Don't Know

34
Journal of Financial Economics 87 (2008) 706–739 Financial distress and corporate risk management: Theory and evidence $ Amiyatosh Purnanandam Ross School of Business, University of Michigan, Ann Arbor, MI 48109, USA Received 28 February 2006; received in revised form 2 April 2007; accepted 10 April 2007 Available online 14 December 2007 Abstract This paper extends the current theoretical models of corporate risk-management in the presence of financial distress costs and tests the model’s predictions using a comprehensive data set. I show that the shareholders optimally engage in ex- post (i.e., after the debt issuance) risk-management activities even without a pre-commitment to do so. The model predicts a positive (negative) relation between leverage and hedging for moderately (highly) leveraged firms. Consistent with the theory, empirically I find a non-monotonic relation between leverage and hedging. Further, the effect of leverage on hedging is higher for firms in highly concentrated industries. r 2008 Elsevier B.V. All rights reserved. JEL classification: G30; G32 Keywords: Hedging; Risk-shifting; Asset substitution; Derivatives 1. Introduction This paper develops and tests a theory of corporate risk management in the presence of financial distress costs. The existing literature shows that hedging can lead to firm value maximization by limiting deadweight losses of bankruptcy (see Smith and Stulz, 1985). 1 These models justify only ex-ante risk-management ARTICLE IN PRESS www.elsevier.com/locate/jfec 0304-405X/$ - see front matter r 2008 Elsevier B.V. All rights reserved. doi:10.1016/j.jfineco.2007.04.003 $ This paper is based on a chapter of my Ph.D. dissertation at Cornell University. I would like to especially thank an anonymous referee for several useful suggestions during the reviewing process. I am grateful to George Allayannis, Warren Bailey, Sugato Bhattacharya, Sreedhar Bharath, Sudheer Chava, Thomas Chemmanur, Wayne Ferson, Ken French, John Graham, Robert Goldstein, Yaniv Grinstein, Jerry Haas, Pankaj Jain, Kose John, Haitao Li, Roni Michaely, M.P.Narayanan, Maureen O’Hara, Paolo Pasquariello, Mitch Petersen, Uday Rajan, William Schwert (the editor), David Weinbaum, Rohan Williamson, and seminar participants at Boston College, Cornell, Darden, Emory, London Business School, University of Michigan, Notre Dame, University of Rochester, The Lehman Brothers Finance Fellowship Competition 2003, and the Western Finance Association’s 2005 meetings for valuable comments and suggestions. I am particularly grateful to Bob Jarrow and Bhaskaran Swaminathan for their advice. All remaining errors are mine. Tel.: +1 734 764 6886; fax: +1 734 936 8715. E-mail address: [email protected] 1 Other motivations for corporate hedging include convexity of taxes, managerial risk-aversion (Stulz, 1984; Smith and Stulz, 1985) underinvestment costs (Froot, Scharfstein, and Stein, 1993), and information asymmetry (DeMarzo and Duffie, 1991, 1995). See also Breeden and Viswanathan (1996) and Stulz (1996).

description

Hmmm

Transcript of Don't Know

Page 1: Don't Know

ARTICLE IN PRESS

0304-405X/$ - s

doi:10.1016/j.jfi

$This paper

for several usef

Sreedhar Bhara

Jerry Haas, Pan

Uday Rajan, W

Darden, Emory

Fellowship Co

particularly gra�Tel.: +1 73

E-mail addr1Other motiv

underinvestmen

Breeden and V

Journal of Financial Economics 87 (2008) 706–739

www.elsevier.com/locate/jfec

Financial distress and corporate risk management:Theory and evidence$

Amiyatosh Purnanandam�

Ross School of Business, University of Michigan, Ann Arbor, MI 48109, USA

Received 28 February 2006; received in revised form 2 April 2007; accepted 10 April 2007

Available online 14 December 2007

Abstract

This paper extends the current theoretical models of corporate risk-management in the presence of financial distress

costs and tests the model’s predictions using a comprehensive data set. I show that the shareholders optimally engage in ex-

post (i.e., after the debt issuance) risk-management activities even without a pre-commitment to do so. The model predicts

a positive (negative) relation between leverage and hedging for moderately (highly) leveraged firms. Consistent with the

theory, empirically I find a non-monotonic relation between leverage and hedging. Further, the effect of leverage on

hedging is higher for firms in highly concentrated industries.

r 2008 Elsevier B.V. All rights reserved.

JEL classification: G30; G32

Keywords: Hedging; Risk-shifting; Asset substitution; Derivatives

1. Introduction

This paper develops and tests a theory of corporate risk management in the presence of financial distresscosts. The existing literature shows that hedging can lead to firm value maximization by limiting deadweightlosses of bankruptcy (see Smith and Stulz, 1985).1 These models justify only ex-ante risk-management

ee front matter r 2008 Elsevier B.V. All rights reserved.

neco.2007.04.003

is based on a chapter of my Ph.D. dissertation at Cornell University. I would like to especially thank an anonymous referee

ul suggestions during the reviewing process. I am grateful to George Allayannis, Warren Bailey, Sugato Bhattacharya,

th, Sudheer Chava, Thomas Chemmanur, Wayne Ferson, Ken French, John Graham, Robert Goldstein, Yaniv Grinstein,

kaj Jain, Kose John, Haitao Li, Roni Michaely, M.P.Narayanan, Maureen O’Hara, Paolo Pasquariello, Mitch Petersen,

illiam Schwert (the editor), David Weinbaum, Rohan Williamson, and seminar participants at Boston College, Cornell,

, London Business School, University of Michigan, Notre Dame, University of Rochester, The Lehman Brothers Finance

mpetition 2003, and the Western Finance Association’s 2005 meetings for valuable comments and suggestions. I am

teful to Bob Jarrow and Bhaskaran Swaminathan for their advice. All remaining errors are mine.

4 764 6886; fax: +1 734 936 8715.

ess: [email protected]

ations for corporate hedging include convexity of taxes, managerial risk-aversion (Stulz, 1984; Smith and Stulz, 1985)

t costs (Froot, Scharfstein, and Stein, 1993), and information asymmetry (DeMarzo and Duffie, 1991, 1995). See also

iswanathan (1996) and Stulz (1996).

Page 2: Don't Know

ARTICLE IN PRESSA. Purnanandam / Journal of Financial Economics 87 (2008) 706–739 707

behavior on the part of the firm; ex-post, shareholders of a levered firm may not find it optimal to engage inhedging activities due to their risk-shifting incentives (Jensen and Meckling, 1976).2 I extend the currentliterature by explaining the ex-post risk-management motivation of the firm.3 I provide a simple model thatgenerates new cross-sectional predictions by relating firm characteristics such as leverage, financial distresscosts, and project maturity to risk-management incentives. I test the key predictions of the model with hedgingdata of COMPUSTAT-CRSP firms meeting some reasonable sample selection criteria for fiscal years1996–1997. The empirical study presents the first large-sample evidence on the determinants of the extent offirms’ hedging activities and provides new findings.

The key assumption underlying my theory is the distinction between financial distress and insolvency.I assume that apart from the solvent and the insolvent states, a firm faces an intermediate state called financial

distress. Financial Distress is defined as a low cash-flow state in which the firm incurs losses without beinginsolvent. The notion that financial distress is a different state from insolvency has some precedence in theliterature. Titman (1984) uses a similar assumption to study the effect of capital structure on a firm’sliquidation decisions.

There are three important sources of financial distress costs. First, a financially distressed firm may losecustomers, valuable suppliers, and key employees.4 Opler and Titman (1994) provide empirical evidence thatfinancially distressed firms lose significant market share to their healthy counterparts in industry downturns.Using data from the supermarket industry, (Chevalier 1995a, b) finds evidence that debt weakens thecompetitive position of a firm. Second, a financially distressed firm is more likely to violate its debt covenants5

or miss coupon/principal payments without being insolvent.6 These violations impose deadweight losses in theform of financial penalties, accelerated debt repayment, operational inflexibility, and managerial time andresources spent on negotiations with the lenders.7 Finally, a financially distressed firm may have to forgopositive NPV projects due to costly external financing, as in Froot, Scharfstein, and Stein (1993). In this paperI focus on the first of these costs, i.e., the product market-related costs of financial distress.

I develop a dynamic model of a firm that issues equity capital and zero-coupon bonds to invest in a riskyasset. The firm makes an initial investment with the consent of its bondholders. At a later date, shareholderscan modify the firm’s investment risk by replacing the existing asset with a new one. The firm’s asset valueevolves according to a stochastic process. The firm is in financial distress if the asset value falls below somelower threshold during its life. In this state, the firm loses market share to its competitors and therefore isunable to realize its full upside potential, even when the industry condition improves at a later date. Insolvencyoccurs on the maturity date if terminal firm value is below the face value of debt, in which case debtholdersgain control of the firm. Shareholders’ final payoffs depend on the terminal asset value as well as on the pathtaken by the firm’s asset over its life.8

2Throughout the paper, I use the terms ex ante and ex post with respect to the time of borrowing.3Other papers analyzing shareholders’ ex-post risk-management decisions include Leland (1998) and Morellec and Smith (2003). Leland

(1998) provides a justification for the firm’s ex-post hedging behavior in the presence of tax-benefits of debt. In Morellec and Smith (2003),

the manager-shareholder conflict reduces shareholders’ ex-post asset-substitution incentives. My model, in contrast, is based on the cost of

financial distress and provides new empirical predictions.4For example, in the mid-1990s Apple Computers had financial difficulties leading to speculation about its long-term survival (see

Business Week, January 29 and February 5, 1996). Software developers were reluctant to develop new application software for Mac-users,

which led in part to a decline of 27% in the unit sales of Mac computers from 1996 to 1997 (see Apple’s 1998 10-K filings with the SEC).

Similarly, when Chrysler faced financial difficulties in the early 1980s, Lee Iacocca (former CEO of the company) observed that ‘‘its share

of new car sales dropped nearly two percentage points because potential buyers feared the company would go bankrupt’’ (quoted from

Titman, 1984).5Lenders often impose debt covenants such as maintenance of minimum networth or maximum debt-to-equity ratio by the borrowing

firms. See Smith and Warner (1979), Kalay (1982), and Dichev and Skinner (2001).6Moody’s Investor Service Report (1998) shows that during 1982–1997 about 50% of the long-term publicly traded bond defaults

(including missed or delayed payment of coupon and principal) didn’t result in bankruptcy filings.7For example, when Delta airlines violated a debt-to-equity ratio covenant in 2002, it was required by its lenders to maintain a minimum

of $1 billion in cash and cash equivalents at the end of every month from October 2002 until June 2003. See Delta’s 2002 10-K filings with

the SEC.8This approach is similar (but not the same) to valuation of equity as a path-dependent (down-and-out call) option. The equity value in

my model differs from the corresponding barrier option by the amount of losses incurred in financial distress. Brockman and Turtle (2003)

provide some empirical evidence in support of equity’s valuation as a path-dependent option.

Page 3: Don't Know

ARTICLE IN PRESSA. Purnanandam / Journal of Financial Economics 87 (2008) 706–739708

The optimal level of ex-post investment risk, from the shareholders’ perspective, is determined by the trade-off between the costs of financial distress and value associated with the limited liability of the firm’s equity.9

Unlike in the risk-shifting models such as Jensen and Meckling (1976), equity value is not always an increasingfunction of firm risk in my model. While a high risk project increases the value of equity’s limited liability, italso imposes a cost on shareholders by increasing the expected cost of financial distress. Due to these losses,the shareholders find it optimal to implement a risk-management strategy ex-post even in the absence of anexplicit pre-commitment to do so.

The optimal investment risk in my model depends on firm leverage, the financial distress boundary, the timehorizon of the project, and the costs of financial distress. As in the extant models (Smith and Stulz, 1985),I show that a firm with high leverage has a higher incentive to engage in hedging activities. However, the risk-management incentives disappear for firms with extremely high leverage. The incentive to hedge arises fromthe product market-related financial distress costs and these costs are more likely to be present when a firm isvulnerable to losing market share to its competitors. Empirical studies by Opler and Titman (1994) andChevalier (1995a, b) show that debt weakens the competitive position of a firm in its industry. Further, theadverse consequences of leverage are more pronounced in concentrated industries. Motivated by these studiesmy model argues that industry concentration provides a good proxy for financial distress costs. Highlyleveraged firms in concentrated industries are more likely to experience a deterioration in their competitiveposition in the event of financial distress i.e., are expected to incur higher financial distress costs. Thus, themodel predicts a stronger hedging incentive for highly levered firms in concentrated industries.

The model shows that hedging incentives increase with project maturity because the likelihood ofexperiencing financial distress as well as the expected loss of default increases with the life of the asset. Risk-management motivation in my model arises from costs incurred by the firm in states in which the firm hits thefinancial distress barrier but remains solvent on the maturity date. If there are no financial distress costs, risk-management incentives disappear. On the other hand, if these costs are very high, the distinction betweenfinancial distress and insolvency diminishes along with any ex-post risk-management motivations.Intermediate levels of losses create risk-management incentives within the firm. Therefore, my model predictsa U-shaped relation between financial distress costs and hedging.

The predictions of my model have important implications for the empirical research. To test the existingtheories, empirical studies regress some measure of financial distress (typically leverage) on firms’ risk-management activities. If firms with extreme distress are less likely to hedge, these models may be misspecified.The bias can be particularly severe in small-sample studies. It is not surprising that existing empirical studiesfind mixed evidence in support of the distress cost-based theories of hedging.10

I contribute to the empirical risk-management literature by analyzing foreign currency and commodity risk-management activities of a comprehensive sample of nonfinancial firms. Since data on firms’ hedging activities(by means of derivatives) are not readily available, empirical studies in this area are based on small samples orinvestigate only the yes–no decision to hedge.11 This has created two major challenges. First, our currentunderstanding is mostly based on analyses that treat firms with different hedging intensities as similar, whichlimits our ability to investigate firms’ hedging motivations. Second, we have been able to gain only limitedinsight into the effect of industry-specific factors on hedging decisions.

I test the predictions of my model with data on the extent of hedging of more than 2,000 firms for the fiscalyear 1996–1997. Due to the large sample size drawn from different industries, I provide new empirical evidencerelating industry structure to hedging decisions. Consistent with the theory, I find strong evidence that firmswith higher leverage hedge more, although the hedging incentives disappear for firms with very high leverage.Also in line with my theory, I find that financially distressed firms in highly concentrated industries hedge

9In the context of swap markets, Mozumdar (2001) demonstrates the trade-off between risk-shifting and hedging incentives in the

presence of information asymmetry about the firm type. His model relates hedging incentives to firm type.10For example, while Haushalter (2000) and Graham and Rogers (2002) find a positive relation between the two variables, Nance, Smith

and Smithson (1993), Mian (1996), and Tufano (1996) fail to find such evidence.11For example, Geczy, Minton, and Schrand (1997) use 372 firms with 154 hedgers; Graham and Rogers (2002) use about 400 firms with

158 hedgers. Studies by Mian (1996) and Bartram, Brown, and Fehle (2003) use large samples to investigate the yes–no decision of

hedging. Tufano (1996) and Haushalter (2000) provide detailed evidence from gold and oil & gas industries, respectively. Brown (2001)

provides evidence from a detailed case study. Purnanandam (2007) investigates the risk-management decisions of commercial banks.

Page 4: Don't Know

ARTICLE IN PRESSA. Purnanandam / Journal of Financial Economics 87 (2008) 706–739 709

more. My empirical results are robust to alternative proxies of financial distress (such as leverage, industry-adjusted leverage and Altman Z-score), alternative ways of measuring the hedging activities (yes–no decisionto hedge and total notional amount of hedging) and various controls for nonderivative-based hedgingstrategies. Further for a subsample of 200 manufacturing firms, I obtain data on the firms’ hedging activitiesfor fiscal years 1997–1998 and 1998–1999 and show that the basic results remain similar for a regression modelinvolving changes in hedging activities. While firms with a moderate increase in leverage increase their hedgingactivities, firms with an extreme increase in leverage decrease their hedging positions. As long as firms do notfrequently change their operational hedging strategies (such as opening plants in foreign countries to hedgetheir foreign currency risk), the analysis based on change regressions provides a robust control fornonderivative-based hedging strategies of the firm. The change regressions also allow me to partiallydisentangle the effects of ex-ante and ex-post hedging incentives.

