Suresh Sundaresan

Dynamic Investment, Capital Structure,
and Debt Overhang*

Suresh Sundaresan
Columbia Business School
Neng Wang
Columbia Business School and NBER
Jinqiang Yang
School of Finance, Shanghai University of Finance and Economics
We develop a dynamic contingent-claim framework to model S. Myerss idea that a
rm is a collection of growth options and assets in place. The rms composition
between assets in place and growth options evolves endogenously with its investment
opportunity set and its nancing of growth options, as well as its dynamic leverage
and default decisions. The rm trades off tax benets with the potential nancial
distress and endogenous debt-overhang costs over its life cycle. Unlike the standard
capital structure models of Leland, our model shows that nancing and anticipated
endogenous default decisions have signicant implications of rms growth-option
exercising decisions and leverage policies. The rms ability to use risky debt to
borrow against its assets in place and growth options substantially inuences its
investment strategies and its value. Quantitatively, we nd that the rm consistently
chooses conservative leverage in line with empirical evidence in order to mitigate the
debt-overhang effect on the exercising decisions for future growth options. Finally,
we nd that debt seniority and debt priority structures have both conceptually important and quantitatively signicant implications on growth-option exercising and
leverage decisions as different debt structures have very different debt-overhang
implications. (JEL E2, G1, G3)
Models of truly intertemporal investments with irreversibility and models

of dynamic nancing with endogenous defaults have proceeded relatively
*We thank Ken Ayotte, Patrick Bolton, Darrell Dufe, Jan Eberly, Paolo Fulghieri, Larry Glosten,
Steve Grenadier, Christopher Hennessy, Hayne Leland, Debbie Lucas, Chris Mayer, Bob McDonald,
Mitchell Petersen, Ilya Strebulaev, Toni Whited, Jeff Zwiebel, an anonymous referee, and seminar participants at Berkeley, Columbia, Northwestern, the NYU Five-star nance research conference,
Stanford, UBC, and Wharton for helpful comments. We thank Sam Cheung and Yiqun Mou for
their research assistance. Neng Wang acknowledges support by the Natural Science Foundation of
China (# 71472117). Jinqiang Yang acknowledges support by Natural Science Foundation of China
(# 71202007), New Century Excellent Talents in University (# NCET-13-0895,) and the Program for
Innovative Research Team of Shanghai University of Finance and Economics (# 2014110345). First
version: December, 2006. Send correspondence to Neng Wang, Columbia Business School and NBER,
New York, NY 10027, USA; telephone: 212-854-3869. E-mail: [email protected].
The Author 2014. Published by Oxford University Press on behalf of The Society for Financial
Studies. All rights reserved. For Permissions, please e-mail: [email protected].
doi:10.1093/rcfs/cfu013
Advance Access publication December 17, 2014
Review of Corporate Finance Studies / v 4 n 1 2015
independent of each other. The literature on intertemporal investments

with multiple rounds of investments often ignores the nancing exibility
possessed by the rms. On the contrary, models of dynamic nancing
have tended to ignore the investment opportunity set to a point whereby
proceeds from each round of new nancing are paid out to equityholders.
Even though considerable insights have emerged from each strand of
literature, much remains to be done in integrating investment theory
with dynamic nancing. Our paper builds on the insights of the realoptions and contingent-claims/credit-risk literature with the objective of
showing the important link between the optimal exercise of growth
options and corporate leverage in a parsimonious and tractable way.
By using our dynamic model of investment and nancing, we show
that a rational rm signicantly lowers its leverage anticipating its
future growth-option exercising decisions. Our numerical exercise generates empirically observed low leverage (e.g., around 1/3) once we incorporate multiple rounds of growth options, indicating the important
interaction of growth-option exercising, leverage, and valuation.
Integration of multiple rounds of investments with multiple rounds of
nancing presents many modeling challenges: rst, the rm must solve
endogenously the upper threshold of its value when it must undertake
new investments. In making this decision the rm must take into account
the level of debt it must use to optimally trade off the expected tax
benets with the possibility of premature termination of the rm when
default occurs, taking into account all future investment and nancing
possibilities. The optimal default-decision constitutes a lower threshold
level of the value of the rm, which must also be decided endogenously.
This paper provides an analytically tractable framework to examine
dynamic endogenous corporate investment, nancing, and default decisions.1 We provide a tractable model of real options in which the rm
makes these endogenous lower (default) and upper (investment) decisions
over time, while choosing its optimal debt level along the way. In so
doing, we provide a methodological framework for assessing how the
life cycle of the rm may inuence the manner in which it makes intertemporal investment, and nancing decisions. Broadly speaking, we use
the term nancing to encompass both the level of debt and the optimal
default decisions.
Several new insights emerge from our analysis. In thinking about of
inter-temporal investments and nancing, we start with the intuitive premise that the rm starts its life as a collection of growth options, much as
in Myers (1977). For simplicity, we assume that the collection of growth
1
See Stein (2003) for a survey on corporate investment, agency conicts, and information. See Caballero
(1999) for a survey on aggregate dynamic investment. See Harris and Raviv (1991) for a survey on
theories of capital structure.
Dynamic Investment, Capital Structure, and Debt Overhang
options possessed by the rm is known and does not change over time.
Then, as the rm moves through time, it optimally decides when to
exercise each growth option and how to nance each growth option,
keeping in mind that several additional growth options may be available
to the rm in the future. When the rm has exercised all its growth
options, it is left only with assets in place. However, it starts its life
with no assets in place. At all other times, it has some future growth
options and some assets in place. The composition of growth options
and assets is endogenously determined in a dynamic optimizing framework. Thus, our model captures the life cycle of the rm in a natural way.
There is an important economic distinction between assets in place and
growth option in terms of what fraction of each is available to residual
claimants upon default. It is reasonable to argue that assets in place are
hard assets, with values that are veriable and hence may provide
greater liquidation value upon default compared with growth options,
which may have embedded human capital and hence may possess a different (and possibly much lower) liquidation value along the lines of Hart
and Moore (1994). We explicitly incorporate the potential differences of
recovery values for growth options and assets in place. The modeling of
this difference is a new contribution to the real-options literature, as well,
because it requires the values of foregone growth options upon default
(which are non-linear functions of primitive states) to bear on optimal
exercise boundaries. Indeed, we provide explicit quantitative and qualitative characterization of the effect of embedded human capital in future
growth options on optimal investment thresholds, default thresholds, and
the level of debt used by the rm at each stage of its life cycle. We believe
that this is a unique contribution of our study.
Our research provides a natural bridge between structural credit risk/
capital structure models, and the dynamic irreversible investment theory.2
We nd that even for rms with only one growth option, integrating
investment and nancing decisions generates new insights, not captured
by either the standard real-options models (e.g., McDonald and Siegel
1986), or credit risk/capital structure models (e.g., Leland 1994). For
example, Leland (1994) shows that the default threshold decreases in
volatility for the standard (put) option argument in a contingent-claim
framework based on the standard trade-off theory of Modigliani and
2
McDonald and Siegel (1985, 1986) and Brennan and Schwartz (1985) are important contributions to
modern real-options approach to investment under uncertainty. Dixit and Pindyck (1994) is a standard
reference on real-options approach toward investment. Abel and Eberly (1994) provide a unied framework integrating the neoclassical adjustment cost literature with the literature on irreversible investment.
Grenadier (1996) studies strategic interactions among agents in real-options settings. Grenadier and
Wang (2005) analyze the effect of informational frictions and agency on investment timing decisions.
Grenadier and Wang (2007) study the effect of time-inconsistent preferences on real option exercising
decisions. In our earlier draft, Sundaresan and Wang (2006), we provide additional detailed analysis on
the impact of alternative debt structures on corporate investment and leverage policies.
Miller (1963). However, the default threshold in our model may either
decrease or increase in volatility. The intuition is as follows: (i) a higher
volatility raises the investment threshold in our model for the standard
(call option) value of waiting argument; (ii) a higher investment threshold
naturally leads to a greater amount of debt issuance. That is, the rm
issues more debt (but at a later time), when volatility is higher. Larger
debt issuance raises the default threshold, ceteris paribus. As a result,
unlike Leland (1994), we have two opposing effects of volatility on the
default threshold owing to endogenous investment in our model.
In developing our analysis, we have made the analytically convenient
assumption that the rm uses its nancing exibility only at times when it
makes its optimal investments. At a rst glance, the reader may think that
this is a strong assumption. Nevertheless, it turns out that this assumption proves to be innocuous for the following reasons: First, when growth
options are economically meaningful, investments occur over time at
frequent (but stochastic) intervals. Hence the real cost of the assumption
is rather slight. In addition, it is well known (Strebulaev 2007) that even
the introduction of low-costs nancing leads to the result that rms will
choose to adjust their capital structure at periodic intervals rather than
continuously. Dudley (2012) shows that when there are xed costs of
adjustment, it is optimal for rms to synchronize capital structure adjustment with the nancing of large investment projects. In our model, the
primary reason for nancing is investment, and investments require a
lump-sum cost. Hence, it is natural to model nancing adjustments
when investments occur. Moreover, the key focus of our paper is on
the effect of nancing on growth-option exercising decisions.
In addition, our study provides several additional insights on the
valuation of equity and credit spreads at different stages in the life
cycle of the rm. We have a natural benchmark to assess of our results:
after all the growth options are optimally exercised, the rm is left with
only assets in place. At this nal stage, our results are exactly the same as
either Leland (1994) (when no dynamic nancing adjustments are
allowed) or Goldstein, Ju, and Leland (2001) (when dynamic nancing
adjustments are allowed). At all previous stages, the rm has a mixture of
assets in place and growth options, and they inuence both equity valuation and credit spreads. The key insight is that the incremental nancial
exibility at times other than actual investments is less important when
there are growth options.
Related literature. Recently, there is a growing body of literature that
extends Leland (1994) to allow for strategic debt service,3 and dynamic
3
Anderson and Sundaresan (1996) use a binomial model to study the effect of strategic debt service on
bond valuation. See Mella-Barra and Perraudin (1997), Fan and Sundaresan (2000), and Lambrecht
capital structure decisions. Fischer, Heinkel, and Zechner (1989);

Goldstein, Ju, and Leland (2001); and Strebulaev (2007) formulate
dynamic leverage decisions with exogenously specied investment policies.4 Leary and Roberts (2005) empirically nd that rms rebalance their
capital structure infrequently in the presence of adjustment costs.
Following Leland (1994), most contingent-claims models of credit risk/
capital structure assume that the rms cash ows are exogenously given
and focus on the rms nancing and default decisions.5 Unlike these
work, our model endogenizes growth-option exercising decisions and
induces dynamic leverage decisions via motives of nancing investment.6
Titman and Tsyplakov (2007) also build a model that allows for dynamic
adjustment of both investment and capital structure. Their model is based
on continuous investment decisions, whereas our model focuses on the
irreversibility of growth-option exercising.7 We solve the model in closed
form (up to coupled nonlinear equations), whereas Titman and
Tsyplakov (2007) have three state variables and numerically solve the
decision rules. Ju and Ou-Yang (2006) show that the rms incentive to
increase rm risk ex post is mitigated if the rm wants to issue debt
periodically. In the interest of parsimony, we abstract from stochastic
interest rates.8 Guo, Miao, and Morellec (2005) develop a model of irreversible investment with regime shifts. Hackbarth, Miao, and Morellec
(2006) study the effects of macro conditions on credit risk and rms
nancing policies. Tserlukevich (2008) studies the effect of real options
on nancing behavior.
1. Model
We rst set up a dynamic formulation where the rm is a collection of
growth options and assets in place. Assume that the rm behaves in the
(2001) for continuous-time contingent-claim treatment. Sundaresan and Wang (2007) study the interactions between investment and nancing decisions when equityholders may ex post strategically force
concessions from debtholders.
4
Early important contributions towards building dynamic capital structure models include Kane, Marcus,
and McDonald (1984, 1985).
Leland (1998) extends Leland (1994) by incorporating risk management with capital structure, and also
allows the rm to engage in asset substitution by selecting volatility of the project.
Our model ties the investment and nancing adjustments to occur at the same time. This assumption is
made for analytical convenience. We leave extensions to allow for separate adjustments of investment
and nancing for future research.
Brennan and Schwartz (1984) is an early, important contribution, which allows for the interaction
between investment and nancing.
See Kim, Ramaswamy, and Sundaresan (1993); Longstaff and Schwartz (1995); and Collin-Dufresne and
Goldstein (2001) for extensions of Merton (1974), to allow for a stochastic interest rate and other
features.
interests of existing equityholders at each point in time.9 At time zero, the

