Beruflich Dokumente
Kultur Dokumente
1 • FEBRUARY 2018
ABSTRACT
I develop a dynamic model of leverage with tax deductible interest and an endoge-
nous cost of default. The interest rate includes a premium to compensate lenders for
expected losses in default. A borrowing constraint is generated by lenders’ unwilling-
ness to lend an amount that would trigger immediate default. When the borrowing
constraint is not binding, the trade-off theory of debt holds: optimal debt equates
the marginal interest tax shield and the marginal expected cost of default. Contrary
to conventional interpretation, but consistent with empirical findings, increases in
current or future profitability reduce the optimal leverage ratio when the trade-off
theory holds.
∗ Andrew B. Abel is with the Department of Finance of the Wharton School of the University
of Pennsylvania and the National Bureau of Economic Research. I thank Jeff Campbell, Marco
Giometti, Vincent Glode, Joao Gomes, Christian Goulding, Richard Kihlstrom, Bob McDonald,
Adriano Rampini, Scott Richard, Michael Roberts, Ali Shourideh, Toni Whited, and seminar par-
ticipants at the Federal Reserve Bank of Chicago, Kellogg Finance, MIT Finance, Shanghai Ad-
vanced Institute of Finance, Wharton Finance, the Tepper/LAEF Conference on Macro-Finance,
and the Tel Aviv University Finance Conference for helpful comments and discussion. I also thank
the Editor, Associate Editor, and two referees for helpful comments. Part of this research was
conducted while I was a Visiting Scholar at Chicago Booth and I thank colleagues there for their
hospitality and support. I have read the Journal of Finance’s disclosure policy and have no conflicts
of interest to disclose.
1 See Robichek and Myers (1966), Kraus and Litzenberger (1973), and Scott (1976).
2 See Scott (1976), Fama and French (2002), and Frank and Goyal (2008).
3 See Fama and French (2002).
DOI: 10.1111/jofi.12590
95
96 The Journal of FinanceR
have lower leverage ratios.4 Notable exceptions to this interpretation are pre-
sented in quantitative models by Hennessy and Whited (2005) and Strebulaev
(2007). In the stochastic dynamic model presented in this paper, I analytically
demonstrate an alternative explanation of the negative relationship between
profitability and leverage that is found empirically.
In this paper, I develop and analyze a model of capital structure that in-
corporates an interest tax shield as well as the possibility of default. In some
situations, the optimal amount of debt is determined by the equality of the
marginal benefit of debt arising from the interest tax shield and the marginal
cost of debt associated with increased probability of default—that is, optimal
debt will be characterized by the trade-off theory. However, in other situa-
tions within the model, optimal debt is not characterized by the equality of the
marginal benefit and the marginal cost that epitomizes the trade-off theory.
Because the model allows for situations in which the trade-off theory holds
and situations in which it does not hold, it has the potential to guide empirical
tests by including both a null hypothesis in terms of the trade-off theory and
an alternative hypothesis that offers an explanation of leverage other than the
trade-off theory. In particular, I show that the model developed here accom-
modates situations in which higher profitability (either current profitability or
expected future profitability) is associated with lower leverage, specifically, a
lower value of market leverage, which is the ratio of debt to the market value of
the firm. Furthermore, if the probability of default is nonzero, these situations
arise when and only when the trade-off theory is operative. Thus, the empirical
finding that more profitable firms tend to have lower leverage ratios, which has
been interpreted by others as evidence against the trade-off theory, is viewed
as evidence in favor of the trade-off theory when seen through the lens of the
model presented here.
As the trade-off theory has been developed over the past half century, it
has become increasingly complex, especially in empirical structural models of
the firm designed to capture realistic features of a firm’s environment.5 The
model I develop here is stripped of these complexities so that I can focus on
its new features and implications in a framework that admits analytic results
without relying on numerical solution. The model’s biggest departure from
standard models of debt concerns the maturity of debt. Many standard models
of debt6 assume that debt has infinite maturity and pays a fixed coupon over the
infinite future, or until the firm defaults. The assumption of infinite maturity
is clearly extreme, but it has been used productively over the years. I also make
an extreme assumption about maturity, but in the opposite direction: I assume
that debt must be repaid an instant after it is issued. The standard specification
with infinite maturity can be viewed as the limiting case of long-term debt,
4 For instance, Myers (1993, p. 6) states that “The most telling evidence against the static trade-
off theory is the strong inverse correlation between profitability and financial leverage . . . . Yet the
static trade-off story would predict just the opposite relationship. Higher profits mean more dollars
for debt service and more taxable income to shield. They should mean higher target debt ratios.”
5 See Hennessy and Whited (2007).
6 See Modigliani and Miller (1958), Leland (1994), and Gorbenko and Strebulaev (2010).
Optimal Debt and Profitability in the Trade-off Theory 97
while my specification with zero maturity can be viewed as the limiting case of
short-term debt, such as commercial paper, much of which has a maturity of
only one to four days,7 as well as overnight repurchase agreements.
The specification of zero-maturity debt is motivated by two considerations.
First, it makes salient the recurrent nature of the financing decision, in contrast
to the once-and-for-all financing decision in many models of debt.8 At each
instant, the firm decides whether to repay its debt or to default; if it decides to
repay its debt, it also chooses the amount of debt to issue anew. Because I do not
include any flotation, issuance, or adjustment costs, the amount of debt issued
responds immediately and completely to changes in the firm’s environment,
and hence does not have the rich dynamics documented and analyzed by Leary
and Roberts (2005). Second, zero-maturity debt is always valued at par, which
alleviates the need to calculate the value of debt that would arise with long-
term debt. Therefore, the firm’s decision about whether to default on its debt,
which depends on a comparison of the total value of the firm and the value of
the firm’s debt, becomes transparent.
Because firms can default on their debt, rational lenders need to account
for the probability of default, as well as their losses in the event of default,
to determine the appropriate interest rate on loans to the firm. In this paper,
risk-neutral lenders have the same information as the firm’s shareholders, and
they require a premium above the riskless rate to compensate for their expected
losses in the event of default. In addition, if the amount of debt is sufficiently
large, it will trigger immediate default. In the current framework with zero-
maturity debt, lenders avoid being subject to immediate default by refusing
to lend an amount greater than the contemporaneous value of the firm, which
itself depends on the amount of debt issued.
An important component of the firm’s financing decision is the stochastic
process for the firm’s pretax preinterest cash flow, that is, earnings before inter-
est and taxes (EBIT). I specify a continuous-time continuous-state process for
EBIT. Instead of a diffusion process, as in Leland (1994), for example, I specify a
Markov process in which EBIT remains unchanged for a random length of time
and then a new value of EBIT arrives at a date governed by a Poisson process
with arrival intensity λ. On these dates, the value of EBIT changes by a discrete
amount, and a discrete decrease in the value of EBIT can lead the firm to default
on its debt. In the event of default, a fraction α > 0 of the firm’s value disappears
as a deadweight loss and creditors take ownership of the remaining fraction
1 − α of the firm. I do not consider renegotiation between shareholders and
creditors. Shareholders and creditors have common knowledge. If they were to
renegotiate when a new realization of EBIT leads to default, the result of that
renegotiation would simply be a function of the amount of debt and the level of
7 In March 2016, the average daily issuance of nonfinancial AA commercial paper was 210
issues, which amounted to $5.3 billion. Of these issues, 53%, which accounted for 55% of the dollar
volume, had a maturity of one to four days. Source: Board of Governors of the Federal Reserve
System, Volume Statistics for Commercial Paper Issuance, Data as of April 12, 2016.
8 See Merton (1974), Leland (1994), and Gorbenko and Strebulaev (2010).
98 The Journal of FinanceR
EBIT at that time. But if shareholders and lenders were to agree on terms that
would apply when a new value of EBIT leads to default, they could simply write
those terms into the debt contract, in which case the debt contract would become
state-contingent in a way that would make the instrument not look like debt.
Since I want to focus on the leverage ratio as measured using conventional debt,
I do not consider renegotiation or the sort of “prenegotiation” just described.9
For simplification, the new values of EBIT are i.i.d. across arrivals. Never-
theless, EBIT displays persistence: EBIT remains constant during each regime,
and regimes have a mean duration of 1λ , which could potentially be quite large.
The unconditional correlation between two values of EBIT that occur x > 0
units of time apart is e−λx , which is positive and declines monotonically in x.
