Dynamic Models and Structural Estimation in Corporate Finance

Ilya A. Strebulaev and Toni M. Whited⇤

first version, June 2012

revised, December 2012

ABSTRACT

We review the last two decades of research in dynamic corporate finance, focusing on capital
structure and the financing of investment. We first cover continuous time contingent claims mod-
els, starting with real options models, and working through static and dynamic capital structure
models. We then move on to corporate financing models based on discrete-time dynamic invest-
ment problems. We cover the basic model with no financing, as well as more elaborate models
that include features such as costly external finance, cash holding, and both safe and risky debt.
For all the models, we o↵er a minimalist, simplified presentation with a great deal of intuition.
Throughout, we show how these models can answer questions concerning the e↵ects of financial
constraints on investment, the level of corporate leverage, the speed of adjustment of leverage to
its target, and market timing, among others. Finally, we review and explain structural estimation
of corporate finance models.

⇤
Strebulaev is from Stanford Graduate School of Business and the National Bureau of Economic Research; Whited
is from the University of Rochester. Yunjeen Kim provided excellent research assistance. We are grateful for help-
ful discussions with Michael Brennan, Harry DeAngelo, Will Gornall, Dirk Hackbarth, Akitada Kasahara, Arthur
Korteweg, Boris Nikolov, Yuri Salitsky, Amit Seru, and Luke Taylor. We also would like to thank the anonymous
reviewer and Franklin Allen (the Editor) for helpful suggestions.

Electronic copy available at: [Link]

Contents
1 Introduction 1

2 Dynamic Contingent Claims Models 3

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2 Real Options and Dynamic Investment . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2.1 Real Option Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2.2 Optimal Investment Policy . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2.3 Discussion and Extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.3 Optimal Static Capital Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3.1 Unlevered firm value . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3.2 Default . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.3.3 Equity and Debt Valuation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.3.4 Optimal Capital Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3.5 Comparative statics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.3.6 Optimal Capital Structure with Reorganization . . . . . . . . . . . . . . . . . 22
2.3.7 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.3.8 Quantitative implications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.4 Optimal Dynamic Capital Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.4.1 Model Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.4.2 Comparative Statics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.4.3 Quantitative Implications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.4.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.5 Cross-Sectional Implications of Dynamic Capital Structure . . . . . . . . . . . . . . 36
2.5.1 Leverage in True Dynamics vs. Leverage at Refinancing Points . . . . . . . . 37
2.5.2 Empirical Capital Structure Studies . . . . . . . . . . . . . . . . . . . . . . . 40
2.6 Macroeconomic Fundamentals and Capital Structure . . . . . . . . . . . . . . . . . . 43
2.7 Debt Structure and Strategic Renegotiations . . . . . . . . . . . . . . . . . . . . . . 47
2.8 Capital Structure and Temporary vs. Permanent Shocks . . . . . . . . . . . . . . . . 51
2.8.1 Impact of Temporary Shocks . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
2.8.2 Modeling Temporary Shocks . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
2.9 Further Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

3 Discrete-Time Investment Models 64

3.1 Basic Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.1.1 First-Order Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
3.1.2 Numerical Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.1.3 Model Intuition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
3.2 Costly External Finance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
3.3 Cash . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
3.4 Risk-Free Debt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
3.5 Cash and Debt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
3.6 Risky Debt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93

Electronic copy available at: [Link]

 3.7 Other Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
3.8 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98

4 Structural Estimation 99
4.1 GMM and Euler Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
4.2 Simulated Method of Moments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
4.2.1 Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
4.2.2 Identification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
4.2.3 Practical Advice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
4.2.4 Model Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
4.3 Simulated Maximum Likelihood . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
4.4 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
4.5 Calibration versus Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
4.6 Other Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118

5 Conclusion 119

References 122

Tables and Figures 132

1 Introduction

Over the last 20 years, research on dynamic corporate finance has witnessed dramatic growth in

both theoretical and empirical directions. Several interrelated factors have led to this advance-

ment. First, although it has long been recognized that most financial decisions involve inherently

dynamic interactions (e.g. Lintner 1956), the development of methods necessary to tackle a number

of important dynamic issues has lagged. However, advances in stochastic dynamic optimization

techniques, contingent claims asset pricing analysis, game theory, and dynamic investment model-

ing have opened avenues for the fruitful study of dynamic problems in corporate finance. Second,

there has been contemporaneous development in structural empirical methods. Third, researchers

have gained access to more and significantly higher-quality data, as well as to dramatically better

computing power, which makes analyzing such data possible in a reasonable amount of time. These

resources have allowed researchers to take dynamic models and empirical methods seriously and,

more importantly, to ask more demanding questions from the data and methods. Fourth, it is

gradually becoming clearer that static models and the intuition they imply often fail to explain

even simple, first-order stylized facts. In contrast, a dynamic paradigm allows the formalization

and exploration of new questions that are either irrelevant or impossible to address in a traditional

set-up.

Given this background, the goal of this survey is threefold. First, we wish to explain the

models and techniques used in this literature as simply as possible, with the goal of making the

literature more accessible. Many of the published papers in dynamic corporate finance contain

models with many details. Although detail adds to the realism and rigor of the research, the

unintended consequence is that the models appear to be black boxes with many indistinguishable

moving internal parts. In reality, the intuition behind the models used in this literature is simple,

and our goal is to reveal this simplicity. As such, we do not claim to o↵er a full-blown and complete

overview of dynamic theoretical and empirical models in dynamic corporate finance. We also do

not claim to be rigorous, precise, or generic. Rather, we o↵er an intuitive presentation with much

less mathematical formalism than has been used in the papers we survey. For technical details and

1
formal proofs of many results, we refer readers to the appropriate studies.

Our second goal is to introduce the reader to the main strands of this literature. Thus, our

paper can be viewed in part as a literature review and in part as a tutorial. We spend a great

deal of time on dynamic capital structure and on corporate investment, both of which have been

the home of much, though certainly not all, of the work in dynamic corporate finance. Looking at

these areas via the lens of a dynamic model helps us understand questions that would be impossible

to tackle in a static framework. For example, the optimal timing of investment projects, equity

issuance, and debt refinancing is inherently dynamic, as is the speed of adjustment of a leverage

ratio to its target. Taking dynamics seriously also helps shed new light on questions that can be

examined in static frameworks. Important examples include the detection of financial constraints,

the corporate diversification discount, the low-leverage puzzle, the e↵ects of agency on leverage and

investment, and the anomalous negative sensitivity of leverage to income, which stands in the face

of static capital structure theories that predict higher leverage for high-income firms.

One particular advantage of using dynamic models is that they can often provide quantitative,

rather than simply qualitative implications. This distinction is especially important in areas such

as capital structure, in which the relative magnitudes of the costs and benefits of leverage have been

the center of much of the research agenda. Throughout we delineate the similarities and di↵erences

between the di↵erent classes of dynamic models that we review.

Our third goal is to explain how to estimate dynamic models. Once again, our intent is not

to provide a rigorous econometric treatment but a practical, ”hands-on” guide to three specific

methodologies that have been used in the literature: generalized method of moments, simulated

method of moments, and maximum simulated likelihood. We also provide a concise guide to the

extant structural estimation literature in corporate finance.

Space constraints necessitate several compromises. We do not cover recent developments in

dynamic principal agent models. We also omit discussion of the relatively widely used class of

dynamic models based on two to three time periods. We also leave out the large, closely related

literature on dynamic models of credit spreads.

2
 The rest of the paper proceeds as follows. Section 2 provides an overview of dynamic corporate

finance models based on techniques developed in the continuous time contingent claims literature.

We start with real options and then move on to static and dynamic capital structure models. We

conclude this section by surveying the rest of the literature based on this class of models. Section

3 covers a separate strand of the literature that stems from a completely di↵erent formal base—

discrete time investment models. Here, we cover the basic model without financing, and then move

through the literature by expanding this basic model one feature at a time. Again, we conclude

this section with a broad overview of the literature. Section 4 then reviews the relatively small

number of di↵erent econometric techniques that have been used to estimate these models, as well

as the studies that have used them. We close with a brief overview of directions for future research.

Because we are reviewing highly diverse sets of models, we need to use a great deal of notation.

We therefore define all of the symbols that we use below in Table 1.

2 Dynamic Contingent Claims Models

We start by providing an informal, intuitive discussion that illuminates the nature of dynamic

contingent claims models and their contribution to our understanding of corporate finance. We

then present the basic structure of several concrete models that serve as fundamental building

blocks and discuss useful implications and applications.

2.1 Introduction

Research that explores dynamic contingent claims models1 sits at the intersection of asset pricing

and corporate finance. It has been subject to particularly rapid development in recent years. These

models borrow from extensive theoretical developments in related fields, such as asset pricing,

macroeconomics, and stochastic processes.

To understand how these models work generically and how they di↵er from other theoretical

frameworks in corporate finance, note that they start with the acknowledgment that any claims
1
These models are often confusingly called structural dynamic models, mostly for historical reasons rather than
for any connection to other uses of “structural” in economics.

3
on corporate cash flow streams are derivatives on underlying firm value or firm cash flows. This

means that we can apply option pricing methods to value these claims, assuming initially that

all financial and investment decisions by economic agents are fixed. Indeed, the e↵orts by Black

and Scholes (1973) were originally aimed at describing the pricing of corporate securities such as

warrants rather than exchange-traded options. In his famous quote, Merton (1974) asserts that

“while options are highly specialized and relatively unimportant financial instruments . . . the same

basic approach could be applied in developing a pricing theory for corporate liabilities in general”

(p. 449).

The exact nature of the underlying variables depends on the specific application (e.g., stock

price in the Black-Scholes model; firm value, firm cash flows, or prices of firm output and input

in other corporate finance settings). These underlying variables are usually the primitives of the

model in the sense that their evolution is (at least partially) exogenous to decision making by

economic agents, and they constitute so-called state variables.2 To value corporate securities, the

model typically requires the same inputs as a standard option pricing model, including a law for

the time variation of state variables that accounts for the distribution of future cash flows. (In the

Black-Scholes model, it is a geometric Brownian motion process for the stock price.)

It is useful to relate this set-up to its counterpart in a typical static model. In the latter,

the distribution of a benchmark one-period return on an investment project is typically given

exogenously; as a natural “state-of-the-world” variable, it thus serves a purpose similar to a law

for the time variation of an exogenous state variable in a dynamic framework. Most of the inputs

are also not unique to a dynamic problem; for example, there is a well-known relation between an

essentially one-period binomial option pricing model and the continuous-time Black-Scholes model,

with a tight link between the parameters that appear in these setups.

The inherently dynamic nature of the state variable process in the Black-Scholes-Merton frame-

work (in the original setup, stock price) is an obvious reason for including “dynamic” in the name

of these models. More substantial grounds for such a name include the flexibility of this framework,
2
Note that state variables do not need to be firm-specific. For example, stochastic interest rates are not.

4
which enables us to study the interrelation of agents’ decisions at di↵erent points in time, as will

become clearer in our subsequent discussion.

A second critical element of these models is the objective function of a single decision maker

(e.g., shareholders, managers, central planners) or many decision makers (e.g., principals and agents,

oligopolists, price-takers in a competitive industry). This feature represents a crucial departure from

derivatives models, though not necessarily from asset pricing models at large. An objective function

summarizes the desires of the economic agent(s) in question and can be expressed succinctly by

combining certain contingent claims. For example, shareholders want to maximize the value of

their equity, whereas a benevolent central planner has the value-weighted utility of all the agents in

mind. Because the model is dynamic, the objective function can change subtly over time to reflect

the changing economic environment (a feature commonly omitted in one-period corporate finance

models but discussed at length in literature on contracting). For example, these models can exploit

di↵erences between ex ante and ex post objectives of decision makers and naturally give rise to

realistic conflicts of interest. They also allow the decision maker’s identity to change over time, as

a function of the state variables’ evolution. For example, equityholders decide on the original firm’s

financial structure, but debtholders can decide on the course of actions in the event of a default.

A third component is the specification of the set of instruments, known as controls, that are

available for the decision maker to maximize the objective function, as well as the set of constraints

imposed on these controls. For example, equityholders maximize equity value by choosing the firm’s

current and future investment and financial policies, subject to a limited liability condition. The

precise set of control(s) depends on the nature of a particular application. In one of the models

we consider in detail later, equityholders choose the amount of debt first, and the timing of default

later.

Finally, the model is solved by finding the optimal set of (generally, time-varying) controls at the

decision maker’s disposal to maximize her objective function. Most technical complications arise

at this stage, because, in most realistic specifications, it is notoriously difficult to find a closed-

form (and sometimes even approximate) solution. Furthermore, the trade-o↵ between complexity

5
and realism in the model and the feasibility of technical methods grows increasingly challenging.

Fortunately though, a dramatic increase in our abilities to solve such models has emerged from

the development of (now well-understood) techniques, which can deal with various technical issues,

in fields such as stochastic dynamic programming. For example, the smooth-pasting and super-

contact conditions (see, among others, Dumas 1991; Dixit and Pindyck 1994; Stokey 2009) have

dramatically facilitated the applicability of such models. In particular, these techniques allow

researchers to characterize the properties of an optimal solution, even though an explicit solution

may not be available.

As should be clear by now, this approach to modeling is very similar in its fundamental qualities

to traditional corporate finance models that specify objective functions, choice variables, and actions

of economic agents in a similar fashion. They also often ask a very similar set of questions. In fact,

any traditional corporate finance model can be reformulated using the language and apparatus of

a contingent claims model. However, whether a resulting model is easy to solve is another matter.

A key di↵erence from the paradigm of traditional corporate finance models, whether they are

static or dynamic in nature, should also be clear at this point. In traditional models, the values

underlying an objective function are separated from any asset pricing implications so that they

are alienated from the prices established in markets. This separation has significant consequences,

because even if a traditional model can produce a directional prediction of the agent’s response

to exogenous changes in primitive parameters, not many people would seriously attempt to claim

that a traditional model can make e↵ective quantitative predictions. By virtue of their tight link to

generic option pricing models, whose primary goal is to price securities, dynamic contingent claims

models can generate internally consistent testable quantitative predictions about the behavior of

the model’s output variables.

Whether quantitative predictions are of interest depends, of course, on the research objective.

However, the dynamic approach can also deliver qualitative results and predictions that are eco-

nomically quite di↵erent and sometimes completely absent from the traditional paradigm. In e↵ect,

the dynamic paradigm broadens our toolkit and enables us to ask new questions, as well as identify

6
and model new economic forces. In what follows, we illustrate both the quantitative and qualitative

sides of dynamic contingent claims models using concrete applications.

2.2 Real Options and Dynamic Investment

One of the first applications of option pricing models to economic decisions gave rise to a new

research area known as real options. An important component of the dynamic contingent claims

paradigm, the real options framework originally referred to activities in the “real” brick-and-mortar

or non-financial world, such as corporate investment policies (hence the name “real”). Research by

McDonald and Siegel (1985, 1986) and Brennan and Schwartz (1985) pioneered the field. As an

example of a real option model, as well as our first contingent claims setup, we consider a modified

version of the investment problem studied by McDonald and Siegel (1985, 1986). As we proceed,

we also informally introduce necessary technical machinery.

2.2.1 Real Option Model

The underlying state variable is the value of an investment project that could be undertaken by a

firm at a certain fixed cost. If undertaken, this investment decision is irreversible (i.e., the fixed

cost is non-recoverable). McDonald and Siegel assume that the state variable follows the same

dynamic process as the underlying state variable in the Black-Scholes model (i.e., the stock price

on which the option is written).

In a traditional corporate finance paradigm (which remains central in any MBA-level finance

textbook, e.g. see Berk and DeMarzo (2011)), an investment decision depends on the net present

value (NPV) of the project. McDonald and Siegel introduce a dynamic element in this decision-

making process. Instead of a take-it-or-leave-it investment decision that characterizes the NPV

decision process, the firm has the flexibility to wait before investing. The (real) option of waiting

to invest is valuable, because the future will bring partial resolution of uncertainty about the value

of the project. Crucially, the firm has control over the decision of when to invest; in other words, the

firm decides the optimal time of investment. The value of the option stems from the firm’s ability

to mitigate the consequences of unfortunate scenarios in which bad news occurs shortly after the

7
firm undertakes what previously appeared to be a positive NPV project. The firm faces a trade-o↵

between delaying investment to minimize the impact of bad news and missing out on the present

value of project’s cash flows in the meantime. Intuitively, as the project’s cash flow level grows, the

option to wait declines in value, and it becomes more attractive to invest today. The firm invests

when the cash flow level reaches a certain upper threshold.

To develop this model formally, we assume that a decision maker (e.g., firm’s manager) contin-

uously observes a potential project’s cash flow X that follows, as in Black and Scholes (1973), a

geometric Brownian motion process:

dXt = µXt dt + Xt dWtQ , (2.1)

where µ and are the constant instantaneous growth rate and volatility of cash flow, respectively,

and dWtQ is a standard Brownian motion process under the risk-neutral measure.3 The cost of

investment is fixed at a constant I. After the investment is made, the firm perpetually generates

cash flow X, whereas prior to the investment no cash flow is generated. The manager’s control

variable is the decision about when to invest in the project. Thus framed, the problem refers to the

choice of the time of investment. However, X is the only state variable, so the manager e↵ectively

chooses the cash flow level at which it is optimal to invest. We denote this level XI .

Then the value of the firm can be written as:

✓ ◆
XI
V (X0 ) = A(X0 , XI ) I , (2.2)
r µ

where A(X0 , XI ) is the date-0 value of a contingent claims security that pays $1 when the cash

flow level X reaches the level XI for the first time from below, and 0 otherwise. Security A is an

example of an Arrow-Debreu-like claim, which is fundamental in any contingent claims model. If

and when the threshold level of cash flow is reached for the first time, the investment takes place,
3
Mapping between physical and risk-neutral measures is economically non-trivial in the case of real assets. In
general, decision makers demand a risk premium to the extent that the project’s risk is systematic, which in turn
exerts an impact on the level of the growth rate µ. All the models we develop, though, can be extended to hold for
a generic risk premium. Alternatively, we might assume that the decision maker and the market are risk-neutral or
that a project’s risk is uncorrelated with systematic risk.

8
and the firm incurs a cost of I. In return, the firm perpetually receives the project’s cash flow,
XI
which equals r µ at that date, where r is the constant risk-free rate.

To relate the decision-making process of the manager to a traditional setup, we denote the NPV
⇣ ⌘
of the project as V N P V (XI ) = rXIµ I . We can thus rewrite the value in Equation (2.2) as:

V (X0 ) = A(X0 , XI )V N P V (XI ). (2.3)

This formulation reveals that the di↵erence created by flexibility is captured entirely by the contin-

gent claim A, so its properties are of great interest to us. Because X follows a geometric Brownian

motion, there is a simple closed-form expression for A(X0 , XI ):

( ⇣ ⌘⇠2
X0
, X 0 < XI ;
A(X0 , XI ) = XI (2.4)
1, X0 XI ,

where ⇠2 is a function of the constant primitive parameters µ, , and r. It is the positive root of

the fundamental quadratic equation and is given by:

0 s✓ 1
2 2
◆2
1 @ 2 rA .
⇠2 = 2
µ µ +2 (2.5)
2 2

2.2.2 Optimal Investment Policy

If X0 XI , the investment occurs instantaneously. In this case, we revert to a traditional NPV

analysis. If X0 < XI , the first-order condition leads to4

⇠2
XI = I(r µ). (2.6)
⇠2 1

Consider the more interesting case of X0 < XI . The traditional NPV rule prescribes an in-
XI
vestment if r µ > I, which implies that the investment threshold is XIN P V = I(r µ). It is easy

to show that ⇠2 > 1, and thus, XI > XIN P V . This fundamental result underscores the dramatic
4
It is straightforward to check that the second-order condition is satisfied as well.

9
di↵erence between essentially static and fully dynamic decision-making modes. By allowing firms

to wait, it dramatically extends the investment opportunity set and modifies the net present value

of investment. For example, empirical observations indicate that firms often find it in their interest

to wait, even though the NPV of the project is positive (Summers 1987). This simple model can

easily explain, at least qualitatively, the otherwise puzzling observation that firms seemingly use

surprisingly high hurdle rates in their investment decisions.

The intuition behind this main result can be understood by analyzing the nature of the option

that the firm possesses. When the future is uncertain, the option to wait is valuable, because over

time firms learn new information and avoid investing when cash flow levels decline. By proceeding

with its investment, the firm gives up this option to wait. Thus, the profit from the investment

must be sufficiently attractive to trade o↵ this option value. The optimal threshold XI is such that

the firm is indi↵erent between holding onto an option by waiting a bit longer and exercising its

option now.

The comparative statics of the investment threshold, XI , with respect to all the primitive

parameters of the model, as we show in Table 2, are intuitive. The threshold increases with the

investment cost I, inasmuch as a larger investment expense requires firms to become more certain

that the state of the world is good. It also increases with cash flow volatility , because greater

volatility increases the marginal value of waiting. That is, the distribution of future cash flows

has fatter tails, and a delay can bring valuable information about changes in the probability of

realizations of increasingly bad states of nature. The threshold also increases with respect to the

discount rate r, because a higher discount rate lowers the present value of future cash flows relative

to the cost of investment and increases the value of the waiting option. Finally, the threshold

decreases with the growth rate µ, because a higher growth rate increases the value of potentially

lost cash flows and thus makes waiting more costly.

A comparison of the optimal decisions in the dynamic and static (NPV) investment models,

which we also include in Table 2, shows that even though both thresholds decrease with the growth

rate, their di↵erence increases: for higher growth opportunities firms wait longer compared to the

10
standard NPV criterion. A comparison of values shows that at higher levels of the asset volatil-

ity, investment cost, and risk-free rate, the flexibility contribution becomes increasingly relatively

valuable. In all these cases, the firm has more to lose by exercising its investment decision earlier.

The quantitative implications of this model are economically significant. For example, consider

a project with the following reasonable characteristics: X0 = 1, I = 25, r = 0.05, µ = 0.02, and

= 0.25. For these parameters, the take-it-or-leave-it NPV is positive. However, it is optimal

for the firm to wait until the cash flow level increases from X = 1 to approximately XI = 2.4 (a

substantial increase of 140%). In this case, the value increases by 84%, from about 8.3 to 15.4.

Another way to look at this example is to consider hurdle rates (or internal rates of return). In the

NPV analysis, the value of the hurdle rate, rH , that produces NPV equal to zero is 6%. If we take

the option into account, the hurdle rate increases to 11.6%.

Figure 1 further shows that the hurdle rate in the presence of flexibility increases at a higher

risk-free interest rate and asset volatility, and quantifies the impact of these two variables (using

the same benchmark parameter we detailed previously). An uncertain environment in particular

has a steep impact. For annual levels of volatility in the range of 0.40-0.60 (practically, reasonable

for many applications), the hurdle rate is between 15% and 25% and can easily fit the ranges that

have been reported empirically (e.g., Summers 1987). Thus, the model has a very good chance of

explaining the stylized fact that firms wait too long (i.e., that implied hurdle rates are too high

relative to the levels we would expect from applying the NPV analysis).

At this stage, it is useful to consider the specific economic ingredients that go into the model,

because they are ubiquitous in a dynamic contingent claims paradigm. First, flexibility is critical:

If the firm cannot wait (e.g., because competitive pressure makes the investment truly a take-it-

or-leave-it-o↵er), the option value is immaterial. Second, the investment cost is fixed. Of course,

what is important here is that the investment cost has a fixed component, as variable costs can be

added easily. The presence of the fixed cost gives rise to an inaction region, in which firms prefer

learning about the state of the world rather than making active investment decisions. Third, the

investment is at least partially irreversible. If the firm can easily retract and salvage the present

11
value of investment costs, then the real option is not valuable. In practice, all three components

are present to some degree, and the framework enables us to study the extent to which a change

in any of them changes the importance of the real option.

2.2.3 Discussion and Extensions

The simple models proposed by McDonald and Siegel (1986) and Brennan and Schwartz (1985)

ushered corporate finance research into a new environment and in turn have been extended in

various dimensions. Methodologically, this paradigm underlines the importance of the dynamic

nature of both the environment and the decisions. In addition, because it is built on a dynamic setup

and realistic option-pricing models, the framework provides the possibility of studying quantitative

implications, such as how long firms actually wait to invest or the implied range of hurdle rates.

All the other examples we consider in this section are based on the fundamental idea that

the option to wait before committing to a financial or investment policy is valuable. We turn

our attention to the realm of investment and financing decisions extensively studied in corporate

finance. Dixit and Pindyck (1994) consider various extensions of the basic real option model to other

investment decisions and environments. A growing literature stream within corporate finance has

been applying the real option framework to explore corporate investment decisions in depth. For

example, Childs, Ott, and Triantis (1998) compare sequential and parallel investments in mutually

exclusive projects. In line with the intuition we have developed, when projects’ values are highly

correlated, sequential investment is superior, because learning about the value of one project allows

one to infer the value and thus the soundness of investment of another project.

Other important extensions that have only recently attracted attention of researchers include

parameter uncertainty (Décamps, Mariotti, and Villeneuve 2005, 2009; Klein 2009), applications

to real estate and leasing (Grenadier 1995), strategic competition (Grenadier 2002), agency cost

and information asymmetry (Grenadier and Wang 2005), time-inconsistent preferences (Grenadier

and Wang 2007), sequential investments and technology adoption (Grenadier and Weiss 1997),

the dynamics of mergers and acquisitions in oligopolistic industries (Hackbarth and Miao 2012),

asymmetric taxation and capital gains (Morellec and Schürho↵ 2010), the impact of uncertainty

12
on the probability of investment (Sarkar 2000), and the role of systemic risk and cost uncertainty

(Sarkar 2003).

2.3 Optimal Static Capital Structure

The next application of the contingent claims paradigm we consider is the optimal financial structure

of the firm. We start by building a formal trade-o↵ capital structure model, then discuss the far-

reaching economic implications of this setup.

2.3.1 Unlevered firm value

As in the case of a real options investment model, we assume that X is cash flow, generated this

time from the assets in place owned by a firm, which also follows a geometric Brownian process:

dXt = µXt dt + Xt dWtQ . (2.7)

To concentrate on the firm’s capital structure policy, we decouple investment and financial

considerations by assuming that investment decisions are exogenous and investment and operating

costs do not a↵ect cash flows. The cash flow X in this setup is then equivalent to earnings before

interest and taxes (EBIT). Because all the firm’s future cash flows are generated by assets in place
Xt
in perpetuity, the total asset value can be written as r µ.

The government levies a corporate tax on income at the constant proportional rate ⌧ . If the

firm’s capital structure consists only of equity, then the value of equity at any time t, St , equals

the value of the unlevered firm:

(1 ⌧ )Xt
St = . (2.8)
r µ

There are many realistic features that can be incorporated in the model at the cost of simplicity.

For example, taxes are typically asymmetric in the real world, so that profits are taxed at a higher

rate than losses, either explicitly or indirectly through carry-back and carry-forward provisions of

the tax code. In addition, a personal tax is levied on distributions to claimholders. Although we

13
ignore these considerations in our exposition, it is important to stress that some of them may have

non-trivial quantitative implications.

Corporate debt provides an obvious way of lowering the tax bill, because interest payments

are universally considered expenses by tax authorities and excluded from income taxation.5 For

simplicity, we consider perpetual debt that promises a constant coupon flow c. Assuming that the

cash flow residual (i.e., free cash flow left after paying interest and taxes) is continuously paid out to

shareholders, the payout flows to equityholders and debtholders are, correspondingly, (1 ⌧ )(Xt c)

and c.6 Thus, the total gain from tax benefits to debt is ⌧ c per unit of time.

2.3.2 Default

The cost of debt is that the firm may default on its debt payments later. The exact specification of

a default model depends on institutional and legal framework. We consider two realistic scenarios

for conditions in which default occurs and two scenarios describing potential firm outcomes upon

default.

Optimal vs. Liquidity Default. Similar to the real options investment model, X is the only

state variable, so default occurs only when X becomes sufficiently low and hits the default threshold,

denoted XD , for the first time from above. The determination of this threshold depends on the

economic mechanism of default. Two contrasting scenarios are frequently considered in theoretical

work. In the optimal or “endogenous” default case, equityholders (or managers representing them)

choose a default threshold to maximize equity value. In other words, default is a real option that

equityholders possess and by defaulting they exercise this option. As will soon become clear, in

this case the e↵ective assumption is that equityholders have “deep pockets”, i.e. they have access
5
The question of why coupon payments to debtholders are considered expenses, and thus receive a favorable tax
treatment, whereas dividends to stockholders do not, is interesting in itself. This distinction seems likely to be the
result of historical forces at the time the tax rules were being developed, rather than any deep economic reasoning
pertaining to contemporary economic or business circumstances. The further discussion of this important issue is
beyond the scope of this review.
6
We thus assume away a proactive cash policy, whereby firms can retain earnings in anticipation of future needs.
Incorporating a cash policy in this setup reduces tractability, though studies of cash policy in the presence of debt in
di↵erent setups suggest that it is an important issue (see e.g., Acharya, Davydenko, and Strebulaev 2012; Anderson
and Carverhill 2012).

14
to external funds to cover coupon payments if needed. The solution method to the problem of

finding the optimal default is exactly the same as for the investment problem of Section 2.2. In

the liquidity or “exogenous” default case, the firm defaults either because it violates net worth

covenants or because the firm and equityholders have no spare cash to cover their current interest

payments.

Both scenarios and their various amalgamations are realistic. For example, even though firms

can raise additional equity funding to pay debtholders, it may be costly or difficult due to timing

constraints or covenants in the debt contract. With some foresight, these elements can be modeled

parsimoniously within one model as XD = c. In the benchmark liquidity case, is equal to 1,

implying that the firm cannot raise any new external funds, and instead defaults the moment its

cash flow is lower than its promised coupon. In the optimal default case, is a control variable

determined by maximizing shareholders. If the optimal is lower than 1, the “deep-pockets”

assumption becomes clear. Shareholders have the means to pay debtholders, as long as they want

to keep their option alive. Alternatively, this assumption can be replaced by the realistic feature

that it is finitely costly to raise additional funds in distress.

Liquidation vs. Reorganization in Default. The economic cost of debt is modeled as the

partial loss of future cash flows in default. We assume that this cost a↵ects all future cash flows

proportionally at the rate ↵.

In default, the firm is liquidated or reorganized, and the proceeds are distributed to the claimants

according to the absolute priority rule (APR). In the simple capital structure case with only two

claimants, debt and equity, debt has seniority over equity, meaning that until it is paid in full,

equityholders are not compensated. For the endogenous default scenarios we consider, the APR in

fact implies that equityholders recover no value in default. Empirically, deviations from the APR

are frequently observed in the U.S., such that junior claimants recover non-zero value in bankruptcy

before more senior claimants are satisfied in full (e.g., Franks and Torous 1989). Default outcomes

also depend in large part on the institutional and legal framework. Such deviations can be easily

15
accommodated as an exogenous model feature (assuming a specific liquidation split) or as an

institutionalized bargaining game in default.

Two alternative scenarios can occur in default. In the case of liquidation, assets are redeployed

elsewhere, and debtholders receive the present value of the after-tax cash flow stream. Our treat-

ment of liquidation implies that debtholders partially recover the value of assets e↵ectively as new

equityholders, but they do not lever up the assets again by issuing new debt. This case closely

corresponds to Chapter 7 of the U.S. Bankruptcy Code or the administrative receivership procedure

in the U.K. Therefore, the value of the firm at default, VD , can be written as:

XD
VD (XD ) = (1 ↵)(1 ⌧) . (2.9)
r µ

Alternatively, default can lead to reorganization rather than liquidation, so that new equity-

holders establish a new firm with the remaining assets in place, and issue debt to maximize its

value. The value in default then depends on the resulting value of the new firm. Such a reorga-

nization is similar to bankruptcy outcomes under the Chapter 11 of the U.S. Bankruptcy Code,

and is considered in Section 2.3.6. For now, we assume that the firm is liquidated. The critical

di↵erence between these two cases is whether the tax benefits of debt get lost in default. If the

firm is liquidated, but the assets are sold to another party that intends to follow optimal capital

structure and pays fair value for it, the resulting value in default should correspond to the value

in the case of reorganization. Practically, the tax shield gets lost only for the duration of the

Chapter 11 bankruptcy process. Thus, liquidation and reorganization can be viewed as generating

low and upper boundaries for the firm value in default, respectively. Broadie, Chernov, and Sun-

daresan (2007) consider an interesting extension that distinguishes between liquidation (Chapter

7) and reorganization (Chapter 11). In their model, the firm that files for bankruptcy is granted

an automatic stay (no debt payments are disbursed, although they accumulate). If the company’s

condition improves, the firm reorganizes and partially pays all the accumulated debt obligations. If

the condition deteriorates, the firm is transferred into liquidation. One realistic implication of their

framework is that although the probability of default and bankruptcy is higher, the probability of

16
eventual liquidation is lower.

2.3.3 Equity and Debt Valuation

As in the real options investment model, it is helpful to consider an Arrow-Debreu contingent claim

that pays $1 only when the level of cash flow reaches the default threshold for the first time from

above. The value of the Arrow-Debreu claim at any time prior to the time of default can be shown

to be:7

✓ ◆⇠1
Xt
A(Xt , XD ) = , (2.10)
XD

where ⇠1 is the negative root of the fundamental quadratic equation:

0 s✓ 1
2 2
◆2
1 @ 2 rA .
⇠1 = 2
µ + µ +2 (2.11)
2 2

At any date prior to default, the value of levered equity can be written as:

✓ ◆ ✓ ◆
Xt c XD c
S(Xt ) = (1 ⌧ ) + A(Xt , XD ) + (1 ⌧ ). (2.12)
r µ r r µ r

The first term in Equation (2.12) is the after-tax levered equity value in perpetuity, assuming

no default. The second term takes default into account. By defaulting, shareholders forgo future

cash flows in exchange for discontinuing interest payments.8 To find the optimal level of XD , we

apply the smooth-pasting condition:9

@S(Xt )
= 0. (2.13)
@Xt
Xt =XD

7
From now on, we assume that t is a time point prior to default: t < min{s : Xs  XD }. Note that the values of
basic contingent claims in this and the real options models are di↵erent, because in the investment model the payo↵
occurs only when the upper trigger is reached, and here it takes place only when the lower trigger is reached.
8
In the exogenous default case, equityholders are also entitled to the residual asset value after debtholders have
been satisfied in full. The residual is not zero for very small values of default costs. Although this slightly changes
the formulation of the problem, we ignore this scenario here for simplicity.
9
It is easy to confirm that, in this simple problem, the smooth-pasting condition is equivalent to the maximization
of shareholder value with respect to XD . Stokey (2009) provides a formal treatment of smooth-pasting in a more
general setup.

17
 The optimal default boundary is then:

⇠1 r µ
XD = c. (2.14)
⇠1 1 r

Thus, the default boundary is proportional to the coupon, so we can write XD = D c, where
⇠1 r µ
D = ⇠1 1 r . Because ⇠1 < 0, it is easy to show that D < 1, and therefore in the endogenous

default case shareholders find it optimal to keep their default option alive longer than they would

in the benchmark liquidity default case, where = 1. For a value of below one, keeping the

default option alive requires external funds to pay the coupon in distress (defined as X < c), which

provides the rationale for the rights issues or, more generally, equityholders’ “deep pockets.” In

practice, it is likely that is the weighted average of the values in these two opposite cases. A

reasonable scenario is a partial liquidity case, where equityholders incur issuance costs to access

external funds in distress.

