This article was downloaded by: [Louisiana State University]On: 20 August 2014, At: 06:22Publisher: RoutledgeInforma Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House,37-41 Mortimer Street, London W1T 3JH, UK

Applied Financial EconomicsPublication details, including instructions for authors and subscription information:http://www.tandfonline.com/loi/rafe20

Bank performance and the financial crisis: evidencefrom KazakhstanAnthony J. Glassa, Karligash Kenjegalievaa & Thomas Weyman-Jonesa

a School of Business and Economics, Loughborough University, Leics LE11 3TU,UKPublished online: 15 Jan 2014.

To cite this article: Anthony J. Glass, Karligash Kenjegalieva & Thomas Weyman-Jones (2014) Bank performance and thefinancial crisis: evidence from Kazakhstan, Applied Financial Economics, 24:2, 121-138, DOI: 10.1080/09603107.2013.868584

To link to this article: http://dx.doi.org/10.1080/09603107.2013.868584

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) containedin the publications on our platform. However, Taylor & Francis, our agents, and our licensors make norepresentations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of theContent. Any opinions and views expressed in this publication are the opinions and views of the authors, andare not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon andshould be independently verified with primary sources of information. Taylor and Francis shall not be liable forany losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoeveror howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use ofthe Content.

This article may be used for research, teaching, and private study purposes. Any substantial or systematicreproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in anyform to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://www.tandfonline.com/page/terms-and-conditions

Bank performance and the financial

crisis: evidence from Kazakhstan

Anthony J. Glass, Karligash Kenjegalieva* and Thomas Weyman-Jones

School of Business and Economics, Loughborough University, Leics LE11 3TU,UK

During the first phase of the financial crisis in 2008/09, after Iceland and Belgium,Kazakhstan experienced the most significant bank failures as a share of banksystem assets. Using rich monthly data for virtually the entire Kazakh bankingindustry for the period March 2007–December 2010, Stochastic FrontierAnalysis (SFA) is used to fit several functions (cost, revenue, standard profit,alternative profit and input distance). Among other things, we estimate the effectsof two measures of the quality and risk of the loan portfolio on the industrybest practice frontiers and bank inefficiencies. We find that an increase in thevolume of bad loans as a ratio of total lending has a desirable effect on the cost,input-distance and alternative profit frontiers, all of which is consistent with the‘skimping’ hypothesis.

Keywords: banking performance; bad loans; Stochastic Frontier Analysis

JEL Classification: C51; G21; G28; P34

There is a better model for Ireland than Iceland ... Peopleshould be looking at Kazakhstan, which didn’t bail out anycreditors and let the three biggest banks fail, yet avoideda recession by letting the currency plunge and usingmonetary stimulus... [The Telegraph, 8 December 2010]

I.Introduction

The recent global financial crisis has resulted in bankingcrises in a range of countries. As a share of total banksystem assets, the biggest bank failures during the firstphase of the financial crisis in 2008/09 were in Iceland(circa 90%) followed by Belgium (53%), Kazakhstan(28%) and the UK (26%) (Laeven and Valencia, 2010).During the financial crisis, the capital of the entire Kazakhbanking sector was negative because of the large negativeequity of two of the country’s largest banks (BTA andAlliance). Because both banks were considered too big to

fail, the government decided to nationalize them but hav-ing done so it turned out they were too big to save and thegovernment declared that it would not guarantee the for-eign liabilities of any banks. Consequently, BTA andAlliance defaulted on their foreign liabilities.

The recent turmoil in the banking sector has focusedattention on banking performance and, in particular, on thequality of bank assets and credit risk. Many studies ofbanking performance have attempted to estimate the effectof loan quality and risk on banking efficiency and produc-tivity. The earlier studies analyse the effect of loan qualityand banking risk on the efficiency of US banks (e.g.Mester, 1996; Berger and Mester, 1997). In general,these studies observe a negative relationship between pro-blem loans and banking efficiency. This relationship, how-ever, may be a phenomenon that relates to periods of badluck, bad management, ‘skimping’ and moral hazard(Berger and DeYoung, 1997). More recent studies inves-tigate the risk-performance nexus for European banks.This literature is wide ranging and includes cross country

*Corresponding author. E-mail: K.A.Kenjegalieva@lboro.ac.uk

Applied Financial Economics, 2014Vol. 24, No. 2, 121–138, http://dx.doi.org/10.1080/09603107.2013.868584

© 2014 Taylor & Francis 121

Dow

nloa

ded

by [

Lou

isia

na S

tate

Uni

vers

ity]

at 0

6:22

20

Aug

ust 2

014

studies (e.g. Iannotta et al., 2007) and studies of bankingefficiency in Greece (Pasiouras, 2008), Japan (Altunbaset al., 2000), emerging economies (Isik and Hassan,2002), East Asian countries (Kwan, 2003; Chiu andChen, 2009; Sun and Chang, 2011) and transition coun-tries (Kraft and Tirtiroglu, 1998; Havrylchyk, 2006;Brissimis et al., 2008; Kenjegalieva et al., 2009).

Using rich monthly data for virtually the entire Kazakhbanking industry for a study period (March 2007–December 2010) which includes the financial crisis, wedraw on the financial intermediation approach to applyStochastic Frontier Analysis (SFA) to fit several functions(cost, revenue, standard profit, alternative profit and inputdistance). The fitted functions are used to investigate theeffect of the quality and risk of the loan portfolio on theindustry best practice frontiers, and the economic andtechnical inefficiencies of individual banks, whilst con-trolling for other bank-specific characteristics. Since thestudy period includes the financial crisis, we pay particularattention to the performance of BTA and Alliance.

This article contributes to the banking literature in fourways. First, while most banking studies use the overalllevel of problem loans and/or provisions for loan losses asproxies for banking risk, we use data which classifiesloans into seven categories according to their quality andriskiness. Specifically, two classifications are used. Thefirst is based on the volume of loans and the second isbased on the level of provision for loan losses. Bothclassifications split the bank loan portfolio into standard,bad and doubtful loans with a further five sub-categoriesof the latter. We observe some interesting differencesbetween the effects of the two classifications on the indus-try best practice frontiers and banking inefficiencies.These differences are likely to be because the loan lossprovisions classification will reflect not only issuancebehaviour but also managerial discretion on loan reservesto manipulate profit. For this reason, we posit that thevolume of bad loans as a share of total assets, rather thanthe ratio of reserves on bad loans to total assets, should beused to monitor the level of nonperforming loans in theKazakh banking sector, which is still a big issue for theindustry despite some improvement in this area sincethe height of the financial crisis.

Second, our approach is similar to Hughes and Mester(1993) and Berger and Mester (1997) but while theyestimate cost and profit functions to analyse the effect ofrisk on the industry best practice frontiers and bank ineffi-ciencies, we fit a further three functions. In addition, herethe modelling of the best practice frontiers takes into

account interbank heteroscedasticity in inefficiencyeffects and unobservable heterogeneity in the banks’ tech-nologies. Third, to the best of the authors’ knowledge thisis the first study to analyse banking performance inKazakhstan. It is the largest landlocked country and hassubstantial oil reserves and in the early 2000s, it was oneof the fastest growing economies in the world with growthcomparable to the East Asian tigers.1

Fourth, the determinants of the industry best practicefrontiers and banking inefficiencies are used to inform adiscussion of the characteristics of Kazakh bank beha-viour during the recent financial crisis. We find that anincrease in the volume of bad loans as a ratio of totallending has a desirable effect on the cost, input-distanceand alternative profit frontiers, all of which is consistentwith the ‘skimping’ hypothesis.

The remainder of this article is organized as follows.Section II provides an overview of banking sector devel-opments in Kazakhstan and in Section III, we set out theempirical methodology. Section IV describes the data setand in Section V, the empirical results are presented andanalysed. In Section VI, we conclude by suggesting howthe empirical findings can be used to assist in acting on thelessons that have been learnt from the Kazakh bankingcrisis.

II. Kazakh Banking Sector Development

A modern banking system typically has a decentralizedtwo-tier structure with second-tier commercial banks thatare independent of the upper-tier central bank. In 1987,several years prior to the collapse of Soviet Union, theSoviet government formally reorganized the monobankinto a two-tier banking system. The State Bank(Gosbank) acted as the central bank. The second tier con-sisted of specialized commercial banks which wereresponsible for providing financial services to SoEs.Many branches of the quasi-independent commercialbanks severed ties with their parent bank and formedautonomous banking units. In addition, privately ownedbanks and financial institutions were allowed to enter theindustry. This resulted in a rapid increase in the number ofregistered commercial banks.

In 1991, Kazakhstan inherited this quasi two-tierbanking system with 72 second-tier banks. In 1993,Kazakhstan left the ruble area and introduced its owncurrency (Kazakh tenge, KZT). In the same year, the

1 In the extant literature, Kazakh banks have featured in a number of cross-country studies on the affect of financial development onbanking efficiency (e.g. Fries and Taci, 2005; De Haas et al., 2010; Turk Ariss, 2010). The first of these studies analyses bankingefficiency across 60 countries, whereas the latter study uses a sample of banks in 15 countries. These studies provide a regional picture ofbank efficiency although it is debatable if it makes sense to compare the findings of such studies because often the samples are verydifferent. Interestingly, in the latter study, on average, Kazakh banks are the most efficient in the sample. This finding provides furthermotivation for an efficiency analysis which focuses exclusively on Kazakh banks.

122 A. J. Glass et al.

Dow

nloa

ded

by [

Lou

isia

na S

tate

Uni

vers

ity]

at 0

6:22

20

Aug

ust 2

014

Gosbank was reorganized and the Kazakh central bank,the National Bank of the Republic of Kazakhstan(NBK), was formed. Initially, as is the case in mosttransition countries, the new central bank had quiteliberal licensing policies which in conjunction withshortcomings in the legal framework and the supervi-sory system fostered further growth in the number ofundercapitalized and potentially nonviable banks. By1994, the number of banks had peaked at 191. Thisrapid growth of the banking system, however, did notresult in an increase in financial intermediation andcompetition. Hoelscher (1998) posits that there arefour reasons for this: (i) in some parts of the countrybanking was highly concentrated; (ii) there were estab-lished working relationships between some banks socompetition was not likely to materialize betweenthese banks; (iii) the formation of financial-industrialgroups deterred entry; (iv) some banks were set up notto compete but for other reasons, e.g., to obtain sub-sidized credit from the NBK.

