Lectures on Corporate Governance

Benjamin E. Hermalin

Draft 5/11/2004 Version 2

 Table of Contents

Contents

1 What is Governance? 1
1.1 Issues of Concern . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Corporate Governance . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 The Study of Governance . . . . . . . . . . . . . . . . . . . . . . 4
1.4 Firm Objectives, Risk, & Return . . . . . . . . . . . . . . . . . . 5
1.4.1 Theory of the Firm . . . . . . . . . . . . . . . . . . . . . . 5
1.4.2 Attitudes Toward Risk . . . . . . . . . . . . . . . . . . . . 9
1.5 Appendix: An Example of Diversication . . . . . . . . . . . . . 13

2 Why Corporations? 15
2.1 What is a Corporation? . . . . . . . . . . . . . . . . . . . . . . . 15
2.2 Centralization of Management . . . . . . . . . . . . . . . . . . . . 17
2.3 Protection of the Enterprise . . . . . . . . . . . . . . . . . . . . . 19
2.3.1 Asset Partitioning . . . . . . . . . . . . . . . . . . . . . . 21
2.4 Liquidity of Ownership . . . . . . . . . . . . . . . . . . . . . . . . 21
2.5 Limited Liability . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

3 Governance and Securities 25

3.1 Costly State Verication . . . . . . . . . . . . . . . . . . . . . . . 25
3.2 Project Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.3 So What is it? Debt or Equity? . . . . . . . . . . . . . . . . . . . 30
3.4 Application: Savings & Loans . . . . . . . . . . . . . . . . . . . . 31
3.5 Control Issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.6 Exploitation of Minority Shareholders . . . . . . . . . . . . . . . 33

4 Agency 35
4.1 Hidden-Action Agency . . . . . . . . . . . . . . . . . . . . . . . . 35
4.1.1 The Basic Model . . . . . . . . . . . . . . . . . . . . . . . 36
4.1.2 A Simple Variant . . . . . . . . . . . . . . . . . . . . . . . 37
4.2 State-Contingent Commodity Theory . . . . . . . . . . . . . . . . 38
4.2.1 Basics of the Theory . . . . . . . . . . . . . . . . . . . . . 38
4.2.2 Ecient Trade . . . . . . . . . . . . . . . . . . . . . . . . 41
4.3 Hidden-Action Agency Continued . . . . . . . . . . . . . . . . . . 42
4.3.1 First-best Benchmark . . . . . . . . . . . . . . . . . . . . 42

i
 Table of Contents

4.3.2 First-best not Attainable . . . . . . . . . . . . . . . . . . 44

4.3.3 Solving for the Second-best Contract . . . . . . . . . . . . 44
4.3.4 Comparative Statics . . . . . . . . . . . . . . . . . . . . . 47
4.3.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.4 Hidden-Information Agency . . . . . . . . . . . . . . . . . . . . . 49
4.4.1 The Model . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.4.2 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.4.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

5 Monitoring 55
5.1 A Simple Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
5.1.1 Blunted Incentives . . . . . . . . . . . . . . . . . . . . . . 56
5.1.2 Free Riding . . . . . . . . . . . . . . . . . . . . . . . . . . 57
5.2 Monitoring Credibility . . . . . . . . . . . . . . . . . . . . . . . . 58
5.3 Monitoring and Contracts . . . . . . . . . . . . . . . . . . . . . . 59
5.3.1 Agency Contracts I . . . . . . . . . . . . . . . . . . . . . . 59
5.3.2 Agency Contract II . . . . . . . . . . . . . . . . . . . . . . 60
5.4 Monitoring for Ability . . . . . . . . . . . . . . . . . . . . . . . . 60
5.4.1 Timing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
5.4.2 Preferences and Ability . . . . . . . . . . . . . . . . . . . 61
5.4.3 Updating Beliefs and Optimal Monitoring . . . . . . . . . 62
5.4.4 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
5.4.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
5.5 Who Monitors? . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
5.5.1 Auditors . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
5.5.2 Government Oversight . . . . . . . . . . . . . . . . . . . . 66
5.5.3 Other Third Parties . . . . . . . . . . . . . . . . . . . . . 67
5.5.4 Other Monitors . . . . . . . . . . . . . . . . . . . . . . . . 67
5.6 Financial Markets as Monitors . . . . . . . . . . . . . . . . . . . 68
5.7 Large Security Holders . . . . . . . . . . . . . . . . . . . . . . . . 69
5.7.1 Large Shareholders . . . . . . . . . . . . . . . . . . . . . . 69
5.7.2 Large Creditors . . . . . . . . . . . . . . . . . . . . . . . . 70

6 Executive Compensation 71
6.1 Executive Compensation: Terminology . . . . . . . . . . . . . . . 71
6.2 Executive Compensation: Theory . . . . . . . . . . . . . . . . . . 72
6.2.1 Intrinsic Incentives . . . . . . . . . . . . . . . . . . . . . . 72
6.2.2 Promotion Seeking . . . . . . . . . . . . . . . . . . . . . . 72
6.2.3 Job Retention . . . . . . . . . . . . . . . . . . . . . . . . . 74
6.2.4 Career Concerns . . . . . . . . . . . . . . . . . . . . . . . 74

Bibliography 78

End Matter 79
Notation Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80

ii
 Enron
Tyco
WorldCom
Global Crossing
Adelphia
South Sea Bubble
Smith,
Adam
governance|textsl
Lecture 1 Smith,
Adam

What is Governance?

C orporate governance is again an issue of popular concern. Enron,

Tyco, WorldCom, Global Crossing, Adelphia, among other companies, have
become household names and synonyms for corporate scandal. Hundreds of
billions of dollars have been lost; by some estimates, WorldComs collapse, alone,
wiped out $126.8 billion.1
Although largely overshadowed by more dramatic world events, issues of
corporate governance have played a role in us politics, with both the current
President and Vice President of the United States facing or having faced in-
quiries into their business dealings.2
Nor is today the rst time corporate governance has been a hot topic. The
South Sea Bubble of the 1720s, for instance, was the Enron of its day. In 1776,
Adam Smith was writing about corporate governance.
But what is corporate governance and what really are the issues of concern?

1.1 Issues of Concern

Governance refers to a number of institutions that serve to regulate how a given
set of individuals manage other individuals property (where property can be
both tangiblee.g., physical assets and moneyand intangiblee.g., corporate
reputation and human capital). Governance issues arise in many settings. For
instance, politicians manage property belonging to the public. A clerk in a store
manages property belonging to the stores owner. And the management team
of a large corporation manage property belonging to investors.
What then are some of the concerns when the people managing the property
are not the same as the people who own the property? As long ago as 1776,
Adam Smith oered one concern:

1 This is the drop in WorldComs stock value from 1999 to 2002. Source: MacAvoy and

Millstein (2003, Table 6.2).

2 See Dean (2004, pp. 3133 and 4253).

1
 2 What is Governance?

Smith,
Adam The directors of [joint stock] companies, however, being the man-
governance!list
of
problems|(
Adelphia agers rather of other peoples money than of their own, it cannot
Tyco well be expected, that they should watch over it with the same anx-
moral hazard|textsl ious vigilance [as owners] . . . Negligence and profusion, therefore,
Enron must always prevail, more of less, in the management of the aairs
Parmalat
Shell
Oil
(Royal
Dutch/Shell
Group) of such a company (Smith, 1776, p. 700).

A more extensive list of potential problems is:

1. Theft. The managers can simply misappropriate the property. This, for
instance, is the allegation in the Adelphia and Tyco scandals.
2. Misdirected effort. The managers can mismanage the property. For
instance, they can fail to work hard at utilizing the property to gener-
ate the largest possible returns. Or they can utilize the property in ways
that are non-optimal for the owners. While such misdirected eort can
be seen as a form of theftafter all, the owners have some entitlement to
the best eorts of the managers given that is what they are presumably
paying them forit is generally seen in the literature as a separate prob-
lem, known as moral hazard. That is, moral hazard is the problem that
potentially arises when the preferences of the owners with respect to the
managers actions do not coincide with the managers preferences.
3. Misinformation. Usually, the owners retain certain control rights over
their property; that is, the managers ability to manage the property is
rarely absolute. For instance, shareholders (owners) of a corporation retain
the right to sell their shares (property). The proper exercise of these re-
tained control rights often depends on the managers providing the owners
with accurate information about the property. For instance, shareholders
often rely on management to provide an accurate picture of their com-
panys status, so that they know whether or not they wish to sell their
shares. If management misleads or misinforms owners, then owners are
likely to suer (e.g., sell [keep] their shares when they would have preferred
to keep [sell] them). Much of the Enron scandal involves allegations that
Enrons management misinformed its shareholders. Similar allegations
have been made in the cases of Parmalat and Shell.
4. Incompetence. Even if the managers are honest, work hard, and keep
the owners accurately apprised of relevant information, there will still be
a problem if the managers are not the most competent people to be man-
aging the property. Thus, one governance issue is determining whether
the managers in place are the best available managers for the property or
whether they should be replaced.
5. Conflicting objectives. Sometimes governance issues arise because
the property owners are themselves divided over how they would like to
see the property used. If some subset of the owners are also the managers
or have greater power over the managers than do the other owners, then
Corporate Governance 3

problems can arise. For instance, both shareholders and debtholders have asset substitution|textsl
governance!list
of
problems|)
ownership claims on the returns of the rm. But, barring certain debt CEO@\CEO\
(chief
executive
office
covenants, it is the shareholders who have the greater control over man- Baskin,
Jonathan Barron
agement. Shareholders and debtholders do not always agree about what Miranti,
Paul
J.
the proper strategy of the rm should be; shareholders are typically more East
India
Company!English
East
India
Company!Dutch
willing to take risks than debtholders. Such conicts are known as asset Baskin,
Jonathan Barron
substitution problems. Miranti,
Paul
J.

1.2 Corporate Governance

As should be clear, governance issues arise anytime owners and managers are
distinct parties. At some level, the issues I face when I hire someone to do
work on my house are the same as those faced by the board of directors of a
Fortune 500 company when hiring a ceo (chief executive ocer). On the other
hand, there are certainly very salient dierences as well.
Beyond the obvious dierences of size, some fundamental dierences have to
do with the corporate form itself. First, in most countries, there is a separate
body of law, corporate law, to govern corporations. Second, a corporation has
a dierent set of instruments with which to deal with governance issues than
I do; for example, a Fortune 500 company can issue its ceo stock options as
an incentive, whereas the set of contingent forms of compensation available for
me to use are more limited. Third, the corporation is, itself, a response to
governance issues that arise with large and complex organizations.
This last point is worth emphasizing. There are many ways to organize
an enterprise. For instance, a rm could be a sole-proprietorship. Or it could
be a partnership, as are many large law rms and accounting rms. In some
instances, it is feasible for the customers to be the owners, as in mutual banks
and insurance rms.3
Indeed, for most of recorded history, there were no corporations. Starting
in medieval Italy, however, various organizational forms emerged that foreshad-
owed the corporate form (see Baskin and Miranti, 1997). The rst joint-stock
companies were established in the early 1600s: The (English) East India Com-
pany, for example, was chartered by Elizabeth i in 1600 and the Dutch East
India Company converted to the joint-stock form in 1602 (Baskin and Miranti).
Why did corporations emerge? Part of the answer is that they proved to be
an ecient means of attracting large amounts of capital for the undertaking of
large enterprises (e.g., the development of trade to the East Indies). However,
that answer is incomplete. After all, the shareholders of the East India company
could have simply formed a giant partnership. Hence, there must be more to a
corporation beyond its ability to attract capital.
One such additional attraction of the corporate form is the relation between
the corporations risk, particularly the risk of default (bankruptcy), and the
investors risk. As we will explore in depth later, the corporate form of organi-

3A particular form of mutual bank in the us is the savings and loan (thrift).
 4 What is Governance?

signaling theory zation oers some attractive means of isolating the investor from some of the
team theory
Modigliani-Miller Theorem risk imposed by the organization and, conversely, isolates the organization from
Modigliani, Franco some of the risk imposed by the investor. However, this isolation will come at
Miller,
Merton a cost.
agency|textsl
agency!agent|textsl
Because the corporate form has many idiosyncratic features relative to other
agency!principal|textsl organizational forms, it is worth studying corporate governance as distinct from
monitoring governance more generally. On the other hand, as noted, there is considerable
commonality among governance issues in many spheres. Hence, while these
lectures will focus on corporate governance, they will certainly touch on some
general aspects of governance as well.