An important observation is that the optimal default boundary XD is independent of the default

cost ↵. In making a default decision, equityholders do not internalize debtholders’ value, and only

debtholders, as residual equity claimants, bear any default costs. As we emphasize soon hereafter,

this ex ante vs. ex post conflict of interest plays an important role in initial capital structure

decisions.

The value of perpetual debt promising coupon c, D(Xt ), is (recall that we assume the case of

liquidation in the case of default):

c c XD
D(Xt ) = A(Xt , XD ) + A(Xt , XD )(1 ↵)(1 ⌧) . (2.15)
r r r µ

The first term is the perpetuity value of risk-free debt. The second term is the present value of

interest payments that debtholders lose in the case of default. The final term is the present value

of assets that debtholders recover in liquidation.

18
2.3.4 Optimal Capital Structure

The shareholders of an all-equity firm maximize their equity value—that is, in this case, firm

value—at the initial date 0 by choosing an optimal capital structure. We assume in this static

capital structure environment that the decision to issue debt is made only once. At any subsequent

date, equityholders may default but are not allowed to refinance. Section 2.4 considers the first-

order extension that allows for refinancing.

Note that at date 0, by maximizing the total value of future equity and debt, shareholders

internalize the value of future debtholders’ claims. This finding is not inconsistent with the result

on shareholders’ default decision. The crucial distinction between debt at date 0 and debt after it

has been issued centers on future versus outstanding claims. In the latter case, debt issuance is a

sunk decision (and the amount contributed by debtholders is a sunk benefit) that is not taken into

account. In the former case, debt is still to be issued and priced by investors, so equityholders take

the debt claim into account by maximizing total firm value. Rational debt markets reflect the future

conflict of interest in pricing the firm’s claims. Intuitively, it would be better for equityholders

to commit ex ante to maximizing firm value rather than equity value at any time, not only at

issuance. We realistically assume that such a commitment is not contractible, though in practice

various covenants in debt indentures attempt to substitute for such a commitment.

Shareholders maximize total firm value, V (X0 ), subject to the constraint XD = c. There are

two approaches to writing firm value, V , after debt has been issued. First, it can be written as the

sum of debt and equity, V = D + S, using Equations (2.12) and (2.15). Second, it can be written

as the sum of all the components that make firm value deviate from the after-tax unlevered asset

value: V = F + T B DC, where F is the unlevered firm value (equivalent to equity value when

c = 0), T B is the present value of the tax benefits of debt, and DC is the present value of default

costs. It is instructive to write the firm value in both ways. In the second approach, the component

values are:

19
 Xt
F (Xt ) = (1 ⌧) ; (2.16)
r µ
XD
DC(Xt ) = A(Xt , XD )↵(1 ⌧) ; (2.17)
r µ
c
T B(Xt ) = ⌧ (1 A(Xt , XD )). (2.18)
r

To find the optimal coupon level, shareholders maximize the date-0 firm value, V (X0 ), with

respect to the coupon c, subject to the default condition XD = c. From the first-order condition,

the solution to this problem is:

" # 1
⇠1
1 1 ⌧
c⇤ = r X0 . (2.19)
1 ⇠1 (1 ⌧ )↵ r µ +⌧

In the endogenous default boundary case, we can find the optimal coupon by substituting for

the value of from Equation (2.14):

" # 1
⇠1
⇤ r ⇠1 1 1 ⌧
c ( D) = ⇠1
X0 . (2.20)
r µ ⇠1 1 ⇠1
⇠1 1 (1 ⌧ )↵ + ⌧

Thus, the optimal coupon is proportional to the cash flow level, implying that size of the firm is

irrelevant in this specification.

2.3.5 Comparative statics

In Table 3 and Figures 2–6, we present comparative statics with respect to the primitive parameters

of the model. To understand the economics that underlie these comparative statics, we need

to emphasize the main economic mechanism at play: Changes in the primitive parameters shift

the term structure of the expected cash flows stemming from the tax benefit and default cost,

because these cash flow streams occur at di↵erent times. The table and figures also allow us

to compare endogenous and exogenous default cases. To understand the intuition behind the

di↵erences between these two cases, note that while in the endogenous case the equityholders exert

control by choosing two parameters (the coupon level at the outset and the default boundary in the

future), in the exogenous case the choice of coupon dictates the choice of the default trigger. For

20
each coupon level, the distance to default is smaller and thus the present value of the default costs

is larger in the exogenous case. This distinction accounts for the di↵erences in their comparative

statics, as well as for the consistently lower values of optimal coupon and default boundary in the

exogenous default case.

A challenging empirical problem for interpreting the coupon level as a main proxy for financial

policy is that firms of di↵erent size are not directly comparable. Instead, both empirically and

theoretically, it is more common to use financial leverage. At any point in time, we can define the

market leverage ratio as:

D(Xt )
L(Xt ) = . (2.21)
V (Xt )
In practice, the market values of debt often are unavailable, and the quasi-market leverage ratio

is used instead:

D(X0 )
QM L(Xt ) = , (2.22)
D(X0 ) + S(Xt )
where D(X0 ) indicates the original (book) value of debt. When the firm is not in distress, the

two ratios are typically very close in value. If debt is issued at par, then the date-0 leverage and

quasi-market leverage ratios are equal. For the purpose of this section, we consider the date-0

leverage only.

We are also interested in measuring the e↵ect of the decision to lever up on the value of the

firm. To do so, we consider the relative value di↵erential between levered and unlevered firms at
V (X0 ) F (X0 )
date 0, F (X0 ) .

Consider a variation in the growth rate parameter. A higher growth rate increases expected

future cash flows and intuitively increases the firm’s desire to take on more debt to increase its

tax benefits. In the endogenous case, the firm responds by implementing exactly this policy, while

simultaneously lowering the default threshold per unit of coupon (as Table 3 shows, D is decreasing

in µ). In the exogenous case, however, the firm does not explicitly control the default trigger, so

a higher coupon implies a greater probability of default early on, which may lead the firm to lose

21
out on the bright (but more distant) future. The firms makes an optimal decision by balancing

this trade-o↵. The e↵ect of a higher growth rate may thus actually lower the optimal coupon. For

the same reason, leverage and the value di↵erential increase (decrease) with the growth rate in the

endogenous (exogenous) case.

In the endogenous default case, the leverage ratio increases with the interest rate, even though

the optimal coupon is lower. To understand the di↵erence, note that the behavior of the leverage

ratio depends on the relative sensitivities of equity and debt values to the variation in the underlying

parameters. As the interest rate increases, all future cash flows are discounted at a higher rate,

but equity is a↵ected more, because most of its value comes from future cash flows. In other

words, it can be said that equity has a greater duration than debt and thus it is more sensitive to

the interest rate variation. Even though the coupon decreases (because tax benefits, expected in

the more distant future, are more a↵ected), the impact of the lower debt value is negated by the

even lower value of equity. The e↵ect is obviously more pronounced in the exogenous default case,

because both forces act in the same direction.

Asset volatility exerts two e↵ects on optimal financial policy. On the one hand, higher volatility

increases the risk of default, which implies a lower optimal coupon. On the other hand, higher

volatility increases the likelihood of the firm finding itself in very good states of the world, where

potential tax benefits are high. As Figure 4 and Table 3 show, in the endogenous case the first

e↵ect dominates for low values of asset volatility and leads to a smaller coupon, while the second

e↵ect dominates for larger values. In the exogenous case, the risk of default naturally plays a more

prominent role. The leverage ratio is a decreasing function of asset volatility, because higher asset

volatility makes equity, as a call option, more valuable, but debt, as a short position in a put option,

less valuable. Finally, as expected, the variables of interest decline with higher default costs and

increase with a higher tax rate.

2.3.6 Optimal Capital Structure with Reorganization

We can derive another empirically relevant case when debtholders reorganize a defaulted firm and

lever it up again. The valuation of equity is unchanged and given by Equation (2.12), because

22
equityholders do not internalize the actions of debtholders upon default after a debt issuance.

Therefore, the default boundary is also the same, XD = D c, and takes the same values we

previously established for the endogenous and exogenous cases.

The value of perpetual debt, D(Xt ), is thus

c c
D(Xt ) = A(Xt , XD ) + A(Xt , XD )(1 ↵) (S(XD ) + D(XD )) , (2.23)
r r

where VD (XD ) = S(XD ) + D(XD ) is the value of the new firm. The first-order homogeneity of the

claims means that we can write debt value as:

c c XD
D(Xt ) = A(Xt , XD ) + A(Xt , XD )(1 ↵) (S(X0 ) + D(X0 )) , (2.24)
r r X0

and therefore, the debt value at date 0 is:

c
r A(Xt , XD ) rc + A(Xt , XD )(1 ↵) X
X0 S(X0 )
D

D(X0 ) = . (2.25)
1 A(Xt , XD )(1 ↵) X
X0
D

Equityholders still maximize the total value of the firm at date 0 by choosing the coupon level,

subject to the appropriate default boundary condition. Compared with the liquidation case, the

comparative statics are qualitatively similar, but a reorganization option e↵ectively reduces the

cost of default and leads to a higher reliance on debt. For the same reason, the di↵erence between

endogenous and exogenous default cases is smaller. We compare these two cases quantitatively in

Section 2.3.8.

2.3.7 Discussion

The trade-o↵ capital structure model is helpful for illustrating the comparison between traditional

static and contingent claims paradigms. For clarity, it is worthwhile to briefly review the historical

background of capital structure studies. Following the landmark indeterminacy result of Modigliani

and Miller (1958) and its extension to cover tax benefits in Modigliani and Miller (1963), the debate

on optimal corporate capital structure centered on what has since become known as static trade-o↵

theory. The general view in the 1960s and the 1970s was that static trade-o↵ theory o↵ered a

23
first-order explanation of the financial structure of firms. Miller (1977) criticized this view, arguing

that bankruptcy costs, an obvious culprit and counterweight to o↵set the tax benefits of debt, are

likely too small to explain why firms appear insufficiently levered. In Miller’s words, tax benefits

versus bankruptcy costs are like “a horse and rabbit stew.” Miller (1977) suggests personal taxes

as one resolution. DeAngelo and Masulis (1980) provide another trade-o↵-based explanation that

relies on alternative tax shields, such as leasing and depreciation. The modified consensus that

emerged as a result of these e↵orts came to the conclusion that optimal capital structure is the

result of the trade-o↵ between the tax advantages of debt and various leverage-related costs.10

No formal model is required to understand the economic forces in this trade-o↵ model. A higher

corporate tax rate should lead to higher optimal leverage, whereas a higher personal income tax

should work in the opposite direction. Higher profitability increases taxable cash flow and thus

increases corporate demand for debt, whereas higher bankruptcy costs reduce the debt advantage.

Even though such a trade-o↵ naturally yields an interior optimal capital structure, a static model

has difficulty quantifying relevant economic forces. At the least, the timing of cash flows stemming

from the tax benefits of debt and the timing of cash flows lost due to bankruptcy costs do not

coincide. A static model is therefore likely doomed in its attempt to quantify this cost-benefit trade-

o↵, because the present values of these asynchronous costs and benefits are difficult to compare.

Therefore, the “horse and rabbit stew” conundrum and subsequent ad hoc attempts to address

it have been based largely on intuitively reasonable, qualitative observations rather than on careful

quantitative analysis. Such qualitative arguments cannot gauge the size of the gap between theory

and reality, nor can they help researchers understand the quantitative e↵ects of relevant forces,

such as asset market volatility or interest rate changes, even though these issues have been widely

recognized as important.

In light of these developments, the contribution of the contingent claims model becomes clear.

At face value, it does not propose new economic mechanisms, because the basic economic forces are

identical to those in the traditional trade-o↵ model of tax benefits and bankruptcy costs. Although
10
It is important to note that Miller (1977) and DeAngelo and Masulis (1980) study adjustment of capital market
prices as a result of the trade-o↵ between net tax benefits of debt and debt-related costs.

24
Bradley, Jarrell, and Kim (1984) discuss dynamic intuition in a static model, the honor of being the

first in this area likely belongs to Kane, Marcus, and McDonald (1984), who apply option pricing

methods to study the extent of the tax advantage of debt. Fischer, Heinkel, and Zechner (1989)

then built a fully-fledged dynamic model that allows for dynamic capital structure. Yet it was

a contribution by Leland (1994) that really started the contingent claims revolution in corporate

finance. The model that Leland (1994) developed is an amalgam of all the preceding contributions.

Its greater importance, in terms of the future development of the field, was in its simple and yet

clear formal treatment of trade-o↵ theory. It developed a relatively robust contingent claims model

of capital structure, based on the intuition of the static model, and then quantified optimal leverage

ratios that firms should choose if they live in the dynamic world described by the trade-o↵ model.

Leland’s model further exploited option pricing models to study corporate credit risk and thus was

fundamental for the subsequent development of contingent claims credit risk models as well.

In contrast to prior models, Leland’s is simple enough to be easily comparable to static trade-o↵

models. It is thus well-suited to illustrate the power of contingent claims analysis. This framework

confers the basic advantage that the Arrow-Debreu-type security A(X, XD ) can be specified using a

well-established asset pricing paradigm. The value of A depends not on ad hoc assumptions about

the costs and benefits of debt in various two-period static models,11 but rather on primitive variables

(e.g., asset volatility, riskless interest rates) that can be estimated or approximated reasonably well.

Consequently, costs, benefits, and primitives determine the value of tax benefits and bankruptcy

costs, the resulting value of the firm, and optimal financial decisions. Contingent claims analysis

enables us to investigate the comparative statics, both qualitatively and quantitatively, that relate

financial structure control variables to the primitive economic parameters. In turn, we can quantify

their e↵ect on capital structure and judge whether reasonable capital structures can be obtained for

any realistic range of parameters. Using this formulation, it is also easy to understand what other

factors might be missing and how their e↵ects can be incorporated. If the calibration shows that

the model performs poorly, not only can the model be rejected, but it is also possible to explore
11
Examples would include quadratic costs and linear benefits, which are used entirely for tractability and to obtain
an interior solution to the optimal capital structure problem, rather than for any reasons pertaining to realism.

25
whether the modeling of the security A needs to be modified and additional economic factors should

be considered.

What is important is that, by modeling the security A in this fashion, we can facilitate modeling

of various costs and benefits time-consistent. Cash flows arising from tax benefits and bankruptcy

costs, as well as other leverage-related costs and benefits, do not occur at the same time. Although

bankruptcy costs potentially are substantial in the event of an eventual default, the expected

present value of bankruptcy costs is much smaller, so they need to be adjusted for the time-

weighted probability of default. Thus, A parsimoniously reflects the term structure of the default

distribution over time, which is an important consideration that separates contingent claims models

from their simple static counterparts.

As we show in Section 2.3.8, for the endogenous default scenarios, and to a lesser extent for

the exogenous case, the model confirms the well-accepted “horse and rabbit stew” conundrum, in

which optimal leverage, as implied by reasonable parameters, is too high to explain actual firm

behavior in the cross-section. For example, in the endogenous default scenario, the model produces

a leverage ratio in excess of 70%, as opposed to the average leverage ratio of 25% that we observe

in reality.

These findings, first clearly demonstrated by Leland (1994), are as important to the develop-

ment of capital structure, and perhaps to the whole field of corporate finance, as the description

of the equity premium o↵ered by Mehra and Prescott (1985). Leland not only showed that the

dynamic version of the static-trade-o↵ model could not replicate observed leverage, but he laid

the groundwork for understanding the sources of this failure by constructing a simple model with

well-defined assumptions. Naturally, most subsequent research based on this and earlier contin-

gent claims models has explored the role of various assumptions. We highlight some of the more

important ones in subsequent sections.

26
2.3.8 Quantitative implications

In this section, we explore the quantitative implications of the capital structure we have developed

thus far. We do this in a couple of ways. First, we start by considering a particular set of

parameters, for which we choose reasonable values relying on existing calibration and empirical

studies. Second, we quantify the comparative statics of the leverage ratio with respect to the

variation in the empirically relevant parameters. We concentrate on the leverage ratio at date 0,

consistent with most of the extant theoretical literature.12

The benchmark set of parameters is as follows: interest rate, r, is 0.05, the growth rate, µ, is

0.02, asset volatility, , is 0.25, the corporate tax rate, ⌧ , is 0.2, and the default cost parameter,

↵, is 0.15. For many of these parameters, the direct empirical estimation is clearly difficult; in

addition, there is substantial cross-sectional heterogeneity. Some empirical evidence enables us to

anchor the benchmark values, though. For example, Schaefer and Strebulaev (2008) estimate the

average asset volatility of firms that issued bonds to be 0.23, and default costs are approximately

10–20% of the total firm value at default, according to Andrade and Kaplan (1998), or 15–30%,

according to Davydenko, Strebulaev, and Zhao (2012). Although the U.S. federal corporate tax

rate is 0.35, the e↵ective tax rate is likely smaller, reduced by, for example, personal tax rates

(Graham 2000).

As Table 4 shows, for this benchmark set of parameters, the leverage ratio in the endogenous

default case varies between 70% (liquidation) and 80% (reorganization) – far in excess of the

observed leverage ratios. For example, the equally-weighted quasi-market leverage ratio in the

COMPUSTAT universe consistently appears within the 20–25% range. The leverage ratios in the

exogenous default case are substantially smaller, at 25–28%. However, it is unlikely that the model

can explain the actual observed capital structure data, because in practice, firms have substantial

flexibility to raise additional funds (or use retained earnings) to pay down their interest. Firms do

not realistically default every time their cash flow drops below the promised interest payment.
12
Needless to say, in the model with the leverage decision only at date 0, the leverage ratio attenuates over time,
conditional on no default. In Section 2.5, we discuss the behavior of the leverage ratio over time in a dynamic model
of capital structure decisions.

27
 At the same time, this model can easily account for di↵erences between large and small firms.

Large firms are likely more flexible in their outside sources of financing, and temporary liquidity

problems should have less significance for them. For smaller firms, cash flows that are too low to

cover interest can be a much more troublesome sign. In turn, large firms likely have more debt

capacity.

Table 4 also shows that the value di↵erential between levered and unlevered firms can be sub-

stantial. In the endogenous case, the firm’s value increases by 11–13%, though the impact is much

more modest in the exogenous case.

The rest of the table illustrates how leverage changes when we vary the main mode parameters.

In the endogenous case, the leverage ratio is substantially in excess of empirically observed ratios,

even for extreme parameter values. For example, leverage is 48–50% when half of the firm is lost

in default. Note that leverage is still high, even if the tax rate is low: when the e↵ective tax rate

is just 10%, the leverage ratio is still 59–65%. These results contrast with the substantially lower

leverage ratios in the exogenous case. Empirically, if it is possible to identify exogenous variation in

the values of , the model provides clear testable predictions. Because the endogenous case is likely

to be more realistic for a large cross-section of firms, these results clearly suggest that the model

produces much higher leverage values than we observe. In the next several sections, we consider

several extensions of the basic model that lead to a considerably lower optimal leverage at date 0.

2.4 Optimal Dynamic Capital Structure

The assumption that firms can raise debt financing only once is unrealistic. In theory, a static debt

policy implies that as firms grow larger over time, their leverage attenuates, leading to an unrea-

sonably negative relationship between leverage and firm size. In practice, firms’ financial policies

are dynamic, and leverage adjusts in response to the changing environment. In this section, we

formalize this important idea by building a benchmark model of optimal dynamic capital structure.

28
2.4.1 Model Setup

The intuition for this model is based on two fundamental properties of the dynamic environment.

First, firms should take into account any expected changes to their future cash flows and opportunity

sets when forming their current financial policy. The opportunity to adjust their leverage at a future

date is likely to change firms’ financial decisions today. Second, firms face various adjustment costs

(e.g., the transaction costs of raising external financing), and even though these costs may be small

relative to the total amount raised, their e↵ect on optimal policies can be substantial. The impact

of adjustment costs on optimal decisions has been studied extensively in macroeconomics, such

as in the context of menu cost models, in which firms do not adjust prices continuously because

of the small transaction costs. The generic outcome of those models is that there is a region of

a state variable (e.g., demand for a firm’s product), called an inaction region, in which the firm

prefers optimally not to make active changes in its e↵ort to reduce the expenses associated with any

adjustment. Mankiw (1985) shows that even very small menu costs can have a profound impact on

aggregate macroeconomic variables. One of the implications of dynamic corporate finance is that

a similar intuition fully applies to corporate decisions and that the magnitude of firms’ response is

similarly large.

To introduce dynamic capital structure decisions to the model in Section 2.3, assume that in

addition to a default option that firms find valuable to exercise in bad states of the world, they also

possess a refinancing option that they may find valuable to exercise in good states.13 As the cash

flow level, X, grows, the present value of the default costs decreases (because firms are further away

from the default threshold). A higher cash flow level also implies a more punitive e↵ective tax rate

(because fewer profits are shielded from the corporate tax by a constant coupon). The refinancing

option is valuable, because it enables firms to restructure their capital structure at some upper

threshold by raising more debt and restoring the balance between the costs of distress and the tax

benefits of debt.
13
Refinancing in a bad state of the world might be considered as well but is less important (default, of course, is
an extreme version of refinancing). Realistically, firms in distress may need to find additional resources at an extra
cost; we do not implement this potentially fruitful extension here.

29
 Because raising new debt and refinancing existing debt are costly, the next logical step is

to introduce the cost of refinancing. In the absence of any refinancing costs, firms adjust their

outstanding amount of debt continuously in good states. We assume that this cost is proportional

to the total amount of debt raised at any refinancing date. This assumption allows us to model

the fixed cost of refinancing in a tractable manner because (as we show below) it preserves the

first-order homogeneity property of the problem, also known as scaling. Under this cost structure,

if the firm raised debt originally at X0 , it will optimally default when X reaches the low value of

XD , and it will refinance when X reaches the upper value of XR , where XD < X0 < XR .

Using the by now familiar strategy, we start by considering the appropriate Arrow-Debreu-type

(hereafter, AD) contingent claims. In the static investment problem of Section 2.2, the appropriate

AD contingent claim pays $1 when the cash flow level reaches the upper threshold; in the capital

structure problem of Section 2.3, the AD contingent claim is the mirror image, such that it pays $1

when the cash flow level reaches the lower threshold. In the dynamic problem with two thresholds,

we therefore introduce two AD contingent claims. The first, denoted A(X, XD |TXD < TXR ), pays

$1 when the cash flow level reaches the lower threshold XD for the first time from above, if and

only if this event happens before it reaches the upper threshold XR (TX indicates the first time it

reaches value X). Similarly, A(X, XR |TXR < TXD ) denotes the value of a contingent claim that

pays $1 when the cash flow level reaches the upper threshold XR for the first time from below, if

and only if this event happens before it reaches the lower threshold XD .

The market value of debt, raised at date 0, can be written at any date t (before either default

or refinancing occurs) as:

c ⇣ c ⌘
D(Xt ) = + A(Xt , XD |TXD < TXR ) + (1 ↵)VD (XD ) (2.26)
r r
+A(Xt , XR |TXR < TXD )R(XR ), TXt  min (TXD , TXR ) .

The second term is a familiar payo↵ to debtholders in the event of default, where VD depends on

the outcome of default: liquidation (Equation (2.9)) or reorganization (Equation (2.23)). The last

30
term is the debt payo↵ at refinancing, R(XR ). There are two alternate assumptions for modeling

R(XR ). In one case, existing debt is recalled (at par, or its original book value at date 0, or at a

premium) and new debt is issued in its place. Taken literally, the firm issues callable debt with

an indenture agreement that prevents it from issuing any further debt. In another case, the old

debtholders’ claim is diluted by the issuance of new debt with a pari passu provision, such that

all outstanding debt issues have equal seniority. Although the exact specifications strongly reflect

the desire for tractability, both of these assumptions are realistic; in addition, many variations can

be considered (likely without significantly a↵ecting any of the model’s economic results). Consider
c
the callability specification, such that R(XR ) = r + (1 + !)D(X0 ), where ! is a premium the

firm must pay for the right to recall its debt. Then, from Equation (2.26), and assuming the firm

is liquidated in default, we can rewrite the value of debt at date 0 as:

⇣ ⌘
c
r + A (X0 , XD |TXD < TXR ) c
r + (1 ↵)(1 ⌧ ) rXDµ + A (X0 , XR |TXR < TXD ) c
r
D(X0 ) = .
1 A (X0 , XR |TXR < TXD ) (1 + !)
(2.27)

The value of equity before either default or refinancing occurs is (not taking into account date-0

transactions):

✓ ✓ ◆◆
Xt c XD c
s(Xt ) = + A (Xt , XD |TXD < TXR ) + (1 ⌧) (2.28)
r µ r r µ r
✓ ◆
XR c
+A (Xt , XR |TXR < TXD ) + (1 ⌧ )
r µ r
+A (Xt , XR |TXR < TXD ) S(XR ), TXt  min (TXD , TXR ) .

The last two terms adjust for the refinancing point. Intuitively, from the equityholders’ point of

view, refinancing reflects two sequential transactions: The equityholders forgo future cash flows in

exchange for coupon payments (the second line), then receive the total residual value of equity (the

third line). Note the di↵erence between s(X) and S(X): whereas S(X) is the total value of equity,

s(X) takes into account the impact of transactions at the beginning of a refinancing cycle (date 0 in

this case). At all values of X (apart from X0 , XR , and other refinancing levels), s(X) = S(X). At

31
refinancing, equityholders pay o↵ existing debtholders and issue new debt, so that the total value

of equity is:

S(XR ) = (1 + !)D(X0 ) + (1 q)D(XR ) + s(XR ), (2.29)

where q is the proportional cost of debt issuance. The proportionality of the cost (as well as the

nature of the cash flow stochastic process) assures the first-order homogeneity property. In other

words, all the securities’ values at refinancing are proportionately larger than the corresponding

values at date 0. We thus can write the value of equity at TR as:

XR XR
S(XR ) = (1 + !)D(X0 ) + (1 q) D(X0 ) + s(X0 ). (2.30)
X0 X0

Equityholders maximize the total value of equity at date 0 (before debt is issued):

S(X0 ) = (1 q)D(X0 ) + s(X0 ). (2.31)

Their controls are c, XD , and XR . A subtle point about this optimization problem is that though

equityholders cannot credibly commit to default at a particular level at date 0 in the endogenous

default case (and thus XD is still found from the smooth-pasting condition), by issuing callable

debt they likely can commit e↵ectively to recall at a certain cash flow level.

Commitment is valuable, so XR can be obtained from the date-0 joint maximization of the total

equity value with respect to c and XR , rather than from the smooth-pasting condition. However,

in the absence of a credible commitment device, XR is obtained from the smooth-pasting condition

(or rather, XR and XD can be determined from the joint smooth-pasting conditions). Here, we

assume the former case, but the latter scenario leads to qualitatively similar results.

Another important point worth emphasizing is that unlike the problem of Section 2.3, default

costs here enter the default decision of equityholders: Although equityholders do not internalize

the value of outstanding debt, they internalize the value of all future debt issues, which of course

depend on default costs.

32
2.4.2 Comparative Statics

Most of the economic forces in the fully dynamic model have not changed, so the qualitative

predictions overall remain similar to those developed in Section 2.3.8. Figure 7 compares the

behavior of the endogenous and exogenous default cases as debt issuance costs vary (we use the

same benchmark set of parameters as in Section 2.3.8). For the endogenous case, the dependence of

the optimal coupon and leverage on issuance costs is non-monotonic. A simple explanation of this

seemingly counterintuitive behavior is that as adjustment costs increase, firms delay refinancing

(as also illustrated), but conditional on refinancing, they issue more debt. This channel works

only because the firm has the refinancing boundary as an additional control. As costs increase,

the overall attractiveness of debt attenuates. In general, an extra degree of freedom implies that

firms constantly face the trade-o↵ of either issuing more at each refinancing point, while delaying

refinancing, or else issuing less but refinancing more often. Interestingly, the refinancing boundary,

XR , is typically uniformly higher in the exogenous default case: Firms respond to a higher default

probability not only by choosing a lower coupon but also by delaying refinancing in good states of

the world.

Figure 8 shows the di↵erence between dynamic and static leverage models with endogenous

default for di↵erent levels of the debt issuance cost. To make these models comparable, we consider

a static leverage model that is slightly di↵erent from the one in Section 2.3: The debt issuance cost

is incurred at date 0, so the firm’s maximization problem is identical in both cases. The first-order

implication is that when firms acquire an additional degree of freedom, they lower their coupon

levels and exhibit lower leverage ratios compared with the static version. This behavior is entirely

intuitive, in that only reason firms pursue an aggressive capital structure policy at date 0 in the

static case is their inability to issue more debt when cash flow levels increase. As an opportunity

to recapitalize becomes available, they issue less debt initially to reduce the likelihood of default.

For the same reason, equityholders opt to default later in the dynamic case, because their upside

option is enhanced by the opportunity to refinance later.

Figure 8 also shows that adding associated with flexibility of adjusting leverage dynamically

33
increases the value of the levered firm. This result is very general. The di↵erence in values between

dynamic and static capital structure firms increases in the growth rate µ, asset volatility , and tax

rate ⌧ , but it decreases with the default cost parameter, ↵, and issuance cost parameter, q. These

relationships follow directly from the economic forces we have already discussed. For example, as

the tax rate increases, future tax benefits become relatively more important, and dynamic flexibility

adds more value.

2.4.3 Quantitative Implications

Table 5 reports the quantitative implications of the dynamic capital structure model for the two

default cases we have considered. We assume that the call premium, !, is zero. For comparison,

we also provide predictions pertaining to the static capital structure model. To make these models

comparable, we again assume that in the static model the firm pays the debt issuance cost at

date 0.14 The introduction of dynamic flexibility lowers the date-0 leverage ratio substantially. In

the benchmark endogenous default case, the di↵erence is 17%, though it ranges between 10% and

23% for the parameter constellations we consider. An important implication is that, although the

leverage ratio is substantially lower, it is still significantly higher than the empirically observed

ratios.

Table 5 also shows that the sensitivity of the leverage ratio to the primitive parameters varies

across the models. At least for the constellation of parameter values we consider, the dynamic

leverage ratio is more sensitive to asset volatility and less sensitive to changes in the tax rate and

default costs. In all cases, the flexibility to adjust in the future makes leverage more valuable. For

example, higher asset volatility introduces substantial kurtosis in the distribution of the future cash

flows and therefore makes dynamic adjustment more valuable. The comparison between dynamic

and static cases is most vivid in the comparison of their value di↵erentials (relative to the unlevered

firm value). In the benchmark endogenous case, adjusting leverage in the future increases the value

di↵erential by 80%, and the magnitude varies between an increase by 50% and a triple rise. For
14
This change explains some minor di↵erences between the otherwise similar results in Tables 4 and 5 for the static
model.

34
example, if the tax benefits are high at the current statutory corporate tax rate of 0.40 (vs. 0.2

assumed in the benchmark case), dynamic flexibility increases the firm value by half, whereas it

increases by one-quarter in the static capital structure case.

If the firm defaults the first time its earnings are less than or equal to coupon, the comparison

between the dynamic and static models remains similar, but the leverage ratios are substantially

smaller. In the benchmark case, the leverage ratio changes from 52% to 17% with a change in the

default policy, and the value di↵erential is substantially smaller.

2.4.4 Discussion

The basic and first-order quantitative implication of allowing firms to refinance in response to an

increase in future profits is a significant decrease in model-implied leverage ratios at the time of debt

issuance, bringing them close to the leverage ratios observed in reality. Note that the importance

of future profitability, an intuitive and economically plausible explanation, is difficult to study even

qualitatively (not to mention quantitatively) using conventional static methods. The extension of

the contingent claims paradigm to dynamic leverage policies allows us to model the impact of future

changes in leverage policy in response to profitability shocks on current financial decisions. It also

enables us to measure the extent to which future interactions explain empirically observed leverage

today. This common theme persists across all the applications of this paradigm.

The impact of introducing dynamic refinancing on leverage ratios has been discussed by Fischer,

Heinkel, and Zechner (1989) and Goldstein, Ju, and Leland (2001). The basic finding of Goldstein,

Ju, and Leland (2001) is that when firms are allowed to adjust their leverage dynamically, the

optimal leverage at the initial date changes (for their benchmark set of parameter values) from

50% to 37%.15

Quantitative results also confirm the extent of the adjustment cost e↵ect, which is very similar
15
The leverage ratio of 50% in their base case (i.e., when firms cannot adjust dynamically) is lower than the leverage
of approximately 70% in our static endogenous default model — mostly because of di↵erent tax assumptions. That
is, Goldstein, Ju, and Leland (2001) assume that firms partially lose their tax benefits when their earnings are
sufficiently low. Although we do not detail the economic reasoning for these assumptions here, we note that their
usage illustrates once again that the introduction of new variables or economic mechanisms can be studied within
the same framework, and their importance can be assessed quantitatively.

35
in nature to that in menu cost models. Even though the transaction costs of debt issuance are

relatively small (in our benchmark case, only 1% of the newly issued debt), their impact on both

optimal leverage and, as we shall see, the expected time between refinancing dates is substantial. In

particular, small adjustment costs produce large inaction regions where firms “passively” observe

changes in their fortunes and leverage without actively adjusting their external financing. Indeed,

in the model we have developed, firms are frequently inactive. Yet, when firms finally refinance,

they do so in a lumpy fashion, issuing a lot of debt (in addition to the debt they have recalled

under the model assumptions). At the heart of the matter is the realization that this “passive”

behavior is entirely optimal and rational.

The observation that adjustment costs lead to lumpy behavior is not new. For example, research

has extensively investigated the relationship between adjustment costs and lumpiness in the context

of investment, hiring, and wage setting (Stokey (2009) discusses these and many other examples).

Studies in various asset pricing contexts has been exploring the impact of transaction costs for

a long time as well. However, inaction and lumpiness have entered corporate finance research

relatively lately, perhaps because any study of such issues must involve a fully fledged dynamic

model. Thus, an example of capital structure serves not only as a means to address an interesting

research question in its own right, but also as an illustration of a wider agenda and methods that

can be applied to many other topics studied under the corporate finance umbrella.

2.5 Cross-Sectional Implications of Dynamic Capital Structure

How well do these models account for the empirically observed behavior of firms? A model with

static capital structure decisions produces leverage that is much too high to be consistent with

what is observed in reality. Allowing firms to adjust debt levels dynamically results in much

lower leverage. Dynamic models thus give rise to a hope that dynamic adjustment can facilitate a

reconciliation of the model with the data.

There is, however, a substantial glitch in this result, as identified first in a corporate finance

context by Strebulaev (2007). He emphasizes that findings about “optimal” capital structure relate

36
only to those moments in time when “active” financial decisions are made and leverage is actively

adjusted. They thus cover only one set of observations. Indeed, as these models predict, and as

empirical evidence suggests, firms do not adjust continuously. Instead, empirical data on leverage

come from the cross-section of firms that spend substantial time in the inaction region between

consecutive refinancings. Therefore, it is inappropriate to compare empirical cross-sectional leverage

ratios with the model-implied leverage ratios at refinancing dates because they typically refer to

di↵erent points in time. The extent of the bias depends on the properties of the cross-sectional

distribution of leverage. To draw comparisons between dynamic models and empirical data, we

need to make sure that the data implied by the model and the data taken from real firms have

mutually consistent structures.