In the 1990s, Kazakhstan experienced severe macroe-conomic and financial instability. GDP fell by over 25%over the period 1992 to 1994 and in 1994 inflation was1160%. Also, private sector confidence in the bankingsector dissipated rapidly. This is evident because thedeposit base as a share of GDP fell from 72% in 1994 toless than 5% in 1996. This coincided with the NBK pre-siding over a series of reforms of the banking and financialsystem. The reforms included, the introduction of liquida-tion proceedings which enabled NBK to initiate the liqui-dation of failing banks. Banking supervision was alsotightened and involved more comprehensive on-siteinspections, tighter licensing agreements and stricter capi-tal requirements. This resulted in an increase in licensewithdrawals and a marked fall in the number of new bankregistrations. This is borne out by the sharp fall in thenumber of banks over the period 1995 to 2005 (seeTable 1). In 1994, nonperforming loans accounted forover 50% of the system’s loan portfolio. In the light ofthis, a big part of the reforms involved transferring non-performing loans from banks to three debt resolutioncompanies (the Rehabilitation Bank, RB, theAgricultural Support Fund, ASF, and Exim Bank) whichwere set up by the NBK (IMF, 1998).

Strong economic growth over the period 2000 to 2007(over 10% for some years during this period), which waslargely due to growth of the oil sector, fuelled confidencein many sectors of the Kazakh economy. The rapid growthof the Kazakh economy was accompanied by an increasein the capacity of the Kazakh banking system. Financialdeepening also increased sharply in Kazakhstan and waswell above the levels in most of the former Soviet Unioncountries and was only slightly below the levels in the EUaccession countries (IMF, 2005). To illustrate, the lendingof commercial banks as a share of GDP increased from 7%in 1999 to 67% in 2007 and over the same period, the ratioof deposits to GDP increased from 8% to 48%.

Although the credit boom in Kazakhstan was accom-panied by an increase in the deposit base, the base was notstrong. This is because the growing confidence in theKazakh banking system and the gradual depreciation ofthe tenge since 2000 created a strong preference for for-eign currency. Moreover, since the demand for credit wasgrowing faster than domestic savings some Kazakh banks,which were rated by international rating agencies, bor-rowed heavily in international markets at competitiverates to finance mortgages, consumer loans, loans forreal estate construction and loans to facilitate internationaltrade. Consequently, the ratio of foreign currency lendingto total lending was high. For example, only 10% ofmortgage loans in 2004 were issued in the domestic cur-rency (IMF, 2005). Because of the availability of foreigncurrency at relatively cheap rates some Kazakh banksincreased their holdings of real estate, investments andfinancial assets abroad. By 2007, the banking sector’sexternal debt was 43.2% of GDP (Barisitz andLahnsteiner, 2010). In addition, by the end of 2006, off-balance sheet items had grown to over three-quarters ofthe banking system’s balance sheet assets. Despite strongeconomic growth in Kazakhstan, the reliance of the bank-ing system on external finance represented a substantialrisk which ultimately proved to be excessive.

The sustained credit expansion in Kazakhstan was fol-lowed by a severe credit collapse in late 2007. The col-lapse was severe because the credit bubble had grown sobig. Typically a credit bubble in an emerging market ispreceded by a credit boom lasting about 3.5 years (IMF,2004). In the case of Kazakhstan, the credit boom lasted

Table 1. Structure of the Kazakh banking industry (1995–2010)

1995 1997 2000 2005 2007 2008 2009 2010

Number of commercial banks 130 81 48 34 35 37 37 39of which have a foreign stakeholder in authorized capital 8 22 16 14 18 13 – –Number of branches of commercial banks 1000 582 471 418 352 379 374 365

Sources: European Bank of Reconstruction and Development; National Bank of Kazakhstan; Financial Supervision Agency of theRepublic of Kazakhstan.Note: – indicates that the data is not available.

Bank performance and the financial crisis 123

Dow

nloa

ded

by [

Lou

isia

na S

tate

Uni

vers

ity]

at 0

6:22

20

Aug

ust 2

014

longer than is typically the case which fuelled the growthof the credit bubble. The Kazakh credit crisis was primar-ily due to moral hazard problems in the Kazakh bankingsector because too many risky loans, in particular mort-gages, were issued in the credit boom. The US subprimeand global financial crises made foreign investors morerisk averse, there was large scale deleveraging by foreigninvestors and there was a shift away from investing in realestate, all of which exacerbated the problems in theKazakh banking sector. This is because Kazakh bankswere heavily dependent on external finance and a largeshare of Kazakh banks’ loan portfolios were financing realestate projects. Kazakh banks were unable to roll overtheir external debts and service sizeable maturing liabil-ities. Confidence in the Kazakh banking sector declinedand consequently, in 2008 there was a marked fall inhousehold deposits as a share of GDP (see Table 2).

Accordingly, domestic liquidity conditions tightenedsignificantly in 2007. To support the liquidity of banks,the NBK reduced reserve requirements and arrangedlarge-scale liquidity provisions through foreign exchangeswaps, repo agreements and early redemption of NBKnotes. Furthermore, the NBK intervened heavily in theforeign exchange market to prop up the tenge. The mainvehicle for crisis relief in the Kazakh banking sector andreal economy was a newly formed state entity, Samruk-Kazyna (SK). At the end of 2007, the assets on SK’sconsolidated balance sheet totalled $47 billion (equivalentto 45% of GDP), which were largely from the Oil Fund(IMF, 2009b).2

Following the sharp depreciation of the Russian ruble,the tenge was devalued by almost a fifth in February 2009to improve the competitiveness of Kazakh exports, saveon foreign currency reserves and decrease the pressure onthe domestic currency. After the devaluation of the tenge,the debt-servicing burdens of Kazakh banks, which werealready struggling to refinance foreign funding, increased

further. In the period 2009 to 2010, two large domesticallyowned banks (the BTA and Alliance banks) and twosmaller domestically owned banks (Temir Bank andAstana Finance) stopped making principal payments andwere forced to restructure their external obligations (IMF,2010). Consequently, international rating agencies down-graded the debt ratings of systemically important Kazakhbanks. In addition, by 2009 the ratio of nonperformingloans to total lending across the sector had increased to38% compared to 7% in 2007. The sharp increase innonperforming loans and hence the rise in loan loss provi-sions resulted in almost zero industry profit in 2008. Largelosses were expected in future years because in 2008 thegap between banks’ assets and liabilities was equivalent to6% of GDP.

In response to the crisis in the Kazakh banking sector,the state holding company, SK, provided direct equitysupport to the three largest ailing banks so they couldrecapitalize. In February 2009, SK acquired a majoritystake in BTA (75%), the country’s largest bank. It alsoacquired minority stakes in the second and third largestbanks, Halyk Bank (21%) and Kazkommertsbank (20%),respectively. In total, SK provided $2.2 billion of supportto the three largest banks. A further $220 million was setaside to purchase more equity in Halyk Bank and $200million was earmarked for a stake in Alliance, the fourthlargest bank. SK used another $4 billion to support strug-gling real estate and construction projects, and to financeSME lending and development of the agricultural andindustrial sectors. Furthermore, SK made deposits in theKazakh banking system which amounted to 1:6% of GDP.

In 2009, BTA and Alliance had massive negative equitywhich amounted to 12.6% of GDP and was the reasontotal banking capital was negative (–915 million KZT).Although BTA and Alliance were nationalized theydefaulted on their foreign liabilities because the Kazakhgovernment only agreed to guarantee domestic liabilities.

Table 2. Kazakh banking sector indicators

2005 2006 2007 2008 2009 2010

Total assets (% of GDP) 60.6 87.5 87.8 74.1 68 61.9Foreign stake (% of total industry assets) 7.3 5.9 15.8 15.8 23.4 34Loans (% of GDP) 41.1 59.1 66.6 57.6 60.3 46.5Household deposits (% of GDP) 33.9 46.5 48.2 28.6 37.6 35.2Return on assets (%) 1.8 1.4 2.6 0.32 –1.4* –1.2*Return on equity (%) 14.1 14.7 22.87 2.6 –12.6* –10.6*Industry concentration (%)** 58.8 57.9 59.6 58.3 54.6 53.6Nonperforming loans (% of total loans) 3.3 2.4 2.7 7.1 37.8 33.7

Sources: National Bank of Kazakhstan and Financial Supervision Agency of the Republic of Kazakhstan.Notes: *Excludes Alyans Bank, Temir Bank and BTA bank.**Concentration is defined as the ratio of assets of the three largest banks to total bank sector assets.

2 The Oil Fund was set up in 2001 to manage Kazakhstan’s liquid surplus oil revenue to avoid a situation resembling the Dutch Disease(IMF, 2001).

124 A. J. Glass et al.

Dow

nloa

ded

by [

Lou

isia

na S

tate

Uni

vers

ity]

at 0

6:22

20

Aug

ust 2

014

Despite this, the dire forecasts of bank runs never materi-alized. In the second half of 2009, the Kazakh bankingsector started to recover and the recovery gatheredmomentum in 2010. This is because: (i) oil and commod-ity prices rebounded; (ii) real output growth remainedpositive because of looser monetary policy; (iii) thedebts of BTA and Alliance which they either defaultedon or restructured did not unduly affect the domesticdeposit base (IMF, 2010); (iv) limits were placed onbanks’ foreign exchange borrowing.

The recovery represented a period of relative stability inthe Kazakh banking system which fostered a steadyimprovement in liquidity and gradual rebuilding of thecapital base. Nevertheless, the volume of nonperformingloans as a share of total lending remains high by interna-tional standards which must be addressed to continue therehabilitation of the Kazakh financial system.

III. Empirical Methodology

The relationship between the quality and risk of the loanportfolio and economic and technical performance in theKazakh banking industry is investigated using the para-metric stochastic frontier framework. Technical efficiencyis analysed by fitting an input distance function, whereaseconomic efficiency is analysed by estimating cost, rev-enue and profit (standard and alternative) functions. Themodelling assumes that a bank’s observed deviation fromthe frontier is caused by random noise (v) and also possi-bly inefficiency (u). The former is a symmetric normallydistributed idiosyncratic error term which captures sam-pling, measurement and specification error. Inefficiency,on the other hand, is a one-sided nonnegative error term.

We model the industry best practice frontiers byaccounting for, first, interbank heteroscedasticity in u àla Kumbhakar et al. (1991) and Battese and Coelli (1995)and second, unobservable heterogeneity in banks’ tech-nologies by following the ‘true SFA’ framework (Greene,2005). In particular, we model the latent heterogeneity inbanks’ technologies using bank fixed effects. The obser-vable heterogeneity in banks’ technologies is modeled byallowing bank-specific characteristics to influence thefrontier. This specification is based on a single-step esti-mation procedure. It is therefore free from the bias asso-ciated with two-stage SFA techniques (Wang andSchmidt, 2002).

