Optimal Currency Hedging

Rui Albuquerque∗
University of Rochester
April 11, 2003

Abstract
This paper characterizes optimal currency hedging in several models of downside
risk. We consider, in turn, three models of hedging: (i) a firm that chooses its
hedging policy in the presence of bankruptcy costs; (ii) an all equity firm that faces
a convex tax schedule; and (iii) a firm whose manager is subject to loss aversion.
In all these models, and contrary to conventional wisdom, we show that forwards
dominate options as hedges of downside risk.
Keywords: Currency hedging, forwards, options, bankruptcy costs, taxes, loss
aversion, downside risk.
JEL Classification: F31, G30.

∗
I thank the comments of Mike Barclay, Luca Benzoni, João Cocco, Hugo Hopenhayn, John Long,
Robert McDonald, Mitchell Petersen, Sérgio Rebelo, Cliff Smith, Alan Stockman, and seminar par-
ticipants at INSEAD, Kellogg School, Simon School, University of Maryland, University of Wisconsin-
Madison, University of Virginia, and the Federal Reserve Bank of Richmond. I thank Andrew Welzel and
Graham Cassidy for the data. I am grateful for financial support from the Banco de Portugal and JNICT,
and for the hospitality of the Finance Department at Kellogg. The usual disclaimer applies. Address:
Simon School of Business, Department of Finance, Carol Simon Hall, University of Rochester, Rochester,
NY 14627. Phone: (585) 275 3956. Fax: (585) 461 3309. E-mail: [email protected].

1
1 Introduction
The globalization of goods and capital markets means that an increasing number of firms
have to make decisions about hedging their foreign exchange exposure. This need is
aggravated by extremely volatile foreign exchange markets. There is a large number
of derivative securities, including forward and options contracts, that can be used to
construct a hedging strategy. How do these instruments interact? And what is the
optimal hedging policy? In this paper we provide answers to these questions in several
economic models of downside risk.
We focus the analysis on models of optimal hedging in the presence of downside risk.
This is for two reasons. First, industry surveys indicate that in designing their hedging
policies CFO’s seek the selective elimination of risks, instead of the elimination of all
risks as dictated by the variance-minimizing framework.1 In fact, Stulz (1996) argues
that this should be the “goal of corporate risk management—namely, the elimination of
costly lower-tail outcomes” (p. 8). Second, swings in exchange rates have normally been
associated with very dramatic industry turnover. Countries such as South Korea and
Thailand have seen their banks become insolvent as a result of currency devaluations
that greatly increased the local currency value of their dollar denominated liabilities. In
the early 1980’s some firms found themselves in financial distress for not hedging against
the strengthening of the dollar. A famous example discussed in Rawls and Smithson
(1993) is Laker Airlines. In September 1992 many European firms were affected by
the devaluations associated with the European Monetary Mechanism crisis. Hedging
downside risk would have helped banks and firms in these episodes stay financially afloat.
We start by discussing two standard models of downside risk that are consistent with
value-maximizing firms. First, we discuss a model of optimal hedging and capital struc-
ture choice in which hedging is driven by the desirability to reduce expected bankruptcy
costs. Second, we analyze an all-equity, risk neutral firm that faces a piecewise linear
convex tax schedule. This model is then generalized to explicitly account for dynamic
tax loss carry-forwards. Our third model focuses on managerial loss aversion (Kahneman
and Tversky (1979, 1992)). All three models describe the excess sensitivity of a firm to
low outcomes, and are appropriate to studying hedging of downside risk. To simplify the
analysis we study an environment in which derivative contracts are priced by risk neutral
agents, but involve transactions costs.
The paper’s main proposition is that, contrary to conventional wisdom (see Stulz
(1996) for a discussion), in all three models, currency forwards are better hedges than
1
Rawls and Smithson (1993) provide a summary of survey evidence on managerial hedging behavior.

2
options against downside risk. The proof of this result relies on the fact that in efficient
markets the net unit payoff of both derivative instruments must have the same expected
value. Suppose an exporting firm faces transactions exposure and wishes to eliminate
the downside risk associated with an abrupt rise on the value of the domestic currency.2
This objective can be accomplished by transferring income from high profit states (those
with domestic currency depreciation) to low profit states (those with domestic currency
appreciation). Because forward contracts pay more than options on the downside (they
also make bigger losses on the upside), less currency needs to be sold forward to achieve
the same degree of hedging. Thus, total transactions costs paid using forwards must be
smaller than those paid using options. This is true even though we have assumed equal
transactions costs per unit of foreign currency traded.
Section 2 develops the model with bankruptcy costs. Section 3 deals with the case of
tax convexity. Section 4 analyzes hedging when managers exhibit loss aversion. Section
5 presents evidence on transactions costs for forwards and options and section 6 presents
the final remarks. Appendices A and B contain proofs of results in sections 2 and 3,
respectively. Appendix C provides a discussion of the plausibility of the tax schedule
used in section 3 and of other tax issues related with hedging.

2 Model with Bankruptcy Costs

A popular model of downside risk is one in which the firm faces bankruptcy costs when
default on debt obligations occurs. Smith and Stulz (1985) have shown that by hedging
the firm is able to reduce the likelihood of default by increasing the income it gets in the
downside. This leads to lower expected bankruptcy costs (possibly higher debt levels)
and higher firm value.3
Consider an infinitely lived, all-equity firm with a fixed investment opportunity set and
production capacity that exports one unit of output per period. The value in domestic
currency of exports’ profits is stochastic because of fluctuations in the exchange rate.
The revenue from exports denominated in foreign currency is determined one period
in advance and normalized to 1. For simplicity of exposition, we assume that there
2
Transaction exposure reflects the changes in the domestic currency value of cash flows from existing
contracts due to changes in the exchange rate. In industry surveys, transaction exposure is typically
referred to as the most important reason to hedge (e.g. Millar (1989), Jesswein, Kwok, and Folks (1995)).
For a comparative study of currency hedging by German and US corporations see Bodnar and Gebhardt
(1998).
3
Bessembinder (1991), Ross (1996), and Leland (1998) develop models of hedging in the presence of
taxes and default costs. Titman (1992) studies how debt maturity is affected by the possibility of buying
interest rate swaps in a model of asymmetric information.