Such an idea is obviously inspired by dynamic considerations, and simply could not have been

entertained within the static paradigm. As a practical implementation of this idea, it is important to

determine the extent to which cross-sectional leverage di↵ers from leverage at times of adjustment.

Such a question can be answered only by quantifying the dynamic intuition using a formal model.

To this end, Strebulaev (2007) simulates leverage dynamics for the cross-section of firms implied by

a dynamic capital structure model. If the model represents the true financial policies of firms in the

real world, then the cross-sectional data it produces should be similar to the cross-sectional data

available to empiricists. as in Bhamra, Kuehn, and Strebulaev (2010a), we refer to the dynamics

of the entire cross-section as “true dynamics,” to distinguish these dynamics from the “dynamics”

pertaining only to times of refinancing. Although the model describes the capital structure decisions

of an individual firm, nothing prevents us from using it for a cross-section of firms, assuming that

all firms make optimal decisions independently of others.

2.5.1 Leverage in True Dynamics vs. Leverage at Refinancing Points

Consider an economy consisting of N firms, each of which is driven by the dynamic capital structure

model in Section 2.4. Assume that all firms are identical replicas of one another at the time of

37
refinancing.16 Although their optimal leverage at refinancing is thus the same, their leverage

at any other time could be quite di↵erent if idiosyncratic shocks hitting firms’ fortunes are not

perfectly correlated. For illustration, let’s assume that the Brownian components of firms’ cash

flow stochastic processes are independent, such that all random shocks are idiosyncratic.

In this case, we are interested in the dynamic behavior of only one variable of interest, the

leverage ratio L, which for the dynamic model at any point in time between refinancing waves is
D(Xt )
expressed as Lt = D(Xt )+s(Xt ) , where D(Xt ) is given by Equation (2.26) and s(Xt ) by Equation

(2.28). We resort to a simulation of this model, though recently Morellec, Nikolov, and Schürho↵

(2012) and Korteweg and Strebulaev (2012) investigated similar models’ performance by deriving

the cross-sectional stationary distributions of leverage. Simulation has the advantages of generality,

because it can be applied equally well to more complicated models and, in our case, simplicity.

Here is how the simulation works: At date 0, N all-equity firms in the economy are “born” and

raise debt for the first time. All firms at that point are identical. Every period after that, shocks

to their cash flows are realized (in our case, only idiosyncratic shocks), and if neither the default

nor the refinancing thresholds are reached, firms optimally take no actions. It is important to note

that though the valuation takes place under the risk-neutral distribution, actual shock realization

falls under the actual distribution, so the risk premium needs to be taken into account.17 If the

refinancing trigger is activated, firms recall existing debt and issue new debt. Because we care

only about the leverage ratio, we assume without loss of generality that at that moment the firm

is “re-born” anew, such that all its cash flows are normalized to X0 . If the firm defaults, a new

firm replaces the defaulted one and raises debt for the first time (this simplifying assumption keeps

the number of firms in the economy constant). We continue the process for T0 years, where T0 is

large enough to ensure the economy converges to its stationary cross-sectional distribution. Then

we simulate the economy for T further years and study the resulting T ⇥ N panel data set. For
16
Note that in the model we have developed, size is irrelevant because of the first-order homogeneity property.
Kurshev and Strebulaev (2006) introduce truly fixed costs of debt issuance that make size an important factor in
decision making.
17
The risk premium accounts for the di↵erence between risk-neutral and actual processes. In the case of Brownian
shocks, the di↵erence is in the instantaneous growth rate.

38
this example, we choose T0 of 150 years, T of 30 years, N equal to 5,000, and T , which is the

length of one period in a simulation, of one quarter.

Table 6 reports the results of this exercise for both endogenous and exogenous default cases,

where the benchmark set of parameters is the same as in Section 2.4. In addition, the risk premium is

assumed to be 5%. The first two columns present the optimal date-0 leverage ratios in the static and

dynamic models for comparison. The next three columns summarize the cross-sectional behavior

of leverage in true dynamics. Consider the endogenous default case. For the benchmark dynamic

capital structure, the date-0 leverage is 0.52. The firm refinances when its leverage decreases to 0.35

and defaults when its leverage reaches 1. The distribution of leverage in true dynamics thus falls

between 0.35 and 1. As expected, there is substantial variation in the cross-sectional distribution

of leverage, but the average leverage in true dynamics is higher than that at date 0.

What can explain the result in which the mean leverage in true dynamics is not equal to the

optimal leverage ratio at refinancing? Figure 9 shows the stationary cross-sectional distribution

of leverage and clearly reveals that the distribution is very asymmetric, with the optimal leverage

ratio as the mode rather than the mean of the distribution. As Korteweg and Strebulaev (2012)

show for a more general case, this result is generic. Other than in special circumstances, the mean

of such a distribution does not equal its mode. Correspondingly, Table 6 shows that the average

leverage in true dynamics never equals the date-0 leverage ratio for all the parameter sets reported;

sometimes, the di↵erence is quite large.

The mean is more likely to be larger than the mode if the date-0 leverage ratio is lower or if

the distribution has fatter tails. Thus, when asset volatility is low or the issuance cost is large,

the mean is more likely to be lower than the mode. If the date-0 optimal leverage ratio is lower

than this model suggests (e.g., if we introduce asymmetric taxation of corporate profits and losses),

the di↵erence between the mean and the mode increases. As soon as we realize that the average

leverage in true dynamics is what we need to compare against the empirically observed average

leverage if we want to be consistent, we also recognize that the model can actually be further away

from accounting for stylized leverage facts.

39
 The intuition behind these results is, of course, very general and not confined to a specific

dynamic capital structure model. For example, firms are not identical even at refinancing, because

they have di↵erent characteristics. In addition, their shocks are partially correlated as a result of

the systematic macro- and industry-wide shocks that a↵ect the entire economy. Strebulaev (2007)

takes these factors into consideration and finds that the average leverage ratio in the dynamic cross-

section is typically substantially higher than that at refinancing for many sets of parameters. In

other words, “true dynamics” largely cancel out the decrease in optimal leverage due to a dynamic

adjustment.

2.5.2 Empirical Capital Structure Studies

To explore the power of true dynamics further, we apply a cross-sectional approach, which we have

used thus far to understand average leverage ratios, to bridge the gap between theory and traditional

empirical studies of capital structure. We start by painting a broad picture of how conventional

empirical studies of capital structure work. Early empirical studies of corporate capital structure,

such as the well-known and careful study by Titman and Wessels (1988), tried to identify factors

that, according to various capital structure theories, should a↵ect corporate financial policies,

and then attempted to pinpoint their explanatory power and sign. What Titman and Wessels

(1988) and the many empirical studies they inspired all found, however, is that leverage is robustly

correlated only with a small set of explanatory variables, such as size, profitability, and the market-

to-book ratio. Thus, an important problem is that this approach typically lacks sufficient power

to distinguish between competing theories of capital structure, because these hypotheses actually

provide the same qualitative conjectures with respect to most of the variables of interest.

A prominent example is a long-standing e↵ort to distinguish between the trade-o↵ theory of

capital structure (the dynamic capital structure model that we have developed here is an example

of such a theory) and the so-called pecking order model (Myers and Majluf 1984), which is rooted in

the idea that there is an inherent information asymmetry between corporate insiders and outsiders.

Although the dynamic version of the pecking order theory has not yet been formalized, many of

its implications might be derived intuitively. As a good illustration of this pure intuitive approach,

40
Fama and French (2002) collect most of the knowledge and methods accumulated by traditional

empirical approaches up to that point. Fama and French (2002) o↵er an an entirely intuitive,

qualitative discussion of the comparative statics implications of static and dynamic versions of

the pecking order and trade-o↵ models. In particular, they assert that the e↵ect of profitability

is perhaps the only really outstanding di↵erence between the two. In the pecking order, higher

profitability leads to less reliance on debt, because internal funds crowd out external debt financing,

and thus lower leverage. The trade-o↵ argument instead intuitively calls for higher leverage, because

higher profitability increases the potential tax benefits of debt and reduces expected bankruptcy

costs. Using a regression analysis, Fama and French (2002) find, consistent with most prior research,

that leverage is negatively related to profitability and thus conclude that this is “the important

failure of the trade-o↵ model” (p. 29). Indeed, this finding is identified as only “one scar on the

trade-o↵ model”(p. 30, added emphasis).

The shortcoming of this intuitive approach should by now be clear. Most authors simply confuse

the predictions of the dynamic trade-o↵ model at the point of refinancing and “true dynamics.”

If our goal is to compare model predictions with the results of a conventional regression analysis,

we should consider only the latter. Hence, these studies cannot hope to measure the extent to

which the trade-o↵ model actually explains the real behavior of firms in an economy. The dynamic

relation between leverage and profitability is a particularly striking example of a compelling test of

the credibility of empirical cross-sectional research.

To do so, we use the same procedure that we adopted to analyze average leverage in the dynamic

cross-section. Profitability is the ratio of earnings before interest and taxes (EBIT, or the sum of

the net payout and retained earnings) to the book value of assets in place. In this model context,

Xt is the net payout to claimholders, and retained earnings refer to the change in the value of

book assets. Because the investment process is fixed, we assume that the book assets grow at a

rate equal to the growth rate of the net payout. This growth rate is the only one, for which the

market-to-book ratio for assets has a finite nonzero expected value at the infinite time horizon.

Table 7 reports the results of a similar experiment from Strebulaev (2007) (see Table IV, p.

41
1771). The regressions, consistent with empirical studies, focus on quasi-market leverage (see

Equation (2.22)), and the experiment introduces heterogeneity across firms at refinancing points

by varying primitive parameters and thus including several controls (e.g., asset volatility, default

costs). It also introduces systematic risk and therefore runs simulations for many economies (for

details see Strebulaev (2007)). The last two columns show the 10th and 90th percentile values

of regression coefficients across economies. The empirical methodology matches several prominent

papers, though we report only the results from running the Fama and French (2002) regressions on

the model-simulated data.

It is important to understand the intuition behind these results: When all firms reach the re-

financing date, the regression of leverage on profitability produces a result that, not surprisingly,

matches the intuition proposed by Fama and French and others, namely, that leverage and prof-

itability are positively related in the world described by the trade-o↵ model. This result is shown

in the first column of Table 7. However, the same regression performed on both actively rebalanc-

ing and passive firms produces a negative relation between leverage and profitability, in line with

previous empirical findings. (A recent empirical work by Danis, Rettl, and Whited (2012) comes

to similar conclusions.) An important implication is that empirical researchers, such as Fama and

French, review the data simulated by this model, they would likely interpret the findings as evidence

in favor of the pecking order argument, contrary to the predictions of the dynamic trade-o↵ model,

even though of course the data set was produced by a trade-o↵ model in the first place! In short,

as Strebulaev (2007) shows, this, and likely many other quantitative, as wells some qualitative,

conclusions of extant literature are unwarranted, mostly because the studies that achieve these

conclusions ignored true dynamics.

The negative relationship between leverage and profitability might instead be understood in the

modeling context by taking dynamic considerations into account. Profitability is persistent, so an

increase in profitability a↵ects future profitability and thus the value of the firm. Both the market

value, through embedded expectations, and the book value, through retained earnings, increase.

But whereas an increase in the value of the firm always lowers leverage, it does not necessarily lead

42
to refinancing in a world with infrequent adjustment. Debt levels, therefore, are not a↵ected unless

a sufficiently high profitability level is reached, triggering refinancing.18 Similar considerations hold

as profitability decreases. As a result, for any individual firm between two refinancing dates the

relationship between leverage and profitability is strictly negative. The positive relationship, on

the other hand, is an outcome of cross-sectional forces at the refinancing date. Although these

arguments seem intuitive, assessing their quantitative relevance requires a dynamic model. The

simulations based on such a model clarify the extent to which the negative relation between leverage

and profitability dominates the positive relation we can observe at refinancing. The result on

the prevalence of the negative e↵ect for a wide range of parameters is implied by a quantitative

assessment of the dynamic behavior of the model, not on the qualitative comparative statics. As

Strebulaev (2007) shows, the quantitative values of the profitability slope coefficient as reported by

Fama and French (2002) are consistent with the coefficient values produced in his model-implied

simulations. In other words, empirically observed profitability coefficients are generated within a

reasonable range of simulated economies.

Overall then, dynamic cross-sectional methods applied to a dynamic trade-o↵ contingent claims

model indicate that the most important and seemingly non-controversial result in empirical capital

structure work in fact o↵ers little if any value.

2.6 Macroeconomic Fundamentals and Capital Structure

How do macroeconomic factors a↵ect corporate financial decisions? How do corporate financial

decisions by individual firms a↵ect the aggregate properties of the economy? Because dynamic

contingent claims models are closely related to asset pricing, they o↵er us an opportunity to consider

issues that traditionally have been the preserve of asset pricing, such as the equity premium puzzle.

These models can address asset pricing issues at the same time as corporate finance issues by

successfully blending contingent claims models and traditional asset pricing models.

The rationale for pursuing this research is straightforward. Consider traditional consumption-
18
To be precise, the book debt values are not a↵ected. The sensitivity of the market value of debt to profitability
is much smaller than that of equity value, and if we replace quasi-market leverage with market leverage, the results
are barely a↵ected.

43
based asset pricing models, starting with Lucas (1978), whose goal is to understand various quan-

titative features of the macroeconomic data, such as the equity premium puzzle of Mehra and

Prescott (1985). These models are typically built around a representative agent, who acts as a

consumer as well as the owner of financial assets in the economy. However, corporations live in the

same world, and firm owners and managers are also consumers. A growing body of empirical work

indicates that common factors may a↵ect the equity risk premium and credit spreads on corpo-

rate bonds. The economic mechanisms behind these “puzzles” (e.g., equity premium, credit risk,

risk-free rate puzzles) are likely linked to the economic mechanisms underlying optimal financial

decisions.

This intuition forms the premise for work by Bhamra, Kuehn, and Strebulaev (2010b) that

builds on the framework developed by Bhamra, Kuehn, and Strebulaev (2010a) and earlier contri-

butions by Hackbarth, Miao, and Morellec (2006) and Guo, Miao, and Morellec (2005). A similar

set of issues is independently considered by Chen (2010). This stream of research develops a

new unified framework that explores asset pricing and corporate finance issues simultaneously and

thereby enables us to study how the optimal financial structure of firms, at both individual and

aggregate levels, is a↵ected by time-varying macroeconomic conditions. Therefore, in this section,

we briefly introduce and discuss this theoretical framework at an intuitive level (a more extensive

formal development requires additional technical apparatus, which is beyond the scope of this re-

view). We also discuss how this framework might reconcile existing empirical evidence about the

relationship between capital structure and macroeconomic factors. Empirically, aggregate leverage

in the economy appears counter-cyclical (e.g., Korajczyk and Levy (2003)), yet parallel evidence

suggests that firm-level leverage is pro-cyclical (e.g., Korteweg (2010)).

In a nutshell, Bhamra, Kuehn, and Strebulaev (2010b) embed a dynamic capital structure

model of the type described in Section 2.4 within a dynamic consumption-based asset pricing

model with a representative agent. From an economic perspective, this approach enables us to

study the impact of macroeconomic factors on the behavior of individual and aggregate corporate

capital structure. The advantage of this type of embedded model over a pure contingent claims

44
model is that it endogenously links unobservable risk-neutral probabilities with observable actual

probabilities through the market price of consumption risk and the agent’s preferences. In turn,

we can investigate the e↵ect of macroeconomic risk on optimal financing decisions endogenously,

because the agent prices the claims on the basis of his or her attitude toward the macroeconomic

risks present in consumption and cash flows. With this framework, we also explain intuitively

how preferences (e.g., risk aversion, the elasticity of intertemporal substitution) a↵ect financing

decisions. Finally, the framework enables us to study actual default probabilities and, in particular,

explore the relation between leverage and default likelihood. All these tasks would have been

impossible with the previous generation of contingent claims models, because the pricing kernel is

given exogenously, and so there are no macro factors to speak of.

The economy consists of consecutive periods of stochastic booms and contractions that a↵ect,

through consumer preferences, asset prices and optimal corporate decisions. Business cycles are

modeled as a two-state Markov chain process, where the probability of switching from one state

into another is asymmetric in such a way that agents expect the economy to spend more time in

booms than in contractions. Booms di↵er from contractions along several dimensions: The growth

rate (volatility) of the cash flow and consumption processes is higher (lower) in booms, whereas

the correlation between the firm’s cash flows and consumption is higher in contractions.

One significant consequence of financial decisions in this model is that optimal leverage at

refinancing is now state-dependent: If firms refinance in bad times, they optimally choose lower,

leverage because cash flows are expected to be lower in the future, and the distance to default

is expected to be shorter. For a similar reason, firms are more likely to default in contractions,

all else being equal. This path dependence, however, leads to a more intricate relation between

macroeconomics and leverage. Firms that lever up in expansions have higher leverage and are more

vulnerable in recessions. As an interesting empirical prediction, we note that firms that choose to

raise debt in booms are more prone to default when macroeconomic conditions worsen. From an

aggregate perspective, optimal levering up in booms leads to default clustering in contractions.

Conversely, firms that refinance in recessions are more likely to be relatively underlevered in booms

45
and thus enjoy lower tax benefits from issuing debt. They also are more likely to refinance in an

expansion. These and many other predictions of the framework have not yet been tested empirically.

As the trade-o↵ intuition implies, optimal leverage is pro-cyclical at the refinancing point. How-

ever, aggregate leverage is actually counter-cyclical in true dynamics. The direction of cyclicality

di↵ers between true dynamics and refinancing points, because when the state of the economy wors-

ens, the market value of equity falls more than the market value of debt, for equity, as a residual

claim, is more sensitive to macroeconomic risk. This mechanism induces counter-cyclicality in

leverage. In true dynamics, the latter e↵ect dominates the optimal choice of debt at refinancing.

The di↵erence between true dynamics and refinancing points explains why it is not an easy task

to study the relation between macroeconomic conditions and financial structure in the data and

why various empirical studies achieve seemingly inconsistent results. The result related to the

counter-cyclicality of the whole cross-section is consistent, for example, with Korajczyk and Levy

(2003), who find leverage to be counter-cyclical for relatively unconstrained firms. Yet, Covas and

Den Haan (2007) and Korteweg (2010) empirically find that when firms choose optimal leverage,

leverage decisions are pro-cyclical, consistent with the refinancing date result. The intuition behind

this result is similar to our approach in Section 2.5, where we contrast the refinancing point (or

date-0) leverage ratio and the cross-sectional leverage ratio in true dynamics.

The framework obviously delivers multiple interesting quantitative results. Macroeconomic

risk leads to substantially lower leverage at refinancing. Under the benchmark set of parameters

considered by Bhamra, Kuehn, and Strebulaev (2010b), the date-0 leverage ratio is 32% in a boom

and 22% in a contraction. The unconditional (i.e., independent of the state of the economy) leverage

ratio is 28%, consistent with stylized facts pertaining to the extent of debt usage. In true dynamics

however, consistent with the results of Section 2.5, leverage is substantially higher in both bad and

good times, rising from 22% to 44% in contractions and from 32% to 38% in booms.

Of course, Bhamra, Kuehn, and Strebulaev (2010b) and others also emphasize how much we do

not yet know about the relation between macroeconomics and corporate financial and investment

decisions. For example, the model assumes that consumption is una↵ected by corporate defaults

46
and that, as in traditional consumption-based models, all decisions are made by a representative

agent. Intuitively, both investment and consumption may interact with corporate financial policies.

Also, financial claims, such as corporate debt and equity, are held by di↵erent agents, which may

prompt conflicts of interest and important implications for default and leverage. Above all, we do

not know how these factors a↵ect leverage dynamics.

2.7 Debt Structure and Strategic Renegotiations

Our treatment of debt has so far assumed that if the firm is unable to cover its debt obligations,

the ensuing default leads either to a reorganization or a bankruptcy. In other words, we have

assumed away the possibility of renegotiation. This scenario is realistic in the case of a firm

issuing public bonds under the U.S. jurisdiction, because, according to the 1939 Trust Indenture

act, any change in the major features of a debt contract, such as principal, interest, or maturity,

require a unanimous consent of all bondholders, a practically infeasible task. However, U.S. public

bond markets represent only one side of the debt market. Another is represented by private debt

contracts, such as bank debt. A critical di↵erence between public and private debt is that firms

can attempt to renegotiate the latter and, if the renegotiation is successful, the firms change the

structure of payments to debtholders to avoid default. This gives rise to shareholders behavior

known as strategic renegotiation: shareholders refuse to honor debt obligations in full and threaten

default, even though they have sufficient resources to fulfill their promises. Such a threat can be

e↵ective only if debtholders face default costs they prefer to avoid.

The importance of strategic renegotiation was explored originally in the credit risk literature,

pioneered by a classical paper of Anderson and Sundaresan (1996) and followed by Mella-Barral

and Perraudin (1997) and Fan and Sundaresan (2000). We consider here a simple version of a static

capital structure model with strategic renegotiation, closely following Hackbarth, Hennessy, and

Leland (2007). The model is similar to the model of Section 2.3 (the case of endogenous default

with liquidation), but in addition to public debt, private (bank) debt can be issued by the firm.

The only di↵erence between public and bank debt is in the possibility that the firm can renegotiate

bank debt at any time

47
 First consider the scenario when only bank debt with principal B is outstanding and the firm

has a strong bargaining position, i.e., it has the ability to make take-it-or-leave-it o↵ers to the bank.

If the bank rejects the o↵er, the firm defaults and the firm is liquidated. In this case, the bank’s

promised payo↵ is given by:


cB X
RB (X) = min , (1 ↵)(1 ⌧) , (2.32)
r r µ

where cB is the initially promised coupon, and (1 ↵)(1 ⌧ ) rXµ is the post-bankruptcy value,

which is similar to that given by Equation (2.9). RB thus represents the bank’s reservation value,

because the bank would reject any o↵er that yields a lower payo↵ than RB . If the shareholders have

all the bargaining power, the bank’s promised claim always equals RB . In the absence of public

non-renegotiable debt, the firm never defaults but rather negotiates its debt payments in bad states

of the world. In this model, the firm renegotiates its debt if the value of cash flow reaches some

value XS for the first time. Because the firm has full bargaining power, the firm renegotiates to

keep the value of bank debt at its reservation value (1 ↵)(1 ⌧ ) rXµ . As a result, the firm pays a

lower coupon rate, given by:

X
s(X) = (r µ)(1 ↵)(1 ⌧) . (2.33)
r µ

Note that the firm never defaults and thus default costs do not occur on equilibrium path. The

total firm value can be written as the sum of the after-tax unlevered firm value and the present

value of bank debt tax benefits:

X
V (X) = (1 ⌧) + ⌧ B(X), (2.34)
r µ

where B(X) is the value of bank debt. Outside of the renegotiation region the value of the bank

debt is determined using the same logic we used in Section 2.3:

✓ ◆✓ ◆⇠1
cB cB XS X
B(X) = (1 ↵)(1 ⌧) , X XS . (2.35)
r r r µ XS

The first term is the value of riskless bank debt and the second term takes into account the

48
impact of renegotiation. Equityholders choose XS by maximizing equity value:

X
E(X) = V (X) B(X) = (1 ⌧) (1 ⌧ )B(X). (2.36)
r µ

The first order condition with respect to the renegotiation threshold XS yields:

c B ⇠1 r µ
XS = . (2.37)
r ⇠1 1 (1 ↵)(1 ⌧)

At date 0, maximizing equity value with respect to cB and taking the threshold value into

account, gives the optimal coupon rate:

⇠1 1 X0
c⇤B = r(1 ⌧ )(1 ↵) . (2.38)
⇠1 r µ

The optimal levered firm value is then:

X0
V (X0 ) = (1 ⌧ )(1 + ⌧ (1 ↵)) . (2.39)
r µ

Is the optimal leverage or firm value higher in public debt or bank debt cases? The comparison

of cB and c (Equations (2.38) and (2.20), respectively) shows that there is no unambiguous answer

to this important question. In the bank debt case, the leverage ratio is constrained by the extent

of default costs. For example, if the asset value is fully destroyed in default (↵ = 1), the firm can

always threaten the bank to default and thus no debt is feasible initially (cB = 0). On the other

hand, if default costs are very low, the firm does not have any credible threat to renegotiate the

terms of the loan and the resulting leverage is very high.

Table 2.7 compares the two cases corresponding to the parameter sets of Section 2.3.5. For the

benchmark set, the bank debt coupon is 2.29 as opposed to 1.36 for public debt, and leverage is

0.76 vs. 0.70. For all the parameter variations in the table, the bank debt coupon is higher than the

coupon on public debt. The leverage ratio, however, can be lower in the bank debt case, because

bank debt leads to lower ex-ante inefficiency and thus higher total value, fully appropriated by

equityholders. The extent of an increase in value is gauged by the ratio of levered to unlevered firm

values, which is substantially large for the bank debt case.

So far we have assumed that firms can use only one source of financing, although empirically

many firms issue both bank and public debt at the same time. To incorporate two types of debt in

49
the same model we need to decide on the seniority of debt claims. Hackbarth, Hennessy, and Leland

(2007) show that in this model it is optimal to give a strict seniority to bank debt, consistent with

empirical evidence. Bank debt seniority protects the firm from ex-ante inefficient renegotiation if

the cash flow level is relatively high and facilitates ex-post efficient renegotiation in the bad states

of the world.

The renegotiation region is modified to take into account the endogenous default on public debt,

which occurs when the cash flow level reaches XD . The renegotiation region is then [XD , XS ) and

the non-distress region is [XS , 1). For simplicity, assume that the recovery of public debtholders

in default is zero. Because the presence of public debt does not change the renegotiation trigger,

the solution for XS is the same. The value of public debt can then be written as:
✓ ◆ !
c X ⇠1
D(X) = 1 . (2.40)
r XD

The smooth-pasting condition gives the following default threshold:

⇠1 c
XD = r(1 (1 ⌧ )(1 ↵)). (2.41)
⇠1 1r µ

Firm’s objective function at date 0 includes all the benefits and costs of both types of debt:
✓ ◆⇠1
X0 XD X
V (X0 ) = (1 ⌧) + ⌧ B(X0 ) + ⌧ D(X0 ) (1 ⌧) (1 (1 ⌧ )(1 ↵)) . (2.42)
r µ r µ XD

As before, the value function is increasing in the value of bank debt and thus the optimal bank

coupon rate is still defined by (2.38). Optimization with respect to public debt coupon yields:
✓ ◆ 1
⇠1 1 r ⌧ ⇠1
c⇤ = (1 (1 ↵)(1 ⌧ ))X0 . (2.43)
⇠1 r µ ⌧ ⇠1

Note that public debt is a complement rather than a substitute for bank debt, because the firm

still wants to issue as much bank debt as possible. The last set of columns in Table 2.7 shows that

the additional public coupon is 0.16 in the benchmark case, with the leverage ratio increasing to

0.79. In this version of the model, however, adding public debt makes little quantitative impact on

the total extent of debt.

50
2.8 Capital Structure and Temporary vs. Permanent Shocks

Empirical evidence strongly suggests that financial managers consider financial flexibility, earnings

and cash flow volatility, and insufficient internal funds the most important determinants of financing

decisions by their firms (e.g., Graham and Harvey (2001)). Although intuitively appealing, these

factors continue to puzzle corporate finance theorists, who find it hard to pin down the exact

mechanism of their influence. We may know the nature of economic frictions that push firms to

adopt more cautious financing behavior, but existing attempts to quantify their e↵ects leave a great

deal to be desired. In Section 3, we introduce endogenous investment models that analyze these

issues directly. Here, we consider a paradigm change within the dynamic contingent claims models

that takes these issues into account.

2.8.1 Impact of Temporary Shocks

If we reconsider the dynamic capital structure model of Section 2.4, we confront some difficulty

explaining these stylized facts. In fact, the model induces several undesirable implications. First,

the non-negativity of market values implies that cash flows in the model can never be negative,

which is, of course, highly unrealistic. For example, between 1987 and 2005, approximately 17% to

25% of all quarterly cash flows for the full Compustat sample were negative. On a broader level,

the model ignores the strong likelihood that firms that are highly profitable today may experience

negative cash flows in the near future. In other words, firms value financial flexibility as a means

to sustain them when cash flows are very low, even though their future is bright. For example,

consider the economic marginal tax rate that Scholes and Wolfson (1992) define to account for the

present value of current and future taxes on an extra dollar of income earned. Existing models

would predict that profitable firms sustain their profitability in the future. They are therefore

assigned high tax rates even though in reality their profitability is less persistent and their true

marginal tax rate should be lower. Their realistic expectations should lower their inclination to

issue debt.

Second, the volatilities of cash flow and asset value growth rates are equal in this model, which

51
makes it difficult to reconcile with the reality in which managers care more about earnings volatility

than about asset volatility. Because the asset volatilities of most firms are relatively low, the

distinction is an important one.19 Third, both changes in and levels of current cash flow and asset

value are perfectly correlated, which is inconsistent with both intuition and evidence.

As it turns out, these problematic features all result from the permanent nature of cash flow

shocks in the model, and this permanence itself is caused in part by reliance on geometric Brow-

nian motion as an underlying cash flow process. Although this process is typically assumed for

tractability, its historical significance should not be understated: The direct predecessor of dy-

namic contingent claims models is the Black-Scholes model, in which the behavior of stock prices

(unlike corporate cash flows) is much more similar to the behavior exhibited by geometric Brownian

motion.

Investigating the importance of the temporary nature of cash flow shocks is the main focus of

Gorbenko and Strebulaev (2010), and we follow their approach closely in this section. 20 Consider

the e↵ect of introducing a temporary component to an otherwise standard cash flow process. The

exact nature or sources of these shocks is irrelevant, because this intuition is generic. For example,

shocks could a↵ect demand for a firm’s product or the production cost structure. Because their

impact is of limited duration, these transitory shocks a↵ect future cash flows disproportionately

over the time horizon. The consequences are many: The total realized current cash flow can

turn negative (even if asset values are large), earnings and asset volatilities di↵er, and levels and

innovations in current cash flows are imperfectly (though positively) correlated with those in asset

values. Overall then, introducing temporary shocks makes the model’s behavior patterns consistent

with stylized facts about corporate earnings, along multiple dimensions.

More importantly, we need to determine the contribution of this mechanism to explanations of

corporate behavior. Intuitively, for any given level of leverage, the expected advantage to debt is

likely of lower value when cash flows are more volatile thus prompt predictions that they will be
19
For example, Schaefer and Strebulaev (2008) estimate annualized asset volatility for an average firm with an
existing investment-grade rating to be approximately 23%.
20
Grenadier and Malenko (2010) explore the impact of similarly modeled temporary shocks in a real options
investment problem.

52
low more frequently. For example, because we expect cash flow often to be lower than the level of

contractual financial obligations, the expected tax benefits of debt should diminish in the absence

of full tax loss o↵set provisions. In other words, incentives to use debt decrease with the probability

that a firm will experience nontaxable states of the world (Kim 1989; Graham 2003).

To quantify this intuition, we build a fully fledged dynamic contingent claims model with both

permanent and temporary shocks. The main finding is that the value of financial flexibility may

indeed increase dramatically when firms face prospects of immediate adverse temporary shocks

of large magnitude. For the benchmark set of parameters, the optimal static leverage ratio de-

creases from 58% to 37% with the introduction of temporary shocks, conditional on retaining the

same total asset volatility. As expected, the model predicts that the e↵ect of temporary shocks

is disproportionately larger for small firms (an almost 30% decrease in leverage, compared with

the permanent-only model), consistent with empirical evidence about the relation between firm

size and leverage. Temporary shocks appear particularly important when asset volatility is low.

For example, when asset volatility is 0.25 per annum (an empirically realistic value), leverage is

lower by 27%, but when asset volatility is 0.5 (considerably higher than asset volatility for the

average Compustat firm), leverage decreases only by approximately 3%. Financially constrained

firms are also more a↵ected by temporary shocks. For example, a firm with a 20% proportional

cost of raising money in financial distress chooses an optimal leverage of 31%, compared with a

financially unconstrained firm, whose optimal leverage is 47%. The model further predicts that if

managers (but not investors) are myopic and take into account only a limited number of shocks,

their remaining desire to issue debt rapidly decreases. For the same benchmark set of parameters,

if the manager expects only one large shock in the future, optimal leverage falls to 20%, similar to

the quasi-market leverage ratio of a typical public U.S. firm.

2.8.2 Modeling Temporary Shocks

Building a framework that fully incorporates both types of shocks entails formidable technical

challenges. The lack of perfect correlation between permanent and temporary shocks gives rise

to two sources of dynamic uncertainty that make all financing decisions intrinsically complicated.

53
Consider a firm whose cash flows are driven by two stochastic processes: one responsible for the

long-term prospects (e.g., a geometric Brownian motion) and another that constitutes the short-

term deviations from long-term cash flow (e.g., any process that mean-reverts to long-term cash

flow levels). In the permanent-only model, the default boundary, as we have seen, is a certain cash

flow threshold level. But with transitory shocks, the knowledge of either total asset value or total

realized cash flow by itself is insufficient to make a well-informed default decision. The current cash

flow could be low because the firm’s future is bleak (and it is optimal to default). Alternatively, it

might be low because the future prospects are bright, conditional on the firm surviving a temporary

period of losses, in which case it is optimal to make debt payments. Only by decomposing the shock

into permanent and temporary components can equityholders make the optimal default decision.

This problem is very difficult to solve, though, because the default boundary is a curve in a two-

dimensional permanent-temporary shock space. Mathematical problems related to the first passage

to such a boundary have not been solved satisfactorily even for the simplest cases.

Gorbenko and Strebulaev (2010) instead build a simplified framework that incorporates all these

intuitive features. To see how their model works, we start by assuming that at any time t, the

realized cash flow, ⇡t , is the sum of two components: the cash flow generated by the permanent

component, Xt , which reflects the long-term prospects of project returns, and the mean-reverting

cash flow, ✏t , which reflects deviations from the permanent component caused by shocks of shorter

duration:

⇡t = X t + ✏ t . (2.44)

If ✏t ⌘ 0, and X follows geometric Brownian motion, we are back to the capital structure model

of Section 2.3. To illustrate the modeling of the temporary shock, we shut down the permanent

component (Xt = 1) and allow for only one transitory shock to arrive in the future. Important

features of the shock include the random timing of the shock’s arrival, the initial magnitude of the

shock, and its persistence. Let the shock arrive at a random moment Ts (s refers to the shock’s

arrival); its size ✏Ts is also random and determined at Ts . The transitory nature of the shock is

reflected in its degree of persistence; that is, the cash flow mean reverts from Xt + ✏Ts to Xt at a

54
random time Tr (r indicates the shock’s reversal).