In this section, for the sake of brevity, we only discussthe cost function and its properties. See Appendix A1 forthe corresponding discussion pertaining to the input dis-tance, revenue, standard profit and alternative profit func-tions.3 The fitted cost functions (i.e. a function for theminimum cost required to produce outputs given theinput prices), c(y, w), for N banks over T periods are ofthe following form:

ln Cit=wKitð Þ ¼ γþ λi þ TL yit;wit=wKitð Þ þ ρ0zit þ vit

þ uit i ¼ 1; :::;N t ¼ 1; :::; T

(1)

where Cit is the observed cost of bank i at time t; yit is avector of output levels;wit is a vector of input prices; zit is avector characterizing the quality and risk of bank i’s loanportfolio at time t as well as other bank-specific features.The λi parameters are bank fixed effects which capture thelatent heterogeneity of banks which is not explained by thez-variables. The TL function in Equation 1 represents thetechnology as the translog approximation of the log of thecost function and is in terms of the output quantities andthe normalized inputs prices.4 vit is a symmetric normallydistributed idiosyncratic error term and uit is a measure ofhow far away bank i’s costs are at time t from the best-practice level associated with the same output quantitiesbeing produced under the same conditions. It is assumedthat uit follows a truncated normal distribution with a meanμit specific to each observation. This is a more flexibleassumption than assuming that uit follows a half-normaldistribution (see Stevenson, 1980, for further details). Theheteroscedastic frontier model assumes that μit is a func-tion of factors specific to bank i (zμ). The determinants ofthe stochastic frontier and uit are therefore simultaneouslyestimated. Specifically, the mean of the inefficiency dis-tribution in Equation 1 is specified as follows:

μit ¼ δ0 þ δ0z μ þ εit (2)

The microeconomic properties of an estimated costfunction, c(y, w), are: (i) non decreasing in outputs, y, ∂ ln c(y,w)=@ ln ym ; eym � 0;m ¼ 1; :::;M ; (ii) non decreasingin input prices, w, ∂ ln c (y, w) =@ ln wk ; ewk � 0;k ¼ 1; :::;K; (iii) homogeneity of degree one in input prices,w, c y;w=wKð Þ ¼ c y;wð Þ=wK ; (iv) a concave and continu-ous function in inputs prices, w.

3 Berger and Mester (1997) favour the alternative profit function over the standard specification if: (i) there are substantial unmeasureddifferences in the quality of services provided by banks because the alternative profit function holds the quantities of outputs constant andcaptures differences in quality by allowing output prices to vary; (ii) a bank cannot achieve every output scale and product mix becausethe quantities of its outputs exhibit very little variability; (iii) the banking industry is imperfectly competitive; (iv) there is likely to bemeasurement error in the data on output prices.4 The translog functional form is used in the model specification as it is more flexible than linear functions and captures cost (profit)behaviour of banks better. We thank the anonymous referee for the note.

Bank performance and the financial crisis 125

Dow

nloa

ded

by [

Lou

isia

na S

tate

Uni

vers

ity]

at 0

6:22

20

Aug

ust 2

014

IV. Data

We use rich monthly data obtained from the Agency of theRepublic of Kazakhstan on the Regulation andSupervision of the Financial Market and FinancialOrganizations for the period March 2007–December2010.5 Using this data, we analyse monthly changes inthe performance of Kazakh banks during the financialcrisis. The data includes financial, ownership and pruden-tial information for 37 second-tier commercial banks. Thepanel is unbalanced and consists of 1566 observationscovering virtually the entire Kazakh banking industry(99.7%). Only a few small new banks were omitted dueto zero values in their deposit and/or loan accounts.

We follow the Sealey and Lindley (1977) intermedia-tion approach. Accordingly, the intermediate deposits ofcommercial banks are split into various categories of earn-ing assets. The inputs in the banking production processare fixed capital (Fixed assets, x1), deposits (Clients’deposits, x2) and labour (Personnel expenses, x3). Theoutputs in the production process reflect both the lendingand nonlending activities of banks. In particular, the out-puts are: total customer loans (Customer loans, y1), invest-ment securities and other securities (Investments, y2), off-balance sheet items which is included as an output tocapture banks’ nontraditional activities (OBS, y3).

6 Thedata is expressed in real terms at March 2007 prices usingthe monthly CPI index which was obtained from theAgency of Statistics of the Republic of Kazakhstan.

In the cost and profit function specifications, we usethree input prices: the cost of physical capital (Price offixed assets, w1) which is calculated as the ratio of depre-ciation charges to the value of fixed assets; the cost ofborrowed funds (Price of clients’ deposits, w2) which istaken to be the ratio of interest expenses on clients’ depos-its to the volume of clients’ deposits; the cost of labour(Price of labour, w3) which is expressed as the ratio ofpersonnel expenses to total assets. The output prices usedin the revenue and profit functions are: the price of loans(Price of customer loans, p1), which is calculated bydividing interest income from loans by total customerloans; the price of other earning assets (Price of invest-ments, p2) which is expressed as the ratio of interestincome on investments and securities to the volume ofinvestments and securities; the price of OBS items (Price

of OBS, p3) which is calculated by dividing income gen-erated from OBS activity by the volume of OBS items.

We capture the quality and riskiness of a bank’s lendingusing two categorizations of its loan portfolio. Both cate-gorizations split a bank’s loan portfolio into standard,doubtful and bad loans. The first categorization is basedon the volume of loans whereas the second uses provisionsfor loan losses. In both cases there are a further five sub-categories of doubtful loans. Doubtful loans are categor-ized according to the score for the loan contract, where theleast risky doubtful contracts are placed in Category 1 andthe most risky are in Category 5.7 Using the first categor-ization, the quality and riskiness of a bank’s loan portfoliois captured using the ratio of the volume of loans in eachcategory to total lending. Similarly, using the second cate-gorization quality and riskiness is modelled using the ratioof loan loss provisions for each category to total loan lossprovisions. It is important to note that there is potentialendogeneity issue related to input prices, risk variablesand profit. The level of input/output prices and bankingrisk are not completely exogenous because they are tosome extent chosen by the bank as part of the bank’smanagement policy. Hence, the possible endogeneity ofthese variables can bias the coefficient estimates.Therefore, as noted by Berger and Mester (1997), thepresented analyses are suggestive but not conclusive.

Along with the risk variables, Z-score of the analysedbanks is included in the analysis to take into accountthe volatility of earnings. It is calculated as a ratio of thesum of the return on assets (ROAit) and the capital ratio(CARit) divided by the SD of the return on assets over theanalysed period betweenMarch 2007 and December 2010(SDROAi): i.e. Z - scoreit ¼ ðROAit þ CARitÞ=SDROAi.In Table 3, we present the descriptive statistics for theinputs, outputs, prices, loan quality variables and theZ-score.

We also include a number of other z-variables to capturethe effect of: (i) bank regulation and supervision i.e. dum-mies which take a value of 1 if a bank complies withprudential capital requirements (Prudential_Req) and lim-its on foreign currency positions (ForCur Limits); (ii) stateownership vis-à-vis private ownership, i.e., a Statedummy which takes a value of 1 if a bank is stateowned; (iii) bank size as a proxy for international financialintermediation and scale of performance, i.e., Small,

5 In May 2011, the NBK took over the responsibilities of the Agency through the newly formed Committee for the Control andSupervision of the Financial Market and Financial Organizations.6 OBS includes total contingent claims which contain letters of credit, guarantees, deposits and loans placed in the future, possible claimson bills, and the purchase and sale of financial derivatives.7 The score for a loan is calculated according to, among other things: the financial condition and rating of the borrower; the qualityof the collateral; any extensions to the repayment period; any write-offs of the borrower by other creditors; any overdue payments.Details of the five categories of doubtful loans are as follows: Category 1 – substandard loans with current payments; Category2 – substandard loans with payments in arrears; Category 3 – unsatisfactory loans with current payments; Category 4 –unsatisfactory loans with payments in arrears; Category 5 – doubtful loans.

126 A. J. Glass et al.

Dow

nloa

ded

by [

Lou

isia

na S

tate

Uni

vers

ity]

at 0

6:22

20

Aug

ust 2

014

Medium and Large bank size dummies8; (iv) negativeequity, i.e., a Negative Equity dummy which takes avalue of 1 if a bank has had negative equity at any pointover the sample period.

V. Results and Analysis

This section consists of four sub-sections. In section‘Returns to scale’, we present the fitted models and discussthe returns to scale estimates. In section ‘Efficiencyresults’, we present and analyse the estimates of economicefficiency (cost, revenue, standard profit and alternativeprofit efficiencies) and technical efficiency (input-orientedefficiency) over the sample period using each categoriza-tion of the quality and risk of the loan portfolio.Furthermore, in sections ‘Impact of loan portfolio qualityand risk on the frontiers’ and ‘Impact of loan portfolioquality and risk on inefficiency’, we discuss the findingson the impact of each categorization of the quality and riskof the loan portfolio on the industry best practice frontiersand the inefficiency estimates. To put the effect of eachcategorization on the best practice frontiers and the

inefficiencies into context, we also discuss the effect ofother z-variables.

Returns to scale

The estimation results for the two model specificationsfor each function – model 1 (quality and risk of theloan portfolio modeled using the volume of loans cate-gorization) and model 2 (quality and risk of the loanportfolio modeled using the loan loss provisions cate-gorization)– are presented in Appendix A2. We can seefrom these results that the input, output and correspond-ing price elasticities for all the fitted functions apartfrom the standard profit function have the expectedsigns where significant, implying that the monotonicityconditions are not disproved at the sample mean. Thereturns to scale estimates at the sample mean from thecost function (2.17 and 1.76 for models 1 and 2,respectively), input distance function (3.31 and 3.01for models 1 and 2, respectively), revenue function(2.15 and 2.04 for models 1 and 2, respectively) andalternative profit function (2.78 and 3.17 for models 1and 2, respectively) are in some cases considerablygreater than unity, i.e., an average Kazakh bank oper-ates at increasing returns to scale in all eight cases.9

Table 3. Descriptive statistics for inputs, outputs, prices and loan quality variables

Mean SD Mean SD

Outputs Quality of loans (%)Customer loans (y1) 158 064 026 337 414 737 DL/TA (z1) 0.254 0.233Investments (y2) 30 018 124 83 761 696 DL Category 1/DL (z2) 0.458 0.332OBS (y3) 226 292 194 540 920 253 DL Category 2/DL (z3) 0.164 0.228Inputs DL Category 3/DL (z4) 0.097 0.158Fixed assets (x1) 4 628 750 8 698 509 DL Category 4/DL (z5) 0.091 0.163Clients’ deposits (x2) 155 545 119 309 677 357 DL Category 5/DL (z6) 0.121 0.196Personnel expenses (x3) 182 666 281 382 BL/TA (z7) 0.061 0.12

BL/SL (z8) 0.319 3.231RDL/TA (z�1) 0.491 0.31

Output prices RDL Category 1/RDL (z�2) 0.247 0.285Price of customer loans (p1) 0.024 0.093 RDL Category 2/RDL (z�3) 0.139 0.218Price of investments (p2) 0.07 0.462 RDL Category 3/RDL (z�4) 0.133 0.187Price of OBS (p3) 0.087 0.694 RDL Category 4/RDL (z�5) 0.134 0.192Input prices RDL Category 5/RDL (z�6) 0.275 0.276Price of fixed assets (w1) 0.017 0.029 Reserves on bad loans/TA (z�7) 0.457 0.311Price of clients’ deposits (w2) 0.024 0.392Price of labour (w3) 0.121 0.205 Z – score 35.025 38.806

Notes: DL – doubtful loans; TA – total assets; BL – bad loans; SL – standard loans; RDL – reserves on doubtful loans. All values are atJanuary 2007 Kazakh tenge prices (000s).

