors of coefficients; εFi isanother error term; and Zi is a vector of exogenousvariables that are correlated with donor behaviorbut do not systematically influence the probabilityof successful implementation. Since the IMF re-sponds to shocks hitting programs by adjusting itseffort and conditionality, εyi and εFi are correlated.We use IV techniques to obtain consistent estimatesof the coefficients in equation (1.1).

Since we do not observe yi*, we cannot estimateequation (1.1) directly. However, we have three in-dicators of implementation, which are correlatedwith yi*. We can relate our observed measures of im-plementation to the unobserved probability of suc-cessful implementation as follows:

yi1 = δ1yi* + Ui1 (1.3)

yi2 = δ2yi* + Ui2 (1.4)

yi3 = δ3yi* + Ui3, (1.5)

where yi1, yi2, and yi3 are our three implementa-tion measures, and Ui1, Ui2, and Ui3 are measure-ment errors that are possibly mutually correlated.

18 This is unfortunate, as economies in transition are good“candidates” for testing the negative impact of special interestgroups on the implementation of IMF-supported programs. Rent-seeking behavior and state capture in transition economies arewell documented in the literature: see Hellman and Kaufmann(2001); Åslund (1999); Odling-Smee (2001); Havrylyshyn andOdling-Smee (2000); and the discussion in the conference ver-sion of this paper available at http://www.imf.org/external/pubs/ft/staffp/2001/00-00/pdf/aiwmgaam.pdf.

19 The MIMIC model is a special case of covariance structuremodel (LISREL), which is a generalization of the factor analysismodel.

171


Equations (1.1)–(1.5) represent a special case of theMIMIC model analyzed in Joreskog and Goldberger(1975). To estimate this model, we first substituteequation (1.2) into (1.1) and (1.1) into (1.3)–(1.5)to obtain a system of equations that can be treatedas seemingly unrelated regressions. This system canbe estimated to obtain reduced form coefficientsthat we can use to recover the parameters γy and βy.To calculate the variance of γy and βy we employ thedelta method. This approach requires normalizationof one of the coefficients δ to one. Since the modelis overidentified, we also had to impose nonlinearconstraints to obtain unique parameter estimates.Because of computational complexity, we estimatethe general form of the MIMIC model (1.1)–(1.5)including only one variable under IMF control,namely the IMF effort.

A computationally convenient version of thismodel arises if the coefficients δ are all unity. In thiscase, substituting (1.1) into (1.3)–(1.5) and settingthe δs to one, we have:

yi1= αy+ γyPi + βyFi+ εyi + Ui1 (1.6)

yi2= αy+ γyPi + βyFi + εyi + Ui2 (1.7)

yi3= αy+γyPi + βyFi + εyi + Ui3. (1.8)

The system (1.6)–(1.8) is a random-effects modelwith random effect εyi. If IMF effort were not simulta-neously determined with the probability of successfulimplementation, then the random effect εyi would beuncorrelated with the set of regressors in Fi and be Pi.We could then obtain consistent estimates of thismodel by pooling the three implementation measuresin one variable and regressing it on the same set ofpolitical economy and IMF effort variables for a par-ticular program using the random-effects estimator.However, since IMF effort is simultaneously deter-mined with the probability of successful implemen-tation, we apply the random-effects IV estimator toobtain consistent estimates of the coefficients onpolitical economy and IMF effort variables.20

To summarize, we proceed as follows: we first esti-mate linear-in-probability and tobit regressions thatcombine three implementation measures in onemodel employing the random-effects estimator—that is, equations (1.6)–(1.8). Two variants are ex-amined, one that ignores the endogeneity of vari-ables under the IMF’s control (Table 4) and anotherdealing with this endogeneity through IV tech-niques (Table 5). The set of IVs employed is speci-fied in Table 6. Table 5 also reports a third, moregeneral, version of the MIMIC model. This is speci-fication (1.1)–(1.5) with only one endogenous vari-able, namely, IMF effort per program year (Table 5,

column 1).21 We then re-estimate our chosen speci-fication of political economy variables with each ofthe implementation indices in isolation, not takinginto account endogeneity of the variables underIMF control (Table 7) and instrumenting for thesevariables (Tables 8–10).22

Results

Main Findings

Program prospects depend on the domestic politicaleconomy. In particular, strong vested interests inparliament, lack of political cohesion, poor qualityof bureaucracy, and ethnic divisions significantlyundermine program implementation. We first esti-mated random-effects regressions on a pooled sam-ple, both for linear-in-probability and tobit specifi-cations (Table 4, column 2). The coefficient on thestrength of special interests is negative and signifi-cant at the 5 percent significance level. The strongempirical evidence of the adverse role of special in-terests on reforms is reassuring because it comesfrom a sample that excludes some transitioneconomies. The coefficients on the index of politi-cal cohesion as well as on the interaction term ofthe quality of bureaucracy and the change of chiefexecutive is positive and significant. Interestingly,once we added to the regression three more politicaleconomy variables,23 which might affect the proba-bility of successful implementation, the coefficientson ethnic fractionalization and ethnic fractionaliza-tion squared became significant.

The impact of ethnic fractionalization on pro-gram performance is nonlinear. Large and small eth-nic divisions are both bad for program implementa-tion.24 The results remain essentially the same whenwe re-estimate the model using the more generalMIMIC specification given by equations (1.1)–(1.5)(Table 5) and when each of the implementationmeasures are considered in isolation (Tables 7–10).25

20 Note that since the question is narrowly focused on a set ofcountries that each had an IMF-supported program, there is noissue of selection bias.

21 The reason for testing the hypothesis about the importanceof IMF effort only in this model is that computing standard errorsusing the delta method with more than one endogenous variablein the MIMIC model is cumbersome.

22 We estimate IV regressions on each of the implementationmeasures separately using Amemiya’s generalized-least-squares(GLS) IV probit/tobit estimators.

23 Table 5 (column 1) presents regressions of our implementa-tion measures on the political economy variables used by Dollarand Svensson (2000). These coefficients are insignificant, bothindividually and jointly.

24 The turning point varies between 44 and 55 on a 0–100scale (Tables 5–7). This is close to the range estimate (44–49)obtained by Dollar and Svensson (2000) in their study of WorldBank programs.

25 In this model, δ =1 only in the equation relating the proba-bility of successful implementation and irreversible interruptiondummy while allowing the other two δs to vary.

172


Table 4

Random-Effects Model: Linear in Probability and Tobit Regressions

Regression Number (1) (2) (3) (4)

Dependent variable: Linear in Linear in Linear in Linear in Program implementation indices probability Tobit probability Tobit probability Tobit probability Tobit

Number of observations 240 240 170 170 179 179 167 167

Political economy variablesEthnic fractionalization 0.12 0.31 1.07 1.99 1.16 2.14 1.31 2.45

(0.25) (0.34) (2.07) (2.34) (2.46) (2.64) (2.70) (2.94)Ethnic fractionalization (squared) 0.00 0.00 –0.01 –0.02 –0.01 –0.02 –0.01 –0.02

–(0.17) –(0.39) –(1.34) –(1.76) –(2.03) –(2.30) –(2.38) –(2.59)Political instability1 –0.39 –1.31 –1.49 –5.47 –2.85 –5.87 –3.11 –5.54

–(0.41) –(0.75) –(0.69) –(1.53) –(1.87) –(2.23) –(2.03) –(2.15)Executive index of electoral

competitiveness2 9.64 15.18 8.23 18.85 12.69 22.36 13.25 23.33(1.27) (1.11) (0.88) (1.28) (1.57) (1.64) (1.67) (1.79)

Time in power 0.34 1.90 1.32 3.70 1.22 3.54 2.24 4.88(0.30) (0.91) (0.73) (1.29) (0.78) (1.33) (1.46) (1.90)

Time in power (squared) 0.00 –0.04 –0.07 –0.18 –0.07 –0.16 –0.10 –0.20–(0.08) –(0.80) –(1.21) –(1.92) –(1.29) –(1.84) –(1.93) –(2.35)

Strength of special interests3 –31.72 –69.73 –34.46 –68.41 –36.39 –70.49–(2.14) –(2.87) –(3.08) –(3.53) –(3.19) –(3.62)

Index of political cohesion4 9.66 19.52 11.49 20.58 13.22 22.85(1.98) (2.50) (2.95) (3.11) (3.20) (3.32)

Quality of bureaucracy interacted with change of chief executive5 12.82 33.25 16.25 31.30 20.77 36.82

(1.19) (1.85) (1.92) (2.13) (2.52) (2.63)

Initial and external conditionsCentral government balance

(percentage of GDP) 0.93 0.98(0.76) (0.50)

Level of reserves 0.00 0.00(0.46) (0.72)

Inflation 0.00 –0.27(0.01) –(0.97)

Current account balance (percentage of GDP) –95 –123

–(1.26) –(1.02)GDP per capita (log) 5.39 4.75

(0.81) (0.44)Debt to the IMF

(percentage of IMF quota) 4.62 13.21(0.37) (0.63)

Terms-of-trade shock6 –0.01 –0.01–(1.15) –(0.34)

Variables under IMF ControlIMF effort per program year 3.81 11.91

(log)7 (0.78) (1.48)

Loan size as percentage of quota (log) 7.01 11.45

(1.63) (1.55)Number of conditions per

program year (log) –6.43 –13.80–(0.99) –(1.29)


173


Table 4 (concluded)

Regression Number (1) (2) (3) (4)

Dependent variable: Linear in Linear in Linear in Linear in Program implementation indices probability Tobit probability Tobit probability Tobit probability Tobit

Share of quantitative PCs waived (percent) –0.49 –1.05

–(1.98) –(2.56)Share of structural conditions

(percent) 0.16 0.21(0.98) (0.77)

Wald chi-square statistics 2.95 5.00 34.29 42.63 33.32 38.31 45.86 48.33p-value 0.81 0.54 0.01 0.00 0.00 0.00 0.00 0.00

Notes: Bold figures indicate significance at the 5 percent level; bold italic figures indicate significance at the 10 percent level. PCs denote per-formance criteria.

The model was estimated on a pooled sample of three implementation measures as left-hand-side variables, ignoring the endogeneity ofvariables under IMF control. The measures of program implementation used are (a) a binary variable indicating no irreversible program inter-ruption; (b) the share of funds committed by the IMF under an arrangement disbursed (we excluded the measure of committed funds dis-bursed for arrangements precautionary on approval; canceled programs that did not have irreversible interruption; and arrangements thatturned precautionary were treated as fully disbursed (100 percent)); and (c) the average share of conditions implemented. Regression also in-cluded the constant term, which is omitted in the table.

1 This index is computed based on the index of internal conflict provided by the International Country Risk Guide (ICRG) on a scale from zeroto 12. Higher values of the index correspond to more internal political instability. We replaced the value of this variable by its maximum score(12) if there was a change of chief executive in the course of the IMF-supported program.

2 Dummy variable equals one if the executive index of electoral competitiveness is equal to seven, and zero otherwise. The executive index ofelectoral competitiveness is from the Database of Political Institutions at the World Bank. It ranges from one to seven, with higher values corre-sponding to more competitive elections.

3 Computed as the maximum share of seats in the parliament held by parties representing special interests (Political Institutions Database,World Bank). Four special interest groups are identified: religious, nationalistic, regional, and rural.

4 The index of political cohesion is defined as follows: in presidential systems a high degree of political cohesion is said to exist if the sameparty is in control of the executive and legislature; in parliamentary systems a high degree of political cohesion means a one-party majority gov-ernment. See the appendix for a more detailed definition.

5 Bureaucracy quality (ICRG) measures the quality of a country's bureaucracy on a four-point scale. See the appendix for a more detailed de-finition. This variable is interacted with the dummy indicating that there was a change of chief executive (Political Institutions Database and CIAWorld Factbook for most recent years).

6 Average growth rate of dollar export prices multiplied by the initial share of exports in GDP minus average growth rate of dollar importprices multiplied by the initial share of imports in GDP over the course of the program

7 IMF effort is the estimated dollar cost of IMF-supported programs computed based on the IMF’s Budget Reporting System (BRS) data onhours spent by the staff on program implementation (which includes both preparation and supervision of the program) and estimated averagesalaries of the staff by grade. We also made use of data provided by the IMF’s Office of Personnel Management (OPM) on the dollar costs of res-ident representatives.

174


Table 5

Random-Effects (IV) and MIMIC Models: Linear-in-Probability Regressions

MIMIC Random Effects IV Regressions1 Model2

Regression Number (1) (2) (3) (4) (5) (6)

Number of observations 165 165 165 165 165 165

Dollar and Svensson variablesEthnic fractionalization 1.08 0.91 0.94 1.25 0.99 2.50

(2.09) (1.70) (1.75) (2.44) (2.05) (3.55)Ethnic fractionalization (squared) –0.01 –0.01 –0.01 –0.01 –0.01 –0.03

–(1.76) –(1.42) –(1.48) –(2.07) –(1.68) –(3.53)Political Instability3 –3.91 –3.82 –3.93 –4.58 –4.48 –4.23

–(2.49) –(2.43) –(2.50) –(2.90) –(2.82) –(2.20)

Other political economy variablesStrength of special interests4 –39.78 –33.86 –34.01 –45.47 –37.38 –38.69

–(3.57) –(2.80) –(2.81) –(3.99) –(3.24) –(2.48)Index of political cohesion5 10.03 10.55 8.78 9.86 9.69 14.02

(2.74) (2.87) (2.17) (2.43) (2.39) (2.86)Bureaucracy quality interacted with

change of chief executive6 21.39 21.36 22.02 24.97 24.50 21.05(2.68) (2.68) (2.75) (3.11) (3.02) (2.16)

Variables under IMF controlIMF effort per program year (log)7,8 1.54 –2.14 –0.49 7.06 –9.29

(0.23) –(0.29) –(0.07) (0.99) –(1.06)Loan size as a percentage of quota (log)8 6.89 7.35 6.58

(1.26) (1.33) (1.30)Number of conditions per program year (log)8 –9.87 –15.77 –13.70

–(1.05) –(1.55) –(1.43)Share of quantitative PCs waived (percent)8 –0.67 –0.47

–(1.59) –(1.17)Share of structural conditions (percent) 0.14 0.11 0.06 0.16 0.14

(0.83) (0.62) (0.32) (0.92) (0.77)

Wald chi-square statistics 26.92 28.50 29.57 31.62 31.93p-value 0.00 0.00 0.00 0.00 0.00

Overidentifying restrictions test 12.67 10.42 8.83 8.04 6.91Degrees of freedom 9 8 7 7 7p-value 0.18 0.24 0.27 0.33 0.44

Hausman t-test 0.38 0.92 1.04 0.84 0.83p-value 0.54 0.63 0.79 0.84 0.84

Notes: Boldfaced figures indicate significance at the 5 percent level; bold italic figures indicate significance at the 10 percent level. PCs denote performance criteria.1 This model was estimated on a pooled sample of three implementation measures as left–hand-side variables, using a random-effects IV estimator with the set of

instruments as specified in Table 6. The measures of program implementation used are (a) a binary variable indicating no irreversible program interruption; (b) theshare of funds committed by the IMF under an arrangement actually disbursed. We excluded the measure of committed funds disbursed for arrangements precaution-ary on approval; canceled programs that did not have irreversible interruption; and arrangements that turned precautionary, which were treated as fully disbursed(100 percent); and (c) the average share of conditions implemented. The regression also included the constant term, which is omitted from the table.

2 This model comprises equations (1.1)—(1.5) in the text. It is essentially a system of seemingly unrelated regressions which can be estimated to obtain reduced-formparameters. Since the model is overidentified, we had to impose nonlinear constraints to obtain unique estimates of coefficients. Then the structural parameters werecomputed using estimates of reduced-form parameters and their variance was estimated using the delta method. (More details are available from the authors upon request.)

3 This index is computed based on the index of internal conflict provided by the International Country Risk Guide (ICRG) on a scale from 0 to 12. Higher values of theindex correspond to more internal political instability. We replaced the value of this variable by its maximum score (12) if there was a change of chief executive in thecourse of the IMF-supported program.


5 The index of political cohesion is defined as follows: in presidential systems a high degree of political cohesion is said to exist if the same party is in control of theexecutive and legislature; in parliamentary systems a high degree of political cohesion means a one-party majority government. See the appendix for a more detaileddefinition.

6 Bureaucracy quality (ICRG) measures the quality of a country's bureaucracy on a four-point scale. See the Appendix for a more detailed definition. This variable isinteracted with the dummy variable indicating that there was a change of chief executive (Political Institutions Database and CIA World Factbook for most recent years).

7 IMF effort is the estimated dollar cost of IMF-supported programs computed based on the IMF’s Budget Reporting System (BRS) data on hours spent by the staffon program implementation (which includes both preparation and supervision of the program) and estimated average salaries of the staff by grade. We also made useof data provided by the IMF’s Office of Personnel Management (OPM) on the dollar costs of resident representatives.

8 Treated as an endogenous variable in this regression.

175


Table 6

First-Stage Regressions

Number of Share of IMF Effort Loan Size as Conditions Quantitative

Per Program Percentage Per Program PCs Waived Dependent Variable Year (log)1 of Quota (log) Year (log) (percent)

Number of observations 57 57 57 57

Political economy variablesEthnic fractionalization –0.021 0.006 0.008 0.031

–(1.56) (0.58) (0.84) (0.12)Ethnic fractionalization (squared) 0.000 0.000 0.000 –0.002

(1.39) –(0.51) (0.620) –(0.57)Political instability2 –0.034 0.008 –0.012 0.316

–(0.66) (0.20) –(0.34) (0.34)Strength of special interests3 0.391 –0.05 –0.042 –9.424

(1.15) –(0.20) –(0.18) –(1.51)Index of political cohesion4 –0.008 –0.092 –0.093 –0.060

–(0.06) –(0.86) –(0.98) –(0.02)Bureaucracy quality interacted with change

of chief executive5 0.007 0.009 0.055 0.391(0.02) (0.04) (0.29) (0.08)

Variables under IMF controlShare of structural conditions (percent) –0.003 0.007 0.009 0.064

–(0.30) (1.11) (1.59) (0.42)

InstrumentsAverage share of bilateral aid by Group of Seven

(G-7) to the country before the program start 0.060 0.153 –0.038 3.009(0.49) (1.60) –(0.45) (1.35)

Approval year 0.119 0.052 0.110 –0.199(1.55) (0.86) (2.04) –(0.14)

Expected program duration 0.123 0.251 0.059 –5.512(0.93) (2.40) (0.63) –(2.27)

IMF quota (log) 0.469 –0.128 0.219 17.421(2.01) –(0.70) (1.33) (4.06)

Dummy for ESAF/PRGF 0.306 0.167 –1.063 16.555(0.80) (0.56) –(3.95) (2.37)

GDP per capita (log) –0.442 0.418 –0.218 –8.443–(2.45) (2.93) –(1.71) –(2.55)

Latin America and Caribbean 1.095 0.784 0.313 0.303(2.29) (2.08) (0.93) (0.03)

Sub-Saharan Africa 0.222 0.968 –0.046 13.041(0.40) (2.24) –(0.12) (1.30)

East Asia 0.724 1.609 –0.005 18.280(1.31) (3.70) –(0.01) (1.81)

Population (log) –0.020 0.291 –0.121 –10.740–(0.11) (2.02) –(0.94) –(3.21)

R2 0.56 0.75 0.57 0.549F-statistic on instruments 3.37 9.48 3.55 2.94p-value 0.00 0.00 0.00 0.01

Notes: PCs denote performance criteria; ESAF denotes Enhanced Structural Adjustment Facility; and PRGF denotes Poverty Reduction and Growth Facility. Estimatedby ordinary least squares (OLS) with robust standard errors. Regression also included the constant term, which is omitted in the table. Bold figures indicate significanceat the 5 percent level; bold italic figures indicate significance at 10 percent level.


2 This index is computed based on the index of internal conflict provided by the International Country Risk Guide (ICRG) on a scale from zero to 12. Higher values ofthe index correspond to more internal political instability. We replaced the value of this variable by its maximum score (12) if there was a change of chief executive inthe course of the IMF-supported program.



5 Bureaucracy quality (ICRG) measures the quality of a country's bureaucracy on a four-point scale. See the appendix for a more detailed definition. This variable isinteracted with the dummy indicating that there was a change of chief executive (Political Institutions Database and CIA World Factbook for most recent years).

176


Table 7

Linear-in-Probability and Probit/Tobit Regressions on Three Implementation MeasuresSeparately Ignoring Endogeneity of Variables Under IMF Control

Our Specification of Political Economy Variables + Variables Under the IMF’s Control

Non-interruption Share of committed Average overall Dependent Variable dummy IMF disbursed1 implementation index

Linear in Linear in Linear in Model probability Probit probability Tobit probability Tobit

Number of observations 57 57 53 53 55 55

Dollar and Svensson variablesEthnic fractionalization 1.50 0.05 1.13 1.98 0.61 0.66

(1.58) (1.43) (2.28) (2.24) (2.47) (2.91)

Ethnic fractionalization (squared) –0.01 0.00 –0.01 –0.02 –0.01 –0.01–(1.18) –(1.06) –(2.05) –(1.91) –(2.34) –(2.84)

Political instability2 –8.13 –0.31 –3.79 –6.31 –0.37 –0.30

–(3.22) –(2.52) –(2.10) –(2.51) –(0.41) –(0.44)

Other political economy variablesStrength of special interests3 –73.98 –2.83 –32.17 –60.48 –17.07 –17.25

–(3.92) –(3.03) –(2.92) –(3.02) –(2.96) –(3.37)

Index of political cohesion4 20.52 0.83 11.67 16.35 –0.83 –1.17

(3.26) (2.50) (3.00) (2.70) –(0.50) –(0.65)

Bureaucracy quality interacted with

change of chief executive5 45.85 1.83 17.28 28.47 4.36 4.28

(4.16) (2.51) (2.29) (2.20) (0.97) (1.21)

Variables under IMF control

IMF effort per program year (log)6 16.93 0.65 –0.88 6.37 2.42 2.88

(1.79) (1.85) –(0.17) (0.77) (1.29) (1.30)

Loan size as percentage of quota (log) 4.92 0.15 3.29 3.10 2.28 2.34

(0.54) (0.56) (0.64) (0.42) (1.63) (1.19)

Number of conditions per program year (log) –10.71 –0.49 –7.53 –13.18 –11.44 –12.93–(0.93) –(0.97) –(1.08) –(1.28) –(4.57) –(4.14)

Share of quantitative PCs waived (percent) –0.87 –0.03 0.09 –0.21 –0.56 –0.58–(1.45) –(2.02) (0.41) –(0.48) –(6.21) –(5.12)

Share of structural conditions (percent) 0.22 0.01 0.15 0.28 –0.12 –0.14(0.73) (0.86) (0.98) (1.00) –(1.60) –(1.78)

R2 0.41 0.46 0.58

Predictive ability of the model

(percent)7 75.44

Notes: Boldfaced figures indicate significance at the 5 percent level; bold italic figures indicate significance at the 10 percent level. PCs denote performance criteria.1 For the regression of the share of committed funds disbursed, we excluded arrangements precautionary on approval. Canceled programs that did not have irre-

versible interruption and arrangements that turned precautionary were treated as fully disbursed (100 percent).2 This index is computed based on the index of internal conflict provided by the International Country Risk Guide (ICRG) on a scale from zero to 12. Higher values of

the index correspond to more internal political instability. We replaced the value of this variable by its maximum score (12) if there was a change of chief executive inthe course of the IMF-supported program.



5 Bureaucracy quality (ICRG) measures the quality of a country's bureaucracy on a four-point scale. See the appendix for a more detailed definition. This variable isinteracted with the dummy variable indicating that there was a change of chief executive (Political Institutions Database and CIA World Factbook for most recent years).

6 IMF effort is the estimated dollar cost of IMF-supported programs computed based on the IMF’s Budget Reporting System (BRS) data on hours spent by the staffon program implementation (including both preparation and supervision of the program) and estimated average salaries of the staff by grade. We also made use dataprovided by the IMF’s Office of Personnel Management (OPM) on the dollar costs of resident representatives.

7 The predictive ability of the model is computed as follows: if the predicted value from the probit regression was higher or equal to 1⁄2, we count this prediction asno interruption; otherwise, we count it as an interruption. We then compare the actual outcome with the predicted outcome and compute the share of correct predictions.

177


Table 8

IV Regressions for Non-Interruption Dummy Taking into Account Endogeneity of Variables Under IMF Control

Dependent Variable Non-Interruption Dummy



Dollar and Svensson variablesEthnic fractionalization 0.042 0.033 0.027 0.037

(1.32) (1.09) (0.86) (1.17)Ethnic fractionalization (squared) 0.000 0.000 0.000 0.000

–(1.03) –(0.70) –(0.49) –(0.72)Political instability1 –0.256 –0.251 –0.255 –0.318

–(2.40) –(2.31) –(2.31) –(2.42)

Other political economy variablesStrength of special interests2 –2.329 –1.887 –1.974 –2.479

–(2.95) –(2.51) –(2.50) –(2.97)Index of political cohesion3 0.636 0.632 0.484 0.856

(2.33) (2.33) (1.54) (2.44)Bureaucracy quality interacted with change

of chief executive4 1.355 1.435 1.345 1.656(2.12) (2.36) (2.16) (2.14)

Variables under IMF controlIMF effort per program year (log)5,6 0.271

(0.57)Loan size as percentage of quota (log)6 0.464

(1.34)Number of conditions per program year (log)6 –0.356

–(0.52)Share of quantitative PCs waived (percent)6 –0.045

–(1.34)Predictive ability of model7 70.49 72.41 66.67 63.93

Notes: Boldfaced figures indicate significance at the 5 percent level. PCs denote performance criteria. ESAF denotes Enhanced Structural Adjustment Facility, andPRGF denotes Poverty Reduction and Growth Facility.

For IV estimation on each of the implementation measures separately, we use shorter sets of IVs:

For IMF effort per program year (log): expected program duration, quota (log), and GDP per capita (log);For loan size as percentage of quota (log): expected program duration, GDP per capita (log), and population (log);For number of conditions per program year (log): approval year, dummy for ESAF/PRGF, and GDP per capita (log); andFor share of quantitative PCs waived (percent): quota (log), GDP per capita (log), and population (log).

IV regression for non-interruption dummy was estimated using the two-stage Amemiya GLS procedure (IV probit).

1 This index is computed based on the index of internal conflict provided by the International Country Risk Guide (ICRG) on a scale from 0 to 12. Higher values ofthe index correspond to more internal political instability. We replaced the value of this variable by its maximum score (12) if there was a change of chief executive inthe course of the IMF-supported program.






7 The predictive ability of the model is computed as follows: if the predicted value from the probit regression is higher or equal to 1⁄2, we count this prediction as nointerruption; otherwise, we count this prediction as an interruption. We then compare the actual outcome with the predicted outcome and compute the share of correctpredictions.

178


Table 9

IV Regressions for the Average Share of Committed Funds Disbursed Taking into Account Endogeneity of Variables Under IMF Control

Dependent Variable Share of Committed Funds Disbursed




(1.38) (1.88) (1.95) (1.45)Ethnic fractionalization (squared) –0.010 –0.012 –0.013 –0.008


–(2.42) –(2.26) –(2.45) –(2.48)


–(2.58) –(2.80) –(2.97) –(2.77)Index of political cohesion3 16.820 17.807 13.194 18.703

(2.95) (3.18) (2.03) (2.87)Bureaucracy quality interacted with

change of chief executive4 26.162 23.426 25.209 27.253(2.00) (1.88) (1.99) (2.03)

Variables under IMF controlIMF effort per program year (log)5,6 –1.311

–(0.10)Loan size as percentage of quota (log)6 4.349

(0.49)Number of conditions per program year (log)6 –22.011

–(1.30)Share of quantitative PCs waived (percent)6 –0.638

–(0.78)

Notes: Boldfaced figures indicate significance at the 5 percent level. PCs denote performance criteria; ESAF denotes Enhanced Structural Adjustment Facility, andPRGF denotes Poverty Reduction and Growth Facility.



IV regression for the share of committed funds disbursed was estimated using two-stage Amemiya (1978) GLS procedure (IV tobit).

For the regression of the share of committed funds disbursed, we excluded arrangements precautionary on approval. Canceled programs that did not have irre-versible interruptions and arrangements that turned precautionary were treated as fully disbursed (100 percent).



3 The index of political cohesion is defined as follows: in presidential systems, a high degree of political cohesion is said to exist if the same party is in control of theexecutive and legislature; in parliamentary systems, a high degree of political cohesion means a one-party majority government. See the appendix for a more detaileddefinition.




179


Table 10

IV Regressions for Average Overall Implementation Index Taking into Account Endogeneity of Variables Under IMF Control

Dependent Variable Average Overall Implementation Index




(1.63) (2.49) (2.32) (2.69)Ethnic fractionalization (squared) –0.006 –0.008 –0.008 –0.007


–(0.38) –(0.09) –(0.09) –(0.56)


–(2.12) –(2.41) –(2.52) –(3.42)Index of political cohesion3 –1.675 –0.747 –0.911 1.143

–(0.77) –(0.35) –(0.34) (0.55)Bureaucracy quality interacted with

change of chief executive4 3.457 3.375 3.069 4.369(0.77) (0.77) (0.64) (1.11)

Variables under IMF controlIMF effort per program year (log)5,6 –5.095

–(1.15)Loan size as percentage of quota (log)6 3.188

(1.01)Number of conditions per program year (log)6 2.126

(0.32)Share of quantitative PCs waived (percent)6 –0.600

–(2.51)

Notes: Boldfaced figures indicate significance at the 5 percent level. PCs denote performance criteria; ESAF denotes Enhanced Structural Adjustment Facility, andPRGF denotes Poverty Reduction and Growth Facility.



IV regression for the average overall implementation index was estimated using the two-stage Amemiya GLS procedure (IV tobit).







Neither incumbents’ democratic credentials northeir newness in office is associated with better pro-gram implementation (Table 4).26 The coefficientson democratic election of a leader (dummy vari-able) and time in power were insignificant in almostall specifications. Likelihood-ratio tests for the tobitspecification confirmed that these exclusions didnot substantially worsen model performance. Thefirst result corroborates anecdotal evidence that theimplementation of IMF-supported programs doesnot suffer in countries with authoritarian regimes.

The magnitude and even direction of impact ofmost reforms is ambiguous, especially at the outset,making them unpopular with policymakers andtheir public even as these reforms enhance welfarein the long run. This may lead to democratic admin-istrations in developing or transition countries hav-ing a harder time than dictators marshaling the sup-port they need to pursue successful reforms. Theabsence of significant correlation between a govern-ment’s length of tenure and the probability of suc-cessful program implementation is also intriguing. Itsuggests that one should not expect too much ofnew, reform-minded governments implementingIMF-supported reforms in countries with adversepolitical economy characteristics. Perhaps the lackof correlation also reflects public sector characteris-tics we have not captured.

Initial and external economic conditions do notseem to influence program implementation muchonce political economy variables are taken into ac-count. The coefficients on all initial and externalconditions in the random-effects regressions cameout individually and jointly insignificant (Table 4).27

Initial conditions were insignificant in the IV re-gressions as well—to save space, we do not presentthese results here. The coefficients on the politicaleconomy variables do not change appreciably whenthe estimation excludes initial conditions (Table 4,column 3). As already mentioned, the fact that ini-tial conditions do not affect the probability of pro-gram implementation does not necessarily implythat IMF-supported programs are optimally de-signed. It does, however, indicate that unfavorableinitial or external conditions per se do not compro-mise programs’ prospects of being successfully imple-mented.

Variables controlled by the IMF, including finan-cial and human effort and the breadth and depth ofconditionality, do not affect program implementa-tion once domestic political economy variables are

taken into account. IMF effort was measured by thedollar cost of staff hours spent on UFR and on tech-nical assistance tasks per program year and the loansize in relation to a country’s IMF quota. The extentof conditionality was captured by the total numberof conditions per program year, the share of quanti-tative performance criteria waived, and the share ofstructural conditions in conditionality.28,29 Oncetheir endogeneity was accounted for, IMF-relatedvariables did not significantly affect the probabilityof successful program implementation (Tables 6,8–10).30 The overidentifying restrictions test con-firmed the validity of including additional IVs inthe regressions. The Hausman test suggests that IVrandom-effects regressions were not much differentfrom the simple random-effects model.

The coefficients on IMF-related variables wereinsignificant in many regressions when their endo-geneity was ignored (Tables 4 and 7, column 4). Wenote two exceptions. First, the share of quantitativeperformance criteria waived was, in several cases,negatively correlated with the probability of suc-cessful program implementation. This partly reflectsthe nature of the implementation index, which isassigned a value of zero if the condition is waived.Second, IMF effort was positively correlated (at the10 percent significance level) with the index ofcompletion of IMF-supported programs (Table 7).This correlation vanished when the endogeneity ofthese two variables was taken into account.

Illustration

It is helpful to illustrate the estimated impacts of po-litical economy variables on the probability of pro-gram implementation. Consider the marginal effectsof improved political stability, political cohesion,and the quality of bureaucracy, based on the IV re-gression of the no-interruption dummy (Table 8,

180


26 It will be recalled that Dollar and Svensson (2000) con-cluded that the implementation of Bank-supported programs im-proves in countries with democratically elected governments.

27 The null hypothesis that the coefficients on all of these vari-ables are jointly insignificant could not be rejected at the 5 per-cent significance level. Likelihood-ratio test statistics for this testgoing from the tobit regression in column 2 of Table 5 to thetobit regression in column 3 of Table 5 was 6.98 with p-valueequal to 0.43.

28 We also tried the number of structural conditions per pro-gram year as an alternative measure of the extent of structuralconditionality. We do not report the results of this estimation, asthe results were essentially the same. Interestingly, the coeffi-cient on the number of structural conditions turned negative inmany cases, although still insignificant.

29 IMF-related variables are included in these regressions tak-ing into account the limitations of our small sample size. See thenotes to Tables 8 through 10 for the description of the shorter IVsets used in the regressions of the implementation measures sepa-rately.

30 We also included the share of prior actions and conditionsfor completion of review in the total number of conditions in ourwork. We tried two regressions, one ignoring the endogeneity ofprior actions and a second one in which we instrumented for thisvariable. In both cases, the coefficient on prior actions was in-significant and is not reported. It appears that more careful studyon prior actions is needed in order to analyze their impact onprogram implementation. One consideration is the lack of infor-mation on programs that are not approved or of program reviewsthat are not completed as a result of failing to meet prior actions.(MONA does not provide this information.) In our view, it is un-likely that this result will change even if the selection bias isproperly accounted for.

181


column 1). For a country that enjoys perfect politi-cal stability and no special interests in parliament,the probability of program implementation is veryhigh (96 percent). If political stability is only aver-age, the chances of successful program implementa-tion decline to 70 percent (Figure 4). If parties rep-resenting special interests occupy 20 percent of theseats in parliament, a program only has a 50-50chance of implementation.

Lack of political cohesion reduces the probabilityof program implementation by 50 percentage points(from 70 percent to 20 percent) when there are nospecial interests. If 20 percent of the seats in parlia-ment are controlled by special interests, the proba-bility of program implementation drops another 10percentage points (Figure 5). The impact of a coun-try’s bureaucracy on program implementation is alsosubstantial. On the one hand, in the absence of spe-cial interests, the probability of program implemen-tation increases from 50 percent when the quality ofthe bureaucracy is low to 74 percent when the bu-reaucracy is of average quality (Figure 6). If, on theother hand, special interests control 20 percent ofthe seats in parliament, the probability of programimplementation increases from 33 percent when thequality of the bureaucracy is low to 50 percent whenthe bureaucracy is of average quality.

Robustness Checks and Limitations

Although our relatively small sample size makes itdifficult to reach definitive conclusions, our findings

appear to be robust to the specification of regres-sions, the choice of left-hand-side variable, and thechoice of the measure of IMF effort. As alreadydemonstrated, our main conclusions regarding theeffect of political economy and IMF-related vari-ables on program implementation are robust to theprecise specification of the econometric model. Esti-mating random-effects models on a pooled datasetand re-estimations using the appropriate probit andtobit technique for each of our three implementa-tion measures separately lead to similar conclusions.

Our basic conclusions also are robust to alterna-tive specifications of IMF effort. We tried various al-ternatives to our primary IMF effort variable (thedollar cost of IMF hours invested in country workbetween the approval and actual end dates of theprogram). Various other measures, such as the num-ber of missions per program year and the number ofmission days per program year, yield qualitativelysimilar results. Even though the number of missionsand mission days are positively and strongly corre-lated with program implementation when their en-dogeneity is not accounted for, this association dis-appears in the proper IV regressions.31 We alsoconsidered a measure of IMF effort scaled by theloan size, correcting for their strong correlation.While the IMF exerts greater effort in monitoringlarger loans (as measured by staff hours per dollar

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Strength of special interests

Prob

abili

ty o

f suc

cess

ful i

mpl

emen

tatio

n

Political stability

Average political instability

Figure 4

Probability of Successful Implementation, Strength of Special Interests and Political Instability

Note: Probabilities are evaluated at the means of other explanatory variables.

31 It will be recalled that this linkage was not present whenIMF effort was proxied by the estimated cost of IMF-supportedprograms.

182


0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1


Lack of political cohesion

Average political cohesion

Prob

abili

ty o

f suc

cess

ful i

mpl

emen

tatio

n

Figure 5

Probability of Successful Implementation, Strength of Special Interests, and Political Cohesion


0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1


Prob

abili

ty o

f suc

cess

ful i

mpl

emen

tatio

n Low-quality

bureaucracy

Average-quality

bureaucracy

Figure 6

Probability of Successful Implementation, Strength of Special Interests, and Quality of Bureaucracy


183


lent), this does not lead to better program imple-mentation. Finally, investing more IMF effort intoprogram preparation, as measured by the dollar costof staff hours and the number of missions or missiondays to a country three and six months before pro-gram approval, does not affect program implementa-tion either.

While useful, our approach is not without limita-tions. To begin with, the linear-in-probability speci-fication may not be an appropriate statistical modelfor the irreversible-interruptions indicator. More-over, the assumption of constant variance needed toapply the random-effects model is hard to justify inthe linear-in-probability model. We believe thatthese drawbacks are outweighed by the substantialinformational advantages from pooling the imple-mentation indicators in one econometric model(see, for example, Lubotsky and Wittenberg, 2001).As additional political economy data become avail-able, it should be possible to extend our dataset andprovide a more thorough check of the robustness ofour results.

Concluding Remarks

This paper makes a start on providing an economet-rically informed assessment of the factors influenc-ing the implementation of IMF-supported programs.This approach fills a gap in a literature that, until re-cently, has evaluated the macroeconomic and struc-tural impacts of these programs without making ade-quate distinctions between implemented and non-implemented programs. The paper presents a varietyof (new and old) statistical indicators of program im-plementation and the groups of factors that could af-fect it, including (a) quantitative measures of the po-litical environment in borrowing countries, (b) theconditionality and financial and human resources in-vested by the IMF in programs, and (c) initial eco-nomic conditions and subsequent shocks in borrow-ing countries. The main findings are as follows:

• Failures in program implementation are associ-ated with a small number of observable politicalindicators in borrowing countries, including thestrength of special interests in parliament, lackof political cohesion in the government, ethnicfragmentation in the broader society, and thecombination of political instability and an inef-ficient bureaucracy.

• Indicators of the IMF’s investment of financialand human effort in programs and the depthand breadth of conditionality are not good pre-dictors of program implementation. This is anuncomfortable conclusion, although it could bepartly due to imprecise measurement of IMF in-puts into programs.

• There is no association between initial and ex-ternal conditions and the probability of pro-gram implementation, indicating that program

targets may incorporate realistic goals and be re-lated effectively to a member’s initial “position.”Interestingly, and despite previous evidence tothe contrary (see Killick, 1998), a member’s ini-tial indebtedness does not affect the outcome ofIMF-supported programs.

The strong empirical link between political vari-ables in borrowing countries and the outcomes ofIMF-supported programs documented in the papersuggests some changes in the way the IMF ap-proaches the extension of its financial support. First,the IMF could take political information and con-straints in borrowing countries into considerationsystematically. With the adoption of new condition-ality guidelines in 2002, the IMF has streamlined itsconditionality and is more carefully tailoring pro-grams to members’ circumstances. The IMF also hascommitted itself to changing its interactions withborrowing countries to put them in the driver’s seatin designing and implementing reforms. Second, tomake systematically informed political judgments,the IMF could methodically collect the growingnumbers of political indicators made available by re-search in quantitative political science. Such infor-mation could be used much like economic informa-tion, as one input in forward-looking quantitativeassessments of program prospects and risks in indi-vidual countries. Third, the close connection be-tween the strength of special interests and weak pro-gram implementation documented in the paperunderscores the need for programs to take measuresto inform and defuse resistance to reforms. These ac-tions are described in detail elsewhere (seeBoughton and Mourmouras, 2004). Related to this,the paper’s results strongly suggest that programsneed to take into account more systematically thanin the past the way legislatures and other key do-mestic players affect the implementation of reforms.While this will undoubtedly make programs morecomplex to design and negotiate, the additionalpayoff in terms of improved implementation may bewell worth the extra effort.

The paper’s results are also relevant in addressingthe issue of selectivity in IMF financing. How highshould the IMF set the bar in approving (or contin-uing) programs if objective political indicators andother evidence (including prior IMF experiencewith failed programs) indicate that these programswould have a low probability of implementation, de-spite the IMF’s anticipated best efforts? In somecases, the IMF may have no choice but to stay in-volved, if only because broader considerations areinvolved. This could be the case, for instance, insome low-income countries in which donor aid, in-cluding support under the debt initiative for HeavilyIndebted Poor Countries (HIPCs), is predicated onthe presence of an active IMF-supported program.In other cases, however, if the probability of imple-mentation is judged to be below some acceptablethreshold, the IMF and its membership might fare

184


better if the IMF exercised greater selectivity in pro-viding financing.

The combination of more selectivity, streamlinedconditionality, and enhanced ownership would en-able the IMF to counter criticisms that it grants toomany waivers or is otherwise lax in its enforcementof conditionality. This combination would also im-prove the quality of IMF-supported programs as sig-nals and catalysts of private investment. The IMFalso could become a better catalyst for change in bor-rowing countries that do not meet the threshold re-quired to receive its assistance. Even though the IMFwould not be providing loans to these countries, itwould continue being active through surveillance,economic education, and technical assistance, andencouraging open debate about policy options andtrade-offs. Especially useful in this regard would bedialogue with reform-oriented groups in borrowingcountries, both explaining the IMF’s points of viewand hearing their perspectives (see Birdsall, 2000).

Future work in this area will involve both a moresystematic collection of information on IMF-supported programs and more careful econometricmodeling of these programs’ impacts. The top prior-ity is establishing on a firmer basis the relation be-tween program implementation and macroeco-nomic impact. Even though this paper presentedsome evidence that improved program implementa-tion was associated with strengthened economicperformance, econometric research on the connec-tion between program implementation and macro-economic success is at an early stage. A more defini-tive econometric study is needed to measure theimpact of improved program implementation on fis-cal and balance of payments outcomes, and inflationand growth. The connection between IMF efforts inborrowing countries and program outcomes needs tobe reassessed as well. The indicators of IMF effortneed to be refined by, among other things, examin-ing in greater detail how missions and staff inputsare related to specific programs and their outcomes.One would hope that the IMF’s Independent Evalu-ation Office would follow the example of the WorldBank’s Operation Evaluations Department in col-lecting and analyzing information on lender effortsat the program design, negotiation, and implemen-tation stages. Such disaggregated information onIMF efforts would permit researchers analyzing IMF-supported programs to ascertain the effectivenesswith which the IMF allocates its resources in ad-dressing the needs of borrowing countries.

Appendix. Detailed Definitions and Data Sources

Program Implementation

An interruption occurs if an SBA program review wasdelayed by more than three months or not com-

pleted at all; if a program review for ESAF/PRGFprograms was delayed by more than six months ornot completed at all; if there was an interval of morethan six months between two subsequent years of amultiyear arrangement; or if at least one of the an-nual arrangements was not approved. Exceptions areprograms that were canceled and replaced by an-other program, in which case noncompleted reviewsand nonapproved annual arrangements are notcounted as interruptions.

An irreversible interruption occurs if either the lastscheduled program review was not completed (allprograms) or all scheduled reviews were completedbut the subsequent annual arrangement was not ap-proved (ESAF/PRGF arrangements).

The Macroeconomic Implementation Index for agiven macroeconomic performance criterion isequal to 100 percent if the macroeconomic perfor-mance criterion was met or met after modification;it is equal to 0 if the macroeconomic performancecriterion was not met, not met after modification,waived, or waived after modification. The Macro-economic Implementation Index for a program thenis computed as the average of Macroeconomic Im-plementation Indices across all macroeconomic per-formance criteria for this program.

The Structural Implementation Index for a givenstructural condition is equal to 100 percent if thestructural condition was met or met with small delayfor structural benchmarks; it is equal to 50 percent ifthe structural condition was partially met or delayedfor performance criteria; and it is equal to 0 if thestructural condition was not met. The Structural Im-plementation Index for a program then is computedas the average of Structural Implementation Indicesacross all structural conditions for this program.

The Average Overall Implementation Index for agiven program is the average of Macroeconomic andStructural Implementation indices over all condi-tions in this program.

Political Indicators

Ethnic fractionalization measures the probability thattwo randomly selected people in a country belong todifferent ethno-linguistic groups. (In regressions,this variable was scaled to range between zero and100.) (See Easterly and Levine, 1997.)

The political instability index is computed based onthe index of internal conflict provided by the ICRGon a scale from zero to 12. Higher values of theindex correspond to more internal political instabil-ity. We replaced the value of this variable by its max-imum score (12) if there was a change of chief execu-tive in the course of an IMF-supported program.

The executive index of electoral competitiveness is adummy variable that is equal to one if the executiveindex of electoral competitiveness is equal to sevenand 0 otherwise. The executive index of electoralcompetitiveness is from the Database of Political Institutions of the World Bank. It ranges from one

185


to seven, with higher values corresponding to morecompetitive elections.

Time in power is the number of years a chief exec-utive has been in power by the approval year of theprogram. We assigned zero to this variable if therewas a change of chief executive in the course of theprogram (Database of Political Institutions and CIAWorld Factbook for the most recent years).

The strength of special interests is computed as themaximum share of seats in the parliament held byparties representing special interests (Database ofPolitical Institutions, World Bank). Four special in-terest groups are identified: religious, nationalistic,regional, and rural.

The index of political cohesion is defined as follows:“For presidential systems, it is zero if different partiesare in control of the executive and legislature (ifmultiple pro-presidential parties, they must not con-trol the legislature); it is one if the same party is incontrol of the executive and legislature (if there aremultiple pro-presidential parties, they must togethercontrol the legislature).”

For parliamentary systems, the index is zero for aminority government, one for a coalition govern-ment with three or more parties, two for a coalitiongovernment with two parties, and three for a one-party majority government (Database of PoliticalInstitutions, World Bank).

Bureaucracy quality (ICRG) measures the qualityof a country’s bureaucracy on a four-point scale.There was a change in scale for this variable, from asix-point to a four-point scale, in August 1997. Werescaled the older series to be measured on a four-point basis. We interact this variable with thedummy indicating that there was a change of chiefexecutive (Database of Political Institutions andCIA World Factbook for the most recent years).

IMF-Related Variables

IMF effort is the dollar cost of IMF programs com-puted based on its Budget Reporting System (BRS)data on hours spent by staff on program implemen-tation (it includes program preparation and supervi-sion) and estimated average salaries of the staff bygrade. Alternative measures of IMF effort were dollarcosts of resident representatives (provided by the Of-fice of Budget and Planning (OBP)), number of mis-sions, and number of mission days (both were pro-vided by the Policy Development and ReviewDepartment (PDR)).

Number of conditions per program year is the totalnumber of conditions (structural and quantitative)divided by the actual duration of the program (theIMF’s Monitoring of IMF Arrangements (MONA)database).

Share of quantitative PCs waived is the number ofquantitative performance criteria waived over thecourse of the program divided by the total numberof quantitative performance criteria for this pro-gram, in percentage points (MONA).

Share of structural conditions is the number ofstructural conditions divided by the total number ofconditions, in percentage points (MONA).

Loan size as percentage of quota is the total com-mitted amount including augmentations divided bythe country’s quota at the IMF (International Fi-nancial Statistics (IFS)).

Debt to the IMF as percentage of IMF quota is ac-tual holdings as percentage of quota from IFS.

Program approval year from MONA.Expected program duration is the number of years

the program was scheduled to last (MONA).IMF quota is from the IMF’s International Finan-

cial Statistics database.

Economic Conditions and Policies

Terms of trade shock is the average growth rate of dol-lar export prices multiplied by the initial share of ex-ports in GDP minus average growth rate of dollar im-port prices multiplied by the initial share of importsin GDP over the course of the program, from IFS.

The following variables are from IFS: central gov-ernment balance, GDP, reserves minus gold, CPI infla-tion, and imports.