Both the timing of the shock arrival Ts and its persistence Tr Ts follow a Poisson process with

intensity that we can write through the following probability density function:

ww
(w) = we , w = {s, r}. (2.45)

Two apparently similar parameters, s and r, play distinctly di↵erent economic roles, such

that s reflects the frequency of “big” transitory shocks hitting the firm, and higher values of s

imply the firm is more often “in the news.” In contrast, r reflects the speed of the shock’s mean

reversion. Indeed, the expected value of the shock size attenuates exponentially with the speed r

after the occurrence date Ts (if the shock has not yet disappeared). At any date t, the expected

value of the transitory shock to the cash flow process at any future date ⌧ > t can be written as:

⇢
0, t < Ts and t Tr ,
Et [✏⌧ ] = r (⌧ t) , (2.46)
✏ Ts e Ts  t < Tr .

The realized transitory shock to the cash flow can similarly be written for any date t as:
⇢
0, t < Ts and t Tr ,
✏t = (2.47)
✏Ts , Ts  t < Tr .

The initial size of the jump is distributed according to the symmetric double exponential distribution

with the following probability density function:

1 |✏Ts |
f (✏Ts ) = e . (2.48)
2

The parameter drives the volatility of the temporary shock, such that smaller leads to shocks

of larger magnitude.

In this model with temporary shocks, the expected cash flow volatility di↵ers from asset value

volatility, and the expected time-series correlation between asset value and cash flows is positive

but less than 1. Finally, this setup naturally accommodates the possibility of negative cash flows.

Even a naive model with temporary shocks thus satisfies all the empirical criteria for a firm’s cash

flows, as outlined in the introduction.

55
 In the presence of negative shocks, the owners of the firm may find it optimal to abandon the

project. Modeling abandonment in this case is similar to its counterpart in real option models.

The threshold level of abandonment is denoted by ✏A , at which point firm owners are indi↵erent

between abandonment and continuation, and this value can be found from the indi↵erence condition

by equating the residual firm value to zero:

Z Tr Z 1
r(s Ts ) Tr r(s Ts )
ETs (1 + ✏A ) e ds + e e ds = 0. (2.49)
Ts Tr

The abandonment level that satisfies Equation (2.49) is:

r +r
✏A = . (2.50)
r

The transitory nature of the shock dictates that the abandonment level is always lower than the

negative value of the permanent cash flow, ✏A < 1. Shorter-lived shocks yield a lower abandonment

threshold, because they are likely to have a marginal impact on asset value. Lower interest rates also

lead to less abandonment, because they render the bright future more valuable after the negative

shock is over.

We next define asset value as the present value of all future cash flows streaming from an asset’s

cash flow rights. It is thus the sum of a perpetuity paying a permanent cash flow and the present

value of transitory deviations, which equals:

Z 1
Xt
Vt = + Et e r(s t) ✏s ds
r t
1 1 s 1
= + e ✏A , t < T s . (2.51)
r 2 r + s (r + r )

The second term is always positive and reflects the asymmetric option-like nature of limited liability.

As we might expect, asset value increases with the variance of the shock’s size, the intensity of the

shock’s occurrence, and its persistence.

As in previous models, it helps to introduce simple Arrow-Debreu-type securities that pay o↵

upon the shock’s arrival. The payo↵ value depends on the region in which the temporary shock

falls. Consider a security A(g(✏T ), , b) that pays g(✏T ) at time T of the shock’s arrival if the size

of the shock ✏T is at least as great as b. These securities do not assume limited liability – their

56
payo↵s, and therefore values, can be negative. The necessary securities have triplets of parameters

(1, s, 1), (1, s , b), and (✏Ts , s , b). The first one pays $1 at time Ts for any value of the shock.

The second and third securities pay correspondingly $1 and ✏Ts if the size of the shock is at least

as great as b.

The values of these securities at time t, t < Ts , are as follows:

h i
r(Ts t) s
s 1)
A(1,
, = E t e = , (2.52)
r+ s
h i 1 b
s
A(1, s , b) = A(1, s , 1)Et 1|✏Ts > b = (1 e ), (2.53)
r+ s 2
h i 1 b
s
A(✏T ( s ) , s , b) = A(1, s , 1)Et ✏Ts |✏Ts > b = e (1 b). (2.54)
r+ s 2

The default boundary, to be optimally determined by equityholders is denoted by ✏D . The

equity value at any time t before the shock occurs can be written as:
✓ ◆
1 c 1 c 1
Et = (1 ⌧) (1 A(1, s, 1)) + A(1, s , ✏D ) + A(✏Ts , s , ✏D ) .(2.55)
r r r+ r

The last term in Equation (2.55) is the present value of short-term deviations, which is strictly

positive given equity’s limited liability. The expected shock thus has an impact on equity value

opposite that on debt value: Equity value increases with the shock’s frequency, persistence, and

magnitude.

In this simple case, the default boundary is determined by the equityholders’ indi↵erence con-

dition when the value of equity is zero:

Z 1
r(s Ts )
ET s = E T s (1 + ✏D · I{s<Tr } c)e ds = 0, (2.56)
Ts

which gives us the optimal default boundary ✏D as a function of the coupon rate:

r +r
✏D = (1 c) . (2.57)
r

Similarly, we can write the debt value as:

c c
Dt = (1 A(1, s, 1)) + A(1, s , ✏D ) (2.58)
r r
⇣ 1 1 ⌘
+(1 ↵)(1 ⌧ ) (A(1, s , ✏A ) A(1, s , ✏D )) + (A(✏Ts , s , ✏A ) A(✏Ts , s , ✏D )) .
r r+ r

57
The first term reveals the riskless bond value if the shock does not happen. The second term shows

that in the case of no default at the time of the shock, the bondholders are entitled to a perpetuity

(because it is a single-shock economy). The last term shows that in the case of bankruptcy, the

value stems from the perpetuity generated by the permanent cash flow (first component of the third

term) and the present value of deviations from the long-term cash flow. To understand the intuition

behind this representation, we can think of the complex security A(✏Ts , s , ✏A ) A(✏Ts , s , ✏D ) as

the present value of the size of the shock if debtholders incur all of the shock’s cost (i.e., the shock

is large enough to default but not severe enough to abandon the firm). Debtholders then must pay

this present value until the shock mean reverts, as taken into account by the perpetuity formula
1
r+ r
.

Because default and abandonment can happen only in response to negative shocks (i.e., ✏B

and ✏A are less than zero), the value of A(✏Ts , s , ✏A ) A(✏Ts , s , ✏B ) is negative, and therefore

the debt value is always lower in the presence of temporary shocks for a given coupon rate and

other model parameters. This is a fundamental observation: the total value of the firm is larger in

the presence of temporary shocks. This apparent inconsistency results from an ex-post conflict of

interest between equity and debt. Debtholders incur costs in bad states of the world but do not

share in the benefits from good states. It then follows naturally that debt value is a decreasing

function of a temporary shock’s frequency, its persistence, and its magnitude.

Gorbenko and Strebulaev (2010) extend this simple model in two obvious directions: first, Xt

follows the geometric Brownian motion, and, second, many temporary shocks are allowed. What

sets the model apart from all previous contingent claims models is that the interest coverage ratio

(ratio of earnings to interest payments) and the leverage ratio are two distinct economic measures

of a firm’s financial health. The current level of cash flow can be low because of a negative

temporary shock, which would imply a low interest coverage ratio and potential liquidity problems,

whereas the future cash flows can be expected to be high, implying a low leverage ratio. This

expectation can explain at least two important stylized facts. First, managers care about cash flow

or earnings volatility, instead of asset volatility. Second, firms that default can be broadly allocated

58
to two types, either those with very low net worth (poor future prospects) or those with very low or

negative interest coverage ratios (liquidity distress), as described by Davydenko (2010), for example.

Overall, accounting for short-term shocks appears crucial for understanding the financial decisions

of corporations.

2.9 Further Applications

Recent research has explored a number of further applications of this framework to corporate finance

that enrich it along various dimensions. Here, we briefly discuss several of them, which we believe

o↵er substantial promise for the future.

Dynamic Investment and Financing. One drawback of optimal capital structure models that

we have considered so far is the absence of a non-trivial investment policy. The separation of invest-

ment and financing means that cash flow generating machines are assumed to exist independently

of financing considerations. Recent research has started exploring the dynamic interactions of fi-

nancing and investment policies within the dynamic contingent claims framework. For example,

Boyle and Guthrie (2003) enrich the real options model of the type we consider in Section 2.2 by

imposing future financing constraints. Intuitively, if in the future the firm is unable to exercise

its real option and invest at the optimal time, because funding is unavailable or costly, the real

option value is lowered and the firm will tend to exercise earlier. As a result, higher cash flow

uncertainty that delays investment in a standard real option setup may have the opposite impact

by accelerating investment, because the window of financial opportunities become more valuable.

Campello and Hackbarth (2012) incorporate financial considerations into the standard real

options setup by adding costly equity financing and renegotiable bank debt (with a strategic rene-

gotiation structure similar to the one in Section 2.7). Firms that face higher cost of equity issuance

invest optimally in projects with lower bankruptcy costs to reduce future renegotiation inefficien-

cies. Hackbarth, Mathews, and Robinson (2012) analyze the boundaries of the firm by exploring

the tension between assets in place and growth options in the presence of the trade-o↵ model of

capital structure. In their model, shareholders have access to both assets in place, which produce

59
cash flows as in the capital structure models of Sections 2.3 and 2.4, and a growth option, similar

to the model in Section 2.2. However, the exercise of the growth option cannibalizes the cash

flow potential of assets in place. The firm can initially choose an organizational form, either by

combining the assets in place and the growth option in one firm or by establishing two di↵erent

firms, with one owning the assets in place and the other having the right to any cash flows from

the future exercise of the growth option.

The trade-o↵ faced by the original founders face emphasizes the fundamental tension between

substitutes and complementarities. Having one firm internalizes the decision to exercise the real

option and thus protects assets in place. In the two-firm setup, exercise of the growth option

does not take into account assets in place. Such an exercise strategy maximizes financial flexibility

but may be suboptimal for the founding shareholders. This setup can be useful, for example, in

studying innovation. Innovative products or business processes negatively a↵ect the profitability

of products and processes that they compete with. The framework enables us to explore how

the decision-maker’s objective a↵ects optimal organizational form (see also Grenadier and Weiss

(1997)).

A number of further studies explore such issues, which include the impact of agency conflicts

on the interaction of investment and financing (Childs, Mauer, and Ott 2005), growth options

and priority structure of debt (Hackbarth and Mauer 2012), undiversified entrepreneurial finance

and capital structure in incomplete markets (Chen, Miao, and Wang 2010), and the interaction

between time-to-build investment lags and capital structure (Tsyplakov 2008). In a study that

merges strategic renegotiation and investment, Pawlina (2010) shows that an ability to renegotiate

debt increases the investment trigger and thus exacerbates underinvestment problem.

Asset liquidity. Firms in distress have a number of tools in their arsenal that they can use to

avoid default. Models that we have considered assume that distressed firms access (potentially

costly) contributions from existing shareholders. In practice, firms often try to sell some of their

assets to pay creditors. To what extent this strategy is feasible depends in part on the covenants

60
in the debt contract. Morellec (2001) is the first paper that carefully explores the impact of asset

liquidity and covenants on optimal leverage. In his model, productivity shocks induce firms to

adjust cash-generating capital stock. In the absence of financing frictions, firms prefer highly liquid

assets that can be adjusted at low cost. However, flexibility has another side: it also makes it

easier to sell assets in distress, potentially at fire-sale prices, to satisfy creditors. Morellec shows

that asset liquidity optimally reduces leverage, because higher liquidity leads to higher frequency

of asset dispositions in distress. However, when the firm includes covenants in the debt contract

that prevent it from selling assets, higher flexibility increases optimal leverage.

Managerial Decisions. In the standard setup we have explored, managers act on behalf of

shareholders to maximize equity value. In practice, corporate managers have their own agendas

and maximize their own utility subject to the constraints imposed by the shareholders. Although

central to traditional corporate finance theory, this topic has yet to be fully explored within this

framework. A paper by Morellec (2004) o↵ers one of the first attempts to incorporate managerial

preferences in capital structure decisions. As in Zwiebel (1996), managers derive utility from

investing and retaining control, a combination that can lead to overinvestment. In choosing debt,

managers face the traditional trade-o↵ between tax benefits and bankruptcy costs. However, they

also confront a new trade-o↵: higher debt lessens the opportunity to invest in the future but at the

same time prevents control challenges by outsiders. An important result of the calibration is that

managers choose relatively low debt levels, which are easy to reconcile with low leverage observed

in practice.

Carlson and Lazrak (2010) incorporate managerial preferences using a di↵erent economic mech-

anism. In their model, managers are risk-averse and have a linear compensation contract, with a

fixed salary component and an equity stake. Their paper is based on the trade-o↵ version of Merton

(1974) with finite debt maturity, where managers choose both capital structure and an asset volatil-

ity parameter. E↵ectively, they choose the distribution of the final-period asset value. In relation

to capital structure, managers choose higher leverage (closer to the first-best case, where managers

61
maximize shareholders value) when the fixed component of compensation is low and interests of

managers and shareholders are closer aligned.

Behavioral approach. Most dynamic contingent-claim models embed standard rationality as-

sumptions for both decision-makers and investors. Several recent papers have started exploring the

impact of deviations from agents’ rationality. For example, Hackbarth (2009, 2010) build a capital

structure model with investment, in which the market holds rational expectations, but the manager

is both “optimistic” and “overconfident” in a sense that the manager believes the expected growth

rate to be higher than the true one (optimism) and the cash flow volatility to be lower than the

true one (overconfidence). Intuitively, in the presence of investment opportunities and constrained

internal funding, managers end up relying on more debt issues, because debt has a lower sensitivity

to a di↵erence in opinions between the manager and the market. Managers also find it optimal to

refinance more frequently.

Mergers, acquisitions, and corporate control. All the models we have discussed in detail

here fix the organizational structure of the firm. An important topic that has only recently started

being analyzed in dynamic contingent-claim models is the market for corporate control. For exam-

ple, Lambrecht (2004) studies the timing and terms of mergers driven by economies of scale. The

potential merger surplus is easy to value using the contingent claims approach. The model finds

the endogenous timing of the merger as a function of firms’ bargaining position, which determines

the surplus. Lambrecht also shows that firms with market power optimally speed up mergers and

that the equilibrium timing of hostile takeovers is inefficient. In a related paper, Lambrecht and

Myers (2007) consider merger activities in a declining industry, in which managers are reluctant to

liquidate optimally because they do not internalize the opportunity cost of assets in place. Using

the contingent claims approach, they consider implications for a number of scenarios, such as hostile

takeovers and golden parachutes.

A paper by Hackbarth and Morellec (2008) studies strategic takeover decisions, in which a pre-

assigned acquirer has assets in place and a takeover option that is modeled similar to a standard

62
real option. The trade-o↵ that the acquirer faces is the improvement in the value of the target (for

example, because of operational improvements) versus the fixed cost of the takeover. The solution

of the model takes a recognizable pattern: the optimal timing of the takeover is determined by a

threshold, which is a function of the ratio of cash flows of the acquirer and the target. The paper

also empirically explores an interesting implication of the model concerning the evolution of betas

of the acquirer. Because the acquirer has a risky takeover option before the acquisition, its beta is

higher. The takeover option exercise triggers a reduction in the acquirer’s beta.

Morellec and Zhdanov (2008) consider the impact of debt financing on the outcome of takeover

contests. Firms with lower leverage are more likely to enter takeover contests and become an

acquirer, because debt financing imposes a financing constraint on the firm and decreases the

maximum price the firm is ready to pay for a target. The model predicts that the leverage of a

winning bidder is lower than the industry average, consistent with empirical evidence, and that

acquirers should lever up after the takeover.

Capital structure and industry competition. Capital structure can be used strategically

by competitors, an issue studied originally in a static setup by, for example, Brander and Lewis

(1986). Incorporating strategic competition in a dynamic setup enriches the options available to

firms and the outcomes of their actions. For example, Lambrecht (2001) considers a duopolistic

industry, in which both firms have pre-determined debt outstanding and thus have a real option to

default and exit the industry. The remaining firm enjoys higher monopoly profits. The paper shows

that financial vulnerability of the incumbent induces earlier entry of the rival and subsequent earlier

default exit by the incumbent. Another approach is to study the interaction of capital structure and

industry competition in a perfectly competitive industry. For example, Miao (2005) incorporates

a capital structure trade-o↵ in the model by Dixit (1989), in which firms optimally enter and exit

in a perfectly competitive industry. Calibrations suggest that the stationary distribution of firms

and firm’s survival probabilities are influenced by capital structure decisions.

63
Information asymmetry. The classical study of Myers and Majluf (1984) applied the lemons

problem to external financing decisions of firms. In their equilibrium, managers of good firms may

choose to bypass positive NPV projects, because they do not want to be pooled with bad firms.

Conditional on issuing external financing, debt financing is preferred to equity issuance, because

debt is a claim with lower sensitivity and thus less dilution for knowledgeable insiders. Morellec

and Schürho↵ (2011) consider a real options problem where insiders are better informed about

the growth option. In equilibrium, good firms separate by deciding to invest too early to prevent

mimicking by bad firms. Good firms also issue debt as an additional separating tool. As a result,

these considerations lead to a substantial erosion in the value of waiting.

Strebulaev, Zhu, and Zryomov (2012) consider a dynamic version of Myers and Majluf (1984),

where, as in the original paper, insiders have both assets in place and a growth option and are

better informed about the value of assets in place than about the growth option. The market

learns the type of the firm over time even if no investment is observed. Low quality firms face the

trade-o↵ between investing now and waiting to pool with high quality firms. In equilibrium, both

pooling and separation can occur, as in the static model, but comparative statics shows that the

possibility of waiting changes the quantitative nature of equilibrium. For example, the probability

of pooling can be substantially lower, because a low type firm finds waiting too expensive. Also,

the model shows substantial delays in investment that can result in the presence of information

asymmetry.

3 Discrete-Time Investment Models

Discrete-time dynamic investment models are also widely used as a backbone for understanding

dynamic financing decisions.21 As in the case of contingent claims models, discrete time investment

models contain three building blocks: exogenous stochastic state variables, an objective function,

and a set of endogenous state variables that can be changed via a set of control variables. The
21
Early papers that develop this class of models include Lucas and Prescott (1971), which is a discrete-time model
and Hayashi (1982), which is a deterministic continuous-time model. Although this section emphasizes the case of
discrete time models, it is, of course, straightforward to write down continuous-time versions of these models (Abel
and Eberly 1994; Hennessy 2004; Hennessy, Levy, and Whited 2007).

64
exogenous state variables are typically shocks to firm profits. As such, they can be interpreted either

as demand shocks or productivity shocks. It is also possible to add further shocks, for example to

production costs (Riddick and Whited 2009) or to financing costs (Jermann and Quadrini 2012).

The objective function is typically to maximize shareholder value, which is the expected present

value of cash flows to shareholders. Once again, it is possible have di↵erent objectives, such as

maximizing CEO utility. The endogenous state variables in dynamic investment models always

include the current capital stock. They can also include other variables such as labor, the stock of

liquid assets, or debt. The paths of these endogenous state variables can then be altered via choice

variables such as next period’s capital, labor, liquid assets, or debt.

As we detail below, these models are more restrictive on some dimensions than are contingent

claims models, and they are more flexible on others. On the one hand, contingent claims models

allow for the pricing of debt and equity claims, whereas investment-based capital structure models

typically do not. On the other hand, investment-based models allow for a richer set of financing

choices. For example, whereas in contingent claims models the firm finances with either debt or

equity, in investment-based models, the firm can finance with debt, cash, equity infusions, and

dividend cuts. A more important advantage is that investment is chosen endogenously along with

financing variables so that it is possible to study the interaction between real and financial decisions.

Although it is also possible to study this interaction in real-options models, as in Morellec (2004),

the class of models outlined below allows for a variety of di↵erent investment policies, as opposed

to the lumpy, discrete projects that are central to real-options models.

These models can shed light on many interesting questions. For example, they have been used

extensively to understand the e↵ects of financial constraints on investment. They have also been

useful for understanding corporate cash holding decisions. Finally, the most prominent example

has been corporate capital structure, and in particular, the nature of the interaction between real

investment and debt-equity choices.

Dynamic investment models are typically cast as infinite horizon problems. At first, the task

of solving such models appears daunting, in that it appears to be necessary to choose an infinite

65
sequence of controls. However, these problems can be greatly simplified by assuming that the

exogenous state variables follow Markov processes. Loosely speaking, if a variable follows a Markov

process, then its expected value given its entire previous history equals its expected value just

given its previous period’s value. In other words, the current value of the stochastic state variable

captures its entire history. In this case, it is usually possible to characterize the model solution

using just two functions. The first is the value function, which is a rule that specifies firm value

as a function of the current exogenous and endogenous state variables. The second is the policy

function, which is a rule that specifies next period’s controls as a function of the current exogenous

and endogenous state variables.

This section is structured in the same way as Section 2 in that we begin with a simple model

and then add complexity. Thus, this section starts with the simplest pure investment model, which

contains no meaningful financing decisions. By building up the analytical framework and highlight-

ing the main intuition behind this type of model, we provide a solid foundation for understanding

the financing models that follow. To formulate these financing models, we layer on top of this basic

dynamic investment framework di↵erent financing features, with each new version of the model

slightly more complex than the previous. We then relate each version of the model to specific

studies in the literature.

With this general overview in mind, we now turn to the development of specific models.

3.1 Basic Model

This basic model is a partial equilibrium model cast in discrete time, with an infinite [Link]

is the only fixed factor of production, and risk-neutral managers, acting on the behalf of sharehold-

ers, choose the capital stock each period to maximize the value of the firm, which is the expected

present value of the stream of future cash flows to (or from) shareholders. These cash flows, which

we denote as e(kt , It , zt ) can be expressed via a sources and uses of funds identity:

e(kt , It , zt ) ⌘ ⇡(kt , zt ) (It , kt ) It . (3.1)

66
Here, kt is the beginning-of-period capital stock; It is investment in capital; ⇡(kt , zt ) is a profit

function with ⇡z > 0, ⇡k > 0, and ⇡kk  0; zt is an exogenous shock to the profit function that is

observed by the manager at time t and that follows a Markov process; and (It , kt ) is an investment

adjustment cost function that is increasing in It and weakly decreasing in kt , the idea being that

investment is less disruptive for larger firms. It is worth noting that in contrast to the real options

models described in Section 2.2, It can be either positive or negative. Disinvestment is costly

because of the adjustment cost function, but it is not completely irreversible.

Several remarks are in order. First, for simplicity, the price of capital is normalized to unity.

Second, expressing profits (not production) as a function only of capital is not restrictive in the

sense that any variable factors of production, such as labor, have already been maximized out of the

problem. Third, the adjustment cost function (·) represents any reduction in profits that comes

from the disruption of operations that can accompany the installation of capital. Thus, equation

(3.1) says that cash flows to shareholders, e(kt , It , zt ), are simply profits minus all expenditures

associated with investment in capital.

In this model, e(kt , It , zt ) can be either positive or negative. If positive, it represents distri-

butions of internal cash flows to shareholders, and if negative, it represents infusions of cash from

shareholders into the firm. Thus, in this model internal funds and external financing from sharehold-

ers are equally costly, the Modigilani-Miller theorem holds, and financing is trivial. Nonetheless, it

is worth spending some time on this simple model because a great deal of the intuition behind the

more complicated models that follow revolves around investment behavior.

With all of these ingredients we can express the value of the firm at time t as:

2 3
1 ✓
X ◆j
1
Vt = max Et 4 e(kt , It , zt )5 , (3.2)
kt+j |j=1,...,1 1+r
j=0

where Et is an expectations operator conditional on information known at time t, and r is the

constant risk-free interest rate.

67
3.1.1 First-Order Conditions

This model has no analytical solution. However, one can characterize the solution via an exam-

ination of the first-order conditions, and doing so helps lend intuition to the solution. The firm

maximizes (3.2) subject to the following capital stock accounting identity:

kt+1 = (1 )kt + It , (3.3)

where is the assumed constant rate of capital depreciation.

An informal way to obtain the first-order condition is simply to form the Lagrangian:

2 3
1 ✓
X ◆j ⇣ ⌘
1
Et 4 e(kt , It , zt ) t (kt+1 kt (1 ) It ) 5 , (3.4)
1+r
j=0

in which t is the Lagrange multiplier on the constraint (3.3). Then di↵erentiating (3.4) with

respect to It yields:

1+ I (It , kt ) = t. (3.5)

Because t is a Lagrange multiplier, it represents the shadow value of capital, so this first-order

condition says that at an optimum the shadow value of capital equals its marginal cost, which has

two components. The first is the price of capital, and the second is the marginal adjustment cost.

It is possible to obtain an intuitive expression for the shadow value of capital, t, from the

Euler equation, which we obtain informally by taking the first-order conditions of the problem of

maximizing (3.4) with respect to kt+1 :


1 ⇣ ⌘
Et ⇡k (kt+1 , zt+1 ) k (It+1 , kt+1 ) + (1 ) t+1 = t. (3.6)
1+r

This expression says that the shadow value of capital today equals its discounted expected value

tomorrow plus any marginal profit flows that today’s capital generate. To solve for t, we substitute

(3.6) into itself recursively and use the law of iterated expectations to obtain:

68
 2 3
1 ✓
X ◆j ⇣ ⌘
1
t = Et 4 (1 )j 1
⇡k (kt+j , zt+j ) k (It+j , kt+j )
5. (3.7)
1+r
j=1

Thus, equation (3.7) shows that the shadow value of capital is the expected stream of future

marginal benefits from using the capital. These benefits include both the marginal additions to

profit (⇡k (kt , zt )) and reductions in installation costs ( k (kt , It )). The quantity t is usually called

“marginal q” (Hayashi 1982).

It is worth noting that if one uses quadratic investment adjustment costs, then the first-order

condition (3.5) turns into a linear relationship between investment and marginal q, which has been

the formal basis for countless published investment regressions.

Further intuition can be gleaned from substituting (3.5) into (3.6) to obtain:


1 ⇣ ⌘
Et ⇡k (kt+1 , zt+1 ) k (kt+1 , It+1 ) + (1 )(1 + I (It+1 , kt+1 )) =1+ I (It , kt ) (3.8)
1+r

This optimality condition can be understood in the form of a perturbation argument. If the

firm is making an optimal choice, it must be indi↵erent on the margin between installing a unit

of capital today and waiting to install that unit of capital tomorrow. Thus, if the firm installs

capital today, it incurs marginal purchasing and adjustment costs, which can be seen on the right

hand side of (3.8). If the firm waits until tomorrow, we can see from examining (3.8) that it

again incurs the same marginal purchasing and adjustment costs, albeit discounted by both the

interest rate and the depreciation rate. However, it also incurs the opportunity cost of waiting to

invest until tomorrow—the foregone marginal product of capital, which is captured by the term

⇡k (kt+1 , zt+1 ) k (kt+1 , It+1 ). As we explain later, this investment Euler equation has been used

to study the e↵ects of finance constraints on investment (Whited 1992; Bond and Meghir 1994).

3.1.2 Numerical Solution

Most recent studies based on variants of this simple model have used numerical model solutions

to characterize both financing and investment decisions. Although it is beyond the scope of this

article to provide a detailed explanation of the relevant numerical methods, it is nonetheless useful

69
to provide an intuitive outline that emphasizes implementation rather than generality. An excellent

treatment of numerical methods can be found in Miranda and Fackler (2002), and proofs of the

existence of solutions to this class of models can be found in Stokey, Lucas, and Prescott (1989).

As a first step in outlining an algorithm for a numerical solution, we drop the time subscripts to

simplify notation. Thus, for any generic variable x, let xt be denoted as x, and let xt+1 be denoted

as x0 .

Next, we assume that the shock z follows a specific Markov process—a first-order autoregressive

process, AR(1), in logs:

ln(z 0 ) = ⇢ ln(z) + "0 . (3.9)

Here, ⇢ is an autoregressive coefficient, and "0 is the error term in this AR(1) process. It has a

variance 2
". We assume that " has a truncated normal distribution, so that z is approximately

lognormally distributed. We also introduce notation for the Markov transition function for z, that

is, the conditional distribution function of z 0 given z. We denote the transition function for z as
R
g(z 0 | z) so that the expectations operator Et (·) can be written as (·)dg(z 0 | z).

We can then heuristically derive the Bellman equation corresponding to (3.2) by splitting (3.2)

into two parts: the term corresponding to time t (the first period’s net profit) and the rest of (3.2).

Substituting the capital stock accounting identity (3.3) into the sources and uses of funds identity

(3.1), we obtain:

⇢ Z
1
V (k, z) = max ⇡(k, z) (k 0 (1 )k, k) (k 0 (1 )k) + V (k 0 , z 0 )dg(z 0 | z) . (3.10)
0
k 1+r

The first three terms of (3.10) constitute the current period flow to shareholders: profits less

investment less adjustment costs. The last term is called the continuation value, that is, the

expectation of next period’s firm value, given that optimal policies will be chosen in the future.

The goal of a numerical solution is to obtain a mapping from (k, z) to V (k, z). The simplest,

possibly the slowest, but often the most robust, method for obtaining a solution is called discrete

70
state value function iteration. Here, we describe the basic algorithm. It starts with a choice of

grids for the state variables (k, z). For example, one might stipulate that the capital stock k can

only take three values, (49, 50, 51), and that the shock can only take on two values, (0.9, 1.1).

In practice, these grids are much larger. Further, it is important to center the capital stock grid

at the steady-state level of the capital stock, which is usually a good approximation to a model

solution corresponding to the mean level of the shock, z. This level is roughly the point at which

the marginal product of capital corresponding to ln(z) = 0 (approximately z = 1) equals the user

cost of capital, r+ . It is also important to have the densest part of the grid correspond to the parts

of the production function with the most curvature. The reason is that numerical solutions are

usually linear approximations to nonlinear functions, and line segments are good approximations

to functions with a great deal of curvature only over short intervals. One way to accomplish this

design is to let the capital stock grid be a multiplicative sequence so that k i+1 = k i /(1 ), in

which k i+1 and ki are adjacent values in the grid.

We need one more ingredient before we can explain the algorithm: the transition function for

z. If z follows an AR(1) in logs as in (3.9), then one can use a variety of methods (Tauchen 1986),

Tauchen and Hussey (1991) to construct a discrete state Markov chain that approximates (3.9).

One then ends up with a grid of possible z values and a transition function in the form of a Markov

transition matrix.

With all of these ingredients we can now describe value function iteration. As a first step, we

characterize the form of the value function. Let nz be the number of points in the shock grid and

nk be the number of points in the capital stock grid. The value function is then an nz ⇥ nk array

of numbers, with each number corresponding to the firm value that is associated with a specific

(k, z) pair.

The simplest form of this algorithm then proceeds as follows: First, guess a value function

V (k 0 , z 0 ); for example, an (nz ⇥ nk ) array of zeros suffices. Second, take the expectation of the

value function by multiplying it by the Markov transition matrix. Third, using this expected value

function, evaluate the right-hand side of (3.10) for all possible (k, k 0 , z) triples. Note that the profit

71
flow in (3.10) needs to be evaluated for every (k, k 0 , z) triple. We denote the result as Ṽ (k, k 0 , z).

Fourth, compute maxk0 Ṽ (k, k 0 , z). The result is a new V (k, z). Fifth, use this result as a new

candidate V (k 0 , z 0 ), and repeat until the value function converges. Once one has obtained a value

function, one then picks out the optimal policies that produce the value function. This mapping of

states, (k, z), to policies, k 0 , is called a policy function, which we denote by h(k, z).

Once one has obtained the policy function, one can generate a time series of optimal capital

stocks. As a first step, one simulates a series of z shocks. Then one uses the optimal policy function

to trace out the optimal capital stock path. For example, suppose the shock z can take on only

two values and that the Markov transition matrix is:

✓ ◆
0.6 0.4
.
0.5 0.5

Thus, if one is in the first state, the probability of remaining in the first state is 0.6, the probability

of transitioning to the second state is 0.4, and so on. To simulate the path of shocks, one generates

a random uniform variable. Suppose the shock is currently in the first state. If the random uniform

number is greater than 0.6, the shock moves to the second state, and otherwise the shock stays in

the first state. This algorithm is straightforward to extend to multiple shocks.

3.1.3 Model Intuition

To illustrate the model intuition, we examine a special case of this basic model in which ⇡(k, z) =

zk ✓ , 0 < ✓ < 1. We consider three forms of the adjustment cost function: no adjustment costs,

convex adjustment costs, and fixed adjustment costs.

No Adjustment Costs

With no adjustment costs, the model has an analytical solution for the policy function, k 0 =

h(k, z). To derive the policy function, we note that without adjustment costs the Euler equation

given by (3.8) simplifies to:

Z
✓z 0 k 0✓ 1
g(z 0 | z) = r + (3.11)

72
Given the assumption that the shock z follows (3.9), we use the properties of a lognormal distri-

bution to write (3.11) as:

✓ ◆
1
exp ⇢ ln(z) + 2
z ✓k 0✓ 1
=r+ (3.12)
2

The exponential term on the left-hand side of (3.12) is the expected value of z 0 , given z, and given

that z has a lognormal distribution. Thus, tomorrow’s optimal capital stock, k 0 , is set so that the

expected marginal product of capital (the left-hand side of (3.12)) equals the user cost of capital

(the right-hand side). For the special case of no adjustment costs, the optimal choice of k 0 only

depends on the shock, z, and does not depend on k. This investment model without adjustment

costs can thus be broken down into a series of one-period problems. Each period, the firm observes

z and then chooses optimal k 0 , regardless of the current value of the capital stock. Later, we will see

that the introduction of investment or financing frictions introduces dependence between periods

and thus real dynamics.

An examination of (3.12) reveals five properties of these models that carry through to the more

complex models we examine later. First, capital (and thus investment) are naturally increasing in

the productivity/demand shock, z. Second, the average rate of investment is largely determined by

because capital depreciates by every period and because the expected value of z is approximately

1. Third, the smaller ✓, the less variable is investment. Intuitively, the more concave the production

function, the less the production function shifts in response to the multiplicative shock, z. In

economic terms, if the firm has sharply decreasing returns, then technology shocks have only a

modest e↵ect on the marginal profit of capital, so the firm responds less strongly to these shocks.

Thus, the optimal capital stock moves less than it would in the case of a less concave production

function. Fourth, the more variable z, the more variable investment. Fifth, if ⇢ equals zero, the

firm never changes its optimal capital stock because it knows that any productivity increases will

be fleeting. As ⇢ increases, the firm responds more strongly to shocks because it knows that the

marginal product of capital is likely to last. This last property would be impossible to obtain from

a static model because in such a model the notion of serial correlation is nonexistent.

73
Convex and Fixed Adjustment Costs

Next, we consider the case of an adjustment cost function that encompasses both fixed and

convex adjustment costs, so that (I, k) = 2

0I 2k + 1 kII6=0 . The first term is the convex

cost term, and the second term is the fixed cost term, which does not depend on the amount

of investment being undertaken. Note that adjustment costs are incurred for both positive and

negative investments. In this case, the model has no analytical solution, so we explain the intuition

behind the model for a special case in which 0 = 0.01, 1 = 0, r = 0.04, = 0.15, the production

function curvature, ✓ is 0.7, the serial correlation, ⇢, of the shock process (3.9) is 0.7, and the

standard deviation of the error term in (3.9), ", equals 0.15. We set the adjustment cost parameters

at low values for expositional purposes, so that the intuition behind the rest of the model is

transparent. The rest of the parameter choices are standard and close to those used in papers such

as Gomes (2001) and Hennessy and Whited (2005, 2007).

