Don't Know

Journal of Financial Economics 87 (2008) 706739
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 nancial distress
costs and tests the models 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 rms. Consistent with the
theory, empirically I nd a non-monotonic relation between leverage and hedging. Further, the effect of leverage on
hedging is higher for rms in highly concentrated industries.
r 2008 Elsevier B.V. All rights reserved.
JEL classication: 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 nancial distress
costs. The existing literature shows that hedging can lead to rm 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.jneco.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 OHara, 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 Associations 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: amiyatos@umich.edu
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 Dufe, 1991, 1995). See also
Breeden and Viswanathan (1996) and Stulz (1996).
behavior on the part of the rm; ex-post, shareholders of a levered rm may not nd it optimal to engage in
hedging activities due to their risk-shifting incentives (Jensen and Meckling, 1976).
2
I extend the current
literature by explaining the ex-post risk-management motivation of the rm.
3
I provide a simple model that
generates new cross-sectional predictions by relating rm characteristics such as leverage, nancial distress
costs, and project maturity to risk-management incentives. I test the key predictions of the model with hedging
data of COMPUSTAT-CRSP rms meeting some reasonable sample selection criteria for scal years
19961997. The empirical study presents the rst large-sample evidence on the determinants of the extent of
rms hedging activities and provides new ndings.
The key assumption underlying my theory is the distinction between nancial distress and insolvency.
I assume that apart from the solvent and the insolvent states, a rm faces an intermediate state called nancial
distress. Financial Distress is dened as a low cash-ow state in which the rm incurs losses without being
insolvent. The notion that nancial distress is a different state from insolvency has some precedence in the
literature. Titman (1984) uses a similar assumption to study the effect of capital structure on a rms
liquidation decisions.
There are three important sources of nancial distress costs. First, a nancially distressed rm may lose
customers, valuable suppliers, and key employees.
4
Opler and Titman (1994) provide empirical evidence that
nancially distressed rms lose signicant market share to their healthy counterparts in industry downturns.
Using data from the supermarket industry, (Chevalier 1995a, b) nds evidence that debt weakens the
competitive position of a rm. Second, a nancially distressed rm is more likely to violate its debt covenants
5
or miss coupon/principal payments without being insolvent.
6
These violations impose deadweight losses in the
form of nancial penalties, accelerated debt repayment, operational inexibility, and managerial time and
resources spent on negotiations with the lenders.
7
Finally, a nancially distressed rm may have to forgo
positive NPV projects due to costly external nancing, as in Froot, Scharfstein, and Stein (1993). In this paper
I focus on the rst of these costs, i.e., the product market-related costs of nancial distress.
I develop a dynamic model of a rm that issues equity capital and zero-coupon bonds to invest in a risky
asset. The rm makes an initial investment with the consent of its bondholders. At a later date, shareholders
can modify the rms investment risk by replacing the existing asset with a new one. The rms asset value
evolves according to a stochastic process. The rm is in nancial distress if the asset value falls below some
lower threshold during its life. In this state, the rm loses market share to its competitors and therefore is
unable to realize its full upside potential, even when the industry condition improves at a later date. Insolvency
occurs on the maturity date if terminal rm value is below the face value of debt, in which case debtholders
gain control of the rm. Shareholders nal payoffs depend on the terminal asset value as well as on the path
taken by the rms asset over its life.
8
ARTICLE IN PRESS
2
Throughout the paper, I use the terms ex ante and ex post with respect to the time of borrowing.
3
Other papers analyzing shareholders ex-post risk-management decisions include Leland (1998) and Morellec and Smith (2003). Leland
(1998) provides a justication for the rms ex-post hedging behavior in the presence of tax-benets of debt. In Morellec and Smith (2003),
the manager-shareholder conict reduces shareholders ex-post asset-substitution incentives. My model, in contrast, is based on the cost of
nancial distress and provides new empirical predictions.
4
For example, in the mid-1990s Apple Computers had nancial difculties 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 Apples 1998 10-K lings with the SEC).
Similarly, when Chrysler faced nancial difculties 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).
5
Lenders often impose debt covenants such as maintenance of minimum networth or maximum debt-to-equity ratio by the borrowing
rms. See Smith and Warner (1979), Kalay (1982), and Dichev and Skinner (2001).
6
Moodys Investor Service Report (1998) shows that during 19821997 about 50% of the long-term publicly traded bond defaults
(including missed or delayed payment of coupon and principal) didnt result in bankruptcy lings.
7
For 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 Deltas 2002 10-K lings with
the SEC.
8
This 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 nancial distress. Brockman and Turtle (2003)
provide some empirical evidence in support of equitys valuation as a path-dependent option.
A. Purnanandam / Journal of Financial Economics 87 (2008) 706739 707
The optimal level of ex-post investment risk, from the shareholders perspective, is determined by the trade-
off between the costs of nancial distress and value associated with the limited liability of the rms equity.
9
Unlike in the risk-shifting models such as Jensen and Meckling (1976), equity value is not always an increasing
function of rm risk in my model. While a high risk project increases the value of equitys limited liability, it
also imposes a cost on shareholders by increasing the expected cost of nancial distress. Due to these losses,
the shareholders nd it optimal to implement a risk-management strategy ex-post even in the absence of an
explicit pre-commitment to do so.
The optimal investment risk in my model depends on rm leverage, the nancial distress boundary, the time
horizon of the project, and the costs of nancial distress. As in the extant models (Smith and Stulz, 1985),
I show that a rm with high leverage has a higher incentive to engage in hedging activities. However, the risk-
management incentives disappear for rms with extremely high leverage. The incentive to hedge arises from
the product market-related nancial distress costs and these costs are more likely to be present when a rm is
vulnerable to losing market share to its competitors. Empirical studies by Opler and Titman (1994) and
Chevalier (1995a, b) show that debt weakens the competitive position of a rm in its industry. Further, the
adverse consequences of leverage are more pronounced in concentrated industries. Motivated by these studies
my model argues that industry concentration provides a good proxy for nancial distress costs. Highly
leveraged rms in concentrated industries are more likely to experience a deterioration in their competitive
position in the event of nancial distress i.e., are expected to incur higher nancial distress costs. Thus, the
model predicts a stronger hedging incentive for highly levered rms in concentrated industries.
The model shows that hedging incentives increase with project maturity because the likelihood of
experiencing nancial 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 rm in states in which the rm hits the
nancial distress barrier but remains solvent on the maturity date. If there are no nancial distress costs, risk-
management incentives disappear. On the other hand, if these costs are very high, the distinction between
nancial distress and insolvency diminishes along with any ex-post risk-management motivations.
Intermediate levels of losses create risk-management incentives within the rm. Therefore, my model predicts
a U-shaped relation between nancial distress costs and hedging.
The predictions of my model have important implications for the empirical research. To test the existing
theories, empirical studies regress some measure of nancial distress (typically leverage) on rms risk-
management activities. If rms with extreme distress are less likely to hedge, these models may be misspecied.
The bias can be particularly severe in small-sample studies. It is not surprising that existing empirical studies
nd 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 nonnancial rms. Since data on rms hedging activities
(by means of derivatives) are not readily available, empirical studies in this area are based on small samples or
investigate only the yesno decision to hedge.
11
This has created two major challenges. First, our current
understanding is mostly based on analyses that treat rms with different hedging intensities as similar, which
limits our ability to investigate rms hedging motivations. Second, we have been able to gain only limited
insight into the effect of industry-specic factors on hedging decisions.
I test the predictions of my model with data on the extent of hedging of more than 2,000 rms for the scal
year 19961997. Due to the large sample size drawn from different industries, I provide new empirical evidence
relating industry structure to hedging decisions. Consistent with the theory, I nd strong evidence that rms
with higher leverage hedge more, although the hedging incentives disappear for rms with very high leverage.
Also in line with my theory, I nd that nancially distressed rms in highly concentrated industries hedge
ARTICLE IN PRESS
9
In 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 rm type. His model relates hedging incentives to rm type.
10
For example, while Haushalter (2000) and Graham and Rogers (2002) nd a positive relation between the two variables, Nance, Smith
and Smithson (1993), Mian (1996), and Tufano (1996) fail to nd such evidence.
11
For example, Geczy, Minton, and Schrand (1997) use 372 rms with 154 hedgers; Graham and Rogers (2002) use about 400 rms with
158 hedgers. Studies by Mian (1996) and Bartram, Brown, and Fehle (2003) use large samples to investigate the yesno 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.
more. My empirical results are robust to alternative proxies of nancial distress (such as leverage, industry-
adjusted leverage and Altman Z-score), alternative ways of measuring the hedging activities (yesno decision
to hedge and total notional amount of hedging) and various controls for nonderivative-based hedging
strategies. Further for a subsample of 200 manufacturing rms, I obtain data on the rms hedging activities
for scal years 19971998 and 19981999 and show that the basic results remain similar for a regression model
involving changes in hedging activities. While rms with a moderate increase in leverage increase their hedging
activities, rms with an extreme increase in leverage decrease their hedging positions. As long as rms do not
frequently change their operational hedging strategies (such as opening plants in foreign countries to hedge
their foreign currency risk), the analysis based on change regressions provides a robust control for
nonderivative-based hedging strategies of the rm. The change regressions also allow me to partially
disentangle 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 3
analyzes the optimal risk-management policy of the rm. The empirical tests are provided in Section 4, and
Section 5 concludes the paper. Without any loss in continuity, readers mostly interested in the empirical part
of the paper can skip to Section 3.1, which provides a self-contained summary of the key features of the
theoretical model.
2. Model
I consider a stylized model of a continuous trading economy with time horizon t
0
; T. There are three
important dates in the model discussed below. Though a discrete time model can also be used to capture the
key feature of my model, the continuous time version allows for an easier analytical solution at the expense of
additional mathematical overhead. In addition, the continuous time model provides additional prediction
relating the time to maturity of the rms project to its hedging incentives.
At t t
0
, the rm makes its capital structure decision and invests in risky asset A
i
(i stands for the initial
investment), which I refer to as an EBIT-generating machine (Goldstein, Ju and Leland, 2001). These
decisions may or may not be made with the consent of the rms debtholders. The risky asset A
i
is acquired
at the market-determined price and nanced through a mix of zero-coupon debt and equity capital. Let L be
the face value of the zero-coupon debt, payable at time T, and E
t
be the time t-value of the rms equity. There
is a tax benet of debt, which provides the incentive to issue debt in my model. For simplicity the tax benet is
assumed to be a fraction t of the face value of debt L. Optimal capital structure is determined by a trade-off
between the tax benet of debt and bankruptcy costs. For simplicity, I do not endogenize the capital structure
decisions. 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 A
i
t
are
driven by a Brownian motion with the usual properties.
At some later time t t
1
t
1
2 t
0
; T, 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 structure
decisions, they have an opportunity to change the assets risk without the bondholders approval. To capture
the risk-shifting incentives, I assume that the bondholders are unable to recontract with the shareholders at
t t
1
. Further, I assume that the two parties cannot contract on the risk-management choice at time t
0
through the use of bond covenants. This latter assumption is what gives rise to the risk-shifting incentive in my
model. This assumption is in the spirit of a large literature on incomplete contracting in economics and nance
(see for example, Bolton and Dewatripont, 2005). The premise here is that it is too costly to specify every state
of the world and write down debt covenants that will limit shareholders behavior with respect to rm risk in
each of those states. Even if such covenants could be written to tie down the managers risk-management
behavior, it would be too costly to implement them especially in very high leverage states when shareholders
have a large incentive to default on covenants.
12
This assumption is in the spirit of Jensen and Mecklings argument that To completely protect the
bondholders from the incentive effects, these provisions would have to be incredibly detailed and cover most
ARTICLE IN PRESS
12
As 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 sufcient to generate the main results of my model.
operating aspects of the enterprise including limitations on the riskiness of the projects undertaken. The costs
involved in writing such provisions, the costs of enforcing them and the reduced protability of the rm
(induced because the covenants occasionally limit managements ability to take optimal actions on certain
issues) would likely be nontrivial. In fact, since management is a continuous decision making process it will be
almost impossible to completely specify such conditions without having the bondholders actually perform the
management function.
After the risk-management decisions have been made, the rm acquires a new EBIT-generating machine.
This EBIT-generating machine generates cashows d
t
forever that evolves according to a geometric Brownian
motion. The value of this EBIT-generating machine, i.e., the value of a similar unlevered rm, is denoted by
A
t
.
13
One can think of d
t
as the state vector representing the state of the rms industry. I assume that the
change in the investment risk of the asset (from A
i
to A has no cashow impact on the rm at t t
1
. This
provides an initial boundary condition in the model, namely A
t
1
A
i
t
1
. Further, for analytical simplicity I
assume that the total payout (to debtholders and shareholders) by the rm is zero during t
0
; T, with the nal
payoffs realized at t T. The shareholders receive the terminal equity value of the rm.
14
The bondholders
receive the face value of debt (L) if the rm remains solvent on the maturity date t T
15
; otherwise they
receive the residual value of the rm. The model can be represented by the following timeline:
m m m
t t
0
t t
1
t 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 of
the rm in the presence of the shareholders risk-shifting incentives. I now discuss the main assumption of the
paper, namely, the distinction between nancial distress and insolvency.
2.1. Financial distress and insolvency
If during (t
0
; T the rms asset value A
t
falls below a boundary KL;
16
the rm is in the state of nancial
distress. Insolvency, on the other hand, occurs on the terminal date T if the terminal rm value V
T
) is less
than the debt obligations. Therefore, in the state of nancial distress, control of the rm does not shift to the
bondholders immediately, but the rm does incur costs that increase with leverage. Opler and Titman (1994)
show that nancially distressed (highly leveraged) rms lose signicant market share to their healthy
competitors during industry downturns. The drop in sales faced by Apple Computers and Chrysler during
periods of nancial difculty provide anecdotal evidence in support of such deadweight losses. In a sample of
31 high-leveraged transactions (HLTs), Andrade and Kaplan (1998) isolate the effect of economic distress
from nancial distress and estimate the cost of nancial distress as 1020% of rm value. Asquith, Gertner
and Scharfstein (1994) show that on average nancially distressed rms sell 12% of their assets as part of their
restructuring plans.
Chevalier (1995a, b) uses detailed information from the local supermarket industry to provide evidence in
support of predatory behavior in this market. She shows that following supermarket leveraged buyouts
ARTICLE IN PRESS
13
The value of the levered rm of my model differs from A
t
by the amount of the tax benet of debt as well as the costs associated with
nancial distress and bankruptcy. Throughout this paper I denote the value of the levered rm by V
t
and the value of its assets (EBIT-
generating machine) by A
t
.
14
For analytical simplicity I assume that the models terminal date corresponds to the maturity date of the rms debt, at t T. This
assumption should not be confused with the assumption that the rms life is nite. It simply states that at time T initial shareholders sell
the rm to some other investors at the fair market value of the rm as an ongoing concern.
15
Other maturity structures are possible. To illustrate the main results of the paper in its simplest form, I prefer to work with zero
coupon debts.
16
I refer to K as the distress barrier in the rest of this paper. K is assumed to be an increasing function of leverage. This denition of
nancial distress is equivalent to assuming that when industry conditions deteriorate, rms with high leverage become nancially
distressed.
(LBOs), prices fall in local markets in which rival rms have low leverage and are concentrated. Further, these
price drops are associated with LBO rms exiting the local market. These ndings suggest that rivals attempt
to prey on LBO chains. Phillips (1995) studies the interactions between product market and nancial structure
for four industries and nds evidence consistent with debt weakening the competitive positions of rms (see
also Kovenock and Phillips, 1997; Arping, 2000). Using deregulation of the trucking industry as an exogenous
shock, Zingales (1998) studies the interplay between nancial structure and product market competition and
provides evidence that leverage reduces the probability of a rms survival after an increase in competition.
The overall message from these papers is that nancial distress may impose a real cost on rms by weakening
their competitive position in the product market.
Motivated by the empirical ndings of above papers and anecdotal evidence, I assume that a rm in
nancial distress loses a fraction of its market share to its healthy competitors.
17
In my model, this is achieved
by assuming that the nancially distressed rms EBIT-generating machine produces less cashow resulting in
a lower value for the distressed rm. If the rm does not experience nancial distress during t 2 t
1
; T, the
terminal rm value is V
T
. However, if the distress boundary is hit, the terminal value falls to f V
T
, where
f V
T
oV
T
(see Fig. 1). The function f represents the losses caused by nancial distress.
2.2. Valuation of equity
The shareholders receive liquidating dividends at T. Due to equitys limited liability, the nal payoff to the
shareholders x
T
is zero if the terminal rm value is below L. Let us dene inf
t
1
ptpT
A
t
m
T
for
the minimum value of the asset during t
1
; T. In the event of no distress (i.e., m
T
4K) and solvency on the
terminal date (i.e., V
T
4L), the shareholders get a liquidating dividend of V
T
L. If nancial distress is
experienced (i.e., m
T
pK), but on the terminal date the rm remains solvent (i.e., f V
T
4L), the shareholders
ARTICLE IN PRESS
Time (t)
A
s
s
e
t

