Dynamic Corporate Finance Theory

Christian Riis Flor

Department of Business and Economics
University of Southern Denmark
February 28, 2012
Contents
Preface ix
I Corporate nance: agents, information, and incentives xi
1 Introduction 1
1.1 What is (dynamic) corporate nance? . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.1 Elements of corporate nance models . . . . . . . . . . . . . . . . . . . . . . 2
1.1.2 Why dynamic corporate nance? . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Structure of this book . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Prerequisites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2 Agents, uncertainty and information 3
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2 Agents in corporate nance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2.1 The economic agent . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2.2 Agents in corporate nance models . . . . . . . . . . . . . . . . . . . . . . . . 3
2.3 Uncertainty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.4 Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.4.1 Simple information structure . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.4.2 Asymmetric information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.4.3 Quality of information: ambiguity aversion . . . . . . . . . . . . . . . . . . . 4
3 Stochastic processes 5
3.1 The continuous stochastic processes . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
3.1.1 The Brownian motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
iii
iv Contents
3.1.2 It o processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
3.1.3 Special processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.1.4 It os Lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.2 Jump processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.2.1 The Poisson process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.2.2 It os Lemma and jumps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.2.3 It os Lemma the general version . . . . . . . . . . . . . . . . . . . . . . . . 22
II Corporate investments real options analysis 23
4 Introduction to corporate investment 25
5 Decision making under uncertainty dynamic programming 27
5.1 Discrete time the general framework . . . . . . . . . . . . . . . . . . . . . . . . . . 27
5.1.1 Discrete time our framework . . . . . . . . . . . . . . . . . . . . . . . . . . 29
5.2 Dynamic programming in continuous time . . . . . . . . . . . . . . . . . . . . . . . . 30
5.3 Binary decision problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
5.3.1 Value matching . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
5.3.2 Smooth pasting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
6 Decision making under uncertainty contingent claims analysis 39
6.1 Contingent claims analysis in continuous time . . . . . . . . . . . . . . . . . . . . . . 39
6.1.1 Decision making under delayed information . . . . . . . . . . . . . . . . . . . 45
6.1.2 Decision making under asymmetric information . . . . . . . . . . . . . . . . . 45
7 Real options analysis 47
7.1 A basic example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
7.2 Combined Brownian motion and Poisson process . . . . . . . . . . . . . . . . . . . . 51
7.3 Research and development projects . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
7.3.1 Valuation of drug development . . . . . . . . . . . . . . . . . . . . . . . . . . 54
7.3.2 Acquisition and development of an information technology project . . . . . . 57
7.4 Other models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
Contents v
8 Competition and the incentive to invest 59
8.1 Limited competition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
8.2 Perfect competition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
III Capital structure theory 61
9 Introduction to capital structure theory 63
10 Introduction to structural models 67
10.1 The Merton model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
10.2 A more general model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
10.2.1 The standard assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
10.2.2 The model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
10.3 Contingent claims pricing by the pde-approach . . . . . . . . . . . . . . . . . . . . . 75
10.4 The probabilistic approach to pricing claims . . . . . . . . . . . . . . . . . . . . . . . 76
10.4.1 Valuing debt with a safety covenant . . . . . . . . . . . . . . . . . . . . . . . 78
10.4.2 Subordinated debt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
10.5 Debt with coupon payments: A main example . . . . . . . . . . . . . . . . . . . . . 81
10.6 Optimal stopping. The high contact or smooth pasting condition . . . . . . . . . . . 83
10.6.1 The case of singular controls . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
10.6.2 Generalized Arrow-Debreu prices and event probabilities . . . . . . . . . . . 86
11 Modeling the tax advantage to debt: trade-o theory 91
11.1 A simple optimal capital structure model . . . . . . . . . . . . . . . . . . . . . . . . 92
11.1.1 The costs and benets of debt . . . . . . . . . . . . . . . . . . . . . . . . . . 92
11.1.2 The interpretation of the state variable . . . . . . . . . . . . . . . . . . . . . 93
11.1.3 Valuation of debt and equity . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
11.1.4 The optimal bankruptcy level . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
11.1.5 A comment on the comparative statics . . . . . . . . . . . . . . . . . . . . . . 98
11.1.6 Optimal coupon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
11.1.7 Introducing a net-worth covenant . . . . . . . . . . . . . . . . . . . . . . . . . 99
11.1.8 Continuous capital restructuring . . . . . . . . . . . . . . . . . . . . . . . . . 100
11.2 A capital structure model with increasing debt payments . . . . . . . . . . . . . . . 100
vi Contents
11.2.1 The interpretation of the drift rate of the earnings . . . . . . . . . . . . . . . 101
11.2.2 Negative net income . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
11.2.3 Increasing the debt service. Upward debt restructurings . . . . . . . . . . . . 103
11.2.4 Reducing the dimension of the model . . . . . . . . . . . . . . . . . . . . . . 105
11.2.5 Main results from allowing more debt . . . . . . . . . . . . . . . . . . . . . . 108
11.3 A dynamic capital structure model without loss of the tax shield . . . . . . . . . . . 110
11.3.1 The model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
11.3.2 Negative net income . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
11.3.3 Some results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
11.4 Dynamic capital structure with debt renegotiations . . . . . . . . . . . . . . . . . . 116
11.4.1 Some results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
11.5 The return premium to leverage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
11.5.1 Another state variable and a comparison . . . . . . . . . . . . . . . . . . . . . 120
11.6 An alternative evolution of earnings . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
12 Analysis of bankruptcy law environments 125
12.1 The possibility to enter and leave chapter 11 . . . . . . . . . . . . . . . . . . . . . . 125
12.1.1 Remembering past defaults . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
12.1.2 Semi rst best bankruptcy and eects on credit spreads . . . . . . . . . . . . 127
12.2 Empirical predictions and results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
13 Other issues on the capital structure of a rm 133
13.1 Strategic default . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
13.1.1 The model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
13.1.2 Debt concessions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
13.2 First best liquidation through renegotiations . . . . . . . . . . . . . . . . . . . . . . . 139
13.2.1 The model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
13.2.2 First best valuation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
13.2.3 Valuation with debt and no renegotiation . . . . . . . . . . . . . . . . . . . . 141
13.2.4 Debt renegotiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
13.2.5 Debt capacity and some remarks . . . . . . . . . . . . . . . . . . . . . . . . . 145
13.3 Dividend policy and renegotiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
13.4 Real assets as an implicit collateral . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
Contents vii
13.4.1 The model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
13.4.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
13.5 Finite maturity with a simple debt policy . . . . . . . . . . . . . . . . . . . . . . . . 155
13.5.1 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
13.6 Finite maturity and dynamic debt policy . . . . . . . . . . . . . . . . . . . . . . . . 158
13.6.1 The model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
14 Investment and Capital Structure 161
14.1 Overinvestment and capital structure . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
14.2 Underinvestment and the option to expand . . . . . . . . . . . . . . . . . . . . . . . 162
14.2.1 The model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
14.3 Underinvestment, overinvestment, and nancial exibility . . . . . . . . . . . . . . . 165
14.4 Underinvestment and strategic debt service . . . . . . . . . . . . . . . . . . . . . . . 168
14.4.1 Underinvestment without strategic debt service . . . . . . . . . . . . . . . . . 168
14.4.2 Underinvestment with strategic debt service . . . . . . . . . . . . . . . . . . . 170
14.4.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
14.5 Risk shifting, investments, and agency costs . . . . . . . . . . . . . . . . . . . . . . 174
14.5.1 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
14.6 Liquidation incentives and agency costs with exogenous debt . . . . . . . . . . . . . 176
14.7 Adjusting capital structure and investment dynamically . . . . . . . . . . . . . . . . 179
15 Takeovers 181
16 Capital Structure and Credit Risk 183
16.1 Credit risk and the simple structural models . . . . . . . . . . . . . . . . . . . . . . 183
16.1.1 Short term yield spreads in diusion models . . . . . . . . . . . . . . . . . . 184
16.2 Credit risk and simple intensity models . . . . . . . . . . . . . . . . . . . . . . . . . 186
16.3 A link between structural and intensity models . . . . . . . . . . . . . . . . . . . . . 191
17 Concluding remarks 193
Bibliography 195
Preface
These lecture notes are used at graduate and doctoral level courses at the University of Southern
Denmark. The notes have evolved from Flor (2003, chapter 1). As the notes have developed over
the years, I have added references to new and interesting applications. Partly, this has been done
to inspire students with subjects for their Masters thesis as well as Ph.D. thesis. Still, however,
the notes are at an incomplete stage and comments or suggestions are most welcome.
As prerequisites I assume that the students have had some introduction to stochastic calculus,
e.g. the students are capable of using It os Lemma.
The notes are a combination of two older lecture notes, Flor (2010b) and Flor (2010a), and the
notes will be revised as we go along.
Odense, 2011
Christian Riis Flor
Department of Business and Economics
University of Southern Denmark
DK5230 Odense M
Denmark
E-mail: crf@sam.sdu.dk
ix
Part I
Corporate nance: agents,
information, and incentives
xi
Chapter 1
Introduction
Financial economics is an economic discipline that deals with issues related to individuals trading
and, hence, exchanging information using nancial markets and nancial instruments. Broadly
speaking, nancial economics can be separated into two major elements: asset pricing and corporate
nance. This book focuses on elements in corporate nance and, in particular, how dynamic eects
impact individuals or rms decision making.
1.1 What is (dynamic) corporate nance?
Corporate nance can be separated into three subelds:
1. Capital budgeting
more explanation to come.
2. Capital structure
more explanation to come.
3. Corporate governance
more explanation to come.
Of course, these subelds can well be interdependent, but it may be benecial to keep the broad
classication in mind when one tries to grasp new theories as well as apply theories in practice. In
this book we are mainly focusing on capital structure and capital budgeting issues.
1
2 Introduction
1.1.1 Elements of corporate nance models
1.1.2 Why dynamic corporate nance?
1.2 Structure of this book
1.3 Prerequisites
This is not a book for rst time students of nancial economics. The aim is to make the book
applicable for students in a Masters or Ph.D. program. In particular, students should be familiar
with basic corporate nance concepts such as agency problems a la debt-equity holder conicts, the
trade-o theory (simple tax implications on capital structure) and pecking order theory etc. Also,
a fundamental understanding of derivatives is necessary. That is, you should know about options,
futures and the like and how such instruments are priced in a dynamic complete market setting,
e.g. the Black-Scholes-Merton framework. Naturally, this implies that you are not completely
unfamiliar with stochastic calculus (It o calculus), so you know something about Brownian motions,
It oa lemma etc.
Chapter 2
Agents, uncertainty, information, and
incentives
2.1 Introduction
2.2 Agents in corporate nance
2.2.1 The economic agent
representation of preferences, utility index or function.
2.2.2 Agents in corporate nance models
rm owners, shareholders, types of creditors (VC, banks, debt holders), nancial intermediations,
the manager, other rms, other stake holders
gives rise to incentive problems.
3
4 Agents, uncertainty and information
2.3 Uncertainty
2.4 Information
2.4.1 Simple information structure
2.4.2 Asymmetric information
2.4.3 Quality of information: ambiguity aversion
When modeling nancial or economic problems we usually assume that there is a given structure
which is known. For example, in the Back-Scholes model, we assume that the stock price has a
known log-normal distribution. But how do we get the parameters? Of course, using econometric
techniques we may come up with estimates for the parameters, but how trust worthy are these
parameters? Perhaps even worse, what if the underlying structure is completely dierent that
what we assume? Going back to the Black-Scholes example, how can we be certain that the true
underlying proces for the stock price does not contain jumps? To handle such issues we need to
extend the standard modeling of preferences. In a few application, we will focus on the so-called
multiplier approach which is also known as robust decision making.
Chapter 3
Stochastic processes, It os Lemma,
and other need-to-know stu
A central issue in dynamic problems is to consider problems in which the information set changes
over time. The information set describes what we know about the random sources which inuence
our problem. These random sources are important because they aect the determinants of the
problem, which we later will denote as state variables. For example, the change in our rm value can
be due to changes in macroeconomic variables, how our competitors behave, changes in production
costs etc. Thus, we need to handle the fact that the uncertainty describing our state variables, e.g.,
the prot ow, changes as time goes by.
1
This chapter contains a subset of the minimum knowledge
we need to have in order to consider valuation of projects. Main references are Dixit and Pindyck
(1994), Bj ork (1994), and Due (2001).
3.1 The continuous stochastic processes
Let us now set up a key ingredient necessary for formulating problems in continuous-time: the
stochastic process. A stochastic process is a variable that evolves randomly over time. An example
could be the temperature, the price of a stock, the risk-free interest rate in the economy, or the
cash ow from a project. In order to introduce the concept of a stochastic process even further,
we consider the following simple example: Suppose that at any point in time, we toss a coin and,
1
A state variable is a variable which is necessary to describe the value function of our problem. For example, the
prot ow and calender time are two important state variables. But, the prot ow can depend on several random
sources.
5
6 Stochastic processes
as a result, we obtain either head or tail. The outcome of our experiment today, t = 0, can be
thought of as a random variable: X(0) head, tail. Now, suppose that we toss the coin at any
point in time, i.e. t [0, ). Then, at time t, the outcome of our experiment is described by the
random variable X(t) head, tail. The complete experiment, i.e. coin tossing from today and
into the indenite future, is then described by the stochastic process X = X(t)

t=0
.
With a bit more formalism, we do the following. Consider a xed probability space (, T, P)
with the ltration F = T
t
: t 0 (satisfying the usual conditions). Here, is the state space,
T is the -algebra, and P is the probability measure. The ltration is a family of non-decreasing
-algebras, where T
t
T. Intuitively, T
t
is the information set at time t. A (stochastic) process,
X, is a measurable function X : [0, ) R. The value of the process X at time t is then
the random variable often written as X(t), X
t
, or X(, t) : R. Finally, x any . Then
the function t X(, t) is denoted the sample path of X given . In words, for a given state in
the state space, the sample path is the evolution of X as a function of time. In the following, we
introduce the most used stochastic processes. These processes are used to describe, for example,
the price of stocks or the evolution of a prot ow.
3.1.1 The Brownian motion
The most fundamental building block is the standard Brownian motion which is often just denoted
as a Brownian motion. It is also commonly denoted as a Wiener process. In the literature, the
Wiener process is often labeled as W, z, or B. We now dene the standard Brownian motion:
Denition 1 (Wiener process, Brownian motion) Assuming a ltered probability space (, T,
(T
t
)
t
, P), the standard Brownian motion is a stochastic process W with the following properties:
1. W
0
= 0, P-almost surely,
2. t, s > t : W
s
W
t
is normally distributed with mean zero and variance s t,
3. t
0
, t
1
, . . . , t
n
: 0 t
0
< t
1
< . . . t
n
< the random variables W(t
0
), W(t
1
) W(t
0
),
. . . , W(t
n
) W(t
n1
) are independently distributed, and
4. the sample path t W(, t) is continuous.
2
2
This assumption is, in fact, a result, since it can be shown that a stochastic process constructed with the rst
three properties always has a continuous version. See e.g. Theorem 2.6 in Jacobsen (1995).
3.1 The continuous stochastic processes 7
3.1.2 Ito processes
Given that we now know what a Wiener process (a Brownian motion) is, we immediately turn to
a more general stochastic process, called an It o process. Such processes are often used to describe
the price of a claim, an interest rate in the economy, a rms cash ow etc. Notice that the drift
and the variance are now functions of the current state and time.
In order to keep some formalism, we let / be the set of adapted processes and dene the spaces
/
1
=
_
/ :
_
T
0
[
t
[dt < , P a.s., T > 0
_
,
/
2
=
_
/ :
_
T
0

2
t
dt < , P a.s., T > 0
_
,
1
1
=
_
/
2
: E
___
T
0

2
t
dt
_
1/2
_
< , T > 0
_
,
1
2
=
_
/
2
: E
_ _
T
0

2
t
dt
_
< , T > 0
_
.
Let a scalar x
0
R, the process /
1
, and the process /
2
be given. A process X of the form
X
t
= x
0
+
_
t
0

s
ds +
_
t
0

s
dW
s
(3.1)
is then called an Ito process. The processes and are usually termed the drift and the diusion
processes of X. Note that X
0
= x
0
. Before we state a very useful result, we also need to introduce
the notion of a martingale. Conceptually, a martingale is a stochastic process with the special
feature that its mean (expected value) is zero. This is extremely useful in calculations as we typically
consider expected values. For example, the present value is the expected value of discounted future
cash ows.
Denition 2 Consider an s and a t < s. Then X is a martingale if
E[X
s
[T
t
] = X
t
, (3.2)
where, of course, X
u
must be integrable for all u (and measurable w.r.t. T
u
). That is, given the
information from time t the expected value of X at any future time s is the same as the value of X
at time t. If the equality in (3.2) is a (), then X is a submartingale (supermartingale).
Occasionally, we will use the notation E
t
[X
s
]

= E[X
s
[T
t
] in order to save space. That is, the
subscript t implicitly refers to the -algebra (information set) at time t < s.
8 Stochastic processes
Example 3.1 Let us see that the Brownian motion, W, is a martingale. Let t be the time of today,
and let s be a later point in time, as above. We now check whether (3.2) is satised:
E
t
[W
s
] = E
t
_
(W
s
W
t
) +W
t

W
t

= E
t
_
W
s
W
t

W
t

+E
t
_
W
t

W
t

= E
t
_
W
s
W
t

W
t

+W
t
= W
t
,
where we use that E
t
_
W
s
W
t

W
t

= 0 due to the second and third property of the Brownian

motion, cf. denition 1. Thus, the Brownian motion is a martingale.
In this example we implicitly assumed that we considered some probability measure, say P.
Some times we need to work with more than one probability measure, e.g., the risk neutral measure
which is often denoted Q. In these cases it is important to remember that the probability measure
is part of the denition of a martingale.
We now know that the Wiener process is a martingale. Since an integral is essentially a sum,
one could guess that the expected value of a stochastic (It o) integral is a martingale it self because
we sum over changes of a Wiener process. Provided some technical condition, this is indeed the
case. From e.g. Due (2001, Chapter 5) we can state the following very useful results that also tell
us how to calculate the variance of a stochastic integral:
Proposition 1 If 1
1
, then
_
dW is a martingale. Thus,
E
t
_ _

t

s
dW
s
_
= 0. (3.3)
If
_
dW is a martingale, then
var
__
T
0

s
dW
s
_
= E
__
T
0

2
s
ds
_
. (3.4)
In addition to the above, we also state a widely used informal dierential notation of an It o
process (3.1). An It o process is often written as
dX
t
=
t
dt +
t
dW
t
, X
0
= x
0
. (3.5)
The informal form in (3.5) is often used in applications or derivations of results because it has
some intuitive features. The increment of X at time t, dX
t
, consists of two parts, namely
a deterministic part dt and a stochastic part dW
t
. We can interpret the dierential
representation further with help of the following result, cf. Due (2001).
3.1 The continuous stochastic processes 9
Lemma 3.1.1 Suppose that and both have continuous sample paths and are in 1
2
. Then for
any time t
d
d
E
t
[X

=t
=
t
, P a.s. (3.6)
and
d
d
var
t
(X

=t
=
2
t
, P a.s. (3.7)
Proof of (3.6). Let t be a xed point in time and choose a > t. From (3.1) we have
X

= X
t
+
_

t

s
ds +
_

t

s
dW
s
,
and taking the conditional expectation at time t we get
E
t
[X

] = X
t
+E
t
_ _

t

s
ds
_
+E
t
_ _

t

s
dW
s
_
.
From Proposition 1 we know that E
t
_ _

t

s
dW
s

= 0, hence
E
t
[X

] = X
t
+E
t
_ _

t

s
ds
_
+ 0.
Taking the derivative w.r.t. yields
d
d
E
t
[X

] =
d
d
E
t
_ _

t

s
ds
_
= E
t
d
d
_ _

t

s
ds
_
= E
t
[

]
and hence
d
d
E
t
[X

] E
t
[
t
] =
t
, t,
where the limit is taken from the right.
In the sense of (3.6) and (3.7) we can interpret
t
as the rate of change of the expectation of X,
conditional on the information available at time t. Similarly, we can interpret
2
t
as the rate of
change of the conditional variance of X at time t. Commonly, and are referred to as the drift
and diusion, respectively.
3.1.3 Special processes
Let us now consider some of the most used stochastic processes in the literature. First, we consider
the immediate extension of a standard Brownian motion. Let
s
= and
s
= in (3.1), where
10 Stochastic processes
2 4 6 8 10
2
1
1
2
3
4
5
Figure 3.1: Sample path of a Brownian motion with drift. dX
t
= dt +dW
t
, = 0.5, = 1.
and are real valued constants. The process
X
t
= x
0
+
_
t
0
ds +
_
t
0
dW
s
= x
0
+t +W
t
, (3.8)
which is often written as
dX
t
= dt +dW
t
,
is called a Brownian motion with drift. Figure 3.1 illustrates a sample path for a such a process in
the case = 0.5, = 1. From Proposition 1 we have
E[X
t
] = E[x
0
+t +W
t
] = x
0
+t +E[W
t
] = x
0
+t
and
var[X
t
] = var[W
t
] =
2
var(W
t
) =
2
t.
Geometric Brownian motion. One of the most used processes for the price of a security or,
more generally, as a state variable is the geometric Brownian motion. The geometric Brownian
motion is obtained by setting
s
= X
s
and
s
= X
s
in (3.1). As noted earlier,
t
is referred to
as the drift and
t
as the diusion. However, for geometric Brownian motions, one often uses the
terminology, that is the drift or drift rate and is the volatility of X. Figure 3.2 illustrates a
sample path for a such a process in the case = 0.05, = 0.2.
3.1 The continuous stochastic processes 11
0 2 4 6 8 10
50
100
150
Figure 3.2: Sample path of a geometric Brownian motion, dX
t
= X
t
dt + X
t
dW
t
, = 0.05,
= 0.2.
For a geometric Brownian motion (3.1) becomes:
X
t
= x
0
+
_
t
0
X
s
ds +
_
t
0
X
s
dW
s
= x
0
+
_
t
0
X
s
ds +
_
t
0
X
s
dW
s
(3.9)
which is often written as
dX
t
= X
t
dt +X
t
dW
t
.
Since the geometric Brownian motion is going to play a central role in later chapters, we would
like to know the expected value of such a process. Let us try to calculate the expected value of a
geometric Brownian motion:
E[X
t
] = E
_
x
0
+
_
t
0
X
s
ds +
_
t
0
X
s
dW
s
_
= x
0
+E
__
t
0
X
s
ds
_
+E
__
t
0
X
s
dW
s
_
= x
0
+E
__
t
0
X
s
ds
_
, (3.10)
where the last equation follows from proposition 1. However, the distribution of X
t
[X
s
= x, s < t,
is not obvious, so we will need to return to the derivation later. In order to derive the distribution
we will use an important lemma stated in the next section.
Ornstein-Uhlenbeck. The Ornstein-Uhlenbeck process satisfy
X
t
= x
0
+
_
t
0
(

X X
s
)ds +
_
t
0
dW
s
(3.11)
12 Stochastic processes
which is often written as the stochastic dierential equation
dX
t
= (

X X
t
)dt +dW
t
.
Here,

X is the level to which X is expected to revert to and is the speed with which the
reversion occurs. Due to Vasicek (1977) this process is often used as a candidate for the risk free
short interest rate. If r
t
is the short rate following (3.11), r
t
exhibits mean reversion, i.e. the interest
rate is expected to approach an given level over time. However, one can show that the distribution
of r
t
is normal (or Gaussian) implying that negative interest rates are possible.
3.1.4 Itos Lemma
A very important result we need to know is It os Lemma. The strength of It os Lemma is that
it tells us how to calculate the evolution of a process which is dened as a function of another
process. For instance, in the Black-Scholes model, the market price of a rms stock follows a
geometric Brownian motion, but how does the price of a European call option evolve? Or, if we
consider an investment described by an earnings and a cost process, It os Lemma tells us how we
can determine the prot process.
Before we state It os Lemma we need to introduce the notion of the quadratic variation of the
process X. On dierential form, the quadratic variation is often written as dX
t
, or perhaps more
commonly as dX
t
= (dX
t
)
2
. In order to calculate the quadratic variation of an It o process, we
need some rules of thumb for calculating products of It o dierentials.
Lemma 3.1.2 Let W
p
and W
q
be two Wiener processes with quadratic (-cross) variation . Then
product dW
p
t
dt
dW
q
t
dt 0
dt 0 0
is a rule of thumb for calculating products of Ito dierentials. In other words
dt dt = 0, dt dW
p
t
= dt dW
q
t
= 0, dW
p
t
dW
q
t
= dt.
Let us now consider dX
t
or (dX
t
)
2
. Being sloppy, we can just derive (dX
t
)
2
from (3.5) as follows:
(dX
t
)
2
= (
t
dt +
t
dW
t
)
2
=
2
t
(dt)
2
+
2
t
(dw
t
)
2
+ 2
t

t
dtdW
t
3.1 The continuous stochastic processes 13
so from Lemma 3.1.2 we obtain
(dX
t
)
2
= dX
t
=
2
t
dt.
We are now ready to state It os Lemma in the one-dimensional case.
Lemma 3.1.3 (It os Lemma) Suppose X is an Ito process of the form (3.1)
dX
t
= (X
t
, t)dt +(X
t
, t)dW
t
and let a function f : R[0, ) R, f C
2,1
be given. Then the process Y dened by Y
t
= f(X
t
, t)
is an Ito process with
df(X
t
, t) =
f
t
(X
t
, t)dt +
f
X
(X
t
, t)dX
t
+
1
2

2
f
X
2
(X
t
, t)dX
t
(3.12)
which from (3.5) and Lemma 3.1.2 can be written as
df(X
t
, t) =
_
f
t
(X
t
, t) +
t
f
X
(X
t
, t) +
1
2

2
t

2
f
X
2
(X
t
, t)
_
dt +
t
f
X
(X
t
, t)dW
t
. (3.13)

Often, we will instead use the notation f

t
, f
x
etc. for the partial derivatives of f.
Example: Distribution of a GBM
Suppose X follows a geometric Brownian motion as in (3.9), i.e.
dX
t
= X
t
dt +X
t
dW
t
.
Consider the process Y dened as Y = f(X) = log(X), where log is the natural logarithm. Let us
use It os Lemma to determine the process for Y . The (partial) derivatives of f are
f
x
(X) =
1
X
, f
xx
(X) =
1
X
2
(and clearly f
t
(X) = 0).
Therefore
dY
t
= f
x
(X)dX
t
+
1
2
f
xx
(X)dX
2
t
=
_
1
X
X
t

1
2
1
X
2

2
X
2
t
_
dt +
1
X
X
t
dW
t
=
_

1
2

2
_
dt +dW
t
.
14 Stochastic processes
On integral form this means
Y
t
Y
0
=
_
t
0
_

1
2

2
_
ds +
_
t
0
dW
s
=
_

1
2

2
_
t +(W
t
W
0
)
and since a Brownian motion begins in 0 we get
Y
t
Y
0
=
_

1
2

2
_
t +W
t
,
whence it follows that Y
t
Y
0
is normally distributed with mean (
1
2

2
)t and variance
2
t. Since
X = expY it also follows that X is lognormally distributed.
Example 3.2 As previously noted, the geometric Brownian motion is going to play a central role
in later chapters, and we would like to know the expected value of such a process, cf (3.10). Let
us try to calculate the expected value of a geometric Brownian motion, assuming that X begins in
x, i.e., X
0
= x:
E
0
[X
t
] = x +E
__
t
0
X
s
ds
_
,
= x +
_
t
0
E
0
[X
s
] ds,
= x +
_
t
0
xexp
__

1
2

2
_
s +
1
2

2
s
_
ds,
= xe
t
. (3.14)
However, the results follows immediately from the result that X
t
log N(
1
2

2
,
2
). We get
E
0
[X
t
] = E
_
xexp
_
(
1
2

2
)t +W
t
__
= xexp
_
E
_
(
1
2

2
)t +W
t
_
+
1
2
V
_
(
1
2

2
)t +W
t
__
= xe
t
, (3.15)
which is the same as in (3.14).
EXERCISE 3.1 Let X be a geometric Brownian motion following
dX
t
= X
t
dt +X
t
dW
t
.
3.1 The continuous stochastic processes 15
Let > 0 be xed and derive X

. Hint: Use It os Lemma on log(x

).
Then derive E
__
X
t
X
0
_

_
.
EXERCISE 3.2 Derive
t
and
2
t
from Lemma 3.1.1 in the case of a geometric Brownian motion.
Hint: Use exercise 3.1.
Multidimensional case The one dimensional case from above can be extended as follows.
Suppose that we have d N independent Wiener processes, W
1
, . . . , W
d
. Dene the process
W = (W
1
, . . . , W
d
) as a Wiener process (standard Brownian motion) in R
d
. Let x
0
R, /
1
,
and
1
, . . . ,
d
/
2
be given. Then
X
t
= x
0
+
_
t
0

s
ds +
d

i=1
_
t
0

i
s
dW
i
s
(3.16)
or, conveniently written,
X
t
= x
0
+
_
t
0

s
ds +
_
t
0

s
dW
s
(3.17)
or written in the stochastic dierential form
dX
t
=
t
dt +
t
dW
t
, X
0
= x
0
(3.18)
is an It o process. Consider now X = (X
1
, . . . , X
N
) as an It o process in R
N
, which we write as
X
t
= x
0
+
_
t
0

s
ds +
_
t
0

s
dW
s
(3.19)
or written in the stochastic dierential form
dX
t
=
t
dt +
t
dW
t
, X
0
= x
0
R
N
, (3.20)
where R
N
and R
Nd
. Following Due (2001), we then have the extended version of It os
Lemma.
Lemma 3.1.4 (It os Formula) Let X be an Ito process like (3.19) and f (
2,1
(R
N
[0, )).
Then f(X
t
, t)
t0
is an Ito process and, for any t
f(X
t
, t) = f(X
0
, 0) +
_
t
0
T
X
f(X
s
, s)ds +
_
t
0
f
x
(X
s
, s)
s
dW
s
, (3.21)
where
T
X
f(X
s
, s) = f
x
(X
s
, s)
s
+f
t
(X
s
, s) +
1
2
tr[
s

T
s
f
xx
(X
s
, s)], (3.22)
where tr(M) denotes the trace of a square matrix M.
16 Stochastic processes
EXERCISE 3.3 Let X and Y be two real-valued It o processes. Dene the product as Z = XY . Is Z an
It o process? How does it look like? What if X and Y ar both geometric Brownian motions?
3.2 Jump processes
3.2.1 The Poisson process
Recall that in Section 3.1.1, we dened the standard Brownian motion by its distributional proper-
ties. We shall also do so for the Poisson process. We dene the Poisson process with the intensity
> 0 as the stochastic process N with the following properties:
1. N
0
= 0, P-almost surely,
2. N
t
= N
t
N
t
0, 1,
3. t, s > t : N
s
N
t
has a Poisson distribution with intensity (t s),
4. t
0
, t
1
, . . . , t
n
: 0 t
0
< t
1
< . . . t
n
< the random variables N(t
0
), N(t
1
) N(t
0
),
. . . , N(t
n
) N(t
n1
) are independently distributed.
Note that condition 4 implies that for two dates t, s with s > t, the random variable N
s
N
t
is
independent of the information set at time t, T
t
. Condition 3 tells us how we can compute the
probability that k N
0
jumps have occurred in the time from t to s. Since N
s
N
t
Poiss((st))
we have
P(N
s
N
t
= k[T
t
) = P(N
s
N
t
= k) =
_
(s t)
_
k
k!
e
(st)
, (3.23)
and in particular, the probability of no jump is
P(N
s
N
t
= 0) = e
(st)
. (3.24)
3.2 Jump processes 17
We shall also need the mean of N. For generality, consider N
s
N
t
. We know that the distribution
of this random variable is Poiss((s t)), hence
E
t
[N
s
N
t
] =

k=0
_
(s t)
_
k
k!
e
(st)
k,
=

k=1
_
(s t)
_
k
k!
e
(st)
k =

k=1
_
(s t)
_
k
(k 1)!
e
(st)
,
= (s t)

k=1
_
(s t)
_
(k1)
(k 1)!
e
(st)
= (s t)

j=0
_
(s t)
_
j
j!
e
(st)
= (s t).
For later purposes it will be important that we can relate the Poisson process N to a martingale.
It is easy to see that N itself is not a martingale:
E
t
[N
s
] = E
t
[N
s
N
t
+N
t
]
= E
t
[N
s
N
t
] +E
t
[N
t
] = E
t
[N
s
N
t
] +N
t
= (s t) +N
t
,= N
t
.
However, from the above calculation we can see how we can modify or compensate the Poisson
process in order to obtain a martingale. If we dene M
t
= N
t
t, then we have
E
t
[M
s
] = E
t
[M
s
M
t
+M
t
] = E
t
[(N
s
s) (N
t
t)] +M
t
= E
t
[(N
s
N
t
)] (s t) +M
t
= (s t) (s t) +M
t
hence
E
t
[M
s
] = M
t
,
so M is a martingale.
More generally, one can under certain technical conditions always compensate a counting pro-
cess, N, with help of its intensity. We omit the technical details, and quote a useful theorem from
Bj ork (1994).
3
3
For instance, we have not dened precisely what an intensity is. We are also (too) sloppy about which -algebra(s)
we are working on.
18 Stochastic processes
Theorem 1 Suppose that the counting process N has the intensity (process) (=
s

s0
), and
that
E[N
t
] < , t 0, (3.25)
then the process M dened as
M
t
= N
t

_
t
0

s
ds (3.26)
is a martingale.
The Poisson process N is a special case of a counting process.
4
In particular, the intensity process
is constant, i.e.
s
= . We also know that E[N
t
] = t < for any t, so we have the following
corollary.
Corollary 1 Let N be a Poisson process with intensity . Then
M
t
= N
t
t
is a martingale.
In applications, we will often work with the informal dierential form, as we already saw when we
considered It o processes. Corollary 1 then says that
dN
t
= dM
t
+dt. (3.27)
Here it is important that dN
t
is understood as a forward looking dierential.
When we discussed It o processes, we interpreted the drift as the rate of change of the conditional
expectation of the process. In the same vein, we will interpret the intensity . In order to rst give
an intuitive argument, consider N
t+h
N
t
for a small h > 0. Then
P(N
t+h
N
t
= 1[T
t
) = h e
h
= h

j=0
(h)
j
j!
and very loosely speaking we get
P(N
t+h
N
t
= 1[T
t
) dt, h dt
4
In the literature, the term
t

t
0

s
ds is often denoted the compensator. Thus, for a Poisson process, the
compensator is simply
t
= t.
3.2 Jump processes 19
which in very sloppy notation is written as
P(dN
t
= 1[T
t
) = dt.
The latter expression is often used to support the intuition that the probability of a jump (an
event) in the short time-interval dt is dt.
3.2.2 Itos Lemma and jumps
In this section, we will follow Bj ork (1994) and make an intuitive development of It os Lemma in
the case of Poisson (and Wiener) processes. Readers must confer other sources in the literature for
a more rigorous treatment of the subject.
Suppose that N is a Poisson process and that X follows a combined It o-Poisson process of the
form
dX
t
= f(X
t
, t)dt +h(X
t
, t)dN
t
+(X
t
, t)dW
t
which we abbreviate to
dX
t
= f
t
dt +h
t
dN
t
+
t
dW
t
. (3.28)
In the above expression for the stochastic dierential equation of X we have used the notation X
t
.
This should be thought of as the value of the process X when we consider the left limit, i.e. for an
s < t we evaluate X
t
lim
st
X
s
.
5
Between the times of a jump, the development of X is simply
dX
t
= f
t
dt +
t
dW
t
, (3.29)
but at a time of a jump we must be a bit careful. At the time of a jump, X has a discontinuity of
magnitude
X
t
= X
t
X
t
= h
t
N
t
= h
t
. (3.30)
Now dene the process Z as
Z
t
= F(X
t
, t).
Between jumps, the dierential of Z is as in the usual It os Lemma
dZ
t
=
_
F
t
(X
t
, t) +F
x
(X
t
, t)f
t
+
1
2

2
t
F
xx
(X
t
, t)
_
dt +F
x
(X
t
, t)
t
dW
t
, (3.31)
5
In the same vein, the Poisson process N is said to be right-continuous with limits from the left. This is commonly
referred to as RCLL from English or cadlag from French.
20 Stochastic processes
but at a jump at time t, Z makes a jump of size
Z
t
= Z
t
Z
t
= F(X
t
, t) F(X
t
, t) = F(X
t
, t) F(X
t
, t),
because we assume that F has continuous derivatives in (x, t). From (3.30) we then obtain
Z
t
= F(X
t
+h
t
, t) F(X
t
, t). (3.32)
Recall that dN
t
= 1 at a jump and 0 otherwise. We can therefore combine (3.31) and (3.32) in
order to get the following intuitive extension of It os Lemma .
Result 3.2.1 (Simple version of It os Lemma with a Poisson process) Suppose that X is
a combined Ito-Poisson process of the form (3.28) and consider the function F : (x, t) F(x, t)
R, where F (
2,1
_
R [0, )
_
. Then the dierential of F is
dF(X
t
, t) =
_
F
t
(X
t
, t) +F
x
(X
t
, t)f
t
+
1
2

2
t
F
xx
(X
t
, t)
_
dt +F
x
(X
t
, t)
t
dW
t
+
_
F(X
t
+h
t
, t) F(X
t
, t)

dN
t
, (3.33)
where we recall that the functions f
t
, h
t
and
t
must be evaluated in the point (X
t
, t). Often one
writes F(X
t
, t) = F(X
t
+h
t
, t) F(X
t
, t). On integral form (3.33) is written as
F(X
t
, t) = F(X
0
, 0) +
_
t
0
_
F
t
(X
s
, s) +
1
2

2
s
F
xx
(X
s
, s)
_
ds +
_
t
0
F
x
(X
s
, s)dX
s
+

st
_
F(X
s
, s) F
x
(X
s
, s)X
s
_
. (3.34)

Example/Exercise We are now going to solve a stochastic dierential equation. Suppose that
X follows
dX
t
=
t
X
t
dt +
t
X
t
dN
t
, X
0
= a, (3.35)
where and are given (predictable) processes.
6
The form of X looks much like the well-known
geometric Brownian motion except, of course, that there is no Brownian motion. But it is natural
to consider the transformation Z
t
= log(X
t
). Now use It os formula and write the result, i.e. dZ
t
,
on integral form. The result is
Z
t
= Z
0
+
_
t
0

s
ds + log
_

T
n
t
[1 +
T
n
]
_
,
6
Meaning that they are suciently nice for our purpose.
3.2 Jump processes 21
where T
n
is the nth jump-time for N. If we now translate this result to be in X-form, by translating
back to X = expZ, we get
X
t
= a

T
n
t
[1 +
T
n
] exp
__
t
0

s
ds
_
.
The moral is that in principle it is not much more dicult to use It os Lemma if we include a
Poisson process.
Example 3.3 The value of a machine. This example is from Dixit and Pindyck (1994), but we
will do it with our notation. Suppose that a machine produces a prot ow of , i.e. the cumulated
prot from time t to t + h is
_
t+h
t
ds = h. However, at some point in time, the machine may
break down. A break down leads to an abandonment of future prot ows. Suppose further that
is the arrival rate of a breakdown, i.e. break downs follow a Poisson process with intensity .
Moreover, let be the appropriate discount rate of return and assume continuous compounding.
This implies that the discount factor from time s to time t is e
(st)
. We want to derive the value
of the machine.
The simplest and immediate procedure is to calculate the expected discounted value of future
prot ows.
7
We rst dene the stopping time, , for the breakdown of the machine:
= infs > 0 : N
s
= 1.
In this case, the value at time t, if there has been no jump (break down), is
8
V
t
() = E
P
t
_ _

t
e
(st)
1
{>s|>t}
ds
_
=
_

t
e
(st)
E
P
t
_
1
{>s|>t}

ds
=
_

t
e
(st)
P
_
> s[ > t
_
ds
=
_

t
e
(st)
P
_
N
s
N
t
= 0[N
t
= 0
_
ds
=
_

t
e
(st)
e
(st)
ds
=

+
. (3.36)
In order to approach a procedure which we shall use later on in the real option analysis, we consider
instead the following intuitive argument. An investor demands a rate of return equal to , so the
7
Note that we are not assuming risk neutrality.
8
Here we write P

> s|F
t

= P

> s| > t

.
22 Stochastic processes
instantaneous dividend rate or the instantaneous payo at time t must be V
t
. The dividend
from the machine consists of the prot ow and the instantaneous expected capital gain. That
is
9
V
t
= +
d
d
E
t
[V

][
=t
. (3.37)
We therefore need the dierential of V . Suppose that there is never going to be a break down.
Then there is never a change in the value of receiving the prot ow forever, i.e. dV
t
= 0. If there
is a breakdown, then the value is zero. Thus,
dV
t
= V
t
dN
t
and if we use the compensator we get
dV
t
= V
t
(dM
t
+dt)
= V
t
dM
t
V
t
dt,
but written in this form we know that
d
d
E
t
[V

][
=t
= V
t
(3.38)
Therefore, condition (3.37) becomes
V
t
= V
t
,
i.e.
V
t
=

+
. (3.39)
Interpret the result!
3.2.3 Itos Lemma the general version
Omitted.
9
In Dixit and Pindyck (1994), terminology such as E[dV ] is used, so e.g.
d
d
E
t
[V

]|
=t
=E[dV ]/dt. But keep in
mind, that E[dV ] does not really make any sense because dV is not a random variable.
Part II
A rms decision making:
Applied stochastic control
dynamic programming
and
real options analysis
23
Chapter 4
Introduction to corporate investment
Companies, managers, and investors face many dierent decision problems. For example, when a
biotechnological company tries to develop a new drug, it must sequentially decide between aban-
donment and continuance of the research and development process. Furthermore, such decisions
can depend on how well the rms competitors are doing. For an another example we could consider
an oil drilling company. The company has the option to temporarily stop the oil production (or
drilling in new oil elds). The decision to stop the production and resume it later on presumably
depends on the current and future oil price, the production costs etc. But just because the current
oil price seems low, it may not be optimal to halt the oil production. Understanding and analyzing
such problems as real options and being able to give advice on the exercise of these options is
the goal of real options analysis.
For many problems there is a huge degree in freedom concerning when to exercise a real option.
Therefore, it is reasonable to work in continuous-time models. This requires some tools and under-
standing of stochastic calculus which we consider for a start. We then need to be able to handle
decision making under uncertainty in this stochastic environment, which is why the next topic is
dynamic programming. The later chapters consider applications of the tools we have developed.
25
Chapter 5
Decision making under uncertainty
dynamic programming
We shall rst consider the problem of optimal decision making in a discrete-time framework. This
is technically easier than the continuous-time framework and we shall be able to obtain the main
ideas.
Only consider Markov processes, i.e. the future evolution depends only on the information
available today.
5.1 Discrete time the general framework
Terminology in discrete time.
x = state variable, i.e. the underlying variable(s) driving the evolution of price, output, etc.,
u = control variable(s), u
t
can depend on x
t
, but not on x
t+1
, . . .; example: u = 0 if wait,
u = 1 if invest,
(x
t
, u
t
) = prot ow,

t
(x
t+1
[x
t
, u
t
) = cumulated density function, i.e. the measure used to calculate the probabil-
ity,
= discount rate,

t
(x
t
) = termination payo at time t,
27
28 Decision making under uncertainty dynamic programming
F
t
(x
t
) = outcome (E[NPV ]) when future decisions are made optimally,
1
E
t
[F
t+1
(x
t+1
)] = continuation value.
In the general multi-period framework we consider rst a problem with a nite time-horizon, T.
Suppose that we split up the period in N periods of equal length. The time-horizon in one period
is then t = T/N and we consider decisions at times t
0
= 0, t
1
= t, . . . , t
n
= nt, . . . , t
N
= T.
Generally, we can think of an individual who can make some decisions which inuences his
periodic consumption or payo and his nal wealth or termination value. Suppose that we
equip the individual with a utility function, U. At time t, we let c
t
be the consumption rate, such
that we can approximate the individuals consumption in the period [t
n
, t
n+1
) as
_
t
n+1
t
n
c
t
n
ds =
c
t
n
t. If the individual has the consumption plan c
t
0
, c
t
1
, . . . , c
t
N1
and a nal wealth of W
T
,
then we assume that the individuals life-time utility of this consumption plan can be written as
2
U(c
t
0
, . . . , c
N1
, W
T
) =
N1

n=0
e
t
n
u(c
t
n
)t +e
T
u(W
T
). (5.1)
Obviously, the individual wants to have the highest attainable expected life-time utility as seen
from time 0. We assume that the individual can decide upon his consumption and also a control,
u. We dene his highest attainable expected life-time utility at time 0 as the so-called indirect (or
derived) utility function, i.e.
F
0
= sup
(c
t
n
,u
t
n
)
N1
n=0
E
_
N

n=0
e
t
n
u(c
t
n
)t +e
T
u(W
T
)
_
(5.2)
and for later periods we dene
F
t
i
= sup
(c
t
n
,u
t
n
)
N1
n=i
E
t
i
_
N

n=i
e
(t
n
t
i
)
u(c
t
n
)t +e
(Tt
i
)
u(W
T
)
_
, i = 0, . . . , N 1. (5.3)
The consumption and controls must be admissible.
The Bellman equation We are now ready to state the fundamental result called the Bellman
equation which gives the dynamic programming property.
Result 5.1.1 (Bellman equation)
F
t
i
= sup
c
t
i
,u
t
i
_
u(c
t
i
)t +E
t
i
_
e
t
F
t
i+1

_
, i = 0, . . . , N 1. (5.4)
1
F could be interpreted as the indirect utility, which is often denoted with the symbol J.
2
The utility function of periodic consumption and the terminal value do not have to be equal.
5.1 Discrete time the general framework 29
Proof. The result follows directly from manipulations of the denition of F
t
i
:
F
t
i
= sup
(c
t
n
,u
t
n
)
N1
n=i
E
t
i
_
N

n=i
e
(t
n
t
i
)
u(c
t
n
)t +e
(Tt
i
)
u(W
T
)
_
= sup
(c
t
n
,u
t
n
)
N1
n=i
E
t
i
_
u(c
t
i
)t +
N

n=i+1
e
(t
n
t
i
)
u(c
t
n
)t +e
(Tt
i
)
u(W
T
)
_
and using iterated expectations
= sup
(c
t
n
,u
t
n
)
N1
n=i
E
t
i
_
u(c
t
i
)t +E
t
i+1
_
N

n=i+1
e
(t
n
t
i
)
u(c
t
n
)t +e
(Tt
i
)
u(W
T
)
__
= sup
(c
t
n
,u
t
n
)
N1
n=i
E
t
i
_
u(c
t
i
)t +e
t
E
t
i+1
_
N

n=i+1
e
(t
n
t
i+1
)
u(c
t
n
)t +e
(Tt
i+1
)
u(W
T
)
__
and seperating the sup we get
= sup
c
t
i
,u
t
i
E
t
i
_
u(c
t
i
)t +e
t
sup
(c
t
n
,u
t
n
)
N1
n=i+1
E
t
i+1
_
N

n=i+1
e
(t
n
t
i+1
)
u(c
t
n
)t +e
(Tt
i+1
)
u(W
T
)
__
now use the denition of F and that u(c
t
i
) T
t
i
F
t
i
= sup
c
t
i
,u
t
i
_
u(c
t
i
)t +e
t
E
t
i
[F
t
i+1
]
_
.