The rest of the paper is organized as follows. In Section 2, I provide the model description. Section 3analyzes the optimal risk-management policy of the firm. The empirical tests are provided in Section 4, andSection 5 concludes the paper. Without any loss in continuity, readers mostly interested in the empirical partof the paper can skip to Section 3.1, which provides a self-contained summary of the key features of thetheoretical model.

2. Model

I consider a stylized model of a continuous trading economy with time horizon ½t0;T �. There are threeimportant dates in the model discussed below. Though a discrete time model can also be used to capture thekey feature of my model, the continuous time version allows for an easier analytical solution at the expense ofadditional mathematical overhead. In addition, the continuous time model provides additional predictionrelating the time to maturity of the firm’s project to its hedging incentives.

At t ¼ t0, the firm makes its capital structure decision and invests in risky asset Ai (i stands for the initialinvestment), which I refer to as an ‘‘EBIT-generating machine’’ (Goldstein, Ju and Leland, 2001). Thesedecisions may or may not be made with the consent of the firm’s debtholders. The risky asset ðAiÞ is acquiredat the market-determined price and financed through a mix of zero-coupon debt and equity capital. Let L bethe face value of the zero-coupon debt, payable at time T , and Et be the time t-value of the firm’s equity. Thereis a tax benefit of debt, which provides the incentive to issue debt in my model. For simplicity the tax benefit isassumed to be a fraction t of the face value of debt L. Optimal capital structure is determined by a trade-offbetween the tax benefit of debt and bankruptcy costs. For simplicity, I do not endogenize the capital structuredecisions. However, the key predictions of the model remain similar for a more general model (unreported)that solves for capital structure decisions as well. The cash generated by the machine and its asset value Ai

t aredriven by a Brownian motion with the usual properties.

At some later time t ¼ t1 ðt1 2 ðt0;TÞÞ, the shareholders (or managers acting on their behalf) make a risk-management decision. At this time, which can be an instant or days or months after the capital structuredecisions, they have an opportunity to change the asset’s risk without the bondholder’s approval. To capturethe risk-shifting incentives, I assume that the bondholders are unable to recontract with the shareholders att ¼ t1. Further, I assume that the two parties cannot contract on the risk-management choice at time t0through the use of bond covenants. This latter assumption is what gives rise to the risk-shifting incentive in mymodel. This assumption is in the spirit of a large literature on incomplete contracting in economics and finance(see for example, Bolton and Dewatripont, 2005). The premise here is that it is too costly to specify every stateof the world and write down debt covenants that will limit shareholders behavior with respect to firm risk ineach of those states. Even if such covenants could be written to tie down the manager’s risk-managementbehavior, it would be too costly to implement them especially in very high leverage states when shareholdershave a large incentive to default on covenants.12

This assumption is in the spirit of Jensen and Meckling’s argument that ‘‘To completely protect thebondholders from the incentive effects, these provisions would have to be incredibly detailed and cover most

12As long as there are nontrivial costs in writing, monitoring and enforcing these contracts, some residual risk-management decisions are

always optimally left with the shareholders/managers, which is sufficient to generate the main results of my model.

Page 5: Don't Know

ARTICLE IN PRESSA. Purnanandam / Journal of Financial Economics 87 (2008) 706–739710

operating aspects of the enterprise including limitations on the riskiness of the projects undertaken. The costsinvolved in writing such provisions, the costs of enforcing them and the reduced profitability of the firm(induced because the covenants occasionally limit management’s ability to take optimal actions on certainissues) would likely be nontrivial. In fact, since management is a continuous decision making process it will bealmost impossible to completely specify such conditions without having the bondholders actually perform themanagement function.’’

After the risk-management decisions have been made, the firm acquires a new EBIT-generating machine.This EBIT-generating machine generates cashflows dt forever that evolves according to a geometric Brownianmotion. The value of this EBIT-generating machine, i.e., the value of a similar unlevered firm, is denoted byAt.

13 One can think of dt as the state vector representing the state of the firm’s industry. I assume that thechange in the investment risk of the asset (from Ai to AÞ has no cashflow impact on the firm at t ¼ t1. Thisprovides an initial boundary condition in the model, namely At1

¼ Ait1. Further, for analytical simplicity I

assume that the total payout (to debtholders and shareholders) by the firm is zero during ½t0;TÞ, with the finalpayoffs realized at t ¼ T . The shareholders receive the terminal equity value of the firm.14 The bondholdersreceive the face value of debt (L) if the firm remains solvent on the maturity date t ¼ T15; otherwise theyreceive the residual value of the firm. The model can be represented by the following timeline:

13The value of the levered fi

financial distress and bankru

generating machine) by At.14For analytical simplicity

assumption should not be con

the firm to some other invest15Other maturity structure

coupon debts.16I refer to K as the distres

financial distress is equivale

distressed.

rm of my model differs from

ptcy. Throughout this paper I

I assume that the model’s term

fused with the assumption tha

ors at the fair market value o

s are possible. To illustrate th

s barrier in the rest of this pa

nt to assuming that when in

m

m m t ¼ t0 t ¼ t1 t ¼ T

Capital structure

Risk-management Payoffs Initial investment decisions

This modeling framework allows me to address the issue of ex-ante vs. ex-post risk-management behavior ofthe firm in the presence of the shareholders’ risk-shifting incentives. I now discuss the main assumption of thepaper, namely, the distinction between financial distress and insolvency.

2.1. Financial distress and insolvency

If during (t0;TÞ the firm’s asset value At falls below a boundary KðLÞ;16 the firm is in the state of financialdistress. Insolvency, on the other hand, occurs on the terminal date T if the terminal firm value ðV T ) is lessthan the debt obligations. Therefore, in the state of financial distress, control of the firm does not shift to thebondholders immediately, but the firm does incur costs that increase with leverage. Opler and Titman (1994)show that financially distressed (highly leveraged) firms lose significant market share to their healthycompetitors during industry downturns. The drop in sales faced by Apple Computers and Chrysler duringperiods of financial difficulty provide anecdotal evidence in support of such deadweight losses. In a sample of31 high-leveraged transactions (HLTs), Andrade and Kaplan (1998) isolate the effect of economic distressfrom financial distress and estimate the cost of financial distress as 10–20% of firm value. Asquith, Gertnerand Scharfstein (1994) show that on average financially distressed firms sell 12% of their assets as part of theirrestructuring plans.

Chevalier (1995a, b) uses detailed information from the local supermarket industry to provide evidence insupport of predatory behavior in this market. She shows that following supermarket leveraged buyouts

At by the amount of the tax benefit of debt as well as the costs associated with

denote the value of the levered firm by V t and the value of its assets (EBIT-

inal date corresponds to the maturity date of the firm’s debt, at t ¼ T . This

t the firm’s life is finite. It simply states that at time T initial shareholders sell

f the firm as an ongoing concern.

e main results of the paper in its simplest form, I prefer to work with zero

per. K is assumed to be an increasing function of leverage. This definition of

dustry conditions deteriorate, firms with high leverage become financially

Page 6: Don't Know

ARTICLE IN PRESS

Time (t)

Ass

et V

alue

At1

f-1(L)

K

t1 Tτ

Financially

Distressed

Insolvent

Healthy

L

Fig. 1. This figure plots three paths for the evolution of the firm’s asset value. I assume zero tax shield of debt for this presentation. These

paths correspond to three states of the firm in my model. In the top-most path, the asset value never hits the financial distress barrier (K).

This corresponds to the ‘Healthy’ state. The middle path represents the state in which the distress barrier is hit (at time tÞ, but the firm

remains solvent at time T . This is the state of ‘Financial Distress.’ In this state the terminal firm value, net of deadweight losses (i.e.,

f ðAT Þ), remains above the face value of debt (i.e., L). Thus, this is the state where f ðAT Þ4L or alternatively AT4f �1ðLÞ, as depicted in the

figure. Finally, the bottom-most path corresponds to the state of ‘Insolvency.’

A. Purnanandam / Journal of Financial Economics 87 (2008) 706–739 711

(LBOs), prices fall in local markets in which rival firms have low leverage and are concentrated. Further, theseprice drops are associated with LBO firms exiting the local market. These findings suggest that rivals attemptto prey on LBO chains. Phillips (1995) studies the interactions between product market and financial structurefor four industries and finds evidence consistent with debt weakening the competitive positions of firms (seealso Kovenock and Phillips, 1997; Arping, 2000). Using deregulation of the trucking industry as an exogenousshock, Zingales (1998) studies the interplay between financial structure and product market competition andprovides evidence that leverage reduces the probability of a firm’s survival after an increase in competition.The overall message from these papers is that financial distress may impose a real cost on firms by weakeningtheir competitive position in the product market.

Motivated by the empirical findings of above papers and anecdotal evidence, I assume that a firm infinancial distress loses a fraction of its market share to its healthy competitors.17 In my model, this is achievedby assuming that the financially distressed firm’s EBIT-generating machine produces less cashflow resulting ina lower value for the distressed firm. If the firm does not experience financial distress during t 2 ½t1;T �, theterminal firm value is VT . However, if the distress boundary is hit, the terminal value falls to f ðVT Þ, wheref ðV T ÞoV T (see Fig. 1). The function f represents the losses caused by financial distress.

2.2. Valuation of equity

The shareholders receive liquidating dividends at T. Due to equity’s limited liability, the final payoff to theshareholders ðxT Þ is zero if the terminal firm value is below L. Let us define inf t1ptpT At � mT forthe minimum value of the asset during ½t1;T �. In the event of no distress (i.e., mT4K) and solvency on theterminal date (i.e., V T4L), the shareholders get a liquidating dividend of ðV T � LÞ. If financial distress isexperienced (i.e., mTpK), but on the terminal date the firm remains solvent (i.e., f ðVT Þ4L), the shareholders

17In a more general industry equilibrium setting, firms can make strategic decisions about their leverage, investment risk, and hedging

(see e.g., Adam, Dasgupta, and Titman, 2004; Nain, 2006). My model abstracts from such considerations and focuses on the firm’s

decision, taking industry structure as given.

Page 7: Don't Know

ARTICLE IN PRESSA. Purnanandam / Journal of Financial Economics 87 (2008) 706–739712

receive liquidating dividends of f ðVT Þ � L. In the event of insolvency, shareholders receive nothing and firmvalue drops by the fraction g 2 ½0; 1�. The shareholders’ payoff under different states is summarized as

18In unreported analysis, I solve the model wi

of this paper on risk-management decisions, I do

without qualitatively changing the results of the19If lo1, then financial distress costs are eve

simplicity.

th tax benefits and obtain the firm’s optimal capital structure

not present these results in the paper. With tax benefits, the

analysis.

n higher and the results become stronger. This assumption

State at t ¼ T

Corresponding firm values Payoff to shareholders

Healthy

VT4L;mT4K V T � L

Financial distress

f ðVT Þ4L;mTpK f ðVT Þ � L

Insolvency

VTpL;mT4K 0 Insolvency f ðV T ÞpL;mTpK 0

Proposition 1. Under mild technical conditions, the equity valuation at t ¼ t1 is given by:

xt1¼ e�rT EQ½ðVT � LÞ � ðVT � f ðVT ÞÞ1ff ðVT Þ4L;mTpKg þ ðL� V T Þf1fVTpLg þ 1

ff�1ðLÞ4VT4L;mTpKgg�.

(1)

Proof. See Appendix A.1. &

The equity value, as shown in Proposition 1, has three components. The first term ðEQ½VT � L�Þ representsthe equity value without the distress costs and the limited liability feature. The second term ðEQ½ðVT �

f ðV T ÞÞ1ff ðVT Þ4L;mTpKg�Þ represents the cost of financial distress. Because the shareholders of a financially

distressed but solvent firm bear this cost, the equity value decreases by this amount. The risk avoidanceincentive results from this cost. The third term ðEQ½ðL� VT Þf1fVTpLg þ 1

ff�1ðLÞ4VT4L;mTpKgg�Þ represents the

savings enjoyed by the shareholders of a levered firm due to the limited liability feature of equity. This termcaptures shareholders’ risk-shifting incentives. By increasing the asset’s risk, the shareholders can makethemselves better off by increasing the call option value (the third term). At the same time, however, theexpected loss in the event of financial distress also increases with an increase in asset risk. The optimal level ofinvestment risk is determined by the trade-off between the two.

2.2.1. Financial distress costs

Proposition 1 provides a general valuation formula in my model. To proceed further I need to beexplicit about the form of financial distress cost that is borne by the shareholders of a financiallydistressed firm. In addition, I make some simplifying assumptions for analytical tractability. I assume that inthe event of distress (i.e., mTpK), the firm’s cashflows drop to ldt; l 2 ð0; 1� and never reach beyond somearbitrary upper bound Uo1 at time T, i.e., dTpU . Therefore, the losses take the form of lost upsidepotential. This representation of financial distress cost is motivated by existing empirical findings andanecdotal evidence, and captures the intuition that distressed firms lose cashflows due to lost sales tocompetitors. If industry conditions improve in the future, the distressed firms continue to feel the negativeeffect of distress due to lost customers. This representation of distress is also consistent with the view thatwhen financially distressed firms restructure themselves by selling assets (Asquith, Gertner and Scharfstein,1994), their EBIT-generating machine produces lower contemporaneous cashflows and in addition it limitstheir ability to capitalize on very good industry conditions in the future. To concentrate on the effect offinancial distress costs (as opposed to tax-motivated incentives of hedging as in Leland, 1998), in the rest of thepaper I set t ¼ 0.18 Under this assumption and the assumption l ¼ 1, the distressed firm’s asset value can berepresented as19:

f ðAT Þ ¼ AT if fdTpUg; and M0 if fdT4Ug for some constant M0. (2)

. However, to keep the focus

firm’s payoffs increase by tL

is made only for analytical

Page 8: Don't Know

ARTICLE IN PRESS

Equ

ity V

alue

L

L

L+M0

Equity Value InHealthy State

Equity Value inFinancial Distress

Equity Valuein my model

Asset Value at T

Fig. 2. This figure plots the equity value as a function of the terminal asset value of the firm. For illustrative purposes I set the tax rate to

zero and g ¼ 1 for this diagram. The equity value in my model is depicted by the solid line. The upper dotted line represents the equity

value for the Healthy state. The lower dotted line depicts the equity value in the state of Financial Distress. The equity value in my model is

a weighted average (weight is decided by the relative likelihood of the two states) of the equity value in these two states.

A. Purnanandam / Journal of Financial Economics 87 (2008) 706–739 713

Let us denote the asset value ðAT Þ corresponding to dT ¼ U by LþM. The shareholders’ liquidatingdividends are given as

States

Payoff to shareholders Firm value

AT4L;mT4K

AT � L AT

AT4L;ATpLþM ;mTpK

AT � L AT

AT4LþM ;mTpK

M LþM

ATpL

0 gAT

The financial distress costs can be expressed as ðAT �MÞ:1fAT4LþM ;mTpKg. A higher value of M

corresponds to lower deadweight losses in the model. In line with Proposition 1, the equity value can beexpressed as follows:

xt1¼ e�rT EQ½ðAT � LÞ1fAT4L;mT4Kg þ ðAT � LÞ1fAT4L;ATpLþM ;mTpKg

þM1fAT4LþM ;mTpKg�. ð3Þ

Fig. 2 plots the equity value as a function of the terminal asset value of the firm. As the diagram shows, theequity value is not a strictly convex function of the underlying firm value as in the classical approach whereequity is valued as a call option on firm value. The deadweight loss of distress introduces a concavity in theequity value, which results in risk-management incentives for the firm.

3. Optimal choice of investment risk

Without loss of generality, I set the risk-free interest rate to zero in the rest of the analysis. At t ¼ t1, theshareholders make a decision about the optimal investment risk of the firm. There are two possibilities forchanging the investment risk: (a) the firm can directly choose an optimal level of s at t ¼ t1, or (b) the asset’srisk, s, may be fixed and the firm can alter its risk profile by buying derivative contracts such as futures andoptions. I analyze the problem of finding optimal s assuming that investment risks can be costlessly modified.

Proposition 2. The shareholders have a well-founded incentive to engage in risk-management activities ex-post.

At t ¼ t1, the shareholders optimally choose a level of risk s� in the interior of all possible risks.