V
a
l
u
e
A
t1
f
-1
(L)
K
t
1 T
Financially
Distressed
Insolvent
Healthy
L
Fig. 1. This gure plots three paths for the evolution of the rms asset value. I assume zero tax shield of debt for this presentation. These
paths correspond to three states of the rm in my model. In the top-most path, the asset value never hits the nancial distress barrier (K).
This corresponds to the Healthy state. The middle path represents the state in which the distress barrier is hit (at time t, but the rm
remains solvent at time T. This is the state of Financial Distress. In this state the terminal rm value, net of deadweight losses (i.e.,
f A
T
), remains above the face value of debt (i.e., L). Thus, this is the state where f A
T
4L or alternatively A
T
4f
1
L, as depicted in the
gure. Finally, the bottom-most path corresponds to the state of Insolvency.
17
In a more general industry equilibrium setting, rms 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 rms
decision, taking industry structure as given.
receive liquidating dividends of f V
T
L. In the event of insolvency, shareholders receive nothing and rm
value drops by the fraction g 2 0; 1. The shareholders payoff under different states is summarized as
State at t T Corresponding rm values Payoff to shareholders
Healthy V
T
4L; m
T
4K V
T
L
Financial distress f V
T
4L; m
T
pK f V
T
L
Insolvency V
T
pL; m
T
4K 0
Insolvency f V
T
pL; m
T
pK 0
Proposition 1. Under mild technical conditions, the equity valuation at t t
1
is given by:
x
t
1
e
rT
E
Q
V
T
L V
T
f V
T
1
ff V
T
4L;m
T
pKg
L V
T
f1
fV
T
pLg
1
ff
1
L4V
T
4L;m
T
pKg
g.
(1)
Proof. See Appendix A.1. &
The equity value, as shown in Proposition 1, has three components. The rst term E
Q
V
T
L represents
the equity value without the distress costs and the limited liability feature. The second term E
Q
V
T

f V
T
1
ff V
T
4L;m
T
pKg
represents the cost of nancial distress. Because the shareholders of a nancially
distressed but solvent rm bear this cost, the equity value decreases by this amount. The risk avoidance
incentive results from this cost. The third term E
Q
L V
T
f1
fV
T
pLg
1
ff
1
L4V
T
4L;m
T
pKg
g represents the
savings enjoyed by the shareholders of a levered rm due to the limited liability feature of equity. This term
captures shareholders risk-shifting incentives. By increasing the assets risk, the shareholders can make
themselves better off by increasing the call option value (the third term). At the same time, however, the
expected loss in the event of nancial distress also increases with an increase in asset risk. The optimal level of
investment 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 be
explicit about the form of nancial distress cost that is borne by the shareholders of a nancially
distressed rm. In addition, I make some simplifying assumptions for analytical tractability. I assume that in
the event of distress (i.e., m
T
pK), the rms cashows drop to ld
t
; l 2 0; 1 and never reach beyond some
arbitrary upper bound Uo1 at time T, i.e., d
T
pU. Therefore, the losses take the form of lost upside
potential. This representation of nancial distress cost is motivated by existing empirical ndings and
anecdotal evidence, and captures the intuition that distressed rms lose cashows due to lost sales to
competitors. If industry conditions improve in the future, the distressed rms continue to feel the negative
effect of distress due to lost customers. This representation of distress is also consistent with the view that
when nancially distressed rms restructure themselves by selling assets (Asquith, Gertner and Scharfstein,
1994), their EBIT-generating machine produces lower contemporaneous cashows and in addition it limits
their ability to capitalize on very good industry conditions in the future. To concentrate on the effect of
nancial distress costs (as opposed to tax-motivated incentives of hedging as in Leland, 1998), in the rest of the
paper I set t 0.
18
Under this assumption and the assumption l 1, the distressed rms asset value can be
represented as
19
:
f A
T
A
T
if fd
T
pUg; and M
0
if fd
T
4Ug for some constant M
0
. (2)
ARTICLE IN PRESS
18
In unreported analysis, I solve the model with tax benets and obtain the rms optimal capital structure. However, to keep the focus
of this paper on risk-management decisions, I do not present these results in the paper. With tax benets, the rms payoffs increase by tL
without qualitatively changing the results of the analysis.
19
If lo1, then nancial distress costs are even higher and the results become stronger. This assumption is made only for analytical
simplicity.
Let us denote the asset value A
T
corresponding to d
T
U by L M. The shareholders liquidating
dividends are given as
States Payoff to shareholders Firm value
A
T
4L; m
T
4K A
T
L A
T
A
T
4L; A
T
pL M; m
T
pK A
T
L A
T
A
T
4L M; m
T
pK M L M
A
T
pL 0 gA
T
The nancial distress costs can be expressed as A
T
M:1
fA
T
4LM;m
T
pKg
. A higher value of M
corresponds to lower deadweight losses in the model. In line with Proposition 1, the equity value can be
expressed as follows:
x
t
1
e
rT
E
Q
A
T
L1
fA
T
4L;m
T
4Kg
A
T
L1
fA
T
4L;A
T
pLM;m
T
pKg
M1
fA
T
4LM;m
T
pKg
. 3
Fig. 2 plots the equity value as a function of the terminal asset value of the rm. As the diagram shows, the
equity value is not a strictly convex function of the underlying rm value as in the classical approach where
equity is valued as a call option on rm value. The deadweight loss of distress introduces a concavity in the
equity value, which results in risk-management incentives for the rm.
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 t
1
, the
shareholders make a decision about the optimal investment risk of the rm. There are two possibilities for
changing the investment risk: (a) the rm can directly choose an optimal level of s at t t
1
, or (b) the assets
risk, s, may be xed and the rm can alter its risk prole by buying derivative contracts such as futures and
options. I analyze the problem of nding optimal s assuming that investment risks can be costlessly modied.
Proposition 2. The shareholders have a well-founded incentive to engage in risk-management activities ex-post.
At t t
1
, the shareholders optimally choose a level of risk s
in the interior of all possible risks.