The result in (5.4) has a form which is very appealing for the following intuition. The decision
problem can be divided in two sub-problem: the optimal decision today and the optimal decision
henceforth.
With the Bellman equation in place we shall now concentrate on the applications of the dynamic
programming principle in real option analysis.
5.1.1 Discrete time our framework
We will now think of the individual as a rm (or an entrepreneur) which considers an investment
problem. Therefore, we will in the following think of F as something like the rms value of the
option to invest. The consumption is now considered as the prot from the investment and the
utility is simply the value. Thus, we replace the utility of the consumption rate u(c
t
n
) with the
prot ow (rate)
t
(x
t
, u
t
), where we have indicated that the prot ow depend on a state variable
x which the company can aect by choosing a control u
t
. Hence, the prot in the period [t
n
, t
n+1
)
30 Decision making under uncertainty dynamic programming
is
t
n
(x
t
n
, u
t
n
)t. Furthermore, where an individual can decide how much to consume in a period,
the rm cannot (in our setup) directly decide the size of the prot. Therefore we do not include the
consumption in the sup. Of course, the rm can still make decisions which inuences on future
prot ows. In our framework, we therefore obtain the following Bellman equation.
Result 5.1.2 (Bellman equation)
F
t
i
= sup
u
t
i
_

t
i
(x
t
i
, u
t
i
)t +e
t
E
t
i
_
F
t
i+1

_
, i = 0, . . . , N 1, (5.5)
and if we consider a given t and t = 1 we can write
F
t
(x
t
) = sup
u
t
_

t
(x
t
, u
t
) +e

E
t
_
F
t+1
(x
t+1
)

_
. (5.6)
Finite horizon: Solving problems.
Innite horizon: some times get rid of t.
5.2 Dynamic programming in continuous time
Since many problems are easier to study in a continuous-time formulation we also want to be able
to use dynamic programming in this case. We will use the terminology from Section 5.1 and if
necessary extend it in the obvious way. One of our goals is to get the Bellman equation (5.5) in a
continuous-time formulation this version is often called the Hamilton-Jacobi-Bellman equation.
To save notation, we consider the problem at time t, i.e. t
i
= t. Then
F
t
i
= sup
u
t
i
_

t
i
(x
t
i
, u
t
i
)t +e
t
E
t
i
_
F
t
i+1

_
= sup
u
t
_

t
(x
t
, u
t
)t +e
t
E
t
_
F(x
t+t
, t + t)

_
now multiply by e
t
and subtract F(x
t
, t) T
t
to get
(e
t
1)F(x
t
, t) = sup
u
t
i
_
e
t

t
(x
t
, u
t
)t +E
t
_
F(x
t+t
, t + t) F(x
t
, t)

_
dividing to t yields
e
t
1
t
F(x
t
, t) = sup
u
t
i
_
e
t

t
(x
t
, u
t
) +
1
t
E
t
_
F(x
t+t
, t + t) F(x
t
, t)

_
, (5.7)
and we now want to let t 0. From the left-hand-side of (5.7) we get
e
t
1
t
, t 0
5.2 Dynamic programming in continuous time 31
because the fraction is just the quotient of dierence for the exponential function, i.e. the limit is
the derivative of e
t
evaluated in the point t = 0.
3
Furthermore, and here we are sloppy,
1
t
E
t
_
F(x
t+t
, t + t) F(x
t
, t)

d
d
E
t
_
F(x

, )

t
, t 0,
and we know that the latter expression is the drift of F(x
t
, t). Plugging the results into equa-
tion (5.7) we now have the Hamilton-Jacobi-Bellman equation:
F(x
t
, t) = sup
u
t
_

t
(x
t
, u
t
) +
d
d
E
t
_
F(x

, )

t
_
. (5.8)
If we assume that x follows an It o process of the form (3.1) or (3.5), i.e.
dx
t
= (x
t
, u
t
, t)dt +(x
t
, u
t
, t)dW
t
, x
0
= x, (5.9)
where we have included the control in the drift and the diusion to explicitly show that those
processes can depend on the control, we can use It os Lemma as in (3.13). This yields
dF(x
t
, t) =
_
F
t
(x
t
, t) +(x
t
, u
t
, t)F
x
(x
t
, t) +
1
2
(x
t
, u
t
, t)
2
F
xx
(x
t
, t)
_
dt +(x
t
, u
t
, t)F
x
(x
t
, t)dW
t
hence the drift of F is
F
t
(X
t
, t) +(x
t
, u
t
, t)F
x
(x
t
, t) +
1
2

2
(x
t
, u
t
, t)F
xx
(x
t
, t)
and from (5.8) we obtain
F(x
t
, t) = sup
u
t
_
(x
t
, u
t
, t) +F
t
(x
t
, t) +(x
t
, u
t
, t)F
x
(x
t
, t) +
1
2

2
(x
t
, u
t
, t)F
xx
(x
t
, t)
_
, (5.10)
We can give an intuitive argument for (5.8) (or (5.10)). On the left hand side, F is the
required instantaneous dividend rate for an investor with as the discount rate of return. On the
right hand side we want to have the total instantaneous (expected) dividend rate of the investment,
i.e. what we receive by holding the asset in an innitely short period of time. The total dividend
rate consists of two parts, namely the (instantaneous) prot, , and the expected rate of capital
gain. Due to this interpretation, one sometimes interpret the Hamilton-Jacobi-Bellman equation
(5.8) as an equilibrium condition.
4
3
It also follows by lHospitals rule.
4
As usual we have omitted the technical conditions, but it is worth to mention that uncertainty must be continuous
form the right and strategies must be continuous from the left cadlag/LLRC-processes.
32 Decision making under uncertainty dynamic programming
Time independent problem. We shall often consider a special case of (5.8) or (5.10) in which
the structure of the problem is particularly simple. This case occur if the time horizon in innite
and if the included functions do not depend explicitly on time. In the language of (5.10) and in
addition to an innite time horizon, we need the functions , , and to be independent of time.
In this case, the Hamilton-Jacobi-Bellman equation becomes
F(x) = sup
u
_
(x, u) +(x, u)F
x
(x) +
1
2

2
(x, u)F
xx
(x)
_
. (5.11)
5.3 Binary decision problems
Many investment (or other decision) problems have the form of deciding one of two things: either
to wait or to invest now. We can think of the decision to wait with the investment as continuing
the management of the rm without actively doing anything. If x is such that we wait, we say that
x is in the continuation region. If our problem is of the simple type just described, we say that we
have a binary decision problem.
When we consider a binary decision problem, we can simplify the Bellman equation. In general,
we must use the control that yields the supremum. However, for a binary decision problem, the
control space can be described by u
t
0, 1, where we can interpret u
t
= 1 as stop or invest
and u
t
= 0 as continue or wait.
Suppose that if the optimal decision at time t is to stop, i.e. to make the investment, then
we obtain the termination value (x
t
, t). In other words we have that F(x
t
, t) = (x
t
, t) if u = 1.
W can rewrite this in the Bellman equation (5.8) as
0 = max
_
(x
t
, t) F(x
t
, t), (x
t
, t) +
d
d
E
t
_
F(x

, )

t
F(x
t
, t)
_
, (5.12)
and if x is an It o process we have
0 = max
_
(x
t
, t) F(x
t
, t), (5.13)
(x
t
, t) +F
t
(x
t
, t) +(x
t
, t)F
x
(x
t
, t) +
1
2

2
(x
t
, t)F
xx
(x
t
, t) F(x
t
, t)
_
.
An important question is how to derive the optimal trigger level of the underlying state variable,
i.e. the level of x where the investment should be made. Denote this level as x

.
5.3 Binary decision problems 33
5.3.1 Value matching
For many applications and in our continuous value function setting, it turns out that a particular
condition must hold when the rm decides to undertake the investment or, more generally, when
the decision maker changes control. Suppose that we stop when x
t
= x
t
, for some x
t
. Note that we
are not claiming that x
t
is an optimal boundary. In our investment problem formulation we can
say that we invest, i.e. exercise the investment option at x
t
, but we are not assuming that x
t
is the
optimal point of investment. But given that we stop (exercise) at x
t
we have the condition
F( x
t
, t) = ( x
t
, t). (5.14)
Example 5.1 For a concrete example think of an ATM American-style put option with an exercise
price K, maturity date T, and the additional condition that the option must be exercised when
the stock price (the state variable) hits the boundary x
t
=
1
2
T+t
T
K (make your own drawing to
illustrate). Then we have the natural condition that if the stock price hits the boundary, the owner
receives the termination value ( x
t
, t) = max0, K x
t
= K
1
2
T+t
T
K =
1
2
Tt
T
K.
5.3.2 Smooth pasting
Above we stated a condition which generally holds at any (decision) boundary. However, we
have a more dicult question to answer, namely, how can we derive the optimal boundary, where
the investment should be initiated? Actually, our problem is to nd the optimal stopping time
(i.e. investment time) in the sense that the rm maximizes its value. It turns out that one can
often reformulate such a problem as a so-called free boundary problem with variational inequalities
describing the boundary. We will not set up all the technical details, but we will try to argue
why an additional condition provides us with a candidate for the optimal boundary and, hence,
the optimal time to undertake the investment. We quote the following verication theorem from
ksendal and Sulem (2001)[Theorem 2.3].
Theorem 2 (HJB-variational inequalities for optimal stopping and control) If a candi-
date, say , for a solution to our maximization problem (of the indirect utility function) and a
candidate u for the control satises
1. C
1
(S
0
) C(

S) (i.e. value-matching and smooth-pasting),

2. and (a lot of ) regularity conditions,
34 Decision making under uncertainty dynamic programming
x
(a) Case 1
x
(b) Case 2
x

t
(x)
(x)
F(x)
x

t
(x)
F(x)
(x)
[
h h
Figure 5.1: Illustration of the two dierent cases where F and do not meet smoothly at the candidate
for the optimal stopping point.
then is a solution to the maximization problem and u is an optimal control.
In order to gain some intuition on why the smooth pasting condition must hold at the optimal
stopping point x

t
we will consider the following argument based on Dixit and Pindyck (1994) and
Dixit (1993). We will restrict our selves to the case, where it is optimal to stop for a sucient large
decrease in x. Thus x
0
> x

0
.
Consider our candidate for an optimal stopping point, x

t
. Suppose that the continuation value
and the termination value do not meet smoothly at x

t
. Then one of two cases are possible. The
cases are illustrated in Figure 5.1. Either the derivative w.r.t. x (evaluated in x

t
) is highest for the
termination function, case 1, or the derivative is highest for the continuation function, case 2.
Suppose we are in case 1. Then x

t
cannot be the optimal switching point because we would
already have stopped. More formally:
> 0 : (x

t
+, t) > F(x

t
+, t),
hence it we would have stopped already at x

t
+ which contradicts our assumption of x

t
being
optimal. Loosely speaking, our candidate is too late.
Suppose instead that we are in case 2. This case is more tricky because we need to consider the
expected value of continuation. In order to handle this we will consider a heuristic discrete time
5.3 Binary decision problems 35
formulation of the problem. We assume that in a continuous time formulation, our state variable
follows the It o process
dx
t
=
t
dt +
t
dW
t
.
In a discrete time setting, we now assume that if we consider continuation for a period of length
t, then the state variable can increase or decrease with a magnitude of h. The increase occurs
with the probability p. Formally,
p =
1
2
_
1 +

t

2
t
h
_
, 1 p =
1
2
_
1

t

2
t
h
_
, h =
t

t.
In order to check wether or not our candidate is optimal, we compare the value we get, if we stop,
with the (expected) value we get, if we continue. If we stop, we get the termination value (x

t
, t).
If we instead continue for a period of length t we can use the Bellman equation to see that in the
point (x
t
, t), we have the continuation value
5
F(x
t
, t) = (x

t
, t)t +e
t
E
t
_
F(x
t+t
, t + t)

,
and the last term we 1st order Taylor expand in (x

t
, t)
= (x

t
, t)t +e
t
_
F(x

t
, t) +
_
pF
x
(x

t
, t)h + (1 p)
x
(x

t
, t)(h) +O
_
(h)
2
__
_
,
and if we also Taylor expand the exponential we get
= (x

t
, t)t
+ (1 t +O
_
(t)
2
_
)
_
F(x

t
, t) +
_
pF
x
(x

t
, t) (1 p)
x
(x

t
, t)
_
(h) +O
_
(h)
2
_
_
,
now since h =
t

t we can use O
_
t
_
= O
_
(h)
2
_
to obtain
= F(x

t
, t) +
_
pF
x
(x

t
, t) (1 p)
x
(x

t
, t)
_
h +O
_
(h)
2
_
,
and pluggin in the probabilities and rearranging we get
= F(x

t
, t) +
1
2
_
F
x
(x

t
, t)
x
(x

t
, t)
_
h +O
_
(h)
2
_
. (5.15)
Now, the rst term in equation (5.15) is equal the (x

t
, t) due to the value matching condition.
The second term in equation (5.15) is larger than zero because we are in case 2 check the gure.
5
We will only give a heuristic derivation.
36 Decision making under uncertainty dynamic programming
Thus we have demonstrated that we obtain a higher value if we continue than if we stop. This
contradicts the conjecture that x

t
is the optimal stopping point. Loosely speaking, our candidate
comes too soon.
In order to conclude, we have considered the two cases illustrated in Figure 5.1. None of the two
cases can occur at an optimal stopping point and, hence, the continuation value and the termination
value must meet smoothly at an optimal stopping point.
EXERCISE 5.1 To provide further intuition regarding the optimality of the smooth-pasting condition
consider a suciently dierentiable function f : (x, y) R
2
R. Suppose that y

is stationary point for f

in the sense that
f
y
(x, y

x=y

= 0. Consider the requirement that f(x, y) = 0 and the total derivative of

f and argue that this leads to the smooth-pasting condition in (x, y) = (y

, y

).
A more precise argument The above argument provides intuition for, why smooth pasting
must hold at an optimal boundary. However, it does not provide much insight into why we have
in fact solved an optimal stopping time problem, nor does it seem obvious that value matching
and smooth pasting can help to nd a candidate for an optimal boundary, if the state variable
can exhibit jumps. Based on Alili and Kyprianou (2005) we now provide another set of conditions
which are linked to the technical conditions omitted in Theorem 2. To exemplify, we consider the
exercise of an American put option with exercise price K and with an underlying stock price process
following expX under the pricing measure; the interest rate is constant and denoted as r > 0.
Let X = X
t
[t 0 be a Levy process; basically we can think of such a process as a nice
jump-diusion process.
6
Also, let y R and consider the rst passage time
y
for y, i.e.

y
inft 0[X
t
y.
The optimal stopping problem for the American put option is
f(x) = sup
T
0,
E
x
_
e
r
(K e
X

)
+

, (5.16)
and we denote the optimal stopping time as

. Let F be a candidate for the value function f, and

y
a candidate for the optimal stopping time of (5.16),

. Then F and
y
characterize f and

,
respectively, if (in addition to some regularity conditions) we have
6
A Levy process, dened on a ltered probability space (, F, {F
t
}, Q) satisfying the usual conditions, is a
continuous-time stochastic process that starts at 0, is c` adl`ag and has stationary independent increments. We
have already seen the two most well-known examples: the Brownian motion and the Poisson process. For more on
Levy processes see e.g. Applebaum (2004).
5.3 Binary decision problems 37
1. e
r(t
y
)
F(X
t
y
) : t 0, is a Q
x
-martingale, x R,
2. e
rt
F(X
t
) : t 0, is a Q
x
-supermartingale, x R,
3. F(x) = (K e
x
)
+
, for all x y,
4. F(x) (K e
x
)
+
, for all x R.
With the above, we then have that y can be used as our optimal boundary x

. Let us now
verify that conditions (1)-(4) are sucient. Since X
0
= x we have
F(x) = e
r0
F(X
0
)
(2)
sup
T
0,
E
x
_
e
r
F(X

(4)
sup
T
0,
E
x
_
e
r
(K e
X

)
+

E
x
_
e
r
y
(K e
X

y
)
+

(3)
= E
x
_
e
r
y
F(X

y
)

(1)
= F(x).
It follows that all inequalities are equalities and, thus,
F(x) = sup
T
0,
E
x
_
e
r
(K e
X

)
+

= E
x
_
e
r
y
(K e
X

y
)
+

.
Before we consider a more concrete application, we rst establish another framework: the
contingent claims analysis. Following this, we then have two frameworks in which we can formulate
our decision problems. In the next chapter we are going to employ these on various applications.
Chapter 6
Decision making under uncertainty
contingent claims analysis
6.1 Contingent claims analysis in continuous time
In this chapter we employ additional assumptions about the nancial market compared to what we
used with dynamic programming. In particular, we assume what is called a spanning condition.
This implies that the nancial market is sucient dynamic complete such that we can spand the
risk for our problem in place by trading in existing securities. Put dierently, we can therefore
construct a replicating portfolio in order to derive a partial dierential equation (pde) for the value
of the project.
Output is a traded asset
In order to motivate the setting consider a rm which produces widgets and the rm is able to sell
the widgets at some price, P, such that the rm has a prot ow equal to (x
t
, t) = x
t
. We now
make the simplifying assumption that the output can be treated as an asset which is traded on
nancial markets and, hence, the prot ow x is known from the nancial markets. As an example
think of copper (or any other frequently traded commodity). We assume that
(i) the price of the output follows a geometric Brownian motion
dx
t
= x
t
dt +x
t
dW
t
, (6.1)
and we can interpret as the expected price appreciation,
39
40 Decision making under uncertainty contingent claims analysis
(ii) there is an explicit or implicit dividend rate related to the output price; if implicit this is
often termed the convenience yield.
From the above it follows that the total expected rate of return from the output is

= + .
Furthermore, we dene F(x
t
, t) as the value of assets with the associated prot ow (x
t
, t).
Alternative 1 Now we make our rst alternative of a replicating portfolio:
1 in the risk free asset,
n units of the rms output.
The price of the portfolio, is then equal to
(x) = 1 +nx,
and the payo from the portfolio, when we hold the portfolio for an innitesimal (short) period of
length dt, is equal to the total dividend rate, i.e.
rdt for the risk free asset,
n(dx
t
+xdt) from the rms output,
where dx
t
is the capital gain and xdt is the dividend per unit. If we plug in x from (6.1), we can
write the total return per invested as
r +n(x +x)
1 +nx
dt +
nx
1 +nx
dW =
r +n( +)x
1 +nx
dt +
nx
1 +nx
dW.
Instead of the portfolio, we can also think of holding the rm, i.e. the right to all future prot
ow. The cost of buying the rm is F(x
t
, t). Over a period of length dt, the payo of the rm is
(x
t
, t)dt +dF(x
t
, t). From It os lemma, we have
dF(x
t
, t) =
_
1
2

2
x
2
F
xx
+xF
x
+F
t
_
dt +F
x
dW
t
,
hence the total return pr invested is
(x
t
, t) +
1
2

2
x
2
F
xx
+xF
x
+F
t
F(x
t
, t)
dt +
xF
x
F
dW.
We now apply the (standard) trick and choose n such that the portfolio risk is equal to the risk of
holding the rm, i.e.
nx
1 +nx
=
xF
x
F
,
6.1 Contingent claims analysis in continuous time 41
which implies
n =
F
x
F xF
x
.
Due to the principle of no arbitrage, we then conclude that the expected risk of the portfolio must
be equal to the expected risk of holding the rm, thus
r +n( +)x
1 +nx
=
(x
t
, t) +
1
2

2
x
2
F
xx
+xF
x
+F
t
F(x
t
, t)
which is reduced to
1
2

2
x
2
F
xx
(x, t) + (r )xF
x
(x, t) +F
t
(x, t) rF(x, t) +(x, t) = 0, (6.2)
and we have obtained our desired dierential equation for the value of the rm.
Now let us consider another replicating portfolio, where we want to hedge the value of the rm, F,
by taking a (short) position in the output, x. Remember that we can write the development in the
price of the state variable as
dx
t
= ( )x
t
dt +x
t
dW
t
, (6.3)
Alternative 2 The second alternative of a replicating portfolio is:
1 unit of the rm (project),
n units of the rms output.
The price of the portfolio is then equal to
(x
t
) = F(x
t
, t) nx
t
,
and the total return (over a short dt-period) of holding one unit of output consists of the dividend
part, xdt, and the capital gain part dx
t
. Hence the total return of the portfolio is
d(x
t
) = ((x
t
, t)dt +dF(x
t
, t)) n(xdt +dx
t
)
and using It os lemma we get
= (x
t
, t)dt +F
x
(x
t
, t)dx
t
+F
t
(x
t
, t)dt +
1
2
F
xx
(x
t
, t)dx
t
nx
t
dt n
_
( )x
t
dt +x
t
dW
t
_
=
_
(x
t
, t) + ( )xF
x
(x
t
, t) +
1
2

2
x
2
t
F
xx
(x
t
, t) +F
t
(x
t
, t) n( )x
t
nx
t
_
dt
+x
t
F
x
(x
t
, t)dW
t
nx
t
dW
t
42 Decision making under uncertainty contingent claims analysis
and we obtain a perfect hedge, if we choose n = F
x
. Then
d(x
t
) =
_
(x
t
, t)dt xF
x
(x
t
, t) +
1
2

2
x
2
t
F
xx
(x
t
, t) +F
t
(x
t
, t)

dt.
Since the portfolio has no risk, its (rate of) return must equal the risk free rate of return, i.e. for
any (x, t) (in the continuation region) we have
(x, t) xF
x
(x, t) +
1
2

2
x
2
F
xx
(x, t) +F
t
(x, t) = r
_
F(x, t) F
x
(x, t)x
t
_
hence
1
2

2
x
2
F
xx
(x, t) + (r )xF
x
(x, t) +F
t
(x, t) rF(x, t) +(x, t) = 0. (6.4)
Note that the two dierential equations (6.2) and (6.4) are identical. Of course, before we have a
well-specied problem we also need to setup the boundary conditions associated with the dierential
equations.
Output is not a traded asset: spanning
If the output cannot be treated as a traded asset, the above analysis fails. However, in the case
where there exists a traded asset which random price uctuations depend on the same Wiener
process as the output, we can still make a replication argument. In other words, the risk of the
other asset span the uncertainty in out state variable x. Suppose that this asset has a market price
process denoted X, and that the state variable and the market price of the replicating asset develop
according to the stochastic dierential equations
dx
t
= (x
t
, t)dt +(x
t
, t)dW
t
, (6.5)
dX
t
= A(x
t
, t)X
t
dt +
X
(x
t
, t)X
t
dW
t
. (6.6)
Let us also assume that the replicating asset (or portfolio) pays a ow dividend at rate D(x
t
, t).
The total total expected rate of return is then
X
(x, t) = D(x
t
, t) + A(x
t
, t). Then we can again
construct a replicating portfolio by buying the rm and selling short n units of X. The result of
the analysis leads to the following pde:
1
2

2
(x, t)F
xx
(x, t) +
_
(x, t)
(x
t
, t)

X
(x
t
, t)

X
(x
t
, t) r
_
F
x
(x
t
, t)
rF(x, t) +F
t
(x, t) +(x
t
, t) = 0. (6.7)
6.1 Contingent claims analysis in continuous time 43
We can rewrite 6.7 further if we introduce the market price of risk. The market price of risk,
denoted , is dened as the total rate of return of a traded asset minus the risk free rate of interest
relative to the volatility of the traded asset, i.e.
=

X
(x, t) r

X
(x, t)
. (6.8)
If fact, it is a well-known result, that if there is only one source of risk i.e. one Wiener process
then the market price of risk is identical for any traded asset, cf. Hull (1997). This is why we do
not denote with a mark for the particular asset. Note that the market price of risk enters in the
big bracket in (6.7) and, hence, we obtain
1
2

2
(x, t)F
xx
(x, t) +
_
(x, t) (x
t
, t)

. .

=
Q
F
x
(x
t
, t) +F
t
(x, t) rF(x, t) +(x
t
, t) = 0. (6.9)
Note that the factor with the market price of risk,
Q
, is exactly the drift of the underlying state
variable, if we state the stochastic dierential equation of the state variable under the risk neutral
measure (the Q measure).
In the case of geometric Brownian motions, we have that,
X
(x, t) =
X
,
X
(x, t) =
X
,
(x, t) = x and (x, t) = x. In the particular case where x spans itself (remember the assumption
of being traded!), we obtain that the market price of risk is
=

X
r

X
=
r

. (6.10)
and the PDE becomes
1
2

2
x
2
F
xx
(x, t) +
_
x x

F
x
(x
t
, t) +F
t
(x, t) rF(x, t) +(x
t
, t) = 0,
i.e.
1
2

2
x
2
F
xx
(x, t) + (r )xF
x
(x
t
, t) +F
t
(x, t) rF(x, t) +(x
t
, t) = 0, (6.11)
where we have used that the total expected rate of return is = . Hence we see that (6.11)
is equal to (6.4).
The dierence between the two approaches
Dierence between dynamic programming and contingent claims analysis? Dynamic programming
is a more general principle. With the additional assumptions used in order to apply the contingent
44 Decision making under uncertainty contingent claims analysis
claim analysis approach we obtain a specication of the discount rate. The appropriate discount
rate is the risk free rate of return, which we denote by r.
Of course, with the additional assumptions the pde from the continuation region part of the
Hamilton-Jacobi-Bellman equation is changed accordingly, and we can still apply the value matching
and smooth-pasting conditions in order to solve our binary decision problems.
Probabilistic solution of the dierential equation Consider the problem of obtaining the
value of a rm which receives the prot ow until some future point in time, say T. The prot ow
depends on some state variable denoted x. At termination, we assume that the rm is terminated
and that the owner of the rm receives the payo (x
T
, T). Thus we have the termination condition
F(x
T
, T) = (x
T
, T).
Furthermore, we assume that the state variable is driven by
dx
t
= ( )x
t
dt +x
t
dW
t
, x
0
= x
o
,
i.e. a geometric Brownian motion, where is the dividend rate or convenience yield and is the
total expected rate of return.
Let us use the dynamic programming principle in order to obtain a dierential equation for the
value of the rm, F. Again, we let be the appropriate discount rate. The value of the rm is
simply the discount value of receiving the expected prot ow until termination, i.e.
F(x, t) = E
t
_ _
T
t
e
(st)
(x
s
, s)ds +e
(Tt)
(x
T
, T)
_
(6.12)
But from (5.13) we know that
max
_
(x
t
, t) F(x
t
, t),
t
+F
t
(x
t
, t) + ( )x
t
F
x
(x
t
, t) +
1
2

2
x
2
t
F
xx
(x
t
, t) F(x
t
, t)
_
= 0,
(6.13)
and since the rm is not terminated before the nal date, we only have to consider the continuation
value, hence we obtain the problem
1
2

2
x
2
F
xx
(x, t) +F
t
(x, t) + ( )xF
x
(x, t) F(x, t) +
t
= 0, t < T (6.14)
subject to
F(x, T) = (x, T), x. (6.15)
6.1 Contingent claims analysis in continuous time 45
Suppose now that we want to use contingent claims analysis. In this case we know that the
appropriate discount rate is the risk free rate of interest and, hence, the dierential equation
becomes
1
2

2
x
2
F
xx
(x, t) +F
t
(x, t) + (r )xF
x
(x, t) rF(x, t) +
t
= 0, t < T (6.16)
subject to
F(x, T) = (x, T), x. (6.17)
This looks formally much like the dierential equation in the dynamic programming case and, thus,
if we literally replace with r and with r we obtain
F(x, t) = E
Q
t
_ _
T
t
e
r(st)
( x
s
, s)ds +e
r(Tt)
( x
T
, T)
_
, (6.18)
where we must change the evolution of the state variable to
d x
t
= (r ) x
t
dt + x
t
dW
Q
t
, x
0
= x
o
,
i.e. we apply the risk neutral measure in order to evaluate the expectation.
The result that the expected value is the solution to the dierential equation is more generally
known as the Feynman-Kac result. A more detailed analysis of this result can be found in e.g.
Due (2001, Appendix E).
6.1.1 Decision making under delayed information
6.1.2 Decision making under asymmetric information
Chapter 7
Real options analysis
In the previous chapter we have seen how to formulate a binary decision problem in a continuous-
time framework. From that chapter it turns out that the central elements in our analysis are the
dierential equation describing our claim F and the boundary conditions. Regarding the conditions
at a particular boundary we have argued that for our type of problem a value matching condition
must prevail and so must a smooth pasting condition, if the boundary is optimally chosen by a
decision maker.
7.1 A basic example
In this section we illustrate an example inspired by the classical paper of McDonald and Siegel
(1986). The example is also inspired by Dixit and Pindyck (1994, chapter 5 and 6). In the
following we assume that the nancial markets are suciently complete such that the contingent
claims analysis is applicable. Consider a rm with a perpetual option to invest in a project. The
project requires the lump-sum payment, the investment cost, I > 0 and in return it yields the
perpetual prot ow P
t
at time t. It turns out that the prot (ow) is the relevant state variable
in this example. The goal is now to establish the rms value of the investment option as well as
when to undertake the investment. Of course, when the rm considers to pay the investment cost,
it needs to know the present value of the future prots generated by the investment. Thus, to solve
this problem, we consider a two stage method and use backward induction. First, we establish the
value of the project, V (P). Second, we subsequent use V (P) in boundary conditions to derive the
47
48 Real options analysis
option value F(P) and the optimal investment trigger level P

.
1
We assume that the prot process P = (P
t
)
t0
follows a geometric Brownian motion under the
physical probability measure P, i.e.
dP
t
= P
t
dt +P
t
dW
P
t
, (7.1)
with initial value P
0
. is the capital gain rate of P and we assume that there is a marginal
convenience yield equal to > 0, such that the total rate of return is
P
= +. With a constant
risk free interest rate r and the simplifying assumption that P also represents the price of a traded
asset, we know that the market price of risk for P is =

P
r

, cf. (6.10). Therefore, the prot also

follows a geometric Brownian motion under a pricing measure Q (the risk neutral measure). The
drift rate under Q is
Q
= =

P
r

= ( + r) = r , thus
dP
t
= P
t
dt +P
t
dW
t
, (7.2)
where W is a Wiener process (standard Brownian motion) under Q. Since calender time plays no
role per se in the model, we know that the value of the project satises the ordinary dierential
equation
1
2

2
P
2
V

(P) +PV

(P) rV (P) +P = 0. (7.3)

Reasonable conditions for the projects value are that it becomes worthless if prots are 0 forever
and that this risk becomes zero as prots tends to innity. Moreover, we demand that since there
are no embedded options in the value of the project it must equal its fundamental component.
The fundamental component of a payment stream is the perpetual value of receiving the payment
stream forever, i.e. embedded options are neglected. Such an assumption implies that we rule out
so-called speculative bubbles. Specically, we calculate the fundamental component as
V (P) = E
_ _

0
e
rt
P
t
dt
_
=
P
r
=
P

, (7.4)
which obviously satises (7.3).
Above we solved step 1 (stage 2), so now we consider step 2 (stage 1). Naturally, the value of
the option to invest must satisfy an ODE basically as in (7.3). The central dierence is that the
1
In other applications than this example, one may prefer to model the project value V directly. In this case, we
can simply skip the rst step.
7.1 A basic example 49
rm receives no payments (positive or negative) before it pays the investment costs. Therefore, the
ODE is

2
2
P
2
F

(P) +PF

(P) rF(P) = 0, (7.5)

with boundary conditions
F(P) 0, P 0, (7.6)
F(P

) = V (P

) I, (7.7)
F

(P)[
P=P
= V

(P)[
P=P
. (7.8)
Before we go on to solve the problem, it is important to interpret the boundary conditions as they
are those which more precisely dene our model. Condition (7.6) says that the value of the option
to invest becomes 0 as the prot ow tends to 0. This is a reasonable assumption, because if the
prot is 0 then it will be 0 forever. The two other conditions are related to the optimal investment
threshold P

, i.e. the level of prot at which it is optimal for the rm to initiate the investment the
rst time this level is hit (or crossed). From section 5.3 we know that we should have value matching
as well as smooth pasting at the investment boundary. Value matching is exactly what is assumed
at (7.7). Here, the option value equals the value of the project minus the investment cost. Of
course, if one makes the analogy to a nancial option the condition is hardly surprising. It simply
says that the value of the option is equal to the value of the underlying minus the cost of getting
the underlying. In relation to value matching, we also need smooth pasting, which states that the
value function must have a smooth t at the exercise boundary. This is precisely condition (7.8).
Consider now the ODE (7.5). This is a second order linear homogenous ordinary dierential
equation with linear coecients, so it well-known that the solution is on the form
F(P) = A
1
P

1
+A
2
P

2
, (7.9)
where A
1
and A
2
are coecients depending on the boundary conditions.
2
Furthermore, the powers
are the roots in the characteristic equation

2
2
( 1) + r = 0, (7.10)
2
An equation like (7.9) is also called a Cauchy-Euler equation. This can be transformed to a second order
dierential equation with constant coecients by using the transformation P = e
t
. I leave this as an exercise for the
mathematically inclined student.
50 Real options analysis
with solutions
_
_

2
_
_
=
_
_
_
(

2
2
)+

2
2
)
2
+2r
2

2
(

2
2
)

2
2
)
2
+2r
2

2
_
_
_
. (7.11)
It can be shown that
1
> 1 and
2
< 0.
Recall that we have the absorbing condition (7.6). Due to the fact that
1
> 1 and
2
< 0,
F(P) will converge to unless A
2
= 0, so the condition implies that A
2
= 0. The value matching
condition (7.7) then implies that
F(P

) = A
1
(P

r
I,
whence it follows that
A
1
=
_
P

r
I
_
(P

1
(7.12)
and using the smooth pasting condition we get
F

(P)[
P=P
=
1
_
P

r
I
_
1
P

1
r
hence
P

=

1

1
1
I(r ). (7.13)
Using (7.13) in (7.12) provides us with the complete solution, i.e. we have now solved the rms
investment problem. From (7.13) we note that the right-hand side is larger than I, so V (P

) > I
implying that the rm should not invest just because the net present value of the investment is
positive. The net present value must be suciently positive such that it appropriately compensates
for the opportunity cost of investing, i.e V (P

) = I +F(P

).
For a concrete example, suppose that the rm must pay I = 20 as investment cost, that the
prot process is characterized by an expected growth rate of = 1% and volatility of =

0.2.
Finally, we let the risk free interest rate be r = 4%, which implies that the convenience yield is
= r = 3%. With these parameters we get A
1
= 20.8185 and P

= 3.25242. The option

value is depicted in Figure 7.1 as the blue curve together with the net present value (purple curve).
As seen in the gure, the option value approaches the net present value smoothly at P

, and for
P > P

the option value is simply the termination value, i.e. the net present value of the project.
7.2 Combined Brownian motion and Poisson process 51
1 2 3 4
20
40
60
80
100
120
F(P)
maxV (P) I, 0
P

Figure 7.1: Value of the option to invest, F. The parameters used are = 1%, r = 4%,
2
= 0.2
and the investment cost is I = 20. The optimal investment threshold, P

, is indicated above the

P-axis.
EXERCISE 7.1 Solve the SDE (7.2) and use this to show that (7.4) is correct.
EXERCISE 7.2 Show that the SDE for V is a geometric Brownian motion. Hint: Recall (7.4) and use
It os lemma.
EXERCISE 7.3 (= this will help you to read coming material and do future exercises) Show that the
example yields A
1
and P

as indicated in the text. Then make a plot like Figure 7.1.

EXERCISE 7.4 (= technical demanding exercise) Restate problem (5.16) for the basic example and show
that the solution to the rms investment problem satises the four suciency conditions (1), (2), (3), and
(4) for the reformulated problem (5.16).
7.2 Combined Brownian motion and Poisson process
N is a Poisson process with intensity . W and N are independent, so dW, N
t
= 0. Jump implies
reduction in the value V
t
of magnitude [0, 1), i.e. the value process can be described as
dV
t
= V
t
dt +V
t
dW
t
V
t
dN
t
. (7.14)
We now use the dynamic programming approach in order to evaluate the option to invest. As
52 Real options analysis
usual, we let be the discount rate.
3
In the continuation region, CR = V : F(V ) > V I, the
Hamilton-Jacobi-Bellman is
F(V ) = drift(F).
We therefore use It os Lemma for jump processes, cf. result (3.33), in order to obtain the dierential
of F:
dF(V
t
) = F

(V
t
)(V
t
dt +V
t
dW
t
) +
1
2
F

(V
t
)
2
V
2
t
dt + [F(V
t
V
t
) F(V
t
)]dN
t
=
_
1
2

2
V
2
t
F

(V
t
) +V
t
F

(V
t
)

dt +V
t
F

(V
t
)dW
t
+ [F(V
t
V
t
) F(V
t
)](dM
t
+dt)
=
_
1
2

2
V
2
t
F

(V
t
) +V
t
F

(V
t
) + [F((1 )V
t
) F(V
t
)]

dt
+V
t
F

(V
t
)dW
t
+ [F((1 )V
t
) F(V
t
)]dM
t
and since the last two terms only involve martingales we have
drift(F)(V
t
) =
1
2

2
V
2
t
F

(V
t
) +V
t
F

(V
t
) + [F((1 )V
t
) F(V
t
)]
hence the Hamilton-Jacobi-Bellman is
0 =
1
2

2
V
2
t
F

(V
t
) +V
t
F

(V
t
) ( +)F(V
t
) +F((1 )V
t
). (7.15)
As in the basic model in Dixit and Pindyck (1994, Chapter 5), we conjecture a candidate for
the solution to the dierential equation
F(V ) = AV

1
, (7.16)
where
1
solves
0 =
1
2

1
(
1
1) + ( )
1
( +) +(1 )

1
, (7.17)
where we have substituted = . The equation must be solved numerically. However, the
solution for the optimal investment triggering level of the value of the project is literally as before,
i.e.
V

=

1

1
1
I.
3
In the equivalent example in Dixit and Pindyck (1994, Chapter 5), it is assumed that the agents are risk neutral.
Therefore = r is applied.
7.3 Research and development projects 53
0.2 0.4 0.6 0.8 1
1.2
1.4
1.6
1.8
2
= 0
= 0.05
= 0.10
= 0.25
= 0.50
= 1.