3
are no costs associated with foreign sales. We introduce production costs in the next
section. Let s denote the exchange rate quoted as units of domestic currency per unit
of foreign currency. The exchange rate follows a first order Markov process and has
bounded support: s ∈ (0, s̄]. The behavior of s is described by the conditional cumulative
distribution function H. To focus on bankruptcy costs we assume a flat tax rate (τ ) on
corporate profits.
Pricing of Debt. The firm borrows one-period debt in local currency units.4 A debt
contract is a pair (B, b), where B is the face value of debt, and b is the market value of
debt. Hence, B/b − 1 is the interest rate paid on debt. Bondholders are risk neutral and
discount their future utility at rate r. If the firm defaults, bondholders pay bankruptcy
costs in the amount of α.5 This makes the level of bankruptcy costs independent of
hedging. This assumption is realistic since hedging contracts typically have priority over
debt contracts. Let primes denote next period’s variables and Π be total profits from
operations and hedging (which is defined explicitly below). Firms are subject to capital
market frictions according to which they cannot borrow against future profits. Thus, the
firm defaults if and only if (1 − τ ) Π0 < B. Under these assumptions the market value of
debt is given by the zero profit condition:
Z s̄ Z
1 0 0 1
b= min [B; (1 − τ ) Π ] H (ds ; s) − α H (ds0 ; s) . (1)
1+r 0 1 + r {s0 :(1−τ )Π0 <B}

Equation (1) says that the market value of debt is equal to the discounted value of
the minimum between the face value of debt and the firm’s net-of-tax profits minus the
expected bankruptcy costs. To conform with the fiscal code, taxes have priority over
debt claims. The interest rate on debt commands a premium due to bankruptcy costs if
the probability of default is positive.
Forward and Put Options Contracts. The firm can hedge its exchange rate risk using
forward or options contracts. Both of these contracts are priced by a risk neutral investor
who faces transactions costs of φ units of local currency per unit of foreign currency
traded. Transactions costs are incurred in the period in which the contract is traded.
This means that our firm can sell foreign currency forward at the (bid) rate:

f = Es (s0 ) − φ(1 + r), (2)

4
Using covered interest parity we know that any debt contract denominated in foreign currency can
be replicated with a forward contract and a debt contract denominated in local currency units.
5
Several papers estimate the costs of financial distress. For example, Andrade and Kaplan (1998)
estimate that 10% to 20% of firm value is lost in direct or indirect costs of financial distress.

4
where Es is the conditional expectations operator. The firm can buy one period put
options with exercise price s∗ at a premium:
Z s∗
1
p= (s∗ − s0 ) H (ds0 ; s) + φ. (3)
1+r 0

For simplicity, the notation ignores the dependence of the options premium (p) on the
strike price. We consider only options with strike price s∗ < s̄, since an option with
s∗ = s̄ is equivalent to a forward contract. For future reference it is useful to note that
the expected profits associated with both a forward contract and an options contract is
negative and equal to −φ(1 + r). The expected value of a forward contract is:

Es (f − s0 ) = −φ(1 + r).

The expected value of an options contract (net of the ask price of the option) is:

Es [max (s∗ − s0 , 0) − p(1 + r)] = −φ(1 + r).

Profits. Total profits from operations and hedging activities are:

Π = s + xf (f − s) + xo max (s∗ − s, 0) − pxo (1 + r)

where pxo is the cost of the options purchased in the previous period.6
Optimal Choice of Hedging Instruments. We now investigate whether the firm should
combine forwards and options as part of its hedging strategy or whether it should use
a single hedging instrument. Without loss of generality we assume that the firm uses a
single type of option. Because of transactions costs it is never optimal to combine options
with different strike prices in this setting.7
Shareholders discount the firm’s after-tax cash flow at interest rate r. Firms maximize
the value of equity. The choice of capital structure and hedging policies is simultaneous.8
6
To make the cash flows from options and forwards comparable we assume that the firm must borrow
to purchase the options. In order not to favor forwards, we assume that the firm can use risk free debt
for this purpose.
7
In particular, a forward contract can be replicated with two options by put-call parity. However,
the synthetic forward produced in this way is dominated by the outright forward because transactions
costs are doubled.
8
This timing assumption is usually nontrivial, in that it affects the amount of hedging done, but it
does not affect the choice of which hedging instruments to use. Shareholders may have an incentive
to hedge less after debt is in place if the only reason to hedge is to reduce the probability of default.
The reason is that although hedging increases firm value it also transfers wealth from shareholders to
bondholders, hence potentially making the former reluctant to pursue hedging policies ex-post (see Smith
and Stulz 1985).

5
For simplicity, interest and principal payments benefit from a tax shield. This assumption
does not affect our qualitative results. The firm’s problem is:
½ Z h i ¾
1 0 0 0
Ṽ (s) = max b+ (1 − τ ) (Π − B) + Ṽ (s ) H (ds ; s) ,
B≥0,xo ,xf ∈[0,1] 1 + r {s0 :(1−τ )Π0 ≥B}

subject to (1). In this formulation the value of equity given by V = (1 − τ ) (Π − B) + Ṽ .

We rule out the possibility of gambling with derivative instruments (i.e. xf , xo < 0, and
xf , xo > 1, are not allowed).
In the absence of debt, the probability of default is zero and hedging does not add
value. The following proposition characterizes optimal hedging for a levered firm.