rm starts with no assets in place, and knows that it has N growth
options. These growth options can only be exercised sequentially. One
way to view these growth options is as the discretized decisions for capital-accumulation decisions.10
The rm observes the demand shock Y for its product, where Y is given
by the following geometric Brownian motion (GBM) process:
dYt Ytdt YtdWt;
11
and W is a standard Brownian motion. Equivalently, we may also

interpret Y as the (stochastic) price process for the rms output.12 The
risk-free interest rate r is constant. For convergence, assume that the
(risk-neutral) expected growth rate is lower than the interest rate, in
that r4: Assume no production cost after the asset is in place.13 When
the rm exercises its n-th growth option, it creates the n-th asset in place,
which generates prot at the rate of mn Y, where mn 40 is a constant. We
may interpret mn as the production capacity, or equivalently the constant
rate of output produced by the rms n-th asset in place. Let the rms
total prot
rate from its rst n assets in place be Mn Y, where
P
Mn nj1 mj .
Let Tin denote the endogenously chosen time at which the rm exercises
its n-th growth option, where 1 n N: Let In denote the xed cost of
exercising its n-th growth option. These exercising costs In are constant
and known at time 0. At each endogenously chosen (stochastic) investment time Tin , the rm issues a mixture of debt and equity to nance the
exercising cost In. As in standard trade-off models of capital structure,
debt has a tax advantage. The rm faces a constant tax rate 40 on its
income after servicing interest payments on debt. To balance the tax
benets, debt induces deadweight losses when the rm does poorly.
The rm dynamically trades off the benets and costs of issuing debt.
For analytical convenience, assume that debt is perpetual and is issued at
9
We ignore the conicts of interests between managers and investors and leave them for future research.
10
One could certainly visualize growth options arriving with some intensity at random times in the future.
In such an economy, the optimal investment decisions would reect the arrival intensity in addition to the
factors that we consider in our formulation. Extension of random arrivals of growth options is clearly an
interesting topic for future research.
11
Let W be a standard Brownian motion in R on a probability space ; F; Q and x the standard

ltration fFt : t 0g of W. Since all securities are traded here, we work directly under the risk-neutral
probability measure Q. Under the innite horizon, additional technical conditions such as uniform
integrability are assumed here. See Dufe (2001).
12
In our model as in many other investment and capital structure models, the process Y captures both
demand and productivity shocks.
13
Our model ignores operating leverage. We may extend our model to allow for operating leverage by
specifying the rms prot from its n-th asset in place as mn Y wn , where wn is the operating cost for the
n-th asset in place.
par. The assumption of perpetual debt simplies the analysis substantially and has been widely adopted in the literature.14 Note that we have
assumed that the rm can only issue debt at investment times
fTin : 1 n Ng. At a rst glance, this may appear to be a strong
assumption. In fact, our assumption is actually rather mild. Strebulaev
(2007) has shown that in the presence of frictions rms adjust their capital
structure only infrequently. Therefore, in a dynamic economy that we
model, the leverage of the rm is likely to differ from the optimum
leverage predicted by models that permit costless adjustment of leverage.
Given this nding it is more natural to recapitalize when optimal investment decisions are warranted. In addition, models that permit re-leveraging in good times, implicitly or explicitly use the debt proceeds to pay
dividends, which is at odds with the basic provision that senior claims
(such as debt) may not be issued to nance junior claims (such as equity).
Let Cn and Fn denote the coupon rate and the par value of the perpetual debt issued to nance the exercising of the n-th growth option at Tin .
Let Tdn denote the endogenously chosen stochastic default time after the
rm exercises n growth options, but before exercising the n 1-th
growth option, where 1 n N. When exercising the new growth
option at Tin1 , the rm calls back its outstanding debt with par Fn and
coupon Cn, and issues new debt with par Fn1 and coupon Cn1 . That is,
at each point in time, there is only one class of debt outstanding.15
Figure 1 describes the decision-making process of the rm over its life
cycle. The rm has (N 1) stages. In stage 0, the rm has no assets in
place. We assume that the initial value of the demand shock is sufciently
low such that the rm always starts with waiting, the economically most
interesting case. If the demand shock fYt : t 0g rises sufciently high
(i.e., greater or equal to an endogenous threshold Yi1 to be determined in
Section 2) at the stochastic (endogenous) time Ti1 , the rm exercises its
rst growth option by paying a one-time xed cost I1 at time Ti1 as in
McDonald and Siegel (1986). Note that since Y(0) is sufciently low, we
have Ti1 40. Notation-wise, we use Yi1 YTi1 . To nance the rst
growth-option exercising cost I1, the rm issues a mixture of equity
and perpetual debt. This completes the description of the rms decision
in its initial stage (stage 0). Next turn to stage 1.
After the rst asset is in place, the rm collects prot ow m1 Y until it
decides to either default on its debt or exercise its (second) growth option.
If the rm defaults before exercising the second growth option (Td1 < Ti2 ),
14
We may extend the model by allowing for a nite average maturity for debt as in Leland (1994b) at the
cost of additional modeling complexity.
15
See Section 5 and also Hackbarth and Mauer (2012) for analysis where more than one class of debt are
outstanding. The design of priority structure of debt and its implications for real-options exercise is a
topic worthy of further research.
Figure 1
The rms decision-making process over its life cycle.
The rm starts with N sequentially ordered growth options. We divide its decision making over its life
cycle into N 1 stages. In stage 0, the rm exercises its rst growth option when the stochastic process
Y given in Equation (1) rises sufciently high (i.e., Y Yi1 YTi1 ). The rm waits otherwise. When
exercising, the rm issues a mixture of equity and the rst perpetual debt with coupon C1 to nance the
exercising cost I1. This completes the description of the rms stage 0 decision. Now move to stage 1.
Provided that Yd1 < Y < Yi2 , the rm generates cash ow m1 Y from its operation. If its cash ow drops
sufciently low, (i.e., Y Yd1 ), the rm defaults. If the cash ow rises sufciently high (i.e. Y Yi2 ), the
rm exercises its second growth option, and issues a mixture of equity and the second perpetual debt with
coupon C2 to nance the exercising cost I2. After the second growth option is exercised, the rm
generates stochastic cash ow m1 m2 Y, provided that Y Yd2 . This process continues. If the rm
reaches the nal stage N, the rm has total N assets in place and collects total cash ow MN Y, where
XN
m . The decision variables include N investment thresholds Yin , N default thresholds Ydn ,
MN
n1 n
and N coupon policies Cn, where n 1; 2; . . . ; N: Notation-wise, we dene the n-th stage as Y, such that
Ydn Y < Yin1 where Yin1 YTin1 and Ydn YTdn .
it ceases to exist. All proceeds from liquidation go to creditors. However,

liquidation is inefcient because it induces value losses from both the
existing assets in place and foregone growth options. We will specify
the liquidation payoff in the next paragraph when discussing the rms
general stage n problem. Intuitively, if the demand shock Y is sufciently
high, then it is optimal for the rm to exercise its second growth option.
By incurring the xed investment cost I2, the rm exercises its second
growth option at endogenously chosen time Ti2 . At the second investment
time Ti2 , the rm calls back its outstanding debt with par F1, and issues
a mixture of equity and the new perpetual debt with par F2 to nance
the second growth option exercising cost I2. This concludes the rms
decision in stage 1.
It is straightforward to describe the rms stage-n decision problem.

After immediately exercising the n-th growth option, the rm operates
its existing n assets in place until the demand shock Y either rises sufciently high, which triggers the rm to call back debt with par Fn,
issue a mixture of new perpetual debt and equity to nance In1 to
exercise the n 1-th growth option, or the demand shock Y drops
sufciently low, which leads the rm to default on its outstanding debt
with par Fn.
Let An Y denote the after-tax present value of all n existing assets in
place (under all equity nancing), in that

1
2
An Y
Mn Y; 1 n N;
r
where Mn is the production capacity for all existing n assets in place and
is given by
Mn
n
X
mj ;
1 n N:
j1
When equityholders default on debt, the rm is liquidated. Let Ln Y

denote the proceeds from liquidation in stage n. Liquidation proceeds Ln
Y has two components: one from the existing assets in place, and the
other from foregone growth options. Following Leland (1994),
we assume that the rm uncovers 1 A fraction of the present value
An Y from existing n assets in place. Unlike Leland (1994), our
model has growth options. Even though growth options may be less
tangible, they still have scrap value upon liquidation. We calculate the
liquidation value for unexercised growth options in an analogous way as
we do for the liquidation value from existing assets in place. That is, we
assume that the rm collects 1 G fraction of the present value of
unexercised foregone growth options. We use the workhorse
real-option model to assess values for unexercised growth options, as if
these options were stand-alone and nanced solely by equity. Let Gk Y
denote the present value of a stand alone growth option (with exercise
cost Ik and cash-ow multiplier mk 40) under all-equity nancing. The
following lemma summarizes the main results (McDonald and Siegel
1986).
Lemma 1
Consider an all-equity financed firm with a single growth option. The
firm may exercise its stand-alone growth option by paying a one-time
fixed cost Ik, and then generate a perpetual stream of after-tax stochastic
cash flow 1 mk Y, where mk 40 is a constant and the stochastic
process Y is given by Equation (1). The firm (option) value is given by

1

Y
1
ae
Gk Y
4
mk Yk Ik ;
Y < Yae
k ;
Yae
r
k
where Yae
k is the optimal growth-option exercising threshold and is
given by
Yae
k
r 1 I k
;
1 1 1 m k
and 1 is the (positive) option parameter and is given by

2
s3

2
1 4
2
2
2r 2 541:
1 2

2
2
The rms liquidation value in stage n, Ln Y, is then given by

Ln Y 1 A An Y 1 G
N
X
Gk Y;
kn1
where An Y given in Equation (2) is the after-tax present value of the

existing n assets in place, and Gk Y given in Equation (4) is the after-tax
present value of k-th unexercised growth options. The specication of
our liquidation payoff is reasonably general and also intuitive. We allow
for different loss rates A and G for assets in place and growth options,
respectively. For example, the growth options may reect the
embedded human capital of current owners, which the new owners
may not be able to replicate after liquidation. This may suggest that G
is greater than A, although our model specication does not require this
condition. In addition to being realistic, the specication for liquidation
payoffs is also quite tractable, and we have closed-form solutions for
both liquidation values of assets in place and of growth options, as
shown above. Finally, we tie liquidation values for assets in place and
growth options to their respective stand-alone values under all-equity
nancing.
Having detailed the rms decision making in stage n, we introduce a
few value functions, and leave the formal mathematical denition of these
value functions to the Appendix. Let En Y denote equity value in stage n,
(i.e., the present discounted value of all future cash ows accruing to the
existing equityholders after servicing debt and paying taxes). Even
though equity value En Y does not internalize the benets and costs of
debt in stage n, it does include the tax benets and distress costs of debt in
future stages. Let Dn Y denote debt value in stage n. Recall that debt
coupon Cn is serviced until debt is either called back at par Fn at
10
investment time Tin1 , or is defaulted at Tdn . At default, creditors collect