Figure 7.2: The optimal level of investment for various down scale possibilities and intensities. The
parameters are = = 0.04( = 0), = 0.2. = 1 is calculated as = 0.999.
Figure 7.2 shows the optimal level, V

for various intensities, , and down scales, . From the gure

we note that a higher down scaling, i.e. , if a jump occurs, yields a lower investment triggering
level. A higher decreases the value of the investment opportunity and, hence, the value the option
to invest decreases. This in turn implies that the investment triggering level decreases. For a given
jump size, Figure 7.2 also shows that the net eect of a higher jump intensity is a decrease in the
investment triggering level. The basic intuition is as before, except that there is an o-setting eect
in the sense that a higher intensity increases the instantaneous relative variance of the option.
Most importantly, however, note that even a small positive jump intensity dramatically changes
the level of the optimal investment boundary.
7.3 Research and development projects
In this section we consider the investment problem of a rm investing in research and development
(R&D) projects. The real options framework provides a conceptual very nice framework to tackle
such seemingly very complicated investment problems in the sense that the real options framework
can help us to structure the problem. The benet of structuring an R&D project as a real options
problem is that we can get a better understanding of the underlying structure and factors and we
can analyze how these inuence the optimal investment strategy. In the following, we are going to
outline how such problems can be analyzed and the reader should obtain an idea of how similar, but
perhaps more complicated problems, can be dealt with. We base our short discussion on Schwartz
54 Real options analysis
and Moon (2000) and Schwartz and Zozaya-Gorostiza (2003).
7.3.1 Valuation of drug development
Schwartz and Moon (2000) outline how many of the complicated features of drug development
can be implemented in a real options approach. Many types of uncertainties are imaginable in
relation to developing (and evaluating) drugs. Schwartz and Moon (2000) focus on three types of
uncertainties
1. uncertainties about the investment cost,
2. uncertainties about the future payos from the investment,
3. the project may be terminated due to a catastrophic event.
Before we continue the valuation, we review the drug development process.
The process of developing a drug When a company decides to go ahead and invest in drug
development it typically faces a lengthy and complex process. This process can be separated into
multiple discrete phases (stages).
4
In the following we employ the drug development phases in the
USA as outlined in Schwartz and Moon (2000), but it seems reasonable that most countries will
have a similar setting.
Suppose the company has undertaken research and has discovered a molecule. The companys
researchers then perform laboratory experiments as well as testing on animals (excluding human
beings). This phase is denoted the preclinical testing phase. In this stage, the company also les
for proprietary position of their invention (e.g. patent rights). This stage takes about two years
and only 1% of the molecules survives from the preclinical testing into the next stage.
Being done with the preclinical testing and various kinds of formalities, the authorities can allow
the rm to initiate testing on human beings. The rst phase involving tests on human beings is
denoted as phase I clinical trials. In this stage 20-80 healthy volunteers are used to test the toxicity
and dosing of the drug. Phase I lasts for about two years, but it may meanwhile become clear
that the drug causes serious damages to human beings and, hence, the drug development should
not continue. About 70% of the phase I drugs survive until phase II. Phase II and phase III also
involves testing on human beings. In phase II the side eects of the drug as well as the dosing are
4
It may not be realistic that all of the phases are separate, but this is assumed to keep the framework tractable.
7.3 Research and development projects 55

0
(2)
1
(4)
2
(6)
3
(9)
4
(11.5)
[ [ [ [ [
1%, 1% 70%, 7 47%; 3.3 82%; 2.7 74%, 2
preclinical
testing
phase I phase II phase III
submission
to FDA
Figure 7.3: Illustration of the assumed structure of a drug development process. The time line
indicates that a change occur at some unknown time. The subsequent numbers in parentheses
provide the expected accumulated time. The numbers in the lower row show the expected success
probability and the accumulated success probability.
analyzed, whereas phase III primarily deals with demonstrating the statistical signicance of the
drug. The studies in phase III are designed to eventually obtain approval by the authorities (Food
and Drug Administration). Such studies are often of a large scale with 1000-3000 ill human beings
and, hence, are often very expensive. Moreover, the rm also builds a plant to manufacture the
drug commercially such that the rm can go ahead with selling the drug as soon as FDA as granted
permission. Expenses in phase III are reported to range from 50-80 millions. When the studies
in phase III are done, the information is handed over to the FDA. The FDAs handling of the
information takes about 2.5 years. The structure, expected time length, and survival probabilities
of the stages are summarized in Figure 7.3.
Valuation The valuation model employed by Schwartz and Moon (2000) is as follows. In order
to model cost uncertainty, the paper follows Pindyck (1993). Thus, let

K be the actual cost of
completion and denote the expected cost as K E[

K]. Let I be the rms rate of investment,
i.e. the rms control and assume that investments are bounded above such that I [0, I
m
]. With
z being a standard Brownian motion (uncorrelated with aggregated wealth), we assume that the
expected cost of completion follows the controlled diusion process
dK
t
= I
t
dt +(I
t
K
t
)
1/2
dz
t
, (7.18)
that is we assume technical uncertainty but not input cost uncertainty. It can be shown that the
variance of the cost to completion is
Var(

K) =
_

2
2
2
_
K
2
.
Since the future payos from production is unknown we model this asset value uncertainty. Let
56 Real options analysis
V be the value of the assets received upon successful completion of the project such that
dV
t
= V
t
dt +V
t
dw
t
, (7.19)
where w is a standard Brownian motion independent of z.
Finally, during the development of the drug it is possible that the process must be stopped
unexpectedly due to e.g. too many side eects of the drug. This catastrophic event is modeled as
a Poisson process with intensity .
As usual we let F be the value of the option to invest. The state variables are the asset value
and the cost to completion so F is a function of V and K. The resulting Hamilton-Jacobi-Bellman
equation is
sup
I
_
1
2

2
V
2
F
V V
+
1
2

2
IKF
KK
+ ( )V F
V
IF
K
(r +)F I
_
= 0
that is
sup
I
_
1
2

2
V
2
F
V V
+ ( )V F
V
(r +)F +
_
1
2

2
KF
KK
F
K
1
_
I
_
= 0, (7.20)
where r is the constant risk free rate of interest and is the risk-adjusted drift for the value of
the assets. Clearly, the investment rate enters in a linear fashion in equation (7.20) so the optimal
control is bang-bang. Hence, either the rm does not invest, or it invests at the maximum rate I
m
.
We refer to Schwartz and Moon (2000) for more details and discussion. However, to show how
the critical asset value changes between the various phases, we provide the base case result. Suppose
the preclinical stage has been done and let r = 5%, = 8%, = 0.5, = 0, = 0.35. Table 7.1
contains the remaining data for the stages as well as the resulting critical asset value. Initially, the
total expected cost to completion is 84 million and if the rm invests at the maximum rate in all
stages the expected time to completion is 9.5 years. The last row in the table demonstrates that
the critical level of the asset value is highly dependent on the stages. Initially, asset value must be
higher than 250 million before investment should take place albeit the expected total cost is only
84 million. This due to the fact that the process is very lengthy; even if there are no catastrophes
the rm must expect to wait about 10 years before the investment starts to yield a positive cash
ow to the rm. Unsurprisingly, the critical asset value decreases as more stages are completed.
Eventually, only the submission to the FDA remains. The rm should proceed to the FDA review
if asset value is higher than just 15 million because there only remains 10 million to be invested
(in expected terms).
7.4 Other models 57
phase I phase II phase III FDA review
expected cost (million) 4 10 60 10
max. investment rate I
m
m/yr 2 5 20 4
expected waiting if max. investment 2 2 3 2.5
probability of failure, 0.15 0.25 0.06 0.10
critical value V

250 215 180 15

Table 7.1: Data for and evaluation of drug development.
The framework by Schwartz and Moon (2000) can be changed to focus on other elements. The
model by Schwartz and Zozaya-Gorostiza (2003) is such an example.
7.3.2 Acquisition and development of an information technology project
In this paper, Schwartz and Zozaya-Gorostiza (2003) study valuation of IT investment projects.
They categorize the projects into two groups: development projects and acquisition projects. A
development project takes time to complete, i.e. investments can not be done faster than at a rate
I
m
. In contrast, an acquisition project allows the rm to complete the investment immediately
(I
m
= ) but the investment cost changes over time.
(incomplete)
7.4 Other models
incomplete...
Murto (2007) considers the investment problem when the investment cost can drop in the future.
This is extended by ?). They consider the basic investment problem in a setting where earnings as
well as the (irreversible) investment cost can suddenly decrease due to technological advances. In
particular, the future decreases in earnings as well as in the investment cost are unknown prior to
their occurrence.
Chapter 8
Competition and the incentive to
invest
[incomplete]
8.1 Limited competition
The eects of monopoly or duopoly in a R&D framework: Miltersen and Schwartz (2004)
Oligopoly
8.2 Perfect competition
Perfect competition in a Cournot-Nash equilibrium is studied in Grenadier (2002).
Se also Novy-Marx (2007) and Back and Paulsen (2009). The latter paper criticizes the work
by Grenadier (2002) in that they argue that the investment boundaries in that paper are not
equilibrium strategies in a closed-loop (feedback) equilibrium.
59
Part III
Capital structure theory: capital
budgeting, trade-o theory and other
issues
61
Chapter 9
Introduction to capital structure
theory
In the following chapters well go through an overview of the literature related to some parts of
capital structure theory. To give a complete and exhaustive treatment of the literature would,
however, be too much. The focus in these notes will therefore be on theoretical capital structure
models in continuous time which use the so-called structural modeling framework. The notes do
not include information or principalagent problems, rather, the models considered in this survey
have symmetric information and management acts on of behalf of the equity holders.
1
Before discussing the various issues, it may be of interest to ask two overall questions. Firstly,
what is understood by a capital structure and secondly, why is a theory about the capital structure
of a rm interesting, i.e. is or should a rm be concerned about its capital structure? If it can
be argued that rms are concerned about their capital structure, one can go a step further and
investigate which factors have major inuence on the capital structure policy of a rm.
By a capital structure one thinks of the composition of the capital of a rm, i.e. which kind
of debt and equity does a rm have and how large a fraction of the rms capital does the debt
or equity account for? For instance, one may observe a rm which capital consists of 33% equity
and 67% debt; in this case one also says that the rm has a leverage of 67%.
2
The equity and the
debt can be further divided, e.g. the debt may be 25% bank debt and 75% corporate bonds, where
again the bonds, say, can belong to dierent rating classes. Further, structural models are usually
1
A future version of the notes plans to include information related corporate nance topics.
2
Here, we use the term leverage for the ratio of the rms (market) value of debt to the rms total (market)
value. In the literature other denitions are used, e.g. the ratio of the rms debt to its equity.
63
64 Introduction to capital structure theory
only directly concerned with capital from external nancing such as traded shares and bonds or,
more to the point, structural models assume that all debt and equity can be treated in essentially
the same way as traded securities.
3
Capital structure theory is then concerned with studying the
inuence of dierent factors on the mix of the rms debt and equity.
In a complete and frictionless world there would be no reason to develop capital structure
theories. This insight is due to Modigliani and Miller (1958). The intuition is that if an investor
who holds bonds in a rm is unhappy with the capital structure of the rm, he can just trade
on the nancial markets so as to bring his positions in accordance with his preferences. But this
result does not hold if one introduces some of the frictions observable, e.g. agency problems, a tax
advantage to debt, and transaction costs. Thus, when one models a world with frictions there may
be scope for interesting capital structure theories. As an example, which is the dominating feature
for the papers considered in the later chapters, suppose that debt has a tax advantage. Intuitively,
this should make rms want to have very high leverage. Suppose, on the other hand, that there is
a cost of debt, for instance if a default of the debt is costly.
4
Then, the rm owners must balance
the trade-o between saving taxes and paying huge yields to the debt holders because the debt
holders will foresee the large default costs. Now one may gain interesting economic insights by
understanding which factors have an inuence on the capital structure of a rm.
Academics interested in capital structure theory have generally used two types of modeling
frameworks in continuous time, either the structural based framework or the reduced form frame-
work. Early examples of structural models are Merton (1974) and Black and Cox (1976). Recent
examples are Leland (1994b), Mella-Barral and Perraudin (1997), and Goldstein, Ju, and Leland
(2001). A few examples of reduced form models are Jarrow, Lando, and Turnbull (1997) and Due
and Singleton (1999). Since the reduced form models take a less detailed view of the rm, hence
the name reduced, the reduced form models are more concerned about the pricing of corporate
bonds than of why a rm has a particular amount of leverage, issues debt with a particular maturity
structure etc. Due and Lando (2001) provide an interesting link between the two frameworks.
We shall only cover structural models in these notes.
3
In a study on data from USA, Japan, Germany, France, Italy, United Kingdom, and Canada, Rajan and Zingales
(1995) report that equity accounts for 2842.2% of the liability of the rm. Moreover, external nancing is in the
area of 2356%.
4
One should think of a default as an event where the rm for some reason or another does not meet its debt
obligation. In practice, a default could happen if an interest payment is missed; this does not necessarily imply that
the rm is liquidated.
65
What characterizes a structural model is that the model builds on assumed knowledge of (at
least) one state variable which is a fundamental of the rm. The state variable can either be the
unlevered rm value, the earnings before interest and taxes, or something similar. However, since
the most of the models assume that the investments are independent of the nancing methods
it seems most natural to use the earnings of the rm as the state variable. Furthermore, if the
earnings of the rm are earnings before taxes, then it is easier to study tax eects of the capital
structure. In most cases it is assumed that the state variable follows a geometric Brownian motion
because of mathematical tractability. Given that there exists a risk neutral measure (also denoted
the equivalent martingale measure), prices of debt and equity can then be evaluated as the solution
to a partial dierential equation (pde). The various structural models have the most obvious
dierences in the boundary conditions associated with the pde. Equivalently, one can use the
Feynman-Kac representation to have the prices expressed as expectations under the martingale
measure, see Due (2001).
As a simple example of how the boundary conditions change with dierent models suppose that
if the state variable hits a certain level written in the debt contracts, then the control rights of the
rm must be handed over from the equity holders to the debt holders. In this case equity has value
0 when the state variable hits the specied level. A more complicated model could try to include
the incentives of the debt and equity holders as part of the decision of when to default. Even if the
equity holders are assumed to get 0 at default, this model has a more complicated analysis because
now the equity holders can decide when to give up the rm, i.e. this model has an associated free
boundary problem.
The rest of the lecture notes is organized as follows. In Section 10.1 we shall use the Black-
Scholes-Merton framework to describe the rst attempt by Merton (1974) to price the debt and
equity of a rm. Section 10.2 starts with a more general description of how to model the debt
and equity of a rm as contingent claims on an underlying state variable related to the rm. A
particular model by Black and Cox (1976) is treated as the main example and the subsequent
papers are related to this representative model.
The structural modeling framework has been used to model many dierent aspects of a rms
capital structure. The authors opinion is that one obtains the best overview of the literature
by categorizing it into major topics rather than beginning with the Merton (1974) model and
then chronologically going through all models until the present day. Inevitably, some models have
connections to more topics, however, I hope that the reader will be satised with my choice of
66 Introduction to capital structure theory
categorization.
The rst topic I look at in Chapter 11 is how to model the tax advantage to debt over equity.
Here, the emphasis will be on the models by Leland (1994b), Goldstein, Ju, and Leland (2001), and
the model presented in Christensen, Flor, Lando, and Miltersen (2002a). The survey then considers
some other issues of the capital structure of the rm in Chapter 13. Here, I turn to the models by
Mella-Barral and Perraudin (1997) and Mella-Barral (1999) in the Sections 13.113.2 which look
at strategic debt policies. Section 13.3 briey looks at the Fan and Sundaresan (2000) model which
analyzes the optimal dividend policy. The model by Fran cois and Morellec (2001) in Section 12.1
is close to Fan and Sundaresan (2000) but focuses on the decision to either restructure or liquidate
the rm. In Section 13.4 I consider the model presented in Flor (2008) which is concerned with the
impact of the outside value of the assets of a rm. The maturity question is treated in the Sections
13.513.6 by looking at Leland and Toft (1996) and the model presented in Flor and Lester (2002).
We close the lecture notes in Chapter 17 by mentioning some of the papers which are not otherwise
discussed and, nally, I conclude.
Before we continue a caveat is in place. In its current form, the notes lacks references to
empirical literature. The reader is therefore encouraged to search for additional empirical results
in order to reect upon the theories presented below.
Chapter 10
Introduction to structural models
The Black-Scholes-Merton framework, i.e. the models in Black and Scholes (1973) and Merton
(1973), is the fundamental framework used when structural models try to study the various aspects
of the capital structure of a rm. The Merton (1974) model initiated the structural approach.
In most of the models considered, the evolution of the rm is described by a single state
variable; let us denote it by x. One can think of the state variable as e.g. the value of the rm or
the earnings of the rm. In this chapter and in Chapter 11 we shall use the earnings as a state
variable. However, because of this change in the interpretation of the state variable compared to
the some of the original articles, the formulae will be slightly changed but the idea of the models
will be the same as in those articles. The main reason for having the same state variable in the
models is that it facilitates comparisons. For models with taxes the (before tax) earnings setup
makes it easy to derive what the dierent claimants receive after tax. We shall elaborate more on
the choice of the state variable in Section 11.1.
Before setting up a general model and discussing the assumptions most commonly used in the
structural models, it will be suitable to give an illustrative example of a simple structural model.
Let us consider the Merton (1974) model.
1
10.1 The Merton model
The state variable is modeled directly under the so-called risk-neutral (or local equivalent martin-
gale) measure, denoted Q, under which all deated gain processes are martingales. It is well known
that the existence and uniqueness of Q will exclude arbitrage and imply complete markets, see
1
In Merton (1974) the state variable is called V and represents the value of the rm.
67
68 Introduction to structural models
Due (1996) for technical details and for a treatment of the theories based on Harrison and Kreps
(1979) and Harrison and Pliska (1981).
For tractability suppose that the earnings under Q develop as the geometric Brownian motion
given below
dx
t
= x
t
dt +x
t
dW
t
, x
0
= x(0). (10.1)
Here is the expected growth rate of the earnings of the rm (under the Q-measure) and is termed
as the drift. W is a standard Brownian motion. is the volatility and can be interpreted as the
instantaneous standard deviation of the earnings.
2
The risk free interest rate, r, is assumed to be
constant.
3
To have well dened price processes the expected growth rate must be less than the risk
free interest rate, i.e. < r (see e.g. (10.2) below). Note, that the drift (rate) is under the
pricing measure. Thus, if for a moment plays the role as the market price of risk, the expected
growth rate under the objective probability measure (often denoted as P) is
P
= + and,
hence, observing growth rates in real life data higher than the interest rate r is not a problem in
this model.
As previously noted, all deated gain processes are martingales under the Q-measure. We now
use this to derive the price process for the claim which gives the owner the x cash ow forever; denote
this price process V . To do so, dene X to be the cumulative dividend process, i.e. X
t
=
_
t
0
x
s
ds
and note that dX
t
= x
t
dt. Furthermore, dene the deator Y
t
= e
rt
, hence dY
t
= rY
t
dt. Finally,
the gain process associated with the claim is
G
t
= V
t
+X
t
,
and the deated gain process is
G
Y
t
= V
Y
t
+X
Y
t
,
where
V
Y
t
= Y
t
V
t
,
X
Y
t
=
_
t
0
Y
s
x
s
dt.
2
More precisely,
2
x
2
t is the quadratic variation at time t of the process in (10.1).
3
Currently there are no taxes in the model. When taxes are introduced, r is used as the appropriate discount rate.
10.1 The Merton model 69
By denition of the Q-measure, G
Y
is a Q-martingale, see (Due 1996, ch. 6K), from which it
follows that for all t > 0, G
Y
0
= E
Q
0
_
G
Y
t

, i.e.
V
Y
0
+X
Y
0
= E
Q
0
_
V
Y
t
+X
Y
t

and since Y
0
= 1 and X
Y
0
= 0 one gets
V
0
= E
Q
0
_
e
rt
V
t
+
_
t
0
dX
Y
s
_
,
= e
rt
E
Q
0
_
V
t

+E
Q
0
_ _
t
0
e
rs
x
s
ds
_
= e
rt
E
Q
0
_
V
t

+
_
t
0
e
rs
E
Q
0
_
x
s

ds
and it follows from (10.1) that
V
0
= e
rt
E
Q
0
_
V
t

+
_
t
0
e
rs
x
0
e
s
ds
= e
rt
E
Q
0
_
V
t

+
x
0
r
_
1 e
(r)t
_
and for t we get
V
0
=
x
0
r
, (10.2)
as the initial value of the price process.
4
One immediately observes that the relation between the
price process and the earnings is just a constant factor. This factor,
1
r
, is the price-earnings
ratio. The linearity in (10.2) implies that the price process is itself a geometric Brownian motion,
i.e.
dV
t
= V
t
dt +V
t
dW
t
. (10.3)
Thus one sees that it is all the same if either x or V is used as the state variable.
Now consider a rm which has a single class of debt outstanding where the time until maturity,
T, is given. The bond is a zero-coupon bond with principal P. There are no costs associated with
a default of the debt. If the equity holders do not pay the promised principal at maturity, then the
4
Note also that when the limit is taken, I have assumed that there are no bubbles in the sense that E
Q
0
[e
rt
V
t
]
0, t .
70 Introduction to structural models
5 10 15 20 25 30
5
10
15
20
25
30
E
v
D
P
x
r
Figure 10.1: The payo diagram for the Merton model. The principal of the debt is 10, as indicated
by P in the gure. The debt, the equity, and the rm values are marked D, E, and v = E + D,
respectively.
rm is immediately handed over to the debt holders. Since equity has limited liability its value is
never negative. Therefore, at maturity, the value of the equity, E, and debt, D, are
E(x
T
, T; T) = max
_
x
T
r
P, 0
_
D(x
T
, T; T) = min
_
P,
x
T
r
_
= P max
_
P
x
T
r
, 0
_
.
The payo structure of the equity and the debt is illustrated in Figure 10.1.
Looking at the pay-o structure immediately reveals how the debt and equity can be interpreted
and priced as contingent claims on the earnings, x.
5
The equity is a European call option on the
rm. The debt is equivalent to having a bank account that pays out P at time T and a written
European put option on the value of the rm. Both the call and the put have P as the strike price.
With this insight and the Black-Scholes model at hand one easily nds the equity and debt prices
at time t as
E(x
t
, t; T) = e
(r)(Tt)
x
t
r
N(d
1
) e
r(Tt)
PN(d
2
),
D(x
t
, t; T) = e
r(Tt)
P
_
e
r(Tt)
PN(d
2
) e
(r)(Tt)
x
t
r
N(d
1
)
_
,
5
As noted above, one could instead use V from (10.2) as the underlying. This will give the same results.
10.1 The Merton model 71
where
d
1
=
log
_
x
t
r
P
_
+ ( +
1
2

2
)(T t)

T t
,
d
2
= d
1

T t.
A particularly nice feature of the Merton model is that it gives an intuition of how to think
of equity and debt because all the comparative static analyses from the Black-Scholes model can
immediately be applied. For instance, if the volatility of the earnings of the rm increases ex post,
then, ceteris paribus, one would expect this to benet the equity holders and hurt the debt holders.
6
The reason is that the equity holders do not bear the downside risk. As a matter of fact this eect
gives rise to the so-called asset substitution problem introduced by Jensen and Meckling (1976)
where the equity holders ex post try to substitute assets to make the rm value more volatile.
Leland (1998) studies this matter in a structural framework and his ndings indicate that agency
costs due to the asset substitution problem are small.
As mentioned earlier (see Footnote 1), we work with the rms earnings and not the value of the
rm as the state variable. To implement the presented model as close to the Merton (1974) model
as possible, we have implicitly assumed that equity holders receive no dividends before the debt
matures. That is, the problem is formulated as a real option investment problem, where equity
holders have to pay the investment costs equal to the debts principal, before they get access
to the rm. In other words, the cash ow between debt issuance and maturity is foregone. As a
consequence, the market value (at time t) of debt and equity is less than the perpetual value of the
rms earnings. It may therefore be more natural to reinterpret the model as follows. The debt
contract is as before, but the equity holders receive the cash ow x until the debt matures. The
time t value of receiving this cashow is
E
Q
t
_
_
T
t
e
r(st)
x
s
ds

=
x
t
r
(1 e
(r)(Tt)
).
Thus the equity value is the above value added to the call option value stated earlier. Using the
reinterpreted model, we nd that the market value of debt and equity is x
t
/(r), i.e. the perpetual
value of the rms earnings.
6
The terms ex ante and ex post are used relative to the point in time when the debt contract is settled. Thus,
ex ante means that the debt contract has not been written and ex post means that the debt contract is xed.
72 Introduction to structural models
10.2 A more general model
The Merton (1974) model gives a nice and simple framework to model the equity and debt of a
rm. However, a closer inspection of the model makes it clear that a lot of important matters are
untouched, e.g. the type of the debt and why there is debt at all. Hence, it seems natural to try to
develop a model which more clearly illustrates how widely the structural framework can be used.
The Black and Cox (1976) model is suitable for this purpose and can serve as a reference for the
other models in the following.
10.2.1 The standard assumptions
Most structural models typically make the same basic assumptions as in Black and Cox (1976).
Later models dier in the economic stories, as e.g. varying maturity of debt or strategic behavior
of the equity holders. Following Black and Cox (1976) it is assumed that
1. assets are completely divisible, and trading takes place in continuous time,
2. individuals are small traders in the sense that they take the prices they observe as given,
3. assets can be sold long or short up to an innite amount,
4. there is a money market account with a constant risk free interest rate, i.e. borrowing and
lending rates are identical on this account. Hence, at time t a risk free zero coupon bond
maturing at time T has a market price of e
r(Tt)
,
5. the nancial markets are complete and free of arbitrage,
6. the evolution of the contingent claims is driven by a single state variable, denoted x, and
perhaps also time dependence. The state variable is given as a geometric Brownian motion
and it is independent of the capital structure of the rm,
7. there are no frictions in the sense of taxes or transactions costs,
8. the value of the rm is invariant to its capital structure (so Modigliani and Miller (1958)
applies).
The above assumptions will be applied henceforth if nothing else is mentioned. However we shall
also assume that
10.2 A more general model 73
9. the modeling is done under the risk neutral measure denoted Q(i.e. a given (locally) equivalent
martingale measure).
Of the above assumption, assumptions 16 can be considered as the basic assumptions. In
a perfect and frictionless world, assumption 8 follows, in fact, from the basic assumptions and
assumption 7. Later we shall consider papers which use a tax advantage to debt. In these papers,
the assumptions 7 and 8 will not hold.
Typically, one is interested in studying the spread between the yield on the debt and the risk
free interest rate, r. This spread is also termed the yield, the credit, or the default spread. To
minimize the complexity of a model, one often assumes that the instantaneous risk free interest
rate is constant. The constant interest rate assumption 4 is relaxed in a few papers, e.g. Shimko,
Tejima, and van Deventer (1993); Collin-Dufresne and Goldstein (2001) have an interesting paper
with stochastic interest rates and where the rm can readjust its capital structure.
As is now common knowledge, the assumptions 5 and 9 are closely related. In fact, if one
restricts oneself to discrete time nite state space models, there exists a unique equivalent martingale
measure if and only if the nancial markets are complete and free of any arbitrage opportunity (i.e.
arbitrage trading strategies). In continuous time technical details disturb this result. To simplify
the modeling we apply assumption 9. This implies that the nancial markets are complete and free
of arbitrage. The completeness assumption essentially means that the price of a new traded asset
can be calculated from the prices of existing traded assets. For more details on these issues, see
Due (1996).
Like Merton (1974) most other models assume that the state variable is a geometric Brownian
motion. The earlier models interpret the state variable as the value of the rm whereas newer
models often interpret the state variable as the earnings of the rm. In these notes we shall usually
interpret the state variable as the earnings of the rm.
If the state variable is the price of a traded asset, e.g. the value of the rm, then the drift is
r under the Q-measure, if there are no dividends. This specication of the drift does not hold,
when the state variable is interpreted as the earnings of the rm. Thus, as in (10.1), one must
exogenously specify the drift. Let us denote the drift under the Q-measure. As in Section 10.1,
we assume that the earnings of the rm evolve according to
dx
t
= x
t
dt +x
t
dW
t
. (10.4)
74 Introduction to structural models
As previously explained, the earnings in (10.4) can be related to the value of perpetually receiving
the earnings. We shall return to this matter in Section 11.1.
When one analyzes the capital structure of a rm, it is evident that the incentives of the agents
with relations to the rm are essential. Agency conicts may have a huge inuence on the valuation
of the claims of the rm. In the models we shall consider the most frequently used agency groups
are equity holders and debt holders. In the standard corporate nance literature there is also a
tradition of looking at equity holders versus managers, i.e. a principal-agent problem. Albeit this
is an interesting matter, we henceforth completely omit any conicts between equity holders and
managers, and also conicts within the group of the equity holders will be ruled out. One may
therefore think of the equity holders and manager(s) acting as a single unit, sometimes termed
as an owner-manager or entrepreneur. Furthermore, we shall also consider the debt holders as a
single group with no internal conicts. This matter is, of course, relevant when the question of
debt renegotiation comes into play; Hege and Mella-Barral (2002) consider this problem.
10.2.2 The model
In the Merton (1974) model the rm issued a single class of debt. The bullet loan type of debt
gives equity and debt a European option character. As a result, equity and debt are easily priced
by the Black-Scholes formula. However, corporate debt often pays interests prior to the maturity
of the debt, and debt contracts can have protective covenants. Black and Cox (1976) recognize
that there are four sources which can contribute to the value of a claim, namely
1. the value of the claim at maturity,
2. the value of the claim at a lower boundary of the state variable,
3. the value of the claim at an upper boundary of the state variable,
4. value added during the life of the claim.
Obviously the boundary of the state variable must be dened according to some particular economic
story. As long as the state variable does not cause any action, the state variable is in the so-called
continuation region. In the Merton (1974) example the continuation region is (0, ). Another
example is that the rm is liquidated due to a protective covenant when x = C, where C is the
coupon to the debt. Then the continuation region is (C, ).
10.3 Contingent claims pricing by the pde-approach 75
Given a fully specied model, one is interested in deriving the values for the contingent claims.
The dierent models use one of two ways to calculate the price of a contingent claim. Either a
model uses a partial dierential equation (pde) which has the price of the contingent claim as the
solution, or a model uses the probabilistic approach which states prices as an expectation under
the risk neutral measure. The two methods may appear very dierent, however, since each method
gives the price of a traded contingent claim, there is a connection between the two methods. Indeed,
as is well known from the Feynman-Kac result, the two methods are equivalent, see Due (1996).
In the following we apply the two dierent methods.
10.3 Contingent claims pricing by the pde-approach
The rm can issue contingent claims depending on the state variable, x. The claims we consider
later on are equity, E, and corporate debt, D. For now, consider a traded contingent claim on the
earnings and time. Denote the price of the claim as F(x
t
, t).
Given the assumed process for the state variable and since it is assumed that there is a unique
(locally) equivalent martingale measure, the nancial markets are complete and free of arbitrage,
see Due (1996). Thus, the claim F can be replicated in the market, and the standard valuation
technique applies, which yields a pde. Let us reveal how one can derive that F is a solution to a
pde. The characteristics of the claim F also determines the boundary conditions that belongs to
the pde. For a start, one needs to assume that F : R
+
R
+
R is suciently smooth, in this
case F C
2,1
, such that It os Lemma can be applied. From It os Lemma one has
7
dF(x
t
, t) =
F(x
t
, t)
x
t
dx
t
+
F(x
t
, t)
t
dt +
1
2

2
F(x
t
, t)
x
2
t
d

x
t
_
=
_
F(x
t
, t)
x
t
x
t
+
F(x
t
, t)
t
+
1
2

2
x
2
t

2
F(x
t
, t)
x
2
t
_
dt +
F(x
t
, t)
x
t
x
t
dW(t)

=
F
dt +
F(x
t
, t)
x
t
x
t
dW(t). (10.5)
In this derivation we now extend the denition of a claim such that the claim holder receives
a continuous payout rate h : R R
+
R with the associated cumulative dividend process H
t
=
_
t
0
h(x
u
, u)du. Thus
dH
t
= h(x
t
, t)dt. (10.6)
7
Henceforth

= symbolizes a denition. In (10.5)
F
is dened from the above equation.
76 Introduction to structural models
Because F is a price process under the Q-measure, it is well known from standard nance theory
that the drift in (10.5) must be equal to the risk free interest rate corrected for the payout rate h.
Therefore one has from (10.5) and (10.6) that

F
= rF(x
t
, t) h(x
t
, t),
hence from (10.5) and rearranging terms
0 =
1
2

2
x
2

2
F(x
t
, t)
x
2
t
+x
t
F(x
t
, t)
x
t
+
F(x
t
, t)
t
rF(x
t
, t) +h(x
t
, t), (10.7)
which is the wanted pde.
8
We have chosen to let the claim holder receive a dividend rate in the derivation above because
in the models where the state variable is the earnings of the rm it is natural to think of part of the
value of the claim as stemming from the dividends paid out from the earnings, cf. source 4 previously
mentioned. The dividend rates most often used are ane in the state variable, i.e. h(x
t
, t) = Ax
t
+B,
where A, B R. In the corporate bond example the debt holders continuously receive the coupon
rate C and the equity holders receive the rest so h
D
(x
t
, t) = C and h
E
(x
t
, t) = x
t
C.
To have a well specied pde it is not enough to know the dierential evolution given by (10.7).
One further needs the boundary conditions. These boundary conditions are associated with the
sources 1 3 previously mentioned. In the next section I consider the probabilistic approach by
Black and Cox (1976).
10.4 The probabilistic approach to pricing claims
The pricing of contingent claims by the probabilistic approach as used in Black and Cox (1976)
directly shows how much each of the four sources contributes to the value of the claim. Dene
X
i
(x
t
, t), i = 1, . . . , 4, to be the value added by the four sources such that a contingent claim on
the earnings, with price denoted F, has value F(x
t
, t) =

4
i=1
X
i
(x
t
, t).
The economic theory dening the boundaries makes it evident that the possibilities 1, 2, and
3 are mutually exclusive. As an example, think of a zero-coupon corporate bond (the Merton
example) which also is callable. This would aect the upper boundary condition 3. But if the bond
has already matured, it cannot be called and if the bond has been called it cannot mature later
on. Source 4 can add positive and negative values to the claim. In the bond example, the coupons
8
For readability I henceforth abbreviate the derivatives as

2
F(x
t
,t)
x
2
= F
xx
,

2
F(x,t)
x
= F
x
, and
F(x
t
,t)
t
= F
t
.
10.4 The probabilistic approach to pricing claims 77
1 2 3 4 5 6 7 8 9 10
0.5
1
1.5
2
2.5
3
3.5
4
x
x
x
T

Figure 10.2: The EBIT process, x, and lower, x, and upper, x, boundaries associated with a
contract. The vertical dashed line at T marks the time of maturity.
paid to the debt holders is a positive source. If the earnings are not sucient to cover the coupon
payments, the equity holders must pay in money to refrain the debt holders from declaring the rm
bankrupt, so in this case the equity has a source which can contribute with negative values.
In this section we assume that the values obtained from any of the sources are given by a known
function written in the contract dening the contingent claim. Furthermore, the agents perfectly
respect their contractual obligations if they can. Thus we exclude models in which the equity
holders may act strategically to achieve concessions from the debt holders or models where it is the
equity holders incentives that determine when the rm is liquidated or the capital is restructured.
In later chapters we shall consider models which study these interesting issues.
Figure 10.2 illustrates how each source contributes to the value of the claim. Figure 10.2 shows
the lower boundary, x, the earnings process, x, and the upper boundary, x. In the example the
debt matures at T = 4 as marked by the left vertical dashed line.
Until the maturity of the debt and as long as the rm is not reorganized, the holder of the
F-claim receives the payout rate x
4
(x
s
, s). If the process of the earnings hits neither the lower nor
the upper boundary, as in the illustrated example, the debt will mature. At maturity the holder
of the F-claim will receive a payment of size x
1
(x
T
), where x
1
is a function which only depends
on the value of the state variable at maturity. If instead the maturity of the debt is T = 6, the
earnings will hit the lower boundary at approximately = 5.4. In such a case the claim holder
receives x
2
() but nothing from the sources 1 and 3. Suppose instead that the debt matured at
T = 10 and that the lower boundary is not reached prior to maturity, then the earnings will hit
the upper boundary at approximately = 9.5. Here, the claim holder receives x
3
() but nothing
78 Introduction to structural models
from the sources 1 and 2.
Consider a > t. Let (x

, [x
t
, t) denote the conditional distribution of x

given that at time

t the earnings are x
t
and that x
s
< x
s
< x
s
, s [t, ], i.e. the lower or upper boundary has not
been hit. The value from the maturity and the intermediate payos can then be written as
9
X
1
(x
t
, t) = e
r(Tt)
_
x
T
x
T
x
1
(x
T
)d(x
T
, T[x
t
, t) (10.8)
X
4
(x
t
, t) =
_
T
t
e
r(st)
_ _
x
s
x
s
x
4
(x
s
, s)d(x
s
, s[x
t
, t)
_
ds. (10.9)
Now, let be the rst passage time to the upper boundary, x, when at time t the earnings are
x
t
. Let

( [x
t
, t) be the distribution of the rst passage time . Similarly, let ([x
t
, t) be the
distribution of the rst passage time to the lower boundary, where the rst passage time is denoted
.
Hence, the values added from the lower or upper boundaries are
X
2
(x
t
, t) =
_
T
t
e
r(t)
x
2
()d([x
t
, t) (10.10)
X
3
(x
t
, t) =
_
T
t
e
r( t)
x
3
( )d

( [x
t
, t). (10.11)
As noted by Black and Cox (1976), (10.10) and (10.11) assume that the boundaries are exogenously
determined. If the boundaries are determined endogenously as part of an optimization problem,
one needs to be more careful about valuing the boundaries. Later on we review some examples
where the liquidation boundary is determined by the incentives of the equity holders. For now we
just assume that the boundary is given in the contracts as in the following example.
10.4.1 Valuing debt with a safety covenant
Suppose that the rm has debt and equity as in the Merton (1974) model. The debt contract has
a safety covenant guaranteeing that if the earnings of the rm decrease to a specied level prior to
maturity, then the debt holders have the right to costlessly take over the rm. The principal of the
debt is P, and for this example suppose that the covenant denes the default level to be
x
t
= P(r )e
r(Tt)
, [0, 1]. (10.12)
The liquidation boundary induced by the covenant, i.e. equation (10.12), states that at maturity
the rm is declared bankrupt if the rm value,
x
r
, is below the -fraction of the principal of the
9
Often the integrals are written in terms of expectations, e.g. X
1
(x
t
, t) = E
Q
t
[e
r(Tt)
x
1
(x
T
)]. Of course, one
must use the conditional distribution to calculate the expectation.
10.4 The probabilistic approach to pricing claims 79
debt, P. Before maturity the covenant guarantees that the debt holders receive the -fraction of
the discounted value of the principal in case of default.
10
As long as the earnings have not reached the lower boundary, the equity holders receive the
earnings. The debt issued is a zero-coupon bond so the debt holders receive no payout prior to
maturity unless the rm defaults. At maturity the equity holders must pay the principal to the
debt holders otherwise the rm will costlessly be handed over to the debt holders.
If one uses the pde-approach, cf. (10.7), to derive the prices for the debt and the equity, one
gets the debt-pde
1
2

2
x
2
t
D
xx
(x
t
, t) +x
t
D
x
(x
t
, t) +D
t
(x
t
, t) rD(x
t
, t) = 0 (10.13)
with the associated boundary conditions
D(x
T
, T) = min
_
x
T
r
, P
_
, (10.14)
D(x
t
, t) =
x
t
r
= Pe
r(Tt)
,
and the equity-pde is
1
2

2
x
2
t
E
xx
(x
t
, t) +x
t
E
x
(x
t
, t) +E
t
(x
t
, t) rE(x
t
, t) +x
t
= 0 (10.15)
with the boundary conditions
E(x
T
, T) =
x
T
r
min
_
x
T
r
, P
_
= max
x
T
r
P, 0
E(x
t
, t) = 0, t T.
The two pdes, (10.13) and (10.15), can be solved numerically quite easily but to derive explicit
solutions seems to be a dicult task. If the necessary conditional distributions can be written
explicitly, the probabilistic approach often reveals a way to get exact formulae for the debt and
the equity prices. Black and Cox (1976) are able to derive expressions for ,

, and due to the
geometric Brownian motion setup and hence they can obtain the values of equity and debt.
11
This example extends the Merton (1974) model with a safety covenant. Therefore many of
main ndings in Merton (1974) are also found in this example. The debt value is increasing in the
earnings and decreasing in the time until maturity, the riskiness of the earnings and the interest
10
Of course, at maturity the rm is also default if the principal cannot be paid.
11
The enthusiastic student is encouraged to try to derive the values for debt and equity.
80 Introduction to structural models
rate. Furthermore, in this model the debt is increasing in the recovery rate stated in the covenant
(for suciently high ). If the earnings of the rm decrease the debt holders will like to force the
rm into bankruptcy as quickly as possible. This conclusion hinges on the fact that at default, the
control right of the rm is handed over to the debt holders without any cost. Suppose namely that
the debt holders would have to pay large costs to take over the rm. In the example considered, if
the debt holders only receive

2
Pe
r(Tt)
at default and the same value is lost to lawyers, say, then
there is scope for a contract renegotiation between the equity and debt holders.
This example from Black and Cox (1976) shows that if bankruptcy is costless, then a safety
covenant may increase the value of the debt. An important matter which is not considered in the
Black and Cox (1976) model is the question of optimizing the debt and equity contracts ex ante,
i.e. what determines the optimal time until maturity, the optimal coupon, and quite importantly,
the optimal recovery rate at default. If one wants to discuss these issues, then the model needs
to have some advantages and disadvantages of debt, otherwise the capital structure is irrelevant.
10.4.2 Subordinated debt
The previous model can be extended to price subordinated bonds if the structure is suciently
simple. Black and Cox (1976) consider an example where the rm has issued two types of zero
coupon bonds. Both bonds mature at the same time, T, and both have the same liquidation
covenant as above, namely x
t
= P(r )e
r(Tt)
. The senior bond has principal P, and the
junior bond has principal Q; the payo at maturity is illustrated in Figure 10.3.
5 10 15 20 25 30
5
10
15
20
25
30
E
v
D
P
O
J
PJ
Q
x
r
Figure 10.3: The payo diagram with subordinated debt. The principal of the senior debt is P = 10
and Q = 5 for the junior debt. E is the equity value, D is the senior debt value, and J is the junior
debt value. The rm value is v = E +D +J.
10.5 Debt with coupon payments: A main example 81
From Figure 10.3 it is obvious that the price of the senior debt and equity, respectively, can be
derived as in the above case with a single debt issue, when the total principal is P +Q. Figure 10.3
further illustrates how the junior bond can be written in terms of the bond prices already calculated.
The junior bond only receives some value if the senior bond is paid in full, i.e. if
x
T
r
> P. Therefore
the junior bond can be seen as the value of a bond with principal P + Q, evaluated as if the rm
only issued this bond, minus the value of the senior bond.
To write up the price of the junior debt, dene D(x
t
, t; P; x
t
) to be the price of a bond calculated
as before with principal P and lower boundary x. Then the price of the junior bond, J, is given as
J(x
t
, t) =
_

_
D(x
t
, t; P +Q; x
t
) D(x
t
, t; P; x
t
), < 1
D(x
t
, t; P +Q; x
t
) e
r(Tt)
P, 1
P+Q
P
e
r(Tt)
Q, >
P+Q
P
.
(10.16)
From (10.16) the value of the junior bond depends on the recovery ratio . When < 1, only
the senior debt holders benet from the covenant and the covenant may actually hurt the junior
debt holders. For a higher recovery ratio the junior debt holders will be protected to a larger extent
by the covenant and when >
P+Q
P
, the covenant gives full protection to the senior and the junior
debt holders.
Furthermore, one notes from Figure 10.3 that the senior debt holders are better o being senior
debt holders than they are by holding the fraction
P
P+Q
of a single debt issue with principal P +Q.
The senior debt has the nal payo structure O, P, D in Figure 10.3 whereas the debt with
principal P +Q has payo O, PJ, D, which is below the senior payo as long as
x
T
r
< P +Q.
An interesting result with subordinated debt with no covenants is that the junior debt holders
will, for a low x, prefer a more risky business than the senior debt holders because the junior debt
is now essentially as equity. Thus, more volatile earnings will only increase the probability that
the junior debt holders get some value at maturity. For high earnings, the junior debt holders will
prefer low volatility.
10.5 Debt with coupon payments: A main example
We now turn to another example in Black and Cox (1976). This example has been used extensively
as the background model in the newer literature on capital structure theory using the structural
modeling framework, e.g. in Leland (1994b) and Mella-Barral (1999).
82 Introduction to structural models
Consider the model in Section 10.4.1, but assume that there is no safety covenant and that the
debt is a perpetual bond which promises the holder a constant coupon rate, C > 0.
12
Thus, there
is no direct time dependence and the values of debt and equity only depends on x. As in Section
10.4.1 one can use the probabilistic approach to derive the price of the debt. However, when the
debt is perpetual, the pde (10.7) becomes an ordinary dierential equation (ode). Hence, it is easier
to use the pde (or ode) approach. We assume that the dividend rate is ane in the earnings of the
rm, i.e. h(x
t
, t) = Ax
t
+B. Then the ode is
13
1
2

2
x
2
F
xx
(x) +xF
x
(x) rF(x) +Ax +B = 0. (10.17)
The ode (10.17) has the general solution
F(x) =
A
r
x +
B
r
+k
1
x

1
+k
2
x

2
, (10.18)
where

1
=
(
1
2

2
) +
_
(
1
2

2
)
2
+ 2
2
r

2
> 1, (10.19)

2
=
(
1
2

2
)
_
(
1
2

2
)
2
+ 2
2
r

2
< 0. (10.20)
The constants k
1
and k
2
will be determined by the boundary conditions associated with a particular
claim.
14
The debt holders receive the dividend rate C, thus A
debt
= 0 and B
debt
= C. Let the constants
in the debt ode be b
1
and b
2
, then
D(x) =
C
r
+b
1
x

1
+b
2
x

2
.
If the debt was risk free, its value would be
_

0
e
rt
Cdt =
C
r
. To exclude bubbles in the economy,
the debt must approach the value of risk free debt, when the earnings grows to innity. Since

1
> 1, it must be the case that b
1
= 0, hence
D(x) =
C
r
+b
2
x

2
.
12
If c
t
is the (possibly time dependent) coupon rate then the coupon paid in one year is

1
0
c
t
dt. With a constant
coupon rate, c
t
= C, the coupon payment per year is

1
0
Cdt = C.
13
We now write F(x) instead of F(x
t
, t) because the time per se is not important.
14

1
and
2
are the roots of the characteristic polynomial
1
2

2
x(x 1) + x r = 0. See also Dixit and Pindyck
(1994).
10.6 Optimal stopping. The high contact or smooth pasting condition 83
When the earnings are not sucient to cover the coupon payments, i.e. when x < C, the equity
holders must pay money to the rm to avoid a default. For a suciently low level of the earnings
of the rm, the equity holders will eventually not nd it optimal to pay in money to service the
debt.
15
Therefore, the equity holders will voluntarily default for a suciently low level of the
earnings, denoted x. Assume that x is a constant larger than zero. At default the value of equity
is 0 and the debt holders costlessly take over the rm, which then has value
x
r
. This implies
D(x) =
C
r
+b
2
x

2

=
x
r
,
so
b
2
=
_
x
r

C
r
_
x

2
,
hence
D(x) =
C
r
+
_
x
r

C
r
__
x
x
_

2
, x (x, ). (10.21)
The equity value is equal to the rm value minus debt value. For earnings equal to x, the rm
value is
x
r
. As the earnings increase, the debt value converges to
C
r
. Thus,
d
dx
E(x)
d
dx
_
x
r

C
r
_
=
1
r
, x .
As noted above E(x) = 0. With calculations similar to those done for the debt one obtains
E(x) =
_
x
r

C
r
_

_
x
r

C
r
__
x
x
_

2
, x (x, ). (10.22)
Note from (10.21) and (10.22) that the sum of the debt and the equity, i.e. the value of the rm,
is
x
r
. This has to be the case when there are no costs nor benets associated with the debt.
10.6 Optimal stopping. The high contact or smooth pasting con-
dition
The expressions in (10.21) and (10.22) are not determined before one has derived the level of the
earnings at which the equity holders will cease to pay the coupons, i.e. x must be determined from
15
Since equity has limited liability, the equity holders can always just abandon their equity claim. In other words,
equity cannot have a negative value.
84 Introduction to structural models
the known parameters of the model. The stopping level of the earnings should be chosen so as to
maximize the value of the equity ex post. In other words, after the debt has been issued, the equity
holders will only take their own incentives into account, when they decide to cease the coupon
payments.
16
Black and Cox (1976) determine x as the level which minimizes the debt value. As
just mentioned this model has the sum of equity and debt being equal to the value of receiving all
of the earnings in the future. Therefore either approach will work. Here, we choose to maximize
the equity value. To show the dependence of the stopping level, x, it is included as an argument
when appropriate. From (10.22) one gets the rst order condition for maximizing the equity value
E(x, x)
x

x=x
= 0,
i.e.