1.3 The Study of Governance

These lectures will utilize the tools of modern microeconomics and nance to
study governance. In particular, much use will be made of game theory and
contract theory. Although some of these tools will be reviewed in the course
of these lectures, these lectures are not intended as a substitute for courses in
these subjects.4
Our study of corporate governance begins with the question of why we have
corporations. As mentioned previously, the corporate form of organization pro-
vides a means of protecting a corporations owners from some risk that they
would otherwise face. It is also true that the corporate form protects the cor-
poration from some of the risk that could otherwise be passed to it from its
owners. Explaining the rationales for these protections will rely, in part, on
signaling theory and the team theory.
Having established rationales for corporations, we turn to the nancing issue;
in particular the debt-vs.-equity issue. The Modigliani-Miller Theorem tells us
that, in a highly stylized world without taxes, the value of the rm is independent
of the division of claims into debt and equity (Modigliani and Miller, 1958). The
possibility of asset substitution and signaling, however, take us from this stylized
world and provide insights into why nancing is done the way it is. As we will
see, however, nancing decisions can also be aected by the agency problems
faced by the corporation.
Agency refers to situations in which one party, known as the agent, is em-
ployed by another party, known as the principal and the two parties have dier-
ent preferences concerning what actions the agent takes and what information
he provides. Much of governance is concerned with how the principal (e.g.,
shareholders) can align the agents (e.g., ceos) incentives with their own.
One way to align the principal and agents interests is for the principal to
monitor the agent. In corporate governance, monitoring of top management
is typically the function of three parties: the board of directors, the security

4 The motivated student interested in a more in-depth overview of game theory should

consult Gibbons (1992). For contract theory, Laont and Martimont (2002) is a good intro-
duction.
Firm Objectives, Risk, & Return 5

markets, and existing or potential large shareholders. Hence one issue is how contingent
compensation
entrenchment|textsl
do these parties monitor and what are the consequences of their monitoring for profits|textsl
the corporation. opportunity costs|textsl
Monitoring is not only done to align interests, but it is also done to make
assessments about the agents (e.g., managements) ability. This type of moni-
toring has implications for the hiring and ring of management and the dynamics
of the board of directors composition.
Beyond monitoring, another way the principal can attempt to improve the
alignment between her interests and the agents is by providing an incentive
contract. Often these contracts are explicit, and involve tying managements
compensation to the performance of the rm. For example, a stock-option plan
is a means of tying managements compensation to the performance of the rms
stock.
Agents are not necessarily passive. A ceo, for instance, could attempt to
entrench himself by taking the company in ways that tend to make him irre-
placeable. Or a ceo could attempt to bargain for less intrusive monitoring (e.g.,
by getting the board of directors to agree to put friends of his on the board).
Although these lectures will primarily consider governance at a general level,
to the extent they deal with specic institutions, the focus will be on Anglo-
American institutions. Some attention will, however, also be given to other
institutions, such as those in Japan and Germany.

1.4 Firm Objectives, Risk, and Return:

A Review
In this section, I will briey review some relevant material from basic economics
and corporate nance.

1.4.1 Theory of the Firm

In neoclassical economics, the rm is assumed to wish to maximize the expected
present value of prots. What does this mean?

Prots
Profits, recall, are the dierence between a rms revenue, what it takes in, and
its costs.

Costs
Costs should be understood to be economic costs; that is, dened by the notion
of opportunity costs. Opportunity costs means that we measure the cost of an
activity by the value of what we are forgoing by doing that activity over the next
best alternative. Often, but not always, this denition of cost coincides with
accounting cost. For example, if a rm buys $1 million worth of a raw material,
 6 What is Governance?

sunk expenditure|textsl then that is the opportunity cost (it is forgoing keeping the $1 million which
imputed costs|textsl
p@$p$
denotes
a
price
(usually) is obviously worth $1 million). Sometimes, however, this denition diers from
present value|textsl accounting cost. One example is when there is an expenditure of resources that
r@$r$ (interest
rate) will be made regardless of what activity is chosen. For instance, if a rm has
discount factor|textsl signed a six-month unbreakable lease, then its rent payments this month are not
a cost (they are a sunk expenditure; sometimes called a sunk cost). Another
dierence has to do with imputed costs: Sometimes a resource is used for which
there is no corresponding expenditure. For instance, a rm builds on land it
already owns. Because the building forecloses other uses of the land, there is
a cost, namely the value of using that land for one of the foreclosed uses. An
example of both sunk expenditures and opportunity costs is the cost of using
raw materials in inventory: Suppose, hypothetically, that a rm purchased raw
materials (e.g., copper wiring) at a price of p0 per unit. Suppose the market for
the raw material has adjusted, so now the price is p1 per unit. Observe that the
p0 is now a sunk expenditure, the true (imputed) cost of using the raw materials
is p1 because by using the materials the rm forgoes reselling the material for
p1 per unit.

Present Value
The notion of opportunity cost lies behind the idea of present value: Suppose
a rm can invest (spend) a dollar today in exchange for $R dollars in a year.
Is this a good investment? Well observe that an alternative for the rm is to
put the dollar in the bank and earn interest r, where r is the annual interest
rate.5 Hence, if it takes this alternative, it will have 1 + r dollars in a year. If
R > 1 + r, then this a good investment (the return in a year outweighs the cost,
which is forgoing 1 + r in a year). If R < 1 + r, then this is a bad investment.
Observe that if R = 1 + r, then we are indierent. Equivalently, we can express
this as saying that today the value of a promise of R dollars in a year is
R
.
1+r
More generally, todays valuethe present valueof an amount of money to be
received in a year is 1/(1 + r) per dollar to be received. The fraction
1
1+r
is known as the (annual) discount factor; that is, the present value of an amount
of money to be received in the future can be calculated by multiplying that
amount by the discount factor.
What if the R dollars were to be received in two years? Well the alternative is
to leave the dollar in the bank for two years. At the end of one year, the rm will
5 This begs the question of where the interest rate comes from, but if one takes it as given

that individual consumers are impatient, then they will demand compensation for money
received in the future or will be willing to pay for money now instead of in the future. By
aggregating such preferences, we can get an interest rate.
Firm Objectives, Risk, & Return 7

have 1 + r dollars. In two years it will earn another r of interest on the principal interest!compounding
of|textsl
(the original dollar) plus it will earn r2 in interest on the interest it earned in t@$t$
is
an
index
of
time
(usuall
the rst year (i.e., it earns interest at the rate r on r dollars = r r = r2 ); this PV@$PV$
(present
value)
is known as compounding. Adding this up, we have 1 + 2r + r2 dollars. Observe present value!PV@$PV$
random variable|textsl
that amount equals (1 + r)2 . More generally, it can readily be shown that a stochastic|textsl
dollar put in the bank today will yield (1 + r)t dollars in t years. This analysis expected value|textsl
tells us that an amount of money to be received in t years has a present value pi@$\pi$
denotes
a
probability
(usually)
of E@$\mathbbE$
(expectation
operato
1
(1.1) expected
value!formula
for
(1 + r)t
per dollar today. In other words, we discount money to be received in t years
by the fraction in expression (1.1).
If the rm is to receive a stream of money, R1 , . . . , RT over T periods, then
the present value, P V , of this stream is the sum of the appropriately discounted
values; that is,

Rt T
R1 RT
PV = + + T
= .
1+r (1 + r) t=1
(1 + r)t

If R1 = = RT (i.e., the money received in each period is the same), then

this formula simplies to
T

 
R 1 1
PV = =R , (1.2)
t=1
(1 + r)t r r(1 + r)T

where R is the common payment each period. If T = (i.e., the payment will
be received forever), then the last fraction in expression (1.2) becomes zero, so
we have


R R
PV = t
= .
t=1
(1 + r) r

Expected Value
The future is typically unknown and there is, therefore, uncertainty about future
prots. In the language of probability theory, we can describe future prots as a
random variable or we can say that future prots are stochastic (i.e., random).
One way to value a random variable is to calculate its expected value. For
random variables that can take a countable number of values, the expected
value is calculated as follows. Assume the values the random variable X can
possibly take can be enumerated as x1 , . . . , xN . Let n be the probability that
the realized value is xn . For example, if X is the random variable value of a
die after a single roll, then xn = n, N = 6, and n (the probability the die
shows n dots) is 1/6 for all n. The expected value of X, denoted EX, is dened
to be

N
EX = 1 x1 + + N xN = n xn .
n=1
 8 What is Governance?

average value|textsl So if X isthe random variable value of a die after a single roll, then its
present
value!formula
for
expectation
of
expected value is

6
n 21
EX = = = 3.5 .
n=1
6 6
One way to interpret the expected value of a random variable is as follows.
Consider T realizations of the random variable (e.g., roll a die T times). The
average value, x, of these realizations is dened to be
T
xt
x = t=1 ,
T
where xt is the realization of the random variable on the tth repetition (e.g., on
the tth throw of the die). It can be shown that as T gets large, the probability
that x is not close to EX vanishes to zero; that is, the average converges to the
expected value as the sample size (i.e., T ) grows large.6
Three facts about expectation are:7
1. If the random variable X can take only one value, (i.e., X is a constant
and not random), then EX = .
2. If X and Y are random variables, then
E(X + Y ) = EX + EY ; (1.3)
that is, the expectation of the sum of random variables equals the sum of
the expectations of those random variables.
3. If a and b are constants, then we can form from the random variable X a
new random variable W = aX + b; that is, the nth possible realization of
W , wn , equals axn + b. For a W formed in this way, EW = aEX + b. This
fact can be summarized as the expectation of an ane transformation of a
random variable equals the same ane transformation of the expectation
of the random variable.8
Using the last two facts, observe that, if cash ows R1 , . . . , RT are each random
variables, then the expected present value of these ows is
 
R1 RT
EP V = E + +
1+r (1 + r)T
ER1 ERT
= + +
1+r (1 + r)T
T

ERt
= .
t=1
(1 + r)t

6 This is known as the law of large numbers.

7 These facts are proved in any good probability theory text.
8 An ane transformation of a variable x is formed by multiplying x by some constant, a,
and adding some constant b, where a and b are real numbers. If b = 0, then the transformation
is said to be linear.
Firm Objectives, Risk, & Return 9

We can also use the second fact to prove a simplied version of the Modig- Modigliani-Miller
Theorem|textsl
liani-Miller Theorem (Modigliani and Miller, 1958). Modigliani, Franco
Miller,
Merton
Theorem 1 (Modigliani-Miller [simplied]) Consider a rm that will liq- residual claimant|textsl
uidate at a future time and payout a total of R, where R is a non-negative
random variable. Assume the rm has both equity and debt, where the total face
value of the debt is D. Then the value of the rm today,9 which is the sum
of the value of the equity and the value of the debt, is the same, namely ER,
regardless of the value of D; that is, the division of claims between debt and
equity is irrelevant to the value of the rm.

Proof: Observe that debt with face value of D pays D if R D and pays R if
R < D (if R < D, the rm is bankrupt and the debtors have priority over the
shareholders in getting paid). We can summarize this as saying that the payo
to the debtholders is the random variable X = min{D, R}.10 The value of the
debt is, thus, its expected value, EX. The shareholders have a residual claim
on the rm (hence, they are known as residual claimants). This means they get
R D if R D and they get 0 if R < D. This can be summarized as saying
the payo to the shareholders is the random variable Y = max{R D, 0}.11
Observe, for future reference, that

X +Y R (observe R + 0 = R and D + (R D) = R) . (1.4)

The value of equity is the expected value of Y , EY . The value of the rm is the
sum of EX and EY . Hence,

Value of rm = EX + EY
 
= E(X + Y ) by equation (1.3)
 
= ER by equation (1.4)

As was to be established.

1.4.2 Attitudes Toward Risk

Consider two gambles. In gamble one, you win $1,000,000 if a fair coin lands
heads and $0 if it lands tails. In the other gamble, you win $500,000 if the coin
lands heads and $499,998 if it lands tails. Which gamble would you rather play?
Most people would elect the second gamble. Note, however, that the second
gamble has an expected value of $499,999, while the rst has an expected value
of $500,000. The reason why most people choose the gamble with the lower ex-
pected value is that it also appears less risky. Hence, when it comes to choosing
9 We will ignore any discounting; alternatively, R and D can be understood to be the

discounted values of future payos.