Proposition 1 For any positive debt level, when bankruptcy costs are positive (α > 0),
and transactions costs are positive (φ > 0) and identical for forwards and options, the
optimal hedging strategy relies solely on forward contracts (i.e. xo = 0, and xf ≥ 0).

Proof. Fix B > 0. Suppose the firm chooses to default for values of the exchange rate
lower than Sd (hence, Sd < B/ (1 − τ )). Thus, Sd determines the firm’s optimal default
probability and expected bankruptcy costs for fixed B. How many forward contracts must
the firm buy? This number is given by setting (1 − τ ) Π = B, which, when forwards alone
are used, becomes:
xf [f − Sd ] + Sd = B/ (1 − τ ) .
And, how many options are needed for the same purpose? Obviously, the strike price
needs to satisfy s∗ ≥ Sd , otherwise the put option would be useless. This leads to:

xo [s∗ − Sd − p(1 + r)] + Sd = B/ (1 − τ ) .

For any level of debt B, strike price Sd ≤ s∗ < s̄, and any default threshold that the
firm chooses, xo > xf , since f > s∗ − p (1 + r).9 Therefore, the total transactions costs
needed to achieve the same benefit (i.e., savings in expected bankruptcy costs) are higher
with options. The reason for this is illustrated in Figure 1. Both contracts share the
same expected value (−φ(1 + r)), but forward contracts have lower payoffs than options
for high values of s. This means that forward contracts must have higher payoffs than
9
To see this last step note that
d ∗
[s − (1 + r) p] = 1 − H (s∗ ) > 0,
ds∗
and f = s∗ − (1 + r) p, when s∗ = s̄.

6
options for low values of s. These higher payoffs for low s mean that we need fewer
forward contracts than options contracts to achieve a certain level of hedging. Thus,
total transactions costs associated with hedging are always lower when forward contracts
are employed. Since the result was proved for any B > 0, it also holds for the optimal
B.
This result runs counter the conventional wisdom that options are an ideal instrument
to hedge downside risk (see Stulz 1996). Keeping a low default probability is the goal
of the firm’s hedging strategy. To do this, hedging instruments must be bought, for
which the firm incurs transactions costs. This objective can be achieved by choosing
options with an appropriately low strike price. First, these options are cheap since they
are unlikely to be exercised. Second, the options payoff has a non-linear shape that can
exactly off-set the one-sided risk the firm faces (given an appropriate strike price). Why
is hedging with cheap options not optimal? This is because a low price just means that
it is more difficult to transfer income across states of nature.

Additional income generated

by the forward contract in low
f profit states

s*-p(1+r)

Payoff of one Put Option

Contract
_ s
f s*
s
-p(1+r)

Payoff of one Forward

Contract

Figure 1: The Forward and Options Payoff.

It is important to note that this result holds very generally and also that we have
taken the conservative view that transactions costs are identical for forwards and options

7
(see Section 5).
We can extend this result to a comparison of two currency put options with different
strike prices. Contrary to conventional wisdom, expected profits are lower when hedging
is implemented with cheaper options. Under our assumptions, all options have the same
transactions costs. Because cheaper options have lower payoffs in low states, a higher
number of these options is required to achieve a given level of hedging (however, see
section 5 for empirical measures of options transactions costs).
Optimal Hedge Ratio and Debt Choice. Setting xo = 0, the first order necessary
conditions for interior solutions on xf and B are:
³ ´ µ ∂ Z sd ¶
0
α + τ B + Ṽ (sd ) − H (ds ; s) = (1 − τ ) φ (1 + r) , (4)
∂xf 0

and Z s̄ ³ ´ ∂ Z sd
0
τ H (ds ; s) = α + τ B + Ṽ (sd ) H (ds0 ; s) ,
sd ∂B 0
where the default threshold is determined by (1 − τ ) Π (sd ) = B. The first of these con-
ditions equates the marginal benefit from hedging, obtained through a reduction in the
Rs
default probability (Pr [(1 − τ ) Π < B] = 0 d H (ds0 ; s)), to the marginal cost resulting
from transactions costs. The second condition says that the firm’s optimal debt pol-
icy trades-off the expected tax benefit of debt (conditional on not defaulting) against
the marginal cost of debt, which arises due to the increased likelihood of paying the
bankruptcy costs, the lost tax shield, and continuation value upon default.10
If f − B/ (1 − τ ) ≥ 0, the firm is able to pay its debt always if it hedges 100% of its
exposure. In this case, an increase in the hedge ratio lowers the default probability and
makes hedging valuable. If f − B/ (1 − τ ) < 0, an increase in the hedge ratio raises the
default probability, and no hedging is done (i.e. xf = 0).
Letting f − B/ (1 − τ ) ≥ 0, we can investigate the impact of bankruptcy costs (α) on
debt and hedging. For simplicity, consider the case in which s is iid and h (s) is uniform.
All the computations are provided in Appendix A. In this model, hedging decreases
when bankruptcy costs increase. The reason for this negative impact is that there are
Rs
decreasing marginal returns to hedging (i.e. ∂ 2 0 d H (ds0 ; s) /∂x2f < 0). That is, more
hedging lowers the default probability, but this impact is smaller the more hedging is
done. On the other hand the impact of higher bankruptcy costs on debt is ambiguous.
There is a direct negative impact, but there is also an indirect effect because less hedging
10
Mello and Parsons (1999) also argue that the use of different instruments (in their case forwards and
futures) may trigger specific debt convenants. We ignore such issues here.

8
is done. This is a somewhat unforeseen possibility that is also present in other models of
capital structure and risk management (e.g. Leland 1998).