Page 9: Don't Know

ARTICLE IN PRESSA. Purnanandam / Journal of Financial Economics 87 (2008) 706–739714

Proof. As shown in Appendix A.2 and A.3, the optimum level of investment risk is obtained by the followingfirst-order condition:

At1fðh1Þ ¼ Kfðc1Þ (4)

where h1 ¼ ðlnðAt1=LÞ þ ðs2=2ÞT 0Þ=s

ffiffiffiffiffiT 0p

, h2 ¼ h1 � sffiffiffiffiffiT 0p

, T 0 ¼ T � t1, c1 ¼ lnðK2=At1ðLþMÞÞ þ ðs2=2ÞT 0=

sffiffiffiffiffiT 0p

, c2 ¼ c1 � sffiffiffiffiffiT 0p

; and f stands for the probability density function of the standard normal distribution.Further simplification leads to the following closed-form solution:

ðs2Þ� ¼1

T 0

lnK2

LðLþMÞ

� �ln

K2L

A2t1ðLþMÞ

!

lnLþM

L

� � : & (5)

As a result of the trade-off between the risk-shifting and risk-avoidance incentives, an interior solution forthe optimal risk is obtained in the model. This result differs from that of the earlier models. In risk-shiftingmodels such as Jensen and Meckling (1976), the shareholders take as much risk as possible, whereas in risk-management models such as Smith and Stulz (1985), the optimal level of risk is obtained at s ¼ 0. Byobtaining an interior solution for the optimal investment risk of the firm, my model provides insights into therisk-management policies of the firm, as discussed below.20

Proposition 3. The firm chooses a lower level of investment risk if (a) it faces a higher distress barrier (K), and (b) it

has a longer project maturity ðT 0 ¼ T � t1Þ. The relation between the deadweight losses and the optimal investment

risk is U-shaped. Let Mc ¼ L exp2ð

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffilnðAt1

=KÞ lnðL=KÞp

Þ� L. When M4Mc, the optimal investment risk decreases

with an increase in the deadweight losses, otherwise it increases with an increase in the deadweight losses.

Proof. The proof follows from direct differentiation of the optimal solution for s given in expression 5(see Appendix A.5). &

The investment risk decreases (i.e., the risk-management incentive increases) with the distress boundary (K). Asexpected, a higher boundary increases the likelihood of financial distress. Therefore, the shareholders optimallychoose a lower investment risk to avoid the financial distress costs. The results show that the firm with a longeroperational horizon ðT 0 ¼ T � t1Þ finds it optimal to engage in increased risk-management activities. With longertime-horizon, the probability of hitting the lower barrier increases. Further, consequent to entering the state ofdistress expected losses increase with time to maturity because there is a higher probability of improvements inindustry conditions and the distressed firm will not be able to capitalize on these opportunities. There isconsiderable empirical evidence that large firms hedge more than small firms. The pursuit of economies of scale hasbeen suggested as one possible explanation for this empirical regularity. My model suggests another explanation:the time horizon of operations. If firms with longer time horizons grow larger over time, the researcher would finda positive association between risk-management activities and firm size at any given point in time.

Finally, I find a U-shaped relation between the risk management incentives and the cost of financial distress.Recall that the deadweight losses in my model are parameterized by M (losses are given byðAT �MÞ:1fAT4LþM ;mTpKg). In the event of financial distress, the firm loses its upside potential beyondLþM. Thus, the higher the M, the lower the lost upside potential and therefore the lower the deadweight losses.If the deadweight losses are absent (i.e., M ¼1), the shareholders lose nothing in the state of financial distressand hence there is no risk-management incentive. On the other hand, if deadweight losses are very high (i.e.,M ¼ 0) the distinction between default and insolvency disappears along with the risk-management incentives.21

It’s the intermediate cases that generate risk-management incentives in the model. Fig. 3 illustrates this relation.

20With nonzero tax rates (in unreported analysis), the optimal s is even lower. The additional incentives for risk reduction, in the

presence of the tax-benefit of debt, comes from the potential loss in the tax shield of debt for a bankrupt firm. This additional effect

generates ex-post hedging as in Leland (1998). See also Fehle and Tsyplakov (2005).21In this case, equity value becomes similar to a down-and-out barrier option. Since the value of this option is increasing in the volatility

of the underlying assets, the shareholders do not have any risk-management incentives at t1.

Page 10: Don't Know

ARTICLE IN PRESS

2.78

2.8

2.82

2.84

2.86

2.88

2.9

2.92

2.94

15.00

14.25

13.50

12.75

12.00

11.25

10.50 9.7

59.0

08.2

57.5

06.7

56.0

05.2

54.5

03.7

53.0

02.2

5

Deadweight Loss Parameter (M)

Inve

stm

ent R

isk

Fig. 3. This figure plots the optimal investment risk as a function of deadweight losses. The model has been calibrated with the following

parameter values: At1¼ 2;L ¼ 1;T 0 ¼ 1 and K ¼ 0:5: On the x-axis, I plot the value of M. M measures the upside potential lost by

the firm in the event of financial distress. I plot M from higher-to-lower value so that the deadweight losses increase as one moves along the

x-axis.

Investment Risk vs. Leverage

0

2

4

6

8

10

12

14

0.11

0.15

0.19

0.23

0.27

0.31

0.35

0.39

0.43

0.47

0.51

0.55

0.59

0.63

0.67

0.71

0.75

0.79

0.83

0.87

0.91

0.95

Leverage

Inve

stm

ent R

isk

Fig. 4. This figure plots the optimal investment risk of the firm against the debt-asset ratio. For this graph I assume the following structure

on the distress boundary and deadweight losses: K ¼ 1� exp�0:1�lev and M ¼ 7� exp2�lev. Amount of debt raised at time zero (L) if fixed

at 1. lev equals L scaled by At1. T is set to one.

A. Purnanandam / Journal of Financial Economics 87 (2008) 706–739 715

Leverage and risk management: To study the relation between leverage and risk management, I differentiatethe optimal s with respect to firm leverage at time 1 ðlev ¼ L=AÞ. The details are provided in Appendix A.5.After some simplification it can be shown that the optimal sigma decreases (i.e., risk-management incentivesincrease) with an increase in leverage for a wide range of specifications of the distress boundary and

Page 11: Don't Know

ARTICLE IN PRESSA. Purnanandam / Journal of Financial Economics 87 (2008) 706–739716

deadweight loss parameter. This relation reverses when leverage is very high due to the risk-shifting incentives.At very high leverage, the value associated with the call option of equity dominates the cost borne byshareholders and thus they lose risk-management incentive. Using a parametric specification of K and M,I solve for optimal risk as a function of leverage and report the results in Fig. 4. The relation is summarizedbelow:

Proposition 4. Risk-management incentives increase with leverage; this relation reverses for extremely high levels

of leverage.

Combining this result with the results between deadweight losses and risk-management, the effect ofleverage on hedging intensity is predicted to be higher for firms operating in industries with a higher incidenceof predatory behavior. The key friction underlying my model consists of the costs incurred by a firm after itenters the state of financial distress. These costs come in the form of lost customers and deterioration in thecompetitive position of the firm within its industry. Based on the empirical studies of Opler and Titman (1994)and Chevalier (1995a, b) such costs are more likely to be incurred by a firm in concentrated industries. Thus, inthe context of my model industry concentration provides a good proxy for the financial distress costs.Accordingly, high leverage firms in concentrated industries are predicted to have greater hedging incentives.

3.1. Summary of theoretical model

In this section I present a self-contained summary of the theoretical part of the paper that serves as the basisfor the empirical tests to follow. In my stylized model, a firm starts with some mix of debt and equity at timezero and buys a productive asset. At this time the capital structure of the firm is determined by trading off thetax benefit of debt against the expected financial distress and bankruptcy costs. I do not solve for the optimalleverage policy in my theoretical model to keep the focus of my analysis on risk-management decisions.However, making capital structure decisions endogenous does not change the key results of the paper. Inunreported analyses, I solve for optimal leverage and as expected show that the debt ratio increases with thetax benefits and decreases with bankruptcy and financial distress costs.22

Given a level of debt determined at time t0, the firm experiences some random shocks to its value till t1,which perturbs its leverage ratio. At this point the shareholders make the key decision in the model, i.e., a risk-management decision so as to maximize equity value. This modeling structure allows me to focus on the ex-post hedging incentives. Subsequent to the risk-management decision at t1, the asset value evolves according toa stochastic process from time t1 to T in the model. If the firm’s asset value breaches a lower threshold beforethe terminal date T, then the firm enters financial distress. Financial distress imposes costs on the firm such aslost customers to the competitors, which in turn prohibits it from capitalizing on its full upside potential.Motivated by the earlier empirical finding, I assume that highly levered firms lose more when they enter thestate of financial distress.

After the distress boundary is hit, the firm can either stay solvent on the terminal date or go bankrupt,depending on whether its value, net of distress costs, is above or below the debt value. The state in which thefirm enters financial distress but remains solvent at time T imposes a real cost on shareholders. In this statethey incur the financial distress costs without being able to use their limited liability option. An increase in firmrisk increases the probability of financial distress and the associated deadweight losses that are borne by theshareholders, not the debtholders. On the other hand, by increasing firm risk they benefit on account of theusual limited liability feature. The optimal risk-management policy trades off these two incentives. Formoderate levels of leverage, the risk-management incentive dominates. But when leverage becomes too high at

22In a rational expectation framework firm value at time t0 should be maximized keeping in mind the expected level of risk that will be

optimally undertaken by the shareholders at time t1. This expected sigma along with the tax benefit of debt and bankruptcy costs will

determine the optimal amount of debt raised by the firm at time t0. Indeed the actual leverage at time t1 will be different from the rationally

expected value of leverage, depending on the shocks experienced by the firm in the intervening period. Depending on the realizations of

these shocks in the interim period, the firm’s leverage at time t1 will be different and shareholders may deviate from the rationally

anticipated risk policy that is based on the expected level of leverage and not on realized leverage. The main result that shareholders will

have risk-management incentives as long as their leverage doesn’t go up too much follows. When the asset value realization is too low (i.e.,

leverage is too high as compared to expectations) the risk-shifting incentive follows.

Page 12: Don't Know

ARTICLE IN PRESSA. Purnanandam / Journal of Financial Economics 87 (2008) 706–739 717

time t1, the value associated with the call option feature of equity dominates the expected financial distresscosts and shareholders find it optimal not to engage in risk-management activities. Thus, the model predicts anonmonotonic relation between leverage and hedging. Further, the positive relation between leverage andhedging is expected to be stronger for firms operating in industries with a higher incidence of predatorybehavior such as concentrated industries. I test these predictions of the model in the rest of the paper.

4. Empirical evidence

There are three important challenges in empirically testing the above theory. First, data on a firm’s hedgingdecision are very limited. Second, leverage and hedging are likely to be determined jointly by firms, leading toan endogeneity problem. Theories based on ex-ante incentives suggest that firms can increase their debtcapacity by engaging in hedging activities, which in turn lead to reverse causation from hedging to leverage.Third, to capture both the ex-ante and ex-post incentives in a clean empirical setting, I need data on the timingof debt issues and hedging decisions, which unfortunately are not available. Below I begin with discussing thesample and data collection procedure followed by econometric strategies used to account for the endogeneityproblem. Empirical results follow these discussions. I then discuss the issues related to ex-ante vs. ex-postincentives in a later section.

4.1. Sample selection and data

I test the key predictions of my model using the foreign currency and commodity derivatives holdings of alarge cross-section of firms during the fiscal years 1996 and 1997. I start with all firms in the intersection of theCRSP and COMPUSTAT with 10-Ks available on the SEC website. I remove financials and utilities since therisk-management incentives of these firms are not necessarily comparable to other industrial firms. From thissample, I exclude firms that fall in the last quartile of the size distribution based on total sales. Earlierempirical studies and survey evidence suggest that such small firms are very unlikely to use derivative productsfor hedging purposes (Dolde, 1993), arguably due to the lack of economies of scale.

For the remaining firms, I collect data on derivative usage from the 10-K filings. In the first step I obtain allavailable 10-K filings of firms in the intersection of COMPUSTAT and CRSP from the SEC for the calendaryear 1997.23 I obtain data by searching the entire 10-K filings for the following text strings: ‘‘riskmanagement,’’ ‘‘hedg,’’ ‘‘derivative’’, and ‘‘swap.’’ If a reference is made to any of these key words, I read thesurrounding text to obtain data on foreign currency and commodity derivatives. I obtain data on the notionalamount of foreign currency derivatives used for hedging purposes across various derivative instruments suchas swaps, forwards, futures, and options.24 For commodity hedging I only obtain data on whether a firm usesderivatives for hedging or not, since the reporting requirement for commodity derivatives doesn’t allow for aneasy quantification in terms of dollar value. If there are no references to the key words, the firm is classified asa nonhedger. I require that data on net sales, leverage and market capitalization be available for a firm to beincluded in the sample.

In addition, to capture the dynamic behavior of a firm’s hedging and leverage decisions, I focus on a smallersubset of 200 manufacturing firms (one-digit SIC code 2) and collect data using the same procedure for twoadditional years, i.e., 1998 and 1999. This smaller subsample allows me to relate the changes in a firm’shedging activities to changes in financial conditions, which in turn allows me to draw sharper inferences asoutlined in the subsequent sections.

I limit my analysis to only those firms that have well-defined exposures to foreign currency and commodityrisks. I conduct my analysis for foreign currency derivatives on the subsample of firms with an exposure toforeign currency risk and similarly commodity derivatives on the subsample of firms with an exposure to

23For some firms (most of the firms with a fiscal year ending in October, November, or December) this corresponds to fiscal year 1996,

while for others this corresponds to the fiscal year 1997.24The break-up of the notional amount across various instrument types was not easy to obtain for some sample firms. For these firms,

I collect data on the aggregate notional amount of derivatives only. Since most of the analysis is conducted with the aggregate amount

of derivatives, this doesn’t create any bias in the study.

Page 13: Don't Know

ARTICLE IN PRESSA. Purnanandam / Journal of Financial Economics 87 (2008) 706–739718

commodity price risk. This sample selection criteria ensures that I can treat the lack of derivative usage as afirm’s choice variable to not hedge rather than an absence of exposure to the risk. I identify a firm’s exposureto these risks in the following manner.

4.1.1. Exposure to foreign currency risk

I closely follow Geczy, Minton, and Schrand (1997) to identify firms with pre-defined exposure to foreigncurrency risks. A firm is classified as being exposed to foreign currency risk if any of the following criteria ismet: (a) it reports foreign currency sales in the COMPUSTAT geographical segment file in the fiscal year ofderivative usage or within þ=� one year; (b) it reports foreign income taxes, deferred foreign currency taxes,or pre-tax foreign income in its annual statements; (c) it reports foreign currency adjustments in its annualreport; or (d) it discloses an exposure hedged with foreign currency derivatives in its footnotes identified byhand-collected data.

Based on these screens, I identify 1,781 firms as exposed to foreign currency risk.25 In the subsequentregression analysis I lose additional firms due to missing data on the explanatory variables used to estimate themultivariate models.

4.1.2. Exposure to commodity price risk

Compared to foreign currency exposure, identifying firms with an exposure to fluctuations in commodityprices is harder to measure. This arises because current accounting standards do not require firms todisclose much information with respect to their exposure to commodity price risk. In the absence of anybalance sheet information, I identify a firm’s exposure to commodity price risk by estimating the sensitivity ofits earnings to movements in various indices of commodity prices. As an alternative specification, one can usea simpler approach and take the set of all commodity-producing industries as the sample of firms that areexposed to commodity price risk. However, with such an approach it would be hard to detect firms that areexposed to commodity price risk on the input side (such as airline industry). Thus, for the sake ofcomprehensiveness, I adopt the more involved methodology of detecting firms with exposure to commodityprice risk.

In particular, I regress the quarterly earnings before interest and taxes obtained from COMPUSTAT’squarterly files on the quarterly changes in several commodity price indices and classify a firm as having anexposure to commodity price risk if the resulting coefficient is significant at the 10% level or better. I take datafrom the last 60 quarters (or the maximum available) to estimate this model. Most of the effect of commodityprice movements is reflected in a firm’s sales or its cost of production, such as raw material or energy costs.Therefore, I take EBIT as the relevant measure of earnings for the purpose of sensitivity analysis.26

There are two important issues with this estimation methodology. First, the use of derivatives can make afirm’s earnings less sensitive to movements in commodity prices, rendering my methodology ineffective forhedger firms. However, I already have hand-collected data on whether these firms use commodity derivativesto hedge a well-specified risk. I, therefore, add the commodity hedgers to the set of firms that I detect as havingan exposure to commodity risk based on the above methodology.