x
__
x
r

C
r
_

_
x
r

C
r
__
x
x
_

2
_

x=x
= 0,
so
1
r
_
x
x
_

2
(
2
)
_
x
r

C
r
__
x
x
_

2
x
1
= 0,
whence
x
r
=

2

2
1
C
r
. (10.23)
Hence, the trigger value of the earnings, at which the equity holders decide to stop paying the
coupon, is determined solely as a function of the given parameters.
In this example one is able to derive the lower boundary by using the standard condition of
nding stationary points, i.e. to take the derivative w.r.t. to the decision parameter and set the
result equal to zero. In other cases one applies the smooth pasting or the high contact condition.
The rst order condition used to derive (10.23) and the high contact condition are closely related.
The following line of reasoning gives the intuition for the relation between the two methods. Let
f(x, x) be a dierentiable function and assume that there exists a dierentiable function g such
that for x = x one has f(x, x) = g(x) (= g(x)). Assume further, that for any x there exists a x

that maximizes f in x. In particular for x = x, x satises the rst order condition

f(x, x)
x

x=x=x

= 0. (10.24)
16
Alternatively, one could consider the ex ante decided stopping level of x. This criteria is relevant if the stopping
level can be contracted upon.
10.6 Optimal stopping. The high contact or smooth pasting condition 85
Now dierentiate f w.r.t. x along the boundary x = x. This gives
f(x, x)
x
dx
dx

x=x
+
f(x, x)
x

x=x
=
dg(x)
dx

x=x
i.e.
f(x, x)
x

x=x
+
f(x, x)
x

x=x
=
dg(x)
dx

x=x
. (10.25)
For the x = x

that satises (10.24) for x = x, equation (10.25) becomes

f(x, x)
x

x=x=x

+
f(x, x)
x

x=x=x

=
dg(x)
dx

x=x=x

thus from (10.24)

f(x, x)
x

x=x=x

=
dg(x)
dx

x=x=x

. (10.26)
Hence, from the rst order condition (10.24) one gets the smooth pasting condition in (10.26).
This line of reasoning shows that there is a link between the usual rst order condition and
the high-contact condition often used in real option analysis. The type of argument above can be
found in (Merton 1973, footnote 60), (Merton 1990, ch. 8, footnote 47), or Brekke and ksendal
(1991). The latter reference contains a more detailed examination of the high contact condition.
In the above example from Black and Cox (1976) the smooth pasting method works in the
following way. The value to the equity holders at default is equal to g(x)

x=x
= 0. Thus,
E(x, x)

x=x
= g(x)

x=x
= 0, (10.27)
i.e. from (10.22)
__
x
r

C
r
_

_
x
r

C
r
__
x
x
_

2
_

x=x
= 0.
Then one takes the (partial) derivative w.r.t. the state variable and evaluates at the boundary, i.e.
E(x, x)
x

x=x
=
dg(x)
dx
,
thus

x
__
x
r

C
r
_

_
x
r

C
r
__
x
x
_

2
_

x=x
= 0,
86 Introduction to structural models
hence
1
r

2
_
x
r

C
r
__
x
x
_

2
x
1

x=x
= 0,
whence
x
r
=

2

2
1
C
r
. (10.28)
and one notes that the lower boundary derived from (10.23) is the same as the one from (10.28).
10.6.1 The case of singular controls
Incomplete....
The above discussion concerned the case where the decision maker like the equity holders
makes an irreversible binary decision. As we shall see, such a setting has wide applications in
nance. In fact, most of the following models will apply to this approach. To be a bit more precise,
the decision need not be irreversible. The key is, that there is some sort of true costs associated with
the decision or, more to the point, that the problem does not imply a singular control. For instance,
a rm may want to shut down production when earnings are low (lower than the production costs)
and resume production when earnings increase suciently. If there are no costs associated with
changing between the production state and the non-production state, we have a case of a singular
control. However, if changing the current state to another state is costly, the decision to change
will a binary decision problem of the kind we considered in the previous section.
17
A more detailed explanation with a focus on nance applications can be found in ksendal and
Sulem (2005).
10.6.2 Generalized Arrow-Debreu prices and event probabilities
In order to interpret the debt and equity values (10.21) and (10.22), and similar expressions that
appear later, it is useful to introduce the following claims. For later reference, suppose that the
continuation region is given by (x, x) and that the initial level of EBIT is x
0
inside the continuation
region. Here, x and x are assumed to be constants. Let p
d
(x; x
0
) be the price of the claim that
pays one unit of account when x is hit, but only if x is not hit before. In other words, the claim is
a (generalized) Arrow-Debreu price of the lower boundary. From the solution (10.18) one has that
p
d
(x; x
0
) = k
1
x

1
+k
2
x

2
,
17
See Dixit and Pindyck (1994, chapter 6 and 7)
10.6 Optimal stopping. The high contact or smooth pasting condition 87
with the boundary conditions
p
d
(x; x
0
) = 1,
p
d
( x; x
0
) = 0.
This system is easily solved and one nds
k
1
=
x

2
x

1
x

2
x

2
x

1
,
k
2
=
x

1
x

1
x

2
x

2
x

1
,
thus
p
d
(x; x
0
) =
x

2
x

1
x

2
x

2
x

1
x

1
x

1
x

2
x

2
x

1
x

2
. (10.29)
Similarly, let p
u
(x; x
0
) be the price of the claim that pays one unit of account when x is hit,
but only if x is not hit before. Then
p
u
(x; x
0
) =
x

2
x

1
x

2
x

2
x

1
x

1
+
x

1
x

1
x

2
x

2
x

1
x

2
. (10.30)
Let us now focus on the cases where there is only an upper or a lower bound. For concreteness,
let us focus on the example from the Black and Cox (1976) model with a lower (default) boundary.
In this case we have x = so one has (by taking limits)
p
d
(x; x
0
) =
_
x

x
_

2
(10.31)
p
u
(x; x
0
) = 0. (10.32)
Note that (10.31) enters into the debt and equity expressions (10.21) and (10.22) which makes
interpretations straightforward. For example, the debt value can be interpreted as the present
value of receiving coupons forever adjusted for the value change at default. In the event of default,
the debt holders receive the right to all future earnings but, on the other hand, they do not receive
coupon payments any more. To adjust this to present value terms we simply apply the event price
in (10.31).
Furthermore, in the present case standard results on rst passage times for a standard Brownian
motion allows us to calculate the probability density function (p.d.f.) for hitting the default barrier.
We denote the p.d.f. for the stopping time for the level x with the process X as f
x
(and similarly
f
y
for a process Y reaching y). From Karatzas and Shreve (1991) we state the following result.
88 Introduction to structural models
Lemma 10.6.1 (First passage time for Brownian motion with drift) Let X be a Brown-
ian motion with drift a, i.e.
dX
t
= adt +dW
t
, (10.33)
where W is a standard Brownian motion (Wiener process). Let
x
be the stopping time for X
reaching the level x ,= 0 the rst time. Then the p.d.f. for the stopping time, f
x
, is
f
x
(t) = Q(
x
dt)
1
dt
=
[x[

2t
3
exp
_

(x at)
2
2t
_
, t > 0.
In order to use this in our formulation, we remark that we are interested in the density for the
default stopping time
x
, i.e. f
x
. Let b = log
_
x
0
x
_
and note that

x
inft > 0 : x
t
= x
= inft > 0 :
x
t
x
= 1
= inf
_
t > 0 : log
_
x
t
x
_
b
. .
Y
t
= b
_
,
thus we seek the p.d.f. f
y
for Y = b y the rst time. We can discover the property of Y from
x due to the fact that
x
t
= x
0
exp(

2
2
)t +W
t
,
and so
x
t
x
=
x
0
x
exp(

2
2
)t +W
t
,
whence it follows that
Y
t
log
_
x
t
x
0
_
b = (

2
2
)t +W
t
,
which we immediately recognize as an arithmetic Brownian motion (a Brownian motion with pos-
sible non-zero drift and non-unit variance per unit of time) with drift rate a

2
2
and volatility
> 0. Dening X Y/ we obtain
inf
_
t > 0 : Y
t
= y
_
= inf
_
t > 0 :
Y
t

=
y

_
= inf
_
t > 0 : X
t
=
y

_
inf
_
t > 0 : X
t
= x
_
,
10.6 Optimal stopping. The high contact or smooth pasting condition 89
and since X is a Brownian motion with drift as in (10.33), where the appropriate drift rate is a/,
we obtain
f
x
(t) =
[x[

2t
3
exp
_

(x
a

t)
2
2t
_
(10.34)
thus
f
y
(t) =
[
y

2t
3
exp
_

(
y

t)
2
2t
_
(10.35)
hence
f
x
(t) =
[
b

2t
3
exp
_

(
b

t)
2
2t
_
=
[b[

2
2
t
3
exp
_

(b at)
2
2
2
t
_
(10.36)
Recall that b = log
_
x
0
x
_
and, thus, b > 0. Hence, for our previous example we obtain
f
x
(t) =
log
_
x
0
x
_

2
2
t
3
exp
_

_
log
_
x
0
x
_
(

2
2
)t

2
2
2
t
_
(10.37)
If we want to calculate the default probability, all we have to do is to integrate over the default
stopping time distribution. In fact, it turns out that there exists a closed-form expression which we
state below. For further details and proofs see e.g. Leland and Toft (1996), Harrison (1985), and
Musiela and Rutkowski (2005). With a and b as above one can show that the default probability
up to time t is
F(t) =
_
t
0
f
x
(s)ds
= N
_
h
1
(t)
_
+
_
x
x
_
2a/
2
N
_
h
2
(t)
_
, (10.38)
where h
1
(t) =
bat

t
and h
2
(t) =
b+at

t
.
Chapter 11
Modeling the tax advantage to debt:
trade-o theory
A rm can have debt for dierent purposes; for instance if there is asymmetric information in the
model, the debt can be used to discipline management. As a disciplinary device the debt can be
issued due to the initial rm owners command, or the manager can alternatively choose to issue
debt to signal his type of commitment to only invest in projects with a positive net present value,
cf. Stulz (1990a) and Zwiebel (1996).
Another reason for corporate debt is if the tax rate on interest payments is lower than the total
tax rate paid by the equity holders. Furthermore, the interest expenses are often tax deductible
when the rm pays corporate tax. In this case one says that there is a tax advantage to debt over
equity. For the structural models the tax advantage is the typical reason for a rm to issue debt.
In this chapter we discuss structural models which are concerned with valuing the tax advantage
to debt.
If there is a tax advantage to debt, there is a dierence between the value of the assets of the
rm and the value of the rm.
1
The default barrier derived from minimizing the debt value will
now be dierent from the barrier derived from equity maximization. Since the equity holders will
take their own incentives into account such that the default decision is incentive compatible (from
the equity holders perspective) the default barrier must now be derived from equity maximization.
The Black and Cox (1976) model in Section 10.2.2 extends the Merton model such that default
can happen before the maturity of the debt of the rm. In this model it is not possible to discuss
1
More to the point, the value of the rm will not equal
x
r
, cf. (10.2).
91
92 Modeling the tax advantage to debt: trade-o theory
optimal capital structure issues because there is no reason for the rm to issue debt. A few years
later Brennan and Schwartz (1978) took a rst step in modeling a tax advantage of debt. An often
cited paper which deals with the tax advantage to debt matter is Leland (1994b), which will be
discussed in some detail in Section 11.1. When the Leland model has been analyzed, it will be
evident that the next step will be to include some degree of decision dynamics in the modeling, i.e.
there is a need for a model which can handle that a rm restructures its capital more than once. In
the Sections 11.211.4 we continue from the Leland (1994b) model with the models by Goldstein,
Ju, and Leland (2001) and the model presented in Christensen, Flor, Lando, and Miltersen (2002a)
in order to treat the dynamic restructuring issue. The model by Fischer, Heinkel, and Zechner
(1989a) will be briey mentioned in Section 11.5.
11.1 A simple optimal capital structure model
In this section we look at the model by Leland (1994b). The model has the basic assumptions 16
and assumption 9 as in Section 10.2.1. The important exception is that the model introduces costs
and benets of debt. Again, we use the earnings of the rm as the state variable and not the value
of the unlevered rm as Leland (1994b) does.
2
11.1.1 The costs and benets of debt
The debt becomes benecial because there is a tax advantage to debt. For concreteness consider
the model in Section 10.5. Thus, the debt is perpetual and the rm pays the coupon rate C.
Suppose that the coupons received by the debt holders are taxed at the rate
i
and that the
coupons are tax deductible for the rm. Of the earnings before taxes, x C, the rm must pay
corporate tax at the rate
c
. The remaining value is immediately paid out to the equity holders as
dividends. The equity holders must pay the dividend tax rate
d
. This implies that altogether, the
earnings before taxes are taxed by the eective tax rate
e
=
c
+
d

c

d
from the point of view
of the equity holders. Table 11.1 illustrates how the tax schedule works in this model. Henceforth
it is assumed that
e
>
i
, i.e. there is a tax advantage to debt.
On the other hand, there are two sources of costs of debt. Assume that if the rm defaults on
the debt, then the value to split between the claimants is reduced by a factor . One can think of
2
The Leland (1994b) model written in earnings terms is similar to the static model part of Goldstein, Ju, and
Leland (2001). There is, however, a dierence in the interpretation of the drift rate. We return to this in Section
11.2.
11.1 A simple optimal capital structure model 93
EBIT x Debt holders The State
- Debt interest -C (1
i
)C
i
C
= Earnings before taxes x C
- Corporate tax
c
(x C) Equity holders
c
(x C)
= Net income (1
c
)(x C) (1
e
)(x C) (
e

c
)(x C)

e
x (
e

i
)C
Table 11.1: The distribution of the earnings of the rm. Debt is paid the coupon C which is taxed
at the rate
i
. The rm must pay the tax rate
c
on the earnings before taxes. The remaining value
is paid to equity as dividends which are taxed at the rate
d
. The last column shows the taxes to
the State. The last row shows the total taxes.
the parameter as an approximation of indirect and direct bankruptcy costs. Furthermore, when
the debt is issued, the issuing party must pay a fraction, denoted k, of the initial value of the debt.
In other words, k is a debt issuing cost parameter.
11.1.2 The interpretation of the state variable
As in the previous sections one takes as given a rm which receives earnings. Since there are taxes
involved, the earnings are before interest and taxes, henceforth denoted EBIT or x. The earnings
are assumed to develop independently of the capital structure of the rm. Under the risk neutral
measure the earnings evolve, as in (10.1), according to
dx
t
= x
t
dt +x
t
dW
t
. (11.1)
Dene the after tax risk free interest rate, r

= (1
i
) r, where r is the risk free interest rate before
interest tax.
3
We assume in (11.1) that the drift of the earnings is a constant < r. Then, in
relation to (11.1), one could dene a capitalized value of x as
V (x
t
) = E
Q
t
_ _

t
e
r(st)
x
s
ds
_
=
x
t
r
, (11.2)
and hence
dV
t
= V
t
dt +V
t
dW
t
, (11.3)
3
In general, the notation is such that r is the appropriate risk free discount rate.
94 Modeling the tax advantage to debt: trade-o theory
cf. (10.2) and (10.3). One could also dene the articial unlevered rm value as
V
U
(x
t
) = (1
e
)
x
t
r
= (1
e
)V (x
t
). (11.4)
Obviously, V
U
is also a geometric Brownian motion with drift < r and volatility .
In Leland (1994b) the state variable is assumed to be the value of the assets of the rm or the
value of the unlevered rm. Therefore the V in Leland (1994b) has the same interpretation as the
V
U
dened above.
4
However, Leland assumes that one can treat the value of the unlevered assets,
V
U
, as the price of a traded asset, and this implies that the drift rate of the value of the unlevered
rm is equal to the after tax risk free interest rate, r. But to assume that V represents the price of
a traded asset yields an arbitrage in the model.
5
Assume namely as in Leland (1994b) that there
are no debt issuance costs and consider the following equation
rm value = asset value + value of tax benets - bankruptcy costs, (11.5)
where all values are stated as market values.
6
Denote the rm value as v(x) = E(x) + D(x). If
v as well as V
U
is supposed to represent prices of traded assets, then it must be the case that
v(x) = (1
e
)
x
r
= V
U
(x). Otherwise all investors will buy the assets of the rm for V
U
, lever
the rm, and sell the rm for v > V
U
and vice versa if v < V
U
. In other words, the tax benets
must exactly outweigh the bankruptcy costs, i.e. there is no reason to have debt in this set-up. On
the other hand, using the EBIT as the state variable yields an extra parameter, in (11.1), which
must be exogenously specied.
11.1.3 Valuation of debt and equity
The debt is issued as a perpetuity promising a constant coupon rate of C 0. With the tax
schedule stated above the debt holders and the equity holders receive payments according to Table
11.1 as long as the rm is servicing the debt.
4
Because of the simpler tax structure in the Leland (1994b) model, there is no distinction between V
U
and V in
the article.
5
It is seen in (Leland 1994b, footnote 11) that Leland is aware of the delicate question of whether V represents
the price of a traded asset or not. However, the fact that this implies an arbitrage was rst brought to my attention
by Michael Brennan. Kane, Marcus, and McDonald (1985) have a model where they try to deal with this problem
by choosing a special interpretation of the state variable. We return to this in Section 11.5.
6
This is equation (12) from Leland (1994b).
11.1 A simple optimal capital structure model 95
Since the debt is perpetual, the pde in (10.7) becomes an ode like (10.17) and one has the
solution from (10.18) for the debt and equity as respectively
7
D(x) =
(1
i
)C
r
+b
1
x

1
+b
2
x

2
, (11.6)
and
E(x) = (1
e
)
_
x
r

C
r
_
+a
1
x

1
+a
2
x

2
. (11.7)
At any point in time the equity holders may abandon the debt contract because they can simply
stop paying the coupon and equity has limited liability. It is assumed that this will trigger default
immediately, i.e. the rm is handed over to the debt holders. Denote the default level of the state
variable as x. At default the value of equity is zero and the debt holders take over the rm for a
proportional cost of . Following a default the rm cannot issue new debt.
8
If the earnings grow to innity, the promised coupon will always be paid, i.e. the debt value
will approach its default free value and equity value cannot increase more than the increase in the
earnings. Thus, the necessary boundary conditions for the debt are
D(x) = (1 )(1
e
)
x
r
,
D(x)
(1
i
)C
r
, x , (11.8)
and for equity
E(x) = 0, (11.9)
dE(x)
dx
(1
e
)
_
1
r
_
, x . (11.10)
As in Section 10.5, the constants b
1
and a
1
in (11.6) and (11.7) are zero to satisfy (11.8) and (11.10),
i.e. to avoid speculative bubbles in the economy. Using the conditions at the default boundary, one
nds
D(x) =
(1
i
)C
r
+
_
(1 )(1
e
)
x
r
(1
i
)
C
r
__
x

x
_

2
, (11.11)
E(x) = (1
e
)
_
x
r

C
r
_
(1
e
)
_
x
r

C
r
__
x
x
_

2
, (11.12)
7
In (10.18)(10.20) the after tax interest rate must be used.
8
If one did a more careful modeling, one would let the principal of the debt be P =
C
r
, which is the perpetual
value of the debt, and let E(x) = max{0, (1 )(1
e
)
x
r
P}. If it is costly to issue new equity in order to pay
coupons, as in Uhrig-Homburg (2005), one could imagine that the equity holders would default when the liquidation
value of the rm was larger than the principal of the debt.
96 Modeling the tax advantage to debt: trade-o theory
and hence the rm value for x > x is
v(x) = E(x) +D(x)
= (1
e
)
x
r
+ (
e

i
)
C
r

_
(1
e
)
x
r
+ (
e

i
)
C
r
__
x
x
_

2
. (11.13)
To make the derivation of the values of debt and equity complete it is also necessary that the
default level, x, is determined as a function of the parameters. To do this one can apply the high
contact condition as outlined above in Section 10.6. Thus, one takes the derivative w.r.t. x in
(11.12) and puts it equal to 0 because of (11.9). Hence
E(x)
x

x=x
=
1
e
r

2
(1
e
)
_
x
r

C
r
__
x
x
_

2
1
x
1
= 0,
i.e.
x
r
=

2

2
1
C
r
, (11.14)
and we return to this expression in a moment.
Instead of deriving the equity and debt values directly, one could use the procedure from Leland
(1994b). Firstly, debt value is derived as above, directly from solving the ode with boundary
conditions. Secondly, one must evaluate the rm. This is done by using equation (11.5) recognizing
that the rm value is equal to the articial unlevered rm value added to tax benets (TB) minus
bankruptcy costs (BC), i.e. v(x) = V
U
(x) + TB(x) BC(x). The tax benets and bankruptcy
costs can be evaluated by pricing a claim that would result in the respective identical payo. The
respective boundary conditions and values are
TB(x)
_

_
0, x x
(
e

i
)
C
r
x
hence
TB(x) = (
e

i
)
C
r
_
1
_
x
x
_

2
_
, (11.15)
and for the bankruptcy costs
BC(x)
_

_
(1
e
)
x
r
, x x
0 x
11.1 A simple optimal capital structure model 97
hence
BC(x) = (1
e
)
x
r
_
x
x
_

2
. (11.16)
Plugging (11.15) and (11.16) into (11.5), one obtains (11.13). Finally, writing the equity value as
E(x) = v(x) D(x), one gets (11.12).
Therefore one can interpret the tax benets in (11.15) as the value of receiving the tax savings
rate (
e

i
)C forever minus the value of losing the tax benets when the lower boundary is hit.
Similarly the bankruptcy costs in (11.16) are interpreted as the value today of receiving the fraction
of the unlevered rm when default occurs. The rm value can then be interpreted as follows. The
rst two terms in (11.13) are the value today of receiving the earnings and the tax shield forever.
The last term expresses the loss at default. The rst term in the square brackets is the loss because
of costly bankruptcy, and the second term is the lost value of exploiting the tax shield following
a default. Furthermore, from (11.13) it is particularly easy to see the role of a higher tax rate on
the dividends to the equity holders, compared to the tax rate on the interest payments to the debt
holders. The equity and debt values can be given interpretations in the same vein.
11.1.4 The optimal bankruptcy level
From the analysis above one has the optimal level of the earnings of the rm for which the equity
holders will declare the rm bankrupt according to their incentives in two models, namely the
Black and Cox (1976) and the Leland (1994b) models. If one compares the bankruptcy boundary
obtained from the two models, the two boundaries (10.28) and (11.14) are apparently the same.
However, if one writes the two boundaries in terms of the value of an unlevered rm, one gets
Black & Cox V
U
=

2

2
1
C
r
,
Leland V
U
=

2

2
1
(1
e
)C
(1
i
)r
.
Thus, the boundaries dier with the tax structure as seen from an (articial) unlevered rm value
perspective. Furthermore, the dierence lies in the coupon rate. In Black and Cox (1976) the capital
structure is irrelevant, i.e. there is no ex ante reason for issuing debt or choosing a particular coupon
rate. In Leland (1994b) the tax structure favors a debt issuance. Therefore one may ask for the
optimal coupon rate in the Leland (1994b) model, denoted C

.
98 Modeling the tax advantage to debt: trade-o theory
11.1.5 A comment on the comparative statics
As a rule we only consider the comparative statics of the models in these notes if they are particularly
interesting. The important dierence between the Leland (1994b) model and the model analyzed
above is that we have not assumed that the unlevered rm is a traded asset and the more complex
tax structure. Therefore the eect of a change in the eective tax rate will be dierent in the two
models.
As in Leland (1994b) let us for a moment x the coupon rate C and consider changes in the
eective tax rate. In the Leland model the state variable is the value of the unlevered assets of
the rm and hence this process is assumed to be independent of a change in the tax structure.
In contrast, if the state variable is the EBIT of the rm, then the (articial) market value of the
unlevered assets of the rm is V
U
= (1
e
)
x
r
, which is obviously aected by the tax structure.
This dierence turns out to make the comparative statics dierent for the value of the rm in the
two frameworks. A higher corporate tax rate will increase the value of the rm in Leland (1994b)
model, but decrease the rm value in the EBIT model.
9
11.1.6 Optimal coupon
Because of the assumed tax advantage to debt, there is scope for asking for the optimal capital
structure in the Leland (1994b) model. When the initial rm owners issue new equity and debt,
they want to maximize the value of the rm minus the cost of issuing the debt. The debt issuance
cost is proportional by a factor k to the initial debt value. Thus, the initial rm owners problem is
max
C0
_
v(x
0
) kD(x
0
)
_
, (11.17)
where the rm value is given in (11.13) and the debt value is given in (11.11). By taking the
derivatives of (11.17) w.r.t. C, one obtains the optimal coupon rate as
C

=

2
1

2
r
_
(1
e
) (1 k)
_
(1 )(1
e
)
2
+ (1
i
)(1
2
)
_
(1
e
) (1 k)(1
i
)
_ 1

2 x
0
r

= c

x
0
. (11.18)
From (11.18) one notes that the optimal coupon rate is a (positive) linear function of the level
of the state variable, when the debt is issued. Moreover, one can calculate that the optimal coupon
9
As earlier noted, this is also seen in the Goldstein, Ju, and Leland (2001) treatment of Leland (1994b).
11.1 A simple optimal capital structure model 99
rate will decrease if it is more costly to issue debt and that the coupon rate will increase if the
eective tax rate on equity increases.
It is intuitive that the optimal coupon rate increases with a larger tax advantage to debt obtained
by increasing the eective tax rate. However, an increase in the eective tax rate also reduces the
equity value for a given coupon rate. Hence, one cannot rule out the possibility that the value to
the initial rm owners in (11.17) can be increasing in the eective tax rate. This result is counter
intuitive and it has to do with the possibility to deduct interest payments even if the net income
of the rm is negative. We return to this point in Section 11.2.
11.1.7 Introducing a net-worth covenant
Rather than allowing the equity holders to declare the rm bankrupt according to their incentives,
Leland considers the possibility of writing debt contracts with a positive net-worth covenant. A
net-worth covenant will force the rm into bankruptcy if, say, the (articial) value of the unlevered
assets of the rm reaches a level specied in the debt contract. Leland chooses the level to be equal
to the principal of the debt. The principal of the debt is derived by issuing the debt at par value.
We do not go into a detailed analysis of the capital structure when the rm issues debt with a
protective covenant. The analysis can be done as above, however, the covenant assumption together
with the debt being issued at par imply that
(1
e
)
x
r
(11.4)
= V
U
(x) = P = D(x
0
)
The value of the debt must be derived numerically. The main implication of issuing debt with a
net-worth covenant is that the debt value is, ceteris paribus, increasing in the coupon rate. Recall
that in the Black and Cox (1976) model the covenant level is
x
t
r
= Pe
r(Tt)
. In Lelands
example the debt is perpetual and all of the principal must be received, i.e. = 1. Thus, Lelands
covenant has a clear resemblance to the covenant in Black and Cox (1976).
Other variations in Leland (1994b) include a loss of the possibility to deduct interest payments
when the earnings are low, a matter we return to in Section 11.2. Furthermore, the analysis can
be extended in a simple way such that the absolute priority rule is not respected when the rm
defaults. Leland also briey discusses the asset substitution problem, which is also an issue in the
Leland and Toft (1996) paper, see Section 13.5.
100 Modeling the tax advantage to debt: trade-o theory
11.1.8 Continuous capital restructuring
The above model assumes that debt is issued initially and following a default there is no debt. A
natural extension of the model is to continuously readjust the capital structure to optimize the tax
deductability of coupons. Leland (1994b) considers this possibility and he argues that continuous
readjustments (with no priority requirements) will be blocked by either equity holders or debt
holders even if there are no transactions costs.
The line of reasoning in Leland (1994b) is the following. Suppose rst that the earnings increase
and that the rm then issues a small additional amount of debt to better exploit the tax shield.
This will dilute the value to the initial debt holders because the default level has increased and the
debt holders do not have higher priority than the new debt holders.
In the same vein, the equity holders will reject to make continuous coupon reductions that are
nanced by selling new equity on the market. The reason is that if some of the debt is bought
by the equity holders, then the remaining debt will be safer because the total coupon rate, to be
paid in the future by the rm, has decreased. Since a single outstanding debt contract is safer but
still receives the initial settled coupon rate, the remaining debt becomes more valuable. Since this
added value comes at the expense of the equity holders, the equity holders will not wish to make
small reductions in the coupon rate.
Altogether, the rm will not make small readjustments of the coupon rate. However, if the
capital structure is changed by rst getting rid of all the outstanding debt and then issuing new
debt, then there can be scope for capital structure changes. The possibility to change the capital
structure should be modelled such that the agents ex ante take this possibility into consideration.
Examples of such models are Goldstein, Ju, and Leland (2001) and the Christensen, Flor, Lando,
and Miltersen (2002a) model. Furthermore, if either the equity holder or the debt holders can
make take-it-or-leave-it oers, then Mella-Barral and Perraudin (1997) and Mella-Barral (1999)
show that there is scope for strategic debt servicing. Another possibility to adjust the debt is to let
the rm issue debt with nite maturity and then, when old debt matures, issue new debt. Leland
and Toft (1996) and also Flor and Lester (2002) consider an extension along this line.
11.2 A capital structure model with increasing debt payments
The static debt policy in the Leland (1994b) model does not fully exploit the tax advantage to
debt. One attempt to allow for a better exploitation of the possibility to deduct interest payments
is done by Goldstein, Ju, and Leland (2001).
11.2 A capital structure model with increasing debt payments 101
As mentioned earlier it is problematic to use the value of the traded unlevered assets as in
Leland (1994b). To overcome this problem Goldstein, Ju, and Leland (2001) use the earnings before
interest and taxes as the state variable. The basic framework is as in Section 11.1. Goldstein, Ju,
and Leland (2001) present their model by rst looking at a static model and then they extend the
model to allow for the possibility to issue more debt. By a static model we think of a model which
only gives the rm one chance to issue debt. When the Leland (1994b) model is interpreted in
the EBIT framework with the extended tax structure, it becomes similar to the static model in
Goldstein, Ju, and Leland (2001).
11.2.1 The interpretation of the drift rate of the earnings
In Section 11.1 we assume that the drift rate of the EBIT process is an exogenously given constant
which is independent of the nancial structure of the rm. However, Goldstein, Ju, and Leland
(2001) argue that the expected growth rate of EBIT, , is dependent on the coupon rate.
For an illustration of how can depend on the debt issued, consider the following example from
(Goldstein, Ju, and Leland 2001, p.496). Suppose that the current level of the EBIT is x
0
= 100
and that the price-earnings ratio =
1
r
= 20. Then the total value of all future earnings is
V
0
= 100 20 = 2000. Let the corporate tax rate be
c
= 0.35. Thus the corporate tax payout
rate is
c
(x C) = 0.35(100 C). Suppose that of the remaining value the rm pays out 35 as a
dividend to the equity holders. This implies that the total payout from the rm is
total initial payout rate = interest + corporate tax + equity dividend
= C + 0.35(100 C) + 35 = 70 + 0.65C,
and since the total payout rate must be equal to EBIT
V
0
_
r
_
= x
0
= 70 + 0.65C. (11.19)
It is now assumed that the total payout ratio,
x
V
, remains constant when the earnings of the
rm vary, i.e.
x
t
V
t
=
x
0
V
0
= r =
70 + 0.65C
V
0
,
hence
= r
70 + 0.65C
V
0
. (11.20)
102 Modeling the tax advantage to debt: trade-o theory
Goldstein, Ju, and Leland (2001) argue that this assumption is reasonable in the EBIT framework
because an increase in the EBIT will result in an increase in the corporate tax payments to the
State.
It seems fair to argue like Goldstein, Ju, and Leland (2001) that the corporate tax payments to
the State should increase when the rm receives higher earnings. However, this eect will always
be prevalent in the EBIT framework for a xed coupon rate because the corporate tax received by
the State is the rate
c
(x C). Thus, if x increases, so will the corporate tax. The particular form
in (11.20) is not necessary for this result. Indeed we nd (11.20) to be an unappealing assumption
because it makes the expected growth rate of the EBIT dependent on the debt contract issued by
the rm.
Perhaps one sees the problem more clearly in (11.19). Here, it was rst assumed that x
0
= 100,
but in (11.19) one gets x
0
= 70 + 0.65C. The missing 30 0.65C stems from the assumption that
the rm pays the amount 35 in dividend to the equity holders. However, the dividend to the equity
holders should be all of the remaining cash in the rm. Otherwise there will be capital gains in
the rm (or free cash ow at the managers disposal) and then one needs to extend the model to
handle this matter.
11.2.2 Negative net income
With the current tax structure the rm can deduct all the interest payments even if it has EBIT
less than the interest payments. This seems not to be in line with how interest payments can be
deducted in practice. Therefore Goldstein, Ju, and Leland (2001) extend the tax structure to allow
for a loss of the tax shelter when the net income is negative. Specically, the equity holders receive
the rate
Ax +B =
_

_
(1
e
)(x C) , x C
(1
e
)x (1
e
)C , x < C,
(11.21)
where [0, 1] is an exogenously chosen parameter that accounts for the loss of tax shelter and
tax carry forwards. The expressions for debt and equity in (11.6) and (11.7) then become
D(x) =
(1
i
)C
r
+b
1
x

1
+b
2
x

2
, and (11.22)
E(x) =
_

_
(1
e
)
_
x
r

C
r
_
+a
u
1
x

1
+a
u
2
x

2
, x C
(1
e
)
x
r
(1
e
)
C
r
+a
d
1
x

1
+a
d
2
x

2
, otherwise.
(11.23)
11.2 A capital structure model with increasing debt payments 103
The boundary conditions are as before but one needs to add two more conditions in order to
have a well specied system. The conditions are that the equity value must be continuous and
dierentiable when x = C.
10
11.2.3 Increasing the debt service. Upward debt restructurings
If there is a tax advantage to debt, it seems plausible that after a large increase in the earnings
(compared to the initial level) the equity holders will wish to increase the coupon rate in order
to better exploit the tax shield. One can imagine many dierent ways to model the increase in
the coupon payments, for instance the rm may issue additional debt of dierent priorities or with
dierent coupon rates. Additional debt could also have protective covenants associated or a nite
maturity.
To have a suciently simple model, Goldstein, Ju, and Leland (2001) restrict their attention to
a restructuring of all of the outstanding debt. The way we present the Goldstein, Ju, and Leland
(2001) model in the following is somewhat dierent from the original article.
There can be only one class of debt outstanding at any time and the callable debt is assumed
to be perpetual. Since the EBIT process is a geometric Brownian motion and since all the cost and
benet parameters are scalable in the EBIT, one has that the initial debt and the equity values
will likewise be scalable as long as the boundary conditions have the same scaling feature. This
positive homogeneity of degree one feature is also the case in the Leland (1994b) model which is
easily seen from the values of debt and equity in (11.11) and (11.12) inasmuch as the default level
is proportional in the coupon rate, cf. (11.14). Furthermore, from (11.18) the optimal coupon rate
is seen to be positive homogenous of degree one in x
0
.
Suppose that the initial coupon rate is C
0
= c
0
x
0
. The positive homogeneity of degree one
property in the EBIT implies that if the rm is restructured when x = x

= u x
0
, for a u > 1, then
the new coupon rate will be u C
0
= uc
0
x
0
and the next upward restructuring will take place at
x = u
2
x
0
. If the initial default level is x

= d x
0
, for a 0 < d < 1, then the new default level will
be u d x
0
. Thus, the initial EBIT level plays a key role in the value function so we add it as an
argument in the value functions.
Homogeneity is useful when the boundary conditions are stated. For high earnings, the rm
can call its debt (at par) and issue new debt to better exploit the tax shield. After calling the
10
See Dixit and Pindyck (1994, chapter 6) or Karatzas and Shreve (1988, theorem 4.4.9).
104 Modeling the tax advantage to debt: trade-o theory
debt, the equity holders become rm owners. Hence, the equity holders will want to maximize the
new rm value minus restructuring costs. This problem is equivalent to the problem faced by the
initial rm owners except that all values are now scaled up by the factor u. Suppose the initial rm
owners have chosen the optimal coupon rate C
0
= c
0
x
0
. When the debt is called, i.e. at x = x,
the homogeneity property thus gives the boundary conditions for debt and equity
D(x; x
0
) = D(x
0
; x
0
)
= D(1; 1)x
0
, (11.24)
E(x; x
0
) = max
C
_
E(x; x) + (1 k)D(x; x)
_
D(x
0
; x
0
)
= u
_
E(x
0
; x
0
) + (1 k)D(x
0
; x
0
)

D(x
0
; x
0
)
=
_
E(1; 1) + (1 k)D(1; 1)

(ux
0
) D(1; 1)x
0
, (11.25)
where, as in (11.17), k is the debt issuance cost factor.
Homogeneity is also useful at the lower boundary. At default the debt holders take over the
rm and the equity holders receive nothing. The change of control rights costs the debt holders a
fraction of the default value of the rm. In Goldstein, Ju, and Leland (2001) it is assumed, as
in Leland (1994b), that the rm cannot have debt given that there has been a default. In other
words, following a default it is no longer possible to exploit the tax shield, which in turn implies
that the rm value will be equal to the unlevered rm value dened earlier, i.e. v(x) = (1
e
)
x
r
.
In the model presented in Christensen et al. (2002b), which is considered next, this assumption will
be relaxed. From the above it follows that
D(x; x
0
) = (1 )(1
e
)
x
r
= (1 )(1
e
)
d x
0
r
(11.26)
E(x; x
0
) = 0. (11.27)
Note that all the four boundary conditions (11.24)(11.27) are linear in the initial EBIT level. This
has to be the case if homogeneity is applicable.
11
When the earnings are low, i.e. when x < x < C
0
, one can account for negative earnings as
previously mentioned. In this case the equity value is described by (11.23). For simplicity we
assume for now that there is full interest deductability even with negative net income.
11
To formally demonstrate the validity of homogeneity we would have to show that the dierential equations and
boundary conditions for debt and equity with homogeneity are identical to the ones without assuming homogeneity,
see e.g. Christensen et al. (2002b).
11.2 A capital structure model with increasing debt payments 105
In Goldstein, Ju, and Leland (2001) the default boundary is derived by the smooth pasting
condition in order to take the equity holders incentives into account. This corresponds to condition
(11.29) below. However, it is assumed that the initial rm owners can pre-commit to a call boundary,
i.e. the call boundary is assumed to be written in the debt contract. For a given coupon rate the
model now has ve equations, (11.24)(11.29) and six unknowns, (d, u, a
1
, a
2
, b
1
, b
2
) and it can be
reduced to one equation with (d, u) as the remaining unknowns. The default boundary is determined
by (11.29) and the call boundary, u, enters into the problem of the initial rm owners as a decision
variable together with the coupon rate.
It is somewhat unclear why Goldstein, Ju, and Leland (2001) choose to let u be determined by
the initial rm owners. Rather one would expect that the call as well as the default boundaries are
determined by the incentives of the equity holders.
12
The boundaries would thus be derived by the
smooth pasting conditions related to (11.25) and (11.27), namely
E(x; x
0
)
x

x=x=ux
0
= E(1; 1) + (1 k)D(1; 1), (11.28)
E(x; x
0
)
x

x=x=ux
0
= 0. (11.29)
Then the six conditions (11.24)(11.28) will close the model with upward restructurings and for a
given coupon rate the problem is to solve six equations with six unknowns. Luckily, the problem
can easily be simplied to two equations in the boundary parameters, d and u. Note that in this
case the only initial decision variable is the coupon rate.
11.2.4 Reducing the dimension of the model
Recall the four sources, mentioned in Black and Cox (1976), that contribute to the value of a claim.
With perpetual debt, the value of debt (or equity) will only depend on the received cash ow, until
one of the boundaries is hit, and on the value of the claim at the boundaries. As in (10.29) and
(10.30) let p
d
(x; x
0
) and p
u
(x; x
0
) be the Arrow-Debreu prices at the lower and upper boundary,
respectively, when the continuation region is dx
0
= x < x < x = ux
0
. Then the debt and equity
12
In (Goldstein, Ju, and Leland 2001, footnote 27) the authors recognize that the call boundary could be determined
without assuming that u is written in the debt contract. What we nd really puzzling is that the authors do not
choose to let either both or none of the boundaries be written in the debt contract.
106 Modeling the tax advantage to debt: trade-o theory
values are
13
D(x; x
0
) =
_
1 p
d
(x; x
0
) p
u
(x; x
0
)
_
(1
i
)
C
0
r
(11.30)
+p
d
(x; x
0
)D(dx
0
; x
0
) +p
u
(x; x
0
)D(ux
0
; x
0
)
E(x; x
0
) =
1
e
r
_
x dx
0
p
d
(x; x
0
) ux
0
p
u
(x; x
0
)
_
(11.31)

_
1 p
d
(x; x
0
) p
u
(x; x
0
)
_
(1
e
)
C
0
r
+p
d
(x; x
0
)E(dx
0
; x
0
) +p
u
(x; x
0
)E(ux
0
; x
0
)
The solution of D(x; x
0
) and E(x; x
0
) depends on the boundary values, stated in (11.24)(11.27),
and in particular on d and u.
In fact one can take this line of thinking a little step further in order to see how the homogeneity
property of x
0
shows up in the equity and debt values in this particular setup.
Lemma 11.2.1 Let
0
,
0
,
1
be constants in R and let G(x;
1
,
0
) be the price of the claim that
continuously pays
1
x
t
+
0
forever. Then
G(x;
1
,
0
) =

1
r
x +

0
r
. (11.32)
Furthermore, let F(x; x
0
,
1
,
0
x
0
) be the price of the claim that continuously pays
1
x
t
+
0
x
0
until
x either hits dx
0
or ux
0
. Thus the continuation region is dx
0
= x < x < x = ux
0
. Assume that the
claim has been issued when x = x
0
. Finally, assume that the claim at the boundaries pays o
F(dx
0
; x
0
,
1
,
0
x
0
) =
d
x
0
,
F(ux
0
; x
0
,
1
,
0
x
0
) =
u
x
0
,
where
d
and
u
are constants. Then F can be written as
F(x; x
0
,
1
,
0
x
0
) =
_

1
r
x +

0
r
x
0
_
(11.33)
+x
0
1

__

1
r
u +

0
r

u
_
d

_

1
r
d +

0
r

d
_
u

2
__
x
x
0
_

1
+x
0
1

_

1
r
u +

0
r

u
_
d

1
+
_

1
r
d +

0
r

d
_
u

1
__
x
x
0
_

2
,
where = d

1
u

2
d

2
u

1
. In a more compact form using the G function stated above, F can be
13
Again, to stress the importance of the initial level of EBIT it has been included as an argument.
11.2 A capital structure model with increasing debt payments 107
written as
F(x; x
0
,
1
,
0
x
0
) = G(x;
1
,
0
x
0
) +
_

d
x
0
G(dx
0
;
1
,
0
x
0
)
_
p
d
(x; x
0
)
+
_

u
x
0
G(ux
0
;
1
,
0
x
0
)
_
p
u
(x; x
0
). (11.34)
Proof. Since G pays
1
x
t
+
0
one has, when x = x
t
at time t, that
G(x
t
;
1
,
0
) = E
Q
_ _

t
e
r(st)
(
1
x
s
+
0
)ds
_
=

1
r
x
t
+

0
r
. (11.35)
From (10.18) the general solution to the ode for F is already known. The two boundary conditions
are
_

1
r
dx
0
+

0
x
0
r
_
+k
1
(dx
0
)

1
+k
2
(dx
0
)

2
=
d
x
0
,
and
_

1
r
ux
0
+

0
x
0
r
_
+k
1
(ux
0
)

1
+k
2
(ux
0
)

2
=
u
x
0
.
This gives a two dimensional linear system in (k
1
, k
2
) which is easily solved. Using the solutions
for k
1
and k
2
gives (11.33). By using the expressions for G and the ArrowDebreu prices, one can
write F as in (11.34).
From (10.29) and (10.30) the Arrow-Debreu prices in the setup of this section are equal to
p
d
(x; x
0
) =
u

_
x
x
0
_

_
x
x
0
_

2
, (11.36)
p
u
(x; x
0
) =
d

_
x
x
0
_

1
+
d

_
x
x
0
_

2
. (11.37)
In particular, Lemma 11.2.1 gives the initial value of the F claim
F(x
0
; x
0
,
1
,
0
x
0
) =
_