3 Tax Convexity
3.1 A Reduced Form Model of Tax Benefits of Hedging
In this subsection we eliminate the debt choice and bankruptcy costs in order to focus
on optimal hedging under tax convexity. To introduce a simple motive for hedging we
assume that the firm faces the following tax schedule:

 τ 1Π , Π > 0
T (Π) = τ 2Π , Π ≤ 0 . (5)

τ2 < τ1

Appendix C provides a discussion of fiscal issues related to hedging, and suggests that this
is a reasonable representation of the US corporate tax schedule. In the next subsection we
present a more general model of hedging with optimal dynamic tax loss carry-forwards.
As before, Π represents the firm’s total profits, including any losses or gains from
hedging activities.11 In this section and the next we assume that there is a cost associated
with foreign sales of c units of domestic currency. With the elimination of debt, this is
required so that profits can be negative. Total profits from operations and hedging
activities are now given by:

Π = s − c + xf (f − s) + xo max (s∗ − s, 0) − pxo (1 + r). (6)

According to the tax schedule T (.) the firm receives a tax rebate when its profits are
negative. However, the tax rate on positive profits is higher than the rebate tax rate. To
simplify we assume that profits on exports are treated as domestic income and are taxed
only in the country of residence. Thus, there is no double taxation of profits.
Optimal Choice of Hedging Instruments. What combination of hedging instruments
is better, if any?12 Firm value conditional on the current level of the exchange rate (s)
11
Notice that for tax purposes the firm is only allowed to deduct p, the option’s premium at historical
values, in computing its taxable income. In practice, this feature of the tax schedule places the option
at a disadvantage with respect to forwards. This disadvantage can be eliminated by financing the option
with risk free debt. The tax benefits associated with debt financing exactly cancel the tax loss implicit
in ignoring the time value of money when deducting the option price at book values.
12
Proposition 2 below holds if instead we allow the firm to write a call option on its export revenue.
Though selling a call might seem more appropriate to hedge tax risk, it still does not provide the kind
of downside profit that the forward is able to provide.

9
and hedging policy is:
· ¸
∗ 1 0 0 ∗ 0
V (xf , xo , s , s) = max Π − T (Π) + Es V (xf , xo , s , s ) .
x0f ,x0o ,s∗ ≥0 1+r

It is easy to show that if profits are always positive, or τ 1 = τ 2 , it is optimal not to

hedge (xf = xo = 0). In this case hedging brings no benefits. Since there are transactions
costs associated with derivative contracts (φ > 0) the expected profits from hedging are
negative.
Let us consider the case in which hedging carries benefits. This requires τ 2 < τ 1 (so
losses are treated asymmetrically from profits), and a positive probability that losses will
be incurred. The following proposition characterizes optimal hedging.

Proposition 2 If the tax schedule is convex, τ 1 > τ 2 , and transactions costs are positive
(φ > 0) and identical for forwards and options, the optimal hedging strategy relies solely
on forward contracts: xo = 0, xf ≥ 0. When the exchange rate follows a stationary
AR(1) process with positive persistence, there is a region of inaction in which xf = 0.
The forward hedge behaves non-monotonically with the exchange rate:
½
dxf ≥ 0, f − c < 0
.
ds ≤ 0, f − c > 0

Proof. Suppose that the firm decides to hedge so that it restricts losses for values
of the exchange rate below S (for any S < c). How many forward contracts would be
necessary to achieve this objective? The answer is given by the equation:

xf (f − S) + S − c = 0.

How can we achieve the same objective with options? Obviously, we have to choose a
strike price s∗ > S, so that the option has a positive payoff when s = S. In addition, the
number of options purchased must be such that the realized profits for s = S are zero:

xo [s∗ − S − p(1 + r)] + S − c = 0.

Using (2) and (3) it is easy to show that f − S > s∗ − S − p(1 + r), which implies that
xf < xo . (Recall footnote 9.) This means that the cost of achieving an arbitrary level of
hedging is always lower with forward contracts than with options. We provide the proof
of the second part of the proposition in Appendix B.

10
 As before, this result runs counter the common perception that options are an ideal
instrument to hedge downside risk. The intuition is similar to the case with bankruptcy
costs. The presence of tax convexity makes it desirable for the firm to avoid making
losses. Because forward contracts pay more than options on the downside (they also
make bigger losses on the upside), less currency needs to be sold forward to achieve
the same degree of hedging. Thus, total transactions costs paid using forwards must be
smaller than those paid using options.
Optimal Hedge Ratio. Now that we have determined that it is optimal to set xo = 0,
we can find the optimal value of xf . The first order condition for an interior solution to
this problem can be written as:
Z
φ(1 + r)(1 − τ 1 ) = (τ 1 − τ 2 ) (f − s0 )H(ds0 ; s). (7)
{s0 :Π≤0}

The left-hand side of this equation is the marginal cost (net of taxes) of entering into
an additional forward contract. The marginal benefit corresponds to the income that is
transferred by the forward contract from high tax rate states (those with positive profits)
to low tax states.
Proposition 2 shows that the effect of an increase in the exchange rate on the dy-
namic forward hedge depends on the magnitude of the current exchange rate. When the
exchange rate is high (and the exchange rate has positive serial correlation), an increase
in s produces high future expected exchange rates and hence reduces the probability
of having negative profits. As the benefit from hedging is reduced so is xf . When the
exchange rate is low, a marginal increase in s raises the marginal benefit from hedging,
and hence xf .
For some values of the current exchange rate (s) the optimal hedging strategy is
to leave the exposure uncovered: xf = 0. When the exchange rate is sufficiently low,
the probability of having negative profits next period is too high (with positive serial
correlation) and the firm has little probability of incurring losses. When the exchange
rate is high enough, the likelihood of positive profits next period is too high and the
convexity of the tax schedule also disappears. Thus, even though there are no fixed costs
of implementing a hedging program it may be optimal not to hedge. This band of inaction
may account for the large percentage of survey respondents that say that they do not
hedge because of the reduced significance of their exposures. It may also account for the
usual policies of selectively hedging perceived exposures. This effect is exacerbated in
cross sectional studies undertaken after periods of high dollar depreciation (when profits
from exports are highest).