ARTICLE IN PRESS
E
q
u
i
t
y

V
a
l
u
e
L
L
L+M 0
Equity Value In
Healthy State
Equity Value in
Financial Distress
Equity Value
in my model
Asset Value at T
Fig. 2. This gure plots the equity value as a function of the terminal asset value of the rm. For illustrative purposes I set the tax rate to
zero and g 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.
Proof. As shown in Appendix A.2 and A.3, the optimum level of investment risk is obtained by the following
rst-order condition:
A
t
1
fh
1
Kfc
1
(4)
where h
1
lnA
t
1
=L s
2
=2T
0
=s
T
0
p
, h
2
h
1
s
T
0
p
, T
0
T t
1
, c
1
lnK
2
=A
t
1
L M s
2
=2T
0
=
s
T
0
p
, c
2
c
1
s
T
0
p
; and f stands for the probability density function of the standard normal distribution.
Further simplication leads to the following closed-form solution:
s
2

1
T
0
ln
K
2
LL M
_ _
ln
K
2
L
A
2
t
1
L M
_ _
ln
L M
L
_ _ : & (5)
As a result of the trade-off between the risk-shifting and risk-avoidance incentives, an interior solution for
the optimal risk is obtained in the model. This result differs from that of the earlier models. In risk-shifting
models 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 0. By
obtaining an interior solution for the optimal investment risk of the rm, my model provides insights into the
risk-management policies of the rm, as discussed below.
20
Proposition 3. The rm chooses a lower level of investment risk if (a) it faces a higher distress barrier (K), and (b) it
has a longer project maturity T
0
T t
1
. The relation between the deadweight losses and the optimal investment
risk is U-shaped. Let M
c
Lexp
2
lnA
t
1
=K lnL=K
_

L. When M4M
c
, 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). As
expected, a higher boundary increases the likelihood of nancial distress. Therefore, the shareholders optimally
choose a lower investment risk to avoid the nancial distress costs. The results show that the rm with a longer
operational horizon T
0
T t
1
nds it optimal to engage in increased risk-management activities. With longer
time-horizon, the probability of hitting the lower barrier increases. Further, consequent to entering the state of
distress expected losses increase with time to maturity because there is a higher probability of improvements in
industry conditions and the distressed rm will not be able to capitalize on these opportunities. There is
considerable empirical evidence that large rms hedge more than small rms. The pursuit of economies of scale has
been suggested as one possible explanation for this empirical regularity. My model suggests another explanation:
the time horizon of operations. If rms with longer time horizons grow larger over time, the researcher would nd
a positive association between risk-management activities and rm size at any given point in time.
Finally, I nd a U-shaped relation between the risk management incentives and the cost of nancial distress.
Recall that the deadweight losses in my model are parameterized by M (losses are given by
A
T
M:1
fA
T
4LM;m
T
pKg
). In the event of nancial distress, the rm loses its upside potential beyond
L M. 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 1), the shareholders lose nothing in the state of nancial distress
and hence there is no risk-management incentive. On the other hand, if deadweight losses are very high (i.e.,
M 0) the distinction between default and insolvency disappears along with the risk-management incentives.
21
Its the intermediate cases that generate risk-management incentives in the model. Fig. 3 illustrates this relation.
ARTICLE IN PRESS
20
With nonzero tax rates (in unreported analysis), the optimal s is even lower. The additional incentives for risk reduction, in the
presence of the tax-benet of debt, comes from the potential loss in the tax shield of debt for a bankrupt rm. This additional effect
generates ex-post hedging as in Leland (1998). See also Fehle and Tsyplakov (2005).
21
In 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 t
1
.
Leverage and risk management: To study the relation between leverage and risk management, I differentiate
the optimal s with respect to rm leverage at time 1 lev L=A. The details are provided in Appendix A.5.
After some simplication it can be shown that the optimal sigma decreases (i.e., risk-management incentives
increase) with an increase in leverage for a wide range of specications of the distress boundary and
ARTICLE IN PRESS
2.78
2.8
2.82
2.84
2.86
2.88
2.9
2.92
2.94
1
5
.
0
0
1
4
.
2
5
1
3
.
5
0
1
2
.
7
5
1
2
.
0
0
1
1
.
2
5
1
0
.
5
0
9
.
7
5
9
.
0
0
8
.
2
5
7
.
5
0
6
.
7
5
6
.
0
0
5
.
2
5
4
.
5
0
3
.
7
5
3
.
0
0
2
.
2
5
Deadweight Loss Parameter (M)
I
n
v
e
s
t
m
e
n
t

R
i
s
k
Fig. 3. This gure plots the optimal investment risk as a function of deadweight losses. The model has been calibrated with the following
parameter values: A
t
1
2; L 1; T
0
1 and K 0:5: On the x-axis, I plot the value of M. M measures the upside potential lost by
the rm in the event of nancial 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
.
1
1
0
.
1
5
0
.
1
9
0
.
2
3
0
.
2
7
0
.
3
1
0
.
3
5
0
.
3
9
0
.
4
3
0
.
4
7
0
.
5
1
0
.
5
5
0
.
5
9
0
.
6
3
0
.
6
7
0
.
7
1
0
.
7
5
0
.
7
9
0
.
8
3
0
.
8
7
0
.
9
1
0
.
9
5
Leverage
I
n
v
e
s
t
m
e
n
t