1
r
+

0
r
_
x
0
(11.38)
+x
0
1

__

1
r
u +

0
r

u
_
d

_

1
r
d +

0
r

d
_
u

2
_
+x
0
1

_

1
r
u +

0
r

u
_
d

1
+
_

1
r
d +

0
r

d
_
u

1
_
,
108 Modeling the tax advantage to debt: trade-o theory
which shows, given the conditions of Lemma 11.2.1, that the initial value of F is scalable in the
initial EBIT level. This was also the case for equity and debt in Section 11.1 with the Leland
(1994b) model.
If one can write equity and debt such that the conditions of Lemma 11.2.1 are fullled, the
equity value is derived from (11.33). It is then easy to write the two equations (11.29) and (11.28)
which only depend on (d, u). Thus there are only two equations two solve. This must usually be
done numerically.
The boundary conditions in Goldstein, Ju, and Leland (2001) are derived by a slightly dierent
reasoning. The idea is that the value of equity, say, depends not only on the cash ow received
in the initial period but also on the cash ow received in all the future periods with coupon
rates uC
0
, u
2
C
0
, . . . etc. until a default occurs. Furthermore, there is value added at the future
restructuring dates. Similarly, the value of debt depends on the current and future coupon payments
and on future restructurings.
14
This line of reasoning requires that the rm has basically the same
equity and debt holders in all periods. Because of the homogeneity property it is possible to derive
conditions equivalent to (11.30) and (11.31).
11.2.5 Main results from allowing more debt
An important result is that when the rm can issue more debt, then the initial leverage decreases.
This result is interesting because the Leland (1994b) model needs very high bankruptcy costs to
obtain leverage levels comparable to what is observed. The reason for the decrease in the leverage
is that the rm initially issues debt with a lower coupon rate because the rm may increase its
coupon payments if EBIT increases. And if EBIT decreases it is better to have a relatively low
coupon rate in order to avoid a default.
Since the coupon rate decreases, one further observes that the default level in the Goldstein,
Ju, and Leland (2001) model is lower than in the Leland (1994b) model. As a consequence one also
nds that the yield spread is lower if future restructurings are possible. Moreover, the dynamic
model better exploits the tax shield. Following Goldstein, Ju, and Leland (2001) we dene the tax
advantage to debt, denoted TAD, as the optimally levered rm value relative to the unlevered rm
value. The optimally levered rm value, when EBIT is x
0
, is E(x
0
; x
0
) + (1 k)D(x
0
; x
0
). For an
14
One can either say that the current debt holders are no longer debt holders after their debt has been called or
one can say that the debt holders all agree to increase their coupon rates to uC
0
, u
2
C
0
, . . . until default.
11.2 A capital structure model with increasing debt payments 109
EBIT level of x we dene the optimally levered rm value as
A(x)

= E(x; x) + (1 k)D(x; x) = A x
with
A = E(1; 1) + (1 k)D(1; 1), (11.39)
where A is the optimally levered rm value per unit of EBIT. Recall that V
U
(x
0
) is the value of
the unlevered rm, dened in (11.4). Hence we dene the tax advantage of debt as
TAD =
E(x
0
; x
0
) + (1 k)D(x
0
; x
0
) V
U
(x
0
)
V
U
(x
0
)
=
A
1
e
r
1. (11.40)
Thus, TAD is the return to the initial rm owners from levering the rm relative to the unlevered
rm value.
If the rm cannot increase its debt, Goldstein, Ju, and Leland (2001) nd that TAD is about
67%. If the rm can restructure upwards TAD increases to about 79%. Thus, one can conclude
that upward restructurings increase the value of the tax shield, however, the eect appears to be
moderate.
15
In Figure 11.1 the debt, equity, and rm values are depicted in the interval x [dx
0
, ux
0
].
The continuously optimally restructured rm value v(x; x) is also illustrated. For the derivation
of the model illustrated in the gure it is assumed that equity holders determine the call as well
as the default boundary. As illustrated, the optimal coupon rate per unit of EBIT is c

= 0.62.
Furthermore, note that equity smooth pastes to zero, its default value, at the lower bound. In
addition, equity is a convex function, whereas debt is a concave function of EBIT. Finally, a tax
refund rate of = 50% is assumed; an explanation for this parameter is given below when the
consequence of negative net income is discussed. Given this tax refund rate, the tax advantage of
debt turns out to be about 10%.
15
The optimal coupon rate appears to be very small in the Goldstein, Ju, and Leland (2001) article. I conjecture
that this is because the coupon rate enters the drift of EBIT with a negative sign. A higher coupon rate thus slows
down the expected growth rate of the earnings. Without this relation between and C, TAD may increase more
when upward restructurings are possible.
110 Modeling the tax advantage to debt: trade-o theory
0.5 1 1.5 2 2.5
10
20
30
40
50
60
x
c

E(x; x) +D(x; x)
E(x; 1) +D(x; 1)
E(x; 1)
D(x; 1)
Figure 11.1: Debt and equity values both with xed coupon rate, c

, (xed optimally when debt

initially is issued at date zero when x
0
= 1) and with continuously optimally determined coupon
rate to the given EBIT level, x, as function of the EBIT level, x, for the model with no possibilities
of renegotiations of the debt terms in the base case: = 2%, = 25%, r = 4.5%,
i
= 35%,

e
= 50%, = 5%, = 25%, k = 3%, and = 50%.
11.3 A dynamic capital structure model without loss of the tax
shield
It seems natural to extend the Goldstein, Ju, and Leland (2001) model such that the rm is allowed
to restructure its debt so as to pay a lower coupon rate without immediately losing the possibility
to exploit the tax shield of the debt. In this section and in the next I consider a model where
the rm can renegotiate its debt downwards and call the debt following prosperous times. In the
next section there is a partly endogenous bargaining game between the equity holders and the debt
holders. In this section the equity holders can enforce a debt renegotiation. This and the next
section are taken from the model in Christensen, Flor, Lando, and Miltersen (2002a).
11.3.1 The model
The model in Christensen, Flor, Lando, and Miltersen (2002a) is related to Goldstein, Ju, and
Leland (2001). The state variable is EBIT and, as in (11.1), the EBIT process develops under the
pricing measure as
dx
t
= x
t
dt +x
t
dW
t
. (11.41)
Contrary to the previous section, assume now that the expected growth rate of the earnings, , is
specied exogenously. In particular, the coupon paid to debt does not inuence on the expected
11.3 A dynamic capital structure model without loss of the tax shield 111
growth rate.
Like Goldstein, Ju, and Leland (2001) the rm can only issue callable perpetual debt which
is issued at par value and there are constant default and call boundaries, denoted x

= d x
0
and
x

= u x
0
, respectively. Unlike Goldstein, Ju, and Leland (2001) both boundaries are determined
in accordance with the incentives of the equity holders.
When the debt is called, the equity holders must pay the principal of the debt and a call premium
to the debt holders. For tractability it is assumed that the premium is proportional to the principal
by the factor . Immediately after the debt has been called, the rm is owned completely by the
equity holders. Therefore the new equity and debt contracts are issued so as to maximize the rm
value less the restructuring costs.
16
Again, for tractability, the restructuring costs are proportional
to the principal of the debt by a factor k. Thus, if the optimal coupon rate is C

= c

x
0
and the
initial level of EBIT is x
0
, the debt and equity values at the call boundary are
D(x; x
0
) = (1 +)D(x
0
; x
0
),
E(x; x
0
) = max
C0
_
E(x; x) + (1 k)D(x; x)
_
(1 +)D(x
0
; x
0
).
Like in Goldstein, Ju, and Leland (2001) this model has the positive homogeneity property; this
property is discussed in more detail in the model in Christensen et al. (2002b). Therefore, when the
equity holders at x = x decide for the optimal coupon rate, this coupon rate will be the same as the
coupon rate chosen initially scaled up by the factor u =
x
x
0
. The conditions at the call boundary
therefore become
D(ux
0
; x
0
) = (1 +)D(1; 1)x
0
(11.42)
E(ux
0
; x
0
) =
_
E(1; 1) + (1 k)D(1; 1)
_
(ux
0
) (1 +)D(1; 1)x
0

= A ux
0
(1 +)D(1; 1)x
0
. (11.43)
We now turn to the lower boundary. For simplicity, suppose rst that when the equity holders
cease to pay the coupon it immediately triggers default. At default, the absolute priority rule
is respected, so equity has value zero.
17
The debt holders take over the rm and lever the rm
optimally. A default is costly, and the fraction of the debt holders value is lost due to default
16
It is assumed that the parameters of the model are chosen so that there does exist an optimal coupon rate. If
the tax advantage to debt is too large, the optimal coupon rate can be innitely high.
17
In fact, equity could have a positive value, however, to keep the notation simple, we omit that possibility here.
The model in Christensen, Flor, Lando, and Miltersen (2002a) deals with this matter.
112 Modeling the tax advantage to debt: trade-o theory
costs. Thus, at the lower boundary, when the absolute priority rule is respected, one has
D(dx
0
; x
0
) = (1 )dx
0
A (11.44)
E(dx
0
; x
0
) = 0. (11.45)
Comparing (11.44) to (11.26) one sees that in (11.44) the debt holders can lever the rm after a
default.
Consider now a dierent outcome at the lower boundary. At the lower boundary, i.e. when
x = dx
0
, the rm will be restructured to have the optimal capital structure. A restructuring
requires a bargaining game between the debt holders and the equity holders. Assume for simplicity
that the outcome of the bargaining over the restructuring gain is common knowledge. Specically,
assume that the equity holders receive a fraction [0, 1] of the value of the restructured rm
minus the restructuring costs. The debt holders get the rest. Hence, the boundary conditions at
the restructuring boundary are
D(dx
0
; x
0
) = (1 )dx
0
A (11.46)
E(dx
0
; x
0
) = dx
0
A. (11.47)
The important thing to note in (11.47) compared to (11.45) is that equity can have a positive
value albeit the debt holders get less than their principal. As in Christensen, Flor, Lando, and
Miltersen (2002a), one can relate to the bankruptcy costs as follows. The bankruptcy costs are
equal to the gain of renegotiating the debt, when a liquidation is avoided, i.e.
G
R
= dx
0
A. (11.48)
At the lower boundary, the equity and debt holders bargain over the saved default costs, G
R
.
Suppose that the equity holders receive the fraction of the gain. Then = . Therefore, the
model with (11.46) and (11.47) implies, that the bankruptcy costs can be avoided through debt
restructurings.
One still needs to determine the relative boundary parameters (d, u). This is done in accordance
with the incentives of the equity holders because it is the equity holders who decide when to
restructure the capital. Thus, the smooth pasting conditions related to (11.43) and (11.45) or
(11.47) are used.
11.3.2 Negative net income
Similar to the Goldstein, Ju, and Leland (2001) model, the model in Christensen, Flor, Lando, and
Miltersen (2002a) considers the possibility of reducing the tax shield for the debt when the earnings
11.3 A dynamic capital structure model without loss of the tax shield 113
0.5 1 1.5
-0.5
0.2
dividends
cflm
gjl
both
x
Figure 11.2: Illustration of the dividends to the equity holders with the changed tax deductability
system in the Goldstein, Ju, Leland 2001 (GJL) and in the Christensen et. al. (CFLM) models. In
the example C = 0.9, = 0.5, and
e
= 0.5. In general, does not have to be equal in the two
models.
are insucient to cover the interest payments. However, the reduction in interest deductability is
done in a slightly dierent way, namely

1
x +
0
x
0
=
_

_
(1
e
)(x c

x
0
) x c

x
0
,
(1
e
)(x c

x
0
) x < c

x
0
.
(11.49)
Compared to (11.21) the above implementation makes the dividends to the equity holders contin-
uous in the earnings of the rm, see Figure 11.2. This is more in line with the idea that the lower
the earnings, the less likely is future use of the tax shield. Actually, in Goldstein, Ju, and Leland
(2001) the equity holders will have to pay a large payment to the rm to avoid cash ow shortage
just because the earnings are marginally lower than the coupon.
11.3.3 Some results
In Figure 11.3 we show the eects of varying the eective tax refund parameter in the model with
no debt renegotiation at the lower bound, i.e. equity has value 0. Negative net income is modeled
as in Christensen et al. (2002b). Observe, that the optimal coupon is very sensitive in the tax
refund rate. Thus, TAD and leverage is also very sensitive to the size of tax refund. Moreover,
the continuation region (d, u) for a given initial EBIT level shrinks as tax refund increases. Firstly,
a higher tax refund induces a higher coupon rate which in turn increases the default boundary.
Secondly, a higher tax refund rate makes in more attractive to increase the coupon payments when
EBIT increases and, hence, the call boundary decreases.
114 Modeling the tax advantage to debt: trade-o theory
0.2 0.4 0.6 0.8 1
0.5
1
1.5
2
2.5
3
E

u
c

d
(a) Upper and lower restructuring boundaries
0.2 0.4 0.6 0.8 1
5
10
15
20
25

E + (1 k)D
TAD
E
D
(b) Debt and equity values
0.2 0.4 0.6 0.8 1
20
40
60
80
100

%
D
E+D
(c) Leverage
0.5 1 1.5 2 2.5
10
20
30
40
50
60
x
E(x; x) +D(x; x)
E(x; 1) +D(x; 1)
c

E
D
(d) Optimal coupon rate and yield spread
Figure 11.3: Upper and lower restructuring boundaries, initial (when x = 1) debt and equity values,
initial leverage, initial optimal coupon rates, and initial optimal yields spreads as function of the
fractional reduction of the tax rate when EBT is negative, , for the model with no possibilities of
renegotiations of the debt terms in the base case: = 2%, = 25%, r(1
i
) = 4.5%,
i
= 35%,

e
= 50%, = 5%, = 25%, and k = 3%.
11.3 A dynamic capital structure model without loss of the tax shield 115
As with Goldstein, Ju, and Leland (2001) the model in Christensen, Flor, Lando, and Miltersen
(2002a) can be reduced to two equations with (d, u) as the unknowns for a given bargaining power,
, and coupon rate, C

. The two equations must be solved numerically.

0.1 0.2 0.3 0.4
0.5
1
1.5
2
2.5
3

u
c
d
(a) Restructuring boundaries
0.1 0.2 0.3 0.4
5
10
15
20
25

E + (1 k)D
TAD
E
D
(b) Security values
0.1 0.2 0.3 0.4
6
8
10
12
14

%
c

D
r
(c) Yield spread
0.5 1 1.5 2 2.5
10
20
30
40
50
x
c
E
D
E(x; x) +D(x; x)
E(x; 1) +D(x; 1)
(d) Value functions
Figure 11.4: Changing the bargaining power in the Christensen et. al. model. In the example
= 0.02, = 0.25, = 25%, k = 0.03, = 5%, = 0.5, r = 4.5%,
i
= 35% and
e
= 50%. The
vertical dotted line marks the optimal = 0.15.
Figure 11.4 shows what happens with the restructuring boundary, TAD, etc. when the equity
holders bargaining power increases. When the equity holders have more bargaining power, they will
be more inclined to make a capital restructuring, as seen in Figure 11.4(a). Figure 11.4(b) shows
the initial equity, debt, TAD, and rm value (minus restructuring costs). The more bargaining
power the equity holders have, the higher value they have initially. The positive eect of more
bargaining power to the equity holders is that the capital structure, and in particular the coupon
rate, is more often adjusted to the earnings of the rm. On the other hand, more bargaining power
to the equity holders decreases the initial debt value because the debt holders foresee that they
receive less at a downward restructuring. Although it is hard to tell from Figure 11.4, it turns out
116 Modeling the tax advantage to debt: trade-o theory
that = 0.150 yields the highest value to the initial rm owners.
Figure 11.4(d) illustrates the value function when EBIT varies inside the continuation region
when = 0.150. This gure should be compared to Figure 11.3(d) (or Figure 11.1). We immediately
observe that the value function generally have the same form. However, in the case where equity
receives some positive value at the lower boundary, equity smooth pastes to a positive value (with
a slope higher than 0). Moreover, the coupon rate is about 50% higher in this case and, hence, the
tax advantage to debt is increases when the tax shield is not lost at default.
11.4 Dynamic capital structure with debt renegotiations
The distribution of the value of the debt and equity claims at the lower boundary in the above
section appears to be very exogenously specied. The model in Christensen et al. (2002a) considers
an extension which modies the restructuring boundary given in conditions (11.46) and (11.47).
The idea is to include the possibility that the equity holders as well as the debt holders can pull out
from the renegotiation. Why is it interesting to include the pull out possibility? Consider the gain
to restructure the debt in (11.48). This gain is calculated under the condition that the alternative
to a debt renegotiation is a liquidation. However, it may not be optimal for the equity holders
to cease paying the coupon; they might rather pay the coupon for a little while longer and then
propose another debt renegotiation. In other words, it may not be a credible strategy to liquidate
the rm.
Suppose that the equity holders propose a capital restructuring to the debt holders. If the
proposal is accepted, the capital of the rm is optimally restructured. However, the equity holders
will have to take into account that the debt holders can reject the proposal. The debt holders will
only accept the restructuring oer if they are as least as well o as they would be by rejecting it.
This simple game is illustrated in Figure 11.5.
Given that the restructuring oer has been given to the debt holders, the alternative to a capital
restructuring is either liquidation or continuation with the pre-oer coupon rate. The equity holders
make this decision in order to maximize the equity value if the restructuring oer is rejected.
If the equity holders decide to continue to service the debt holders as stated in the debt contract,
the debt holders cannot enforce a liquidation of the rm. However, if it is not costly to propose
a debt renegotiation the equity and debt values will not change at x, where a renegotiation is
proposed but rejected. To calculate the continuation values of equity and debt with the pre-oer
11.4 Dynamic capital structure with debt renegotiations 117
Ask
Restructure
Liquidate
Continue
Yes
No
E
E
D
Figure 11.5: The renegotiation game between the equity holders and the debt holders in the model
from Christensen et. al., when it is possible to pull out from the renegotiation.
coupon rate, it is necessary to apply an iteration procedure. In economic terms, one can think
of the number of iterations as being the number of options the rm has left for a restructuring
proposal. Let the number of options be n. If n = 0 there can be no restructuring oer and the
model is similar to the model in Section 11.1.
Suppose that initially there are n > 0 options left for a restructuring oer. The debt is issued
with the optimal coupon rate C

n
. Following a renegotiation with a debt restructuring the optimal
coupon rate is C

n1
. Similarly, let the subscript n 1 written elsewhere in this section have the
interpretation that there are n 1 restructuring proposals left. Thus E
n1
(x; C

n
) is the value of
the equity when the EBIT is x, given that there are n 1 options left and given that the coupon
rate has been set when there were n options left.
18
The Reject Continuation (RC) value of the
equity and debt can now be written as
19
E
RC
n
(dx
0
; C

n
) = E
n1
(dx
0
; C

n
) (11.50)
D
RC
n
(dx
0
; C

n
) = D
n1
(dx
0
; C

n
). (11.51)
If the equity holders decide not to pay the coupons, the rm is liquidated. Those who buy
the liquidated rm are not committed to any contracts so they will issue new equity and debt to
18
For simplicity, I exclude the argument for the level of EBIT, when the debt was issued. Due to the homogeneity
property, E
n1
(x; C

n
) can be calculated (approximately) if the model with n 1 restructuring options has been
solved for. The trick is to have c

n
x
0,n

= C

n
= C

n1
, so the initial EBIT level with n1 restructuring options should
be x
0,n1
=
c

n
x
0,n
c

n1
.
19
The terminology Reject Continuation comes from the actions in Figure 11.5: rst rejection of the oer by the
debt holders which is followed by the equity holders decision to continue to service the debt. Reject Liquidation
mentioned below should be though of in a similar manner.
118 Modeling the tax advantage to debt: trade-o theory
maximize the rm value minus restructuring costs. However, there are only n 1 options now.
As above, the cost of liquidation is the proportion of the new rm value minus restructuring
costs. The remaining liquidation value is divided between the equity holders and the debt holders
according to the absolute priority rule. Hence, the Reject Liquidation (RL) value is
E
RL
n
(dx
0
; C

n
) = max
_
(1 )
_
E
n1
(dx
0
, C

n1
) + (1 k)D
n1
(dx
0
, C

n1
)
_
D
n
(x
0
, C

n
), 0
_

= max
_
(1 )
n1
dx
0
D
n
(x
0
, C

n
), 0
_
(11.52)
D
RL
n
(dx
0
; C

n
) = min
_
(1 )
n1
dx
0
, D
n
(x
0
, C

n
)
_
. (11.53)
Suppose that the outcome of the renegotiation process is that there is No Renegotiation (NR).
The equity holders will then follow the strategy which maximizes their value. From (11.50)(11.53)
one has
E
NR
n
(dx
0
; C

n
) = max
_
E
RL
n
(dx
0
; C

n
), E
RC
n
(dx
0
; C

n
)
_
(11.54)
D
NR
n
(dx
0
; C

n
) = 1
ERL
D
RL
n
(dx
0
; C

n
) + 1
ERC
D
RC
n
(dx
0
; C

n
), (11.55)
where 1
ERL
= 1 if the equity holders choose to liquidate and zero otherwise. Equivalently, 1
ERC
is
the indicator function equal to 1 if the equity holders choose to continue to service the debt holders.
Now, the gain from restructuring is dened as the dierence between the optimally restructured
rm value minus the restructuring costs and the no renegotiation value, i.e.
G
R
n
(dx
0
; C

n
) = dx
0

n1

_
E
NR
n
(dx
0
; C

n
) +D
NR
n
(dx
0
; C

n
)
_
, (11.56)
because of the homogeneity property. Note the dierence between (11.56) and (11.48). In (11.56),
the gain to a renegotiation is not just the saved liquidation costs. Instead, the optimal alternative
for the equity holders determines the restructuring gain.
The equity and debt holders may now bargain over the gain G
R
n
to optimally restructure the
rm. For simplicity the model assumes that the outcome of the bargaining over the restructuring
gain is a priori given. Suppose, as before, that the equity holders receive the fraction [0, 1] of the
restructuring gain and that the debt holders receive the rest. The conditions at the restructuring
boundary are therefore
E
n
(dx
0
; C

n
) = G
R
n
(dx
0
; C

n
) +E
NR
n
(dx
0
; C

n
), (11.57)
D
n
(dx
0
; C

n
) = (1 )G
R
n
(dx
0
; C

n
) +D
NR
n
(dx
0
; C

n
). (11.58)
11.4 Dynamic capital structure with debt renegotiations 119
Model c d u A E D Lev(%) s(bps) TAD(%)
APR 0.63 0.21 2.63 22.04 14.37 7.91 35.5 107 10.2
43 58 -12 4 -11 34 28 46 49
-split 0.90 0.31 2.25 23.03 12.77 10.58 45.3 156 15.2
Table 11.2: Comparison of restructuring boundaries, coupon rates etc. of the models with the
absolute priority rule (APR) and with the -split, when EBIT = x
0
. The initial rm value is
v = E + D, so A = v kD is the rm value minus restructuring costs. The leverage is Lev =
D
v
and the (before tax) par yield spread is s = y r/(1
i
) =
cx
0
D
r/(1
i
). The italic numbers
show the percentage dierence between the numbers above and below. The parameters used are
x
0
= 1, = 0.02, = 25%, r = 4.5%,
i
= 35%,
e
= 50%, = 5%, = 25%, k = 3%, = 50%,
and = 50%.
The model is now solved by rst considering the case when the rm initially has zero restruc-
turing options. Given this solution, the model with one restructuring option can be solved and so
forth. For a suciently large number of options it turns out that there is essentially no dierence.
When this is the case, a xed point is achieved, and the model is solved.
11.4.1 Some results
Table 11.2 compares the dierent models: the model with the absolute priority rule respected and
the model where equity receives the fraction of the renegotiation gain. It is interesting to note
that the possibility to break the absolute priority rule results in an increase of TAD of about
50% and a much higher coupon rate. Allowing for downward restructurings, so that the equity
holders also receive some value, is benecial because it is possible to avoid paying the liquidation
costs. Furthermore, Table 11.2 shows that the equity holders are much more willing to enter a
restructuring, i.e. u decreases and d increases, when the absolute priority rule can be broken.
Finally, one may compare the two models which break with the absolute priority rule, i.e. the
models where the equity holders receive the fraction or , respectively, of the renegotiation gain.
The model in Christensen et al. (2002a) nds that there will be a lower coupon, that the debt will
be called earlier, and that the equity holders will propose a debt renegotiation for a lower EBIT
level in the model with the split.
120 Modeling the tax advantage to debt: trade-o theory
11.5 The return premium to leverage
Kane, Marcus, and McDonald (1985) put forward the idea that the tax advantage to debt can
be measured by looking at the drift rate. This idea has also been used by Fischer, Heinkel, and
Zechner (1989a) and by Fischer, Heinkel, and Zechner (1989b).
An important insight in the above mentioned papers is that the market value of real assets
must be equal to the market value of an optimally levered rm holding these assets; the problem
of considering V as the price of a traded asset, as in Leland (1994b), is thus not apparent. Using
this insight one can derive the return premium to leverage. The following discussion is based on
Fischer, Heinkel, and Zechner (1989a).
11.5.1 Another state variable and a comparison
Suppose, contrary to the other models considered so far, that the state variable is the value today
of the continuously optimally levered rm, denoted A. A follows a geometric Brownian motion
dA
t
= A
t
dt +A
t
dW
t
, (11.59)
under the risk neutral measure.
The rm issues perpetual debt at par as in the previous models. Since < r = (1
i
)r,
Fischer, Heinkel, and Zechner (1989a) can dene the expected return dierence as
= (1
i
)r . (11.60)
One may think of as the convenience yield of having access to optimally lever the assets of the
rm.
Fischer, Heinkel, and Zechner (1989a) consider the model from a (basically) rst best point
of view, i.e. one can write contracts on the level of A.
20
Specically suppose that when either
A = A < A
0
or A = A > A
0
, the debt is called without a premium. If the debt is risky, then there
is a default when A = A. In this case the equity holders receive nothing and the debt holders take
over the rm and issue new securities as in Goldstein, Ju, and Leland (2001). Since it is possible to
contract on A = A and A = A, the boundary levels are a part of the initial optimization problem.
Second best optimization, i.e. when the restructuring boundaries are inferred from the incentives
of the equity holders, is done in Fischer, Heinkel, and Zechner (1989b).
20
In Fischer, Heinkel, and Zechner (1989a) the rst best solution is to pay innitely high coupons and receive
innitely high tax subsidiaries.
11.6 An alternative evolution of earnings 121
An interesting observation as regards this particular choice of the state variable is done in the
model presented in Christensen, Flor, Lando, and Miltersen (2002a) by comparing the state variable
A with the state variable x (EBIT). As in (11.41) let x be a geometric Brownian motion. Denote
the optimal coupon rate C

= c

x
t
, when the debt is issued for EBIT equal to x
t
. Then, with the
notation from Section (11.3), A and x are related as
A
t
= E(x
t
; x
t
) + (1 k)D(x
t
; x
t
)
=
_
E(1; 1) + (1 k)D(1; 1)
_
. .

=A
x
t
, (11.61)
because of positive homogeneity of degree one. It follows that A
t
derived this way is also a geometric
Brownian motion as in (11.59) and in fact = . This implies that the return premium to leverage,
, is exogenously specied.
A natural question then is how Fischer, Heinkel, and Zechner (1989a) can search for an optimal
return premium and hence derive . If A is the state variable given by (11.59), then (11.61) becomes
A
t
=
_
E
A
(1; 1) + (1 k)D
A
(1; 1)
_
. .

=
A
( ,c

)
A
t
hence
1 =
A
( , c

), (11.62)
where the superscript refers to the assumption that A is given by (11.59). It seems reasonable to
assume that a higher monotonically yields a higher initial rm value minus restructuring costs
and that hence there is a monotonic relationship between and the coupon rate c

; this conjecture
has only been veried numerically. But then it follows that there is at most one pair ( , c

) which
satises the identity in (11.62). So, as above, it turns out that is uniquely and, in fact, exogenously
given.
11.6 An alternative evolution of earnings
So far we have assumed that the earnings of the rm, EBIT, follows a geometric Brownian motion.
The key benet of this assumption is that we obtain the homogeneity property which makes it
possible to handle upward and downward debt restructuring in a tractable manner. On the other
hand, this assumption implies that EBIT is always positive, an assumption which seems somewhat
122 Modeling the tax advantage to debt: trade-o theory
unrealistic. One way to circumvent this is to let gross prot be modeled as above and then introduce
a constant ow cost in production. We return to this in a later chapter.
21
Alternatively, one can consider EBIT to follow an arithmetic Brownian motion. This is done
in e.g. Genser (2006) and Bank and Lawrence (2005). In order to illustrate, suppose that EBIT
follows
dx
t
= dt +dW
t
. (11.63)
As before we are working directly under a pricing (risk neutral) measure Q. We also assume that
r > (/)
2
, which seems to be a reasonable assumption. We can apply the standard technique
to obtain the PDE for the price of a traded asset, denoted F(x), which receives no payment ow
before a boundary is hit. We get (assuming no time dependence)
1
2

2
F
xx
(x) +F
x
(x)

= rF(x) (11.64)
which has the solution
F(x) = A
1
e
k
1
x
+A
2
e
k
2
x
, (11.65)
where
k
1
=
+
_

2
+ 2r
2

2
, k
2
=

_

2
+ 2r
2

2
. (11.66)
Proof. We conjecture the solution (11.65). It follows that
F

(x) = k
1
A
1
e
k
1
x
+k
2
A
2
e
k
2
x
,
F

(x) = k
2
1
A
1
e
k
1
x
+k
2
2
A
2
e
k
2
x
,
whence it follows
1
2

2
(k
2
1
A
1
e
k
1
x
+k
2
2
A
2
e
k
2
x
) +(k
1
A
1
e
k
1
x
+k
2
A
2
e
k
2
x
) = r(A
1
e
k
1
x
+A
2
e
k
2
x
).
Rearranging terms yields
_
1
2

2
k
2
1
+k
1
r

A
1
e
k
1
x
+
_
1
2

2
k
2
2
+k
2
r

A
2
e
k
2
x
= 0.
21
This is the Mella-Barral and Perraudin (1997) model.
11.6 An alternative evolution of earnings 123
Dene the fundamental quadratic as
Q(k) =
1
2

2
k
2
+k r,
and let k
1
be the positive root and k
2
the negative root.
Furthermore, we can derive the perpetual value of receiving the payo ow
1
x+
0
. Since EBIT
in (11.63) follows an arithmetic Brownian motion we nd the fundamental component (perpetual
value) of EBIT as
E
_ _

0
e
rt
x
t
dt
_
=
_

0
e
rt
E[x
t
]dt =
_

0
e
rt
[x
0
+t]dt =
x
0
r
+

r
2
and, hence, we get
V

(x
0
) = E
_ _

0
e
rt
(
1
x
t
+
0
)dt
_
=
1
_
x
0
r
+

r
2
_
+

0
r
. (11.67)
If we consider the unlevered rm, keeping this rm a live forever would thus have the value
V

U
(x) = (1
e
)
_
x
r
+

r
2
_
. (11.68)
Note that in contrast to the geometric Brownian motion case, it is possible that the expression
in (11.68) becomes negative. In the arithmetic Brownian motion framework, it can therefore be
optimal to liquidate (abandon) the rm (receiving zero in return) in nite time. If equity holders
have the option to abandon the rm we should take this option into account. Let V
U
(x) be the
unlevered rm value in this case. Then we can use (11.68) and (11.65) to get
V
U
(x) = (1
e
)
_
x
r
+

r
2
_
+A
1
e
k
1
x
+A
2
e
k
2
x
.
Ruling out speculative bubbles gives us A
1
= 0
V
U
(x) = (1
e
)
_
x
r
+

r
2
_
+A
2
e
k
2
x
,
and value matching at abandonment implies
V
U
(x) = (1
e
)
_
x
r
+

r
2
_
+A
2
e
k
2
x

= 0,
A
2
= (1
e
)
1
r
(x +

r
2
)e
k
2
x
.
Collecting terms yields
V
U
(x) = (1
e
)
_
x
r
+

r
2
(
x
r
+

r
2
)e
k
2
(xx)
_
, (11.69)
124 Modeling the tax advantage to debt: trade-o theory
Finally, smooth pasting yields
x =
1
k
2

r
(11.70)
as the optimal EBIT level for liquidation. Since we have assumed r > (/)
2
, k
2
is indeed a root
less than zero and, hence, x < 0, i.e. abandonment occurs at a negative EBIT.
Clearly, one could easily extend this framework to a capital structure model a long the lines
of Leland (1994b), an exercise for the reader. However, since we cannot hope for homogeneity
when EBIT follows an arithmetic Brownian motion a model with dynamic capital adjustments is
generally a harder task.
Chapter 12
Analysis of bankruptcy law
environments
The research on bankruptcy law is mainly inspired by the law system in USA., that is the U.S.
Bankruptcy Code. For our purposes (studying corporate default) the most relevant parts to know
about are Chapter 7 and Chapter 11. Essentially, the U.S. Bankruptcy Code Chapter 7 corresponds
to liquidating the rm by selling all assets and divide the proceeds according to priority. For Chapter
11, see footnote 1. In the models in chapter 11, where we focused on tax shield exploitation, we also
distinguished between liquidation or reorganization. However, the outcome of a particular model
in chapter 11 is either always liquidation, e.g. Leland (1994b) or Goldstein, Ju, and Leland (2001),
or always a reorganization of the rms capital structure with a new debt issuance, e.g. Christensen
et al. (2002a) or Flor and Lester (2002). However, before a rm enters Chapter 7 it has often been
in Chapter 11 distress for a while, see Wruck (1990) or Gilson (1997).
Features of Chapter 11. Automatic stay, absolute priority, potential debt forgiveness.
The purpose of Chapter 11 is to allow the rm to reorganize its capital such that it can continue
its operation, i.e. the rm survives as a going concern (perhaps in a modied form). Thus, Chapter
11 is a mechanism with which the right to various claims and the associated incentives of the rms
stake holders can be resettled.
12.1 The possibility to enter and leave chapter 11
Fran cois and Morellec (2001) have an interesting model which approximates a situation where the
rm is not immediately liquidated but rst goes through a period of default.
125
126 Analysis of bankruptcy law environments
(Update to Fran cois and Morellec (2004))
The framework is closest in spirit to Leland (1994b) and Fan and Sundaresan (2000). The
rm issues perpetual debt and, as in Leland (1994b), the tax structure is simplied to a single
parameter which accounts for the subsidiary coupon rate from the State to the rm. Let V be the
value of the assets of the rm and V the total instantaneous payout from the rm as in Section
13.3. Similar to Mella-Barral and Perraudin (1997) and Fan and Sundaresan (2000) the debt may
be renegotiated for a suciently low level of the value of the assets, V
D
. As in Fan and Sundaresan
(2000) the debt holders do not receive the promised coupon rate when V < V
D
but only a part of
the payout rate from the rm.
The cost structure of renegotiation is as in Fan and Sundaresan (2000). The debt renegotiation
is implicitly costly because the tax shield is lost during the renegotiation period and renegotiations
also have a direct cost. The direct renegotiation cost rate is V +l. The extension in this model is
that the rm can only be in a renegotiation process for d consecutive years. Thus, if the value of
the assets of the rm gets below V
D
and then exceeds V
D
before the d years have passed, the rm
continues as if nothing has happened. If V remains below V
D
for d years, the rm is liquidated
according to the absolute priority rule and a fraction is lost due to liquidation costs.
An interesting result in Fran cois and Morellec (2001) is that because of the renegotiation costs
it is not always possible to have renegotiation. In fact, if V + l > V , i.e. if the instantaneous
renegotiation costs are higher than the liquidation costs, then there will never be renegotiation but
only liquidation. On the other hand, if renegotiation is costless, there will never be liquidation.
Moreover, the default level V
D
and the initial leverage will decrease with higher renegotiation
costs. The reason is that the equity holders also have to pay the renegotiation costs and these cost
will be delayed with a lower default level. Because the equity holders lose the tax shield during
the renegotiation period, a longer renegotiation period also decreases the leverage if renegotiation
is suciently costly.
When the value of the assets of the rm is below the default level V
D
, it is interpreted as the rm
being in chapter 11 distress. Thus, if the rm is in chapter 11 for d years, it automatically enters
chapter 7. Empirical studies by Franks and Torous (1989), Eberhart, Moore, and Roenfeldt (1990),
Gilson (1997), Helwege (1999), and Weiss and Wruck (1998) nd that the average period spend in
chapter 11 is about two years. According to Gertner and Scharfstein (1991) the U.S. Bankruptcy
Code is written so that a judge has to approve that the chapter 11 period is extended from the
initial 120 days which the rm is entitled to. The empirical evidence cited above indicates that the
12.1 The possibility to enter and leave chapter 11 127
period is often extended. Presumably, the judge will be more reluctant to extend the chapter 11
period if the rm has performed badly. A simple extension to account for this could be to include
a liquidation level, V
L
< V
D
. Then if the rm performs suciently badly so that V = V
L
, the rm
will immediately be liquidated albeit the d years have not passed. Moreover, a rm typically enters
chapter 11 to have its capital structure changed, and it could be interesting to implement this in
the Fran cois and Morellec (2001) model.
12.1.1 Remembering past defaults
Full memory: Moraux (2002). Partial memory with a distress clock: Galai, Raviv, and Wiener
(2005).
12.1.2 Semi rst best bankruptcy and eects on credit spreads
The various features of the U.S. Bankruptcy Code, such as the automatic stay, priority rules,
and (potential) debt forgiveness, implies that the various stakeholders of the rm have dierent
incentives concerning the triggering and use of the various features. Potentially, the discrepancies
in incentives, control rights, and claims to value (future cash ow) are the basis for suboptimal
decisions, because the stakeholder with the prevailing control rights (the decision maker) maximizes
the value he can obtain which need not coincide with rm value maximizing. In turn, this forms
the basis of potential debt renegotiation.
In a framework without debt renegotiation, Broadie, Chernov, and Sundaresan (2007) analyze
how control rights with respect to declaring bankruptcy (Chapter 11) and liquidation (Chapter 7)
inuence debt and equity values. The outline of their model is illustrated in the sequence of events
in Figure 12.1.
The rm has issued a single perpetual non-callable bond with a xed coupon rate c. As usual, we
denote the rms cash ow as x
t
.
1
When the cash ow becomes lower than the promised coupon,
x
t
< c, the rm enters the equity dilution state. If the cash ow decreases further to the level
x
B
= V
B
(r ), the rm enters the bankruptcy (or default) state. In bankruptcy three cases are
possible:
1. The rms cash ow increases suciently and the rm leaves the bankruptcy state, or
1
x = in Broadie, Chernov, and Sundaresan (2007).
128 Analysis of bankruptcy law environments
Clearing bankruptcy:
(1 )A is forgiven
S is used to pay A
Liquid
state
Default Cost:
V ; A : ct, S :
Liquidation
Cost: V
Equity
dilution
A
t
= 0

t
< c
A
t
= c d
V
t
= V
L
E(
t
) = 0
V
t
V
B
, A
t
> 0
V
t
< V
B
Figure 12.1: Sequence of distress events and the resulting restructuring of the rm.
2. the rm stays in bankruptcy (Chapter 11) too long and is liquidated, i.e. the time in bankruptcy
is longer than a granted grace period, d, or
3. the cash ow deteriorate to a low level at which the rm leaves Chapter 11 and enters Chapter
7. As a result, the rm is liquidated and claimants receive payments according to the priority
of their claim. This occurs at x
L
= V
L
(r ).
The framework in Broadie, Chernov, and Sundaresan (2007) is standard, where the underlying
state variable is the rms cash ow x
t
. The authors only consider taxes and, hence, optimal capital
structure at the end, but we can interpret x as EBIT. However, since we introduce time dependence
in the following, we cannot obtain the usual analytical expression for debt and equity values. In
fact, we also need two additional state (or counting) variables.
The counting variable A
t
is introduced to account for accumulated arrears. If the rm survives
Chapter 11, say at time T, the debt holders receive a lump sum payment equal to A
T
, [0, 1]
(and subsequent coupons). In other words, 1 is the debt forgiveness as a result of a successful
Chapter 11. While in Chapter 11, the equity holders do also not receive any dividends. Instead
the entire EBIT is accumulated in a separate account S
t
. Thus, when Chapter 11 is left at time
T, the debt holders receive the non forgiven arrears A
T
and the equity holders receive the slack
minus the arrears, i.e. S
T
A
T
. If S
T
A
T
< 0 this implies that the equity holders dilute their
claim (or pay in money to cover the decit). Of course, one should think about if leaving Chapter
11 is incentive compatible for the various claimants. If the rm enters Chapter 7, the rms assets
12.1 The possibility to enter and leave chapter 11 129
are liquidated with proportional liquidation costs equal to .
Let us now turn to the specic conditions in the bankruptcy state. The arrears evolve according
to
dA
t
=
_

_
rA
t
dt +cdt, V
L
V
t
< V
B
A
t
dt, V
t
= V
B
0, V
t
> V
B
.
(12.1)
The rst part in (12.1) corresponds to the case where arrears accrue due to missing coupon payments
and compensation with the risk free rate of interest. When the rm leaves (or enters) Chapter 11,
the arrears are cleared to 0, and nally the arrears are 0 when the rm is outside Chapter 11. In
a similar vein, the automatic stay account evolves due to
dS
t
=
_

_
rS
t
dt +x
t
dt, V
L
V
t
< V
B
S
t
dt, V
t
= V
B
0, V
t
> V
B
,
(12.2)
and assume that A and S are c`adl`ag (LLRC) processes. Dene the stopping time associated with
the most resent crossing of the bankruptcy boundary as

V
B
t
= sups t : V
s
= V
B
, (12.3)
the stoping time due to liquidation because of hitting the grace period restriction as

d
t
= infs t : s
V
B
s
d, V
s
V
B
, (12.4)
and liquidation due to hitting the liquadation boundary where the value of equity is zero is
determined as

V
L
t
= infs t : V
s
= V
L
, (12.5)
and, hence, the liquidation time is
T
t
=
d
t

V
L
t
. (12.6)
Given the stopping times, the market value of debt is
D(t, x
t
,
V
B
t
,
V
L
t
) = E
Q
t
__
T
t
t
e
r(st)
_
c1
{V
s
V
B
}
+A
s
1
{V
s
=V
B
}
_
ds
_
+ (1 )E
Q
t
_
e
r(T
t
t)
(V
T
t
+S
T
t
)
_
(12.7)
130 Analysis of bankruptcy law environments
and the market value of equity is
E(t, x
t
,
V
B
t
,
V
L
t
) = E
Q
t
__
T
t
t
e
r(st)
_
(x
s
c)1
{V
s
V
B
}
V
s
1
{V
L
<V
s
<V
B
}
+ (S
s
A
s
)1
{V
s
=V
B
}
_
ds
_
, (12.8)
and by adding debt and equity we obtain the market value of the rms assets
v(t, x
t
,
V
B
t
,
V
L
t
) = E
Q
t
__
T
t
t
e
r(st)
_
x
s
1
{V
s
V
B
}
V
s
1
{V
L
<V
s
<V
B
}
+S
s
1
{V
s
=V
B
}
_
ds
_
+ (1 )E
Q
t
_
e
r(T
t
t)
(V
T
t
+S
T
t
)
_
. (12.9)
Recall, that x
t
= V
t
(r ), so the rm value terms in the debt and equity values could be
rewritten in x terms only. The second term in (12.8) implies that the equity holders pay the costs
of being in Chapter 11. One could consider the alternative, where the Chapter 11 costs are born
by the cash ow, i.e. to change the evolution of the S account. It is also important to note that the
bankruptcy or Chapter 11 level, V
B
, is constant, and Broadie, Chernov, and Sundaresan (2007)
determine the Chapter 11 level in dierent cases: rm value maximization (ex ante optimization or
rst best), debt value optimization, and equity value optimization (second best). The most realistic
case is when the equity holders decide when to trigger Chapter 11. Finally, the liquidation level,
V
L
is determined endogenously by the equity holders limited liability. The liquidation boundary is
not constant, since it depends on the actual size of the automatic stay account.
An interesting extension of the Broadie, Chernov, and Sundaresan (2007) model is to let the
grace period d be determined endogenously as well. Thus, we could obtain a model with several
embedded options:
1. equity holders have an option to declare Chapter 11 or Chapter 7 (the latter would presumably
be irrelevant),
2. this implies three dierent options
debt holders have a European option to extend the grace period (suppose d
1
= 1/2 by
default when entering the Chapter 11 state) which expires at t
V
B
t
= d if the rm is
still in Chapter 11,
equity holders have an American option to liquidate (Chapter 7), i.e. the V
L
t
(lower)
bound,
12.2 Empirical predictions and results 131
equity holders have an American option to pay arrears to the debt holders and resume
coupon payments, i.e. the V
B
t
(upper) bound.
In this extended model, the upper bound related to the Chapter 11 region is not constant (at least
one must argue why it should be). Thus, the entry and exit of Chapter 11 do not have the same
boundary in the x-dimension. In fact, this is what one should expect from real options analysis
because there are no entry costs related to Chapter 11, whereas the equity holders must pay exit
costs (S