11
 We prove in Appendix B that the optimal hedge ratio, xf , is non-monotonic both
with respect to costs of exports (c) and to the transactions cost parameter (φ). An
increase in c produces two effects. An increase in c increases the set of negative profit
states, thus making hedging more desirable. However, when c is very high the benefits
of hedging decline because profits are low, and thus there is not much income that can
be transferred to negative profit states to reduce the taxes paid by the firm.
The optimal hedge is generally not monotonic in transactions costs for f − c > 0 (this
is the interesting case since the firm has positive profits if it fully hedges its exposure).
An increase in φ increases the marginal cost of hedging, hence lowering the hedge ratio.
However, because the forward bid rate decreases with transaction costs a higher hedge
ratio is needed to keep a low probability of negative profits. The net effect is ambiguous.
This ambiguity disappears when f − c < 0 where the second effect is eliminated.

3.2 An Explicit Model of Tax Loss Carry-forwards

Here we generalize the analysis of the previous subsection to explicitly model tax loss
carry-forwards. As before, we allow for stationary, serially correlated exchange rate
processes.
Without loss we assume the firm can carry losses forward indefinitely.13 Let the tax
schedule be given by T (Π (st ) − Lt−1 ) = τ (Π (st ) − Lt−1 )+ , where Lt−1 is the tax loss
carry-forward at the beginning of fiscal year t, and the notation (X)+ = X if X ≥ 0 and
X = 0 otherwise. The law of motion of the carry-forwards is Lt = (Lt−1 − Π (st ))+ .
Consider hedging with forward contracts only. A firm with tax loss carry-forwards of
Lt−1 ≥ 0, current hedge ratio of 100xt−1 %, and level of exchange rate st solves:
· ¸
1
V (xt−1 , Lt−1 ; st ) = max Π (st ) − T (Π (st ) − Lt−1 ) + Es V (xt , Lt ; st+1 ) .
xt ,Lt 1+r
If Lt = 0 then no hedging is done. The first order necessary condition when Lt > 0 is
(after some simplifications):
Z
1
φ (1 − τ ) (1 + r) + VL (xt+1 , Lt+1 ; st+1 ) (ft+1 − st+1 ) H (dst+1 ; st )
1 + r {Π(st+1 )−Lt <0}
Z
= τ (ft+1 − st+1 ) H (dst+1 ; st ) ,
{Π(st+1 )−Lt <0}

with
1
VL (xt−1 , Lt−1 ; st ) = τ 1(Π(st )−Lt−1 ≥0) + 1(Π(st )−Lt−1 <0) Es VL (xt , Lt ; st+1 ) , (8)
1+r
13
Imposing finite carry-forward of losses limits the usefulness of options even more since with options
it is more likely that some of the losses will never be used to offset gains.

12
where VL (.) is the partial derivative of the value function with respect to L (the marginal
value of tax loss carry-forwards), and 1 (ω) is an indicator function taking the value of
1 if the event ω is true and zero otherwise. The first term on the left-hand-side and the
term on the right-hand-side are the same as in equation (7) above. There is a new term:
It characterizes the value of postponing the exercise of the carry-forwards. Rewrite the
first order condition as:

Z · ¸
1 1
φ (1 − τ ) = τ− VL (xt+1 , Lt+1 ; st+1 ) (ft+1 − st+1 ) H (dst+1 ; st ) .
1+r {Π(st+1 )−Lt <0} 1+r
(9)
Condition (9) is very similar to the one we have derived in the main text for the reduced
form tax schedule (equation 7). Using (8) it is easy to see that VL (xt+1 , Lt+1 ; st+1 ) ≤ τ .
Intuitively, postponing the tax shield from the tax loss carry-forward cannot yield a
greater benefit than using the tax shield today when interest rates are positive.
Consider now the choice between forwards and options. Since with forwards the firm
achieves the largest reduction in tax loss carry-forwards today, the firm is better off by
doing so than postponing the realization of the tax shield, which the option does in
greater proportion. Thus, forwards dominate options.

4 Loss Aversion
Loss aversion plays a central role in the prospect theory developed in Tversky and Kah-
neman (1979) and Kahneman and Tversky (1992). Loss aversion describes the excess
sensitivity of a firm to low outcomes. Mathematically, the firm’s utility function displays
a kink. This leads to risk neutrality with respect to profits’ volatility above a certain
threshold and risk aversion for volatility of outcomes below that threshold. Thus, the
model is appropriate to studying hedging of downside risk, i.e. the elimination of these
undesirable outcomes.
Let manager’s compensation be a linear function of profits. The lifetime utility of the
firm’s manager can be written as:
· ¸
∗ 1 0 0 ∗ 0
U(xf , xo , s , s) = 0max u(Π) + Es U (xf , xo , s , s )
xf ,x0o ≥0 1+r

where the utility function u is given by:

½
Π, Π > 0
u(Π) = , (10)
λΠ, Π ≤ 0

13
and Π is given by (6). The parameter λ > 1 measures the degree of loss aversion
(e.g. Benartzi and Thaler, 1995), which refers to the psychological tendency to be more
sensitive to low outcomes than to high ones. Note that the same formulation is valid for a
different threshold level, in which case profits (Π) would have to be adjusted accordingly.
With this in mind we refer to negative profits as the firm’s downside.
It is now obvious that the model of a risk neutral firm that maximizes its value while
facing the tax structure (5), presented in section 3, is formally equivalent to the model
with loss aversion if one sets λ = (1 − τ 2 ) / (1 − τ 1 ) > 1.