Figure 10 illustrates the intuition behind the role of adjustment costs by plotting the policy

function for the ratio of investment to the capital stock, (h(k, z) (1 )k)/k, for three models: one

with no adjustment costs, one with smooth adjustment costs, and another with fixed adjustment

costs. The policy function is evaluated at the steady-state capital stock and is expressed as a

function of the log shock, ln(z). The magnitudes can be interpreted approximately as the percent

increase or decrease in profits relative to k ✓ . We see that investment rises sharply when there are

no adjustment costs, with a two-standard deviation shock producing a rate of investment of over

100%. This behavior is damped substantially in the presence of smooth adjustment costs. Fixed

costs produce a large range of inaction for shocks ranging from four standard deviations below the

mean to almost two standard deviations above the mean. At that point the firm optimally invests

a great deal.

Figure 11 fleshes out this intuition by depicting comparative statics exercises in which we show

how various moments of the simulated data change with respect to the parameters. We construct

each panel of Figure 11 as follows. We start with the baseline parameterization, and then allow one

parameter to vary over a range of 20 values, which is shown on the horizontal axis of each figure.

74
For each of these 20 parameter values, we solve the model and generate simulated data. We then

calculate the moments of interest. The first moment is average investment, which is defined as

I/k. The next is the financing surplus/deficit (as a fraction of the capital stock), which is defined
⇣ ⌘.
as ⇡ (k, z) (k 0 (1 ) k, k) (k 0 (1 ) k) k and thus equals cash inflows minus cash

outflows. External financing equals the absolute value of the surplus/deficit variable, when it is

negative, and zero otherwise. It too is scaled by the capital stock.

The first panel shows that production function curvature has a strong impact on investment

and financing policies. In particular, when ✓ is small and the firm has sharply decreasing returns

to scale, it does not respond strongly to shocks, and its optimal investment rarely requires external

financing. Indeed, it invests exactly enough to replace depreciated capital 45% of the time, and this

amount of investment is easily met by internally generated funds. This behavior occurs because

the e↵ect of the shock, z dies out fairly rapidly: after t periods, the e↵ect is ⇢t . Therefore, with ✓

small, the marginal product of capital does not move much with z, so the firm does not respond

strongly, if at all, to small movements in z. In contrast, when the firm has nearly constant returns

to scale, its investment equals capital depreciation only 3% of the time and is highly variable.

This behavior is the result of two model features. First, because the production function is nearly

linear, movements in z change expected marginal profitability a great deal. Second, the concavity

of the profit function means that the multiplicative shocks have an asymmetric e↵ect. Positive

shocks entail large shifts in the marginal product of capital and thus in the optimal capital stock,

and negative shocks entail smaller shifts. Thus, we see many instances in which the firm finds it

optimal to make big investments, which require external financing. Therefore, for high values of ✓,

on average the firm runs a financing deficit and uses a great deal of external financing. Mitigating

but not reversing these e↵ects is the fact that a higher level of ✓ implies higher average profits,

which in turn imply a lower need for external financing.

The second panel shows that the depreciation rate has a more modest e↵ect on investment and

financing. Average investment is slightly higher than the depreciation rate, and average external

financing rises with investment, but on average the firm generates more internal funds than are

75
needed for investment. The reason for the muted response is that although changes in the depre-

ciation rate do a↵ect the value of capital and the average rate of investment, these changes do not

a↵ect the sensitivity of the marginal product of capital to the shock z.

The third panel shows that the standard deviation of the shock process also matters a great

deal for average investment and financing in much the same way as does the curvature of the profit

function. In this case the intuition is similar. With highly volatile shocks, the firm occasionally

receives a great investment opportunity and requires outside financing, but the concavity of the

profit function implies that the same e↵ect does not occur for negative shocks, which entail no or

negative investment. Thus, investment and external financing increase with the standard deviation,

", and the financing deficit decreases.

The fourth panel appears to show that an increase in the serial correlation of the shock process,

⇢, has a similar but somewhat more muted e↵ect on average investment and financing. The intuition

is similar. With highly serially correlated shocks, when the firm receives a positive shock, it is more

likely to make a large investment than when it has i.i.d shocks because it expects the marginal

product of capital to last. Thus when ⇢ increases, the probability of large investments increases,

as does the need for financing.

The fifth and sixth panels show that convex and fixed investment adjustment costs have opposite

e↵ects on investment and financing. When convex costs increase, the optimal policy of the firm is to

smooth investment over time, and for very high costs, the firm simply invests to replace depreciated

capital so that the average rate of investment equals the depreciation rate. In this case the firm

never needs outside financing and always runs a financing surplus. In contrast, in the case of high

fixed costs, the firm invests in a lumpy fashion. Thus, the distribution of investment becomes highly

skewed, which mechanically raises the mean of investment. Also, on average the firm needs to raise

finance because the large bursts of investment are larger than internally generated funds.

Three key takeaways emerge from this exercise. First, because of a standard sources-and-

uses-of-funds identity, the productivity shock has a profound e↵ect on financing decisions via its

e↵ect on investment decisions. Second, technology and uncertainty matter greatly for optimal

76
financial policies because they dictate the variability and lumpiness of investment policy. Thus,

understanding basic investment dynamics is essential for understanding financing decisions. Third,

the level of external flow financing is much higher than what one observes in the data because the

marginal cost of external financing is the same as that of internal financing. The average amount

of financing raised by firms in Compustat is typically less than 10% of assets in any given year,

whereas in the simulated data, it is around 20% of assets.

3.2 Costly External Finance

We now introduce costly external financing into this basic model. The simplest formulation comes

from Gomes (2001), in which external finance takes the form of equity injections from current share-

holders. These injections carry a fixed and proportional cost, and are thus a more expensive source

of funds than internally generated cash flows. This structure follows Altinkilic and Hansen (2000),

which estimates that underwriting fees are characterized by both fixed and variable components.

The firm does not use debt financing, and the firm cannot retain earnings. If cash inflows exceed

optimally chosen cash outflows, the firm must pay the entire excess out to shareholders. On the

other hand, if outflows exceed inflows, then to fill the gap, the firm pays transactions costs. Define:

e(k, k 0 , z) ⌘ ⇡(k, z) (k 0 (1 )k, k) (k 0 (1 )k), (3.13)

that is, cash inflows (⇡(k, z)) minus cash outflows ( (k 0 (1 )k, k) (k 0 (1 )k)). First,

note that (3.13) is identical to (3.1), except with simpler notation. It is also important to note that

this sources and uses of funds identity implies that if one chooses capital, then one also implicitly

chooses e(·). It is impossible to choose both separately. Next, as in Gomes (2001), we define the

cost of external finance as:

⌘(e(k, k 0 , z)) ⌘ ⌘0 + ⌘1 e(k, k 0 , z) Ie<0 , (3.14)

in which I is an indicator function. The cost of external finance is assumed to be zero if the firm

is distributing funds to shareholder, but it is strictly positive if the firm is receiving injections of

77
funds from shareholders. In particular, the function contains a fixed component, ⌘0 , as well as a

linear component, ⌘1 .

With costly external finance the Bellman equation for this model is:

⇢ Z
0 0 1
V (k, z) = max e(k, k , z) + ⌘(e(k, k , z)) + V (k 0 , z 0 )dg(z 0 | z) (3.15)
k0 1+r

The model in Gomes (2001) is substantially richer than this simple partial equilibrium frame-

work. His model excludes capital adjustment costs but includes a fixed cost of production, and

he explicitly models both labor and capital choices. More importantly, his model is an industry

equilibrium model. A representative consumer consumes the firm’s output and supplies labor to

the firm. Then the real wage adjusts so that the supply of labor equals the demand for labor.

Firms enter an industry when they get a shock that is sufficiently high for them to break even.

Firms exit when the proceeds from selling capital are greater than the value of the firm as a going

concern. The result of this entry and exit is a steady state distribution of firms.

In this setting, it is possible to characterize four types of firms: those that exit, those that

do not need external finance, those that both need and choose to obtain external finance, and

those that would obtain external finance if it were costless but that choose not to do so when it

is costly. In general, firms with extremely low productivity exit. For those firms that do not exit,

their financing status depends on both their size and productivity, with large, lower productivity

firms not needing finance, intermediate sized, intermediate productivity firms being constrained,

and small, extremely productive firms getting finance.

Although the model in Gomes (2001) is rich enough to explore a wide variety of questions,

one particularly important part of the paper is an attempt to answer the question of whether the

sensitivity of investment to cash flow is a good measure of finance constraints. This question is

from Fazzari, Hubbard, and Petersen (1988), who argue that the investment of constrained firms

should be highly correlated with movements in internal funds. They then test this proposition by

running panel regressions of the ratio of investment to capital on the ratio of cash flow to capital

and Tobin’s q, which is the ratio of the market value of the capital stock to the replacement value.

78
In our notation, this regression can be expressed as:

I V (k, k 0 , z) zk ✓
= b0 + b1 + b2 + u, (3.16)
k k k

in which b0 , b1 , and b2 are regression coefficients and u is a regression disturbance. Fazzari, Hubbard,

and Petersen (1988) find that the sensitivity of investment to cash flow (b2 ) is highest for groups

of firms they categorize as constrained.

The model in Gomes (2001) is ideal for evaluating whether investment regressions are useful

for detecting finance constraints for two reasons. It is possible to run this precise regression on

simulated data, and it is possible to know exactly which simulated firms are constrained and which

are not. The main result in Gomes (2001) is that investment-cash flow sensitivity is neither sufficient

nor necessary for the existence of finance constraints.22 The intuition is that in a dynamic model

with entry, exit, and costly external finance, the first-order conditions of the model are only weakly

approximated by a regression of investment on Tobin’s q and cash flow. If one could estimate the

first-order conditions of this model, one might be able to identify constrained firms, but using an

approximation falls short in this dimension.

We now depart from the analysis in Gomes (2001) to show that this point can also be seen in

a slightly di↵erent way in the simpler framework here. We do so by examining what happens to

investment-cash flow sensitivity when we change the parameters that govern the cost of external

finance, ⌘0 and ⌘1 . To this end we initially set ⌘0 = 0.08 and ⌘1 = 0.028, as in Gomes (2001).

The top two panels of Figure 12 presents the results from this comparative statics exercise.

Each panel is constructed exactly as for Figure 11, with the exception of the top two panels. In

these panels, which explore the costs of external finance, when we vary ⌘0 , we set ⌘1 = 0, and vice

versa. On the vertical axis of each panel is the regression coefficient b2 .

Several results are of note. First, in both panels, even when the cost of external finance is zero,

the cash flow coefficient is positive. The reason is that with decreasing returns to scale, Tobin’s

q is not a sufficient statistic for investment (Hayashi 1982). Thus, as pointed out in Cooper and

Ejarque (2003), because cash flow contains incremental information about investment opportunities,
22
See also Alti (2003), which examines related partial equilibrium model with learning.

79
it enters into the regression. Second, the cash flow coefficient is decreasing in the cost of external

finance, not increasing, as predicted by Fazzari, Hubbard, and Petersen (1988). The reason, as

pointed out in Moyen (2004) and Hennessy and Whited (2007), is that when the firm cannot freely

access external finance, it cannot respond strongly to profit shocks by investing. Thus, when the

cost of external finance rises, investment becomes less highly correlated with the information about

investment opportunities in cash flow, and therefore with cash flow itself. Note that this intuition

is dramatically di↵erent from the simple, and static, argument in Fazzari, Hubbard, and Petersen

(1988) that revolves around increasing the firm’s liquid resources in a one-time, lump-sum fashion.

The realistic and nuanced intuition from a dynamic model is thus difficult to obtain from a static

model. Third, the coefficients from the simulated data are approximately 10 times as large as the

coefficients one finds in real data. This result, which is also in Gomes (2001), is to be expected

in this simple model for two reasons. It contains minimal investment frictions, and with only one

choice variable, no other choices break the strong correlation between profit shocks, investment,

and cash flow.

The next six panels of Figure 12 make a di↵erent point. Financial frictions are by no means

the only firm characteristics that matter for investment-cash flow sensitivity. As in Figure 11, we

plot the cash flow coefficient as a function of the six model parameters that describe technology

and uncertainty. One guiding piece of intuition informs all of these comparative statics exercises:

any parameter that makes investment respond more strongly to z increases investment-cash flow

sensitivity, and vice versa.

The third panel shows that cash flow sensitivity increases as the production function becomes

closer to linear, i.e., as ✓ approaches one. A nearly linear production function implies large swings

in the optimal capital stock when z fluctuates. Therefore, investment responds more strongly to z,

which leads to increased cash flow sensitivity. The fourth panel shows that changes in the depre-

ciation rate have little to do with the response of investment to shocks, because the depreciation

rate helps determine the average level of investment—not investment dynamics.

The fifth and sixth panels show how investment-cash flow sensitivity changes with the param-

80
eters that govern the shock process (3.9). First, we consider the standard deviation of the profit

shock. With a low variance shock, investment varies little with movement in the shock, and with

a high variance shock investment responds strongly to z. Recall that this pattern stems from the

inherently temporary nature of the z shocks, so that the firm optimally ignores small shocks. There-

fore, cash flow sensitivity increases with the shock standard deviation, although most of this e↵ect

occurs at low shock standard deviations. Now we consider the shock serial correlation. Equation

(3.12) implies that investment responds to z more when the shock process is more highly serially

correlated, so cash flow sensitivity increases with ⇢.

Finally, the seventh and eights panels show that increasing either smooth or fixed capital stock

adjustment frictions almost always lowers investment cash flow sensitivity by making investment

insensitive to shocks. In addition, as fixed costs become more important, investment responds

negatively to movements in cash flow. This result comes from the extreme lumpiness observed in

investment as fixed costs rise. At points where the firm invests, investment opportunities plummet

because of decreasing returns. Because in the regression (3.16) both Tobin’s q and cash flow pick

up movements in investment opportunities, investment-cash flow sensitivity is negative.

Two main takeaways emerge from this figure. First, financial frictions actually lower rather than

increase investment cash flow sensitivity. Second, financial frictions are but one of many di↵erent

factors that a↵ect this correlation, so that it is difficult to attribute di↵erential investment-cash

flow sensitivity to di↵erences in financial frictions.

3.3 Cash

We now add another layer to the model by introducing a new source of financing, the stock of

liquid assets, which we denote as p, with p 0. We assume that cash balances are held as a one

period discount bond. To model the cost of holding cash, we also add corporate taxation, so that

both profits and interest on cash balances are taxed at a rate ⌧ . We also assume that depreciation

expense it tax deductible. In this case the sources and uses of funds identity becomes:

81
 e(k, k 0 , p, p0 , z) ⌘ (1 ⌧ )⇡(k, z) + ⌧ k

(k 0 (1 )k, k) (k 0 (1 )k)

p0 /(1 + r(1 ⌧ )) + p. (3.17)

The first two terms, ((1 ⌧ )⇡(k, z) + ⌧ k), are simply after tax cash flows, with the term ⌧ k

representing the extra cash flow the firm receives because of the depreciation deduction. The next

two terms, ( (k 0 (1 )k, k) (k 0 (1 )k)), represent investment in physical assets, and the

last two terms, ( p0 /(1 + r(1 ⌧ )) + p), represent investment in liquid assets.

The Bellman equation looks much as it did in the previous section:

⇢ Z
1
V (k, p, z) = max e(k, k 0 , p, p0 , z) + ⌘(e(k, k 0 , p, p0 , z)) + V (k 0 , p0 , z 0 )dg(z 0 | z) . (3.18)
0 0
k ,p 1+r

In comparing this model of cash holding with a simple investment model, the main di↵erences

are in the sources and uses of funds identity and in the existence of costly external finance. This

model is a slightly simpler version of those studied in Riddick and Whited (2009), and it is closely

related to the continuous time model studied in Bolton, Chen, and Wang (2011). The numerical

solution procedure is similar to that described in Section 3.1.2, except that one must optimize over

a pair of choice variables, (k 0 , p0 ), instead of one, k 0 . Therefore, if np is the number of grid points

for the cash state variable, the value and policy functions have dimensions (nz ⇥ nk ⇥ np ), and one

must optimize over (nk ⇥ np ) pairs of state variables. The increased computational burden from

having just one extra state variable is nontrivial, and models quickly become intractable when the

number of state variables exceeds five or six.

Although this model also has no analytical solution, it is instructive to write down the first-order

conditions. The first-order conditions for optimal cash balances can be obtained by di↵erentiating

the Bellman equation (3.18):

Z
1 + r (1 ⌧ )
1 + ⌘1 Ie<0 = Vp k 0 , p0 , z 0 + ⇣ 0 dg z 0 , z , (3.19)
1+r

82
in which ⇣ is a Lagrange multiplier on the constraint that the state variable p be positive. The

right-hand side of (3.19) represents the expected discounted shadow value of cash balances, and

the left-hand side represents the marginal cost of external equity finance. Thus at an optimum the

firm equates the shadow value of cash with the opportunity cost of using cash, which is the use of

external finance.

Equation (3.19) highlights several pieces of intuition behind this model. First, cash derives value

because it is an alternative to costly external finance. We call this benefit the “financial flexibility”

benefit. Indeed, in this model the firm optimally holds no cash if external finance is free. Second,

the term 1 ⌧ on the right-hand side of (3.19) indicates that because interest is taxed, holding cash

is costly for the firm. Thus, an optimal interior cash policy balances the flexibility benefit with the

tax cost. In general, this tax cost can represent a variety of costs from holding too much cash, such

as agency costs.

To glean further intuition, we again use a numerical solution to the model to examine the policy

function and to conduct comparative statics exercises with respect to the various model parameters.

For these exercises, we use the same parameterization as the one used to construct Figure 11, except

that we set ⌘0 = 0, ⌘1 = 0.07, and ⌧ = 0.2.

Figure 13 plots optimal investment, cash, and distributions to shareholders as a function of the

productivity shock, z. All three variables are scaled by the capital stock, and as in Figure 10, the

policy functions are evaluated at the steady state capital stock. As in Figure 10 optimal investment

rises with the productivity shock, and optimal cash balances fall with the productivity shock. Thus,

we see that financial and physical assets are substitutes. For high levels of the productivity shock,

the firm draws its cash balances down to zero because capital becomes highly productive. For high

values of z, distributions to shareholders become negative, that is, the firm raises external equity

finance. The e↵ect of costly external finance is also apparent in the muted response of investment to

the productivity shock once optimal cash balances are zero and the firm needs to fund investment

with external finance.

Next we turn to comparative statics exercises. We examine how the ratio of cash to book

83
assets, p/(p + k) responds to each model parameter. Once again, when we perform a comparative

statics exercise for ⌘0 , we set ⌘1 = 0. Figure 14 presents these exercises. The central intuition

behind the results in this figure is that any model feature that raises the probability of needing

external finance in the future also raises the shadow value of cash and thus optimal cash levels.

Thus, the intuition from the simplest investment model plays a central role in understanding cash

accumulation because optimal investment dynamics determine the probability of needing outside

finance.

Of course, a necessary condition for holding cash is that external finance be costly. This feature

of the model can be seen in the top two panels of Figure 14. In both cases, optimal cash holding

is zero when external finance is costless. However, cash increases sharply for low costs of external

finance but then flattens out. In other words, firms do not change their cash balances sharply if

they face more costly external finance.

The next six panels show that other features of the firm do matter a great deal for average cash

balances. The second row of panels shows that profit function curvature, ✓, has a nonmonotonic

e↵ect on cash balances. Two opposing forces are at work. As the production function becomes

more linear, the firm tends to invest in a lumpier fashion. These large investments are likely to

require costly outside finance, so the firm holds precautionary cash balances. On the other hand,

as the production function becomes more linear, the firm becomes more profitable, and it has more

internally generated cash flows to finance investment. The first force is stronger when ✓ is low,

but the second force is stronger when ✓ is high. The second row of panels also shows that optimal

cash balances increase with the depreciation rate. The intuition is simple. Firms whose capital

depreciates rapidly are more likely to need external financing. Further, a higher depreciation rate

makes capital less valuable relative to the other asset held by the firm—cash.

The third row of panels shows that the variance and serial correlation of the profit shock

process have strong e↵ects on cash balances, once again, because of investment dynamics. Both

high variance and highly serially correlated shocks increase the probability of needing external

funds for investment, so average cash balances are increasing in both of these parameters. Not

84
surprisingly, final row of panels again shows that investment dynamics matter. When the firm has

high smooth adjustment costs, it invests only to replace depreciated capital and thus never expects

to need outside funding. It optimally holds no cash. When the firm has high fixed adjustment

costs, it invests in a lumpy investment, expects to need outside financing, and holds a great deal

of cash.

This model has some limitations. First, because the firm has no debt, it never su↵ers financial

distress and thus never needs to hold cash to avoid financial distress. We relax this restriction

below. Second, in reality a great deal of the demand for cash derives from the role of cash as a

form of working capital. For example, it is useful to have cash on hand in order to pay suppliers if

sales revenues fall short of expectations. This simple model abstracts from this source of liquidity

demand. Third, there is no strategic demand for cash that might arise if a firm wants to use cash to

purchase a patented technology. These last two sources of the demand for cash are an interesting

avenue for future quantitative research.

3.4 Risk-Free Debt

The next layer we add to this basic structure is risk-free debt as in Hennessy and Whited (2005) and

DeAngelo, DeAngelo, and Whited (2011). We drop the state variable for cash, p, and we denote

the stock of debt as b, where debt is held as a one-period discount bond. In these two models, the

state variable b takes both positive and negative values, with a positive value denoting debt, and a

negative value denoting cash. One can therefore interpret cash as negative debt—a model feature

that we relax later. In order for the debt to be risk free, the lender requires that the firm be able

to repay the debt by selling capital or by using its current after-tax cash flows, even in the worst

state of the world. For illustrative purposes, we assume that only a fraction of the capital stock,

s = 0.5 can be liquidated to repay capital. The collateral constraint can thus be expressed as:

b0  (1 ⌧ )zk 0✓ + ⌧ k 0 + sk 0 , (3.20)

in which z is the lowest possible value that the shock z can attain. In this model, the sources and

85
uses of funds identity looks almost identical to its counterpart in the cash model from section 3.3:

b0
e(k, k 0 , b, b0 , z) ⌘ (1 ⌧ )⇡(k, z) (k 0 (1 )k, k) (k 0 (1 )k) + b. (3.21)
1 + r(1 ⌧)

The only substantive di↵erence is one of interpretation. In the cash model from Section 3.3,

e(k, k 0 , p, p0 , z) < 0 means that the firm is raising generic external finance: there is no distinction

between debt and equity. Here, e(k, k 0 , p, p0 , z) < 0 means specifically that external equity finance

can only be attained at a premium. In this case, the Bellman equation becomes:
⇢ Z
1
V (k, b, z) = max e(k, k 0 , b, b0 , z) + ⌘(e(k, k 0 , b, b0 , z)) + V (k 0 , b0 , z 0 )dg(z 0 | z) , (3.22)
0 0
k ,b 1+r

subject to the constraint given by (3.20):

This simple model is only slightly more complex than the model from Section 3.3, with the

main additions being a collateral constraint and a financing state variable, b, that can take both

positive and negative values.

The model closely resembles that in DeAngelo, DeAngelo, and Whited (2011). One important

di↵erence is that in DeAngelo, DeAngelo, and Whited (2011), debt is constrained to be less than

a fixed fraction of the steady state capital stock given by (3.12). Our model also resembles that in

Hennessy and Whited (2005), except that the latter is substantially richer in detail. In particular,

the model in Hennessy and Whited (2005) contains both corporate and personal taxation, as well as

a strictly convex corporate tax schedule. As is the case here, debt can be collateralized by profits,

but the rest of the collateral constraint is slightly more elaborate. Instead of assuming that a fixed

fraction of the capital stock can be used as collateral, Hennessy and Whited (2005) employ the

closely related assumption that the firm can sell up to its entire capital stock to repay debt, but

that the price is lower than 1. In other words, “fire sales” of capital occur at a discounted price

occur if the firm does not have enough internally generated cash flow to repay its one-period debt.

Although the model features risk-free debt, the fire sales capture the notion of financial distress.

The model also does not contain capital adjustment costs, a feature that simplifies the fire sales in

86
the model.

Nonetheless, the main intuition behind our simple model is largely the same as that behind

the models in Hennessy and Whited (2005) and DeAngelo, DeAngelo, and Whited (2011), and

the discussion that follows draws from both of these papers. To begin the development of this

intuition, we examine the first-order condition for optimal debt/cash accumulation, which we obtain

by di↵erentiating (3.22) with respect to b0 . As in Section 3.3, let ⌘0 = 0 and ⌘1 > 0. Also, let ⇣ be

the Lagrange multiplier on the constraint given by (3.20), then the first-order condition is:

Z
1 + r (1 ⌧ )
1 + ⌘1 Ie<0 = Vb k 0 , b0 , z 0 + ⇣ 0 dg z 0 , z , (3.23)
1+r

This first-order condition conveys much of the same intuition as the first-order condition for

optimal cash holding in Section 3.3. As seen on the left-hand side of (3.23), debt derives value

because external equity financing is costly. As seen on the right-hand side, debt also derives value

because of the tax benefit of debt.

Further intuition can be gleaned from using the envelope condition,23 which we can use to

rewrite (3.23) as:

Z
1
1 + ⌘1 Ie<0 + ⇣ = (1 + r (1 ⌧ )) 1 + ⌘1 Ie0 <0 + ⇣ 0 dg z 0 , z . (3.24)
1+r

To interpret this condition, we follow Hennessy and Whited (2005) by taking the current state

(k, b, z) as given, and then imagining that the optimal investment decision has already been made.

In this case, all of the terms in (3.21) are constant, except for the one involving b0 . Therefore, a

dollar increase in debt either increases distributions or decreases equity issuances. Further, (3.21)

implies a dollar increase in debt today leads to more debt repayment in the future. These two

simple points are crucial to understanding the intuition behind the model.

The left-hand side of (3.24) represents the marginal benefit of using an extra unit of debt. To

see this point, note that if the firm’s current state and optimal investment policy imply that it has
23
At an optimum, the derivative with respect to the first period profit flow must equal the derivative with respect
to the value function.

87
a financing deficit (i.e. e < 0), then an extra dollar of debt financing keeps the firm from having to

use costly external equity financing today, and the benefit of this extra dollar of debt is 1 + ⌘1 . On

the other hand, if the firm has a financing surplus, then using a dollar of debt financing means that

the firm can distribute an extra dollar to shareholders. Because we are not modeling distribution

taxes, the benefit of debt is simply 1.

The right-hand side of (3.24) represents the marginal cost of debt financing, which can be seen

to be the expected principal and interest on debt that must be repaid tomorrow. Two extra terms

add texture to this simple intuition. First, the term ⌘1 Ie0 <0 implies that the marginal cost of debt is

higher when the firm expects to have a financing deficit next period, (e0 < 0). In other words, raising

an extra dollar of debt today implies debt repayment tomorrow and therefore a higher likelihood

of needing external equity financing tomorrow. Second, the presence of the Lagrange multiplier

in (3.24) shows that the marginal cost of debt is also higher when the firm expects to bump up

against its collateral constraint next period, that is, exhaust its debt capacity. As explained in

DeAngelo, DeAngelo, and Whited (2011), the intuition is that choosing a high level of debt today

implies lowers financial flexibility in the future, so the cost of borrowing today includes the value

lost when a firm fails to preserve the option to borrow later. This marginal cost schedule is thus

increasing in b0 , because raising b0 increases the likelihood of resorting to positive equity issuance

next period and increases the option value of debt capacity preservation.

Further intuition can be gleaned from plotting the first-order condition (3.24). Figure 15,

adapted from Hennessy and Whited (2005), does so by plotting the marginal cost and marginal

benefit of debt. Along the horizontal axis is b0 . We do not distinguish between b0 > 0 and b0 < 0,

although points to the left are more likely to represent cash and points to the right are more likely

to represent debt. The upward sloping line is the marginal cost of debt, and the two horizontal lines

represent the two possible values of the marginal benefit of debt, with the solid portion representing

the entire marginal benefit schedule. Note that the solid portion jumps down at the point b0e=0 ,

which represents the (possibly suboptimal) level of b0 at which the financing deficit is equal to zero.

For points to the left of b0e=0 , the firm is not generating enough resources via debt and profits to

88
fund its optimal investment policy. It therefore must resort to external equity issuance, and the

marginal benefit of debt is therefore 1 + ⌘1 . At the point b0e=0 , the marginal benefit schedule jumps

down to 1 because for levels of debt beyond this point an extra unit of debt serves to increase

distributions to shareholders.

Optimal debt policy is then given by the point at which the marginal costs and benefits of debt

are equal. Figure 16 depicts a situation in which the firm has abundant sources of funds or few

optimal uses of funds, so that the level of debt that sets the financing surplus equal to zero is low.

The point b0e=0 can be low for two reasons. First, the firm might have low (or even negative) debt,

so that it requires few funds to repay debt in the future. Second, the firm might receive a low

productivity shock, which implies a low marginal product of capital, which in turn implies that it

is preferable to distribute profits to shareholders than invest them in the firm. In this case, the

optimal level of debt is L, and shareholder distributions are then the di↵erence between L and b0e=0 .

Figure 16 is an analogous figure that depicts optimal debt policy when the firm has limited

sources of funds but many optimal uses. In this case, b0e=0 is high either because the firm has

high debt going into the period, which requires repayment, or because the firm receives a high

productivity shock so that it is optimal to invest profits rather than distribute them. The firm’s

optimal debt level is given by H, and equity issuance is the di↵erence between b0e=0 and H.

These two figures highlight two important pieces of intuition. First, leverage is likely to be

positively serially correlated; that is, if one were to run an autoregression on leverage, one ought

to find a high coefficient on lagged leverage. Higher lagged debt causes the firm to occupy the high

portion of the marginal benefit schedule over a longer stretch. With higher lagged debt, more debt

must be issued this period before the marginal unit of debt serves to increase distributions rather

than to replace external equity. Second, highly liquid firms are likely to be debt conservative. In

this case debt issuance serves to finance higher distributions to shareholders, rather than replacing

costly external equity. Since high liquidity firms occupy the lower portion of the marginal benefit

schedule, debt issuance is less attractive.

To elaborate on the model intuition, we again plot the model policy function, which expresses

89
optimal investment, debt (or cash, if negative), and distributions as a function of the log produc-

tivity shock. As before, the policy function is evaluated at the steady state capital stock, and

all variables are scaled by the capital stock. As seen in Figure 17, investment rises with the pro-

ductivity shock. Interestingly, for low levels of the productivity shock, investment is financed by

dissaving (when the state variable b < 0) or by increases in debt (when b > 0). Distributions stay

roughly constant. It is only for extremely high productivity shocks that the firm finds it optimal to

tap costly external equity financing. In this figure we also plot the right-hand side of the collateral

constraint (3.20). As explained in DeAngelo, DeAngelo, and Whited (2011), the firm prefers to stay

below its debt capacity. The intuition is that costly external equity implies that the firm optimally

preserves debt capacity so that it can have sufficient slack to take advantage of future productivity

shocks.

Finally, we conduct comparative statics exercises, such as those in DeAngelo, DeAngelo, and

Whited (2011). For these exercises, which can be found in Figure 18, we use the parameterization

used to construct Figure 14, except that we set the risk free rate to 0.01 so that we end up with

realistic leverage ratios. The panels in this figure depict average leverage and average debt capacity

utilization as functions of the various model parameters. Debt capacity utilization is described as

the ratio of debt to the collateral constraint given by (3.20). We plot debt capacity utilization in

order to highlight the e↵ect of the Lagrange multiplier in (3.24) on optimal leverage policy. As

emphasized in DeAngelo, DeAngelo, and Whited (2011), this e↵ect can be quite strong.

The main piece of intuition behind this figure is that lower optimal debt accompanies any model

parameter that increases the probability of needing external equity finance (equivalently, of getting

too close to debt capacity) or that increases the cost of equity finance. This intuition is highlighted

in the first two panels, which depict leverage policy as a function of the parameters that govern

the cost of external finance. The striking result is that a higher cost of equity finance lowers

optimal leverage and decreases debt capacity utilization. This model prediction is exactly opposite

of what one would find in a typical static model, in which firms substitute away from high cost and

toward low cost financial instruments. The reason for the seemingly counterintuitive result in the

90
dynamic model is that these static models cannot embody the option value of maintaining financial

flexibility. In other words, in this sort of dynamic models, as equity financing becomes more costly,

the firm foregoes the tax benefits of debt in favor of “keeping its powder dry” by lowering leverage

in order to take advantage of possible future investment opportunities.

In general, this last e↵ect is striking. In the baseline simulation, the average ratio of debt to

debt capacity is only 0.63, because optimal policy only dictates exhausting debt capacity when

the firm encounters a series of high productivity shocks. In the simulation the firm rapidly pays

down the debt after it has funded its investment projects. In general, we only see high debt

capacity utilization when equity finance is “free” or when optimal investment policy rarely requires

outside funding. For example, firms with low volatility or low serial correlation shocks or firms

with smooth investment tend to take advantage of the tax benefit of debt and have high leverage

(full debt capacity utilization), whereas firms with lumpier or more volatile investment conserve

debt capacity and keep leverage optimally low.

We conclude this section by describing further results in the literature that have been produces

by this class of models. Hennessy and Whited (2005) contains two further important results.

First, the paper provides explanations for the stylized fact that leverage is negatively correlated

with lagged measures of profitability.24 This fact appears to stand in the face of static trade-o↵

theory, which predicts that highly profitable firms should lever up to shield their profits from taxes.

However, in a dynamic trade-o↵ model with endogenous investment, highly profitable firms are

likely to have low leverage because they are more likely to have financing surpluses. They therefore

simply do not need leverage to fund optimal investment. In addition, they are likely to resemble

the type of firm in Figure 15, whose marginal benefit of debt is low.

Second, Hennessy and Whited (2005) o↵ers an explanation for the stylized fact, documented

in Baker and Wurgler (2002) that leverage varies negatively with a variable called lagged weighted

q. This variable is the weighted average of the most recent values of a firm’s Tobin’s q, with

more weight given to those observations accompanied by security issuance. Baker and Wurgler
24
See, for example, Rajan and Zingales (1995).

91
(2002) interpret this finding as indicative of firms’ attempts to time the equity issuance market. In

contrast, the model in Hennessy and Whited (2005) shows that firms with high productivity shocks

simultaneously have high qs and finance high desired investment with equity. Increased investment

raises next period’s firm value. The debt to assets ratio then falls and stays low because of leverage

serial correlation. Therefore, in simulated data, one observes a negative correlation between current

low leverage and those values of Tobin’s q that were accompanied by security issuance.