V
H

V
H

) when the rm exits Chapter 11 and enters the healthy state at time

V
H
. Furthermore, Broadie, Chernov, and Sundaresan (2007) do not consider a restructuring of
the rms capital. Clearly, the incentives to restructure in this model are not particularly dierent
from those in chapter 11 and, hence, there is scope for debt renegotiation in this model as well.
The key results of the Broadie, Chernov, and Sundaresan (2007) model are that it is important
who is in control of the bankruptcy triggering rights. This result is obvious, and the authors
also shows that if equity holders control the decision, they appropriate most of the benets of
Chapter 11. The debt holders can restore the (pseudo) rst-best outcome by having control over
Chapter 11 or by having the liquidation (Chapter 7) control rights. The authors consider Leland
(1994b) as their benchmark models and show that their model generate higher probabilities of
(Chapter 11) default, i.e. higher credit spreads, and less liquidation. One could argue that a more
appropriate benchmark choice is e.g. Fran cois and Morellec (2004) because they also distinguish
between default and liquidation.
12.2 Empirical predictions and results
highly incomplete
Tests of trade-o theory, pecking order theory, pricing of corporate debt and stocks (related to
credit risk as well as asset pricing), agency costs.
Chapter 13
Other issues on the capital structure
of a rm
The previous Chapter 11 reviews structural models which as (one of) their main purposes have tried
to model the tax advantage to debt. Obviously there exists a number of other issues which have an
inuence on the capital structure of a rm. For instance, it was earlier discussed in connection to
the model in Christensen, Flor, Lando, and Miltersen (2002a) that dierent methods for specifying
the bargaining game, when the debt is restructured, are important for the optimal choice of the
coupon rate and the restructuring policy.
In this section I turn to some other examples of capital structure issues which have been ad-
dressed in the literature. The following Sections 13.113.3 continue with alternative ways of rene-
gotiating the debt. Section 12.1 addresses the problem of under which conditions to liquidate the
rm following a chapter 11 period.
1
Section 13.4 considers an extension where the asset value
uctuates. At the end, I consider debt with nite maturity in Section 13.5.
13.1 Strategic default
For a rm with low EBIT the model in Christensen, Flor, Lando, and Miltersen (2002a) only allows
the current equity holders to propose a complete debt restructuring. The equity holders will take a
renegotiation game into consideration, and they will oer the current debt holders just enough to
1
Here, chapter 11 refers to the U.S. Bankruptcy Code. A chapter 11 bankruptcy allows the current management
to run the rm while a debt restructuring is renegotiated. More about chapter 11 can be found in Gertner and
Scharfstein (1991). For chapter 7, see footnote 12.
133
134 Other issues on the capital structure of a rm
have the proposal accepted. As an outcome the capital structure of the rm is optimally renewed.
A dierent debt renegotiation procedure is proposed by Anderson and Sundaresan (1996) and
Mella-Barral and Perraudin (1997). Here the equity holders try to obtain concessions from the
current debt holders due to the costs associated with a formal default. As a result the equity
holders will gradually pay lower coupons as the EBIT declines. Often this renegotiation procedure
is termed as write downs or strategic debt service. In the following I review the model by Mella-
Barral and Perraudin (1997).
13.1.1 The model
Let the state variable be denoted p and interpreted as the price of the good which the rm produces.
Let w be the production costs per unit of good and time. Thus, the net earnings ow to the rm is
p w. Suppose that w is constant and that the price follows a geometric Brownian motion under
the risk neutral measure
dp
t
= p
t
dt +p
t
dW
t
.
Note that in contrast to the model in Christensen, Flor, Lando, and Miltersen (2002a) and Gold-
stein, Ju, and Leland (2001) EBIT does not follow a geometric Brownian motion unless w = 0.
In Mella-Barral and Perraudin (1997) bankruptcy is costly because the post-bankrupt price is
reduced by a factor
1
[0, 1] and the production cost is increased by a factor
0
1. Hence, the
post bankrupt rm receives the net earnings
1
p
t

0
w. In addition, the current equity holders may
at any point in time liquidate the rm and in return receive the constant value K.
2
Furthermore,
assume that there are no taxes in the model.
An important assumption is, similar to Leland (1994b) and others, that it is assumed that
the contingent claims considered are independent of the time itself. The pde for pricing therefore
becomes an ode and because the production cost is constant, it is clear that the ode has a solution
similar to the one given in (10.18). Suppose that the rm has no debt and denote the unlevered
rm as V
E
U
if the rm is held by the initial equity holders and as V
D
U
if the rm is held by other
2
In Mella-Barral and Perraudin (1997) the net earnings of the post bankrupt rm are denoted
1
p
t
x
0
w and
the liquidation value is . Since I have used these symbols elsewhere with a dierent interpretation, I have decided
to change the original notation. Later I review the model presented in Flor (2008) in which the liquidation value K
is stochastic.
13.1 Strategic default 135
owners after a default has occurred.
3
Then
V
E
U
(p) =
p
r

w
r
+
_
K
p
c
r
+
w
r
__
p
p
c
_

2
, p p
c
(13.1)
where
p
c
=

2

2
1
w +rK
r
(r ), (13.2)
and
V
D
U
(p) =

1
p
r


0
w
r
+
_
K

1
p
c
r
+

0
w
r
__
p
p
x
_

2
, p p
x
, (13.3)
where
p
x
=

2

2
1

0
w +rK

1
r
(r ). (13.4)
Here,
2
is similar to (10.20) in Chapter 11. The liquidation boundaries p
c
and p
x
have been found
using the smooth pasting condition.
Consider now the case where the rm has issued perpetual debt which receives the coupon rate
C as in Leland (1994b). The principal of the debt is
C
r
. Let the equity value be E and the debt
value D. The equity holders will cease to pay the coupon rate when the price of the good reaches
the level p
b
, a level determined by the smooth pasting condition. At the bankruptcy point
E(p
b
) = max0, V
E
U
(p
b
)
C
r

D(p
b
) = minV
D
U
(p
b
),
C
r
,
so the values of the equity and the debt are
E(p) =
p
r

w +C
r

_
p
b
r

w +C
r
__
p
p
b
_

2
D(p) =
C
r
+
_
V
D
U
(p
b
)
C
r
__
p
p
b
_

2
,
where
p
b
=

2

2
1
w +C
r
(r ), (13.5)
3
One could imagine that the rm initially had debt but after a default the debt holders take over and have a pure
equity rm. Note that if taxes are present in the model, one knows from the previous discussion that one must be
careful with interpreting claims as unlevered values.
136 Other issues on the capital structure of a rm
if the debt is risky, i.e. if
C
r
K. Note that the bankruptcy level in (13.5) has been derived under
the condition that the value of the equity is in fact 0.
4
If the debt is risk free, then D(p) =
C
r
and E(p) = V
E
U
(p)
C
r
. Further note, that the risky debt (and equity) value can be interpreted in
the usual way, as in Chapter 11:
C
r
is the value of receiving the coupon rate forever. At default,
future coupons are lost but the debt holders take over the rm. Finally, the default value must be
discounted. Adding equity and debt, the total rm value with risky debt is
v(p) = E(p) +D(p)
=
p
r

w
r

_
p
b
r

w
r
V
D
U
(p
b
)
__
p
p
b
_

2
, (13.6)
and with risk free debt the rm value is v(p) = V
E
U
(p).
In this model one is interested in studying the ineciencies of introducing debt. There is a
cost of debt inasmuch as the net earnings to the post bankrupt rm are lower than for the pre-
bankrupt rm. Mella-Barral and Perraudin (1997) investigate the ineciency of debt by looking at
the levered rm value for a given coupon rate. As noted, C < rK yields risk free debt. So suppose
the debt is risky, hence C > rK. At p = p
b
the equity holders will default and hence the debt
holders will take over the rm. The post bankrupt rm will be liquidated when p = p
x
. However,
from (13.4) and (13.5) one nds that p
b
< p
x
if C <
(
0

1
)w+rK

1
. Hence, the levered rm value
can be illustrated as in Figure 13.1.
For low coupon rates, C < rK, the debt is risk free and the rm value is equal to the all-equity
rm value. For slightly higher coupon rates the debt holders will immediately liquidate the rm
in case of a default because p
b
< p
x
. This is termed liquidation bankruptcy. With liquidation
bankruptcy the debt causes an ineciency because the rm is liquidated too early in the sense that
p
b
> p
c
. The debt holders liquidate the rm earlier because the postdefault earnings are lower due
to
0
> 1 and
1
< 1. For higher coupon rates p
b
> p
x
, hence following a default the debt holders
will continue operating the rm until p = p
x
. In this case there is operating concern bankruptcy.
With operating concern bankruptcy the ineciency stems from the decrease in the net earnings
and because the rm is liquidated at p
x
rather than p
c
.
4
p
b
has been derived with the condition
dE(p)
dp

p=p
b
= 0. If the equity holders receive V
E
U
(p
b
)
C
r
at default, the
smooth pasting condition is
dE(p)
dp

p=p
b
=
dV
E
U
(p)
dp

p=p
b
.
13.1 Strategic default 137
0.5 1 1.5 2
21
22
23
24
25
Levered rm value
V
D
U
risk free
operating concern
C
Figure 13.1: The levered rm value in the Mella-Barral and Perraudin (1997) model when there
is no renegotiation. In the example, w = 0.15, = 0.1, = 0.2, r = 0.05,
1
= 0.9,
0
= 1.1,
K = 15, and p = 1. When C [0.75, 0.87] there is liquidation bankruptcy; with these parameters
C = 0.87 yields p
b
= p
x
.
13.1.2 Debt concessions
Suppose now that the equity holders can make take-it-or-leave-it oers to the debt holders con-
cerning the coupon rate. Thus instead of always paying the coupon rate C, the equity holders pay
according to a piecewise right-continuous debt service function s(p).
The debt holders have the right to declare bankruptcy if s(p) < C. For a high price, p, the
equity holders will service the debt as stated in the contract, i.e. s(p) = C. If s(p) < C and
the debt holders declare bankruptcy, then the debt holders obtain the value of the all equity post
bankrupt rm, V
D
U
(p). Due to the loss in the net earnings the debt holders will not necessarily
declare the rm bankrupt if the oered coupon is not too low. The equity holders take the debt
holders alternative value into account when oering s(p). Mella-Barral and Perraudin (1997) show
that the optimal debt service function is
s(p) =
_

_
rK if p [p

c
, p
x
),

1
p
0
w if p [p
x
, p
s
),
C if p [p
s
, ),
(13.7)
where p

c
is the price level which makes the equity holders liquidate the rm. For p < p
x
the
best alternative for the debt holders is to liquidate the rm, and for p > p
x
the debt holders will
continue operation. Finally, p
s
is the minimum price level at which the equity holders will pay the
contracted coupon rate.
138 Other issues on the capital structure of a rm
With the debt service function from (13.7) one can derive the price for the equity and the debt.
Note that it is the alternative of the debt holders to accept the lower coupon rate that determines
the trigger levels p
c
and p
s
. For risky debt the values are
E(p) = V
E
U
(p) D(p) (13.8)
D(p) =
_

_
V
D
U
(p) , p p
s
C
r
+ [V
D
U
(p
s
)
C
r
]
_
p
p
s
_

2
, p > p
s
. (13.9)
It is worth noting that there is no ineciency associated with debt when this type of debt
renegotiation is possible. The reason is that the equity holders are now always the residual claimants
whence it follows that the rm will be liquidated at the optimal price level. As a result p

c
= p
c
,
and from (13.8) and (13.9) one notes that the total levered rm value is equal to the unlevered rm
value.
For risky debt Mella-Barral and Perraudin (1997) show that p
c
< p
s
, i.e. there do in fact exist
price levels at which the debt holders will be paid less than the contracted coupon rated. From
(13.7) one further notes that
lim
pp
x
s(p) lim
pp
x
s(p) < 0,
lim
pp
s
s(p) lim
pp
s
s(p) > 0.
Thus, for prices getting slightly lower than p
s
the equity holders will decrease the coupon rate.
However, for prices getting slightly lower than p
x
the equity holders will increase the debt service.
Mella-Barral and Perraudin (1997) note that for p (p
x
, p
s
) there exist parameter values at which
the debt holders will actually pay money to the equity holders to avoid a costly liquidation.
It is also possible to let the debt holders make take-it-or-leave-it oers. The debt holders
must take into account that the equity holders have limited liability so the equity value cannot
be negative. Mella-Barral and Perraudin (1997) show that in this case the debt service function
becomes
S(p)
debt
=
_

_
p w , p [p
c
, p
b
[
C , p [p
b
, [
.
It is interesting to note that for a price just higher than the liquidation price level the debt service
is lower, when the debt holders have all the bargaining power. This observation conrms that it is
the value of a claim that is important and not the current payment per se.
13.2 First best liquidation through renegotiations 139
To study how much more risky the debt becomes when there is renegotiation, Mella-Barral and
Perraudin (1997) introduce the risk premium
R =
D(p) D(p)
neg
C
r
D(p)
neg
.
R is the additional value of negotiable debt relative to the additional value of default free debt. The
risk premium due to renegotiation is strongly decreasing in the volatility. This happens because
a higher volatility generally decreases the debt value, and thus the equity holders cannot lower
the coupon payments as much. Furthermore, a higher volatility increases the value of the post
bankrupt rm, i.e. the debt holders will have a higher alternative value if a bankruptcy is enforced.
This also works against the equity holders wish to cut the actually paid coupon rates. The authors
note that the introduction of renegotiation may therefore partly explain why high quality corporate
bonds (low volatility) can have a relatively large default premia; something which has not been
captured by the classical structural models.
The above analysis shows that it is possible to obtain the rst best solution, namely liquidation
at p
c
, if the debt can be renegotiated. In the model the debt can be costly because a bankruptcy
decreases the net earnings of the rm. Unfortunately, Mella-Barral and Perraudin (1997) have not
considered what happens if there is also an advantage to debt. The authors mention that it is easy
to incorporate a tax advantage to debt. Another cost could be costly renegotiation. I nd that it
would be interesting to see an extension such that an optimal capital structure can be analyzed. In
such a case the volatility eect mentioned above may not be so obvious because the chosen coupon
rate will depend on the volatility. Moreover, the liquidation point triggered by an all-equity rm
need no longer be the optimal ex ante choice.
13.2 First best liquidation through renegotiations
Empirical evidence shows that the absolute priority rule is not always respected. The equity holders
occasionally manage to get some value out of a default even though the creditors have not been
paid all their claims.
5
One way to take this fact into account is to simply accept it and then subtract a fraction of the
value which the creditors can receive in default; this is done in Longsta and Schwartz (1995) and
5
Evidence on deviations from the absolute priority rule can be found in Franks and Torous (1989) and Eberhart,
Moore, and Roenfeldt (1990). The latter paper reports that the deviation has an average of 9.87%.
140 Other issues on the capital structure of a rm
Leland (1994b).
6
Albeit this modeling method is easy to implement, it is not quite satisfactory because it does not
explain why the equity holders get some value. The purpose of Mella-Barral (1999) is to show that
liquidation occurs at the rst best liquidation level if the debt can be renegotiated. As a consequence
this explains why and when the debt holders accept either a deviation from the absolute priority
rule at liquidation or a decrease in the agreed coupon payments. Thus, Mella-Barral (1999) is also
related to the debt renegotiation literature.
13.2.1 The model
Suppose that there is a single state variable, x, and that the state variable follows a geometric
Brownian motion
dx
t
= x
t
dt +x
t
dW
t
, (13.10)
which is specied under the risk neutral measure.
7
As in Chapter 11, x is in this section interpreted
as the earnings of the rm.
The rm has physical assets in place. Besides the assets the rm has human capital from the
incumbent entrepreneur and this combination results in an income ow to the rm which may be
positive or negative. Let the perpetual value of this income ow be denoted (x
t
) at time t. Then
(x
t
) = E
Q
t
_ _

t
e
r(st)
x
s
ds

=
x
t
r
.
8
Note that the value (x
t
) is related to the incumbent entrepreneurs human capital. If, say,
a competitor has access to the same physical assets, he could generate a value of

(x
t
). The
incumbent entrepreneur can therefore sell the physical assets to the competitor for

(x
t
) at time
t.
To simplify the analysis Mella-Barral (1999) assumes that the rm cannot be sold in fractions,
or rather that it is not optimal to sell the rm in fractions, and that the decision to sell is an
irreversible decision. Therefore the incumbent will not sell the assets just when (x
t
) =

(x
t
),
but he will take the liquidation option into account.
6
Suppose the default value at x is (x). After bankruptcy costs the debt holders should get (1 )(x) according
to the absolute priority rule; a violation of the absolute priority rule can then be modelled as a fraction which goes
to equity holders, and thus the debt holders only receive (1 )(1 )(x).
7
Mella-Barral (1999) assumes that agents are risk neutral and that (13.10) is a general diusion process. As
elsewhere in the survey I have used a dierent notation than the one in the original paper.
8
Albeit there are no taxes, the interest rate used for discounting is still denoted by r.
13.2 First best liquidation through renegotiations 141
The point in Mella-Barral (1999) is to study when the rm will be liquidated. The liquidation
level of x can be determined from rst best liquidation, i.e. the liquidation level is written in the
debt contract. Otherwise, the liquidation level is derived by second best optimization, i.e. it is the
equity holders incentives that determine liquidation.
13.2.2 First best valuation
Suppose that the rm has no debt and that the liquidation sale occurs at time
y

= infs t[x
s
=
y, i.e. when x
t
hits the level y. Before the rm is liquidated one can write the rm value in a
Black and Cox (1976) or Section 11 style
V (x
t
[y) = (x
t
) +
_

(y) (y)

(x
t
, y), (13.11)
where
(x
t
, y) =
_

t
e
r(
y
t)
f
t
(
y
)d
y
.
Here, f
t
(
y
) is the density of
y
at time t. Note that (x
t
, y) can be interpreted as the (generalized)
ArrowDebreu price.
The rst best solution of the liquidation level y = x maximizes the rm value when the liqui-
dation level can be written in the debt contract. Thus the liquidation level solves the rst order
condition
V (x[x)
x
= 0, (13.12)
which in particular holds for x = x.
The option value of the decision to trigger liquidation at y is
_

(y) (y)

(x
t
, y). Mella-
Barral (1999) assumes that this option is strictly concave in y and that there exists a maximizing
level x < x
0
. This implies that there exists a unique solution to (13.12) which is less than x
0
.
13.2.3 Valuation with debt and no renegotiation
Suppose that the rm issues perpetual debt as e.g. in Leland (1994b) with coupon rate C and
principal
C
r
. However, there is no tax advantage of the debt. Mella-Barral (1999) argues that the
initial entrepreneur needs funding to nance the investment in the physical assets.
9
Hence, if the
rm is never liquidated, the equity holders have a claim worth (x)
C
r
.
9
Unfortunately, the need for an initial investment is not explicitly modelled in Mella-Barral (1999).
142 Other issues on the capital structure of a rm
Moreover, the debt contracts specify that if the equity holders repudiate the debt contract,
then the debt holders can enforce a liquidation sale and get their collateral, denoted C(x). If the
proceeds from this sale,

(x), must be split in accordance with the absolute priority rule, then
C(x) = min

(x), P, where P is the par value of the debt.

The equity holders continuously decide whether to pay the coupon or to default. As in Leland
(1994b), the model in Christensen, Flor, Lando, and Miltersen (2002a), etc. the default level is
computed via the smooth pasting condition. Denote the default level x
eh
. Then equity and debt,
when there is No Renegotiation (NR), can be valued similar to (13.11) as respectively
E
NR
(x
t
[x
eh
) = (x
t
)
C
r
+
_

(x
eh
) C(x
eh
)
_
(x
eh
)
C
r
__
(x
t
[x
eh
) (13.13)
D
NR
(x
t
[x
eh
) =
C
r
+
_
C(x
eh
)
C
r
_
(x
t
[x
eh
), (13.14)
where the rst order condition is
E
NR
(x[x
eh
)
x
eh

x=x
eh
= 0. (13.15)
As above, the interpretation of the equity and debt values are straightforward. The rst two terms
in the equity value in (13.13) are the values if there is never a default. At default the equity holders
receive the new rm value, but rst the debt holders must receive their collateral. Furthermore,
the equity holders lose the old rm value, but they also do not have to pay the coupon rate after
a default. Multiplying the default value with the Arrow-Debreu price yields the value today of a
default. The interpretation for the debt value in (13.13) is similar. If there never is a default, the
debt holders always receive the coupon rate C. At default the future coupons are lost but the debt
holders receive the collateral. Finally, the default value must be properly discounted.
The default term in (13.13) is also equal to the value of the equity holders option to default.
As above, Mella-Barral (1999) assumes that the option value is strictly concave and maximized at
x
eh
< x
0
. This implies existence and uniqueness of the default level.
By comparing the two dierent liquidation triggering levels from (13.12) and (13.15) Mella-
Barral (1999) obtains the following result.
Result 13.2.1 The equity holder liquidation level, x
eh
, is increasing in the coupon rate. Given that
the debt holders receive C(x) at default, there exists a unique coupon rate, C

, such that x
eh
= x.
The threshold coupon rate is
C

= r
_
C(x) +
(

(x) (x))
dC
dx
d(x)
dx

d

(x)
dx
_
, (13.16)
13.2 First best liquidation through renegotiations 143
where x is the rst best liquidation level.
From Result 13.2.1 one gets that there are three possibilities:
1. C < C

. The equity holders trigger liquidation at a lower level than the rst best (if C

> 0),
i.e. x
eh
< x,
2. C = C

. Liquidation occurs at the rst best level even though it is the equity holders who
trigger liquidation,
3. C > C

. The coupon rate is too high compared to the rst best coupon rate, so the equity
holders trigger liquidation too early, i.e. x
eh
> x.
By denition, the rst best liquidation level generates the maximum rm value at time zero, i.e.
V (x
t
[x) > E(x
t
[y) +D(x
t
[y), y ,= x.
Only by chance will it be the case that x
eh
= x. Hence, it is natural to investigate if it is possible
to renegotiate the debt such that the rst best solution is achieved. Mella-Barral (1999) considers
this problem subject to the size of the coupon rate.
13.2.4 Debt renegotiation
First of all, note that the renegotiation surplus in this model is the dierence between the ex
ante optimal value and the value obtained when the debtors choose the liquidation point, i.e.
V (x
t
[x)
_
E(x
t
[x
eh
) +D(x
t
[x
eh
)
_
.
Suppose that C > C

, which yields a too early liquidation. Hence, there is scope for lowering the
coupon rate and Mella-Barral (1999) terms this as a deferring concession. In contrast, if C < C

,
then liquidation is triggered too late. Therefore the equity holders have an incentive to ask the
debt holders to make an inducive concession, i.e. to bribe the equity holders into a liquidation of
the rm when x = x. To focus on the renegotiation, assume that liquidation does not occur due to
liquidity problems. Furthermore, the renegotiation procedure is costless.
Another important aspect of the renegotiation game is the relative bargaining power between
the equity holders and the debt holders. For simplicity Mella-Barral (1999) considers two cases.
Either the debt holders can propose the renegotiation oer or the equity holders can propose the
oer. The former is termed a self-imposed concession, the latter is termed a forced concession.
Thus, there are four cases altogether as depicted in Table 13.1.
144 Other issues on the capital structure of a rm
Oer initiator: C < C

C > C

Debt holders self-imposed inducive concession self-imposed deferring concessions

Equity holders forced inducive concession forced deferring concessions
Table 13.1: Possible renegotiations in Mella-Barral (1999). The renegotiation depends on the size
of the coupon, C, and if it is the debt or equity holders who initiates take-it-or-leave-it oers.
Selfimposed deferring concessions. Let the initial coupon rate be C
0
> C

so x < x
eh
0
and
the debt holders have an interest in reducing the coupon rate just before liquidation at x
eh
0
.
10
Naturally, the debt holders wish to make as small concessions as possible. Since the renegotiation
is costless, the debt holders oer to marginally write down the coupon from C
0
to C
1
, which in turn
marginally reduces the liquidation triggering level derived by the equity holders to x
eh
1
. If the state
variable increases, the latest agreed coupon rate, C
1
, prevails until the state variable falls to x
eh
1
. If
the state variable decreases to x
eh
1
, a new coupon reduction occurs and C = C
2
, etc. Note that this
renegotiation is dierent from the one in Mella-Barral and Perraudin (1997) but somewhat similar
to the model in Christensen, Flor, Lando, and Miltersen (2002a). In Mella-Barral and Perraudin
(1997) the coupon reduction is only temporary, cf. the service ow function (13.7). In the model
in Christensen, Flor, Lando, and Miltersen (2002a) the new coupon rate is also xed until a new
renegotiation, however, due to a tax advantage of debt the equity holders may wish to call the debt
in order to increase the coupon rate.
Eventually, the coupon rate will be reduced to the rst best level, C
n
= C

, and hence x
eh
n
= x.
When the state variable falls to this level, the debt holders will no longer have an incentive to
decrease the coupon and the rm will be liquidated.
In this setup equity and debt claims can be easily derived using the same pricing techniques as
above. Between the n
th
and the n + 1
st
renegotiation the values are
E(x
t
[x
eh
n
) = (x
t
)
C
n
r
+
_

(x
eh
n
) C(x
eh
n
)
_
(x
eh
n
) +
C
n
r
__
(x
t
, x
eh
n
)
D(x
t
[x
eh
n
) =
C
n
r
+
_
V (x
eh
n
[x)

(x
eh
n
) +
_
C(x
eh
n
)
C
n
r
__
(x
t
, x
eh
n
).
When x
eh
n
= x, there are no further renegotiations. For any n note that at x
t
= x
eh
n
the value to the
equity holders is

(x
t
) C(x
t
) and the value to the debt holders is V (x
t
[x)
_

(x
t
) C(x
t
)
_
.
If the absolute priority rule is used, one has C(x
eh
n
) = min

(x
eh
n
), P =

(x
eh
n
). Thus the value
10
The subscript in x
eh
0
refers to the number of already completed concessions.
13.2 First best liquidation through renegotiations 145
to the equity holders is 0 and the debt holders get V (x
t
[x). This is because the debt holders are
the ones who have all the bargaining power.
Self-imposed induced concession. An inducive concession occurs if C < C

so liquidation
occurs too late, i.e. x > x
eh
. Clearly, the surplus which can be bargained about at the rst best
liquidation level is

(x) (E(x[x
eh
) +D(x[x
eh
)).
Because the debt holders have all the bargaining power, the debt holders will simply give
the equity holders their current value, which is E(x[x
eh
). The remaining value at liquidation,

(x) E(x[x
eh
), goes to the debt holders. Therefore,
E(x
t
[x) = (x
t
)
C
r
+
_
E(x[x
eh
) (x) +
C
r
_
(x
t
, x)
D(x
t
[x) =
C
r
+
_

(x) E(x[x
eh
)
C
r
_
(x
t
, x).
Note that the debt holders voluntarily choose to bribe the equity holders to liquidate the rm.
Thus, this can be interpreted as an explanation for a violation of the absolute priority rule.
Forced deferring and inducive concessions. Above, the renegotiation has been initiated by
the debt holders. Suppose now that the equity holders have all the bargaining power and can make
take-it-or-leave-it oers to the debt holders. For deferring concessions, the debt holders alternative
is to accept the oer is the collateral. For an inducive concession, the equity holders can threat
to keep paying the low coupon, even when x < x, unless the debt holders accept to just get the
collateral when x = x. Since the valuation of equity and debt can be done as above the valuation
is not included in the survey.
13.2.5 Debt capacity and some remarks
It is clear that the bargaining power between the equity and debt holders matters for the debt
valuation. Therefore, the maximum amount of debt the entrepreneur can get, i.e. the debt capacity,
depends on which situation is prevailing.
11
The more the debt holders have to say, the more they will be willing to lend and hence maximum
debt capacity is obtained when the debt holders can make self-imposed renegotiations. In this case
11
It seems as if Mella-Barral (1999) has in mind that the entrepreneurs have the constraint D(x
0
) I, where I is
the size of the initial investment. He is, however, not explicit about this constraint.
146 Other issues on the capital structure of a rm
there can be a 100% leverage. On the contrary, when the equity holders make forced concessions,
the debt holders only get their collateral, and thus their willingness to lend is limited to this level.
Before I turn to the next paper, let me make some nal comments about the Mella-Barral
(1999) paper. The paper is interesting because it shows that a capital irrelevance result holds
when the debt can be renegotiated, i.e. it is possible to obtain rst best liquidation with but not
without debt renegotiation. This result does not depend on who of the two claimant groups has the
bargaining power. Furthermore, depending on the size of the coupon, the paper gives a rationale
for either a sequence of debt write downs or a violation of the absolute priority rule. An interesting
extension of this paper is to consider multiple classes of debt holders, which is the focus of Hege
and Mella-Barral (2002) but I do not discuss that model in further detail.
As Mella-Barral recognizes, the irrelevance result should hold when the renegotiation is costless.
However, the existence of debt is not obvious from an ex ante point of view because there is no
benet from issuing the debt. The rationale in the paper is that the entrepreneur needs initial
outside debt funding to start the production. But if there are no information complications, the
entrepreneur should just issue equity and, as a result, there will only be rst best liquidation.
12
Otherwise one needs to establish a model with asymmetric information, which in turn could inu-
ence the renegotiation process. The reason for debt in papers like Leland (1994b) and the model in
Christensen, Flor, Lando, and Miltersen (2002a) is that there is a tax advantage to debt. I believe
that it would be interesting to extend the Mella-Barral (1999) paper in this direction to see how
the renegotiation process is aected. Probably the equity holders will like to be able to increase
the coupon, if the prot of the rm increases. Furthermore, a costly debt renegotiation will disturb
the possibility to obtain rst best liquidation.
13.3 Dividend policy and renegotiation
The point in Mella-Barral (1999) is to obtain liquidation at the rst best level. Fan and Sundare-
san (2000) consider renegotiations in a dierent way than Mella-Barral (1999) and the model in
Christensen et al. (2002a) or Section 11.3 but close to Mella-Barral and Perraudin (1997). The Fan
and Sundaresan (2000) paper also introduces a cash ow covenant and discusses how the equity
holders should choose the dividend rate. In the following I briey review the Fan and Sundaresan
12
To me, the case C < C

is a bit hard to explain. If the debt holders can be paid the coupon C

, they will not

have to bribe the equity holders in order to obtain rst best liquidation.
13.3 Dividend policy and renegotiation 147
(2000) model.
As in the Leland (1994b) paper it is the value of the assets of the rm, denoted V , that is the
state variable. Under the risk neutral measure V follows
dV
t
= (r )V
t
dt +V
t
dW
t
. (13.17)
As in the previous sections, r is the after tax risk free interest rate.
13
V is the total cash payout
from the rm to cover interest payments and dividends. If the rm is liquidated, the equity holders
receive nothing and the debt holders take over the rm. As in Section 11 liquidation is costly. Let
be the proportional and l the constant cost factor, respectively. Thus, at default, the debt holders
receive (1 )V
B
l.
There are two ways of renegotiating the debt. Either through a debt-equity swap or through
strategic debt service. An important assumption is that following a debt-equity swap it is not
possible to exploit the tax shield, i.e. the rm becomes and continues as an all-equity rm. This
assumption implies that the equity and debt holders will bargain over the value of the (unlevered)
assets of the rm.
The bargaining is set up a bit dierently than in Section 11.3. For concreteness, let be the
bargaining power of the equity holders and 1 the bargaining power for the debt holders. Let
be the sharing rule, i.e. the fraction of the value at bargain that goes to the equity holders. The
debt holders must at least receive their outcome from a liquidation. Thus [0, +
l
V
S
] where V
S
is the swap level of the value of the assets. Given the bargaining power Fan and Sundaresan (2000)
can solve for the Nash solution of and they nd

= min
_

V
S
+l
V
S
,
_
,
where

= if the costs are so high that the debt holders receive nothing at liquidation.
This setup is essentially as in Black and Cox (1976) so from (10.17) one already has the odes
used for pricing the equity and the debt. Here, the equity holders continuously receive the rate
V (1 )C and the debt holders receive the rate C.
14
For simplicity, assume that there are no
xed costs, i.e. l = 0. Then

= , so when the debt-equity swap is triggered, the equity holders

13
In Fan and Sundaresan (2000) the r in (13.17) is a . The discussion concerning the drift in Section 11.1.2 applies
here as well.
14
Related to the tax structure in Chapter 11, one can think of as
c
and
d
=
i
= 0. This is similar to the tax
structure in Leland (1994b).
148 Other issues on the capital structure of a rm
receive
E(V
S
) = V
S
. (13.18)
Note, that this has a similarity to (11.47) in Section 11.3, where = , i.e. there is bargaining
over the saved bankruptcy costs.
Since the equity holders are the ones who can trigger the swap, the swap level is determined by
the smooth pasting condition related to (13.18). Let p
S
(V ) be the Arrow-Debreu price for receiving
one unit of account when the swap is triggered. Hence, from (10.31), p
S
(V ) = (
V
V
S
)

2
. Thus, from
(10.18) one has
E(V ) = V (1 )
C
r

_
(1 )V
S
(1 )
C
r
_
p
S
(V ),
D(V ) =
C
r
(1 p
S
(V )) + (1 )V
S
p
S
(V ),
and the swap level is
V
S
=

2

2
1
(1 )
C
r
1
1
. (13.19)
With this, Fan and Sundaresan (2000) can relate the swap level to other papers in the literature.
Note that [0, ] because l = 0. Then, if the equity holders have all the bargaining power, i.e.
= 1, this model is essentially like Anderson and Sundaresan (1996) or Anderson, Sundaresan,
and Tychon (1996). The swap level is in this case denoted V
B,AST
=

2

2
1
(1 )
C
r
1
1
which is the
indierence point for the debt holders. In contrast, if the debt holders have all the bargaining power,
the equity holders indierence point is V
B,Leland

=

2

2
1
(1)
C
r
, as in Leland (1994b). Hence, the
actual swap level varies between the two other trigger levels depending on the renegotiation power,
i.e. V
S
= V
B,Leland
+ (1 )V
B,AST
, where =
1

(
1
1

1
1
).
Consider now a debt contract as above but extended with a cash ow covenant. Assume that
it is impossible to renegotiate the covenant. Hence, if the cash ow of the rm becomes lower than
the coupon rate, there is an immediate liquidation. Note that the swap level in (13.19) is increasing
in the liquidation costs. If liquidation is suciently costly so that V
S
> C, then there will never
be a covenant enforced liquidation and vice versa.
Fan and Sundaresan (2000) also consider a second renegotiation procedure along the lines of
Anderson and Sundaresan (1996) and Mella-Barral and Perraudin (1997). For a suciently low
level, the debt will not be serviced as well as written in the debt contract. This is clearly costly for
the debt holders. However, if the performance of the rm improves the debt holders will receive
13.3 Dividend policy and renegotiation 149
their full coupon rate. As long as the debt holders receive less than the agreed coupon rate, the
rm cannot deduct the interest payments from the tax bill. This makes strategic renegotiation
costly for the equity holders, as well. A main dierence between the debt-equity swap case and the
strategic debt servicing case is that the object of bargaining in the former case is the value of the
assets, whereas it is the value of the (initially levered) rm in the latter case. For simplicity, I only
consider the debt-equity swap case in the following.
The optimal dividend policy is determined so that the equity holders choose the payout rate
which maximizes their value. As long as there is no cash ow constraint, i.e. when V C, (13.17)
is changed to
dV
t
= (r )V
t
dt +V
t
dW
t
.
The total payout is divided into coupon payments and dividends, i.e. V = C +(V C), and the
total payout rate satises the restriction (1 )C
1
V
; the equity holders also receive the
tax shield C.
The nding in Fan and Sundaresan (2000) is that if there is no cash ow covenant, then it is
optimal for the equity holders to have the maximum dividend, i.e. = , as long as cash ows can
cover the interest payments. For low asset values, V < C, there are no dividends to the equity
holders. This result is intuitive. There is no reason for the equity holders to save money in the
rm for a default because this will only benet the debt holders. However, if there is a cash ow
covenant, Fan and Sundaresan (2000) obtain that for a suciently low it is optimal for the equity
holders to receive no dividends. With strategic debt service a more interesting result is obtained.
The total payout parameter will be chosen by the equity holders so that the renegotiation level
is exactly equal to the cash ow constraint level. In this way the highest possible dividends are
extracted without entering a covenant enforced liquidation.
The objective in Fan and Sundaresan (2000) is clearly interesting: will the equity holders
voluntarily cut dividends to avoid costs in the future? In my opinion, the paper would have
benetted from using the cash ow to the rm, e.g. EBIT, as the state variable. The EBIT
approach seems to be particularly well suited when a cash ow covenant is discussed. Furthermore,
if the equity holders ex post can change the drift of the state variable, then the modeling technique
may not be convincing because the choice of depends on the level of the state variable, V . In
the Fan and Sundaresan setting, the problem of a state dependent dividend rate can be solved
because they obtain that the optimal control, , is bang-bang. The reason for this type of control is
150 Other issues on the capital structure of a rm
that the dierential equation that is, the left-hand-side in the Hamilton-Jacobi-Bellman equation
is linear in the dividend rate.
15
However, the approach based on the asset value, V does not
explicitly recognize how retained earnings are used. If earnings are used as the state variable this
line of thinking becomes more clear. The key is that the relationship between the current earnings
and the rm or asset value is (in this simple setting) hidden in the initial asset value. One may
therefore wonder if there is consistency between the total cash payout, V , and the initial value of
the assets.
16
13.4 Real assets as an implicit collateral
In Mella-Barral and Perraudin (1997) and Mella-Barral (1999) the alternative to continue the
current production was either to liquidate the rm and in return receive a constant, or to change the
production and continue operation. The change of production in Mella-Barral (1999) is somewhat
special because it is the same variable that drives the earnings from each type of production.
Therefore one may as well interpret the change from to

as a change in management rather

than a production shift.
One may wonder how the restructuring policy is aected if it is possible to sell the assets of
the rm to an outsider who is not interested in continuing the current production. One attempt to
address this matter is done in the model presented in Flor (2008). This paper can also be seen as
an extension of the model in Christensen, Flor, Lando, and Miltersen (2002a) or in Section 11.4.
Now the equity and debt holders do not only have to take the earnings of the rm into account but
also the value of the assets of the rm. The model in Flor (2008) can also be seen as an attempt to
model whether a rm should choose chapter 11 or chapter 7 (U.S. bankruptcy code) corresponding
to the regions I and II mentioned below.
15
In the notation of Fan and Sundaresan (2000), the relevant terms in the LHS of the HJB is (V )(1E
G
V
(V ; ))V ,
hence the optimal dividend rate depends on whether E
G
V
< 1.
16
Indeed, if retained earnings is taken into consideration, it may be needed to introduce a second state variable
e.g. a savings account or the size of the rms assets. In this case, it may be hard to obtain closed-form solutions
for debt and equity values.
13.4 Real assets as an implicit collateral 151
13.4.1 The model
In the following I use the terminology from Section 11.4. The rms assets and current human
capital generate EBIT, x. However, the assets of the rm can also be sold to an outsider. There is
a pool of outsiders, who at any given time are ready to buy the assets. At each instant of time the
value which the rm can receive by liquidating its assets is known and the value is K. Under the
risk neutral measure EBIT and the sales value of the assets follow
dx
t
= x
t
dt +x
t
dW
1
t
, (13.20)
dK
t
= K
t
dt +K
t
dW
2
t
, (13.21)
and the Wiener processes W
1
and W
2
have the correlation coecient .
The rm can issue callable debt to take advantage of the tax shield. Initially, however, it may
be more benecial for the rm owners to immediately sell the assets of the rm to the outsider
with the highest bid.
Since the model has two state variables, the EBIT and the alternative value of the assets, it
is not possible to reduce the pricing pde to an analytically solvable ode. Thus the model must be
solved numerically and to keep matters as simple as possible, the debt is assumed to be perpetual.
The model could, in principle, be extended to debt with nite maturity, however, the computations
would be quite cumbersome and time consuming. With EBIT and the value of the assets as in
(13.20) and (13.21), the price of a contingent claim, denoted F, which continuously receives the
rate Ax +B is
1
2

2
x
2
F
xx
+xF
x
+
1
2

2
K
2
F
KK
+KF
K
+xKF
xK
(1
i
)rF + (Ax +B) = 0. (13.22)
The arguments in F have been omitted and the subscripts correspond to partial derivatives in the
obvious way. To make the pde (13.22) complete, one further needs to state the specic boundary
conditions to the contingent claim.
17
17
For modeling purposes the model in Flor (2008) uses the state variables x and Z

=
K
x
. Note that Z also follows
a geometric Brownian motion. In this case the pde is
1
2

2
x
2
F
xx
+xF
x
+
1
2

2
+
2
2

Z
2
F
ZZ
+

+
2

ZF
Z

xZF
xZ
(1
i
)rF + (Ax +B) = 0.
152 Other issues on the capital structure of a rm
As in Section 11.4 this model involves an iteration procedure which one may think about as the
rm having an a priori given number of restructuring options. Due to the larger complexity of the
model it is assumed that the rm initially only has one restructuring option.
18
When the debt has been issued the equity holders have the right to call the debt. In this case
the debt holders receive the principal, P, and a call premium, P. The principal is such that the
debt is issued at par. Following a call the equity holders become the rm owners. The rm can
now either be sold to the winning outside bidder, or it can be optimally issued as new equity and
debt. Thus the call possibility yields the boundary conditions
E
sell
(x, K; C
1
1
) = K (1 +)P
1
,
D
sell
(x, K; C
1
1
) = (1 +)P
1
,
and
E
new
(x, K; C
1
1
) = E(x, K; C
0
0
) + (1 k)D(x, K; C
0
0
) (1 +)P
1
,
D
new
(x, K; C
1
1
) = (1 +)P
1
.
As regards the coupon rate, the superscript denotes the number of options when the debt was
issued and the subscript denotes the currently remaining number of options. For later use C
1
0
is
the coupon rate issued when the rm had one option to restructure its capital but now the rm
has no options left to restructure its capital.
Although it has already been mentioned I remind the reader about the renegotiation game in
Section 11.4. An extension of this game is illustrated in Figure 13.2.
In this model the extended strategy space will inuence the outcome of the renegotiation. Now,
the equity holders do not only have the possibility to issue new debt but they can instead choose to
sell the assets of the rm. Similarly, if the debt holders reject the proposal and the equity holders
cease to pay the coupon, the debt holders will take over the rm and either change the capital
structure or sell the assets of the rm.
Let G
R
be the gain to renegotiation and E
NR
and D
NR
the No Renegotiation values of the
equity and the debt, respectively. With being the bargaining power of the equity holders one has
18
In principle it is easy to extend the model to more than one option. However, with the current computer power
available it is unlikely that more than a two or three options model could be obtained within reasonable time.
13.4 Real assets as an implicit collateral 153
Ask
Restructure
Liquidate
Continue
Yes
No going concern
going concern
sell
sell
E
E
D
Figure 13.2: The structure of the renegotiation game between the equity holders and the debt
holders in the model in Flor (2003). The restructuring and the liquidation decision may result in
either selling the assets of the rm or issuing new securities.
the conditions (going from one to zero restructuring options)
G
R
(x, K) = max
C
0
0
_
K, E(x, K; C
0
0
) + (1 k)D(x, K; C
0
0
)
_