The advantage of using the i.i.d. specification is that it allows for persistence
in EBIT as well as analytic tractability.10 In particular, as I show below, the
optimal amount of debt is invariant to contemporaneous EBIT if the value
of EBIT is high enough that the trade-off theory of debt is operative. As a
result, the leverage ratio, which is the ratio of debt to the value of the firm,
is a decreasing function of EBIT, since the value of the firm is increasing in
EBIT. The stark result that optimal debt is invariant to EBIT when the trade-
off theory is operative is a consequence of the assumption that EBIT is i.i.d.
across regimes. In Section VII, I extend the model to allow for persistence in
EBIT across regimes. I show that optimal debt is no longer invariant to EBIT
when the trade-off theory is operative. I present a simple example in which
the conditional distribution of EBIT is uniform and show that, in this example,
optimal debt and the leverage ratio are both decreasing in contemporaneous
EBIT, which reinforces the major result of the paper. I also show that the con-
ditional probability of default is a decreasing function of EBIT when and only
when the trade-off theory is operative.
When EBIT is i.i.d. across regimes, the firm’s optimal capital structure de-
pends on whether EBIT exceeds or falls short of an endogenously determined
critical value. For values of EBIT that exceed the critical value, the trade-
off theory is operative, provided that the firm faces a positive probability of
default. When the trade-off theory is operative, the optimal level of debt is in-
variant to EBIT; since an increase in EBIT increases the total value of the firm,
the optimal leverage ratio falls as profitability, measured by EBIT, increases.
This negative relationship between profitability and the leverage ratio, which
9 The debt contract that I analyze is not an optimal contract, but rather is of the form commonly
used in practice and in analytical models. Although it is not optimal, it does have the important
feature that all participants in the debt instrument participate voluntarily and are not made worse
off by participating. Biais et al. (2007) derive “performance sensitive” debt as an optimal contract
in a dynamic framework.
10 In their analysis of the impact of temporary shocks on optimal debt issuance, Gorbenko and
Strebulaev (2010) specify temporary shocks as Poisson arrivals from a given distribution; when
they consider multiple shocks over time, they assume that they are i.i.d., as in the framework used
here. The temporary shocks in Gorbenko and Strebulaev are drawn from a double exponential dis-
tribution. In contrast, the shocks in the current model are drawn from a more general distribution
on finite support for which the density is finite and nondecreasing and the restriction in equation
(2) is satisfied.
Optimal Debt and Profitability in the Trade-off Theory 99
11 To the extent that creditors recover a portion of their loans in the event of default, the social
cost of default is smaller than the loss to current shareholders. I assume that creditors receive a
fraction 1 − α of the continuation value of the firm in the event of default, so the social cost of default
is a fraction α of the continuation value. Therefore, the social cost of default, as well as private
cost to current shareholders, is increased by a favorable shift of the unconditional distribution of
profitability.
12 Section II in the Internet Appendix, which is available in the online version of this article
on the Journal of Finance website, shows that this finding holds more broadly within the class of
truncated exponential distributions for density functions that do not slope upward too steeply. Only
for sufficiently steeply upward-sloping density functions will optimal debt increase in response to
a rightward translation of the distribution function when the trade-off theory is operative.
100 The Journal of FinanceR
of shareholders’ equity, which together equal the total valuation of the firm.
An important component of this analysis is the critical value of current EBIT,
which is the boundary between low values of EBIT for which the borrowing con-
straint binds and high values of EBIT for which the borrowing constraint does
not bind. Depending on whether this critical value of EBIT is at the minimum
value of the support of EBIT, the maximum value of the support of EBIT, or in
between, the firm will find itself in one of three scenarios, which I characterize
in Section III. Although optimal behavior in two of the three scenarios can be
derived easily, optimal behavior in one of the scenarios (Scenario II) entails
more extensive analysis, which I present in Section IV. Of particular interest
is Section IVA, which interprets the analytic results in terms of the trade-off
theory. Section V demonstrates that whether Scenario I, II, or III will prevail
depends on whether the tax rate is low, intermediate, or high, and it provides
explicit expressions for the values of the tax rate associated with the bound-
aries between adjacent scenarios. Section VI analyzes the impact of a rightward
translation of the unconditional distribution of EBIT on optimal debt, share-
holder equity, the critical value of EBIT, and the probability of default. Section
VII extends the model to allow for persistence in EBIT across regimes. In the
focal case of a uniform distribution, persistence strengthens the major result
of the paper that the optimal leverage ratio is a decreasing function of EBIT
if the trade-off theory is operative. Section VIII presents concluding remarks.
To avoid disruption in the narrative flow of the paper, the proofs of all lemmas,
propositions, and corollaries are in Appendix A. Appendix B presents details on
the calculation of the value of the firm. An Internet Appendix presents closed-
form solutions for optimal debt and the value of the firm in the special case in
which the unconditional distribution of EBIT is uniform and creditors recover
zero in the event of default. It also contains a section analyzing the impact
of a rightward translation of the unconditional distribution of EBIT when the
distribution is a truncated exponential distribution.
13 In contrast to this assumption, Gorbenko and Strebulaev (2010) allow the unlevered value of
the firm to become negative, in which case shareholders would abandon the firm.
14 Leland (1994, p. 1217) effectively makes the same assumption by specifying a stochastic
∞
process for the “asset value,” which corresponds to W (φ(t)) ≡ Et { t (1 − τ )φ(s)e−ρ(s−t) ds}, that is
bounded away from zero.
102 The Journal of FinanceR
ownership interest in the firm and receive zero. Creditors take ownership of
the firm in default, but the act of default imposes a deadweight loss equal to
a fraction α of the value of the firm, where 0 < α ≤ 1. Therefore, in default
at time t + dt, creditors receive (1 − α)V (φ(t + dt)), where V (φ(t + dt)) is the
value of the firm, which I describe more formally in equation (7) below. Since
creditors lose the value of the bonds, Dt , but recover (1 − α)V (φ(t + dt)) in
default, the expected
loss to creditors due to default in the interval from t
to t + dt is λ V (φ(t+dt))<Dt [Dt − (1 − α)V (φ(t + dt))]dF(φ(t + dt)). Therefore, the
interest rate, r(t), that compensates lenders for the expected loss in default is
V (φ)
r(t) = ρ + λ 1 − (1 − α) dF(φ) ≥ ρ, (3)
V (φ)<Dt Dt
where the first term, ρ, is the riskless rate in the absence of any chance of
default, and the second term is the expected loss to creditors in default for each
dollar of debt they hold.
In addition to requiring an interest rate that includes compensation for the
expected loss in default, potential lenders to the firm never lend an amount
that is so large that the firm would default immediately upon receiving the
funds from issuing the bond. This limit on the amount the firm can borrow is
Dt ≤ V (φ(t)). (4)
The tax rate on taxable income is constant and equal to τ , where 0 < τ < 1. If
taxable income is negative, the firm receives a tax rebate of −τ y(t) > 0.
I assume that shareholders do not retain any earnings in the firm, that is,
the firm pays out all net cash flows as dividends, dX(t), so
where dDt is the net increase in bonds at time t, which represents a net inflow
of funds to the firm. The notations dX(t) and dDt allow dividends and net
increases in bonds to be discrete amounts, rather than flows, at a point in time.
For instance, as I show below, the arrival of a new regime that increases φ(s)
by a discrete amount may increase the amount of bonds by a discrete amount
at a point in time, so that dDt and dX(t) are both positive.
Optimal Debt and Profitability in the Trade-off Theory 103
where T is the date at which shareholders decide to default on their debt, and
hence is endogenous. In equation (7), V (φ(t)) is the value of the firm at time t
if it arrives at time t with zero bonds outstanding. Equivalently, V (φ(t)) is the
value of the firm at time t immediately after it has paid off (with interest) its
outstanding debt issued at t − dt and before it issues new debt at time t.
To describe the timing of events around time t, it is useful to define Dt− ≡
limdt 0 Dt−dt as the amount of bonds issued by the firm an instant before time
t. The timing of events can then be summarized as follows.
where t1 is the date of the first arrival of a new regime after time t . The value of
the firm in equation (8) is the sum of two terms. The first term is the expected
present value of dividends, dX, over the interval between time t and the arrival
of a new regime at time t1 . The second term is the expected present value of the
firm at time t1 , after observing the new value of profitability, φ(t1 ). When the
15 As it turns out, the firm will change the amount of outstanding debt only at times of regime
change, when φ(s) changes. At these times, the amount of debt will change by a discrete amount,
dDt .
104 The Journal of FinanceR
firm observes φ(t1 ), it can choose to become a firm with value V (φ(t1 )), but to
do so, it must repay its outstanding debt Dt1− , so the net value to shareholders
of continuing their ownership in the firm is V (φ(t1 )) − Dt1− . Provided that this
net value of continued ownership of the firm is nonnegative, shareholders will
repay their outstanding bonds and issue a new amount of bonds, Dt1 . Otherwise,
the firm will default on its bonds and shareholders lose their ownership of the
firm.