8 The bank size dummies are based on a size categorization of banks according to total assets. In particular, banks are classified as: small– if their total assets are less than 10 000 000 (000s) KZT; medium – if their total assets are between 10 000 000 (000s) KZT and1 000 000 000 (000s) KZT; large – if their total assets exceed 1 000 000 000 (000s) KZT.9 Cost (Alternative profit) returns to scale (RTS) can be defined as the percentage change in cost (alternative profit) as a result of a onepercent increase in all outputs. In other words, cost (alternative profit) RTS are equal to the reciprocal of the sum of the cost (alternativeprofit) elasticities with respect to the outputs, i.e., ðP eyÞ�1. It follows therefore from a fitted input distance function thatRTSIDF ¼ �ðP eyÞ�1. Revenue RTS can be defined as the percentage change in revenue when there is a one percent increase in allinputs, i.e., RTSRF ¼ �ðP exÞ�1.

Bank performance and the financial crisis 127

Dow

nloa

ded

by [

Lou

isia

na S

tate

Uni

vers

ity]

at 0

6:22

20

Aug

ust 2

014

The above scale elasticities are computed at the samplemean of the data and are therefore unweighted averagemeasures. A country’s banking industry, however, isgenerally characterized by a large number of smallbanks where the scale elasticity is high and a smallnumber of large banks where the scale elasticity is low,possibly below one. Therefore, the above unweightedscale elasticities are most probably overestimates due toa large number of small banks in the sample as we usedata for virtually the entire Kazakh banking industry.Asset weighted average scale elasticities would belower than the corresponding unweighted elasticities.

Interestingly, revenue returns to scale from models1 and 2 are very similar. In contrast, there is a markedchange in cost, alternative profit and input distance econo-mies of scale depending on whether the measure of loanportfolio quality and risk is the volume of loans categor-ization or the loan loss provisions categorization. Therobustness or sensitivity of economies of scale to themeasure of loan portfolio quality and risk is most probablybecause the economic frontiers are sensitive to the affectof managerial discretion on loan loss provisions which is atool to manipulate bank profit. At the sample mean, invest-ments and OBS items are not the source of the differencebetween cost returns to scale for models 1 and 2 andbetween the estimates of alternative profit returns toscale because the parameter estimates for these outputsare all small. Our results suggest that the large differencebetween the cost returns to scale and the alternative profitreturns to scale from models 1 and 2 is due to financialintermediation activities. More specifically, we attributethe differences in cost returns to scale and alternative profit

returns to scale to differences in the marginal effects ofcustomer loan issuance in models 1 and 2.

Efficiency results

Average economic and technical efficiency scores forKazakh banks over the sample period are reported inTable 4. In general, the average efficiency scores overthe sample period using the volume of loans categoriza-tion range from 69% to 88%, the exception being theaverage efficiency score from the revenue function(47%). More specifically, the average efficiency scorefrom the cost model over the sample period suggests thatthe costs of an average bank are 12% above the bestpractice level. Similarly, mean profit efficiency scoresover the sample period of 79% and 80% suggest that, onaverage, banks make about 80% of the profit which thebest practice bank would make under the same conditions.

The Kruskal–Wallis tests of the null that the efficiencyscores from corresponding models (e.g. the two cost mod-els) in Table 4 do not differ is always rejected. That said, thedifference between the corresponding average standardprofit and alternative profit efficiencies is negligible. Overthe sample period, there is a 14% difference between theaverage efficiency scores from the cost functions and thereis a 6% difference between the average efficiency scoresfrom the revenue functions. The corresponding averageinput-oriented and average revenue efficiency scores arecharacterized by a somewhat downward trend which ismore evident in profit efficiency scores. A downwardtrend in average economic efficiency and average technicalefficiency is to be expected over the financial crisis.

Table 4. Economic and technical efficiency scores

CF IDF RF SPF APF

Model 1 (volume of loans categorization)2007 0.711 0.843 0.475 0.841 0.8162008 0.741 0.834 0.453 0.812 0.7912009 0.710 0.814 0.449 0.786 0.7812010 0.691 0.793 0.408 0.763 0.756Total sample 0.713 0.819 0.444 0.798 0.784

Model 2 (loan loss provisions categorization)2007 0.854 0.758 0.526 0.849 0.7842008 0.875 0.749 0.517 0.825 0.8262009 0.845 0.743 0.498 0.803 0.8022010 0.832 0.720 0.477 0.787 0.776Total sample 0.851 0.742 0.503 0.814 0.794

Model 1 versus Model 2Kruskal–Wallis test 47.34** 22.83** 23.08** 4.79* 12.57**

Notes: CF – cost function; IDF – input distance function; RF – revenue function; SPF – standard profit functions; APF –alternative profit function.The Kruskal–Wallis test statistic is Chi-square distributed with 1 degree of freedom.*Denotes significance at the 5% level and **denotes significance at the 1% level.

128 A. J. Glass et al.

Dow

nloa

ded

by [

Lou

isia

na S

tate

Uni

vers

ity]

at 0

6:22

20

Aug

ust 2

014

Average cost efficiencies are presented in Fig. 1. Theplot suggests that although there is a considerable differ-ence between the average cost efficiencies from models1 and 2, there is little difference between the average costefficiencies at the start of the sample period compared tothe end of the period. It is evident from Fig. 2 that this isalso the case for the average input-oriented and the aver-age revenue efficiencies. It is apparent, however, that theaverage efficiencies from the standard and alternativeprofit functions at the start of the sample period are well

above the levels which we observe at the end of the sampleperiod. Summarizing, this suggests that the financial crisisonly had a sustained detrimental effect on the averageefficiencies from the standard and the alternative profitfunctions.

We can see that average technical efficiency frommodel 1is well above that frommodel 2 for the entire sample period.Conversely, average revenue efficiency from model 2 ismuch higher than that from model 1 throughout the sampleperiod. The average profit (standard and alternative)

0.950

0.900

0.850

0.800

0.750

0.700

0.650

CE (model 1) CE (model 2)

0.600

Mar

-07

Jun-0

7

Sep-0

7

Dec-07

Mar

-08

Jun-0

8

Sep-0

8

Dec-08

Mar

-09

Jun-0

9

Sep-0

9

Dec-09

Mar

-10

Jun-1

0

Sep-1

0

Dec-10

Fig. 1. Average cost effciency (CE) of Kazakh banks

0.900

0.850

0.800

0.750

0.700

0.650

0.900

0.850

0.800

0.750

0.700

0.650

0.900

0.850

0.800

0.750

0.700

0.650

0.550

0.500

0.450

0.400

0.350

0.300

0.250

0.200

0.150

0.100

0.900

0.850

0.800

0.750

0.700

0.650

0.850

0.800

0.750

0.700

0.650

0.600

SPE (model 1) SPE (model 2) APE (model 1) APE (model 2)

Mar

-07

Jun-

07

Sep-

07

Dec

-07

Mar

-08

Jun-

08

Sep-

08

Dec

-08

Mar

-09

Jun-

09

Sep-

09

Dec

-09

Mar

-10

Jun-

10

Sep-

10

Dec

-10

Mar

-07

Jun-

07

Sep-

07

Dec

-07

Mar

-08

Jun-

08

Sep-

08

Dec

-08

Mar

-09

Jun-

09

Sep-

09

Dec

-09

Mar

-10

Jun-

10

Sep-

10

Dec

-10

Mar

-07

Jun-

07

Sep-

07

Dec

-07

Mar

-08

Jun-

08

Sep-

08

Dec

-08

Mar

-09

Jun-

09

Sep-

09

Dec

-09

Mar

-10

Jun-

10

Sep-

10

Dec

-10

Mar

-07

Jun-

07

Sep-

07

Dec

-07

Mar

-08

Jun-

08

Sep-

08

Dec

-08

Mar

-09

Jun-

09

Sep-

09

Dec

-09

Mar

-10

Jun-

10

Sep-

10

Dec

-10

SPE (model 1) APE (model 1) SPE (model 2) APE (model 2)

RE (model 1) RE (model 2)

Mar

-07

Jun-

07

Sep-

07

Dec

-07

Mar

-08

Jun-

08

Sep-

08

Dec

-08

Mar

-09

Jun-

09

Sep-

09

Dec

-09

Mar

-10

Jun-

10

Sep-

10

Dec

-10

TE (model 1) TE (model 2)

Mar

-07

Jun-

07

Sep-

07

Dec

-07

Mar

-08

Jun-

08

Sep-

08

Dec

-08

Mar

-09

Jun-

09

Sep-

09

Dec

-09

Mar

-10

Jun-

10

Sep-

10

Dec

-10

Fig. 2. Average economic and technical efficiencies of Kazakh banks

Bank performance and the financial crisis 129

Dow

nloa

ded

by [

Lou

isia

na S

tate

Uni

vers

ity]

at 0

6:22

20

Aug

ust 2

014

efficiencies from models 1 and 2, however, are very similarover the whole sample period. It is also evident that theaverage profit (standard and alternative) efficiencies aremuch more volatile than the other average efficiencies.That said, we observe at least one spike in the other averageefficiencies. Over the course of the financial crisis there arelarge isolated temporary changes in the average efficienciesfrom the cost, revenue and input distance functions. Forexample, the average cost efficiency from model 1 fell to66% in August 2008. Average cost efficiency from model 1once again fell to 64% in October 2009 but on this occasionis was the result of a steady decline over several months. InOctober 2009, the average cost efficiency from model 2 hadfallen to 79%. It is not surprising that the lowest average costefficiencies are observed in 2009 because it was a veryturbulent year for the Kazakh banking industry. As wasnoted in Section 2, in 2009, the level of nonperformingloans increased dramatically and in July 2009 the equitycapital of the banking system was negative.