The following variables are from the IMF’s WorldEconomic Outlook (WEO) database: current accountbalance; initial population.

Finally, GDP per capita is from the World Bank’sWorld Development Indicators (WDI).

Other

The average share of bilateral aid given by the Group ofSeven (G-7) to the country before the program start wascomputed as the average of the shares of gross offi-cial transfers that each of the G-7 countries allocatedto a particular country one year prior to the approvalyear of the IMF program for that country (Organiza-tion for Economic Cooperation and Development(OECD)).

ReferencesÅslund, Anders, 1999, “Why Has Russia’s Economic

Transformation Been So Arduous?” paper preparedfor the Annual World Bank Conference on Develop-ment Economics (Washington: Carnegie Endow-ment for International Peace), April 28–30.

Beck, Thorsten, George Clarke, Alberto Groff, PhilipKeefer, and Patrick Walsh, 2001, “New Tools inComparative Political Economy: The Database ofPolitical Institutions,” World Bank Economic Review,Vol. 15 (September), pp. 165–76.

Bird, Graham, 2002, “Conditionality and Ownership:How to Improve the Implementation of IMF Pro-grammes,” Working Paper (Guildford, Surrey, UnitedKingdom: Surrey Center for International EconomicStudies, University of Surrey).

Birdsall, Nancy, 2000, “The World Bank of the Future:Victim, Villain, Global Credit Union?” remarks deliv-ered at the conference “Promoting Dialogue: Global

186


Challenges and Global Institutions” (Washington:American University), April 13, available on the Webat http://www.odc.org/commentary/bird-au.html.

Boughton, James M., and Alex Mourmouras, 2002,“Whose Programme Is It? Policy Ownership WithConditional Lending,” Chapter 9 in The IMF and itsCritics, ed. by David Vines and Christopher L.Gilbert (Cambridge, United Kingdom: CambridgeUniversity Press), pp. 225–53.

Bredenkamp, Hugh, and Susan Schadler, eds., 1999, Eco-nomic Adjustment and Reform in Low-Income Countries(Washington: International Monetary Fund).

Bulír, Ales, and Soojin Moon, 2003, “Do IMF-SupportedPrograms Help Make Fiscal Adjustment MoreDurable?” IMF Working Paper 03/38 (Washington:International Monetary Fund). This paper also ap-pears, in somewhat different form, as Chapter 14 inthis volume.

Conway, Patrick, 1994, “IMF Lending Programs: Partici-pation and Impact,” Journal of Development Econom-ics, Vol. 45, pp. 365–91.

———, 1998a, “Evaluating Fund Programs: Methodologyand Empirical Estimates” (unpublished workingpaper: University of North Carolina at Chapel Hill).

———, 1998b, “IMF Programs and External Balance:The Crisis-Participation Dynamic” (unpublishedworking paper: University of North Carolina atChapel Hill).

Dollar, David, and Jakob Svensson, 2000, “What Explainsthe Success or Failure of Bank-Supported Adjust-ment Programs,” Economic Journal, Vol. 110 (Octo-ber), pp. 894–917.

Drazen, Allan, 2002, “Conditionality and Ownership inIMF Lending: A Political Economy Approach,” StaffPapers, International Monetary Fund, Vol. 49, Spe-cial Issue, pp. 36–67.

Easterly, William, and Ross Levine, 1997, “Africa’sGrowth Tragedy: Policies and Ethnic Divisions,”Quarterly Journal of Economics, Vol. 112, pp. 203–50.

Haque, Nadeem Ul, and Mohsin S. Khan, 1998, “Do IMF-Supported Programs Work? A Survey of the Cross-Country Empirical Evidence,” IMF Working Paper98/169 (Washington: International Monetary Fund).

Havrylyshyn, Oleh, and John Odling-Smee, 2000, “Politi-cal Economy of Stalled Reforms,” Finance & Develop-ment, Vol. 37 (September), pp. 7–10.

Hellman, Joel, and Daniel Kaufmann, 2001, “Confrontingthe Challenge of State Capture in TransitionEconomies,” Finance & Development, Vol. 38 (Sep-tember), pp. 31–35.

International Monetary Fund (IMF), 1998, External Eval-uation of the ESAF: Report by a Group of IndependentExperts (Washington: International Monetary Fund).

———, 2001a, “Strengthening Country Ownership ofFund-Supported Programs” (Washington: Interna-tional Monetary Fund), available on the Web athttp://www.imf.org/external/np/pdr/cond/2001/eng/strength/120501.pdf.

———, 2001b, “Conditionality in IMF-Supported Pro-grams—Policy Issues” (Washington: InternationalMonetary Fund), March, available on the Web athttp://www.imf.org/external/np/pdr/cond/2001/eng/policy/021601.pdf.

———, 2001c, “Structural Conditionality in IMF-Supported Programs” (Washington: InternationalMonetary Fund), March, available on the Web athttp://www.imf.org/external/np/pdr/cond/2001/eng/struct/cond.pdf.

———, 2002, “Guidelines on Conditionality” (Washing-ton: International Monetary Fund), September,available on the Web at http://www.imf.org/External/np/pdr/cond/2002/eng/guid/092302.htm.

Joreskog, Karl G., and Arthur Goldberger, 1975, “Estima-tion of a Model with Multiple Indicators and Multi-ple Causes of a Single Latent Variable,” Journal of theAmerican Statistical Association, Vol. 70 (September),pp. 631–39.

Joyce, Joseph P., 2002, “Through a Glass Darkly: WhatWe Know (and Don’t Know) About IMF Programs,”paper presented at the IMF Conference on Researchon IMF-Supported Programs, Washington, June.

Killick, Tony, with Ramani Gunatilaka and Ana Marr,1998, Aid and the Political Economy of Policy Change(London: Routledge).

Lubotsky, Darren, and Martin Wittenberg, 2001, “Inter-pretation of Regressions with Multiple Proxies,” (un-published working paper; Princeton, New Jersey:Princeton University).

Mayer, Wolfgang, and Alex Mourmouras, 2002, “VestedInterests in a Positive Theory of IFI Conditionality,”IMF Working Paper 02/73 (Washington: Interna-tional Monetary Fund).

Mecagni, Mauro, 1999, “The Causes of Program Interrup-tions,” in Economic Adjustment and Reform in Low-Income Countries, ed. by Hugh Bredenkamp andSusan Schadler (Washington: International Mone-tary Fund).


Odling-Smee, John, 2001, “Belarus: Recent Experienceand Challenges Ahead,” speech given at the Belaru-sian Academy of Management, Minsk, November 6,available on the Web at http://www.imf.org/external/np/speeches/2001/110601.htm.

Schadler, Susan, Adam Bennett, Maria Carkovic, LouisDicks-Mireaux, Mauro Mecagni, James H. J.Morsink, and Miguel A. Savastano, 1995a, IMF Con-ditionality: Experience Under Stand-By and ExtendedArrangements, Part I: Key Issues and Findings, IMFOccasional Paper No. 128 (Washington: Interna-tional Monetary Fund).

———, 1995b, IMF Conditionality: Experience UnderStand-By and Extended Arrangements, Part II: Back-ground Papers, IMF Occasional Paper No. 129 (Wash-ington: International Monetary Fund).

Thomas, Alun, 2003, “Prior Actions—True Repentance?An Evaluation Based on IMF Programs over the1992–1999 Period” (unpublished paper; Washington:Policy Development and Review Department, Inter-national Monetary Fund).

Tsebelis, George, 2001, Veto Players: How Political Institu-tions Work (Princeton, New Jersey: Princeton Uni-versity Press and Russell Sage Foundation).

Part III. Effectiveness

189

11

The large international bailouts of the 1990s have beencriticized for generating moral hazard at the expense of theglobal taxpayer. We argue that this criticism is misleadingbecause international bailouts create either no or very fewcosts to the international community. Instead, the prob-lem is that bailouts may be used to facilitate bad domesticpolicies, thus creating moral hazard at the expense of do-mestic taxpayers. Ensuring that this does not happen mayrequire a shift toward ex ante conditionality, in the sensethat the availability and size of official crisis lending needto be conditional on government policies before the crisis.

Introduction

The 1990s have seen international official crisislending of unprecedented size (Table 1). Advocatesjustified these bailouts as a form of international fi-nancial safety net to deal with new risks arising fromthe international financial integration of emergingeconomies, most notably the rollover risk on foreigncurrency debt (Fischer, 1999; and Summers, 1999).At the same time, critics argued that internationalbailouts were instead a factor contributing to finan-cial crises, because of the moral hazard that they cre-ate. There has been a lively debate on whether thepractice of large international bailouts should becontinued, discontinued, or reformed.2

That financial safety nets create moral hazard,and possibly excessive moral hazard, seems fairly un-controversial.3 However, much of the policy debateon how to reduce moral hazard seems to hinge on anerroneous view of how moral hazard is created by in-ternational bailouts, namely, the analogy betweeninternational bailouts and a standard insurancearrangement (for example, fire insurance or medicalinsurance). We dismiss this view on empiricalgrounds and offer an alternative interpretation ofhow moral hazard might be created by internationalcrisis lending. We then draw the policy implications.4

In a standard insurance contract, the conse-quences of a bad outcome are mitigated by a state-contingent transfer. For example, if an insured per-son gets sick, medical insurance will cover part ofher treatment costs. Moral hazard arises because theexistence of this transfer makes her less anxious toavoid the bad state, at the expense of those who ul-timately finance the transfer. One widespread viewof international bailouts is based on the transposi-tion of this argument to the international level: acountry gets “sick” and receives a transfer, in theform of a subsidized loan from the internationalcommunity. Moral hazard arises at the expense ofthose who supposedly pay for that transfer, namely,the global taxpayers. As noted in an April 23, 1998Wall Street Journal editorial, “What really happens isthat the U.S. ends up subsidizing the IMF’s growingpractice of making large loans at low interest ratesto very risky economies ... and U.S. governmentmoney comes from taxpayers.”

This view is inconsistent with the available evi-dence on the repayment record on international of-ficial lending. The IMF’s large-scale crisis loans ofthe 1990s were made at a low interest rate, but inmost cases they have been repaid or are in theprocess of being repaid, as Figure 1 shows. This is

International Bailouts, Moral Hazard, and Conditionality

Olivier Jeanne and Jeromin Zettelmeyer1

1 This paper was first published in Economic Policy in 2001 andhas not subsequently been revised or updated. It appears here withthe permission of Blackwell Publishing, the journal’s publisher.An earlier version was presented at the 33rd Economic PolicyPanel Meeting, which was held in Stockholm April 6–7, 2001.We are grateful to our two discussants at the Panel, Philippe Bac-chetta and Andrew Scott, as well as the editor, Charles Wyplosz,for very useful comments and suggestions. We also thank MarkAllen, Reza Baqir, Eduardo Borensztein, Barry Eichengreen, Gas-ton Gelos, Ken Kletzer, Tim Lane, Paul Masson, Carlos Medeiros,Christian Mulder, Mike Mussa, Sanjaya Panth, Steve Phillips,Andrea Richter, Ken Rogoff, Nouriel Roubini, Alexander Swo-boda, Tessa van der Willigen, Alexander Wolfson, and especiallyDaniel Cohen for discussions and/or comments on earlier drafts.Our thanks also go to several colleagues at the IMF—includingJim Boughton, Christina Daseking, Leyla Ecevit, Doris Ross, andBarry Yuen—for data and information on IMF financial opera-tions. Any remaining errors, however, are our sole responsibility.

2 See Shultz, Simon, and Wriston (1998); Schwartz (1998);Edwards (1998); Feldstein (1998); Eichengreen (1999, 2000);Council on Foreign Relations Task Force (1999); de Gregorioand others (2000); IFIAC (2000); Goldstein (2000b); andCalomiris (2001), among many others.

3 The distinction between moral hazard and excessive moralhazard is emphasized by Mussa (1999). The relevant questionfrom a policy perspective is not whether international bailoutsgenerate moral hazard per se, but whether this moral hazard is ex-cessive in view of the benefits of the international financialsafety net.

4 Note that we are not concerned with quantifying the extentto which international bailouts create moral hazard. This is thesubject of a recent set of papers by Nunnenkamp (1999); Zhang(1999); Lane and Phillips (2000); Dell’Ariccia, Gödde, andZettelmeyer (2000); and McBrady and Seasholes (2000).

CHAPTER

190

11 � INTERNATIONAL BAILOUTS, MORAL HAZARD, AND CONDITIONALITY

0

5

10

15

20

25

1994

Mexico

Brazil

Thailand

Korea,Rep. of

Indonesia

Russia

1995 1996 1997 1998 1999 2000 2001

Figure 1

Use of IMF Credit by Mexico, Thailand, Indonesia, the Republic of Korea, Brazil, and the Russian Federation, 1994–2001(in billions of U.S. dollars)

typical of the IMF repayment record with countriesthat have access to international financial markets.Thus, the international safety net does not involveany state-contingent transfers from the global tax-payer to crisis countries. Instead, it involves state-contingent loans, which do not embody any signifi-cant subsidy.

The fact that the simple insurance analogy iswrong and moral hazard does not arise at the ex-pense of the international taxpayer does not imply

that international bailouts cannot create excessivemoral hazard. However, it does imply that one needsto think harder about the mechanisms that might becausing it. If official crisis lending comes at no costto the rest of the world, and is voluntary on the sideof the crisis countries—which are free to reject theoffer of assistance, and sometimes do—where is theproblem? The problem, in our view, lies in the in-teraction between international bailouts and do-mestic policies, as pointed out by Calomiris (1998).

Table 1

IMF-Supported Crisis Packages of the 1990s: Financing Commitments and Disbursements1

(in percentage of initial GDP)

Financing Commitments Disbursements

Date of Approval Total IMF Total IMF

Mexico Feb. 1995 18.3 6.3 9.1 4.6Thailand Aug. 1997 11.5 2.7 9.6 2.3Indonesia Nov. 1997 19.6 5.2 9.3 4.8Korea, Rep. of Dec. 1997 12.3 4.4 6.5 4.1Brazil Dec. 1998 5.4 2.3 3.4 1.4Memorandum itemRussia n.a.2 … … … 6.63

1 GDP in year of program approval (for Russia, we used 1995).2 Russia had several consecutive IMF programs during the 1990s. The first large-scale loan was a stand-by arrangement approved in April of

1995.3 Total disbursements in the 1990s.

191

Olivier Jeanne and Jeromin Zettelmeyer

The international financial safety net increases thescope for bad policies as well as good policies. Inter-national bailouts can help countries implementgood policies, but they could also make the conse-quences of bad domestic policies much worse.

For example, consider the design of domestic fi-nancial safety nets in a financially integrated emerg-ing market economy. A good policy, in this context,could be defined as a well-designed system of depositinsurance that would provide a limited guarantee onboth domestic currency and foreign currency de-posits in the domestic banking system. An exampleof bad policy could be an implicit government guar-antee on the debt of the policymaker’s friends andrelatives. The international safety net could be usedto back one or the other. The fact that its operationcosts nothing to the global taxpayer does not implythat it is going to be used to back good policies. Itsimply implies that the consequences of domesticpolicies, good or bad, are going to be borne by thedomestic taxpayer. Excessive moral hazard mightarise not because international bailouts involve aninternational transfer, but because they facilitate adomestic transfer from the domestic taxpayer to theborrowers that benefit from implicit guarantees(Jeanne and Zettelmeyer, 2001a).

A similar story can be told about sovereign debtcrises. The expectation of full repayment even in aneconomic crisis could make international investorsmore lenient in granting new loans or rolling overexisting loans to a government (“investor moralhazard”). Official crisis lending can provide govern-ments with liquidity to implement such a bailout.However, if it does not carry a subsidy, and if thegovernment maximizes domestic welfare—with dueweight given to the domestic taxpayers who will payfor the bailout in the event of a crisis—this cannotgenerate excessive moral hazard. For moral hazard tobe excessive, there must be a discrepancy betweenthe policymaker’s objective and the domestic tax-payers’ long-term interests (Jeanne and Zettelmeyer,2001b).

If the problem is that the international commu-nity creates moral hazard by playing the role of ac-complice in bad domestic policies, what is the ap-propriate policy response? Increasing the interestrate on international loans, or collateralizing IMFlending, will clearly not do—such proposals miss thepoint. Instead, our preferred answer is to make theinternational community an accomplice in the im-plementation of good domestic policies. We argue inthis paper that this requires shifting the weight ofconditionality from traditional conditionality expost, as an accompanying element of official lendingafter the crisis, to conditionality ex ante, in thesense that the availability and size of official lendingneed to be conditional on government policies be-fore the crisis. The international community’s lend-ing policies must create a link between domestic ef-forts at preventing crises ex ante and the extent ofinsurance ex post. In our conclusion, we discuss

some problems involved in making further progressin this direction, as well as possible solutions.

The IMF Lending Record

In this section, we address a question involving a(hypothetical) emerging economy that must repayon its sovereign or private debt 10 percent of itsGDP to foreign creditors—the rough order of mag-nitude of the large bailouts of the 1990s (seeTable 1, or the average fiscal cost of banking crisesin Frydl, 1999). The country could default, or, alter-natively, it could repay its foreign creditors using aloan from the international community amountingto 10 percent of GDP. Based on the historicalrecord, how much of this 10 percent could the coun-try hope to transfer to the global taxpayer by not re-paying the loan? The answer, we are going to argue,is “not very much”—at most a fraction of a percent-age point of GDP.

Our answer is based on the historical lendingrecord of the IMF. The justification for concentrat-ing on the IMF is its role as the preeminent officialcrisis lender. While other multilaterals (in particular,the World Bank, the Asian Development Bank, andthe Inter-American Development Bank) also haveplayed a role, their financial involvement in a crisiscontext has been much more modest than that ofthe IMF (whose share in multilateral crisis lendinghas ranged from about 63 percent for Thailand to100 percent for Mexico). Large-scale bilateral crisislending has taken place in only three cases: Mexico(1995), Thailand (1997), and Brazil (1998). In twoof these—Mexico and Brazil—bilateral lenders haveby now been repaid in full. In Thailand’s case, the bi-laterals are being repaid in tandem with the IMF. Thetwo most recent crisis packages, for Turkey and Ar-gentina, were largely financed by the IMF on its own.

Ignoring bilateral lending might be misleading inone important sense. Conceivably, the repayment ofIMF loans could be financed by bilateral loans—that is, the IMF could be bailed out by its ownshareholders. Bulow, Rogoff, and Bevilaqua (1992)present evidence suggesting that the repayment ofnonconcessional World Bank loans to a group oflow- and middle-income countries during the 1980swas financed either by bilaterals or by the WorldBank’s own concessional lending. Thus, for thesecountries, the repayment record of international fi-nancial institutions (IFIs) could give an overly opti-mistic picture of the repayment record of the officialsector as a whole. However, we think that this is un-likely to be a problem in the class of countries thatare candidates for large international bailouts—rela-tively advanced emerging economies with access tointernational capital markets. In Appendix I, wepresent evidence indicating that repayments to theIMF are not financed by new bilateral debt to thisgroup of countries, although the same analysis sup-ports Bulow and others’ (1992) results for the group

192


of heavily indebted poor countries (HIPCs). Giventhat the focus of this paper is on IMF lending to thefirst group, we ignore bilateral financing in the re-mainder of the paper.

How do we measure the potential transfer en-tailed in IMF lending? We say that IMF loans in-volve a transfer if their interest rate is not highenough to offset the default risk faced by the IMF. Inother words, we define the absence of transfer as asituation where the IMF makes zero profit on aver-age. Conceivably, the IMF could charge a higher in-terest rate and make a strictly positive profit. Somewould argue that the failure to do so—to exploit theIMF’s monopoly rent in crisis lending—constitutesan implicit transfer, or a subsidy. This is not, how-ever, what the IMF critics mean when they claimthat its loans are subsidized.

The IMF’s default risk is not necessarily the sameas that faced by private lenders. The ability to im-pose conditionality ex post, which is typically justi-fied by the need to ensure repayment, might implythat the IMF has a more direct influence on mea-sures that improve country solvency than do privateinstitutions. In addition, the IMF may be in a betterposition to deter default. It can deny further IMFlending and thus effectively bar access to the multi-lateral financial safety net in the future (this occursautomatically if a country accumulates arrears to theIMF). It can also exert pressure on a countrythrough the actions of its major shareholders, whichmay link bilateral lending, or other bilateral poli-cies, to a country’s good standing with the IMF.

It follows that the question of whether IMF lend-ing implies a transfer cannot be answered by simplycomparing IMF interest rates to market interestrates prevailing during crises, as has been argued in,among other places, the Wall Street Journal’s editor-ial dated April 23, 1998 (quoted in the previous sec-tion) and Calomiris (2001). Instead, the relevantcomparison is between the interest rate charged bythe IMF and the default risk it faces.

What Interest Rate Does the IMF Charge?

The relevant interest rate for the bulk of IMF lend-ing is the “basic rate of charge on ordinary re-sources.”5 The rate of charge is set as low as possiblewhile covering the IMF’s financing costs, its admin-istrative budget, and the accumulation of some “pre-cautionary balances” (IMF, 1998). The IMF’s fi-nancing costs, in turn, depend on the portion ofmember currency holdings that are remunerated,and the rate of remuneration that is applied. Prior to

1969, the IMF did not compensate its membercountries for their currency holdings. The remuner-ation of member holdings was introduced after 1969and was increased in various steps during the 1970sand 1980s (Boughton, 2001, Chapter 17, providesthe details). Since 1987, the rate of remunerationhas been equal to the SDR interest rate (a weightedaverage of three-month money market interest ratesof the five major currencies). Since all but a smallportion of member currency holdings are remuner-ated, this implies that since 1987, the basic rate ofcharge is effectively set as a markup over the SDRinterest rate.

Figure 2 shows that prior to the mid-1980s, theIMF’s basic rate of charge was lower than both theU.S. three-year government bond rate—which is auseful benchmark in view of the average maturity ofIMF nonconcessional loans (of two to five years)—and the market-determined SDR interest rate.Thus, in this period, IMF interest rates generally im-plied a subsidy, in the sense that a borrowing coun-try could have made a small, risk-free profit on IMFloans by maintaining reserves borrowed from theIMF in liquid money market instruments denomi-nated in SDR currencies. Since the mid-1980s, theIMF’s rate of charge has fluctuated around the U.S.three-year government bond rate. Moreover, sincethe introduction of the Supplemental Reserve Facil-ity (SRF) in December 1997, the IMF has begun toimpose surcharges for large-scale lending. Since No-vember 2000, surcharges apply to most nonconces-sional IMF lending that exceeds 200 percent ofquota—see IMF (2000b) for details.

The IMF Repayment Record: Is the Past aGood Guide for the Future?

Does the interest rate charged by the IMF since1987 reflect the default risk to which it is exposed?If so, the default risk facing the IMF would need tobe very low, since the SDR and U.S. interest ratescomparable with the IMF basic rate of charge arevirtually risk free. Given that the IMF typicallylends to crisis countries that appear very shaky—and are often cut off from private financing or faceprohibitively high market interest rates precisely forthat reason—this seems a tall order. However, it isexactly what the IMF’s repayment record suggests, atleast on first appearance. The recent HIPC Initia-tive is the first case of IMF claims reduction, and ar-rears cases to the IMF are few and far between. Mostof them are settled eventually, and they generallyinvolve countries that go through wars or violentinternal conflicts.6

5 This is the interest rate that applies to the standard (noncon-cessional) IMF lending facilities. The interest rate charged by theIMF on its concessional facilities—the Enhanced Structural Ad-justment Facility (ESAF) and its successor, the Poverty Reduc-tion and Growth Facility (PRGF)—is much lower (currently0.5 percent). Concessional lending targets very poor countriesand has not been used to deal with the international financialcrises that we focus on in this paper.

6 Total arrears to the IMF in 2000 amounted to US$3 billion,about 4.5 percent of its total loans and credits outstanding. Thefollowing countries were in arrears: Afghanistan, the DemocraticRepublic of the Congo, Iraq, Liberia, Somalia, Sudan, and theFederal Republic of Yugoslavia. For a comprehensive documenta-tion of IMF arrears cases, see Aylward and Thorne, 1998.

193


However, the relatively small number and volumeof arrears cases are evidence of low default risk onlyif we assume that the IMF’s repayment record overthe last 50 years will continue to apply in the future.Some skeptics have expressed doubts about this.They argue that the IMF’s past repayment recordwas made possible by the IMF’s practice of refinanc-ing its loans to countries that were unable to repay.If this is true, bad loans will increasingly tend tocrowd out good ones in the IMF loan portfolio, andwe eventually should observe either an increase inthe rate of default on IMF lending, or debt reduc-tion, as in the case of the HIPC initiative.

The fraction of IMF programs that end up inlong-term lending relationships is examined inTable 2. We report the number of “complete” and“incomplete” lending cycles over IMF history. Alending cycle is defined as an uninterrupted periodof strictly positive “Total Fund Credits and LoansOutstanding,” as reported in International FinancialStatistics. A lending cycle is “complete” if it endsprior to 2000. We refer to countries with outstand-ing debt in 2000 as being in an “incomplete” lend-ing cycle.

Overall, and in most regions, the number of com-plete lending cycles exceeds that of incompleteones. There are two main exceptions: Africa andnon-OECD (Organization for Economic Coopera-tion and Development) Europe. The situation inthe latter is due to a number of relatively recent

lending cycles with transition economies that bor-rowed from the IMF for the first time in the 1990s.In contrast, the large proportion of incomplete lend-ing cycles in Africa reflects the prevalence of suc-cessive borrowing arrangements from the IMF overmany years.

Unsurprisingly, the occurrence of long incom-plete cycles is especially high in HIPCs. Almost allof these countries (38 out of 42) have ongoing lend-ing relationships with the IMF that started morethan twenty years ago, on average, and incompletelending cycles far outnumber completed cycles. TheIMF experience with HIPCs illustrates the possibil-ity that very long lending cycles might turn into“bad loans” that the international community willeventually have to forgive, at least in part. However,this experience concerns very poor countries thathave little in common with the group of countriesthat are the main focus of this paper—the relativelyadvanced emerging economies that have benefitedfrom the large international bailouts of the 1990s.The question is how widespread the long-term con-tinuous lending relationships with the IMF are inthe latter group.

To answer this question, we look at the length oflending cycles in the emerging market–countrygroup tracked by J.P. Morgan’s “EMBI Global”Index (EMBIG). This is a group of 27 relatively ad-vanced emerging market countries—the most inter-nationally financially integrated countries in the

0

2

4

6

8

10

12

14

16

1950

Basic rate (includes "burden sharing" after 1987)

Market-determined SDR interest rate (since 1981)

U.S. three-year government bond rate

1955 1960 1965 1970 1975 1980 1985 1990 1995 2000

Figure 2

IMF Basic Rate of Charge and Market Interest Rates(in percentage points)

194


nonindustrial country group. Appendix II has a fulllist of these countries. All major capital accountcrises of the 1990s involved members of the EMBIGgroup. Table 2 shows that the number of incompletelending cycles in the EMBIG group (15) is low rela-tive to the number of completed cycles, and that theincomplete cycles’ average length is substantiallyshorter than that of the non-EMBIG developingcountries (13.8 versus 18.8). However, as the lastline of the table shows, even in the EMBIG group,about half of the incomplete lending cycles in 2000have been ongoing for, on average, more than 20years. This is driven mainly by four long-term lend-ing relationships: Côte d’Ivoire (since 1974),Panama (since 1974), Peru (since 1976), and thePhilippines (since 1968). Thus, among the more ad-vanced developing countries, which are the focus ofthis paper, long-term lending relationships with theIMF are not the rule, but neither are they nonexis-tent. On this basis, we proceed under the assump-tion that the IMF’s repayment record so far need notnecessarily reflect the steady state even in this groupof countries.

Estimating an Upper Bound for the TransferElement in IMF Lending

If one accepts that, in principle, the IMF’s excellentrepayment record need not continue indefinitely,estimating the subsidy element in IMF lendingraises a fundamental difficulty. An estimate must be

conditional on an assessment of the future repay-ment prospects of countries that are currently in-debted to the IMF, and such an assessment is neces-sarily highly subjective. Rather than speculatingabout default probabilities for individual countries,we address these issues by computing a crude upperbound to the redistributive element implicit in IMFlending.

Our methodology relies on the notion of a hypo-thetical long-run IMF lending regime. This long-runregime involves two types of IMF lending cycles:those that are eventually repaid, and those thatwould go on forever unless the country either de-faulted or received some multilateral debt relief. Werefer to the first type as “finite,” and to the second as“infinite,” cycles. The latter are motivated by thelong lending cycles observed in Table 2, and theworry that such cycles could reflect a process bywhich the IMF refinances its own obligations, lead-ing to increasing country debt levels that need to beeventually recognized as unsustainable—as in thecase of HIPCs. We estimate the fraction µ of “infi-nite” cycles, using the method presented in Box 1.

Table 3 summarizes our results. It presents pointestimates and p-values for µ for the countries inTable 2, along with δ, the maximum amount thatcountries can borrow from the IMF based on theirquotas (see Box 2). The maximum ex ante transferfor each subsample (assuming a zero recovery ratefor infinite cycles) is then given by the product ofthe two (τ = µδ). Thus, τ is the maximum transfer

Table 2

Completed and Incomplete Debt Cycles, 1947–20001

Average Duration of

Number of Complete Incomplete Cycles (years)

Countries Debt Cycles Debt Cycles Complete Incomplete

All countries 186 165 88 7.1 17.9Industrial countries 25 31 0 4.7 n.a.Developing countries 161 134 88 7.6 17.9

Africa 52 28 38 6.1 22.7Asia 29 23 13 9.0 21.2Europe 28 12 21 10.2 7.9Middle East 14 14 2 6.5 9.5Western Hemisphere 37 57 14 7.6 18.1HIPC countries2 42 25 38 6.1 23.5Non-HIPC developing countries 119 109 50 8.0 13.6EMBIG countries3 27 38 15 7.8 13.8Non-EMBIG countries 134 96 73 7.6 18.8

Memorandum item (excluding cycles initiated after 1991)EMBIG countries3 27 37 8 7.9 20.6

1 Number of countries with outstanding debt in 2000.2 HIPCs (see Appendix II for full list).3 Countries whose bond spreads are tracked by J.P. Morgan's “EMBI Global” Index (see Appendix II).

195


that a country can expect ex ante, i.e., based on ourbest estimate of the fraction of infinite cycles andwithout knowing whether its lending cycle is of theinfinite type or not. For purposes of comparison withδ, we also show the actual outstanding average debtto the IMF in 2000 for each subsample, in terms ofthat year’s GDP.

For the whole sample, the estimated fraction of“infinite” cycles, µ is about 25 percent. However,

this masks enormous disparities across regions andincome groups. In the HIPC sample, the estimatedfraction is almost 60 percent—which is perhaps notsurprising given that we know that countries inthese groups are eligible for multilateral debt relief.For the 119 non-HIPC developing countries, the es-timated proportion of cycles that look like theycould go on forever is only 11 percent. For theEMBIG countries, the fraction falls to just 5 percent,

Box 1. Estimating the Fraction of “Infinite” Cycles

Suppose that a fraction µ of lending cycles is of theinfinite type, and denote the average net present valueof debt owed to the IMF at the time the lending cyclesare recognized as unsustainable as δ. Then, given thatthe IMF lends at approximately the riskless interest rate,the average ex ante transfer is equal to the frequency ofinfinite lending cycles, times the country’s indebtednessto the IMF, times the loss rate on infinite cycles:

τ = µδ(1− ρ)

where ρ is the average IMF “recovery rate” for infinitelending cycles. The strategy is to estimate an upperbound for τ by assuming an upper bound for δ and alower bound for ρ, while attempting to estimate µ fromthe data. The most pessimistic assumption one canmake about the recovery rate is obviously that ρ= 0,i.e., that debt owed on infinite cycles is forgiven in full.

To set δ, we make the radical assumption that all coun-tries on infinite lending cycles will be allowed to bor-row up to the full amount of their cumulative debtceilings prior to defaulting or receiving debt relief,namely, 300 percent of quota. This is about three timesas high as the average actual indebtedness to the IMFin 2000. While the 300 percent cumulative debt ceil-ing can be waived in exceptional circumstances, thishas occurred only in rare cases—which in the 1990shave mostly coincided with the large crisis packageswe briefly discussed in the introduction.

We estimate µ using a simple parametric durationmodel (Greene, 1993). To do so, we need to assume adistribution for the duration of the finite lending cy-cles. We use a standard generalization of the constanthazard rate model, the Weibull model, which allowsfor increasing or decreasing the hazard rate as a func-tion of duration.

Table 3

Maximum Average Ex Ante Subsidy(in percentage of debtor country GDP in 2000)

Actual Maximum Total Number p-Value Debt Debt in Subsidy of Countries µ1 (H0: µ = 0) Limit2 20003 τ = µδ

All countries, average 186 0.25 0.00 11.2 4.4 2.81Industrial countries, average 25 0.00 1.00 4.1 n.a. 0.00Developing countries, average 161 0.28 0.00 12.3 4.4 3.46

Africa 52 0.58 0.00 17.1 5.7 9.88Asia 29 0.28 0.00 8.9 2.2 2.48Europe 28 0.12 0.28 11.5 4.7 1.32Middle East 14 0.00 0.89 7.2 4.6 0.00Western Hemisphere 37 0.09 0.16 10.5 2.5 0.99HIPC countries4 42 0.59 0.00 18.4 5.6 10.84Non-HIPC developing countries 119 0.11 0.00 9.0 1.3 0.96EMBIG countries5 27 0.05 0.66 7.0 1.7 0.37Non-EMBIG countries 134 0.29 0.00 12.5 2.6 3.58

1 Maximum-likelihood estimate of fraction of "infinite" lending cycles for various subsamples.2 300 percent of quota in 2000, as a percentage of 2000 GDP.3 In percentage of 2000 GDP.4 Heavily indebted poor countries (see Appendix II for full list).5 Countries whose bond spreads are tracked by J.P. Morgan's "EMBI Global" Index (see Appendix II).

196


and is insignificantly different from zero. The morerestrictive the definition of the subsample in termsof income levels and access to capital markets, thelower the estimated fraction of “infinite cycles.”

In the last column, we multiply cumulative debtlimits with the estimated fraction of “infinite cy-cles.” As one would expect, this results in a furthermagnification of the cross-regional variations thatwere noted earlier. For the HIPC group, the upperbound for the ex ante transfer is more than 10 per-cent of debtor country GDP. Of course, the actualtransfer that these countries can expect from theIMF as a result of the HIPC initiative is far lower,both because the actual debt levels are much belowthe assumed debt ceilings, and because the HIPCInitiative does not envisage a complete debt write-down, as assumed in Table 3; see IMF (1998). Forthe group of emerging economies that are our mainfocus here, i.e., the EMBIG countries, the estimatedupper bound is only 0.37 percent of GDP. This isdriven by the fact that the estimated fraction of “in-finite” cycles is very small in this group.

We can now answer the question that was askedat the beginning of this section. Based on the histor-ical record, an EMBIG economy that borrows10 percent of its GDP would fail to repay with aprobability of at most 5 percent—the probability ofan “infinite cycle” for this class of countries. Theimplicit transfer, which results because the interestrate charged by the IMF fails to reflect this default

risk, is thus less than 0.5 percent of the country’sGDP. If the country represents 1 percent of theworld population and GDP (this corresponds to alarge emerging economy, between Argentina andBrazil in size), the per capita cost of the bailout forthe global taxpayer would amount to less than0.0005 times that borne by the domestic taxpayer. Itbears emphasizing that even these small numbersare based on an extreme assumption underpinningour hypothetical worst-case scenario, namely, thatnone of the outstanding debt on “infinite” lendingcycles will be recovered. Thus, a reasonable estimateof the ex ante subsidy implicit in IMF lending islikely to be much smaller.

Policy Discussion

In light of the evidence presented in the previoussection, we now return to the main question moti-vating this paper—namely, how the internationalcommunity can best limit moral hazard. In additionto the papers and editorials cited in the introduc-tion, a number of reports have been written on thereform of international financial institutions in theaftermath of the Asian crisis (see Williamson, 2000for a review). One report that has been especiallyinfluential in the policy debate was prepared by theInternational Financial Institution Advisory Com-mission established by the U.S. Congress and

Box 2. Quotas and IMF Credit Limits

A “quota” is a country’s capital contribution to theIMF. It determines voting power within the institutionas well as access to IMF financing. Quotas are set as afunction of a country’s GDP, openness, the volatility ofexports (or current receipts), and the level of reserves(see IMF, 1998, Box 3). Large countries with diversi-fied exports tend to have much smaller quotas, as ashare of GDP, than do small countries whose exportsare concentrated in a few sectors. For example, the me-dian quota in 2000 was about 2.7 percent of GDP, butthe United States’ quota was only about 0.5 percent,and Zambia’s was about 18 percent. Quotas are re-viewed periodically (about once every five years).

Until 1963, a country’s total borrowing limit was100 percent of quota. This was gradually relaxed dur-ing the 1960s and 1970s, in step with the creation ofnew lending facilities, reaching about 300 percent by1978 (see Boughton, 2001). Following the second oilshock, access limits were briefly raised much higher,and remained at approximately 450 percent until1992, when the Ninth Review of Quotas increasedquotas by 50 percent. Since then, the limit on total cu-mulative borrowing from the IMF has been 300 per-cent of quota, with the understanding that this could

be exceeded in “exceptional circumstances.” Recently,two new facilities have been created: the Supplemen-tal Reserve Facility (SRF), for cases of “exceptionalbalance of payments difficulties,” and the ContingentCredit Lines (CCL), as a defense against contagion.These facilities are exempted from the formal accesslimit, but are also subject to much shorter repaymentperiods and much higher rates of charge.

Since 1992, the 300 percent cumulative ceiling hasso far been exceeded in six cases: Mexico (peak:607 percent of quota in 1995), Thailand (400 percentin 1998), the Republic of Korea (1,500 percent in 1998),Indonesia (431 percent in 1998), Turkey (445 percentin early 2001), and briefly Russia, whose outstandingdebt reached 318 percent of quota in 1998. TheFY 1998/99 Brazil package stayed within the 300 per-cent cumulative limit, in spite of the use of the SRF.

At present (in mid-2001) two countries still have li-abilities to the IMF that exceed the standard cumula-tive debt ceiling, namely Turkey and Indonesia. Formost countries, actual levels of debt outstanding in2000 are one-third to one-fourth, on average, of thetheoretical debt ceilings (see Table 3).

chaired by Allan Meltzer (IFIAC, 2000, hereinafterreferred to as the Meltzer report). We present a briefreview of the different policy approaches in these re-ports, before presenting the case for ex ante condi-tionality and concluding with a discussion of someobjections against it.

Policy Approaches

Proposals to reform IMF lending policies for the pur-poses of reducing moral hazard have varied alongthree main dimensions:

First, increasing the price (making it more costly forcountries to borrow from the IMF). The Meltzer re-port argues that the IMF should charge a penaltyrate (above the borrower’s recent market rate) andsecure its loan by a clear priority claim on the bor-rower’s assets. As documented in the previous sec-tion, the IMF increased some of its lending rates inNovember 2000.

Second, decreasing the quantity (limiting the size ofbailouts). The Council on Foreign Relations TaskForce (1999) proposes that the IMF return to nor-mal lending limits (100 percent to 300 percent ofquota) for “country crises,” that is, for crises that donot threaten the performance of the world economy.Larger loans could be made under a systemic facility,which would require a supermajority of creditors tobe activated. Arguments that there should be more“private sector involvement” in financing crises(Haldane, 1999; and Roubini, 2000) are in the samevein. Eichengreen (1999, 2000) advocates improv-ing crisis resolution mechanisms through changes inthe law governing private debt contracts, or throughofficially sanctioned standstills as a way to resolveinvestor panics. These can be interpreted as propos-als that seek to reduce the need to resort to large of-ficial bailouts as the main way of mitigating the con-sequences of an international financial crisis.

Third, introducing ex ante conditionality (sometimesalso referred to as “selectivity” or “prequalifica-tion”). The idea is to make the generosity of inter-national bailouts in a crisis conditional on the qual-ity of domestic policies, particularly in the financialsector, before a crisis erupts. The Council on For-eign Relations Task Force suggests that the IMFshould distinguish among three categories of coun-tries on the basis of their compliance with a set ofstandards and good practices, and publish regular re-ports assessing countries’ progress in meeting thesestandards. (The standards and practices include,among others, the IMF’s Special Data Dissemina-tion Standard and the Basel Committee’s Core Prin-ciples of Effective Banking Supervision.) A coun-try’s class would then determine the availability ofofficial assistance and the interest rate at which itwould be charged in the event of a crisis. TheMeltzer report takes a similar approach, but in amore extreme form. It recommends that IMF lend-ing be restricted to a group of countries selected for

the soundness of their banking policies, with othercountries being altogether ineligible for official cri-sis lending. The idea of ex ante conditionality is alsoembodied in a new IMF facility, the ContingentCredit Lines (CCL), which provided for exceptionalaccess to IMF resources for countries that qualifiedex ante on the basis of sound policies and progresstoward meeting internationally accepted standards.(For a precise statement of the conditions that wererequired to qualify for the CCL, see IMF, 2000a.)7

The Case for Ex Ante Conditionality

We do not think that there is much to be gainedfrom measures that increase the price of official crisislending—at least in terms of reducing moral hazard.It amounts to fixing a problem that does not exist.The welfare losses stemming from excessive moralhazard, if any, are in the countries receiving interna-tional financial assistance, not in the contributingcountries. Nor will the proposals in this categorywork as an incentive mechanism: in countries withsevere policy failures, governments are unlikely tobe deterred by higher interest rates, since they ulti-mately do not bear the costs of such interest rates.

Institutional and legal changes that reduce the“demand” for big bailouts by making debt restructur-ings less costly are clearly desirable from the point ofview of reducing moral hazard. Moreover, even ifsuch changes are not successful in eliminating theneed for an international financial safety net, sys-tematically smaller bailouts would be optimal if oneviewed domestic policy failures as pervasive. Inother words, it would be optimal to systematicallyrestrict the policy options of countries if onethought that they would systematically use them toimplement bad policies. This is a possible view ofthe world, but we see it as overly pessimistic. Theremight be “good countries”—that is, those that putthe international financial safety net to good use—and their welfare would be decreased by a one-size-fits-all reduction in bailouts.

Ideally, one would like to reserve the interna-tional financial safety net for countries with goodpolicies and deny it to countries with bad ones.This, in essence, is what ex ante conditionality at-tempts to achieve. There are two benefits. First, thistargets financial assistance to countries that arelikely to make the best use of it. Second, it gives allcountries an incentive to improve their policies(Jeanne and Zettelmeyer, 2001a).

The conclusion that crisis lending should belinked to policies before the crisis does not rely on a

197


7 No country has applied for the CCL, in part because it wasviewed as offering few advantages when compared with the Sup-plemental Reserve Facility (SRF), which has been used for large-scale lending without formal prequalification. In response, inNovember 2000, the IMF lowered the rate of charge on the CCLrelative to the SRF, and made several other changes designed toenhance the CCL’s attractiveness.

specific model of how policy failures arise. Rather, itfollows from the general trade-off between incen-tives and insurance. The better the institutions andpolicies that prevail before the crisis, the more “in-surance” can be provided, in the form of an interna-tional financial safety net, without destroying pri-vate sector incentives. Conversely, the worseinstitutions are, the greater the degree of market dis-cipline required to offset their adverse incentive ef-fects.

An additional benefit of linking crisis lending toex ante policies is that they could reduce the extentand intrusiveness of ex post conditionality. Afterthe Asian crisis, IMF structural conditionality wascriticized for being excessive (Goldstein, 2000a).One reason why more conditions may have beenimposed than were perhaps necessary to ensure re-payment (the standard justification for IMF condi-tionality), is that the IMF sought to take advantageof the time window in which it had some leverageover the governments—that is, while the countriesneeded IMF support. Imposing some conditions forcrisis lending ex ante rather than ex post would givethe IMF more leverage in good times, allowing it toallocate conditionality over time in a more efficientand balanced way.

Problems with Ex Ante Conditionality

Greater reliance on ex ante conditionality raises anumber of questions related to design and imple-mentation. One issue is precisely what ex ante con-ditions should be imposed. This is beyond the scopeof our paper, but the basic principle is clear: themaximum level of crisis assistance ex post (and pos-sibly the interest rate at which it is provided) shoulddepend on the sustainability of fiscal policies, publicdebt management, and the extent to which a coun-try implements financial sector policies that miti-gate excessive risk taking. The last factor includespolicies for effective banking supervision along thelines described in the Basel Committee’s “CorePrinciples” (Basel Committee on Banking Supervi-sion, 1997), accounting standards, and public dis-closure requirements.

One important problem with ex ante conditional-ity is that it may not be time-consistent (De Grego-rio and others, 2000). The temptation to bail out allinvestors is typically greater ex post than ex ante.This is a real problem, but it is not specific to inter-national bailouts; it arises, to some extent, with allfinancial safety nets (Rochet, 2000). The institu-tional design of these safety nets is aimed preciselyat solving this problem. The solution, as is the casewith time-consistency problems in monetary policy,involves rules, delegation, and reputation. Nor isthe time-consistency problem specific to ex anteconditionality. Conventional IMF conditionallyfaces a similar problem, since it is often optimal expost to release the next disbursement even if preced-

ing performance criteria have been violated. Inpractice, this is dealt with through rules that deter-mine in which circumstances program requirementscan be waived.

A related issue concerns the room that should beleft for “constructive ambiguity.” The need to pre-serve a measure of constructive ambiguity is themost serious objection against ex ante conditional-ity. On the one hand, it is important that the linkbetween the quality of domestic policies and the ex-tent of crisis lending be governed by explicit rules—both to deal with the time-consistency problem andto ensure universality of treatment ex ante. On theother hand, there will need to be exceptions tothese rules. For example, a crisis that is clearly trig-gered by a financial panic may call for a rescuepackage even when the policy record prior to thecrisis was mixed, or the risk of international conta-gion may require intervention even when the coun-try would not, in isolation, have qualified for aloan. Conversely, limited fund availability mayprevent the IMF from fulfilling its commitments toprequalified countries if a large number of them aresimultaneously hit by a crisis. These circumstancescannot always be incorporated into the rules be-cause they are difficult to describe ex ante or to ver-ify ex post.

However, the need for constructive ambiguitydoes not in itself invalidate the case in favor of someex ante conditionality. Domestic attempts at recon-ciling the two are instructive in this regard. For ex-ample, the reform of the U.S. deposit insurance sys-tem that followed the saving and loan crisisattempted to mitigate the moral hazard resultingfrom constructive ambiguity by establishing a sys-tem of checks and balances that places stringent ac-countability conditions on economic officials whenthey decide to rescue a bank because it is “too largeto fail” (Council on Foreign Relations Task Force,1999; Tirole, 2001).8 This approach is consistentwith an important lesson from the theory of incom-plete contracts: when contracts cannot fully specifythe actions of the contracting parties, they shouldinstead focus on the rules by which these actionswill be decided—the allocation of “decision rights.”Thus, behind the move toward ex ante conditional-ity in international bailouts looms the broader ques-tion of the governance structure—the rules and pro-cedures by which the international communitydetermines its response to international financialcrises.

198


8 The Federal Deposit Insurance Corporation ImprovementAct (FDICIA) of 1991 makes it harder for regulators to bail outuninsured creditors. An extension of the guarantee to all bankcreditors is possible in exceptional circumstances, but this re-quires the explicit consent of the secretary of the treasury in con-sultation with the president of the United States, two-thirds ofthe governors of the Federal Reserve System, and two-thirds ofthe directors of the FDIC.

199


Table A1

Correlations Between IMF and Bilateral Debt Stock Changes, 1975–99

Regression Sample

Independent variable All developing HIPC1 Non-HIPC Non-EMBIG EMBIG2

OECD debt change –0.001 0.001 –0.001 0.000 –0.098(–0.28) (0.29) (–0.22) (–0.05) (–1.21)

OECD debt change (–1) –0.026 –0.016 –0.032 –0.098 0.027(–1.00) (–3.73) (–0.93) (–5.27) (0.39)

OECD debt change (–2) 0.128 –0.019 0.139 0.008 0.207(4.64) (–4.41) (3.79) (0.39) (2.84)

Regression constant –153.6 –14.3 –235.8 –37.3 –577.6(–7.40) (–12.7) (–6.82) (–3.58) (–5.63)

Number of countries 117 40 77 93 24Number of observations 954 370 584 771 183

Notes: Dependent variable is the change in total IMF credits and loans outstanding; panel regressions are estimated using fixed effects, with t-values in parentheses. OECD denotes the Organization for Economic Cooperation and Development.

1 Group of heavily indebted poor countries.2 Group of emerging market countries whose sovereign bond spreads are tracked in J.P. Morgan's "Emerging Market Bond Index Global."

(See Appendix II.)