R
i
s
k
Fig. 4. This gure plots the optimal investment risk of the rm against the debt-asset ratio. For this graph I assume the following structure
on the distress boundary and deadweight losses: K 1 exp
0:1lev
and M 7 exp
2lev
. Amount of debt raised at time zero (L) if xed
at 1. lev equals L scaled by A
t
1
. T is set to one.
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 by
shareholders and thus they lose risk-management incentive. Using a parametric specication of K and M,
I solve for optimal risk as a function of leverage and report the results in Fig. 4. The relation is summarized
below:
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 of
leverage on hedging intensity is predicted to be higher for rms operating in industries with a higher incidence
of predatory behavior. The key friction underlying my model consists of the costs incurred by a rm after it
enters the state of nancial distress. These costs come in the form of lost customers and deterioration in the
competitive position of the rm 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 rm in concentrated industries. Thus, in
the context of my model industry concentration provides a good proxy for the nancial distress costs.
Accordingly, high leverage rms 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 basis
for the empirical tests to follow. In my stylized model, a rm starts with some mix of debt and equity at time
zero and buys a productive asset. At this time the capital structure of the rm is determined by trading off the
tax benet of debt against the expected nancial distress and bankruptcy costs. I do not solve for the optimal
leverage 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. In
unreported analyses, I solve for optimal leverage and as expected show that the debt ratio increases with the
tax benets and decreases with bankruptcy and nancial distress costs.
22
Given a level of debt determined at time t
0
, the rm experiences some random shocks to its value till t
1
,
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 t
1
, the asset value evolves according to
a stochastic process from time t
1
to T in the model. If the rms asset value breaches a lower threshold before
the terminal date T, then the rm enters nancial distress. Financial distress imposes costs on the rm such as
lost customers to the competitors, which in turn prohibits it from capitalizing on its full upside potential.
Motivated by the earlier empirical nding, I assume that highly levered rms lose more when they enter the
state of nancial distress.
After the distress boundary is hit, the rm 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 the
rm enters nancial distress but remains solvent at time T imposes a real cost on shareholders. In this state
they incur the nancial distress costs without being able to use their limited liability option. An increase in rm
risk increases the probability of nancial distress and the associated deadweight losses that are borne by the
shareholders, not the debtholders. On the other hand, by increasing rm risk they benet on account of the
usual limited liability feature. The optimal risk-management policy trades off these two incentives. For
moderate levels of leverage, the risk-management incentive dominates. But when leverage becomes too high at
ARTICLE IN PRESS
22
In a rational expectation framework rm value at time t
0
should be maximized keeping in mind the expected level of risk that will be
optimally undertaken by the shareholders at time t
1
. This expected sigma along with the tax benet of debt and bankruptcy costs will
determine the optimal amount of debt raised by the rm at time t
0
. Indeed the actual leverage at time t
1
will be different from the rationally
expected value of leverage, depending on the shocks experienced by the rm in the intervening period. Depending on the realizations of
these shocks in the interim period, the rms leverage at time t
1
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 doesnt 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.
time t
1
, the value associated with the call option feature of equity dominates the expected nancial distress
costs and shareholders nd it optimal not to engage in risk-management activities. Thus, the model predicts a
nonmonotonic relation between leverage and hedging. Further, the positive relation between leverage and
hedging is expected to be stronger for rms operating in industries with a higher incidence of predatory
behavior 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 rms hedging
decision are very limited. Second, leverage and hedging are likely to be determined jointly by rms, leading to
an endogeneity problem. Theories based on ex-ante incentives suggest that rms can increase their debt
capacity 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 timing
of debt issues and hedging decisions, which unfortunately are not available. Below I begin with discussing the
sample and data collection procedure followed by econometric strategies used to account for the endogeneity
problem. Empirical results follow these discussions. I then discuss the issues related to ex-ante vs. ex-post
incentives 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 a
large cross-section of rms during the scal years 1996 and 1997. I start with all rms in the intersection of the
CRSP and COMPUSTAT with 10-Ks available on the SEC website. I remove nancials and utilities since the
risk-management incentives of these rms are not necessarily comparable to other industrial rms. From this
sample, I exclude rms that fall in the last quartile of the size distribution based on total sales. Earlier
empirical studies and survey evidence suggest that such small rms are very unlikely to use derivative products
for hedging purposes (Dolde, 1993), arguably due to the lack of economies of scale.
For the remaining rms, I collect data on derivative usage from the 10-K lings. In the rst step I obtain all
available 10-K lings of rms in the intersection of COMPUSTAT and CRSP from the SEC for the calendar
year 1997.
23
I obtain data by searching the entire 10-K lings for the following text strings: risk
management, hedg, derivative, and swap. If a reference is made to any of these key words, I read the
surrounding text to obtain data on foreign currency and commodity derivatives. I obtain data on the notional
amount of foreign currency derivatives used for hedging purposes across various derivative instruments such
as swaps, forwards, futures, and options.
24
For commodity hedging I only obtain data on whether a rm uses
derivatives for hedging or not, since the reporting requirement for commodity derivatives doesnt allow for an
easy quantication in terms of dollar value. If there are no references to the key words, the rm is classied as
a nonhedger. I require that data on net sales, leverage and market capitalization be available for a rm to be
included in the sample.
In addition, to capture the dynamic behavior of a rms hedging and leverage decisions, I focus on a smaller
subset of 200 manufacturing rms (one-digit SIC code 2) and collect data using the same procedure for two
additional years, i.e., 1998 and 1999. This smaller subsample allows me to relate the changes in a rms
hedging activities to changes in nancial conditions, which in turn allows me to draw sharper inferences as
outlined in the subsequent sections.
I limit my analysis to only those rms that have well-dened exposures to foreign currency and commodity
risks. I conduct my analysis for foreign currency derivatives on the subsample of rms with an exposure to
foreign currency risk and similarly commodity derivatives on the subsample of rms with an exposure to
ARTICLE IN PRESS
23
For some rms (most of the rms with a scal year ending in October, November, or December) this corresponds to scal year 1996,
while for others this corresponds to the scal year 1997.
24
The break-up of the notional amount across various instrument types was not easy to obtain for some sample rms. For these rms,
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 doesnt create any bias in the study.
commodity price risk. This sample selection criteria ensures that I can treat the lack of derivative usage as a
rms choice variable to not hedge rather than an absence of exposure to the risk. I identify a rms exposure
to these risks in the following manner.
4.1.1. Exposure to foreign currency risk
I closely follow Geczy, Minton, and Schrand (1997) to identify rms with pre-dened exposure to foreign
currency risks. A rm is classied as being exposed to foreign currency risk if any of the following criteria is
met: (a) it reports foreign currency sales in the COMPUSTAT geographical segment le in the scal year of
derivative usage or within = 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 annual
report; or (d) it discloses an exposure hedged with foreign currency derivatives in its footnotes identied by
hand-collected data.
Based on these screens, I identify 1,781 rms as exposed to foreign currency risk.
25
In the subsequent
regression analysis I lose additional rms due to missing data on the explanatory variables used to estimate the
multivariate models.
4.1.2. Exposure to commodity price risk
Compared to foreign currency exposure, identifying rms with an exposure to uctuations in commodity
prices is harder to measure. This arises because current accounting standards do not require rms to
disclose much information with respect to their exposure to commodity price risk. In the absence of any
balance sheet information, I identify a rms exposure to commodity price risk by estimating the sensitivity of
its earnings to movements in various indices of commodity prices. As an alternative specication, one can use
a simpler approach and take the set of all commodity-producing industries as the sample of rms that are
exposed to commodity price risk. However, with such an approach it would be hard to detect rms that are
exposed to commodity price risk on the input side (such as airline industry). Thus, for the sake of
comprehensiveness, I adopt the more involved methodology of detecting rms with exposure to commodity
price risk.
In particular, I regress the quarterly earnings before interest and taxes obtained from COMPUSTATs
quarterly les on the quarterly changes in several commodity price indices and classify a rm as having an
exposure to commodity price risk if the resulting coefcient is signicant at the 10% level or better. I take data
from the last 60 quarters (or the maximum available) to estimate this model. Most of the effect of commodity
price movements is reected in a rms 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 a
rms earnings less sensitive to movements in commodity prices, rendering my methodology ineffective for
hedger rms. However, I already have hand-collected data on whether these rms use commodity derivatives
to hedge a well-specied risk. I, therefore, add the commodity hedgers to the set of rms that I detect as having
an exposure to commodity risk based on the above methodology.
Second, rms may be exposed to various types of commodity risks, ranging from oil price shocks to metals
to farm produce. Based on the rms disclosure in the footnotes of their annual statements as well as the
contract volume of various futures contracts on the futures exchanges, it is clear that the main sources of
commodity risk facing U.S. nonnancial rms are the following: (a) crude oil and related products; (b) metals
such 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
ARTICLE IN PRESS
25
Geczy, Minton, and Schrand (1997) also consider rms with a high concentration of foreign importers in the industry as exposed to
foreign currency risk. I dont consider this screening criterion since they show that very few rms are identied as having a foreign currency
exposure based solely on this criterion. In their sample of about 370 rms, only three rms are identied as having exposure based solely
on this criterion.
26
I also repeat my analysis with other measures such as cashows, EBIT/TA, NI/TA, the seasonally adjusted earnings, and obtain
similar set of rms. Note that scaling EBIT by total assets doesnt make any qualitative difference because the regression is estimated on a
rm-by-rm 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.
Studies. Further, I obtain data on quarterly changes in the aggregate Producer Price Index (PPI), which
reects 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 ve price indices (crude
oil, metals, farm products, chemicals, and all commodities) and I estimate a rms commodity price sensitivity
with respect to each of these indices separately. Since my analysis doesnt distinguish rms based on the
specic source of risk they face, I consider a rm as exposed to commodity price risk if I obtain a signicant
coefcient in any of the ve regressions. This methodology identies 1,238 rms as having an exposure to
commodity price risk in my sample. When I merge this sample with the sample of rms with ex-ante exposure
to foreign exchange movements, I nd that I have a total of 2,256 rms with an exposure to at least one of
these sources of risk.
4.1.3. Derivatives as a proxy for hedging
I use two denitions of hedging based on derivative usage. The rst denition is based on the rms binary
decision of whether to use derivatives for hedging purposes. This specication uses both types of derivative
contractsforeign currency and commodity. In the second specication, I use the total notional amount of
foreign currency derivatives. The notional amount-based denition of hedging captures the rms total
ownership of risk-management instruments and is thus able to distinguish between rms with different
hedging 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 classied as risk-management tools, there may still be a concern
about their intended useare rms indeed using these instruments for hedging purposes or not? Earlier
empirical studies nd strong evidence in support of risk-reducing (i.e., hedging) effects of derivatives on
various measures of a rms risk. Guay (1999) nds that the new users of derivatives experience a decline in
their earnings and stock price volatility after the initiation of derivatives contracts. Similarly Allayannis and
Ofek (2001) show that using derivatives reduces currency exposure, and Hentschel and Kothari (2001) do not
nd any evidence that derivatives are used for speculative purposes. Thus, there is enough evidence in the
literature to suggest that the majority of rms use derivative instruments for hedging purposes and not for
speculative reasons.
The second concern with the use of derivatives data relates to the importance of derivatives on the
overall cashows of the rms. Allayannis and Weston (2001) and Graham and Rogers (2002) nd a
signicant impact of derivative instruments on rm value and the rms debt capacity, respectively.
These ndings suggest that derivative instruments have a signicant impact on rm performance and
thus are good instruments for the rms risk-management activities. Guay and Kothari (2003) show
that the median rms derivatives cashow sensitivity (dened as the level of cashows that derivative
instruments can generate in extremely adverse scenarios of interest rate, foreign currency or commodity
prices) is modest at only about 10% (mean of 45%) of the average years operating cash-ows of the
rms.
27
At an extreme, if the median rms operating cash-ows drops to 25% of its normal level, the
impact of derivative instruments can be as high as 40% of a bad years operating cash-ows. However,
at the same time the study by Guay and Kothari underscores the importance of nonderivative
based risk-management strategies for rm-value. The study by Petersen and Thiagarajan (2000) illustrates
the importance of nonderivative based hedging strategies for a rms overall risk-management decisions.
In my empirical study I provide various robustness tests to account for nonderivative-based methods of
hedging.
4.1.4. Descriptive statistics of hedging variables
Table 1 provides the descriptive statistics of hedging activities. In Panel A, I provide the frequency
distribution of hedgers of different risks. Out of a total of 1,781 rms with an exposure to foreign currency
risk, 497 (about 28% of the rms) use derivatives to hedge their exposure to movements in foreign exchange
rates. For commodity price risk, there are 211 hedgers (about 20% of the rms) out of a total sample size of
1,238 rms. If I consider exposure to either type of risk, I nd a total of 645 hedgers from a sample of 2,256
ARTICLE IN PRESS
27
The sensitivity varies from 9% to 39% depending on the scaling variable used (see Table 4 of Guay and Kothari, 2003).
rms. Panel B provides the summary statistics for the aggregate notional amount of foreign currency
derivatives used for risk-management purposes. The mean (median) notional amount of foreign currency
derivatives is $359.15 million ($40 million). The average level of derivatives holdings in my sample is smaller
than that of earlier studies such as Graham and Rogers (2002). This is not surprising, since these studies focus
mostly on large rms, whereas my sample contains many medium and small rms as well. The notional value
of derivatives scaled by the book value of the rms total assets (sales) amounts to 8.62% (10.74%) for the
average rm 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 rms use forward and futures contracts. In unreported
analyses, I nd that there are comparable levels of transactions for both buying and selling in the foreign
currency forward markets.
My main tests are based on the relation between leverage and hedging. In the next section, I briey describe
the 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 rms
hedging incentives. My control variables are motivated by these studies. First, I control for rm size (size) as
measured by log of total sales to capture the well-known size effects in derivative usage (see Dolde, 1993). I use
the ratio of research and development (R&D) expenses to total sales as a proxy for rms growth
opportunities. Froot, Scharfstein, and Stein (1993) predict a positive relation between growth opportunities
and hedging incentives since hedging can minimize the underinvestment problem in low cash-ow states of the
world. I also use a rms market-to-book ratio as an additional control variable for growth opportunities and
obtain similar results. However, I do not include it in my base model since market-to-book has been taken as a
measure of rm-value in several studies in corporate nance and rm value may itself depend on derivative
usage. Second, I model leverage in an endogenous setting, which requires regressing leverage on all
ARTICLE IN PRESS
Table 1
Descriptive statisticsderivatives usage
This table provides the descriptive statistics of derivatives usage by sample rms. Panel A provides the number of rms that use foreign
currency (FX) or commodity (CM) derivatives for hedging purposes for the scal year ending between September 1996 to August 1997.
The Any column represents the number of rms 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
explanatory variable in the rst stage regression. Given the illusionary nature of relation between leverage and
market-to-book ratio, this specication poses additional challenges.
28
The underinvestment problem of a rm
can be reduced by keeping more liquid assets. I include the quick ratio of the rm (quick) as a measure of the
rms liquid assets. The quick ratio is constructed as a ratio of cash and short-term investments to the current
liabilities of the rm.
29
Motivated by earlier studies (see Geczy, Minton, and Schrand, 1997; Graham and Rogers, 2002), I include
institutional shareholdings as an explanatory variable in the model to control for risk-management incentives
due to information asymmetry between rms insiders and outsiders. DeMarzo and Dufe (1991) and Breeden
and Viswanathan (1996) argue that rms with higher information asymmetry between managers and
shareholders should hedge more. Assuming that higher institutional share-holdings leads to lower information
asymmetry between the managers and shareholders of the rm, the coefcient on this variable should be
negative as predicted by these theories. The inst variable measures the fraction of common shares of the rm
held by institutional investors. The data are obtained from the 13-F lings. In an alternative unreported
specication, I also use the number of analysts following the rm as a proxy of (inverse) information
asymmetry and obtain similar results.
Next, I control for tax convexity-based hedging incentives. If a rm faces a progressive tax structure, then its
post-tax value becomes a concave function of its pre-tax value. The rm can lower its expected tax liability by
engaging 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 provided
in Appendix A.6. The tax-convexity variable measures the expected tax benets (in dollars) from a 5%
reduction in the rms income volatility. I scale this measure by the total sales of the rm. Since this variable is
estimated by using other accounting variables of the rm, 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 interest
remain 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 rms total sales
as an additional control variable (fsale). Jorion (1991) shows that foreign currency sales is a good proxy of the
rms exchange rate risk exposure. Thus, this variable controls for two effects. First, it controls for the extent
of exposure faced by the sample rms, and second, it proxies for economies of scale that can be exploited in
hedging foreign currency risks. High exposure rms should have a lower cost of hedging if there are signicant
economies of scale in these activities.
Firms can achieve signicant reductions in their foreign currency risk exposure by operating in multiple
geographical locations around the world (see Allayannis, Ihrig, and Weston, 2001). A rm with more
diversied geographical operations has a natural foreign currency hedge if currencies in different markets are
not highly correlated. I control for these effects by including the number of geographical segments reported by
sample rms as a control variable. In an unreported analysis, I also control for the entropy of a rms foreign
sales in diverse geographical regions and obtain similar results.
30
ARTICLE IN PRESS
28
In 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
rm. Since my results remain qualitatively similar, I dont report the results of this model.
29
See also Acharya, Almeida, and Campello (2004), who argue that cash can serve as a hedge against future cash shortfalls for
nancially constrained rms.
30
If a rm operates in n foreign segments (as dened in the COMPUSTAT segments les) and the percentage share of sales of foreign
segment i is P
i
, then the entropy is computed as follows:
Entropy
n
i1
P
i
ln1=P
i
. (6)
The entropy measure is a proxy of the rms geographical diversicationrms with higher entropy values have more diversied
operations across various foreign markets. As an additional robustness check, I also experiment with the rms geographical Herndahl
index to control for this effect. All results remain similar to these alternative control variables.
4.2. Endogenous modeling of leverage and hedging
My theory predicts a positive relation between leverage and hedging for rms with moderate levels of
leverage, and a negative relation at extremely high levels of leverage. In addition, the relation between leverage
and hedging is expected to be stronger for rms operating in industries with greater likelihood of predatory
behavior such as concentrated industries. Thus, my key tests are based on the relation between hedging
and leverage.
For expositional simplicity and analytical tractability, at the time the hedging decision is made in the
theoretical model (i.e., at time t
1
in the model) the debt level is pre-determined. However, we know from prior
theoretical work that a rms debt capacity and hence leverage can itself increase due to hedging. For example,
consider a variant of my model where rms hedge rst and obtain debt at a later date. In the context of my
model, hedging lowers the volatility of rm value, which in turn lowers the probability of bankruptcy and thus
allows rms to borrow more for a given level of the tax benet of debt. This leads to endogeneity between
leverage and hedging. It, therefore, becomes important for my empirical study to explicitly account for this
endogeneity bias. To do so, I need a structural model for the capital structure choice and hedging decisions of
the rm. In the absence of a consensus on an ideal model for debt choices, it is advantageous to have a
theoretical model linking capital structure and hedging choices. I keep the empirical estimation tightly linked
to the theoretical model. In particular, I estimate the following structural model:
leverage b
0
b
1
hedging Sg X
i
e
i
, (7)
hedging a
0
a
1
leverage a
2
leverage
2
Sy Y
i