The BTA and Alliance cost efficiencies from models 1and 2 are presented in Figs 3 and 4. The cost efficienciesadd credence to the anecdotal evidence on the troubles of

BTA and Alliance during the financial crisis. Apart fromthe sharp fall in the BTA cost efficiency scores inSeptember 2008, BTA was performing well relative tothe industry average up until April 2009, which waswhen BTA defaulted on its foreign liabilities. From April2009, the cost efficiency of BTA from both models dete-riorates rapidly over a period of several months. By July2009, BTA’s ratio of bad loans to total lending (54%) wasthe highest in the industry. According to the cost models,in June 2010 BTA was the worst performing bank in theindustry with an efficiency score of 6% from model 1 and31% from model 2.

The Alliance cost efficiencies over the sample per-iod follow similar paths to those for BTA. It should benoted, however, that although the Alliance and BTAcost efficiencies started to decline in April 2009, thedecline was much more sustained for BTA. This isevident because the Alliance cost efficiencies startedto rise in November 2009, whereas the BTA costefficiencies did not begin to rise until July 2010.This corroborates the view that BTA suffered moreduring the financial crisis than Alliance.

1.000

0.800

0.600

0.400

0.200

CE (model 1) CE (model 2)

0.000

Mar-0

7

Jun-07

Sep-07

Dec-07

Mar-0

8

Jun-08

Sep-08

Dec-08

Mar-0

9

Jun-09

Sep-09

Dec-09

Mar-1

0

Jun-10

Sep-10

Dec-10

Fig. 3. Cost efficiency (CE) of BTA bank

1.000

0.800

0.600

0.400

0.200

0.000

Mar-0

7

Jun-07

Sep-07

Dec-07

Mar-0

8

Jun-08

Sep-08

Dec-08

Mar-0

9

Jun-09

Sep-09

Dec-09

Mar-1

0

Jun-10

Sep-10

Dec-10

CE (model 1) CE (model 2)

Fig. 4. Cost efficiency (CE) of Alliance bank

130 A. J. Glass et al.

Dow

nloa

ded

by [

Lou

isia

na S

tate

Uni

vers

ity]

at 0

6:22

20

Aug

ust 2

014

Impact of loan portfolio quality and risk on the frontiers

The estimates of the effect of each z-variable on theindustry best practice frontiers and the bank inefficienciesare presented in Tables 5 and 6. Specifically, the estimatesin Table 5 are for model 1 (i.e. when loan portfolio qualityand risk is measured using the volume of loans categor-ization) and the estimates in Table 6 are for model 2 (i.e.when loan portfolio quality and risk is measured using theloan loss provisions categorization). A cursory glance atthe results indicates that theMedium bank size dummy hasbeen omitted to avoid perfect collinearity. This means thatthe effects of Small and Large banks are relative to theeffect of the Medium size category. Other z-variableswhich are omitted for the same reason are: the ratio ofthe volume of doubtful loans in the fifth category to thetotal volume of doubtful loans; the ratio of loan lossprovisions in the fifth category to total loan loss provi-sions; the ratio of the volume of standard loans to thevolume of total lending; the ratio of loan loss provisionsfor standard loans to total loan loss provisions.

Turning our attention to the effects of the z-variables onthe industry best practice frontiers (see the top panels of

Tables 5 and 6). It is evident that the ratio of the volume ofbad loans to the volume of standard loans has a smalleffect on each of the frontiers, which is similar in magni-tude in corresponding models, e.g., both the cost func-tions. It is surprising that all the BL/SL parameters are sosmall but this is because the effect of the quality and risk ofthe loan portfolio is being captured by other variables. Thelarge and significant effects of other variables which mea-sure the quality and risk of the loan portfolio often varybetween corresponding models. For example, it is appar-ent that relative to DL Category 5/DL, DL Category 3/DLhas a large negative effect on the revenue frontier andrelative to RDL Category 5/RDL, RDL Category 3/RDLhas a small positive effect on the revenue frontier. Also,we can see that DL/TA has quite a large negative effect onthe alternative profit frontier, whereas RDL/TA has quite alarge positive effect on the alternative profit frontier.

Interestingly, the fitted models suggest that if there is anincrease in BL/TA, there will be a big improvement in thecost and input distance best practice frontiers. Also, wefind that an increase in RBL/TAwill have a large positiveeffect on the alternative profit best practice frontier. All

Table 5. Impact on the frontiers and inefficiency (model 1)

Variable CF IDF RF SPF APF

Effect on the industry best practice frontierDL/TA 0.356*** 0.138* –0.665*** –0.162* –0.575***DL Category 1/DL –0.264** 0.007 –1.483*** –0.258*** –0.192**DL Category 2/DL 0.139* –0.01 –1.431*** –0.174** –0.029DL Category 3/DL 0.06 –0.015 –1.263*** –0.062 0.283*DL Category 4/DL 0.043 –0.104* –1.667*** –0.001 –0.005BL/TA –1.343*** 1.100*** –0.273 0.141 0.342BL/SL 0.004 0.007** –0.016** –0.001 –0.002Foreign –2.048*** 2.606*** –5.946*** –1.685*** 0.131State –1.862*** 1.986*** –1.332*** –2.186*** –0.311Small –0.901*** 0.940*** 3.494 0.099 0.169Large –0.097 –0.003 0.015 0.327*** 0.299*Negative Equity –1.057 0.155 0.511 –0.517*** –0.244*Z-score –0.565*** 0.526*** 0.372*** –0.006*** –0.005***

Effect on banking inefficiencyDL/TA –0.822*** 0.166 –1.601*** –3.581 –15.584**DL Category1/DL 0.344** 0.058 –1.872*** –7.689 –5.732**DL Category2/DL –0.407*** 0.13 –1.861*** –4.664 –2.047DL Category3/DL 0.062 0.051 –1.843*** –3.874 1.348DL Category4/DL –0.287* 0.026 –2.276*** –1.109 –0.457BL/TA 1.426*** 2.977*** 1.241* 4.723 6.588*Foreign 1.235 0.632*** 0.475 –2.421 –3.119State 0.049 –1.068*** 0.362* –0.308 0.762Small 0.410** –5.625 2.002*** –34.517 –31.464Large 0.443*** 2.353*** 4.419 0.037 –0.127Negative Equity 0.280** –0.174 –0.272 3.09 5.903**Prudential_Req –0.063 0.03 –0.127 –0.231 0.375ForCur Limits 0.189** 0.031 –0.275 –1.476 0.659

Notes: *Denotes significance at the 5% level, **denotes significance at the 1% level and *** denotes significance at the 10% level.DL – doubtful loans; TA – total assets; BL – bad loans; SL – standard loans. Monthly unbalanced data consists of 46 time periods (fromMarch 2007 to Dec 2010) of 37 banks (1566 observations).

Bank performance and the financial crisis 131

Dow

nloa

ded

by [

Lou

isia

na S

tate

Uni

vers

ity]

at 0

6:22

20

Aug

ust 2

014

these findings are consistent with ‘skimping’ behaviourwhich is where banks allocate less resources to loanscreening because they have short-run objectives of costminimization and revenue maximization. This can serve toimprove cost and revenue performance in the short termbut in the medium to long term it will have a detrimentaleffect on cost and revenue performance. ‘Skimping’ hasalso been used to explain the findings of other studies ofbank performance in both advanced and emerging econo-mies (e.g. Berger and DeYoung, 1997; Wheelock andWilson, 2000; Berger et al., 2009; Delis et al., 2011). Inthe other models where BL/TA or RBL/TA are significantthe parameters have the expected signs. This suggests thatan increase in BL/TA or RBL/TAwill have an undesirableeffect on industry best practice.

While the effect of the BL/SL variable on the bestpractice frontiers is similar in magnitude in the corre-sponding models in Tables 5 and 6, this if often not thecase for other z-variables. Negative Equity, for example,has quite a large positive and significant effect on the costfrontier in model 1 but it does not have a significant effecton the cost frontier in model 2. That said, the significantcoefficients on Negative Equity consistently suggest that it

has a detrimental effect on industry best practice.Furthermore, the significant Small and Large bank sizeparameters suggest that industry best practice willimprove if, relative to the number of medium-sizedbanks, there are more small and large banks. The Z-scoreparameter is significant in all functions for both models. Inparticular, in the input-distance and revenue function forboth models it has expected positive impact. That is bankswith higher Z-score, i.e. more stable banks with fewer risk,tend to perform better in terms technical and revenueactivities. The negative sign of Z-score in the cost functionis also expected as it suggests that more stable banks havelower costs. However, when it comes to profit functions,the negative sign is at odds with the expectations but inline with the findings relating to negative equity in the nextsub-section. It suggests that less stable banks with higherrisks tend to have significantly higher profits.

Impact of loan portfolio quality and risk on inefficiency

Moving on to consider the effects of the z-variables oneconomic inefficiency and technical inefficiency (see thebottom panels of Tables 5 and 6). It is evident that, in

Table 6. Impact on the frontiers and inefficiency (model 2)

Variable CF IDF RF SPF APF

Effect on the industry best practice frontierRDL/TA –0.027 –0.051 –2.788*** –0.326** 0.316*RDL Category 1/RDL 0.052 –0.001 –0.069 –0.100* –0.142*RDL Category 2/RDL 0.091* 0.018 0.197* –0.117* –0.11RDL Category 3/RDL –0.06 0.159*** 0.411*** 0.069 0.091RDL Category 4/RDL –0.005 –0.086* 0.072 0.096 0.059RBL/TA –0.064 0.036 –2.871*** –0.334** 0.405**Negative Equity –0.142 –0.182*** 1.024 –0.409*** –0.116BL/SL 0.001 0.005** –0.015* –0.003 –0.007*Foreign –0.416*** 0.821*** –0.823*** –1.567*** 0.323State –0.979*** 1.318*** –2.147*** –0.538*Small –0.710*** 1.256*** 3.275 0.135 0.215*Large –0.02 –0.012 –0.142 0.251** 0.253*Z-score –0.535*** 0.455*** 0.320*** –0.004*** –0.003*

Effect on banking inefficiencyRDL/TA 8.112* 0.197 –4.176*** –1.970* –4.608*RDL Category 1/RDL 0.022 –0.057 0.077 –4.854** –11.851*RDL Category 2/RDL –1.195* 0.036 0.237 –6.128* –7.908RDL Category 3/RDL 0.514* 0.305*** 0.323 –2.204* –3.061RDL Category 4/RDL –0.504 –0.064 0.042 –0.095 –0.487RBL/TA 8.258* 0.666*** –4.079*** –0.414 0.477Negative Equity 1.272*** –0.009 2.214*** –0.694 1.025Foreign 0.503** –0.567*** 0.2 –0.799 –0.156State 1.199*** –6.014 2.051*** –27.7 –36.452Small 0.851*** 1.340*** 4.734 0.828 0.993Large –0.065 –6.195 –0.646** 1.202* 1.798Prudential Req –0.1 0.023 –0.182 0.189 2.155ForCur Limits –0.062 0.214** –0.418* –1.946* –2.089

Notes: *Denotes significance at the 5% level, **denotes significance at the 1% level and *** denotes significance at the 0.1% level.RDL – reserves on doubtful loans; TA – total assets; RBL – reserves on bad loans; BL – bad loans; SL – standard loans.Monthly unbalanced data consists of 46 time periods (from March 2007 to Dec 2010) of 37 banks (1566 observations).