Appendix I. IMF Lending and Bilateral Lending

This appendix deals with Bulow, Rogoff, andBevilaqua’s (1992) argument that “any ability IFIs[international financial institutions] have to extractrepayments ahead of private creditors may come al-most entirely at the expense of bilateral governmentcreditors.” We undertake a simple empirical exerciseshowing that the relationship between repaymentsto the IMF and bilateral lending by OECD members(both concessional and nonconcessional) looks quitedifferent depending on the level of income and capi-tal market integration of the creditor countries.

Consider a regression of changes in IMF debtstocks on contemporaneous and lagged changes inthe stock of bilateral developing country debt heldby OECD countries (these data, going back to 1975,are available from the OECD). In the extreme case,in which repayments to the IMF are entirely fi-nanced by new bilateral debt, the contemporaneousand recent lagged changes in OECD debt shouldhave coefficients that add up to –1. In the less ex-treme case, in which new OECD debt contributesto, but does not fully finance, IMF repayments, thecoefficients should be negative, and add up to lessthan unity in absolute value. Table A1 presents re-sults from such a regression based on a panel of 161developing countries, restricting attention toepisodes of repayments to the IMF (i.e., condition-ing on negative changes in the IMF debt stock).

The results indicate that the Bulow, Rogoff, andBevilaqua (1992) hypothesis is not supported on theentire sample of developing countries, where we see

small and insignificant coefficients on the contem-poraneous OECD debt change and its first lag, and asignificant positive coefficient on the second lag.But the picture is strikingly different if one considersonly heavily indebted poor countries. Both the firstand second lags are now negative and highly signifi-cant, and add up to about –0.25. If attention is re-stricted to all “non-emerging” developing countries,the negative effect is smaller (about –0.1), but stillsignificant. In contrast, the relatively richer and“emerging” developing countries show a significantpositive relationship between repayments to the IMFand to OECD members.

These findings may not be entirely surprising. AsBulow, Rogoff, and Bevilaqua point out, one wouldexpect bilateral transfers to facilitate multilateraldebt repayment if bilaterals cared about multilater-als being repaid, and the amounts involved weresmall. But crisis lending to middle-income countriesis often large, and may be politically much morecostly than lending through specialized multilateralinstitutions. This is suggested by the shrinking roleof bilateral crisis lending after the Mexico crisis andits complete absence in the recent lending packagesto Argentina and Turkey.

Appendix II. Countries in HIPC and EMBIG Groups: Definitions

Used in Previous Section

Heavily Indebted Poor Countries (HIPCs): Angola,Benin, Bolivia, Burkina Faso, Burundi, Cameroon,the Central African Republic, Chad, Congo,

Côte d’Ivoire, the Democratic Republic of theCongo, Equatorial Guinea, Ethiopia, The Gambia,Ghana, Guinea, Guinea-Bissau, Guyana, Honduras,Kenya, the Lao People’s Democratic Republic,Liberia, Madagascar, Malawi, Mali, Mauritania,Mozambique, Myanmar, Nicaragua, Niger, Rwanda,São Tomé and Príncipe, Senegal, Sierra Leone, So-malia, Sudan, Tanzania, Togo, Uganda, Vietnam,the Republic of Yemen, and Zambia.

EMBI Global Index (EMBIG) countries: Algeria,Argentina, Brazil, Bulgaria, Chile, China, Colom-bia, Côte d’Ivoire, Croatia, Ecuador, Hungary, theRepublic of Korea, Lebanon, Malaysia, Mexico, Mo-rocco, Nigeria, Panama, Peru, the Philippines,Poland, Russia, South Africa, Thailand, Turkey,Ukraine, and República Bolivariana de Venezuela.

BibliographyAndrews, David, Anthony R. Boote, Sitad Rizavi, and

Sukhwinder Singh, 1999, Debt Relief for Low-IncomeCountries. The Enhanced HIPC Initiative, IMF Pam-phlet No. 51 (Washington: International MonetaryFund). Also available on the Web at http://www.imf.org/external/pubs/ft/pam/pam51/contents.htm.

Aylward, Lynn, and Rupert Thorne, 1998, “An Econo-metric Analysis of Countries’ Repayment Perfor-mance to the International Monetary Fund,” IMFWorking Paper 98/32 (Washington: InternationalMonetary Fund).

Basel Committee on Banking Supervision, 1997, “CorePrinciples for Effective Banking Supervision,” docu-ment available on the Web at http://www.bis.org.

Boughton, James M., 2001, Silent Revolution: The Interna-tional Monetary Fund, 1979–1989 (Washington: International Monetary Fund).

Bulow, Jeremy, Kenneth Rogoff, and Alfonso S. Bevilaqua,1992, “Official Creditor Seniority and Burden-Sharing in the Former Soviet Bloc,” Brookings Paperson Economic Activity: 1, Brookings Institution.

Calomiris, Charles W., 1998, “The IMF’s Imprudent Roleas Lender of Last Resort,” Cato Journal, Vol. 17, No. 3.

________, 2001, “How New Is the ‘New IMF?’” Testi-mony Before the Joint Economic Committee of theUnited States Congress (March 8).

Council on Foreign Relations Task Force, 1999, Safe-guarding Prosperity in a Global Financial System: TheFuture International Financial Architecture (Washing-ton: Institute for International Economics).

De Gregorio, José, Barry Eichengreen, Takatoshi Ito, andCharles Wyplosz, 2000, An Independent and Account-able IMF, Geneva Reports on the World EconomyReport, Vol. 1 (September).

Dell’Ariccia, Giovanni, Isabel Gödde, and JerominZettelmeyer, 2000, “Moral Hazard and InternationalCrisis Lending: A Test” (unpublished; Washington:International Monetary Fund).

Edwards, Sebastian, 1998, “Abolish the IMF,” FinancialTimes, November 13.

Eichengreen, Barry, 1999, Toward a New International Fi-nancial Architecture. A Practical Post-Asia Agenda(Washington: Institute for International Economics).

_________, 2000, Can the Moral Hazard Caused by IMFBailouts be Reduced? Geneva Reports on the WorldEconomy, Special Report No. 1.

Feldstein, Martin, 1998, “Refocusing the IMF,” Foreign Af-fairs, Vol. 77.

Fischer, Stanley, 1999, “On the Need for an InternationalLender of Last Resort,” Journal of Economic Perspec-tives, Vol. 13 (Fall).

Frydl, Edward, 1999, “The Length and Cost of BankingCrises,” IMF Working Paper 99/30 (Washington: International Monetary Fund).

Goldstein, Morris, 2000a, “IMF Structural Programs” (un-published; Washington: Institute for InternationalEconomics).

______, 2000b, “Strengthening the International Finan-cial Architecture: Where Do We Stand?” Institutefor International Economics Working Paper No. 00-8(Washington).

Greene, William H., 1993, Econometric Analysis, 2nd ed.(New York: Macmillan).

Haldane, Andy, 1999, “Private Sector Involvement in Financial Crisis: Analytics and Public Policy Approaches,” Bank of England Financial Stability Review (November).

International Financial Institutions Advisory Commis-sion (IFIAC), 2000, “Report of the International Financial Institution Advisory Commission,” AllanH. Meltzer, Chairman (Washington) (also referred toas the Meltzer report).

International Monetary Fund (IMF), 1998, Financial Or-ganization and Operations of the IMF, IMF PamphletNo. 45 (Washington: International Monetary Fund).Also available on the Web at http://www.imf.org/external/pubs/ft/pam/pam45/contents.htm.

______, 2000a, “Summing Up by the Acting Chairman ofthe IMF Executive Board: Contingent Credit Lines,”IMF Executive Board Meeting, November 17 (Wash-ington: International Monetary Fund). Also avail-able on the Web at http://www.imf.org/external/np/pdr/fac/2000/02/111700.htm.

______, 2000b, “Review of Fund Facilities: Proposed Deci-sions and Implementation Guidelines,” prepared by theLegal, Policy Development and Review, and Trea-surer’s Departments, November 2. Available on theWeb at http://www.imf.org/external/np/pdr/fac/2000/02/index.htm.

Jeanne, Olivier, and Jeromin Zettelmeyer, 2001a, “Inter-national Safety Nets, Private Debt Crises and MoralHazard” (unpublished; Washington: InternationalMonetary Fund), forthcoming as an IMF WorkingPaper.

______, 2001b, “International Bail-Outs, Domestic PolicyFailures and Moral Hazard” (unpublished; Washing-ton: International Monetary Fund), forthcoming asan IMF Working Paper.

Lane, Timothy, and Steven Phillips, 2000, “Moral Hazardin IMF Financing,” IMF Working Paper 00/168(Washington: International Monetary Fund).

McBrady, Matthew, and Mark S. Seasholes, 2000, “Bailing-In” (unpublished; University of California, Berkeley).

Meltzer report—see IFIAC.Mussa, Michael, 1999, “Reforming the International Fi-

nancial Architecture: Limiting Moral Hazard andContaining Real Hazard” (unpublished; Washington:International Monetary Fund).

200


201


Nunnenkamp, Peter, 1999, “The Moral of IMF Lending:Making a Fuss About a Minor Problem?” Kiel Discus-sion Paper No. 332 (Kiel, Germany: Kiel Institute ofWorld Economics).

Rochet, Jean-Charles, 2000, “The Future of Banking Supervision: Market Discipline vs. Centralized Super-vision” (unpublished; Toulouse, France: Institut d’Economie Industrielle, Université des Sciences Sociales).

Roubini, Nouriel, 2000, “Bail-In, Burden-Sharing, PrivateSector Involvement (PSI) in Crisis Resolution andConstructive Engagement of the Private Sector” (un-published; Department of Economics, New YorkUniversity).

Schwartz, Anna J., 1998, “Time to Terminate the ESF andthe IMF,” Cato Institute Policy Briefing No. 48(Washington: Cato Institute).

Shultz, George, William E. Simon, and Walter B. Wris-ton, 1998, “Who Needs the IMF?” Wall Street Journal,February 3.

Summers, Lawrence H., 1999, “The Right Kind of IMF fora Stable Global Financial System,” remarks to theLondon School of Business, December 14.

Tirole, Jean, 2001, “Financial Crises, Liquidity Provisionand the International Monetary System,” manuscriptprepared for the Paolo Baffi Lecture on Money andFinance.

Williamson, John, 2000, “The Role of the IMF: A Guideto the Reports,” IIE Policy Brief No. 00-5 (Washing-ton: Institute for International Economics).

Zhang, Xiaoming Alan, 1999, “Testing for ‘Moral Hazard’in Emerging Markets Lending,” IIF Research PaperNo. 99-1 (Washington: Institute of International Finance).

202

12

During the 1990s, the concept of “catalytic official fi-nance” (COF) gained prominence in policy debates.The concept revolves around the idea that the propensityof investors to lend to a country increases when the IMFprovides its “seal of approval”—backed up by only lim-ited official financing—on the country’s economic pro-gram. COF aims at avoiding, on the one hand, the mas-sive use of public money to bail out private investorsand, on the other, the recourse to coercive bailing-inmechanisms. This paper concludes that COF, while pos-sibly useful in other contexts, is less reliable when used tomanage capital account crises.

Chi non può quel che vuol, quel che può voglia2

—Leonardo da Vinci

Introduction

Ever since Walter Bagehot’s times, it has been rec-ognized that emergency liquidity support, to be ef-fective, must in principle be unlimited. Of course,the precept should not be taken too literally. WhatBagehot meant, and central banks duly understood,was that the lender of last resort (LOLR) shouldprovide money in amounts large enough to makegood all the ailing bank’s short-term (and therefore“liquid”) liabilities, so as to reassure their holdersthat all claims would ultimately be met. In a panic,Bagehot reasoned, partial insurance is no insuranceat all. To meet the precept, sometimes central bankshave acted on their own, availing themselves of thecapacity to create money out of nothing. In otherinstances, exploiting their authority as bank super-visors, they have managed to arrange “lifeboats,” orconcerted rescues, involving the participation of a

group of private institutions with sufficient liquidityto soothe the markets (Freixas and others, 2000).

Generalizing Bagehot’s precept to internationalliquidity support directed at countries undergoing aforeign exchange crisis has proved a daunting task.Some argue that this was only natural, given thelack of a truly supranational currency that could besupplied in unlimited amounts (Capie, 2002). Butthis explanation sounds unconvincing. Surely, thelack of a world currency notwithstanding, sufficientresources could be mustered to cope even with rela-tively large countries’ foreign exchange crises, if thepolitical will of a group of powerful nations could beharnessed? (Fischer, 2002). Thus, to explain thelack of a Bagehotian LOLR at the internationallevel, one must look for something deeper.

More plausibly, attention could be directed at thefragmentation of the regulatory and political envi-ronment in a world of sovereign nations. As CharlesGoodhart and Gerhard Illing have recently argued,at the international level,

…the whole process is far more complex than inthe national case, with a multiplicity of partici-pants (governments, central banks, creditors anddebtors), of legal systems and infrastructures. So,besides the standard concerns about contagionon one side, and moral hazard on the other(which remain just as strong in the internationalas in the national contexts), questions about thedesign and conduct of an international LOLRare complicated by externalities due to coordi-nation problems among many agents and thelack of enforcement mechanisms among sover-eign states. (Goodhart and Illing, 2002, p. 21)

Operationally, this greater complexity is felt atvarious levels (Giannini, 1999): in the greater diffi-culty of reaching agreement within the official com-munity as to when, how, and in what amount LOLRoperations should be conducted; in the lack of regu-latory (i.e., externally enforced) means to hold sov-ereign debtors true to their promise to reimburse theemergency finance they have received—hence theneed to “safeguard” the resources of the interna-tional LOLR through conditionality; in the relativeineffectiveness of moral suasion, especially when the

Bedfellows, Hostages, or Perfect Strangers? Global Capital Markets and the Catalytic Effect of IMF Crisis Lending

Carlo Cottarelli and Curzio Giannini1

1 This paper has benefited from the comments of Michael Bell,James Boughton, Marco Committeri, Stanley Fischer, RexGhosh, Rakia Moalla-Fetini, Pier Carlo Padoan, Dani Rodrik,Alina Rubino, Massimo Russo, and the participants in seminarsheld at the European I Department of the IMF and at the Bancad’Italia. This is a slightly revised version of a paper presented atthe conference on “The Lender of Last Resort—Experiments,Analyses, Controversies,” held in Paris September 23–24, 2002and first published in Cahiers d’économie politique, No. 45 (2003)(Paris: Editions L’ Harmattan).

2 Translation: “He who cannot get what he wants had betterwant what he can get.”

CHAPTER

203

Carlo Cottarelli and Curzio Giannini

bulk of a country’s external debt takes the form ofbond financing, since bondholders are hard to bindin the “mutual hostage” relationship that is the typi-cal effect of moral suasion.

These impediments to the working of an interna-tional LOLR were less significant in the postwarworld of pervasive capital controls and relativelyunderdeveloped international capital markets. It isbecause of this supportive institutional setup that atBretton Woods the IMF could be assigned a limitedLOLR role—that of providing short-term financingto overcome current account imbalances (Helleiner,1994).

With the almost universal move toward financialliberalization in the 1980s, and the associatedreprise in capital mobility, this “reductionist” strat-egy could no longer work. Hence, the task of adapt-ing domestic LOLR practices to the internationalenvironment had to be faced squarely. Major over-hauls of the Bretton Woods framework were ruledout by the political impracticality of reaching agree-ment on a new international treaty, which meantthat adjustment had to take place at the margin.Under the pressure of events, there thus emerged anew kind of LOLR, in the form of catalytic official fi-nance (COF). The underlying idea of COF is that theprovision of official resources—typically the IMF’s—to a country in the context of a full-fledged programmight increase the propensity of private investors tohold financial assets in the country concerned. Inother words, COF postulates that under appropriateconditions private capital flows may be expected tobehave like dependable “bedfellows” of official fi-nance, thanks to the various services provided bythe IMF through its lending and other activities.

The notion of COF has been invoked in practicallyall the capital account crises of the period 1997–2002(Ghosh and others, 2002), and as a consequence ref-erences to “catalytic” forces have also multiplied untilrecently in policy documents. However, after the dif-ficulties encountered by IMF programs in Argentinaand Turkey—where private and official financialflows have behaved more like perfect strangers thandependable bedfellows—attitudes toward COF havebegun to change. Thus, no reference to “catalysis”can be found in the press communiqués issued bythe International Monetary and Financial Commit-tee (IMFC) at the 2002 and 2003 Annual Meetingsof the IMF and the World Bank.

The time therefore seems ripe for a comprehen-sive assessment of COF, its potentialities, and itslimitations. In this paper, we have set ourselvesthree tasks. The first is to trace the process throughwhich (and the reasons why) the notion of COF hastaken shape. The second is to survey the empiricalevidence on the existence and magnitude of cat-alytic effects. And the third is to suggest possibleways to rationalize COF in theoretical terms,thereby also identifying the conditions that must bemet in order for COF to work.

Three things ought to be stressed right at the out-set to avoid misunderstandings. First, the notion ofCOF does not pertain only to the handling of confi-dence crises—the typical domain of domesticLOLRs. One may also find catalytic effects at workunder more ordinary circumstances, such as in help-ing a country gain access to international capitalmarkets under a prolonged period of financial isola-tion. Indeed, as we shall try to show in the sectionon the history of COF, the notion of COF originallyhad nothing to do with capital account crises. Sec-ond, under crisis conditions the notion of COF at-tempts to cover the gray area between the two ex-tremes of full bail-out (the typical outcome ofBagehotian lending) and full bail-in (the immediateimpact of either default or a temporary suspension ofdebt service) of the country’s creditors. This grayarea is the domain on which we focus. Our analysistherefore does not directly deal with the pros andcons of the two polar opposites that delimit the cho-sen turf, although of course much of what we will besaying might be seen as relevant for that purpose aswell. Third, the catalytic effect is by no means theonly channel through which IMF programs work(see Mussa and Savastano, 1999, for a comprehen-sive assessment of the economics of IMF programs).Conclusions regarding the strength—or weakness—of catalytic effects are directly relevant only for pro-grams that can be expected to work only if catalyticforces are at play. Regarding other programs, the fol-lowing analysis has very little to say.

Defining Catalytic Official Finance

Despite the frequency with which the notion is in-voked, no precise definition of COF has ever beenprovided. This is hardly surprising. COF is builtaround a metaphor, much like the invisible hand orthe inconsistent trinity. Like all such mental con-structs, this too is hard to pin down into an opera-tional concept. Looking for guidance, we have de-cided to take the metaphor at face value andconsulted a Webster’s dictionary, where a (chemi-cal) catalyst is defined as “any substance that initi-ates a reaction and enables it to take place undermilder conditions than in [its] absence.”

Three features of this definition ought to bestressed. First, the reaction, though clearly intendedby the chemist, is spontaneous. In our context, thismeans that COF should be distinguished from otherpotential ways of addressing unstable capital flows,ranging from direct intervention (suspension of pay-ments or capital controls) to milder actions (moralsuasion, concerted lending), which have also foundapplication at the international level in recenttimes. Second, the purpose of catalysis is to alleviatethe burden associated with the intended event. Wetake the “intended event” of IMF financial packa-ges as being the restoration of the medium-term

204

12 � GLOBAL CAPITAL MARKETS AND THE CATALYTIC EFFECT OF IMF CRISIS LENDING

sustainability of the financial profile of a givencountry. The purpose of COF should therefore beconstrued as bringing about such an event at lowercost for the country (i.e., with less domestic adjust-ment) and for the international community (i.e.,with less direct exposure and less moral hazard) thanwould otherwise be the case. Third, even if not ex-plicit in Webster’s definition, the amount of the cat-alyst substance is supposed to be limited with respectto the substance that reacts to it. For our purposes,this can only mean that the amount of IMF re-sources should be limited compared with the poten-tial capital outflow a country is subject to.

Accordingly, the IMF’s involvement in a countryhas a catalytic effect to the extent that the an-nouncement of an economic program backed up bya limited amount of IMF resources (compared withthe size of the potential capital outflow) increasesthe propensity of private investors to lend to thecountry concerned, thereby reducing the adjust-ment burden falling on the debtor country with re-spect to the no-catalysis scenario.

This definition is not yet operational, as it leavesunspecified the exact meaning of the expressions,“potential capital outflow” and “limited.” As regardsthe former, a useful benchmark is the total amountof the country’s external short-term liabilities. But itought to be borne in mind that this is just a bench-mark, more useful under ordinary conditions thanduring a generalized confidence crisis. In fact, withfull capital account convertibility, residents might atany time decide to convert the entire stock of short-term assets into foreign exchange. This would in-crease the “potential” capital outflow dramatically.

As to the meaning of “limited,” it is clearly tied tothe potential outflow, but the relationship is notmechanical, the more so in view of the inevitableambiguity we have just pointed out in the bench-mark. Zettelmeyer (2000), for example, considers“limited” any package that is not large enough to fi-nance outflows assuming a zero rollover rate of theexisting short-run external liabilities. But a “lim-ited” package of, say, US$40 billion might have astronger impact on the private sector’s investmentdecisions than would a US$4 billion one, especiallyif the intervention is appropriately timed. That is,definitional ambiguities cannot simply be dispelledaltogether, and we must live with them. All that cansensibly be said is that the smaller the amount of fi-nancing compared with the size of the potentialcapital outflow, the larger the reliance on catalyticforces. Conversely, a large package will more closelymimic an “unlimited” package.

IMF lending may have a catalytic effect regardlessof whether the country is already facing a capital ac-count crisis, but clearly under crisis conditions theconcept assumes particular importance. It would beuseful if one could have a precise definition of whatshould count as a “crisis.” Alas, after decades of re-flection on LOLR practices at the national level the

subject is still a matter of great controversy, and weshy away from the task. Suffice it to say that we usea notion of crisis broad enough to encompass notonly situations in which a country is experiencing amassive capital outflow but also precrisis situations,such as when the country’s foreign exchange re-serves no longer exceed scheduled amortization ofexternal debt during the next year or so.3 WhenCOF is used for countries that find themselves insuch circumstances, we refer to it as being used forcrisis management purposes. This is to differentiateit from two other possible uses of COF, namely for acrisis prevention purpose (as in “precautionary” IMFprograms) or for crisis resolution (as in the contextof debt restructuring following default).

The History of COF: A Tale of Two Crises

Until the 1970s, the notion of catalytic lending hadnever been entertained, either at the domestic or atthe international level. As already remarked, Bage-hot’s principle dictated that domestic emergencylending should be provided to whatever extent wasneeded to meet the demand for liquidity expressedby the market. As for international practices, theBretton Woods architecture was conceived to ruleout by construction any form of LOLR in excess oftemporary current account financing. To make therestriction tighter, each member’s access to IMFcredit was bound by its quota. Moreover, the IMF’sArticles of Agreement explicitly prohibited grant-ing credit to finance a sustained capital outflow.

While domestic practices have not substantiallychanged to this date, international practices haveevolved over time, especially since the late 1970s.The seeds of catalytic emergency lending were sownin the late 1970s in the context of the recycling ofoil-surplus funds, but the notion came to full blos-som only in the following decade, with the pro-tracted attempt to find a way out of the developingcountry debt crisis. By the end of the decade, how-ever, the effectiveness of catalytic lending was in-creasingly being questioned. A new change in atti-tudes occurred during the late 1990s, essentially onthe initiative of the Group of Seven (G-7), whichhad grown concerned about the drawbacks of thelarge financial package assembled to overcome theMexican crisis of 1994–95. Catalytic lending wasthen made a central pillar of the official strategy for crisis management at the IMF 2000 AnnualMeeting in Prague. Ever since then, the pendulumseems to have swung backward, with the merits (and

3 This is, for example, the benchmark proposed by the formerArgentinean Deputy Treasury Minister, Pablo Guidotti, as a mea-sure of prudent risk management on the part of domestic author-ities. See Financial Stability Forum (2000).

effectiveness) of catalytic lending once again beingincreasingly called into question.

A New Notion Begins to Take Shape (Late 1970s to Early 1980s)

In the context of the limited capital mobility of thepostwar period, the IMF recipe for addressing bal-ance of payments difficulties was fairly straightfor-ward. When a country faced current account pres-sures, the IMF would typically arrange a programfeaturing a combination of official financing and do-mestic policy adjustment, the latter including a cur-rency depreciation in cases of “fundamental disequi-librium.” Within this mechanism, quotas—that is,each country’s contribution to IMF resources—played the dual role of a constraint on the capacityof each member to borrow and of a ceiling on the re-sources each member would have to contribute tothe common pool.

The ability of the IMF to play its intended rolecame under stress during the 1970s, as a conse-quence of worldwide inflation, which reduced thereal value of existing quotas, and larger current ac-count imbalances in the aftermath of the oil crises.Faced with the prospect of being unable to meet themembers’ needs arising from the first oil shock, theIMF was forced to contemplate at least a partialtransformation from a credit union into a financialintermediary. This essentially meant bringing bor-rowing into the panoply of funding means. The pos-sibility of direct borrowing on financial markets wasruled out at the outset, because it would have funda-mentally altered the intergovernmental nature ofthe institution. As a consequence, the IMF’s financ-ing needs would have to be satisfied by borrowingfrom member countries. To be sure, this transforma-tion already had begun in 1963, with the creation ofthe General Arrangements to Borrow (GAB). Butthe departure from previous practice remained mod-est until the 1970s. In February 1977, IMF Manag-ing Director H. Johannes Witteveen proposed aSupplementary Financing Facility that would linktraditional lending based on quotas with funds ob-tained through borrowing from countries.

The transformation of the IMF into a financialintermediary, however, did not go unchallenged.First, this looked much more controversial in thelate 1970s, after international financial markets hadtaken up the key role in “recycling” the current ac-count surpluses of Organization of Petroleum Ex-porting Countries (OPEC) members. Second, andmore important from our perspective, the notionthat IMF lending could cause moral hazard andtherefore that “taxpayers’ money” should not beused to bail out private investors was increasinglyattracting political attention.

It is in connection to this latter problem that theidea that IMF resources, rather than substituting forprivate capital, could be used to sustain financial

markets’ confidence and catalyze private capitalflows began to take shape. This is most evident inthe testimony given in 1977 by Anthony Solomon,Undersecretary of the Treasury for Monetary Affairs,before the Committee on Banking, Finance andUrban Affairs of the U.S. House of Representatives.To defend the proposed Witteveen facility, Solomonmade the case that the new IMF window would notbe used to bail out the private banks, as the lattertypically follow the IMF in their lending decisions:

The very fact that [countries] are meeting theIMF’s performance criteria and thus continue tobe eligible to draw from the IMF tends to repre-sent a kind of Good Housekeeping seal of ap-proval. Good performance under an IMF pro-gram tends to result in private capital inflows,private banks being willing to lend more to thecountry concerned. (U.S. House of Representa-tives, 1977, p.72).4

The notion that private capital may tend to “fol-low” IMF lending had also begun to gain currencywithin the IMF itself. Margaret Garritsen de Vries,for many years the IMF’s historian, described themechanism as follows:

A close tie between the willingness of commer-cial banks to lend to a country and the country’shaving a stand-by arrangement with the IMF be-came almost standard after commercial banksencountered repayment problems with Zaïre in1976 and 1977…. Commercial bank officialswere beginning to be more familiar with IMFparlance and policy, and the test which they de-vised to judge the creditworthiness of a countrywas whether it could meet the conditionsneeded for a stand-by arrangement with the IMFin the upper credit tranches. This test inducedprivate commercial banks to have prospectiveborrowing countries obtain the stamp of ap-proval of the IMF by obtaining a stand-byarrangement in the upper credit tranches.5

De Vries might have overstated the importanceattached to the mechanism in the “IMF parlance

205


4 Solomon produced as supporting evidence a short empiricalstudy, showing that between 1970 and 1975 a number of IMFprograms had been followed by an increase in private lending(see the section on the history of COF). Interestingly, virtuallythe same wording is included in an oft-quoted remark madetwenty years later during the debate on the role of the IMF thattook place in the Treasury Committee of the U.K. House ofCommons: “But is there not an all pervasive conventional wis-dom that if you do sign up to an IMF programme you get theGood Housekeeping Seal of Approval and away you go!” (quotedby Bird and Rowlands, 1997).

5 de Vries (1985), pp. 493–94. A reference to banks followingthe IMF seal of approval in this period, made by a private banker,is Guth (1979), member of the board of the Deutsche Bank andformer Executive Director for Germany at the IMF.

and policy,” however. Indeed, references to what wenow call catalytic effects in the writings of IMF staffremained for a while episodic.6 Moreover, it seemsclear from reading such texts that at that time theseal-of-approval mechanism was not seen as havingmajor implications for the way the IMF operated.Thus, Manuel Guitián, who would later becomehead of the Monetary and Exchange Affairs Depart-ment of the IMF, argues, in a paragraph remarkablyentitled “The Fund as Catalyst,” that “a critical sideeffect of the mix of adjustment and financing that istypically built into the programs supported by Fundresources has been to help members attract flows ofcapital from sources other than the Fund” (Guitián,1982, p. 91).

This passage solicits two remarks. First, contraryto Dooley’s (1994) suggestion that the notion of cat-alytic lending is to be attributed to the Baker Plan of1985, one can see that not only the notion, but alsothe label, was known to IMF staff several years ear-lier. Second, Guitián does not argue that the effectlies at the core of the IMF’s role. Rather, he is quiteexplicit that this is but a “side effect” (although a“critical” one) of IMF programs. The reason proba-bly is that Guitián was still contemplating a worldin which balance of payments crises were predomi-nantly being determined by the behavior of the cur-rent account.

The Debt Crisis, 1982–93

Increased financial liberalization and capital mobil-ity during the 1980s were making the latter assump-tion increasingly questionable. The capital accountwas rapidly becoming as important a source of bal-ance of payments pressure as the current account.But capital account pressures are far more compli-cated to deal with. The problem arose for the firsttime in August 1982, when Mexico started havingtrouble in servicing its external debt, in the after-math of a generalized increase in international in-terest rates. The problem soon spread to at leasteight other countries in Europe, Africa, and LatinAmerica. The IMF was called in to help solve theproblem, but it soon became apparent that “the oldapproaches would not work, because new financingfrom the Fund would quickly be siphoned off as theindebted countries would have to repay other credi-tors” (Boughton, 2000a, p. 171).

In the old approach the IMF could work out theneeded amount of “policy adjustment” and “financ-ing” on the assumption that the overall financinggap the country faced was given and easy to mea-sure. Now, the possibility of a drawdown in the for-eign sector’s holdings of domestic debt—as well as ofan increase in residents’ assets abroad—implied thatthe size of the overall financing gap depended onthe credibility of the promised policy adjustment.Unless one found a way to convince investors not toliquidate their holdings, financing a country’s bal-ance of payments crisis might turn into trying to fillup a bottomless pit. Of course, the problem couldhave been obviated by providing a full bail-out of, atleast, foreign creditors, but this option was never se-riously contemplated in policy circles, on thegrounds that it would conflict with both the creditunion structure and the monetary nature of the IMF(Boughton, 2001), not to mention obvious politicaldifficulties in raising sufficient funds.

With the “Bagehotian” solution ruled out, the of-ficial sector’s strategy took at first the form of con-certed lending.7 The strategy was inaugurated in No-vember 1982, when the IMF’s then–ManagingDirector, Jacques de Larosière, informed a gatheringof bankers in New York that the IMF would not ap-prove a request by Mexico for a US$4 billion loanuntil private banks provided written assurances thatthey would as a group increase their Mexican expo-sure by US$5 billion.

With respect to domestic experiments with or-chestrated rescues, all of which are ultimately basedon the power of the central bank to hold privatebanks “hostage” through moral suasion, this interna-tional version’s main novelty lay in its formalizedcharacter. Under the “financing assurances policy,”as it came to be known, the IMF would suspend ac-cess to its own money if creditor banks refused toreschedule the country’s debt (Boughton, 2001,p. 477). Such a high degree of formalization was re-quired to make up for the IMF’s lack of direct powersover commercial banks. It had a serious drawback,though, in that it virtually gave banks veto powerover the approval and financing of adjustment pro-grams. More bluntly, under the new practice it wasunclear who was being made hostage to whom.8

Yet, the concerted lending strategy was appliedwith some success to a dozen cases between 1982and 1986 (Boughton, 2000a, p. 172). By 1985, how-ever, the strategy was no longer working, essentiallybecause of commercial banks’ growing unwillingness

206


7 Hereinafter we refer to a Bagehotian solution in a ratherloose sense, since, as already remarked, international rescuepackages have always involved a measure of conditionality,something that Bagehot never contemplated.

8 And, in fact, to encourage banks to take a cooperative stance,authorities in creditor countries made some concessions as tohow the new loans would have to be treated for supervisory pur-poses. See Sgard (2002).

6 Bird and Orme (1981) argue that “the Fund itself clearly be-lieves … that borrowing from it, particularly on the basis of astand-by agreement, acts as a catalyst for the generation of pri-vate capital inflows” and quote Sturc (1978), Gold (1979), andde Larosière (1980) as relevant references. However, only the lat-termost includes an explicit reference to the fact that the adoptionand pursuit of suitable adjustment programs in cooperation with theFund generally tends through enhancement of the creditworthiness ofborrowing countries to facilitate the attraction of private capital. Notealso that the reference in de Vries (1985) quoted above covers lessthan one page in her 600-page history of the IMF in the 1970s.

207


to be “concerted.” Financial and macroeconomicdevelopments converged to harden bankers. On thefinancial side, the time gained through the con-certed packages had allowed banks to build up con-siderable provisions against sovereign credit risk.Feeling less exposed to the threat of outright default,banks gradually assumed a tougher negotiating posi-tion. On the macroeconomic side, the policy adjust-ment of the countries benefiting from concertedlending had been insufficient, particularly in thestructural area, resulting in protracted stagnation.This made concerted lending increasingly unpopu-lar, as banks feared they would end up throwinggood money after bad. Indeed, net bank lending todeveloping countries declined dramatically in 1985,and turned negative the next year.

All this provided the setting for the U.S. TreasurySecretary James Baker’s famous October 1985 speechlaying out a “Program for Sustained Growth”—laterto be known as the Baker Plan. The plan’s underly-ing idea was that to avert massive cuts in lending todeveloping countries, growth had to be revived. Tothis end, what was needed was a combination ofstructural reform in recipient countries and the mo-bilization of financial resources on a large scale.How to achieve the first was fairly straightforward:one needed only to extend the scope of IMF condi-tionality to structural aspects, something the IMFhad already begun doing. The latter task was morecomplicated. It is here that the notion of “catalysis”came in handy.9 The Baker Plan posited that theprovision of sufficient amounts of official financingin the context of structural reform aiming at eco-nomic liberalization would catalyze private sectorlending to the extent needed to fill whatever fi-nancing gap was created in the process of spurringeconomic growth. It was estimated that the moneyneeded to finance the plan during 1986–88 wasUS$17–18 billion for official money and US$20 bil-lion for private lending (Boughton, 2001, p. 428).

This first experiment with using COF in crisismanagement yielded mixed results. Actual officiallending was very close to the planned figure, but thecatalytic effect on private lending sought was hardlyvisible.10

Leaving figures aside, the experiment led to wide-spread frustration. As Paul Volcker puts it, barely a

year after the launching of the plan “any sense ofenthusiasm was very much gone” (Volcker and Gy-ohten, 1992, p. 215). Even the most ardent advo-cates of the plan had to recognize that “by 1988there was no question that the banks’ willingness toprovide new financing was essentially finished”(Boughton, 2001, p. 428). The symbolic coup degrâce to the Baker Plan came when, in 1988, Citi-corp provisioned US$3 billion against Citibank’sexposure toward developing countries. While theprice of Citicorp stock skyrocketed, the prices of de-veloping countries’ loans on secondary marketsplummeted. Not surprisingly, new lending to devel-oping countries nearly dried up over the next twoyears.

In this environment, proposals to deal with thedeveloping countries’ problems by resorting to someform of debt relief began to gain currency, first inEurope and Japan, then in the United States. TheIMF itself began to show dissatisfaction with bothconcerted lending and catalytic financing by adopt-ing a new operational framework called the “menuapproach,” which entailed dealing with individualcrises by presenting creditors with a range of op-tions, all of which implied replacing old loans withnew instruments that more accurately reflected sec-ondary market values.

There was a critical problem, however, before thenew strategy could be translated into concrete ac-tions. Under the “financing assurances” policy, theIMF was for all practical purposes unable to forcedebt relief onto recalcitrant creditors. This policyhad to be modified. Thus, the plan concocted byU.S. Treasury Secretary Nicholas Brady in the late1980s, envisaging a comprehensive scheme for debtreduction, also encouraged the IMF to reconsiderits financing assurances policy to reduce the riskthat some creditors might hold out for a protractedperiod, thereby undermining the success of thedebt-reduction strategy in individual cases. TheIMF Board endorsed the Brady Plan in May 1989.Under the new operational framework, an accumu-lation of arrears to banks could be tolerated evenpending agreement between the country and itscreditors, provided “negotiations continue and thecountry’s financing situation does not allow (ar-rears) to be avoided.” The new policy meant adapt-ing COF to the needs of crisis resolution, as theIMF, through its lending into arrears, would nowpressure creditors to reach agreement with thecountry’s authorities on the appropriate amount ofdebt reduction.

The Brady Plan can be counted among the successes of international cooperation. By 1993,eight Latin American countries, through a variety ofoperations, reduced their external obligations byUS$42 billion, out of an initial stock of obligationsof about US$104 billion (Boughton, 2001, p. 552).Over the same period, rescheduling agreements withbilateral official creditors achieved through the

9 It is indeed in the mid-1980s that use of the word “catalytic”becomes more frequent in the literature on international finan-cial arrangements. See, for example, Kenen (1986); Buira(1987); and the empirical literature cited in this paper’s sectionentitled “A Survey of the Empirical Evidence.”

10 On the basis of stock data, IMF staff estimated zero net pri-vate lending over 1986–88. Other analysts, such as Cline (1989),whose views appear to have been shared by the U.S. Treasury, es-timated actual flows at some US$13 billion, still well below theplanned US$20 billion. As Boughton (2001, p. 428) explains,the difference was to be attributed to some offsetting transactions(in the form of swaps, write-downs, and buybacks) that the majorbanks put in place to contain their overall exposure while in-creasing direct lending.

208


Paris Club totaled about US$35 billion. The devel-oping country debt crisis was finally over.

The Emerging Market Crisis, 1994–2002

After the launch of the Brady Plan, discussions oninternational crisis lending stopped for severalyears—because of the lack of major new balance ofpayment crises, among other things. The idea thatIMF lending could catalyze private financial inflowshad not lost its appeal, however, particularly withinthe staff and management of the IMF (see, for exam-ple, IMF, 1992; and Guitián, 1992, p. 20).

As a matter of fact, the catalytic effect lay at theheart of “precautionary” programs aimed specificallyat promoting financial markets’ confidence throughthe IMF seal of approval, rather than providingmember countries with official finance. “Precau-tionary” programs—programs in which the countryauthorities state their intention not to draw—became quite popular in the early 1990s, especiallyin those countries with a recent history of govern-ment intervention in the economy, and thereforewith a structural credibility deficit (Dhonte, 1997).By the time of the outbreak of the Asian crises,such programs accounted for one-third of all Stand-By and Extended Fund Facility (EFF)Arrangements. Moreover, the ratio between actualand potential borrowing in all outstanding IMF arrangements—excluding those off track—declined significantly between the early 1980s andthe middle of the 1990s (Cottarelli and Giannini,1997).

Interestingly, however, when Mexico was hit by adevastating capital account crisis in late 1994, thecatalytic effect was not invoked. Indeed, as Kenen(2001, p. 22) remarks, the package that wasmounted to cope with the crisis (totaling aboutUS$50 billion, mostly from the IMF) had all thefeatures of a Bagehot-style LOLR operation:

The official financing for Mexico was unique inmore than its unprecedented amount. Althoughother crisis-stricken countries obtained large-scale official financing in subsequent years, theamounts involved were not intended—nor werethey sufficient—to pay off those countries’short-term foreign debts. The size of the Mexi-can package, by contrast, was meant to extin-guish the whole stock of tesobonos, to helpMexico cope with large dollars withdrawals [sic]from its banking system, and to help it rebuild itsreserves.

Of course, no one at either the IMF or the U.S.Treasury really thought that the package should beused to pay back the tesobonos. Only, the packagewas big enough to ensure holders of tesobonos thatthe Mexican authorities would have sufficient

money to honor their short-term debt, even if thatimplied redeeming it in full.

But why was COF not tried out for Mexico?11 Animportant chunk of the answer is probably politicalin nature (one could maliciously ask why a Bagehot-style operation of such a size proved so easy toarrange, after all). But this is hardly the whole story.One should add to it that until then no one had yetcontemplated the idea of using COF as a crisis man-agement tool while financial markets stayed open. Aswe have seen, the Baker Plan had been concocted inquite different circumstances, since all the crisiscountries had already suspended external debt ser-vice long before.

But the Mexican full-rescue package raised anumber of problems that would later generate theattempt to use catalytic lending as a tool for crisismanagement also for major emerging markets. First,the sheer size of the package was hard to replicate,should more than one country be hit by a crisis. Sec-ond, it generated moral hazard concerns. Third, andperhaps most important, the idea that taxpayers’money should be used to bail out private banks washard to sell. Thus, it is not surprising that immedi-ately after the announcement of the Mexican pro-gram, references to COF as a market-friendly crisismanagement tool became more frequent.

For example, in April 1995 the Interim Commit-tee of the IMF already refers explicitly to the “cat-alytic role” that the IMF could play in assistingmembers in coping with sudden market distur-bances. Later that year Michel Camdessus made itplain that the Mexican package was to be regardedas an exception:

We have agreed on procedures through whichthe Fund can respond rapidly—as it did in Mex-ico—to help put appropriate adjustment mea-sures in place and avoid spill-over effects. Wehave made clear, however, that the use of suchprocedures will be limited to truly exceptionalcircumstances. The extent of our support willdepend, as always, on the strength of the coun-try’s own adjustment effort, and we will take ap-propriate steps to ensure that its support remainscatalytic in nature. (Camdessus, 1995)

More extreme measures, reminiscent of the waythe developing country debt crisis had been handled,

11 The amount the IMF was willing to lend to Mexico was in-creased by US$10 billion in February 1995, after the initial fore-casts on the behavior of private investors had proved overly opti-mistic. Was this yet another example of a failed catalytic effect?Our reading of the episode is that the problems initially encoun-tered by the program had more to do with the uncertainty sur-rounding the amount of the U.S. contribution to the package(which turned out to be less than originally envisaged) than withthe program itself. See Boughton (1997) and De Long and Eichen-green (2001) for detailed accounts of the Mexican packages.

were also contemplated. Thus, at the Halifax sum-mit, in June 1995, the G-7 recommended an in-depth reflection on crisis management tools. Thetask was taken up by the Group of Ten (G-10),which set up a working group whose report (the so-called Rey report, named after the group’s chair-man) was published in May 1996. The report did ac-cept the importance of catalytic lending.12 But italso warned that, given the difficulty of arranging aconcerted package when a large share of “lending”takes place through bonds held by a myriad of in-vestors, a temporary suspension of debt payments (astandstill) might sometimes prove desirable, as away to gain time and better assess the sources ofmarket turbulence. The Rey report, however, metwith fierce opposition from the private sector. Acounter-report published by the Institute of Interna-tional Finance in September 1996 called its approach“misguided,” on the grounds that the approach wouldface enormous implementation problems and, if im-plemented, would fuel moral hazard on the debtorside. As a result, the report’s recommendations neverwon enough consensus to be adopted.

With the concerted approach increasingly re-garded as difficult to implement, and the standstillapproach facing strong opposition, by default the cat-alytic approach saw its appeal as a crisis managementtool unquestionably rise.13 The Asian crises provideda first testing ground for the new technique.

Several excellent descriptions of the Asian crisesare already available (see, for example, IMF, 2001a;and Kenen, 2001). Here, it will suffice to note thatCOF—with IMF resources provided in largeamounts, yet falling short of Bagehot-style LOLR fi-nancing—featured in all the main Asian crises, no-tably in Thailand, Indonesia, and the Republic ofKorea. Indeed, as noted by IMF staff, “most of thecapital account crisis programs [of the late 1990s]were predicated on just such a catalytic effect of offi-cial financing” (Ghosh and others, 2002, p. 8). As ithappens, the catalytic effect failed to materialize,leading to sharper domestic adjustment, further in-jection of IMF resources and, in some cases (such asthe Republic of Korea) the recourse to strongerforms of concerted lending.

Russia’s crisis, in the summer of 1998, althoughdifferent in both causes and ending, confirmed thatthe effectiveness of COF in the new environmentcould not be taken for granted. There, too, the exis-tence of an IMF-supported program failed to providesufficient confidence to markets. Yet, for want ofbetter alternatives, the official community contin-ued to bet on the idea. Thus, the report on interna-tional financial crises published by the so-calledGroup of Twenty-Two (1998), reiterates that “inmost cases when a country experiences payment dif-ficulties, the combination of adjustment and financ-ing of a typical IMF programme can be expected torestore market confidence and catalyze private capi-tal flows.”

While the new report was being published, a newcrisis broke out, this time in Brazil. The financialpackage the official community mounted this timetotaled about US$42 billion, almost half of whichcame from the IMF. The package was heavily front-loaded, as about 90 percent of the sum was to bemade available within 13 months. Both the size ofthe package and its time profile were clearly un-precedented. Yet, the package remained, accordingto our definition, catalytic in nature, since Brazil’sshort-term debt exceeded US$100 billion at thetime. After the announcement of the program, mar-kets stabilized for a while, but then, in late Decem-ber 1998, new pressures emerged in the foreign ex-change market. This time, financial conditionsimproved only after the floating of the real, and theannouncement of a rollover agreement between theBrazilian government and international banks. Suchdeals, contrary to the Republic of Korea’s, had notbeen actively promoted through moral suasion atthe international level, but this one proved success-ful. Brazil’s external problems subsided. Eventually,only a fraction of the US$42 billion package was ac-tually drawn. The whole episode was widely read asa success story for COF, although some commenta-tors invited caution in drawing lessons from it.14

After Brazil, official thinking on crisis manage-ment proceeded along two parallel paths. A firstpath led to giving center stage to the notion of cat-alytic lending, which was in fact enshrined in re-peated public statements. The other led to the no-tion of “private sector involvement” (or PSI, ageneral term encompassing a range of options, fromconcerted lending to mandatory solutions), whichclearly underscored the belief that catalytic lendingmight not be sufficient in a number of cases. Indeed,some debt-restructuring operations were activelypromoted, if somewhat implicitly, in a number ofsmall countries with limited financial ramifications.

209


12 For example, it noted that “the official community has sev-eral objectives, including … catalyzing finance in support of ad-justment efforts, when the latter are credible” and that “the maininstruments of the official sector are signaling confidence in thegood faith of the debtor and the economic soundness of its ad-justment programme and providing the prospects of limited fi-nance, subject to conditionality, to foster the resumption ofspontaneous inflows.”

13 At this juncture, references to catalytic lending can, oncemore, be found in the communiqués of the IMF Interim Com-mittee. The one issued on September 21, 1997 states that “theFund will continue to play a critical role in helping to mobilizefinancial support for members’ adjustment programs. In such en-deavors, the Fund will continue its central catalytic role whileminimizing moral hazard.”

14 De Long and Eichengreen (2001, p. 55), for example, re-mark that “the very different aftermath of Brazil’s crisis remainsone of those late-20th centuries mysteries to be unraveled by fu-ture historians.”

This two-pronged approach culminated in the“Prague Framework” issued at the 2000 IMF AnnualMeeting, which hinged around three propositions.The first simply reiterated the belief that, in somecases, catalytic lending would be sufficient to over-come a crisis. The second pointed to the usefulnessof voluntary PSI in many circumstances. The thirdannounced that “in exceptional cases, countriesmight impose capital or exchange controls as part ofpayments suspensions or standstills, in conjunctionwith IMF support for their policies and programmes,to provide time for an orderly debt restructuring.” Insuch a context, “IMF lending into arrears might beappropriate.”

In practice, however, experimentation continuedfor only the first two propositions of the PragueFramework. Thus, COF was again invoked, thistime coupled with PSI, in two prominent cases,Turkey and Argentina. Both countries had re-quested an IMF program in December of 1999, forprecautionary purposes in Argentina, and tostrengthen the credibility of disinflationary policiesin Turkey. Both programs initially worked, but thenfor a number of reasons, including policy slippages,financial conditions deteriorated. In line with thePrague Framework, in the course of 2000 the pro-grams were increased in size (in IMF technical jar-gon, they were “augmented”) and expanded to in-clude a PSI component. But this provided only abrief respite. Financial conditions stabilized inTurkey only after a further “augmentation,” at theend of 2001, brought financial support up to Bage-hotian levels, namely about 1,500 percent ofTurkey’s quota in the IMF. In Argentina, a thirdaugmentation, in August 2001, proved unable to re-store calmer financial conditions. The following De-cember, the authorities then declared a moratoriumon foreign debt to private creditors.