5 Empirical Evidence on Transactions Costs

In the model, transactions costs on forwards and options are equal, proportional to the
amount traded, and constant across strike prices. This section shows that we have taken
a conservative view of identical transactions costs for forwards and options in favor of
options, independently of the strike price. Thus, our results are robust to modelling trans-
actions costs on options that change with the strike price (which is a better description
of the data) once we account for higher options transactions costs.
We analyze data on transactions costs for American options and forwards for three
exchange rates: the Yen/US$, Euro/US$, and BP/US$. The contract size is 10,000 BP
and Euros, and 1,000,000 Yen. The data was collected from March/31/99-May/27/99.
There are 20 days in the sample, and as many observations per business day as there
are traded strike prices for 3- and 6-month options contracts. Bid and ask spreads were
obtained from Bank One for forward rates, and from the Philadelphia Stock Exchange
for put options. These quotes are roughly at the same time of day, and measure only the
cost of establishing a position.
Figure 2 plots the annualized transactions costs for 3-month put options, in excess of
that for forwards, measured in basis points for the three currencies considered. Figure
3 reproduces the annualized excess transactions costs for 6-month contracts.14 Average
annualized transactions costs for 3-month forward contracts are roughly 30 basis points
across the currencies, with a minimum of 8.5 b.p. and a maximum of 40.8 b.p. Since
transactions costs are measured by the bid-ask spread, these numbers compare to the
theoretical value of 2φ (1 + r). To get a sense of these numbers, a firm buying a one year
put option for 1,000,000 BP would pay 1,000 pounds over the forward in transactions
costs alone (half spread, or φ).
14
There are no put options on the Euro at 6-months due to absence of trades in the sample. Note
that the illiquidity of these contracts is also an obvious cost to them.

14
 Excess Transactions Costs for the BP on 3 Month
Contracts

60
Annualized Basis Points

50
40
30
20
10
0
150 152 154 156 158 160 162 164 166
Strike Prices (US cents per BP)

Excess Transactions Costs for the Yen on 3 Month

Contracts

50
Annualized Basis Points

40

30

20

10

0
75 80 85 90 95
-10
Strike Prices (US hundredths of a cent per Yen)

Excess TransaExcess Transactions Costs for the

Euro on 3 Month Contracts

60
Annualized Basis Points

45
30
15
0
-15 98 100 102 104 106 108 110 112 114

-30
-45
Strike Prices (US cents/EURO)

Figure 2: Excess transactions costs for Sterling, the Yen, and Euro on 3-month contracts.

15
 Excess Transactions Costs for the BP on 6 Month
Contracts

25

Annualized Basis Points

20

15

10

0
152 154 156 158 160 162 164 166
-5
Strike Prices (US cents per BP)

Excess Transactions Costs for the Yen on 6 Month

Contracts

25
Annualized Basis Points

20
15
10
5
0
-5 72 77 82 87 92

-10
Strike Prices (US hundredths of a cent per Yen)

Figure 3: Excess transactions costs for Sterling, and Yen on 6-month contracts.

The evidence collected in these figures says that options transactions costs are higher
at any traded strike price, and that transactions costs vary with moneyness of the op-
tions.15 These findings broadly suggest that besides the mechanism that we develop,
according to which forwards are better hedges of downside risk for identical costs, the
difference in transactions costs is another reason to hedge with forwards.

6 Final Remarks
We have analyzed several models of optimal currency hedging of downside risks. Two of
the models consider value maximizing firms that either face (i) bankruptcy costs, or (ii)
a piece-wise linear convex tax schedule. The third model focuses on loss aversion, which
makes the manager more sensitive towards low outcomes, hence more willing to hedge
15
Further out-of-the-money options have even larger bid-ask spreads due to, for example, thin markets.

16
against downside risks.
The main contribution of the paper is to highlight the type of derivative contracts
whose payoff structure produces a better hedge against downside risk. In particular, we
show that forward contracts dominate options as hedges of transactions exposure when
the firm is concerned with downside risks. The main reason for this is the forward’s ability
to transfer income across states of nature, concentrating the payments on the firm’s
downside. Thus, a smaller hedge ratio is required, and less costly is risk management.
A preliminary analysis of hedging downside risk in the presence of tax convexities
reveals that there are significant non-linearities in hedging behavior. These non-linearities
include zones of inaction and non-monotonicity in the relation between hedge ratios
and exchange rates. Inaction, or zero hedging, can account for the usual policies of
selectively hedging perceived exposures. Also, it is likely that this effect is exacerbated
in cross sectional studies/surveys undertaken after periods of high dollar depreciation
(when profits from exports are highest).
Indirect evidence from industry survey studies suggests that forwards are more widely
used as hedging derivatives than options. Rawls and Smithson (1993) compare several
surveys regarding the question of what instruments firms use to hedge their foreign
exchange exposure. In Millar (1989), 99% of the respondents indicate that they have
used forwards whereas only 48% say they have used options (see also Jesswein, Kwok,
and Folks 1995, and Bodnar and Gebhardt 1998). Direct evidence from the gold mining
industry indicates that roughly 70% of hedging is done with forward contracts (Adams,
1999). Of course, this evidence does not necessarily provide a conclusive test of our results
for two main reasons. First, risk management strategies need not be based on a desire to
eliminate downside risk, or may relate to other forms of exchange rate exposure. Second,
proportional transactions costs on forwards tend to be smaller than those associated
with options. Note that our analysis is not restricted to hedging exchange rate exposure.
Given the availability of liquid markets for financial hedging instruments, our results can
be applied to corporate risk management in general.