132 A. J. Glass et al.

Dow

nloa

ded

by [

Lou

isia

na S

tate

Uni

vers

ity]

at 0

6:22

20

Aug

ust 2

014

general, where the DL/TA and RDL/TA parameters aresignificant, the effect on inefficiency is negative. Theexception is the positive effect which RDL/TA has oncost inefficiency.

Where the BL/TA parameter is significant, the effect oninefficiency is positive. Sun and Chang (2011) also find apositive relationship between credit risk and inefficiencyfor banks in emerging Asian countries. Specifically, we findthat BL/TA has the biggest positive effect on alternativeprofit inefficiency and smaller positive effects on cost inef-ficiency and revenue inefficiency. This is most probablybecause, as Berger et al. (2009) note, profit efficiency is amore encompassingmeasure of economic performance andincorporates both cost and revenue efficiency. Furthermore,it is interesting to note that in both model specificationsforeign ownership has a significant negative effect on tech-nical inefficiency and a significant positive effect on costinefficiency. There is no consensus in the banking perfor-mance literature on whether foreign-owned banks are moreefficient than their domestic counterparts. For example,Jemric and Vujcic (2002) find that foreign-owned banksin Croatia are far more efficient than domestic banks.Similarly, Weill (2003) finds that foreign-owned banks inPoland and the Czech Republic are more efficient thandomestic banks, which he attributes to foreign banks havingmore human capital and better corporate governance.Conversely, other studies such as Hasan and Marton(2003) find that foreign banks in transition economies areless efficient than domestic banks. Lensink et al. (2008)analyse the performance of over 2000 banks in 105 coun-tries and conclude that foreign banks are less efficient thandomestic banks. They attribute the difference between theefficiency of domestic and foreign banks to banking systemconditions and the economic climate in the relevantcountries.

Since the most distressed banks in Kazakhstan weredomestic banks, our finding that foreign-owned bankswere less cost-efficient during the financial crisis couldwell be because they did not engage in ‘skimping’. Ifbanks skimp they are likely to allocate fewer resourcesto underwriting and monitoring loans, thereby savingcosts in the short term. Having said this, the quality ofthe loan portfolio declines because of ‘skimping’ and thesubsequent rise in nonperforming loans is likely toincrease a bank’s costs in the medium to long term. Ourresults are consistent with foreign banks allocating moreresources than domestic banks to screening loans, apprais-ing collateral and monitoring borrowers, rather than‘skimping’ on these costs. This would explain why overthe sample period foreign banks have a lower ratio ofnonperforming loans to total lending.

Both of the fitted input distance functions suggest that,relative to medium-sized banks, small banks are moretechnically inefficient. The second specification of therevenue function also suggests that small banks are

relatively inefficient. In contrast, the second specificationof the standard profit function and the alternative profitfunction suggest that large banks are more inefficient thantheir medium-sized counterparts. Comparing and contrast-ing all the above findings on the effect of bank size oninefficiency, with the estimates of the effect of bank sizeon the frontier suggests that an improvement in industrybest practice (i.e. the availability of better technology)does not necessarily lead to more efficient bankperformance.

The Negative Equity parameter is significant in the firstspecification of the input distance function. The NegativeEquity parameters are also significant in the second speci-fication of the cost function and revenue function. Asexpected all the significant Negative Equity parametersare positive, which suggests that Negative Equity leadsto more inefficient performance. This finding could wellbe a feature of banking performance during the financialcrisis because in an earlier study by Hasan and Marton(2003) a positive relationship between the level of equityand inefficiency is observed. Interestingly, our resultssuggest that Negative Equity does not affect standardprofit inefficiency, whereas we noted in section “Impactof loan portfolio quality and risk on the frontiers” thatNegative Equity has a detrimental effect on the standardprofit frontier in both of the fitted models and on thealternative profit frontier in the second specification.This suggests that the large negative equity of a smallnumber of banks during the financial crisis affected profitbest practice at the industry-wide level but not the profitinefficiency of individual banks.

Turning our attention to the effect on bank performanceof compliance with prudential capital requirements andlimits on foreign currency positions. In general, our find-ings suggest that compliance with prudential capitalrequirements or limits on foreign currency positions hasno implications for banking performance. There are, how-ever, four interesting exceptions- the positive and signifi-cant ForCurr Limits parameter in the cost function(model 1), and the significant positive coefficient onForCurr Limits in the second specification of the inputdistance function and negative coefficients in the revenueand standard profit function in model 2.

The positive and significant ForCurr Limits parameterin the fitted model 1 specification of the cost functionsuggests that more risk averse banks which comply withthe limits on foreign currency positions are more cost-inefficient. Being more cost-efficient in the short term bynot complying with the regulations is consistent withbanks being less risk averse and ‘skimping’, which willultimately erode the quality of the loan portfolio.Hellmann et al. (2000) develop a theoretical modelwhich shows that financial liberalization gives rise tosuch problems. The intuition behind their model is asfollows. Financial liberalization promotes entry to the

Bank performance and the financial crisis 133

Dow

nloa

ded

by [

Lou

isia

na S

tate

Uni

vers

ity]

at 0

6:22

20

Aug

ust 2

014

banking industry which intensifies competition and erodesbank profit. The capitalized value of expected profits (i.e.the franchise value) falls and there is less incentive to issuegood loans to preserve the franchise value. In their model,banks choose between investing in a prudent asset whichyields a high expected return or investing in a gamblingasset which yields a very high return if the gamble pays-off but if it doesn’t depositors bear the cost of the invest-ment. The conclusion which is reached from the model isthat following financial liberalization there will be a biggerincentive to invest in gambling assets. This model and, inparticular, the scenario where gambles do not pay-off isakin to developments in the Kazakh banking sector duringthe financial crisis.

Upon receiving favourable credit ratings from majorinternational credit agencies, many Kazakh banksengaged in a range of new foreign exchange activities.The second specification of the input distance functionsuggests that banks are more technically inefficient ifthey are less risk averse and exceed limits on foreigncurrency positions. Along similar lines, the second speci-fication of the revenue and standard profit functions sug-gests that banks which exceed limits on foreign currencypositions are more inefficient. This is most probablybecause more risk averse banks which comply with limitson foreign currency positions are less exposed to riskyforeign currency operations.

VI. Lessons from the Kazakh Experience

Using rich monthly data for virtually the entire Kazakhbanking industry over the financial crisis, SFA is appliedto the financial intermediation approach to fit a number offunctions. Among other things, from the fitted functionswe estimate the effects of two measures of the quality andriskiness of the loan portfolio on the industry best practicefrontiers and the technical and economic inefficiencies ofindividual banks. The determinants of the bank technolo-gies and bank inefficiencies shed light on Kazakh bankbehaviour over the financial crisis. For example, we findthat an increase in the volume of bad loans as a ratio oftotal lending has a desirable effect on the cost, input-distance and alternative profit frontiers, all of which isconsistent with the ‘skimping’ hypothesis. In the loanloss provisions specifications, an increase in the ratio ofreserves on bad loans to total reserves has the expectedsignificant detrimental effect on the cost, input distanceand revenue frontiers, which is at odds with the ‘skimping’hypothesis. Given there is a lot of anecdotal evidence of‘skimping’ by Kazakh banks during the financial crisisand nonperforming loans is still a big concern in theKazakh banking industry, the ratio of the volume of badloans to total lending, rather than the ratio of reserves on

bad loans to total reserves, should be used by the regula-tory authorities to monitor nonperforming loans. To con-clude, an empirical analysis of Kazakh banking efficiencyusing nonparametric frontier models with bootstrappingwould be a worthwhile and interesting area for furtherwork.

Acknowledgements

The authors would like to thank an anonymous referee forconstructive comments which helped to improve thearticle substantially.

References

Altunbas, Y., Liu, M., Molyneux, P. et al. (2000) Efficiency andrisk in Japanese banking, Journal of Banking and Finance,24, 1605–28.

Barisitz, S. and Lahnsteiner, M. (2010) From stormy expansionto riding out the storm: banking development inKazakhstan, Austrian Central Bank Financial StabilityReport, 19, 62–71.

Battese, G. E. and Coelli, T. J. (1995) A model for technicalinefficiency effects in a stochastic frontier production func-tion for panel data, Empirical Economics, 20, 325–32.

Berger, A. N. and DeYoung, R. (1997) Problem loans and costefficiency in commercial banks, Journal of Banking andFinance 21, 849–70.

Berger, A. N., Hasan, I. and Zhoud, M. (2009) Bank ownershipand efficiency in China: what will happen in the world’slargest nation?, Journal of Banking and Finance, 33,113–30.

Berger, A. N. and Mester, L. J. (1997) Inside the black box: whatexplains differences in the efficiencies of financial institu-tions?, Journal of Banking and Finance, 21, 895–947.

Brissimis, S. N., Delis, M. D. and Papanikolaou, N. I. (2008)Exploring the nexus between banking sector reform andperformance: evidence from newly acceded EU countries,Journal of Banking and Finance, 32, 2674–83.

Chiu, Y.-H. and Chen, Y.-C. (2009) The analysis of Taiwanesebank efficiency: incorporating both external environmentrisk and internal risk, Economic Modelling, 26, 456–63.

De Haas, R., Ferreira, D. and Taci, A. (2010) What determinesthe composition of banks. Loan portfolios? Evidence fromtransition countries, Journal of Banking and Finance, 34,388–98.