In light of these experiences, it is not surprisingthat the emphasis on COF as a crisis managementtool has more recently been declining. As Eichen-green (2001) puts it:

The main lesson of the Turkish and Argentineexperiences is that the markets are likely to dis-appoint official hopes. In the climate of uncer-tainty that invariably surrounds a crisis, waitinghas option value. Investors have an incentive towait and see whether the commitment to reformis sustained instead of being first to provide newmoney.

A Survey of the Empirical Evidence

The narrative in the previous section casts doubtson the magnitude of catalytic effects, at least forcapital account crises, but clearly more systematicevidence is needed to enable us to rule on this issue.We have surveyed 26 papers—to our knowledge the

whole empirical literature on catalytic effects. Theyare here classified in three branches—case studies,statistical studies of capital inflows, and statisticalstudies of interest rate spreads in international capi-tal markets—which will be reviewed in turn. Oneshould be aware before proceeding that severalmethodological difficulties make the measurementof catalytic effects particularly troublesome. Here isa tentative list of empirical hurdles.

• First of all, one must allow for the so-calledcounterfactual problem. The IMF seal of ap-proval may not raise capital inflows, but onecould argue that, in its absence, capital inflowsmight have been even weaker. This argument issometimes used to explain the absence of obvi-ous catalytic effects. Case studies are liable tothis criticism—much less so those studies thatmodel the behavior of capital movements,which are able to control for a variety of factorsbefore and after the crisis.

• Second, one should properly specify the rele-vant time horizon over which the catalytic ef-fect is expected to take place. Some studiesfocus on the short term, which is more relevantfor assessing the IMF role in stemming a suddenconfidence crisis. In such a context, however, itis critical to use high-frequency data, so as to beable to distinguish preprogram from postpro-gram observations. Failure to do so would biasthe results toward the rejection of the existenceof catalytic effects.

• Third, it may be difficult to distinguish betweendemand-driven and supply-driven capitalmovements. In particular, under a floating ex-change rate with no intervention (although notnecessarily under fixed exchange rates or in thepresence of intervention) a decline in the cur-rent account deficit—a frequent goal of IMF-supported programs—must be associated with adecline in net capital inflows (as the balance ofpayments is by definition in equilibrium). Thisagain may create a bias against finding catalyticeffects. The studies focusing on interest ratespreads, rather than on capital movements, are,however, not affected by this problem.

• Fourth, in assessing whether it is the IMF sealof approval that makes the difference, oneshould control for the adjustment in economicpolicies. A strengthening of investors’ confi-dence may be due to a policy change, ratherthan to the IMF seal of approval on those poli-cies. Thus, not controlling for policy changescreates a bias in favor of the existence of cat-alytic effects.

• Fifth, one should control for steps taken in support of catalytic programs. Some programswere supported by various forms of private sec-tor involvement, in which the private sectorwas nudged into action through more or less

210


211


formalized regulatory means. Failure to controlfor these supporting steps also creates a bias infavor of catalytic effects.

• Sixth, one should control for the purpose COFis used for (whether for crisis prevention, man-agement, or resolution), if truly operational im-plications are to be drawn from the analysis. Butclassifying programs according to purpose isclearly no straightforward matter.

These difficulties, as mentioned, affect to variousdegrees the three empirical methodologies that havebeen followed in the literature. Thus, by contrastingthe results obtained on the basis of each of them,one should be able to arrive at a better picture of theactual extent of catalytic effects. This is what wewill try to do in the rest of this section. Our readingof the evidence is that, while catalysis may be foundat work in quite specific circumstances, such as inprecautionary IMF programs or in the context of cri-sis resolution, overall its magnitude is small, andtends to become yet smaller as a country approachesthe crisis stage.

Case Studies

Few papers have focused on case studies. An earlyexample of the case study approach is Killick(1995), who discusses the experience with IMF-supported programs in 17 countries, concluding thatthere is no evidence of catalytic effects on privatecapital inflows.15

A comprehensive discussion of country experi-ences is also presented in Bird, Mori, and Rowlands

(2000).16 This paper looks at the experience of 17countries during the 1980s and 1990s and draws fourconclusions: (a) involvement by the multilateralswill not guarantee an inflow of capital from othersources; (b) what really matters is the perceivedcommitment by a government to a policy agendathat is seen as sound and internally consistent;(c) catalysis with the multilaterals is likely to bestronger and more positive in the case of bilateralaid flows; and (d) the nature and extent of catalysismay differ between the IMF and the World Bank,with the IMF having a stronger but negative effect.

A significant contribution to the case study ap-proach has recently been given by the IMF staff intheir study of IMF-supported programs in capital ac-count crises (Ghosh and others, 2002). This paperlooks at the experience of eight programs for largeemerging markets during the 1990s and pointsmostly to the absence of strong catalytic effects:

It is striking that, in every [emphasis in original]instance in our sample, the outcome [for privatecapital inflows] was worse than projected. Evenin cases where the magnitude of the error wassmall, such as Argentina (1995), the programdoes not appear to have had a strong catalytic ef-fect, at least in the very short run. (Ghosh andothers, 2002, p. 11)

The resulting financing gap had to be filled withlarger current account adjustment than originallyenvisaged. The amounts were not negligible, as Fig-ure 1 testifies. The discrepancy reached a peak in1998 (in the Republic of Korea and Thailand, whereactual current account adjustment was 13–15 per-centage points larger than planned), but remained

–0.2

1.40.7–0.9–0.5

1.12.92.7

1.5

7.1

0.4

14.9

5.9

14.4

5.76.5

–2

0

2

4

6

8

10

12

14

16

Mexico(1995)

Thailand(1997)

Korea(1998)

Indonesia(1998)

Thailand(1998)

Brazil(1999)

Turkey(2001)

Argentina(2001)

Program *

Outcome **

Figure 1

Current Account Adjustment in Selected Crisis Countries(As percentages of GDP)

16 The same results are also reported in Bird and Rowlands(2000).

Sources: IMF, World Economic Outlook database; Ghosh and others (2002).

15 Another early paper sometimes credited with reporting evi-dence of catalytic effects is de Vries (1986). It should be noted thatin the de Vries paper the term catalytic is used with reference to“concerted lending,” rather than to the meaning used in this paper.

sizable also afterwards (in Turkey and Argentina).The only exception to this pattern is Brazil in 1999,where the outturn was about 1 percentage pointbelow the planned figure.17

Effects on Capital Inflows

Several papers have focused on testing statisticallyhow the existence of an IMF-supported program, orthe size of IMF financing, has affected private capi-tal inflows. Various methodologies have been used,with the predominant result being that catalytic ef-fects, if any, are small. Three approaches have beenfollowed (Table 1).

Analysis of the demand for IMF credit. Several pa-pers have explored the issue of why countries needIMF credit. These papers do not typically focus onthe existence of catalytic effects, but are oftenquoted in this context because the right-hand sideof the econometric equations describing the de-mand for IMF credit usually include the amount ofprivate capital inflows. The sign of the coefficienton this variable is seen as having a bearing on thecatalytic issue: a positive sign would indicate com-plementarity between IMF credit and private credit(as one would expect when catalytic effects are inplace); a negative sign would indicate substitution.The first of these papers (Bird and Orme, 1981) doesfind a positive and significant correlation for a crosssection of 1976 programs, but no correlation for the1977 cross section. A positive correlation, albeit aweak one, is found also in Bird (1994; published alsoin Bird, 1995). Other papers (Cornelius, 1987;Joyce, 1992) find negative (and typically nonsignifi-cant) coefficients.18

Before-after tests. This approach focuses onwhether the inception of an IMF-supported programinvolves a statistically significant increase of capitalinflows (or of variables related to them, such as for-eign exchange reserves) with respect to the prepro-gram period. The first of these studies is the shortpaper (see the section entitled “The History ofCOF: A Tale of Two Crises) prepared by the U.S.Treasury in 1977 as background for the discussionon the Witteveen facility (United States, Depart-ment of the Treasury, 1977). This analysis of 36Stand-By Arrangements (SBAs) during 1970–75finds that in most cases SBAs were followed by anincrease in bank lending. Reichmann and Stillson(1978), although not directly aiming at measuringthe existence of catalytic effects, also find that netforeign assets typically increase as a result of IMF-supported programs, but note that in only one-fourth of their sample is this increase statistically

significant. Some evidence of catalytic effects inalso found by McCauley (1986): in only one-third ofthe programs considered in his sample did bank lend-ing decline in the year following the inception of aprogram. The relative frequency of outflows, how-ever, increases to 50 percent by the end of his sample(the year 1981). A more negative assessment of theexistence of catalytic effects can be found in Killick,Malik, and Manuel (1991).19 They find that, on atwo-year time horizon, the effect on private capitalinflow is negative or small: the improvement in theoverall balance of payments position, which they alsoobserve, is thus generally due to the improvement ofthe current account.

They also note that the programs do not affectnet inflows or gross flows much (there is no increasein gross disbursements). Another paper findingweak catalytic effects was prepared by IMF staff(Schadler and others, 1995). This paper, covering45 IMF loans from mid-1988 to mid-1991, finds thatthere were large increases in capital inflows in onlyone-third of the cases. A more sophisticated variantof this approach is found in Conway (1998), whoexplores whether the probability of a foreign ex-change crisis is reduced by the existence of an IMF-supported program. However, as the “crisis” is de-fined as foreign exchange reserves falling below acertain threshold, and as the behavior of foreign ex-change reserves is modeled through a simple time-series model, this approach essentially focuses againon whether on average the reserve position in-creases after the inception of a program. The resultsshow that reserves do increase when a program is inplace, but that this is no longer true for countriesunder a program for an extended period.

Capital inflows models. The before-after approachdoes not address the “counterfactual” problem. Asolution to this problem is to model directly the be-havior of capital inflows, and evaluate whether thisbehavior was changed by the existence of an IMF-supported program (or by the amount of financingprovided through them). Early results in this areawere obtained by Hajivassiliou (1987), who, whilenot focusing directly on the existence of catalytic ef-fects, does find a nonsignificant (and, indeed, nega-tive) effect of the lagged presence of an IMF pro-gram on the supply of new private loans. In thispaper, however, the dummy measuring the existenceof a program also measures the existence of a previ-ous rescheduling (being built to signal the existenceof previous payment difficulties). A more direct at-tempt to assess the existence of catalytic effects oncapital inflows is Rodrik (1996),20 who finds thatthe effect of IMF lending on net private capital

212


17 On the Brazilian crisis and its peculiarities, see the subsec-tion entitled “The Emerging Market Crisis, 1994–2002.”

18 Faini and others (1995) compute, outside a regression ap-proach, the correlation between capital inflows and the amountof IMF credit. They find a negative and significant correlation.

19 The same results are also reported in Killick (1995).20 Rowlands (1994) is an earlier contribution, but we could not

acquire a copy of this paper in time for the current draft. Accord-ing to Bird and Rowlands (1997), this paper concludes for theabsence of catalytic effects on private capital inflows.

213


Bird and Orme(1981)

Cornelius (1987)

Joyce (1992)

Bird (1994);results alsopublished in Bird(1995)

In 1976 significant positivecorrelation betweendrawings from the IMF andborrowing from euromarkets

Negative and nonsignificantcorrelation betweendrawings and borrowingfrom foreign andinternational capital markets

The commitment of newfunds from private creditorsrelative to imports has anegative impact on requestsfor IMF assistance, but it isnot significant.

Positive but weakcorrelation betweendrawings and netdisbursement of debt-creating private finance

Regression analysis

Regression analysis

Regression analysis

Regression analysis

What determines thedrawings by programcountries?

What determines the IMFdrawings?

What determines thedecision of a country toenter an IMF-supportedprogram?

What determines thedrawings in programcountries?

31 developingcountries that madedrawings in 1976and (an unspecifiednumber of countrieswith drawings in)1977 (annual data)

11 sub-Saharancountries, during1975–77 and1981–83

45 countries, annualdata for 1980–94

1980–85 for 27countries

I. Papers Estimating the Demand for IMF Credit

Main FindingsMethodology Question Sample

U.S. Treasury(1977)

Reichmann andStillson (1978)

McCauley (1986)

Killick, Malik, andManuel (1991);same resultsreported in Killick(1995)

Hajivassiliou(1987)

In most cases, bank lendingincreased.

Improvement observed intwo-thirds of cases, butstatistically significant only inless than one-fourth ofcases

Generally supportive of thecatalytic approach, with onlyone-third of countriesshowing a decline in privatelending. The frequency ofthese cases, however,increases over time (to halfby 1981).

Small effects and negativeon net capital accounts, withnet repayment of foreignloans and no increase indisbursements

There is a negative (althoughbarely significant) correlationbetween the supply of newloans and the laggedpresence of IMF support orrequest for a rescheduling.

Inspection ofchanges in capitalinflows

Mann-Whitney testfor difference infrequencydistribution

Before-after analysis

Before-after analysisof differences frombase (preprogramvalue); time horizon:four-year period

Regression analysis

Does bank lending toprogram countries increaseor decrease with theinception of a program?

Is the increase in netforeign assets observedstatistically significant?

Does a program affectbank lending to a countryin the year following theinception of a program?

Is the size of private long-term capital inflowsdifferent from thepreprogram period over afour-year horizon?

What determines thesupply of new loans?

36 programcountries during1970–75

85 programs during1963–72 (quarterlydata)

99 programscovering 56countries during1976–81

16 developingcountries with IMFprogramscommenced during1979–85

79 countries during1970–82, annualdata


Table 1

Private Capital Inflows and IMF Lending

II. Papers Based on Before-After Tests

214


Rodrik (1996)

Adji and others(1997)

Bird andRowlands (1997)

Bird andRowlands (2000)

Edwards (2000)

Marchesi (2001)

Schadler andothers (1995)

Conway (1998)

Effect of IMF lending notsignificantly different fromzero, negative in some cases

No significant effect of theamount of IMF lending onforeign direct investment

The catalytic effect onprivate flows is eithernegative (and significant) ornonsignificant; there is somepositive and significanteffect on official inflows, notprivate.

No evidence of a positivecatalytic effect and someevidence of a negative onefor the private inflows; some(weak) effect on bilateralinflows; no evidence thatthe catalytic effect hasstrengthened over time

No catalytic effect from theexistence of an IMFprogram, even in countrieswith a good record ofcompliance with pastprograms; negative effectsfor countries with IMFprograms that experiencedcompliance problems

The presence of an IMFprogram increases theprobability that a countrybenefits from debtrestructuring.

Large increases in capitalinflows in only one-third ofcases

Countries participating inIMF programs are less likelyto enter crisis, but thisbenefit is lost by countrieswhose IMF programscontinue for extendedperiods.

Regression analysis

Regression analysis

Regression analysis

Regression analysisand questionnaires

Regression analysis

Regression analysis

Inspection of capitalinflows

Regression analysis

What determines netprivate capital inflows?

What determines foreigndirect investment?

What determines newlending commitments (netof repayments)?

What affects private(portfolio, foreign directinvestment, and otherdebt) and bilateral capitalinflows?

Do private inflows(commitments) andforeign direct investmentreact positively to IMFprograms?

Does an IMF programfacilitate debtrescheduling bycommercial creditors?

What is the effect of aprogram on capitalinflows?

Is the probability of exitinga crisis (defined as adeviation of the reserve-to-import ratio from acertain threshold) affectedby the existence of an IMFprogram?

87 developingcountries, averagingsix-year periods(1970–93)

Annual data for 23developing countriesduring 1970–81

90 developingcountries, annualdata for 1974–89

96 developingcountries during1980–95, annualdata; 15 interviewswith financial marketmanagers

106 countries,annual data for1979–95

93 developingcountries, annualdata during 1983–95

45 IMF loans frommid-1988 to mid-1991

84 countries,quarterly data for1973–92

III. Papers Estimating Capital-Inflow Models


Table 1 (concluded)

215


inflows was not significantly different from zero andwas negative in many cases. No effect of IMF lend-ing on foreign direct investment is found by Adjiand others (1997). Similar conclusions werereached by Bird and Rowlands (1997 and 2000),using a more sophisticated capital inflows model, al-though the authors do find some evidence of cat-alytic effects on bilateral official inflows. Oneshould note that the dynamic specification of thesepapers is absent or very basic, and one wonderswhether the use of longer lags, and perhaps higher-frequency data, would have helped. The fact thatannual data do not reveal catalytic effects suggests,however, that if those effects are present, they donot show up very rapidly and are thus not very valu-able for crisis management.21 Edwards (2000) alsofails to identify positive catalytic effects, after con-trolling for the record of past implementation ofIMF programs: indeed, he finds that while a programdoes not have catalytic effects even following arecord of good program implementation, it doeshave negative catalytic effects if coupled with a re-cent history of nonperformance. The only paperthat does find catalytic effects on private capitalflows is Marchesi (2001). Marchesi focuses on a veryspecific form of such flows, however, namely the de-cision to reschedule existing external obligations: aprogram would raise the probability of a reschedul-ing. In principle, the decision to reschedule a loancan be seen as equivalent to granting a new loan—and therefore as providing a new capital inflow.However, the paper does not make allowance forrescheduling decisions that were part of concertedstrategies as distinct from decisions that were due togenuine catalytic effects. And there is no doubt thatthe IMF did play a major role in several cases of con-certed rescheduling (Milivojevic, 1986). Moreover,the paper does not control for the effect of a changein policies on the decision to reschedule; the latter,rather than the IMF seal of approval, may be respon-sible for the increased probability of a rescheduling.22

Effects on Spreads

One obstacle in estimating capital inflows equationsis that capital movements are typically very noisy. Inaddition, as noted, a decline in net capital inflowsmay merely reflect an improvement in the externalcurrent account. Thus, some papers have focused onhow the interest rates charged on a country’s debtare affected by a program (Table 2).

The first paper to include a dummy for the exis-tence of an IMF-supported program is Özler (1993).She finds that programs are associated with an in-crease in spreads, a result that is interpreted as indi-cating that IMF-supported programs signal (or areassociated with) payment difficulties. The effect isstatistically significant, albeit less so in specifica-tions including more variables. More recently a sim-ilar conclusion was reached by Haldane (1999) whofocuses on the behavior of secondary market spreadson the bonds issued by large emerging economies.He finds that a large spread persists after the initia-tion of a program and that, indeed, the spread “typi-cally widens further in the immediate aftermath of aprogram.”

Haldane’s paper, however, looks only at sevencountries in a relatively short time frame, and doesnot test statistically his conclusions, reaching themfrom visual inspection of the spreads. A more sys-tematic approach is followed in Eichengreen andMody (2000). This paper has two advantages overothers: it is based on a large sample and on high-frequency data (thus reducing the risk of averagingbetween before- and after-program data). The paperlooks at the spread at launch on bond issues duringthe 1990s. The strongest catalytic effects are foundfor EFF loans (Extended Fund Facility, which pro-vides the IMF’s longer-term loans) and, to a lesserextent, SBA loans (the traditional shorter-termloans) in equations when the number of quartersunder a program is also included with a positive co-efficient. This means that a program lowers thespread but the effect declines over time.

The authors conclude that their results “willhearten official observers concerned to documentthe ‘catalytic effect’ of IMF programs,” althoughthey also suggest caution about the type of programsthat exercise this effect.23 Two qualifications shouldbe made in interpreting these results. First, the re-gressions do not control for the change in economicpolicies that typically characterizes an IMF-supported program. This means that it is impossibleto distinguish between the decline in the spreadthat is due to sounder policies, and that arising fromthe IMF seal of approval: as noted, catalytic effects

21 It would also be necessary to control for simultaneity issues:the IMF may step in to stem capital outflows, but the use of an-nual data (mixing pre- as well as post-IMF intervention periods)may preclude the identification of the catalytic effect. However,Bird and Rowlands (2000) also present results based on a ques-tionnaire sent to financial managers. The sample would showthat financial managers do find that signing an agreement withthe IMF or the World Bank makes a country more attractive, butnone of them regarded IMF or World Bank involvement as oneof the five principal reasons for investing in a country. The sam-ple of respondents was, however, quite small (15 managers). Bird(1995, p.175, n. 51), also reports mixed evidence on catalytic ef-fect from interviews with bankers held during the 1980s.

22 One paper that does try to distinguish between the effect ofchanges in policies and the seal-of-approval effect (or “creden-tialing,” in the author’s terminology) is Conway (1994). Thispaper focuses directly on the effect of programs on economic per-formance, concluding that there is no evidence of the IMF’s sealof approval having improved performance, after controlling forpolicy changes.

23 Indeed, the authors find that Enhanced Structural Adjust-ment Facility (ESAF) loans (those addressed to poorer countries)do not have a significant effect on spreads (and the coefficient ispositive). Moreover, they find that the effect is significant onlyfor countries with an intermediate credit rating.

216


should refer only to the latter.24 Second, though sig-nificant, the effect of the programs remains small:the spread is lowered by some 9 percent, or 24 basispoints in the sample average.

This is not much compared with the averagevolatility of spreads in international markets. Simi-lar conclusions are reached in Mody and Saravia(2003), with the presence of a program loweringspreads on average by 8 percent. In addition, this

paper shows that the size of a program does matter:an additional program size of 10 percent of a coun-try’s long-term debt lowers spreads by about 20 per-cent. While this effect remains fairly low, it under-scores that what matters more than the seal ofapproval is the amount of financial support that theIMF makes available. A second interesting findingrelates to precautionary programs (those for whicheither the authorities declared that there was no in-tention of actually borrowing, or those for which noborrowing took place): the paper shows that thewhole negative effect on spreads comes from theseprograms. This, of course, may be tautological: if aprogram succeeds, this should be reflected in both adecline in spreads and the fact that the country does

Özler (1993)

Haldane (1999)

Eichengreen andMody (2000)

Mody and Saravia(2003)

The effect is positive (i.e.,the existence of a programraises the spread). The effectis statistically significant,although less so inspecifications with morevariables.

A large spread persists afterthe program initiation.Indeed, the spread “typicallywidens further in theimmediate aftermath of aprogram.”

Negative effect of theexistence of a program onthe spread. The effect isstatistically significant insome regressions, inparticular for EFF programsand, to a lesser extent, SBAs,but only in equationsincluding the duration ofprograms with a positivesign (i.e., the effect isnegative, but declines overtime). The size of the effectis small, however.

Broadly in line with theprevious paper, but, inaddition, the size of aprogram matters, with strongereffects for larger programs;precautionary programs arethose where the effect isstronger (and, indeed, thecatalytic effect does notmatter for nonprecautionaryprograms).

Regression analysis

Inspection ofspreads

Regression analysis

Regression analysis

Does the existence of anIMF-supported programaffect the spread overLIBOR on commercialbank loans?

Does a program affectspreads on secondarybond markets?

Does a program affect thespread at launch oninternational bonds?

Does a program affect thespread at launch oninternational bonds?

1,525 commercialbank loan contractsfor 26 countries over 1968–81

7 large emergingmarkets during1996–99

All bond issues byemerging marketsfrom 1991–I to1999–IV

All bond issues byemerging marketsfrom 1991–I to1999–IV


Table 2

Spreads and IMF Lending

24 Of course, it could be argued that, without the IMF, thosepolicies would not have been implemented, as policymakersneeded, for internal political reasons, an outside “policy en-forcer.” However, this points to an entirely different channelthrough which the IMF might have a role, even in the absence ofcatalytic effects.

Note: LIBOR denotes the London interbank offered rate.

217


not need to borrow from the IMF. In any case, thisresult suggests that if catalytic effects are present,they do not arise under crises situations (when pro-grams are not precautionary).

In Search of Theoretical Underpinnings

Where does all this leave us? None of the papers sur-veyed finds evidence of strong catalytic effects, al-though some of them do find the existence of mod-erate effects. Besides, this overall unfavorableoutcome has been reached in spite of the failure tocontrol for policy changes, which should have bi-ased results in favor of COF. Moreover, there are in-dications that to the extent that catalytic effects doexist, they are diminishing over time, are weaker for countries that went through a series of IMF-supported programs, and are weaker in crisis situa-tions.

Why are catalytic effects empirically weak? Isthere anything that the IMF could do to enhancethem? Has the trend toward increased capital mobil-ity affected their magnitude? Are any developmentsin the international financial markets likely to makethem stronger? Conversely, why should there be anycatalytic effect at all? Answering these questions re-quires discussing the analytical bases of the very no-tion of COF.

If IMF support to a country boosts the confidenceof private investors, it must be because the IMF issupplying something to the private sector, andsomething that, presumably, cannot be provided bymarket forces. An IMF program does indeed provide“services,” and, based on these services, one canidentify five channels through which the catalyticeffect, in principle, operates. While these channelsare not mutually exclusive, it is useful to discussthem separately, not only because their effectivenessdepends on different conditions but also becausethere are cases in which they may conflict with oneanother, thus weakening potential catalytic forces.

For each channel or service, we first identify somekey “requirements” (seven altogether) needed for itseffectiveness. Then, we assess whether these re-quirements are likely to be met under ordinary andextraordinary circumstances, and point out a num-ber of adverse interactions among the various condi-tions. Channels and requirements are summarized inTable 3.25

Five Channels for the Catalytic Effect

Channel I: Policy Design

Bird and Rowlands (1997) note that “for the [inter-national financial institutions] to have a possiblecatalytic effect there must be a presumption thateconomic policy will be better designed and moreappropriate to a country’s existing economic situa-tion and needs.”

The IMF—an agency specialized in macroeco-nomic adjustment policies—provides key advice inthe design of programs, thus supplementing thehuman resources of the program country. Indeed,IMF-supported programs are often perceived as hav-ing been designed by the IMF. The catalytic effectwould arise because the private sector would be reas-sured that, thanks to the “technical assistance” ofthe IMF, the best policy design has been adopted.

In principle, this service could be provided by pri-vate sector consultants. There are three reasons whythe IMF may have a comparative advantage in thisarea. First, as a public agency that member countriescommonly “own,” the IMF is in a better position toguarantee the confidentiality of (at least some) ofthe information on which the process of policy de-sign is based. Thus, it may be easier for the IMF toobtain the full collaboration of the government indisclosing all the relevant data. Second, it may alsobe politically more acceptable to receive advicefrom an international agency than from privateagents. Third, as a lending institution that puts itsown resources at risk and as administrator of condi-tionality, the IMF may have stronger incentives indelivering high-quality advice. That is, the uniquecombination of services it provides tends to give theIMF a competitive edge in the advisory business.

For this channel to operate effectively, two re-quirements must be met:26

• The IMF should have all necessary informationavailable to optimize the policy design: thereshould be no information asymmetries betweenthe program country and the IMF.

• The IMF should provide the best advice for thecountry, regardless of other considerations. Thismeans that the program should be the “best”

26 One additional condition, for this as well for other channelsto operate, is that the IMF be endowed with sufficient resourcesand skills to perform its tasks. Rodrik (1996) makes this assump-tion explicitly. Other authors have challenged it, blaming thefailure of some IMF-supported programs in restoring confidenceon errors in policy design, possibly because of poor judgment(see, for example, Stiglitz, 2000). There has been no systematicanalysis of whether the human resources available to the IMF areadequate to the task. Nevertheless, this line of criticism is notparticularly relevant from an analytical standpoint, as it points toa problem that, at least in principle, could be easily handled byenhancing the IMF’s human resources.

25 It has been argued (see, for example, Bird, 1994; Killick,1995; and Bird and Rowlands, 2001) that the catalytic effect failsto materialize because the record on IMF-supported programs inachieving their objectives is poor. Leaving aside the difficultissue of assessing the effectiveness of these programs, however,programs may not achieve their results precisely because the in-tended catalytic effects fail to materialize. Explaining this failurewith the programs’ alleged poor record thus involves a logical cir-cularity.

218


that can be designed under the circumstancesfor the specific country. It is important to stressthat the focus here is on the absence of constraints that might arise from the fact thatthe IMF is a multicountry institution and may,as such, have to take into consideration coststhat certain policies bring about for other countries. As we will see, problems of this typemay arise.

Channel II: Information

Rodrik (1996) notes the following:

Information about the broader investment envi-ronment and the quality of government policy-making is a public good: such information bene-fits all potential investors, regardless of specificprojects.… [I]n view of the public nature of thebenefit, individual investors have inadequate in-centives to devote resources to informationgathering of this particular kind and certainlylittle incentive to share with others the informa-tion they do gather.

Thus, it should be possible for the IMF to investmore resources on information gathering than theprivate sector and, consequently, to have an infor-

mation edge over the private sector. Consequently,private investors should be willing to follow the IMFwhen the latter provides a positive assessment ofmacroeconomic conditions (a catalytic effect).27

Playing this monitoring role, however, requires aset of stringent requirements (Table 3):

• The IMF should, not only in theory but also inpractice, have better information than do themarkets.

• The monitoring should be transparent. The IMFcould, of course, simply signal that the program ison track. However, a mere seal of approval notbacked up by additional information is notequally effective. On this basis, Rodrik (1996)and other IMF watchers have argued that theIMF should play this role more transparently.And the IMF has responded to this call, increas-ing its transparency.28

Table 3

Conditions for Existence of Catalytic Effects

(4) (6) (7) (2) No Conditionality Reasonable No Cost in Mechanism and

(1) Constraint Playing and Predictable Sufficient on Policy (3) This Role (5) Negotiation Costs for

Information1 Design Transparency Well Lending Process Slippages

Channel I(provision of assistance in policy design) � �

Channel II(provision of information) � � � �

Channel III (provision of commitment technology) � � � � �

Channel IV(provision of ascreening device) � � �

Channel V (provision of liquidity insurance) � �

1 This condition requires (a) for Channel II, that the IMF have better information than the market; and (b) for all other channels, that there beno information asymmetries between the IMF and the program country.

27 The critical role that the availability of adequate informa-tion to the IMF can play in enhancing catalytic forces is alsohighlighted in Corsetti, Guimarães, and Roubini (2003).

28 Until the mid-1990s, the provision of information by theIMF was limited to statements that the program was on track,occasional public statements by IMF staff and management, andoccasional contacts with the business community. The IMF iscurrently publishing virtually all letters of intent, as well as moststaff reports for program countries.

219


• The IMF should not suffer major costs in pro-viding information transparently, whether thisinformation is “good news” or “bad news.”

• The IMF should lend—monitoring can, ofcourse, be decoupled from lending. But, as Rodrik (1996) notes, the word of the IMF ac-quires more weight if it “puts its money where itsmouth is” (see also Masson and Mussa, 1995).29

Channel III: Commitment

This channel—perhaps the one that has attractedthe most attention until recently—focuses on themechanism of conditionality and its implications forthe government’s credibility in promising “good”policies. Conditionality in IMF-supported programsmeans that if these policies are not implemented theprogram country suffers an immediate cost.30 In en-tering into a program that involves a clear and im-mediate penalty for noncompliance, a governmentties its own hands, thus addressing the typical in-tertemporal consistency problem of economic poli-cymaking (Sachs, 1989; Diwan and Rodrik, 1992;Masson and Mussa, 1995; Rodrik, 1996; Dhonte,1997; Fischer, 1997; Cottarelli and Giannini, 1999;Dornbusch, 2001). A catalytic effect would in thiscase arise because the private sector would be reas-sured that policies will be implemented. The privatesector could not play this conditionality-centeredrole, mainly because direct negotiation with the private sector is politically unacceptable: a public,multilateral institution such as the IMF is needed(Rodrik, 1996; Masson and Mussa, 1995).

Several requirements need to be met for thischannel to operate smoothly. These can be moreeasily identified by interpreting the relation betweenthe IMF and the program country as a principal-agent relationship (Killick, 1997). The IMF is theprincipal that wants to see certain policies imple-mented because this is good not only for the pro-gram country but also for the international commu-nity. The program country is the agent that receivesa reward (IMF support) if the program is imple-mented. For a principal-agent relationship to workeffectively, some key requirements must be met:

• The IMF and the program country should sharethe same information, so as to allow effectivemonitoring of the agent’s policies.

• Lending (the reward for the agent) is needed asthis channel is based on the program condition-ality arising from lending.

• There should be a process for negotiations thatallows the definition of the “contract” betweenthe principal and the agent, as well as a processfor renegotiation to take new information intoaccount. This renegotiation is not a trivial as-pect of an IMF-supported program, which in-volves essentially iterative processes, sometimesbased on trial and error (Mussa and Savastano,1999).

• The penalty for the program country in not de-livering should be clear. It should also be “rea-sonable”: it would not make sense to sever theprincipal-agent relation (i.e., stop the program)for minor deviations in implementation.

• There should be no large cost for the principal(the IMF) when the agent (the country) doesnot deliver: this is typically the situation inmost principal-agent relationships in which theprincipal (say, the employer) has a large numberof agents (employees).31

Channel IV: Screening

Marchesi and Thomas (2001) argue that only coun-tries where the authorities’ policies are fundamen-tally good can afford sustaining the costs arisingfrom IMF policies. The authors do not say what thecosts of entering into an IMF-supported programare, but one can make the case that countries suffera short-run economic hardship under a program.Moreover, some measures are unpleasant to under-take, not least because they harm powerful politicalconstituencies. Only countries with viable long-term plans are willing to face these costs. Thus, en-tering a program helps the private sector to singleout (or to screen) the countries that (upon enteringa program) will follow good policies, a point alsomade by Fischer (1997).

What is peculiar, and paradoxical, about thischannel is that it does not require that the condi-tions “imposed” by the IMF be strictly necessary for the program’s success (or even that they be

29 Another requirement that is frequently mentioned is thatthe inception of an IMF program should not send a negative sig-nal. The provision of information should not be distorted by thevery fact that the country has entered an IMF-supported program.Many authors have noted that this may not be the case: the an-nouncement of a program may be interpreted as signaling that thecountry is in trouble, which may weaken any catalytic effects(Killick, 1995; Eichengreen and Mody, 2000; and Bird and Row-lands, 2001). We do not find this argument convincing, as typi-cally markets know already that a country is in trouble when itreaches the point of starting program discussions with the IMF.

30 Note that the interruption of a program involves differenttypes of costs for a country, ranging from the direct loss of IMF fi-nancial support, to political costs, to the loss of private sectorsupport (assuming the existence of catalytic effects).

31 It is worth noting that a principal-agent relationship willwork less effectively, the stronger are the differences in the utilityfunctions of principal and agent. With respect to IMF-supportedprograms, this means that a program works better if the programcountry owns the program. This has been long acknowledged.Some authors have raised the issue of whether the imposition ofprogram conditionality does not in itself signal lack of owner-ship. We do not see this as a major issue: a policymaker that tiesits own hands does not necessarily do it because the country doesnot own the program, but to reassure outsiders that the programwill in any case be implemented.

220


economically useful): as long as they are not plainlyinappropriate (which would signal only stupidityand thus reduce credibility), all that is required isthat they involve costs—and thus are “difficult” toimplement. While this is by no means the channelthat is most frequently mentioned, it is not, at ananalytical level, unreasonable, since it focuses on aform of signaling that is not uncommon in social be-havior (acting “tough” is a signaling device; see, forexample, Diamond, 1992). Critics of the IMF havealso sometimes argued that certain measures (in par-ticular those of a structural nature) are not clearlylinked to program goals. One oft-heard rebuttal ofthis criticism is that, while not directly linked, thesemeasures show the authorities’ resolve to undertakedifficult steps. This is the essence of the screeningchannel.

Although Marchesi and Thomas (2001) stress thedifference between this channel and the more tradi-tional commitment channel, the effectiveness of thescreening mechanism is enhanced by many of thesame factors required for Channel III to operate,namely those pertaining to the effectiveness of the“bargaining process” between the IMF and thecountry authorities, and to program monitoring: noinformation asymmetry, no costs to the IMF in en-forcing the program conditions, and the existence ofa negotiation/renegotiation process (see Table 3).

Channel V: Insurance

This is perhaps the trickiest channel of all, or atleast the channel that has been most affected by thetransition to capital account convertibility (and hasthus attracted more attention recently). Knowledgethat the IMF might step in should a country fallshort of external finance might act as implicit insur-ance for investors that their claims will be honored.Indeed, such a mechanism lies at the root of the oft-heard criticism that the possibility that a countrycould rely on an IMF program to stem a capital ac-count problem generates moral hazard—a typicalby-product of any form of insurance. Leaving asidethe issue of moral hazard, which need not concern ushere, the gist of the argument is that the very exis-tence of the IMF might reduce the probability of aself-fulfilling run caused purely by illiquidity prob-lems (Haldane, 1999; Zettelmeyer, 2000; Miller andZhang, 2000). The catalytic effect would arise fromthe provision to the private sector of the insuranceservice that is implicit in the LOLR function.

This service could in principle also be provided bythe private sector: contingent credit lines could bemade available by banks, against the payment of afee. However, the cases in which this has happenedso far are rare, and the results are hardly encourag-ing32—probably because it is hard to identify, or

even define, cases of pure liquidity crises, in whichthe line of credit would become available. The tradi-tional lending facilities of the IMF are not directly af-fected by this problem, because IMF programs alwaysinvolve elements of conditionality.33 It is thus some-what academic to discuss the “insurance” channel asa separate one. Nevertheless, for the sake of com-pleteness, it is worth pointing out the two require-ments that are needed for this channel to operate:

• Once again, there should be no asymmetric in-formation that would prevent the IMF from as-sessing whether the crisis is one of liquidity, orfrom requesting measures that would eliminateany solvency issues.

• Lending should be forthcoming in amounts suf-ficiently large to convince the private sectorthat the risk of being locked in is small.

Channel-Specific Problems

We now turn to the discussion of why some of therequirements summarized in Table 3 are unlikely tobe realized in a number of practical cases, and espe-cially under conditions prevailing in capital accountcrises. Four requirements may be particularly hard tomeet in some circumstances.

Requirement 1: Sufficient Information

This requirement takes two forms. The informationchannel requires that the IMF have better informa-tion than the market. All other channels requirethat there be no major information asymmetries be-tween the IMF and the program country. As to thefirst aspect, the process of financial consolidationand technical improvements in information tech-nology have reduced the private sector’s disadvan-tage in gathering and processing macroeconomic in-formation. Consequently, the resources invested inthese tasks by the private sector have increased andthe information edge of the IMF may have declinedin comparison with, say, the late 1970s. Note alsothat some information that becomes more critical incapital account crises (that is, that related to finan-cial transactions and intermediaries) may be moreeasily accessible to financial analysts than to theIMF. And concerning the assumption of no majorinformation asymmetries between the IMF and theprogram country, there are also some problems. TheIMF does not control directly the information pro-duced by the program country and has to rely on thedata provided by country authorities. Nevertheless,

33 This is not entirely true of the CCL, the facility establishedin 1999 to deal with “pure” liquidity problems, which, to someextent, worked on the basis of preselection. The CCL attractedno applicants, however, and was allowed to expire on its sched-uled sunset date in November 2003. See Giannini (2002) for acriticism of the CCL.

32 See Kenen (2001) for an account of the experience with pri-vate contingent credit lines.

221


while there have been cases of data misreporting, itis fair to say that, under ordinary conditions, coun-tries do provide timely and accurate information forprogram design and monitoring. Problems may,however, arise under crisis conditions, when eventsdevelop rapidly and the timeliness of the policy re-sponse becomes as important as the nature of the re-sponse. In these circumstances, complete sharing ofall relevant information between the IMF and coun-try authorities obviously becomes more difficult andinformation asymmetries may arise.

Requirement 2: No Constraint on Policy Design

The policy design channel (Channel I) assumes thatthe IMF can identify and promote the best programfor each country. Leaving aside the issue of possibleerrors in policy design, a more intrinsic problemarises from the very nature of the IMF as a multi-country agency. For lack of a better term, we willrefer to this problem as “policy contagion.” Supposethe best program that could be designed for a coun-try requires a measure that markets fear may also beintroduced in other countries. If the IMF had todeal only with one country, its unconstrained advicewould be to introduce that measure. However, theIMF may not go ahead because of the implicationsthat this may have for other countries. To make thecase more concrete, the measure in question may bethe introduction of some form of capital control orof debt restructuring. The fact that the IMF is sup-porting this measure in one country (where it couldbe implemented rapidly, thus minimizing disrup-tions) might be perceived by markets as indicatingthat the same approach would be followed for otherprogram countries. The latter perception could trig-ger speculative attacks. Fearing this, the IMF wouldstop short of including in a program a measure thatwould be optimal for that country. This is an ex-treme case, involving a strong (and very contagion-prone) measure. But the point is more general (it affects, for example, advice regarding the choice ofthe exchange rate regime). IMF advice to a countrydoes not take place in a vacuum, and this may con-strain its policy advice, particularly in crisis manage-ment packages where more radical (and contagion-prone) measures may be needed.

Requirement 5: Lending

Lending, of course, takes place (at least potentially)in all IMF programs. The real question, however, iswhether the “limited” lending that characterizedcatalytic programs is enough to raise the propensityof the private sector to invest in the program coun-try, particularly in a capital account crisis. This issueis particularly problematic for the insurance channel(Channel V), as highlighted by Zettelmeyer (2000),Eichengreen and Mody (2000), and Ghosh and

others (2002). In particular, the analytical model inZettelmeyer (2000) demonstrates that limited fi-nancing does not eliminate the risk for creditors ofbeing locked into pre-default situations; indeed,limited financing may prompt a speculative attack asinvestors take the opportunity of closing their posi-tions.

Requirement 7: Reasonable andPredictable Costs for Policy Slippages

The costs of noncompliance in the execution ofIMF-supported programs could be clearly defined inthe “old days.” Lack of implementation of criticalmeasures would cause the interruption of a programand of IMF financial support. The latter could be re-sumed after the implementation of corrective mea-sures. This is still the way things work in a numberof countries—those less involved in international fi-nancial markets, mostly pure cases of “external current account crises.” But in capital accountcrises, the penalty for noncompliance cannot beset a priori: it depends on the markets’ reaction.The latter may be disproportionately large owingto bandwagon effects. Consequently, the magni-tude of the penalty is also difficult to predict.34

The unpredictability of the penalty can have se-vere effects on the IMF’s willingness to promptlyreact to lack of compliance and, hence, on its cred-ibility and the strength of catalytic effects. Fearingan excessive market response, the IMF may not re-spond in a timely way to inappropriate policies,thus undermining its role both as provider of infor-mation (Channel II), provider of credibility (Chan-nel III), and provider of a screening device (Chan-nel IV).

Channel Inconsistencies

It would be somewhat reassuring if the seven re-quirements for effectiveness we have identifiedcould be shown to have no adverse interaction.However, this is not the case: some requirements,while making some channels stronger, tend toweaken the others. Thus, attempts to enhance onechannel, for example through changes in the waythe IMF operates, may yield ambiguous results over-all. There appear to be three main “channel incon-sistencies,” respectively associated with “conflicts ofinterest in blowing the whistle,” “transparency and

34 On this issue, Ghosh and others (2002) note,

There is no country that has not at some point of time ex-perienced hesitations and lapses in policy implementa-tion, mixed political signals, untimely release of bad news,and uncertainties in particular elements of financing. Thedifference is, in a capital account crisis, the country’s en-tire macroeconomic prospects may be hostage to suchevents—and the markets are unforgiving of any lapses.

the conditionality process,” and “conditionality andinsurance services.”35

Conflicts of Interest in Blowing the Whistle

The information channel (Channel II), the com-mitment channel (Channel (III), and the screeningchannel (Channel IV) all require that the IMF bean unbiased observer and that “blowing the whistle”not be costly for the IMF. However, when the IMF isinvolved in policy design (Channel I) and insurance(Channel V), blowing the whistle is costly.

Channels II, III, and IV require that the IMF playits role fairly: it should provide the relevant infor-mation to the market (Channel II) and enforce theimplementation of the program policies, interrupt-ing the program in case of noncompliance (Chan-nels III and IV). However, the larger the cost thatthe IMF faces when it blows the whistle, the largerwill be the disincentive in playing its role effectively.

Unfortunately, the IMF does face large costs inblowing the whistle. According to some IMF watch-ers, because of the IMF’s political nature—the IMFis administered by a board of Executive Directorswho are political appointees—these costs are pri-marily political. If this were the whole story, the so-lution would be to give the IMF more indepen-dence, as indeed is proposed in De Gregorio andothers (1999). But there are costs that arise from thevery function the IMF is required to play:

• No matter how much stress is put on programownership, IMF-supported programs are per-ceived as IMF programs. In a way, this is desir-able, as the fact that IMF expertise is used tobuild the program should support the catalyticeffect (Channel I). But this implies that theIMF suffers a cost in terms of reputation if theprogram does not work. Note that the cost isthere even if the program fails because of lack ofimplementation: the IMF could in this case beaccused of having supported a government thatdid not deserve credit, possibly because of a“bias toward lending.”36

• As mentioned earlier, stopping a program cantrigger a crisis, with major costs for the programcountry. A crisis involves costs for the IMF aswell—not only because the program country isan IMF member but also because of possiblecontagion effects to other IMF members. Thus,

from the perspective of an international agency,it may be preferable to wait to “blow the whis-tle” as this allows the agency to “buy time,” giv-ing financial markets time to differentiate andreducing, in this way, contagion effects. Thecost of this for the IMF is a weakening of itscredibility, and thus a weakening of catalytic ef-fects. Needless to say, this cost grows as morecountries are exposed to the risk of a capital ac-count crisis.37

Note also that these costs exist not only when a pro-gram is already in place but also when the IMF hasto decide whether to start a new program.

Transparency and the ConditionalityProcess

Channel II requires transparency, but transparencyenhances problems in the process of conditionalityand negotiation required by Channels III and IV.

As stressed by Mussa and Savastano (1999), theconditionality mechanism is not based on a “con-tract,” but is rather a “process”:

A typical IMF-supported program is not set instone at its inception, either proceeding subse-quently in exact accord with the initial plan, orterminated because of some minor deviation.[D]isbursements proceed automatically if all theperformance criteria are met as initially speci-fied. This rarely happens all the way through anarrangement. Instead, if various conditions arenot met, deviations may be accommodated with“waivers,” projections may be revised, and nu-merical targets changed. Those who participatein the process of IMF-supported programs, fromboth sides, do so with full awareness of their fun-damental iterative, open-loop character.

The “constructive imperfection” of this process isplain in these words. Economic policymaking is notan exact science. The process of steering an econ-omy back on track is a continuous one and may re-quire, for example, waiving some conditions thathad been initially regarded as critical, or, instead,halting the program if a sufficient number of condi-tions initially regarded as not critical are not met,or, again, changing conditions. In virtually all ofthese cases, judgment and negotiation will play acritical role.

222


35 Some of the problems highlighted in this section are also“channel specific,” i.e., would arise even in the absence of incon-sistencies. For example, this is true for some of the costs that theIMF suffers when it blows the whistle on a member country.

36 The existence of a conflict between monitoring activities (theso-called surveillance), on one side, and program design and negoti-ation, on the other, is probably at the root of the proposal byChancellor Gordon Brown (Brown, 2001) “to make the IMF’s sur-veillance and monitoring functions independent of the intergov-ernmental decisions about financial support for crisis resolution.”

37 Rodrik (1996) points to the fact that the IMF can no longerbe regarded as an unbiased observer merely because it lends (an-other form of channel inconsistency as, at the same time, lendingis needed for the working of all of the above channels; seeTable 3). It is not clear whether this is a severe problem, how-ever, taking into account that IMF lending has seniority over pri-vate sector lending, so that the IMF should not be concernedabout the effect that stopping a program would have on the like-lihood of recovering its investment.

Transparency, in particular the publication of let-ters of intent, even though critical to enhancing theinformation channel, makes this process more diffi-cult to manage. For example, failure to implement acertain measure may trigger a crisis, even when theIMF would have been willing to grant a waiver.Conversely, the granting of a waiver may be seen asa watering down of the program, which could beconstrued as being done for political reasons.

It may be argued that these problems could besolved through increased transparency. For example,the IMF could be more explicit or more prompt inclarifying what is critical and what is not, and underwhich circumstances it would be willing to support achange in policies. This is a naïve view. Not only isthe range of events that a program can face too largeto be subject to pre-specification, but, more funda-mentally, pre-specification would be inconsistentwith the negotiation aspects of an IMF-supportedprogram. Suppose financial markets reacted nega-tively to the lack of implementation of a measureincluded in a published letter of intent. If the IMFimmediately stated publicly its willingness to waivethe condition, so as to stem a speculative attack atits inception, it would lose any leverage in negotiat-ing remedial actions. Thus, in these circumstances,what the IMF might end up doing would be to pro-vide some halfhearted support, which would not befully convincing for the markets (it might indeed beeven counterproductive).

The critical point here is that the process of nego-tiation and conditionality requires time, and trans-parency, coupled with high capital mobility, dramat-ically cuts the time available for the process tocontinue effectively and to lead to an optimal re-design of programs as they are implemented. In thisenvironment, conditionality may become perverse:not only does it not provide credibility, it may endup being used as a coordinating device for specula-tive attacks, with program benchmarks and perfor-mance criteria inappropriately assuming the role oftriggering devices.38

Conditionality and Insurance Services

The commitment and screening channels presup-pose conditionality, and this creates uncertaintyabout the availability of IMF resources; but certaintythat sufficient resources will be made available iskey to the working of the insurance channel.