Ln YTdn given in Equation (7), which is less than Cn =r in equilibrium.16
Let Vn Y denote stage-n rm value, which is the sum of equity and debt
values, in that Vn Y En Y Dn Y.
The rm follows the decision-making process sketched earlier during
each stage of its life cycle until it defaults. If the rm survives to exercise
its last growth option (i.e., t TiN ), then the rm has exercised all of its
growth options. The rm then collects MN Y in prot ow from its N
assets in place, servicing debt payment CN and paying taxes, until prot
drops sufciently low, which triggers the rm to default on its outstanding debt with par FN at time TdN . The last-stage default-decision problem
for the mature rm is the one analyzed in Leland (1994).
Having described the decision-making process over the life cycle of the
rm, we next solve the model using backward induction.
2 Solution
We solve our model in four steps. The rst three steps take debt coupon
levels fCn : 1 n Ng and liquidation payoff fLn Y : 1 n Ng in
each stage n as given, and analyze the rms growth option and
default-option-exercising decisions. To be specic, rst, we study the
default decision in stage N (no investment decision in the last stage).
This is effectively the classic capital structure/default problem treated
in Leland (1994). Second, we characterize the rms optimal growthoption and default-option exercising decisions when the rm is in the
intermediate stages of its life cycle (stage 1 to stage N 1). Third, we
analyze the rms initial growth-option exercising decision (no
default decision in stage 0). After solving the investment and default
decisions, we provide optimality conditions for the rms nancing policies fCn : 1 n Ng over its life cycle (stage 0 to stage N 1).
2.1 The nal stage (stage N) in the rms life cycle
In the nal stage, the rm has exercised all its growth options and hence
operates N assets in place, which generate the prot ow at the rate of
P
MN Y with MN N
k1 mk being the total production capacity.
Conditioning on the optimal choice of the investment threshold YiN , we
have the classic Leland (1994) formulation. In addition, the corresponding details of proof are provided in the Appendix.
16
Because default is endogenous, equityholders will never default when liquidation value exceeds the riskfree value of debt.
11
Leland (1994) and Goldstein, Ju, and Leland (2001) show that equity
value EN Y is given by
2

1 CN
1 CN
Y
d
EN Y AN Y
AN YN
; Y YdN ;
r
r
YdN
8
where AN Y given in Equation (2) is the after-tax present value of N
existing assets in place, and the optimal default threshold YdN for a given
coupon CN is given by
YdN
r 2 CN
;
MN 2 1 r
and 2 is given by
2
s3

14
2
2 2

2r 2 5 < 0:
2 2

2
2
10
Equity value EN Y is given by the after-tax present value of all N assets

in place, AN Y, minus the after-tax present value of the risk-free perpetual debt 1 CN =r, plus the option value of default, the last term in
Equation (8). The standard option-value argument implies that the default
threshold YdN decreases with volatility , and equity value EN Y is convex
in Y. Naturally, when Y YdN , equity is worthless, i.e., EN Y 0.
Similarly, the market value of debt DN Y is given by

2
CN
CN
Y
d
DN Y

LN YN
; Y YdN ;
11
r
r
YdN
where the second term captures the default risk. In stage N, the rm only
has assets in place, and therefore, LN YdN 1 A AN YdN . Note that
DN Y is concave in Y because the creditor is short a default option.
The second term in Equation (11) measures the discount on debt owing
to the risk of default, which has two components: the loss given default
CN =r LN YdN for the creditor, and Y=YdN 2 , the present discounted
value for a unit payoff when the rm hits the default boundary YdN .
Intuitively, rm value is VN Y EN Y DN Y which is given by

2
CN
CN
Y
d
VN Y AN Y
A AN YN
; Y YdN : 12
r
r
YdN
Firm value VN Y is given by the after-tax value of the N assets in
place AN Y, plus the perpetuity value of the risk-free tax shield CN =r,
minus the cost of liquidation. Importantly, rm value VN Y is concave
in Y, because the rm as a whole is short in a liquidation option.
12
Intuitively, after TiN , the rm is long in the N assets in place and the riskfree tax shield perpetuity CN =r, and short in the liquidation option.
Upon liquidation, the rm as a whole loses A fraction of assets in
place value AN YdN and also the perpetual value of tax shields, the sum
of the two terms in the square bracket in Equation (12).
2.2 Intermediate stages (stage N 1 to stage 1)
Having analyzed the rms optimization problem in stage N, we now use
backward induction to analyze the rms decision problem in stage
N 1. As Figure 1 indicates, the key decisions are (i) the N-th
growth-option exercising and (ii) the default decision on the existing
debt with par FN1 . For generality, we solve the rms decision problem
for its intermediate stage n, including stage N 1 as a special case.
2.2.1 Equityholders decisions and equity pricing. For given investment
threshold Yin1 and default threshold Ydn in stage n, equity value En Y
solves the following ODE:
rEn Y 1 Mn Y Cn YEn 0Y
Yin1 :
2 2 00
Y En Y;
2
Ydn Y
13
Now consider boundary conditions for investment. When exercising the

n 1-th growth option, equityholders are required to call back the old
debt at the par value Fn. Importantly, we will determine the value of Fn as
part of the model solution that depends on the rms endogenous investment, default, and coupon decisions.
Note that since the rm has to call back the debt at its par, the total cost
of exercising the n 1-th growth option is given by In1 Fn , the sum
of investment cost In1 and the face value of the debt Fn. And part of this
exercising cost is nanced by new debt, which has market value Dn1 Yin1
at issuance time Tin1 . The remaining part In1 Fn Dn1 Yin1 is
nanced by equity. Therefore, the net payoff to equitysholders right
after exercising is En1 Yin1 In1 Fn Dn1 Yin1 : The valuematching condition for the threshold Yin1 is then given by
En Yin1 Vn1 Yin1 In1 Fn ;
14
where Vn1 Y En1 Y Dn1 Y is rm value in stage n 1. Because

equityholders optimally choose the threshold Yin1 , the following smoothpasting condition holds:
E0n Yin1 V0n1 Yin1 :
15
Now turn to the default boundary conditions. Using the same arguments
as those for equity value EN Y in the last stage, equityholders choose the
13
default threshold Ydn to satisfy the value-matching condition En Ydn 0

and the smooth-pasting condition E0n Ydn 0.
Unlike the decision problem in the last stage, we now have a double
(endogenous) barrier option exercising problem, where the upper boundary is primarily about the real-option exercising decision as in McDonald
and Siegel (1986), and the lower boundary is effectively the nancial
default-option-decision as in Leland (1994). Of course, the upper (investment) and the lower (default) boundaries are interconnected. This is
precisely how the investment and default decisions affect each other.
Next, we formally characterize this interaction between investment and
default decisions.
Let in Y denote the present discounted value of receiving a unit
payoff at Tin1 if the rm invests at Tin1 , namely, Tin1 < Tdn . Similarly,
let dn Y denote the present discounted value of receiving a unit payoff at
Tdn if the rm defaults at Tdn , namely Tdn < Tin1 . The closed-form expressions for in Y and dn Y are given by
h
i
1
i
in Y Et erTn1 t 1Tdn 4Tin1
Ydn 2 Y1 Ydn 1 Y2 ;
16
n
h
i
1
d
dn Y Et erTn t 1Tdn <Tin1
Yin1 1 Y2 Yin1 2 Y1 ; 17
n
and
n Ydn 2 Yin1 1 Ydn 1 Yin1 2 40:
18
Using these formulas, we may write equity value En Y as follows:

En Y An Y
1 Cn
ein in Y edn dn Y;
r
Ydn Y Yin1 ;
19
where
ein
Vn1 Yin1

1 Cn
i
In1 Fn An Yn1
40;
r

1 Cn
40:
edn An Ydn
r
20
21
Equity value En Y is given by the after-tax present value of assets in

place An Y minus the after-tax perpetuity value of risk-free debt with
coupon Cn, (i.e. 1 Cn =r) plus two option values: the (real) growth
option and the (nancial) default option. The third term in Equation (19)
measures the present value of the growth option, which is given by the
14
product of in Y, and the net payoff ein from exercising the growth
option. The net payoff ein is the difference between the payoff
from growth-option exercise Vn1 Yin1 In1 Fn and An Yin1
1 Cn =r, the forgone unlevered equity value when investing at the
threshold Yin1 . Note that the forgone un-levered equity value appears
as an additional cost term in the net payoff ein because the option payoff
Vn1 Yin1 In1 Fn contains cash ows from the existing assets in
place. Similarly, the fourth term in Equation (19) is the present value of
the (nancial) default option, which is given by the product of dn Y and
the net payoff edn upon default. Because equityholders receive nothing
at default, the net payoff edn is given by the savings, An Ydn
1 Cn =r40, from avoiding the loss of running the un-levered
equity value at the default threshold Ydn .
2.2.2 Debt pricing.
In the Appendix, we show that debt value Dn Y in stage n where Ydn Y
Yin1 is given by:

Cn
dn Yin
Cn
i
d
d

Y n Y
Ln Yn ;
Dn Y
22
1 in Yin n
r
r
where in Y and dn Y are given in Equation (16) and Equation (17),
respectively. Creditors incur losses when the rm default
(i.e., Cn =r4Ln Ydn ). The second term in Equation (22) gives the
value discount on debt due to the risk of default. We may obtain
the par value Fn of this debt by evaluating Dn Y at the investment
threshold Yin .
Because debt is priced at par Fn at issuance time Tin , we have the
following valuation equation for the par value Fn:

Cn
dn Yin
Cn
d

Ln Yn :
Fn
23
1 in Yin r
r
Default is costly in that Cn =r4Ln Ydn . The second term in Equation (23)
gives the value discount of debt at issuance due to default risk.
2.2.3 Firm valuation. Now, we may calculate rm value Vn Y as the
sum of debt value Dn Y and equity value En Y, in that
Vn Y An Y
Cn
vin in Y vdn dn Y;
r
Ydn Y Yin1 ;
24
where

Cn
vin Vn1 Yin1 In1 An Yin1
;
r
25
15

Cn
vdn Ln Ydn An Ydn
:
r
26
Having described the details to solve for the default threshold Ydn and
the investment threshold Yin1 for stage n 1, we now turn to the investment decision for the initial stage. Unlike the intermediate stages, the
initial stage (stage 0) has no default decision, and hence simplies the
analysis.
2.3 The initial stage (stage 0) in the rms life cycle
As in standard real-option models, equity value E0 Y in stage 0 solves
the following ODE:
rE0 Y YE00 Y
2 2 00
Y E0 Y;
2
Y Yi1 ;
27
subject to the following boundary conditions

E0 Yi1 V1 Yi1 I1 ;
28
E00 Yi1 V01 Yi1 :
29
The intuition behind the value-matching Condition (28) builds and

extends the one in McDonald and Siegel (1986). Without any assets
and liability, the rm raises D1 Yi1 in debt to partially nance the exercising cost I1. Immediately after exercising the rst growth option at the
threshold Yi1 , equityholders collect E1 Yi1 I1 D1 Yi1 giving rise to
the value-matching Condition (28). The smooth-pasting Condition (29)
states that the investment threshold Yi1 is chosen optimally. Finally,
equity value E0 Y also satises the standard absorbing barrier condition
at the origin, in that E0 Y ! 0, when Y ! 0.
Equity value E0 Y, the solution to the above optimization problem, is
given by
1

Y
i
E0 Y
Y

I
30
V
; Y Yi1 ;
1
1
1
Yi1
where 1 is given by Equation (6), and the investment threshold Yi1 solves
the following implicit equation:

1 r 1
C1
1 2 i 2 d 1 i
i
i 1 d
I1
Y1 Y1 v1 Y2 v1 ;
Y1
1 m1 1 1
r
1 1
31
and 1 is a strictly positive constant given in Equation (18) with n 1.
16
Unlike in the standard equity-based real-options models (e.g., McDonald

and Siegel 1986), the payoff from investment in our model is total rm
value V1 Y, which includes the present values of cash ows from both
operations and nancing. Note that equity value E0 Y is convex in Y, a
standard result in the real-options literature.
Having analyzed the rms investment and default thresholds, we now
analyze the rms dynamic nancial (debt) policies, and summarize the
rms integrated dynamic decision making over its life cycle.
2.4 Dynamic debt policies and a summary of the rms life cycle decisions
First, review the decision problem in stage N. The rm chooses its last
default threshold YdN as a function of coupon CN by maximizing equity
value EN Y; CN . The solution for YdN as a function of CN is given by
Equation (9), a well-known problem treated in Leland (1994). Then,
equityholders choose CN to maximize VN Y and then evaluate the
rst-order condition (FOC) for CN at Y YiN . Intuitively, equityholders
internalize all benets and costs of debt issuance at Tin1 and pay fair
market value DN YiN FN when choosing coupon CN.17 Because
rm value VN Y given by Equation (12) is known in closed form, we
obtain the following explicit solution for CN in terms of YiN :
CN
r 2 1 1
MN YiN ;
r 2 h
32
where h is given by
h

i1=2
h 1 2 1 A A
41:
33
Using Formula (9) for YdN for a given coupon CN, we obtain the relationship between the last default threshold YdN and the last growth-option
exercising: YdN YiN =h.
Now consider stage N 1. Equityholders choose the thresholds YiN
and YdN1 to maximize equity value EN1 Y; CN1 , taking the default
threshold YdN in Equation (9) and optimal coupon CN in Equation (32)
in stage N as given. Because equityholders internalize the tax benets
from issuing debt at TiN1 , equityholders choose coupon CN1 to maximize VN1 Y and then evaluate at YiN1 .
Next turn to stage n, where 1 n < N 1. As in stage N 1, the
rm chooses thresholds Yin1 and Ydn to maximize equity value En Y; Cn ,
taking into account the rms future optimality conditions described earlier. Then, the rm chooses the optimal coupon policy Cn to maximize
Vn Y and then evaluate at Yin .
17
The optimality for CN and YiN and the envelope condition jointly imply that we do not need to consider
the feedback effects between the investment threshold YiN and the coupon policy CN.
17
Finally, stage 0 is a special case of stage n. The rm chooses the rst

investment threshold Yi1 to maximize equity value E0 Y. Note that Yd0
0 (no debt and no default). We have shown that equity value E0 Y is
given by Equation (30) and the investment threshold is given by the
implicit nonlinear Equation (31).
Our model thus have predictions for the dynamics of leverage choice
by the rm, and how the leverage dynamics relate to the life cycle of the
rm. Unlike most existing dynamic nancing models, which ignore
investments, our model explicitly incorporates investment frictions,
which are potentially important. The leverage dynamics under investment
frictions will reect the importance of remaining future growth options,
and the potential for premature liquidation from excessive leverage. We
explore this tension later in the paper.18
Having outlined the solution methodology for the general model specication, we next summarize the setting where the rm is all-equity
nanced (i.e., McDonald and Siegel 1986) with multiple growth options
and taxes.
2.5 All-equity-nancing model
We now consider an all-equity setting with multiple rounds of growth
options. Recall that Lemma 1 gives the growth option value Gk Y and
the exercise threshold Yae
k for a rm with a stand-alone growth option.
When the rm has N sequentially ordered growth options, the technology
constraint requires that growth option k can only be exercised if and only
if all previous k 1 growth options have been exercised. Intuitively, this
sequential exercising constraint binds when future growth options
are worthy immediately exercising after growth options are exercised.
We can show that simultaneously exercising growth options k and k 1
is optimal if and only if mk =Ik mk1 =Ik1 . Then we may combine these
two consecutive growth options into one, with an exercise cost Ik Ik1
and a new cash ow multiplier mk mk1 for the combined growth
option. By redening growth options, we can always focus on the setting
where mn =In strictly decreases in n, in that
m1 m2
mN
4
4...4
:
34
I1
I2
IN
Under this condition, the option value of waiting is strictly positive
between any two consecutive growth options in the rst-best (all-equity
nancing) benchmark. The following lemma summarizes the main results
for the equity nancing benchmark.
18
Our characterization of leverage dynamics only requires us to solve a system of non-linear equations for
investment and default thresholds, and coupon policies. This substantially simplies our analysis, in that
we have solved out the endogenous default and investment thresholds up to a set of nonlinear equations.
18
Lemma 2
The firms investment decisions follow stopping time rules Tin inff
t 0 : Yt Yae
n g for n 1; 2; . . . ; N, where the investment threshold
Yae
are
given
by
Equation (5), and the constant 1 is given by Equation
n
(6). Firm (equity) value En Y in stage n is given by the sum of assets in
place An Y and its unexercised growth options, in that
En Y An Y
N
X
Gk Y;
Y Yae
n ; 1 n N;
35
kn1
where Gk Y is the k-th growth option value and is given in Equation (4).
For any stage n, the investment threshold Yae
n is the same as the one if the
n-the growth option were stand-alone. Taxes reduce cash ows but do
not provide benets under all-equity nancing. This explains the factor
1=1 for the investment threshold Yin given in Equation (5). As in
standard real-options models (McDonald and Siegel 1986), the investment threshold Yin increases in volatility. For the ease of future reference,
let Yn denote the n-th investment threshold without taxes 0.
We have
Yn r
1 I n
;
1 1 mn
1 n N:
36
Next, we analyze the investment and nancing decisions for the onegrowth-option setting (N 1).
3 Benchmark: One-Growth-Option Setting
Before delving into the details of the general model where the rm has
multiple rounds of growth options and leverage choices, we rst provide
explicit solutions for the one-growth-option setting in Subsection 3.1
and then highlight important economic insights in Subsection 3.2.
Importantly, we show that this simple one-growth-option setting yields
novel insights that can only be obtained by jointly analyzing the rms
investment and default decisions.
3.1 Closed-form solution
When the rm has only one growth option, we obtain closed-form formulas for the joint investment, leverage, and default decisions. Our onegrowth-option setting can be viewed as a model setting, where McDonald
and Siegel (1986), the seminal real-options model in a Modigliani-Miller
world, meet Leland (1994), the classic contingent-claim tradeoff model of
19
capital structure. The following proposition summarizes the main

results.19
Proposition 1
The firms investment decision follows a stopping time rule
Ti1 infft : Yt Yi1 g, where the investment threshold Yi1 is given by

1 1 ae
Yi1 1
Y1 ;
37
h 1
where the constant h is given in Equation (33) and Yae
1 is the all-equity
investment threshold given in Equation (5). The default time Td1 is given
by Td1 infft4Ti1 : Yt Yd1 g, where the default threshold Yd1 is given by
Yd1 Yi1 =h < Yi1 . The optimal coupon C1 for debt issued at the
investment time Ti1 is given by

r
2 1
1
1

I1 :
38
C1
h
1
2
1
1 1
Firm value V1 Y (after investing at Ti1 ) is given by

2
C
C
Y
V1 Y A1 Y 1 A A1 Yd1 1
;
r
r
Yd1
Y Yd1 :
39
Firm (equity) value E0 Y (before investing at Ti1 ) is given by

Equation (30).
We make the following observations. The investment threshold Yi1 , the
default threshold Yd1 , and the optimal coupon C1 are all proportional to
the investment cost I1. At investment time Ti1 , equity value E1 Yi1 , debt
value D1 Yi1 , and rm value V1 Yi1 are all proportional to the investment cost I1. This implies that the market leverage at the moment of
investment Ti1 ; D1 Yi1 =V1 Yi1 is independent of the size of the investment
cost I1. Next, we turn to the model analysis.
3.2 Model analysis and insights
One of the most important results of real-options analysis is that both the
investment hurdle and option value increase with volatility (by drawing
the analogy to the standard Black-Scholes-Merton option pricing
insight.) We show that debt nancing invalidates this well-known result
19
Mauer and Sarkar (2005) derive similar results for the one-growth-option setting. Their focus on the
results and economic interpretations is very different. We derive explicit formulae and provide explicit
link between investment and default thresholds, while they do not. They contain operating leverage
(variable production costs), and we do not.
20
Option value E0(Y=0.05)
Option value E (Y=0.1)

0
0.65
1.4
0.6
1.35
0.55
0.5
1.3
0.45
0.4
1.25
0.35
0.3
1.2
0.25
0.2
0.1
0.2
0.3
0.4
0.5
1.15
0.1
0.2
0.3
0.4
0.5
Figure 2
Two opposing effects of volatility on the rms option value E0 Y.
Panel A pots the monotonic relation of E0 Y in for Y 0.05. However, for a higher value of Y where
the growth option is deeper in the money, we nd that volatility has a non-monotonic relation on the
value of real option E0 Y. Panel B shows that E0 Y rst decreases and then increases with for Y 0.1.
in the real-options literature, in that the value of growth option may

decrease with volatility. The intuition is as follows. Before debt nancing
and exercising the growth option, the rm is a growth option. What is the
underlying asset for this growth option? Unlike in the standard realoptions setting such as McDonald and Siegel (1986), the value of the
underlying asset is given by the sum of (i) the value of the unlevered
equity and (ii) the (stochastic annuity) value of tax shields (prior to
default) minus (iii) the present value of nancial distress because of equityholders optimal exercising of the ex post default option as in Leland
(1994). The underlying asset value, given by the sum of these three components of the rms value, is therefore concave in Y as we have
shown (because of the short position in the inefcient liquidation loss.)
Therefore, increasing volatility may lower the payoff value from growthoption exercising. This negative effect on the payoff (upon the real-option
exercising) partially mitigates the standard positive-volatility effect on the
real-option value, causing the total effect of volatility on the rms
option value E0 Y to be non-monotonic.
Figure 2 highlights these two opposing effects of volatility on the
option value E0 Y. Panel A shows that the option value E0 Y is increasing in volatility for sufciently low values of Y (e.g., Y 0.05). In this
case, the standard real-option positive-volatility effect on the option
value dominates, because the option value is deep out of the money
and hence the standard real-option convexity argument applies.
However, as Y increases, the growth option becomes sufciently in the
money. In this case, the standard real-option volatility effect becomes less
21
The investment decision Y1
The default decision Y1
0.25
0.06
Debt issue @ endogenous Y i
Debt issue @ Y = 0.07

0
0.2
0.05
0.04
0.15
0.03
0.1
0.02
0.05
0.1
0.2
0.3
0.4
0.5
0.01
0.1
0.2
0.3
0.4
0.5
Figure 3
Volatility effects on the investment threshold Yi1 and the default threshold Yd1 .
This gure shows that even though the investment threshold Yi1 is monotonically increasing in as in
standard real-option models, the default threshold Yd1 is non-monotonic in unlike in the standard
contingent-claim tradeoff model of capital structure due to the endogenous choice of the investment
threshold Yi1 . To demonstrate the endogenous investment threshold effect on Yd1 , we plot the dashed line
in Panel B, which corresponds to the default threshold in Leland (1994), where leverage is chosen at time
zero at a xed value of Y0 for all values of volatility . For the Leland case, we set Y0 0:07, which is the
optimal investment threshold for the case with 10% in our one-growth-option setting (i.e.,
Yi1 0:1 Y0 0:07).
important compared with the negative effect of volatility on the optionexercise payoff becomes more important. As a result, we see that the
option value E0 Y is non-monotonic in volatility in Panel B of
Figure 2, where Y 0.1 is sufciently large. For our example, we nd
that E0 Y 0:1 decreases in for values of < 0:23 and increases in
for 40:23 (i.e., when volatility is sufciently high).20
Next, we turn to the volatility effects on the investment threshold Yi1 and
the default threshold Yd1 . Panel A of Figure 3 shows that the investment
hurdle Yi1 is increasing with volatility , which can be shown by using the
closed-form solution (37). Moreover, this result is consistent with the standard real-options intuition that the higher the volatility , the longer the
rm waits before investing, and hence a higher threshold Yi1 is the result.
However, the default threshold Yd1 is non monotonic in as shown in
Panel B of Figure 3 because the two opposing effects of volatility on the
rms default threshold Yd1 . First, for a given value of Y0, Leland (1994)
and Goldstein, Ju, and Leland (2001) show that the default threshold Yd1
decreases with , consistent with the standard real-option result, despite an
intuitive but involved argument.21 The dashed line in Panel B of Figure 3
20
Miao and Wang (2007) show the opposing effects of volatility on real options in an incomplete-markets
setting where the entrepreneur cannot fully diversify the idiosyncratic risk of the underlying project.
21
To be precise, in Leland (1994), volatility also has two opposing effects on the default threshold
Yd1 . Given coupon C, the higher the volatility , the lower the default threshold due to the standard
22
shows the monotonically decreasing function of Yd1 in volatility . Second,

as debt coupon C is chosen at the moment when the rm exercises its
growth option and the optimal investment threshold Yi1 is increasing in
volatility as shown in Panel A, the rms optimal debt coupon C is also
increasing with , thus providing a channel for the default boundary Yd1 to
potentially increase with . Combining the standard Leland mechanism
with the newly introduced channel (via the endogenous Yi1 at which the
rm issues debt), we see that the default threshold Yd1 rst decreases in
(when the Leland mechanism dominates) but then increases in as the
investment threshold Yi1 signicantly increases.
Our results, that real option value does not necessarily increase with
volatility and the timing of exercising (default/put) options may not
decrease with volatility are more than theoretical possibilities. Indeed,
we think that our results have important practical implications for
rms capital budgeting. Growth option/capital investments are often
nanced by a mixture of debt and equity, but current real-optionsbased capital budgeting recommendations covered in standard MBA
textbooks and practitioners journals hold only under all-equity-nanced
rms. When the M&M theorem does not hold and there is room for debt
nancing, we need to be careful in providing real-option-style capitalbudgeting recommendations. It is no longer clear that we should emphasize the positive effect of volatility on the value of real options nor should
we emphasize that the higher the volatility, the longer we should wait
before exercising the real options.
Having analyzed the two benchmarks, we next turn to the feedback
effects between investment and nancing when the rm has multiple
rounds of growth options.
4 Analysis for the General Model
To highlight the rich interactions between a rms investment and nancing decisions over its life cycle, we rst analyze the two-growth-option
setting (N 2), and then generalize our model to settings with multiple
growth options (e.g., N 3; 4; 5; 6).
4.1 Parameter choices
For the baseline calculation, we use the following annualized parameter
values summarized in Table 1. As in Leland (1994) and the follow-up
dynamic capital structure literature, we choose the risk-free interest rate
real-option convexity argument. However, coupon C is endogenous. Indeed, the higher the volatility ,
the lower the debt coupon C. Therefore, the second (endogenous coupon) effect mitigates the rst option
effect on the default threshold Yd1 , but the overall, the standard real-option effect still dominates the
endogenous coupon effect causing the rms default threshold Yd1 to decrease with .
23
Table 1
Parameter values (annualized whenever applicable)
r
0.05