i
. (8)
This model is estimated in a two-stage instrumental variable (IV) regression framework. The rst-stage
equation is an OLS model for the leverage decision, whereas the second equation models a rms hedging
(derivative) decisions. In the second stage, the risk-management equation is estimated using the predicted value
of the leverage ratio as the explanatory variable in the Logit or Tobit estimation. I try alternative econometric
specications to this model in later sections.
31
The leverage (leverage) of a rm is dened as the ratio of total
debt (long-term debt plus debt included in the current liabilities) to the book value of total assets. To
investigate the effect of extreme leverage on hedging, I include leverage
2
as an additional explanatory variable
in the second equation. I expect a positive sign on leverage and a negative sign on leverage
2
in the regression
involving various measures of hedging as the dependent variable. X and Y represent control variables affecting
rms leverage and hedging decisions, respectively.
As argued earlier, industry concentration provides a good measure of nancial distress costs in my model.
In such industries, highly levered rms are more vulnerable to losing their competitive position in the industry
in the event of nancial distress. Opler and Titman (1994) provide empirical evidence in support of this
assumption. Based on this argument, my model predicts a positive relation between hedging and industry
concentration for highly levered rms. To capture this effect empirically, I include industry concentration
measure and its interaction with leverage in the hedging model. This measure is constructed by summing the
market shares (based on sales in 1996) of the top four players in the rms three-digit SIC code. Then I create a
dummy variable (concd) that equals one if the concentration ratio is above the median, and zero otherwise.
4.2.1. Identication strategy
To estimate this model I need to nd proper instrument(s) for the rst-stage leverage regression. A large
literature studies corporations capital structure determinants (see Frank and Goyal, 2003 for a survey) and
researchers have proposed several determinants of a rms leverage such as size, tangible assets, the book-to-
market ratio, earnings volatility, protability, and marginal tax rates (see Bradley, Jarrell, and Kim, 1984;
Titman and Wessels, 1988; Lang, Ofek, and Stulz, 1996; Graham, Lemmon, and Schallheim, 1998 among
others). For my identication strategy to work, one has to argue that one or more of these variables affect a
rms hedging decision only through their impact on leverage and not independently by themselves. Finding a
ARTICLE IN PRESS
31
In 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., leverage
2
) in the second stage.
truly exogenous instrument for leverage is an extremely challenging task. Given this, I propose an
identication strategy that is motivated by the theoretical model itself.
In the theoretical section, the rms capital structure follows a trade-off model. As in standard trade-off
models, the advantage of debt nancing is its tax benet, whereas its cost is nancial distress and deadweight
losses of bankruptcy. The ex-ante debt ratio is determined by the relative costs and benets of this trade-off.
This provides a dispersion in the debt ratio at time zero in the model. Subsequently, in the intervening time
period the debt ratio is further perturbed by random shocks to the rms protability. Thus, at the time the
hedging decision is made (at time t
1
in the model), the cross-sectional leverage ratio is an outcome of the rms
marginal tax benet of debt, the bankruptcy and distress costs of debt, as well as its recent prot history. The
theoretical model focuses on how the leverage ratio affects hedging decisions at this point in time, which I call
the ex-post hedging decision. In that sense, leverage becomes pre-determined in the model at the time the
hedging decision is made. At this time, shareholders engage in hedging activities as long as the rms leverage
is not too high, beyond which point the risk-shifting incentives begin to dominate.
I rst note that in my model the rst-order effect of hedging on leverage (i.e., the concern about reverse
causation) is through its affect on the cost of leverage and not through its effect on its tax benet. Thus, at
least in the context of my stylized model the marginal tax benet of debt provides one key source of dispersion
in the ex-ante debt ratio that remains largely unaffected by the extent of hedging. It is the deadweight cost of
nancial distress and bankruptcy that decreases due to hedging, allowing rms to borrow more. Thus,
marginal benet of debt seems like a reasonable instrument for identifying the leverage equation in my
empirical model. Motivated by this logic, I consider two instrumentsthe before-nancing simulated
marginal tax rate (MTR) of Graham, Lemmon, and Schallheim (1998) and a rms nondebt tax shield
(NDTS).
Marginal tax rates provide a reasonably direct proxy for the tax benet of debt. Thus, it is directly in the
spirit of my theoretical motivation. To avoid problems associated with negative spurious correlation between
the leverage and marginal tax rates, I use the before-nancing simulated tax rates. Further, I take the historical
averages of MTRs under the assumption that the current level of debt is an outcome of historical incremental
capital structure decisions. I take the past 10 years average MTR as a proxy for a rms current leverage ratio.
As discussed later, this variable has a signicant explanatory power in leverage regression. I also repeat my
analysis with last ve years average and the current level of MTR (without averaging) as instruments and
obtain similar results.
32
My second instrument is the nondebt tax shield enjoyed by a rm. Following earlier literature, I use
depreciation and amortization (da) scaled by the total assets of the rm as a measure of the rms nondebt tax
shield. This instrument measures the disincentive of using debt rather than directly measuring the incentive to
use debt based on tax considerations as captured by MTR. Thus, it is capable of detecting the rms leverage
ratios in response to the tax incentives as proposed in my model. At least controlling for rms size, PPE, and
other key characteristics, it can be argued that nondebt tax shield is a reasonable instrument for leverage in my
leverage-hedging model.
Both these instruments (MTR and DA/TA) possess good statistical properties for an instrument. They both
are signicant determinants of rms debt ratios in the rst stage regression reported in the next section. I also
check for their strength and nd that they do not suffer from any weak instrumentation bias in the sense of
Bound, Jaeger, and Baker (1995) and Staiger and Stock (1997). I repeat all my analyses after considering only
MTR and NDTS (one at a time) as my instrument and all results remain qualitatively similar. In order to save
space and due to the statistical advantage of having more instruments, I consider them both in my leverage
model for the results that I present in the paper. In addition, I use a rms net income to sales ratio (ni) in the
leverage regression as an additional instrument to capture the effect of recent protability on a rms capital
structure at the time of hedging in the spirit of my theoretical model. As I show later, this variable performs
well in the rst stage regression as well.
I include additional control variables in the spirit of Titman and Wessels (1988) and Graham, Lemmon and
Schallheim (1998) to control for well-known drivers of cross-sectional dispersion in leverage ratios. First,
ARTICLE IN PRESS
32
Historical 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.
I include Property, Plant and Equipment (ppe) scaled by total assets to control for the collateral available for
borrowing. I include a rms modied Z-score (see Graham, Lemmon, and Schallheim, 1998) to control for
the effect of rms that may currently be in nancial distress. The modied Z-score (modz) excludes the effect
of leverage from the original Altman Z-score to avoid a mechanical relation between leverage and this
variable. I also include two-digit SIC codes to control for industry-specic drivers of capital structure in my
leverage model. In addition, rm size and R&D-to-Sales ratio enter both the hedging and leverage
specications. Overall, my model is in line with the theoretical arguments and is also close to earlier empirical
studies in corporate risk-management such as Geczy, Minton, and Schrand (1997) and Graham and Rogers
(2002). The base model is presented below:
lev b
0
b
1
hedging b
2
size b
3
rnd b
4
MTR b
5
ppe b
6
modz b
7
ni b
8
da
Ind e
i
.
hedging a
0
a
1
lev a
2
lev
2
Sy Y
i