10 The function min{, } equals the smaller of the two values.
11 The function max{, } equals the larger of the two values.
 10 What is Governance?

utility|textsl among risky alternatives, most individuals are concerned with risk as well as
utility function|textsl
expected value (return).
U@$U(\cdot)$
is
a
utility
function
(usually)
expected utility|textsl One way to allow an individuals preferences over risk to enter into our
expected analysis is to assume that what matters to an individual is not the amount of
utility!maximizers|textsl
certainty equivalent|textsl money he or she receives per se, but the happiness or utility he or she derives
CE@$CE$
(certainty
equivalent from that money. A complete review of utility theory is outside the scope of
value) these lectures.12 Here, we are primarily concerned with expected utility. Let
diminishing marginal
utility|textsl U (y) be the amount of utility an individual gets from y dollars; that is, U () is
a utility function.13 The function U () has the property that, if y > y  , then
U (y) > U (y  ); that is, more money leads to greater utility. If Y is a random
variable that has possible realizations y1 , . . . , yN , where the probability that yn
is realized is n , then the individuals expected utility is
N


EU (Y ) = 1 U (y1 ) + + N U (yN ) = n U (yn ) .
n=1

A standard behavioral assumption is that decision makers are expected-

utility maximizers;14 that is, if given the choice between two gambles represented
by random variables X and Y , a decision maker with utility function U () will
choose X if and only if
EU (X) EU (Y ) .
For any gamble X, we dene the certainty equivalent value, denoted CEX , to
be the amount of money that, were it to be paid with certainty, would yield the
decision maker the same expected utility as X; that is,

EU (CEX ) = EU (X) .

Recalling that the expectation of a constant is the constant (fact #1 on page 8),
this last expression can be rewritten as

U (CEX ) = EU (X) .

Another way to describe the certainty equivalent value of a gamble is that it

is the smallest amount of money that the decision maker would be willing to
accept in exchange for giving up the right to play the gamble.
A standard assumption about individual decision makers is that their utility
functions exhibit diminishing marginal utility of money. Specically, if > 0
is a common increment, then the individual has diminishing marginal utility if

U (x + ) U (x) U (y + ) U (y) for all pairs of x and y , x < y ,

12 Any decent microeconomics text will provide an account.

13 Technically, U : R R, where R is the set of real numbers.
14 Howgood an assumption expected-utility maximization is a matter of debate in economics
and nance. See Rabin (1998) for a survey of some of the issues.
Firm Objectives, Risk, & Return 11

and diminishing marginal

utility!strict|textsl
risk aversion|textsl
U (x + ) U (x) > U (y + ) U (y) for some pair of x and y , x < y . (1.5)
In other words, giving someone an increment of increases his or her utility by
the same or greater when he or she is starting with less than when he or she is
starting with more.15 For example, the increase in your happiness from going
from an income of $50,000 a year to an income of $100,000 is probably greater
than the increase in your happiness from going from an income of $1,000,000
to an income of $1,050,000. If the inequality in expression (1.5) holds for all x
and y pairs, x < y, then the decision maker has strictly diminishing marginal
utility of money.16
If a decision maker has diminishing marginal utility of money, then for any
gamble X,
CEX EX .
Moreover, if the decision maker has strictly diminishing marginal utility of
money and X is not a constant (i.e., X is a real gamble), then17
CEX < EX . (1.6)
Expression (1.6) can be read as saying that an individual with strictly dimin-
ishing marginal utility of money would be willing to sell a true gamble, X, for
less than its expected value. Why would he or she do this? Because a dimin-
ishing marginal utility of money induces a distaste for risk; under such a utility
function the value put on smaller payos is disproportionately large given their
size, while the value put on larger payos is disproportionately small given their
size. A decision maker for whom expression (1.6) holds is said to be risk averse.
Note that risk aversion oers an explanation for the choices that people tend
to make with respect to the two gambles considered on page 9: In gamble one,
you win $1,000,000 if a fair coin lands heads and $0 if it lands tails. In gamble
two, you win $500,000 if the coin lands heads and $499,998if it lands tails.
Suppose, for instance, that the utility function were U (x) = x. It is readily
veried that this utility function exhibits strictly decreasing marginal utility of
money. What are the expected utilities oered by the two gambles:
1 1 1 1
EUone = 1, 000, 000 + 0 = 1000 + 0 = 500
2 2 2 2
1 1 1 1
EUtwo = 500, 000 + 499, 998 = 707.107 + 707.105 = 707.106 .
2 2 2 2
Because gamble two yields the larger expected utility, it will be preferred to
gamble one.
15 Mathematically, diminishing marginal utility of money corresponds to U () being a con-

cave function.
16 Mathematically, if U () is twice dierentiable, then strict diminishing marginal utility is

equivalent to U

() < 0.
17 This follows from Jensens Inequality (see, e.g., van Tiel, 1984, page 11).
 12 What is Governance?

risk neutrality|textsl If, for all gambles X, CEX = EX, then a decision maker is risk neutral. A
diversification|textsl
risk-neutral
agency!principal!risk
neutrality decision maker does not care about the risk inherent in a gamble,
of he or she cares about its expected value only. It can be shown that there is no
agency!agent!risk
aversion
of loss of generality in taking a risk-neutral decision makers utility function to be
the identity function; that is, U (x) = x for a risk-neutral decision maker.
While individuals are typically thought to be risk averse, the security holders
of a company are typically thought of as risk neutral; even though the security
holders are, of course, individuals. The reason for this apparent contradiction is
that security holders can typically diversify. Diversifying is, essentially, follow-
ing the adage dont put all your eggs in one basket, meaning that you want
to spread your risk around. A complete study of diversication is beyond the
scope of these notes, but Section 1.5 considers a simple example of diversi-
cation that illustrates how diversication can make diversied security holders
essentially risk neutral with respect to the behavior of the rms in which they
hold securities.
Of course some stakeholders cannot readily diversify. For instance, em-
ployeesespecially managementtypically have considerable human capital in-
vested in the rm; so much so, in fact, that it is not feasible for them to diversify
away that risk.
Putting the last two insights together, it is clear why, in most agency mod-
els of governance, the principal (e.g., shareholders) is assumed to be risk neu-
tral, while the agent (e.g., the ceo) is assumed to be risk averse.
Another point is to note that if there is a gamble, X, a risk-neutral party,
and a risk-averse party, then it is never ecient to allocate the gamble to the
risk-averse party. When monetary transfers are possible, eciency requires
that the party that values a good more ends up owning it (regardless of initial
allocation). The risk-neutral party values X at EX, while the risk-averse party
values it at CEX < EX; the gamble is worth more to the risk-neutral party than
the risk-averse party. For instance, recall the rst gamble on page 9: Heads you
win $1,000,000,
tails you win $0. If the risk-averse partys utility function is
U (x) = x, then, as we saw above, the gamble oers him an expected utility
of 500. From this we can calculate his certainty equivalent value:

U (CEX ) = EU (X)

CEX = 500

so, squaring both sides,

CEX = $250, 000 .

As noted earlier, the expected value of this gamble is $500,000. Hence, if the risk-
averse party initial owns the gamble, he can nd a price between $250,000 and
$500,000 such that the risk-neutral party would be happy to purchase the gamble
from him. For instance, if he sells at $375,000, then he gains $125,000. The
risk-neutral person now owns the gamble X 375, 000, which has an expected
value of $125,000, so she, too, gains.
Appendix: An Example of Diversification 13

1.5 Appendix: An Example of Diversication

To clarify the issue of diversication, consider the following example. There are
two companies in which you can invest. One sells ice cream. The other sells
umbrellas. Ice cream sales are greater on sunny days than on rainy days, while
umbrella sales are greater on rainy days than on sunny days. Suppose that, on
average, one out of four days is rainy; that is, the probability of rain is 14 . On
a rainy day, the umbrella company makes a prot of $100 and the ice cream
company makes a prot of $0. On a sunny day, the umbrella company makes a
prot of $0 and the ice cream factory makes a prot of $200. Suppose you invest
in the umbrella company only; specically, suppose you own all of it. Then you
face a gamble: on rainy days you receive $100 and on sunny days you receive
nothing. Your expected value is
1 3
$25 = $100 + $0.
4 4
Suppose, in contrast, that you sell three quarters of your holdings in the
umbrella company and use some of the proceeds to buy one eighth of the ice
cream factory. Now on rainy days you receive $25 from the umbrella company
(since you can claim one quarter of the $100 prot), but nothing from the ice
cream company (since there are no prots). On sunny days you receive $25
from the ice cream company (since you can claim one eighth of the $200 prot),
but nothing from the umbrella company (since there are no prots). That is,
rain or shine, you receive $25your risk has disappeared! Your expected value,
however, has remained the same (i.e., $25). This is the magic of diversication.
Moreover, once you can diversify, you want your companies to make expected-
value-maximizing decisions. Suppose, for instance, that the umbrella company
could change its strategy so that it made a prot of $150 on rainy days, but
lost $10 on sunny days. This would increase its daily expected prot by $5the
new EV calculation is
1 3
$150 + ($10) = $30 .
4 4
It would also, arguably, increase the riskiness of its prots by changing its
strategy in this way. Suppose, for convenience, that 100% of a company trades
18
on the stock exchange for 100 times its expected daily earnings.
1 The entire ice

cream company would, then, be worth $15,000 (= 100 4 $0 + 34 $200 )

18 The price-to-earnings ratio is 100 here, but the value of the price-to-earnings ratio does

not matter for the conclusions reached here. If the ratio were r, then decreasing your holdings
of the umbrella company to 16 th of the company and increasing your holdings of the ice cream
2
company to 15 th would yield a trading prot of
30r 150r 5
= r > 0.
12 120 4
Now you might wonder whether it is appropriate to use the same price-to-earnings ratio
for both rms. In this case it is, at least if you believe that the stock price is driven by
fundamentals (that is, future prots).
14 What is Governance?

and the entire umbrella company would, then, be worth $3000. To return to your
position of complete diversication and earning $25 a day, you would have to
reduce your position in the umbrella company to hold one sixth of the company
2
and you would have to increase your holdings of the ice cream company to 15 th
of the company:
1 2
Earnings on a rainy day : $150 + $0 = $25; and
6 15
1 2
Earnings on a sunny day : ($10) + $200 = $25.
6 15
Going from holding one fourth of the umbrella company to owning one sixth
1
of the umbrella company means selling 12 th of the umbrella company,19 which
would yield you $250 (= 12 $3000). Going from holding one eighth of the ice
1
1
cream company to owning 2/15ths means buying an additional 120 th of the ice
cream company, which would cost you $125 (= 120 $15, 000). Your prot
20 1

from these stock market trades would be $125. Moreover, you would still receive
a riskless $25 per day. So because you can diversify, you benet by having your
umbrella company do something that increases its expected value, even if it is
riskier.

19 Since 1 1 = 3 2 = 1 .
4 6 12 12 12
20 Since 2 1 = 16 15 = 1 .
15 8 120 120 120
 California!incorporation
in

Lecture 2

Why Corporations?

A sa rm
noted in the last lecture, a corporation is only one way in which
can be organized. Alternatives include sole proprietorships (which
could include family rms), mutuals, and partnerships. Why, then, do we need
the corporate form of organization and what advantages does it oer over other
forms?

2.1 What is a Corporation?

What a corporation is depends largely on whether you want a legal answer or an
economic answer. The legal answer is simply that a corporation is an entity that
has been incorporated under the laws of the relevant jurisdiction. For an entity
to be incorporated, it must meet certain conditions spelled out by the relevant
corporate law. For instance, in the state of California, to be incorporated a rm
must draft articles of incorporation satisfying certain requirements, including
the naming of an initial agent for service of process and setting the total number
of shares the corporation will be authorized to issue. You must also pay the
state a $100 ling fee.
From an economic perspective, what is distinctive about a corporation over
other forms of organization is the way in which shareholders, the legal owners of
a corporation, and the corporation are isolated from each other with respect to
certain liabilities. There are also other features of a corporation that, although
not necessarily distinct to a corporation, are nonetheless easier to utilize for
practical and legal reasons than they are with other forms. These features in-
clude transferability of investor interests, an indenite life span, and centralized
management. Under some jurisdictions there could be tax advantages as well.
To understand some of the advantages of a corporation, consider a situation
in which an inventor wishes to build a rm to market her new invention. If
such a rm requires little capital, so that it is feasible for the inventor either
to fund the rm out of her pocket or use retained earnings to quickly grow the
rm to the desired size, then the inventor has the option of a sole proprietorship

15
16 Why Corporations?