3.5 Cash and Debt

One of the drawbacks of the model in Section 3.4 is that it can only really make predictions

about net debt because of the assumption that cash is negative debt. Therefore, the next layer of

complication that one can add to this class of models is a firm that can simultaneously hold both

cash, p, and debt, b. Gamba and Triantis (2008) study this model. The central insight is that

without any transactions costs associated with issuing debt, the firm never optimally holds both

debt and cash. However, as long as the firm incurs costs when it changes its level of debt, then it

can be optimal for the firm to have positive stocks of debt and cash simultaneously. The intuition

behind this insight is the same intuition behind the model from Section 3.3: firms hold cash to

avoid having to resort to costly external finance. The main di↵erence here is that external finance

can take the form either of debt or of equity.

To outline this model, we let q(b0 , b) represent the costs associated with debt issuance. In this

case, the sources and uses of funds identity can be written as:

e(k, k 0 , p, p0 , b, b0 , z) ⌘ (1 ⌧ )⇡(k, z) (k 0 (1 )k, k) (k 0 (1 )k)

b0 p0
+ b q(b0 , b) + p. (3.25)
1 + r(1 ⌧) 1 + r(1 ⌧)

Once again, the e↵ects of adding both cash and debt are seen primarily in the sources and uses

of funds identity (3.25). The rest of the dynamic maximization problem is largely similar, except

that the firm chooses three variables, cash, debt, and capital, instead of just two.

As in the case of the model in Hennessy and Whited (2005), the model in Gamba and Triantis

92
(2008) is substantially more complex. It too has a personal taxation and fire sales of assets when

the firm has insufficient internally generated funds to repay debt. The existence of these fire sales

increases the value of cash by making external finance in the form of debt potentially more costly.

The main result in Gamba and Triantis (2008) is that when the firm experiences low profitability,

if it wants to decrease its net debt position, it should increase its cash balance rather than paying

down debt. The reason is that the cash investment is costlessly reversible, whereas paying down

debt is not. Conversely, if the firm wished to increase its net debt position, it is usually optimal

for it to do so via manipulation of its cash levels. These results in turn imply that, ceteris paribus,

two firms with the same net debt level can have very di↵erent valuations, with the one with more

cash, and therefore more flexibility, usually being more valuable.

3.6 Risky Debt

One of the less attractive features of the model we have built thus far is that debt is risk-free.

Several papers in the literature have relaxed this assumption. In the finance literature Moyen

(2004), Hennessy and Whited (2007), and Titman and Tsyplakov (2007) have used what are called

“endogenous default” models to study and quantify, respectively, the e↵ects of financial frictions

on investment cash flow sensitivity, the cost of external finance, and the e↵ects of agency conflicts

on leverage dynamics.25

To outline this type of model, we revert back to the assumption that cash equals negative debt,

drop the state variable, p, and allow the state variable, b, to assume both positive and negative

values. We also drop the collateral constraint (3.20) so that the firm can take on as much debt as

it desires. The sources and uses of funds identity then looks very similar to (3.21), with two small

changes:

b0
e(k, k 0 , b, b0 , z) ⌘ (1 ⌧ )⇡(k, z) (k 0 (1 )k, k) (k 0 (1 )k) +
1 + r̃(z, k 0 , b0 )
⌧ r̃(z, k 0 , b0 )b0
+ b (3.26)
(1 + r̃(z, k 0 , b0 ))(1 + r)
25
Related models in the macroeconomics literature include Pratap and Rendon (2003) and Cooley and Quadrini
(2001), which examine the e↵ects of financial frictions on capital accumulation and industry dynamics, respectively.

93
 The first change is that the interest rate on debt is a risky rate, r̃(z, k 0 , b0 ), which a function

of the current shock and the firm’s next-period choice of capital and debt. To see why the risky

interest rate depends on these three variables, first note that in these kinds of models the firm

optimally defaults when its equity value equals zero. Thus, the lender knows the default states

corresponding to each possible (k 0 , b0 , z 0 ) triple. The firm chooses (k 0 , b0 ), so that given the current

shock, z, the lender also knows the probability distribution over the default states corresponding

to the firm’s (k 0 , b0 ) choice. The risky interest rate is then set so that banks earn zero expected

profits, and it is therefore a function of (z, k 0 , b0 ).

The second change is the term (⌧ r̃(z, k 0 , b0 )b0 )/(1 + r̃)/(1 + r), which implies that the firm takes

the present value of the interest tax deduction in the period in which it issues debt. Although

clearly not what happens in reality, this feature greatly simplifies the determination of the risky

interest rate. Otherwise, the tax deduction would depend on last period’s shock, and current profits

would then depend on four state variables, which would include last period’s shock.

If the firm defaults, the lender recovers the firm’s profits and assets less any deadweight default

costs, ↵, so that total recovery is R(k, z) ⌘ (1 ↵)((1 ⌧ )⇡(k, z) + (1 )k)).26 At this point the

firm is left to start over again. The endogenous risky interest rate is then set so that the expected

rate of return on debt equals the risk free rate, given that the bank receives less than 100% recovery

in default. This zero-profit condition is

Z Z
b0 (1 + r) = R(k 0 , z 0 )dg z 0 , z + b0 (1 + r̃)dg z 0 , z (3.27)
default states solvent states

The Bellman equation for the problem can then be written as:

⇢ ⇢ Z
1
V (k, b, z) = max 0, max e(k, k 0 , b, b0 , z) + ⌘(e(k, k 0 , b, b0 , z)) + V (k 0 , b0 , z 0 )dg(z 0 | z) , (3.28)
0 0k ,b 1+r

This expression captures the default option by expressing equity value as the maximum of zero
26
In Hennessy and Whited (2007) the lender also extracts all of the bargaining surplus from the firm, which equals
the (opposite of) current period cash flow that would produce what equity value would be if the firm did not have
limited liability.

94
and the equity value of the firm as a going concern. Although the Bellman equation appears almost

identical to (3.22), the firm’s problem is more complicated because it interacts with a competitive

lender to set the risky interest rate. To solve the model, one needs to know firm value in order to

calculate the default states and thus the interest rate. However one needs to know the interest rate

to calculate firm value. One feasible solution algorithm is a “loop-within-a-loop,” which proceeds

as follows. First, assume that the interest rate on debt is the risk-free rate and solve for the value

function. With this estimate of the value function, one can calculate the default states and the

interest rate, which is a function of the model’s state variables. Then, using this new interest rate,

one repeats this procedure until the value function converges.27 The complexity of these models is

one disadvantage they have with respect to contingent claims models, in which default thresholds

can often be found analytically and can almost always be characterized as a di↵erential equation.

These models convey much of the intuition from the simpler risk-free debt model of Section 3.4.

In particular, optimal leverage declines with the serial correlation and variance of the productivity

shock, as well as with profit function curvature and the cost of external equity finance. Notwith-

standing these similarities, this type of model contains richer intuition than collateral-constraint

models. In particular, leverage is not constrained by available collateral. Instead, firms optimally

keep leverage at a moderate level to avoid default and the consequent deadweight default costs. Of

course, as the deadweight costs increase, leverage falls. Less obviously, because default occurs when

equity value is zero, optimal leverage increases with any technological feature that helps the firm

more efficiently transform capital into equity value. For example, a high capital depreciation rate

lowers average leverage because a high depreciation rate implies that any given amount of capital

creates less value. This last piece of intuition cannot be gleaned from simpler models in which the

size of the capital stock is the main determinant of debt capacity. Thus, the direct e↵ects of value

on debt capacity is unique to endogenous default models

Moyen (2004) uses this type of a model to reexamine the question from Gomes (2001) and

Alti (2003) of whether the sensitivity of investment to cash flow is a good measure of finance
27
See Gomes and Schmid (2010) and Moyen (2004) for other algorithms.

95
constraints. To this end, the paper compares two models: a full-blown endogenous default model

with costly external finance and an endogenous risky interest rate and an analogous model in

which all sources of outside financing have been shut down. Moyen (2004) finds that shutting o↵

external finance lowers firm value and reduces investment cash flow sensitivity. The intuition is

that firms that cannot access external finance have limited investment opportunities. As in the

simplest investment model, even conditioning on Tobin’s q, cash flow is a proxy for investment

opportunities because of production function concavity. When faced with constraints on external

finance, the firm invests less aggressively when hit with a positive cash flow shock.

This e↵ect operates in Hennessy and Whited (2007) as well, with regard to increasing the

cost of external equity financing. The intuition with regard to increasing the deadweight costs

of bankruptcy is di↵erent. In the face of high deadweight default costs, the firm hoards financial

assets, thereby allowing it to respond strongly to shocks to cash flow. Thus investment cash flow

sensitivity rises with the deadweight costs of default.

Hennessy and Whited (2007) also attempt to estimate the costs of external finance by performing

a structural estimation. The goal is to infer these costs by examining firm behavior through the

lens of the model. The main result is that there exist large indirect costs of external equity that

exceed simple underwriting fees. In addition, deadweight bankruptcy cost estimates are in line

with previous studies.

3.7 Other Models

The goal of this section has been to provide a strong basis for understanding the workings of

dynamic models of financing and investment. In so doing, we have not provided an exhaustive

survey of all of the papers that have used these kinds of models. This section attempts to clean up

these loose ends.

Dynamic models of investment and financing are a natural place to start understanding the

related areas of corporate diversification and restructuring. For example, the model in Gomes and

Livdan (2004) features a firm that can expand into an additional industry at a fixed cost. The

96
main result is that firms optimally expand when their original division becomes unproductive, so

that it is worth it to the firm to pay the fixed cost of diversifying. Thus, the widely documented

“diversification discount” emerges from the model as a natural consequence of diminishing returns

to scale. Warusawitharana (2008) examines the related question of asset purchases and sales, which,

in contrast to acquisitions, generate substantial shareholder value. Using a related model, he shows

that these gains are an endogenous outcome of firm value maximization.

The area of behavioral finance related to market timing has recently seen studies that use dy-

namic investment models. For example, Bolton, Chen, and Wang (2012) essentially add stochastic

costs of equity issuance to the model in Bolton, Chen, and Wang (2011) to understand when issuing

overvalued stock or repurchasing undervalued stock is optimal. One of many interesting results is

that firms use market timing to alleviate financial constraints and can thus smooth investment

much more than they would in the absence of any market timing ability. Another paper in this

area is Warusawitharna and Whited (2012), which models actual equity misvaluation instead of

stochastic issuance costs. This paper is also a structural estimation paper instead of a pure theory

paper. Their main conclusions are that managers do appear to time the market and add value for

long-term shareholders. However, this issuance and repurchasing activity has more of an e↵ect on

firm financial policies than on real investment policies.

One area of corporate finance that has seen little formal dynamic research is payout policy.

We are aware of one recent paper that fits loosely into the class of models treated in this section:

Lambrecht and Myers (2012). Risk-averse, habit-forming managers maximize their utility, which

is a function of rents. Rents, in turn, are a residual after the firm invests, borrows, and distributes

funds to shareholders. There are no taxes or financial frictions in the model, so that Modigliani-

Miller holds. However, managers are subject to a threat of intervention by outside shareholders.

Managers therefore pay out just enough dividends to keep their jobs. This threat ties dividends to

managerial rents. In this setting the original partial adjustment model from Lintner (1956) arises

from relatively primitive assumptions about preferences. The main result is that risk-aversion

causes managers to smooth rents over time. Because dividends are tied to rents via the control

97
threat, managers also smooth dividends. Managers’ habit formation also causes rents and payouts

only partially to adjust to changes in permanent income.

Finally, other literatures besides corporate finance have used this class of models to understand

a wide variety of phenomena. The first is asset pricing. At this point several papers have employed

dynamic models of financing and investment to try to understand phenomena ranging from the

e↵ects of finance constraints on expected returns (Whited and Wu 2006; Li, Livdan, and Zhang

2009; Livdan, Sapriza, and Zhang 2009), the low expected returns of highly levered firms (Gomes

and Schmid 2010), and credit risk (Kuehn and Schmid 2011). The second literature is macroeco-

nomics. In the wake of the 2007–2009 financial crisis, studies such as Covas and Den Haan (2011)

and Jermann and Quadrini (2012) have used versions of these models to understand the cyclical

properties of leverage and how financial shocks a↵ect real activity.

3.8 Summary

We close this section by pointing out that although dynamic models of investment and financing

are rich in their treatment of dynamic e↵ects, they are less rich in their treatment of fundamental

reasons behind the existence of financial frictions. For example, in these model debt and equity

issuance costs are specified exogenously, so that the firm is powerless to influence its own cost of

external finance. In addition, the form of financial contracts (usually equity and one-period debt) is

exogenous. Finally, most (but not all) of these models are implicitly specified under the risk-neutral

measure, which implies that one cannot disentangle the e↵ects of risk from the e↵ects of financial

frictions on corporate policies.

Of course, many of these simplifications have been made because the purpose of these models is

not to explain why financial frictions exist but to understand the consequences of financial frictions.

In addition, many of these simplifications have been adopted because one of the main purposes of

these models is to serve as a vehicle for using actual data to quantify economic phenomena. There

is a sharp trade-o↵ between the feasibility of taking a model directly to the data and the stringency

of the assumptions regarding financial contracts. A few studies such as Hennessy, Livdan, and

98
Miranda (2010) and Schmid (2011) have made some progress in trying to understand whether these

simplifications have important quantitative e↵ects; however, further work would be interesting.

4 Structural Estimation

One feature of all of the studies thus far reviewed that we have yet to discuss is whether and

how the models can be estimated. Some of these studies, such as Gomes (2001), Moyen (2004),

and Gamba and Triantis (2008) are pure theoretical pieces. Others, such as Hennessy (2004) and

Riddick and Whited (2009) contain both models and reduced form empirical evidence. Still others,

such as Hennessy and Whited (2005, 2007) have a great deal of empirical content in the form of

structural estimation exercises. We therefore turn to the next broad topic—structural estimation.

Structural estimation is an attempt to fit a model directly to data, to assess the quality of

the fit, to identify parameters that govern technology, preferences, and (thus far in corporate

finance) largely time-invariant institutional features. In particular, structural estimation ascertains

whether optimization models generate data that resemble data from real-world firms. As such,

structural estimation is an exercise in using a realistic theoretical structure to interpret the data.

Estimating models is useful because it allows estimation of parameters that can be used to quantify

the primitives that shape firm behavior. Thus, parameter estimates allow us to measure quantities

that, as financial economists, we find interesting, such as the cost of external finance or managerial

preferences over shirking. These parameters can also be used to study counterfactual situations

that can be useful for policy evaluation. Estimating models is also useful because the connection

between theory and tests of theory is extremely tight, thereby allowing a transparent interpretation

of any results. In contrast, interpretation of many reduced-form regressions is more difficult because

the assumptions of the underlying, often verbal, model are not spelled out.

Interestingly, structural estimation may or may not require a dynamic—as opposed to a static—

model. However, most modern incarnations of structural estimation, at least since Hansen and Sin-

gleton (1982), have employed dynamic optimization models to generate equations to be estimated.

In corporate finance structural estimation starts with Whited (1992), but despite this almost 20

99
year history, the total number of structural estimation papers in corporate finance is tiny. Promi-

nent examples include Bond and Meghir (1994), Love (2003), Hennessy and Whited (2005, 2007),

Sorensen (2007), DeAngelo, DeAngelo, and Whited (2011), Kang, Liu, and Qi (2010), Lin, Ma, and

Xuan (2011), Taylor (2010, 2012), Nikolov and Whited (2012), Albuquerque and Schroth (2010),

Morellec, Nikolov, and Schürho↵ (2012), Matvos and Seru (2011), Gantchev (2011), Glover (2011),

Page (2012), Korteweg (2010), Dimopoulos and Sacchetto (2011) and Coles, Lemmon, and Meschke

(2012), Schroth, Suarez, and Taylor (2012).

As in the case of theoretical dynamic models, we conjecture that the paucity of studies stems

from the perception that structural estimation is too complex, so that lack of transparency makes

it difficult to learn anything from these sorts of exercises. Therefore, one goal of this section is to

explain the intuition behind these techniques to dispel this perception of complexity. Once again,

the intent is not to provide mathematical rigor, which can be found in any econometrics textbook,

but to provide insight into how these techniques work and “advice from the trenches” designed to

help researchers avoid common pitfalls.

The second goal of this section is therefore not only to review the literature, which we structure

around the four main methods that have been used so far in corporate finance for structural

estimation: generalized method of moments (GMM), simulated method of moments (SMM), and

simulated maximum likelihood, (SMLE). We then survey the literature. We close with a comparison

between calibration and structural estimation, and with a brief overview of the wealth of other

structural estimation techniques that have not yet been widely applied in corporate finance.

4.1 GMM and Euler Equations

The papers using GMM, starting in corporate finance with Whited (1992), typically estimate

investment Euler equations, which are variants of (3.8).28 In reviewing this literature, we do not

review the basics of GMM, which can be found in any graduate econometrics textbook. Instead,

we look at assumptions necessary to apply GMM to specific problems.

28
See Bond and Meghir (1994), Love (2003), Kang, Liu, and Qi (2010), Lin, Ma, and Xuan (2011), and Liu, Whited,
and Zhang (2009). One important example that does not involve investment Euler equations is Albuquerque and
Schroth (2010), which uses exactly identified GMM on a static model of a takeover premium.

100
 Hansen and Singleton (1982) point out that estimating any Euler equation for any dynamic

decision requires an assumption of rational expectations. This assumption allows the empirical

researcher to replace the unobservable expected cost of delaying investment in estimating equa-

tions such as (3.8) with the observable realized cost of delaying investment plus an expectational

error. The intuition behind this replacement is straightforward: as a general rule, what happens

is equal to what one expects plus one’s mistake. Further, the mistake has to be orthogonal to any

information available at the time that the expectation was made; otherwise, the expectation would

have been di↵erent. This last observation implies that lagged endogenous variables are orthogonal

to the expectational error and that they therefore can be used as instruments to estimate Euler

equations.29

One useful feature of GMM is that it is accompanied by a general specification test—the test

of the over identifying restrictions. Of course, the validity of this test relies on the assumption

that one has as many instruments as parameters, and this assumption (also known as an exclusion

restriction) cannot be tested. Finally, it is worth reemphasizing that one should have “strong”

instruments. In the case of nonlinear GMM, instrument strength cannot be cast in terms of the

degree of correlation between the instruments and endogenous regressors. Instead, it is usually

cast as full rank of the gradient of the moment conditions with respect to the parameters. Wright

(2003) o↵ers a useful diagnostic test of instrument strength for nonlinear models. Euler estimation

with panel data can be accomplished with a variety of di↵erent statistical packages, such as Stata

or SAS, so that implementing this kind of estimation is straightforward.

We now turn to a brief review of this literature. Whited (1992) estimates a nonlinear investment

Euler equation with data on U.S. firms, finding that the overidentifying restrictions of the model are

not rejected for financially healthy firms but that they are rejected for financially unhealthy firms.

By parameterizing a Lagrange multiplier on a constraint that limits firm financing (for example,

such as (3.20)), and by substituting this parameterization for the Lagrange multiplier in the Euler
29
It is worth pointing out that using lagged instruments on a generic regression does not constitute structural
estimation, which specifically requires that an economic model directly produce the estimating equation. In the case
of Euler equations, the model also produces the assumptions allowing the validity of lagged instruments.

101
equation, the paper also finds that financially constrained firms tend to behave as if they have

a higher discount rate than their financially unconstrained counterparts; that is, they postpone

investment until the future. This result uncovers a specific economic e↵ect of financing constraints,

and it could not have been found in a static regression because the concept of delay is by definition

a dynamic concept.

Kang, Liu, and Qi (2010) and Lin, Ma, and Xuan (2011) conduct similar exercises in order

to understand, respectively, the e↵ect of the Sarbanes-Oxley Act on corporate investment and the

e↵ect of ownership structure on financial constraints and thus investment. In contrast, Bond and

Meghir (1994) and Love (2003) linearize the Euler equation. They demonstrate using models such

as that in Section 3.4 that financial variables enter into the specification in the presence of financial

frictions such as an increasing interest rate schedule. They then straightforwardly test for the

significance of these variables. They conclude that financial frictions are important for investment.

4.2 Simulated Method of Moments

To motivate this section, we note that one serious drawback with estimating the first-order condi-

tions from dynamic models is the assumptions one needs to make in order to produce a closed-form

estimating equation. For example, investment Euler equations make the assumption of quadratic

investment adjustment costs, which, as explained in Section 3.1.3, imply that the firm optimally

wants to smooth investment over time. This assumption is at odds with the lumpiness often ob-

served in the time-series of firm investment. More realistic models that relax this unpalatable

assumption, however, require much more computationally intensive techniques if one wants to

estimate them.

To sum up, there exists a tension between realism and the sorts of models that can produce

closed-form estimating equations. Better models that can explain more phenomena may not lend

themselves to closed-form solutions or to smooth, di↵erentiable first-order conditions. Fortunately,

the development of the econometrics of simulation estimators, such as Pakes and Pollard (1989),

Ingram and Lee (1991), Duffie and Singleton (1993), and Gourieroux, Monfort, and Renault (1993),

102
has allowed researchers to bring better models to the data. Although many di↵erent simulation

estimators are used in other fields such as industrial organization, macroeconomics, and labor

economics, we review the two main types that have been used in corporate finance. The first is

simulated method of moments (SMM) and the second is simulated maximum likelihood. The rest

of this section outlines how to do an SMM estimation and reviews the literature that uses this

technique.

4.2.1 Outline

SMM is conceptually simple. As detailed in Section 3.1.2, discrete-time dynamic models typically

have a solution in the form of a rule that prescribes optimal policy tomorrow, given the firm’s state

today (such as its levels of debt and capital) and a random shock received today. This rule can be

used to generate a panel of simulated data by simulating a series of shocks and tracing out the firm’s

optimal choices. The researcher then calculates interesting moments using both the simulated data

and a real data set and then methodically searches for model parameters that make the simulated

moments as close as possible to their corresponding real moments. Estimation requires at least as

many moments as model parameters. Examples of parameters in the models in Section 3 include

profit function curvature, the serial correlation and variance of the z shock, etc.

An SMM estimation usually proceeds in two separate set of steps.

First Set of Steps:

1. Choose a set of moments to match. The list can include means, variances, covariances,

regression coefficients, and so on, but one must have at least as many moments as parameters

to estimate.

2. Calculate these moments with real data and stack them in a vector. We denote the estimated

vector of real moments as M (x), in which x is an i.i.d. data sample.

3. Calculate the covariance matrix of the moments. Invert it. This matrix is the GMM weight

matrix. We denote the estimated weight matrix as W .

Second Set of Steps:

103
1. Pick starting values for the set of parameters, which we call 0.

2. Using 0, solve the model, create simulated data using the policy function, and calculate

the same moments that were calculated with the real data. Stack them in a vector. We

denote the simulated moments as m(y, ), in which y is a simulated data sample. Note that

the simulated moments are a function of the parameter vector, , and can change when

changes.

3. Form the SMM moment vector as the stack of simulated moments minus the stack of real

moments.

4. Form the SMM objective function exactly as one would form a GMM objective function, that

is, as a weighted sum of squared errors, in which the “errors” are given by the moment vector,

and the weights are given by the weight matrix. Let the objective function be denoted as

Q(x, y, ):

Q(x, y, ) ⌘ (M (x) m(y, ))0 W (M (x) m(y, )) . (4.1)

5. Find the parameter vector, ˆ, that minimizes the SMM objective function.

6. Adjust the standard errors and test statistics for simulation error. Let J be the ratio of the

number of observations in the simulated data set to the number observations in the real data
p
set, which we denote as N . Then the covariance matrix of N ( ˆ ) is given by:

✓ ◆⇣ ⌘
1 1
1+ (@m(y, )/@ )0 W (@m(y, )/@ ) (4.2)
J

The term 1 + 1/J is the adjustment for simulation error, which approaches 1 as J ! 1.

The rest is a standard GMM covariance matrix.

7. Use either a general test of the overidentifying restrictions of the model as a specification

test or, as in DeAngelo, DeAngelo, and Whited (2011), test the equality of the moments

individually. The test of the overidentifying restrictions can be written as:

104
 NJ
Q(x, y, ) (4.3)
1+J

The t-statistics on the individual moment conditions are calculated exactly as pricing errors

in a standard GMM framework. See, for example, Cochrane (2005).

From this recipe, it is clear that the “data” part of SMM can be done separately from any

computations involving the model. Thus, the hardest part of doing SMM is solving the economic

model. Once one is armed with a solution, it is straightforward to simulate data and calculate

simulated moments. Nonetheless, SMM can be time intensive because one has to solve the model

with every step of the econometric minimization routine.

4.2.2 Identification

The success of SMM relies fundamentally on picking moments that can identify the structural

parameter vector, . “Identification” means something di↵erent in this context than it does in

an instrumental variables setting, so it is useful to elaborate on this di↵erence. In an IV setting,

“identification” typically means finding a source of exogenous variation in the data. This exogenous

variation allows one to ascribe a causal interpretation to any regression coefficients one obtains.

In other words, because IV or OLS can fit lines through most data scatters, identification in this

context can be thought of as making enough assumptions about the source of data variation to

interpret estimated elasticities as causal relations. In other words, it amounts to finding a suitable

lens through which to interpret the data. In structural estimation, finding a suitable lens involves

writing down a model, and the word “identification” is used in the econometric sense that an

econometric objective function have a unique minimum.

Both the mathematical and intuitive conditions for identification of an SMM estimator are

similar to those for GMM. We focus on the intuitive aspect, which requires that the moments be

informative about the structural parameters, or equivalently, that the sensitivity of the moments

to the parameters be high. This requirement is obviously analogous to the importance of using

high quality instruments in a standard IV framework, that is, instruments highly correlated with

105
the endogenous regressors.

Ideally, each parameter would be identified by a single moment. Changing that parameter would

cause that moment, and only that moment, to vary. Of course, this hurdle is very high and not

actually necessary. In fact, if it were, then an economic model probably would not be necessary.

What is much more likely to occur can be intuitively described as staggered identification, which is

what we exploit whenever we solve linear systems of equations. A simple example helps illustrate

how identification works. Suppose we are using three moments (a, b, c) to identify three parameters

(x, y, z), and that the economic model implies the following moment conditions:

a = x+y

b = y+z

c = x+z

If we know a, b and c, we can uniquely identify the (x, y, z), even though each moment is a↵ected by

several parameters and each parameter a↵ects several moments. SMM exploits this same feature

in a non-linear way.

How does one ensure that the model is identified? First, check the parameter standard errors.

If one has chosen bad moments, the standard errors will be large because the sensitivity of the

moments to the parameters enters into the calculation of the standard errors. Second, and similarly,

if the estimation does not converge, the model parameters are likely not identified.

Third, and most important, the best way to pick good moments is to understand the economic

forces that drive the model, just as the best way to pick good instruments is to understand the

economics of the question being asked. The economic intuition behind SMM thus comes from

the identification conditions, which can only come from a careful understanding of the model.

Understanding the economics of a model that has no closed form solution requires time because

it requires conducting comparative statics exercises and plotting moments versus parameters. The

ideal moment has a steep, monotonic relation to a parameter, and moments should move in di↵erent

directions for di↵erent parameters. Ensuring meaningful identification thus entails an attempt to

106
disentangle which parameters a↵ect which moments. Finally, it is important not to “cherry pick”

moments. A model should be able to make sense of many features of the data, and it is useful and

often instructive to determine where the model does and does not do a good job of matching the

data.

To understand the importance of identification, we construct a simple counterexample of an

unidentified parameter as follows. Suppose one were to add operating leverage to the simple model

from section 3.1 and remove capital adjustment costs so that the sources and uses of funds identity

(3.1) becomes

e(k, k 0 , z) ⌘ ⇡(k, z) (k 0 (1 )k) 'k, (4.4)

in which the term, 'k, is operating leverage. It is straightforward to see that it impossible to

identify the parameters and ' separately using data on the capital stock. It is only possible to

estimate their sum, + '. However, if one were to find separate data on capital depreciation, it

might be possible to identify ' using data on profitability.

To close our discussion of identification, we add one word of caution. Just as there does not

exist any perfectly exogenous source of data variation in observational studies, structural estima-

tion does not magically solve all endogeneity problems. Structural estimation accounts for any

endogeneity within the model. However, just as any linear econometric model su↵ers from omit-

ted variables problems to one degree or another, there will always be elements omitted from an

estimated structural model. Therefore, one needs to worry about whether omitting those elements

biases the parameter estimates. If one understands the model one is estimating, this type of im-

portant thought experiment is usually feasible, and it often opens doors for future research.

4.2.3 Practical Advice

Several pieces of practical advice are in order. The first is to use real data to estimate the weight

matrix. In both GMM and SMM, the weight matrix is the inverse of the covariance matrix of

the moment conditions. In this type of an SMM exercise, the weight matrix is therefore just the

107
covariance matrix of the data moments alone. This weight matrix (roughly) puts the most weight

on the most precisely estimated moments.

It is not necessary to use any model simulations to estimate this covariance matrix. Instead, it

can be estimated by calculating all of the moments jointly as a large GMM system. Another, often

computationally simpler, method is to calculate the moments separately, calculate the influence

function for each moment, stack the influence functions of the moments, and take their inner

product.30 Heuristically speaking, an influence function of an estimator is a function of the data

whose mean has the same asymptotic distribution as the estimator. For example, suppose an i.i.d.

sample of a random variable, x, has a sample average x̄. Then the influence function is trivially

x x̄. See Newey and McFadden (1994) for more precise definitions of influence functions. One final

important issue is accounting for time-series dependence in the data when estimating the weight

matrix. In panel data, this issue is usually accomplished via a clustering algorithm.

The second piece of advice is not to use an identity matrix as a weight matrix because it

mechanically puts the most weight on the moment that is largest in absolute value. This advice lies

in contrast to that given in Cochrane (2005) regarding tests of asset pricing models. The di↵erence

is that returns on di↵erent portfolios are all of the same order of magnitude. In contrast, moments

used in a corporate finance simulated moments exercise can be of very di↵erent magnitudes.

The third piece of advice is a set of small pointers concerning the minimization of the SMM

objective function. It is computationally infeasible to use gradient based methods such as Newton-

Raphson. One can use a grid search if one is only estimating one or two parameters. For more

complex problems, other alternatives include various linear-programming algorithms such as Nelder-

Meade or simulation based algorithms such as Metropolis-Hastings or the closely related simulated

annealing. As a general rule, simulation based algorithms do a good job of efficiently getting

close to a global minimum, but their final convergence is often glacial. On the other hand, linear-

programming based algorithms work much more quickly but are also more likely to get stuck in

a local minimum. Therefore, a useful approach is to start with a simulation based algorithm and
30
See Erickson and Whited (2012).

108
then switch to a linear-programming algorithm when the estimation is near convergence. A formal

switching rule can be cast in terms of a loose convergence criterion. Next, it is important to pick a

good starting value for the minimization algorithm. This choice is often a byproduct of comparative

statics exercises that one should do. It is equally important to try several starting values to increase

the probability of finding a global minimum of the SMM objective function. Next, one must use

the same pseudo-random draws for each simulation of the artificial data. Otherwise, it is possible

to attribute a change in the SMM objective function to a change in a parameter, when it is actually

due to a change in the pseudo-random draws. Finally, simulate data sets that are several times as

large as the size of one’s actual data set. The simulation step is computationally cheap, so it makes

sense to lower one’s standard errors as much as possible by using large simulations to remove as

much simulation error as possible.

The fourth piece of advice concerns heterogeneity. The models we have considered here are

usually of a single firm or at best of an industry equilibrium, in which heterogeneity comes from

realizations of shocks. Thus, SMM estimates the parameters of an average firm, not the average

parameter across firms. These two quantities are equal in linear models, but not necessarily in

the nonlinear models considered here. However, corporate finance data are generated by heteroge-

neous firms in di↵erent industries with di↵erent technologies, di↵erent sources of uncertainty, and

di↵erential access to financial markets. Thus, interpretation of any results is easiest on relatively

homogeneous subpopulations in firms. For example, DeAngelo, DeAngelo, and Whited (2011) es-

timate their model on di↵erent industries. Whether this type of sample splitting is feasible or not,

it is nonetheless important to extract as much heterogeneity from the data as possible, usually

by removing fixed e↵ects. Often in estimating dynamic models, it is natural to use autoregressive

coefficients as moments. In this case, one cannot, of course, demean data on a firm by firm basis.

A feasible alternative is the double di↵erencing method in Han and Phillips (2010).

4.2.4 Model Design

The most important piece of advice concerns model design. It is essential to write down a realistic

model that can shed light on the questions one wants to answer. To illustrate this point, we turn

109
to the careful model design in Taylor (2010). This paper asks three related questions. It asks

why so few CEOS are fired every year (2%). It also asks how much firing one should expect from

a well-functioning board, and how much shareholder value is destroyed if boards do not function

well. The second two questions are particularly well suited to structural estimation because they

require calculating a counterfactual, that is, asking what would happen if the model parameters

were di↵erent from the estimated parameters.

The paper then postulates four possible reasons for the low firing rate. The costs to shareholders

of turning over the CEO may be large. If the next best CEO is as good as the current one, why

bother? Boards can learn slowly about CEO ability. CEOs can be entrenched. What makes this

paper useful is that the model is tailored specifically to answer these questions.

The model can be summarized as follows. In each period the board decides whether or not

to fire the CEO. The CEO also retires or quits with a certain probability, and the firm generates

profits. These profits are the sum of three components: industry profits, firm-specific profits, and

CEO turnover costs if the CEO is fired. Firm specific profitability can be a↵ected by the CEO’s

ability, but the board cannot distinguish the industry and firm-specific components of profits.

New CEOs are drawn from a talent pool characterized by normally distributed ability. The

board of directors has priors over this distribution, so that when the CEO starts, the board at-

tributes mean CEO talent to the CEO. However, the board updates its beliefs about CEO ability

each period when it observes profitability. The board then maximizes its utility, which consists of

the expected discounted value of future profit flows minus any personal costs incurred if the board

fires the CEO. The board optimally fires the CEO as soon as posterior mean skill drops below an

endogenous threshold

The model’s parameters capture the four possible reasons for firing CEOs. This feature of the

design of the model implies that estimating these parameters sheds light on the importance of

the di↵erent reasons. The model contains an entrenchment parameter that takes the form of the

board’s personal costs (loss of a golf partner) of firing the CEO. It contains a parameter quantifying

the costs to shareholders of CEO turnover. Lack of better possible replacement CEOs is measured

110
by the variance of the CEO talent pool. Finally, slow learning on the part of the board is captured

by the statistical properties of the profit signals that the board does observe.

The empirical results are interesting. The parameter estimates indicate that boards have high

personal costs of firing CEOs—so high that boards behave as if firing a CEO costs shareholders

5.9% of assets, whereas it really only costs them 1.3% of assets. Also, in counterfactual experiments

that set the board’s personal costs to zero, the model predicts that 13% of CEOs would be fired

every year if boards were to act in shareholders’ interests.31 This type of counterfactual exercise is

one of the main reasons for doing structural estimation because counterfactuals that revolve around

preferential di↵erences are difficult to do in any other way.

4.3 Simulated Maximum Likelihood

Very recently, a new methodology has been introduced into structural estimation in empirical

corporate finance: simulated maximum likelihood. Simulated maximum likelihood comes in many

di↵erent flavors, so we focus on the particular flavor that has recently been used in corporate finance

in Morellec, Nikolov, and Schürho↵ (2012) (MNS, hereafter).