i
.
I add several other variables to this specication in additional tests. As an alternative test, I use three-year
panel data of 200 manufacturing rms 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 an
association between the variables of interest. My results are qualitatively similar for both the cross-sectional
IV regression model and the change regression model. Further, it should be noted that the nonlinear
specication that I use in my modeling approach gives additional condence that my results are not driven by
reverse causality. This obtains because the endogeneity in my model comes from the fact that hedging can lead
to higher debt levels; this is true for all levels of leverage and especially so for higher levels. Thus, the
endogeneity argument will predict a positive relation between hedging and both leverage and leverage
2
,
something that is opposite to what my theory predicts.
4.3. Univariate tests
Table 2 presents the median values of key rm-level variables across hedgers and nonhedgers. To prevent
outliers from affecting my analysis, all variables used in this paper are winsorized at 1% from both tails. In
Panel A, I present the median characteristics of hedgers and nonhedgers of foreign currency risk in the sample
of rms with exposure to this risk. Panel B provides the same statistics for hedgers and non-hedgers for rms
with exposure to commodity price risk. Panel C is based on pooled observations across both types of risk.
I nd that the hedgers have signicantly different characteristics from the nonhedgers. The hedgers are
signicantly larger rmsthe median hedger rm is about four to ve times bigger than the nonhedger rm in
terms of market capitalization or total sales. The median leverage for hedgers is signicantly higher than the
median leverage of nonhedgers, with stronger results for commodity hedging sample. Hedgers keep less liquid
assets as compared with the nonhedgers as shown by the quick ratios of the two groups. As expected, the
foreign 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 rms with
commodity exposure. I also nd that hedgers have signicantly larger institutional shareholdings than non-
hedgers. While foreign currency hedgers have higher growth opportunities (as proxied by R&D to sales and
market-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 rms hedging incentives to leverage and other
control variables.
4.4.1. First stage estimation
As a starting point, I present the regression results from the rst-stage estimation of leverage as reported in
the rst panel of Table 3. I nd a positive and signicant coefcient on MTR indicating that rms with higher
tax benets obtain higher debt. As expected the coefcient on depreciation and amortization, da=ta the proxy
ARTICLE IN PRESS
for nondebt tax shields, is negative and signicant. Further, consistent with my model rms with higher
protability have lower leverage as indicated by a negative and signicant coefcient on net income to sales
ni=sales. These results are consistent with the motivations behind the use of these variables in the leverage
regression model. Other results are in line with the earlier empirical literature. Once I obtain the predicted
values of leverage from the rst-stage model, I use it in the second-stage model to explain a rms foreign
currency and commodity hedging decision. To save space, I do not present the results from the rst stage
estimation in the rest of the paper.
4.4.2. Foreign currency hedging
I start with the rms foreign currency hedging decision and subsequently analyze the commodity hedging
decisions.
Yes/No decision: I present the results of a second-stage Logit regression in Table 3. The dependent variable
equals one if a rm uses foreign currency derivatives and zero otherwise. The model is estimated with only
those rms that have a pre-dened exposure to foreign currency risks. In the rst model, leverage is positive
and signicant at 1% whereas leverage
2
is negative and signicant at the 1% level. For easier interpretation, I
present the marginal effect (on the probability of hedging) of the explanatory variable evaluated at the mean
rather than the raw estimated coefcient from the logit model. In the next model, I include the interaction of
leverage and a dummy indicating whether the rm belongs to a highly concentrated industry or not. I nd a
positive coefcient on the interaction of leverage and industry concentration (concd). As expected, the
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 identied as rms 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 rm 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 rm. The foreign sales data are obtained from the COMPUSTAT geographical segments le. inst measures the
percentage institutional ownership in the rm. rnd stands for percentage research and development expenses (item 46) scaled by the sales of
the rm (item 12). mtb stands for the market-to-book ratio of the rms assets (COMPUSTAT (item 6 minus 60 plus 25 199) 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
introduction of this interaction term lowers the signicance of leverage variable, but it still remains signicant
at almost the 5% level. These ndings are consistent with the key predictions of the model.
Other results indicate that rms with higher foreign currency sales are more likely to use hedging products,
indicating that highly exposed rms have higher incentives to hedge. There is a strong relation between growth
opportunities as measured by R&D expenses and hedging. This nding is consistent with the theoretical
predictions of Froot, Scharfstein, and Stein (1993) and earlier empirical ndings of Geczy, Minton, and
Schrand (1997). I nd a positive relation between institutional shareholdings and hedging. Assuming an
inverse relation between institutional shareholdings and the extent of information asymmetry between the
insiders and outsiders of the rm, this result is inconsistent with information asymmetry-based models of
hedging. However, more analysis is needed to draw stronger inferences for this theory since the measurement
of information asymmetry remains a difcult task for empirical researchers. In the nal model I include three
more control variables: tax convexity, market-to-book mtb, and the number of geographical segments
(segno) in which the rm operates. All key results remain similar. I dont nd evidence in support of tax-based
ARTICLE IN PRESS
Table 3
Foreign currency hedgingyes/no decision
This table presents logistic regression results for foreign currency hedging by means of derivatives. In the rst stage I estimate an OLS
regression model for leverage. The estimation results from this regression are presented in the rst two columns. In addition to the
coefcients reported in this table, this regression also includes industry dummies based on two-digit SIC code (coefcients suppressed). In
the second stage, a logistic model is estimated with rms foreign currency derivative usage as the dependent variable (one for hedgers and
zero for nonhedgers). lev
denotes the predicted value of leverage from the rst-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 rm. 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 rm. concd is
a dummy variable based on the four-rm concentration ratio of the rms industry (based on three-digit SIC code). concd equals one if the
rm 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 rm. mtr stands for the historical average of rms marginal tax
rates. ppe/ta stands for plant, property, and equipment scaled by total assets. Modied 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 benet from a 5% volatility reduction in the
rms income scaled by the sales of the rm. mtb stands for the market-to-book ratio of the rm. segno stands for the number of
geographical segments in which the rm operates. The number of observations and R
2
(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)
R
2
0.402
N 1,421 1,421 1,421 1,418
motivations for hedging. Finally, rms operating in more diverse foreign markets hedge more as evident by a
positive and highly signicant coefcient on this variable. This result points toward the possibility that
derivative instruments act as complements to a rms natural hedging strategies.
Extent of hedging: In Table 4, I present results from the Tobit estimation with the notional amount of
foreign currency derivatives scaled by total sales of the rm as the dependent variable. Since the estimated
coefcients in a Tobit model dont represent the marginal effect of explanatory variables on the observed
dependent variable, for easier economic interpretation I report the slope coefcients at the mean level.
I present results from three different model specications and nd that rms with high leverage hedge more
and the relation between hedging and leverage reverses at very high levels of leverage. Highly levered rms in
concentrated industries have higher hedging incentives as well. These results are in line with both the models
predictions and the results obtained from the logit model described earlier, as well as with the earlier
theoretical models based on bankruptcy costs (Smith and Stulz, 1985). Economically, these results suggest that
if leverage increases from 10% to 20%, the rm increases its foreign currency derivative holdings by
approximately 6.4%, which is about 60% of the average level of foreign currency derivatives held by the
sample rms (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 nancial distress costs. Mian
(1996) studies the binary (i.e., yesno) hedging decision for a large sample of rms and nds no support for the
distress cost theories.
33
My results suggest that linear models seeking to test theories of risk-management,
ARTICLE IN PRESS
Table 4
Foreign currency hedgingextent of hedging
This table presents the Tobit regression results for foreign currency hedging by means of derivatives. In the rst stage (unreported), I
estimate an OLS regression model for leverage. In the second stage, a Tobit model is estimated with rms 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 rm (zero for nonhedgers). I provide the second-stage estimation results for three different model specications in the table below.
lev
denotes the predicted value of leverage from the rst-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 rm. 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 rm. concd is a dummy variable based on the four-rm concentration ratio of the rms industry (based
on three-digit SIC code). concd equals one if the rm 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 rm. taxconvexity
measures the dollar tax benet from a 5% volatility reduction in the rms income scaled by the sales of the rm. mtb stands for the
market-to-book ratio of the rm. segno stands for the number of geographical segments in which the rm 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)
mtb 0.0005 (0.38)
segno 0.0086 (4.99)
N 1,421 1,421 1,418
33
There 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 doesnt test for nonlinearity and therefore it doesnt
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.
especially if conducted on small sample sizes, may fail to detect a positive relation between nancial distress
and derivative activities for moderately leveraged rms.
To understand the nonmonotonic relation between hedging and leverage, I conduct a semi-parametric test.
I break the sample of rms into two groups based on whether their predicted leverage is below or above the
70th percentile of the empirical distribution of leverage in my sample. For this estimation I run a Tobit
regression with the same set of variables as in Model 1 of Table 4 after dropping leverage
2
. The spline
regression results show that for the rst group, i.e., for the group with moderate leverage the marginal effect of
leverage on hedging is positive with a slope coefcient of 0.0452, which is signicant at the 1% level. However,
the marginal effect of leverage on hedging becomes negative for rms in the other group, i.e., for rms with
leverage in top 30% of the sample. For this group, the marginal effect of leverage on hedging is estimated to be
0:1504 with a signicance level of 2%. The semi-parametric test conrms the non-monotonic relation
obtained in parametric regressions.
4.4.3. Commodity hedging
Table 5 provides logistic regression results for the commodity hedging decision. This regression is estimated
on a sample of rms with exposure to commodity price risk only. As in the foreign currency derivative
regression, I nd a positive and signicant coefcient on leverage, and a negative and signicant coefcient on
leverage
2
. Both these relations are signicant at the 1% level. When I include the interaction of leverage with
high concentration industry, I nd the coefcient on the interaction term to be positive and signicant at the
6% level. These results show that the predictions of my theory are supported by both foreign currency and
commodity hedging data. As in the case of foreign currency hedging, larger rms are more likely to use
commodity derivatives as well. However, in this regression the coefcient on the quick ratio becomes positive
and signicant, while it is positive but insignicant in the foreign currency hedging models. Commodity
hedgers keep more liquid assets as well, which can be taken as evidence that hedgers complement their hedging
policies with liquid assets.
The most noticeable difference between the two models is the coefcient on the R&D variable. This variable
has a positive and highly signicant coefcient in the logit and Tobit model of foreign currency hedging
ARTICLE IN PRESS
Table 5
Commodity hedgingyes/no decision
This table presents logistic regression results for commodity hedging by means of derivatives. In the rst stage (unreported) I estimate an
OLS regression model for leverage. In the second stage, a logistic model is estimated with rms commodity derivative usage as the
dependent variable (one for hedgers and zero for non-hedgers). lev
denotes the predicted value of leverage from the rst 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 rm. 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 rm. concd is a dummy variable based on the four-
rm concentration ratio of the rms industry (based on three-digit SIC code). concd equals one if the rm belongs to an industry with
concentration ratio above the median, zero otherwise. inst measures the percentage institutional ownership in the rm. taxconvexity
measures the dollar tax benet from a 5% volatility reduction in the rms income scaled by the sales of the rm. mtb stands for the
market-to-book ratio of the rm. The number of observations is provided at the end of the table.
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)
mtb 0.0013 (0.15)
N 948 948 947
decisions. However, here it is negative and insignicant. While high growth rms manage their foreign
currency hedging more aggressively, they are less likely to manage their commodity risk exposure. Although
exploring the differences in hedging incentives across different types of risks is beyond the scope of this paper,
this nding is suggestive of rms facing conicting incentives in managing various forms of risk. These
conicting incentives may be driven by factors such as differences in the correlation between the risk being
hedged and the rms investment opportunity set. For example, the argument behind a positive relation
between growth opportunities and hedging relies on the assumption that high growth rms may need funds to
undertake projects in bad cashow states. If rms do not have good investment opportunities in states with
poor realizations of cashows, then this incentive disappears. At the extreme, if the investment opportunity set
is highly positively correlated with the realizations of cashows, then such rms may have a disincentive to
hedge. A better understanding of these issues is left for future research that explicitly incorporates these
correlations in the analysis.
4.5. Alternative model specications
Industry-adjusted leverage ratios: The results presented so far in the paper are based on a two-stage
specication that requires an assumption about the structural model determining a rms leverage choice. For
robustness, I test a model that does not require such a specication. Specically, I conduct the analysis with
rms industry-adjusted leverage ratios as industry adjustment provides a simpler and perhaps more robust
way to classify rms into moderately and highly leveraged. The industry-adjusted leverage ratio of a rm is
dened as the difference between the rms leverage and the industry median based on the two-digit SIC code.
ARTICLE IN PRESS
Table 6
Alternative models
This table presents the results for derivatives usage for various alternative model specications. The rst two models use alternative
denitions of nancial distress, whereas the last two models use leverage as the measure of nancial distress but use bootstrapped standard
errors in estimating the t-Values. FD stands for the measure of nancial distress for the given model and FD
2
is its squared term. The rst
model uses the industry-adjusted leverage ratio based on two-digit SIC codes as a measure of FD. For this model FD
2
equals leverage-
squared if the rms leverage is above industry average, zero otherwise. In the second model, I use Altmans Z-score as a measure of the
rms nancial distress and estimate a logistic model using the rms 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 rms closer to nancial 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 rst-stage regression, and FD
2
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 rm. 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 rm. concd is a dummy variable based on the four-rm concentration ratio of
the rms industry (based on three-digit SIC code). concd equals one if the rm 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 rm. Number of observations used for the estimation is provided in the last row.
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)
FD
2
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
I reestimate all my results with these ratios. For this model, leverage variable can be either positive or negative
depending on whether the rm is above or below industry-median. Thus, I cannot use leverage
2
as an
explanatory variable to test for the nonlinearity predicted by the model. Instead, I use a variable that equals
leverage
2
for rms with higher than industry median leverage and zero otherwise. To conserve space, I pool
foreign currency and commodity hedging decisions to estimate these robustness results. I estimate a logit
model on a sample of rms that are exposed to either of these two types of risks and present the results in
Model 1 of Table 6.
34
My key results remain the same with this denition of leverage.
Altman Z-score: I use the Altman Z-score as an alternative proxy of nancial distress. Lower Z-score values
correspond to nancially weaker rms. I, therefore, transform them by taking their inverse to be consistent in
the presentation of results. Results are presented in Model 2 of Table 6. I nd a nonmonotonic relation based
on this measure as well.
Bootstrapped standard errors: Since I use a two-stage estimation methodology in the logit and Tobit
regressions, there is a potential for overstated t-statistics due to the sampling error of rst-stage estimation (see
Maddala, 1983). To account for this possibility I reestimate my models with bootstrapped standard errors. In
every replication I create a pseudo-random sample by drawing observations from the base sample with
replacement. Thus, in every replication some of the observations appear more than once and some do not
appear at all. With 100 such replications, I generate an empirical distribution of estimated coefcients in
the logit and Tobit models. The standard deviations of these estimates are then used to obtain bootstrapped
p-Values for my base estimation. This methodology does not rely on any structural form for the estimation of
the variance-covariance matrix and has the advantage of benchmarking base estimates against their empirical
distributions. In Models 3 and 4 of Table 6, I present the Logit and Tobit model estimates with bootstrapped
errors. As shown, all my key results are robust to bootstrapped standard error estimation.
Alternative IV regression specication: Wooldridge (2002) suggests an alternative instrumental variable
regression model for models involving functions of an endogenous variable (such as leverage
2
in the second-
stage estimation). Potentially, this technique provides econometrically better estimates than the model
that uses predicted values of leverage and its function in the second stage. In this method, rather than
using the squared value of predicted leverage in the second-stage regression, both leverage and leverage
2
are treated as endogenous variables and instrumented with their own instruments. To achieve identication,
I add the squared terms of all exogenous variables entering the leverage model as instruments for leverage
2
.
These instruments are the squared values of: mtr, da=ta, ppe=ta, ni=sales and modified_z. In addition, the
squared predicted value of leverage from the rst stage estimation is used as an additional instrument for
leverage
2
.
With both leverage and leverage
2
as endogenous variables and these instruments in hand, I estimate an
instrumental variable model in a two-stage regression framework. I estimate the binary decision to hedge
foreign currency or commodity derivatives with an IV Probit model and the extent of foreign currency hedging
with an IV Tobit model. The results are presented in Table 7. Note that the parameter estimates in this model
are not directly comparable to the earlier tables since, due to computational simplicity, I report the coefcients
from the regressions directly rather than the slope coefcients presented earlier. I nd that my key results
remain robust to this alternative IV estimation technique. All other results remain similar to the earlier base-
case specication.
4.6. Dynamic analysis
Change regression: I focus on a three-year panel of 200 manufacturing rms to exploit the variations in the
individual rms leverage and hedging intensities in a dynamic setting. This empirical strategy closely
resembles my theoretical model and presents several econometric advantages. The change regression controls
for unobserved rm-specic factors. In particular, unless a rms nonderivative-based hedging strategies have
changed substantially over this time period, change regressions control for operational and natural hedging in
a more precise manner. Second, the endogeneity argument is less severe for change regressions since by
construction it removes rm-specic unobservable effects that could be correlated with both hedging and
ARTICLE IN PRESS
34
Individual regressions estimated separately on foreign currency and commodity samples are qualitatively similar.
leverage at any given point in time. Third, this model allows me to relate changes in this years hedging
intensities to both current and past years leverage changes, allowing me to establish some evidence on ex-ante
and ex-post hedging incentives.
I obtain data on derivative holdings for a random subsample of 200 manufacturing rms (one-digit SIC
code 2) for the scal years ending in 19971998 and 19981999, i.e., for two more years after the initial sample
period. I focus on manufacturing rms to ensure that the sample is homogenous. The choice of a smaller
subsample is purely dictated by the requirements of manually collecting data. After dropping one rm-year
observation due to the non-availability of its 10-K on EDGAR and restricting attention to rms with
nonmissing observations, I obtain 394 rst-differenced rm-year observations. The accounting characteristics
such as size, leverage and market-to-book ratio of this subsample are qualitatively similar to those of the
overall sample (unreported) and thus the sample is a reasonable representation of COMPUSTAT-CRSP
rms.
For this sample, I obtain data on their foreign currency derivatives and investigate a subset of rms
that have either increased or decreased the intensity of their foreign currency hedging. Since there are
very few initiators or terminators of commodity derivatives in my sample, the change regression is not
feasible for commodity hedges. In this sample, there are 42 rm-year observations with an increase and
39 rm-year observations with a decrease in the extent of hedging (notional amount of foreign currency
derivatives as a percentage of sales). The remaining rm-year observations have no changes mostly due to zero
hedging positions across the period. In my analysis, I focus on only those observations that have either
increased or decreased their hedging positions to avoid inferences based on a majority of rms that remain
nonhedgers during the sample period. Focussing on the sample of active hedging decisions (increase or
decrease) allows for a sharper identication of hedging incentives in response to changes in rm-level
variables.
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 specication 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), modied Z-score and ni/sales (net income to
total sales). The squared terms of each of these variables are used as additional instruments for leverage
2
. In addition, the squared value of
the predicted leverage is also used as an instrument for leverage
2
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 rms. Size represents the log of total sales
of the rm. rnd stands for research and development expenses scaled by the sales of the rm. quick is the ratio of cash and short-term
investments to current liabilities. concd is a dummy variable based on the four-rm concentration ratio of the rms industry (based on
three-digit SIC code). concd equals one if the rm 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 rm. 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.
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
I conduct logit as well as OLS regressions (linear probability model) to estimate the impact of changes in
leverage on changes in derivatives holding. The following model is estimated:
Dhedge
j;t
a
0
a
1
Dlev
j;t
a
2
Dlev
2
j;t