0.2

0.01

0.2
In
1
mn
0:8
n1
A
G
0.25
0.5
r 0.05, the expected growth rate 0:01, annual volatility 0:2,

and the effective tax rate 0:2. The default cost parameters are A
0:25 for assets in place and G 0:5, implying that the recovery rate for
assets in place is 75% of the unlevered asset value and 50% for the
(unlevered) unexercised growth option value, respectively. Whenever
applicable, all parameter values are annualized.
Without loss of generality, we normalize the cost of exercising the
growth option to unity, in that In 1, for all stages 1 n N. Instead
we capture the net value of each sequential growth option via the the
production capacity (the rate at which output is generated from each
asset in place) mn. We normalize the production capacity in the rst
stage to be unity, m1 1, and decrease the production capacity mn at
an exponential rate (i.e., mn mn1 1 gn ). We choose gn 0:2 for
n 2; 3; . . . , which implies that the protability of each new asset in
place decays at 20%. Therefore, m2 0:8 and m3 0:82 0:64;
m4 0:83 , and m5 0:84 .
4.2 The settings with N 1, 2, 3 growth options
Our framework is general enough that we can allow the rm to have
multiple growth options. Earlier, we treated the case of a rm with two
growth options. We now extend our analysis to the case of a rm that has
three growth options. Panels A, B, and C in Table 2 report the rms
decisions in three models (with 1, 2, and 3 growth options in total,
respectively.)
One growth option (N 1). Panel A of Table 2 summarizes the closedform solution for the one-growth-option setting (N 1 and
m1 1; 0:8; 0:64). First, we analyzes the baseline case with m1 1. The
rm optimally chooses to invest when Y exceeds Yi 0:099, and defaults
when its earning Y falls below Yd 0:038. When exercising its growth
option at Yi 0:099, the rm issues perpetual risky debt with coupon
rate C 0.082 at a credit spread of 108 basis points. The implied initial
leverage is 62.1%. For lower production capacity m1, the rm chooses a
higher investment threshold and also a higher default threshold. For the
case with m1 0:8, we have Yi 0:124 and Yi 0:047, and for the case
with m1 0:64, we have Yi 0:155 and Yi 0:059. Importantly, the
production capacity has no effect on the optimal coupon C and leverage.
24
Table 2
Models where the number of growth options N 1, 2, 3
Stage
Panel A. N 1
1st
1st
1st
Panel B. N 2
1st
2nd
Panel C. N 3
1st
2nd
3rd
Production
capacity m
Investment
threshold Yi
Default
threshold Yd
Coupon
rate C
Leverage
Lev
Credit spreads
cs (bps)
1
0.8
0.64
0.099
0.124
0.155
0.038
0.047
0.059
0.082
0.082
0.082
62.1%
62.1%
62.1%
108
108
108
1
0.8
0.095
0.126
0.030
0.048
0.077
0.188
43.8%
62.1%
129
108
1
0.8
0.64
0.093
0.117
0.158
0.028
0.036
0.060
0.080
0.156
0.319
39.1%
46.6%
62.1%
127
123
108
This table reports results for three settings with N 1, 2, 3. For the setting with N 1, we consider three
subcases with m1 1; 0:8; 0:64. For the setting with N 2, we set m1 1 and m2 0:8. Finally, for the
case with N 3, we set m1 1; m2 0:8, and m3 0:64.
This is due to the endogenous adjustment of investment and default

thresholds such that the rm achieves its optimal leverage ratio at
62.1% at the moment of debt issuance.
Two growth options (N 2). Panel B of Table 2 reports the results for
the two-growth-option setting. First consider the decision rules in the
second (last) stage. After the rst debt is in place, the rm exercises its
second growth option when its earning exceeds Yi2 0:126, larger than
0.124, the investment threshold for the setting with N 1 and m1 0:8
(see Panel A and B of Table 2). This reects the effect of debt overhang,
in that the rm exercises its investment option later than an otherwise
identical rm with this stand-alone growth option (with m1 0:8) does.
The cost of exercising the second growth option is higher now because the
equityholders need to call back the existing debt at par, which potentially
involves the wealth transfer to creditors.22 However, conditioning on
calling back the existing debt and exercising the growth option, equityholders optimally choose their leverage and maximize the rm value
going forward. This again gives rise to 62.1% leverage ratio, the same
level as in the stand-alone one-growth-option setting, which is due to the
scaling invariance property of the leverage ratio for assets in place as in
Leland (1994). Given that the equityholders are investing at a higher
threshold and needs to call back the existing debt, the coupon is naturally
much higher (i.e., C2 0:188) than the coupon for stand-alone onegrowth-option setting (i.e., C 0.082). Despite a higher debt coupon,
22
There is another truncation effect that effects the credit-spread calculation, but that effect is dominated.
25
the credit risk of the second-stage debt is the same as the one in the standalone one-growth option setting. This is in the spirit of Leland (1994) and
our one-growth-option benchmark result.
Now, we turn to the rst-stage decision making and see how the presence of the second growth option inuences the optimal-exercising and
nancing decisions of the rst growth option. First, note the anticipation
effect of the subsequent debt overhang as we have discussed in the previous paragraph. Equityholders anticipate future debt overhang and thus
rationally lowers the leverage and take into account the future conicts of
interest between equityholders and debtholders. This is reected via a
lower leverage, 43.8%, in stage 1 compared with 62.1% in stage 2.
Moreover, the presence of the second growth option raises the rms
debt capacity. Therefore, benets from exercising the rst growth
option are higher in the two-growth-option setting than in the standalone one-growth-option setting with m1 1. Therefore, in the twogrowth-option setting, the payoffs from investing in the rst round are
greater, and hence the rm optimally invests earlier (i.e., Yi1 0:095
compared with Yi1 0:099 for the stand-alone one-growth-option setting
with m1 1). Additionally, the rm defaults later, as we can see from
Yd 0:03, which is lower than 0.038 for the setting with N 1 and
m1 1.
Three growth options (N 3). Here, we show that the exercising timing
decisions for early-rounds growth options are even earlier with more
growth options in the future. One effect of having more future-growth
options is that the rm can raise more debt against future cash ows,
which effectively raises the rms immediate ability to issue debt and
makes investment more attractive. This explains the result that Yi1
0:093 in three-growth-option setting, which is lower than Yi1 0:095
in the two-growth-option setting. Additionally, the leverage chosen by
the rm when it exercises its rst growth option is now lowered to 39.1%
from 43.8%. We see the pattern for eventual convergence as we increase
the number of growth options.
By comparing our results for the two-growth-option setting with those
for the three-growth-option setting, we see that quantitatively the threegrowth-option problem can be somewhat approximately decomposed
into two two-growth-option optimization problems. The intuition is as
follows. The additional effect of current growth-option exercising and
nancing effect on any future growth-option exercising and nancing
decisions beyond the immediate one is quantitatively less signicant.
Using this logic, we may simplify an N-stage growth-option exercising/
nancing problem into N 1 two-growth-option exercising problem.
Of course, our approximation based on the economic insight only
holds for growth options that are sufciently close to each other.
26
For growth options that are somewhat different from each other, our
insights for this approximation may work better if we decompose the
N-growth option problem into a collection of three-growth-option
problems. Based on what we have shown and also what we will show
in the following subsection, we conjecture that for some sophisticated
real-world corporate-investment/nancing-decision problem with many
growth options, we may obtain a good understanding at rst pass (and
potentially also in terms of quantitative analyses) by using a tractable and
plausible setting with only a few growth options (perhaps as few as three
or four.)
4.3 Multiple growth options
In principle, we can extend our analysis to treat a rm with any number,
N, of growth options. Such a treatment will allow us to see, how a rm
optimally decide on its dynamic leverage strategy, while taking into its
cognizance that it may have to issue additional debt to nance future
growth options. Intuitively, we would expect such a rm to start with
fairly low to moderate levels of debt in its early stages and slowly ramp up
the debt level, By doing so, the rm can mitigate the debt-overhang
effects on its growth options in earlier stages, and exploit its steadier
cash ows from assets in place to service a higher level of debt in its
later stages.
4.3.1 N growth options. In Table 3, we present the investment thresholds Yi1 , default thresholds Yd1 , the optimal coupon rate C1 and optimal
leverage Lev1 when the rm is in its rst stage and at the the moment of
exercising its rst growth option, in six setting where the rm faces one to
as many as six growth options into the future. Note that the rst growth
option in all cases has the identical parameter values. Thus, the differences in these models only arise from the future growth options and
assets in places across them.
Table 3
The rst-stage decisions in models with N growth options
Model with N
growth options
Investment
threshold Yi1
Default
threshold Yd1
Coupon
rate C1
Leverage
ratio Lev1
1
2
3
4
5
6
0.099
0.095
0.093
0.093
0.092
0.092
0.038
0.030
0.028
0.027
0.026
0.026
0.082
0.077
0.080
0.082
0.083
0.083
62.1%
43.8%
39.1%
37.0%
35.3%
34.5%
This table reports the rst-stage decisions in a model with N growth options.
We increase N from 1 to 6. Our parameter values are m1 1; m2 0:8;
m3 0:64; m4 0:512; m5 0:410, and m6 0:328. Others are reported in
Table 1.
27
It is worth making the following observations. First, the investment

threshold Yi1 monotonically decreases from Yi1 0:099 for the case with
N 1 to Yi1 0:092 for the case with N 6. Intuitively, the additional
benet of not distorting future investment options in models with more
growth options (a higher value of N) encourages the rm to exercise its
growth options in earlier stages sooner. Second, the optimal leverage level
Lev1 decreases from 62.1% for the case with N 1 to 34.5% for the case
with N 6. This partly reects the debt conservatism as the rm worries
about the debt-overhang costs in the future if leverage is too high. Third,
a rm with more growth options tends to default at a much later point
than a rm with fewer growth options. For example, the default threshold Yd1 decreases from Yd1 0:038 for the case with N 1 to Yd1 0:026
for the case with N 6. This makes intuitive sense: for the rm with many
growth options, the cost of default should include the forgone opportunities associated with the loss of all future growth options, and hence the
rm chooses to default much later in the case with N 6 than that with
N 1, in line with the predictions from the leverage pattern that we
discussed earlier.
To reiterate, leverage, default, and investment thresholds in early
stages all monotonically decrease as the rm has more growth options.
This life-cycle pattern driven by the endogenous composition between
growth options and assets in place is important. Quantitatively, in our
model with N 3 or more growth options, we obtain leverage in the
empirically plausible range of 1/3 for U.S. corporations. Moreover, the
optimal investment and default, as well as coupon decisions, essentially
all converge as we increase the number of growth options to six (i.e.,
N 6). Intuitively, the additional growth option (most likely to be exercised in the distant future if the rm has not defaulted by then) has little if
any effect on the decision making in stage 1 because of the discounting
effect.
4.4 Growth-option liquidation recoveries 1 gG and leverage