Second, firms may be exposed to various types of commodity risks, ranging from oil price shocks to metalsto farm produce. Based on the firm’s disclosure in the footnotes of their annual statements as well as thecontract volume of various futures contracts on the futures exchanges, it is clear that the main sources ofcommodity risk facing U.S. nonfinancial firms are the following: (a) crude oil and related products; (b) metalssuch as copper and iron; (c) farm products such as corn; and (d) various industrial chemicals. Noting this,I obtain data on the quarterly price changes for a basket of these commodities from the Bureau of Labor

25Geczy, Minton, and Schrand (1997) also consider firms with a high concentration of foreign importers in the industry as exposed to

foreign currency risk. I don’t consider this screening criterion since they show that very few firms are identified as having a foreign currency

exposure based solely on this criterion. In their sample of about 370 firms, only three firms are identified as having exposure based solely

on this criterion.26I also repeat my analysis with other measures such as cashflows, EBIT/TA, NI/TA, the seasonally adjusted earnings, and obtain

similar set of firms. Note that scaling EBIT by total assets doesn’t make any qualitative difference because the regression is estimated on a

firm-by-firm basis with fairly stable total asset values (as compared to EBIT). Therefore, I only present results with the EBIT-based

sensitivity analysis to conserve space.

Page 14: Don't Know

ARTICLE IN PRESSA. Purnanandam / Journal of Financial Economics 87 (2008) 706–739 719

Studies. Further, I obtain data on quarterly changes in the aggregate Producer Price Index (PPI), whichreflects price changes based on a basket of commodities including oil, farm products, industrial chemicals,metals, and other commonly used products by the industrial producers. Thus I have five price indices (crudeoil, metals, farm products, chemicals, and all commodities) and I estimate a firm’s commodity price sensitivitywith respect to each of these indices separately. Since my analysis doesn’t distinguish firms based on thespecific source of risk they face, I consider a firm as exposed to commodity price risk if I obtain a significantcoefficient in any of the five regressions. This methodology identifies 1,238 firms as having an exposure tocommodity price risk in my sample. When I merge this sample with the sample of firms with ex-ante exposureto foreign exchange movements, I find that I have a total of 2,256 firms with an exposure to at least one ofthese sources of risk.

4.1.3. Derivatives as a proxy for hedging

I use two definitions of hedging based on derivative usage. The first definition is based on the firm’s binarydecision of whether to use derivatives for hedging purposes. This specification uses both types of derivativecontracts—foreign currency and commodity. In the second specification, I use the total notional amount offoreign currency derivatives. The notional amount-based definition of hedging captures the firm’s totalownership of risk-management instruments and is thus able to distinguish between firms with differenthedging intensities.

There are two important concerns associated with the use of derivatives as a proxy for hedging activities.First, though I obtain data on derivatives classified as risk-management tools, there may still be a concernabout their intended use—are firms indeed using these instruments for hedging purposes or not? Earlierempirical studies find strong evidence in support of risk-reducing (i.e., hedging) effects of derivatives onvarious measures of a firm’s risk. Guay (1999) finds that the new users of derivatives experience a decline intheir earnings and stock price volatility after the initiation of derivatives contracts. Similarly Allayannis andOfek (2001) show that using derivatives reduces currency exposure, and Hentschel and Kothari (2001) do notfind any evidence that derivatives are used for speculative purposes. Thus, there is enough evidence in theliterature to suggest that the majority of firms use derivative instruments for hedging purposes and not forspeculative reasons.

The second concern with the use of derivatives data relates to the importance of derivatives on theoverall cashflows of the firms. Allayannis and Weston (2001) and Graham and Rogers (2002) find asignificant impact of derivative instruments on firm value and the firm’s debt capacity, respectively.These findings suggest that derivative instruments have a significant impact on firm performance andthus are good instruments for the firm’s risk-management activities. Guay and Kothari (2003) showthat the median firm’s derivatives cashflow sensitivity (defined as the level of cashflows that derivativeinstruments can generate in extremely adverse scenarios of interest rate, foreign currency or commodityprices) is modest at only about 10% (mean of 45%) of the average year’s operating cash-flows of thefirms.27 At an extreme, if the median firm’s operating cash-flows drops to 25% of its normal level, theimpact of derivative instruments can be as high as 40% of a bad year’s operating cash-flows. However,at the same time the study by Guay and Kothari underscores the importance of nonderivativebased risk-management strategies for firm-value. The study by Petersen and Thiagarajan (2000) illustratesthe importance of nonderivative based hedging strategies for a firm’s overall risk-management decisions.In my empirical study I provide various robustness tests to account for nonderivative-based methods ofhedging.

4.1.4. Descriptive statistics of hedging variables

Table 1 provides the descriptive statistics of hedging activities. In Panel A, I provide the frequencydistribution of hedgers of different risks. Out of a total of 1,781 firms with an exposure to foreign currencyrisk, 497 (about 28% of the firms) use derivatives to hedge their exposure to movements in foreign exchangerates. For commodity price risk, there are 211 hedgers (about 20% of the firms) out of a total sample size of1,238 firms. If I consider exposure to either type of risk, I find a total of 645 hedgers from a sample of 2,256

27The sensitivity varies from 9% to 39% depending on the scaling variable used (see Table 4 of Guay and Kothari, 2003).

Page 15: Don't Know

ARTICLE IN PRESS

Table 1

Descriptive statistics—derivatives usage

This table provides the descriptive statistics of derivatives usage by sample firms. Panel A provides the number of firms that use foreign

currency (FX) or commodity (CM) derivatives for hedging purposes for the fiscal year ending between September 1996 to August 1997.

The ‘Any’ column represents the number of firms that use either FX or CM (or both) derivatives for hedging purposes. Panel B provides

the details on the notional amount of FX derivatives. Panel C provides the instrument-wise break-up of FX derivatives across swaps,

forwards/futures, and options. This panel is based on a smaller subsample of 435 foreign currency hedgers for which the instrument-wise

break-up is available. Statistics in Panel C are based on only those observations that have nonzero values for the respective hedging

instrument.

Panel A

Foreign Currency Commodity Any

Firms with Exposure 1781 1238 2256

Number of Derivative Users 497 211 645

Number of non-users 1284 1027 1611

Panel B

Mean Median Std. Dev.

Notional Amount 359.15 40.28 1010.99

As a % of Assets 8.62 4.43 22.36

As a % of Sales 10.74 4.07 50.48

Panel C

Frequency Mean % of Sales

Swap 63 344.82 8.07

Forward/Futures 360 259.72 6.73

Options 82 316.83 21.61

A. Purnanandam / Journal of Financial Economics 87 (2008) 706–739720

firms. Panel B provides the summary statistics for the aggregate notional amount of foreign currencyderivatives used for risk-management purposes. The mean (median) notional amount of foreign currencyderivatives is $359.15 million ($40 million). The average level of derivatives holdings in my sample is smallerthan that of earlier studies such as Graham and Rogers (2002). This is not surprising, since these studies focusmostly on large firms, whereas my sample contains many medium and small firms as well. The notional valueof derivatives scaled by the book value of the firm’s total assets (sales) amounts to 8.62% (10.74%) for theaverage firm in the sample. These numbers are comparable to earlier studies.

Table 1 (Panel C) also provides the break-up of foreign currency derivatives across instrument types.Forward and futures contracts are the most widely used instruments for managing foreign currency risk.Among the foreign currency hedgers, about 80% of firms use forward and futures contracts. In unreportedanalyses, I find that there are comparable levels of transactions for both buying and selling in the foreigncurrency forward markets.

My main tests are based on the relation between leverage and hedging. In the next section, I briefly describethe control variable used in the analysis before turning to the issue of endogenous modeling of risk-management and leverage decisions.

4.1.5. Control variables

Earlier theoretical and empirical work in this literature proposes several variables that can explain a firm’shedging incentives. My control variables are motivated by these studies. First, I control for firm size (size) asmeasured by log of total sales to capture the well-known size effects in derivative usage (see Dolde, 1993). I usethe ratio of research and development (R&D) expenses to total sales as a proxy for firm’s growthopportunities. Froot, Scharfstein, and Stein (1993) predict a positive relation between growth opportunitiesand hedging incentives since hedging can minimize the underinvestment problem in low cash-flow states of theworld. I also use a firm’s market-to-book ratio as an additional control variable for growth opportunities andobtain similar results. However, I do not include it in my base model since market-to-book has been taken as ameasure of firm-value in several studies in corporate finance and firm value may itself depend on derivativeusage. Second, I model leverage in an endogenous setting, which requires regressing leverage on all

Page 16: Don't Know

ARTICLE IN PRESSA. Purnanandam / Journal of Financial Economics 87 (2008) 706–739 721

explanatory variable in the first stage regression. Given the illusionary nature of relation between leverage andmarket-to-book ratio, this specification poses additional challenges.28 The underinvestment problem of a firmcan be reduced by keeping more liquid assets. I include the quick ratio of the firm (quick) as a measure of thefirm’s liquid assets. The quick ratio is constructed as a ratio of cash and short-term investments to the currentliabilities of the firm.29

Motivated by earlier studies (see Geczy, Minton, and Schrand, 1997; Graham and Rogers, 2002), I includeinstitutional shareholdings as an explanatory variable in the model to control for risk-management incentivesdue to information asymmetry between firm’s insiders and outsiders. DeMarzo and Duffie (1991) and Breedenand Viswanathan (1996) argue that firms with higher information asymmetry between managers andshareholders should hedge more. Assuming that higher institutional share-holdings leads to lower informationasymmetry between the managers and shareholders of the firm, the coefficient on this variable should benegative as predicted by these theories. The inst variable measures the fraction of common shares of the firmheld by institutional investors. The data are obtained from the 13-F filings. In an alternative unreportedspecification, I also use the number of analysts following the firm as a proxy of (inverse) informationasymmetry and obtain similar results.

Next, I control for tax convexity-based hedging incentives. If a firm faces a progressive tax structure, then itspost-tax value becomes a concave function of its pre-tax value. The firm can lower its expected tax liability byengaging in hedging activities (Smith and Stulz, 1985). I use the methodology suggested by Graham and Smith(1999) to measure the tax-convexity incentive of hedging. A brief description of their methodology is providedin Appendix A.6. The tax-convexity variable measures the expected tax benefits (in dollars) from a 5%reduction in the firm’s income volatility. I scale this measure by the total sales of the firm. Since this variable isestimated by using other accounting variables of the firm, in my base-case analysis I do not control for the tax-convexity measure to ensure that my key results are not driven by the inclusion of this imputed variable.Subsequently, I control for this effect and show that the results with respect to the key variables of interestremain robust to the inclusion of this control variable in the model.

In the foreign currency hedging model, I include foreign currency sales as a percentage of a firm’s total salesas an additional control variable (fsale). Jorion (1991) shows that foreign currency sales is a good proxy of thefirm’s exchange rate risk exposure. Thus, this variable controls for two effects. First, it controls for the extentof exposure faced by the sample firms, and second, it proxies for economies of scale that can be exploited inhedging foreign currency risks. High exposure firms should have a lower cost of hedging if there are significanteconomies of scale in these activities.

Firms can achieve significant reductions in their foreign currency risk exposure by operating in multiplegeographical locations around the world (see Allayannis, Ihrig, and Weston, 2001). A firm with morediversified geographical operations has a natural foreign currency hedge if currencies in different markets arenot highly correlated. I control for these effects by including the number of geographical segments reported bysample firms as a control variable. In an unreported analysis, I also control for the entropy of a firm’s foreignsales in diverse geographical regions and obtain similar results.30

28In one of the unreported analyses, I also use the analyst growth forecast obtained from I/B/E/S as a proxy for the growth option of the

firm. Since my results remain qualitatively similar, I don’t report the results of this model.29See also Acharya, Almeida, and Campello (2004), who argue that cash can serve as a hedge against future cash shortfalls for

financially constrained firms.30If a firm operates in n foreign segments (as defined in the COMPUSTAT segments files) and the percentage share of sales of foreign

segment i is Pi , then the entropy is computed as follows:

Entropy ¼Xn

i¼1

Pi lnð1=PiÞ. (6)

The entropy measure is a proxy of the firm’s geographical diversification—firms with higher entropy values have more diversified

operations across various foreign markets. As an additional robustness check, I also experiment with the firm’s geographical Herfindahl

index to control for this effect. All results remain similar to these alternative control variables.

Page 17: Don't Know

ARTICLE IN PRESSA. Purnanandam / Journal of Financial Economics 87 (2008) 706–739722

4.2. Endogenous modeling of leverage and hedging

My theory predicts a positive relation between leverage and hedging for firms with moderate levels ofleverage, and a negative relation at extremely high levels of leverage. In addition, the relation between leverageand hedging is expected to be stronger for firms operating in industries with greater likelihood of predatorybehavior such as concentrated industries. Thus, my key tests are based on the relation between hedgingand leverage.

For expositional simplicity and analytical tractability, at the time the hedging decision is made in thetheoretical model (i.e., at time t1 in the model) the debt level is pre-determined. However, we know from priortheoretical work that a firm’s debt capacity and hence leverage can itself increase due to hedging. For example,consider a variant of my model where firms hedge first and obtain debt at a later date. In the context of mymodel, hedging lowers the volatility of firm value, which in turn lowers the probability of bankruptcy and thusallows firms to borrow more for a given level of the tax benefit of debt. This leads to endogeneity betweenleverage and hedging. It, therefore, becomes important for my empirical study to explicitly account for thisendogeneity bias. To do so, I need a structural model for the capital structure choice and hedging decisions ofthe firm. In the absence of a consensus on an ideal model for debt choices, it is advantageous to have atheoretical model linking capital structure and hedging choices. I keep the empirical estimation tightly linkedto the theoretical model. In particular, I estimate the following structural model:

leverage ¼ b0 þ b1 � hedgingþ Sg � X i þ ei, (7)

hedging ¼ a0 þ a1 � leverageþ a2 � leverage2 þ Sy � Y i þ �i. (8)

This model is estimated in a two-stage instrumental variable (IV) regression framework. The first-stageequation is an OLS model for the leverage decision, whereas the second equation models a firm’s hedging(derivative) decisions. In the second stage, the risk-management equation is estimated using the predicted valueof the leverage ratio as the explanatory variable in the Logit or Tobit estimation. I try alternative econometricspecifications to this model in later sections.31 The leverage (leverage) of a firm is defined as the ratio of totaldebt (long-term debt plus debt included in the current liabilities) to the book value of total assets. Toinvestigate the effect of extreme leverage on hedging, I include leverage2 as an additional explanatory variablein the second equation. I expect a positive sign on leverage and a negative sign on leverage2 in the regressioninvolving various measures of hedging as the dependent variable. X and Y represent control variables affectingfirms’ leverage and hedging decisions, respectively.

As argued earlier, industry concentration provides a good measure of financial distress costs in my model.In such industries, highly levered firms are more vulnerable to losing their competitive position in the industryin the event of financial distress. Opler and Titman (1994) provide empirical evidence in support of thisassumption. Based on this argument, my model predicts a positive relation between hedging and industryconcentration for highly levered firms. To capture this effect empirically, I include industry concentrationmeasure and its interaction with leverage in the hedging model. This measure is constructed by summing themarket shares (based on sales in 1996) of the top four players in the firm’s three-digit SIC code. Then I create adummy variable (concd) that equals one if the concentration ratio is above the median, and zero otherwise.

4.2.1. Identification strategy

To estimate this model I need to find proper instrument(s) for the first-stage leverage regression. A largeliterature studies corporations’ capital structure determinants (see Frank and Goyal, 2003 for a survey) andresearchers have proposed several determinants of a firm’s leverage such as size, tangible assets, the book-to-market ratio, earnings volatility, profitability, and marginal tax rates (see Bradley, Jarrell, and Kim, 1984;Titman and Wessels, 1988; Lang, Ofek, and Stulz, 1996; Graham, Lemmon, and Schallheim, 1998 amongothers). For my identification strategy to work, one has to argue that one or more of these variables affect afirm’s hedging decision only through their impact on leverage and not independently by themselves. Finding a

31In particular, I estimate an alternative econometric model suggested by Wooldridge (2002) for IV estimations involving the presence of

a function of the endogenous variable (i.e., leverage2) in the second stage.

Page 18: Don't Know

ARTICLE IN PRESSA. Purnanandam / Journal of Financial Economics 87 (2008) 706–739 723

truly exogenous instrument for leverage is an extremely challenging task. Given this, I propose anidentification strategy that is motivated by the theoretical model itself.