Channel IV requires that liquidity be available“with no strings attached.” Ever since Bagehot(1873), it has been recognized that the attempt totighten the type of “security” required for grantingliquidity support during a panic would have disas-

trous results. But this is inconsistent with condition-ality.39 It could be argued that this inconsistencycould be addressed by eliminating the conditionalityin pure “liquidity” crises. This is the logic underlyingthe recommendations of the Meltzer Commission:liquidity support should be restricted to countriesthat have prequalified by meeting a (limited) num-ber of performance criteria ahead of a crisis. There isa difficulty, though, in this logic, since countries,unlike the representative individual of most eco-nomic models, may be subject to sudden regimechanges. No matter how good the historical recordhas been (and typically for most IMF borrowers,“historical” means a decade or so), the domestic po-litical process may engender a coalition that is sim-ply “unwilling” to accept the burden of past policies.It is not accidental, in our view, that in most casescapital account crises burst out in the vicinity of amajor election, in the run-up to which one or moreof the candidates has won large popular support pre-cisely by promising a “regime change.” A measure ofconditionality, and of discretion by the crisis man-ager in assessing the credibility of the ruling coali-tion, is therefore inevitable. But the insurance ser-vice provided by an insurer that feels the need toinsure itself against the adverse event is unlikely tomeet with success in the eyes of private investors.

Conclusions

In this paper, we have reviewed the historical evolu-tion, empirical record, and theoretical underpin-nings of the notion of “catalytic official finance.”Our reading of the history of COF has yielded threeinsights.

First, the idea that official lending might play acatalytic role was recognized and accepted by IMFstaff before the generalized move toward capital ac-count liberalization that took place during the1980s. But in the era of capital account inconvert-ibility this was considered just a side effect of IMFprograms, not the main hinge around which IMF ac-tivity revolved.

Second, with the opening up of the capital ac-count, the emphasis on catalytic effects drasticallychanged. Attitudes toward COF appear to have fol-lowed a similar pattern in the course of the twomajor crisis episodes that have marked the lasttwenty years. In both the developing country debtcrisis of the 1980s and the capital account crisis of the1990s and early 2000s, COF emerged as the officialcommunity’s preferred option out of dissatisfaction

223


39 As remarked in the IMF staff ’s account of the recent crises,“the IMF’s own financing was phased and conditional on pro-gram implementation, implying that markets could not count onthe availability of this financing” (Ghosh and others, 2002). Thestudy mentions this as one of the main reasons why, in thosecases, the IMF program failed.

38 This is consistent with the empirical results in Edwards(2000) discussed earlier. The possibility of a negative catalytic ef-fect (triggering a crisis) related to IMF discussions is also men-tioned in IMF (1999, p. 35).

224


with the alternatives. In the 1980s, it was the ero-sion of the concerted approach that prompted cre-ation of the Baker Plan. In the 1990s, the proximatecause was the moral hazard concern generated bythe “Bagehotian” Mexican package. In both cases,private flows and official finance were postulated tobe dependable “bedfellows” only for want of a betteralternative. However, since more recently it eventu-ally became evident that private capital flows werereacting as “perfect strangers” to policy impulses, thefocus on catalytic lending as a tool to address capitalaccount crises has declined.

Third, the major difference between the develop-ing country debt crisis and the emerging marketscrisis, apart from the composition of external debt,is that in the former, crisis management took placein the context of a suspension of debt service,whereas in the latter the official packages’ primaryobjective was precisely to avoid such a suspension.This difference is responsible both for the greatervariety of recent packages (as they had to be fine-tuned to the circumstances of time and place) andfor the greater size of IMF lending and ultimate ex-posure (since the crisis package had to be bigenough to “impress” the market).

On the basis of this historical review, we thenturned to a more systematic evaluation of the empir-ical record of COF. Here the evidence appearsmixed, but consistent overall with the reading wehave proposed. All in all, catalytic effects do not ap-pear to be strong, which is all the more remarkablesince most empirical studies fail to control for actualpolicy change, thereby biasing results in favor ofCOF. Slightly stronger results tend to be obtainedwhen COF is used either in the context of precau-tionary programs (that is, for crisis prevention) or inassociation with debt rescheduling and/or restruc-turing (that is, for crisis resolution).

The theoretical review contained in the last partof the paper tried to make sense of this rather dismalrecord. It so happens that the conditions for havingstrong catalytic effects, always rather demanding,become yet more severe when COF is used as a crisismanagement tool. Reviewing theoretical rationalesfor COF, we have identified five channels throughwhich the announcement of an IMF program mightprop up the propensity of private investors to lendto a financially distressed country. We have labeledthese channels “policy design,” “information,”“commitment,” “screening,” and “insurance.” Thedependability of each channel appears weaker underthe conditions that typically accompany a capitalaccount crisis, impending or actual. The main prob-lems stem from (a) the costs that the IMF sufferswhen programs go off track (in terms of reputation,and of the risk of being accused of precipitating acrisis through the announcement that the programhas derailed); (b) the uncertainty regarding thepenalty for lack of compliance, which depends onthe markets’ reaction (an important preconditionfor the conditionality mechanism to work properly);

(c) the inadequacy as an insurance mechanism of aform of lending that is limited (compared with thepotential capital flight) and conditional; and (d) theasymmetric information between the IMF andcountry authorities.

On the basis of these considerations, we concludethat COF should be handled with great care. Its po-tential as a crisis management tool, in particular, ap-pears at best limited. This by no means implies that,if and when catalysis fails, IMF programs will likelybe ineffectual, because catalytic effects are not theonly channel through which programs work. Our ar-gument solely implies that one should not have cri-sis management programs that can be expected towork only if strong catalytic forces are at play.

BibliographyAdji, S.S., Y.S. Ahn, C.M. Holsey, and T.D. Willett,

1997, “Political Capacity, Macroeconomic Factors,and Capital Flows,” in Political Capacity and EconomicBehavior, ed. by Marina Arbetman and Jacek Kugler(Boulder, Colorado: Westview Press).

Bagehot, Walter, 1873, Lombard Street (London: Murray).Bird, Graham, 1994, “The Myths and Realities of IMF

Lending,” World Economy, Vol. 17 (September),pp. 759–78.

_______, 1995, IMF Lending to Developing Countries: Issues and Evidence (London: Routledge).

_______, Antonella Mori, and Dane Rowlands, 2000,“Do the Multilaterals Catalyze Other Capital Flows?A Case Study Analysis,” Third World Quarterly,Vol. 21, No. 3, pp. 483–503.

Bird, Graham, and Timothy Orme, 1981, “An Analysis ofDrawings on the International Monetary Fund byDeveloping Countries,” World Development, Vol. 9,No. 6, pp. 563–68.

Bird, Graham, and Dane Rowlands, 1997, “The CatalyticEffect of Lending by the International Financial In-stitutions,” World Economy (November), pp. 967–92.

_______, 2000, “The Catalyzing Role of Policy-BasedLending by the IMF and the World Bank: Fact or Fic-tion?” Journal of International Development, Vol. 12,pp. 951–73.

_______, 2001, “Catalysis or Direct Borrowing: The Roleof the IMF in Mobilizing Private Capital,” WorldEconomy (January), pp. 81–98.

Boughton, James M., 1997, “From Suez to Tequila: TheIMF as Crisis Manager,” IMF Working Paper 97/90(Washington: International Monetary Fund).

_______, 2000a, “Historical Perspectives on FinancialDistress: A Comment,” Carnegie-Rochester ConferenceSeries on Public Policy, Vol. 53, pp. 169–77.

_______, 2000b, “Northwest of Suez: The 1956 Crisis andthe IMF,” IMF Working Paper 00/192 (Washington:International Monetary Fund).

_______, 2001, Silent Revolution: The International Mone-tary Fund, 1979–1989 (Washington: InternationalMonetary Fund).

Brown, Gordon, 2001, speech given at the Federal Re-serve Bank of New York, November 16.

Buira, Ariel, 1987, “Adjustment with Growth and theRole of the IMF in the 1980s,” in The InternationalMonetary System and Its Reform, Paper Prepared for the

225


Group of Twenty-Four, ed. by Sidney Dell (Amster-dam: Elsevier Science).

Camdessus, Michel, 1995, “The International MonetaryFund and the Challenges of Globalization,” ad-dress given at the Free University, Amsterdam, November 28.

_______, 1999a, “Camdessus Outlines Role for PrivateSector in New Financial Architecture,” IMF Survey,Vol. 28 (March 8), p. 76.

_______, 1999b, “Report of the Managing Director to theInterim Committee on Progress in Strengthening theArchitecture of the International Financial System”(Washington: International Monetary Fund). Avail-able on the Web at http://www.imf.org/external/np/omd/1999/092499.htm.

Capie, Forrest, 2002, “Can There Be an InternationalLender-of-Last-Resort?” in Financial Crises, Conta-gion, and the Lender of Last Resort: A Reader, ed. byCharles Goodhart and Gerhard Illing (Oxford Uni-versity Press).

Cline, William R., 1989, “The Baker Plan and Brady Re-formulation: An Evaluation,” in Dealing with the DebtCrisis, ed. by Ishrat Husain and Ishac Diwan (Wash-ington: World Bank).

Conway, Patrick, 1994, “IMF Lending Programs: Partici-pation and Impact,” Journal of Development Econom-ics, Vol. 45, pp. 365–91.

_______, 1998, “IMF Programs and Economic Crisis: AnEmpirical Study of Transition” (unpublished; Wash-ington: World Bank).

Cornelius, Peter, 1987, “The Demand for IMF Credits bySub-Saharan African Countries,” Economics Letters,Vol. 23, pp. 99–102.

Corsetti, Giancarlo, Bernardo Guimarães, and NourielRoubini, 2003, “The Tradeoff Between an Interna-tional Lender of Last Resort to Deal with LiquidityCrisis and Moral Hazard Distortions. A Model of theIMF’s Catalytic Finance Approach” (unpublished;New York University).

Cottarelli, Carlo, and Curzio Giannini, 1997, CredibilityWithout Rules? Monetary Frameworks in the Post-Bretton Woods Era, IMF Occasional Paper No. 154(Washington: International Monetary Fund).

_______, 1999, “Inflation, Credibility, and the Role ofthe International Monetary Fund,” International Economic Policy Review.

De Gregorio, José, Barry Eichengreen, Takatoshi Ito, andCharles Wyplosz, 1999, An Independent and Account-able IMF (Geneva: International Center for Mone-tary and Banking Studies).

de Larosière, Jacques, 1980, “Managing Director TellsECOSOC He Foresees Sufficient Resources for Ex-panded Fund Recycling Role,” IMF Survey (July),pp. 202–206.

De Long, J. Bradford, and Barry Eichengreen, 2001, Be-tween Meltdown and Moral Hazard: The InternationalMonetary and Financial Policies of the Clinton Adminis-tration, paper presented at the conference on “TheEconomic Policies of the Clinton Administration,”Kennedy School of Government, held June 26–29 inCambridge, Massachusetts.

de Vries, Margaret Garritsen, 1985, The International Mon-etary Fund, 1972–1978: Cooperation on Trial, Vol. I(Washington: International Monetary Fund).

_______, 1986, “The Role of the International MonetaryFund in the World Debt Problem,” in World Debt Cri-

sis, ed. by Michael P. Claudon (Cambridge, Massa-chusetts: Ballinger).

Dhonte, Pierre, 1997, Conditionality as an Instrument ofBorrowed Credibility, IMF Paper on Policy Analysisand Assessment 97/2 (Washington: InternationalMonetary Fund).

Diamond, Jared M., 1992, The Third Chimpanzee: TheEvolution and the Future of the Human Animal (NewYork: Harper Collins).

Diwan, Ishac, and Dani Rodrik, 1992, External Debt, Adjustment, and Burden Sharing: A Unified Frame-work, Princeton Studies in International FinanceNo. 73 (Princeton, New Jersey: International FinanceSection, Department of Economics, Princeton Uni-versity).

Dooley, Michael P., 1994, “A Retrospective on the DebtCrisis,” NBER Working Paper No. 4963 (Cambridge,Massachusetts: National Bureau of Economic Re-search).

Dornbusch, Rudi, 2001, A Primer on Emerging MarketCrises (unpublished; Cambridge, Massachusetts:Massachusetts Institute of Technology, EconomicsDepartment).

Edwards, Martin S., 2000, “Reevaluating the Catalytic Effect of IMF Programs,” paper for presentation at theannual meetings of the American Political ScienceAssociation held in Washington in September.

Eichengreen, Barry, 2001, “Crisis Prevention and Man-agement: Any New Lessons from Argentina andTurkey?” (unpublished).

_______, and Ashoka Mody, 2000, “Bail-Ins, Bailouts,and Borrowing Costs,” paper presented at the FirstIMF Annual Research Conference held Novem-ber 9–10 in Washington.

Faini, Riccardo, Jaime de Melo, Abdel Senhadji-Semlali,and Julie Stanton, 1995, “Macro Performance UnderAdjustment Lending,” paper prepared for the WorldBank symposium on “Adjustment Lending and Be-yond,” held in Washington in April.

Financial Stability Forum, 2000, Report on Capital Flows(Basel: Bank for International Settlements).

Fischer, Stanley, 1997, “Applied Economics in Action:IMF Programs,” American Economic Review, Papersand Proceedings, Vol. 87 (May), pp. 23–27.

_______, 2000, Strengthening Crisis Prevention: The Role ofContingent Credit Lines, speech given at the Banco deMéxico in Mexico City on November 15.

_______, 2002, “On the Need for an International Lenderof Last Resort” in Financial Crises, Contagion, and theLender of Last Resort: A Reader, ed. by Charles Good-hart and Gerhard Illing (Oxford, Oxford UniversityPress).

Freixas, Xavier, Curzio Giannini, Glenn Hoggarth, andFarouk Soussa, 2000, “Lender of Last Resort: WhatHave We Learned Since Bagehot?” Journal of Finan-cial Services Research, Vol. 18, No. 1.


Giannini, Curzio, 1999, “Enemy of None but a CommonFriend of All?” An International Perspective on theLender-of-Last-Resort Function,” Princeton Essays inInternational Finance No. 214 (Princeton, New Jersey: International Finance Section, Department ofEconomics, Princeton University).

226


_______, 2002, “Pitfalls in International Crisis Lending,”in Financial Crises, Contagion, and the Lender of LastResort: A Reader, ed. by Charles Goodhart and Ger-hard Illing (Oxford University Press).

Gold, Joseph, 1979, Conditionality (Washington: Interna-tional Monetary Fund).

Goodhart, Charles, and Gerhard Illing, 2002, “Introduc-tion,” in Financial Crises, Contagion, and the Lender ofLast Resort: A Reader, ed. by Charles Goodhart andGerhard Illing (Oxford: Oxford University Press).

Group of Ten, 1996, The Resolution of Sovereign LiquidityCrises (the Rey report) (Basel).

Group of Twenty-Four, 2000, “Intergovernmental Groupof Twenty-Four on International Monetary Affairs Communiqué (April 15).” Available on the Web athttp://esa.un.org/ffd/policydb/PolicyTests/G24-1.htm.

Group of Twenty-Two, 1998, International Financial Crises(Washington).

Guitián, Manuel, 1982, “Economic Management and theInternational Monetary Fund Conditionality,” in Adjustment and Financing in the Developing World:The Role of the International Monetary Fund, ed. by Tony Killick (Washington: International Mone-tary Fund).

_______, 1992, The Unique Nature of the Responsibilities ofthe International Monetary Fund, IMF Pamphlet SeriesNo. 46 (Washington: International Monetary Fund).

Guth, Wilfried, 1979, “Sources of Finance for Develop-ment: The View of a Developed Country,” paper pre-sented at the 32nd International Banking SummerSchool, “The Financing of Long-Term Develop-ment,” held in August at St. John’s College, Cam-bridge, United Kingdom.

Hajivassiliou, Vassilis A., 1987, “The External Debt Re-payments Problems of LDC’s,” Journal of Economet-rics, Vol. 36, pp. 205–30.

Haldane, Andy, 1999, “Private Sector Involvement in Financial Crises: Analytics and Public Policy Ap-proaches,” Financial Stability Review, November.

Helleiner, Eric, 1994, States and the Reemergence of GlobalFinance (Ithaca, New York: Cornell University Press).

Institute of International Finance (IIF), 1996, ResolvingSovereign Financial Arrears (Washington: Institute ofInternational Finance).

International Monetary and Financial Committee (IMFC),2001, “Communiqué of the International Monetaryand Financial Committee of the Board of Governorsof the International Monetary Fund,” IMF Press Re-lease No. 01/20, April. Available on the Web athttp://www.imf.org/external/npsec/pr/2001/pr0120/htm.

International Monetary Fund (IMF), 1992, Press ReleaseNo. 2/181, November 9.

_______, 1999, “Involving the Private Sector in Fore-stalling and Resolving Financial Crises” (unpub-lished; Washington: Policy Development and ReviewDepartment, International Monetary Fund).

_______, 2001a, “Assessing the Determinants and Prospectsfor the Pace of Market Access by Countries Emergingfrom Crises—Country Cases” (Washington: Interna-tional Monetary Fund). Available on the Web athttp://www.imf.org/external/npsec/pr/2001/pr0120/htm.

_______, 2001b, “The IMF and the Private Sector,” available on the Web at http://www.imf.org/external/np/exr/facts/privsec.htm.

Joyce, Joseph P., 1992, “The Economic Characteristics ofIMF Program Countries,” Economics Letters, Vol. 38,pp. 237–42.

Kenen, Peter B., 1986, Financing, Adjustment, and the In-ternational Monetary Fund (Washington: BrookingsInstitution).

_______, 2001, The International Financial Architecture.What’s New? What’s Missing? (Washington: Institutefor International Economics).

Killick, Tony, 1995, IMF Programmes in Developing Coun-tries—Design and Impact (London: Routledge).

_______, 1997, “Principals, Agents, and the Failings ofConditionality,” Journal of International Development,Vol. 9, No. 4, p. 483.

_______, Moazzam Malik, and Marcus Manuel, 1991,“What Can We Know About the Effects of IMF Programmes?” ODI Working Paper No. 47 (London:Regent’s College, Overseas Development Institute).

Marchesi, Silvia, 2001, “Adoption of an IMF Programmeand Debt Rescheduling: An Empirical Analysis,” Development Studies Working Paper No. 152 (Uni-versity of Oxford/Queen Elizabeth House).

_______, and Jonathan P. Thomas, 2001, “IMF Condi-tionality as a Screening Device,” Economic Journal,Vol. 109 (March), pp. C111–C125.

Masson, Paul, R., and Michael Mussa, 1995, The Role ofthe IMF: Financing and Its Interactions with Adjustmentand Surveillance, IMF Pamphlet Series, No. 50(Washington: International Monetary Fund).

McCauley, R., 1986, “IMF: Managed Lending,” in WorldDebt Crisis: International Lending on Trial, ed. by M.Claudon (Cambridge, Massachusetts: Ballinger).

Milivojevic, Marko, 1986, The Debt Rescheduling Process(London: Frances Pinter).

Miller, Marcus, and Lei Zhang, 2000, “Sovereign LiquidityCrises: The Strategic Case for a Payment Standstill,”Economic Journal, Vol. 110 (January), pp. 335–62.

Mody, Ashoka, and Diego Saravia, 2003, “CatalyzingCapital Flows: Do IMF-Supported Programs Work asCommitment Devices?” IMF Working Paper 03/100(Washington: International Monetary Fund).


Niinistö, M. Sauli, 1999, “Statement by M. Sauli Niinistöin His Capacity as Chairman of the EU Presidency to the IMF and World Bank Annual Meetings in September 1999,” available on the Web athttp://www.imf.org/external/am/1999/icstate/eu.htm.

Özler, Sule, 1993, “Have Commercial Banks Ignored History?” American Economic Review, Vol. 83 (June),pp. 608–20.

Reichmann, Thomas M., and Richard T. Stillson, 1978,“Experience with Programs of Balance of PaymentsAdjustment: Stand-By Arrangements in the HigherTranches, 1963–72,” Staff Papers, International Mon-etary Fund, Vol. 25 (June), pp. 49–79.

Rodrik, Dani, 1996, “Why Is There Multilateral Lend-ing?” in Annual World Bank Conference on Develop-ment Economics, ed. by Michael Bruno and BorisPleskovic (Washington: World Bank).

Rowlands, Dane, 1994, “The Response of New Lending tothe IMF,” Norman Paterson School of InternationalAffairs Development Studies Working Paper No. 7(Ottawa: Carleton University).

Sachs, Jeffrey D., 1989, “Conditionality, Debt Relief, andthe Developing Country Debt Crisis,” in DevelopingCountry Debt and Economic Performance, Vol. I, ed. byJeffrey Sachs (University of Chicago Press).

227


Schadler, Susan, Adam Bennett, Maria Carkovic, LouisDicks-Mireaux, Mauro Mecagni, James H.J. Morsink,and Miguel A. Savastano, 1995, IMF Conditionality:Experience Under Stand-By and Extended Arrange-ments, IMF Occasional Paper No. 128 (Washington:International Monetary Fund).

Sgard, Jérôme, 2002, L’Économie de la Panique. Faire Faceaux Crises Financières (Paris: Éditions La Décou-verte).

Stiglitz, Joseph, 2000, “The Insider,” New Republic,April 17.

Sturc, Ernest, 1978, “The Role of the International Monetary Fund in International Development Fi-nancing,” paper presented at the World ScientificBanking Meeting, “International Financing of Eco-nomic Development,” held in Dubrovnik, Croatia,October–November 1977.

United States, Department of the Treasury, 1977, “Whythe Witteveen Facility Is Not a ‘Bail Out’ for theBanks,” in U.S. Participation in the Supplementary Financing Facility of the International Monetary Fund,

in United States, House of Representatives (1977),pp. 80–84.

United States, House of Representatives, 1977, U.S. Par-ticipation in the Supplementary Financing Facility of theInternational Monetary Fund, Hearings Before theSubcommittee on International Trade, Investment,and Monetary Policy of the Committee on Banking,Finance and Urban Affairs, House of Representatives,Ninety-Fifth Congress, First Session, on H.R. 9214(Washington: U.S. Government Printing Office).

Volcker, Paul A., and Toyoo Gyohten, 1992, ChangingFortunes: The World’s Money and the Threat to Ameri-can Leadership (New York: Times Books).

Zettelmeyer, Jeromin, 2000, “Can Official Crisis LendingBe Counterproductive in the Short-Run?” EconomicNotes, Vol. 29, No.1.

228

13

In this chapter, we examine the IMF’s role in maintainingthe access of emerging market economies to internationalcapital markets. We find evidence that both macroeco-nomic aggregates and capital flows improve following theadoption of an IMF program, although they may initiallydeteriorate somewhat. Consistent with theoretical pre-dictions and earlier empirical findings, we find that IMFprograms are most successful in improving capital flowsto countries with bad, but not very bad fundamentals. Insuch countries, IMF programs are also associated withimprovements in the fundamentals themselves.

Introduction

Recent research on the role of the InternationalMonetary Fund can, broadly speaking, be dividedinto two strands. Studies focusing on the immediateresponse of the IMF to financial crises—the firststrand of literature—have assessed whether IMF-supported programs stemmed the crises, preventedcontagion, and helped countries regain access to in-ternational capital markets—without inducing theharmful side effect of imprudent lending by interna-tional creditors.2 Others—constituting the secondstrand—have been concerned with the longer-termmacroeconomic implications of IMF programs.3

Future research on the role of the IMF faces threechallenges. First, the two separate strands of litera-ture need to be brought closer together. In particu-lar, the following question needs to be asked: doIMF programs serve primarily as a source of emer-gency finance or do they, by improving macroeco-

nomic fundamentals, also help to ensure longer-term access to international capital markets? Second,a greater appreciation of differences in countries’initial economic conditions is necessary to under-stand the effects of IMF intervention. Initial condi-tions influence subsequent dynamics, as well asstrategic behavior of the country authorities and in-ternational creditors. Third, the channels—ormechanisms—through which the IMF is able to in-fluence economic outcomes need to be identified forthe forward-looking design of IMF policy and theinternational financial architecture.

This paper presents an initial investigation of thisambitious agenda by focusing on the relationshipbetween IMF programs and private internationalcapital flows. This is a natural setup to address thefirst challenge—bringing the two strands of litera-ture closer together—since, to the extent that IMFprograms succeed in keeping capital flowing, theycan help to both contain financial crises and stimu-late longer-term investment and growth. To take onthe second challenge, we examine the possibility ofa nonlinear response of capital flows to initial condi-tions. Finally, we deal with the third challenge byidentifying the main mechanisms that allow the IMFto maintain capital flows. We argue that a “goodhousekeeping seal of approval” may not be operative,leaving in contention the “delegated monitoring”and “catalytic lending” roles of the IMF.4

The IMF itself has increasingly viewed the stabi-lization and fostering of capital flows as crucial to itsevolving role. In the 1990s, private capital flows todeveloping economies resumed after the hiatus fol-lowing the debt crises of the 1980s. This resumptionin capital flows represented both an opportunity toaccelerate economic development and a threat insituations where a “sudden stop” of capital flows wasassociated with a sharp contraction in economic ac-tivity. As the fluctuations in countries’ capital ac-counts acquired greater importance—and sometimeseven dominated current account fluctuations—theIMF increasingly came to view the facilitation andmaintenance of capital flows to developing countriesas one of its essential functions. A clear indicationof this was the formation of the IMF’s International

Keeping Capital Flowing: The Role of the IMF Michael Bordo, Ashoka Mody, and Nienke Oomes1

1 This paper was previously published in 2004 in InternationalFinance, Vol. 7, No. 3, pp. 421–50 and appears here with the per-mission of Blackwell Publishing, the journal’s publisher. It hasalso been issued as IMF Working Paper 04/197. The authors aregrateful to Barry Eichengreen, Stephen Morris, two anonymousreferees, and reviewers from the IMF’s Policy Development andReview Department for very helpful comments on earlier drafts.They are also indebted to Tom Walter for excellent editorialcomments and to Adrián de la Garza and Wellian Wiranto forvaluable research assistance. The authors are, however, solely re-sponsible for any remaining errors.

2 See, for example, Boorman and Allen (2000), Bordo andSchwartz (2000), Meltzer (2002), and Ghosh and others (2002).

3 An early survey of this literature was carried out by Haqueand Khan (1998). More recent articles on this subject includePrzeworski and Vreeland (2000), Hutchison (2001), Barro andLee (2003), and Hajro and Joyce (2004).

4 Cottarelli and Giannini (2002) discuss in greater detail theIMF’s channels of influence.

CHAPTER

229

Bordo, Mody, and Oomes

Capital Markets Department in 2001, which was in-troduced by former Managing Director Horst Köhleras follows: “Because private capital flows are an in-dispensable source of financing for development,another crucial function of the IMF’s new CapitalMarkets Department will be to strengthen our abil-ity to help countries gain access to internationalcapital markets” (Köhler, 2001, paragraph 13).

The main goal of this paper is to study whetherIMF programs have indeed helped countries gain orregain access to international capital markets and, ifso, through which channels. Since we focus on capi-tal flows, we limit our empirical analysis to 29emerging market economies.5 We thus exclude IMFprograms supported by the low-interest Poverty Re-duction and Growth Facility (PRGF), which are tar-geted at low-income countries. We cover the period1980–2002, during which these 29 countries con-tracted 119 programs with the IMF. The unit ofanalysis is a program-country pair. Since the numberof observations declines considerably as we controlfor country conditions, we mainly present descrip-tive evidence on the effects of IMF programs on ag-gregate performance (growth, inflation, the currentaccount, and capital flows).

A key issue in empirical analysis of IMF programsis the characterization of the appropriate counter-factual.6 That is, in addition to studying how eco-nomic conditions changed after the adoption of anIMF program, it is necessary to benchmark thosechanges against the evolution that would have oc-curred in the absence of an IMF program.

We proceed in three steps. As a first step, we com-pare the average macroeconomic performance ofemerging market countries before and after IMFprogram initiation. The results suggest that macro-economic aggregates—GDP growth, inflation, andthe current account—and capital flows improve fol-lowing the adoption of an IMF program, albeit oftenwith a lag. In particular, we find that GDP growthrates and capital flows display a “dip and recovery”pattern—that is, performance deteriorates some-what in the first year of the program but improvessignificantly in the second year.

We are careful not to commit the post hoc, ergopropter hoc fallacy. That is, from the observation thatperformance improved after adoption of the IMFprogram, we cannot necessarily conclude that per-formance improved because of program adoption. Infact, one would expect a certain amount of selection

bias, in that countries that adopt IMF programs arelikely to have worse initial conditions than countriesthat do not adopt IMF programs (i.e., selection intoIMF programs is not random). Given such selectionbias, it would be quite natural to observe a recoveryfor the group of program countries, simply because ofmean reversion, that is, an inbuilt tendency for im-provement in performance following a setback. Inorder to reduce this selection bias, one should com-pare the performance of program and nonprogramcountries with similar initial conditions.

Before doing that, we investigate, as a second step,the role of initial conditions. In particular, we exam-ine whether the response of capital flows depends onthe state of a country’s external fundamentals justprior to the start of an IMF program. We show thatthe improvement in aggregate performance follow-ing program adoption reflects primarily success inmaintaining capital flows to countries with interme-diate initial external fundamentals. That is, in mea-suring fundamentals in terms of four states (verybad, bad, good, and very good), we find that IMFprograms are more effective in countries with bad,but not very bad, fundamentals. Interestingly, thisfinding is consistent with the empirical results inMody and Saravi (2003), as well as with the predic-tions of recent models of catalytic lending (Morrisand Shin, 2003; and Corsetti, Guimarães, andRoubini, 2003) and of the IMF’s delegated monitor-ing role (Tirole, 2002). The results are generally robustto using four different external fundamentals: the cur-rent account balance as a share of GDP, the reserves-to-imports ratio, the ratio of short-term debt to re-serves, and the ratio of total external debt to GDP.

Finally, in our most ambitious attempt at provid-ing a counterfactual, we compare the performancesof countries that adopted IMF programs with thoseof countries that did not adopt IMF programs, butthat had similar initial conditions. We do this by es-timating the probabilities of transitioning betweenthe four different states and by comparing the esti-mated transition probabilities of program countrieswith those of nonprogram countries. For most initialconditions, except very good ones, we are able to re-ject the null hypothesis that nonprogram countriesperform similarly, given similar initial conditions.That is, we show that the observed increase in capitalflows to program countries with intermediate initialfundamentals is accompanied by an improvement inthe fundamentals themselves, an improvement thatis not observed for nonprogram countries with simi-lar intermediate initial fundamentals.7

5 Our definition of “emerging market economies” is the defini-tion used by the IMF’s International Capital Markets Depart-ment for the purpose of producing “early warning signals,” andincludes the following 29 countries: Argentina, Bolivia, Brazil,Chile, Colombia, Cyprus, the Czech Republic, Egypt, Hungary,India, Indonesia, Israel, Jordan, the Republic of Korea, Lebanon,Malaysia, Mexico, Pakistan, Peru, the Philippines, Poland, theSlovak Republic, South Africa, Sri Lanka, Thailand, Turkey,Uruguay, República Bolivariana de Venezuela, and Zimbabwe.

6 For early discussions of counterfactuals in the context of IMFprograms, see Goldstein and Montiel (1986); and Khan (1990).

7 Goldstein and Montiel (1986) propose simulating a policy re-sponse to a crisis without IMF presence and, further, an outputresponse to policy to generate a counterfactual. Estimating suchfunctions can be complicated and the results unreliable. An al-ternative is to run regressions while controlling for factors likelyto explain macroeconomic performance and to include a dummyvariable for an IMF program. For recent examples, see Barro andLee (2003) for the analysis of growth; and Mody and Saravia(2003) for the analysis of bond spreads.

230

13 � KEEPING CAPITAL FLOWING: THE ROLE OF THE IMF

We conclude that countries in severe distress (i.e.,in a very bad state) are unable to revive their capitalflows under an IMF program. Countries with goodinitial conditions possibly benefit from IMF pro-grams by maintaining relatively stable capital flowsafter being hit by temporary liquidity shocks. How-ever, IMF programs play their most important rolewhen countries are in an intermediate regime (withbad, but not very bad, initial conditions). In sucheconomies, capital flows and external fundamentalsimprove over a two- or three-year time period fol-lowing the start of the program.

The remainder of the paper is organized as follows.We begin with a discussion of the three channelsthrough which the IMF can help maintain capitalflows. Next, we describe the evolution of macroeco-nomic performance—GDP growth, inflation, thecurrent account balance, and capital flows—follow-ing the commencement of an IMF program. We thendisaggregate the response in capital flows by thestate of a country’s initial external fundamentals.Next, we examine whether IMF programs influencethe transition probabilities across states. We synthe-size the principal results in the concluding section.

The Role of the IMF in Facilitating Capital Flows

Under the Bretton Woods par value system, charac-terized by pervasive capital controls, the IMF wasprimarily concerned with surveillance of the ex-change rate system and provision of temporary fi-nance to manage current account deficits. Followingthe breakdown of the Bretton Woods system in1973, the IMF has faced a new environment encom-passing a significant increase in its membership(both developing countries and transition coun-tries), a myriad of exchange rate arrangements, and,especially, new problems arising from open capitalaccounts and financial globalization. With theopening of international capital markets to ad-vanced countries in the 1970s and to emerging mar-kets in the 1980s, the focus of the IMF has shiftedtoward new market failures and the new publicgoods needed to maintain financial stability.8

How can the IMF add value in this new environ-ment? We begin with the premise that maintainingthe flow of international capital is valuable to thecountries that receive those flows, as well as to theinternational financial system as a whole. Capitalflows, as the evidence of the 1990s amply bears out,are subject to sudden stops and sharp reversals(Calvo, 1998). The IMF can be particularly impor-tant to the international financial and economic

order if it can mitigate these stops and reversals and,thereby, maintain the flow of capital even when acountry is vulnerable to a loss of confidence.

This section discusses three possible channelsthrough which the IMF can help maintain capitalflows (for a more extensive discussion, see Cottarelliand Giannini, 2002). First, if the IMF has informationthat the private sector does not have, it can provide avaluable signal that can act as a good housekeepingseal of approval. Second, when the IMF does nothave an informational advantage but a country lackscredibility in being able to honor its external debtobligations, the IMF can act as a delegated monitor.Finally, catalytic lending by the IMF can inducelenders to roll over their credits and, hence, preventan exodus of capital from the country. However, IMFlending could also encourage moral hazard, inducingprivate lenders to be less careful in their credit deci-sions in the expectation of being bailed out.

In the rest of this section, we examine in more de-tail these three channels through which the IMF canredirect capital flows to an emerging market economyat risk of losing access. We analyze the logic of eachchannel and highlight the likely observable implica-tions of influence exercised through each channel.

Good Housekeeping Seal of Approval

A good housekeeping seal of approval from a re-spectable institution can facilitate market access foremerging country borrowers. It can act as a signalthat members have sound financial institutions andfollow sensible policies. The credibility of such aseal of approval depends, however, on an informa-tional advantage (Rodrik, 1996). For the IMF toprovide such a seal, it should possess superior infor-mation and be able to communicate its assessmentin a transparent manner.

An earlier successful example of an internationalinstitution serving as a seal of approval was the clas-sical gold standard in the pre-1914 era of financialglobalization. Under the institution of the gold stan-dard, adherence to gold convertibility served as asignal to international lenders that a country hadhealthy financial institutions and followed sound fi-nancial policies (Bordo and Rockoff, 1996). It alsoserved as a credible commitment mechanism, inthat gold adherence required that countries keep in-flation low and budgets balanced (Bordo and Kyd-land, 1995 and 1996). Countries that successfullyadhered to gold convertibility were able to borrowfrom London at rates significantly lower than thosethat did not (Bordo and Rockoff, 1996; and Obst-feld and Taylor, 2003).9

8 Bordo and James (2000) refer to several such public goods,including short-to-medium-term capital for credit-constrainedcountries, coordination of lenders in debt crises, internationalliquidity in financial crises, and “early warning signals.”

9 Other institutional factors may have also served as a goodhousekeeping seal, such as being part of the British empire (Fer-guson, 2002). However, Obstfeld and Taylor (2003) report evi-dence that the gold standard played a significantly more promi-nent role.

231


Does the gold standard have resonance for today?There are several key differences between the opera-tion of the gold standard and of the IMF. The goldstandard, unlike the IMF, was an informal institu-tion that had evolved from centuries of market ex-perience. As such, “membership” in the standardwas at the initiative of the individual countries, andcountry policies would determine if a country wason or off the standard.

When international financial crises occurredunder the gold standard, as they frequently did, do-mestic lenders of last resort, individually and coop-eratively, provided emergency liquidity. At the sametime, financial institutions adopted preventive mea-sures, such as maintaining high capital and liquidityratios (Bordo, 2003). Moreover, the advanced(core) countries could temporarily leave the goldstandard during a crisis or a war to follow counter-cyclical policies—the belief that they were credibleadherents of the gold standard allowed this depar-ture from “the rules of the game.”

Even at that time, the emerging countries faced adifferent environment. In the face of shocks (termsof trade and political instability), they often left thegold standard, devalued their currencies, and didnot return except at a devalued parity. They also oc-casionally defaulted on their debt (Reinhart, Rogoff,and Savastano, 2003). Because of their inability toadhere consistently to the gold standard, they paid asignificant currency risk premium and lost the sealof approval.

Thus, the seal of approval worked because a set ofcredible policies and institutions backed up the ad-herents of the gold standard. Where gold was merelya temporary prop, its seal was of little value.

It is unclear whether the seal of approval is an im-portant mechanism through which the IMF helpsmaintain capital flows, because the IMF’s informa-tional advantage over the private sector appears tohave declined. First, transparency has increasedboth at the IMF, which now makes most of its dataand reports publicly available on its website, and inmany emerging market countries, especially follow-ing the Asian crisis. Second, while in some in-stances IMF reports remain unpublished at the re-quest of the member country, such “privilegedinformation” tends to convey economic concernsand vulnerabilities that could scarcely constitutethe basis for a positive signal. Third, there is no evi-dence that the IMF’s economic forecasts are superiorto those of other institutions. For example, Loun-gani (2001) finds that IMF forecasts of economicgrowth are no better—though typically no worse ei-ther—than those of the private sector. Finally, theremay be some concern that IMF signals are influ-enced by political considerations and, hence, not re-liable.

Nevertheless, there are some reasons to assumethat the IMF still has an informational edge overthe private sector. Disparate market actors may be

better at assessing and synthesizing data than theIMF, but it is not obvious that they have better in-formation than the IMF. For example, an invest-ment bank research team typically has one analystcovering multiple small countries, while the IMF usually has multiple economists covering evensmall countries. In addition, one could argue thatthe IMF is better placed to verify data and to make judgments regarding the intentions of the au-thorities.

Even if IMF surveillance could constitute a seal ofapproval, it is difficult to develop an empirical strat-egy through which the importance of the IMF’s in-formational advantage can be inferred. The IMF, inprinciple, monitors all countries and the worldeconomy on an ongoing basis, and virtually allcountries have been members of the IMF for a longtime. As such, creating a counterfactual for inferringan IMF informational advantage is not straightfor-ward. There exists a better possibility of identifyingthe effects of IMF programs, conditional on the in-formation that the IMF surveillance process gener-ates. We thus turn to the effects of IMF programs,through which the IMF plays the roles of delegatedmonitor and lender.

Delegated Monitor

While the IMF may not have an informational ad-vantage over private creditors and investors, it canplay the role of a monitor in situations where gov-ernments have more information about their eco-nomic condition and their intentions than the restof the world. In such situations of asymmetric infor-mation, problems related to contract enforcementhave impeded the free movement of financial capi-tal, resulting in frequent financial crises character-ized by sudden stops in foreign lending and, on oc-casion, debt defaults. The IMF can serve as adelegated monitor (Tirole, 2002), that is, it cancomplete contracts between international lendersand both sovereign and private borrowers in emerg-ing market economies through its lending condi-tionality.

According to Tirole (2002), the market failure ininternational financial markets arises from twoproblems identified in the corporate finance litera-ture: the dual and common agency problems. Thedual agency problem occurs because emerging marketgovernments are potential spoilers of internationalfinancial arrangements between foreign lenders andprivate domestic borrowers. A contract between anadvanced country lender and an emerging countryprivate borrower can be threatened ex post by gov-ernment policies such as devaluation, capital levy,or opportunistic behavior in the political interest ofthe decision makers. These actions will reduce theex post returns received by the lender, which, inturn, will discourage future lending. A delegatedmonitor, such as the IMF, can reduce this problem

232


by constraining opportunistic behavior throughconditionality imposed ex post.10

The common agency problem arises when an emerg-ing country borrows from a myriad of lenders; whenbooking the loans, these lenders do not pay adequateattention to the loans already issued, which leads tooverborrowing. In the face of a crisis, the lenders rushto the exits to avoid being caught by the default, andthus ensure its eventual occurrence. A delegatedmonitor can serve to coordinate lenders and preventthe overborrowing problem or, after a crisis has oc-curred, arrange an orderly workout. These actions canreduce the disruption in access to capital markets.

While the delegated monitoring role can followfrom basic IMF membership (which serves as a com-mitment to follow certain rules of conduct)11 andIMF surveillance (using the information gatheredfrom members via Article IV consultations), itcomes into play especially in the context of an IMFprogram. Mere membership and surveillance do notinduce implementation of prudent policies and fi-nancial stability. In contrast, an IMF program, withthe carrot of access to financial resources and thestick of conditionality, creates the conditions forcountry commitment that the IMF can then moni-tor. In particular, the fact that IMF programs pro-vide credit in tranches, in accordance with specificpolicy commitments, helps the IMF limit the risk ofopportunistic debtor behavior.

In turn, country commitment to good perfor-mance and, hence, ability to service external debtarises from three overlapping sources: policy design,conditionality, and the costs associated with thesepolicies and conditionality. Private investors will bemore likely to invest their capital when they believethat an IMF program will improve the country’s eco-nomic policy (e.g., Rodrik, 1996; Masson andMussa, 1995; and Bird and Rowlands, 1997) or thatloans will be repaid (Tirole, 2002). Conditionalityworks if the country can, in fact, overcome politicaland other implementation barriers. Even if previousprograms have not been fully implemented, in-vestors may still view the adoption of a new programfavorably if the borrowing government must enactpolicy changes to access new credit (Marchesi andThomas, 1999).

In a recent paper, Mody and Saravia (2003) arguethat the delegated monitor role, although impor-tant, is effective only under particular country con-ditions. If a country has apparently sound economic

fundamentals, entering into an IMF program mayactually signal a problem and, hence, make marketaccess worse. At the other extreme, if country fun-damentals already place it in a crisis mode, the com-mitment implied by an IMF program may not becredible and thus, once again, be of limited value.Indeed, Powell and Arozamena (2003) argue thatwhen a country is in a crisis, “gambling for resurrec-tion” may lead countries in IMF programs to behavein a manner contrary to that required for achievingstabilization and recovery. Mody and Saravia (2003)find empirical support for their proposition that theIMF’s delegated monitor role is likely to work bestwhen fundamentals are in an intermediate range.

Catalytic Lending

By providing liquidity to a country, the IMF can be acatalyst, in that it reduces the probability of defaultand can solve the coordination problem betweenprivate creditors (e.g., Miller and Zhang, 2000; Mor-ris and Shin, 2003; and Corsetti, Guimarães, andRoubini, 2003). That is, even if the seal of approvaland delegated monitor channels do not work, themere provision of credit by the IMF can give privatecreditors an incentive to roll over their existingloans and possibly supply new loans. If IMF pro-grams are catalytic, they should be expected to in-crease gross capital flows.

For this channel to work, the IMF should havesufficient information to determine whether theproblem is mainly illiquidity, as opposed to insol-vency, and it should be able to decide the optimalamount of lending—that is, the amount of lendingthat is large enough to convince private creditors toroll over their loans. Successfully performing thisrole requires balancing the benefits of the programagainst the costs of moral hazard.12

Morris and Shin (2003); and Corsetti, Guimarães,and Roubini (2003) present models that show thatIMF lending lowers the threshold for preventing de-fault (that is, it lowers the minimum rate of returnnecessary to prevent default and thereby lowers theprobability of default).13 Their main conclusions,summarized in Table 1, are as follows:

10 However, as Ivanova (2003) shows, IMF conditionality maynot be effective when there is opposition to reforms by domesticlobbies. If domestic lobbies are strong enough, the governmentmay not be able to commit itself to “good” policies even in thepresence of the IMF (and even more so if the IMF has imperfectinformation about government decisions).

11 These rules of conduct are given in the IMF’s Articles ofAgreement, which are available on the Web at http://www.imf.org/external/pubs/ft/aa.

12 Lerrick and Meltzer (2003) offer a specific proposal thatwould enable the IMF to preclude panic and contagion whilelimiting moral hazard. The IMF would perform its lender-of-last-resort role by making a cash support bid at a discount to a gov-ernment’s minimum offer for the restructured value of the debt.The IMF floor would guarantee liquidity and create a functioningmarket in the defaulted debt, while lenders would be forced tobear the risks they undertake, thereby reducing moral hazard andfuture instability. In this proposal, however, there is no guaranteethat the IMF would not overbid.

13 The model by Corsetti, Guimarães, and Roubini (2003) alsosuggests that the IMF’s catalytic impact is greater if it is able todistinguish between an insolvent and an illiquid debtor, and thatits lending serves as a signal to the market. This relates to the“seal of approval” discussed previously.

233


• For very bad fundamentals (bad enough that, evenwith IMF support and a maximum adjustment ef-fort, the country cannot prevent default), IMFprograms have no effect. This is because thecountry’s benefits from preventing default donot exceed the cost of the necessary adjustmenteffort needed to improve fundamentals.

• For bad fundamentals (bad enough that, withoutthe IMF, it would not be worthwhile for thecountry to make the adjustment effort neededto prevent default), IMF programs can have apositive effect on both the country’s adjustmenteffort and private creditors’ willingness to rollover their loans. Absent the IMF, the country is“vulnerable.” With IMF support, default isavoided, private capital flows are catalyzed, andfundamentals improve.

• For good fundamentals (good enough that, evenwithout the IMF, the country would be willingto put in the adjustment effort needed to pre-vent default), IMF programs can have a nega-tive effect on the country’s adjustment effort,and fundamentals may actually worsen. At thesame time, there is no effect on private capitalflows (i.e., there is no catalysis).

• For very good fundamentals (good enough that,even without the IMF, the country can preventdefault without any adjustment effort), IMF pro-grams have no effect on the adjustment effort(since there would be no effort in either case)and no effect on capital flows (since creditorswould have rolled over their loans in any case).

Aggregate Trends Before and After Program Adoption

In this section, we present descriptive evidence onaggregate performance following the start of an IMFprogram. Our main finding is that macroeconomicfundamentals (growth, inflation, and the current ac-count) and capital flows all improve after a programhas been adopted, albeit often with a lag. We ob-serve a “dip and recovery” pattern for growth andcapital flows, with an initial deterioration in first-year performance but a significant improvement inthe second year and stabilization in the third year.

As noted in the introduction, our definition ofemerging market economies is the one used by theIMF’s International Capital Markets Departmentand comprises 29 countries. The period covered is1980–2002.14 Performance is measured in terms ofreal GDP growth, inflation, the current account bal-ance (in percentage of GDP), and capital flows (alsoin percentage of GDP). Our measure of capital flowsis the aggregate of gross bond, equity, and loanflows.15 We use gross flows, rather than net flows,since we are interested in a measure of access to in-ternational capital markets. Net flows could besmall either because a country has lost contact withinternational capital markets or because it is repay-ing old loans that are being rolled over.

Average Performance

We start by comparing the average performance inthe 12 months prior to the start of a program (“YearBefore IMF Program”) with the average perfor-mance in the first 12 months (“First Year”), the sec-ond 12 months (“Second Year”), and the third 12months (“Third Year”) after the start of each pro-gram.16 We are interested in studying a country’sperformance immediately following the start of an

14 By merging data from the 1980s and the 1990s, we implicitlyassume that the behavior of creditors has not changed over time.For future research, however, it may be useful to study the 1980sand 1990s separately, given the changes in the type of lending(from sovereign syndicated bank lending to a sovereign bondmarket and short-term cross-border bank lending), the growingtrend of capital account liberalization in the 1990s, and the grad-ual decline in the IMF’s role in explicitly coordinating creditorbehavior.

15 The source of our bond, equity, and loan data is Dealogic. Allother data are from the IMF’s International Financial Statisticsand World Economic Outlook databases and other IMF sources.

16 We are able to date the start of the IMF program precisely.For example, if a program were started in March 1998, but beforeMarch 15, then the first year of the program would include themonths March 1998 through February 1999, while if a programwere started after March 15, the first year would include themonths April 1998 through March 1999. Using this definition ofyears, the data on capital flows and inflation, being available at amonthly frequency, can be timed quite precisely. For data thatare not available at a monthly frequency (e.g., real GDP growthand the current account balance), we follow the convention ofconstructing monthly values by interpolation (i.e., assuming aconstant growth rate during a quarter or year).