Delis, M. D., Molyneux, P. and Pasiouras, F. (2011) Regulationsand productivity growth in banking: evidence from transi-tion economies, Journal of Money, Credit and Banking, 43,735–64.

Fries, S. and Taci, A. (2005) Cost efficiency of banks in transi-tion: evidence from 289 banks in 15 post-communist coun-tries, Journal of Banking and Finance, 29, 55–81.

Greene, W. (2005) Reconsidering heterogeneity in panel dataestimators of the stochastic frontier model, Journal ofEconometrics, 126, 269–303.

Hasan, I. and Marton, K. (2003) Development and efficiency ofthe banking sector in a transitional economy: Hungarianevidence, Journal of Banking and Finance, 27, 2249–71.

134 A. J. Glass et al.

Dow

nloa

ded

by [

Lou

isia

na S

tate

Uni

vers

ity]

at 0

6:22

20

Aug

ust 2

014

Havrylchyk, O. (2006) Efficiency of the Polish banking industry:foreign versus domestic banks, Journal of Banking andFinance, 30, 1975–96.

Hellmann, T. F., Murdock, K. C. and Stiglitz, J. E. (2000)Liberalization, moral hazard in banking and prudential reg-ulation: are capital requirements enough?, AmericanEconomic Review, 90, 147–65.

Hoelscher D. S. (1998) Banking system restructuring inKazakhstan, IMF Working Paper WP/98/96, IMF,Washington, DC.

Hughes, J. P. andMester, L. J. (1993) A quality and risk-adjustedcost function for banks: evidence on the ‘Too-Big-to-Fail’doctrine, Journal of Productivity Analysis, 4, 292–315.

Iannotta, G., Nocera, G. and Sironi, A. (2007) Ownership struc-ture, risk and performance in the European banking indus-try, Journal of Banking and Finance, 31, 2127–49.

IMF (1998) Republic of Kazakhstan – recent economic devel-opments, IMF Staff Country Report No. 98/84, IMF,Washington, DC.

IMF (2001) Republic of Kazakhstan, IMF Staff Country ReportNo. 01/20, IMF, Washington, DC.

IMF (2004) Are credit booms in emerging markets a concern?,World Economic Outlook, April, 103–46. IMF,Washington, DC.

IMF (2005) Republic of Kazakhstan: selected issues, IMF StaffCountry Report No. 05/240, IMF, Washington, DC.

IMF (2009a) Initial lessons of the crisis, Prepared by theResearch, Monetary and Capital Markets, and Strategy,Policy and Review Departments, IMF, Washington, DC.

IMF (2009b) Republic of Kazakhstan: 2009 article IVconsultation — staff report; supplement; and public infor-mation notice on the executive board discussion. CountryReport No. 09/300, IMF, Washington, DC.

IMF (2010) Republic of Kazakhstan: 2010 article IVconsultation – staff report; public information notice onthe executive board discussion; and statement by the execu-tive director for Kazakhstan. Country Report No. 10/241,IMF, Washington, DC.

Isik, I. and Hassan, M. K. (2002) Technical, scale and allocativeefficiencies of Turkish banking industry, Journal ofBanking and Finance, 26, 719–66.

Jemric, I. and Vujcic, B. (2002) Efficiency of banks in Croatia: aDEA approach, Comparative Economic Studies, 44,169–93.

Kenjegalieva, K., Simper, R., Weyman-Jones, T. et al. (2009)Comparative analysis of banking production frameworks inEastern European financial markets, European Journal ofOperations Research, 198, 326–40.

Kraft, E. and Tirtiroglu, D. (1998) Bank efficiency in Croatia: astochastic-frontier analysis, Journal of ComparativeEconomics, 26, 282–300.

Kumbhakar, S. C., Ghosh, S. and McGuckin, J. T. (1991) Ageneralized production frontier approach for estimatingdeterminants of inefficiency in US dairy farms, Journal ofBusiness and Economic Statistics, 9, 279–86.

Kwan, S. H. (2003) Operating performance of banks amongAsian economies: an international and time series compar-ison, Journal of Banking and Finance, 27, 471–89.

Laeven, L. and Valencia, F. (2010) Resolution of banking crises:the good, the bad, and the ugly, IMFWorking PaperWP/10/146, IMF, Washington, DC.

Lensink, R., Meesters, A. and Naaborg, I. (2008) Bank efficiencyand foreign ownership: do good institutions matter?,Journal of Banking and Finance, 32, 834–44.

Mester, L. J. (1996) A study of bank efficiency taking intoaccount risk-preferences, Journal of Banking andFinance, 20, 1025–45.

Pasiouras, F. (2008) Estimating the technical and scale efficiencyof Greek commercial banks: the impact of credit risk, off-balance sheet activities, and international operations,Research in International Business and Finance, 22,301–18.

Sealey, C. and Lindley, J. T. (1977) Inputs, outputs and a theoryof production and cost at depository financial institutions,Journal of Finance, 32, 1251–66.

Stevenson, R. E. (1980) Likelihood functions for generalizedstochastic frontier estimation, Journal of Econometrics,13, 58–66.

Sun, L. and Chang, T. (2011) A comprehensive analysis of theeffects of risk measures on bank efficiency: evidence fromemerging Asian countries, Journal of Banking and Finance,35, 1727–35.

Turk Ariss, R. (2010) On the implications of market power inbanking: evidence from developing countries, Journal ofBanking and Finance, 34, 765–5.

Wang, H. and Schmidt, P. (2002) One-step and two-step estima-tion of the effects of exogenous variables on technicalefficiency levels, Journal of Productivity Analysis, 18,129–44.

Weill, L. (2003) Banking efficiency in transition economies: therole of foreign ownership, Economics of Transition, 11,569–92.

Wheelock, D. C. and Wilson, P. W. (2000) Why do banksdisappear? The determinants of US bank failures andacquisitions, Review of Economics and Statistics, 82,127–38.

Bank performance and the financial crisis 135

Dow

nloa

ded

by [

Lou

isia

na S

tate

Uni

vers

ity]

at 0

6:22

20

Aug

ust 2

014

Appendix A

A1 Four Further Functions and TheirProperties

In addition to cost efficiency which was discussed inSection III, four other efficiency measures are estimatedin this article. The four other efficiency measures areestimated by fitting an input distance function (IDF),revenue function (RF) and two specifications of the profitfunction, standard (SPF) and alternative (APF). Thesefunctions have the following forms:

� ln xKit ¼ γþ λi þ TL yit; xit=xKitð Þ þ ρ0zit þ vit � uit(IDF)

�ln Rit=pKitð Þ ¼ γþ λi þ TL xit; pit=pKitð Þþ ρ0zit þ vit � uit

(RF)

�ln πit þ θ=wKitð Þ ¼ γþ λi þ TL pit;wit=wKitð Þþ ρ0zit þ vit � uit

(SPF)

�ln πit þ θ=wKitð Þ ¼ γþ λi þ TL yit;wit=wKitð Þþ ρ0zit þ vit � uit;

(APF)

where x is a matrix of inputs in the banking technology; yis a matrix of outputs; w is a matrix of input prices; p is amatrix of output prices; R is a vector of banks’ revenues; πis a vector of banks’ profits; θ is a constant added to everybank’s profit so that the natural log is a positive number.

As was the case for the cost function the error compo-nents are v and u. Unlike for the cost function, however,the inefficiency term u is subtracted for the other func-tions. This is because cost inefficiency measures the dis-tance to the minimum attainable cost whereas standardprofit, alternative profit and revenue inefficiency measurethe distance to the maximum achievable level. Technicalefficiency from a fitted input distance function measureshow close a bank’s usage of resources is to the minimumattainable level of inputs. Given the input distance func-tion is an input requirement function multiplied by –1,inefficiency is subtracted rather than added.

Information on the quantities of inputs and outputs isneeded to represent the production technology using theinput distance function,DI y; xð Þ.DI y; xð Þ is defined accord-ing to the input set and satisfies the following properties:

(i) nondecreasing in inputs x,@ ln DI=@ ln xk ; exk � 0, k ¼ 1; :::;K;

(ii) nonincreasing in outputs y,@ ln DI=@ ln ym ; eym � 0, m ¼ 1; :::;M ;

(iii) homogeneity of degree one in inputsx, DI y; x=xKð Þ ¼ DI y; xð Þ=xK ;

(iv) concave and continuous function in inputs x.

Estimating the revenue function, rðx; pÞ, is basedon the assumption that producers attempt to maxi-mize their revenue. Information on inputs and outputprices is required to estimate the revenue function.The revenue function satisfies the followingproperties:

(i) nondecreasing in output prices p,@ ln r x; pð Þ=@ ln pm ; epm � 0, m ¼ 1; :::;M ;

(ii) nondecreasing in inputs x,@ ln r x; pð Þ=@ ln xk ; exk � 0, k ¼ 1; :::;K;

(iii) homogeneity of degree one in output pricesp, r y; p=pMð Þ ¼ r y; pð Þ=pM

(iv) convex and continuous function in outputprices p.

To fit the standard profit function, πðp;wÞ, informationon input and output prices is required. The standard profitfunction has the following properties:

(i) nondecreasing in output prices p,@ ln π w; pð Þ=@ ln pm ; epm � 0, m ¼ 1; :::;M ;

(ii) nonincreasing in input prices w,@ ln π w; pð Þ=@ ln wk ; ewk � 0, k ¼ 1; :::;K;

(iii) homogeneity of degree one in output prices p andinput prices w, π w; p=pMð Þ ¼ π w; pð Þ=pMand π w=wK ; pð Þ ¼ π w; pð Þ=wK ;

(iv) convex and continuous function in output pricesp and input prices w.

The right-hand side of the alternative profit functionis the same as the right-hand side of the cost function.The standard and alternative profit functions have thesame dependent variable. The alternative profit func-tion, π�ðy;wÞ, therefore holds output levels constant,as is the case in the cost function, and allows outputprices to vary and influence profit. The properties ofthe alternative profit function are:

(i) nondecreasing in outputs y,@ ln π� y;wð Þ=@ ln ym ; eym � 0, m ¼ 1; :::;M ;

(ii) nonincreasing in input prices w,@ ln π� y;wð Þ=@ ln wk ; ewk � 0, k ¼ 1; :::;K;

(iii) homogeneity of degree one in input pricesw, π� y;w=wKð Þ ¼ π� y;wð Þ=wK ;

(iv) convex and continuous function in inputprices w.