or incorporation. What are the advantage of incorporating? Possibly she gains

some limited liability protection should the rm be sued (e.g., her invention
proves to be hazardous to consumers health).1 Such added protection reduces
the riskiness of her venture. On the other hand, consumers, knowing that they
have less ability to collect damages if they are injured, could be less willing to
buy from the company or buy only at a lower price. We will discuss the costs
and benets of limited liability later in Section 2.5. Another advantage is that
she can sell shares in the company. Because she can use the proceeds from
such sales to purchase other securities, she can diversify away some or most
of her risk. That is a benet. The cost is that she will cease to own 100%
of the rm, which means her ability to control the rm will be less. She may
have to accept management by others to induce investors to purchase shares. In
addition, because the problems that corporate governance is meant to ameliorate
are typically not fully corrected by corporate governance, separating ownership
from control will impose costs on the rm that will be reected in the price
of the shares she sells; they will sell for less than their corresponding expected
present value were she to maintain sole proprietorship.
If the inventor needs to raise capital to launch her business, then she raises
capital by issuing claims on the future prots of the rm. She could retain sole
proprietorship and attempt to get her funding from a bank loan or by issuing
debt; that is, she can borrow the necessary funds. Such borrowing, however,
exposes her to risk should the rm default on its loans (fail to pay them back).
As a sole proprietorship, her creditors can seek to seize not only the assets of
the rm, but also her personal assets. If, instead, she had incorporated, then
her personal assets could potentially be protected from creditors (but recall
footnote 1).
Alternatively, she can raise capital by forming a partnership. Wouldbe in-
vestors provide capital and share in the prots as partners. Partnerships, how-
ever, have a number of bad properties vis-a-vis the isolation of risk between
individuals and the rm. First, partnership law may allow an outside claimant
(e.g., a creditor or a winning plainti in a law suit) to collect his due from just
a subset of partners if the other partners are unable to pay their share. That
is, losses need not be share proportionately. Moreover, there is typically no
limited liability. Further, it is dicult to trade rights in a partnership, meaning
that an investor can have diculty liquidating his position or diversifying his
portfolio. In addition, each partner has management rights. Management by
committee can be less ecient than centralized, hierarchical management, so
such democratization of management rights could prove costly. A nal problem
is that, under some circumstances, a partner can unilaterally impose liabilities
on the partnership.
Relative to a partnership, incorporation does much better in these regards:

1 Whether or not she gains this protection depends on how tightly she controls the rm

and her ownership of its shares. In a closely held corporation, it is often feasible to pierce
the corporate veil; that is, make the shareholders personally responsible for the payment of
damages. See Clark (1986) on closely held corporations.
Centralization of Management 17

The shareholders cannot be made to put up more money to cover the corporate
form!advantages
of
Clark,
Robert
C.
debts or obligations of the corporation (i.e., they enjoy limited liability). Hansmann, Henry
Kraakman,
Reinier
All shareholders are treated proportionately to their ownership. limited
liability!corporations
predatin
Transfer, including up to liquidation, of ownership claims is easy. East
India
Company!English
California!limited
liability
law

The rm is independent of its shareholders in the sense that the assets of
the rm cannot be attached or seized to cover the debts or other obliga-
tions of any individual shareholder.
The centralization of management in a corporation can also be a plus, although
it comes at a cost of separating ownership from control. In other words, the
drawback to centralizing management is that it creates governance problems.
To summarize, the corporate form oers some advantages over other forms,
at least in some contexts. Four features in particular make the corporate form
attractive:
1. Centralization of management;
2. Protection of the enterprise from the personal risks imposed by owners;
3. Ease of transferring ownership shares;
4. Limited protection of the individual owners from the risks imposed by the
rm (limited liability).
We will study each of these in more depth. Before doing so, it is worth making
an observation about the fourth: Many writers on corporations focus on the
advantages of limited liability (see, e.g., Clark, 1986). However, it is important
to recognize that limited liability cannot be the raison detre of the corporation.
As Hansmann and Kraakman (2000a,b) note, the existence of the corporation
predates limited liability. For instance, while the East India Company was
chartered in 1600, the shareholders of English manufacturing companies did
not enjoy limited liability until 1855. Shareholders of California incorporated
rms didnt enjoy limited liability until 1931; well after the founding of many
California corporations.

2.2 Centralization of Management

Although most of these lectures will be concerned with the adverse consequences
of the separation of ownership and control, it is worthwhile to rst ask why
such a separation exists; in particular, why is it advantageous to centralize
management in the hands of a few?
As we saw, even if there is no need to attract capital, an entrepreneur will
have a motive to sell claims on her rms returns in order to diversify her own
portfolio. If she does need to raise capital, then she will have no choice but to
sell claims on her rms returns. Either way, there are reasons for a rm to end
up with many claimants.
 18 Why Corporations?

team theory!free-riding Although all claimants could, in theory, form a giant committee to run the
problem|textsl
team theory|textsl rm, such a practice would have a number of drawbacks:
1. Coordination and communication among the committee members would
impose considerable cost on the organization. As anyone who has ever
tried to schedule a committee meeting knows, it takes a lot of eort to get
even a modest-sized group of people together.
2. Conict among claimants could be a problem. If some decisions advan-
tage one group of claimants over another, then it could be hard to reach
agreement, which could impose costly delays on decision making.
3. There can be a dilution of expertise. Some people are simply better at
running rms than others. Unless the set of claimants is limited to the
best managers, there is the risk that the expertise of the better managers
is diluted by the ill-informed opinions of the worst managers.
4. Related to the last point, the people with expertise in running companies
might not be the people who wish or are able to invest in the rm. Hence,
there could be division of labor reasons for employing managers.
5. There could be inecient duplication of eort; multiple individuals could,
for instance, do the same analysis.
6. Conversely, committees or teams can suer from free riding; that is, each
team member is tempted to take it easy in the hopes that some other team
member will do his work.
The last two points can be illustrated with a simple team theory model:
Suppose there are 10 investors. At a personal cost of 10, an investor can de-
termine what the right course of action is for the rm. If the rm pursues the
right course, it pays o a total of 110 (or 11 to each investor). If it doesnt
pursue the right course of action, it pays o 0. Clearly, it is inecient for more
than one investor to determine the right action; the total surplus from the rm
is 110 10n when n 1 investors expend eort determining the right course.
This total surplus is maximized by n = 1.
Suppose, however, that no one investor is put in charge (i.e., assigned the
duty of determining the right course of action). Consider the reasoning of each
investor. Suppose he reasons that each of the other investors will expend eort
with probability 1 and not expend eort with probability . Then the
probability that none of the other investors determine the right course of action
is 9 . Should this investor expend eort? His payo if he does is 1 (= 11 10).
His expected payo if he doesnt is
11 (1 9 ) + 0 9
(assume everyone is risk neutral for convenience). The investor in question will
denitely expend eort if that expected value is less than 1. He denitely wont
if that expected value exceeds 1. If, however,
11 (1 9 ) = 1 , (2.1)
Protection of the Enterprise 19

then he is indierent between expending eort and not. If he is indierent, then Hansmann, Henry
Kraakman,
Reinier
he is willing to choose what he does randomly.2 Since there is no reason to expect
that this given investor is any dierent than his fellow investors, if they are
randomizing in their choice of action (expending eort with probability 1 and
not expending eort with probability ), then he presumably utilizes the same
probabilities. Hence, we have an equilibrium if all investors are randomizing
using the that solves equation (2.1).3 Solving, we have = .989. Observe that
the probability that no one bothers to determine the right course is .98910 .9;
that is, 90% of the time no one would bother to gure out what the right course
is! Moreover, the probability that exactly one person will determine the right
course (the ecient outcome) occurs with only probability .096.
Clearly, the investors would do better if they put just one of their number in
charge of determining the rms right course. Indeed, they could even hire an
outsider as their ceo: Provided this outsider is paid at least 10 if the rm returns
110 and nothing if it return 0, she will have the right incentive to determine the
right course. Realized surplus will be the maximum, namely 100.

2.3 Protection of the Enterprise

If I go bankrupt, my creditors can seize the shares of stock I own (or have
them sold and retain the proceeds). What they cant do, however, is seize the
proportion of the assets of the rms in which Ive invested that corresponds to
the proportion of my ownership of those rms. That is, if I own 5% of some
company, they cant seize 5% of that companys assets.
As Hansmann and Kraakman (2000a,b) argue, this is can be an important
protection. Moreover, it is a protection that has three components:

1. As an individual, I cannot pledge the assets of the rm as collateral for

personal loans. This is true whether I am just a shareholder or the ceo.

2. My proportional ownership of the rm does not give my personal creditors

a proportional claim on the assets of the rm.

3. Because the rm is a judicial person, it can own property; that is, its
property is its own and partitioned from the property of its shareholders.

The value of the rst protection is as follows. Suppose three friends and
I decide to start a company. Each of us puts in $250,000 and the company
purchases a $1,000,000 of necessary assets to operate. Suppose these assets will
return a total of $1,200,000 (in present value); that is, each of us will receive
$300,000 (for a prot of $50,000). Assume having done this, I have no personal
assets. Suppose another party invites me to invest in a more speculative propo-
sition: If I invest $500,000, I have a 50% chance of getting paid $1,000,000 and

2 In the language of game theory, he is willing to play a mixed strategy.

3 That is, we have a Nash equilibrium in mixed strategies.
20 Why Corporations?

a 50% chance of getting paid nothing. Suppose I could borrow the $500,000 by
pledging $500,000 of my companys assets. Ignoring ethical issues and assuming
Im risk neutral (or not too risk averse), should I do so? Yes, because in the
good state, when I get paid $1,000,000, my wealth will be:

$1, 000, 000 $500, 000 + $300, 000 = $800, 000 .


 
 

payment repaying loan my share

In the bad state, when I get paid $0, half the assets of the rm will be seized,
putting it out of business, so I dont get my $300,000 share. I dont personally
repay the loan, it is repaid by the seized assets. Suppose that my friends can keep
me from getting any share of the liquidated value of the remaining assets (this
assumption isnt critical; if I got some share than this would make borrowing
even more attractive). So my wealth in the bad state is $0. My expected wealth
if I borrow the $500,000 and pledge the rm assets is $400,000 (= 12 $800, 000 +
2 0). Because $400,000 exceeds the $300,000 I will have for sure if I dont
1

borrow the $500,000, I should borrow the money. Each of my (former) friends
expected wealth will be $233,333, which is less than the guaranteed $300,000
each would have had if Id not borrowed the $500,000.
Indeed, I dont even need to invest the $500,000. Suppose some bank would
lend me $500,000, with $500,000 of company assets as collateral, just for my
consumption. I get $500,000 worth of consumption versus the $300,000 I would
have got had I not borrowed the money.
The point is that allowing a shareholder to pledge the rms assets to collat-
eralize his personal borrowing is a clear recipe for disaster. No one would invest
if such pledging were allowed. Hence, unless we protect the rms assets from
the individuals creditors, there can be no investment. Without investment, it
would often be impossible for rms to raise the capital they need.
What if we limited shareholders to pledging no more than a proportional
amount of the rms assets (in the preceding example, I owned 25% but pledged
50%)? This too would create problems. First, it would be dicult to plan
if, at any moment, some number of assets could be seized to satisfy various
shareholders creditors. Second, it would raise the cost of borrowing by the
rm: Suppose in the earlier example, the rm suddenly needed an additional
$100,000 in capital. If none of the shareholders had pledged the rms assets,
then there is no risk in this loan: The rm will return $1,200,000, more than
enough to cover the $100,000; moreover, the rm would still be worth more as
an ongoing concern than liquidated. However, if shareholders had pledged their
share of the assets, then whatever risk they have undertaken now passes to the
rm, which would (i) raise its cost of debt because the lender would demand a
higher face value to compensate for this risk and (ii) incur additional costs as
the lender would need to assess what the risk pass-through was. Item (ii) could
be prohibitively expensive if there are a large number of shareholders.
Finally, if the company did not own its assets as a separate entity (i.e., a
judicial person), then the company couldnt pledge those assets as collateral for
borrowing (you cant pledge property to which you dont have clear title), which
Liquidity of Ownership 21

would either preclude it borrowing or raise its cost of borrowing (uncollateralized Hansmann, Henry
Kraakman,
Reinier
loans cost more than collateralized loans). asset partitioning|textsl
market liquidity|textsl

2.3.1 Asset Partitioning

As Hansmann and Kraakman (2000b) point out, asset partitioning issues also
arise in contexts other than those partitioning the private assets of investors and
the assets of the rm. For instance, to use their example, consider an airline
company, hkair, that decides to go into the car rental business. One option is
for hkair to simply make the car rental business a division of hkair. Another
option is to make the car rental business a separate company whose stock is
completely owned by hkair (alternatively, make both the airline and the car
rental company separate subsidiaries of single holding company, the hk group).
Why have separate subsidiaries?
The answer is that by partitioning the assets of the car rental business from
the airline in this way (and vice versa), there is no question of the assets of one
subsidiary being seized to pay the debts incurred by the other. For example,
suppose the car rental subsidiary is extended sales credit by the automobile
manufacturers. While the automobile manufacturers likely have a good sense
of the car rental business from their dealing with other car rental rms, they
probably are not up on the airline industry. Hence, they would have to do
more checking and analysis before extending credit to the car rental subsidiary
if there were a risk that losses in the airline subsidiary could lead to the seizure
of assets in the car rental subsidiary or would otherwise imperil the car rental
subsidiarys ability to repay its loans. In essence, subsidiaries function like re
doors in a buildingthey keep problems from spreading from one business to
another.