aControl
j;t

j;t
. (9)
The dependent variable takes a value of one for an increase in hedging and zero for a decrease; Dlev
j;t
measures the change in leverage of rm j in year t; Dlev
j;t
2
is the squared change in leverage for rms with an
increase in leverage, zero otherwise, and all other control variables in the model are also rst differenced. I
include all control variables in this model that enter the earlier regression except for variables that are unlikely
to 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 rst (logit) and third (OLS) model, I nd that rms with a
moderate increase in leverage are more likely to increase their hedging positions as evident by a positive and
signicant coefcient on Dlev. In contrast, rms with a very high increase in leverage are more likely to
decrease their hedging intensities. Though the statistical signicance of the coefcient on Dlev is weaker as
compared to the cross-sectional case, it remains signicant at the 7% level. The coefcient on Dlev
2
on the
other hand remains statistically strong (at 2%) as in the cross-sectional case. In fact changes in leverage remain
the most signicant determinant of changes in hedging intensities as compared to other covariates that enter
this 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 rms hedging position, which is a mixture of both ex-ante
and ex-post actions (i.e., hedging decisions taken before/together with debt issuance and those taken after debt
issuance). Ex-ante theory predicts a positive association between leverage and hedging, whereas ex-post theory
predicts a nonmonotonic relation. In cross-sectional data, therefore, ex-ante decisions bias my study against
nding a nonmonotonic relation. This happens because, if all decisions are taken ex-ante, then the hedging
motivations should be strongly positively associated with leverage even at very high levels of leverage, making
the task of nding a negative relation between the two variables harder. Though change regressions lead to an
improvement over the cross-sectional model, it is still possible that changes in leverage and hedging occur at
the same time in the spirit of ex-ante hedging theory. To analyze this issue further, I regress Dhedge
j;t
on Dlev
j;t
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 rms (SIC code 2) for 19961997 to 19981999 period. Changes in derivative holding is regressed on changes in leverage
and various other rm characteristics. The dependent variable is one if rm 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. Dlev
2
equals the
square of Leverage change if leverage has increased over the year and zero otherwise. Dlaglev is the previous years 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 rm. 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 rm. fsale represents foreign sales as a percentage of total sales. Models 1 and 2 are logistic estimations (slope
coefcients 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 rm level to account for correlation in error terms of
same rms across multiple years.
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)
Dlev
2
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
along with Dlev
j;t1
(i.e., the lagged value of leverage change) and other control variables used in the earlier
regression.
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 coefcient
on the lag change can be reasonably attributed to the hedging decisions consequent to debt issuance. Given
that we do not observe reporting of hedging at a high frequency, establishing a link between past leverage
change and current change in derivatives based on annual data is a challenging task. Still, the results from
columns 2 and 4 in Table 8 show that the past years leverage change is a signicant predictor of the current
years hedging activities. The coefcient on the lagged value of leverage change (i.e., Dlaglev) is positive and
signicant at the 2% level. This is encouraging since, due to the sequence of decision making (i.e., last years
leverage decision and this years hedging), this specication is very likely to detect causation between the
hedging and leverage variable.
5. Conclusion
This paper develops a theory of corporate risk management in the presence of nancial distress costs. By
distinguishing nancial distress from insolvency, I provide a justication for the ex-post risk-
management behavior of the rm. Due to nancial 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 between
shareholders risk-shifting incentives due to equitys limited liability and their risk-avoidance incentives due to
nancial distress costs. I obtain a closed-form solution for the optimal level of investment risk based on this
trade-off. The model generates several testable predictions. It predicts a nonmonotonic relation between
leverage and hedging and a U-shaped relation between nancial distress costs and hedging. Financially
distressed rms 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. I
model a rms leverage and hedging in an endogenous framework using a sample of more than 2,000
nonnancial rms. I nd evidence in support of a positive relation between leverage and foreign currency and
commodity hedging. Consistent with the theory, this relation becomes negative for rms with very high
leverage. Financially distressed rms in highly concentrated industries hedge more. Finally, I show that the
key results remain similar for a dynamic analysis based on a change regression for a smaller subset of rms.
Appendix A
In the theoretical model of the paper I consider a continuous trading economy with a time horizon t
0
; T
and ltered probability space O;
t
; ; P satisfying the usual regularity conditions. I assume a complete and
arbitrage-free market. This guarantees the existence of an equivalent martingale measure Q. I assume a
deterministic short interest rate process given by r. In what follows, I denote the indicator function of an event
X by 1
fXg
. I assume that the unlevered asset value of the rm can be expressed as a Q-Brownian Motion (under
a martingale measure) as follows:
dA
t
rA
t
dt sA
t
dW
t
.
A.1. Proof of Proposition 1
Proof. Shareholders payoff (CF) on the terminal date T is given by
CF
T
V
T
L1
fV
T
4L;m
T
4Kg
f V
T
L1
ff V
T
4L;m
T
pKg
V
T
L1
fV
T
4Lg
1
fV
T
4L;m
T
pKg
f V
T
L1
ff V
T
4L;m
T
pKg
V
T
L1
fV
T
4Lg
V
T
L1
fV
T
4L;m
T
pKg
f V
T
L1
ff V
T
4L;m
T
pKg
ARTICLE IN PRESS
V
T
L1
fV
T
4Lg
V
T
L1
ff V
T
4L;m
T
pKg
1
ff
1
L4V
T
4L;m
T
pKg
f V
T
L1
ff V
T
4L;m
T
pKg
V
T
L1
fV
T
4Lg
V
T
f V
T
1
ff V
T
4L;m
T
pKg
L V
T
1
ff
1
L4V
T
4L;m
T
pKg
V
T
L V
T
f V
T
1
ff V
T
4L;m
T
pKg
L V
T
f1
fV
T
pLg
1
ff
1
L4V
T
4L;m
T
pKg
g.
Under mild technical restrictions, the equity value at t t
1
x
t
1
is simply the expectation of
this payoff under the martingale measure. Taking the expectation of the terminal payoff gives the desired
result. &
A.2. Equity valuation
As shown in expression 3 in the paper, the equity valuation is given by the following:
E
t
1
E
Q
A
T
Lf1
fA
T
4L;m
T
4Kg
1
fA
T
4L;A
T
pLM;m
T
pKg
g M1
fA
T
4LM;m
T
pKg
E
Q
A
T
Lf1
fA
T
4L;m
T
4Kg
1
fA
T
pLM;m
T
pKg
1
fA
T
pL;m
T
pKg
g M1
fA
T
4LM;m
T
pKg
E
Q
A
T
Lf1
fA
T
4Lg
1
fA
T
4LMg
1
fA
T
4LM;m
T
4Kg
g M1
fA
T
4LM;m
T
pKg
.
A:1
The rst two components of the equity value, namely, E
Q
A
T
L1
fA
T
4Lg
and E
Q
A
T
L1
fA
T
4LMg
,
can be computed using the standard Black-Scholes formula for the valuation of European call options. Let F
and f stand for the normal cumulative density function (cdf) and probability density function (pdf),
respectively. For notational simplicity I set A
t
1
A
0
and T
0
T t
1
. Then the rst two terms result in the
following expression:
E
Q
A
T
L1
fA
T
4Lg
A
0
Fh
1
LFh
2
and
E
Q
A
T
L1
fA
T
4LMg
A
0
Fd
1
LFd
2
,
where
h
1

ln
A
0
L
_ _
s
2
2
T
0
s
T
0
p and h
2
h
1
s
T
0
p
,
d
1

ln
A
0
L M
_ _
s
2
2
T
0
s
T
0
p and d
2
d
1
s
T
0
p
.
The last two terms, i.e., E
Q
A
T
L1
fA
T
4LM;m
T
4Kg
and E
Q
M1
fA
T
4LM;m
T
pKg
, require the knowledge
of 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 two
path-dependent expressions:
Lemma. Let KoL and KoA
0
, then the joint density of the terminal asset value A
T
and the running
minima of the geometric Brownian motion m
T
, under the martingale measure, is provided by the following
formula:
QA
T
4L; m
T
XK F
ln
A
0
L
_ _
s
2
2
T
0
s
T
p
0
_
_
_
_
_
_
_
_
A
0
K
_ _
F
ln
K
2
A
0
L
_ _
s
2
2
T
0
s
T
0
p
_
_
_
_
_
_
_
_
. (A.2)
ARTICLE IN PRESS
Using (A.2), I compute the expectation of the last two terms of expression (A.1) as follows:
E
Q
A
T
L1
fA
T
4LM;m
T
4Kg
fA
0
Fd
1
LFd
2
g KFc
1

A
0
L
K
Fc
2
_ _
,
E
Q
M1
fA
T
4LM;m
T
pKg

A
0
M
K
Fc
2
,
where d
1
and d
2
are as given before and
c
1

ln
K
2
A
0
L M
_ _
s
2
2
T
0
s
T
0
p and c
2
c
1
s
T
0
p
.
Collecting the above results and simplifying the expressions further, I get the following expression for the
valuation of the rms equity at t t
1
:
E
t
1
fA
0
Fh
1
LFh
2
g KFc
1

A
0
L M
K
Fc
2
_ _
. (A.3)
A.3. Proof of Proposition 2
At t t
1
; the shareholders choose an optimal risk level such that it maximizes the equity value given in
expression (A.3). I am assuming that the rm is not in nancial distress at t t
1
, i.e., KoA
0
. I also assume
that the distress barrier is below the face value of debt, i.e., KoL. At the optimum:
qE
t
1
qs
0.
Differentiating expression (A.3) gives the following:
qE
t
1
qs
A
0
fh
1
qh
1
qs
Lfh
2
qh
2
qs
_ _
Kfc
1
qc
1
qs

A
0
L
K
fc
2
qc
2
qs
_ _
A
0
M
K
fc
2
qc
2
qs
. (A.4)
Note that:
fh
2

A
0
L
fh
1
,
fc
2

K
2
A
0
L M
fc
1
. A:5
Eqs. (A.4) and (A.5) lead to
qE
t
1
qs
A
0
fh
1
qh
1
qs
A
0
fh
1
qh
1
qs