In the theoretical section, the firm’s capital structure follows a trade-off model. As in standard trade-offmodels, the advantage of debt financing is its tax benefit, whereas its cost is financial distress and deadweightlosses of bankruptcy. The ex-ante debt ratio is determined by the relative costs and benefits of this trade-off.This provides a dispersion in the debt ratio at time zero in the model. Subsequently, in the intervening timeperiod the debt ratio is further perturbed by random shocks to the firm’s profitability. Thus, at the time thehedging decision is made (at time t1 in the model), the cross-sectional leverage ratio is an outcome of the firm’smarginal tax benefit of debt, the bankruptcy and distress costs of debt, as well as its recent profit history. Thetheoretical model focuses on how the leverage ratio affects hedging decisions at this point in time, which I callthe ex-post hedging decision. In that sense, leverage becomes pre-determined in the model at the time thehedging decision is made. At this time, shareholders engage in hedging activities as long as the firm’s leverageis not too high, beyond which point the risk-shifting incentives begin to dominate.

I first note that in my model the first-order effect of hedging on leverage (i.e., the concern about reversecausation) is through its affect on the cost of leverage and not through its effect on its tax benefit. Thus, atleast in the context of my stylized model the marginal tax benefit of debt provides one key source of dispersionin the ex-ante debt ratio that remains largely unaffected by the extent of hedging. It is the deadweight cost offinancial distress and bankruptcy that decreases due to hedging, allowing firms to borrow more. Thus,marginal benefit of debt seems like a reasonable instrument for identifying the leverage equation in myempirical model. Motivated by this logic, I consider two instruments—the before-financing simulatedmarginal tax rate (MTR) of Graham, Lemmon, and Schallheim (1998) and a firm’s nondebt tax shield(NDTS).

Marginal tax rates provide a reasonably direct proxy for the tax benefit of debt. Thus, it is directly in thespirit of my theoretical motivation. To avoid problems associated with negative spurious correlation betweenthe leverage and marginal tax rates, I use the before-financing simulated tax rates. Further, I take the historicalaverages of MTRs under the assumption that the current level of debt is an outcome of historical incrementalcapital structure decisions. I take the past 10 year’s average MTR as a proxy for a firm’s current leverage ratio.As discussed later, this variable has a significant explanatory power in leverage regression. I also repeat myanalysis with last five year’s average and the current level of MTR (without averaging) as instruments andobtain similar results.32

My second instrument is the nondebt tax shield enjoyed by a firm. Following earlier literature, I usedepreciation and amortization (da) scaled by the total assets of the firm as a measure of the firm’s nondebt taxshield. This instrument measures the disincentive of using debt rather than directly measuring the incentive touse debt based on tax considerations as captured by MTR. Thus, it is capable of detecting the firm’s leverageratios in response to the tax incentives as proposed in my model. At least controlling for firm’s size, PPE, andother key characteristics, it can be argued that nondebt tax shield is a reasonable instrument for leverage in myleverage-hedging model.

Both these instruments (MTR and DA/TA) possess good statistical properties for an instrument. They bothare significant determinants of firm’s debt ratios in the first stage regression reported in the next section. I alsocheck for their strength and find that they do not suffer from any weak instrumentation bias in the sense ofBound, Jaeger, and Baker (1995) and Staiger and Stock (1997). I repeat all my analyses after considering onlyMTR and NDTS (one at a time) as my instrument and all results remain qualitatively similar. In order to savespace and due to the statistical advantage of having more instruments, I consider them both in my leveragemodel for the results that I present in the paper. In addition, I use a firm’s net income to sales ratio (ni) in theleverage regression as an additional instrument to capture the effect of recent profitability on a firm’s capitalstructure at the time of hedging in the spirit of my theoretical model. As I show later, this variable performswell in the first stage regression as well.

I include additional control variables in the spirit of Titman and Wessels (1988) and Graham, Lemmon andSchallheim (1998) to control for well-known drivers of cross-sectional dispersion in leverage ratios. First,

32Historical average MTR has better statistical properties in explaining the leverage-ratio than the current MTR. Therefore, I prefer the

average MTR on statistical grounds as well.

Page 19: Don't Know

ARTICLE IN PRESSA. Purnanandam / Journal of Financial Economics 87 (2008) 706–739724

I include Property, Plant and Equipment (ppe) scaled by total assets to control for the collateral available forborrowing. I include a firm’s modified Z-score (see Graham, Lemmon, and Schallheim, 1998) to control forthe effect of firms that may currently be in financial distress. The modified Z-score (modz) excludes the effectof leverage from the original Altman Z-score to avoid a mechanical relation between leverage and thisvariable. I also include two-digit SIC codes to control for industry-specific drivers of capital structure in myleverage model. In addition, firm size and R&D-to-Sales ratio enter both the hedging and leveragespecifications. Overall, my model is in line with the theoretical arguments and is also close to earlier empiricalstudies in corporate risk-management such as Geczy, Minton, and Schrand (1997) and Graham and Rogers(2002). The base model is presented below:

lev ¼ b0 þ b1hedgingþ b2sizeþ b3rnd þ b4MTRþ b5ppeþ b6modzþ b7ni þ b8daþX

Ind þ ei.

hedging ¼ a0 þ a1levþ a2lev2 þ Sy � Y i þ �i.

I add several other variables to this specification in additional tests. As an alternative test, I use three-yearpanel data of 200 manufacturing firms and regress changes in hedging activities on changes in leverage ratios.Change regressions are less likely to suffer from endogeneity biases and face a tougher hurdle in detecting anassociation between the variables of interest. My results are qualitatively similar for both the cross-sectionalIV regression model and the change regression model. Further, it should be noted that the nonlinearspecification that I use in my modeling approach gives additional confidence that my results are not driven byreverse causality. This obtains because the endogeneity in my model comes from the fact that hedging can leadto higher debt levels; this is true for all levels of leverage and especially so for higher levels. Thus, theendogeneity argument will predict a positive relation between hedging and both leverage and leverage2,something that is opposite to what my theory predicts.

4.3. Univariate tests

Table 2 presents the median values of key firm-level variables across hedgers and nonhedgers. To preventoutliers from affecting my analysis, all variables used in this paper are winsorized at 1% from both tails. InPanel A, I present the median characteristics of hedgers and nonhedgers of foreign currency risk in the sampleof firms with exposure to this risk. Panel B provides the same statistics for hedgers and non-hedgers for firmswith exposure to commodity price risk. Panel C is based on pooled observations across both types of risk.

I find that the hedgers have significantly different characteristics from the nonhedgers. The hedgers aresignificantly larger firms—the median hedger firm is about four to five times bigger than the nonhedger firm interms of market capitalization or total sales. The median leverage for hedgers is significantly higher than themedian leverage of nonhedgers, with stronger results for commodity hedging sample. Hedgers keep less liquidassets as compared with the nonhedgers as shown by the quick ratios of the two groups. As expected, theforeign currency hedgers have higher foreign currency sales as compared with the nonhedgers. Not surprising,there is no difference in the extent of foreign sales across hedgers and nonhedgers in the sample of firms withcommodity exposure. I also find that hedgers have significantly larger institutional shareholdings than non-hedgers. While foreign currency hedgers have higher growth opportunities (as proxied by R&D to sales andmarket-to-book ratio) than their nonhedger counterparts, this pattern reverse for the commodity hedgers.I explore these effects more carefully in the multivariate models presented below.

4.4. Regression analysis

In this section I present the regression results relating a firm’s hedging incentives to leverage and othercontrol variables.

4.4.1. First stage estimation

As a starting point, I present the regression results from the first-stage estimation of leverage as reported inthe first panel of Table 3. I find a positive and significant coefficient on MTR indicating that firms with highertax benefits obtain higher debt. As expected the coefficient on depreciation and amortization, da=ta the proxy

Page 20: Don't Know

ARTICLE IN PRESS

Table 2

Summary statistics

This table presents the descriptive statistics for the key explanatory variables used in the analysis. Panel A presents the median

characteristics of users and nonusers of foreign currency (FX) derivatives based on 1,781 observations that are identified as firms with

exposure to foreign currency risk. Panel B is based on commodity (CM) derivatives (1,238 observations with exposure to commodity price

risk), and Panel C is based on the usage of any of these two derivatives (2,256 observations). In every panel, I provide the median

characteristics of hedgers and nonhedgers as well as the entire sample. The last row in each sample gives the p-Value for the test that

median characteristics for hedger and nonhedger groups are equal based on a Wilcoxon-Mann-Whitney test. Sales represent the total sales

of the firm as reported under item 12 of COMPUSTAT tapes. mv stands for market value obtained by multiplying COMPUSTAT item 25

by item 199. lev measures the ratio of total liabilities (sum of COMPUSTAT items 9 and 34) to total assets (item 6). Quick ratio is

constructed as the ratio of cash and short-term investments (item 1) to current liabilities (item 5). fsale represents the ratio of foreign sales

to total sales of the firm. The foreign sales data are obtained from the COMPUSTAT geographical segments file. inst measures the

percentage institutional ownership in the firm. rnd stands for percentage research and development expenses (item 46) scaled by the sales of

the firm (item 12). mtb stands for the market-to-book ratio of the firm’s assets (COMPUSTAT (item 6 minus 60 plus ð25 � 199Þ) scaled by

item 6).

sales mv lev quick fsale inst rnd mtb

Panel A: FX derivatives

Nonhedgers 228.5150 256.6120 0.1744 0.2632 0.0685 44.1235 0.0000 1.6588

Hedgers 1147.0000 1313.4053 0.2082 0.2159 0.3480 58.0480 2.1435 1.7048

All 334.4900 392.5571 0.1877 0.2460 0.1416 47.7017 0.7773 1.6723

p-Value 0.01 0.01 0.05 0.04 0.01 0.01 0.01 0.02

Panel B: Commodity derivatives

Nonhedgers 212.0220 214.9539 0.2194 0.2378 0.0000 38.3827 0.0000 1.5692

Hedgers 768.4550 798.7656 0.2829 0.1412 0.0000 53.0766 0.0000 1.5024

All 244.8135 263.7081 0.2331 0.2091 0.0000 41.1219 0.0000 1.5501

p-Value 0.01 0.01 0.01 0.01 0.33 0.01 0.01 0.09

Panel C: Any derivatives

Nonhedgers 201.7550 207.9030 0.1983 0.2505 0.0000 39.3921 0.0000 1.6002

Hedgers 917.1540 977.1712 0.2267 0.1995 0.2828 56.2881 1.0701 1.6659

All 285.0805 305.7399 0.2071 0.2271 0.0241 44.4876 0.0000 1.6136

p-Value 0.01 0.01 0.04 0.02 0.01 0.01 0.01 0.06

A. Purnanandam / Journal of Financial Economics 87 (2008) 706–739 725

for nondebt tax shields, is negative and significant. Further, consistent with my model firms with higherprofitability have lower leverage as indicated by a negative and significant coefficient on net income to salesðni=salesÞ. These results are consistent with the motivations behind the use of these variables in the leverageregression model. Other results are in line with the earlier empirical literature. Once I obtain the predictedvalues of leverage from the first-stage model, I use it in the second-stage model to explain a firm’s foreigncurrency and commodity hedging decision. To save space, I do not present the results from the first stageestimation in the rest of the paper.

4.4.2. Foreign currency hedging

I start with the firm’s foreign currency hedging decision and subsequently analyze the commodity hedgingdecisions.

Yes/No decision: I present the results of a second-stage Logit regression in Table 3. The dependent variableequals one if a firm uses foreign currency derivatives and zero otherwise. The model is estimated with onlythose firms that have a pre-defined exposure to foreign currency risks. In the first model, leverage is positiveand significant at 1% whereas leverage2 is negative and significant at the 1% level. For easier interpretation, Ipresent the marginal effect (on the probability of hedging) of the explanatory variable evaluated at the meanrather than the raw estimated coefficient from the logit model. In the next model, I include the interaction ofleverage and a dummy indicating whether the firm belongs to a highly concentrated industry or not. I find apositive coefficient on the interaction of leverage and industry concentration (concd). As expected, the

Page 21: Don't Know

ARTICLE IN PRESS

Table 3

Foreign currency hedging—yes/no decision

This table presents logistic regression results for foreign currency hedging by means of derivatives. In the first stage I estimate an OLS

regression model for leverage. The estimation results from this regression are presented in the first two columns. In addition to the

coefficients reported in this table, this regression also includes industry dummies based on two-digit SIC code (coefficients suppressed). In

the second stage, a logistic model is estimated with firm’s foreign currency derivative usage as the dependent variable (one for hedgers and

zero for nonhedgers). lev� denotes the predicted value of leverage from the first-stage regression. The marginal effect of explanatory

variables (evaluated at the mean) on the probability of hedging along with associated t-Values are presented in the table. Columns 3-8

present results from the second-stage estimation of hedging model. size represents the log of total sales of the firm. quick is the ratio of cash

and short-term investments to current liabilities. rnd stands for research and development expenses scaled by the sales of the firm. concd is

a dummy variable based on the four-firm concentration ratio of the firm’s industry (based on three-digit SIC code). concd equals one if the

firm belongs to an industry with a concentration ratio above the median, zero otherwise. fsale represents foreign sales as a percentage of

total sales. inst measures the percentage institutional ownership in the firm. mtr stands for the historical average of firm’s marginal tax

rates. ppe/ta stands for plant, property, and equipment scaled by total assets. Modified Z is the Altman Z-score without the leverage effect.

ni/sales stands for the ratio of net income to total sales. taxconvexity measures the dollar tax benefit from a 5% volatility reduction in the

firm’s income scaled by the sales of the firm. mtb stands for the market-to-book ratio of the firm. segno stands for the number of

geographical segments in which the firm operates. The number of observations and R2 (for OLS regression) are provided at the end of the

table.

Leverage FX Derivatives

Estimate t-Value Estimate t-Value Estimate t-Value Estimate t-Value

size 0.0116 (3.61) 0.1348 (12.70) 0.1346 (12.65) 0.1170 (9.80)

lev� 1.1181 (3.04) 0.7657 (1.94) 0.9519 (2.52)

lev�2 �2.3759 (�3.34) �2.3041 (�3.23) �2.1029 (�2.95)

lev � concd 0.5295 (2.42)

quick �0.0295 (�5.73) 0.0203 (1.10) 0.0155 (0.82) 0.0149 (0.78)

rnd �0.0060 (�6.52) 0.0103 (5.20) 0.0104 (5.19) 0.0101 (4.95)

concd 0.0112 (1.01) �0.0479 (�1.75) �0.1706 (�2.87) �0.0499 (�1.83)

fsale �0.0123 (�0.94) 0.3457 (9.34) 0.3497 (9.39) 0.1918 (3.95)

inst �0.0006 (�2.91) 0.0007 (1.20) 0.0007 (1.16) 0.0006 (1.00)

mtr 0.2634 (2.79)

ppe=ta 0.1105 (2.56)

modifiedz �0.0783 (�12.90)

ni=sales �0.1729 (�2.95)

da=ta �0.6776 (�2.64)

taxconvexity �0.7850 (�1.54)

mtb �0.0074 (�0.62)

segno 0.0670 (4.60)

R2 0.402

N 1,421 1,421 1,421 1,418

A. Purnanandam / Journal of Financial Economics 87 (2008) 706–739726

introduction of this interaction term lowers the significance of leverage variable, but it still remains significantat almost the 5% level. These findings are consistent with the key predictions of the model.