To begin explaining the technique used here in general terms, suppose one has written down an

economic model that can produce the density of a variable, x, of interest. This density is completely

characterized by a parameter and can thus be denoted as f (xi | ), in which xi is an observation

from the random sample, x. One example of such a density comes from a real options model, such

as the one outlined in Section 2.2, in which the marginal product of capital fluctuates between

two bounds. The firm invests in a lumpy fashion when the marginal product hits a bound, thus

driving the marginal product back to a return point. Another example can be found in the dynamic

capital structure model in Section 2.4, in which cash flows (and hence leverage) fluctuate between

two bounds. For many models in this class, the density is usually piecewise exponential.

Of course, if one has a closed-form solution for the density f (x | ), it is possible to estimate the

model parameters via standard MLE. However, because one has a closed-form density, it is possible
31
Of course, some entrenchment may be optimal for shareholders ex ante if, for example, it improves the pool of
potential new CEOs.

111
to add parameter heterogeneity to the econometric model. The ability to address heterogeneity

directly is one important advantage of simulated MLE over moment-based estimators. One adds

heterogeneity by expressing as i ⌘ + ui , in which ui is a random variable that varies across

the observations in the data set. In this case the density becomes f (x | , ui ).

If one is willing to make distributional assumptions about ui , such as ui ⇠ f (ui ), then one can

integrate ui out of the density to arrive at the marginal density of x. In this kind of a mixture

model, it is rare that one can derive a closed form solution for the marginal density. Instead,

one can use Monte Carlo integration as follows. Simulate H random draws, uhi , from f (ui ) and

approximate the integral by:

1 X ⇣ ⌘ ⇣ ⌘
h
f xi , | uhi f uhi (4.5)
H
j=1

Now the SMLE estimator is defined as:

!
N
X H
1 X ⇣ ⌘
ˆ ⌘ arg max ln f (xi , | ui ) f uhi , (4.6)
H
i=1 h=1
and the usual MLE distribution theory applies. Unlike an SMM estimator, which is consistent even

for small H, the SMLE is consistent only if both N and H go to infinity. Thus, it is important to

use many simulated observations in the Monte Carlo integration.

Thinking about identification is just as important here as it is for SMM. The technical conditions

for identification of an SMLE estimator are identical to those for an MLE estimator. Heuristically,

one cannot have a flat likelihood around the true parameter value. This requirement implies that

the likelihood has to change shape when a parameter moves, and it has to change shape in di↵erent

ways for di↵erent parameters. Therefore, just as in the case of SMM, it is important to do many

comparative statics exercises to determine whether one’s likelihood does in fact change shape with

di↵erent parameter values.

MNS use this technique to estimate a contingent claims model of leverage that is closely related

to the model in Morellec (2004), in which managers can steal cash flows. Shareholders can only

stop managers by conducting costly control challenges; so managers steal just enough to keep the

112
shareholders from ousting them. Without agency problems, as explained in Section 2, this class

of models often produces leverage ratios that are too high relative to those observed in the real

world. The tax-advantages of debt loom large in comparison to the expected bankruptcy costs. The

converse of this problem is that excessively high debt issuance costs are required to match model

generated leverage with real world leverage. The reason is that equity values drift upward, thereby

mechanically causing leverage to drift downward. With high issuance costs, firms restructure their

debt infrequently, and the downward drift keeps model and real leverage in line. In MNS, however,

the distribution of optimal model-generated leverage ratios (including the mean) can be reconciled

with the distribution of actual leverage ratios primarily because managers keep leverage low in

order to be able to divert cash flows to themselves rather than to bondholders.

The estimation starts with a derivation of the distribution of leverage in the model. At this

point MNS add parameter heterogeneity as described above. First, they assume that each estimated

model parameter has a normal distribution. Therefore, for each possible set of model parameters,

leverage must have a di↵erent distribution. This joint distribution is intractable, but they solve

this problem by doing maximum likelihood with the marginal distribution of leverage, which they

calculate by using Monte Carlo integration to integrate out the parameter heterogeneity. This last

step is what adds a numerical dimension to their estimation problem.

MNS get two benefits from all of this added complexity. First, they quantify managers’ average

perceived costs of control (private benefit from resource diversion) to be approximately 2% of firm

value. These kinds of quantitative predictions are hard to make in the absence of a theoretical lens

through which to view the data. Second, the addition of random e↵ects, ui allows them to see

whether parameter estimates vary in sensible ways with observed firm characteristics, which is akin

to an out-of-sample test. They find that the most important determinant of managerial leeway to

steal is institutional ownership.

This kind of exercise is much less time consuming than SMM, and the problem of picking

moments to match disappears if one is maximizing a likelihood. However, one does need to be able

to derive a closed form likelihood function from one’s model. This requirement puts restrictions

113
on the classes of models that can be estimated with this technique. Further, these types of models

tend to be harder to identify because one needs to extract a great deal of information from the

distribution of one variable, in this case, leverage. Thus SMM and SMLE have di↵erent advantages

and disadvantages, and the choice of technique is likely to be dictated by the model being estimated.

4.4 Literature Review

In the process of describing the di↵erent techniques that can be used for structural estimation, we

have reviewed several of the extant studies. However, we have also omitted many more, and the

intent of this section is to summarize the rest of the literature. First, as detailed in Section 4.1,

the literature on investment Euler equations has been almost exclusively been interested in the

e↵ects of some sort of financial friction on investment. The subsequent literature that uses more

computationally intensive techniques has been able, in contrast, to address a much wider range of

interesting questions. Thus, although Hennessy and Whited (2005, 2007) and DeAngelo, DeAngelo,

and Whited (2011) examine the kinds of capital structure and financial frictions questions on which

we have focused, other studies have ventured into closely related areas such as cash holdings, as

well as very di↵erent areas such as shareholder activism and executive compensation.

In the broad area of capital structure, Nikolov and Whited (2012) estimate a dynamic model

of investment and cash via SMM to understand the extent to which agency issue a↵ect corporate

cash holdings. Their main result from this estimation is that agency issues related to self dealing

are more important for explaining corporate cash balances but that agency issues related to firm

size are more important for firm value.

Again in capital structure, Korteweg (2010) uses Markov Chain Monte Carlo methods to es-

timate the net benefits to debt, that is, the tax benefits less financial distress costs. The general

challenge in this literature is to isolate financial distress costs from economic distress that usually

leads to financial distress. The idea inKorteweg (2010) to derive estimating equations from robust

identities for levered and unlevered betas and a linear leverage model, which can be derived from

the Leland (1994) model. These equations express the benefits of leverage as a function of the

114
level of leverage, and the the benefit beta as a function of debt and equity betas. These equations,

along with the assumption that similar firms have identical asset betas allow the recovery of the

unobservable net benefits. Bayesian estimation techniques are useful for this exercise because they

easily allow estimation in the presence of missing data. The main results are that the maximum

net benefit for the median firm is about 5% of value, and that the average firm is under levered.

However, the second result mostly comes from the zero leverage firms in the sample.

One further structural estimation study dealing with financial distress is Glover (2011). This

paper is based on the clever observation that using observed defaults to estimate the costs of

financial distress, understates these costs substantially. The reason is that firms with high costs

optimally choose low leverage and thus do not default. The only way to tackle this self selection

problem is to have an estimate of optimal leverage, and one feasible way to estimate this quantity

is with a model. Glover (2011) uses SMM to estimate a variant of the contingent claims model in

Goldstein, Ju, and Leland (2001). His main result is that firms expect to lose 45% of their value

in default—a number much higher than previous estimates.

Executive compensation has also been home to structural estimation studies. As detailed above

Taylor (2010) uses structural estimation of a dynamic learning model to understand why CEOs

are seldom fired. Taylor (2012) uses estimation of a related model to ask a di↵erent question: how

does CEO pay respond to news about CEO ability. One of the more interesting results from this

exercise is that CEOs have downwardly rigid pay when bad news arrives. Using a model helps

make sense of this result. It does not appear to be driven by weak governance. Instead, the model

estimates imply that CEOs accept low pay in return for the insurance provided by downward wage

rigidity. Page (2012) asks the converse question of why firm value appears to be insensitive to

changes in CEO incentive (as opposed to total) pay.32 To answer the question, he estimates a

model in which a risk averse CEO and a board interact to set the CEO’s contractual incentive pay.

The main message from the estimation is that performance is insensitive to pay because CEO’s

have an extremely high marginal utility of leisure; that is, they are already working very hard.
32
See also Coles, Lemmon, and Meschke (2012).

115
 Structural estimation has also been used to understand takeovers. For example, Albuquerque

and Schroth (2010) estimate the private benefits of control by examining the discounts and premia

associated with block trades. The intuition starts with a model in which that block trades transfer

private benefits of control and that the block trades at a premium when the owner can fend o↵ a

possible subsequent tender o↵er. Therefore, Albuquerque and Schroth (2010) can extract estimates

of the private benefits of control from the premia and discounts. The main result is that private

benefits represent about 3% of equity value. Dimopoulos and Sacchetto (2011) tackles a related

question by trying to understand whether large takeover premia represent takeover resistance or

initial preemptive bidding on the part of one of several potential acquirers. To separate these two

possible explanations, Dimopoulos and Sacchetto (2011) estimate an auction model of takeovers

that encompasses both explanations. The most important result from this estimation is that target

resistance is the main determinant of takeover premia.

Several papers in a smorgasbord of other areas of corporate finance have also used structural

estimation. Sorensen (2007) asks whether better venture capitalists (VCs) are more likely to take

their entrepreneurs public. This question is difficult because good entrepreneurs end up getting

matched with better VCs. This self selection problem cannot be treated with a standard Heckman

correction because the matching between VCs and entrepreneurs is not a zero-one decision. Instead

Sorensen (2007) uses Markov Chain Monte Carlo methods to estimate a two-sided matching model

of VCs and entrepreneurs. The results from this estimation can then be used to correct for the

sample selection bias in the estimation of the e↵ect of VC quality on entrepreneurial outcomes.

The main result is that a standard profit estimation of the e↵ect of VCs on IPO probabilities is

overstated by a factor of two. Gantchev (2011) estimates a sequential decision model of a hedge

fund’s decision to conduct a proxy contest. The intent is to understand whether the returns earned

by shareholder activists cover their monitoring costs. Interestingly, the mean net activist return

is near zero, but the top quartile of activists earn substantial net returns. Schroth, Suarez, and

Taylor (2012) estimate a dynamic model of bank runs to measure the fragility that results from

financing long-term assets using dispersed, short-term debt. The estimates of the model then allow

116
evaluation of various policies designed to prevent bank runs. The main finding is that interventions

targeting asset liquidity and conduit leverage are most e↵ective. Warusawitharana (2011) estimates

a dynamic model in which innovation improves firm productivity in order to understand variation

in R&D intensity across industries.

4.5 Calibration versus Estimation

We now comment on the di↵erences between structural estimation and an exercise called “calibra-

tion,” which is similar to SMM inasmuch as calibration also tries to match model-generated stylized

facts with stylized facts in the data. Both exercises are useful, but they serve di↵erent purposes,

so it is important to keep them separate. From a purely technical point of view, calibration is

di↵erent from structural estimation mostly because model calibrations are not accompanied by

standard errors for the model parameters, and structural estimations do provide standard errors.

The reason is that calibration only tries to match a few stylized facts with many model parameters,

and whenever there are too many degrees of freedom, inference is impossible. In contrast, SMM

matches at least as many stylized facts as model parameters, and therefore standard errors can be

calculated in the same way as GMM standard errors. Calibration is especially useful in situations

in which model estimation is infeasible but in which one nonetheless wants to learn from the model.

However, the distinction between calibration and estimation goes deeper. The main purpose

of calibration is to ensure that a model with a numerical solution provides (usually directional)

predictions that are likely to be empirically relevant. This type of exercise can be extremely helpful

for rounding out intuition. For example, Strebulaev (2007) picks parameters for his model in a

variety of ways: primarily by matching moments and distributions from di↵erent data sets, and by

using estimates from previous studies. As discussed above, the calibration in this paper advances

understanding of extant empirical results by running regressions on simulated data. Thus, although

calibration is not an empirical exercise, it can yield useful results as a part of a theoretical exercise.

In contrast, structural estimation tries to find the unique set of model parameters that reconcile

the model with the data. It is therefore an explicitly empirical exercise that attempts to “stress

117
test” a single model to compare the performance of competing models. Any SMM estimation that

uses more moments than parameters is an example of stress testing. If one uses enough moments,

any model will fail, but it is useful to know where. As an example of model comparisons, Whited

(1992) compares investment models with and without financing frictions. Structural estimation also

is useful as an empirical exercise because it can be used to measure quantities that are of interest

to financial economists, such as the cost of external financing (Hennessy and Whited 2007).

4.6 Other Methods

The three methods here are those that have been used in empirical corporate finance. A wealth

of other methods exists in asset pricing, macroeconomics, labor economics, and industrial organi-

zation. For example, macroeconomics and asset pricing researchers use a method closely related

to SMM called indirect inference (Gourieroux, Monfort, and Renault 1993). The idea behind this

technique is that it is often impossible to characterize the likelihood of the data from a dynamic

model. However, it is often possible to use what is called an “auxiliary” model to approximate

the likelihood. As long as the mapping between the model parameters and the parameters of the

auxiliary model is unique, one can then estimate the structural parameters of a model by minimiz-

ing the di↵erence between the parameters of the auxiliary model estimated on real data and the

parameters of the auxiliary model estimated on model simulated data.

The empirical industrial organization literature contains examples of many di↵erent techniques.

For example, one can estimate the parameters of a dynamic discrete choice model by constructing

a likelihood function directly from the model, as in the seminal work of Rust (1987). Other

methods sidestep the necessity of solving the model in the first place. For example, Hotz and

Miller (1993) show that one can estimate the parameters of a dynamic discrete choice model using

a two-step method. First, one estimates, nonparametrically, the choice probabilities from the data

and then one uses these estimates to back out the model parameters from the model without

having to solve it repeatedly. Aguirregabiria and Mira (2007) build upon this idea to formulate a

method for estimating dynamic discrete games that have multiple equilibria. One final, but by no

means exhaustive, example is in Bajari, Benkard, and Levin (2007), which uses an estimated policy

118
function in conjunction with the optimality conditions of the model to estimate the structural

model parameters.

At this point, we are aware of only two examples of the use of these methods in corporate finance.

Matvos and Seru (2011) uses the methods in Bajari, Benkard, and Levin (2007) to examine the

extent to which inefficient resource allocation contributes to the diversification discount. Kang,

Lowery, and Wardlaw (2012) uses the methods in Hotz and Miller (1993) to estimate the direct

and indirect costs to taxpayers of bailing out failed banks. Clearly, the scope is enormous for

incorporating these methods into corporate finance to answer interesting questions.

5 Conclusion

This survey has attempted to cover the broad area of dynamic corporate finance. One lesson that

emerges is that the field is extraordinarily diverse. For example, we have covered three broad

areas—contingent claims models, discrete-time financing and investment models, and structural

estimation—and each area is markedly di↵erent in terms of the types of analytical devices it em-

ploys. Structural estimation can be performed on either class of theoretical models, but the two

classes of theoretical models are distinct. Thus, at this point, a brief comparison is in order.

Contingent claims models o↵er two important advantages over discrete-time investment models.

First, they allow the pricing of claims, being based on techniques developed in the derivatives pricing

literature. Investment-based models, in contrast, have only just started to be used to price claims

via the addition of a pricing kernel (Gomes and Schmid 2010). Second, dynamic contingent claims

models allow for unbounded firm growth because of the homogeneity property that allows large

firms to be thought of as scaled up versions of small firms. Investment-based models, in contrast,

study firm behavior in a bounded set of possible values for the state variables. Thus, while they are

extremely useful for studying ratios (leverage ratios, investment rates, etc.), they are less useful for

studying long-term growth, as they can, at best, be interpreted as representing fluctuations around

long-term deterministic trends.

Discrete-time financing and investment models also have a unique set of advantages. In these

119
models, investment is endogenous and the firm faces a period-by-period sources and uses of funds

identity. This set-up allows the study of several phenomena that are outside the realm of contingent

claims models. For example, it is possible to investigate the joint dynamics of investment and

finance—a concept that is not well-defined in a contingent claims model. Further, it is possible to

examine the interplay between a pecking-order-like incentive to use the cheapest source of funds

versus the classic trade-o↵ between tax benefits and distress costs. Both of these motives operate

in a dynamic investment and financing model, thus giving rise to behavior that resembles neither

a pecking order nor a static trade-o↵. Only the trade-o↵ motive operates in a contingent claims

model. Finally, it is possible to model explicitly an array of financing vehicles that is larger than

just simple “debt” and “equity,” such as distributions to shareholders and cash.

These two classes of models constitute di↵erent types of theory. In contrast, structural estima-

tion is a specific type of empirical exercise that fits theoretical (as opposed to statistical) models

directly to data. In corporate finance, structural estimation has mainly been used on the types of

models surveyed here, but the only requirement for estimation is that the model parameters can be

identified with the relevant data. The main advantage of structural estimation is that interpreting

any empirical results—even the sign of a regression coefficient—usually requires a model. Struc-

tural estimation puts the model first and makes the model explicit, and thus the results are often

simpler to interpret than the results from reduced form regressions, which are often the product of

a verbal model.

Further work would be interesting in other areas as well. For largely historical reasons, the areas

of investment, leverage, and finance constraints have seen the bulk of research in dynamic corporate

finance. As noted above, very little work has been done on payout, and the studies that have tackled

issues in executive compensation are few (e.g., Taylor 2010, 2012; Brisley 2006; He 2009; Noe and

Rebello 2012). Other areas that have seen scant work include product markets (Lambrecht 2001,

2004; Morellec and Zhdanov 2008) and entrepreneurship (Wang, Wang, and Yang 2012). In general,

any area of corporate finance could be amenable to dynamic models and structural estimation.

Our essay should be seen in the proper light. In our review of dynamic corporate finance we

120
do not argue against conventional modeling apparatus, or worse, intuition developed in the process

of working on such models. We are very much in favor of non-technical economic intuition and

consider it to be of utmost importance. On the one hand, simple static models are useful for

understanding economic mechanisms in their simplest forms and getting benchmark results. On

the other hand, every dynamic model, however complicated it may seem, should be explained by

straightforward economic intuition. Ability to explain intuition to one’s grandmother is a non-

trivial test of scientific inquiry in social sciences. Moreover, we certainly do not claim that there is

no benefit from simple intuitive thinking without any formal models. Instead, we believe it should

be, and most often is, the starting stage for more complicated and realistic models.

We also do not argue against conventional reduced form or quasi-experimental empirical meth-

ods. Instead, we wish to explain how structural methods can add to our understanding of corporate

finance questions in new and interesting ways. In particular, parameter estimates obtained from

structural estimations are useful for counterfactual (i.e. “what-if”) analysis that can be used at a

minimum to further our understanding of how firms respond to primitives or at best to evaluate

policies. In addition, the number of high-quality natural experiments available to corporate finance

researchers is likely much smaller than the number of interesting questions to be asked, so that

structural estimation often o↵ers a more feasible alternative for understanding these interesting

questions.

Our biggest claim is twofold. First, however penetrating static models can sometimes be, they

are only a starting block in our understanding of any phenomenon. The first question that should

be asked is whether the results are robust in a dynamic world. Second, pure qualitative judgment

cannot take us too far in our desire to distinguish between various economic mechanisms, and

therefore a closer link between modeling framework and empirical methods should be established.

In a way, we should not only take our models more seriously, and expect more from them, but also

be able to criticize them with better precision.

121
References
Abel, Andrew B., and Janice C. Eberly, 1994, A unified model of investment under uncertainty,
American Economic Review 84, 1369–1384.

Acharya, Viral, Sergei A. Davydenko, and Ilya A. Strebulaev, 2012, Cash holdings and credit risk,
Review of Financial Studies forthcoming.

Aguirregabiria, Victor, and Pedro Mira, 2007, Sequential estimation of dynamic discrete games,
Econometrica 75, 1–53.

Albuquerque, Rui, and Enrique Schroth, 2010, Quantifying private benefits of control from a struc-
tural model of block trades, Journal of Financial Economics 96, 33–55.

Alti, Aydoḡan, 2003, How sensitive is investment to cash flow when financing is frictionless?, Journal
of Finance 58, 707–722.

Altinkilic, Oya, and Robert S. Hansen, 2000, Are there economies of scale in underwriter spreads?
Evidence of rising external financing costs, Review of Financial Studies 13, 191–218.

Anderson, Ronald W., and Andrew Carverhill, 2012, Corporate liquidity and capital structure,
Review of Financial Studies 25, 797–837.

Anderson, Ronald W., and Suresh Sundaresan, 1996, Design and valuation of debt contracts, Review
of Financial Studies 9, 37–68.

Andrade, Gregor, and Steven N. Kaplan, 1998, How costly is financial (not economic) distress?
evidence from highly leveraged transactions that became distressed, Journal of Finance 53, 1443–
1493.

Bajari, Patrick, C. Lanier Benkard, and Jonathan Levin, 2007, Estimating dynamic models of
imperfect competition, Econometrica 75, 1331–1370.

Baker, Malcolm, and Je↵rey Wurgler, 2002, Market timing and capital structure, Journal of Finance
57, 1–32.

Berk, Jonanthan, and Peter DeMarzo, 2011, Corporate Finance (Pearson).

Bhamra, Harjoat S., Lars-Alexander Kuehn, and Ilya A. Strebulaev, 2010a, The aggregate dynamics
of capital structure and macroeconomic risk, Review of Financial Studies 23, 4187–4241.

, 2010b, The levered equity risk premium and credit spreads: A unified framework, Review
of Financial Studies 23, 645–703.

Black, Fisher, and Myron Scholes, 1973, The pricing of options and corporate liabilities, Journal
of Political Economy 1981, 637–654.

Bolton, Patrick, Hui Chen, and Neng Wang, 2011, A unified theory of tobin’s q, corporate invest-
ment, financing, and risk management, Journal of Finance 66, 1545–1578.

, 2012, Market timing, investment, and risk management, Journal of Financial Economics
forthcoming.

122
Bond, Stephen, and Costas Meghir, 1994, Dynamic investment models and the firm’s financial
policy, Review of Economic Studies 61, 197.

Boyle, Glenn W., and Graeme A. Guthrie, 2003, Investment, uncertainty, and liquidity, Journal of
Finance 58, 2143–2166.

Bradley, Michael, Gregg A. Jarrell, and E. Han Kim, 1984, On the existence of an optimal capital
structure: Theory and evidence, Journal of Finance 39, 857–878.

Brander, James A., and Tracy R. Lewis, 1986, Oligopoly and financial structure: The limited
liability e↵ect, American Economic Review pp. 956–970.

Brennan, Michael J., and Eduardo S. Schwartz, 1985, A new approach to evaluating natural resource
investments, Midland Corporate Finance Journal 3, 37–47.

Brisley, Neil, 2006, Executive stock options: Early exercise provisions and risk-taking incentives,
Journal of Finance 61, 2487–2509.

Broadie, Mark, Mikhail Chernov, and Suresh Sundaresan, 2007, Optimal debt and equity values in
the presence of chapter 7 and chapter 11, Journal of Finance 62, 1341–1377.

Campello, Murillo, and Dirk Hackbarth, 2012, The firm-level credit multiplierq, Journal of Finan-
cial Intermediation 21, 446–472.

Carlson, Murray, and Ali Lazrak, 2010, Leverage choice and credit spreads when managers risk
shift, Journal of Finance 65, 2323–2362.

Chen, Hui, 2010, Macroeconomic conditions and the puzzles of credit spreads and capital structure,
Journal of Finance 65, 2171–2212.

, Jianjun Miao, and Neng Wang, 2010, Entrepreneurial finance and nondiversifiable risk,
Review of Financial Studies 23, 4348–4388.

Childs, Paul D., David C. Mauer, and Steven H. Ott, 2005, Interactions of corporate financing
and investment decisions: The e↵ects of agency conflicts, Journal of Financial Economics 76,
667–690.

Childs, Paul D., Steven H. Ott, and Alexander J. Triantis, 1998, Capital budgeting for interrelated
projects: A real options approach, Journal of Financial and Quantitative Analysis 33.

Cochrane, John H., 2005, Asset Pricing: (Revised) (Princeton University Press: Princeton, NJ).

Coles, Je↵rey L., Michael L. Lemmon, and J. Felix Meschke, 2012, Structural models and endo-
geneity in corporate finance: The link between managerial ownership and corporate performance,
Journal of Financial Economics 103, 149 – 168.

Cooley, Thomas F., and Vincenzo Quadrini, 2001, Financial markets and firm dynamics, American
Economic Review 91, 1286–1310.

Cooper, Russell, and Joao Ejarque, 2003, Financial frictions and investment: Requiem in q, Review
of Economic Dynamics 6, 710–728.

123
Covas, Francis, and Wouter J. Den Haan, 2007, Cyclical behavior of debt and equity using a panel
of canadian firms, Manuscript, Bank of Canada.

Covas, Francisco, and Wouter J. Den Haan, 2011, The cyclical behavior of debt and equity finance,
American Economic Review 101, 877–899.

Danis, András, Daniel Rettl, and Toni M. Whited, 2012, Testing dynamic tradeo↵ theory: Evidence
from rebalancing points, Manuscript University of Rochester.

Davydenko, Sergei, Ilya A. Strebulaev, and Xiaofei Zhao, 2012, A market-based study of the cost
of default, Review of Financial Studies 25, 2959–2999.

Davydenko, Sergei A., 2010, What triggers default? a study of the default boundary, Manuscript,
University of Toronto.

DeAngelo, Harry, Linda DeAngelo, and Toni M. Whited, 2011, Capital structure dynamics and
transitory debt, Journal of Financial Economics 99, 235–261.

DeAngelo, Harry, and Ronald W. Masulis, 1980, Optimal capital structure under corporate and
personal taxation, Journal of Financial Economics 8, 3–29.

Décamps, Jean-Paul, Thomas Mariotti, and Stéphane Villeneuve, 2005, Investment timing under
incomplete information, Mathematics of Operations Research 30, 472–500.

, 2009, Investment timing under incomplete information: Erratum, Mathematics of Opera-

tions Research 34, 255–256.

Dimopoulos, Theodosios, and Stefano Sacchetto, 2011, Preemptive Bidding, Target Resistance, and
Takeover Premiums, Manuscript, Carnegie Mellon University.

Dixit, Avinash, 1989, Entry and exit decisions under uncertainty, Journal of Political Economy pp.
620–638.

Dixit, Avinash K., and Robert S. Pindyck, 1994, Investment under Uncertainty (Princeton Uni-
versity Press: Princeton, NJ).

Duffie, Darrell, and Kenneth J Singleton, 1993, Simulated moments estimation of markov models
of asset prices, Econometrica 61, 929–52.

Dumas, Bernard, 1991, Super contact and related optimality conditions, Journal of Economic
Dynamics and Control 15, 675–685.

Erickson, Timothy, and Toni M. Whited, 2012, Treating measurement error in Tobin’s q, Review
of Financial Studies 25, 1286–1329.

Fama, Eugene F., and Kenneth R. French, 2002, Testing trade-o↵ and pecking order predictions
about dividends and debt, Review of Financial Studies 15, 1–33.

Fan, Hua, and Suresh Sundaresan, 2000, Debt valuation, renegotiation, and optimal dividend policy,
Review of Financial Studies 13, 1057–1099.

Fazzari, Steven M., R. Glenn Hubbard, and Bruce C. Petersen, 1988, Financing constraints and
corporate investment, Brookings Papers on Economic Activity 1988, 141–206.

124
Fischer, Edwin O., Robert Heinkel, and Josef Zechner, 1989, Dynamic capital structure choice:
Theory and tests, Journal of Finance 44, 19–40.

Franks, Julian R., and Walter N. Torous, 1989, An empirical investigation of us firms in reorgani-
zation, Journal of Finance pp. 747–769.

Gamba, Andrea, and Alexander Triantis, 2008, The value of financial flexibility, Journal of Finance
63, 2263–2296.

Gantchev, Nickolay, 2011, The costs of shareholder activism: Evidence from a sequential decision
model, Manuscript, University of North Carolina.

Glover, Brent, 2011, The expected cost of default, Manuscript, Carnegie Mellon University.

Goldstein, Robert, Nengjiu Ju, and Hayne Leland, 2001, An ebit-based model of dynamic capital
structure, Journal of Business 74, 483–512.

Gomes, Joao, and Dmitry Livdan, 2004, Optimal diversification: Reconciling theory and evidence,
Journal of Finance 59, 507–535.

Gomes, Joao F., 2001, Financing investment, American Economic Review 91, 1263–1285.

, and Lukas Schmid, 2010, Levered returns, Journal of Finance 65, 467–494.

Gorbenko, Alexander S., and Ilya A. Strebulaev, 2010, Temporary versus permanent shocks: Ex-
plaining corporate financial policies, Review of Financial Studies 23, 2591–2647.

Gourieroux, Christian S., Alain Monfort, and Eric Renault, 1993, Indirect inference, Journal of
Applied Econometrics 8, S85–S118.

Graham, John R., 2000, How big are the tax benefits of debt?, Journal of Finance 55, 1901–1941.

, 2003, Taxes and corporate finance: A review, Review of Financial Studies 16, 1075–1129.

, and Campbell R. Harvey, 2001, The theory and practice of corporate finance: Evidence
from the field, Journal of Financial Economics 60, 187–243.

Grenadier, Stecen R., 1995, Valuing lease contracts a real-options approach, Journal of Financial
Economics 38, 297–331.

, 2002, Option exercise games: An application to the equilibrium investment strategies of

firms, Review of Financial Studies 15, 691–721.

Grenadier, Steven R., and Andrey Malenko, 2010, A Bayesian approach to real options: The case
of distinguishing between temporary and permanent shocks, Journal of Finance 65, 1949–1986.

Grenadier, Steven R., and Neng Wang, 2005, Investment timing, agency, and information, Journal
of Financial Economics 75, 493–533.

, 2007, Investment under uncertainty and time-inconsistent preferences, Journal of Financial

Economics 84, 2–39.

Grenadier, Steven R., and Allen M. Weiss, 1997, Investment in technological innovations: An option
pricing approach, Journal of Financial Economics 44, 397–416.

125
Guo, Xin, Jianjun Miao, and Erwan Morellec, 2005, Irreversible investment with regime shifts,
Journal of Economic Theory 122, 37–59.

Hackbarth, Dirk, 2009, Determinants of corporate borrowing: A behavioral perspective, Journal of

Corporate Finance 15, 389–411.

, 2010, Managerial traits and capital structure decisions, Journal of Financial and Quanti-
tative Analysis 43, 843.

, Christopher A. Hennessy, and Hayne E. Leland, 2007, Can the trade-o↵ theory explain
debt structure?, Review of Financial Studies 20, 1389–1428.

Hackbarth, Dirk, Richmond Mathews, and David Robinson, 2012, Capital structure, product mar-
ket dynamics, and the boundaries of the firm, Manuscript, Duke University.

Hackbarth, Dirk, and David C. Mauer, 2012, Optimal priority structure, capital structure, and
investment, Review of Financial Studies 25, 747–796.

Hackbarth, Dirk, and Jianjun Miao, 2012, The dynamics of mergers and acquisitions in oligopolistic
industries, Journal of Economic Dynamics and Control 36, 585–609.

, and Erwan Morellec, 2006, Capital structure, credit risk, and macroeconomic conditions,
Journal of Financial Economics 82, 519–550.

Hackbarth, Dirk, and Erwan Morellec, 2008, Stock returns in mergers and acquisitions, Journal of
Finance 63, 1213–1252.

Han, Chirok, and Peter C. B. Phillips, 2010, GMM estimation for dynamic panels with fixed e↵ects
and strong instruments at unity, Econometric Theory 26, 119–151.

Hansen, Lars Peter, and Kenneth J. Singleton, 1982, Generalized instrumental variables estimation
of nonlinear rational expectations models, Econometrica pp. 1269–1286.

Hayashi, Fumio, 1982, Tobin’s marginal q and average q: A neoclassical interpretation, Economet-
rica 50, 213–224.

He, Zhiguo, 2009, Optimal executive compensation when firm size follows geometric brownian
motion, Review of Financial Studies 22, 859–892.

Hennessy, Christopher A., 2004, Tobin’s q, debt overhang, and investment, Journal of Finance 59,
1717–1742.

, Amnon Levy, and Toni M. Whited, 2007, Testing Q theory with financing frictions, Journal
of Financial Economics 83, 691–717.

Hennessy, Christopher A., Dmitry Livdan, and Bruno Miranda, 2010, Repeated signaling and firm
dynamics, Review of Financial Studies 23, 1981–2023.

Hennessy, Christopher A., and Toni M. Whited, 2005, Debt dynamics, Journal of Finance 60,
1129–1165.

, 2007, How costly is external financing? Evidence from a structural estimation, Journal of
Finance 62, 1705–1745.

126
Hotz, V. Joseph, and Robert A. Miller, 1993, Conditional choice probabilities and the estimation
of dynamic models, Review of Economic Studies 60, 497.

Ingram, Beth F., and Bong-Soo Lee, 1991, Simulation and estimation of time series models, Journal
of Econometrics 47, 197–205.

Jermann, Urban, and Vincenzo Quadrini, 2012, Macroeconomic e↵ects of financial shocks, Ameri-
can Economic Review, forthcoming.

Kane, Alex, Alan J. Marcus, and Robert L. McDonald, 1984, How big is the tax advantage to
debt?, Journal of Finance 39, 841–853.

Kang, Ari, Richard Lowery, and Malcolm Wardlaw, 2012, Bank failures and regulatory incentives:
A structural estimation, Manuscript, University of Texas, Austin.

Kang, Qiang, Qiao Liu, and Rong Qi, 2010, The Sarbanes-Oxley act and corporate investment: A
structural assessment, Journal of Financial Economics 96, 291–305.

Kim, E. Han, 1989, Optimal capital structure in Miller’s equilibrium, in Sudipto Bhattacharya, and
George Constantinides, ed.: Financial Markets and Incomplete Information . pp. 36–48 (Rowman
and Littlefield: Totowa, NJ).

Klein, Manuel, 2009, Comment on “Investment timing under incomplete information”, Mathematics
of Operations Research 34, 249–254.

Korajczyk, Robert A., and Amnon Levy, 2003, Capital structure choice: Macroeconomic conditions
and financial constraints, Journal of Financial Economics 68, 75–109.

Korteweg, Arthur G., 2010, The net benefits to leverage, Journal of Finance 65, 2137–2170.

, and Ilya A. Strebulaev, 2012, An (s, s) model of capital structure: Theory and evidence,
Manuscript, Stanford University.

Kuehn, Lars Alexander, and Lukas Schmid, 2011, Investment-based corporate bond pricing,
Manuscript, Duke University.

Kurshev, Alexander, and Ilya A. Strebulaev, 2006, Firm size and capital structure, Manuscript,
Stanford University.

Lambrecht, Bart M., 2001, The impact of debt financing on entry and exit in a duopoly, Review of
Financial Studies 14, 765–804.