Growth options and assets in place generally have different recoveries
during the liquidation process. Intuitively, growth options tend to have
lower debt capacity than assets in place, for various reasons, such as
different degrees of tangibility and also inalienability of human capital
embedded in growth options. To capture this important feature and
analyze its effect on leverage, we allow for growth options and assets
in place to have different recovery rates in liquidation. For simplicity,
we have assumed that the recovery value of an unexercised growth option
is equal to a constant fraction, 1 G , of Gk Y given by Equation (4),
which is the value for an otherwise identical (all-equity) stand-alone
growth option.
28
Table 4
Effects of liquidation loss parameter gG on leverage
Model with N
G 0:25
G 0:5
G 0:75
G 1
1
2
3
4
5
6
62.1%
45.5%
40.1%
37.6%
35.9%
34.9%
62.1%
43.8%
39.1%
37.0%
35.3%
34.5%
62.1%
42.6%
38.0%
36.5%
34.8%
33.8%
62.1%
41.8%
36.9%
35.9%
34.2%
33.2%
This table reports the effect of the growth-option liquidation-loss parameter, G,

on leverage in the rst stage for models with N growth options, where N ranges
from 1 to 6. The liquidation loss parameter for assets in place, A, is xed at 0.25
for all six models. Note that the recoveries of assets in place are weakly greater
than the recoveries of growth options for all six models. Other parameter values
are reported in Table 1.
In this section, we explore the comparative static effects of varying

liquidation loss parameter G on leverage. First, we recall that in our
baseline case (reported in Table 1), we x the liquidation loss parameter
for assets in place, A, to be 25% and the loss parameter for growth
options, G, to be 50%. That is, the recovery value for growth option
is 50% of its market value (if it were stand-alone and all-equity nanced),
which is signicantly smaller than the recovery value for assets in place,
which is 75% of its market value (if it were stand-alone and all-equity
nanced.)
Next, in Table 4, we report leverage in the rst stage by considering
four values for the growth-option liquidation-loss parameter G: 25%,
50% (baseline case), 75%, and 100%. First, we note that for the case with
only one growth option (i.e., N 1), the parameter G has no effect on
leverage choice because the only growth option has already been exercised at the moment of debt nancing and hence changing the effect o G
cannot have any effect on leverage. Second, the higher the loss parameter
for the growth option, G, the more costly the rms default is, and hence
the lower the leverage. For all ve cases with N 2 where the rm still
has un-exercised growth options after making the rst leverage decision,
we see that leverage (in the rst stage) clearly decreases with G. For
example, the rst-stage leverage ratio decreases from 45.5% to 41.8%
when G increases from 0.25 to 1 for N 2.
Third, the rst-stage leverage decreases substantially with N.
Intuitively, a rm with more growth options tends to take a lower leverage (especially at early stages) in order to mitigate the debt-overhang
burdens in the future. The exibility of increasing leverage in the future
when the rm has more assets in place is a valuable option for the rm.23
23
Additionally, by construction, a rm with more growth options in our model also has more nancing
options as we require the rm to choose leverage when exercising a growth option. However, quantitatively,
29
In a model with informational asymmetry and real options, Fulghieri,

Garcia, and Hackbarth (2013) show that high-growth rms may prefer
equity over debt, and then switch to debt nancing as they mature. Their
predictions under a different model formulation are in line with our
models prediction in that leverage increases over the rms life cycle in
our model, as the rm sequentially exercises its growth options and
becomes one with more assets in place.
Does a rm with more growth options have a larger leverage response
to the change of G? The answer is not obvious. For example, for the case
with N 2, leverage in the rst stage decreases from 45.5% to 41.8% by
3.7%, and for the case with N 6, it decreases from 34.9% to 33.2% by
1.7% as we decrease recovery, 1 G , from 75% to zero. It is not easy to
assess the signicance of this difference (3.7% versus 1.7%).
5 Alternative Debt Structures
In our baseline model, we assume that the existing debt is called back and
new debt is issued when the rm exercises its growth option. In reality,
we often see different types of debt in terms of seniority and priority
coexisting. To analyze the effects of debt seniority and priority, we
next consider two widely used alternative debt structures: the APR and
the pari passu structure. For simplicity, we only consider the cases with
two rounds of growth options.
5.1 Formulation and solution under alternative debt structures
Let c1 and c2 denote the coupon rate on the rst and second perpetual
debt, respectively. And let Df2 Y and Ds2 Y denote the market value of
the rst debt and that of the second debt issued at the second investment
time Ti2 , respectively. These debt values (after the second growth option is
exercised (i.e., t Ti2 )) are given by
"Z d
#
T2
f
rst
rTd2 t f
d
D2 Yt E
e
c1 ds e
D2 YT2 ; Ti2 t Td2 ;
t
40
given that it is not uncommon that investment and nancing take place at the same time or within very
close time window, this approximation (of tying a rms investment and nancing decisions at the same
time) appears a second-order issue for a rm that has recurrent and sufciently regular investment
opportunities. Put differently, the additional exibility of having a timing for nancing different from
that for investment may have a second-order effect on rm value (at least for some parameter values).
30
Ds2 Yt
"Z
Td2
rst
c2 ds e
rTd2 t
Ds2 YTd2
Ti2 t Td2 ;
41
where the residual values of the rst and second debt, Ds2 YTd2 and
Df2 YTd2 are given by the debt structures to be discussed later.
The total market value of debt after exercising both growth options is
then given by D2 Y Df2 Y Ds2 Y. Let D1 Y denote the market
value of the rst debt after the rst growth option is exercised, but
before the second growth option or the rst default option is exercised
(i.e., Ti1 < t < Td1 ^ Ti2 . We have
"Z d i
T1 ^T2
d
erst c1 ds erT1 t D1 YTd1 1Td <Ti
D1 Yt E
2
1
t
#
i
erT2 t Df2 YTi2 1Td 4Ti :

1
42
Under the new debt structure, the total coupon level for all outstanding
debt in stage 1 and 2 are C1 c1 and C2 c1 c2 , respectively. The
valuation for equity remains the same as that in the baseline model,
given the specied coupon levels for all outstanding debt.
When the rm defaults, it splits the recovery values to the rst and the
second debt valued at Df2 Yd2 and Ds2 Yd2 , depending on the debt covenants. Assume that the debt covenants will be strictly enforced by the
court without any deviation. Given these endogenous values at the
chosen default boundary Yd2 , we may write the market value of the seasoned debt issued at Ti1 and that of the second debt issued at Ti2 , before
default at Td2 , as follows:
i Y 2
c1 hc1
Df2 Y
Df2 Yd2
; Y Yd2 ;
43
r
r
Yd2
Ds2 Y
i Y 2
c2 hc2
s
d
D2 Y2

;
r
r
Yd2
Y Yd2 ;
44
where Df2 Yd2 and Ds2 Yd2 are valued differently under different debt
structures as we have noted. Thus, the total debt value is D2 Y
Df2 Y Ds2 Y. In addition, the total debt value at default D2 Yd2 is
equal to the total rms liquidation value at default, because equity is
worthless at default.
Now we turn to analyze the effect of the debt structure on the nancing
decision in different stages. First, consider stage 2. Dene Vs2 Y as the
31
sum of equity value E2 Y and Ds2 Y, the value of debt issued when the
rm exercises its second growth option, in that Vs2 Y E2 Y Ds2 Y.
Using Equation (8) and Equation (44), we have
C2 C1
Vs2 Y A2 Y
r

2
C1 C2
Y
s
d
D2 Y2 A2 Yd2
;
r
Yd2
Y Yd2 :
45
The distinction between V2 Y and Vs2 Y is essential for our analysis.

Equityholders no longer care about the payoffs to the rst-round debtholders after collecting the proceeds from the debt issuance at Ti2 . This
creates conicts of interests between equityholders and the existing (rstround) debtholders. Equityholders choose the investment threshold Yi2
and the coupon policy c2 to maximize Vs2 Y, not V2 Y. The rst debt
issued at Ti1 to nance the exercise of the rst growth option generates a
debt-overhang problem and distorts the exercising decision for the second
growth option. Of course, the rst-round debtholders anticipate the equityholders incentives in stage 2 and hence price the debt accordingly in
stage 1 at the moment of debt issuance. As a result, equityholders eventually bear the cost of this debt overhang induced by debt issued in stage
1. Unlike most papers in the literature on debt overhang, the amount of
preexisting debt and hence the severity of debt-overhang in our model is
determined endogenously. We show that different debt structures affect
the debt overhang problem in different ways.
Next, we consider two commonly used debt structures: the absolute
priority rule (APR) and the pari passu structure.
5.2 Absolute priority rule
The long maturity of debt allows us to generate debt overhang in a
convenient way (Myers 1977; Hennessy 2004). Because debt is perpetual
and not callable, the rst debt continues to exist even after exercising the
second growth option.
Now, we consider the APR. For expositional simplicity and concreteness, we assume that the rst debt has seniority over the second debt,
unless otherwise noted. Smith and Warner (1979) document that 90.8%
of their sampled covenants contain some restrictions on future debt issuance. As in Black and Cox (1976), at default, the junior debtholders will
not get paid at all until the senior debtholders are completely paid off.
At the second default threshold Yd2 , the senior debtholders collect
Df2 Yd2 minfF1 ; 1 A A2 Yd2 g;
46
where F1 is the par value of the rst debt and is equal to F1 D1 Yi1 .
32
The payoff Function (46) states that either the senior debtholders receive
F1 at Td2 or collect the total recovery value of the rm 1 A A2 Yd2 at
Td2 . It is immediate to see that under this seniority structure, the junior
debt value at default time Td2 is given by
Ds2 Yd2 maxf1 A A2 Yd2 F1 ; 0g:
47
Let F2 denote the par value of the second debt issued at Ti2 . The second
debt is also issued at par, and thus we have F2 Ds2 Yi2 . Equityholders
receives nothing at default, and hence in equilibrium we have
1 A A2 Yd2 F1 F2 . Even when the senior debtholders receive
par F1 at default time Td2 , senior debtholders still prefer that the rm
does not default. This is intuitive, because the par value F1 < C1 =r.
Debt seniority structure matters not only for payoffs at default boundaries Yd2 as in Black and Cox (1976), but also for the real investment and
nancial-leverage decisions. The costs and benets of issuing debt depend
on the seniority and payoff structures. Moreover, the equityholders
interests and incentives also change over time and after each nancing
and investment decisions. How equityholders incentives change over
time naturally depends on the debt seniority structure.
5.3 Pari passu
Now, we turn to another debt structure, pari passu, which requires that
debt issued at Ti1 and that issued at Ti2 have equal priority in default at
stochastic time Td2 . The total debt recoveries at default Y Yd2 are
proportional to 1 A A2 Yd2 , the total liquidation value of the rm.
Because both types of debt are perpetual, the residual values at the
default threshold Yd2 are thus given by
c1
1 A A2 Yd2 ;
48
Df2 Yd2
c1 c2
Ds2 Yd2
c2
1 A A2 Yd2 :
c1 c2
49
Here, we assume that the payments to debtholders are based on the debt
values at the second investment time Ti2 . This assumption captures the
key feature of the pari passu structure, and substantially simplify the
analysis.24
Equityholders choose c2 at stochastic time Ti2 to maximize Vs2 Y given
in Equation (45), the sum of equity value E2 Y and newly issued debt
value Ds2 Y. The following implicit function characterizes the optimal
24
Under this assumption, we do not need to carry the face values F1 and F2 for both classes of debt. A more
realistic way to model pari passu seniority structure is to make the payment at default proportional to the
face values F1 and F2.
33
Table 5
The effects of debt structures
Stage
Panel A. Baseline
1st
2nd
Panel B. APR
1st
2nd
Panel C. pari passu
1st
2nd
Production
capacity m
Investment
threshold Yi
Default
threshold Yd
Coupon
rate C
Leverage
Lev
Credit spreads
cs (bps)
1
0.8
0.095
0.126
0.030
0.048
0.077
0.188
43.8%
62.1%
129
108
1
0.8
0.096
0.138
0.031
0.044
0.079
0.173
48.8%
54.8%
75
82
1
0.8
0.096
0.115
0.017
0.053
0.041
0.208
23.7%
71.1%
120
151
This table reports results for three commonly used debt restructures: the baseline, APR, and pari passu.
We set m1 1 and m2 0:8 for a model with N 2.
coupon c2 for a given level of the rst coupon c1:

r
2 1
c2 c1
r
2

1=2
1

1
2
2 1 A A =
c1
1
M2 Yi2 :
h
2 1 1 2 1 A A = c1 c2
50
5.4 The effects of debt structures