_
E
NR
(x, K; C
0
1
) +D
NR
(x, K; C
0
1
)
_
,
E(x, K; C
0
0
) = G
R
(x, K) +E
NR
(x, K; C
0
1
),
D(x, K; C
0
0
) = (1 )G
R
(x, K) +D
NR
(x, K; C
0
1
).
The no renegotiation values are similar to those from Section 11.4 written in (11.54) and in (11.55),
but now the possibility to sell the assets of the rm must be taken into account.
13.4.2 Results
Using the numerical analysis one is able to see which restructuring strategy will be the consequence
when the rm is allowed to sell its assets. Furthermore, the impact on the equity and the debt values
can be analyzed. There are four dierent restructuring strategies, denoted I IV, as illustrated in
Table 13.2. The actual appearance of the regions on the boundary is illustrated in Figure 13.3.
The new possibilities compared to the model in Section 11.4 are the regions II and III. It is
quite interesting that even though the equity holders want to sell the assets of the rm, they may
have to go into a renegotiation process with the debt holders (region II). A critical assumption here
is that the new rm owner buys the rm without having any future obligations to the debt holders.
154 Other issues on the capital structure of a rm
New capital Sell
Renegotiate I II
Call IV III
Table 13.2: Possible strategies when the boundary is reached.
1 2 3 4 5
25
50
75
100
125
150
175
200
*
--
--
--
I
II
III
IV
K
x
Figure 13.3: The optimal free boundary with one option. The parameters used are (1
i
)r = 4.5%,

e
= 50%,
i
= 35%, = 0.02, = 0.25, = 0.02, = 0.2, = 0.25, = 25%, = 0.5, k = 2%,
and = 5%. Initially (x
0
, K
0
) = (1, 25), indicated by a in the gure.
13.5 Finite maturity with a simple debt policy 155
The debt holders have already been paid by the equity holders, when the rm is sold.
From the numerical comparative statics analysis done in Flor (2008) it turns out that the initial
rm value minus restructuring costs behaves quite as in the model in Section 11.4. An interesting
dierence is when the volatility of the earnings of the rm is changed. If it is not possible to sell
the assets of the rm, then higher volatility results in a lower initial rm value and higher yield
spreads. If the assets can be sold the rm value increases and the yield spread increases only slowly
in the volatility. A higher earnings risk is benecial for the equity holders because their claim is
related to a call option on the earnings of the rm. If the assets cannot be sold, then the debt
holders will be hurt by higher risk of the earnings because their claim acts as a written put option.
However, in this model the possibility to sell the assets gives the debt holders an implicit collateral
because the equity holders must pay the debt holders before selling the rm. For the parameters
considered in Flor (2008) it is the positive value of the equity that dominates.
The value of the implicit collateral to the debt holders is studied in an example which compares
two rms with the same initial rm value minus restructuring costs. The rms are identical except
that one rm has assets with an outside value of 25 and an earnings drift of 2% and the other rm
has no outside value of its assets and an earnings drift of 2.788% (under the risk neutral measure).
For interpretation one may think of the former rm as a classical industry rm and of the latter
rm as an Internet rm. A renegotiation occurs in the Internet rm when the EBIT falls from
the initial 1 to 0.41. In this case the equity value has had a 84% decrease, and the debt value
has had a 49% decrease. Keeping the ratio of the EBIT and the sale value of the assets xed, the
industry rm will be renegotiated when (x, K) = (0.72, 17.65). Here the equity value has lost 76%
but the debt value has only lost 17%. Thus the possibility to sell the assets of the rm is an implicit
collateral for the debt holders which is quite valuable.
13.5 Finite maturity with a simple debt policy
In Sections 1113.4 it is assumed that the rm issues bonds which never expire albeit corporate
perpetuities are rarely observed on the nancial markets. The main reason for this maturity
assumption is that in single state variable models perpetuities often allow for explicit solutions of
contingent claims such as debt and equity.
A natural question, then, is how important the maturity assumption is when a model allows the
rm to change its capital structure over time. In particular, one is interested in the inuence on the
156 Other issues on the capital structure of a rm
initial rm value, the yield spread, and the restructuring boundaries. One paper which introduces
nite maturity is Leland and Toft (1996). This article takes the Leland (1994b) model as its point
of departure but changes the debt structure of the rm.
It is assumed that the capital of the rm consists of equity and debt. There are multiple debt
contracts outstanding. At each point in time the rm issues a new debt contract. All the bonds
are issued with the same time until maturity, the same promised coupon rate payment, and the
same principal.
As an important consequence the aggregated coupon rate, C, and total principal, P, are both
constant over time. A new bond is issued with time until maturity equal to T, a coupon rate c =
C
T
,
and a principal rate of p =
P
T
. At the same time, outstanding debt is retired at the same rate with
which new debt is issued. Therefore the total debt service payment is constant over time and equal
to C +
P
T
.
The state variable is the value of the assets of the rm, V , which follows a geometric Brownian
motion, as in Leland (1994b).
19
The model has only a corporate tax rate,
c
, and debt interest
payments are fully tax deductible. Hence, the tax benet rate is
c
C. The drawback of debt is
that the equity holders may decide to stop paying the coupons, which forces the rm into a costly
default. Default occurs at the asset level V
B
. Once defaulted, the company becomes a pure equity
rm forever in the sense that the pre-default debt holders take over the whole company and they
cannot issue new debt. To take over the rm, the debt holders must pay a cost proportional to the
default value of the rm, V
B
. The remaining value, (1 )V
B
, is equally divided between all the
debt classes.
For a given level of the asset value which triggers default and with the above mentioned debt
policy, Leland and Toft (1996) are able to derive explicit formulae for the value of a single debt
class, the rm, and for aggregated debt and equity.
The next task is to make the default boundary endogenous, i.e. default happens when the equity
holders decide to stop paying the coupons. This is done by a smooth pasting condition on equity
as in Leland (1994b). It is possible to obtain an explicit expression for the optimal default level
given the coupon rate because at the aggregated level the capital structure remains constant until
default. Finally, the debt is restricted to be issued at par, which yields a necessary constraint on
the coupon rate. The model can now be studied numerically.
19
The earlier discussion about the state variable should be kept in mind, cf. Section 11.1.2.
13.5 Finite maturity with a simple debt policy 157
13.5.1 Results
Notably, Leland and Toft (1996) nd that the initial rm value and leverage are both increasing in
the time until maturity. The paper therefore calls for the conclusion that in terms of initial rm
value it is optimal to issue debt as perpetuities. In this case the Leland and Toft (1996) model is
equivalent to the Leland (1994b) model.
For a given maturity of the debt, Leland and Toft (1996) also discuss potential agency problems.
In particular, the authors pay attention to the asset substitution problem introduced by Jensen
and Meckling (1976). This problem is essentially that the equity holders often have an incentive to
increase the riskiness of the business when the debt is issued.
To analyze this problem, Leland and Toft (1996) consider the sensitivity of equity and debt
with respect to the volatility of the assets for dierent maturities. They nd that the equity and
debt holders will have dierent incentives when the maturity is extended, except when default is
very likely. In fact, for a short maturity neither the equity nor the debt holders wish to increase the
riskiness of the assets of the rm. Contrary to this, with a very long maturity the equity holders
always wish to increase the risk. This nding is interesting because it challenges the Jensen and
Meckling (1976) view of the equity holders willingness to increase the risk. Moreover, this result
related to agency problems may explain why one observes debt with short maturity even though
the model has the result that the optimal maturity is innite.
There are some points in the Leland and Toft (1996) model which deserve further attention.
First of all it seems strange that if the initial rm owners optimize over rm value, then they will
always issue perpetual debt. One would expect that in good times the rm would increase the
coupon rate on newly issued debt in order to increase the tax benets and, similarly, in bad times
that the rm would lower the coupon rate on the new debt to avoid paying default costs exactly as
in Sections 11.311.4. Therefore, I nd that the xed coupon rate on each debt contract is a very
strong assumption which more or less removes the advantage of being able to issue new debt. The
xed coupon assumption is relaxed in the model presented in Flor and Lester (2002) which will be
discussed below in Section 13.6.
The xed debt contract assumption also makes the asset substitution discussion much easier
but less attractive. I nd it fair to argue that if the debt holders observe a change in the volatility
of the assets of the rm, then the debt holders will take this action into account when new debt is
issued. The asset substitution problem is further discussed in Leland (1998).
158 Other issues on the capital structure of a rm
13.6 Finite maturity and dynamic debt policy
A natural response to the Leland and Toft (1996) paper is to allow the rm to include the maturity
decision as a part of the debt policy and more importantly to let the rm choose the optimal coupon
size whenever new debt is issued. The survey now briey review the model presented in Flor and
Lester (2002) which deals with this.
13.6.1 The model
The framework of the model is basically as in Goldstein, Ju, and Leland (2001) and in Section 11.3.
However, the model assumes that it is not possible to renegotiate the debt contract.
Suppose rst that the rm issues non-callable debt with a time until maturity of T years and
a coupon rate C. The principal of the debt, P, is set such that the debt is issued at par, i.e.
D(x
0
; x
0
) = P.
For readability, the coupon rate, C, the time until maturity, T, and the principal, P, are omitted
in the expressions.
If the rm defaults prior to maturity, equity has value zero and the debt holders take over the
rm. The debt holders then optimally restructure the rm. Because of the homogeneity property
discussed in Chapter 11, the conditions at the default boundary, x
t
, are
E(x
t
; x
0
) = 0,
D(x
t
; x
0
) = (1 )d
t
_
E(x
0
; x
0
) + (1 k)D(x
0
; x
0
)

,
where d
t
=
x
t
x
0
, 0 < d
t
< 1.
At maturity, the debt holders must receive the principal. When the principal has been paid,
the equity holders issue new debt. Due to homogeneity, the new debt will be issued with the same
time until maturity as the old debt, and the coupon and principal will be scaled up by the same
factor, u =
x
T
x
0
. Therefore, at maturity,
E(x
T
; x
0
) = u
_
E(x
0
; x
0
) + (1 k)D(x
0
; x
0
)

P,
D(x
T
; x
0
) = P.
Since the debt cannot be called, the capital structure is only changed when debt matures or at
default. Due to the maturity choice, the model is not time-independent. Therefore the equity and
debt prices must be solved by the pde-approach.
13.6 Finite maturity and dynamic debt policy 159
25 50 75 100 125 150
30
40
50
60
70
80
90
v kD
T
E
D
Figure 13.4: The term structure for the equity, debt, and rm value less restructuring costs. For
each time until maturity the optimal coupon is derived and used.
One is interested in nding the optimal coupon and time until maturity, i.e. the optimal (C, T).
The maturity trade-o is illustrated in Figure 13.4. For long maturities the present value of the
future restructuring costs is low. Shorter maturities will increase the present value of the restruc-
turing costs. On the other hand, a shorter maturity allows for matching the coupon rate to the
actual level of the EBIT more often so the tax shield is better exploited. Balancing between these
two eects will determine the optimal maturity. In the example in Figure 13.4 the optimal time
until maturity is 15.8 years. The model in Flor and Lester (2002) nd (numerically), contrary to
Leland and Toft (1996), that it always pays for the initial rm owners to issue the debt with nite
maturity.
Consider now an extension of the model where the debt is callable as in Section 11.3. In
particular, assume that the initial rm owners can decide the call premium rate, denoted . The
default and maturity conditions are as before. The conditions at the call boundary, x
t
, are
E( x
t
; x
0
) =
_
x
t
x
0
_
_
E(x
0
; x
0
) + (1 k)D(x
0
; x
0
)

(1 +)P,
D( x
t
; x
0
) = (1 +)P.
This model can be solved basically as the one without callable debt, however, the complexity
is increased because one also needs to nd the optimal call premium factor. Figure 13.5 shows the
rm value minus restructuring costs for dierent levels of the call premium and time until maturity
with the optimal coupon rate. With callable debt the model nds that the optimal call premium
factor is low. In fact, the call premium rate is so low that it is not optimal to have debt with
nite maturity. With the parameters in Figure 13.5, the optimal call premium factor is 2.8%. The
160 Other issues on the capital structure of a rm
50 100 150 200
82
84
86
88
90
T
= 0.1
= 0.2
= 0.25
= 0.3
=
E + (1 k)D
Figure 13.5: Term structure of rm value minus restructuring costs for dierent values of . The
parameters used are: the initial EBIT level is x
0
= 1. The (before tax) risk free interest rate is
r = 0.07, and the drift of EBIT is = 0.04. The instantaneous standard deviation of EBIT is
= 0.20. In bankruptcy the fraction = 0.25 of the newly restructured rm value is lost. The
restructuring cost fraction for new debt is k = 0.01. The eective tax rate is
e
= 0.40 and the tax
rate on interest payments is
i
= 0.30. The after tax interest rate is (1
i
)r = 0.049. There is full
tax deductability i.e. = 1.
reason for choosing a low call premium factor is that the equity holders now see nite maturity as a
constraint because it is suciently cheap to call the debt when an upward restructuring is wanted,
and the equity holders are willing to pay for this possibility initially.
Chapter 14
Investment and Capital Structure
A signicant amount of decision making within a rm is related to investments. Traditionally,
investments are divided into two parts: nancial investments and real investments. The former
type of investments are concerned with the rms risk management, i.e. asset liability management,
swapping interest rates etc. For non-nancial rms, however, real investments are the key decision
of the rm without any real investments there would be no rm at all. Given real investments
there may be a number of interesting problems, e.g. risk management, design of capital structure
to exploit tax shield and so forth. Of course, the latter type of decisions may inuence future
real investment and, hence, it is dicult perhaps even meaningless to completely separate the
various decisions.
In this chapter we shall focus on how the rms capital structure decisions inuence its invest-
ment policy. In a Modigliani-Miller world with perfect and complete capital markets, full informa-
tion, and a single stakeholder there is no investment problem. In this setting, the rms problem is
uniquely to maximize its market value which is obtained by implementing the projects with highest
attainable net present value. Central for this result is that incentive problems are omitted. If there
are several stakeholders of the rm, then the discrepancy of incentives between the rst-mover de-
cision maker and the remaining stakeholders can easily result in investment ineciencies compared
to the rst best investment policy, albeit the markets are perfect and complete.
Introduction to over- and underinvestment; literature references...Stulz (1990b), Myers (1977),
Jensen and Meckling (1976)...
161
162 Investment and Capital Structure
14.1 Overinvestment and capital structure
So far we have not considered the investment part of the rm, i.e. we have separated the nancing
and investment decisions. Mauer and Sarkar (2005) build a model where the rm undertakes a
project, which is similar to the basic model in chapter 11. The dierence to the usual optimal
capital structure models is that the rm signs a loan commitment. A loan commitment is an
agreement between the borrower (the rm) and the lender (the bank or debt holders) such that
the borrower can obtain a specied amount, K, in a given time interval, and following the receipt
of K, the borrower pays back the loan with an initially specied payment schedule.
The time line of the model is illustrated in Figure 14.1.
0
I

D

A
[ [ [ [
rm signs loan
commetment
Invest in project:
Pay investment I
Receive cash ow P
equity holders
decide to default
bank takes over
project, no new debt
project is
abandoned
by bank (or
equity holders)
Figure 14.1: Time line of the model.
14.2 Underinvestment and the option to expand
Most companies have the possibility to adjust their production to some extent. A natural question
arises concerning the rms capital structure when some of this production exibility is taken into
account. Mauer and Ott (2000) study the eect of a growth option. At the onset the rm has
already assets in place, but the rm has in addition an option to expand its operating activities with
some exogenously specied scale. The analysis of this growth option is very similar to a standard
real options problem analysis. What complicates the matter is the fact that the rm initially has
incentive to issue debt and, hence, the exercise strategy of the option is now based on maximizing
equity holders value and not the rms value (i.e. the joint value of debt and equity).
14.2 Underinvestment and the option to expand 163
14.2.1 The model
The rm produces a commodity (say widgets) using its assets in place. The cost of producing one
unit is w, and the selling price per unit is P which follows the geometric Brownian motion
dP
t
= P
t
(dt +dz
t
), (14.1)
under the physical probability measure. Assuming a constant convenience yield of > 0 yields the
price process under the risk neutral measure as
dP
t
= P
t
(r )dt +P
t
dz
t
, (14.2)
that is r is what is usually denoted as in the notes.
1
As usual the rm has the option to abandon production at a cost equal to zero. Since production
is costly in the present setting, a suciently low price will induce abandonment. On the other hand,
the rm has a growth option which allows the rm to expand its production at a scale q > 1. The
exercise cost (or expansion cost) is a xed constant equal to I. Since a down scale of production is
assumed to be innitely costly the growth decision is de facto irreversible.
The model includes symmetric corporate taxation which provides a rationale for corporate debt.
In case of default, the equity holders receives nothing and the debt holders take over the rm (with a
fractional reduction due to bankruptcy costs); the growth option does not disappear due to default.
However, the debt holders are not allowed to introduce debt into the rm subsequent to a default.
This assumption presumably aects the rms optimal capital structure ex ante, but it may not
blur the key insights of the paper.
Initially, the nancing of the growth option must be undertaken by infusion of cash from equity
holders. Obviously, since equity holders bear the full costs of expansion, but only partly gain by
the additional production the debt holders claim becomes more secure the equity holders are
less willing to expand compared to the case of an unlevered rm. Put dierently, the equity holders
wait for a higher output price before they exercise the option. Thus, this paper deals with an
underinvestment problem. In order to quantify the agency costs of the underinvestment problem,
the paper considers a number of comparative statics (initially without using the optimal coupon)
and extensions to the basic setting.
For a xed coupon payment, the paper rst establishes that equity holders do postpone exer-
cising the growth option relative to the rm value maximizing strategy. In the same analysis, the
1
The unit production cost is denoted C in the paper, and R is the interest payment. We keep the notation used
otherwise in the notes.
164 Investment and Capital Structure
results demonstrate that the default threshold is signicantly decreased when the growth option
has been exercised. This is not surprising as the debt contract is not changed by the growth option
exercise. Eectively, following the growth option exercise the rm is essentially less levered and,
thus, the default threshold is lower. The paper also studies agency costs, where
AC(P) = V
L
F
(P) V
L
S
(P), P (P
D
, maxP
F
1
, P
S
1
),
where V
L
is the levered rm value and the subscript reects whether the growth option is exercised
as rm value maximizing rst-best or equity value maximizing second best.
2
Based on Table
9.4 Mauer and Ott (2000) the agency costs appear low. Indeed, when the the optimal coupon is
taken into account Table 9.5 revels that agency costs are of the size 1-3%. More interestingly is
the observation that the rms capital structure is clearly inuenced by agency costs. For example,
with low volatility ( = 0.10), the coupon decreases from 2.07 in the rst best case to 1.38 in the
second best case. Moreover, the leverage is strongly negatively dependent on volatility in the rst
best case, whereas it is slightly increasing (basically independent) in the second best case.
At the end, the paper considers mechanisms to reducing the agency costs. Of course, since the
rm value is only moderately inuenced by agency problems in this model, one should not expect
the various mechanisms to improve second best rm value dramatically. Based on Myers (1977) the
agency problem is studied when the growth option is partially nanced with additional debt and
when the maturity of the debt can be shortened. Finally, referring to Anderson and Sundaresan
(1996) and Mella-Barral and Perraudin (1997) the eect of strategic debt is considered. In order
to study the eect of partial debt nancing the authors consider four exogenous scenarios (interest
payment equal to 0, 0.5, 1, and 1.5); somewhat surprisingly this is not derived endogenously. The
paper concludes that partially nancing the growth option with new debt can mitigate the under-
investment problem to some extent, because more additional debt nancing seems to decrease the
agency costs. However, since the new and the initial debt contracts are not determined optimally,
one should be careful with the conclusions. Also, the priority of the new debt is something which
should be taken into account initially.
Concerning shortening the debt maturity this is done `a la Leland (1994a) and Leland and Toft
(1996), i.e. the rm continuously retires a fraction m of its outstanding principal P and replaces
it with new debt with the same coupon, principal, priority, and maturity structure; note that
2
Note that the same coupon is used in both cases and that the default strategy is maximizing equity value also in
the rst best case.
14.3 Underinvestment, overinvestment, and nancial exibility 165
the average maturity is 1/m.
3
Introducing shorter maturity opens up for another trade o
mechanism. For a given coupon and principal a lower maturity increases the default probability,
so rm the decreases its debt to overcome this eect. Since the rm becomes less levered the
growth policy becomes closer to the one undertaken by an all equity rm. However, the lower debt
decreases the rms exploitation of the tax shield. It turns out that the latter eect dominates, i.e.
the rm optimally chooses debt with innite maturity.
4
. The authors also discuss the implications
of introducing strategic debt service and they argue that this would not eliminate the equity holders
incentive to underinvest in the growth option. As previously, the argument lies on the fact that
equity holders bears the full cost of investment in the growth option whereas debt holders gain.
Finally, Mauer and Ott (2000) consider the eect of competition. This is modeled such that a
competitive rm enters into the markets according to the jump of a Poisson process with intensity
. Unfortunately, optimal capital structure issues are not studied in detail in this setting, but the
results indicate that agency cost increase as the intensity of entry by a competitor increases.
To summarize the paper, it is shown that equity holders postpone the decision to exercise the
growth option relative to an all equity rm, i.e. underinvest is present in the model. Allowing the
investment to be partially nanced with new debt signicantly decreases the postponement and
agency costs, whereas allowing for debt with short maturity and strategic debt service has only
minor or no eect. If product market competition is taken into account, it seems that this increases
the agency costs of underinvestment.
14.3 Underinvestment, overinvestment, and nancial exibility
Childs, Mauer, and Ott (2005) present a paper with the purpose of studying agency conicts in
a dynamic model which allow for interactions between exible nancing and investment decisions.
As in Mauer and Ott (2000) the rm is initially equipped with a growth option. However, the
modeling of the growth option is more elaborate in the present model. Specically, the assets-in-
place, denoted A, follows the geometric Brownian motion
dA(t) = (
A

A
)A(t)dt +
A
A(t)dZ
A
(t),
3
A caveat with this method has been elucidated by [Decamps and Villenueve] who show that simply applying
smooth pasting to a constant barrier in this setting, as is commonly done, is wrong.
4
See also Flor and Lester (2002) for a discussion of the maturity issues
166 Investment and Capital Structure
and the asset underlying the growth option follows
dG(t) = (
G

G
)G(t)dt +
G
G(t)dZ
G
(t),
where the two underlying Brownian motions have instantaneous correlation Z
A
, Z
G
(t) = t. In
the paper A is interpreted as the present value of the expected after-tax cash ows generated by
the rms assets-in-place. The payout rates
A
and
G
are assumed to be constant and independent
of the rms leverage. Note that the above evolutions are specied under the physical probability
measure. Under the pricing measure the drift rates are going to be r
A
and r
G
, respectively.
In contrast to most other dynamic models, the present model is build on a nite time horizon T.
Moreover, the investment decision is reversible for any t [0, T], although reversion is costly. When
the growth option is exercised the rm must substitute a fraction [0, 1] of its assets-in-place
and additionally pay xed investment cost K
I
. As extreme cases, = 0 implies that the rm has a
pure expansion option, whereas = 1 implies that the assets-in-place are substituted to the asset
underlying the growth option. If the rm desires to reverse to its initial assets-in-place, it can do
so by paying the xed cost K
O
.
Due to corporate taxes at a xed rate and because interest rate payments are tax deductable,
the rm initially decides upon a debt contract with initial principal equal to F
0
0 and maturity
m T. Debt issuance cost is a proportion k of the market value of the newly issued debt. With c
as the xed coupon rate, the coupon payment is cF
0
per unit of time. At maturity, the rm must
pay the principal to the debt holders.
5
The rm has nancial exibility in the sense that previously
issued debt can be replaced with newly issued debt. This is done by letting debt being callable at
par, but the new debt must have the same maturity as the initial debt. This seems as a somewhat
restrictive assumption in a nite time horizon model, but it is partially compensated by allowing
the new debt to have another principal. In the event of default, equity holders entirely give up their
claim and the debt holders take over the rm after having paid bankruptcy cost (a proportional
fraction b of unlevered rm value). Importantly, the post-default rm is not allowed to become
levered. As a simplifying assumption this may be acceptable in order to get a rst order impression
of the agency conicts. However, since the initial debt holders recognize that their claim is worth
less in default (they cannot exploit the tax regime), it is not unthinkable that agency costs can be
signicantly aected by this assumption; in particular if the time horizon is long.
6
5
Note that this setting has some resemblance to Flor and Lester (2002).
6
Finding the correct time horizon in a model like the one discussed here is a somewhat subtle question. Of course,
14.3 Underinvestment, overinvestment, and nancial exibility 167
Before we consider the results, it is worth noting that with the present modeling the equity
holders receive the payo
A
A cF(1 ), if the growth option is not exercised, and the payo
(1 )
A
A + GG cF(1 ) if the growth option is exercised. At the nal point in time, the
equity value must obey terminal conditions depending on the exercise of the growth option
E
O
(G, A, F
Tm
, m, T) = max
_
AF
Tm
+ maxGAK
I
, 0, 0
_
,
E
I
(G, A, F
Tm
, m, T) = max
_
(1 )A+GF
Tm
+ maxAGK
O
, 0, 0
_
.
Thus, it appears as if the equity holders at termination must make a nal decision regrading the
growth option.
In order to study how nancing and investment decisions play together, Childs et al. (2005)
consider two special cases regarding the growth option. The rst case is the asset substitution
case in which the growth option implies more risk. Specically, the assets-in-place satisfy r
A
=
6 2 = 4%,
A
= 15% whereas the growth option satisfy r
G
= 6 5 = 1% and
G
= 40%; the
correlation is = 0.5. At substitution the rm replaces = 75% of its assets-in-place and pays a
cost K
I
= 0.03. Thus, as in Flor (2011), asset substitution implies an irreversible lower expected
growth rate and higher volatility. In this setting the equity holders will exercise the growth option
earlier than what is optimal from a rm value maximizing point of view, i.e. this case studies an
overinvestment problem.
The other growth option case concerns an asset expansion. Here, the assets-in-place and the
growth option have the same parametrization, i.e.
A
=
G
= 6%, so r = 0, and
A
=
G
= 20%.
The correlation between the two Wiener processes is set to = 1 so the model is essentially only
working in a one dimensional setting. The cost of expansion is K
I
= 1.00, and since there is no
gain by down-scaling, the exercise decision is eectively also irreversible.
In both of the above cases the results reveal that in a static debt case (no adjustment of debt
level at maturity dates) the initial leverage is less in the second best case compared to the rst
best case. This changes in the dynamic debt case in which the agency costs are very small. In the
dynamic setting, the rm optimally chooses short debt maturities. In the asset substitution (asset
one can argue that most assets are not expected to last forever, but, on the other hand, when such arguments are
present most rms will either take depreciation into account or expect that new assets are bought in the future.
Thus, single projects may well have nite time horizons, but the life of the rm may not be nite. One can close the
discussion by introducing a truncation date at which the rm is scrapped with some value, but to derive a correct
scrap value at the truncation date may be far from straightforward. For more on truncation in terms of valuation see
for example Christensen and Feltham (2003, chapter 9).
168 Investment and Capital Structure
expansion) case the leverage is higher (lower) in the dynamic case. For the substitution case the
argument is that in the dynamic case the rm can adjust its leverage downward in low rm value
states. More importantly, if the rm cannot adjusts its debt contract, substitution makes the debt
more risky and in the static setting, the rm must take this into account initially and, therefore,
leverage is lower with static debt.
To summarize, the paper nds that when the rm has a exible nancing environment (it is
possible to dynamically adjust the principal of new debt), it is optimal for the rm to undertake
debt with short maturity in order to reduce agency costs. Whether the lower agency costs imply
higher or lower leverage depends on the type of the growth option and eect which must be taken
into account when empirical studies are performed.
Although the model presented seems very exible some comments are in place. One issue
concerns how dynamically the rm can choose its debt, i.e. the nancing exibility. It seems
restrictive that the rm can only adjust the principal at maturity dates; why should the maturity
not be endogenous? In a dierent setting Flor and Lester (2002) show that the rm optimally
chooses the same maturity length and coupon (relative to the level of earnings) when the time
horizon is innite. Thus, the nite time horizon assumption in the present paper is potentially
important for the maturity conclusions.
14.4 Underinvestment and strategic debt service
The debt overhang underinvestment problem put forth by Myers (1977) illustrates that rms may
deviate from the rst best investment policy due to conict of interests between debt holders and
equity holders. Pawlina (2010) sheds further light on the underinvestment issue by including the
possibility of strategic debt renegotiation between debt and equity holders along the lines of Mella-
Barral and Perraudin (1997) and others. In the following, we outline the model and discuss its
implications.
14.4.1 Underinvestment without strategic debt service
The structure of the model is closely related to the models in section 11.1 and section 13.1. Under
the pricing measure, the rms EBIT stemming from assets in place follow
dx
t
= x
t
dt +x
t
dW
t
.
14.4 Underinvestment and strategic debt service 169
In addition to the assets already in place, the rm has one investment option with constant exercise
price I. Subsequent to the investment, the rms earnings increase to x, where > 1 is a priory
known. Furthermore, the rm has an abandonment or liquidation option. Liquidation yields the
value
i
, where i = 1 if the investment option has been exercised and i = 0 is it has not.
The rm issues perpetual debt with tax deductible coupon rate c and the principal is xed
to c/r, where r is the (after personal tax) constant risk free rate of interest. The rm pays
corporate tax rate
c
and bankruptcy is declared according to equity holders incentives. In case of
bankruptcy proceedings are split according to the absolute priority rule, and the creditors take over
the rm. By assumption, the creditors are less ecient rm managers (or there are some fractional
default costs) such that the earnings after bankruptcy decreases to
i1
x, where (0, 1) is the
creditors eciency. In order to avoid immediate liquidation subsequent to bankruptcy assume

i
<
i1
(1
c
)b/r; it follows that debt is risky. Three sources of bankruptcy costs are imposed:
1. creditors ineciency, ,
2. the investment option disappears,
3. the rm cannot issue new debt and, hence, the tax shield is lost.
In the case where the investment has been undertaken it follows from the standard model in
section 11.1 that the values of debt (D
1
), equity (E
1
), creditor managed rm (R
1
), and the rm
are
R
1
(x) = (1
c
)
x
r
+
_

1
(1
c
)
x
lc
1
r
__
x
x
lc
1
_

2
,
D
1
(x) =
b
r

_
b
r
R
1
(x
b
1
)
_
_
x
x
b
1
_

2
,
E
1
(x) = (1
c
)
_
x
r

b
r
_
(1
c
)
_ x
b
1
r

b
r
_
_
x
x
b
1
_

2
,
= (1
c
)
_
x
r

b
r
_
+B
1
x

2
,
V
1
(x) = E
1
(x) +D
1
(x),
until EBIT gets lower than the bankruptcy level x
b
1
the rst time. From that time on, the equity
value is 0, the debt value becomes R
1
(x), and R
1
(x) is equal to the liquidation value
1
when EBIT
has hit the creditors liquidation trigger x
lc
1
. Applying the usual smooth pasting method yields
x
b
1
r
=

2

2
1
c
r
, (14.3)
(1
c
)
x
lc
1
r
=

2

2
1

. (14.4)
170 Investment and Capital Structure
The above bankruptcy trigger assumes that the equity holders obtain zero if the rm is liquidated,
i.e.
1
< b/r. This follows from the assumption
1
< (1
c
)b/r. Naturally, the bankruptcy
level (14.3) is similar to the one in the Leland model in section 11.1.
In order to determine the optimal investment level, x, we need to derive the debt and equity
values prior to investment. The investment level, the bankruptcy level, x
b
0
, and the creditors
liquidation level, x
lc
0
, follow from the standard procedure. The results are
R
0
(x) = (1
c
)
x
r
+
_

0
(1
c
)
x
lc
0
r
__
x
x
lc
0
_

2
,
D
0
(x) =
b
r
+G
(d)
0
x

1
+B
(d)
0
x

2
,
E
0
(x) = (1
c
)
_
x
r

b
r
_
+G
0
x

1
+B
0
x

2
,
V
0
(x) = E
0
(x) +D
0
(x),
where G
0
, B
=
, G
(d)
0
and B
(d)
0
as usual follow from the boundary conditions. Moreover,
x
r
=

1
I (
1

2
)
2
(B
1
B
0
)x

2
(
1
1)( 1)(1
c
)
,
x
b
0
r
=

2

2
1
_
b
r

(
1

2
)G
0
x

1
b
0
1
c
_
, (14.5)
(1
c
)
x
lc
0
r
=

2

2
1

.
Note, that in the rst two triggers must be solved simultaneously. In the debt expression the
term G
(d)
0
x

1
is the present value of the wealth transfer from equity holders to debt holders when
the investment is undertaken. This wealth transfer follows inasmuch as the investment makes the
existing debt less risky. Of course, if the investment resulted in another type of change in the
earnings, then the wealth transfer could be negative. This is, for example, the intuition underlying
the asset substitution problem. The bankruptcy condition (14.5) reveals how bankruptcy depends
on the growth option. In fact, Pawlina (2010) argues that x
b
1
< x
b
0
, and he obtains the result
that risky debt nancing leads to underinvestment in the sense of excessive waiting by the equity
holders, i.e. the investment level x is too high compared to the rst best.
14.4.2 Underinvestment with strategic debt service
The above model simply sets up an underinvestment problem due to debt overhang in a standard
structural stetting. The idea is now to introduce strategic debt service which is done as a solution
14.4 Underinvestment and strategic debt service 171
to a Nash bargaining game. Let V
r
(x) be the value of the rm when debt renegotiation is possible,
and let

i
(x) be the fraction of the rm received by the equity holders. Then Pawlina (2010) nds

i
(x) = arg max

i
(x)
__

i
(x)V
r
i
(x) 0
_

_
(1
i
(x))V
r
i
(x) R
i
(x)
_
1

=
V
r
i
(x) R
i
(x)
V
r
i
(x)
. (14.6)
More generally, we can relate the sharing rule (14.6) to the split of the gain of debt renegotiation
used in section 11.4. Suppose for a moment, that the total rm value is equal to V
r
(x) if debt
renegotiation takes place. In addition, suppose that if debt renegotiation is oered, but rejected,
then the debt holders outside option (rejection) value is R
debt
(x) and the equity holders outside
option value is R
equity
(x). The solution to the Nash bargaining game is

(x) = arg max

(x)
__
(x)V
r
(x) R
equity
(x)

_
(1 (x))V
r
(x) R
debt
(x)
_
1
_
.
Dierentiating w.r.t. and equating to zero yields

(x) =

_
V
r
(x) R
debt
(x)
_
+ (1 )R
equity
(x)
V
r
(x)
. (14.7)
In order to check for a maximum we derive the second order derivative
d
2
d
2
= (1 )[(x)V
r
(x) R
equity
(x)]

[(1 (x))V
r
(x) R
debt
]
(+1)
(V
r
(x))
2

_
[(x)V
r
(x) R
equity
(x)] + [(1 (x))V
r
(x) R
debt
(x)]
_
2
< 0,
and, thus,

(x) in (14.7) solves the maximization problem. Similar to (11.56), we introduce the
gain of debt renegotiation G
R
(x) = V
r
(x) (R
debt
+R
equity
). Thus, the equity holders receive

(x)V
r
(x) =
_
V
r
(x) R
debt
(x)
_
+ (1 )R
equity
(x) = G
R
(x) +R
equity
(x),
which is identical to (11.57) with = .
Returning to (14.6), when strategic debt renegotiation takes place, the equity holders receive
the payo stream (V
r
i
(x) R
i
(x)). The value of the creditors outside option, R
i
(x), is obviously
important for the value received. If the creditors are extremely inecient ( = 0) and there is no
liquidation value (
i
= 0), then R
i
(x) = 0. In contrast, if the creditors option value is equal to the
renegotiation rm value, then the equity holders receive nothing during debt renegotiation. Since
the debt holders only receive the fraction (1

i
(x)) during renegotiation, it follows from (14.6)
that the coupon payments are
b
r
i
=
_

_
(1 )x
i
(1
c
) +r
i
, x (x
l
i
, x
lc
i
[
(1 (1 ))x
i
(1
c
), x (x
lc
i
, x
r
i
]
b, x > x
r
i
.
(14.8)
172 Investment and Capital Structure
In the lowest earnings regime, debt holders receive a weighted average of an eciently operated rm
after tax cash ow and their collateral, r
i
. For EBIT higher than the creditors liquidation level,
but still in renegotiation, the debt holders receive the eciently operated rm after tax cash ow
according to their bargaining power and their eciency. For higher EBIT levels, the equity holders
optimally pay the full contractual coupon rate b. As noted by Pawlina (2010), (
c
= 0, 0, 1)
corresponds to the take-it-or-leave-it renegotiation outcome in Mella-Barral and Perraudin (1997),
whereas
i
= 0 corresponds to Fan and Sundaresan (2000). Moreover, note that b
r
i
is independent
of the growth opportunity per se.
The market values of debt, equity, the rm etc. can be derived in a similar manner as was done
earlier. We only state the debt renegotiation and liquidation levels from Pawlina (2010) in the case
where the investment has been undertaken.
7
Strategic debt renegotiation takes place when EBIT
gets lower than
x
r
1
r
=

2

2
1
b
r
1 (1 )
c
(1 (1 ))(1
c
)
,
and liquidation occurs at
x
l
1
r
=

2

2
1
1
(1
c
)
_

1

b
c
r
_
x
l
1
x
r
1
_

1
_
.
Clearly, the growth option inuences the equity holders decision to trigger strategic debt renegoti-
ation and liquidation. Pawlina shows that liquidation occurs later (i.e. at a lower cash ow level),
when the growth option is present. The conclusion concerning the renegotiation level is less clear
cut because the result depends on the equity holders bargaining power . If equity holders have a
suciently high bargaining power, it is optimal to trigger debt renegotiation earlier (for a higher
EBIT level) when the investment opportunity is present compared to the case in which the growth
option is absent. Thus, the growth option may make the debt holders more exposed to strategic
debt service when the equity holders bargaining power is high.
14.4.3 Results
The key result in Pawlina (2010) is that the possibility to renegotiate the debt exacerbates the
underinvestment problem illustrated in Myers (1977). That is, the equity holders optimal EBIT
level for investment is higher, when debt renegotiation is possible, compared to the case in which
7
The renegotiation and liquidation levels have been found by smooth pasting. Since the renegotiation level can
be crossed freely, one may wonder about a higher contact condition.
14.4 Underinvestment and strategic debt service 173
strategic debt service is not possible. The intuition for this result is that when the equity holders
exercise the investment option, they, of course, have to pay the investment cost, I, but in addition,
the investment reduces the equity holders value of the debt renegotiation option. The latter eect
is due to the fact that the increased cash ow subsequent to investment, x, provides the debt
holders with a more valuable rejection claim in the case of debt renegotiation.
Given the coupon rate b, the results also reveal that more ecient creditors in the sense of a
higher decrease the equity holders value of the renegotiation option and, hence, the (absolute)
dierence between this option value with or without the growth option, respectively, reduces. This
results in lower underinvestment. In a similar vein, more bargaining power to the equity holders
increases their value of the debt renegotiation option, and, hence, higher equity bargaining power
increases the underinvestment problem. Pawlina measures the agency costs of debt as the dierence
between rst best rm value and second best rm value divided with the value of debt. The result is
that the higher the equity holders bargaining power, the higher are the agency costs of debt. This
suggests that it is benecial to reduce equity bargaining power when growth options are available.
8
Again, the higher ex post exibility due to renegotiable debt leads to ex ante costs. A related result
is obtained in Flor (2011) who analyzes asset substitution in a setting in which debt renegotiation
results in an optimal restructuring of the rms capital structure. However, Flor focuses more on
overinvestment than underinvestment.
Turning to the case of optimal capital structure for non-renegotiable debt, Pawlina obtains
the intuitive result that the presence of the investment option increases the optimal principal b

/r
and, thus, the rms debt capacity. However, leverage decreases moderately when a growth option
is introduced. The nding is related to the result in Myers (1977), namely that the rms debt
capacity decreases the more of the total rm value stems from growth options relative to assets
in place. Introducing the debt renegotiation option dampens the ex ante incentives to undertake
debt, i.e. the renegotiation possibility amplies the eect of the growth option in the sense that the
optimal coupon rate (principal) increases less and the decrease in leverage is stronger. This holds
in particular when the equity holders have a high bargaining power.
To summarize, the central points in Pawlinas model are that debt renegotiation exacerbates
the underinvestment (debt overhang) problem and, in addition, that the growth option has a
signicant negative eect on leverage when the renegotiation option is present. Thus, by analyzing
8
Of course, this calls for a model in which the exercise of investment options and equity bargaining power restriction
are endogenized in order to truly analyze the ex ante eects.
174 Investment and Capital Structure
the interdependency between the investment option and the debt renegotiation option, Pawlina
provides one explanation for a lower leverage in the structural models.
14.5 Risk shifting, investments, and agency costs
The analysis in Mao (2003) studies how investments may inuence on the equity holders incentives
to increase the riskiness of the rms cash ow.
The main idea is to collect the issues in Jensen and Meckling (1976) and Myers (1977). Thus,
the risk-shifting or asset substitution problem is combined with the underinvestment problem which
follows from the fact that equity holders (in Myers (1977)) bear the full investment costs but only
receive a fraction of the return.
An isolated view of one of the two theories implies that higher leverage exacerbate the debt
agency problems (agency costs). However, interactions between the two theories may change the
immediate conclusion. The key is if the cash ows from the new investment is positively correlated
with cash ows from assets in place, then more investment may increase the total volatility of the
rms cash ows.
9
14.5.1 Model
Assume that all agents are risk neutral and the risk free rate of interest is zero, r = 0. The time
line of the model is illustrated in Figure 14.2.
1 0 1
[ [ [
equity and
debt with principal
D issued
Invest in project:
buy asset amount k
0
investment I = C
0
k
0
control: k
0
liquidation value =
cash ow = c
pay principal
Figure 14.2: Time line of the Mao (2003) model.
9
In Jensen and Meckling (1976) equity holders want more investment, if there is leverage, in order to increase
volatility of cash ows, whereas in Myers (1977) the equity holders are more reluctant to invest with leverage.
However, Myers (1977, p. 167) has a comment related to point of increasing investment.
14.5 Risk shifting, investments, and agency costs 175
The return (cash ow) from the investment is assumed to follow
c = (k
0
) +(k
0
), (14.9)
where is such that
c > 0, and E() = 0, V() = 1,
i.e. there is a lowest possible noise term
>

= min
_

(k
0
)
(k
0
)

. (14.10)
In addition, the noise term has density f() with support (, ). The above implies that
E(c) = (k
0
) and V(c) =
2
(k
0
). Assume the expected cash ow is increasing and concave in the
invested asset, i.e.

(k
0
) > 0,

(k
0
) < 0 and dene MVI as the marginal volatility of investment,

(k
0
).
Consider rst the case where the rm has no debt. Thus, the value of equity at time t = 0 is
E(k
0
) =
_

_
(k
0
) +(k
0
)
_
f()d C
0
k
0
= (k
0
) C
0
k
0
, (14.11)
and the rst order condition is

(k
0
) C
0
= 0. (14.12)
We now turn to the case where the rm at an earlier point in time has issued debt with principal
P maturing at time t = 1. If cash ows are insucient to fulll the debt obligation, the equity
holders receive no value (due to limited liability). The cash ow cut-o level for default is
h(k
0
) =
D (k
0
)
(k
0
)
.
Note that the principal of debt is already given. In an ex ante consideration of the debt, we of
course have to take the ex post dependence of k
0
into account. The value of equity follows
E(k
0
) =
_

h(k
0
)
_
(k
0
) +(k
0
) D
_
f()d C
0
k
0
,
= ((k
0
) D)(1 F(h(k
0
))) +(k
0
)E[[ > h(k
0
)] (14.13)
and the rst order condition is
E(k
0
)
k
0
=
_

h(k
0
)
_

(k
0
) +

(k
0
)
_
f()d C
0
(14.14)
176 Investment and Capital Structure
Example
In order to get a better understanding of the model, we may consider a special case of the model
presented in Mao (2003). Suppose that (k
0
) = k
1/2
0
, (k
0
) = a + bk
0
(this follows Mao), and
f() =
1

, i.e. is uniformly distributed over [, ].