Other than deciding whether to default, the only decision that the firm makes
is how many bonds to issue at each point in time. The structure of the stochastic
process for φ(t) implies that the optimal value of Dt is simply a function of
contemporaneous φ(t). I use the notation D(φ(t)) to denote the optimal value
of Dt . Profitability is constant, and equal to φ(t), over the interval between
times t and t1 , so after-tax income, (1 − τ )y(s), and the optimal amount of
bonds outstanding, D(φ(s)), are constant and equal to (1 − τ )y(t) and D(φ(t)),
respectively, for all s in [t, t1 ). With
D(φ(s)) constant over this interval of time,
dDs = 0 for all s in (t, t1 ), but dDt =
D(φ(t)) for a firm at time t after it has repaid
its debt Dt1− .
It is straightforward but tedious to calculate the value of the firm, so I
relegate the derivation to Appendix B, where I show that
1
V (φ(t)) = max [(1 − τ )φ(t) + λv + A(Dt )], (9)
ρ+λ Dt ≤V (φ(t))
where
A(D) ≡ τ ρ + λ dF(φ) D − [α + τ (1 − α)]λ V (φ)dF(φ), (10)
V (φ)<D V (φ)<D
and v ≡ V (φ)dF(φ) is the unconditional mean of V (φ). To interpret the func-
tion A(D) in equation (10), which arises naturally in the derivation of the value
of the firm, use equation (3) to rewrite equation (10) as
A(Dt ) = τr(t)Dt − αλ V (φ)dF(φ). (11)
V (φ)<Dt
I refer to A(D) as the “trade-off function” because it contains the elements of the
trade-off theory of debt. The first term on the right-hand side of the expression
for A(Dt ) in equation (11), τr(t)Dt , is the interest tax shield, which is simply
the tax rate, τ , multiplied by interest payments, r(t)Dt . The second term is the
expected deadweight loss associated with default, where the deadweight loss
is αV (φ) if default occurs when profitability equals φ.
Equation (9) can be expressed in terms of flows at time t by multiplying
both sides by ρ + λ and then subtracting λV (φ(t)) from both sides to obtain the
Hamilton-Jacobi-Bellman equation
The left-hand side of equation (12), ρV (φ(t)), is the required return on the
firm, which is the product of the rate of return required by shareholders and the
value of the firm. The expected return to the firm, on the right-hand side, has
three components: the after-tax profit from operations, (1 − τ )φ(t); the expected
change in the value of the firm if the regime changes, which is the product
of λ, the instantaneous probability of a regime change, and v − V (φ(t)), the
expected change in the value of the firm when a new value of φ is drawn from
its unconditional distribution; and A(Dt ), the contribution of bond financing,
which consists of the interest tax shield, τr(t)Dt , less the expected deadweight
loss associated with default.
The optimal value of debt is the value of Dt that attains the maximum on the
right-hand side of equation (12). Since Dt enters the maximand on the right-
hand side of equation (12) only through A(D), the optimal amount of debt is16
D(φ(t)) ≡ arg max A(Dt ). (13)
Dt ≤V (φ(t))
Since A(Dt ) is independent of φ(t), the optimal value of debt, D(φ(t)), is indepen-
dent of φ(t) if the borrowing constraint, Dt ≤ V (φ(t)), is not binding. However, if
the borrowing constraint is binding, then D(φ(t)) = V (φ(t)), which is increasing
in φ(t), since, as shown in Proposition 1 below, V (φ(t)) is strictly increasing in
φ(t).
To obtain the value function, substitute the optimal amount of debt, D(φ(t)),
from equation (13) into equation (9), which yields
(1 − τ )φ(t) + λv + A(
D(φ(t)))
V (φ(t)) = . (14)
ρ+λ
16 Equivalently, the optimal value of D attains the maximum on the right-hand side of equation
t
(9), which, since ρ + λ > 0, is also given by equation (13).
106 The Journal of FinanceR
Proposition 2 states that the optimal value of debt never falls below previous
optimal values of debt as long as the firm does not default. This behavior is
consistent with the time path of debt in Goldstein, Ju, and Leland (2001), in
which it is assumed that, outside of default, the firm does not decrease its
debt.17
Since shareholders owe a liability of
D(φ(t)), the value of shareholders’ equity
in the firm, S(φ(t)), is
S(φ(t)) = V (φ(t)) −
D(φ(t)) ≥ 0, (15)
17 The second paragraph of Goldstein, Ju, and Leland (2001, p. 483) begins “Below, we consider
only the option to increase future debt levels. While in theory management can both increase and
decrease future debt levels, Gilson (1997) finds that transactions costs discourage debt reductions
outside of Chapter 11.”
Optimal Debt and Profitability in the Trade-off Theory 107
The strict concavity of A(D) in D implies that D∗ is unique. Since A(D) is inde-
pendent of φ(t), D∗ is invariant to φ(t). If D∗ ≤ V (φ(t)), the borrowing constraint
does not bind, and optimal debt, D(φ(t)), equals D∗ . However, if D∗ > V (φ(t)),
the firm cannot borrow as much as D∗ and D(φ(t)) = V (φ(t)).
Define φ ∗ as the critical value of φ ∈ [ L, H ] above which V (φ)>D∗ and
below which V (φ)<D∗ . Formally,
φ ∗ ≡ V −1 (D∗ ). (17)
Figure 1 illustrates optimal debt, shareholders’ equity, and the total value
of the firm, each as a function of φ, for V ( L) < φ ∗ < V ( H ). For φ(t) ≤ φ ∗ ,
condition for this strict concavity is that default imposes a deadweight cost, that is, α > 0. To see the
necessity of this condition, set α = 0 in the definition
of A(D) in equation (10) and differentiate twice
with respect to D to obtain A (D) = τ [ρ + λ V (φ)<D dF(φ)] and A (D) = τ λV −1 (D) f (V −1 (D)) > 0
because f (V −1 (D)) > 0 and V −1 (D) > 0. Therefore, if α = 0, then A(D) is strictly convex in D.
Thus, α > 0 is a necessary condition for strict concavity of A(D) in D.
108 The Journal of FinanceR
Figure 1. Optimal debt, equity, and firm value. (Color figure can be viewed at wileyonlineli-
brary.com)
studies find a negative relationship between the leverage ratio and productiv-
ity, consistent with Corollary 3 when the trade-off theory is operative.
Proposition 2 states that, outside of default, D(φ(t)) is a weakly increasing
function of the state φ(t) and of time t. The leverage ratio, however, is weakly
decreasing in the state φ(t) and is not monotonic in time t. To see that the lever-
age ratio can either increase or decrease over time, consider several consecutive
regimes in which φ(t) ≥ φ ∗ . The optimal value of bonds, D(φ(t)) , remains con-
stant and equal to D∗ over time since φ(t) ≥ φ ∗ . When φ(t) increases from one
of these regimes to the next regime, V (φ(t)) increases and the leverage ratio
falls; when φ(t) decreases from one of these regimes to the next regime, V (φ(t))
falls and the leverage ratio increases.
Under the optimal debt policy, D(φ(t)), the instantaneous probability of de-
fault is
P(
D(φ(t))) ≡ λ dF(φ), (18)
V (φ)<
D(φ(t))
which is the probability that a new regime arrives that induces a value of the
firm lower than the currently outstanding debt,
D(φ(t)).
PROPOSITION 6: The instantaneous probability of default P(
D(φ(t))) =
λ min[F(φ(t)), F(φ ∗ )].
In the framework analyzed here, default occurs only at times of regime
change.19 If φ(t) < φ ∗ , then D(φ(t)) = V (φ(t)), so default occurs at the next
regime change if the new value of profitability is lower than φ(t), in which
case the new value of the firm is less than V (φ(t)) = D(φ(t)). Alternatively,
if φ(t) ≥ φ ∗ , then
D(φ(t)) = D∗ = V (φ ∗ ), so default occurs at the next regime
change if the new value of profitability is less than φ ∗ , in which case the new
value of the firm is less than V (φ ∗ ) = D(φ(t)). Thus, regardless of whether φ(t)
is above or below the critical value φ ∗ , default occurs at the next regime change
if and only if the new value of profitability is less than min[φ(t), φ ∗ ].
PROPOSITION 7: The instantaneous probability of default, P(
D(φ(t))), and the
interest rate, r(t), are
In principle, there are two endogenous channels that can increase the prob-
ability of default at the time of the next regime change, t : channel (1) is an
increase in the current amount of bonds outstanding, D(φ(t)), and channel (2)
is a shift in the distribution of φ(t ) that shifts the distribution of V (φ(t )) in an
unfavorable way. Statement 1 of Proposition 7 captures channel (1). Channel
19 The model in Gorbenko and Strebulaev (2010) has a similar feature. Specifically, in their
model, default occurs only when a new value of the temporary shock has a Poisson arrival.
110 The Journal of FinanceR
A (D∗ ) = ω2 − ω1 , (20)
and
ω2 (V ( H ) − D∗ ) = 0. (22)
A. Scenario I
In Scenario I, ω1 > 0 and ω2 = 0, so the first-order condition in equation
(20) implies A (D∗ ) = −ω1 < 0. Thus, the marginal interest tax shield, which
provides an incentive to increase the amount of debt, is overwhelmed by the
marginal expected cost of losing the continuation value of the firm in default.