2.4 Liquidity of Ownership

An advantage of asset partitioning between owners private assets and those
of the rm is that it facilitates the trade of shares. If each owners private
debts could spill over onto the rm, then existing owners would want to vet any
prospective owner carefully before allowing him or her to purchase shares. This
would be a big burden and would vastly increase the transactions costs involved
in trading shares.
A market with low transactions coststhat is, a liquid marketoers a
number of advantages. First, someone has to bear those transactions costs, so
it makes shares less valuable and, thus, the rm gets less capital from selling
shares than it otherwise would. Moreover, it also increases the risk associated
with holding shares. Future events may cause me to want to sell shares (e.g., I
lose my job and need money for necessities). If markets are illiquid, then I risk
delays in being able to sell my shares. To the extent I may need to cash out
immediately, this is a risk that I will need to be compensated for bearing. This
compensation will come in the form of lower stock prices. Or, put dierently,
 22 Why Corporations?

liquidity premium|textsl there is a liquidity premium; I am willing to pay more for liquid assets than
Aghion, Philippe
Hermalin,
Benjamin
E. illiquid assets ceteris paribus.
limited
liability!signaling
justification
for
signaling|textsl
type!in
a
2.5 Limited Liability
signaling
model|textsl
Aghion, Philippe As noted earlier, limited liability is not essential for corporations to exist.
Hermalin,
Benjamin
E. Nonetheless, limited liability can be a valuable restriction.
The most compelling justication for limited liability is the signaling model
of Aghion and Hermalin (1990). In a signaling model, one party, the informed
party, knows payo-relevant information that another party, the uninformed
party, doesnt know.4 For instance, the owner-manager of a rm could know
more about how likely the rm is to be successful than a potential lender does.
Lets call a rm that is very likely to be successful a good-type rm and lets
call a rm that is unlikely to be successful a bad-type rm. A potential lender
doesnt know if it is dealing with a good or bad-type rm. Because it might be
dealing with a bad-type rm, which is more likely to fail to repay a loan, the
lender will demand a higher interest rate from any random rm than it would if
it knew it was dealing with a good-type rm (i.e., one likely to repay the loan).
In an attempt to get a more favorable interest rate, a good-type rm will seek
to signal that is the good type by oering terms that, should they convince the
lender that the rm is good, will result in a lower interest rate than it would
get if it failed to convince the lender. Importantly, these terms have to be such
that a bad-type rm would be unwilling to mimic a good-type rm; that is,
to be a successful signal, the terms a good-type rm oers must be such that
a bad-type rm would not want to oer them even if oering them fooled the
lender into thinking it was a good-type rm.
As Aghion and Hermalin note, one term that can potentially separate good
from bad-type rms is the amount of collateral that is oered. Collateral works
as a signal because a bad-type rm knows it is more likely to default on its loan
and, thus, forfeit the collateral than is a good-type rm. At some point the
amount of collateral at risk is so great that the bad-type rm is unwilling to
oer it, even if, by failing to oer it, it reveals itself to be a bad-type rm.
A problem with signaling in this way, however, is that the good-type could
have to oer an excessive amount of collateral in equilibrium. Unfortunately,
barring an legal limitation, such as limited liability, there is nothing a good-type
rm can do about it. If, however, there is a limited liability law, so that there
is a limit on the amount of collateral that can be pledged, then a more ecient

4 Signaling games, rst studied by Spence (1973), are games of asymmetric information in

which the better informed party takes actions that have the potential to conveysignal
her information to the less well informed party. The classic example (Spence) is a worker who
signals information about her quality to potential employers through the amount of education
she acquires. An equilibrium of a signaling game is called separating if the equilibrium actions
of the informed player vary with her information (e.g., workers who know themselves to be
more talented acquire more education than workers who know themselves to be less talented).
A pooling equilibrium is one in which the equilibrium actions of the informed player do not
vary with her information (e.g., all workers get the same level of education).
Limited Liability 23

outcome can be attained: Because, given the law, not oering an excessive Aghion, Philippe
Hermalin,
Benjamin
E.
amount of collateral is no longer evidence that a rm is a bad-type rm, a
good-type rm can oer less collateral without being seen to be a bad-type
rm. See Aghion and Hermalin for details.
In terms of corporations, limited liability precludes rms from pledging the
assets of their shareholders as collateral; that is, there is a limit on the amount
of collateral that can be pledged, which could yield more ecient outcomes than
would otherwise be possible.
24 Why Corporations?
 Modigliani,
Franco
Miller, Merton
Modigliani-Miller
Theorem
costly state
verification|textsl
Townsend,
Robert
M.

Lecture 3

Governance and Securities

W einsaw with the Modigliani-Miller Theorem (Theorem 1 on page 9) that,

a simplied, highly stylized world, the division of claims on the rms
returns between debt and equity was irrelevant for the value of the rm. When,
however, governance issues exist, the Modigliani-Miller theorem may no longer
hold; that is, governance issues mean that some division of claims lead to higher
rm value than other divisions.

3.1 Costly State Verication

One of the assumptions underlying the Modigliani-Miller Theorem is that all
claims will be honored without problem. If, instead, there was a potential
for some claims not to be honored, then some securities would be intrinsically
less valuable than others. The value of the rm (the sum of the value of the
securities) would, therefore, vary with the division of claims between securities.
One source of potential problems in paying claims is determining what the
returns are. If, for instance, investors do not automatically observe the rms re-
turns, then a problem can arise if the managers can, then, hide (retain) returns.
This is a problem known as costly state verification and was rst analyzed by
Townsend (1979).
Specically, there is a manager/owner and an investor. The rm cannot
produce unless the manager/owner acquires I > 0 in nancing from the investor.
If nanced, the rm returns R [R0 , R1 ], where R is stochastic and R0 < I <
R1 . Assume that while the manager/owner observes the realized value of R
without error, the investor does not. At a cost, c, however, the manager/owner
can verify the returns to the investor (e.g., have an audit performed). Assume
R0 > c > 0.1 For convenience, take all parties to be risk neutral.
A security in such a world has two components. One is a repayment function
() : [R0 , R1 ] R, which states the amount that the manager/owner must
1 Making the verication cost less than the minimum possible return is done, here, for

convenience. It is possible to have a costly state verication model without this assumption.

25
 26 Governance and Securities

revelation principle|textsl repay the investor if she announces that the rms returns are R. For example,
Pr@$\Pr\\mathcalE\$
(probability
of
event
$\mathcalE$) a simple equity function would be (R) = R, where 0 < < 1 and R is the
E announced return. Because the investor cannot observe the return himself, it
conditional@$\E\X"|\mathcalE\$
(expectation
of
$X$
conditional
is possible that the announced value, R, and the actual value, R, are not the
on
event
$\mathcalE$
occurring)
same. For that reason, a security has a second component, namely a verication
trigger. Specically, the contract stipulates a range of returns, R, such that if
R R [R0 , R1 ], then the manager/owner must verify.
There is a theorem in economics known as the revelation principle that
establishes that there is no loss of generality in restricting attention to contracts
that, in equilibrium, induce the informed party (here, the manager/owner) to tell
the truth (here, report R honestly).2 Using the revelation principle, consider an
R outside of R; that is, an R R, where R is the set of R not in R (i.e., R is the
set of R for which the manager/owner does not have to verify). For any R R,
the repayment function must be a constant; that is, (R ) = (R ) for any
R , R R. Why? Well suppose not, and suppose that (R ) < (R ). Then
when R is realized, the manager/owner would do better to lie and claim R was
realized (she would have to repay less). Hence, such a repayment schedule would
not induce the manager/owner to tell the truth. But, by the revelation principle,
we know we can restrict attention to contracts that do induce her to tell the
truth. By contradiction, then, we have that it must be that (R ) = (R ) for
any R , R R.
Consider, next, security design. An equity contract requires that the man-
ager/owner repay a constant share of the net returns (returns minus any veri-
cation costs). Because such a repayment function varies with R for all R, such
a security can be consistent with the revelation principle only if R = [R0 , R1 ];
that is, only if there is always verication. Hence, the value of the equity sold
to the investor is (ER c) and the value of the equity retained by the man-
ager/owner is (1 )(ER c), so the total value of the rm is

ER c . (3.1)

Consider, instead, a simple debt contract with a face value of D. This means
that the rm is to repay the investor D if feasible (i.e., R D) and to turn over
the net returns to the investor otherwise (i.e., if R < D). Observe that there is
no need to require verication for any R Dall a creditor needs to know is
that the debtor has enough to repay him, not how much the debtor has. Hence,
[D, R1 ] R. Because the investor gets all the returns if R < D and the returns
obviously vary with R, the revelation principle means that we are required to
have verication for R < D. Hence, R = [R0 , D) and R = [D, R1 ]. The value
of debt is

VD Pr{R D}D + Pr{R < D}(E{R|R < D} c) ,

where Pr{} denotes probability of and E{R|R < D} is the expected value of
R conditional on R being less than D. The value of the equity retained by the
2 See, for example, Gibbons (1992, 3.3) for a discussion.
Costly State Verification 27

manager/owner is SEC@\SEC\
(Securities
and
Exchang
Commission,
\US)
Enron
VE Pr{R D}(E{R|R D} D) + Pr{R < D} 0 .

The value of the rm is VD + VE , which is3

V Pr{R D}(D + E{R|R D} D) + Pr{R < D}(E{R|R < D} c)

= Pr{R D}E{R|R D} + Pr{R < D}E{R|R < D} Pr{R < D}c
= ER Pr{R < D}c . (3.2)

Because Pr{R < D} 1, expression (3.2) is never less than expression (3.1).
Indeed, unless D R1 (in which case, note, a debt contract becomes equivalent
to an equity contract with = 1), it must be that expression (3.2) is strictly
greater than expression (3.1). In other words, the value of the rm is greater if
it nances itself with debt than with equity. Because of costly state verication,
the Modigliani-Miller theorem no longer holds.
We have established:
Proposition 1 Given costly state verication, the value of the rm is greater
under debt nancing than under an equity nancing.
In fact, it can be shown that debt nancing of the sort considered here is superior
to any other form of nancing (see Townsend, 1979, for details).
Taken literally, Proposition 1 suggests that we should never see equity nanc-
ing. That conclusion would, however, be to take Proposition 1 too seriously.
There are a number of real-life complications that limit its applicability:

1. Verification can be obligatory. There are numerous regulations,

such as those imposed by regulatory bodies such as the sec (Securities
and Exchange Commission) in the us, that essentially require rms to
verify no matter what. If verication must occur no matter what, then
it is not an economic cost (recall the discussion of opportunity costs in
Lecture 1) and, thus, irrelevant for decision making (e.g., whether to debt
or equity nance).
2. Taxes. The tax authorities may require verication, which can then
be used as verication for nancing purposes. If, then, verication is
obligatory for tax reasons, then the argument is similar to point 1.
3. Laws against fraud and the difficulty of concealment. What
motivated our analysis is the fear that the manager/owner would lie about
what the returns are. Legally, that would be a form of fraud and subject
the manager/owner to prosecution. Moreover, it would be hard in a large
organization for such a fraud to be concealed, as for instance Enron illus-
trates. On the other hand, Enron does illustrate that real-life managers
3 Observe that (3.2) follows from the preceding line because EX = Pr{E}E{X|E} +

Pr{E}E{X|E}, where X is any random variable, E is any event, and E is the event that
event E does not occur.
 28 Governance and Securities

free cash flow|textsl do, apparently, misreport returns; hence, there is clearly some validity to
Jensen,
Michael
C.
Modigliani-Miller
Theorem the costly state verication model (i.e., Proposition 1).

4. Equity retains pluses relative to debt. Finally, keep in mind that

we have focussed on only one dimension along which to compare debt
against equity. There are other dimensions along which equity is superior
(e.g., it does not expose the investor to asset-substitution problems).

Although, as just discussed, Proposition 1 is not 100% denitive on the

matter, it does, nonetheless, oer important insight in the relation between
governance and security design. As Enron and numerous other scandals illus-
trate, there are real-life verication problems. Moreover, even if we dont take
literally the idea that the manager/owner is pocketing funds in excess of those
reported, it could well be that managers mis-allocate funds within the rm
(free cash flow), spending them in ways that dont necessarily benet investors
(e.g., fancy headquarters, donations to charities, empire building, etc.). What
Proposition 1 demonstrates is that debt can be a cost-eective way to limit such
agency problems. This is, in part, what lies behind Michael Jensens call for
rms to increase their leverage (see, e.g., Jensen, 1986).