Other results indicate that firms with higher foreign currency sales are more likely to use hedging products,indicating that highly exposed firms have higher incentives to hedge. There is a strong relation between growthopportunities as measured by R&D expenses and hedging. This finding is consistent with the theoreticalpredictions of Froot, Scharfstein, and Stein (1993) and earlier empirical findings of Geczy, Minton, andSchrand (1997). I find a positive relation between institutional shareholdings and hedging. Assuming aninverse relation between institutional shareholdings and the extent of information asymmetry between theinsiders and outsiders of the firm, this result is inconsistent with information asymmetry-based models ofhedging. However, more analysis is needed to draw stronger inferences for this theory since the measurementof information asymmetry remains a difficult task for empirical researchers. In the final model I include threemore control variables: tax convexity, market-to-book ðmtbÞ, and the number of geographical segments(segno) in which the firm operates. All key results remain similar. I don’t find evidence in support of tax-based

Page 22: Don't Know

ARTICLE IN PRESS

Table 4

Foreign currency hedging—extent of hedging

This table presents the Tobit regression results for foreign currency hedging by means of derivatives. In the first stage (unreported), I

estimate an OLS regression model for leverage. In the second stage, a Tobit model is estimated with firm’s foreign currency derivative

usage as the dependent variable. This variable takes the value of the notional amount of foreign currency derivatives scaled by total sales

of the firm (zero for nonhedgers). I provide the second-stage estimation results for three different model specifications in the table below.

lev� denotes the predicted value of leverage from the first-stage regression. The marginal effect of explanatory variables (evaluated at the

mean) on the expected value of uncensored observations along with associated t-Values are presented in the table. size represents the log of

total sales of the firm. quick is the ratio of cash and short-term investments to current liabilities. rnd stands for research and development

expenses scaled by the sales of the firm. concd is a dummy variable based on the four-firm concentration ratio of the firm’s industry (based

on three-digit SIC code). concd equals one if the firm belongs to an industry with concentration ratio above the median, zero otherwise.

fsale represents foreign sales as a percentage of total sales. inst measures the percentage institutional ownership in the firm. taxconvexity

measures the dollar tax benefit from a 5% volatility reduction in the firm’s income scaled by the sales of the firm. mtb stands for the

market-to-book ratio of the firm. segno stands for the number of geographical segments in which the firm operates. The number of

observations is provided at the end of the table.

Estimate t-Value Estimate t-Value Estimate t-Value

size 0.0117 (10.51) 0.0116 (10.45) 0.0104 (8.47)

lev� 0.1435 (3.21) 0.1074 (2.27) 0.1459 (3.19)

lev�2 �0.2741 (�3.19) �0.2699 (�3.14) �0.2721 (�3.17)

lev� � concd 0.0572 (2.21)

quick 0.0032 (1.46) 0.0029 (1.33) 0.0036 (1.66)

rnd 0.0010 (4.21) 0.0010 (4.17) 0.0009 (3.55)

concd �0.0033 (�1.05) �0.0162 (�2.37) �0.0041 (�1.31)

fsale 0.0370 (8.27) 0.0372 (8.33) 0.0176 (3.06)

inst 0.0001 (1.37) 0.0001 (1.36) 0.0001 (1.49)

taxconvexity 0.0226 (0.46)

mtb 0.0005 (0.38)

segno 0.0086 (4.99)

N 1,421 1,421 1,418

A. Purnanandam / Journal of Financial Economics 87 (2008) 706–739 727

motivations for hedging. Finally, firms operating in more diverse foreign markets hedge more as evident by apositive and highly significant coefficient on this variable. This result points toward the possibility thatderivative instruments act as complements to a firm’s natural hedging strategies.

Extent of hedging: In Table 4, I present results from the Tobit estimation with the notional amount offoreign currency derivatives scaled by total sales of the firm as the dependent variable. Since the estimatedcoefficients in a Tobit model don’t represent the marginal effect of explanatory variables on the observeddependent variable, for easier economic interpretation I report the slope coefficients at the mean level.I present results from three different model specifications and find that firms with high leverage hedge moreand the relation between hedging and leverage reverses at very high levels of leverage. Highly levered firms inconcentrated industries have higher hedging incentives as well. These results are in line with both the model’spredictions and the results obtained from the logit model described earlier, as well as with the earliertheoretical models based on bankruptcy costs (Smith and Stulz, 1985). Economically, these results suggest thatif leverage increases from 10% to 20%, the firm increases its foreign currency derivative holdings byapproximately 6.4%, which is about 60% of the average level of foreign currency derivatives held by thesample firms (see Table 1; these are only rough estimates with linear extrapolation around the mean.)

Earlier studies provide mixed evidence in support of hedging theories based on financial distress costs. Mian(1996) studies the binary (i.e., yes–no) hedging decision for a large sample of firms and finds no support for thedistress cost theories.33 My results suggest that linear models seeking to test theories of risk-management,

33There are three possible explanations for the different results between my study and Mian (1996). First, his sample comes from 1992—

a period before the strict FASB regulation on derivatives disclosure. Second, his model doesn’t test for nonlinearity and therefore it doesn’t

have any quadratic terms. Finally my modeling technique is different as I use a two-stage estimation technique that considers leverage and

hedging as endogenous variables.

Page 23: Don't Know

ARTICLE IN PRESSA. Purnanandam / Journal of Financial Economics 87 (2008) 706–739728

especially if conducted on small sample sizes, may fail to detect a positive relation between financial distressand derivative activities for moderately leveraged firms.

To understand the nonmonotonic relation between hedging and leverage, I conduct a semi-parametric test.I break the sample of firms into two groups based on whether their predicted leverage is below or above the70th percentile of the empirical distribution of leverage in my sample. For this estimation I run a Tobitregression with the same set of variables as in Model 1 of Table 4 after dropping leverage2. The splineregression results show that for the first group, i.e., for the group with moderate leverage the marginal effect ofleverage on hedging is positive with a slope coefficient of 0.0452, which is significant at the 1% level. However,the marginal effect of leverage on hedging becomes negative for firms in the other group, i.e., for firms withleverage in top 30% of the sample. For this group, the marginal effect of leverage on hedging is estimated to be�0:1504 with a significance level of 2%. The semi-parametric test confirms the non-monotonic relationobtained in parametric regressions.

4.4.3. Commodity hedging

Table 5 provides logistic regression results for the commodity hedging decision. This regression is estimatedon a sample of firms with exposure to commodity price risk only. As in the foreign currency derivativeregression, I find a positive and significant coefficient on leverage, and a negative and significant coefficient onleverage2. Both these relations are significant at the 1% level. When I include the interaction of leverage withhigh concentration industry, I find the coefficient on the interaction term to be positive and significant at the6% level. These results show that the predictions of my theory are supported by both foreign currency andcommodity hedging data. As in the case of foreign currency hedging, larger firms are more likely to usecommodity derivatives as well. However, in this regression the coefficient on the quick ratio becomes positiveand significant, while it is positive but insignificant in the foreign currency hedging models. Commodityhedgers keep more liquid assets as well, which can be taken as evidence that hedgers complement their hedgingpolicies with liquid assets.

The most noticeable difference between the two models is the coefficient on the R&D variable. This variablehas a positive and highly significant coefficient in the logit and Tobit model of foreign currency hedging

Table 5

Commodity hedging—yes/no decision

This table presents logistic regression results for commodity hedging by means of derivatives. In the first stage (unreported) I estimate an

OLS regression model for leverage. In the second stage, a logistic model is estimated with firm’s commodity derivative usage as the

dependent variable (one for hedgers and zero for non-hedgers). lev� denotes the predicted value of leverage from the first stage regression.

The marginal effect of explanatory variables (evaluated at the mean) on the probability of hedging along with associated t-Values are

presented in the table. size represents the log of total sales of the firm. quick is the ratio of cash and short-term investments to current

liabilities. rnd stands for research and development expenses scaled by the sales of the firm. concd is a dummy variable based on the four-

firm concentration ratio of the firm’s industry (based on three-digit SIC code). concd equals one if the firm belongs to an industry with

concentration ratio above the median, zero otherwise. inst measures the percentage institutional ownership in the firm. taxconvexity

measures the dollar tax benefit from a 5% volatility reduction in the firm’s income scaled by the sales of the firm. mtb stands for the

market-to-book ratio of the firm. The number of observations is provided at the end of the table.

Estimate t-Value Estimate t-Value Estimate t-Value

size 0.0360 (5.20) 0.0355 (5.25) 0.0354 (4.79)

lev� 0.9815 (3.23) 0.7556 (2.47) 0.9680 (3.13)

lev�2 �1.7470 (�3.46) �1.7267 (�3.50) �1.7337 (�3.40)

lev�concd 0.3107 (1.94)

quick 0.0229 (2.13) 0.0229 (2.20) 0.0224 (2.04)

rnd �0.0216 (�6.76) �0.0220 (�7.19) �0.0215 (�6.56)

concd 0.0206 (1.23) �0.0737 (�1.27) 0.0212 (1.26)

inst 0.0008 (2.16) 0.0008 (2.20) 0.0008 (2.02)

taxconvexity �0.0764 (�0.33)

mtb �0.0013 (�0.15)

N 948 948 947

Page 24: Don't Know

ARTICLE IN PRESS

Table 6

Alternative models

This table presents the results for derivatives usage for various alternative model specifications. The first two models use alternative

definitions of financial distress, whereas the last two models use leverage as the measure of financial distress but use bootstrapped standard

errors in estimating the t-Values. FD stands for the measure of financial distress for the given model and FD2 is its squared term. The first

model uses the industry-adjusted leverage ratio based on two-digit SIC codes as a measure of FD. For this model FD2 equals leverage-

squared if the firm’s leverage is above industry average, zero otherwise. In the second model, I use Altman’s Z-score as a measure of the

firm’s financial distress and estimate a logistic model using the firm’s usage of foreign currency or commodity derivatives as the dependent

variable (one for users, zero for nonusers). For consistency with other models, I set FD equal to the inverse of the Z-score such that a

higher value of FD corresponds to firms closer to financial distress. The third model replicates the base-case logistic regression with

bootstrapped standard errors. The dependent variable is one for the users of commodity or foreign currency exposure, zero otherwise. In

the fourth model I estimate a Tobit model for the extent of foreign currency derivatives using bootstrapped standard errors. For these last

two models FD stands for predicted leverage from the first-stage regression, and FD2 is simply the squared predicted leverage. All

regression results provide the slope estimates evaluated at the mean of the explanatory variable along with corresponding t-statistics. size

represents the log of total sales of the firm. quick is the ratio of cash and short-term investments to current liabilities. rnd stands for

research and development expenses scaled by the sales of the firm. concd is a dummy variable based on the four-firm concentration ratio of

the firm’s industry (based on three-digit SIC code). concd equals one if the firm belongs to an industry with a concentration ratio above the

median, zero otherwise. fsale represents foreign sales as a percentage of total sales. inst measures the percentage institutional ownership in

the firm. Number of observations used for the estimation is provided in the last row.

Estimate t-Value Estimate t-Value Estimate t-Value Estimate t-Value

Ind Leverage Z-score LOGIT TOBIT

size 0.1089 (13.20) 0.0433 (1.98) 0.1167 (12.05) 0.0111 (5.88)

FD 0.2536 (3.07) 0.0812 (3.83) 1.3070 (4.01) 0.1162 (2.57)

FD2 �0.7713 (�2.70) �0.0265 (�11.11) �2.4002 (�4.18) �0.2276 (�2.71)

quick 0.0111 (0.89) 0.0039 (0.79) 0.0415 (2.63) 0.0029 (1.41)

rnd 0.0039 (2.35) 0.0019 (1.70) 0.0057 (2.81) 0.0009 (3.58)

concd �0.0255 (�1.16) �0.0095 (�0.95) �0.0250 (�1.09) �0.0034 (�1.47)

fsale 0.2833 (9.07) 0.1168 (1.97) 0.3056 (8.47) 0.0345 (4.56)

inst 0.0010 (2.05) 0.0004 (1.48) 0.0012 (2.07) 0.0001 (1.49)

N 2,089 2,049 1,769 1,421

A. Purnanandam / Journal of Financial Economics 87 (2008) 706–739 729

decisions. However, here it is negative and insignificant. While high growth firms manage their foreigncurrency hedging more aggressively, they are less likely to manage their commodity risk exposure. Althoughexploring the differences in hedging incentives across different types of risks is beyond the scope of this paper,this finding is suggestive of firm’s facing conflicting incentives in managing various forms of risk. Theseconflicting incentives may be driven by factors such as differences in the correlation between the risk beinghedged and the firm’s investment opportunity set. For example, the argument behind a positive relationbetween growth opportunities and hedging relies on the assumption that high growth firms may need funds toundertake projects in bad cashflow states. If firms do not have good investment opportunities in states withpoor realizations of cashflows, then this incentive disappears. At the extreme, if the investment opportunity setis highly positively correlated with the realizations of cashflows, then such firms may have a disincentive tohedge. A better understanding of these issues is left for future research that explicitly incorporates thesecorrelations in the analysis.

4.5. Alternative model specifications

Industry-adjusted leverage ratios: The results presented so far in the paper are based on a two-stagespecification that requires an assumption about the structural model determining a firm’s leverage choice. Forrobustness, I test a model that does not require such a specification. Specifically, I conduct the analysis withfirms’ industry-adjusted leverage ratios as industry adjustment provides a simpler and perhaps more robustway to classify firms into moderately and highly leveraged. The industry-adjusted leverage ratio of a firm isdefined as the difference between the firm’s leverage and the industry median based on the two-digit SIC code.

Page 25: Don't Know

ARTICLE IN PRESSA. Purnanandam / Journal of Financial Economics 87 (2008) 706–739730

I reestimate all my results with these ratios. For this model, leverage variable can be either positive or negativedepending on whether the firm is above or below industry-median. Thus, I cannot use leverage2 as anexplanatory variable to test for the nonlinearity predicted by the model. Instead, I use a variable that equalsleverage2þ for firms with higher than industry median leverage and zero otherwise. To conserve space, I poolforeign currency and commodity hedging decisions to estimate these robustness results. I estimate a logitmodel on a sample of firms that are exposed to either of these two types of risks and present the results inModel 1 of Table 6. 34 My key results remain the same with this definition of leverage.

Altman Z-score: I use the Altman Z-score as an alternative proxy of financial distress. Lower Z-score valuescorrespond to financially weaker firms. I, therefore, transform them by taking their inverse to be consistent inthe presentation of results. Results are presented in Model 2 of Table 6. I find a nonmonotonic relation basedon this measure as well.

Bootstrapped standard errors: Since I use a two-stage estimation methodology in the logit and Tobitregressions, there is a potential for overstated t-statistics due to the sampling error of first-stage estimation (seeMaddala, 1983). To account for this possibility I reestimate my models with bootstrapped standard errors. Inevery replication I create a pseudo-random sample by drawing observations from the base sample withreplacement. Thus, in every replication some of the observations appear more than once and some do notappear at all. With 100 such replications, I generate an empirical distribution of estimated coefficients inthe logit and Tobit models. The standard deviations of these estimates are then used to obtain bootstrappedp-Values for my base estimation. This methodology does not rely on any structural form for the estimation ofthe variance-covariance matrix and has the advantage of benchmarking base estimates against their empiricaldistributions. In Models 3 and 4 of Table 6, I present the Logit and Tobit model estimates with bootstrappederrors. As shown, all my key results are robust to bootstrapped standard error estimation.

Alternative IV regression specification: Wooldridge (2002) suggests an alternative instrumental variableregression model for models involving functions of an endogenous variable (such as leverage2 in the second-stage estimation). Potentially, this technique provides econometrically better estimates than the modelthat uses predicted values of leverage and its function in the second stage. In this method, rather thanusing the squared value of predicted leverage in the second-stage regression, both leverage and leverage2

are treated as endogenous variables and instrumented with their own instruments. To achieve identification,I add the squared terms of all exogenous variables entering the leverage model as instruments for leverage2.These instruments are the squared values of: mtr, da=ta, ppe=ta, ni=sales and modified_z. In addition, thesquared predicted value of leverage from the first stage estimation is used as an additional instrument forleverage2.

With both leverage and leverage2 as endogenous variables and these instruments in hand, I estimate aninstrumental variable model in a two-stage regression framework. I estimate the binary decision to hedgeforeign currency or commodity derivatives with an IV Probit model and the extent of foreign currency hedgingwith an IV Tobit model. The results are presented in Table 7. Note that the parameter estimates in this modelare not directly comparable to the earlier tables since, due to computational simplicity, I report the coefficientsfrom the regressions directly rather than the slope coefficients presented earlier. I find that my key resultsremain robust to this alternative IV estimation technique. All other results remain similar to the earlier base-case specification.

4.6. Dynamic analysis

Change regression: I focus on a three-year panel of 200 manufacturing firms to exploit the variations in theindividual firm’s leverage and hedging intensities in a dynamic setting. This empirical strategy closelyresembles my theoretical model and presents several econometric advantages. The change regression controlsfor unobserved firm-specific factors. In particular, unless a firm’s nonderivative-based hedging strategies havechanged substantially over this time period, change regressions control for operational and natural hedging ina more precise manner. Second, the endogeneity argument is less severe for change regressions since byconstruction it removes firm-specific unobservable effects that could be correlated with both hedging and

34Individual regressions estimated separately on foreign currency and commodity samples are qualitatively similar.