Table 1

Summary of Predictions by Morris and Shin (2003) and Corsetti, Guimarães, and Roubini (2003)

Fundamentals

Very bad Bad Good Very good

Effect of IMF lending on adjustment effort None Positive Negative None(moral hazard)

Effect of IMF lending on capital flows None Positive (catalysis) None None

234


IMF program, since both the seal of approval andcatalytic lending channels would suggest that the ef-fect of IMF program adoption on performance is im-mediate, while the delegated monitoring channelbecomes effective during the course of a program.Since it takes time for macroeconomic fundamen-tals to adjust, we study their evolution up to threeyears after the start of the program.

We treat each program as a separate observationand average across programs. To make the averagescomparable across each year, we use a balanced sam-ple, in which the set of programs over which we av-erage is the same for all years. We thus exclude pro-grams where no data are available for the year beforeor any of the three years following the program.17

Our main finding here is that both macroeco-nomic aggregates and capital flows improve follow-ing the adoption of an IMF program. This is consis-tent with Morris and Shin’s (2003) hypothesis thatthe catalytic effect on capital flows and adjustmenteffort go hand in hand. For growth and capital flows,we observe a dip and recovery pattern: in the firstyear of the program, performance fails to improveand may even worsen; the second year tends to see alarge improvement; and, in the third year, a weakerimprovement or even a slight fallback may occur,possibly because of mean reversion. However, inspite of the possible fallback in the third year, theoverall effect after three years for the different indi-cators is generally positive.

Panel A of Figure 1 illustrates the dip and recov-ery pattern for real GDP growth: growth dips in thefirst year of the program, recovers in the secondyear, and falls somewhat in the third year while re-maining well above the low point and even abovethe level before the start of the program. Inflationperformance (Panel B) fails to improve in the firstyear, but then exhibits a large improvement (i.e., areduction in inflation) in both the second and thirdyears following the start of the IMF program. Coun-tries with IMF programs, not surprisingly, tend tostart with a high current account deficit (Panel C);this deficit tends to be reduced immediately in thefirst year of the program and even turns into a surplus

17 Data may not be available because the program started onlyrecently or because a new program started within three years ofthe old program. Dropping programs followed by other programswithin the balanced-sample period could lead to selection bias,because such programs are likely to have failed. However, even ifwe did not drop these programs, average performance would stillbe overestimated in situations where performance improvedunder the second programs, since this improved performancewould then be considered a result of both the first and secondprograms. While we could have considered two consecutive pro-grams simply as one program, we decided not to do so for the fol-lowing reasons: (a) the new program may have a separate signal-ing effect; (b) the amount under the second program may bedifferent from the first; and (c) the second program may be of adifferent type.

A: Real GDP growth

1

2

3

4

5

6 B: Inflation

6

10

14

18

22

26

30

C: Current account to GDP

–4

–3

–2

–1

0

1

2

3

4

0

D: Gross capital flows to GDP

0.8

1.2

1.6

2.0

2.4

1 2 3

0 1 2 3 0 1 2 3

0 1 2 3

years years

years years

Figure 1

Average Performance of Emerging Market Economies with IMF Programs(in percent)

235


in the second year (presumably, as import compres-sion continues). However, as conditions becomemore normal in the third year, the current accountreturns to a lower (possibly, more sustainable)deficit than at the start.

The pattern for capital flows is similar to that ofreal GDP. As Panel D of Figure 1 shows, gross capi-tal flows tend to fall somewhat in the first year ofthe program—especially for bonds and loans, whichwe do not show separately here. However, flows in-crease thereafter. By the second year of the program,the different components of capital flows tend to behigher than in the year preceding the program, andthe bond component continues to increase in subse-quent years (while equity flows “overshoot” in thesecond year and then decline).

Program Duration

We next compare performance across two differentIMF programs: the Stand-By Arrangement (SBA)and the Extended Fund Facility (EFF). Althoughthese programs share several common features, animportant difference is that programs under theSBA are shorter than programs under the EFF.While SBAs are designed to address short-term bal-ance of payments problems and typically last 12–18months, EFFs are designed to address more struc-tural balance of payments problems and last about 3years. Also, repayment for SBAs is expected within21⁄4 to 4 years, while repayment for EFFs is expectedwithin 41⁄2 to 7 years.

Interestingly, we find some evidence that EFFs areassociated with better performance than are SBAs

(Table 2). Given the small number of observationsin each category (90 SBA programs versus 29 EFFprograms), these results should be interpreted withcaution. However, they do suggest that longer-termIMF lending for structural reform is more effectivethan short-term lending for balance of payments ad-justment. A possible explanation for this result isthat EFFs require a stronger—or at least longer—commitment by the IMF, implying a stronger seal ofapproval, a longer period of delegated monitoring,and often—but not always—more lending. An al-ternative explanation is that EFF programs workbetter because they give countries more time for re-payment.18

Another interesting result is that EFF programsare associated with higher capital flows (again, witha slight dip in the first year), while the impact ofSBA programs on capital flows is roughly nil on av-erage. An explanation for this could be that coun-tries that enter into an EFF program tend to startfrom a low base and have larger chances of being re-warded with high capital flows than do countriesthat enter an SBA program. The latter countriessupposedly have fewer deep-seated structural prob-lems and therefore are already able to borrow more,although they have to struggle to regain market con-fidence if they are experiencing temporary liquidity

Table 2.

Average Performance of Emerging Market Economies with EFF and SBA Programs(in percent)

A. Average Performance of Emerging Market Economies with EFF Programs

Year Before Fundamental IMF Program First Year Second Year Third Year

Current account to GDP –3.8 –2.7 3.2 –3.2Gross capital flows to GDP 1.1 1.0 1.7 1.5Inflation 23.5 18.7 13.5 11.0Real GDP growth 1.4 2.6 5.2 4.3

B. Average Performance of Emerging Market Economies with SBA Programs

Year BeforeFundamental IMF Program First Year Second Year Third Year

Current account to GDP –3.7 –1.5 1.7 –0.5Gross capital flows to GDP 2.1 2.1 2.2 2.1Inflation 21.8 25.4 18.0 15.3Real GDP growth 3.0 1.8 3.9 3.9

Notes: EFF denotes Extended Fund Facility; SBA denotes Stand-By Arrangement.

18 For example, in August 1998 Indonesia requested to replaceits 1997 SBA program with an EFF program, with the explicitgoal of lengthening the repayment period. In this case, theamount of credit available under the EFF and its duration wereidentical to those remaining under the SBA; hence, the EFFcould not necessarily be regarded as reflecting a stronger IMFcommitment than the SBA.

236


problems.19 As Table 2 shows, countries with EFFprograms start with an initially lower level of capitalflows than do countries with SBA programs, buttheir recovery is subsequently stronger.

Impact of Initial Fundamentals andLending Levels on Performance

In the previous section, we reported some prelimi-nary evidence that IMF programs are associatedwith improvements in both macroeconomic perfor-mance and capital flows, typically exhibiting a dipand recovery pattern. However, we cannot excludethe possibility that these observed improvementssimply reflect mean reversion. That is, to the extentthat economic conditions may naturally improveafter a crisis even without a program, the observedimprovement cannot necessarily be attributed tothe program alone.

Although it is the case that the initial conditionsof countries that enter into an IMF program arelikely to be worse than those of nonprogram coun-tries, in practice we find considerable variation inthe initial conditions of program countries them-selves. To test whether initial conditions matter,this section disaggregates the response in capitalflows by the state of a country’s external fundamen-tals in the year prior to the adoption of an IMF pro-gram. The goal here is to examine whether there is adifferential response to an IMF program dependingon initial conditions: if there is mean reversion, is itthe same irrespective of the starting point, or arethere systematic differences? We also briefly discusswhether the level of IMF lending matters in helpingto maintain capital flows.

Defining Initial Conditions

To operationalize the predictions by Morris andShin (2003) and others in Table 2, we use four indi-cators to assess the state of those external funda-

mentals that can affect the probability of a country’sdefault. We use two indicators of liquidity risk (ra-tios of short-term debt to reserves, and reserves toimports), one indicator of insolvency (total externaldebt to GDP), and one indicator that is a measure ofboth illiquidity and insolvency (current accountdeficit to GDP). The indicators of illiquidity mea-sure the risk of running out of reserves, while the in-dicators of insolvency assess whether, with IMF lend-ing and external adjustment, the country will be ableto pay its overall debt. Although the four indicatorsare obviously related, the response of capital flows isexpected to be different depending on whether themain problem is illiquidity or insolvency.

Next, we define the thresholds defining whetherthe state of fundamentals is very bad, bad, good, orvery good. While the choice of thresholds is neces-sarily ad hoc, we base them, as far as possible, oncommonly accepted thresholds used in the financialcrises literature and in IMF surveillance.20 Thus, forexample, we consider a country to be in a very badstate during the 12 months preceding the start of anIMF program if the current account deficit is higherthan 6 percent of GDP; if reserves are less than 1.25months of imports; if short-term debt is more thanfour times reserves; or if total external debt exceeds60 percent of GDP. In these cases, a country is gen-erally not expected to be able to prevent default,even with IMF support and maximum adjustmenteffort. Table 3 summarizes this information, and alsocontains similar definitions of “bad,” “good,” and“very good” states.

Results

Table 4 summarizes the evolution of capital flowsconditional on the four states defined in Table 3.

19 For example, as Argentina’s debt levels built up in the 1990s,it had to struggle just to attract the capital flows needed to refi-nance its external debt—particularly after Brazil’s devaluation.

Table 3

Definitions of States in Terms of Fundamentals(in percent)

Current Account Reserves Short-Term External Debt Fundamentals to GDP to Imports Debt to Reserves to GDP

1 (Very bad) Deficit larger than 6 Between 0 and 1.25 Larger than 4 Larger than 0.62 (Bad) Deficit between 3 and 6 Between 1.25 and 3 Between 2 and 4 Between 0.4 and 0.63 (Good) Deficit between 0 and 3 Between 3 and 7 Between 1 and 2 Between 0.2 and 0.44 (Very good) Surplus More than 7 Between 0 and 1 Between 0 and 0.2

20 As an alternative to predefining thresholds for each state,more sophisticated ways could be used to determine the thresh-olds. For example, using a Markov switching model, one coulddefine the states by an unobserved composite state variable thatdepends in certain known ways on the observed fundamentals.Another possibility is to define, for example, the threshold for avery bad current account deficit as a deficit above two standarddeviations from its mean. We leave these as suggestions for futureresearch.

237


Our general findings, explained below in more de-tail, are as follows. First, conditional on very bad ini-tial external fundamentals, IMF programs are typi-cally unable to reverse the slide in capital flows:capital flows either fail to increase or continue tofall. Second, IMF programs seem most successful inmaintaining capital flows to countries with bad ex-ternal fundamentals before the start of a program,when fundamentals are stronger than in the verybad case, but the country is still vulnerable to an ex-ternal crisis. IMF programs in this state are typicallyfollowed by a dip and recovery pattern, suggestingthat countries and the IMF are able to enter into ajoint commitment to a policy effort that is credibleto foreign investors, although the effort may takesome time to bear fruit. Finally, as external funda-mentals improve further (good and very good

states), the role of the IMF diminishes, although itcan help in situations where a temporary shock re-quires short-term external assistance.

As Table 4 shows, countries starting out in a verybad state generally experience a slide in capitalflows despite the introduction of an IMF program.This is most clear for the reserves-to-imports mea-sure: just before the program starts, a low reserves-to-imports ratio is already associated with relativelylow capital inflows, and three years after the start ofthe program, the capital flows-to-GDP ratio hasfallen even further, to only one-third of its initiallevel. When the ratio of short-term debt to reserves—the second measure—is in a very badstate, capital flows increase somewhat in the firstyear of a program but fall thereafter. When the prob-lem is a very high current account deficit—the third

Table 4

Average Capital Flows to Emerging Market Economies Conditional on External Fundamentals(in percent of GDP)

A. Conditional on Reserves to Imports

Year Before Number of State IMF Program First Year Second Year Third Year Observations

Very bad 1.5 1.2 0.5 0.5 8Bad 1.5 1.2 2.1 2.2 16Good 2.2 2.4 2.5 2.0 11Very good 2.3 2.4 2.8 2.8 6

B. Conditional on Short-Term Debt to Reserves



C. Conditional on Current Account Deficit to GDP



D. Conditional on External Debt to GDP



238


measure—capital flows are substantial in the run-upto the program (presumably because the deficit wasbeing financed by external flows), but once the pro-gram has started, capital flows decline for as long astwo years before bumping back up. In countriesstarting out in a very bad state, only in the case ofthe external debt-to-GDP ratio—the fourth mea-sure—is there some slight indication that capitalflows increase after the start of a program. But thisincrease is unlikely to be statistically significant,given the small number of observations.21

Starting from a bad state before program adop-tion, a dip and recovery pattern for capital flows isobserved for all external fundamentals, except forexternal debt, conditional on which capital flowsincrease steadily without a dip. For all other funda-mentals, however, capital flows fall in the first year,suggesting that IMF programs are initiated at themoment when things are worsening. In the secondand third years, capital flows for all four fundamen-tals recover and are above their level before the startof the program. It is as if IMF programs slow the de-terioration and help these countries transit to amore benign state. The indicators that most clearlyrepresent this dip and recovery pattern are the liq-uidity risk indicators: the ratios of reserves to im-ports and of short-term debt to reserves. The patternalso holds for the current account, where the secondyear sees a sharp recovery following a shallower dip.

For countries with good or very good fundamen-tals, the evidence is mixed. When countries start ina good state, IMF programs seem to make very littledifference in the trend of capital flows, as predictedby Morris and Shin (2003). Surprisingly, however,when countries start in a very good state, there issome indication that capital flows actually improvefollowing the start of an IMF program (except forvery good external debt conditions, but this is basedon only two observations). Most likely, these coun-tries with very good external fundamentals had ex-perienced temporary liquidity difficulties that led toan IMF program, but the sound fundamentals re-asserted themselves soon thereafter.22

Finally, we consider the relationship between theamount of IMF lending and the evolution of capitalflows. The evidence, presented in Table 5, is incon-clusive. With a larger lending amount, capital flowsincrease over the course of the following three years;however, the initial level of capital flows is also lowin cases where the IMF has made large loans, so thatit is unclear whether the increase is a response toIMF lending or to the relatively low initial level ofcapital flows.23

Transition Probabilities

Our third step, and most ambitious attempt at acounterfactual, is the comparison of the evolution ofexternal fundamentals for program and nonprogramcountries, disaggregated by initial conditions. Thisallows us to test the null hypothesis that the ob-served improvements in performance following IMFprogram adoption, reported in the paper’s third sec-tion, might be simply the result of mean reversion.

In the previous section, we studied whether andhow the evolution of capital flows following a coun-try’s adoption of an IMF program depends on thestate of the country’s initial external fundamentals.Our results seem to confirm that IMF programs aremost successful at keeping capital flowing to coun-tries with bad, but not very bad fundamentals. How-ever, the fact that countries with bad fundamentalsappear to be better at escaping from near-crisis situa-tions than countries with very bad fundamentalsmay be a general phenomenon and not necessarilythe result of entering into an IMF program. To de-termine whether IMF programs were the main fac-tor in keeping capital flowing, we would need toknow what would have happened to these samecountries if they had not adopted an IMF program.While it is impossible to answer this counterfactualquestion, we can approximate it quite closely bycomparing the performance of program countrieswith the performance of countries that did not adoptan IMF program but had similar fundamentals.

Our methodology is as follows. First, we considera country’s initial state to improve if the probabilityof moving to a better state is raised, or if the proba-bility of moving to a worse state is lowered. Next, todetermine whether IMF programs are associatedwith an improvement in a country’s initial state, wecalculate a transition probability matrix for the fourstates, both for countries with IMF programs and forcountries without IMF programs. We then testwhether the transition probabilities for countries

21 Formal tests for statistical significance would require estima-tion of the distribution of the external fundamentals, which isbeyond the scope of this paper.

22 Another possibility is that, in such cases, liquidity problemswere not the main reason for starting an IMF program. For exam-ple, prior to its July 1996 SBA program, República Bolivariana deVenezuela was in very good shape, judged by (a) its current ac-count, which was in surplus because of a devaluation at the endof 1995; (b) its reserves, which amounted to over seven monthsof imports; and (c) its short-term debt, which was less than50 percent of its reserves. However, an SBA was considered necessary because of banking sector problems, high inflation(about 70 percent in the year prior to the program), high unem-ployment (over 10 percent in 1995), and low growth in the non-oil sector. The program was successful at lowering inflation andimproving confidence, which likely contributed to a rebound incapital flows, although in this case the rebound was also helpedby high oil prices.

23 The results, in this case, are quite sensitive to the specifica-tion. When we consider two-year samples, the picture is lessclear, suggesting that countries that start a new program within athree-year period do not obviously gain from larger amounts ofIMF lending (results available from the authors upon request).

239


with IMF programs are significantly different fromthose for countries without IMF programs.

To calculate these transition probabilities, we de-fine a state variable ST that specifies whether exter-nal fundamentals are very bad (ST =1); bad (ST =2);good (ST=3); or very good (ST= 4) during period T.Since IMF programs take at least a year to have aneffect, we are interested in comparing the funda-mental states in the year before and in the secondyear after the start of a program. Thus, for eachcountry-program pair, we calculate the following:

• ST = average state of external fundamentals dur-ing the year before the start of a program (i.e., Tencompasses the months t –11 through t, wheret is the month in which the program starts); and

• ST + 2 = average state of external fundamentalsduring the second year following the start of theprogram (i.e., T+ 2 encompasses the monthst +13 through t +24).

For nonprogram countries, there is no naturalstarting point, t. Nevertheless, it would be mislead-ing to calculate transition probabilities using pairs ofST and ST+2 for every month in a given year, as thiswould generate a large amount of similar observa-tions and, therefore, would lead to artificially lowstandard errors when testing for differences betweenthe two transition probability matrices. Instead, wecalculate only the states for one month (January) ina given year for each nonprogram country. In addi-tion, for nonprogram countries, we use only the ob-servations of ST and ST+2 if no IMF program was inplace during any of the months t – 11 through t +24.This gives us around 250 nonprogram observations,compared with some 100 program observations.

The null hypothesis is that IMF programs have noeffect even after two years, that is, Pr(ST + 2|ST) isthe same regardless of whether an IMF program waspresent in month t. The alternative hypothesis isthat these transition probabilities are different for

Table 5

Average Capital Flows to Emerging Market Economies Conditional on IMF Lending Amount(in percent of GDP)

A. Amount to Reserves


Low 1.8 2.6 1.3 2.4 6Medium 2.1 1.9 2.4 1.9 22High 1.1 0.9 1.5 1.6 13

B. Amount to Short-Term Debt


Low 1.5 2.1 0.9 1.1 8Medium 2.4 2.2 2.3 2.1 12High 1.8 1.5 2.9 2.6 15

C. Amount to Imports


Low 2.2 1.9 1.7 2.2 11Medium 2.2 2.4 2.3 2.0 12High 1.1 1.1 1.9 1.5 18

D. Composite Amount Index


Low 2.2 2.6 1.6 2.1 11Medium 1.8 1.4 2.1 1.8 27High 1.1 1.2 2.2 1.7 10

countries with and without IMF programs. To deter-mine whether any observed differences are signifi-cant, we conduct a two-sided test and consider dif-ferences as “significant” when they are statisticallysignificant at the 5 percent level. (For the calcula-tion of p-values, see the appendix).

The main result (Table 6) is that IMF programsare generally associated with improvements in ex-ternal fundamentals, except when these fundamen-tals are already in very good shape (State 4). AsTable 6 shows, even for countries that start out in avery bad state (State 1), IMF programs significantlyincrease the probability of moving to a better state.For countries that start out in a bad state (State 2),the programs also significantly increase the proba-bility of moving to a better state after two years, except for the indicator of short-term debt-to-reserves.24 For countries that start out in a goodstate (State 3), fundamentals significantly improvefollowing the start of an IMF program, that is, all ex-ternal fundamentals are significantly more likely tohave improved from State 3 to State 4 after twoyears.25

Synthesis and Conclusion

The IMF has evolving functions in the changingworld economy. In this paper, we have put the spot-light on a set of emerging market economies and theIMF’s role in maintaining their access to interna-tional capital markets. Our results reinforce theoret-ical predictions and earlier empirical findings on thevalue of IMF intervention for countries in vulnera-ble—as distinct from extreme distress—situations.

We identified three channels through which theIMF potentially helps emerging market economiesmaintain access to international capital markets:(a) by providing a good housekeeping seal of ap-proval, (b) by means of delegated monitoring, and(c) through catalytic lending. We argued that, al-though the good housekeeping seal presupposes su-periority of information held by the IMF, it is notevident that the IMF has significantly more compre-hensive or more timely information on a sustainedbasis than, say, investment banks or credit ratingagencies. The IMF does have an informational roleto play, but this more likely occurs in the context ofagency problems, where a country, as an “agent” ofinternational investors, needs to be monitored by acredible external agency. In this context, the secondand third channels are more important than thefirst. As a delegated monitor, the IMF can constrain

opportunistic behavior through its program bench-marks and conditionality imposed ex post. As alender, the IMF can reduce the probability of defaultand solve the coordination problem between privatecreditors, simply through its ability to provide finan-cial resources.

We started by making a simple before-after com-parison (in the third section) suggesting that, on av-erage, both macroeconomic aggregates and capitalflows improve following the adoption of an IMFprogram, albeit with lags. The first year of IMF pro-grams is typically associated with some initial de-cline in performance, but by the second year growthis up and inflation is down. The current account isinitially compressed but by the third year appears toreturn to a more sustainable deficit than at the startof the program. Capital flows follow a path similarto growth: an initial small dip followed by a modestrecovery. These findings are, of course, based on av-eraging across programs.

We also found that three-year programs (EFFs)are associated with better performance than areshorter-term programs (SBAs), suggesting that theimprovement in performance depends positively onthe duration of the program. A possible explanationfor this result is that EFFs imply a stronger seal of ap-proval, a longer period of delegated monitoring, andoften—but not always—more lending.

We next attempted to test whether the effect ofIMF programs on capital flows was conditional onthe state of a country’s initial external fundamentals(in the fourth section), and whether these funda-mentals themselves improved following programadoption. Table 7 summarizes our findings on capitalflows and improvements in fundamentals, as as-sessed by changes in the probability of transitionsacross states.

Our results suggest that when a country starts outin a situation of economic distress (with very badexternal fundamentals), an IMF program is at bestassociated with stemming the decline in capitalflows and is at worst associated with the country’scontinued loss of access to international capitalmarkets. External fundamentals, starting from thisdistressed situation, do improve, as the current ac-count deficit declines and reserves rise. However,when these improvements do occur, they are possi-bly the consequence of significant import compres-sion and, hence, are not necessarily sustainable.Thus, when initial fundamentals are extremelyweak, IMF programs may be associated with someadjustment effort, but this is not sufficient to ensurea significant turnaround in capital flows.

In contrast, we found that when a country is vul-nerable (with bad, but not very bad external funda-mentals), an IMF program is associated with bothbetter market access and improvements in the coun-try’s external fundamentals. It may well be possiblethat, in such vulnerable initial conditions, thecountry is itself able to engineer the recovery. It

240


24 We also calculated the transition probabilities between STand ST+1 and found that IMF programs do not have a significanteffect in one year (results available from the authors upon request).

25 Even within one year, the reserves-to-imports and currentaccount-to-GDP ratios are significantly more likely to improvefrom State 3 to State 4 under an IMF program. However, reservesmay be increasing in some cases just because of IMF disbursements.

241


Table 6

Two-Year Transition Probabilities for Emerging Market Economies With and Without IMF Programs(in percent)

A. Current Account Balance to GDP

ST+2 = 1 ST+2 = 2 ST+2 = 3 ST+2 = 4

ST = 1 Program 0.23 0.15 0.38 0.23(Very bad) Nonprogram 0.29 0.49 0.17 0.06

P-value 0.35 0.01 0.05 0.03

ST = 2 Program 0.14 0.14 0.48 0.24(Bad) Nonprogram 0.26 0.43 0.28 0.03

P-value 0.12 0.01 0.04 0.00

ST = 3 Program 0.10 0.10 0.55 0.25(Good) Nonprogram 0.08 0.27 0.55 0.10

P-value 0.37 0.05 0.50 0.04

ST = 4 Program 0.00 0.11 0.33 0.56(Very good) Nonprogram 0.10 0.13 0.36 0.41

P-value 0.15 0.44 0.44 0.21

B. Reserves to Imports

ST+2 = 1 ST+2 = 2 ST+2 = 3 ST+2 = 4


P-value 0.04 0.30 0.00 n.a


P-value 0.49 0.01 0.01 n.a


P-value 0.01 0.15 0.00 0.00


P-value n.a 0.24 0.37 0.49

C. Short-Term Debt to Reserves

ST+2 = 1 ST+2 = 2 ST+2 = 3 ST+2 = 4


P-value 0.01 0.15 0.03 n.a


P-value 0.07 0.05 0.32 0.44


P-value 0.22 0.24 0.00 0.00


P-value n.a 0.24 0.12 0.23

Note: The abbreviation n.a indicates that the p-value could not be calculated.

242


may, however, also be the case that, although the ef-fort is essentially undertaken by the country, theIMF is used as a commitment mechanism or dele-gated monitor. Either way, this is where we have thestrongest evidence that IMF programs are associatedwith increased capital flows and improved funda-mentals.

Finally, for countries with relatively strong initialconditions (good or very good external fundamen-tals), we found that IMF programs can sometimeshelp to stabilize capital flows following temporaryshocks, while their effect on fundamentals is am-biguous. While Morris and Shin (2003) make a dis-tinction between good and very good states, such adistinction is not clear in the data examined. More-over, there is no clear evidence that in countrieswith good fundamentals, IMF lending may lead tomoral hazard (that is, reduced adjustment effort bythe country itself).

Our results are subject to several caveats. First, al-though we have attempted to construct the bestcounterfactual possible, by comparing the perfor-mance of program countries with that of nonpro-gram countries with similar initial conditions, theremay be other important observed or unobserved ini-tial conditions that we have not controlled for.While it would be useful to extend our analysis bywidening our definition of initial conditions, thefact remains that we will never know what wouldhave happened to program countries had they notadopted a program. Hence, one can never concludewith certainty that IMF programs necessarily hadthe determining influence in the observed improve-ment in performance. Second, since we limited our-selves to emerging market economies between 1980and 2002, our results cannot necessarily be general-ized to apply to all IMF programs. Finally, while wedistinguished between EFF and SBA programs, weotherwise treated all programs as more or less uni-form. In reality, however, IMF programs differ inboth their design and their execution, and these dif-ferences may affect the private sector’s response interms of capital flows.

We have several suggestions for future research.First, the list of variables used to characterize acountry’s initial condition could be expanded to in-clude real GDP growth, inflation, and even capitalflows themselves. Second, more sophisticatedeconometric methods could be used to derive, ratherthan predefine, the thresholds for the states of exter-nal fundamentals. For example, one could use aMarkov switching model to estimate an unobservedcomposite state variable that depends in certain as-sumed ways on the observed fundamentals. Third,rather than treating IMF programs as a “black box,”one could further differentiate between IMF pro-grams in terms of their design (e.g., precautionaryversus nonprecautionary programs) and execution(e.g., canceled versus completed programs). Fourth,the determinants of capital flows could be bettermodeled. While this paper focused on the supply-side determinants of capital flows (i.e., decisions onthe part of creditors as to whether to roll over theirloans or issue new loans), it would be useful to cap-ture demand-side determinants as well by, for exam-ple, scaling capital flows in terms of gross financingneed. Similarly, it would be useful to try to distin-guish between “pull” and “push” factors as determi-nants of capital flows.

Appendix. Calculation of p-Values forDifferences in Transition Probabilities

Let p1 denote the transition probability between twostates for countries with an IMF program, and let p0denote the transition probability between the sametwo states for countries without an IMF program.Let D = p1 – p0 denote the difference between thetwo transition probabilities. The standard deviationof D is then given by

Table 7

Summary of Main Results

Initial Fundamentals Capital Flow Trends Change in Fundamentals

Very bad (State 1) Either continued decline after initiation of Some improvement in external fundamentals, but program or, at best, no improvement related to domestic recessionary conditions

Bad (State 2) Dip and recovery, as the decline in capital Strong evidence of improvement in external flows is stemmed and a rise is initiated fundamentals

Good and very good Either no change or evidence consistent Ambiguous(States 3 and 4) with stabilization of flows following a

temporary shock

σD

p pn

p pn

=−

+−

0 0

0

1 1

1

1 1( ) ( ),

243


where n0 and n1 are the number of observations cor-responding to each row in the transition probabilitymatrix (e.g., for p11, this is the number of caseswhere ST = 1). Under the null hypothesis, we haveH0: p1= p0= p. Let p0 , p1,and p denote the estimatesof the respective transition probabilities, i.e., thesample proportions. We then have

where X0 and X1 are the number of transitions with-out and with an IMF program, respectively. Thisgives the following estimate of σD under the null hy-pothesis:

If n0 and n1 are large, the standardized differencez = (p1 – p0)/SEDp is approximately N(0,1). The p-value for a test of the null hypothesis (H0: p1=p0)against the alternative hypothesis Ha: p1≠ p0 is thusgiven by Pr(Z ≠ z), where Z is a standard normal ran-dom variable.

ReferencesBarro, Robert J., and Jong-Wha Lee, 2003, “IMF Programs:

Who Is Chosen and What Are the Effects?” WorkingPapers in Trade and Development No. 2003/09 (Can-berra, Australia: Australian National University).Available on the Web at http://post.economics.harvard.edu/faculty/barro/papers/IMF_Programs.pdf.

Bird, Graham, and Dane Rowlands, 1997, “The CatalyticEffect of Lending by the International Financial Insti-tutions,” World Economy, Vol. 20, No. 7, pp. 967–91.

Boorman, Jack, and Mark Allen, 2000, “A New Frame-work for Private Sector Involvement in Crisis Preven-tion and Crisis Management,” in Reforming the Inter-national Financial System: Crisis Prevention andResponse, ed. by Jan Joost Teunissen (The Hague:Forum on Debt and Development), pp. 101–23.

Bordo, Michael D., 2003, “Market Discipline and Finan-cial Crisis Policy: A Historical Perspective,” in Research in Financial Services: Private and Public Policy,Vol. 15, ed. by George G. Kaufman (Greenwich,Connecticut: JAI Press), pp. 157–82.

———, and Harold James, 2000, “The International Mon-etary Fund: Its Present Role in Historical Perspective,”NBER Working Paper No. 7724 (Cambridge, Massa-chusetts: National Bureau of Economic Research).

Bordo, Michael D., and Finn E. Kydland, 1995, “The Gold Standard as a Rule: An Essay in Explo-ration,” Explorations in Economic History, Vol. 32,No. 4, pp. 423–64.

———, 1996, “The Gold Standard as a CommitmentMechanism,” in Modern Perspectives on the Gold Stan-dard, ed. by Tamim Bayoumi, Barry Eichengreen, andMark P. Taylor (Cambridge, New York, and Mel-bourne: Cambridge University Press), pp. 55–100.

Bordo, Michael D., and Hugh Rockoff, 1996, “The GoldStandard as a ‘Good Housekeeping Seal of Approval,’ ”Journal of Economic History, Vol. 56, No. 2,pp. 389–428.

Bordo, Michael D., and Anna J. Schwartz, 2000, “Measur-ing Real Economic Effects of Bailouts: Historical Per-spectives on How Countries in Financial DistressHave Fared With and Without Bailouts,” NBERWorking Paper No. 7701 (Cambridge, Massachu-setts: National Bureau of Economic Research).

Calvo, Guillermo A., 1998, “Capital Flows and Capital-Market Crises: The Simple Economics of SuddenStops,” Journal of Applied Economics, Vol. 1, No. 1,pp. 35–54.

Corsetti, Giancarlo, Bernardo Guimarães, and NourielRoubini, 2003, “International Lending of Last Resortand Moral Hazard: A Model of IMF’s Catalytic Fi-nance,” NBER Working Paper No. 10125 (Cam-bridge, Massachusetts: National Bureau of EconomicResearch).

Cottarelli, Carlo, and Curzio Giannini, 2002, “Bedfel-lows, Hostages, or Perfect Strangers? Global CapitalMarkets and the Catalytic Effect of IMF Crisis Lend-ing,” IMF Working Paper 02/193 (Washington: International Monetary Fund). This paper also ap-pears, in somewhat different form, as Chapter 12 inthis volume.

Ferguson, Roger W., 2002, “Understanding FinancialConsolidation,” Economic Policy Review, Federal Re-serve Bank of New York, Vol. 8 (May), pp. 209–13.


Goldstein, Morris, and Peter Montiel, 1986, “EvaluatingFund Stabilization Programs with Multicountry Data:Some Methodological Pitfalls,” Staff Papers, Interna-tional Monetary Fund, Vol. 33 (June), pp. 304–44.

Hajro, Zlata, and Joseph P. Joyce, 2004, “A True Test: DoIMF Programs Hurt the Poor?” Wellesley College De-partment of Economics Working Paper No. 04-01(Wellesley, Massachusetts: Wellesley College).

Haque, Nadeem Ul, and Mohsin Khan, 1998, “Do IMF-Supported Programs Work? A Survey of the Cross-Country Empirical Evidence,” IMF Working Paper98/169 (Washington: International Monetary Fund).

Hutchison, M., 2001, “A Cure Worse Than the Disease?Currency Crises and the Output Costs of IMF-Supported Stabilization Programs,” NBER WorkingPaper No. 8305 (Cambridge, Massachusetts: National Bureau of Economic Research).

Ivanova, Anna, 2003, “IMF Conditionality and Imple-mentation of IMF-Supported Programs” (Ph.D. dissertation; Madison, Wisconsin: University of Wisconsin-Madison).

Khan, Mohsin, 1990, “The Macroeconomic Effects of Fund-Supported Adjustment Programs,” Staff Papers, International Monetary Fund, Vol. 37 (June),pp. 195–231.

Köhler, Horst, 2001, “Promoting Stability and Prosperityin a Globalized World,” remarks by IMF ManagingDirector, Council of the Americas, Washington, May 7. Available on the Web at http://www.imf.org/external/np/speeches/2001/050701.htm.

ˆ ,pX Xn n

=++

0 1

0 1

SE p pn nDp = − +

ˆ( ˆ) .11 1

0 1

244


Lerrick, Adam, and Allan H. Meltzer, 2003, “Blueprint foran International Lender of Last Resort,” Journal ofMonetary Economics, Vol. 50 (January), pp. 289–303.

Loungani, Prakash, 2001, “How Accurate Are PrivateForecasts? Cross-Country Evidence from ConsensusForecasts of Output Growth,” International Journal ofForecasting, Vol. 17 (July–September), pp. 419–32.

Marchesi, Silvia, and Jonathan P. Thomas, 1999, “IMFConditionality as a Screening Device,” EconomicJournal, Vol. 109 (March), pp. 111–25.

Masson, Paul R., and Michael Mussa, 1995, “Long-TermTendencies in Budget Deficits and Debt,” IMF Work-ing Paper 95/128 (Washington: International Mone-tary Fund).

Meltzer, Allan H., 2002, “New International FinancialArrangements,” Monetary and Economic Studies,Vol. 20 (Special Edition, December), pp. 11–22.

Miller, Marcus, and Lei Zhang, 2000, “Sovereign LiquidityCrises: The Strategic Case for a Payments Standstill,”Economic Journal, Vol. 110 (January), pp. 335–62.

Mody, Ashoka, and Diego Saravia, 2003, “CatalyzingCapital Flows: Do IMF-Supported Programs Work asCommitment Devices?” IMF Working Paper 03/100(Washington: International Monetary Fund).

Morris, Stephen Edward, and Hyun Song Shin, 2003,“Catalytic Finance: When Does It Work?” CowlesFoundation Discussion Paper No. 1400 (New Haven,

Connecticut: Yale University). Available on the Webat http://ssrn.com/abstract=383604.

Obstfeld, Maurice, and Alan M. Taylor, 2003, “SovereignRisk, Credibility and the Gold Standard: 1870–1913versus 1925–31,” Economic Journal, Vol. 113 (April),pp. 241–75.

Powell, Andrew, and Leandro Arozamena, 2003, “Liquid-ity Protection Versus Moral Hazard: The Role of theIMF,” Journal of International Money and Finance,Vol. 22 (Special Issue, December), pp. 1041–63. Avail-able on the Web at http://www.utdt.edu/~apowell.

Przeworski, Adam, and James Raymond Vreeland, 2000,“The Effect of IMF Programs on Economic Growth,”Journal of Development Economics, Vol. 62 (August),pp. 385–421.

Reinhart, Carmen M., Kenneth Rogoff, and MiguelSavastano, 2003, “Debt Intolerance,” Brookings Papers on Economic Activity: 1, Brookings Institution,pp. 1–74

Rodrik, Dani, 1996, “Why Is There Multilateral Lending?”in Annual World Bank Conference on Development Eco-nomics, 1995, ed. by Michael Bruno and BorisPleskovic (Washington: World Bank), pp. 167–93.

Tirole, Jean, 2002, Financial Crises, Liquidity, and the Inter-national Monetary System (Princeton, New Jersey andOxford: Princeton University Press).

245

14

The paper investigates fiscal developments in 112 coun-tries during the 1990s. It finds that although the overallfiscal balance improved in most of them, the composi-tion of this improvement differed. In countries withoutIMF-supported programs, revenues increased modestlyand expenditure declined sharply, while in programcountries both postprogram revenue and expenditure de-clined. In countries with programs that included fiscalstructural conditions, however, the adjustment was ef-fected primarily through sharp expenditure compression.No evidence of a statistically significant impact of IMFconditionality was found. Moreover, fiscal developmentswere influenced by cyclical factors and by the generalstance of macroeconomic policies.

Introduction

What determines the composition of fiscal adjust-ment, and does it differ between countries withIMF-supported programs and those without sucharrangements? Moreover, how relevant is IMF struc-tural conditionality for postprogram fiscal develop-ments? This paper attempts to answer these ques-tions by investigating the fiscal developments in112 countries during the 1990s, some with and somewithout IMF-supported programs.

A central objective of IMF-supported programshas been to reduce external imbalances. This oftenrequires bringing the budget under control: first, fis-cal profligacy often causes current account deficits,and, second, even if the initial budgetary position issustainable, additional fiscal tightening may beneeded if the domestic currency comes under pres-sure. This adjustment has been part of broadermedium-term macroeconomic programs that alsoencompass supply-side structural reforms relevantfor external stability.

This paper examines postprogram fiscal develop-ments in countries with and without an IMF-supported program. It finds significant differences inthe composition of adjustment between programand nonprogram countries as well as large differ-ences among program countries. In nonprogramcountries, revenue increased modestly and expendi-ture declined sharply, while in program countriesboth revenue and expenditure declined during thepostprogram period. Moreover, in IMF-supportedprograms that included structural conditions, theadjustment was effected primarily through expendi-ture compression in order to offset revenue declines.We did not find any evidence that fiscal structuralconditions improved revenue performance after theend of the program. Fiscal developments were af-fected by the business cycle and general stance ofmacroeconomic policies.

The paper is organized as follows. First, we reviewthe stylized facts and define the sample. Second, wedescribe the techniques used in our estimations.Third, we present and discuss our results. The finalsection offers concluding remarks.

IMF Programs and Fiscal Developments

How to Measure the Impact of IMF-Supported Programs

What is the impact of IMF-supported programs onfiscal adjustment? In the literature, three influenceshave been construed. One view is that those pro-grams provide external resources beyond the financ-ing provided by the IMF itself to the extent thatthey have a catalytic effect—thus adjustments takeplace at lower costs than in the absence of such anarrangement. Hence, IMF-supported programs canbe associated with either smaller or larger fiscaldeficits, depending on the nature of the shock andthe design of the program (Bird and Rowlands,2002). A second view is that those programs pre-scribe fast adjustment by uniformly requiring exces-sive monetary and fiscal tightening, hurting boththe poor and businesses in the process. A thirdview is that IMF-supported programs delay funda-mental reforms by merely treating the symptoms of

Long-Term Fiscal Developments and IMF Conditionality: Is There a Link?Ales Bulír and Soojin Moon1

1 This article was first published in Vol. 46 (2004) of Compara-tive Economic Studies. It appears here with the permission of Pal-grave Macmillan, the journal’s publisher. The authors are in-debted to Tim Lane and Alex Mourmouras for discussions and toChistina Daseking, Kamil Dybczak, Gervan Fearon, Rex Ghosh,Sanjeev Gupta, Javier Hamann, Eduardo Ley, Alun Thomas, andparticipants in the 2002 Atlantic Economic Society conference,the 2002 Czech Economic Society conference, the 2003 Interna-tional Institute of Public Finance conference, and InternationalMonetary Fund seminars for comments.

CHAPTER

246

14 � LONG-TERM FISCAL DEVELOPMENTS AND IMF CONDITIONALITY: IS THERE A LINK?

financing needs by repeated lending to crisis-proneand structurally unstable countries (Bird, 1996).

In assessing the impact of IMF-supported pro-grams, we ask two questions. First, what are the fac-tors that lead to IMF-supported programs? Eco-nomic variables, such as the current accountbalance, inflation, international reserves, debt ser-vice, GDP per capita, and so on, together with par-ticipation in previous programs, explain reasonablywell the approval of an IMF-supported arrangement(Conway, 1994; and Knight and Santaella, 1997).Policy commitments made by recipient govern-ments matter as well—if the authorities promisestronger adjustment, the IMF is more likely to ap-prove a bigger loan. Barro and Lee (2002) foundthat “better connected” countries are likely to getmore money with fewer strings attached. In con-trast, the literature found no relationship betweenpolitical economy variables (political institutions,quality of bureaucracy, and so on) and participationin an IMF-supported program.

Second—and this is the question we are primarilyinterested in—what are the macroeconomic effectsof IMF-supported programs? This strand of the liter-ature has a few well-established stylized facts as well.IMF-supported programs were found to be associ-ated with an improved postprogram current accountbalance. Inflation slowed down and real growth re-covered, however, typically by less than what wasprojected under the program (Conway, 1994;Schadler and others, 1995; and Ghosh and others,2002). In contrast, Barro and Lee (2002) reportedopposite results—participation in an IMF-supportedprogram was found to lower growth and investment.

Macroeconomic effects of IMF-supported pro-grams depended, on the one hand, on borrowingcountries’ domestic political economy (Ivanova andothers, 2003; Khan and Sharma, 2001), and on theother hand, on the technical design of the program(conditionality) or the amount of money borrowed(Schadler and others, 1995). Regarding the former,strong special interests, political instability, ineffi-cient bureaucracies, lack of political cohesion, andethno-linguistic divisions weakened program imple-mentation. Adjustment programs were more suc-cessful in countries where they augmented home-grown reforms than in countries where the donorstried to impose them on unwilling authorities. Re-garding the latter, the impact of conditionalityseems governed by a “Laffer-curve” relationship,whereby a few, well-targeted conditions had a posi-tive impact on economic performance, but toomany or too intrusive conditions hindered such per-formance (Collier and others, 1997; Goldstein,2000; and Bird, 2001).

To this end, we will use the IMF’s Monitoring of IMF Arrangements (MONA) database, whichcollects information on conditionality under IMF-supported programs and which was first utilized inIMF (2001). Surprisingly, assessments of structuralconditionality have been rare, and this paper is one

of a few empirical exercises to address this condi-tionality’s role in macroeconomic adjustment.

What Is IMF Conditionality?

Conditionality is an explicit link between the ap-proval (or continuation) of IMF financing and theimplementation of certain aspects of the authorities’policy program (Guitián, 1981). The conditionsmay be either quantitative (say, a limit on reservemoney growth) or structural (say, the introductionof a value-added tax). In general, conditionality isdesigned to encompass policy measures that are crit-ical to program objectives or key internal data tar-gets that sound warning bells if policies veer offtrack. Whereas in the mid-1980s structural condi-tionality in IMF-supported arrangements was rare,by the mid-1990s about half of all programs includedstructural conditions. The average number of struc-tural conditions in a program year increased from 2in 1987 to more than 16 in 1997 (IMF, 2001).

These developments were the result of severalforces. First, the IMF gradually placed more empha-sis on supply-side reforms compared with demandmanagement. Second, the IMF’s involvement inlow-income and transition countries was focused onthe alleviation of structural imbalances and rigidi-ties prevalent in these economies. Finally, the expe-rience with monetary and fiscal policies indicatedthat their success depends critically on structuralconditions. Indeed, most structural conditions werein the core area of IMF expertise.

In this paper, we focus on three main types ofstructural conditions tabulated in the MONA data-base: prior actions, which are stipulated as precondi-tions to the approval of an IMF-supported program;structural performance criteria, fulfillment of which isa formal precondition for program continuation;and structural benchmarks, which are agreed with theauthorities and monitored by IMF staff, but are not aformal precondition for program continuation. Themajority of conditions were structural benchmarks,while structural performance criteria were the leastnumerous conditions. The extent of structural con-ditionality was in part determined endogenously—countries with a large reform agenda or history ofpoor reform performance tended to get more condi-tions, although no clear-cut answers as to why somecountries have many more conditions than othersare available (IMF, 2001). If anything, distributionof structural conditions was positively correlatedwith the length of the programs.

All but two IMF-supported programs with struc-tural conditionality in our sample contained at leastone fiscal condition.2 Indeed, fiscal structural condi-tionality was the most common area of structural

2 Throughout the paper, we used a sample of 112 countries, ofwhich 48 countries did not have a program during the sample pe-riod, and 31 and 33 countries had programs without and withstructural conditions, respectively.

247

Ales Bulír and Soojin Moon

conditionality, comprising about 50 percent of allconditions. While most fiscal conditions were designed as neutral with regard to the overall fiscal balance, some conditions were geared toward eitherhigher revenue or lower expenditure. We classify allof those measures according to their expected rev-enue or expenditure impact (Table 1). Based on IMFcountry team assessments, close to 4⁄5 of all fiscalconditions were met.

Some Stylized Facts About FiscalDevelopments in 1990s

Fiscal developments—besides the immediate, short-term impact of IMF-supported programs—are af-fected by the business cycle, political economy, and

debt-sustainability factors. First, the impact of cycli-cal conditions was strong in our sample—while realGDP grew on average by 1.5 percent during 1993–94,the rate more than doubled to almost 4 percent dur-ing 1997–99. Second, the components of the overallfiscal balance were public choice variables, and votersdecided how much tax they wanted to contribute andhow they wanted the proceeds to be spent (Drazen,2000). Third, debt sustainability constrained the fis-cal stance: the deficits preferred by the electorate maynot be sustainable (Hansson and Stuart, 2003).

The fiscal balance improved in 2⁄3 of all countriesby an average of 2 percentage points of GDP be-tween the preprogram and postprogram periods orbetween 1993 and 1999 for the nonprogram coun-tries (Figure 1 and Table 2). The magnitude of the

Table 1

Frequency of Fiscal Structural Conditionality

Total Number of Conditions1 Implementation Ratio2

All conditions 15.4 77.4Revenue conditions3 4.7 78.5Expenditure conditions3 1.8 81.3Neutral conditions4 8.7 71.4

Sources: IMF, Monitoring of IMF Arrangements (MONA) database; and authors’ calculations.1 Sample average, per program, not adjusted for program length.2 Sample average, implemented conditions/total conditions, in percent.3 Conditions with identified impact on the overall balance.4 Revenue and expenditure conditions without a clear impact on the overall balance.

Table 2

Change in Fiscal Outcomes Three Years After End of IMF-Supported Programs1

(in percent of GDP)

Expenditure Overall Balance Revenue and Grants and Net Lending

Initial Initial InitialChange balance Change revenues Change expenditures

All countriesAverage 1.8 –4.4 –0.3 25.2 –2.0 29.2Median 1.5 –3.7 0.0 24.2 –2.0 28.9Of which:Nonprogram countries2

Average 3.2 –4.5 0.4 27.0 –2.8 31.5Median 2.4 –3.7 0.3 27.2 –2.6 31.5

Program countriesAverage 0.4 –4.2 –1.0 23.4 –1.2 27.0Median 0.5 –3.7 –0.5 20.6 –0.8 24.7Of which:Without structural conditions

Average –1.9 –2.9 –0.3 24.0 1.6 26.9Median –0.8 –2.7 0.2 22.2 1.0 24.2

With structural conditionsAverage 3.7 –6.3 –2.1 22.6 –5.2 27.1Median 2.6 –5.9 –1.5 19.3 –4.7 25.4

Sources: IMF, World Economic Outlook database; authors’ calculations.1 Three years after the end of the IMF-supported program minus the preprogram observation.2 1999 for nonprogram countries. The median initial observation is 1993 and the median end-period observation is 1999.