The complementary slackness condition in equation (21) implies D∗ = V ( L).
Therefore, the optimal value of debt is V ( L), which is the highest value of
debt that the firm can issue without facing a positive probability of default.
PROPOSITION 9: Define D0 ≡ ρ+λ 1
[ L + ρλ E{φ}]. In Scenario I (where ω1 > 0 and
ω2 = 0), for all φ(t) ∈ [ L, H ],
1.
D(φ(t)) = D∗ = D0 , which is invariant to φ(t) and τ ,
2. S(φ(t)) = ρ+λ
1−τ
(φ(t) − L), and
3. V (φ(t)) = D0 + ρ+λ
1−τ
(φ(t) − L).
B. Scenario III
In Scenario III, ω1 = 0 and ω2 > 0, so the first-order condition in equation
(20) implies that A (D∗ ) > 0. The complementary slackness condition in equa-
tion (22) implies that D∗ = V ( H ). Therefore, since A(D) is strictly concave in
D, A (D) > 0 for all D ∈ [V ( L), V ( H )]. The marginal interest tax shield asso-
ciated with an increase in debt overwhelms the increased exposure to default
associated with an increase in debt. Therefore, the firm will issue as much debt
as it can, driving shareholders’ equity to zero, so that D(φ(t)) ≡ V (φ(t)) for all
φ(t) ∈ [ L, H ].
The expression for the value of the firm at φ(t) = H depends on v as well
as on the exogenous parameters ρ, λ, H , and α. In the special case in which
default destroys the entire value of the firm (α = 1), Proposition 10 implies
V ( H ) = ρ+λ
1
H , which is simply the expected value of the flow of H from the
current time until the next regime change, which triggers default because the
new value of V (φ(t)) will be less than D( H ) = V ( H ). In this special case, firm
value, V ( H ), is completely insulated from taxes.
IV. Scenario II
Suppose that the firm is in Scenario II, so ω1 = ω1 = 0. Then, equation (20)
implies that A (D∗ ) = 0. Differentiating A(D) in equation (10) with respect to D
and evaluating A (D) at D = D∗ yields
A (D∗ ) = τ [ρ + λF(V −1 (D∗ ))] − (1 − τ )αλV −1 (D∗ )D∗ f (V −1 (D∗ )) = 0. (23)
The following lemma provides a simple expression for V −1 (D∗ ), which ap-
pears in the second term in equation (23).
Optimal Debt and Profitability in the Trade-off Theory 113
ρ+λ
LEMMA 1: If ω1 = ω2 = 0, then V (φ ∗ ) = 1−τ
ρ+λ
and V −1 (D∗ ) = 1−τ
, where φ ∗ ≡
V −1 (D∗ ).
Use Lemma 1 to rewrite equation (23) as
A (D∗ ) = τ [ρ + λF(V −1 (D∗ ))] − (ρ + λ)αλD∗ f (V −1 (D∗ )) = 0. (24)
Because A(D) is strictly concave, there is at most one value of D∗ that sat-
isfies equation (24).22 For φ(t) ≥ φ ∗ ≡ V −1 (D∗ ), the borrowing constraint does
not bind and the optimal amount of debt is D∗ ; for φ(t) < φ ∗ ≡ V −1 (D∗ ), the
borrowing constraint binds and the optimal value of debt equals the value of
the firm, V (φ(t)).
be the cost of issuing D units of bonds, and assume that c(0) = 0, c (D) ≥ 0, and c (D) ≥ 0 for D ≥ 0.
Subtracting these flotation costs from operating profits, the value of the firm in equation (9) can
be written as V (φ(t)) = ρ+λ 1
max Dt ≤V (φ(t)) [(1 − τ )φ(t) + λv +
A(Dt )], where
A(Dt ) ≡ A(Dt ) − c(Dt ) is a
“modified trade-off function.” Since A(Dt ) is strictly concave and c(Dt ) is convex, the modified trade-
off function,
A(Dt ), is strictly concave. Define
D∗ ≡ max Dt ≤V ( H )
A(Dt ). Suppose that c (V ( L)) <
τρ, so that if Dt ≤ V ( L), then A (Dt ) = A (Dt ) − c (Dt ) = τρ − c (Dt ) > 0, where the first equality
follows from the definition of
A(Dt ), the second equality follows from the definition of A(D), which
implies A(D) ≡ τρ D for D ≤ V ( L), and the final inequality follows from c (V ( L)) < τρ and the
convexity of c(D). Therefore, V ( L) <
D∗ ≤ V ( H ) and for any φ(t) ≥ V −1 (
D∗ ), the optimal amount
of debt,
D(φ(t)), equals
D∗ and hence is invariant to φ(t). The optimal amount of debt for these
values of φ(t) satisfies
A (
D(φ(t))) = 0, which can be viewed as a “modified trade-off theory.” Note
that, when
A (D) = 0, A (D) = c (D) > 0. In this case, the marginal interest tax shield is equal to
the marginal expected cost of default, plus the marginal flotation cost, c (D). When this modified
trade-off theory holds, the optimal leverage ratio will be a declining function of φ(t), as in the text.
114 The Journal of FinanceR
∗
dt of time by 1 − (1 − τ )(ρ + λF(φ ∗ ))dt − e−ρdt e−λF(φ )dt , which equals the left-
hand side of equation (25) for small dt. Thus, the left-hand side of equation
(25) is the marginal interest tax shield associated with an additional dollar of
debt.
The right-hand side of equation (25) is the marginal cost associated with the
increased probability of default resulting from an additional dollar of debt. By
increasing the probability of default, a one-dollar increase in debt increases
the default premium, and hence increases the interest rate, ρ + λF(φ ∗ ), paid
by the firm. To measure the increase in the probability of default, define ψ(D)
as the threshold value of φ that triggers default when outstanding debt is
D. Formally, D ≡ V (ψ(D)), so 1 = V (ψ(D))ψ (D). Since V (φ ∗ ) = 1−τ r+λ
(Lemma
∗ ρ+λ
1), ψ (D ) = 1−τ . Therefore, a one-unit increase in D increases the threshold
level of φ at which default occurs by ψ (D∗ ) = ρ+λ 1−τ
and thus increases the prob-
ability of default by λ f (φ ∗ )ψ (D∗ ) = λ f (φ ∗ ) ρ+λ
1−τ
, which, in the case in which
∗ ρ+λ
α = 1, increases the interest rate by λ f (φ ) 1−τ and increases the total flow of
after-tax interest payments at time t0 by λ f (φ ∗ )(ρ + λ)D∗ , which is the right-
hand side of equation (25) when α = 1. Therefore, equation (25) represents
the equality of the marginal interest tax shield and the marginal cost associ-
ated with increased exposure to default, which is the essence of the trade-off
theory.
In the more general case in which α ≤ 1, the interpretation of equation (25)
in terms of the trade-off theory is more nuanced. Define R(D, φ0 ) as the flow of
(pretax) interest payments at time t if the amount of outstanding debt is D and
if the firm defaults if and only if φ < φ0 . Then,
φ0
R(D, φ0 ) ≡ (ρ + λF(φ0 ))D − λ(1 − α) V (φ)dF(φ), (26)
L
so that R(Dt , ψ(Dt )) = r(t)Dt , where r(t) is the interest rate in equation (3). If
the value of φ that triggers default, φ0 , were to remain
∗
unchanged, an increase
in debt would increase interest payments by ∂ R(D ∂D
,φ0 )
= ρ + λF(φ0 ). Therefore,
when D = D∗ and φ0 = φ ∗ , a one-dollar increase in D increases the tax shield
associated with interest deductibility by
∂ R(D∗ , φ ∗ )
τ = τ (ρ + λF(φ ∗ )). (27)
∂D
Therefore, the left-hand side of equation (25) is the marginal tax shield.
Now consider the impact on interest payments of an increase in φ0 , holding
the amount of debt unchanged. Partially differentiating equation (26) with
respect to φ0 , evaluating this derivative at φ0 = φ ∗ , and using V (φ ∗ ) = D∗ yields
∂ R(D∗ , φ ∗ )
= αλ f (φ ∗ )D∗ . (28)
∂φ0
Optimal Debt and Profitability in the Trade-off Theory 115
Equation (28) implies that if φ0 were to increase by ψ (D∗ ), as would be the case
if D were to increase by one dollar, after-tax interest payments would increase
by
∂ R(D∗ , φ ∗ ) ∗
(1 − τ ) ψ (D ) = (ρ + λ)αλ f (φ ∗ )D∗ . (29)
∂φ0
The left-hand side of equation (29) is the marginal expected default cost, mea-
sured in flow terms, which reflects the increased probability of default when
D increases by one dollar. Specifically, a one-dollar increase in D increases the
default threshold φ0 by ψ (D∗ ) = ρ+λ1−τ
, thereby increasing total interest costs by
∂ R(D∗ ,φ ∗ ) ∗ ∗
∂φ0
ψ (D ) and after-tax interest costs by (1 − τ ) ∂ R(D
∗
∂φ0
,φ )
ψ (D∗ ), which is the
left-hand side of equation (29). The right-hand side of equation (29) is identical
to the right-hand side of equation (25), so the right-hand side of equation (25)
is the marginal expected cost of default associated with a one-dollar increase
in D. Thus, equation (25) is a statement of the trade-off theory, equating the
value of the interest tax shield associated with an additional dollar of debt, ex-
pressed per unit of time, and the marginal cost of increased exposure to default
resulting from this increased debt, also expressed per unit of time.