Table 5 reports the effects of debt structures on investment and default
decisions, optimal coupon and leverage choices, and the equilibrium
credit spread.25 We compare the results from the three debt structures:
APR, pari passu, and our baseline case (where new debt is only issued if
and only if existing debt is retired and paid back in full.)
First, we note that debt structures have signicant implications on the
second growth-option exercising timing. The investment threshold Yi2
0:138 under the APR, which is signicantly larger than Yi2 0:126 under
our baseline case, which in turn is also signicantly larger than Yi2
0:115 under the pari passu case. This sequencing is consistent with our
intuition. As APR offers the strongest protection for the existing
debtholders, the debt-overhang problem is most severe, and hence the
investment threshold is the largest among the three. pari passu gives
the most favorable seniority treatment for the new debt, and hence the
25
Hackbarth and Mauer (2012) also study the priority-structure choice by considering the trade-off
between pari passu and APR.
34
equityholders are least concerned about the debt-overhang burden causing the investment threshold to be even lower than our baseline case.
Second, the optimal leverage Lev and coupon rate C also reect the
severity of debt overhang that depends on the underlying debt structure.
Intuitively, as APR offers the strongest protection for the existing debt
and pari passu offers least protection for debt issued in the rst stage, the
second-stage leverage and coupon rate are the lowest under APR (54.8%
and 0.173, respectively) and the highest under pari passu (71.1% and
0.208, respectively). Because of the anticipated debt-overhang problem
in the second stage, the equityholders choose the rst-stage leverage and
debt coupon in the reverse order, in that leverage and coupon in stage 1
are the highest under APR (48.8% and 0.079) and the lowest under pari
passu (23.7% and 0.041).
Third, the protection of the existing debt also implies that credit spread
for the risky debt is lowest under APR and highest under pari passu in
stage 2.
In summary, we nd that debt structures have substantial effect on
investment timing decisions and leverage through the life cycle of the
rm mostly through the important endogenous debt-overhang channel.
6 Conclusions
Our paper provides an integrated framework for thinking about multiple
rounds of sequential investments simultaneously with dynamic nancing.
Our modeling approach is tractable and provides a coherent way to think
about a complex, dynamic problem that is at the heart of both investments theory and dynamic nancing. Importantly, we show that the rm
substantially lower its leverage in order to take advantage of its growth
options going forward, and indeed we obtain empirically plausible leverage (around 1/3) with as few as three growth options. Besides providing a
natural way to model the life cycle of the rm, our model also highlights
the need to distinguish between the residual values of assets in place
and that of remaining live-growth options upon default.
In addition, we nd that debt seniority and debt priority structures
have conceptually important and quantitatively signicant implications
on growth-option exercising and leverage decisions, because different
debt structures (e.g., APR versus pari passu) have very different endogenous debt-overhang implications.
Finally, mainly for tractability reasons, we have assumed that nancing and investment timing decisions coincide and also side-stepped
from some important frictions that may inuence a rms nancing
and investment decisions over its life cycle. We see generalizing our
model to separate investment and nancing decisions as a natural
next step. More broadly, we expect that future research will incorporate
35
important frictions, such as dynamic corporate liquidity considerations,

contractual frictions because of moral hazard, and informational frictions, into a corporate life cycle framework similar to ours. Over the
last decade or so, we have seen signicant progress in the development
of dynamic corporate nance models that can be structurally estimated,
such as Hennessy and Whited (2005, 2007). We have also seen a fast
growing dynamic nancial contracting literature, including DeMarzo
and Fishman (2007), DeMarzo and Sannikov (2006), and DeMarzo
and others (2012). Our model builds on the classic McDonald-SiegelLeland contingent-claims real-options framework. We expect to see fruitful cross-overs among the three different but highly complementary
approaches from which researchers can draw to further deepen our
standing of the dynamics of corporate nancial and investment decision
making. For example, Asvanunt, Broadie, and Sundaresan (2007) and
Bolton, Wang, and Yang (2014) study the effect of liquidity on real
option decisions and valuation.
A Appendix
We provide derivations and proofs for various results used in the main text.
A.1 Derivations of Main Results in Section 2

We solve the rms decision problem and its valuation equations for debt, equity, and rm
using backward induction. First, note that the decision problem in the last stage is the
standard model analyzed in Leland (1994).
A.2 The Final Stage (Stage N)

A.2.1 Equity pricing. Using the standard valuation argument, we may value equity
EN Y using the following ordinary differential equation (ODE):
rEN Y 1 MN Y CN YE0N Y
2 2 00
Y EN Y;
2
Y YdN ;
A:1
subject to the following conditions at the endogenously chosen default boundary YdN :
EN YdN 0;
A:2
E0N YdN 0:
A:3
The value-matching (A.2) states that equity value is zero when equityholders default. The
smooth-pasting (A.3) implies that equityholders optimally choose the default boundary YdN .
Moreover, the default option is completely out of money when Y approaches 1.
A.2.2 Debt pricing. Similarly, using the standard valuation argument, we may value
debt DN Y using the following ODE:
rDN Y CN YD0N Y
36
2 2 00
Y DN Y;
2
YdN Y;
A:4
subject to the following conditions:

DN YdN LN YdN ;
A:5
CN
:
r
A:6
lim DN Y
Y!1
A.3 Intermediate Stages (Stage N 1 to Stage 1)

Now consider the rms intermediate stage n, where 1 < n < N. Given the default threshold
Ydn and the investment threshold Yin1 , we may write down equity value as in Equation (19),
using in Y and dn Y, which are the present discounted value of receiving a unit payoff
contingent on the n 1-th growth option exercised before the rm defaults at Tdn or not,
respectively. Formally,
i
in Y Et erTn1 t 1Tdn 4Tin1 ;
and
dn Y Et erTn t 1Tdn <Tin1 ;
A:7
Tdn 4Tin1 ,
we have 1Tdn 4Tin1 1:

where 1Tdn 4Tin1 and 1Tdn <Tin1 are the indicator functions. If
Otherwise,
1Tdn 4Tin1 0:
It
is
important
to
see
that
dn Ydn in Yin1 1; dn Yin1 in Ydn 0,
and
dn Y40; in Y40,
for
Ydn < Y < Yin1 .
Using the equity value Formula (19) and the rm value Formula (24), we have
YE0n Y An Y ein i0n YY edn d0
n YY;
A:8
YV0n1 Y An1 Y vin1 i0n1 YY edn1 d0

n1 YY:
A:9
Applying the smooth pasting condition E0n Yin1 V0n1 Yin1 to (A.8) and (A.9) gives
"
#

ei Yd 2 edn Yin1 2 vin1 Ydn1 2 vdn1 Yin2 2
1
mn1 Yin1 1 Yin1 1 n n

r
n
n1
"
#

i
d
i
d 1
d
i
1
vn1 Yn1 1 vdn1 Yin2 1
2 en Yn en Yn1
i

2 Yn1
:
n
n1
A:10
Similarly, the smooth pasting condition E0n Ydn 0 gives
0 An Ydn 1 Ydn 1
ein Ydn 2 edn Yin1 2

ei Yd 1 edn Yin1 1
2 Ydn 2 n n
:
n
n
A:11
For given investment threshold Yin1 and default threshold Ydn in stage n, debt value Dn
Y solves the following ODE:
rDn Y Cn YD0n Y
2 2 00
Y Dn Y;
2
Ydn Y Yin1 ;
A:12
subject the following boundary conditions:

Dn Ydn Ln Ydn ;
A:13
Dn Yin1 Fn :
A:14
37
Using the ODE (A.12) for debt value Dn Y and the corresponding boundary conditions
(A.13) and (A.14), we have the debt value is given by:

Cn
Cn
Cn

Fn in Y
Ln Ydn dn Y;
Dn Y
Ydn Y Yin1 :
A:15
r
r
r
Intuitively, debt value is given by the risk-free debt value Cn =r, minus the present value of
the discount when debt is called back at Tin1 , and minus the present value of the loss when
the rm defaults at Tdn . Because debt is priced at par Fn at issuance time Tin , using the debt
pricing Formula (A.15), we have the par value is given by Equation (23). And then using
Equation (23), debt value Dn Y is then given by Equation (22).
Now turn to the rms decision making in stage 0. Substituting the conjectured equity
value Equation (30) into the ODE Equation (27) and applying the endogenous default
boundary conditions (28) and (29) give the following implicit equation for the rst investment threshold Yi1 :

1
C1 i01 Yi1 Yi1 1 i1 Yi1 i d0
Yi Yi 1 d1 Yi1 d
I1
v1 1 1 1
v1 :
A1 Yi1
1
1
1 1
r
A:16
Simplifying the Equation (A.16) gives Equation (31). Note that we may also obtain the
same results directly using the general Formulation (A.10) with the following properties:
Yd0 0, ed0 0; ei0 V1 Yi1 I1 , and 0 Yd0 2 Yi1 1 .
A.4 Proof of Lemmas 1 and 2

Following the standard real-option analysis (e.g., McDonald and Siegel 1986), the value of
the growth option, Gk Y, for a stand-alone investment opportunity with one-time exercise
cost Ik that generates cash ow 1 mk Y, solves the following ODE:
rGk Y YG0k Y
2 2 00
Y Gk Y;
2
Y Yae
k ;
A:17
subject to the following boundary conditions

Gk Yae
k
1 mk Yae
k
Ik ;
r
A:18
1 mk
:
r
A:19
G0k Yae
k
In addition, we have the absorbing barrier condition Gk 0 0 because Y is a GBM process. Using the standard guess-and-verify procedure, we obtain the option value
ae
Formula (4) for Gk Y for Y Yae
k , and the growth-option exercise threshold Yk given
in Equation (5).
When the rm is all-equity nanced and holds a sequence of decreasingly attractive
growth option, the exercising decisions of each option is independent of the exercising
decisions of other options. This is a robust result under all-equity nancing. Therefore,
when m1 =I1 4m2 =I2 4 . . . 4mN =IN holds, all-equity-nanced rm value is given by the sum
of assets in place and unexercised growth options. That is, in stage n, the rm has n existing
assets in place valued at An Y and N n unexercised growth options. Each growth option is
valued at Gk Y with exercising cost Ik and cash-ow multiple mk. Total rm value is then
given by Equation (35). Note that the growth options are ordered sequentially from the
most attractive (1st growth option) to the least attractive is without loss of generality. See
the main text for discussions on how we may redene the growth options if the preceding
growth option is more attractive. (For example, if m2 =I2 m1 =I1 , we can combine the rst
38
two growth options and relabel the option with exercising cost I1 I2 and cash-ow multiple m1 m2 .)
A.5 Proof of Proposition 1

When the rm has only one growth option, by denition, the exercise threshold for the
second growth option is innite, in that Yi2 1. Therefore, Equation (16) and Equation
(17) imply i1 Y 0 and d1 Y Y=Yd1 2 , for Y Yd1 . Since the rm only has one
growth option, we have

C1
C1
vd1 L1 Yd1 A1 Yd1
A A1 Yd1
:
A:20
r
r
Equation (31) for the rst investment threshold Yi1 thus implies
"

#
1 r 1
C1
1 2
C1 Yi1 2
i
d
:
I1
Y1
A A1 Y1
1 m1 1 1
r
1
r
Yd1
A:21
The optimal coupon policy is given by

C1
r 2 1 1
m1 Yi1 :
r 2 h
A:22
Rearranging and simplifying Equation (A.21) gives the following implicit equation for the
investment threshold:

C1
C1
2
1 2
A 1
1 1A1 Yi1 1 I1 1

r
r
2 1

C1
C1 h 2
A:23
1 2
h2 ; h2 ;
1 I1 1
r
r 1 2
1 I1 1 1
1 m1 Yi1
;
hr
where the rst, second, and third line uses the explicit formulae for Yd1 as a function of C1
given in Equation (9) (when stage 1 is the last stage (i.e. N 1)), h given in Equation (33),
and coupon C1 given in Equation (A.22), respectively. Finally, re-arranging the last expression gives Yi1 in Equation (37). Substituting Equation (37) into Equation (A.22) gives the
coupon policy Equation (38) and the default threshold Yd1 Yi1 =h.
If the initial value Y0 is below the investment threshold Yi1 given in Equation (37),
the rm will wait before investing. Equity value before investment E0 Y is given by
Equation (30).
References
Abel, A., and J. Eberly. 1994. A unied model of investment under uncertainty. American Economic
Review 84:136984.
Anderson, R., and S. Sundaresan. 1996. Design and valuation of debt contracts. Review of Financial
Studies 9:3768.
Asvanunt, A., M. Broadie, and S. Sundaresan. 2007. Growth options and optimal default under liquidity
constraints: The role of corporate cash balances. Research Paper, Columbia Business School.
Black, F., and J. Cox. 1976. Valuing corporate securities: Some effects of bond indenture provisions.
Journal of Finance 81:35167.
Black, F., and J. Scholes. 1973. The pricing of options and corporate liabilities. Journal of Political
Economy 81:63759.
39
Bolton, P., N. Wang, and J. Yang. 2014. Investment under uncertainty and the value of nancial and real
exibility. Research Paper, Columbia Business School.
Brennan, M., and E. Schwartz. 1984. Optimal nancial policy and rm valuation. Journal of Business 39:
593609.
. 1985. Evaluating natural resource investments. Journal of Business 58:13557.
Caballero, R. 1999. Aggregate investment. In J. Taylor, and M. Woodford (Eds.), Handbook of macroeconomics, vol. 1B, pp. 81362. Amsterdam: North-Holland.
Collin-Dufresne, P., and R. Goldstein. 2001. Do credit spreads reect stationary leverage ratios? Journal
of Finance 56:192957.
DeMarzo, P., and M. Fishman. 2007. Optimal long-term nancial contracting. Review of Financial
Studies 20:2079128.
DeMarzo, P., M. Fishman, Z. He, and N. Wang. 2012. Dynamic agency and the q theory of investment.
Journal of Finance 67:2295340.
DeMarzo, P., and Y. Sannikov. 2006. Optimal security design and dynamic capital structure in a continuoustime agency model. Journal of Finance 61:2681724.
Dixit, A., and R. Pindyck. 1994. Investment under uncertainty. Princeton, NJ: Princeton University
Press.
Dudley, E. 2012. Capital structure and large investment projects. Journal of Corporate Finance 18(5):
116892.
Dufe, D. 2001. Dynamic asset pricing theory. 3rd ed. Princeton, NJ: Princeton University Press.
Fan, H., and S. Sundaresan. 2000. Debt valuation, renegotiation, and optimal dividend policy. Review of
Financial Studies 13:105799.
Fischer, E., R. Heinkel, and J. Zechner. 1989. Dynamic capital structure: Theory and tests. Journal of
Finance 44:1940.
Fulghieri, P., D. Garcia, and D. Hackbarth. 2013. Asymmetric information and the pecking (dis)order.
Working Paper, University of North Carolina and Boston University.
Goldstein, R., N. Ju, and H. Leland. 2001. An EBIT-based model of dynamic capital structure. Journal
of Finance 74:483512.
Graham, J. 2000. How big are the tax benets of debt? Journal of Finance 55:190141.
Grenadier, S. 1996. The strategic exercise of options: development cascades and overbuilding in real
estate markets. Journal of Finance 51:165379.
Grenadier, S., and N. Wang. 2005. Investment timing, agency and information. Journal of Financial
Economics 75:493533.
. 2007. Investment under uncertainty and time-inconsistent preferences. Journal of Financial
Economics 84:239.
Guo, X., J. Miao, and E. Morellec. 2005. Irreversible investment with regime shifts. Journal of Economic
Theory 122:3759.
Hackbarth, D., and D. Mauer. 2012. Optimal priority structure, capital structure, and investment.
Review of Financial Studies 25:74796.
Hackbarth, D., J. Miao, and E. Morellec. 2006. Capital structure, credit risk, and macroeconomic
conditions. Journal of Financial Economics 82:51950.
Harris, M., and A. Raviv. 1991. The theory of capital structure. Journal of Finance 44:297355.
40
Hart, O., and J. Moore. 1994. A theory of debt based on the inalienability of human capital. Quarterly
Journal of Economics 109:84179.
Hennessy, C. 2004. Tobins q, debt overhang, and investment. Journal of Finance 59:171742.
Hennessy, C., and T. Whited. 2005. Debt dynamics. Journal of Finance 60:112965.
. 2007. How costly is external nancing? Evidence from a structural estimation. Journal of Finance
62:170545.
Ju, N., and H. Ou-Yang. 2006. Asset substitution and underinvestment: A dynamic view, Working
Paper, Duke University and HKUST.
Kane, A., A. Marcus, and R. McDonald. 1984. How big is the tax advantage of debt? Journal of Finance
39:84155.
. 1985. Debt policy and the rate of return premium to leverage. Journal of Financial and
Quantitative Analysis 20:47999.
Kim, I., K. Ramaswamy, and S. Sundaresan. 1993. Does default risk in coupons affect the valuation of
corporate bonds?: A contingent claims model. Financial Management 22:11731.
Lambrecht, B. 2001. The impact of debt nancing on entry and exit in a duopoly. Review of Financial
Studies 14:765804.
Leary, M., and M. Roberts. 2005. Do rms rebalance their capital structures? Journal of Finance 60:
2575619.
Leland, H. 1994. Corporate debt value, bond covenants, and optimal capital structure. Journal of
Finance 49:121352.
. 1994b. Bond prices, yield spreads, and optimal capital structure with default risk, Working Paper,
University of California at Berkeley.
. 1998. Agency costs, risk management, and capital structure. Journal of Finance 53(4):121343.
Longstaff, F., and E. Schwartz. 1995. A simple approach to valuing risky xed and oating rate debt.
Journal of Finance 50(3):789819.
Mauer, D., and S. Sarkar. 2005. Real options, agency conicts, and optimal capital structure. Journal of
Banking and Finance 29:140528.
McDonald, R., and D. Siegel. 1985. Investment and the valuation of rms when there is an option to
shut down. International Economic Review 26(2):33149.
. 1986. The value of waiting to invest. Quarterly Journal of Economics 101:70728.
Mella-Barra, P., and W. Perraudin. 1997. Strategic debt service. Journal of Finance 52:53156.
Merton, R. 1973. Theory of rational option pricing. Bell Journal of Economics and Management Science
4:14183.
. 1974. On the pricing of corporate debt: The risk structure of interest rate. Journal of Finance 29:
44969.
Modigliani, F., and M. Miller. 1963. Corporate income taxes and the cost of capital: A correction.
American Economic Review 53:43343.
Miao, J., and N. Wang. 2007. Investment, consumption, and hedging under incomplete markets. Journal
of Financial Economics 86:60842.
Myers, S. 1977. Determinants of corporate borrowing. Journal of Financial Economics 5:14775.
Smith, C., and J. Warner. 1979. On nancial contracting: An analysis of bond covenants. Journal of
Financial Economics 7:11761.
41
Stein, J. 2003. Agency, information, and corporate investment. In G. M. Constantinides, M. Harris, and
R. Stulz (Eds.), Handbook of the Economics of Finance, pp. 10963. Amsterdam: North-Holland.
Strebulaev, I. 2007. Do tests of capital structure theory mean what they say? Journal of Finance 62(4):
174787.
Sundaresan, S., and N. Wang. 2006. Dynamic investment, capital structure, and debt overhang, Working
Paper, Columbia Business School.
. 2007. Investment under uncertainty with strategic debt service. American Economic Review
Papers and Proceedings 97(2):25661.
Titman, S., and S. Tsyplakov. 2007. A dynamic model of optimal capital structure. Review of Finance 11:
40151.
Tserlukevich, Y. 2008. Can real options explain nancing behavior? Journal of Financial Economics 89:
23252.
42

Suresh Sundaresan

Uploaded by

Copyright:

Available Formats

Suresh Sundaresan

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Suresh Sundaresan

Uploaded by

Copyright:

Available Formats

Dynamic Investment, Capital Structure,

and Debt Overhang*

Models of truly intertemporal investments with irreversibility and models

Review of Corporate Finance Studies / v 4 n 1 2015

independent of each other. The literature on intertemporal investments

Dynamic Investment, Capital Structure, and Debt Overhang

Review of Corporate Finance Studies / v 4 n 1 2015

Dynamic Investment, Capital Structure, and Debt Overhang

capital structure decisions. Fischer, Heinkel, and Zechner (1989);

Review of Corporate Finance Studies / v 4 n 1 2015

interests of existing equityholders at each point in time.9 At time zero, the

and W is a standard Brownian motion. Equivalently, we may also

Let W be a standard Brownian motion in R on a probability space ; F; Q and x the standard

Dynamic Investment, Capital Structure, and Debt Overhang

Review of Corporate Finance Studies / v 4 n 1 2015

it ceases to exist. All proceeds from liquidation go to creditors. However,

Dynamic Investment, Capital Structure, and Debt Overhang

It is straightforward to describe the rms stage-n decision problem.

When equityholders default on debt, the rm is liquidated. Let Ln Y

Review of Corporate Finance Studies / v 4 n 1 2015

process Y is given by Equation (1). The firm (option) value is given by

and 1 is the (positive) option parameter and is given by

The rms liquidation value in stage n, Ln Y, is then given by

where An Y given in Equation (2) is the after-tax present value of the

Dynamic Investment, Capital Structure, and Debt Overhang

investment time Tin1 , or is defaulted at Tdn . At default, creditors collect

Review of Corporate Finance Studies / v 4 n 1 2015

Equity value EN Y is given by the after-tax present value of all N assets

Dynamic Investment, Capital Structure, and Debt Overhang

Now consider boundary conditions for investment. When exercising the

where Vn1 Y En1 Y Dn1 Y is rm value in stage n 1. Because

Review of Corporate Finance Studies / v 4 n 1 2015

default threshold Ydn to satisfy the value-matching condition En Ydn 0

Using these formulas, we may write equity value En Y as follows:

Equity value En Y is given by the after-tax present value of assets in

Dynamic Investment, Capital Structure, and Debt Overhang

Review of Corporate Finance Studies / v 4 n 1 2015

subject to the following boundary conditions

E00 Yi1 V01 Yi1 :

The intuition behind the value-matching Condition (28) builds and

Dynamic Investment, Capital Structure, and Debt Overhang

Unlike in the standard equity-based real-options models (e.g., McDonald

Review of Corporate Finance Studies / v 4 n 1 2015

Finally, stage 0 is a special case of stage n. The rm chooses the rst

Dynamic Investment, Capital Structure, and Debt Overhang

Review of Corporate Finance Studies / v 4 n 1 2015

capital structure. The following proposition summarizes the main

Firm (equity) value E0 Y (before investing at Ti1 ) is given by

Dynamic Investment, Capital Structure, and Debt Overhang

Option value E0(Y=0.05)

Option value E (Y=0.1)

in the real-options literature, in that the value of growth option may

Review of Corporate Finance Studies / v 4 n 1 2015

The investment decision Y1

The default decision Y1

Debt issue @ endogenous Y i

Debt issue @ Y = 0.07

Dynamic Investment, Capital Structure, and Debt Overhang

shows the monotonically decreasing function of Yd1 in volatility . Second,

Review of Corporate Finance Studies / v 4 n 1 2015

and 1 is the (positive) option parameter and is given by

This table reports the effect of the growth-option liquidation-loss parameter, G,

in Y Et erTn1 t 1Tdn 4Tin1 ;

dn Y Et erTn t 1Tdn <Tin1 ;

ein Ydn 2 edn Yin1 2