248


–30.0

–20.0

–10.0

0.0

10.0

20.0

30.0

40.0

–25.0 –20.0 –15.0 –10.0 –5.0 0.0 5.0 10.0 15.0 20.0

Initial overall balance

Impr

ovem

ent

in t

he b

alan

ce

Lebanon

Singapore

Kuwait

Malawi

Lesotho

Figure 1

Change in Overall Fiscal Balance(in percent of GDP, 112 countries)

Sources: IMF, World Economic Outlook database; authors' calculations.Note: The figure shows the change three years after the end of the IMF-supported program(s) or 1999 for nonprogram countries, compared

with the initial observation. The median initial observation is 1993, and the median end-period observation is 1999.

249


postprogram fiscal improvement was not uniform,however, and nonprogram countries improved theirfiscal balances more than program countries—by3 percentage points versus 1⁄2 of a percentage point ofcyclically nonadjusted GDP. Differences prevailedamong program countries: while nonstructural pro-gram countries worsened their balances by some 2 per-centage points of GDP, those with structural condi-tionality improved it by more than 3 percentagepoints of GDP. These findings are robust to the choiceof the end-period observation: our results change littlewhether we assess them one, two, or three years afterthe end of the IMF-supported program.

How was the fiscal adjustment achieved? First,revenue adjustment was much weaker than expen-diture adjustment. Revenue and grants declined inprogram countries and increased somewhat in non-program countries. The difference could not be ac-counted for by either the lowering of trade taxes orlower aid receipts. Regarding the former, we did notfind any quantitative link between trade taxes andrevenues. Regarding the latter, the contribution ofgrants is too small to account for the fall in the ag-gregate variable (Bulír and Hamann, 2003). Sec-ond, the expenditure compression was strong innonprogram and structural program countries (by

3 percentage points and 5 percentage points of GDP,respectively), while in nonstructural program coun-tries postprogram expenditures expanded by 11⁄2 per-centage point of GDP.

The variability of program country results suggeststhat we control for exogenous and program-specificfactors. First, the initial fiscal deficits in nonstruc-tural program countries were smaller than those instructural program countries and did not pose such athreat to macroeconomic stability (Table 2). Sec-ond, the nature of the initial disequilibrium differedacross countries: in nonstructural program coun-tries, GDP declined more sharply prior to the pro-gram, and the countries’ rates of inflation and GDPper capita were higher (Table 3). Third, structuralconditionality programs had a higher incidence ofprogram interruptions.3 Finally, programs that didnot include structural conditions were mostly short-term in nature, typically Stand-By Arrangements.In contrast, structural conditions were mostly applied in the context of the Enhanced Structural

3 More conditions obviously increase the risk of missing someof them. However, missing one of the conditions does not stop aprogram—provided that the macroeconomic program has re-mained on track, the missed condition is typically waived.

Table 3

Selected Characteristics of Program and Nonprogram Countries

Preprogram Developments

PostprogramGDP per Current Real Terms of Program Real capita1,2 account1,3 GDP1,4 trade Inflation1,4 Stoppage5 GDP4,6

All countriesAverage 6,882 –4.4 1.5 0.8 229.0 n.a. 3.9Median 1,954 –2.8 2.7 0.4 11.2 n.a. 3.5Of which:Nonprogram countriesAverage 12,751 –2.3 3.9 0.9 6.1 n.a. 3.6Median 12,772 –1.2 3.3 0.1 2.9 n.a. 3.1

Program countriesAverage 1,134 –6.6 –0.7 0.6 447.3 57.1 4.1Median 774 –3.6 1.5 0.5 23.9 n.a. 3.7Of which:Without structural conditionsAverage 1,511 –7.6 –1.5 2.0 610.2 48.3 3.5Median 1,239 –3.2 1.2 0.5 28.1 n.a. 4.1

With structural conditionsAverage 587 –5.2 0.4 –1.5 211.1 70.0 5.0Median 367 –3.9 2.8 1.0 19.9 n.a. 3.5

Sources: IMF, World Economic Outlook and Monitoring of IMF Arrangements (MONA) databases; authors' calculations.Note: The abbreviation n.a. denotes not applicable.1 Average for 1993–94.2 In 1995 U.S. dollars.3 In percent of GDP.4 Percentage change.5 Program stoppage occurs if either (a) the scheduled program review was not completed, or (b) all scheduled reviews were completed but

the subsequent annual arrangement was not approved. If a country had more than one program during this period, one stoppage overrides oneor more successes.

6 Average for 1997–99 for nonprogram countries.

Adjustment Facility (ESAF), which was succeededby the Poverty Reduction and Growth Facility(PRGF), or the Extended Fund Facility (EFF).

Specification of Model

Fiscal developments are affected by various exoge-nous and country-specific effects and, therefore, wereexamine them in multivariate panel and cross-country regressions. The econometric investigationof the role of IMF-supported programs has tradition-ally been motivated by the following question: “Didthe involvement of the IMF significantly improvethe macroeconomic outcomes relative to what they would have been in the absence of an IMF-supported program?”4 Macroeconomic outcomes,such as inflation or external balance, were describedas a function of (a) policies that would have beenobserved in the absence of an IMF-supported pro-gram; (b) exogenous variables, such as terms-of-trade shocks or wars, and political economy vari-ables, such as the stability of the government; and(c) the presence of an IMF-supported program.5

The simple model we have described has twodrawbacks. First, “macroeconomic policies in the ab-sence of an IMF-supported program” is an unobserv-able variable that has to be constructed in an ad hocfashion. Second, the impact of IMF programs is am-biguous: an identical macroeconomic outcome canbe achieved because of the confidence effect of aprogram, a cumulative impact of policies and IMF fi-nancing (the catalytic effect), or structural reforms.

The key empirical issue is the formulation of poli-cies adopted in the absence of IMF involvement.These policies can be observed only for nonprogramcountries and a counterfactual has to be estimatedfor program periods. Goldstein and Montiel (1986)suggested constructing a policy reaction functionlinking the changes in macroeconomic policies tothe deviations of observed lagged outcomes fromtheir preannounced target values, and lagged exoge-nous variables.

Our modification of the model is twofold. We at-tempted to separate the impact of the country’s per-formance under the program, and structural condi-tionality. First, we checked compliance withprogram conditions. Successful programs were de-fined as those that either disbursed all committedresources without interruptions or those that weredesigned and executed as precautionary arrange-ments (see Ivanova and others, 2003). A statisti-cally significant parameter would indicate that the

IMF’s emphasis on program implementation hassome bearing on postprogram performance.

Second, we separated out the role of fiscal struc-tural conditionality. We tested whether the pres-ence and implementation of IMF fiscal structuralconditionality led to fiscal outcomes that were sta-tistically different from those without such condi-tionality. There was no need to establish counterfac-tual structural policies: similar fiscal structuralreforms were introduced irrespective of the presenceof an IMF-supported program.

Sample Selection and Estimation

We estimated the model in three steps. First, usingdata for nonprogram countries only, we estimatedthe policy reaction function for the relevant macro-economic variables. Second, using the estimated pa-rameters, we simulated macroeconomic policies inprogram countries to reflect what those policieswould have been in the absence of an IMF-supported program. Hence, the vector of policiescomprised actual observed policies in nonprogramcountries and counterfactual policies in programcountries. Third, we estimated the model for bothprogram and nonprogram countries, capturing theimpact of IMF-supported programs and structuralconditionality residually.

We selected the 1993–96 period because of threeconsiderations. First, this four-year period followedthe IMF membership of transition economies in1991–92, but preceded the “Asian” crisis of1997–98. Second, during this period the IMF wasdeeply involved in structural reforms in developingeconomies. Third, we needed three years of after-program data for the General Evaluation Estimator(GEE) estimation, which made 1996 the latest per-missible cutoff point in our sample.

Policy Reaction Function

The policy reaction function determined the stanceof monetary, external, and incomes policies, respec-tively, as a function of the preannounced fiscal ad-justment. The fiscal targets were derived from one-year-ahead World Economic Outlook (WEO)projections based on the annual policy discussionsbetween the authorities and IMF staff, which reflectthe authorities’ policy stance for the period ahead.6

250


4 Such a question can be answered using the General Evalua-tion Estimator (GEE), owing to Goldstein and Montiel (1988),who construct counterfactual economic policies first and thentest the importance of IMF-supported programs. This approachwas applied by, among others, Khan (1990); Conway (1994); andDicks-Mireaux, Mecagni, and Schadler (2000).

5 For a description of the model, see Bulír and Moon (2004).

6 The estimation was for the period 1992–97 with data for 48countries that did not have an IMF-supported program duringthe 1991–97 period, or two years prior to 1991: Australia, Aus-tria, The Bahamas, Bahrain, Belgium, Belize, Botswana, Canada,China, Colombia, Cyprus, Denmark, Fiji, Finland, France, Ger-many, Greece, Grenada, Hong Kong SAR, Ireland, Israel, Italy,Japan, Kuwait, Lebanon, Maldives, Malta, Mauritius, Myanmar,the Netherlands, New Zealand, Norway, Oman, Paraguay, Portu-gal, Qatar, Samoa, Singapore, the Solomon Islands, SouthAfrica, Spain, St. Lucia, Swaziland, Sweden, Switzerland, theUnited Kingdom, the United States, and Vanuatu.

251


The difference between this projection and the cur-rent fiscal outcome then measured the fiscal disequi-librium to which the authorities reacted withchanges in policy instruments in the coming year.Three policy variables were used: (a) the ex post realinterest rate (the representative nominal interestrate minus the consumer price index, CPI); (b) thenominal effective exchange rate (NEER); and(c) the current account balance as a percentage ofGDP (Table 4).

The set of potential endogenous policy variableswas—using the general-to-specific approach—nar-rowed to five variables: (a) the change in the overallfiscal balance in percentage of GDP; (b) the terms-of-trade index; (c) the oil price (the international

crude oil price in U.S. dollars); (d) the political co-hesion index (a measure of political stability); and(e) an Organization for Economic Cooperation andDevelopment (OECD) intercept dummy (Table 5).The estimated coefficients were statistically signifi-cant and corresponded to basic intuition: higher fis-cal deficits were associated with higher current ac-count deficits; improvements in the terms of tradewere associated with narrower current accountdeficits; looser fiscal policy was associated withtighter monetary policy; developed countries tendedto lower real interest rates; and so on. Only one po-litical economy variable was significant: if one partycontrolled the government, the current account bal-ance was more likely to improve and vice versa.

Table 4

Definitions of Variables

Variable Description Source1

Overall balance Change from the preprogram year; in percentage of GDP WEORevenue and grantsExpenditure and net lending

Real GDP growth Gross domestic product at constant prices; year-on-year change, in percent WEO

GDP per capita Gross domestic product, in constant U.S. dollars WEO

Aid-to-GDP ratio External aid; change from the preprogram period WDI

Inflation rate Consumer price index (CPI); year-on-year change, in percent WEO

Terms of trade Terms of trade of goods and services; year-on-year change, in percent WEO

Index of political cohesion This variable measures the extent to which one party controls both the DPIlegislative and executive branches of the government.

Program stoppage Program stoppage occurs if either (a) the scheduled program review was IMMAnot completed; or (b) all scheduled reviews were completed but the subsequent annual arrangement was not approved.

Current account balance Estimated from the policy reaction function for program countries, actual WEOdata for nonprogram countries; in percentage of GDP

Nominal effective exchange rate Estimated from the policy reaction function for program countries, actual WEOdata for nonprogram countries; change from the preprogram period, in percent

Real interest rate Ex post real money market interest rate; deflated by the CPI; estimated IFSfrom the policy reaction function for program countries, actual data for nonprogram countries; in percent

IMF program dummy 1 if the country had an IMF-supported program during 1993–96, MONA0 otherwise

Measures (count) Number of fiscal measures (narrowly or broadly defined) adjusted for MONAprogram duration

Measures (implementation) Number of implemented fiscal measures (narrowly or broadly defined) MONAadjusted for program duration

1 The abbreviations stand for the following data sources: IMF, World Economic Outlook (WEO); World Bank, World Development Indicators(WDI); World Bank, 2001, Database of Political Institutions (DPI), Version 3.0; Ivanova and others (2003) (IMMA); IMF, International FinancialStatistics (IFS) and Monitoring of IMF Arrangements (MONA) databases.

252


Generalized Evaluation Estimator

We consider three target variables measuring fiscaldevelopments: (a) the overall central governmentbalance; (b) central government revenue andgrants; and (c) central government expenditure andnet lending, all expressed as percentages of GDP, in64 countries that operated under IMF-supportedprograms7 and 48 nonprogram countries during1993–96. While the first target variable is intu-itively preferable to the other variables as a measureof the fiscal stance, revenue and expenditure regres-sions are useful checks of government policies. Theendogenous policy variables stemmed from the pol-icy reaction function and the exogenous variables

were two-year averages, lagged one period: the termsof trade, GDP per capita in constant U.S. dollars,foreign aid in percentage of GDP, the rate of infla-tion, and real GDP growth. Given the inclusion ofthe preprogram fiscal observation, the model in lev-els can be rewritten into one with the dependentvariables in first differences.

This paper is primarily interested in the long-termeffects of IMF-supported programs, knowing that inthe short run, fiscal developments could be affectedby temporary budgetary adjustment in the contextof an IMF-supported arrangement.8 We want tomeasure the impact of IMF-supported programs be-yond the initial, short-term impact; and, hence, weconsidered fiscal variables one, two, and three yearsafter the initial program ended, with 112, 109, and97 observations, respectively. For example, if acountry had a three-year program from January 1993to December 1995, our fiscal variables in the one-,two-, and three-year GEE estimation were dated1997, 1998, and 1999, respectively, with a preprogramobservation of 1992. Thus, we compared program pe-riods of different lengths: the time span between the

Table 5

Estimates of Policy Reaction Function(heteroskedasticity-consistent, feasible GLS regression estimates, t-statistics in parentheses)

Current Nominal Effective Real Dependent Variable Account Balance Exchange Rate Interest Rate

Overall fiscal balance (yt*– yt–1) 0.21346*** 4.56403*** –3.59938**

(6.10) (2.59) (2.50)Terms of trade –0.00005** –0.11601*** –0.01507*

(2.05) (3.50) (1.64)International oil prices 0.66082*** 0.11747**

(3.56) (2.31)Political cohesion 0.00170**

(2.30)Dummy for OECD membership –1.12769***

(5.34)

Wald test of joint parameter significance (χ2) 43.08*** 13.35** 61.96***Log-likelihood 667.0774 –1,012.315 –630.634Number of observations 288 288 288

Source: Authors’ estimates.Notes: All variables, except the Organization for Economic Cooperation and Development (OECD) dummy, are in first differences. The super-

scripts ***, **, and * denote the rejection of the null hypothesis that the estimated coefficient is zero at the 1 percent, 5 percent, and 10 percentsignificance levels, respectively. GLS denotes generalized least squares.

7 The following 31 countries’ IMF-supported program did notcontain any structural conditions: Azerbaijan, Belarus, the Re-public of Congo, Costa Rica, Croatia, the Czech Republic, theDominican Republic, Egypt, El Salvador, Estonia, Georgia, Haiti,Hungary, Jordan, Kazakhstan, Latvia, Lesotho, the former Yu-goslav Republic of Macedonia, Mexico, Moldova, Nicaragua,Panama, Peru, the Philippines, Poland, Romania, Sierra Leone,the Slovak Republic, Turkey, Uganda, and Uzbekistan. Thirty-three countries with structural conditions were as follows (num-bers of fiscal conditions are in parentheses): Albania (10), Alge-ria (3), Benin (8), Bolivia (8), Bulgaria (0), Burkina Faso (14),Cambodia (16), Cameroon (5), the Central African Republic(7), Chad (10), Côte d’Ivoire (8), Ecuador (1), EquatorialGuinea (4), Gabon (1), Ghana (8), Guinea-Bissau (11), Guyana(3), Kenya (4), the Kyrgyz Republic (9), the Lao People’s Demo-cratic Republic (10), Lithuania (0), Malawi (8), Mauritania(13), Mongolia (3), Niger (1), Pakistan (4), Papua New Guinea(5), the Russian Federation (2), Senegal (8), Togo (6), Ukraine(1), Vietnam (3), and Zambia (4).

8 Gupta and others (2002) reported that the probability of a re-versal in fiscal adjustment was as high as 70 percent at the end ofthe second postprogram year. Three possible explanations areavailable for this finding. First, poor fiscal discipline or a lack ofownership caused the reversal. Second, the initial fiscal tighten-ing was excessively tight, necessitating a subsequent fiscal stimu-lus. Finally, the initial adjustment was a mirage: the fiscal author-ity ran arrears vis-à-vis its suppliers, improving the cash balanceand worsening the accrual balance.

preprogram and first postprogram observations wasas short as two years and as long as four years. Fornonprogram countries, we used 1997–99 data and atwo-year average for the “preprogram” period in1991–92.

Results in Full Sample

In general, we find that cyclical variables drove thefiscal developments and that the impact of macro-economic policy variables was comparatively small(Tables 6–8). In all cases, the robust estimators werethe autoregressive terms, real GDP growth, and thereal rate of interest, the stance of monetary policybeing a good measure of the general tightness ofmacroeconomic policies. In some cases, we alsofound inflation and certain conditionality variablesto be significant. The dummy measuring programparticipation was statistically insignificant, implyingthat past IMF-supported programs did not make themedium-term fiscal adjustment softer or stronger—on average, program countries adjusted as much asnonprogram countries. Countries in programs with-out interruptions adjusted somewhat more, butthese results were statistically insignificant.

The lack of in-sample variability in the structuralconditionality variables and their overall substi-tutability suggest that these variables operated morelike a dummy variable. Unlike Ivanova and oth-ers (2003), who looked at performance during IMF-supported programs, we did not find any statisticallysignificant postprogram impact of the political sta-bility variables. Neither did we find any systematicimpact of the type of IMF-supported program, itslength, or the repeated use of IMF credit. The onlystatistically significant regional dummy was the sub-Saharan Africa dummy.

Overall Balance

The change in the postprogram overall balance waspredicted reasonably well by the preprogram overallbalance (a bigger initial deficit was associated with abigger improvement), lagged GDP growth (fastergrowth improved the balance), and the level of de-velopment (countries with higher GDP per capitaimproved their overall balance more than did coun-tries with low GDP per capita)—see Table 6. Thesevariables accounted for almost all of the explainedvariance of the dependent variable (50–60 percent).

Several other variables were either marginally sig-nificant or significant only in some regressions. Oneof them was the aid-to-GDP ratio, indicating somestabilizing impact of foreign aid inflows.9 Moderateinflation was associated with improvements in theoverall balance, while countries with average an-nual inflation of more than 50 percent worsened

their fiscal position. Countries with tighter mone-tary policies had a stronger improvement in theiroverall balances, presumably as a result of generallytighter macroeconomic policies. The IMF programperformance variables were statistically insignificantfor the postprogram period, although the signs oftheir parameters were intuitive. Countries with pro-gram stoppages did worse than the average, whilethose without interruptions did better. The condi-tionality variables were all statistically insignificant.

Revenue and Grants

Revenue regressions explained much less of thevariance of the dependent variable (20–30 percent),even though the results were also dominated by thepreprogram revenue levels and cyclical effects(Table 7). The revenue-to-GDP ratio worsened incountries with larger-than-average initial revenueand it was inversely related to real GDP growth.Both results were intuitive: on the one hand, the taxburden peaked in many countries in the late 1980s,and on the other hand, fast-growing economies didnot need to increase their tax-to-GDP ratios.

The aid-to-GDP ratio was positive, but statisti-cally insignificant in all but the one-year-after-the-program estimates. Inflation worsened revenue inmost regressions—presumably through the Tanzi-Oliveira collection lag—and no nonlinearity in theinflationary impact was found. The real interest ratewas significant and negative, indicating that tightmacroeconomic conditions were not conducive torevenue collection.

We did not find any statistically significant im-pact of IMF-supported programs, although the para-meter signs were consistently negative. Good perfor-mance under the program was linked to improvedrevenue collection by some 2 percent of GDP, butthis marginally significant effect disappeared in thethird year after the program. All but one of the vari-ables describing the quantity of structural measureswere statistically insignificant, although they allcame with a negative sign. The latter results suggestthat revenue-enhancing measures, and perhaps also technical assistance provided to program coun-tries, failed to provide a sustainable increase in therevenue-to-GDP ratio.

Expenditure and Net Lending

The variance of the expenditure-to-GDP ratio wasmostly explained by preprogram expenditure levels,the real rate of growth, and monetary policy(20–30 percent) (Table 8). Unlike in previous regres-sions, we found strong nonlinearity compared withpast expenditure-to-GDP ratios: the expenditure-to-GDP ratio declined in countries with lower-than-average preprogram expenditure ratios, but in-creased in countries with higher-than-average levelsthereof. The former group comprised mostly poorer

253


9 The improvement in the overall balance was partly tautolog-ical, because total revenues included grants, a part of foreign aid.

254

14 � LONG-TERM FISCAL DEVELOPMENTS AND IMF CONDITIONALITY: IS THERE A LINK?Ta

ble

6

The

Ove

rall

Bal

ance

Aft

er t

he E

nd o

f the

Pro

gram

: Est

imat

es o

f GEE

(het

eros

keda

stic

ity-c

onsi

sten

t OLS

, t-s

tatis

tics

in p

aren

thes

es)

One

Yea

r Af

ter

End

of P

rogr

amTw

o Ye

ars

Afte

r En

d of

Pro

gram

Thre

e Ye

ars

Afte

r En

d of

Pro

gram

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(11)

(12)

Con

trol

va

ria

ble

sC

onst

ant

–0.0

274

–0.0

264

–0.0

203

–0.0

372

–0.0

296

–0.0

332

–0.0

308

–0.0

297

–0.0

356

–0.0

433

–0.0

429

–0.0

487

(2.0

9)(2

.47)

(1.9

5)(3

.99)

(1.8

6)(2

.48)

(2.2

3)(4

.52)

(2.9

2)(4

.41)

(4.0

4)(6

.78)

Initi

al v

alue

of t

he d

epen

dent

var

iabl

e–0

.727

6–0

.754

1–0

.721

9–0

.733

7–0

.901

9–0

.907

6–0

.873

4–0

.901

1–1

.053

6–1

.055

3–1

.037

9–1

.039

4(8

.67)

(7.4

7)(9

.21)

(8.2

0)(9

.84)

(9.2

1)(9

.45)

(8.9

1)(7

.45)

(6.9

9)(7

.18)

(7.3

3)La

gged

rea

l GD

P gr

owth

0.00

390.

0040

0.00

390.

0047

0.00

160.

0017

0.00

190.

0020

0.00

250.

0025

0.00

250.

0024

(3.0

9)(3

.12)

(3.3

1)(4

.47)

(1.4

6)(1

.65)

(1.6

6)(2

.36)

(4.0

2)(3

.96)

(3.9

1)(3

.88)

GD

P pe

r ca

pita

6.73

E-7

5.87

E-7

4.32

E-7

9.79

E-7

6.05

E-7

7.13

E-7

6.70

E-7

9.47

E-7

1.27

E-6

1.27

E-6

1.54

E-6

(1.6

8)(1

.67)

(1.0

7)(3

.52)

(1.0

2)(1

.32)

(1.1

8)(1

.46)

(2.1

5)(2

.04)

(4.6

1)Ai

d-to

-GD

P ra

tio0.

0011

0.00

110.

0009

0.00

100.

0011

0.00

110.

0009

0.00

060.

0006

0.00

07(1

.09)

(1.1

1)(0

.95)

(1.6

4)(1

.81)

(1.7

1)(2

.60)

(0.6

67(0

.63)

(0.7

0)La

gged

infla

tion

rate

0.00

020.

0001

0.00

010.

0002

–5.6

2E-5

9.22

E-5

–1.7

4E-5

0.00

010.

0001

0.00

010.

0001

(1.0

9)(2

.14)

(2.1

4)(3

.08)

(0.1

2)(0

.00)

(0.0

4)(2

.39)

(4.1

8)(3

.77)

(3.9

8)H

igh-

infla

tion

dum

my1

–0.0

474

–0.0

366

–0.0

386

–0.0

441

–0.0

015

–0.0

024

–0.0

027

–0.0

043

–0.0

097

–0.0

111

(2.7

1)(2

.43)

(2.5

6)(3

.77)

(0.0

4)(0

.06)

(0.0

7)(0

.17)

(0.4

4)(0

.50)

Lagg

ed te

rms

of tr

ade

–0.0

007

–0.0

007

–0.0

008

–0.0

001

–0.0

001

–0.0

002

0.00

020.

0002

0.00

010.

0002

(1.2

6)(1

.16)

(1.4

4)(0

.42)

(0.5

0)(0

.68)

(2.8

1)(2

.42)

(0.9

5)(3

.41)

Pol

icy

vari

ab

les

Real

inte

rest

rat

e1.

68E-

61.

41E-

61.

48E-

63.

10E-

62.

24E-

62.

20E-

63.

09E-

63.

73E-

63.

15E-

63.

04E-

62.

92E-

6(1

.18)

(1.1

3)(1

.19)

(3.2

4)(2

.92)

(2.9

2)(4

.03)

(2.8

8)(4

.12)

(3.5

8)(5

.26)

Nom

inal

exc

hang

e ra

te2.

20E-

55.

11E-

62.

13E-

5–6

.75E

-5–7

.25E

-5–6

.23E

-50.

0002

0.00

020.

0002

0.00

02(0

.33)

(0.0

9)(0

.37)

(0.9

8)(0

.93)

(0.8

3)(3

.88)

(3.7

8)(3

.43)

(3.9

1)C

urre

nt a

ccou

nt b

alan

ce0.

1227

0.12

340.

1062

0.06

360.

0636

0.05

060.

0505

0.04

560.

0419

(1.1

8)(1

.20)

(1.1

3)(0

.92)

(0.9

0)(0

.78)

(0.6

0)(0

.53)

(0.5

3)

IMF

pro

gra

m p

erfo

rma

nce

IMF

prog

ram

dum

my

0.01

570.

0132

0.00

23(0

.89)

(0.8

2)(0

.14)

Prog

ram

sto

ppag

e–0

.011

4–0

.017

9–0

.017

7(0

.85)

(1.6

2)(1

.07)

“Suc

cess

ful I

MF

prog

ram

” du

mm

y20.

0155

0.01

450.

0049

(1.0

1)(1

.35)

(0.4

2)C

ond

itio

na

lity

va

ria

ble

sFi

scal

mea

sure

s (c

ount

)3–0

.021

8–0

.031

50.

0301

(0.5

5)(0

.86)

(0.5

7)

R2

0.51

50.

515

0.50

90.

453

0.58

30.

579

0.57

30.

529

0.63

00.

621

0.62

20.

614

Log-

likel

ihoo

d19

5.5

195.

619

4.8

188.

820

5.01

204.

520

3.7

198.

417

3.6

172.

517

2.6

171.

7N

umbe

r of

obs

erva

tions

112

112

112

112

109

109

109

109

9797

9797

Nor

mal

ity te

st [

χ2(2

,2)]

64.2

564

.72

75.9

013

1.68

9.73

12.5

711

.98

26.6

632

.55

32.9

434

.24

48.0

6H

eter

oske

dast

icity

test

(F)

1.70

1.89

2.10

0.81

0.95

0.97

0.69

0.66

0.50

0.46

0.54

0.67

Sour

ce: A

utho

rs’ e

stim

ates

.N

otes

: G

EE d

enot

es G

ener

aliz

ed E

valu

atio

n Es

timat

or. O

LS d

enot

es o

rdin

ary

leas

t sq

uare

s. V

ery

smal

l num

bers

are

den

oted

as

E-k,

whe

re E

indi

cate

s th

e ba

se o

f 10

and

k in

dica

tes

the

posi

tion

of t

he

deci

mal

poi

nt.

1 Th

e du

mm

y ta

kes

the

valu

e of

1 if

the

lagg

ed, t

wo-

year

ave

rage

infla

tion

was

hig

her

than

50

perc

ent a

yea

r; an

d 0

othe

rwis

e.2

The

dum

my

is e

qual

to 1

if e

ither

all

com

mitt

ed r

esou

rces

wer

e di

sbur

sed

or if

the

prog

ram

was

pre

caut

iona

ry; a

nd 0

oth

erw

ise.

3 In

clud

es a

ll st

ruct

ural

mea

sure

s w

ith fi

scal

impl

icat

ions

.

255


Tabl

e 7

Rev

enue

and

Gra

nts

Aft

er E

nd o

f Pro

gram

: Est

imat

es o

f GEE

(het

eros

keda

stic

ity-c

onsi

sten

t OLS

, t-s

tatis

tics

in p

aren

thes

es)

One

Yea

r Af

ter

End

of P

rogr

amTw

o Ye

ars

Afte

r En

d of

Pro

gram

Thre

e Ye

ars

Afte

r En

d of

Pro

gram

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(11)

(12)

Con

trol

va

ria

ble

sC

onst

ant

0.04

520.

0395

0.04

380.

0321

0.04

830.

0462

0.04

710.

0425

0.05

820.

0420

0.04

200.

0514

(2.9

5)(2

.98)

(3.1

6)(2

.74)

(4.3

1)(3

.85)

(3.9

6)(3

.99)

(4.1

4)(3

.65)

(3.7

1)(4

.35)

Initi

al v

alue

of t

he d

epen

dent

var

iabl

e–0

.120

3–0

.127

8–0

.118

0–0

.111

5–0

.139

0–0

.139

0–0

.139

0–0

.131

7–0

.140

3–0

.131

0–0

.135

5–0

.143

4(2

.24)

(2.4

0)(2

.20)

(1.9

8)(3

.16)

(3.1

2)(3

.11)

(2.9

2)(2

.89)

(2.6

0)(2

.72)

(2.9

0)La

gged

rea

l GD

P gr

owth

–0.0

016

–0.0

014

–0.0

017

–0.0

015

–0.0

015

–0.0

015

–0.0

019

–0.0

019

–0.0

019

–0.0

017

–0.0

022

(0.8

5)(0

.75)

(0.8

8)(1

.43)

(1.4

4)(1

.43)

(1.6

3)(2

.42)

(2.2

7)(1

.91)

(3.0

1)G

DP

per

capi

ta4.

28E-

82.

97E-

77.

16E-

83.

38E-

84.

71E-

86.

36E-

10–3

.85E

-81.

64E-

71.

77E-

7(0

.09)

(0.7

0)(0

.18)

(0.0

7)(0

.09)

(0.0

0)(0

.75)

(0.3

4)(0

.40)

Aid-

to-G

DP

ratio

0.00

180.

0019

0.00

160.

0017

0.00

060.

0007

0.00

060.

0008

0.00

070.

0007

(3.2

1)(3

.60)

(2.9

0)(3

.71)

(0.9

1)(0

.99)

(0.8

0)(1

.41)

(1.0

9)(1

.04)

Lagg

ed in

flatio

n ra

te–0

.000

2–0

.000

2–0

.000

2–0

.000

2–0

.000

6–0

.000

6–0

.000

6–0

.000

6–2

.07E

-5–4

.28E

-5–3

.63E

-5(1

.53)

(1.4

3)(1

.58)

(2.0

6)(2

.15)

(2.3

5)(2

.37)

(2.7

4)(0

.52)

(0.9

3)(0

.82)

Lagg

ed te

rms

of tr

ade

0.00

020.

0003

0.00

01–0

.000

3–0

.000

3–0

.000

23.

51E-

56.

29E-

68.

17E-

5(0

.34)

(0.6

1)(0

.14)

(1.9

5)(2

.10)

(1.5

3)(0

.33)

(0.0

5)(0

.55)

Pol

icy

vari

ab

les

Real

inte

rest

rat

e–5

.13E

-6–4

.96E

-6–4

.50E

-6–4

.53E

-6–4

.95E

-6–4

.91E

-6–4

.75E

-6–4

.82E

-6–3

.69E

-6–3

.06E

-6–2

.89E

-6–4

.16E

-6(2

.36)

(2.3

6)(2

.34)

(2.5

2)(4

.20)

(3.8

6)(3

.87)

(3.6

1)(3

.54)

(3.6

9)(3

.72)

(3.8

7)N

omin

al e

xcha

nge

rate

–4.1

4E-5

–6.3

1E-5

–4.7

6E-5

–7.7

5E-5

–7.8

9E-5

–8.1

4E-5

5.46

E-5

7.40

E-5

7.24

E-5

(0.8

1)(1

.06)

(0.8

9)(1

.55)

(1.5

3)(1

.62)

(0.7

7)(1

.01)

(0.9

7)C

urre

nt a

ccou

nt b

alan

ce0.

0303

0.05

320.

0246

0.02

120.

0253

0.02

010.

0328

0.03

460.

0391

(0.6

7)(1

.19)

(0.5

2)(0

.44)

(0.5

3)(0

.41)

(0.9

3)(0

.93)

(1.0

5)

IMF

pro

gra

m p

erfo

rma

nce

IMF

prog

ram

dum

my

–0.0

051

–0.0

027

–0.0

216

–0.0

210

(0.4

6)(0

.23)

(1.9

9)(2

.34)

“Suc

cess

ful I

MF

prog

ram

” du

mm

y10.

0170

0.00

20–0

.005

3(1

.31)

(1.1

7)(0

.48)

Con

dit

ion

ali

ty v

ari

ab

les

Reve

nue

mea

sure

s (c

ount

)2–0

.026

1–0

.009

1–0

.018

8(1

.46)

(0.7

3)(0

.81)

R2

0.21

40.

228

0.23

90.

189

0.23

50.

234

0.23

80.

195

0.26

60.

239

0.24

20.

240

Log-

likel

ihoo

d19

1.8

192.

819

3.6

190.

019

7.9

197.

919

8.1

195.

217

5.1

173.

317

3.5

173.

4N

umbe

r of

obs

erva

tions

112

112

112

112

109

109

109

109

9797

9797

Nor

mal

ity te

st [

χ2(2

,2)]

32.5

529

.71

29.2

144

.14

20.6

219

.71

20.6

622

.14

18.3

416

.13

15.7

719

.40

Het

eros

keda

stic

ity te

st (

F)0.

990.

920.

742.

480.

300.

300.

251.

340.

180.

182.

750.

77

Sour

ce: A

utho

rs’ e

stim

ates

.N

otes

: G

EE d

enot

es G

ener

aliz

ed E

valu

atio

n Es

timat

or;

OLS

den

otes

ord

inar

y le

ast

squa

res.

Ver

y sm

all n

umbe

rs a

re d

enot

ed a

s E-

k, w

here

E in

dica

tes

the

base

of

10 a

nd k

indi

cate

s th

e po

sitio

n of

the

de

cim

al p

oint

.1

The

dum

my

is e

qual

to 1

if e

ither

all

com

mitt

ed r

esou

rces

wer

e di

sbur

sed

or if

the

prog

ram

was

pre

caut

iona

ry; a

nd 0

oth

erw

ise.

2 In

clud

es a

ll st

ruct

ural

mea

sure

s w

ith r

even

ue-im

prov

ing

impl

icat

ions

.

256


Tabl

e 8

Expe

nditu

re a

nd N

et L

endi

ng A

fter

End

of P

rogr

am: E

stim

ates

of G

EE(h

eter

oske

dast

icity

-con

sist

ent O

LS, t

-sta

tistic

s in

par

enth

eses

)

One

Yea

r Af

ter

End

of P

rogr

amTw

o Ye

ars

Afte

r En

d of

Pro

gram

Thre

e Ye

ars

Afte

r En

d of

Pro

gram

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(11)

(12)

Con

trol

va

ria

ble

sC

onst

ant

–0.3

782

–0.3

864

–0.4

089

–0.4

362

–0.1

969

–0.1

958

–0.2

240

–0.2

897

–0.2

080

–0.2

133

–0.2

382

–0.2

695

(3.6

6)(3

.88)

(4.0

6)(4

.09)

(1.6

7)(1

.72)

(1.8

0)(2

.45)

(1.8

1)(1

.90)

(2.0

7)(2

.69)

Initi

al v

alue

of d

epen

dent

var

iabl

e–1

.912

7–1

.857

0–2

.024

0–2

.171

1–1

.216

2–1

.190

8–1

.367

5–1

.582

7–1

.377

3–1

.350

4–1

.573

7–1

.635

4(4

.26)

(4.6

0)(4

.82)

(4.8

4)(2

.90)

(2.8

6)(3

.07)

(3.7

1)(3

.32)

(3.3

4)(3

.90)

(4.7

9)In

itial

val

ue o

f dep

ende

nt v

aria

ble,

squa

red

1.76

201.

7113

1.88

822.

0133

1.03

741.

0112

1.19

441.

4028

1.16

971.

1443

1.35

501.

4177

(3.9

8)(4

.27)

(4.4

9)(4

.48)

(2.3

5)(2

.32)

(2.5

2)(3

.07)

(3.9

8)(2

.66)

(3.1

3)(3

.77)

Lagg

ed r

eal G

DP

grow

th–0

.003

5–0

.003

2–0

.003

6–0

.002

4–0

.001

6–0

.001

8–0

.001

7–0

.004

5–0

.004

5–0

.003

8–0

.003

6(2

.05)

(2.0

3)(2

.20)

(1.8

0)(0

.97)

(1.0

2)(1

.15)

(4.1

4)(4

.15)

(4.2

7)(5

.94)

GD

P pe

r ca

pita

–9.8

8E-9

6.69

E-7

–1.7

6E-7

–3.5

8E-7

–1.4

2E-7

–8.5

3E-7

–8.4

2E-7

–4.0

8E-7

–1.4

1E-7

(0.0

2)(1

.33)

(0.3

7)(0

.52)

(0.2

5)(1

.40)

(1.4

3)(0

.85)

(2.9

5)La

gged

infla

tion

rate

–0.0

001

–0.0

001

–0.0

001

–0.0

005

–0.0

005

–0.0

005

–0.0

001

–0.0

001

–1.1

6E-5

(1.3

1)(1

.33)

(1.7

0)(1

.31)

(1.3

0)(1

.51)

(1.5

8)(1

.73)

(0.2

2)P

olic

y va

ria

ble

sRe

al in

tere

st r

ate

–5.2

6E-6

–4.9

7E-6

–3.4

8E-6

–1.7

3E-6

–4.0

3E-6

–4.1

3E-6

–2.7

2E-6

–1.8

0E-6

–3.6

5E-6

–3.5

4E-6

–1.4

7E-6

–2.5

7E-6

(2.7

8)(2

.90)

(2.7

4)(1

.70)

(2.6

5)(2

.57)

(2.5

0)(1

.72)

(2.5

9)(2

.65)

(1.6

5)(2

.47)

Nom

inal

exc

hang

e ra

te–1

.58E

-5–2

.94E

-5–1

.97E

-53.

63E-

52.

95E-

51.

95E-

5–4

.34E

-5–3

.96E

-5–2

.53E

-5(0

.21)

(0.3

8)(0

.36)

(0.9

0)(0

.69)

(0.5

8)(0

.44)

(0.3

8)(0

.29)

Cur

rent

acc

ount

bal

ance

–0.1

054

–0.0

887

–0.0

953

–0.0

361

–0.0

246

–0.0

282

–0.0

131

0.00

98–0

.016

5(1

.66)

(1.2

4)(1

.71)

(0.7

5)(0

.51)

(0.6

4)(0

.22)

(0.1

8)(0

.40)

IMF

pro

gra

m p

erfo

rma

nce

IMF

prog

ram

dum

my

–0.0

191

–0.0

043

–0.0

108

(1.0

0)(0

.25)

(0.5

4)“S

ucce

ssfu

l IM

F pr

ogra

m”

dum

my1

0.01

380.

0107

0.01

80(0

.88)

(0.6

7)(0

.85)

Con

dit

ion

ali

ty v

ari

ab

les

Stru

ctur

al c

ondi

tiona

lity

(dum

my)

2–0

.040

7–0

.038

9–0

.036

8–0

.032

4–0

.062

0–0

.049

7(3

.10)

(3.1

9)(3

.20)

(4.0

1)(4

.49)

(4.1

2)

R2

0.45

40.

449

0.50

50.

470

0.31

90.

322

0.37

10.

337

0.44

80.

452

0.53

40.

503

Log-

likel

ihoo

d17

6.8

176.

318

2.3

178.

416

6.7

166.

917

1.1

168.

214

4.1

144.

615

2.5

149.

3N

umbe

r of

obs

erva

tions

112

112

112

112

109

109

109

109

9797

9797

Nor

mal

ity te

st [

χ2(2

,2)]

16.4

124

.12

9.26

22.8

226

.83

25.5

423

.63

23.2

221

.23

16.3

418

.94

26.6

2H

eter

oske

dast

icity

test

(F)

1.36

1.05

1.02

1.55

0.52

0.77

0.64

0.96

0.72

1.48

0.92

0.39

Sour

ce: A

utho

rs’ e

stim

ates

.N

otes

: G

EE d

enot

es G

ener

aliz

ed E

valu

atio

n Es

timat

or;

OLS

den

otes

ord

inar

y le

ast

squa

res.

Ver

y sm

all n

umbe

rs a

re d

enot

ed a

s E-

k, w

here

E in

dica

tes

the

base

of

10 a

nd k

indi

cate

s th

e po

sitio

n of

the

de

cim

al p

oint

.1

The

dum

my

is e

qual

to 1

if e

ither

all

com

mitt

ed r

esou

rces

wer

e di

sbur

sed

or if

the

prog

ram

was

pre

caut

iona

ry; a

nd 0

oth

erw

ise.

2 Th

e du

mm

y ta

kes

the

valu

e of

1 if

the

IMF-

supp

orte

d pr

ogra

m in

clud

ed a

ny s

truc

tura

l con

ditio

ns; a

nd 0

oth

erw

ise.

257


countries with structural conditionality programs,while the latter group comprised richer countrieswith nonstructural conditionality. Countries thatgrew faster and those with tight monetary policiesalso lowered their expenditure-to-GDP ratios.

We did not find any statistically significant im-pact of IMF-supported programs on expenditure de-velopments. The structural conditionality variableswere negative and significant, suggesting relativeexpenditure compression in countries with struc-tural conditionality of 2 percentage points of GDPor more. It is problematic to distinguish whether ex-penditures that were cut were wasteful or whetherthe compression was excessive. We can only conjec-ture that the gradually increasing value of the struc-tural conditionality parameter points to the formerexplanation, as expenditure compression acceler-ated after the end of the IMF arrangement. This ob-servation is also consistent with a body of evidencethat social and capital spending were protected dur-ing the program’s existence (Abed and others, 1998).

Are Countries with Programs Containing Structural Conditionality “Different”?

The finding that conditionality variables were in-significant for all but the expenditure regressions ispuzzling. We do not see a unique explanation forthese findings, as they can be justified by alternativerelationships. First, these results may imply thatIMF-supported programs mechanically compensated

with additional conditionality for historically poorperformance, owing to deep-rooted structural weak-nesses, or persistent shocks, or a lack of a reformdrive, or a combination of all of those. Without ad-dressing the causes of the past performance, addi-tional conditions would not affect the fiscal perfor-mance. Second, IMF conditionality and donortechnical assistance in the fiscal area may have failedto bring about sustained fiscal improvements, espe-cially if the reforms were not supported by the public.

To understand better the developments in struc-tural conditionality countries, we reestimated our re-gressions for the program countries only (Table 9).10

While the size and signs of the individual coeffi-cients were broadly unchanged compared to Ta-bles 6–8, their statistical significance declined pre-dictably with the loss of degrees of freedom. Wefound that the overall balance improvement waslarger in countries with structural conditionalitythan in other program countries by about 1⁄2 and3 percentage points of GDP. At the same time, rev-enue and grants declined by 2 additional percentagepoints of GDP in structural conditionality countries.Finally, the expenditure and net lending compres-sion increased in structural conditionality countriesover time—from about 2 percentage points of GDPone year after the program to 8 percentage points of

10 The sample sizes for one-, two-, and three-year-after-the-program regressions were 64, 61, and 49 observations, respectively.The full set of results is available on request from the authors.

Table 9

Fiscal Developments in Structural Conditionality Countries Relative to NonstructuralConditionality Countries(heteroskedasticity-consistent OLS regression estimates of structural conditionality and sub-Saharan Africadummies, t-statistics in parentheses)

Overall Revenue Expenditure and Balance and Grants Net Lending

SC dummy –0.0018 –0.0106 –0.0192One-year-after- (0.17) (0.72) (1.15)program sample Africa dummy 0.0327** 0.0258* –0.0116

(2.33) (1.92) (0.79)

SC dummy 0.0163 –0.0224* –0.0564***Two-years-after- (1.28) (1.73) (2.94)program sample Africa dummy –0.0035 0.0203 0.0229

(0.26) (1.32) (1.02)

SC dummy 0.0325** –0.0286*** –0.0810***Three-years-after- (2.46) (3.30) (4.42)program sample Africa dummy –0.0126 0.0565*** 0.0714***

(0.54) (4.27) (3.33)

Source: Authors’ estimates.Notes: The superscripts ***, **, and * denote the rejection of the null hypothesis that the estimated coefficient is zero at the 1 percent, 5 per-

cent, and 10 percent significance levels, respectively. OLS denotes ordinary least squares. SC denotes structural conditionality.

GDP three years after the program—and these re-sults were statistically significant.

We also checked for the presence of fiscal rever-sals in low-income countries and found this effect tobe at work only for the sub-Saharan Africa region.While African countries started with a better-than-average postprogram overall balance of more than3 percent of GDP, this result disappeared in the sec-ond year after the end of the program period. Onthe expenditure side, the sub-Saharan average wasstatistically indistinguishable from the rest initially,but by the third year expenditures were higher thanthe average by 7 percentage points of GDP. Revenueperformance in sub-Saharan Africa was better thanaverage, although not sufficient to offset the expen-diture increase.

These results seem to suggest that countries withstructural conditionality were indeed different fromthe other program countries. First, they were subjectto more pronounced shocks than were other pro-gram countries; for example, their terms of tradewere twice as volatile. Second, the effort to addressrevenue weaknesses in those countries throughstructural conditionality failed, most likely becauseconditionality was a poor substitute for homegrownreform. Finally, postprogram fiscal performance inthose countries was driven by accelerating expendi-ture compression, which may not be a bad thing,provided, for example, that the preprogram level ofspending was wasteful or that a statist budget was re-placed with a less intrusive one.

Concluding Remarks

This paper presents empirical tests of the relevanceof IMF structural conditionality for postprogram fis-cal performance in a large sample of countries dur-ing the 1990s. Although the overall balance improvedin most countries, the impact of IMF-supported pro-grams was not statistically significant, owing to thelarge variance in the sample of program countries.In structural conditionality countries, revenue de-clined slightly and expenditure declined signifi-cantly. In contrast, in countries that had nonstruc-tural conditionality programs, revenue remainedstable and expenditure increased somewhat. Thepostprogram statistical insignificance of IMF-sup-ported programs indicates that program participa-tion did not make the fiscal adjustment automati-cally softer—on average, program countries adjustedas much as nonprogram countries and fiscal adjust-ment continued in most countries even after thecompletion of the IMF-supported arrangement. Thebusiness cycle strongly influenced all fiscal variables,and an impact of the general macroeconomic stancewas detectable as well.

Our results highlight the difficulty in identifyingthe impact of structural conditionality. Several ef-fects seem to be in play. First, we found some evi-

dence that programs with too many structural con-ditions had worse postprogram results than thosewith fewer program conditions. Second, we foundno quantitative evidence that structural condition-ality aimed at raising revenue was successful. Third,postprogram expenditure compression clearly wasmuch stronger in countries with structural condi-tionality, but the risk of reversal was higher too, es-pecially in sub-Saharan Africa.

The findings in this paper are not definitive andthe possibilities for further research are extensive.First, more work is needed to examine the role ofinitial shocks, structural weaknesses, political econ-omy, and regime-specific effects, such as the choiceof the exchange rate regime. Second, the policy re-action function can be specified differently, reflect-ing, for example, policies that would stabilize thedebt-to-GDP ratio or that would be based on “fiscalrules.” Finally, some of the issues, such as the appro-priateness of the initial revenue and expenditurelevels, cannot be addressed adequately in a cross-country model and need to be investigated in casestudies.

ReferencesAbed, George T., Liam Ebrill, Sanjeev Gupta, Benedict

Clements, Ronald McMorran, Anthony Pellechio,Jerald Schiff, and Marijn Verhoeven, 1998, Fiscal Re-forms in Low-Income Countries, IMF Occasional PaperNo. 160 (Washington: International Monetary Fund).

Barro, Robert J., and Jong-Wha Lee, 2002, “IMF Pro-grams: Who Is Chosen and What Are the Effects?”NBER Working Paper No. 8951 (Cambridge, Massa-chusetts: National Bureau of Economic Research).

Bird, Graham, 1996, “Borrowing from the IMF: The Policy Implications of Recent Empirical Research,”World Development, Vol. 24, No. 11, pp. 1753–60.