3.2 Project Selection

In the Modigliani-Miller Theorem there is no issue of where the rms returns
come from. It is, implicitly, assumed that the rm pursues the same strategy
regardless of its nancing. This, however, need not be true.
Consider again a manager/owner who needs to obtain nancing for her rm.
Assume that returns are veriable without cost. Assume her rm again requires
I > 0 in nancing to start.
Once started, the manager/owner has a choice between two projects (strate-
gies), a safe project, that returns S for sure; and a risky project, that returns
R, where R is stochastic. For simplicity, assume that R = R1 with probability
and that R = R0 with probability 1 , where 0 < < 1. Assume the
following:

R0 < I < S < R1 and

I < ER < S.

The second point tells us that rm value is maximized by the rm pursuing the
safe project.
Consider what happens if the manager/owner nances her rm using debt.
The face value of the debt, D, must be at least I. Consider, then, the expected
payos for the manager/owner between the two projects. If she chooses the safe
project her payo is

YS S D . (3.3)
Project Selection 29

If she chooses the risky project her payo is asset substitution|textsl

YR (R1 D) . (3.4)
Observe, absent any additional information, we cant be sure that YR is less
than YS . For example, were it the case that R0 = 0, = 3/5, I = 60, S = 90,
R1 = 120, and D = I, then
YS = 30 and
YR = 36 .
That is, if the face value of the debt were 60, which would certainly be appro-
priate if the investor thought the manager/owner would pursue the safe project,
then the manager/owner would, in fact, have an incentive to pursue the risky
project.
Of course, the investor wouldnt be so foolish as to not recognize this problem
known as an asset substitution problem. Because his expected return is 3D/5
(or 36 in this example if D = I) should the manager/owner pursue the risky
project, the investor wont lend 60 in exchange for debt with a face value of 60.
The investor will insist on a higher face value. Observe, from expressions (3.3)
and (3.4), that as D increases, YS falls faster than YR . Hence, there is no higher
face value such that the manager/owner will choose the safe project over the
risky project.
Knowing, then, that at any D 60 = I, the manager/owner will choose the
risky project, the only question is whether there is a D such that the investor
would be willing to invest knowing that the manager/owner will choose the risky
project. The expected value of debt in this example is 3D/5 so
5
D I = 100
3
if the investor is willing to lend. Because 100 < 120, the manager/owner is
willing to accept the loan under those terms (YR > 0).
So, were the manager/owner to raise money through debt, the outcome
would be that she issues debt with a face value of 100 and takes the risky
project. The value of the rm is ER = 72,4 which is less than its value were the
safe project taken, 90.
Suppose, instead, that the manager/owner raised her funding through equity.
Now, the investor gets of the returns and the manager/owner gets 1 ,
0 < < 1. If the manager/owner chooses the safe project her payo is
ZS (1 )S . (3.5)

If she chooses the risky project her payo is

ZR (1 )R1 + (1 )(1 )R0 = (1 )ER . (3.6)

4 Expected value of the manager/owners equity is Y = (120 100) 3/5 = 12 and the
R
expected value of debt is 60.
 30 Governance and Securities

Modigliani-Miller
Theorem Because S > ER, it follows that expression (3.5) is always greater than expres-
sion (3.6); that is, the manager/owner will always choose the safe project.
Understanding this, the investor will be willing to fund the rm if S I.
Using the numbers of our example, this means = 2/3. Observe that the
manager/owners payo is 1/3 90 = 30; hence, she does better with equity
nancing over debt nancing. Moreover, the value of the rm is greater; its 90
under equity nancing, but only 72 under debt nancing.
We have established:

Proposition 2 It is possible, given the danger of asset substitution, that the

value of a rm is greater with equity nancing than with debt nancing.

Note that if the numbers were dierent, so that YS YR , then the man-
ager/owner would choose the safehere, value-maximizingproject regardless
of how the rm is nanced. Given that her action is, now, the same regardless
of nancing, we can then employ the Modigliani-Miller Theorem to conclude
that the value of the rm is the same regardless of nancing. So, if YS YR ,
the value of the rm is no less with equity nancing, while, as seen, the value is
greater with equity nancing if YS < YR , we can conclude:5

Proposition 3 Given the possibility of asset substitution and assuming costless

state verication, then the value of the rm is always at least as great under
100% equity nancing as it is under any other nancing.

3.3 So What is it? Debt or Equity?

Costly state verication argues for using 100% debt nancing. Project selection
(asset substitution) argues for using 100% equity nancing. How to reconcile
these two forces?
Using the numerical example from the last section, observe that the value
of the rm, when state verication is c > 0, is 72 2c/5 (from expression (3.2))
if the rm is debt nanced. If the rm is equity nanced, then it is 90 c.6
Therefore, if c < 30, equity nancing yields greater rm value; if c > 30, debt
nancing yields greater rm value (despite asset substitution); and if c = 30,
then either kind of nancing works. In other words, the choice of nancing
method depends on weighing verication costs against asset-substitution costs.

5 To be precise, this argument isnt a full proof because it relies on a simple model. Nonethe-

less, a full proof can readily be established: Because the manager/owners payo under equity
nancing is always (1 )E{R|strategy chosen}, it is clear that the manager/owner always
has the incentive to maximize expected return given equity nancing, which means that max-
imum rm value is always attained under equity nancing. Because the best is achieved, no
other form of nancing can do better.
6 Of course, if the safe project always returns 90, one might ask why verication is needed.

But in a more elaborate model we could have the return from the safe(r) project be stochastic
too.
Application: Savings & Loans 31

In a more general setting (one beyond the scope of these lectures), one savings & loan|textsl
thrift|seesavings
\&
loan
can develop models in which, because of asset substitution and costly state savings
and
loan@\SnL\
(savings
a
verication, both debt and equity nancing are utilized in equilibrium.
The basic idea, however, that the tradeo is between state verication and
asset substitution suggests the following about nancing decisions in a world in
which corporate governance matters:

If managerial project selection is relatively limited or the project choice

set narrow, while returns are highly variable, then the presumption is that
this rm should be mostly debt nanced.

If managerial project selection is wide open and the project choice set
large, while returns given project selection readily audited and veried,
then the presumption is that this rm should be mostly equity nanced.

If managerial project selection is wide open and the project choice set
large, while returns are quite variable and costly to verify, then there
could be some optimal internal mix of debt and equity nancing.

If managerial project selection is relatively limited or the project choice

set narrow, while returns are fairly constant and easy to verify, then this
rm lives, approximately, in the world of Modigliani-Miller, and, from a
governance perspective,7 it is fairly immaterial how this rm is nanced.

3.4 Application: Savings & Loans

Financial institutions, such as banks and savings & loans, are highly leveraged
rms. Deposits in such institutions are debt; that is, your nancial institution
is borrowing the money youve deposited. Although there are, no doubt, many
reasons for why nancial institutions are highly leveraged (as opposed, say,
to selling you shares in exchange for your deposits, as money-market funds
do), having to do with history and subsequent regulation, one explanation is
that, historically, many nancial institutions were engaged in limited lines of
business. The dangers of asset substitution were relatively small, while the cost
of verication relatively large.
A particular form of nancial institution in the us is the savings & loan (also
called a thrift). Until deregulation in the 1980s, savings & loans were restricted
to the business of taking deposits and making real estate loans, primarily for
housing purchases. Within the us, savings & loans (s&ls) are organized in one
of two ways. Some are stock rms, that is they have shareholders (although their
primary source of capital remains debti.e., deposits). Others are mutuals. In
a mutual, the depositors and, in some instances, the borrowers are the titular
owners.
7 There can be other issues beyond governance, such as tax treatment of debt, dividends

(equity), and capital gains (equity), that can cause one form of nancing to dominate another.
 32 Governance and Securities

Hermalin,
Benjamin E. Given deposit insurance, which protects them against downside risk, and the
Wallace,
Nancy
E.
debt
overhang lack of any upside gain, there is little motive for depositor/owners of a mutual
to engage in any monitoring or oversight of the mutuals management. Hence,
it is to be expected that mutuals will be run less well than stock s&ls.
Hermalin and Wallace (1994) investigated that expectation using quarterly
data on us s&ls from 1986 to 1988 (the period following deregulation). Con-
trary to that expectation, they found, instead, that stock s&ls were both more
likely to fail and more likely to perform poorly (appear not to maximize rm
value). Further investigation oered an explanation: When Hermalin and Wal-
lace also controlled for the types of investments (loans, etc.) made by the s&ls
that is, their project choicesthen, consistent, with expectation, stock s&ls
did better than mutuals. The explanation for these results is asset substitution.
Once deregulated, the stock s&ls had an incentive to pursue strategies that
were good for shareholders, but which did not necessarily increase total rm
value. In terms of the example in Section 3.2, the stock s&ls were pursuing
the risky strategy over the safe (traditional) strategy. Not surprisingly, then,
they were both more likely to fail and they were not maximizing rm value. As
Hermalin and Wallace summarize their results:

. . . stock institutions are better at resolving the standard agency

conict between owners and managers, but worse at resolving the
asset-substitution conict between shareholders and debtholders
(depositors).

In other words, deregulation, which allowed s&ls to go into many more lines of
business (expanded their choice of projects) led to an asset-substitution problem.
This asset-substitution problem was, in turn, behind much of the us s&l crisis
of the late 1980s and early 1990s.

3.5 Control Issues

One of the events that can occur if a rm defaults on its debts is that control of
the rm passes from shareholders to debtholders. That is, as was noted earlier,
debtholders have a senior claim on the rms assets in the event of default.
A problem with the change of control that default can bring is that it can
be inecient. There are a number of potential reasons for this:

1. Asset substitutionthe other way. Just as the manager/owner

above could take decisions that were to her benet at the expense of
debtholders, debtholders, once in control, can take decisions that are to
their benet at the expense of shareholders. Typically, this would mean
taking safe, but non-value-maximizing actions. Consequently, the rms
value may not be maximized once debtholders are in control.

2. Debt overhang. Generally, new debt or nancing is subordinate to

existing debt; that is, the new investors will be paid after the claims of
Exploitation of Minority Shareholders 33

the existing debtholders. Consequently, the following problem can arise: minority shareholder|textsl
self dealing
Suppose, if liquidated, the rm is worth V
, where V
< D, D being the face
value of existing debt. If, however, I new funds are invested in the rm, it
will return Vh , where Vh I > V
; that is, it is ecient to invest more in
the rm. A debt overhang problem arises, however, if Vh D < I, because
then the new investors (those supplying I) cannot be repaid adequately.
Hence, the rm will be ineciently liquidated.

3. Bargaining under asymmetric information. One possible resolu-

tion to the debt overhang problem would be for the existing debtholders
to negotiate away some of their seniority. But if they have superior infor-
mation about the prospects of the rm, then new investors may reasonable
remain worried that they wont get a fair return on their investment. Sim-
ilarly, asymmetries of information between management and debtholders
can prevent the parties from pursuing ecient ways out of default.

Such control problems represent additional motives for rms to prefer equity
to debt; that is, they are additional concerns to be balanced against costly state
verication and asset substitution.

3.6 Exploitation of Minority Shareholders

Another potential governance issue that aects nancing has to do with the
treatment of minority shareholders. A simple example illustrates the danger.
Suppose there are two shareholders. As before, one is the manager/owner. The
other holds a share < 1/2; that is, he is a minority shareholder. Suppose
that the rm in question has the choice of buying from one of two suppliers.
One, supplier is independent of the rms two shareholders. The other supplier,
M , is owned entirely by the manager/owner. Observe that the manager/owner
has a motive to steer the rms business to M : Every additional dollar that
the rm spends at M is dollar of prot for her (the value of M goes up by a
dollar, the value of the rm goes down by a dollar, of which only 1 is the
manager/owners, so the manager/owners net is $1 $1(1 ) = $).
This problem could also arise with debt, to the extent that a creditor cant
pierce the corporate veil. But, while laws protect both shareholders and debthold-
ers against such self dealing, the laws protecting debtholders are typically stronger.
To the extent that laws are imperfect protections against the exploitation
of minority shareholders, this will reduce the value of minority shareholdings.
Such a reduction means that an entrepreneur selling a minority fraction of her
business could have to sell those shares at a discount vis-a-vis the price they
would command if there werent the danger of minority exploitation. Hence,
if she wants a better price, she may have to yield more control. On the other
hand, as we will see, there can also be advantages to large-bloc shareholding,
which counter-balances this eect.
34 Governance and Securities
 agency!principal|textsl
agency!agent|textsl
agency!hidden action|textsl
agency!hidden
information|textsl

Lecture 4

Agency

A employs
gency refers to a situation in which one party, called the principal,
a second party, called the agent, to perform some task for her. 1

There are two kinds of agency problems: hidden action and hidden information.
In the former, the action taken by the agent is unobservable bythat is, hidden
fromthe principal. To the extent that the principal and agent have dierent
inherent preferences concerning the action that the agent takes, the fact that the
action is hidden will prove to be a problem. In a hidden-information problem,
the issue is that the agent possesses relevant information that the principal
does not observethat is, which is hidden from her. Because this information
is relevant for decisions, a problem can arise.
An example of a hidden-action agency problem is where a ceo prefers one
course of action to another because it requires less work, while the shareholders
prefer the second course of action to the rst because it yields greater expected
prots.
An example of a hidden-information agency problem is where a ceo knows
how dicult it will be to achieve cost savings, but the shareholders do not.
Because the shareholders must compensate the ceo more, the more dicult is
his job, the ceo has an incentive to claim that cost savings are very dicult,
regardless of the truth. The shareholders would prefer to achieve cost savings
while avoiding overpaying the ceo.