(ρ + λ)
D(φ(t)) − A(
D(φ(t))) = (1 − τ )φ(t) + λv, if φ(t) ≤ φ ∗ . (30)
Differentiate both sides of equation (30) with respect to φ(t) to obtain the
ordinary differential equation (ODE)24,25
(ρ + λ − A (
D(φ(t))))
D (φ(t)) = 1 − τ . (31)
The boundary condition for the ODE in equation (31), which takes the form of
a value-matching condition, is
D(φ ∗ ) = D∗ . (32)
24 This ODE is solved in the Internet Appendix in the special case in which α = 1 and the
distribution F is uniform.
25 Differentiate equation (31) with respect to φ(t) to obtain (ρ + λ − A ( D(φ(t))))
D (φ(t)) −
A (
D(φ(t)))(
D (φ(t)))2 = 0. Differentiating A(D) in equation (10) with respect to D implies that
A (D) < τ (ρ + λ) < ρ + λ, so that D (φ(t)) has the same sign as A (
D(φ(t))), which is negative if
α
τ < 1+α , and f (φ) ≥ 0.
116 The Journal of FinanceR
Evaluate equation (31) at φ(t) = φ ∗ and use A (D∗ ) = 0 and the boundary con-
dition in equation (32) to obtain
1−τ
D (φ ∗ ) = . (33)
ρ+λ
Since equation (33) was derived from the ODE that holds for φ ≤ φ ∗ , the
derivative in equation (33) is actually the left-hand derivative. Since V (φ(t)) ≡
D(φ(t)) for L ≤ φ(t) ≤ φ ∗ , equation (33) implies that the left-hand derivative of
V (φ(t)) at φ(t) = φ ∗ is ρ+λ
1−τ
. Proposition 5 implies that the right-hand derivative
∗
of V (φ(t)) at φ(t) = φ is also ρ+λ
1−τ
. Therefore, the right- and left-hand derivatives
∗
of V (φ(t)) at φ(t) = φ are equal to each other.
The value-matching condition in equation (32) is illustrated in Figure 1 at
point L, where the curve through K and L, which represents V (φ) for φ ≤ φ ∗ ,
meets the line segment LM, which represents V (φ) for φ ≥ φ ∗ . As discussed,
the left- and right-hand derivatives of V (φ) are equal to each other at φ = φ ∗ , so
the meeting of the curve through K and L and the line segment LM is smooth,
that is, differentiable, at point L.26
26 This smooth meeting of the curve through K and L and the line segment LM is superficially
similar to the smooth-pasting condition that arises in optimal stopping problems with an under-
lying diffusion process. In those problems, the value-matching and smooth-pasting conditions are
two separate boundary conditions that pin down two parameters in the solution. In the current
framework, the fundamental stochastic variable has finite variation, whereas in optimal stopping
problems with a diffusion process, the stochastic variable has infinite variation. In the current
framework, equality of the left- and right-hand derivatives of V (φ(t)) at φ = φ ∗ arises as a conse-
quence of the value-matching condition. That equality does not impose any additional structure or
restriction on the solution.
27 The interval τ ≤ τ ≤ τ τL
L H is nonvacuous if τ ≤ 1. Using the definitions of τ L and τ H ,
H
λ λ λ λ
τL ρ ( L + ρ E{φ}) f ( L ) ρ ( L + ρ E{φ}) f ( L )
τH = λ ( +(1−α)λv) f ( H )
≤ λ f ( H )
, where the inequality follows from (1 − α)λv ≥ 0 .
ρ+λ H ρ+λ H
τL
Therefore, τH ≤ (1 + γ ) L+γ
E{φ} f ( L )
f ( H )
, where γ ≡ λ
ρ . As an example, if f (φ) is the density of a
H
f ( L ) 1 τL 1 (2 L +γ ( L + H ))
uniform distribution, then f ( H )
= 1 and E{φ} = 2 ( L + H ), so τ H ≤ 2 (1 + γ ) H =
Optimal Debt and Profitability in the Trade-off Theory 117
Figure 2 illustrates the three scenarios and displays the behavior of debt
and the leverage ratio in each scenario. The tax rate τ is measured along the
horizontal axis and φ is measured along the vertical axis. Scenario I prevails
for τ < τ L, Scenario II prevails for τ L ≤ τ ≤ τ H , and Scenario III prevails for
α
τ H < τ < 1+α . The ordinate of each point on the thick line that is horizontal at
φ = L in Scenario I, upward sloping in Scenario II, and horizontal at φ = H
in Scenario III is the value of φ ∗ for each value of τ . Everywhere above this line
(that is, in Scenario I and in the upper portion of Scenario II, which is labeled
IIA), φ(t) > φ ∗ , so the borrowing constraint is not binding and D(φ(t)) = D∗ .
Since D(φ(t)) is invariant to φ and V (φ(t)) is strictly increasing in φ, the optimal
leverage ratio, L(φ(t)) ≡ VD(φ(t))
(φ(t))
, is strictly decreasing in φ throughout Scenarios I
and IIA. Everywhere below this line (that is, in the lower portion of Scenario II,
which is labeled IIB, and in Scenario III), φ(t) < φ ∗ , so the borrowing constraint
is binding and the leverage ratio is invariant to φ.
L
1
+ γ ). A sufficient condition for ττ L ≤ 1 in this example is 12 (1 + γ )((2 + γ )
2 (1 + γ )((2 + γ ) H
L
+
H H
γ ) ≤ 1, which is satisfied if, for instance, L ≤ 0 and (1 + γ )γ ≤ 2, which is satisfied for 0 ≤ γ ≤ 1.
118 The Journal of FinanceR
28 However, the entire distribution of φ(t) shifts to the right, so the unconditional expected value
1−τ
of S(φ(t)), which is ρ+λ (E{φ} − L) in Scenario I, is unchanged since ddmL (m)
= dE{φ}
dm = 1.
Optimal Debt and Profitability in the Trade-off Theory 119
translation of F(φ) increases the total value of the firm for any given φ(t)
(Statement 3). Finally, the increase in debt and the decrease in the value of
equity increase the leverage ratio for any given φ(t) (Statement 4).
α
LEMMA 2: If 0 < τ < 1+α
and f (φ) ≥ 0 for all φ, then φ ∗ (m) < 1.
To understand why φ ∗ (m) < 1, suppose that φ ∗ (m) = φ ∗ (0) + m, so that a
rightward translation of the distribution by m increases φ ∗ by m, which would
leave F(φ ∗ (m) − m) and f (φ ∗ (m) − m) unchanged but would increase V (φ ∗ (m); m)
through the direct effect of an increase in φ on the firm’s value and through
the effect in Proposition 12. Then, the marginal interest tax shield in equation
(34) would remain unchanged but the marginal expected cost of default would
increase, which would reduce the optimal amount of debt and hence reduce the
associated critical value φ ∗ (m) below φ ∗ (0) + m.
To further explore the impact of a translation of the distribution F(φ − m),
differentiate
∗
equation (34) with respect to m and use equation (34) to substitute
τ ρ+λF(φ∗
(m)−m)
f (φ (m)−m)
for α(ρ + λ)λV (φ ∗ (m), m) to obtain 29
dV (φ ∗ (m); m)
τ [φ ∗ (m) − 1]χ (φ ∗ (m) − m) = α(ρ + λ)λ f (φ ∗ (m) − m) , (35)
dm
where
f (φ)
χ (φ) ≡ λ f (φ) − (ρ + λF(φ)) . (36)
f (φ)
The following proposition uses equation (35) to examine the impacts of a
rightward translation of the distribution F(φ) on P(φ(t); m), φ ∗ (m), and D∗ (m).
α
PROPOSITION 14: If (1) 0 < τ < 1+α
and (2) f (φ) ≥ 0 for all φ, then
1. dP(φ(t);m)
dm
< 0,
∗
2. sign( dV (φdm(m);m)
) = sign(D∗ (m)) = −sign(χ (φ ∗ (m) − m)), and
3. φ (m) < 0, if χ (φ ∗ (m) − m) ≥ 0.