______, 2001, “IMF Programmes: Is There a Conditional-ity Laffer Curve?” World Economics, Vol. 2, No. 2,pp. 29–49.

______, and Dane Rowlands, 2002, “Do IMF ProgrammesHave a Catalytic Effect on Other Capital Flows?” Oxford Development Studies, Vol. 30, No. 3, pp. 229–49.

Bulír, Ales, and A. Javier Hamann, 2003, “Aid Volatility:An Empirical Assessment,” Staff Papers, Interna-tional Monetary Fund, Vol. 50, No. 1, pp. 64–89.

Bulír, Ales, and Soojin Moon, 2004, “Is Fiscal Adjust-ment More Durable When the IMF is Involved?”Comparative Economic Studies, Vol. 46 (September),pp. 373–99.

Collier, Paul, Patrick Guillaumont, Sylviane Guillau-mont, and Jan W. Gunning, 1997, “RedesigningConditionality,” World Development, Vol. 25, No. 9,pp. 1399–1407.



258


259


Drazen, Allen, 2000, Political Economy in Macroeconomics(Princeton, New Jersey: Princeton University Press).

Ghosh, Atish, Timothy Lane, Marianne Schulze-Ghattas,Ales Bulír, A. Javier Hamann, and Alex Mour-mouras, 2002, IMF-Supported Programs in Capital Account Crises, IMF Occasional Paper No. 210(Washington: International Monetary Fund).

Goldstein, Morris, 2000, “IMF Structural Programs,”paper prepared for National Bureau of Economic Re-search Conference on “Economic and FinancialCrises in Emerging Market Economies,” held inWoodstock, Vermont in October.

_______, and Peter J. Montiel, 1986, “Evaluating FundStabilization Programs with Multicountry Data:Some Methodological Pitfalls,” Staff Papers, Interna-tional Monetary Fund, Vol. 33 (June), pp. 304–44.

Guitián, Manuel, 1981, Fund Conditionality: Evolution ofPrinciples and Practices, IMF Pamphlet Series, No. 38(Washington: International Monetary Fund).

Gupta, Sanjeev, Benedict Clements, Emanuele Baldacci,and Carlos Mulas-Granados, 2002, “ExpenditureComposition, Fiscal Adjustment, and Growth inLow-Income Countries, IMF Working Paper 02/77(Washington: International Monetary Fund). Thispaper also appears, in somewhat different form, asChapter 5 in this volume.

Hansson, Åsa, and Charles Stuart, 2003, “Peaking of Fis-cal Sizes of Government,” European Journal of PoliticalEconomy, Vol. 19 (November), pp. 669–84.

International Monetary Fund (IMF), 2001, Conditionalityin Fund-Supported Programs—Overview. Available on

the Web at http://www.imf.org/external/np/pdr/cond/2001/eng/overview/index.htm.

Ivanova, Anna, Wolfgang Mayer, Alex Mourmouras, andGeorge Anayiotos, 2003, “What Determines theSuccess or Failure of Fund-Supported Programs?”IMF Working Paper 03/8 (Washington: InternationalMonetary Fund). This paper also appears, in some-what different form, as Chapter 10 in this volume.

Khan, Mohsin S., 1990, “The Macroeconomic Effects of Fund-Supported Adjustment Programs,” Staff Papers, International Monetary Fund, Vol. 37 (June),pp. 195–231.

________, and Sunil Sharma, 2001, “IMF Conditionalityand Country Ownership of Programs,” IMF WorkingPaper 01/142 (Washington: International MonetaryFund). This paper also appears, in somewhat differentform, as Chapter 7 in this volume.

Knight, Malcolm, and Julio A. Santaella, 1997, “Eco-nomic Determinants of IMF Financial Arrange-ments,” Journal of Development Economics, Vol. 54(December), pp. 405–36.

Schadler, Susan, Adam Bennett, Maria Carkovic, LouisDicks-Mireaux, Mauro Mecagni, James H.J. Morsink,and Miguel A. Savastano, 1995, IMF Conditionality:Experience Under Stand-By and Extended Arrange-ments, IMF Occasional Paper No. 128 (Washington:International Monetary Fund).

World Bank, 2001, “Database of Political Institutions,”Version 3.0 (Washington: World Bank).

260

15

This paper contributes to the literature on evaluating theimpact of IMF-supported programs on three key macro-economic variables: inflation, the budget position, andgrowth. The paper documents the importance of distin-guishing between completed programs and those that ter-minated prematurely, showing that they have signifi-cantly different impacts on these targets. While stoppedprograms are associated with higher inflation and budgetdeficits, and lower growth relative to periods without aprogram, completed programs are marginally associatedwith increased growth three years after the terminationof the program.

Introduction

During the past two decades, a number of studieshave explored whether IMF programs are effectivein improving participating countries’ current ac-count, inflation, and growth outcomes. These stud-ies developed different methodologies and used vari-ous datasets, and while coming to differentconclusions on specific economic variables, the gen-eral thrust of the results is that IMF programs do notgenerally affect growth or inflation. This paper aimsto provide some new insights on the effectiveness ofIMF programs by using a new database for countriesthat did not participate in any IMF programs duringthe 1980s but were engaged in one or more pro-grams in the 1990s. The sample characteristics ofthe data were chosen to better capture the indepen-dent effect of IMF programs on three target vari-ables: inflation, budget conditions, and growth.

One of the major novelties of this approach is thatit emphasizes the importance of implementing IMFprograms by distinguishing between programs thatwere implemented successfully and those that werestopped prematurely. Surprisingly, this distinctionhas been largely absent from previous analyses of theeffects of IMF programs. Indeed, stoppages are fairlyprevalent in IMF programs, amounting to almost

40 percent of all programs during the 1992–2001 pe-riod. Distinguishing between IMF programs thatwere implemented successfully and those that brokedown is essential in properly evaluating the effec-tiveness of IMF programs. The only study that previ-ously addressed this issue was Killick (1995), whichdistinguished between these two groups by using theratio of disbursements to committed amounts ascontrols.

The question of whether IMF-supported programshave significant independent effects on the macro-economic outcomes of particular countries is diffi-cult to answer because it requires the constructionof a counterfactual indicating what policies and out-comes would have resulted in the absence of an IMFprogram. Since the mid-1980s, several papers haveconsidered how to construct a counterfactual forsuch exercises through differentiating the effects ofcounterfactual policies from exogenous develop-ments, initial conditions, and IMF support. Themethodology that has been most widely applied, de-veloped by Goldstein and Montiel (1986), uses pol-icy reaction functions estimated for countries thatdid not have support from a particular internationalfinancial institution (IFI) to approximate the coun-terfactual for countries that did have IFI backing fortheir programs. Unfortunately, several recent studieshave cast doubt on the appropriateness and reliabil-ity of this method. For instance, Dicks-Mireaux,Mecagni, and Schadler (2000) did diagnostic tests toshow that the assumptions underlying the methodol-ogy may not hold.

Another method of evaluating the effects of IMFprograms is the before-after approach, in whichcountries are evaluated for a number of years beforeand after the onset of an IMF program to identifywhether IMF advice is able to improve the macro-economic situation. The major problem with thesimple before-after approach is that the economiccondition of the country might have improved any-way without the presence of an IMF-supported pro-gram. Failure to take this into account leads to bi-ased estimation of the coefficient associated withIMF programs.

This paper uses the two-step Heckman (1979)procedure to control for the bias caused by partici-pation in an IMF program. In the first step, a probit

Evaluating IMF-Supported Programs in the 1990s: The Importance of Taking Explicit Account of Stoppages Chuling Chen and Alun Thomas1

1 The authors are grateful for helpful comments from the par-ticipants of various seminars held at the IMF and at the WesternEconomic Association International Annual Meetings held inDenver, Colorado in July 2003. Any errors that remain, however,are their sole responsibility.

CHAPTER

261

Chuling Chen and Alun Thomas

model is used to capture the choice of participatingin a program based on changes in initial economicand political conditions. A composite variable, theinverse Mill’s ratio (IMR), summarizing this choice isthen introduced into the second-stage least-squaresregression of the determinants of the various macro-economic targets. This variable gives an estimatedprobability of program participation and thereforecontrols for nonrandom selection into IMF programsby holding fixed an estimated probability for this se-lection. Similar treatment is also given to the stop-page variable to control for the possible endogeneityproblem—that is, the possibility that a programmight be terminated prematurely owing to unsatis-factory performance of the macroeconomic targetsthat are under examination. This distinction sepa-rates out the effects of IMF programs from the tim-ing of a program, and provides insights on evaluat-ing whether stopped programs are any different.

Since the sample splits between the 1980s and1990s, it is possible that any improvement in themacroeconomic targets during the 1990s could berelated to a more stable economic environmentrather than to the availability of IMF advice. Tocontrol for this effect, time dummies are included inthe specification so that any improvement associ-ated with an IMF-supported program is independentof the strength of the world economic cycle.

Our findings demonstrate significant differencesbetween successful and stopped programs, althoughthe macro targets associated with successful pro-grams are no different from their preprogram values.While stopped programs raise inflation and the bud-get deficit, and lower growth relative to periodswithout a program, completed programs are margin-ally associated with increased growth three yearsafter the termination of the program.

Our paper is organized as follows: the second sec-tion reviews the related literature; the third sectiondescribes the data, model specification, and esti-mates; and the fourth section presents our conclu-sion.

Related Studies

Studies that have tried to identify the factors thatinduce countries to initiate IMF-supported programshave found that deteriorating external conditionssuch as the balance of payments, the debt position,and rapid exchange rate movements play an impor-tant role in the timing of programs. Conway (1994)used both probit and tobit models to analyze theparticipation of 74 countries in Stand-By Arrange-ments (SBAs) and Extended Fund Facilities (EFFs)during 1976–86 and found that past participation ofIMF programs, real GDP growth, and external fac-tors (terms of trade, current account, long-term ex-ternal debt) were significant determinants of thetiming of IMF involvement. Joyce (2002) used a

panel dataset for 45 countries for 1980–84 and iden-tified the ratio of government expenditures to GDPand the reserves-import ratio as significant factors.Edwards and Santaella (1992) used 48 devaluationperiods in developing countries and identified GDPper capita and the ratio of net foreign assets to themoney supply as the most important factors. Knightand Santaella (1997) employed a bivariate probitmodel to examine 91 non-oil countries for 1973–91.They concluded that the level of international re-serves and GDP per capita were the most importantfactors in influencing a county’s decision to negoti-ate programs while revenue and expenditurechanges, domestic credit, and exchange rate move-ments were highlighted by the IMF.

In terms of the effects of IMF programs on keymacroeconomic indicators, studies have found thatIMF programs lead to immediate improvements inthe current account and overall balance of pay-ments, but do not have a consistent impact on infla-tion and economic growth. For example, Khan(1990), Schadler and others (1993), Conway(1994), Killick (1995), and Dicks-Mireaux,Mecagni, and Schadler (2000) all find a negative re-lationship between IMF programs and inflation, butthe estimated effect is significant only in Killick’sanalysis. Some studies find a significant positive re-lationship with respect to growth in the short term(Killick, 1995; Bagci and Perraudin, 1997; andDicks-Mireaux, Mecagni, and Schadler, 2000) andin the long term (i.e., three years after the pro-gram—see, for example, Conway, 1994) whereasothers, in particular, Khan (1990) and Przeworskiand Vreeland (2000), find significant negativegrowth effects both in the short and long term. Con-cerning budgetary conditions, Schadler and others(1993) find that fiscal deficits fall during IMF pro-grams, while Bulír and Moon (2002) are unable toidentify significant effects. A summary of the find-ings can be found in Table 1.

Data and Empirical Results

Data

The data choice to highlight differences betweenstopped and completed programs was based on re-quiring a fairly long period to evaluate the effective-ness of programs. In the past, a number of studieshave isolated specific years in which an IMF pro-gram did or did not take place. However, since themacroeconomic objectives of IMF programs gener-ally have a much longer duration than one year, itseemed appropriate to ensure a fairly long periodbetween engaging and not engaging in IMF pro-grams. With these objectives in mind, the countriesused in this paper are those that had IMF programsduring the 1990s but did not have any type of IMF

262

15 � EVALUATING IMF-SUPPORTED PROGRAMS IN THE 1990S

program between 1980 and 1988: Algeria, Benin,Bulgaria, Burkina Faso, Cambodia, Cape Verde,Colombia, Djibouti, Indonesia, Jordan, Mongolia,Nicaragua, Papua New Guinea, Poland, Rwanda,República Bolivariana de Venezuela, and the Re-public of Yemen. Since Nicaragua and Cambodiawere involved in civil wars during the 1980s, andthe Republic of Yemen became a republic only in1990, data points for these countries during the1980s are excluded from the sample. Djibouti wasalso excluded because of lack of data during the1980s. In the end, our sample covers 17 countriesand 35 programs for the period from 1980 to 1999.2

Since the sample is restricted, it is important todetermine whether it is representative of all IMFprograms. To aid in this process, Table 2 providesvarious benchmarks for this sample as well as for allIMF programs negotiated between 1992 and 2000.The average per capita income levels in the twosamples are close, at about $1,000–$1,100. More-over, the type of conditions that were applied in theprograms are comparable in terms of number anduse of prior actions, with conditions in the restrictedsample slightly more numerous. Interestingly, thepercentage of stoppages is slightly lower in the re-stricted sample. In terms of the target variables, theinitial conditions are also broadly similar, with in-flation averaging about 30–40 percent and the gen-eral government budget deficit averaging 31⁄2 per-cent of GDP. The initial current account andgrowth estimates differ more significantly betweenthe two samples, with the current account deficit inthe restricted sample at 3.1 percent of GDP and thecorresponding deficit in the full sample at 5.7 per-cent of GDP. Moreover, growth in the initial periodwas considerably higher in the restricted sample at1.7 percent, compared with 1.1 percent for the fullsample.

The behavior of the various target variables overtime is also comparable between the two samples.The inflation rate declines through the program pe-riod and three years after the program to averageabout 11 percent in both samples, and the growthrate picks up to over 3 percent in both samples threeyears after the end of the program. In contrast, whilethe budgetary position improves in the reduced sam-ple, the improvement shown in the full sample istemporary, since the budget deficit returns to its ini-tial average value three years out. Similarly, whilethe current account position improves in the re-duced sample, it initially deteriorates in the fullsample and only returns to its initial value threeyears out.

First-Stage Probit Specification

As mentioned in the introduction it is necessary tocorrect for the timing of an IMF program so that thisdecision can be isolated from the effects of IMF poli-cies. To this end, a probit model was estimated withthe dependent variable representing the observed 0-1 dummy capturing whether a program was initi-ated or not. The independent variables fall into twocategories: macroeconomic indicators at the begin-ning of the program (GDP per capita, inflation, gov-ernment budget position, debt/GDP ratio, exchangerate depreciation) and political economy variables(the democracy index, a dummy for the change ofgovernment within three years of the commence-ment of the program, and time in power). The rea-son for including political variables is that recentstudies have found that they are closely related tothe success of programs (Dollar and Svensson, 2000;Ivanova and others, 2002; and Thomas, forthcoming)and they are therefore suitable instruments for iden-tifying the timing of programs. Barro and Lee (2002)also emphasize the need for appropriate instrumentsin conducting this type of analysis but rely on politi-cal economy variables to achieve this objective,

2 The appendix gives a list of program countries, years, andprogram types.

Table 1

Summary of Recent Studies Analyzing the Impact of IMF Programs on Inflation, Budgetary Conditions, and Growth

Study Sample Inflation Growth Budget

Khan (1990) 1973–88, 259 programs – –*Schadler and others (1993) 1983–93, 55 programs – + –Conway (1994) 1976–86, 217 programs – –, +*Killick (1995) 1979–85 –* +*Bagci and Perraudin (1997) 1973–92 - +*Dicks-Mireaux, Mecagni, and Schadler (2000) 1986–91, 88 programs - +*Przeworski and Vreeland (2000) 1951–90, 226 programs –*Bulír and Moon (2002) 1993–96, 64 programs +

Note: A single asterisk (*) indicates significance at the 10 percent level.

263


including each country’s share of quotas and profes-sional staff, as well as voting patterns in the UnitedNations.

The political economy variables used in thispaper are constructed from the World Bank’s Data-base of Political Institutions. The democratic indexis coded from a democratic score on a 0–12 scale,with higher numbers associated with more democra-tic governments. For the democratic variable, thescore was reclassified as a 1-0 dummy variable de-pending on whether the score was 9–12 or less; forthe autocratic variable, the score was reclassified asa 1-0 dummy variable depending on whether thescore was less than 3. The dummy variable for thechange in government is 1 if there is a change ingovernment within three years prior to the programand 0 otherwise. The variable, time in power, repre-

sents the log value of the years in power of the gov-ernment at the time of program entry, assuming nochange in the previous three years, and 0 otherwise.The dummy for the change in government is hy-pothesized to capture the raised likelihood of condi-tionality when a new government comes to power.In addition, separating out longer political regimesallows us to determine whether length of leadershipinfluences the likelihood of having an IMF program.No dummy is included for past programs because thesample was identified on the basis that no programtook place during the 1980s.

The results of the first-stage probit estimation canbe found in Table 3. The overall performance of thevarious specifications is good in predicting the tim-ing of an IMF program with at least 80 percent accu-racy. Moreover, the accuracy of the predictions is

Table 2

Comparing the Restricted and Full Sample Averages(standard errors in parentheses)

GDP Per Capita Total Number Total Number Percentage of

(In U.S. dollars) of Conditions of Prior Actions Stoppages

Restricted sample 1,010.41 25 9.9 36.4(733.2) (21.7) (12.9) (48.9)

Full sample 1,091.52 22 6.6 41.8(1,315) (23.2) (12.8) (49.5)

Values at Specific Points in Program Cycle

At beginning At end 3 years following of program of program program

Inflation (in percent)Restricted sample 32.4 13.2 11.3

(50.2) (15.1) (18.0)Full sample 42.3 22.4 11.8

(61.1) (38.2) (18.1)

Government Budget Balance (in percent of GDP)Restricted sample –3.3 –2.6 –2.3

(3.6) (4.1) (5.0)Full sample –3.9 –3.5 –3.9

(4.2) (4.0) (4.2)

Current Account (in percent of GDP)Restricted sample –3.1 –3.3 –2.5

(9.7) (8.6) (11.1)Full sample –5.7 –5.9 –5.7

(9.4) (11.1) (8.7)

Growth (in percent)Restricted sample 1.7 3.8 3.3

(6.1) (3.8) (3.8)Full sample 1.1 2.7 3.9

(7.2) (6.2) (5.8)

264


quite stable across the specifications (Table 4). Thepseudo R-square3 is also high compared with somestudies, ranging from 53 percent to 65 percent.

Among the economic variables, we found thatGDP per capita, the debt/GDP ratio, and the depre-ciation of the exchange rate were significant at the5 percent level across all specifications, corroborat-

ing the results of previous studies. The higher theGDP per capita, the less likely that a country willenter a program, whereas higher debt/GDP ratiosand exchange rate depreciations raise the likelihoodthat a country will come to the IMF for financial as-sistance. The initial inflation rate and the govern-ment budgetary position are insignificant in allcases.

The political economy variables also provide sig-nificant explanatory power. Indeed, including themraises the pseudo R-square by about 10 percentagepoints and a likelihood ratio test rejects the jointhypothesis that they are all equal to zero. Interest-ingly, highly democratic and autocratic countries

Table 3

Probit Model(maximum-likelihood estimates)Dependent Variable: I =1 if there is an IMF program

I = 0 if there is no IMF program

Independent variables (1) (2)

Constant 4.1509 1.0072 1.6528 0.3607(1.07) (0.45) (0.58) (0.19)

Economic variablesGDP per capita –1.4485* –0.6466** –0.9085* 0.5697*

(–2.03) (–1.89) (–2.18) (–2.02)Initial inflation 0.0519 –0.0164

(0.84) (–0.46)Budget/GDP 0.0860 0.0930

(0.77) (1.34)Debt/GDP 5.6267* 4.9920* 5.7400* 4.8957*

(2.99) (3.48) (3.26) (3.75)Exchange rate depreciation 0.0630** 0.0771* 0.0741* 0.0636*

(1.85) (2.72) (2.57) (2.99)Reserves/imports 0.3087 1.0624 0.1091 0.5228

(0.17) (0.86) (0.07) (0.48)

Political variablesDemocratic –1.1778 –1.1849

(–1.25) (–1.36)Autocratic –1.3405 –0.8158

(–1.18) (–0.87)Change of government 3 years

before program 2.3233 1.2412(1.44) (1.13)

Years in office ( > 3 years ) 0.9892 0.4009(1.26) (0.77)

Observations 61 66 64 69Pseudo R-square 0.6408 0.5478 0.6303 0.5382Log-likelihood –15.1833 –20.6884 –16.3871 –22.0553Lr test of political variables 11.01*** 11.34***

Notes: The t-statistics are in parentheses. One asterisk (*) denotes significance at 5 percent; two asterisks (**) denote significance at 10 per-cent. The Lr test is for the joint hypotheses that the coefficients of the four political variables are all zero; the chi-square (4) test statistic at 5 per-cent is 9.49. Therefore, the political variables are significant, which is denoted by three asterisks (***).

3 Pseudo R-square is calculated as 1–[log(Lur) – log(Lr)], whereLur is the maximum-likelihood value of the unrestricted-likelihoodfunction and Lr is the maximum-likelihood value of the restricted-likelihood function (constant only). Conway (1994) reports a 90percent ratio whereas Dicks-Mireaux, Mecagni, and Schadler(2000) only records a 3.5 percent ratio for their best fitted probitmodel.

265


are less likely to enter IMF programs. A possible ra-tionale for democratic countries not coming to theIMF is that considerable bargaining between politi-cal interest groups might be needed before an agree-ment is reached and this takes time. While this isnot an issue for autocratic regimes, they may try todissuade foreign institutions from closely scrutiniz-ing their economic policies. Interestingly, Thomas(forthcoming) finds that once autocratic govern-ments acquiesce to an IMF program, the program ismuch less likely to stop. New governments are morelikely to enter an IMF program; and for those notchanging office, longer incumbents are more likelyto come to the IMF, although this effect tapers offafter some time.

The stoppage variable is also an endogenous vari-able and therefore may require instruments. If a pro-gram is stopped because of unsatisfactory macroeco-nomic performance, it is inappropriate to use programstoppages as an exogenous variable when evaluatingthe effects of IMF programs on these same targetvariables.4 Mecagni (1999) argues based on countryreport documentation that, in most cases, programinterruptions were not associated with IMF condi-tionality that was too stringent. Rather, the interrup-tions were associated with domestic dissatisfactionwith existing institutional and political arrange-ments or with preexisting problems. The robustnessof this assumption is evaluated by comparing the es-timates assuming that program stoppages are exoge-nous with those obtained from instrumenting forthis variable.

Instruments for the stoppage variable comprisedummies for autocratic regimes and countries en-

gaging in guerrilla warfare, a dummy representing achange in government within three years of the pro-gram, the total number of conditions, the real ex-change rate depreciation during the year prior to theprogram, the initial inflation level, and the loga-rithm of GDP per capita. The estimation results canbe found in Table 5. On the whole, we find thatamong the instruments used, the political variableshave stronger effects on the likelihood of a stoppagethan the economic variables. The hypothesis thatthe political variables are all equal to zero cannot beaccepted at the 10 percent level by the likelihoodratio test. However, the hypothesis that the eco-nomic variables are all equal to zero can be ac-cepted. In terms of the coefficients, the economicvariables are all insignificant, while the democracyvariable is consistently significant. The politicalvariables suggest that the higher probability of stop-pages is associated with less democracy and wars,but, perhaps surprisingly, not a change in govern-ment. The relatively weak effects of the economicvariables mitigate somewhat concerns about the en-dogeneity of the stoppage variable. However, tomake our analysis more robust, we experiment withboth exogenous and endogenous stoppage variables.

Second-Stage Specification

One of the main subjects of interest in IMF pro-grams is how they succeed in improving macroeco-nomic target variables during the duration of pro-grams and following their termination. To analyzethis issue effectively, the dependent variable is de-fined in two ways for each target variable. First, it isdefined as the average change in the value of thetarget between the first year of the program and thefinal year of the program. Second, it is defined as theaverage change in the value of the target betweenthe first year of the program and three years after theend of the program. For periods during which no

Table 4

Accuracy of Probit Predictions

(1) (2)

Total predicted I =1 28 27 27 27Total actual I =1 31 33 31 33Accuracy (percent) 90.32 81.82 87.10 81.82

Total predicted I =1 27 26 28 30Total actual I =1 30 33 33 36Accuracy (percent) 90 78.78 84.85 83.33

Overall correct predictions 55 53 55 57Overall actual programs 61 66 64 69Overall accuracy (percent) 90.16 80.30 85.94 82.61

4 The stoppage variable used in this paper represents programsthat were abandoned because the authorities did not follow IMFpolicy recommendations. If IMF targets were not met because ofexogenous shocks, this would not warrant a stoppage of the pro-gram but would be accommodated through the granting of awaiver.

266


program was conducted (1980–92), three- and six-year intervals are used for the dependent variable.This time interval roughly corresponds to the aver-age duration of IMF programs during the 1990s of21⁄2 years.

In many studies, the effects of IMF involvementin particular countries are identified through the in-clusion of a dummy variable for the macroeconomicprograms that were supported by the IMF. However,as indicated previously, it is also necessary to ac-count for stoppages. In our sample of 35 programs,12, or 36 percent, were stopped. Killick (1995) dif-ferentiates between countries that completed pro-grams and those that terminated early by using athreshold value of 80 percent of the initial commit-ted loan that was disbursed. He argues that this cut-off point is likely to be closely associated with suc-cessful completions based on a survey made ofprograms over the 1980–92 period. Consistent withhis hypothesis, he finds that the inflation rate andthe current account improved significantly forcountries with completed programs relative to coun-tries with noncompleted programs. This paper con-

trols for the possible different impacts of completedprograms and those that terminated early with astoppage variable. Although as recorded most of thestoppages happened as a result of the program coun-tries’ failure to follow IMF recommendations, to ac-count for the possible endogeneity problem, we con-sider both exogenous and endogenous stoppages.5Moreover, following Killick (1995), we constructthe disbursement ratio as another proxy for stop-pages. By doing so, the effectiveness of IMF programson the target macroeconomic variables is evaluatedmore accurately, and the effects of stoppages are alsoanalyzed. For the sample of observations in the1980s when no program took place, the dummyvariables are recorded as zero.

Since timing can make a large difference in theinterpretation of results, annual time dummies wereincluded in each specification to control for this fac-tor. Owing to insufficient degrees of freedom, both

Table 5

Probit Model for StoppageDependent Variable: I =1 if there is an IMF program

I = 0 if there is no IMF program

Independent Variables (1) (2) (3)

Constant –0.7245 0.3733 0.0034(–0.31) (0.20) (0.01)

GDP per capita 0.1637 –0.0349(0.2671) (–0.16)

Initial inflation –0.0099 –0.0008(–0.94) (–0.12)

Exchange rate depreciation 0.0293 0.0169(1.32) (1.10)

Total number of conditions –0.1875 –0.3023(–0.48) (–0.82)

Democratic –1.2279 –1.0714(–2.05) (–2.03)

Guerrilla wars 1.1445 0.6922(0.92) (0.72)

Change of government 3 years before program –0.3559 –0.1881(–0.61) (–0.39)

Observations 35 35 35Pseudo R-square 0.2177 0.0502 0.1404Log-likelihood –16.92 –20.54 –18.59Restricted log-likelihood –21.63 –21.63 –21.63Lr test 7.25* 3.34

Note: For the joint test that all institution variables are zero, the chi(3) statistic at 10 percent is 6.25; for the joint test that all macroeconomicvariables are zero, the corresponding chi(4) statistic is 7.78. Therefore, the institution variables are significant, which is denoted by an asterisk (*).

5 In the endogenous version, the variable is instrumented bythe predicted stoppage probability from the probit regression inTable 5.

267


the dependent and time-varying independent vari-ables were purged of the timing effect before beingintroduced into the second-stage equation.

For the inflation equation, the economic determi-nants include the level of inflation at the beginningof the period and the change in the local cur-rency/U.S. dollar exchange rate. The variables rep-resenting IMF involvement include a dummy forIMF programs, a dummy for stoppages, and the IMRcalculated from the first-stage probit estimation.

The first two columns of Table 6 present the re-sults without correcting for the timing of IMF in-volvement. In this case, distinguishing betweencompleted and stopped programs makes no differ-ence because both coefficients are insignificant.Similar results hold when a dummy variable for pro-grams with disbursed amounts over 80 percent ofthe committed total (comparable to Killick’s defini-tion) is substituted for the stoppage variable. Turn-ing to the other variables, high-inflation countrieshave difficulty reducing inflation because the coeffi-cient on the initial inflation level is significantlypositive. Moreover, the change in the exchange rateis positive although significantly below unity. Theinclusion of the IMR has little effect on the othervariables in this specification (columns (3) and (4)).The major change in results occurs when the stop-page variable is instrumented, because the coeffi-cient becomes significantly positive at the 10 per-cent level. The coefficient estimate implies that the

inflation rate rises by 61⁄2 percent a year during theprogram period in stopped programs.

In the regression of inflation changes over alonger horizon (Table 7), the coefficients on theprogram dummies are broadly similar to the previousspecification, with the instrumented stoppage vari-able the only significant variable. The coefficientestimate of IMF programs is comparable in magni-tude to that presented in the short-run equation,suggesting that the inflation rate keeps decliningafter the termination of the program. Both the ini-tial inflation rate and the change in the exchangerate remain significant, with the exchange rate coef-ficient now insignificantly different from unity.

For the equation explaining the change in thebudgetary position, the independent variables com-prise the initial budget position, the change in infla-tion, and the change in the terms of trade. Thechange in the terms of trade proxies strong develop-ments in GDP, which would likely lower the budgetdeficit. The change in the inflation rate is includedto control for the Tanzi-Olivera effect, which postu-lates that as the inflation rate is lowered, the bud-getary position would be expected to improve be-cause the real value of conventional tax revenuerises on account of collection lags. The variablesrepresenting IMF involvement include the dummyfor IMF programs, a dummy for IMF programs thatterminated prematurely, and the IMR calculatedfrom the first-stage probit estimation.

Table 6

Inflation Equation During Program

Independent Variables (1) (2) (3) (4) (5)

Initial inflation 0.3304* 0.3469* 0.3255* 0.3422* 0.3202*(5.92) (5.89) (5.48) (5.51) (5.40)

Change in exchange rate 0.6243* 0.6246* 0.6158* 0.6175* 0.6169*(6.59) (6.67) (6.03) (6.15) (6.08)

IMF program –0.7319 –0.0217 –0.4574 0.1047 –1.9763(–0.48) (–0.01) (–0.21) (0.04) (–0.85)

Stoppage1 1.9962 2.2145 6.6876**(1.12) (1.09) (1.75)

Disbursement ratio 0.0407 0.3952(0.02) (0.19)

Inverse Mill's ratio –0.5546 –0.4308 –0.3682(–0.37) (–0.29) (–0.24)

Number of observations 69 69 60 60 60R-square 0.7569 0.7544 0.7485 0.7457 0.7525

Notes: The t-statistics appear in parentheses; one asterisk (*) denotes significance at the 5 percent level; two asterisks (**) denote signifi-cance at the 10 percent level.

1 In column (5), the stoppage variable is instrumented using the following variables: dummies for autocracy, guerrilla warfare, changes ingovernment, the total number of structural conditions, the change in the real exchange rate one year prior to the program, the initial inflationlevel, and the log of GDP per capita.

268


In the regression for the budgetary change duringthe program (Table 8), the initial budget level isconsistently significantly negative, implying thatthe budgetary improvement is greater the larger theinitial size of the deficit. But the budgetary positionalso weakens if the initial position is in surplus. Thechange in the inflation rate is significantly negative,suggesting that the Tanzi-Olivera effect holds. In-deed, a 10 percent decline in the inflation ratewould raise the budget surplus by about 1⁄2 percent ofGDP. In contrast, the change in the terms of tradedoes not play a significant role in budgetary devel-opments. For the IMF dummies, while the coeffi-cient on the completed IMF program is significantlypositive in the specification without the inverseMill’s ratio, the coefficient is insignificant when theIMR is included. In contrast, the coefficient on thestoppage variable is significantly negative in eachspecification, with a particularly high value wheninstrumented.

Three years after the program has terminated(Table 9), initial conditions remain significant, although the impact is less strong. The dummy forcompleted IMF programs is insignificant in all speci-fications, so that the budgetary position is not af-fected by an IMF program. In contrast, for programsthat stop, the budgetary position deteriorates rela-tive to periods without an IMF program. Similar re-sults hold when the dummy variable reflecting thedisbursement ratio is used.

While improvements in the inflationary environ-ment and in the budgetary position are important

targets in themselves, they are not the ultimategoals of economic policy. The ultimate objective ofa country’s economic policy is to improve the wel-fare of its citizens, which is normally measured asthe change in per capita income. There is a huge lit-erature on the determinants of growth, and there-fore this paper has chosen to be selective in decidingwhich controls to include in the specification. Thepaper takes as its point of departure the analysis ofDoppelhofer, Miller, and Sala-i-Martin (2000), whohas tried to detect the variables that are the most ro-bust determinants of growth using the Bayesian up-dating technique. He finds that a dummy variablefor sub-Saharan Africa, the initial level of GDP percapita (the convergence effect), and the primaryschool enrollment rate (measuring human capital)are robust to changes in specification. This paper in-cludes the first two variables and substitutes the il-literacy rate for the primary school enrollment rate.We also include the change in the budget balanceand the inflation rate to capture direct effects fromthe policy changes, and export market growth tocapture exogenous effects.

In the regression for the change in growth duringthe program (Table 10), the initial convergenceterm is generally significant, indicating that eachyear poor countries make up 1 percent of the dispar-ity in per capita GDP with the richest country, orthat, holding all other factors constant, it wouldtake them 100 years to fully catch up. Perhaps surprisingly, both policy variables represented by thechange in the inflation rate and the budgetary

Table 7

Inflation Equation Three Years After Program


Initial inflation 0.1717* 0.1789* 0.1209* 0.1354* 0.1245*(3.95) (4.24) (3.93) (4.58) (4.98)

Change in exchange rate 0.8509* 0.8503* 0.9124* 0.9119* 0.9037*(11.09) (10.98) (13.61) (13.47) (13.44)

IMF program –0.4122 –0.1270 –0.0946 0.1684 –1.6231(–0.26) (–0.08) (–0.06) (0.10) (–0.95)

Stoppage1 1.0600 2.6809 5.8508**(0.61) (1.55) (1.76)

Disbursement ratio 0.2857 1.4460(0.18) (0.99)




1 In column (5), the stoppage variable is instrumented using variables specified in footnote 1 to Table 6.

269


Table 8

Budgetary Positions During Program


Initial budget level –0.1782* –0.1657* –0.2242* –0.2108* –0.2220*(–3.02) (–2.74) (–4.85) (–4.20) (–4.58)

Change in inflation –0.0429* –0.0529* –0.0471* –0.0572* –0.0495*(–3.12) (–3.89) (–3.61) (–4.31) (–3.74)

Change in terms of trade 0.0055 0.0096 0.0094 0.0137 0.0132(0.41) (0.67) (0.75) (0.97) (0.99)

IMF program 0.6094** 0.3872 0.3608 0.1858 0.6476(1.78) (0.96) (0.84) (0.38) (1.27)

Stoppage1 –1.1169* –1.1380* –2.0330*(–2.58) (–2.25) (–2.43)

Disbursement ratio –0.3753 –0.4550(–0.86) (–0.92)





Table 9

Budgetary Positions Three Years After Program


Initial budget level –0.1303* –0.1264* –0.1456* –0.1400* –0.1419*(–8.03) (–7.26) (–12.44) (–9.43) (–7.70)

Change in inflation –0.0050 –0.0076 –0.0060 –0.0090 –0.0074(–0.82) (–1.19) (–0.98) (–1.37) (–1.09)

Change in terms of trade 0.0102 0.0091 0.0185 0.0185 0.0176(0.68) (0.60) (1.26) (1.10) (1.10)

IMF program 0.3831 0.3559 0.2819 0.2703 0.3330(1.56) (1.26) (1.05) (0.87) (0.99)

Stoppage1 –0.8651* –0.9318* –0.8644(–3.10) (–3.03) (–1.60)

Disbursement ratio –0.6930* –0.7108*(–2.22) (–2.07)





270


position are insignificant, although the dummy variable for countries in sub-Saharan Africa is sig-nificantly negative in the specifications with theIMR. The other two macroeconomic variables, theexport market growth and the illiteracy rate, are in-significant.

Turning to the direct effects of IMF programs, thedummy variable for completed programs is insignifi-cant in all specifications whereas the stoppage vari-able is significantly negative in all specifications ex-cept the instrumented equation.

Over the longer run (Table 11), the effect of thechange in inflation on growth is significantly nega-tive, with a 10 percentage point decline in the infla-tion rate leading to a 0.5–0.6 percentage point im-provement in the growth rate. This is comparable tothe findings of Ghosh and Phillips (1998). GDP percapita is generally significantly negative, indicatingthat a lower growth rate is associated with a higherlevel of GDP per capita. The dummy for the sub-Saharan region is significantly negative at 5 percentwhen the inverse Mill’s ratio is included, while the illiteracy rate is significantly negative when

the IMR is excluded. It appears therefore that sub-Saharan countries and countries with high illiteracyrates have difficulty in raising their standard of liv-ing over the longer run.

In the long run, the IMF dummy for completedprograms remains insignificant except in the specifi-cation including the IMR in which it yields positiveeffects on growth (Table 12). In contrast, the stop-page variable is significant except in the instru-mented equation, and the coefficient implies nega-tive effects on growth. The different results point tothe possibility that a failure in macroeconomic per-formance could trigger a stoppage.

Conclusion

This paper has presented empirical evidence rele-vant to evaluating whether IMF programs havebeen effective in the 1990s in influencing threemajor macroeconomic variables: inflation, the bud-get position, and growth. To properly identify the ef-fect of IMF policies on these aggregates, variables

Table 10

Growth Equation During Program


Change in inflation –0.0353 –0.0489 –0.0261 –0.0380 –0.0375(–0.77) (–1.02) (–0.51) (–0.72) (–0.70)

Change in budget position 0.0869 0.1247 0.1483 0.1786 0.1833(0.49) (0.69) (0.73) (0.90) (0.96)

GDP per capita –0.6307** –0.5604 –0.9826* –1.0287* –0.9078**(–1.65) (–1.30) (–2.03) (–2.04) (–1.76)

Export market GDP growth 0.2431 0.2582 0.1999 0.2273 0.1683(0.84) (0.82) (0.61) (0.62) (0.48)

Sub-Saharan region –0.7704 –0.3896 –2.5127** –2.4202* –2.1550**(–0.74) (–0.39) (–1.94) (–1.96) (–1.84)

Illiteracy –0.0070 –0.0137 0.0123 0.0068 0.0048(–0.30) (–0.58) (0.44) (0.26) (0.18)

IMF program 0.5904 0.2574 1.0676 0.8866 1.0665(0.92) (0.29) (1.46) (0.88) (1.22)

Stoppage1 –1.6959** –1.7196** –1.6672(–1.78) (–1.66) (–1.13)

Disbursement ratio –0.4399 –0.8266(–0.48) (–0.86)





271


were included to control for the timing of IMF pro-grams. The paper finds that the timing of an IMFprogram can be represented well by the level ofGDP per capita, the debt/GDP ratio, and the mag-nitude of the exchange rate depreciation, with over90 percent of the timing decisions correctly identi-fied. Although the inverse Mill’s ratio is insignifi-

cant in all specifications, its inclusion does changethe significance of the dummy for completed IMFprograms in some cases, demonstrating the impor-tance of its inclusion in the analysis.

The paper highlights the importance of distin-guishing between completed and stopped programsbecause they are associated with significantly different

Table 11

Growth Equation Three Years After Program


Change in inflation –0.0606* –0.0668* –0.0543* –0.0603* –0.0582*(–4.04) (–4.17) (–4.20) (–4.50) (–3.92)

Change in budget position –0.0331 –0.0265 –0.0553 –0.0505 0.0307(–0.13) (–0.10) (–0.22) (–0.20) (0.13)

GDP per capita –0.7701** –0.7734** –1.3182* –1.3717* –1.3237*(–1.93) (–1.82) (–2.73) (–2.78) (–2.64)

Export market GDP growth 0.3544 0.3678 0.3126 0.3213 0.2352(1.08) (1.10) (0.93) (0.89) (0.66)

Sub-Saharan region –0.2987 –0.2148 –2.6695* –2.5925* –2.4166*(–0.34) (–0.25) (–2.57) (–2.53) (–2.28)

Illiteracy –0.0347** –0.0391* –0.0083 –0.0136 –0.0136(–1.82) (–2.07) (–0.39) (–0.66) (–0.64)

IMF program 0.8097 0.7888 1.0505** 1.0283 1.1199(1.35) (1.27) (1.71) (1.52) (1.26)

Stoppage1 –1.7342* –1.6221* –1.4884(–2.57) (–2.23) (–1.03)

Disbursement ratio –1.4352* –1.2934*(–2.23) (–2.00)


Number of observations 53 53 47 47 47R–square 0.4078 0.3961 0.5160 0.5035 0.4852

Notes: The t–statistics appear in parentheses; one asterisk (*) denotes significance at the 5 percent level; two asterisks (**) denote signifi-cance at the 10 percent level.


Table 12

IMF Dummy Only Versus Dummies Distinguishing Stoppage

Inflation Budget Growth

During Three During Three During Three program years later program years later program years later

Completed programs – – + + + +**Stopped programs + + –* –* –** –*Stopped programs (using instruments) +** +** –* – – –

Notes: One asterisk (*) denotes significance at the 5 percent level; two asterisks (**) denote significance at the 10 percent level.

272


macroeconomic outcomes. Since stoppages arelikely endogenous, they were instrumented withboth political and economic variables. Our analysisshows that political variables seem to have more im-pact than do economic variables on stoppages. Onthe whole, endogenizing the stoppage variable doesnot change our conclusion that stopped programsraise inflation, worsen the budgetary position, andimpede growth, but it deepens the negative effectsthat a stopped program can have on inflation whileit lessens the negative impact on budget conditionsand growth. On IMF programs that end successfully,the only variable that appears to be affected in apositive way is growth, and the significance of this

variable breaks down when instruments are foundfor stoppages.

One of the limitations of this analysis is the sam-ple size. We avoid the problems with cross-countrycomparisons by making a clear split between pro-gram periods and nonprogram periods, but thetrade-off is a relatively short sample that mightmake our results less convincing and less robust. Wehope to tackle this problem in the future when moredata become available. We also intend to dig deeperinto whether IMF programs in the 1990s have suc-ceeded in fostering growth by distinguishing amongtypes of programs and between normal and pro-longed users of IMF resources.

273


Appendix. List of Countries and Programs Considered in 1990s

Country Type of Program Start End

Algeria SBA 1994 1995Algeria EFF 1995 1998Benin ESAF 1993 1996Benin ESAF 1996 1999Bulgaria SBA 1992 1993Bulgaria SBA 1994 1995Bulgaria EFF 1996 1998Bulgaria SBA 1997 1998Bulgaria EFF 1998 2001Burkina Faso ESAF 1993 1996Burkina Faso ESAF 1996 1999Burkina Faso ESAF 1999 2002Cambodia ESAF 1994 1997Cambodia PRGF 1999 2002Cape Verde SBA 1998 1999Colombia EFF 1999 2002Djibouti ESAF 1996 1999Djibouti ESAF 1999 2002Indonesia SBA 1997 1998Indonesia EFF 1998 2000Jordan SBA 1992 1993Jordan EFF 1994 1996Jordan EFF 1996 1999Jordan EFF 1999 2002Mongolia ESAF 1993 1996Mongolia ESAF 1997 2000Nicaragua ESAF 1994 1997Nicaragua ESAF 1998 2001Papua New Guinea SBA 1995 1997Poland SBA 1993 1994Poland SBA 1994 1996Rwanda ESAF 1998 2001Venezuela, República Bolivariana de SBA 1996 1997Yemen, Republic of SBA 1996 1997Yemen, Republic of ESAF 1997 2000

Notes: SBA denotes a Stand-By Arrangement; EFF denotes the Extended Fund Facility; ESAF denotes the Enhanced Structural Adjustment Fa-cility; and PRGF denotes the Poverty Reduction and Growth Facility.

274


ReferencesBagci, Pinar, and William Perraudin, 1997, “The Impact

of IMF Programs,” CEPR Working Paper No. 24(London: Center for Economic Policy Research).

Barro, Robert, and J. W. Lee, 2002, “IMF Programs: WhoIs Chosen and What Are the Effects?” NBER Work-ing Paper No. 8951 (Cambridge, Massachusetts: National Bureau of Economic Research).

Bulír, Ales, and Soojin Moon, 2002, “The Composition ofFiscal Adjustment and Structural ConditionalityUnder IMF-Supported Programs,” (unpublished;Washington: International Monetary Fund). Thispaper also appears, in somewhat different form, asChapter 14 in this volume.


Dicks-Mireaux, Louis, Mauro Mecagni, and SusanSchadler, 2000, “Evaluating the Effect of IMF Lend-ing to Low-income Countries,” Journal of Develop-ment Economics, Vol. 61 (April), pp. 495–526.

Dollar, David, and Jakob Svensson, 2000, “What Explainsthe Success or Failure of Structural Adjustment Pro-grammes?” Economic Journal, Vol. 110 (October),pp. 894–917.

Doppelhofer, Gernot, Ronald I. Miller, and Xavier Sala-i-Martin, 2000, “Determinants of Long-Term Growth:A Bayesian Averaging of Classical Estimates (BACE)Approach,” NBER Working Paper No. 7750 (Cam-bridge, Massachusetts: National Bureau of EconomicResearch).

Edwards, Sebastian, and Julio Santealla, 1992, “Devalua-tion Controversies in the Developing Countries:Lessons from the Bretton Woods Era,” NBER Work-ing Paper No. 4047 (Cambridge, Massachusetts: National Bureau of Economic Research).

Ghosh, Atish, and S. Phillips, 1998, “Inflation, Disinfla-tion and Growth,” Staff Papers, International Mone-tary Fund, Vol. 45 (December), pp. 672–710.

Goldstein, Morris, and Peter Montiel, 1986, “EvaluatingFund Stabilization Programs with Multi-CountryData: Some Methodological Pitfalls,” Staff Papers, International Monetary Fund, Vol. 33 (June),pp. 304–44.

Haque, Nadeem Ul, and Mohsin Khan, 1998, “Do IMF-Supported Programs Work? A Survey of theCross-Country Empirical Evidence,” IMF WorkingPaper 98/169 (Washington: International MonetaryFund).

Heckman, James, 1979, “Sample Selection Bias as a Specification Error,” Econometrica, Vol. 47 (January),pp. 153–62.

Ivanova, Anna, Wolfgang Mayer, Alex Mourmouras, andGeorge Anayiotos, 2002, “What Determines theSuccess or Failure of Fund-Supported Programs?”(unpublished: Washington: International MonetaryFund). This paper also appears, in somewhat differentform, as Chapter 10 in this volume.

Joyce, James, 2002, “Through a Glass Darkly: What WeKnow (and Don’t Know) About IMF Programs,”Wellesley College Working Paper No. 2002-04(Wellesley, Massachusetts: Wellesley College).

Khan, Mohsin, 1990, “The Macroeconomic Effects ofFund-Supported Adjustment Programs,” Staff Papers,International Monetary Fund, Vol. 37 (June),pp. 195–231.

Killick, Tony, 1995, IMF Programmes in DevelopingCountries: Design and Impact (London: Routledge).

Knight, Malcolm, and Julio A. Santaella, 1997, “Eco-nomic Determinants of IMF Financial Arrange-ments,” Journal of Development Economics, Vol. 54(December), pp. 405–36.

Mecagni, Mauro, 1999, “The Causes of Program Interrup-tions,” in Economic Adjustment and Reform in Low-Income Countries (Washington: International Mone-tary Fund), pp. 215–76.

Przeworski, Adam, and James R. Vreeland, 2000, “The Effect of IMF Programs on Economic Growth,” Jour-nal of Development Economics, Vol. 62 (August),pp. 385–421.

Schadler, Susan, Franek Rozwadowski, Siddharth Tiwari,and David Robinson, 1993, Economic Adjustment inLow-Income Countries: Experience Under the EnhancedStructural Adjustment Facility, IMF Occasional PaperNo. 106 (Washington: International Monetary Fund).

Thomas, Alun, forthcoming, “Prior Actions-True Repen-tance? An Evaluation Based on Active Programsover the 1992–99 Period,” IMF Working Paper(Washington: International Monetary Fund).

IMF-Supported Programs: Recent Staff Research; Ashoka Mody ... · Wolfgang Mayer Professor of...

Documents

Transcript of IMF-Supported Programs: Recent Staff Research; Ashoka Mody ... · Wolfgang Mayer Professor of...