17
A Some Comparative Statics in the Model with Bank-
ruptcy Costs

For notational convenience let Πf = f −B/ (1 − τ ). Assume that Πf > 0, so that hedging
is desirable, and that xf < 1. Let the default threshold (sd ) be

−Πf + (1 − xf ) f
sd = .
1 − xf

Given our assumptions, Ṽ exists and is unique, bounded and continuous. Assuming s
is iid and follows a uniform distribution, and after some simplifications, the first order
necessary conditions can be written as (note that Ṽ is constant when s is iid):
³ ´
α + τ B + Ṽ Πf = s̄ (1 − τ ) φ (1 + r) (1 − xf )2 ,

and ¡ ¢
τ (s̄ − f ) (1 − xf ) + Πf = α + τ B + Ṽ .
Totally differentiating the first order conditions we obtain the derivatives of B, and xf ,
with respect to α:
τ Πf +(α+τ B+Ṽ )
dxf − τ (1−τ )
= h i ,
dα τ (s̄ − f ) (1 − xf ) + α+τ B+Ṽ
(s̄ − f ) + 2 (2 − τ ) s̄φ (1 + r) (1 − xf )
1−τ

and
dB 11−τ 1−τ dxf
=− − (s̄ − f ) .
dα τ 2−τ 2−τ dα
Thus, hedging always decreases with higher bankruptcy costs, whereas debt may decrease
or increase.

B Proofs of the Main Results with Tax Convexity

To simplicity the notation define σ ∗ = s∗ − p (1 + r).
Proof of Proposition 2. Here we prove the second part of the proposition. (An
algebraic proof of the first part of the proposition is available upon request.) Given
our assumptions, V exists and is unique, bounded and continuous. Because the net
profit function is concave in xf , V is also concave for any fixed s, and so differentiable

18
almost everywhere (see Stokey, Lucas, and Prescott (1989)). The first order necessary
and sufficient condition for 0 < x∗f < 1 is
Z s̃f
1
φ (1 + r) (1 − τ 1 ) = (τ 1 − τ 2 ) (f − s0 )H(ds0 ; s). (11)
1+r 0

A unique solution with positive hedging exists so long as:

Z c
1
φ (1 + r) (1 − τ 1 ) < (τ 1 − τ 2 ) (f − s0 )H(ds0 ; s). (12)
1+r 0
When this condition is violated x∗f = 0. When, f −R c = 0, and (12) holds, the marginal
1 c
benefit of an extra unit of forwards (τ 1 − τ 2 ) 1+r 0
(f − s0 )H(ds0 ; s) is invariant with
xf , and so the full hedge is optimal. Let s follow an AR(1) process, with correlation
ρ ∈ (0, 1), and unconditional average µ = E (s0 ):

s0 = ρs + (1 − ρ) µ + v,

with v being an iid shock with zero mean and support (−µ(1 − ρ), (s̄ − µ)(1 − ρ)). This
implies that the unconditional support of s is (0, s̄). The density function of v is given
by γ > 0. The distribution of s0 conditional on s has support:

S (s) = (ρs, ρs + (1 − ρ) s̄) .

The assumption of serially correlated exchange rate and bounded innovations is equivalent
to a model in which the mean of the exchange rate shifts up or down over time. To
characterize the impact on the optimal hedge of changes in s, let the first order condition
be written as ϕ (xf , s) = 0. Because the value function is concave, for an interior optimum
the sign of dx∗f /ds is given by that of ∂ϕ/∂s. First, by change of variables:
Z ṽ
1
ϕ (xf , s) = −φ (1 + r) (1 − τ 1 ) − (τ 1 − τ 2 ) ((1 + r) φ + v) γ (v) dv,
1+r −(1−ρ)µ

where ṽ = s̃f − ρs − (1 − ρ) µ. It is now easy to see that

∂ϕ 1 f −c
= − (τ 1 − τ 2 ) ρ¡ ¢ γ(ṽ).
∂s 1 + r 1 − x∗f 2

Therefore, with positive persistence (ρ > 0):

½
dx∗f ≥ 0, f ≤ c
.
ds ≤ 0, f ≥ c

19
Some Additional Comparative Statics Aside from when f − c = 0, it must be
that x∗f < 1. Thus, let f − c 6= 0, in the following comparative statics exercise. Totally
differentiating (11), and after some simplifications, we obtain:
"µ ¶ ¡ ¢ #
∗ 2
dx∗f 1+r ¡ ¢ 1 − τ 1 1 − xf
=− 1 − x∗f + H(s̃f ; s) − x∗f ,
dφ f −c τ1 − τ2 (f − c) h(s̃f ; s)

which is negative if f < c, and

½
dx∗f 1 − x∗f > 0, f − c > 0
= .
dc f −c < 0, f − c < 0

C Discussion of Tax Issues Related to Hedging

Convexity of the US Corporate Tax Schedule. The usual theoretical presentation

of the tax hypothesis focuses on strict convexity of the tax schedule, despite the fact that
most firms in the US face a flat corporate income tax. In fact, the evidence suggests that
the US corporate income tax schedule is convex, but Graham and Smith (1998, page
19) estimate that “the asymmetric treatment of profits and losses drives much of the
observed convexity”. This asymmetric treatment comes from the use of tax loss carry-
backs and carry-forwards. Other features normally associated with the convexity of the
tax schedule are: (i) a small interval of progressive taxation (the lowest tax rate is 15%
for corporations with positive income lower than $50,000, and the highest tax rate is 34%
for incomes higher than $100,000); (ii) the foreign tax credit, and general business credit;
and (iii) the alternative minimum tax. Another more subtle form of progressive taxation
is the possibility of a firm becoming regulated if its value grows unexpectedly large (see
Smith and Stulz, 1985). Furthermore, Graham and Smith simulate their model and find
that the convexity of the tax schedule can generate significant savings from hedging to
an average firm: a reduction on the volatility of taxable income by 5% brings about a
5.4% reduction in the expected tax liability. As of August 1997, the carry-back period of
operating losses was shortened from 3 to 2 years and the carry-forward period extended
from 15 to 20 years (U.S. Master Tax Guide, 1998). These changes in the tax code
were not considered in Graham and Smith (1998). We suspect that the increase in the
carry-forward period, being too far off in the future, is not as important as the reduction
in the carry-back period. If this is correct these recent changes increased the convexity
of the tax schedule. The quantitative importance of the asymmetric treatment of losses
versus gains in the U.S. tax schedule has been detected also in past tax codes. Altshuler
and Auerbach (1990) show, for the 1982 tax code, that the existence of tax loss carry-
forwards alone can significantly lower the marginal income tax rate. Depending on the

20
income range of a particular firm tax loss carry-backs can reduce the marginal income
tax rate by 18%-26% whereas tax loss carry-forwards can reduce the marginal income
tax rate by 41%-48%.