∗
Figure 3. Rightward translation of F(φ) in Scenario II. (Color figure can be viewed at wiley-
onlinelibrary.com)
30 The text shows that, when χ (φ ∗ ) > 0, D∗ (m) < D∗ (0) = D(φ ∗ (0); 0). Note that D∗ (m) =
D(φ ∗ (m); m) = V (φ ∗ (m); m) > V (φ ∗ (m); 0) =
D(φ ∗ (m); 0), where the final equality follows from the fact
that the borrowing constraint binds for φ(t) = φ ∗ (m) under the original distribution F(φ). Therefore,
D(φ ∗ (m); 0) <D∗ (m) < D(φ ∗ (0); 0). Since D(φ; 0) is increasing in φ for φ < φ ∗ (0), there is a unique
∗ ∗
φ ∈ (φ (m), φ (0)) for which D(φ ; 0) = D (m). ∗
31 Section II in the Internet Appendix analyzes the case in which the unconditional distribution
Table I
Effects of Increased Profitability
This table summarizes the impact of increased profitability (an increase in φ(t) or a rightward
translation of F(φ)) on optimal debt when χ (φ) > 0. B.C. refers to the borrowing constraint D(φ) ≤
V (φ).
Characteristics of Scenarios
B.C. not binding B.C. not binding B.C. binds B.C. binds
Pr{default}=0 Pr{default}>0 Pr{default}>0 Pr{default}>0
Not trade-off Trade-off operative Not trade-off Not trade-off
Increase in φ(t)
D unchanged
D unchanged
D↑ D↑
S↑ S↑ S≡0 S≡0
V ↑ V ↑ V ↑ V ↑
L↓ L↓ L≡1 L≡1
Rightward Translation of F(φ)
D↑
D ↓ if χ > 0
D↑ D↑
S↓ S↑ S≡0 S≡0
V ↑ V ↑ V ↑ V ↑
L↑ L ↓ if χ > 0 L≡1 L≡1
reduces the optimal amount of debt and hence reduces the optimal leverage
ratio. When φ(t) < φ ∗ , the trade-off theory is not operative and D(φ(t); m) =
V (φ(t); m). Therefore, Proposition 12 implies that the optimal value of debt
increases in response to a rightward translation of F(φ) in Scenario II when
φ(t) < φ ∗ . Of course, whenever φ(t) ≤ φ ∗ , the optimal leverage ratio equals one
and thus is invariant to a translation of F(φ).
32 Formally, define h( p, φ(t)) implicitly by p ≡ F(h( p, φ(t)), φ(t)). Stable positive first-order serial
Under the conditional distribution F(φ(t ), φ(t)), the trade-off function A(D)
in equation (10) is
The two terms on the right-hand side of equation (42) correspond to channels
(1) and (2), respectively, introduced in the discussion following Proposition 7.
The first term captures the effect on the optimal amount of debt of an increase
in φ(t), and the consequent effect on the probability of default. The second term
Optimal Debt and Profitability in the Trade-off Theory 125
When the borrowing constraint binds, p(φ(t)) ≡ λF(V −1 ( D(φ(t))), φ(t)) = λF(V −1 (V (φ(t))), φ(t)) =
λF(φ(t), φ(t)), which is (weakly) increasing in φ(t) under the general condition for stable persis-
tence in equation (37).
126 The Journal of FinanceR
Appendix A: Proofs
PROOF OF LEMMA A1: Assume that V (φ(t)) is concave and strictly in-
creasing. Twice differentiate V −1 (V (φ)) ≡ φ with respect to φ to obtain
V −1 (V (φ))V (φ) ≡ 1 and V −1 (V (φ))[V (φ)]2 + V −1 (V (φ))V (φ) = 0. Since
V (φ(t)) is concave and strictly increasing, V (φ) > 0, V −1 (V (φ)) > 0, and
V (φ) ≤ 0, so that V −1 (Dt ) ≥ 0. Twice partially differentiating A(Dt ) ≡ τ [ρ +
λ V (φ)<Dt dF(φ)]Dt − [α + τ (1 − α)]λ V (φ)<Dt V (φ)dF(φ) with respect to Dt yields
A (Dt ) = τ [ρ + λ V (φ)<Dt dF(φ)] − α(1 − τ )λV −1 (Dt )Dt f (V −1 (Dt )) and A (Dt )
= [(1 + α)τ − α]λV −1 (Dt ) f (V −1 (Dt )) − α(1 − τ )λDt (V −1 (Dt ) f (V −1 (Dt )) +
α
f (V −1 (Dt ))[V −1 (Dt )]2 ) < 0 for Dt ≥ 0 since τ < 1+α implies that the first
term, [(1 + α)τ − α]λV (Dt ) f (V (Dt )), is negative, and V −1 (Dt ) ≥ 0 and
−1 −1
f (V −1 (Dt )) ≥ 0 imply that the second term, which follows (but does not in-
clude) a minus sign, is nonnegative. Therefore, A (Dt ) < 0, so A(Dt ) is strictly
concave in Dt for Dt ≥ 0.
(1 − τ )φ(t) + λv i + Ai (
Di (φ(t)))
T Vi (φ(t)) = , (A2)
ρ+λ
where Di (φ(t)) is the optimal value of debt when the value function is Vi (φ) and
φ = φ(t). Using the definition of A(D) in equation (10), rewrite equation (A2) as
(1 − τ )φ(t) + gi (φ, D)dF(φ)
T Vi (φ(t)) = , (A3)
ρ+λ
where
τρ D + τ λD + (1 − α)(1 − τ )λVi (φ), if Vi (φ) < D
gi (φ, D) ≡ . (A4)
τρ D + λVi (φ), if Vi (φ) ≥ D
where Da (φ(t)) is the optimal amount of debt under the value function V (φ(t)) +
a, and, following equation (A4),
ga (φ,
Da (φ(t))) − g0 (φ,
Da (φ(t) − a))
[τρ + (1 − α(1 − τ ))λ]a, if V (φ) + a < D̂a (φ(t))
= . (A12)
[τρ + λ]a, if V (φ) + a ≥ D̂a (φ(t))
D(φ(t)) = D∗ . Therefore, for any φ(t) ≥ φ ∗ , the value of V (φ(t)) defined in equa-
(1−τ )φ(t)+λv+A(D∗ )
tion (14) is V (φ(t)) = ρ+λ
, so
1−τ
V (φ2 ) − V (φ1 ) = (φ2 − φ1 ) for φ1 ≥ φ ∗ and φ2 ≥ φ ∗ . (A15)
ρ+λ
Setting φ1 in equation (A15) equal to φ ∗ and using V (φ ∗ ) = D∗ yields
1−τ
V (φ(t)) = [φ(t) − φ ∗ ] + D∗ , for φ(t) ≥ φ ∗ . (A16)
ρ+λ
Since
D(φ(t)) = D∗ when φ(t) ≥ φ ∗ , equations (A16) and (15) imply
1−τ
S(φ(t)) = [φ(t) − φ ∗ ], for φ(t) ≥ φ ∗ . (A17)
ρ+λ
For values of φ(t) < φ ∗ , setting D = D∗ would violate the constraint D ≤
V (φ(t)) because V (φ(t)) ≤ V (φ ∗ ) = D∗ , so the optimal value of D,
D(φ(t)), will
be less than D∗ . Since A(D) is strictly concave in D (Corollary 2), the constraint
V (φ(t)) ≥ D will strictly bind for any φ(t) < φ ∗ so that
V (φ) ≡
D(φ), for φ(t) ≤ φ ∗ ,
and
PROOF OF COROLLARY 3: For φ(t) ≥ φ ∗ , use Proposition 5 to substitute D∗ for
1−τ
D(φ(t)) and ρ+λ [φ(t) − φ ∗ ] + D∗ for V (φ(t)) in the definition of the leverage ra-
tio, L(φ(t)) ≡ D(φ(t))
V (φ(t))
, and divide numerator and denominator by D∗ to obtain
L(φ(t)) = 1
φ(t)−φ ∗ . For φ(t) ≤ φ ∗ , Proposition 5 implies
D(φ(t)) = V (φ(t)), so
1−τ
ρ+λ D∗
+1
the leverage ratio equals one.
PROOF OF PROPOSITION 6: If φ(t) < φ ∗ , then D(φ(t)) = V (φ(t)), so V (φ) < D(φ(t))
if and only if V (φ) < V (φ(t)), that is, if and only if φ < φ(t), since V (φ) is strictly
increasing. Therefore, P(φ(t)) = λF(φ(t)) < λF(φ ∗ ). If φ(t) ≥ φ ∗ , then D(φ(t)) =
D∗ , so V (φ) < D(φ(t)) = D∗ = V (φ ∗ ) if and only if V (φ) < V (φ ∗ ), that is, if and
only if φ < φ ∗ , since V (φ) is strictly increasing. Therefore, P(φ(t)) = λF(φ ∗ ) ≤
λF(φ(t)). It thus follows that P(φ(t)) = λ min[F(φ(t)), F(φ ∗ )].