Page 26: Don't Know

ARTICLE IN PRESS

Table 7

Alternative IV model

This table presents the second-stage results for derivatives usage for an alternative instrumental variable regression model following

Wooldridge (2002). Rather than using the square of predicted values of leverage in the second-stage regression, this specification uses both

leverage and its squared term as endogenous variables. Leverage is instrumented with da/ta (depreciation and amortization scaled by total

assets), mtr (marginal tax rates), ppe (property, plant and equipment scaled by total assets), modified Z-score and ni/sales (net income to

total sales). The squared terms of each of these variables are used as additional instruments for leverage2. In addition, the squared value of

the predicted leverage is also used as an instrument for leverage2 as suggested by Wooldridge (2002). Models 1 and 2 provide the second-

stage results from the instrumental variable Probit models with foreign currency derivatives and commodity derivatives, respectively. In

Model 3, an instrumental variable Tobit model is estimated. The dependent variable in Tobit regressions is the notional amount of foreign

currency derivatives (scaled by total sales) for the users of derivatives and zero for the rest of the firms. Size represents the log of total sales

of the firm. rnd stands for research and development expenses scaled by the sales of the firm. quick is the ratio of cash and short-term

investments to current liabilities. concd is a dummy variable based on the four-firm concentration ratio of the firm’s industry (based on

three-digit SIC code). concd equals one if the firm belongs to an industry with concentration ratio above the median, zero otherwise. fsale

represents foreign sales as a percentage of total sales. inst measures the percentage institutional ownership in the firm. For all three

regressions, the parameter estimate and p-Values are provided. Number of observations used for the estimation is provided in the last row.

Estimate t-Value Estimate t-Value Estimate t-Value

FX Yes/No Commodity Yes/No FX-Extent

size 0.3628 (9.10) 0.1976 (4.20) 0.0412 (6.59)

lev 6.6451 (2.93) 9.5649 (4.22) 1.0180 (2.77)

lev2 �11.0198 (�3.12) �11.4499 (�3.65) �1.5277 (�2.71)

rnd 0.0348 (4.72) �0.0890 (�3.40) 0.0047 (3.91)

quick 0.0937 (1.36) 0.3032 (3.53) 0.0186 (1.63)

concd �0.1774 (�1.94) 0.2141 (1.57) �0.0187 (�1.27)

fsale 1.0809 (8.25) 0.1584 (7.32)

inst 0.0020 (0.93) 0.0045 (1.59) 0.0004 (1.24)

N 1421 948 1421

A. Purnanandam / Journal of Financial Economics 87 (2008) 706–739 731

leverage at any given point in time. Third, this model allows me to relate changes in this year’s hedgingintensities to both current and past year’s leverage changes, allowing me to establish some evidence on ex-anteand ex-post hedging incentives.

I obtain data on derivative holdings for a random subsample of 200 manufacturing firms (one-digit SICcode 2) for the fiscal years ending in 1997–1998 and 1998–1999, i.e., for two more years after the initial sampleperiod. I focus on manufacturing firms to ensure that the sample is homogenous. The choice of a smallersubsample is purely dictated by the requirements of manually collecting data. After dropping one firm-yearobservation due to the non-availability of its 10-K on EDGAR and restricting attention to firms withnonmissing observations, I obtain 394 first-differenced firm-year observations. The accounting characteristicssuch as size, leverage and market-to-book ratio of this subsample are qualitatively similar to those of theoverall sample (unreported) and thus the sample is a reasonable representation of COMPUSTAT-CRSPfirms.

For this sample, I obtain data on their foreign currency derivatives and investigate a subset of firmsthat have either increased or decreased the intensity of their foreign currency hedging. Since there arevery few initiators or terminators of commodity derivatives in my sample, the change regression is notfeasible for commodity hedges. In this sample, there are 42 firm-year observations with an increase and39 firm-year observations with a decrease in the extent of hedging (notional amount of foreign currencyderivatives as a percentage of sales). The remaining firm-year observations have no changes mostly due to zerohedging positions across the period. In my analysis, I focus on only those observations that have eitherincreased or decreased their hedging positions to avoid inferences based on a majority of firms that remainnonhedgers during the sample period. Focussing on the sample of active hedging decisions (increase ordecrease) allows for a sharper identification of hedging incentives in response to changes in firm-levelvariables.

Page 27: Don't Know

Table 8

Change regression

This table presents logistic and OLS regression results based on yearly changes in foreign currency derivatives holding of a sample of

manufacturing firms (SIC code 2) for 1996–1997 to 1998–1999 period. Changes in derivative holding is regressed on changes in leverage

and various other firm characteristics. The dependent variable is one if firm increases its hedging intensity as measured by the ratio of

foreign currency derivatives to total sales, zero if decreases it. Dlev is the change in book leverage over the same year. Dlev2 equals the

square of Leverage change if leverage has increased over the year and zero otherwise. Dlaglev is the previous year’s change in book

leverage. All other variables used in the regression are based on changes for the corresponding year. size represents the log of total sales of

the firm. quick is the ratio of cash and short-term investments to current liabilities. rnd stands for research and development expenses

scaled by the sales of the firm. fsale represents foreign sales as a percentage of total sales. Models 1 and 2 are logistic estimations (slope

coefficients evaluated at mean are reported in the table), whereas Models 3 and 4 are OLS (a linear probability model) estimates. Number

of observations is provided at the end of the table. All standard errors are clustered at firm level to account for correlation in error terms of

same firms across multiple years.

Estimate t-Value Estimate t-Value Estimate t-Value Estimate t-Value

Logit OLS

Dsize 0.5618 (0.75) 0.4991 (0.76) 0.4124 (0.68) 0.2502 (0.48)

Dlev 1.0473 (1.84) 1.3793 (2.14) 0.8890 (1.81) 0.9642 (1.92)

Dlev2 �12.0573 (�2.31) �15.5853 (�2.72) �9.5736 (�3.01) �8.9984 (�2.69)

Dlaglev 1.2980 (2.40) 0.8826 (2.34)

Dquick 0.1807 (0.75) 0.1849 (0.78) 0.0554 (1.03) 0.0444 (0.91)

Drnd 0.5362 (0.16) 0.1901 (0.06) 0.7597 (0.20) 0.3883 (0.11)

Dfsale �0.0314 (�0.14) 0.1794 (0.64) �0.0532 (�0.26) 0.0339 (0.16)

N 81 81 81 81

A. Purnanandam / Journal of Financial Economics 87 (2008) 706–739732

I conduct logit as well as OLS regressions (linear probability model) to estimate the impact of changes inleverage on changes in derivatives holding. The following model is estimated:

Dhedgej;t ¼ a0 þ a1Dlevj;t þ a2Dlev2j;t þX

aControlj;t þ �j;t. (9)

The dependent variable takes a value of one for an increase in hedging and zero for a decrease; Dlevj;t

measures the change in leverage of firm j in year t; ½DðlevÞj;t�2 is the squared change in leverage for firms with an

increase in leverage, zero otherwise, and all other control variables in the model are also first differenced. Iinclude all control variables in this model that enter the earlier regression except for variables that are unlikelyto change much on a yearly basis, namely the industry concentration ratio and institutional shareholdings.Including these variables in the model does not change any results.

The results are provided in Table 8. In the first (logit) and third (OLS) model, I find that firms with amoderate increase in leverage are more likely to increase their hedging positions as evident by a positive andsignificant coefficient on Dlev. In contrast, firms with a very high increase in leverage are more likely todecrease their hedging intensities. Though the statistical significance of the coefficient on Dlev is weaker ascompared to the cross-sectional case, it remains significant at the 7% level. The coefficient on Dlev2 on theother hand remains statistically strong (at 2%) as in the cross-sectional case. In fact changes in leverage remainthe most significant determinant of changes in hedging intensities as compared to other covariates that enterthis model.

Ex-ante vs. ex-post incentives: Due to data limitations, my base model is estimated with cross-sectional data.At any given point in time, an empiricist observes a firm’s hedging position, which is a mixture of both ex-anteand ex-post actions (i.e., hedging decisions taken before/together with debt issuance and those taken after debtissuance). Ex-ante theory predicts a positive association between leverage and hedging, whereas ex-post theorypredicts a nonmonotonic relation. In cross-sectional data, therefore, ex-ante decisions bias my study againstfinding a nonmonotonic relation. This happens because, if all decisions are taken ex-ante, then the hedgingmotivations should be strongly positively associated with leverage even at very high levels of leverage, makingthe task of finding a negative relation between the two variables harder. Though change regressions lead to animprovement over the cross-sectional model, it is still possible that changes in leverage and hedging occur atthe same time in the spirit of ex-ante hedging theory. To analyze this issue further, I regress Dhedgej;t on Dlevj;t

Page 28: Don't Know

ARTICLE IN PRESSA. Purnanandam / Journal of Financial Economics 87 (2008) 706–739 733

along with Dlevj;t�1 (i.e., the lagged value of leverage change) and other control variables used in the earlierregression.

Thus, I regress innovations in hedging intensities on contemporaneous as well as lagged changes in leverage.While the contemporaneous change in leverage contains a mix of ex-ante and ex-post decisions, the coefficienton the lag change can be reasonably attributed to the hedging decisions consequent to debt issuance. Giventhat we do not observe reporting of hedging at a high frequency, establishing a link between past leveragechange and current change in derivatives based on annual data is a challenging task. Still, the results fromcolumns 2 and 4 in Table 8 show that the past year’s leverage change is a significant predictor of the currentyear’s hedging activities. The coefficient on the lagged value of leverage change (i.e., Dlaglev) is positive andsignificant at the 2% level. This is encouraging since, due to the sequence of decision making (i.e., last year’sleverage decision and this year’s hedging), this specification is very likely to detect causation between thehedging and leverage variable.

5. Conclusion

This paper develops a theory of corporate risk management in the presence of financial distress costs. Bydistinguishing ‘‘financial distress’’ from ‘‘insolvency’’, I provide a justification for the ex-post risk-management behavior of the firm. Due to financial distress costs, the shareholders engage in ex-post risk-management activities even without a pre-commitment to do so. The theory is based on a trade-off betweenshareholders’ risk-shifting incentives due to equity’s limited liability and their risk-avoidance incentives due tofinancial distress costs. I obtain a closed-form solution for the optimal level of investment risk based on thistrade-off. The model generates several testable predictions. It predicts a nonmonotonic relation betweenleverage and hedging and a U-shaped relation between financial distress costs and hedging. Financiallydistressed firms in highly concentrated industries are predicted to have higher hedging incentives.

I test the key predictions of my model with one of the most comprehensive samples used in the literature. Imodel a firm’s leverage and hedging in an endogenous framework using a sample of more than 2,000nonfinancial firms. I find evidence in support of a positive relation between leverage and foreign currency andcommodity hedging. Consistent with the theory, this relation becomes negative for firms with very highleverage. Financially distressed firms in highly concentrated industries hedge more. Finally, I show that thekey results remain similar for a dynamic analysis based on a change regression for a smaller subset of firms.

Appendix A

In the theoretical model of the paper I consider a continuous trading economy with a time horizon ½t0;T �and filtered probability space ðO; ð tÞ; ;PÞ satisfying the usual regularity conditions. I assume a complete andarbitrage-free market. This guarantees the existence of an equivalent martingale measure Q. I assume adeterministic short interest rate process given by r. In what follows, I denote the indicator function of an eventX by 1fXg. I assume that the unlevered asset value of the firm can be expressed as a Q-Brownian Motion (undera martingale measure) as follows:

dAt ¼ rAt dtþ sAt dW t.

A.1. Proof of Proposition 1

Proof. Shareholders’ payoff (CF) on the terminal date ðTÞ is given by

CF T ¼ ðV T � LÞ1fVT4L;mT4Kg þ ðf ðVT Þ � LÞ1ff ðVT Þ4L;mTpKg

¼ ðV T � LÞð1fVT4Lg � 1fVT4L;mTpKgÞ þ ðf ðV T Þ � LÞ1ff ðVT Þ4L;mTpKg

¼ ðV T � LÞ1fVT4Lg � ðVT � LÞ1fVT4L;mTpKg þ ðf ðV T Þ � LÞ1ff ðVT Þ4L;mTpKg

Page 29: Don't Know

ARTICLE IN PRESSA. Purnanandam / Journal of Financial Economics 87 (2008) 706–739734

¼ ðV T � LÞ1fVT4Lg � ðV T � LÞ½1ff ðVT Þ4L;mTpKg þ 1ff�1ðLÞ4VT4L;mTpKg

þ ðf ðV T Þ � LÞ1ff ðVT Þ4L;mTpKg

¼ ðV T � LÞ1fVT4Lg � ðV T � f ðVT Þ1ff ðVT Þ4L;mTpKg þ ðL� VT Þ1ff�1ðLÞ4VT4L;mTpKg

¼ ðV T � LÞ � ðV T � f ðV T Þ1ff ðVT Þ4L;mTpKg þ ðL� VT Þf1fVTpLg þ 1ff�1ðLÞ4VT4L;mTpKg

g.

Under mild technical restrictions, the equity value at t ¼ t1 ðxt1Þ is simply the expectation of

this payoff under the martingale measure. Taking the expectation of the terminal payoff gives the desiredresult. &

A.2. Equity valuation

As shown in expression 3 in the paper, the equity valuation is given by the following:

Et1¼ EQ½ðAT � LÞf1fAT4L;mT4Kg þ 1fAT4L;ATpLþM;mTpKgg þM1fAT4LþM ;mTpKg�

¼ EQ½ðAT � LÞf1fAT4L;mT4Kg þ 1fATpLþM ;mTpKg � 1fATpL;mTpKgg þM1fAT4LþM ;mTpKg�

¼ EQ½ðAT � LÞf1fAT4Lg � 1fAT4LþMg þ 1fAT4LþM;mT4Kgg þM1fAT4LþM;mTpKg�.

ðA:1Þ

The first two components of the equity value, namely, EQ½ðAT � LÞ1fAT4Lg� and EQ½ðAT � LÞ1fAT4LþMg�,can be computed using the standard Black-Scholes formula for the valuation of European call options. Let Fand f stand for the normal cumulative density function (cdf) and probability density function (pdf),respectively. For notational simplicity I set At1

¼ A0 and T 0 ¼ T � t1. Then the first two terms result in thefollowing expression:

EQ½ðAT � LÞ1fAT4Lg� ¼ A0Fðh1Þ � LFðh2Þ and

EQ½ðAT � LÞ1fAT4LþMg� ¼ A0Fðd1Þ � LFðd2Þ,

where

h1 ¼

lnA0

L

� �þ

s2

2T 0

sffiffiffiffiffiT 0p and h2 ¼ h1 � s

ffiffiffiffiffiT 0p

,

d1 ¼

lnA0

LþM

� �þ

s2

2T 0

sffiffiffiffiffiT 0p and d2 ¼ d1 � s

ffiffiffiffiffiT 0p

.

The last two terms, i.e., EQ½ðAT � LÞ1fAT4LþM;mT4Kg� and EQ½M1fAT4LþM ;mTpKg�, require the knowledgeof the joint distribution of the running minima and the terminal value of the geometric Brownian motion.Such distributions have been widely used for the pricing of path-dependent options. I use the following lemma(see Harrison, 1985 or Musiela and Rutkowski, 1998) to obtain the expression for the valuation of these twopath-dependent expressions:

Lemma. Let KoL and KoA0, then the joint density of the terminal asset value ðAT Þ and the running

minima of the geometric Brownian motion ðmT Þ, under the martingale measure, is provided by the following

formula:

QðAT4L;mTXKÞ ¼ Fln

A0

L

� ��

s2

2T 0

sffiffiffiffiTp 0

0BB@

1CCA� A0

K

� �F

lnK2

A0L

� ��

s2

2T 0

sffiffiffiffiffiT 0p

0BB@

1CCA. (A.2)

Page 30: Don't Know

ARTICLE IN PRESSA. Purnanandam / Journal of Financial Economics 87 (2008) 706–739 735

Using (A.2), I compute the expectation of the last two terms of expression (A.1) as follows:

EQ½ðAT � LÞ1fAT4LþM ;mT4KgÞ� ¼ fA0Fðd1Þ � LFðd2Þg � KFðc1Þ �A0L

KFðc2Þ

� �,

EQ½M1fAT4LþM;mTpKgÞ� ¼A0M

KFðc2Þ,

where d1 and d2 are as given before and

c1 ¼

lnK2

A0ðLþMÞ

� �þ

s2

2T 0

sffiffiffiffiffiT 0p and c2 ¼ c1 � s

ffiffiffiffiffiT 0p

.