4.1 Hidden-Action Agency

Hidden action refers to situations in which the agent takes actions that are
payo-relevant for the principal, but which the principal cannot observe. Table
4.1 oers some examples.

1 Following convention, principals are shes and agents are hes. If youre married, youll

understand why.

35
 36 Agency

agency!hidden
action!timing Principal Agent Problem Solution

a@$a$
is
an
action (usually)
Employer
A@$\mathcalA$
(set
of
feasible
actions) Employee Induce employee to take Base employees
common knowledge actions that increase compensation on employers
employers prots, but prots.
which he nds personally
costly.
Plainti Attorney Induce attorney to expend Make attorneys fee
costly eort to increase contingent on damages
plaintis chances of awarded plainti.
prevailing at trial.
Homeowner Contractor Induce contractor to Give contractor bonus for
complete work (e.g., completing job on time.
remodel kitchen) on time.
Landlord Tenant Induce tenant to make Pay the tenant a fraction of
investments (e.g., in time or the increased value (e.g.,
money) that preserve or share-cropping contract).
enhance propertys value to Alternatively, make tenant
the landlord. post deposit to be forfeited
if value declines too much.

Table 4.1: Examples of hidden-action agency problems.

4.1.1 The Basic Model

Although there are many variations of the hidden-action agency model,2 we will
focus here on one. It has the following timing:

1. A principal and agent meet and enter into a contract with each other. At
the time they enter into this contract, they are symmetrically informed.

2. The agent then takes an action, a, from a set A. While the set A is
common knowledge (that is, known by both parties),3 the action chosen
by the agent is known only to the agent.

3. An outcome, x, drawn from the set X R is realized. The x that is

realized is a stochastic function of a. The outcome x is observable by
both parties and veriable (i.e., can serve as a contract contingency).

4. The principal pays the agent according to the initial contract, which typ-
ically stipulates a wage, w, that is contingent on the realized value of
x.
2 See Laont and Martimont (2002) for more details. A somewhat idiosyncratic analysis,

which can be found on the web, is Caillaud and Hermalin (2000).

3 To be precise, information is common knowledge if it is known by both parties, each party

knows the other party knows, each party knows that the other party knows that he or she
knows, and so forth.
Hidden-Action Agency 37

The bargaining game in step one is typically assumed to be take-it-or-leave- bargaining!take-it-or-leave-it

UR@$U_R$
(reservation
it bargaining in which the principal proposes a contract to the agent that the utility)
agent can either accept (take) or reject (leave).4 Well assume that bargaining reservation utility|textsl
game here. Observe that if the agent rejects the oered contract, the game ends. agency!principal!risk
neutrality

agency!agent!risk aversion
of
To know whether the agent accepts or rejects, we need to know what the individual rationality
agent receives if he rejects the oered contract. Let his payo if he rejects be constraint|textsl
UR , a quantity known as his reservation utility. incentive compatibility
constraint|textsl
To know how the parties will behave, we need to know the parties utility
functions. As is typically done, assume that the principal is risk neutral, with a
utility function xw, where w is her payment to the agent.5 As is also typically
done, assume that the agent is risk averse, with a utility function U (w) K(a),
where U () exhibits the properties of risk aversion and K() is the agents cost-
of-action function.6
Let W (x) be the amount the contract stipulates the agent will be paid if
outcome x is realized.7 In equilibrium, W () will induce the agent to take some
action a A. Knowing this, the agent will accept the contract in step one if
and only if   
E U W (x) |a K(a) UR , (IR)
  
where E U W (x) |a means the expected utility over money conditional on
the agent having chosen action a. Expression (IR) is known as the individual
rationality constraint; it simply says that the agent will accept the contract if
and only if he does at least as well by accepting as he would do by rejecting.
Which action does the contract W () induce? Well the one that maximizes
the agents expected utility; that is, the agent will choose action a if and only if
     
E U W (x) |a K(a) U W (x) |a K(a) for all a = a . (IC)

This last expression is known as the incentive compatibility constraint.

The principals problem can be seen as choosing an action, a , to induce and
a contract, W (), that will (i) induce that action; (ii) be acceptable to the agent
(satisfy (IR)); and (iii) maximize her expected utility, which is

E{x|a } E{W (x)|a } .

4.1.2 A Simple Variant

For our purposes in this lecture, we dont need as general a model as the one
set forth so far. Hence, we will consider the following version:

4 A notable exception is Hermalin (1992), in which the agent is given all the bargaining

power.
5 A more general formulation would make the principals utility bw, where b is some benet

that is a function (possibly stochastic) of a. We have no need for that greater generality here,
however.
6 That is, U : R R and K : A R.
7W : X R.
 38 Agency

A
state-contingent
commodities!theory
of = {0, 1}; that is, there are only two possible actions, 0 and 1.
state-contingent
commodities|textsl
K(a) = a.

X = {xL , xH }; that is, there are only two possible outcomes, xL and xH .
Assume xL < xH .

UR = 0; that is, the agents reservation utility is just zero.

In addition, we will assume that the stochastic relation between a and x is

Pr{x = xL |a} = 1 qa and Pr{x = xH |a} = qa ,

where 0 < q < 1. Observe that if a = 0, then the low outcome, xL , is always
realized, whereas if a = 1, then the low outcome is realized with probability
1 q and the high outcome, xH , is realized with probability q.

4.2 A Digression: State-Contingent Commodity

Theory
It is helpful, to facilitate a graphical analysis of the hidden-action agency prob-
lem, to understand state-contingent commodity theory.

4.2.1 Basics of the Theory

A state-contingent commodity is a commodity, or contract, or security that pays
o dierent amounts in dierent states of the world. For example, a share of
stock is a state-contingent commodity, the amount it pays in dividends depends
on how well the rm that issued that share does (i.e., the prot state of that
rm). Another example is an insurance contract: In an accident state of the
world, the contract pays L p + R(L), where L is the loss in the state of the
world have an accident that does damage L, p is your insurance premium,
and R(L) is the reimbursement for loss you receive as a function of loss. There
is also a no accident state, in which the contract pays p.
We will here consider only two-state state-contingent commodities. They
have the form (x, y), where x R is what the commodity pays o if the rst
state of the world is realized and y R is what the commodity pays o if the
second state of the world is realized.
What is the value of a state-contingent commodity? The answer depends on
two issues: The underlying probability of the two states and the risk preferences
of the commoditys owner (i.e., whether the owner is a risk-neutral or risk-averse
party). Let 1 be the probability of the rst state occurring and let be
the probability of the second state occurring. If the commodity owner is risk
neutral, then the value of the commodity is its expected value,

(1 )x + y .
State-Contingent Commodity Theory 39

If the commodity owner is risk averse, with utility function V (w), where w is state-contingent
commodities!riskless
money, then the value of the commodity to the owner is his expected utility, state-contingent
commodities!risky
(1 )V (x) + V (y) .

Rather than continue to do all the analyses separately for the risk-neutral
and risk-averse cases, let U () be a generic utility function. If the commodity
owner is risk neutral, then U (w) = w. If he is risk averse, then U () exhibits
diminishing marginal utility of money (recall Section 1.4.2).
Suppose a party has a choice between two state-contingent commodities,
(x, y) and (x , y  ). He will choose (x, y) if

(1 )U (x) + U (y) > (1 )U (x ) + U (y  ) ;

he will choose (x , y  ) if

(1 )U (x) + U (y) < (1 )U (x ) + U (y  ) ;

and he will be indierent if

(1 )U (x) + U (y) = (1 )U (x ) + U (y  ) .

An important kind of state-contingent commodities are the riskless state-

contingent commodities. A state-contingent commodity is riskless if it pays
o the same amount in both states; that is, has the form (z, z). Observe the
value of a riskless state-contingent commodity is just U (z). State-contingent
commodities that are not riskless will be called risky.
Suppose a party likes the risky state-contingent commodity (x, y) as much
as the riskless state-contingent commodity (z, z). Then it must be that

(1 )U (x) + U (y) = U (z) . (4.1)

We can say that equation (4.1) denes the indierence curve that passes through
the point (z, z); that is, that intersects the 45 -degree line at coordinates (z, z).8
Hence, if the party is risk neutral, we have, from equation (4.1),

(1 )x + y = z ; (4.2)

that is, the indierence curve through (z, z) is the set of all state-contingent
commodities that have an expected value of z. Observe, too, from expression
(4.2), that the indierence curves for a risk-neutral party over state-contingent
commodities are straight lines. Rewriting equation (4.2) in slope-intercept form,
we have
1 1
y= x+ z.

8 Recall that an indierence curve denes a set of points in a commodity space that the

relevant decision maker likes equally well; that is, which all give her the same utility.
 40 Agency

fair-odds ratio|textsl The minus one times the slope of that line,
fair-odds line|textsl
1
,

is known as the fair-odds ratio. A line with slope minus one times the fair-odds
ratio is known as a fair-odds line. Note a fair-odds line is an indierence curve
for a risk-neutral party. Because all fair-odds lines share the same slope, they
are all parallel to each other.
Consider expression (4.1) for a risk-averse party:
(1 )V (x) + V (y) = V (z) ; (4.3)
that is, the indierence curve through (z, z) is the set of all state-contingent
commodities that have an expected utility of V (z). Observe, this means that the
certainty equivalent value of any state-contingent commodity on the indierence
curve through (z, z) is z. That is, if we use to denote indierent to and we
write (x, y; ) for the gamble that pays x with probability 1 and y with
probability , then we have
(x, y) (z, z) if and only if CE(x,y;) = z .
What about the shape of a risk-averse partys indierence curves? Observe,
from equation (4.3), that the slope of the curve is
(1 )V  (x) 1 V  (x)

=  , (4.4)
V (y) V (y)
where V  () is the derivative of V () or marginal utility. Because risk aversion
means the utility function exhibits diminishing marginal utility, observe y > x
implies V  (y) < V  (x) and y < x implies V  (y) > V  (x). Hence, the slope of
the indierence curve at (x, y), x < y, is steeper than the fair-odds line through
(x, y); and the slope of the indierence curve at (x, y), x > y, is atter than the
fair-odds line through (x, y). See Figure 4.1.
Observe from expression (4.4) that if x = y, then the slope of the indierence
curve is just minus one times the fair-odds ratio. This is clearly true regardless
of the utility function, provided its dierentiable, so we have:
Result 1 At the 45 , all indierence curves derived from dierentiable utility
functions have the same slope, namely minus one times the fair-odds ratio.
Another insight into the shape of a risk-averse partys indierence curves
can be gained by considering a risky state-contingent commodity (x, y). Let z
satisfy equation (4.3). Then, because z is the certainty equivalent of the gamble,
we know
(1 )x + y > z ;
that is, (x, y) must lie on a higher fair-odds line than the fair-odds line through
(z, z). Again see Figure 4.1. Yet another way to put this is if a risk-averse
party were indierent between (x, y) and (z, z), then a risk-neutral party would
strictly prefer (x, y) to (z, z).
State-Contingent Commodity Theory 41

payoff
in state 2 risk-averse
indifference 45
curve

(x,y)
(z',z')

(z,z) fair-odds lines

(risk-neutral
indifference curves)

payoff
in state 1

Figure 4.1: As illustrated, the risk-neutral is indierent between (x, y) and (z, z).
His certainty equivalent value for (x, y) is thus z. The expected value
of (x, y) is, however, greater; it is z  .

4.2.2 Ecient Trade between a Risk-Averse and a Risk-

Neutral Party
I noted at the end of Section 1.4.2 that eciency required that, when trade is
possible, risk be passed from a risk-averse party to a risk-neutral party. This
can readily be shown using state-contingent commodities.
Consider Figure 4.2. In this gure, suppose there are two parties, a risk-
averse party who owns state-contingent commodity (x, y) and a risk-neutral
party who owns state-contingent commodity (z, z) (equivalently, the risk-neutral
party simply has z units of currency). Observe that if they swapped (equiva-
lently, the risk-neutral party purchased (x, y) from the risk-averse party for z
in cash), then both parties would better o. The point (z, z) lies on a higher
indierence curve for the risk-averse party than does (x, y). For the risk-neutral
party, the point (x, y) lies on a higher indierence curve than (z, z). On other
way to see this last point is that the state-contingent commodity (x, y) has an
expected value of z  , which exceeds z.
As noted earlier, this is a general result:

Result 2 If there is a risky state-contingent commodity, a risk-neutral party,

and a risk-averse party, then the ecient allocation is for the risk-neutral party
to own the state-contingent commodity. A nal allocation in which the risk-
42 Agency

payoff
in state 2 risk-averse
indifference
45
curves

(x,y)
(z',z')

(z,z)
fair-odds lines
(risk-neutral
indifference curves)

payoff
in state 1

Figure 4.2: The transaction in which the risk-averse party transfers the state-
contingent commodity (x, y) to the risk-neutral party in exchange for
z in cash. Observe this trade makes both parties better o.

averse party is left owning a state-contingent commodity is inecient.