, 2004, The timing and terms of mergers motivated by economies of scale, Journal of
Financial Economics 72, 41–62.

, and Stewart C. Myers, 2007, A theory of takeovers and disinvestment, Journal of Finance
62, 809–845.

, 2012, A Lintner model of payout and managerial rents, Journal of Finance 67, 1761–1810.

Leland, Hayne E., 1994, Corporate debt value, bond covenants, and optimal capital structure,
Journal of Finance 49, 1213–1252.

127
Li, Erica X. N., Dmitry Livdan, and Lu Zhang, 2009, Anomalies, Review of Financial Studies 22,
4301–4334.

Lin, Chen, Yue Ma, and Yuhai Xuan, 2011, Ownership structure and financial constraints: Evidence
from a structural estimation, Journal of Financial Economics 102, 416–431.

Lintner, John, 1956, Distribution of incomes of corporations among dividends, retained earnings,
and taxes, American Economic Review 46, 97–113.

Liu, Laura Xiaolei, Toni M. Whited, and Lu Zhang, 2009, Investment-based expected stock returns,
Journal of Political Economy 117, 1105–1139.

Livdan, Dmitry, Horacio Sapriza, and Lu Zhang, 2009, Financially constrained stock returns, Jour-
nal of Finance 64, 1827–1862.

Love, Inessa, 2003, Financial development and financing constraints: International evidence from
the structural investment model, Review of Financial Studies 16, 765–791.

Lucas, Robert E., 1978, Asset prices in an exchange economy, Econometrica 46, 1429–1446.

, and Edward C Prescott, 1971, Investment under uncertainty, Econometrica 39, 659–81.

Mankiw, N. Gregory, 1985, Small menu costs and large business cycles: A macroeconomic model
of monopoly, Quarterly Journal of Economics 100, 529–537.

Matvos, Gregor, and Amit Seru, 2011, Resource allocation within firms and financial market dis-
location: Evidence from diversified conglomerates, manuscript.

McDonald, Robert, and Daniel Siegel, 1986, The value of waiting to invest, Quarterly Journal of
Economics 101, 707.

McDonald, Robert L., and Daniel R. Siegel, 1985, Investment and the valuation of firms when there
is an option to shut down, International Economic Review 26, 331–349.

Mehra, Rajnish, and Edward C. Prescott, 1985, The equity premium: A puzzle, Journal of Mone-
tary Economics 15, 145–161.

Mella-Barral, Pierre, and William Perraudin, 1997, Strategic debt service, Journal of Finance 52,
531–556.

Merton, Robert C., 1974, On the pricing of corporate debt: The risk structure of interest ratess,
Journal of Finance 29, 449–470.

Miao, Jianjun, 2005, Optimal capital structure and industry dynamics, Journal of Finance 60,
2621–2659.

Miller, Merton H., 1977, Debt and taxes, Journal of Finance 32, 261–275.

Miranda, Mario J., and Paul L. Fackler, 2002, Applied Computational Economics and Finance
(MIT Press: Cambridge, MA).

Modigliani, Franco, and Merton H. Miller, 1958, The cost of capital, corporation finance and the
theory of investment, American Economic Review 48, 261–297.

128
 , 1963, Corporate income taxes and the cost of capital: A correction, American Economic
Review 53, 433–443.

Morellec, Erwan, 2001, Asset liquidity, capital structure, and secured debt, Journal of Financial
Economics 61, 173–206.

, 2004, Can managerial discretion explain observed leverage ratios?, Review of Financial
Studies 17, 257–294.

, Boris Nikolov, and Norman Schürho↵, 2012, Corporate governance and capital structure
dynamics, Journal of Finance 67, 803–848.

Morellec, Erwan, and Norman Schürho↵, 2010, Dynamic investment and financing under personal
taxation, Review of Financial Studies 23, 101–146.

Morellec, Erwan, and Norman Schürho↵, 2011, Corporate investment and financing under asym-
metric information, Journal of Financial Economics 99, 262–288.

Morellec, Erwan, and Alexei Zhdanov, 2008, Financing and takeovers, Journal of Financial Eco-
nomics 87, 556–581.

Moyen, Nathalie, 2004, Investment-cash flow sensitivities: Constrained versus unconstrained firms,
Journal of Finance 59, 2061–2092.

Myers, Stewart C., and Nicholas S. Majluf, 1984, Corporate financing and investment decisions
when firms have information that investors do not have, Journal of Financial Economics 13,
187–221.

Newey, Whitney, and Daniel McFadden, 1994, Large sample estimation and hypothesis testing, in
Robert Engle, and Daniel McFadden, ed.: Handbook of Econometrics, Vol. 4 . pp. 2111–2245
(North-Holland: Amsterdam).

Nikolov, Boris, and Toni M. Whited, 2012, Agency conflicts and cash: Estimates from a structural
model, Manuscript University of Rochester.

Noe, Thomas H., and Michael J. Rebello, 2012, Optimal corporate governance and compensation
in a dynamic world, Review of Financial Studies 25, 480–521.

Page, Beau, 2012, CEO ownership and firm value: Evidence from a structural estimation,
Manuscript, University of Rochester.

Pakes, Ariel, and David Pollard, 1989, Simulation and the asymptotics of optimization estimators,
Econometrica 57, 1027–1057.

Pawlina, Grzegorz, 2010, Underinvestment, capital structure, and strategic debt restructuring,
Journal of Corporate Finance 16, 679–702.

Pratap, Sangeet, and Silvio Rendon, 2003, Firm investment in imperfect capital markets: A struc-
tural estimation, Review of Economic Dynamics 6, 513–545.

Rajan, Raghuram G., and Luigi Zingales, 1995, What do we know about capital structure? Some
evidence from international data, Journal of Finance 50.

129
Riddick, Leigh A., and Toni M. Whited, 2009, The corporate propensity to save, Journal of Finance
64, 1729–1766.

Rust, John, 1987, Optimal replacement of GMC bus engines: An empirical model of Harold Zurcher,
Econometrica pp. 999–1033.

Sarkar, Sudipto, 2000, On the investment-uncertainty relationship in a real options model, Journal
of Economic Dynamics and Control 24, 219–225.

, 2003, The e↵ect of mean reversion on investment under uncertainty, Journal of Economic
Dynamics and Control 28, 377–396.

Schaefer, Stephen M., and Ilya A. Strebulaev, 2008, Structural models of credit risk are useful:
Evidence from hedge ratios on corporate bonds, Journal of Financial Economics 90, 1–19.

Schmid, Lukas, 2011, A quantitative dynamic agency model of financing constraints, Manuscript,
Duke University.

Scholes, Myron S., and Mark A. Wolfson, 1992, Taxes and Business Strategy: A Planning Approach
(Prentice Hall: Englewood Cli↵s, NJ).

Schroth, Enrique, Gustavo Suarez, and Lucian A. Taylor, 2012, Dynamic debt runs: Evidence from
a structural estimation, Manuscript, University of Pennsylvania.

Sorensen, Morten, 2007, How smart is smart money: A two-sided matching model of venture
capital, Journal of Finance 62, 2725–2762.

Stokey, Nancy L., 2009, The Economics of Inaction: Stochastic Control Models with Fixed Costs
(Princeton University Press: Princeton, NJ).

, Robert E. Lucas, and Edward C. Prescott, 1989, Recursive Methods in Economic Dynamics
(Harvard University Press: Cambridge, MA).

Strebulaev, Ilya A., 2007, Do tests of capital structure theory mean what they say?, Journal of
Finance 62, 1747–1787.

, Haoxiang Zhu, and Pavel Zryomov, 2012, Dynamic information asymmetry, financing, and
investment, Manuscript, Stanford University.

Summers, Lawrence, H., 1987, Investment incentives and the discounting of depreciation allowances,
in Martin Feldstein, ed.: The E↵ects of Taxation on Capital Accumulation (Chicago University
Press: Chicago).

Tauchen, George, 1986, Finite state markov-chain approximations to univariate and vector autore-
gressions, Economics Letters 20, 177–181.

, and Robert Hussey, 1991, Quadrature-based methods for obtaining approximate solutions
to nonlinear asset pricing models, Econometrica 59, 371–396.

Taylor, Lucian A., 2010, Why are CEOs rarely fired? evidence from structural estimation, Journal
of Finance 65, 2051–2087.

130
 , 2012, CEO pay and CEO power: Evidence from a dynamic learning model, Journal of
Financial Economics forthcoming.

Titman, Sheridan, and Sergey Tsyplakov, 2007, A dynamic model of optimal capital structure,
Review of Finance 11, 401–451.

Titman, Sheridan, and Roberto Wessels, 1988, The determinants of capital structure choice, Journal
of Finance pp. 1–19.

Tsyplakov, Sergey, 2008, Investment frictions and leverage dynamics, Journal of Financial Eco-
nomics 89, 423–443.

Wang, Chong, Neng Wang, and Jinqiang Yang, 2012, Dynamics of entrepreneurship under incom-
plete markets, Journal of Financial Economics, forthcoming.

Warusawitharana, Missaka, 2008, Corporate asset purchases and sales: Theory and evidence, Jour-
nal of Financial Economics 87, 471–497.

, 2011, Research and development, profits and firm value: A structural estimation,
Manuscript, Federal Reserve Board.

Warusawitharna, Missaka, and Toni M. Whited, 2012, Equity market misvaluation and firm finan-
cial policies, Manuscript, University of Rochester.

Whited, Toni M., 1992, Debt, liquidity constraints, and corporate investment: Evidence from panel
data, Journal of Finance pp. 1425–1460.

, and Guojun Wu, 2006, Financial constraints risk, Review of Financial Studies 19, 531–559.

Wright, Jonathan, 2003, Detecting lack of identification in GMM, Econometric Theory 19, 322–330.

Zwiebel, Je↵rey, 1996, Dynamic capital structure under managerial entrenchment, American Eco-
nomic Review pp. 1197–1215.

131
 Table 1: Symbol Definitions
Symbol Definition
Section 2
Xt project or firm cash flows
µ the instantaneous growth rate of cash flows
the instantaneous volatility of cash flows
dWtQ a standard Brownian motion process under the risk-neutral measure
I the cost (amount) of investment
r the risk-free interest rate
V (·) firm value
A(·) the date-0 value of an Arrow-Debreu security
⇠ 1 , ⇠2 roots of the fundamental quadratic equation
St the value of equity
Dt the value of debt
XD the default threshold
c the coupon rate of perpetual debt
a multiplicative constant defining the default threshold
↵ deadweight default costs
F the unlevered firm value
TB the present value of the tax benefits of debt
DC the present value of default costs
L(·) the market leverage ratio
QM L(·) quasi-market leverage ratio
XR refinancing threshold
TX time that X is reached for the first time
R(·) value of the debt payo↵ at refinancing
! premium associated with the right to recall debt
q the proportional cost of issuing debt
⇡t realized cash flow
✏t temporary component of realized cash flow
intensity of a Poisson process
(·) Poisson density
Ts time of arrival of a temporary shock
Tr time of reversal of a temporary shock

132
Section 3
e(·) cash flows to equity holders
k capital stock
z productivity/demand shock
⇡(·) profit function
(·) investment adjustment cost function
I investment
V (·) equity or firm value
r risk free interest rate
Et expectations operator conditional on information known at time t
rate of capital depreciation
shadow value of capital
" innovation to the AR(1) process for z
" standard deviation of "
⇢ serial correlation of the AR(1) process for z
g(z 0 | z) transition function for z
✓ profit function curvature
h(·) policy function
0, 1 fixed and quadratic investment adjustment cost parameters
⌘(·) external financing/equity issuance costs
⌘ 0 , ⌘1 fixed and linear issuance cost parameters
p cash
b debt
s fraction of debt that can be collateralized
⌧ corporate tax rate on profits and interest
⇣ Lagrange multiplier
r̃ interest rate on risky debt
Section 4
x real data vector
M (x) vector of real data moments
y simulated data vector
m(y, ) vector of simulated moments
parameter vector to be estimated
W weight matrix
Q(x, y, ) SMM objective function
N number of observations in x
S ratio of the dimension of y to x
ui random e↵ect
f (·) generic symbol for a density

133
 Table 2: Comparative Statics of the Dynamic Investment Model
This table presents the comparative statics of the dynamic investment model. To derive
the table, we assume that X0 < XI . For subsequent comparison with the standard NPV
criterion, we either assume that XIN P V X0 , where XIN P V is the cash flow level at which
NPV is zero, or that the + and signs indicate weak (greater or equal, smaller or equal)
rather than strong relations.

Sign of change in variable

for an increase in:
Variable µ I r
XI + + +
XIN P V n.a. + +
XI XIN P V + + + +
V (X0 ) + +
V N P V (X0 ) + n.a.
V (X0 ) V N P V (X0 ) + + +

134
Table 3: Comparative Statics of the Optimal Static Capital Structure Model

Sign of change in variable

for an increase in:
Variable µ r ↵ ⌧
Endogenous case
D + n.a. n.a.
c +. /+ +
L + + +
XD + + +
V (X0 ) F (X0 )
F (X0 ) + + +
Exogenous case
c /+ + +
L + +
XD /+ + +
V (X0 ) F (X0 )
F (X0 ) + +

135
 Table 4: Static Capital Structure

This table reports the optimal coupon, c, date-0 leverage ratio, L, and the di↵erence between
levered and unlevered firm values for the optimal static capital structure model. The benchmark
set of parameters is: r = 0.05, µ = 0.02, = 0.25, ⌧ = 0.2, and ↵ = 0.1. The first three columns
are for the exogenous default case, and the last three columns are for the endogenous default case.

V F V F
c L F % c L F %
Benchmark 0.32 0.25 0.03 1.36 0.70 0.11
= 0.15 0.50 0.37 0.05 1.28 0.75 0.14
= 0.40 0.16 0.13 0.01 1.73 0.66 0.09
⌧ = 0.35 0.40 0.34 0.07 1.49 0.77 0.25
⌧ = 0.10 0.22 0.16 0.01 1.14 0.59 0.04
↵ = 0.5 0.13 0.10 0.01 0.84 0.48 0.07
↵ = 0.05 0.39 0.31 0.03 1.48 0.74 0.12

Table 5: Dynamic Capital Structure

This table reports the optimal coupon, c, date-0 leverage ratio, L, and the di↵erence between
levered and unlevered firm values for the optimal dynamic and static capital structure models. The
benchmark set of parameters is: r = 0.05, µ = 0.02, = 0.25, ⌧ = 0.2, and ↵ = 0.1. The first three
columns are for the endogenous default case of the static capital structure model and the last three
columns are for the dynamic capital structure model.

Static Dynamic
V F V F
c L F % c L F %
Benchmark 1.33 0.69 0.10 1.12 0.54 0.18
= 0.15 1.26 0.75 0.13 1.13 0.63 0.20
= 0.40 1.67 0.65 0.08 1.24 0.45 0.19
⌧ = 0.35 1.48 0.77 0.24 1.32 0.57 0.51
⌧ = 0.10 1.08 0.57 0.03 0.88 0.46 0.06
↵ = 0.5 0.82 0.48 0.06 0.67 0.37 0.11
↵ = 0.05 1.45 0.73 0.11 1.22 0.57 0.19
q = 0.005 1.35 0.69 0.10 1.06 0.51 0.19
q = 0.05 1.19 0.66 0.07 1.06 0.57 0.11

136
 Table 6: True Dynamics of Leverage in Dynamic Capital Structure

This table reports the true dynamics of leverage in the dynamic capital structure model. The
benchmark set of parameters is: r = 0.05, µ = 0.02, = 0.25, ⌧ = 0.2, and ↵ = 0.1. The first
three columns are for the exogenous default case, and the last three columns are for the endogenous
default case.

Static L L(X0 ) mean LT D st. dev. LT D min LT D

Benchmark 0.70 0.54 0.66 0.17 0.35
= 0.15 0.76 0.63 0.65 0.12 0.44
= 0.40 0.66 0.45 0.68 0.20 0.28
⌧ = 0.35 0.77 0.56 0.70 0.15 0.42
⌧ = 0.10 0.60 0.46 0.59 0.18 0.27
↵ = 0.5 0.49 0.37 0.54 0.18 0.27
↵ = 0.05 0.74 0.57 0.68 0.16 0.37
q = 0.005 0.70 0.51 0.66 0.16 0.37
q = 0.05 0.72 0.57 0.59 0.20 0.23

Table 7: Cross-Sectional regressions of Leverage on Profitability

This table reports the results of cross-sectional regressions on the level of the quasi-market leverage
ratio, QM L, on profitability and control variables. The results are from Strebulaev (2007).

QM L(X0 ) QM LT D QM LT D , 10% QM LT D , 90%

Constant 0.24 0.62 0.55 0.71
Profitability 5.88 -0.78 -1.53 -0.22
Control variables Yes Yes NaN NaN
R2 0.89 0.08 0.06 0.10

137
 Table 8: Capital Structure with Strategic Renegotiation

This table reports the optimal bank coupon, cB , optimal public debt coupon, c, date-0 leverage
ratio, L, and the di↵erence between levered and unlevered firm value for the optimal static capital
structure model with only public debt, only bank debt, and both bank and public debt. The
benchmark set of parameters is: r = 0.05, µ = 0.02, = 0.25, ⌧ = 0.2, and ↵ = 0.1. The first three
columns are for the case of only public debt; the next three columns are for case of only bank debt;
and the last three columns are for the case of both bank and public debt.

Only public debt Only bank debt Both bank and public debt
c L (V F )/F cB L (V F )/F c cB L (V F )/F
Benchmark1.36 0.70 0.11 2.29 0.76 0.18 0.16 2.29 0.79 0.19
µ = 0.005 0.94 0.69 0.10 1.68 0.76 0.18 0.10 1.68 0.78 0.19
µ = 0.035 2.65 0.71 0.11 4.21 0.76 0.18 0.35 4.21 0.80 0.20
= 0.15 1.28 0.75 0.14 1.67 0.76 0.18 0.23 1.67 0.83 0.20
= 0.40 1.73 0.66 0.09 3.60 0.76 0.18 0.11 3.60 0.77 0.19
⌧ = 0.35 1.49 0.77 0.25 1.86 0.68 0.32 0.36 1.86 0.65 0.38
⌧ = 0.10 1.14 0.59 0.04 2.58 0.83 0.09 0.06 2.58 0.85 0.09
↵ = 0.5 0.84 0.48 0.07 1.27 0.45 0.10 0.35 1.27 0.53 0.13
↵ = 0.05 1.48 0.74 0.12 2.42 0.80 0.19 0.14 2.42 0.82 0.20

138
 0.12

0.1
rH

0.08

0.06
0.02 0.025 0.03 0.035 0.04 0.045 0.05
r

0.8

0.6

0.4
rH

0.2

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
σ

Figure 1: Real options investment model. Comparative statics of the hurdle rate, rH .
This figure shows the comparative statics of the hurdle rate, rH , with respect to the interest rate,
r, and asset volatility, , for both the NPV model (smaller values) and the real options model.

139
 XD
c

r r
(V−F)/F

r r

Figure 2: Optimal static capital structure model. Comparative statics with respect to
the risk-free interest rate. This figure shows the comparative statics of the optimal coupon
rate, c, the default boundary, XD , optimal leverage, L, and the di↵erence in values between levered
and unlevered firms, V F F , all with respect to the risk-free interest rate, r, in the optimal static
capital structure model. The dashed lines show the endogenous default case ( = ⇠1⇠1 1 r r µ ), and
the solid lines show the exogenous default case ( = 1).

140
 XD
c

µ µ
(V−F)/F

µ µ

Figure 3: Optimal static capital structure model. Comparative statics with respect
to the growth rate. This figure shows the comparative statics of the optimal coupon rate, c,
the default boundary, XD , optimal leverage, L, and the di↵erence in values between levered and
unlevered firms, V F F , all with respect to the growth rate, µ, in the optimal static capital structure
model. The dashed lines show the endogenous default case ( = ⇠1⇠1 1 r r µ ), and the solid lines show
the exogenous default case ( = 1).

141
 XD
c

σ σ
(V−F)/F

σ σ

Figure 4: Optimal static capital structure model. Comparative statics with respect
to asset volatility. This figure shows the comparative statics of the optimal coupon rate, c,
the default boundary, XD , optimal leverage, L, and the di↵erence in values between levered and
unlevered firms, V F F , all with respect to asset volatility, , in the optimal static capital structure
model. The dashed lines show the endogenous default case ( = ⇠1⇠1 1 r r µ ), and the solid lines show
the exogenous default case ( = 1).

142
 XD
c

α α
(V−F)/F

α α

Figure 5: Optimal static capital structure model. Comparative statics with respect
to bankruptcy costs. This figure shows the comparative statics of the optimal coupon rate, c,
the default boundary, XD , optimal leverage, L, and the di↵erence in values between levered and
unlevered firms, V F F , all with respect to the bankruptcy cost parameter, ↵, in the optimal static
capital structure model. The dashed lines show the endogenous default case ( = ⇠1⇠1 1 r r µ ), and
the solid lines show the exogenous default case ( = 1).

143
 XD
c

τ τ
(V−F)/F

τ τ

Figure 6: Optimal static capital structure model. Comparative statics with respect to
corporate tax rate. This figure shows the comparative statics of the optimal coupon rate, c,
the default boundary, XD , optimal leverage, L, and the di↵erence in values between levered and
unlevered firms, V F F , all with respect to the corporate income tax rate, ⌧ , in the optimal static
capital structure model. The dashed lines show the endogenous default case ( = ⇠1⇠1 1 r r µ ), and
the solid lines show the exogenous default case ( = 1).

144
 1.5 0.4
1

XD
0.2
c

0.5
0 0
0 0.05 0.1 0 0.05 0.1
q q
10 0.2
(V−F)/F
XR

5 0.1

0 0
0 0.05 0.1 0 0.05 0.1
q q
1

0.5
L

0
0 0.05 0.1
q

Figure 7: Optimal dynamic capital structure model. Comparative statics with respect
to debt issuance cost. This figure shows the comparative statics of the optimal coupon, c, the
default boundary, XD , the refinancing boundary, XR , optimal leverage, L, and the di↵erence in
values between levered and unlevered firms, V F F , all with respect to the debt issuance cost param-
eter, q, in the optimal dynamic capital structure models. The dashed lines show the endogenous
default case, and the solid lines show the exogenous default case.

145
 1.5 1

XD
1 0.5
c

0.5 0
0 0.05 0.1 0 0.05 0.1
q q
10 0.2
(V−F)/F
XR

5 0.1

0 0
0 0.05 0.1 0 0.05 0.1
q q
0.7

0.6
L

0.5
0 0.05 0.1
q

Figure 8: Optimal dynamic and static capital structure models. Comparative statics
with respect to debt issuance cost. This figure shows the comparative statics of the optimal
coupon, c, the default boundary, XD , optimal leverage, L, and the di↵erence in values between
levered and unlevered firms, V F F , all with respect to the debt issuance cost parameter, q, in
the optimal dynamic and static capital structure models. For the dynamic model, the results on
refinancing boundary, XR , are also shown. The dashed lines show the static capital structure model
(with debt issuance costs), and the solid lines show the dynamic capital structure model.

146
 2.5

1.5

0.5

0
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
L

Figure 9: Dynamic capital structure model. The cross-sectional distribution of the

leverage ratio in true dynamics. The figure shows the cross-sectional distribution of the
leverage ratio, L, in true dynamics in the dynamic capital structure model.

147
 Figure 10: Policy Functions for Investment Models

3.5
3
2.5
2
1.5
1
0.5
0
-0.5 -0.63 -0.50 -0.38 -0.25 -0.13 0.00 0.13 0.25 0.38 0.50 0.63
-1
Smooth Fixed None

Figure 10 depicts three policy functions for optimal investment. These functions map the current productivity shock
and the steady state capital stock into optimal future investment. The plots are evaluated at the steady state capital
stock, the log productivity shock is on the horizontal axis, and the rate of investment is on the vertical axis. Each
policy function is for a version of the simple investment model from Section 3.1. One version contains no adjustment
costs, one contains convex, smooth adjustment costs, and one contains fixed adjustment costs.

148
 Figure 11: Comparative Statics: Investment Model with Adjustment Costs

0.40 0.4 0.40 0.4

Investment / External Finance

Investment / External Finance

0.35 0.3 0.35 0.3
0.30 0.30

Financing Deficit

Financing Deficit
0.2 0.2
0.25 0.25
0.1 0.1
0.20 0.20
0 0
0.15 0.15
0.10 -0.1 0.10 -0.1

0.05 -0.2 0.05 -0.2

0.00 -0.3 0.00 -0.3
0.40 0.48 0.56 0.64 0.72 0.79 0.87 0.03 0.06 0.08 0.11 0.14 0.16 0.19
Profit Function Curvature Depreciation Rate

Investment External Finance Financing Deficit Investment External Finance Financing Deficit

0.40 0.4 0.40 0.4

Investment / External Finance

Investment / External Finance

0.35 0.3 0.35 0.3
0.30 0.30
Financing Deficit

Financing Deficit
0.2 0.2
0.25 0.25
0.1 0.1
0.20 0.20
0 0
0.15 0.15
0.10 -0.1 0.10 -0.1

0.05 -0.2 0.05 -0.2

0.00 -0.3 0.00 -0.3
0.01 0.06 0.12 0.17 0.22 0.28 0.33 0.30 0.39 0.49 0.58 0.68 0.77 0.87
Shock Standard Deviation Shock Serial Correlation

Investment External Finance Financing Deficit Investment External Finance Financing Deficit

0.40 0.4 0.40 0.4

Investment / External Finance

Investment / External Finance

0.35 0.3 0.35 0.3

0.30 0.30
Financing Deficit

Financing Deficit
0.2 0.2
0.25 0.25
0.1 0.1
0.20 0.20
0 0
0.15 0.15
0.10 -0.1 0.10 -0.1

0.05 -0.2 0.05 -0.2

0.00 -0.3 0.00 -0.3
0.00 0.04 0.08 0.12 0.16 0.20 0.24 0.00 0.04 0.08 0.12 0.16 0.20 0.24
Smooth Adjustment Costs Fixed Adjustment Costs

Investment External Finance Financing Deficit Investment External Finance Financing Deficit

Figure 11 depicts average investment, external financing, and financing deficits as a function of six model parameters:
profit function curvature, the depreciation rate of capital, the standard deviation of the profit shock, the serial
correlation of the profit shock, a smooth adjustment cost parameter, and a fixed adjustment cost parameter. All
quantities are scaled by total firm assets. External financing is defined to be zero if the firm distributes funds to
shareholders and positive otherwise. The financing deficit is defined as internal funds minus investment.

149
 Figure 12: Investment-Cash Flow Sensitivity in a Dynamic Investment Model

14.00 14.00

12.00 12.00
Cash Flow Coefficient

Cash Flow Coefficient

10.00 10.00

8.00 8.00

6.00 6.00

4.00 4.00

2.00 2.00

0.00 0.00
0.00 0.79 1.58 2.37 3.16 3.95 4.74 0.00 0.08 0.16 0.24 0.32 0.39 0.47
Fixed Cost of External Finance Linear Cost of External Finance

14.00 14.00

12.00 12.00
Cash Flow Coefficient

Cash Flow Coefficient

10.00 10.00

8.00 8.00

6.00 6.00

4.00 4.00

2.00 2.00

0.00 0.00
0.40 0.48 0.56 0.64 0.72 0.79 0.87 0.03 0.06 0.08 0.11 0.14 0.16 0.19
Profit Function Curvature Depreciation Rate

14.00 14.00

12.00 12.00
Cash Flow Coefficient

Cash Flow Coefficient

10.00 10.00

8.00 8.00

6.00 6.00

4.00 4.00

2.00 2.00

0.00 0.00
0.01 0.06 0.12 0.17 0.22 0.28 0.33 0.30 0.39 0.49 0.58 0.68 0.77 0.87
Shock Standard Deviation Shock Serial Correlation

8.00
14.00
6.00
12.00
Cash Flow Coefficient

4.00
Cash Flow Coefficient

10.00
2.00
8.00
0.00
6.00 0.00 0.04 0.08 0.12 0.16 0.20 0.24
-2.00
4.00 -4.00
2.00 -6.00
0.00 -8.00
0.00 0.04 0.08 0.12 0.16 0.20 0.24
-10.00
Smooth Adjustment Costs Fixed Adjustment Costs

Figure 12 depicts the coefficient on cash flow in a regression of investment on Tobin’s q, and cash flow, using simulated
data. This coefficient is plotted as a function of the fixed and linear costs of external finance, as well as a function
of profit function curvature, the depreciation rate of capital, the standard deviation of the profit shock, the serial
correlation of the profit shock, a smooth adjustment cost parameter, and a fixed adjustment cost parameter.

150
 Figure 13: Policy Functions for a Dynamic Cash Model

1
0.8
0.6
0.4
0.2
0
-0.2 -0.63 -0.50 -0.38 -0.25 -0.13 0.00 0.13 0.25 0.38 0.50 0.63
-0.4
-0.6
-0.8
Investment Cash Distributions

Figure 13 depicts the policy functions for the model from Section 3.3 These functions map the current log productivity
shock and the steady state capital stock into optimal future investment, optimal future cash balances, and optimal
future distributions to shareholders. The plots are evaluated at the steady state capital stock, and the log productivity
shock is on the horizontal axis. Each variable is scaled by the capital stock, and negative distributions are equivalent
to equity issuances.

151
 Figure 14: Average Cash Balances

0.25 0.25

0.20 0.20
Cash/Assets

Cash/Assets
0.15 0.15

0.10 0.10

0.05 0.05

0.00 0.00
0.00 0.79 1.58 2.37 3.16 3.95 4.74 0.00 0.08 0.16 0.24 0.32 0.39 0.47
Fixed Cost of External Finance Linear Cost of External Finance

0.25 0.25

0.20 0.20
Cash/Assets

Cash/Assets
0.15 0.15

0.10 0.10

0.05 0.05

0.00 0.00
0.40 0.48 0.56 0.64 0.72 0.79 0.87 0.03 0.06 0.08 0.11 0.14 0.16 0.19
Profit Function Curvature Depreciation Rate

0.25 0.25

0.20 0.20
Cash/Assets

Cash/Assets

0.15 0.15

0.10 0.10

0.05 0.05

0.00 0.00
0.01 0.06 0.12 0.17 0.22 0.28 0.33 0.30 0.39 0.49 0.58 0.68 0.77 0.87
Shock Standard Deviation Shock Serial Correlation

0.25 0.35

0.30
0.20
0.25
Cash/Assets

Cash/Assets

0.15 0.20

0.10 0.15

0.10
0.05
0.05

0.00 0.00
0.00 0.04 0.08 0.12 0.16 0.20 0.24 0.00 0.04 0.08 0.12 0.16 0.20 0.24
Smooth Adjustment Costs Fixed Adjustment Costs

Figure 14 depicts the average ratio of cash to assets. This ratio is plotted as a function of the fixed and linear costs
of external finance, as well as a function of profit function curvature, the depreciation rate of capital, the standard
deviation of the profit shock, the serial correlation of the profit shock, a smooth adjustment cost parameter, and a
fixed adjustment cost parameter.

152
 Figure 15: Optimal Debt Policy: Financing Surplus

MC

1 + ⌘1

-
cash b0e=0 L H debt

Figure 15 depicts the first-order conditions for optimality from the model in Section 3.4. The upward sloping line
is the marginal cost (MC) of debt financing. The solid portion of the two horizontal lines is the marginal benefit
schedule. b0e=0 is the level of debt at which the sources of funds equals the uses of funds. Optimal leverage can be
found at the point L.

153
 Figure 16: Optimal Debt Policy: Financing Deficit

MC

1 + ⌘1

-
cash L H b0e=0 debt

Figure 16 depicts the first-order conditions for optimality from the model in Section 3.4. The upward sloping line
is the marginal cost (MC) of debt financing. The solid portion of the two horizontal lines is the marginal benefit
schedule. b0e=0 is the level of debt at which the sources of funds equals the uses of funds. Optimal leverage can be
found at the point H.

154
 Figure 17: Policy Functions for a Dynamic Debt Model

1.5

0.5

0
-0.63 -0.50 -0.38 -0.25 -0.13 0.00 0.13 0.25 0.38 0.50 0.63
-0.5

-1
Investment Debt Distributions Collateral Constraint

Figure 17 depicts the policy functions for the model from Section 3.4 These functions map the current log productivity
shock and the steady state capital stock into optimal future investment, optimal future debt, and optimal future
distributions to shareholders. The plots are evaluated at the steady state capital stock, and the log productivity
shock is on the horizontal axis. Each variable is scaled by the capital stock, and negative distributions are equivalent
to equity issuances.

155
 Figure 18: Average Leverage and Debt Capacity

0.7 1 0.7 1

0.6 0.6
0.8 0.8

Debt Capacity Utilization

Debt Capacity Utilization

0.5 0.5
Net Debt/Assets

Net Debt/Assets
0.4 0.6 0.4 0.6

0.3 0.4 0.3 0.4

0.2 0.2
0.2 0.2
0.1 0.1

0 0 0 0
0.00 0.79 1.58 2.37 3.16 3.95 4.74 0.00 0.08 0.16 0.24 0.32 0.39 0.47
Fixed Cost of External Finance Linear Cost of External Finance

0.8 1 0.5 1
0.7 0.45
0.8 0.4 0.8
Debt Capacity Utilization

Debt Capacity Utilization

0.6
0.35
Net Debt/Assets

Net Debt/Assets
0.5 0.6 0.3 0.6
0.4 0.25
0.3 0.4 0.2 0.4
0.15
0.2
0.2 0.1 0.2
0.1 0.05
0 0 0 0
0.40 0.48 0.56 0.64 0.72 0.79 0.87 0.03 0.06 0.08 0.11 0.14 0.16 0.19
Profit Function Curvature Depreciation Rate

0.8 1 0.6 1
0.7
0.5
0.8 0.8
Debt Capacity Utilization

Debt Capacity Utilization

0.6
Net Debt/Assets

Net Debt/Assets

0.4
0.5 0.6 0.6
0.4 0.3
0.3 0.4 0.4
0.2
0.2
0.2 0.2
0.1
0.1
0 0 0 0
0.01 0.06 0.12 0.17 0.22 0.28 0.33 0.30 0.39 0.49 0.58 0.68 0.77 0.87
Shock Standard Deviation Shock Serial Correlation

0.6 1 0.3 1

0.5 0.25
0.8 0.8
Debt Capacity Utilization

Debt Capacity Utilization

Net Debt/Assets

Net Debt/Assets

0.4 0.2
0.6 0.6
0.3 0.15
0.4 0.4
0.2 0.1

0.2 0.2
0.1 0.05

0 0 0 0
0.00 0.04 0.08 0.12 0.16 0.20 0.24 0.00 0.04 0.08 0.12 0.16 0.20 0.24
Smooth Adjustment Costs Fixed Adjustment Costs

Figure 18 depicts the average ratio of net debt to assets on the left axis and the ratio of debt to debt capacity on
the right axis. These ratios are plotted as a function of the fixed and linear costs of external finance, as well as a
function of profit function curvature, the depreciation rate of capital, the standard deviation of the profit shock, the
serial correlation of the profit shock, a smooth adjustment cost parameter, and a fixed adjustment cost parameter.

156