Collecting the above results and simplifying the expressions further, I get the following expression for thevaluation of the firm’s equity at t ¼ t1:

Et1¼ fA0Fðh1Þ � LFðh2Þg � KFðc1Þ �

A0ðLþMÞ

KFðc2Þ

� �. (A.3)

A.3. Proof of Proposition 2

At t ¼ t1; the shareholders choose an optimal risk level such that it maximizes the equity value given inexpression (A.3). I am assuming that the firm is not in financial distress at t ¼ t1, i.e., KoA0. I also assumethat the distress barrier is below the face value of debt, i.e., KoL. At the optimum:

qEt1

qs¼ 0.

Differentiating expression (A.3) gives the following:

qEt1

qs¼ A0fðh1Þ

qh1

qs� Lfðh2Þ

qh2

qs

� �� Kfðc1Þ

qc1

qs�

A0L

Kfðc2Þ

qc2

qs

� �þ

A0M

Kfðc2Þ

qc2

qs. (A.4)

Note that:

fðh2Þ ¼A0

Lfðh1Þ,

fðc2Þ ¼K2

A0ðLþMÞfðc1Þ. ðA:5Þ

Eqs. (A.4) and (A.5) lead to

qEt1

qs¼ A0fðh1Þ

qh1

qs� A0fðh1Þ

qh1

qs�

ffiffiffiffiffiT 0p

� �� Kfðc1Þ

qc1

qs�

KL

ðLþMÞfðc1Þ

qc1

qs�

ffiffiffiffiffiT 0p

� �� ��

þKM

ðLþMÞfðc1Þ

qc1

qs�

ffiffiffiffiffiT 0p

� �.

Simplification leads to the following:

qEt1

qs¼ A0fðh1Þ

ffiffiffiffiffiT 0p� fðc1Þ K

qc1

qs�

KL

ðLþMÞ

qc1

qs�

ffiffiffiffiffiT 0p

� �� ��

KM

ðLþMÞ

qc1

qs�

ffiffiffiffiffiT 0p

� �� �.

Thus, I have the following first-order condition for the optimal investment risk of the firm,

A0fðh1ÞffiffiffiffiffiT 0p� fðc1ÞK

ffiffiffiffiffiT 0p¼ 0. (A.6)

Page 31: Don't Know

ARTICLE IN PRESSA. Purnanandam / Journal of Financial Economics 87 (2008) 706–739736

Thus,

A0fðh1Þ � Kfðc1Þ ¼ 0. (A.7)

A.4. Second-order condition

Differentiating expression (A.6) gives the second-order optimality condition:

q2Et1

qs2¼ �h1A0fðh1Þ

ffiffiffiffiffiT 0p qh1

qsþ c1fðc1ÞK

ffiffiffiffiffiT 0p qc1

qs. (A.8)

Using the first-order condition and simplifying, the above expression, at the optimum, reduces to

q2Et1

qs2¼ � h1fðc1ÞK

ffiffiffiffiffiT 0p qh1

qsþ c1fðc1ÞK

ffiffiffiffiffiT 0p qc1

qs

¼ fðc1ÞKffiffiffiffiffiT 0p

�h1

ffiffiffiffiffiT 0p�

h1

s

� �þ c1

ffiffiffiffiffiT 0p�

c1

s

� �

¼fðc1ÞK

ffiffiffiffiffiT 0pðh1 � c1Þðh1 þ c1 � s

ffiffiffiffiffiT 0pÞ

s.

Using the following equalities,

ðh1 � c1Þ ¼

lnA2

0ðLþMÞ

K2L

� �sffiffiffiffiffiT 0p 40 and ðh1 þ c1 � s

ffiffiffiffiffiT 0pÞ ¼

lnK2

LðLþMÞ

� �sffiffiffiffiffiT 0p o0,

it follows that

q2Et1

qs2o0. (A.9)

Thus, the second-order condition for the maximization problem is satisfied.

A.5. Comparative statistics

Comparative statistics are obtained by a direct differentiation of the optimal solution for s given inexpression 5.

(a)

Sensitivity with respect to default boundary (K):

qðs2Þ�

qK¼

2

T 0K lnLþM

L

� � lnK2

LðLþMÞ

� �þ ln

K2L

A2t1ðLþMÞ

!( )o0.

The inequality follows from the facts that KoL;KoAt1and M40.

(b)

Sensitivity with respect to time-to-maturity ðT 0 ¼ T � t1Þ:

qðs2Þ�

qT 0¼ �ðs2Þ�

T 0o0.

(c)

Sensitivity with respect to the deadweight loss parameter M: Direct differentiation of the optimalinvestment risk leads to the following:

qðs2Þ�

qM40 if M4Lexp

2ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffilnðAt1

=KÞ lnðL=KÞp

� L.

Page 32: Don't Know

ARTICLE IN PRESSA. Purnanandam / Journal of Financial Economics 87 (2008) 706–739 737

Sensitivity with respect to leverage: Let X ¼ lnðK2=LðLþMÞÞ, Y ¼ lnðK2L=A2t ðLþMÞÞ and Z ¼ ln

(d) 1

ððLþMÞ=LÞ. Denote leverage by lev ¼ L=At1. After some algebra it can be shown that

qs2T

ql¼

2

K

qK

ql�

qM

ql

1

LþM� A

1

1

LþM

� �� �Y

2

K

qK

ql�

qM

ql

1

LþMþ A

3

L�

1

LþM

� �� �

�X

Z�

qM

ql

1

LþM�

AM

LðLþMÞ

� �XY

Z2.

In this model, leverage affects investment risk via its affect on distress boundary ðKÞ and deadweight lossparameter ðMÞ. Consider K ¼ 1� exp�0:1

�lev and M ¼ 7� exp2�lev. This specification corresponds to a

concave distress boundary. A highly levered firm faces higher distress boundary, meaning such firms are morelikely to get into distress and become the target of predatory behavior of their rivals even for relatively smalladditional downturn in their financial health. The specification of M corresponds to a model in which higherleverage imposes higher costs on the firm again due to reasons such as lost customers to rivals. The model hasbeen solved for different values of leverage using the analytically derived formula for s. For this specification,I set debt to one and vary the asset value to obtain different levels of leverage in the model. T has been set to 1.The results are plotted in Fig. 4 and show the nonmonotonic relation between leverage and investment risk.I also use several other parametric specifications on K and M and obtain similar results.

A.6. Tax-convexity measure

I use the methodology suggested by Graham and Smith (1999) to measure the tax-convexity incentive ofhedging. Using the simulation methods and considering the various features of tax codes, they compute theexpected tax-benefits that would result from a 5% reduction in income volatility. Subsequently they perform aregression analysis that relates tax savings to the following explanatory variables: (i) an indicator variableidentifying taxable income between �$500; 000 and zero (TI(NEG)), (ii) an indicator variable identifyingtaxable income between zero and $500,000 (TI(POS)), (iii) income volatility measured as the absolutecoefficient of variation (VOL), (iv) first-order serial correlation in income (RHO), (v) a dummy variableindicating the existence of investment tax credit (ITC), (vi) a dummy variable indicating the existence of NetOperating Loss (NOL) carryforwards, and finally (vii) NOL dummy interacted with the small-loss(NOL�TI(NEG)) and small-gain (NOL�TI(POS)) indicator variables. The regression estimate is given asfollows:

Tax�convexity ¼ 4:88þ 7:15 � TIðNEGÞ þ 1:60 � TIðPOSÞ þ 0:019 � VOL� 5:50 � RHO

� 1:28 � ITC þNOL � ð3:29� 4:77 � TIðNEGÞ � 1:93 � TIðPOSÞÞ.

I obtain the predicted tax savings in dollars for each firm in the sample by using the above equation. This isscaled by the sales of the firm to get the tax-convexity measure.

References

Acharya, V., Almeida, H., Campello, M., 2004. Is cash negative debt? A hedging perspective on corporate financial policies. Working

paper, London Business School.

Adam, T., Dasgupta, S., Titman, S., 2004. Financial constraints, competition and hedging in industry equilibrium. Journal of Finance,

forthcoming.

Allayannis, G., Ofek, E., 2001. Exchange rate exposure, hedging, and the use of foreign currency derivatives. Journal of International

Money and Finance 20, 273–296.

Allayannis, G., Weston, J., 2001. The use of foreign currency derivatives and firm market value. Review of Financial Studies 14, 243–276.

Allayannis, G., Ihrig, J., Weston, J., 2001. Exchange-rate exposure: financial vs. operating strategies. American Economic Review 91,

391–398.

Andrade, G., Kaplan, S., 1998. How costly is financial (not economic) distress? Evidence from highly leveraged transactions that became

distressed. Journal of Finance 53, 1443–1493.

Arping, S., 2000. Debt and product market fragility. Working paper, HEC, Lausanne.

Page 33: Don't Know

ARTICLE IN PRESSA. Purnanandam / Journal of Financial Economics 87 (2008) 706–739738

Asquith, P., Gertner, R., Scharfstein, D., 1994. Anatomy of financial distress: an examination of junk-bond issuers. Quarterly Journal of

Economics 109, 625–658.

Bartram, S.M., Brown, G.W., Fehle, F.R., 2003, International evidence on financial derivatives usage. Working paper, University of

North Carolina.

Bolton, P., Dewatripont, M., 2005. Contract Theory. MIT Press.

Bound, J., Jaeger, D.A., Baker, R.M., 1995. Problems with instrumental variables estimation when the correlation between the

instruments and the endogenous explanatory variable is weak. Journal of the American Statistical Association 90, 443–450.

Bradley, M., Jarrell, G.A., Kim, E.H., 1984. On the existence of an optimal capital structure: theory and evidence. Journal of Finance 39,

857–878.

Breeden, D., Viswanathan, S., 1996. Why do firms hedge? An asymmetric information model. Working paper, Duke University.

Brockman, P., Turtle, H.J., 2003. A barrier option framework for corporate security valuation. Journal of Financial Economics 67,

511–529.

Brown, G., 2001. Managing foreign exchange risk with derivatives. Journal of Financial Economics 60, 401–448.

Chevalier, J., 1995a. Capital structure and product market competition: empirical evidence from the supermarket industry. American

Economic Review 85, 415–435.

Chevalier, J., 1995b. Do LBO supermarkets charge more? An empirical analysis of the effects of LBOs on supermarket pricing. Journal of

Finance 50, 1112–1195.

DeMarzo, P.M., Duffie, D., 1991. Corporate financial hedging with proprietary information. Journal of Economic Theory 53, 261–286.

DeMarzo, P.M., Duffie, D., 1995. Corporate incentives for hedging and hedge accounting. Review of Financial Studies 8, 743–771.

Dichev, I.D., Skinner, D.J., 2001. Large-sample evidence on the bond covenants hypothesis. Working paper, University of Michigan.

Dolde, W., 1993. The trajectory of corporate financial risk management. Journal of Applied Corporate Finance 6, 33–41.

Fehle, F., Tsyplakov, S., 2005. Dynamic risk management: theory and evidence. Journal of Financial Economics 78, 3–47.

Frank, M.Z., Goyal, V.K., 2003. Capital structure decisions. Working paper, University of Minnesota.

Froot, K.A., Scharfstein, D.S., Stein, J.C., 1993. Risk management: coordinating corporate investments and financing policies. Journal of

Finance 5, 1629–1658.

Geczy, C., Minton, B.A., Schrand, C., 1997. Why firms use currency derivatives? Journal of Finance 52, 1323–1354.

Goldstein, R., Ju, N., Leland, H., 2001. An EBIT-based model of dynamic capital structure. Journal of Business 74, 483–512.

Graham, J.R., Lemmon, M.L., Schallheim, J.S., 1998. Debt, leases, taxes and the endogeneity of corporate tax status. Journal of Finance

53, 131–162.

Graham, J.R., Rogers, D.A., 2002. Do firms hedge in response to tax incentives? Journal of Finance 57, 815–839.

Graham, J.R., Smith, C.R., 1999. Tax incentives to hedge. Journal of Finance 54, 2241–2262.

Guay, W., 1999. The impact of derivatives on firm risk: an empirical examination of new derivatives users. Journal of Accounting and

Economics 26, 319–362.

Guay, W., Kothari, S.P., 2003. How much do firms hedge with derivatives? Journal of Financial Economics 70, 423–461.

Harrison, J.M., 1985. Brownian Motion and Stochastic Flow Systems. Wiley, New York.

Haushalter, D.G., 2000. Financing policy, basis risk, and corporate hedging: evidence from oil and gas producers. Journal of Finance 55,

107–152.

Hentschel, L., Kothari, S.P., 2001. Are corporations reducing or taking risks with derivatives? Journal of Financial and Quantitative

Analysis 36, 93–118.

Jensen, M.C., Meckling, W.H., 1976. Theory of the firm: managerial behavior, agency costs and ownership structure. Journal of Financial

Economics 3, 305–360.

Jorion, P., 1991. The pricing of exchange rate risk in the stock market. Journal of Financial and Quantitative Analysis 363–376.

Kalay, A., 1982. Stockholder-bondholder conflict and dividend constraints. Journal of Financial Economics 10, 211–233.

Kovenock, D., Phillips, G., 1997. Capital structure and product market behavior: an examination of plant exit and investment decisions.

Review of Financial Studies 10, 767–803.

Lang, L., Ofek, E., Stulz, R., 1996. Leverage, investment and firm growth. Journal of Financial Economics 40, 3–29.

Leland, H.E., 1998. Agency costs, risk management, and capital structure. Journal of Finance 53, 1213–1243.

Maddala, G.S., 1983. Limited dependent and qualitative variables in econometrics. Econometric Society Monographs. Cambridge

University Press, Cambridge.

Mian, S.L., 1996. Evidence on corporate hedging policies. Journal of Financial and Quantitative Analysis 31, 419–439.

Moody’s Investors Service Report, 1998. Historical Default Rates of Corporate Bond Issuers, 1920–1997.

Morellec, E., Smith, C.W., 2003. Investment policy, financial policies, and the control of agency conflicts. Working paper, University of

Rochester.

Mozumdar, A., 2001. Corporate hedging and speculative incentives: implications for swap market default risk. Journal of Financial and

Quantitative Analysis 36, 221–250.

Musiela, M., Rutkowski, M., 1998. Martingale Methods in Financial Modelling. Springer, Berlin.

Nain, A., 2006. Corporate risk management in an industry setting: an empirical investigation. Working paper, McGill University.

Nance, D.R., Smith, C.W., Smithson, C.W., 1993. On the determinants of corporate hedging. Journal of Finance 48, 267–284.

Opler, T., Titman, S., 1994. Financial distress and corporate performance. Journal of Finance 49, 1015–1040.

Petersen, M., Thiagarajan, S.R., 2000. Risk measurement and hedging: with and without derivatives. Financial Management 29,

5–30.

Phillips, G., 1995. Increased debt and industry product markets: an empirical analysis. Journal of Financial Economics 37, 189–238.

Page 34: Don't Know

ARTICLE IN PRESSA. Purnanandam / Journal of Financial Economics 87 (2008) 706–739 739

Purnanandam, A., 2007. Interest rate derivatives at commercial banks: an empirical investigation. Journal of Monetary Economics 54,

1769–1808.

Smith, C.W., Stulz, R., 1985. The determinants of firms’ hedging policies. Journal of Financial and Quantitative Analysis 28, 391–405.

Smith, C.W., Warner, J.B., 1979. On financial contracting: an analysis of bond covenants. Journal of Financial Economics 7, 117–161.

Staiger, D., Stock, J.H., 1997. Instrumental variables regression with weak instruments. Econometrica 65, 557–586.

Stulz, R., 1984. Optimal hedging policies. Journal of Financial and Quantitative Analysis 19, 127–140.

Stulz, R., 1996. Rethinking risk management. Journal of Applied Corporate Finance 9, 8–24.

Titman, S., 1984. The effect of capital structure on a firm’s liquidation decision. Journal of Financial Economics 13, 137–151.

Titman, S., Wessels, R., 1988. The determinants of capital structure choices. Journal of Finance 43, 1–19.

Tufano, P., 1996. Who manages risk? An empirical examination of risk management practices in the gold mining industry. Journal of

Finance 51, 1097–1137.

Wooldridge, J.M., 2002. Econometric Analysis of Cross Section and Panel Data. MIT Press, Cambridge, MA.

Zingales, L., 1998. The survival of the fittest or the fattest: exit and financing in the trucking industry. Journal of Finance 53, 905–938.