Results
Mao (2003) establishes
Proposition 2 For rms with MVI> 0 and high unit cost of capital, C
0
, the optimal debt level is
increasing in MVI. If C
0
low, then the optimal debt level decreases in MVI.
In the case of MVI 0, the result is that the optimal debt level decreases with the magnitude of
MVI. For the empirical analysis, underinvestment is measured with a positive measure of growth;
this seems to be founded in the agency literature.
By analyzing data from Compustat (1976-1995), Mao (2003) nds support for her predictions:
For high growth rms there is a positive relationship between leverage and MVI, and for low growth
rms there is negative relationship between leverage and MVI when MVI > 0.
14.6 Liquidation incentives and agency costs with exogenous debt
We have already touched on the issue of optimal liquidation of a rm. This was done in section 13.1,
the model of Mella-Barral and Perraudin (1997), and in section 13.2, the model of Mella-Barral
(1999). In the former model the rm could be liquidated at an a priori known price, K, in the
latter model the liquidation value of the rm was (x
t
) the value a competitor is willing to pay
for the rm when current earnings (or a state variable) is equal to x
t
at time t. In section 13.4 we
considered the Flor (2008) model which is also related to optimal liquidation. Here, the liquidation
value stems from outside entrepreneurs who are willing to buy the rms assets in order to include
them in another production. In particular, the outside asset value is modeled as a K
t
which is
uncertain.
Related to the above literature is Decamps and Faure-Grimaud (2002). They consider the ques-
tion of optimal liquidation in a model where the equity holders or the entrepreneur (the skipper
of a boat) at discrete points in time must decide whether or not to abandon current production.
14.6 Liquidation incentives and agency costs with exogenous debt 177
The liquidation or scrap value of the rms assets is modeled as a geometric Brownian motion.
Flor (2008) follows a similar approach, but Decamps and Faure-Grimaud (2002) do not consider
a positive explanation for debt issued by the rm. Thus, there is no tax (or other debt related
friction) reason to distinguish between the productive value of the rm and the earnings of the
rm. In particular, under a pricing measure Q the productive value of the rm follows
dV
t
= rV
t
dt +
V
V
t
dW
1
(t), V
0
> 0,
and the scrap value follows
dK
t
= K
t
dt +
K
K
t
dW
2
(t), K
0
> 0,
where the two Brownian motions are correlated (usually argued to have positive correlation).
10
An important assumption is that the equity holders only have a limited number of abandonment
options with nite maturity. The rm can only be abandoned at dates 0 < t
1
< . . . < t
n
< T,
where T is the end of the model. This implies the tendency to abandon the rm even if there is
no debt increases: at the maturity of the nal abandonment option the equity holders recognize
that beyond this date they can only obtain value from production per se. However, the rst best
scrapping policy is always implemented as long as the group of equity holders is the unique decision
maker.
Consider the introduction of zero coupon debt with principal M maturing at the end date T.
If the rm is scrapped, the debt becomes immediately due (with a risk free discounted principal).
Clearly, the equity holders incentives to scrap the rm at date t (t
i
)
n
i=1
is inuenced by the
introduction of debt. The equity holders decision to scrap the rm is inuenced by at least two
eects. Firstly, there is a price dierence eect at the nal abandonment date t
n
. Without debt, the
equity holders would essentially be concerned about the dierence between the production value, V ,
and the scrapping value K, but with debt the discounted value of the principal matters; thus, this
eect is compounded when the abandonment decision is considered at an earlier date. Secondly,
at a date t
i
t
n
, there is a price eect because the equity holders recognize that they can buy the
production value at the price V (T) for the expected debt repayment at the nal date. This eect
results in excessive continuation at any date, that is continuation can occur albeit S > V . The
debt contract is basically a put option. If we denote the debt value as P(V ; M, T t
n
) at the last
10
To keep the previous notation I apply K for the scrap value. Originally, the paper uses the notation S for scrap
value.
178 Investment and Capital Structure
abandonment date, the equity holders will continue as long as S V + P(V ; M, T t
n
); relating
to investment problems in real options analysis this also illustrates a value matching condition. At
a previous date, the value matching condition has the same interpretation with the extension that
there is a compound option eect stemming from the fact the the rm may be abandoned in the
future. For example, at date t
n1
, rst best continuation occurs when
S(1 e
(r)(t
n
t
n1
)
)

=

S C
1
(V, S, t
n1
),
whereas second best continuation happens when

S C
1
(V +P(V ; M, T t
n1
), S, t
n1
).
Above, C
1
is the value of a compound option when there is one scrapping date left.
Apparently, the number of possible scrapping dates should inuence the agency costs of debt.
However, Decamps and Faure-Grimaud (2002) show that these costs are independent of the number
of scrapping dates if r , i.e. if the risk free interest rate is lower than the expected growth rate of
the scraping value. The reason is that the two continuation regimes (rst best or second best) only
dier at the nal date and, hence, agency costs are independent of the number of dates. Intuitively,
the scrap value appreciates at a high rate in this case and, hence, there is no reason to scrap before
the last possible scrapping date. This also shows up in the above expression for

S. If r , then

S 0 and, thus, C
1
V > 0

S, i.e. continuation is always optimal in both regimes except
perhaps at the nal scrap date. On the other hand, the agency costs are dependent on the number
of scrapping dates if r > , and the agency costs are not monotonic in the number of dates.
Furthermore, a higher face value increases the agency costs of debt. This is not surprising
inasmuch as there is no benet of debt. However, if we consider two potential scrapping dates
(n = 2) and the total debt burden M is split into two, M = M
1
e
r(Tt
1
)
+ M
2
, i.e. a coupon
eect is introduced, then the agency costs decrease, the rm value increases, and the debt capacity
increases in the rst payment M
1
. One way to see the intuition for this result is the following.
A higher coupon payment, M
1
, increases the probability of default. However, there are no costs
associated with default. Hence, a higher default probability is also a higher probability that the
rm is taken over by the debt holders. This in turn implies that the cause of ineciency is removed
because following default the debt holders becomes equity holders, i.e. default reestablishes the rst
best scrapping strategy.
14.7 Adjusting capital structure and investment dynamically 179
14.7 Adjusting capital structure and investment dynamically
Titman and Tsyplakov (2007)
Chapter 15
Takeovers
incomplete...
Takeovers and corporate control is one of the central issues in corporate nance. Until recently,
most research has been based on discrete-time models with only a few periods. Such models can
be well suited to illustrate important points, however, their timing is almost always exogenous.
One of the reasons why many central corporate nance ideas have not received a larger attention
in the continuous-time framework is presumably that the modeling often becomes very complex.
However, if one works in a perfect (symmetric) information setting, or a very special asymmetric
information setting it is possible to apply standard It o calculus.
Lambrecht and Myers (2007), Margsiri, Mello, and Ruckes (2008)...
181
Chapter 16
Capital Structure and Credit Risk
In this chapter we shall only briey touch upon a very important topic, namely credit risk. The
central issue in credit risk is the pricing of corporate bonds. Of course, since the price of a corporate
bond is the present value of a future promised payment stream, an elaborate pricing of corporate
bonds partly consists of a good interest rate model, but the dominant problem is to gure out
how likely the promised payments are going to be met and, if not, to which extend the payments
are changed (reduced or extended maturity). Due to the risk of default, the market evaluates a
corporate bond with a higher discount rate than a similar bond issued by a government (sovereign
debt).
1
This chapter is primarily based on Lando (2004).
16.1 Credit risk and the simple structural models
So far we have focussed on the price of a corporate bond. When the focus is pricing of default free
bonds, the key input is the term structure of interest rates and, hence, default free bond prices are
typically modeled by a specic interest rate model. Similarly, we could equally well speak of the
yield of a corporate bond.
Consider the model by Merton (1974) introduced in section 10.1. Here, the debt holders are
promised a principal payment equal to P at date T. Suppose the current date is t and that the
bond trades at the price D(t). Then we dene the (continuously compounded) yield y(t, T) as the
1
Conditioning on the government being though of as trustworthy.
183
184 Capital Structure and Credit Risk
Figure 16.1: Credit spreads as a function of time until maturity in the Merton model for various
levels of total rm value, V . r = 5%, P = 100, = 0.2
rate which equates the market value with the present value calculated on the basis of the yield, i.e.
D(t, T) = expy(t, T)(T t)P,
whence
y(t, T)

=
1
T t
log
P
D(t, T)
(16.1)
It is important to remember, that the yield dened in 16.1 is in fact based on a promised payment.
This should not be misinterpreted as the expected rate of return of the bond. As we know, under
the risk neutral pricing measure the expected rate of return is always equal to the risk free rate of
interest, denoted r. Due to the risk of default, i.e. credit risk, the bond will typically trade at prices
such that the yield is higher than the (locally) risk free rate of return. The dierence (spread=s)
between the yield of the corporate bond and a corresponding treasury bond is denoted as the yield
spread or the credit spread, i.e.
s(t, T)

= y(t, T) r. (16.2)
Often, we will write s(0, T) = S(T). The risk structure of interest rates is obtained by viewing
s(T) as a function of (time until) maturity T.
16.1.1 Short term yield spreads in diusion models
In the preceding chapters we have considered a variety of structural models. These models are based
on some assumptions about the basics (fundamentals) of the rm such as earnings, investments etc.
16.1 Credit risk and the simple structural models 185
However, as seen in the Merton (1974) model the credit spread becomes very small when the time
until maturity becomes small. In fact, the credit spread converges to zero. In order to establish
this result, we rst write up a lemma.
Lemma 16.1.1 Let B be a Brownian motion under a probability measure Q. Then
> 0 : lim
h0
Q([B
t+h
B
t
[ )
h
= 0. (16.3)
Proof. Let > 0 be given and introduce the random variable X N(0, 1). Since B
t+h
B
t

N(0, h) we have
Q([B
t+h
B
t
[ )
h
=
Q([h
1/2
X[ )
h
=
Q([X[ h
1/2
)
h
=
Q
_
X (, h
1/2
] [h
1/2
, )
_
h
=
N(h
1/2
) + (1 N(h
1/2
)))
h
=
2(1 N(h
1/2
))
h
.
It is clear that the numerator and the denominator both converges to 0 as h 0. Applying
lHospitals rule yields
d
dh
2(1 N(h
1/2
))
d
dh
h
=
2n(h
1/2
)(1/2)h
3/2
1
= n(h
1/2
)h
3/2
=

2

exp
1
2

2
h
1
h
3/2
. (16.4)
This leaves us with an exponential factor which converges to 0 and a polynomial factor converging
towards innity. Since exponential growth dominates polynomial growth, we have the desired
result. (Otherwise, just try and make a higher order Taylor expansion of the exponential term and
see what happens).
We will in fact need that this lemma holds for any diusion process which we assume henceforth.
The reader is referred to Bhattacharya and Waymire (1990). Recall the denition of the yield and
the credit spread in (16.1) and (16.2). With h = T t we now consider a bond paying o 1 if
V
h
> P i.e. no default at maturity and 0 otherwise. Such a bond is termed a zero-recovery
bond and we denote the market price as D
0
. Clearly, the zero-recovery bond is less valuable than
the corporate bond in the Merton (1974) model and, thus, it has a higher yield. The value of the
zero-recovery bond with time until maturity h is
D
0
(h) = E
Q
[exp(rh)P1
{V
h
P}
] = exp(rh)PQ(V
h
P)
186 Capital Structure and Credit Risk
Whence it follows that the credit spread for the zero-recovery bond is
s(h) = y(0, T t) r =
1
h
log
P
D
0
(h)
r =
1
h
log
_
P
exp(rh)PQ(V
h
P)
_
r
=
1
h
log
_
1
Q(V
h
P)
_
=
1
h
log(Q(V
h
P))

1
h
(Q(V
h
P) 1) =
1
h
Q(V
h
P).
The probability that V
h
< P is the same as the probability that the (random) distance [V
h
V
0
[ is
greater than [P V
0
[. Hence
s(h)
1
h
Q([V
h
V
0
[ [P V
0
[
. .

)
h0
0. (16.5)
On the other hand, if the rm is close to default, i.e. rm value is lower than the principal and
the time until maturity is short, then the yield spread becomes very large. This seems obvious. For
example, consider a bond with face value 100 maturing in two weeks (=1/24 year). If the current
value of the rm is 95 then it is clear that default is very likely. Indeed, the market value of the
debt (according to the Merton model) is 94.78 and as a results the credit spread is 12368 bps.
Obviously, the extremely high yield of the bond is due to the fact that default is almost certain.
There are a number of empirical papers studying the classic structural models, e.g. Merton
(1974) and Leland and Toft (1996), in terms of the predicted credit spread from the models in
comparison with the observed credit spreads. Of course, empirical work in complicated by the type
of assumptions made in most structural models, but generally the structural models seems to lack
some features in order to provide satisfactory results regarding credit spreads. For instance, Eom,
Helwege, and Huang (2004) nd support for the classic criticism that the predicted spreads are
too low for bonds with low risk, i.e. bonds close to maturity or with a high rating. On the other
hand, they also nd that the model by Leland and Toft (1996) overestimates spreads. For further
readings on empirical issues see e.g. Lando (2004), Eom, Helwege, and Huang (2004), and Genser
(2006).
16.2 Credit risk and simple intensity models
The above mentioned criticism of the structural models have led to another class of models: the
so-called intensity models. As earlier noted, the following is based on Lando (2004). The reader is
referred to Lando (2004), Schonbucher (2003), and Due and Singleton (2003) for further readings
on intensity models and, more generally, on pricing credit derivatives.
16.2 Credit risk and simple intensity models 187
As we have seen, the structural diusion models have the feature that the credit spreads converge
to zero in the short end (close to maturity). Intuitively, the reason is that diusion processes have
continuous sample paths and, therefore, default occurs with no surprise over the next short time
interval (it is predictable). It is therefore necessary to introduce some type of surprise eect into
the models. Of course, we could extend the state variable (earnings) to include jumps. However,
it may turn out to be more convenient to abstract from the fundamentals of the rm and instead
directly model the probability of default. Since we want the possibility that there is a positive
probability of default during the next instant of time, we in some sense assume that there are
jumps in some of the fundamental variables of the rm. As may be well known, a jump process is
characterized with the jump size and the instantaneous probability of a jump. This probability is
usually termed the intensity of the jump process. Accordingly, we are now going to set up a simple
version of an intensity model. In the credit risk literature the default intensity is often referred to
as the hazard rate, so we use this terminology henceforth.
Consider a positive random variable which we may interpret as the time of the rst default
event. A random variable is characterized by its distribution which we assume can be described by
a hazard function h. This implies that the probability being greater than a number t > 0 is
2
P( > t) = exp
_

_
t
0
h(s)ds
_
.
One can show that
1
t
P( t + t[ > t)
t0
h(t).
Hence, we can interpret h(t)t as the conditional probability of a default in a small interval (t)
after t given survival up to and including time t. This should be familiar from basic knowledge
about Poisson processes which have a constant intensity. Note that we only condition on the passing
of time. In realistic examples it is usually necessary to be able to condition on a more general
information set, say T
t
at time t. Intuitively, we want something like the above interpretation of
conditional probabilities, i.e.
P( t + t[T
t
) 1
{>t}
(t)t.
In this sense (t) (predictable with respect to the ltration F) is a stochastic pre-default intensity
for the jump time . However, for modeling we need to know what we are working with, i.e. we
need a formal denition. The exact construction of the default intensity is rather technical. At
2
It is standard to let denote a random variable. It should not be confused with any kind of tax rate.
188 Capital Structure and Credit Risk
a rst glance, the reader should try to focus on the main implications of the construction given
below.
A jump time constructed by the Cox process
Consider a given probability space (, T, Q), where we conveniently think of Q as our pricing
measure. We introduce the following:
X process of state variables in R
d
nonnegative measurable function, R
d
R
N
t
the constructed jump process with the property that (X
t
) is the T
t
-intensity of N
the rst jump time of the process N
((
t
)
t0
the ltration generated by X, i.e. (
t
= X
s
[0 s t
E
1
exponential random variable with mean 1, independent of ((
t
)
t0
1
t
-algebra generated by the jump process, N
s
[0 s t
T
t
information set about the state variable X and the jump process, T
t
= (
t
1
t
.
Note that the assumption about E
1
implies that for some number e
1
> 0 we have
P(E
1
> e
1
) = expe
0
.
In addition, dene
= inf
_
t :
_
t
0
(X
s
)ds E
1
_
.
It can then be showed that (X) is a pre-default intensity for the jump time and that 1
{t}

_
t
0
(X
s
)1
{s}
ds is an T
t
-martingale.
In addition to the above, we will assume that the interest rate is stochastic and we denote the
short-rate process as r(X
s
). At time 0 the price of a default-free zero-coupon bond maturing at
time t is
p(0, t) = E
_
exp
_

_
t
0
r(X
s
)ds
_
_
.
We now want to price a corporate zero-coupon zero-recovery bond maturing at time t. First note
that since E
1
is independent of (
t
and
_
t
0
(X
s
)ds (
t
we obtain
E[1
{>t}
[(
t
] = Q( > t[(
t
) = Q
__
t
0
(X
s
)ds < E
1

(
t
_
= exp
_

_
t
0
(X
s
)ds
_
.
16.2 Credit risk and simple intensity models 189
Thus enables us to calculate the price of the zero-recovery bond as
v(0, t) = E
_
exp
_

_
t
0
r(X
s
)ds
_
1
{>t}
_
= E
_
E
_
exp
_

_
t
0
r(X
s
)ds
_
1
{>t}

(
t
__
= E
_
exp
_

_
t
0
r(X
s
)ds
_
E
_
1
{>t}

(
t

_
= E
_
exp
_

_
t
0
r(X
s
)ds
_
exp
_

_
t
0
(X
s
)ds
__
that is
v(0, t) = E
_
exp
_

_
t
0
_
r +
_
(X
s
)ds
__
.
The above example can be extended to the case where a contingent claim promises a payment equal
to f(X
t
), but the actual payment is f(X
t
)1
{>t}
.
The pricing building blocks
From the previous section we have already obtained the value of a zero-recovery zero-coupon bond.
In this section, we state (without proofs) the fundamental building blocks for pricing contingent
claims in the intensity setting. What we want is the price today (at time t) of receiving a future
payment. We assume that maturity happens at time T. Let f, g and m be functions satisfying
sucient integrability assumptions. A future promised payment f(X
T
) if default does not occur
then has the price
E
_
exp
_

_
T
t
r(X
s
)ds
_
f(X
T
)1
{>T}

T
t
_
= 1
{>t}
E
_
exp
_

_
T
t
_
r +
_
(X
s
)ds
_
f(X
T
)

(
t
_
. (16.6)
A claim continuously paying g(X
s
)1
{>s}
until default or maturity has the price
E
_ _
T
t
g(X
s
)1
{>s}
exp
_

_
s
t
r(X
u
)du
_
ds

T
t
_
= 1
{>t}
E
_ _
T
t
g(X
s
) exp
_

_
s
t
_
r +
_
(X
u
)du
_
ds

(
t
_
, (16.7)
whereas todays price of receiving the payment m(X

) at default is
E
_
exp
_

_

t
r(X
u
)du
_
m(X

)1
{t<T}

T
t
_
= 1
{>t}
E
_ _
T
t
m(X
s
)(X
s
) exp
_

_
s
t
_
r +
_
(X
u
)du
_
ds

(
t
_
. (16.8)
190 Capital Structure and Credit Risk
Recovery assumptions
Clearly, the price of a corporate bond is sensible to the change of future cash ow received if default
occurs. In the literature there are three standard assumptions.
1. Recovery of face value (RFV). This assumption implies that the bond holders receive a
fraction of the principal (face value) of the bond at the time of default. Of the three recovery
assumptions, it is the one which resembles legal practice the most because in practice, future
debt principals are accelerated at default. Hence, future coupons and the maturity structure
is canceled, see e.g. Gertner and Scharfstein (1991). For empirical purposes, only the post-
default market price is needed. Mathematically, an integral expression like (16.8) must be
calculated.
2. Recovery of treasury. At default, the corporate bond is replaced with a similar treasury
bond, but with a reduced principal. Note, that this keeps the maturity aspect of payments
alive. In principle, this is a neat assumption when we want to derive implied default prob-
abilities (from market prices). In order to see this, suppose that the bond holder recovers
the fraction of a similar treasury bond at default. Hence, the value of the bond is equal to
receive 1 with certainty but loose (1 ) at maturity if default occurs. That is
v(0, T) = E
_
exp
_

_
T
0
r(X
u
)du
_

T
0
_
+E
_
exp
_

_

0
r(X
u
)du
__
(1 ) exp
_

_
T

r(X
u
)du
_
1
{0<T}

T
0
_
= p(0, T) (1 )E
_
exp
_

_
T
0
r(X
u
)du
_
1
{0<T}

T
0
_
and if we assume independence between r and we get
= p(0, T) (1 )p(0, T)E
_
1
{0<T}

T
0
_
= p(0, T)
_
1 (1 )Q(0 < T)
_
= p(0, T)
_
1 (1 )[1 Q( > T)]
_
= p(0, T)
_
1 (1 )Q(0 < T)
_
= p(0, T)
_
+ (1 )Q( > T)
_
.
Rearrranging terms we obtain the implied default probability
E
_
exp
_

_
T
0

s
ds
_
_
= Q( > T) =
v(0, T) p(0, T)
(1 )p(0, T)
.
16.3 A link between structural and intensity models 191
With real data one must pay attention to the possibility that the right-hand side need not be
between 0 and 1.
3. Recovery of market value (RMV). In this case, the bond holders receive a fraction of
the bonds pre-default market value, see the example below. This assumption has a natural
economic interpretation and is convenient in terms of modeling. Unfortunately, for estimation
purposes the assumption requires knowledge about pre- and post-default prices which may
not be easy to obtain in a meaningful way.
Example: Recovery of market value. Suppose that a defaultable contingent claim promises
a payment equal to f(X
T
) at time T. Let the price process of this claim be V . If (the rst) default
occurs at time then we assume that the claim holders receives
m(X

) = V (), T,
at the time of default. Here [0, 1) is the fractional recovery and V () is the value of the rm
at the time of default, if we take the limit from the left. If there has been no default until t, it
turns out that the value of this defaultable contingent claim is
V (t) = E
t
_
exp
_

_
T
t
_
r + (1 )
_
(X
s
)ds
_
f(X
T
)
_
(16.9)
The recovery assumption is important because, as seen above, it implies how estimation can be
done. The reader is referred to the above mentioned literature for further readings.
It is obvious that the assumed intensity process plays a crucial role for the prices derived. For
pure pricing purposes, it may be an acceptable approach to make a short cut and directly model a
specic intensity process dependent on the particular case (claim) considered. However, it appears
somewhat unsatisfactorily that the intensity process may be xed without a deeper understanding
of the implied assumptions on the fundamentals of the rm. Indeed, default is obviously linked to
the rms investment, dividend, and capital structure decision.
16.3 A link between structural and intensity models
An important link between the structural models and the intensity models has been demonstrated
in Due and Lando (2001). The key is to recognize that information incompleteness induces
potential surprises for market participants and, hence, default of a rm may truly be a surprise.
192 Capital Structure and Credit Risk
This opens up for the possibility that the credit spread of a corporate bond need not converge to
zero as time until maturity vanishes.
We have already argued that the credit spread in structural models becomes zero when time
until maturity converges to zero (for a solvent rm). Consider a rm with asset value process V
and let be the rst hitting time of the default boundary v. Suppose that the value process starts
at V
0
= v
0
. Then the credit spread result stems from the fact in (16.5), i.e.
Q
v
0
( h)
h

h0
0. (16.10)
But what happens if the initial value of V is unknown to market participants? In this case,
they use a density function f(x) to handle their beliefs about the starting point V
0
= x. Inasmuch
as the rm has not yet defaulted, the density should have support on (v, ]. In this case we could
make the analog to (16.10) which is
lim
h0
1
h
_

v
Q
x
( h)f(x)dx.
A brutal guess would be that the limit is also zero in this case. This is incorrect in general and it is
this result which brings us to an introduction of intensities in structural models. Due and Lando
(2001) show that if (i) f(x)
xv
0 and (ii) f

(v) (with limit from the right) exists, then

lim
h0
1
h
_

v
Q
x
( h)f(x)dx =
1
2

2
(v)f

(v). (16.11)
Here,
2
(v) is the squared diusion coecient (satisfying Lipschitz conditions) of V at the default
boundary. (16.11) is a key result because Due and Lando (2001) show that it is in fact the default
intensity.
The result tells us that what we need is a reasonable model which provides us with a density
function f. Due and Lando (2001) apply the above in a setting similar to Leland (1994b). In
this model the capital structure is optimally chosen ex ante and the equity holders endogenously
decide upon default. The idea is that there is full information initially, but immediately following
the issuance of debt, the debt holders only have partial information about the evolution of the true
value process V based on accounting reports. The reader is referred to Due and Lando (2001)
for more on their interesting ndings.
A drawback in the Due and Lando (2001) model is that the default intensity evolves de-
terministically in the time between accounting reports. Interesting guidelines for future research
would be to extend the framework to a stochastic information-intertemporal class and to handle
information incompleteness for equity as well as debt holders.
Chapter 17
Concluding remarks
incomplete...
In this lecture note I have discussed a number of continuous time models related to the capital
structure of a rm. Many other interesting papers have not been studied or mentioned. For instance
Nielsen, Sa a-Requero, and Santa-Clara (1993), Longsta and Schwartz (1995), and Collin-Dufresne
and Goldstein (2001) study the valuation of debt when the boundary is constant or stochastic and
the risk free interest rate is stochastic. Related to the Merton (1974) model Kim, Ramaswamy, and
Sundaresan (1993) and Shimko, Tejima, and van Deventer (1993) also study the eect of stochastic
interest rates. A newer reference on corporate bond valuation is Acharya and Carpenter (2001)
but they do not consider the optimality of the capital structure. Berens and Cuny (1995) have a
discussion related to measuring the advantage of debt with dierent growth rates. Morellec (2001)
analyzes the eect of asset liquidity and the impact bond covenants. In Uhrig-Homburg (2000)
the Leland (1994b) model is extended to analyze what happens when it is costly to issue new
equity. Dangl and Zechner (2001) have a model similar to Goldstein, Ju, and Leland (2001) and
the model in Christensen, Flor, Lando, and Miltersen (2002a) with dynamic capital structure where
default frequencies are studied. A recent empirical study of changes in credit spreads is done in
Collin-Dufresne, Goldstein, and Martin (2001).
The survey starts with the classical models of Merton (1974) and Black and Cox (1976) which
give a basic framework for the structural models. I then focus on models which try to model the
tax advantage to debt. To relate the dierent papers in this category I have rewritten the models
to have the earnings of the rm as the state variable. During this process it has become clear that
there is a problem in assuming that the unlevered asset value can be considered as representing the
193
194 Concluding remarks
price of a traded asset if there are taxes.
The tax advantage modeling concludes with a review of the model presented in Christensen,
Flor, Lando, and Miltersen (2002a) which considers debt renegotiation. An important part of the
renegotiation game is that it is possible to with draw from the renegotiation process and keep the
already settled coupon rate. A main result of the model is that it provides a rationale for ex post
deviation from the absolute priority rule.
The notes also considers some models which focus on other issues than the tax advantage to
debt. A renegotiation game somewhat dierent from the model in Christensen, Flor, Lando, and
Miltersen (2002a) is studied in Mella-Barral and Perraudin (1997), Mella-Barral (1999), and Fan
and Sundaresan (2000). Here the debt can be served strategically to avoid paying the full coupon
rate.
The liquidation or change of production in Mella-Barral and Perraudin (1997) and Mella-Barral
(1999) is related to the analysis in Fran cois and Morellec (2001) and the model in Flor (2008). The
latter paper nds that assets can be important for the valuation of the debt because the possibility
to sell the assets for low earnings give the debt holders an implicit collateral.
Should also mention Pawlina (2010), Chen (2003) Huang, Ju, and Ou-Yang (2003), Hackbarth,
Hennessy, and Leland (2007), Frank and Goyal (2005).
Finally, the notes has covered the issue of debt with nite maturity. Leland and Toft (1996)
nd that perpetual debt is optimal, but the model presented in Flor and Lester (2002) shows that
the optimality of nite maturity depends on the debt being callable or not.
Bibliography
Acharya, V. V. and J. N. Carpenter (2001). Corporate Bond Valuation and Hedging with Stochas-
tic Interest Rates and Endogenous Default. Working paper, Stern School of Business, Depart-
ment of Finance, New York University, New York, NY 10012. Forthcoming in The Review of
Financial Studies.
Alili, L. and A. E. Kyprianou (2005). Some remarks on rst passage of Levy processes, the
American put and pasting principles. The Annals of Applied Probability 15(3), 20622080.
Anderson, R. W. and S. Sundaresan (1996, Spring). Design and Valuation of Debt Contracts.
The Review of Financial Studies 9(1), 3768.
Anderson, R. W., S. Sundaresan, and P. Tychon (1996). Strategic analysis of contingent claims.
European Economic Review 40, 871881.
Applebaum, D. (2004, December). Levy Processes From Probability to Finance and Quantum
Groups. Notices of the AMS 51(11), 13361347.
Back, K. and D. Paulsen (2009). Open-loop equilibria and perfect competition in options exercise
games. The Review of Financial Studies 22(11), 45314552.
Bank, M. and J. Lawrence (2005, January). Informational asymmetry between managers and
investors in the optimal capital structure decision. Working paper, University of Innsbruck,
Universitatsstrae 15, A-6020 Innsbruck, Austria.
Berens, J. I. and C. J. Cuny (1995). The Capital Structure Puzzle Revisited. The Review of
Financial Studies 8, 11851208.
Bhattacharya, R. and E. Waymire (Eds.) (1990). Stochastic processes and applications. New
York: John Wiley & Sons, Inc.
Bj ork, T. (1994, February). Stokastisk kalkyl och kapitalmarknadsteori Del II Specialstudier.
Undervisningsnoter, Matematiska Institutionen, Kgl. Tekniska H ojskolan, 100 44 Stockholm,
195
196 Bibliography
Sweden.
Black, F. and J. Cox (1976). Valuing Corporate Securities: Some Eects of Bond Indenture
Provisions. The Journal of Finance XXXI, 351367.
Black, F. and M. Scholes (1973, May-June). The Pricing of Options and Corporate Liabilities.
Journal of Political Economy 81(3), 637654.
Brekke, K. A. and B. ksendal (1991). The high Contact Principle as a Suciency Condition
for Optimal Stopping. In Stochastic Models and Option Value, pp. 187208. Amsterdam, The
Netherlands: Elsevier Science Publishers, B. V., North-Holland.
Brennan, M. J. and E. S. Schwartz (1978). Corporate Income Taxes, Valuation, and the Problem
of Optimal Capital Structure. Journal of Business 51, 103114.
Broadie, M., M. Chernov, and S. Sundaresan (2007). Optimal Debt and Equity Values in the
Presence of Chapter 7 and Chapter 11. The Journal of Finance 62(3), 13411377.
Chen, N. (2003, June). Chapter 11, Private Workouts and Corporate Debt Pricing under Asym-
metric Information. Working paper, Division of Finance and Economics, Columbia University,
New York, New York 10027, USA.
Childs, P. D., D. C. Mauer, and S. H. Ott (2005). Interactions of corporate nancing and invest-
ment decisions: The eects of agency conicts. Journal of Financial Economics 76, 667690.
Christensen, P. O. and G. A. Feltham (2003). Economics of Accounting: Volume I Information
in Markets. Kluwer Academic Publishers.
Christensen, P. O., C. R. Flor, D. Lando, and K. R. Miltersen (2002a, November). Dynamic
Capital Structure with Callable Debt and Debt Renegotiation. Working paper, University of
Southern Denmark, Campusvej 55, 5230 Odense M, Denmark.
Christensen, P. O., C. R. Flor, D. Lando, and K. R. Miltersen (2002b, November). Dynamic
Capital Structure with Callable Debt and Debt Renegotiation. Working paper, obtained from
papers.ssrn.com, University of Southern Denmark.
Collin-Dufresne, P. and R. S. Goldstein (2001, October). Do Credit Spreads Reect Stationary
Leverage Ratios? The Journal of Finance LVI (5), 19291957.
Collin-Dufresne, P., R. S. Goldstein, and J. S. Martin (2001, December). The Determinants of
Credit Spread Changes. The Journal of Finance LVI (6), 21772207.
Bibliography 197
Dangl, T. and J. Zechner (2001, August). Credit Risk and Dynamic Capital Structure Choice.
Working paper, Department of Business Studies, University of Vienna, Br unner Strae 72,
A-1210 Vienna.
Decamps, J.-P. and A. Faure-Grimaud (2002). Excessive continuation and dynamic agency costs
of debt. European Economic Review 46, 16231644.
Dixit, A. K. (1993). The Art of Smooth Pasting. Harwood Academic Publishers.
Dixit, A. K. and R. S. Pindyck (1994). Investment under Uncertainty. Princeton, New Jersey,
USA: Princeton University Press.
Due, J. D. (1996). Dynamic Asset Pricing Theory (2 ed.). Princeton, New Jersey, USA: Prince-
ton University Press.
Due, J. D. (2001). Dynamic Asset Pricing Theory (3 ed.). Princeton, New Jersey, USA: Prince-
ton University Press.
Due, J. D. and D. Lando (2001, May). Term Structures of Credit Spreads with Incomplete
Accounting Information. Econometrica 69(3), 633664.
Due, J. D. and K. J. Singleton (1999). Modeling Term Structures of Defaultable Bonds. The
Review of Financial Studies 12, 687720.
Due, J. D. and K. J. Singleton (2003). Credit risk: pricing, measurement, and management.
Princeton, New Jersey, USA: Princeton University Press.
Eberhart, A., W. Moore, and R. Roenfeldt (1990). Security pricing and deviations from the
absolute priority rule in bankruptcy proceedings. The Journal of Finance XLV, 14571469.
Eom, Y. H., J. Helwege, and J.-z. Huang (2004). Structural Models of Corporate Bond Pricing:
An Empirical Analysis. The Review of Financial Studies 17(2), 499544.
Fan, H. and S. M. Sundaresan (2000). Debt Valuation, Renegotiation, and Optimal Dividend
Policy. The Review of Financial Studies 13(4), 10571099.
Fischer, E. O., R. Heinkel, and J. Zechner (1989a, March). Dynamic Capital Structure Choice:
Theory and Tests. The Journal of Finance XLIV, 1940.
Fischer, E. O., R. Heinkel, and J. Zechner (1989b, December). Dynamic Recapitalization Policies
and the Role of Call Premia and Issue Discounts. Journal of Financial and Quantitative
Analysis 24, 427446.
198 Bibliography
Flor, C. R. (2003). Dynamic Capital Structure. Ph.D. Dissertation from the University of South-
ern Denmark. Odense, Denmark: University Press of Southern Denmark.
Flor, C. R. (2008, July). Capital Structure and Assets: Eects of an Implicit Collateral. European
Financial Management 14(2), 347373.
Flor, C. R. (2010a). Dynamic Capital Structure. Lecture notes, University of Southern Denmark,
Department of Business and Economics.
Flor, C. R. (2010b). Real Option Analysis: Stochastic Calculus, Dynamic Programming, and Real
Option Analysis. Lecture notes, University of Southern Denmark, Department of Business
and Economics.
Flor, C. R. (2011, September/October). Asset Substitution and Debt Renegotiation. Journal of
Business Finance & Accounting 38(7).
Flor, C. R. and J. Lester (2002, January). Debt Maturity, Callability, and Dynamic Capital
Structure. Working paper, Department of Accounting and Finance, University of Southern
Denmark, Campusvej 55, 5230 Odense M, Denmark.
Fran cois, P. and E. Morellec (2001, January). Debt valuation, renegotiation, and liquidation.
Working paper, University of Rochester, William E. Simon Graduate Shool of Business Ad-
ministration, University of Rochester, Rochester NY14627, USA.
Fran cois, P. and E. Morellec (2004). Capital Structure and Asset Prices: Some Eects of
Bankruptcy Procedures. Journal of Business 77, 387411.
Frank, M. Z. and V. K. Goyal (2005). Trade-o and Pecking Order Theories of Debt. Working
paper, Sauder School of Business, University of British Columbia, Vancouver BC, Canada,
V6T 1Z2. To appear in Eckbo (ed.) Handbook of Corporate Finance: Empirical Corporate
Finance, Elsevier/North-Holland.
Franks, J. and W. Torous (1989). An empirical investigation of rms in reorganization. The
Journal of Finance XLIV, 747779.
Galai, D., A. Raviv, and Z. Wiener (2005, August). Liquidation Triggers and the Valuation of
Debt and Equity. Working paper, Hebrew University of Jerusalem, Israel.
Genser, M. (2006). A Structural Framework for the Pricing of Corporate Securities. Lecture
Notes in Economics and Mathematical Systems, 566. Springer-Verlag.
Bibliography 199
Gertner, R. and D. S. Scharfstein (1991, September). A theory of workouts and the eects of
reorganization law. The Journal of Finance XLVI (4), 11891222.
Gilson, S. C. (1997, March). Transactions Costs and Capital Structure Choice: Evidence from
Financially Distressed Firms. The Journal of Finance LII (1), 161196.
Goldstein, R., N. Ju, and H. E. Leland (2001). An EBIT-Based Model of Dynamic Capital
Structure. Journal of Business 74(4), 483512.
Grenadier, S. R. (2002). Option Exercise Games: An Application to the Equilibrium Investment
Strategies of Firms. The Review of Financial Studies 15(3), 691721.
Hackbarth, D., C. A. Hennessy, and H. E. Leland (2007, September). Can the Tradeo Theory
Explain Debt Structure? The Review of Financial Studies 20(5), 13891428.
Harrison, M. J. (1985). Brownian Motion and Stochastic Flow Systems. Toronto, Ontario,
Canada: John Wiley & Sons, Inc.
Harrison, M. J. and D. M. Kreps (1979). Martingales and Arbitrage in Multiperiod Securities
Markets. Journal of Economic Theory 20, 381408.
Harrison, M. J. and S. R. Pliska (1981). Martingales and Stochastic Integrals in the Theory of
Continuous Trading. Stochastic Processes and their Applications 11, 215260. Addendum:
(Harrison and Pliska 1983).
Harrison, M. J. and S. R. Pliska (1983). A Stochastic Calculus Model of Continuous Trading:
Complete Markets. Stochastic Processes and their Applications 15, 313316.
Hege, U. and P. Mella-Barral (2002, April). Repeated Dilution of Diusely Held Debt. Work-
ing paper, HEC School of Management, CentER and CEPR, HEC School of Management,
Department of Finance and Econmics, 1 rue de la Liberation, F-78351 Jouy-en-Josas Cedex,
France.
Helwege, J. (1999, February). How Long Do Junk Bonds Spend in Default? The Journal of
Finance 54, 341357.
Huang, J.-z., N. Ju, and H. Ou-Yang (2003). A Model of Optimal Capital Structure with Stochas-
tic Interest Rates. Working paper, Smeall College of Business, Penn State University, Uni-
versity Park, PA 16802, USA.
Hull, J. (1997). Options, Futures, and other Derivatives (3 ed.). Upper Saddle River, New Jersey
07458, USA: Prentice-Hall, Inc.
200 Bibliography
Jacobsen, M. (1995). Videregaende Sandsynlighedsregning. Universitetsparken 5, 2100 Kbenhavn
: IMSKU.
Jarrow, R. A., D. Lando, and S. M. Turnbull (1997, Summer). A Markov Model for the Term
Structure of Credit Risk Spreads. The Review of Financial Studies 10(2), 481523.
Jensen, M. C. and W. H. Meckling (1976). Theory of the Firm: Managerial Behavior, Agency
Costs, and Capital Structure. Journal of Financial Economics 3, 305360.
Kane, A., A. J. Marcus, and R. L. McDonald (1985, December). Debt policy and the rate of
return premium to leverage. Journal of Financial and Quantitative Analysis 20(4), 479499.
Karatzas, I. and S. E. Shreve (1988). Brownian Motion and Stochastic Calculus. Number 113 in
Graduate Texts in Mathematics. New York, New York, USA: Springer-Verlag.
Karatzas, I. and S. E. Shreve (1991). Brownian Motion and Stochastic Calculus. Number 113 in
Graduate Texts in Mathematics. New York, New York, USA: Springer-Verlag.
Kim, I. J., K. Ramaswamy, and S. Sundaresan (1993). Does Default Risk in Coupons Aect
the Valuation of Corporate Bonds?: A Contingent Claims Model. Financial Management 22,
117131.
Lambrecht, B. and S. C. Myers (2007). A Theory of Takeovers and Disinvestment. The Journal
of Finance 62(2), 809845.
Lando, D. (2004). Credit risk modeling. Princeton, New Jersey, USA: Princeton University Press.
Leland, H. E. (1994a). Bonds Prices, Yield Spreads, and Optimal Capital Structure with Default
Risk. Working paper, Univeristy of California, Berkeley, USA.
Leland, H. E. (1994b, September). Corporate Debt Value, Bond Covenants, and Optimal Capital
Structure. The Journal of Finance XLIX(4), 12131252.
Leland, H. E. (1998, August). Agency Costs, Risk Management, and Capital Structure. The
Journal of Finance LIII (4), 12131243.
Leland, H. E. and K. B. Toft (1996, June). Optimal Capital Structure, Endogenous Bankruptcy,
and the Term Structure of Credit Spreads. The Journal of Finance LI, 9871019.
Longsta, F. A. and E. S. Schwartz (1995, June). A simple approach to valuing risky xed and
oating rate debt. The Journal of Finance L(3), 789819.
Bibliography 201
Mao, C. X. (2003, June). Interaction of Debt Agency Problems and Optimal Capital Structure:
Theory and Evidence. Journal of Financial and Quantitative Analysis 38(2), 399423.
Margsiri, W., A. Mello, and M. Ruckes (2008). A Dynamic Analysis of Growth via Acquisition.
Review of Finance 12(4), 635671.
Mauer, D. C. and S. H. Ott (2000). Agency Costs, Underinvestment, and Optimal Capital
Structure: The Eect of Growth Options to Expand. In Project Flexibility, Agency, and
Competition: New Developments in the Theory and Application of Real Options, Chapter 9,
pp. 151179. USA: Oxford University Press.
Mauer, D. C. and S. Sarkar (2005). Real Options, agency conicts, and optimal capital structure.
Journal of Banking and Finance 29, 14051428.
McDonald, R. and D. Siegel (1986, November). The Value of Waiting to Invest. The Quarterly
Journal of Economics 101, 707728.
Mella-Barral, P. (1999). The Dynamics of Default and Debt Reorganization. The Review of
Financial Studies 12, 535578.
Mella-Barral, P. and W. R. M. Perraudin (1997, June). Strategic Debt Service. The Journal of
Finance LII, 531556.
Merton, R. C. (1973, Spring). Theory of Rational Option Pricing. Bell Journal of Economics
and Management Science 4, 141183. Reprinted in (Merton 1990, Chapter 8).
Merton, R. C. (1974, May). On the pricing of corporate debt: The risk structure of interest rates.
The Journal of Finance XXIX, 449470.
Merton, R. C. (1990). Continuous-Time Finance. Padstow, Great Britain: Basil Blackwell Inc.
Miltersen, K. R. and E. S. Schwartz (2004). R&D Investments with Competitive Interactions.
Review of Finance 8(3), 355401.
Modigliani, F. and M. H. Miller (1958, June). The Cost of Capital, Corporation Finance, and
the Theory of Investment. The American Economic Review 48(3), 261297.
Moraux, F. (2002). Valuing Corporate Liabilities When the Default Threshold is not an Absorb-
ing Barrier. Working paper, Universite de Rennes I, France.
Morellec, E. (2001). Asset liquidity, capital structure, and secured debt. Journal of Financial
Economics 61(2), 173206.
202 Bibliography
Murto, P. (2007). Timing of investment under technological and revenue-related uncertainties.
Journal of Economic Dynamics and Control 31, 14731497.
Musiela, M. and M. Rutkowski (2005). Martingale Methods in Financial Modelling (2 ed.).
Stochastic Modelling and Applied Probability No. 36. Berlin, Heidelberg, NewYork: Springer-
Verlag.
Myers, S. C. (1977). Determinants of Corporate Borrowing. Journal of Financial Economics 5,
147175.
Nielsen, L. T., J. Sa a-Requero, and P. Santa-Clara (1993). Default risk and interest rate risk:
The term structure of default spreads. Working paper, INSEAD, France.
Novy-Marx, R. (2007). An equilibirum model of investment under uncertainty. The Review of
Financial Studies 20(5), 14611502.
ksendal, B. and A. Sulem (2001). Introduction to Impulse Control Theory and Applications to
Economics. Lecture notes, University of Oslo.
ksendal, B. and A. Sulem (2005). Applied Stochastic Control of Jump Diusions. Berlin Hei-
delberg, Germany: Springer-Verlag.
Pawlina, G. (2010). Underinvestment, capital structure, and strategic debt restructuring. Journal
of Corporate Finance 16(5), 679702.
Pindyck, R. S. (1993). Investments of uncertain cost. Journal of Financial Economics 34, 5376.
Rajan, R. G. and L. Zingales (1995). What Do We Know about Capital Structure? Some
Evidence from International Data. The Journal of Finance L(5), 14211460.
Schonbucher, P. J. (2003). Credit derivatives pricing models: models, pricing and implementation.
Chichester, UK: John Wiley & Sons, Inc.
Schwartz, E. S. and M. Moon (2000). Evaluating research and development investments. In
Project Flexibility, Agency, and Competition, eds. M. J. Brennan and L. Trigeorgis. Oxford
University Press. New York, 85106.
Schwartz, E. S. and C. Zozaya-Gorostiza (2003, January). Investment Under Uncertanty in In-
formation Technology: Asquisition and Development Projects. Management Science 49(1),
5770.
Shimko, N. Tejima, and D. van Deventer (1993, September). The Pricing of Risky Debt when
Interest Rates are Stochastic. The Journal of Fixed Income.
Bibliography 203
Stulz, R. M. (1990a). Managerial discretion and optimal nancing policies. Journal of Financial
Economics 26, 327.
Stulz, R. M. (1990b). Managerial discretion and optimal nancing policies. Journal of Financial
Economics 26, 327.
Titman, S. and S. Tsyplakov (2007). A Dynamic Model of Optimal Capital Structure. Review of
Finance 11(3), 401451.
Uhrig-Homburg, M. (2000, November). Cash-ow Shortage as an Endogenous Bankruptcy Rea-
son. Working paper, Finance Department, University of Mannheim, D 68131 Mannheim.
Uhrig-Homburg, M. (2005). Cash-ow Shortage as an Endogenous Bankruptcy Reason. Journal
of Banking and Finance 29, 15091534.
Vasicek, O. (1977). An Equilibrium Characterization of the Term Structure. Journal of Financial
Economics 5, 177188.
Weiss, L. and K. H. Wruck (1998, April). Information Problems, Conicts of Interest, and Asset
Stripping: Ch.11s Failure in the Case of Eastern Airlines. Journal of Financial Economics 48,
5597.
Wruck, K. H. (1990). Financial distress, reorganization, and organizational eciency. Journal of
Financial Economics 27, 419444.
Zwiebel, J. (1996, December). Dynamic Capital Structure under Managerial Entrenchment. The
American Economic Review 86(5), 11971215.
Index
contingent claim, 75
covenant
exogenous, 78
credit spread, 184
debt
perpetual
constant coupon, 81
market value, 83
subordinated bonds, 80
hazard rate, 187
Merton, 67
partial dierential equation, 75
derivation, 75
price-earnings ratio, 69
state variable, 93
Merton, 68
yield spread, 73
204