Reporting of Hedging Profits. Kramer and Heston (1993, page 74) discuss when
are gains/losses from hedging reported as ordinary versus capital income. The October
1993 decision by the Supreme Court over Arkansas Best Corp. vs. Comm’r determines
that “hedges of any ‘ordinary’ property or obligation will give rise to ordinary gains or
losses,” as long as the taxpayer is able to identify what is being hedged. Since we are
analyzing transaction exposure this is certainly possible. The main difference between
capital and income gains/losses is that the capital losses can only be deducted to capital
gains. Otherwise, capital gains are treated as ordinary income when computing taxable
income (see also Kramer (1997)). The IRS defines hedging transactions as those that
reduce the risk of currency fluctuations with respect to ordinary income, written options,
and trades that counteract existing hedging transactions (paragraph 1.1221-2 of the 26
CFR, Chapter 1, 1997 edition).
For practical purposes, we are assuming that the firm uses ‘hedge accounting’ to
report these transactions as opposed to ‘mark-to-market’ accounting. This means that
the firm only has to report any gains or losses from trading in forwards at the time the
hedged exposure is recorded, instead of reporting any changes in value as they occur (see
DeMarzo and Duffie 1995, and Moffett and Skinner 1995).

21
References
[1] Adam, Tim, “Do Firms Use Derivatives to Reduce their Dependence on External
Capital Markets?,” Hong Kong University of Science and Technology, working paper,
1999.

[2] Altshuler, Rosanne and Alan J. Auerbach, “The Significance of Tax Law Asymme-
tries: An Empirical Investigation,” Quarterly Journal of Economics, 61-86, 1990.

[3] Andrade, Gregor, and Steven Kaplan (1998), “Evidence from Highly Leveraged
Transactions that Became Distressed,” Journal of Finance, 53: 1443-1494.

[4] Benartzi, Shilomo, and Richard Thaler, “Myopic Loss Aversion and the Equity
Premium Puzzle, ” Quarterly Journal of Economics, 110: 73-92, 1995.

[5] Bessembinder, Hendrik, “Forward Contracts and Firm Value: Investment Incentive
and Contracting Effects,” Journal of Financial and Quantitative Analysis, 26: 519-
532, 1991.

[6] Bodnar, Gordon M., and Günther Gebhardt, “Derivatives Usage in Risk Manage-
ment by US and German Nonfinancial Firms: A Comparative Survey,” NBER Work-
ing paper 6705, 1998.

[7] DeMarzo, Peter M., and Darrel Duffie, “Corporate Incentives for Hedging and Hedge
Accounting,” The Review of Financial Studies, 8:743-771, 1995.

[8] Engel, Charles, 1996, “The Forward Discount Anomaly and the Risk Premium: a
Survey of Recent Evidence,” Journal of Empirical Finance 3: 123-192.

[9] Graham, John R., and Clifford W. Smith Jr., “Tax Incentives to Hedge,” forthcom-
ing Journal of Finance, 1998.

[10] Jesswein, Kurt, Chuck C. Y. Kwok, and W. R. Folks, Jr., “What New Currency
Products are Companies Using and Why?,” Journal of Applied Corporate Finance,
8: 103-114, 1995.

[11] Kahneman, Daniel, and Amos Tversky, “Prospect Theory: An Analysis of Decision
Under Risk,” Econometrica 47:263-291, 1979.

22
[12] Kramer, Andrea S. and J. Clark Heston, “An Overview of Current Tax Impediments
to Risk Management,” Journal of Applied Corporate Finance, 6:73-80, 1993.

[13] Kramer, Andrea S., Financial Products: Taxation, Regulation, and Design,” 1997
Cumulative Supplement. John Wiley and Sons, Inc. 1997.

[14] Leland, Hayne E., “Agency Costs, Risk Management, and Capital Structure,” Jour-
nal of Finance 53, 1998.

[15] Mello, A. and J. Parsons (1999), “Hedging and Liquidity, ” Review of Financial
Studies, forthcoming.

[16] Millar, William (1989), “New Directions in Financial Risk Management,” Strategic
Risk Management. Business International.

[17] Moffet, Michael H., and Douglas J. Skinner, “Issues in Foreign Exchange Hedge
Accounting,” Journal of Applied Corporate Finance, 8:82-94, 1995.

[18] Rawls, S. W., and C. W. Smithson, “Strategic Risk Management,” in The New Cor-
porate Finance, Where Theory Meets Practice, Donald H. Chew, Jr. ed., McGraw-
Hill, 1993.

[19] Ross, Michael, “Corporate Hedging: What, Why, and How?,” Working paper, Uni-
versity of California at Berkeley, 1996.

[20] Smith, Clifford W., and René M. Stulz, “The Determinants of Firm’s Hedging Poli-
cies,” Journal of Financial and Quantitative Analysis, 4:391-405,1985.

[21] Stokey, Nancy L., Rober E. Lucas, with Edward C. Prescott, Recursive Methods in
Economic Dynamics, Harvard University Press, 1989.

[22] Stulz, René M., “Rethinking Risk Management,” Journal of Applied Corporate Fi-
nance, 9:8-24, 1996.

[23] Titman, Sheridan, “Interest Rate Swaps and Corporate Financing Policies,” Journal
of Finance XLVII:1503:1516, 1992.

[24] Tversky, Amos, and Daniel Kahneman, “Advances in Prospect Theory: Cumulative
Representation of Uncertainty, ” Journal of Risk and Uncertainty 5:297-323, 1992.

[25] U.S. Master Tax Guide, 1998.

23