PROOF OF PROPOSITION 7: From equation (18), P( D(φ(t))) ≡ λ V (φ)<D(φ(t)) dF(φ)
= λ φ<V −1 (D(φ(t))) dF(φ), so dD(φ(t)) = V ( D(φ(t))) f (V −1 (
dP −1
D(φ(t)))) > 0. If φ(t) < φ ∗ ,
d d
then D(φ(t))
dφ(t)
> 0, so dφ(t)
dP
= dD(φ(t))
dP D(φ(t))
dφ(t)
> 0. If φ(t) ≥ φ ∗ , then dD(φ(t))
dφ(t)
≡ 0,
d
D(φ(t))
so dφ(t) = dD(φ(t)) dφ(t) ≡ 0. From equation (3), r(t) = ρ + λ V (φ)<D(φ(t)) [1 −
dP dP
V (φ)
V (φ)
(1 − α) D(φ(t)) ]dF(φ) = ρ + λ φ<V −1 (D(φ(t))) [1 − (1 − α) D(φ(t)) dr(t)
]dF(φ), so dD(φ(t)) =
Optimal Debt and Profitability in the Trade-off Theory 133
λV −1 (
D(φ(t)))α f (V −1 (
V (φ)
D(φ(t)))) + λ D(φ(t))) [(1 − α) [
φ<V −1 ( D(φ(t))]2
]dF(φ) > 0. If
∗ dr(t) d
φ(t) < φ , then dD(φ(t))
dφ(t)
> 0, so dφ(t)
dr(t)
= d
D(φ(t))
D(φ(t)) dφ(t)
> 0. If φ(t) ≥ φ ∗ , then dD(φ(t))
dφ(t)
≡
dr(t) dD(φ(t))
dr(t)
0, so dφ(t) = dD(φ(t)) dφ(t)
≡ 0.
− V (φ)]dF(φ). (A21)
Substitute [V (φ) − V (φ(t))]dF(φ) for v − V (φ(t)) in equation (A21) and rear-
range to obtain
(1 − τ )[φ(t) − r(t)V (φ(t))] = −λ [V (φ) − V (φ(t))]dF(φ) ≤ 0. (A22)
V (φ)≥V (φ(t))
1−τ
V (φ(t)) = [φ(t) − L] + V ( L). (A25)
ρ+λ
1−τ
v= [E{φ} − L] + V ( L), (A26)
ρ+λ
(1 − τ ) L + λv + τρV ( L)
V ( L) = . (A27)
ρ+λ
Equations (A26) and (A27) are two linear equations in V ( L) and v, which
can be solved to obtain V ( L) = ρ+λ 1
[ L + ρλ E{φ}]. Therefore,
D(φ(t)) = D∗ =
V ( L) = ρ+λ
1
[ L + ρλ E{φ}]. Since φ ∗ = L, Proposition 5 implies that S(φ(t)) =
1−τ
ρ+λ
[φ(t) − L] and V (φ(t)) = D∗ + S(φ(t)).
(I) Assume that τ < τ L. Suppose that D∗ = V ( L), ω1 > 0, and ω2 = 0. Since
D∗ = V ( L), the complementary slackness condition in equation (21) is
satisfied, and since ω2 = 0, the complementary slackness condition in
equation (22) is satisfied. Since ω2 − ω1 < 0, the first-order condition
in equation (20) implies that A (D∗ ) < 0. Thus, it suffices to show that
A (V ( L)) < 0. Differentiating A(D) in equation (10) with respect to D,
evaluating the derivative at D = D∗ = V ( L), and using φ ∗ = V −1 (D∗ ) =
V −1 (V ( L)) = L yields
A (D∗ ) = τρ − α(1 − τ )λV −1 (V ( L)) f ( L)V ( L). (A28)
where the inequality follows from ρ > 0 and the assumption that τ < τ L.
Therefore, D∗ = V ( L), ω1 > 0, and ω2 = 0 satisfy the first-order con-
dition in equation (20) and the complementary slackness conditions in
equations (21) and (22), and the firm is in Scenario I.
(II) Assume that τ L ≤ τ ≤ τ H . Suppose that ω1 = ω2 = 0. Therefore, the com-
plementary slackness conditions in equations (21) and (22) are satis-
fied. Since ω2 − ω1 = 0, the first-order condition in equation (20) implies
that A (D∗ ) = 0. Thus, it suffices to show that A (D) = 0 for some D ∈
[V ( L), V ( H )]. The proof proceeds by showing that if D∗ = V ( L), so
that φ ∗ = L, then A (D∗ ) ≥ 0, and if D∗ = V ( H ), so that φ ∗ = H , then
A (D∗ ) ≤ 0, so there is a D∗ ∈ [V ( L), V ( H )] that satisfies A (D∗ ) = 0.
Differentiate A(D) in equation (10) with respect to D and use Lemma 1
to obtain
1 λ/ρ
V ( L) = L + E{φ}. (A34)
ρ+λ ρ+λ
Now evaluate A (D∗ ) in equation (A31) at D∗ = V ( L) and use the ex-
pression for V ( L) in equation (A34) to obtain
λ
A (V ( L)) = τρ − αλ L + E{φ} f ( L). (A35)
ρ
where the inequality follows from ρ > 0 and the assumption that τ ≥ τ L.
Evaluate A (D) at D = V ( H ) to obtain
A (V ( H )) = (τ − τ H )(ρ + λ) ≤ 0, (A39)
Use V ( H ) = ρ+λ1
[ H + (1 − α)λv] from Proposition 10 and the definition
λ
τ H ≡ α ρ+λ f ( H )( H + (1 − α)λv) to rewrite equation (A41) as
where the second inequality follows from ρ + λ > 0 and τ > τ H . Thus,
D∗ = V ( H ), ω1 = 0, and ω2 > 0 satisfy the first-order condition in equa-
tion (20) and the complementary slackness conditions in equations (21)
and (22), and the firm is in Scenario III.
PROOF OF PROPOSITION 12: Let y(φ(t); 0) be the optimal taxable income under
the original distribution (m = 0) and recall from Statement 2 of Proposition
8 and footnote 20 that y(φ(t); 0) is strictly increasing in φ(t). Consider the
following feasible financing plan under Gm(φ). (1) If φ(t) ∈ [ L(m), H (0)], set
D(φ(t); m) = D(φ(t); 0), where D(φ(t); 0) is the optimal amount of debt under
G0 (φ), and pay the same default premium that would be paid under G0 (φ),
so that taxable income y(φ(t); m) = y(φ(t); 0). I verify below that the default
premium under the new distribution, Gm(φ), is no higher than under the origi-
nal distribution, G0 (φ). (2) If φ(t) ∈ ( H (0), H (m)], set D(φ(t); m) = D( H (0); 0)
and pay the same default premium that would be paid under G0 (φ) when
φ(t) = H (0), so that y(φ(t); m) =y( H (0); m) + φ(t) − H (0) = y( H (0); 0) +
φ(t) − H (0) > y( H (0); 0). (3) At any future date of regime change, t j , (a) if
φ(t j ) ∈ [ L(m), H (0)], default if and only if the firm would optimally default
under G0 (φ) when φ = φ(t j ) and (b) if φ(t j ) ∈ ( H (0), H (m)], do not default. For
φ(t0 ) ∈ [ L(m), H (0)], the firm will have the same cash flow Gm(φ) as it would
under the optimal policy under G0 (φ). However, with the financing plan under
Gm(φ), the continuation value under Gm(φ) will exceed the continuation value
(m) (m)
under G0 (φ) by HH(0) (1 − τ )y(φ(t); m)dF(φ − m) − LL(0) (1 − τ )y(φ(t); 0)dF(φ)
(m) (m)
> (1 − τ )y( H (0); 0) HH(0) dF(φ − m) − (1 − τ )y( L(m); 0) LL(0) dF(φ) = (1 −
(0) (0)+m
τ )y( H (0); 0) HH(0)−m dF(φ) − (1 − τ )y( L(m); 0) LL(0) dF(φ) > 0 , where the
first inequality follows from y(φ(t); m) > y( H (0); 0) if φ(t) > H (0), as shown
above, and from the fact that y(φ(t); 0) is strictly increasing in φ(t) as shown ear-
lier in this proof; the second inequality follows from y( H (0); 0) > y( L(m); 0)
(0) (0)+m
and HH(0)−m dF(φ) ≥ LL(0) dF(φ) since f (φ) ≡ F (φ) is nondecreasing. There-
fore, the continuation value under Gm(φ) exceeds the continuation value under
138 The Journal of FinanceR
G0 (φ), and hence V (φ(t); m) > V (φ(t); 0) for m > 0. Finally, because the contin-
uation value of the firm for any given φ(t) is higher under Gm(φ) than under
G0 (φ), the default premium under Gm(φ) is no higher than under G0 (φ).