T
0
p
_ _
Kfc
1
qc
1
qs

KL
L M
fc
1
qc
1
qs

T
0
p
_ _ _ _ _
KM
L M
fc
1
qc
1
qs

T
0
p
_ _
.
Simplication leads to the following:
qE
t
1
qs
A
0
fh
1
T
0
p
fc
1
K
qc
1
qs

KL
L M
qc
1
qs

T
0
p
_ _ _ _
KM
L M
qc
1
qs

T
0
p
_ _ _ _
.
Thus, I have the following rst-order condition for the optimal investment risk of the rm,
A
0
fh
1
T
0
p
fc
1
K
T
0
p
0. (A.6)
ARTICLE IN PRESS
Thus,
A
0
fh
1
Kfc
1
0. (A.7)
A.4. Second-order condition
Differentiating expression (A.6) gives the second-order optimality condition:
q
2
E
t
1
qs
2
h
1
A
0
fh
1
T
0
p
qh
1
qs
c
1
fc
1
K
T
0
p
qc
1
qs
. (A.8)
Using the rst-order condition and simplifying, the above expression, at the optimum, reduces to
q
2
E
t
1
qs
2
h
1
fc
1
K
T
0
p
qh
1
qs
c
1
fc
1
K
T
0
p
qc
1
qs
fc
1
K
T
0
p
h
1
T
0
p
h
1
s
_ _
c
1
T
0
p
c
1
s
_ _
_ _
fc
1
K
T
0
p
h
1
c
1
h
1
c
1
s
T
0
p

s
.
Using the following equalities,
h
1
c
1

ln
A
2
0
L M
K
2
L
_ _
s
T
0
p 40 and h
1
c
1
s
T
0
p

ln
K
2
LL M
_ _
s
T
0
p o0,
it follows that
q
2
E
t
1
qs
2
o0. (A.9)
Thus, the second-order condition for the maximization problem is satised.
A.5. Comparative statistics
Comparative statistics are obtained by a direct differentiation of the optimal solution for s given in
expression 5.
(a) Sensitivity with respect to default boundary (K):
qs
2
qK

2
T
0
K ln
L M
L
_ _ ln
K
2
LL M
_ _
ln
K
2
L
A
2
t
1
L M
_ _ _ _
o0.
The inequality follows from the facts that KoL; KoA
t
1
and M40.
(b) Sensitivity with respect to time-to-maturity T
0
T t
1
:
qs
2
qT
0

s
2
T
0
o0.
(c) Sensitivity with respect to the deadweight loss parameter M: Direct differentiation of the optimal
investment risk leads to the following:
qs
2
qM
40 if M4Lexp
2
lnA
t
1
=K lnL=K
_
_ _
L.
ARTICLE IN PRESS
(d) Sensitivity with respect to leverage: Let X lnK
2
=LL M, Y lnK
2
L=A
2
t
1
L M and Z ln
L M=L. Denote leverage by lev L=A
t1
. After some algebra it can be shown that
qs
2
T
ql

2
K
qK
ql

qM
ql
1
L M
A
1
L

1
L M
_ _ _ _
Y
Z

2
K
qK
ql

qM
ql
1
L M
A
3
L

1
L M
_ _ _ _
X
Z

qM
ql
1
L M

AM
LL M
_ _
XY
Z
2
.
In this model, leverage affects investment risk via its affect on distress boundary K and deadweight loss
parameter M. Consider K 1 exp
0:1
lev
and M 7 exp
2
lev
. This specication corresponds to a
concave distress boundary. A highly levered rm faces higher distress boundary, meaning such rms are more
likely to get into distress and become the target of predatory behavior of their rivals even for relatively small
additional downturn in their nancial health. The specication of M corresponds to a model in which higher
leverage imposes higher costs on the rm again due to reasons such as lost customers to rivals. The model has
been solved for different values of leverage using the analytically derived formula for s. For this specication,
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 specications 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 of
hedging. Using the simulation methods and considering the various features of tax codes, they compute the
expected tax-benets that would result from a 5% reduction in income volatility. Subsequently they perform a
regression analysis that relates tax savings to the following explanatory variables: (i) an indicator variable
identifying taxable income between $500; 000 and zero (TI(NEG)), (ii) an indicator variable identifying
taxable income between zero and $500,000 (TI(POS)), (iii) income volatility measured as the absolute
coefcient of variation (VOL), (iv) rst-order serial correlation in income (RHO), (v) a dummy variable
indicating the existence of investment tax credit (ITC), (vi) a dummy variable indicating the existence of Net
Operating Loss (NOL) carryforwards, and nally (vii) NOL dummy interacted with the small-loss
(NOLTI(NEG)) and small-gain (NOLTI(POS)) indicator variables. The regression estimate is given as
follows:
Taxconvexity 4:88 7:15 TINEG 1:60 TIPOS 0:019 VOL 5:50 RHO
1:28 ITC NOL 3:29 4:77 TINEG 1:93 TIPOS.
I obtain the predicted tax savings in dollars for each rm in the sample by using the above equation. This is
scaled by the sales of the rm to get the tax-convexity measure.
References
Acharya, V., Almeida, H., Campello, M., 2004. Is cash negative debt? A hedging perspective on corporate nancial 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, 273296.
Allayannis, G., Weston, J., 2001. The use of foreign currency derivatives and rm market value. Review of Financial Studies 14, 243276.
Allayannis, G., Ihrig, J., Weston, J., 2001. Exchange-rate exposure: nancial vs. operating strategies. American Economic Review 91,
391398.
Andrade, G., Kaplan, S., 1998. How costly is nancial (not economic) distress? Evidence from highly leveraged transactions that became
distressed. Journal of Finance 53, 14431493.
Arping, S., 2000. Debt and product market fragility. Working paper, HEC, Lausanne.
ARTICLE IN PRESS
Asquith, P., Gertner, R., Scharfstein, D., 1994. Anatomy of nancial distress: an examination of junk-bond issuers. Quarterly Journal of
Economics 109, 625658.
Bartram, S.M., Brown, G.W., Fehle, F.R., 2003, International evidence on nancial 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, 443450.
Bradley, M., Jarrell, G.A., Kim, E.H., 1984. On the existence of an optimal capital structure: theory and evidence. Journal of Finance 39,
857878.
Breeden, D., Viswanathan, S., 1996. Why do rms 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,
511529.
Brown, G., 2001. Managing foreign exchange risk with derivatives. Journal of Financial Economics 60, 401448.
Chevalier, J., 1995a. Capital structure and product market competition: empirical evidence from the supermarket industry. American
Economic Review 85, 415435.
Chevalier, J., 1995b. Do LBO supermarkets charge more? An empirical analysis of the effects of LBOs on supermarket pricing. Journal of
Finance 50, 11121195.
DeMarzo, P.M., Dufe, D., 1991. Corporate nancial hedging with proprietary information. Journal of Economic Theory 53, 261286.
DeMarzo, P.M., Dufe, D., 1995. Corporate incentives for hedging and hedge accounting. Review of Financial Studies 8, 743771.
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 nancial risk management. Journal of Applied Corporate Finance 6, 3341.
Fehle, F., Tsyplakov, S., 2005. Dynamic risk management: theory and evidence. Journal of Financial Economics 78, 347.
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 nancing policies. Journal of
Finance 5, 16291658.
Geczy, C., Minton, B.A., Schrand, C., 1997. Why rms use currency derivatives? Journal of Finance 52, 13231354.
Goldstein, R., Ju, N., Leland, H., 2001. An EBIT-based model of dynamic capital structure. Journal of Business 74, 483512.
Graham, J.R., Lemmon, M.L., Schallheim, J.S., 1998. Debt, leases, taxes and the endogeneity of corporate tax status. Journal of Finance
53, 131162.
Graham, J.R., Rogers, D.A., 2002. Do rms hedge in response to tax incentives? Journal of Finance 57, 815839.
Graham, J.R., Smith, C.R., 1999. Tax incentives to hedge. Journal of Finance 54, 22412262.
Guay, W., 1999. The impact of derivatives on rm risk: an empirical examination of new derivatives users. Journal of Accounting and
Guay, W., Kothari, S.P., 2003. How much do rms hedge with derivatives? Journal of Financial Economics 70, 423461.
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,
107152.
Hentschel, L., Kothari, S.P., 2001. Are corporations reducing or taking risks with derivatives? Journal of Financial and Quantitative
Analysis 36, 93118.
Jensen, M.C., Meckling, W.H., 1976. Theory of the rm: managerial behavior, agency costs and ownership structure. Journal of Financial
Jorion, P., 1991. The pricing of exchange rate risk in the stock market. Journal of Financial and Quantitative Analysis 363376.
Kalay, A., 1982. Stockholder-bondholder conict and dividend constraints. Journal of Financial Economics 10, 211233.
Kovenock, D., Phillips, G., 1997. Capital structure and product market behavior: an examination of plant exit and investment decisions.
Review of Financial Studies 10, 767803.
Lang, L., Ofek, E., Stulz, R., 1996. Leverage, investment and rm growth. Journal of Financial Economics 40, 329.
Leland, H.E., 1998. Agency costs, risk management, and capital structure. Journal of Finance 53, 12131243.
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, 419439.
Moodys Investors Service Report, 1998. Historical Default Rates of Corporate Bond Issuers, 19201997.
Morellec, E., Smith, C.W., 2003. Investment policy, nancial policies, and the control of agency conicts. 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, 221250.
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, 267284.
Opler, T., Titman, S., 1994. Financial distress and corporate performance. Journal of Finance 49, 10151040.
Petersen, M., Thiagarajan, S.R., 2000. Risk measurement and hedging: with and without derivatives. Financial Management 29,
530.
Phillips, G., 1995. Increased debt and industry product markets: an empirical analysis. Journal of Financial Economics 37, 189238.
ARTICLE IN PRESS
Purnanandam, A., 2007. Interest rate derivatives at commercial banks: an empirical investigation. Journal of Monetary Economics 54,
17691808.
Smith, C.W., Stulz, R., 1985. The determinants of rms hedging policies. Journal of Financial and Quantitative Analysis 28, 391405.
Smith, C.W., Warner, J.B., 1979. On nancial contracting: an analysis of bond covenants. Journal of Financial Economics 7, 117161.
Staiger, D., Stock, J.H., 1997. Instrumental variables regression with weak instruments. Econometrica 65, 557586.
Stulz, R., 1984. Optimal hedging policies. Journal of Financial and Quantitative Analysis 19, 127140.
Stulz, R., 1996. Rethinking risk management. Journal of Applied Corporate Finance 9, 824.
Titman, S., 1984. The effect of capital structure on a rms liquidation decision. Journal of Financial Economics 13, 137151.
Titman, S., Wessels, R., 1988. The determinants of capital structure choices. Journal of Finance 43, 119.
Tufano, P., 1996. Who manages risk? An empirical examination of risk management practices in the gold mining industry. Journal of
Finance 51, 10971137.
Wooldridge, J.M., 2002. Econometric Analysis of Cross Section and Panel Data. MIT Press, Cambridge, MA.
Zingales, L., 1998. The survival of the ttest or the fattest: exit and nancing in the trucking industry. Journal of Finance 53, 905938.
ARTICLE IN PRESS

Don't Know

Hochgeladen von

Dokumentinformationen

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Don't Know

Hochgeladen von

Copyright:

Verfügbare Formate

Journal of Financial Economics 87 (2008) 706739

Financial distress and corporate risk management:

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

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

in the interior of all possible risks.

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

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

concd 0.0572 (2.21)

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

concd 0.3107 (1.94)

Das könnte Ihnen auch gefallen

Don&#39;t Know

Hochgeladen von

Dokumentinformationen

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Don&#39;t Know

Hochgeladen von

Copyright:

Verfügbare Formate

Journal of Financial Economics 87 (2008) 706739

Financial distress and corporate risk management:

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

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

in the interior of all possible risks.

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

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

concd 0.0572 (2.21)

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

concd 0.3107 (1.94)

Das könnte Ihnen auch gefallen

Don't Know

Don't Know