All the functions fitted here contain the translog approx-imation of the relevant function. The general form of the

136 A. J. Glass et al.

Dow

nloa

ded

by [

Lou

isia

na S

tate

Uni

vers

ity]

at 0

6:22

20

Aug

ust 2

014

translog approximation of each function for panel data attime t is as follows:

TL a; bð Þ ¼ α0laþ β0lbþ 1

2Ala0laþ 1

2Blb0lbþ laΓlb

þ η1t þ1

2η2t

2 þ ψ0lat þ f0lbt;

where la ¼ lnðaÞ; lb ¼ lnðbÞ; η1 and η2 are parameters to beestimated; α′, β′, ψ′ and ϕ′ are vectors of parameters to beestimated; A, B and Γ are matrices of parameters to beestimated. The signs of the elements of α′ and β′ indicatewhether the monotonicity conditions of TLða; bÞ aresatisfied.

Estimation results for model 1 (volume of loans categorization)

CF IDF RF SPF APF

TC=w1 LHS � lx1 LHS TR=p1 LHS π�=w1 LHS π�=w1 LHS

y1 0.385*** y1 –0.274*** x1 0.033 p1 –0.112*** y1 0.344***y2 0.032*** y2 –0.013** x2 0.346*** p2 –0.006 y2 0.002y3 0.044*** y3 –0.015* x3 0.087** p3 0.024*** y3 0.014w2=w1 0.198*** x2=x1 0.487*** p2=p1 0.058*** w2=w1 0.092*** w2=w1 –0.051**w3=w1 0.602*** x3=x1 0.138*** p3=p1 0.307*** w3=w1 0.502*** w3=w1 0.603***

ðy1Þ2 0.048*** ðy1Þ2 –0.034*** ðx1Þ2 –0.060*** ðp1Þ2 0.003 ly11 0.077***ðy2Þ2 0.004*** ðy2Þ2 –0.002* ðx2Þ2 –0.008 ðp2Þ2 0.005 ly22 0.001ðy3Þ2 –0.001 ðy3Þ2 –0.002* ðx3Þ2 –0.02 ðp3Þ2 0.010*** ly33 0.008***y1y2 –0.002 y1y2 0.003 x1x2 0.117*** p1p2 –0.049*** ly12 –0.014**y1y3 0.003 y1y3 –0.004 x1x3 –0.007 p1p3 0.034*** ly13 –0.024***y2y3 –0.004*** y2y3 0.002* x2x3 0.017 p2p3 –0.007** ly23 0.006**ðw2=w1Þ2 0.004*** ðx2=x1Þ2 0.055*** ðp2=p1Þ2 0.000 ðw2=w1Þ2 0.021*** ðw2=w1Þ2 0.037***ðw3=w1Þ2 0.055*** ðx3=x1Þ2 0.006 ðp3=p1Þ2 0.028*** ðw3=w1Þ2 0.084*** ðw3=w1Þ2 0.147***ðw2=w1Þ� ðx2=x1Þ� ðp2=p1Þ� ðw2=w1Þ� ðw2=w1Þ�ðw3=w1Þ –0.043*** ðx3=x1Þ –0.014* ðp3=p1Þ 0.025*** ðw3=w1Þ –0.058*** ðw3=w1Þ –0.113***y1ðw2=w1Þ 0.034*** y1ðx2=x1Þ 0.002 x1ðp2=p1Þ 0.016 p1ðw2=w1Þ 0.033*** y1ðw2=w1Þ –0.020**y1ðw3=w1Þ –0.070*** y1ðx3=x1Þ 0.01 x1ðp3=p1Þ –0.026* p1ðw3=w1Þ –0.058*** y1ðw3=w1Þ –0.001y2ðw2=w1Þ –0.004*** y2ðx2=x1Þ 0.003* x2ðp2=p1Þ 0.011 p2ðw2=w1Þ –0.015*** y2ðw2=w1Þ 0.002y2ðw3=w1Þ 0.023*** y2ðx3=x1Þ 0 x2ðp3=p1Þ 0.042*** p2ðw3=w1Þ 0.037*** y2ðw3=w1Þ 0.006y3ðw2=w1Þ 0.009*** y3ðx2=x1Þ 0.004 x3ðp2=p1Þ –0.020* p3ðw2=w1Þ 0.003 y3ðw2=w1Þ –0.021***y3ðw3=w1Þ –0.017*** y3ðx3=x1Þ –0.014** x3ðp3=p1Þ 0.013 p3ðw3=w1Þ –0.061*** y3ðw3=w1Þ 0.032**

t 0.009*** t –0.009*** t 0.006*** t –0.001 t 0.000t2 –0.000*** t2 0.000 t2 0.000*** t2 0.000 t2 0.000y1t 0.002*** y1t 0.001** x1t –0.003** p1t –0.003** y1t –0.002y2t 0.000 y2t –0.000** x2t –0.002* p2t –0.001* y2t 0.000y3t –0.001*** y3t 0.000 x3t 0.002 p3t 0.001 y3t 0.003***ðw2=w1Þt –0.003*** ðx2=x1Þt –0.002*** ðp2=p1Þt 0.002*** ðw2=w1Þt –0.003*** ðw2=w1Þt –0.006***ðw3=w1Þt 0.000 ðx3=x1Þt –0.001* ðp3=p1Þt 0.003*** ðw3=w1Þt 0.000 ðw3=w1Þt 0.004***

Log-like-lihood288.79 466.92 –967.21 –327.06 –666.74

Notes: CF – cost function; IDF – input distance function; RF – revenue function; SPF – standard profit function; APF – alternative profitfunction. All regressions include individual fixed effects for banks (not reported in the table).*Denotes significance at the 5% level, **denotes significance at the 1% level and ***denotes significance at the 10% level.

A2 The Fitted Functions

Bank performance and the financial crisis 137

Dow

nloa

ded

by [

Lou

isia

na S

tate

Uni

vers

ity]

at 0

6:22

20

Aug

ust 2

014

Estimation results for model 2 (volume of loans categorization)

CF IDF RF SPF APF

TC=w1 LHS � lx1 LHS TR=p1 LHS π�=w1 LHS π�=w1 LHSy1 0.502*** y1 –0.288*** x1 –0.027 p1 –0.146*** y1 0.289***y2 0.028*** y2 –0.023*** x2 0.387*** p2 0.000 y2 0.012y3 0.037*** y3 –0.021*** x3 0.129*** p3 0.025*** y3 0.014w2=w1 0.225*** x2=x1 0.465*** p2=p1 0.034*** w2=w1 0.115*** w2=w1 –0.064***w3=w1 0.566*** x3=x1 0.127*** p3=p1 0.292*** w3=w1 0.490*** w3=w1 0.614***

ðy1Þ2 0.068*** ðy1Þ2 –0.054*** ðx1Þ2 –0.074*** ðp1Þ2 –0.013* ly11 0.075***ðy2Þ2 0.003*** ðy2Þ2 –0.004*** ðx2Þ2 –0.018* ðp2Þ2 0.007** ly22 0.001ðy3Þ2 –0.001 ðy3Þ2 –0.002* ðx3Þ2 –0.022 ðp3Þ2 0.008*** ly33 0.007***y1y2 0.001 y1y2 0.007** x1x2 0.168*** p1p2 –0.053*** ly12 –0.008y1y3 –0.010** y1y3 0.000 x1x3 –0.02 p1p3 0.038*** ly13 –0.024***y2y3 –0.001 y2y3 0.000 x2x3 0.018 p2p3 –0.004 ly23 0.006**ðw2=w1Þ2 0.007*** ðx2=x1Þ2 0.041*** ðp2=p1Þ2 0.002 ðw2=w1Þ2 0.021*** ðw2=w1Þ2 0.041***ðw3=w1Þ2 0.062*** ðx3=x1Þ2 0.014*** ðp3=p1Þ2 0.031*** ðw3=w1Þ2 0.087*** ðw3=w1Þ2 0.172***ðw2=w1Þ� ðx2=x1Þ� ðp2=p1Þ� ðw2=w1Þ� ðw2=w1Þ�ðw3=w1Þ –0.037*** ðx3=x1Þ –0.024*** ðp3=p1Þ 0.016*** ðw3=w1Þ –0.060*** ðw3=w1Þ –0.138***y1ðw2=w1Þ 0.056*** y1ðx2=x1Þ –0.026** x1ðp2=p1Þ 0.020* p1ðw2=w1Þ 0.035*** y1ðw2=w1Þ –0.039***y1ðw3=w1Þ –0.047*** y1ðx3=x1Þ 0.014 x1ðp3=p1Þ –0.004 p1ðw3=w1Þ –0.058*** y1ðw3=w1Þ 0.029y2ðw2=w1Þ –0.004*** y2ðx2=x1Þ 0.008*** x2ðp2=p1Þ 0.011 p2ðw2=w1Þ –0.021*** y2ðw2=w1Þ 0.003y2ðw3=w1Þ 0.021*** y2ðx3=x1Þ 0.007** x2ðp3=p1Þ 0.047*** p2ðw3=w1Þ 0.038*** y2ðw3=w1Þ 0.011*y3ðw2=w1Þ 0.009** y3ðx2=x1Þ 0.010* x3ðp2=p1Þ –0.020* p3ðw2=w1Þ 0.003 y3ðw2=w1Þ –0.021***y3ðw3=w1Þ –0.030*** y3ðx3=x1Þ –0.004 x3ðp3=p1Þ –0.018 p3ðw3=w1Þ –0.059*** y3ðw3=w1Þ 0.024*

t 0.009*** t –0.006*** t 0.003* t 0.000 t 0.000t2 –0.000*** t2 0.000 t2 0.000*** t2 0.000 t2 0.000y1t 0.003*** y1t 0.000 x1t 0.000 p1t –0.003** y1t –0.001y2t 0.000 y2t 0.000 x2t –0.003*** p2t –0.001** y2t 0.000y3t –0.001* y3t 0.000 x3t 0.001 p3t 0.001 y3t 0.003***ðw2=w1Þt –0.003*** ðx2=x1Þt –0.003*** ðp2=p1Þt 0.001* ðw2=w1Þt –0.003*** ðw2=w1Þt –0.005***ðw3=w1Þt 0.001* ðx3=x1Þt –0.001* ðp3=p1Þt 0.002** ðw3=w1Þt 0.000 ðw3=w1Þt 0.003**

Log-like-lihood333.22 520.47 –778.63 –267.06 –614.56

Notes: CF – cost function; IDF – input distance function; RF – revenue function; SPF – standard profit function; APF – alternative profitfunction. All regressions include individual fixed effects for banks (not reported in the table).*Denotes significance at the 5% level, **denotes significance at the 1% level and ***denotes significance at the 10% level.

138 A. J. Glass et al.

Dow

nloa

ded

by [

Lou

isia

na S

tate

Uni

vers

ity]

at 0

6:22

20

Aug

ust 2

014