τρ+λ
L(φ(t); m) ρ+λ ] > 0, where the inequality follows from L(φ(t); m) ≤ 1 and τ < 1
(Statement 4).
∂ V (φ ∗ (m);m)
inequality follows from Proposition 12. Finally, since ∂φ
> 0 (Proposition
∂ V (φ ∗ (m);m)
1), φ ∗ (m) ≤ − ∂m
∂ V (φ ∗ (m);m) < 0 (Statement 3).
∂φ
Note that
g(φ, D) is nondecreasing in φ.36 Therefore, T V (φ2 )
(1−τ )φ2 + g(φ(t ), D(φ1 )) f (φ(t ),φ2 )dφ(t ) (1−τ )φ1 + g(φ(t ),
D(φ1 )) f (φ(t ),φ2 )dφ(t )
≥ ρ+λ
> ρ+λ
≥
(1−τ )φ1 + g(φ(t ),
D(φ1 )) f (φ(t ),φ1 )dφ(t )
ρ+λ
= T V (φ1 ), where the first inequality follows
from the fact that D(φ1 ) is feasible when φ(t) = φ2 , the second inequality fol-
lows from φ2 > φ1 , and the third inequality follows from the facts that g(φ, D)
is nondecreasing in φ and F(φ(t ), φ2 ) first-order stochastically dominates
F(φ(t ), φ1 ). Therefore, T maps nondecreasing functions into strictly increasing
functions.
Monotonicity: Suppose that V2 (φ(t)) ≥ V1 (φ(t)). Define gi (φ, D) to be the
function g(φ, D) defined in equation (A43),
where V (φ) is replaced by Vi (φ),
(1−τ )φ(t)+ g2 (φ(t ),
D1 (φ(t))) f (φ(t ),φ(t))dφ(t )
i = 1, 2. Therefore, T V2 (φ(t)) ≥ ρ+λ
and T V1 (φ(t))
(1−τ )φ(t)+ g1 (φ(t ), D1 (φ(t))) f (φ(t ),φ(t))dφ(t )
= ρ+λ
, where
D1 (φ(t)) attains the maximum of
(1−τ )φ(t)+ g1 (φ(t ),D) f (φ(t ),φ(t))dφ(t )
ρ+λ
, subject to D ≤ V1 (φ(t)), and D1 (φ(t)) is feasible
under V2 (φ(t))
because D1 (φ(t)) ≤ V1 (φ(t)) ≤ V2 (φ(t)). Therefore, T V2 (φ(t)) −
[g2 (φ(t ),
D1 (φ(t)))−g1 (φ(t ),
D1 (φ(t)))] f (φ(t ),φ(t))dφ(t )
T V1 (φ(t)) ≥ ρ+λ
. To prove that T V2 (φ(t)) −
T V1 (φ(t)) ≥ 0, it suffices to prove that g2 (φ(t ), D1 (φ(t))) − g1 (φ(t ),
D1 (φ(t))) ≥ 0
for all φ(t) in the support of F. The definition of g(φ, D) implies that
α)λD ≤ τρ D + λD, or equivalently that 0 ≤ (1 − τ )λD − (1 − τ )(1 − α)λD = α(1 − τ )λD, which fol-
lows from α > 0, τ < 1, λ > 0, and D ≥ 0.
140 The Journal of FinanceR
Set y(s) = y(t) for all s in [t, t1 ), dDt = Dt , dDs = 0 for all s in (t, t1 ), and
Dt1− = Dt , to obtain
t1
V (φ(t)) = max Et (1 − τ )y(t) e−ρ(s−t) ds + Dt + e−ρ(t1 −t) max[V (φ(t1 ))
Dt ≤V (φ(t)) t
− Dt , 0] . (B2)
Use the expectations in equations (B3) and (B4) to calculate the expectation
in equation (B2) and rearrange to obtain
1
V (φ(t)) = max Et (1 − τ )yt + (ρ + λ)Dt + λ [V (φ)
ρ + λ Dt ≤V (φ(t)) V (φ)≥Dt
− Dt ]dF(φ) . (B5)
142 The Journal of FinanceR
Substitute the expression for taxable income, y(t), from equation (5) into
equation (B5) to obtain
1 (1 − τ ) φ(t) − ρ Dt − λ V (φ)<Dt [Dt − (1 − α)V (φ)]dF(φ)
V (φ(t)) = max Et .
ρ+λ Dt ≤V (φ(t)) + (ρ + λ)Dt + λ V (φ)≥Dt [V (φ) − Dt ]dF(φ)
(B6)
(B7)
Use 1 − V (φ)≥Dt dF(φ) = V (φ)<Dt dF(φ) and V (φ)≥Dt V (φ)dF(φ) = V (φ)dF
(φ) − V (φ)<Dt V (φ)dF(φ) and rearrange to rewrite equation (B7) as
⎧ ⎫
⎪
⎪ (1 − τ )φ(t) ⎪
⎪
1 ⎨
⎬
V (φ(t)) = max Et τ ρ + λ V (φ)<D dF(φ) Dt .
ρ + λ Dt ≤V (φ(t)) ⎪
⎪ t
⎪
⎪
⎩ ⎭
+ λ V (φ)dF(φ) − [1 − (1 − τ )(1 − α)]λ V (φ)<Dt V (φ)dF(φ)
(B8)
Use the definitions A(D) ≡ τ [ρ + λ V (φ)<D dF(φ)]D − [α + τ (1 − α)]λ
V (φ)<D
V (φ)dF(φ) and v ≡ V (φ)dF(φ) to rewrite equation (B8) as
(1 − τ )φ(t) + λv + A(Dt )
V (φ(t)) = max . (B9)
Dt ≤V (φ(t)) ρ+λ
REFERENCES
Abel, Andrew B., 2016, Investment and leverage, Working paper, Wharton School, University of
Pennsylvania.
Biais, Bruno, Thomas Mariotti, Guillaume Plantin, and Jean-Charles Rochet, 2007, Dynamic
security design: Convergence to continuous time and asset pricing implications, Review of
Economic Studies 74, 345–390.
Fama, Eugene F., and Kenneth R. French, 2002, Testing trade-off and pecking order predictions
about dividends and debt, Review of Financial Studies 15, 1–33.
Frank, Murray Z., and Vidhan K. Goyal, 2008, Trade-off and pecking order theories of debt, in
Espen Eckbo, ed.: Handbook of Corporate Finance: Empirical Corporate Finance, Volume 2
(Elsevier Science, North-Holland, Amsterdam), 135–202.
Gilson, Stuart C., 1997, Transactions costs and capital structure choice: Evidence from financially
distressed firms, Journal of Finance 52, 161–196.
Goldstein, Robert, Nengjiu Ju, and Hayne Leland, 2001, An EBIT-based model of dynamic capital
structure, Journal of Business 74, 483–512.
Gorbenko, Alexander S., and Ilya A. Strebulaev, 2010, Temporary versus permanent shocks: Ex-
plaining corporate financial policies, Review of Financial Studies 23, 2591–2647.
Hennessy, Christopher A., and Toni M. Whited, 2005, Debt dynamics, Journal of Finance 60,
1129–1165.
Optimal Debt and Profitability in the Trade-off Theory 143
Hennessy, Christopher A., and Toni M. Whited, 2007, How costly is external financing? Evidence
from a structural estimation, Journal of Finance 62, 1705–1745.
Kraus, Alan, and Robert H. Litzenberger, 1973, A state-preference model of optimal financial
leverage, Journal of Finance 28, 911–922.
Leary, Mark T., and Michael R. Roberts, 2005, Do firms rebalance their capital structures? Journal
of Finance 60, 2575–2619.
Leland, Hayne E., 1994, Corporate debt value, bond covenants, and optimal capital structure,
Journal of Finance 49, 1213–1252.
Merton, Robert C., 1974, On the pricing of corporate debt: The risk structure of interest rates,
Journal of Finance 29, 449–470.
Modigliani, Franco, and Merton H. Miller, 1958, The cost of capital, corporation finance and the
theory of investment, American Economic Review 48, 261–297.
Myers, Stewart C., 1993, Still searching for optimal capital structure, Journal of Applied Corporate
Finance 6, 4–14.
Robichek, Alexander A., and Stewart C. Myers, 1966, Problems in the theory of optimal capital
structure, Journal of Financial and Quantitative Analysis 1, 1–35.
Scott, James H., Jr., 1976, A theory of optimal capital structure, Bell Journal of Economics 7,
33–54.
Strebulaev, Ilya A., 2007, Do tests of capital structure theory mean what they say? Journal of
Finance 62, 1747–1787.
Supporting Information
Additional Supporting Information may be found in the online version of this
article at the publisher’s website:
Appendix S1: Internet Appendix.