4.3 Hidden-Action Agency Continued

We are now in position to analyze the hidden-information agency problem set
forth earlier. Observe that if the agents compensation is contingent on per-
formance (i.e., the realization of x), then his employment contract is a state-
contingent commodity: If the L state is realized, he is paid W (xL ); if the H
state is realized,
  W (xH ). So the contract is the state-contingent com-
he is paid
modity, W (xL ), W (xH ) , with the probability of the second (high) state being
qa.

4.3.1 First-best Benchmark

To begin, lets suppose that the agent does what the principal asks without the
need for monetary incentives. This does not, of course, describe the real world,
but it is a useful benchmark against which to judge the more realistic situation
in which the agent requires monetary incentives.
Because the principal is risk neutral, while the agent is risk averse, we know
that, to have a fully ecient outcome, the state-contingent commodity that the
Hidden-Action Agency Continued 43

 
contract represents, W (xL ), W (xH ) , must be riskless. That is, we can only
have full eciency if W (xL ) = W (xH ).
Consider such a riskless contract, where W (xL ) = W (xH ) = w. Because the
agent simply does what he is told, we can ignore the (IC) constraint; the only
constraint we need to worry about now is the (IR) constraintwill the agent
actually agree to work for the principal?
The (IR) constraint is, recall,
  
E U W (x) |a K(a) UR (IR)

if action a will be chosen; which, in our simple variant, is

   
(1 qa)U W (xL ) + qaU W (xH ) a 0 (4.5)

or, if W (xL ) = W (xH ) = w, just

U (w) a 0 . (4.6)

The principal wants to maximize her expected payo. She is constrained in

that the agent must agree to work for her; that is, she wishes to solve:

max (1 qa)(xL w) + qa(xH w) (4.7)

a{0,1},wR

subject to expression (4.6).

Clearly, the principal wants to make her payment to the agent as small as
possible. Hence, the constraint (4.6) must bind; that is, be an equality. So we
know,

U (w) = a ; or
w = C(a) , (4.8)

where C() is the inverse

of U (); that is, if U (w) = u, then C(u) = w. For
instance, if U (w) = w, then C(u) = u2 . Because utility functions are increas-
ing, C() is an increasing function. Because the agent is assumed to be risk
aversethat is, have a diminishing marginal utility of incomemarginal C()
is also increasing.9
Because the (IR) constraint is binding, we can substitute it into the prin-
cipals optimization program. Consequently, the principals problem can be
expressed as
max (1 qa)xL + qaxH C(a) . (4.9)
a{0,1}

The agency problem wouldnt be very interesting if the solution to (4.9) was
a = 0; such a solution would mean that the principal doesnt want the agent to

9 The inverse of a concave function is convex; hence, because U () is concave (equivalently,

exhibits diminishing marginal utility), C() is convex (equivalently, has an increasing margin).
 44 Agency

fixed-wage contract work hard. The agent, not surprisingly has no desire to work hard. Thus, in a
world in which the principal didnt want the agent to work hard, there wouldnt
be any tension between principal and agent. Because our focus is on situations
in which there is a tension, we will assume that a = 1 is the solution to (4.9);
that is, the assumption is that
 
q (xH xL ) C(1) C(0) > 0 . (4.10)

In words, the expected incremental gain from a = 1 over a = 0, q(xH xL ), is

assumed to be greater than the incremental cost of a = 1 over a = 0, C(1)C(0).
In terms of state-contingent commodities, this solution corresponds
 to the
principal providing the state-contingent commodity C(1), C(1) . See Figure 4.3
below.

4.3.2 First-best not Attainable

Unfortunately, there is no reason to expect the agent will simply do what the
principal asks without the appropriate monetary incentives. That is, it is more
realistic to suppose that if the principal wishes the agent to work hard, that
is choose a = 1, then the contract oered will have to be incentive compatible
(satisfy (IC) on page 37).
Observe, rst, that a xed-wage contract will fail to satisfy (IC). In our
simple variant, if the agent is to work hard, (IC) requires
     
(1 q)U W (xL ) + qU W (xH ) 1 U W (xL ) . (4.11)

But if we substitute w for both W (xL ) and W (xH ), we see

(1 q)U (w) + qU (w) 1 = U (w) 1

< U (w) .

Hence, it will clearly be necessary to have W (xH ) = W (xL ) if the agent is to

be induced to work hard (i.e., choose a = 1).
But this means that the incentive contract will be equivalent to a risky state-
contingent commodity. And we know from previous analysis, that it cannot
be fully ecient to leave the risk-averse party with a risky state-contingent
commodity. We can, therefore, conclude:

Result 3 When the principal seeks to induce the agent to take the harder action,
then no feasible contract (i.e., a contract satisfying (IC)) can be fully ecient,
because any feasible contract exposes the agent to risk.

4.3.3 Solving for the Second-best Contract

Observe that expression (4.11) can be rewritten as
    1
U W (xH ) U W (xL ) + .
q
Hidden-Action Agency Continued 45

Hence, inverting both sides, we have

 
  1
W (xH ) C U W (xL ) + . (4.12)
q
Because 1/q > 0 and C() is an increasing function, expression (4.12) tells us
that   
W (xH ) > C U W (xL ) = W (xL ) .
In other words, under any incentive compatible contract, the agent must get
paid more for the good outcome than for the bad outcome.
The idea that the better outcomes lead to higher pay is a fairly standard
result and tends to hold generally in models in which there are more than two
possible outcomes.10
Recall that a feasible contract must also satisfy the (IR) constraint, which
is expression (4.5) with a = 1:
   
(1 q)U W (xL ) + qU W (xH ) 1 0 . (4.13)
Observe that if we move the 1 to the other side, then the (IR) constraint,
expression (4.13), can be read as saying that the set of feasible contracts must
be those that yield expected utility at least 1. Or, equivalently, they must be
contracts that have a certainty equivalent value of C(1).
We can now solve for the optimal contract for inducing a = 1. Consider
Figure 4.3. On it are plotted both the (IC) constraint, expression (4.12), and
the (IR) constraint, expression (4.13). To satisfy the (IC) constraint, a contract
must lie on the (IC) curve or above it (the areas shaded blue and green in
Figure 4.3). To satisfy the (IR) constraint, a contract must lie on the (IR) curve
or beyond it (the areas shaded green and yellow). Observe that the green region
and its borders are the only contracts that satisfy both constraints. Hence, a
feasible contract must lie in the green area or on its borders.
Consider the principals preferences. Her expected wage bill is
(1 q)W (xL ) + qW (xH ) .
Observe that the set of contracts that produce a given expected wage (e.g., z  )
all lie on the fair-odds line through that expected wage (e.g., through (z  , z  )).
Because the principal prefers to the pay agent less rather than more, all else
equal, it must be that the principals utility is greater for contracts on lower
fair-oddsline. For instance,
 in Figure 4.3, she is better o on the fair-odds line
through C(1), C(1) than the one through (z  , z  ).
Therefore, to determine the optimal contract, we need to nd the contract
in the space of feasible contracts (i.e., the contracts in the green area or its
borders) that is on the lowest possible fair-odds line. From Figure 4.3, we see
that is the contract (wL , wH ), which corresponds to the point where the (IR)
and (IC) curves intersect. For this model with two actions and two outcomes,
this is a general result:
10 Although, to be precise, some additional assumptions are sometimes required to get this

monotonicity result. See Grossman and Hart (1983).

46 Agency

W(xH) IC
IR
45

^ ,w
(w ^
L H)

(z',z')

(C(1),C(1))
Principal's
indifference
curves
C(U(0)+1/q)

W(xL)

Figure 4.3: The set of feasible contracts are those that both lie on or above (IC)
and on or beyond (IR). From the principals perspective, the optimal
feasible contract is (wL , wH ).

Result 4 Suppose there are only two actions and only two possible outcomes.
Suppose too that both outcomes can occur with positive probability if the agent
takes the harder of the two actions. Then the optimal contract for inducing the
agent to take the harder action is the contract that solves both the (IR) and (IC)
constraints as equalities.

The nal step

Were not quite done, however. Observe from Figure 4.3 that the principals
expected wage bill if she induces the agent to choose a = 1 is z  . We need to
check whether, given this cost, the principal still wishes to induce a = 1. Her
expected prots from inducing a = 1 are

(1 q)xL + qxH z  . (4.14)

Her expected prots from inducing a = 0 is

xL C(0) . (4.15)

From expression (4.10), we know only that

(1 q)xL + qxH C(1) > xL C(0) ;

Hidden-Action Agency Continued 47

but because z  > C(1), we dont which of expressions (4.14) and (4.15) is larger tradeoff
between
incentives
\&
in
without checking.
Note that it is possible that z  is so large that the principal opts to induce
a = 0; hence, a potential cost of agency is that the principal gives up and
doesnt try to get eort from the agent.

4.3.4 Comparative Statics

As already observed, the principals expected wage bill if she seeks to induce
a = 1 is greater than C(1), what she would pay if the agent could be trusted to
do what he was supposed to or if the principal could observe the agents action
and force him to take the right action. What explains this dierence?
We can read the answer o of Figure 4.3. The agent is risk averse. Hence, to
get him to accept a risky state-contingent commodity, which is what his com-
pensation contract entails, he must receive compensation for that risk. That
compensation comes in the form of a higher expected wage. Note this risk com-
pensation just osets his loss from the risk, so the agent is not made happier
from this increase in  his expected
 wage (observe (wL , wH ) is on the same in-
dierence curve as C(1), C(1) ). This illustrates a general tradeo associated
with incentive contracts: increasing incentives increases risk and the principal
must pay the agent, in expectation, to compensate for this increased risk. This
tradeo is sometimes described as there being a tradeo between incentives
and insurance (compensation for risk). In Figure 4.3, this additional expected
compensation is z  C(1).
What are the factors that inuence this cost, z  C(1)? From Figure 4.3,
it is clear that this cost is being driven by two factors: How far above the 45
line the (IC) constraint lies and how curved is the (IR) constraint.
Consider the second, if the (IR) constraint got atter, which would corre-
spond to the agent being less risk averse, then the intersection of (IR) and
(IC)
 would take place on fair-odds lines closer to the fair-odds line through
C(1), C(1) ). Hence, the cost goes down as well. Indeed, if the agent were
risk
 neutral, so that the (IR) curve was identical to the fair-odds line through
C(1), C(1) ), then we see there would be no cost of incentives. This makes
sense, the dierence z  C(1) arises because the principal must compensate the
agent for bear risk. If, however, the agent doesnt care about riskhes risk
neutralthan he requires no such compensation.
What about the impact of the (IC) constraint? Observe from expression
(4.12) that, as q gets smaller, the (IC) curve shifts up. Conversely, if q gets
larger, (IC) shifts down. How do we interpret this? We can see q as representing
the correlationor, more accurately, the informativenessof outcome to action
taken.11 At rst pass, it might seem odd to be worried about the informativeness
of the outcome because, in equilibrium, the principal can accurately predict the
agents choice of action from the structure of the game and her knowledge of the

11 To see this, in a loose sense, suppose that agent is randomizing over whether to choose

a = 1 with probability that he does being (0, 1). [Note: in equilibrium, the agent chooses
48 Agency

contract. But thats not the point: The principal is forced to design a contract
that pays the agent based on performance measures that are informative about
the variable upon which she would truly like to contract, namely the agents
action. The more informative these performance measures areloosely, the
more correlated they are with actionthe closer the principal is getting to the
ideal of contracting on the agents action.
In terms of incentive contracting, this last insight tells us that we prefer to
base compensation on more informative signals (performance measures) than
on less informative signals. For instance, one could in theory base a salesper-
sons compensation on the stock price of the company. However, the stock price
moves very little with the eorts of the salesperson, so stock price is not very in-
formative. Indeed, to have such a contract inuence the salespersons behavior,
the expected compensation would have to be huge. In contrast, the number or
value of sales achieved by the salesperson is certainly quite informative about
his actions. Therefore the expected compensation wouldnt have to be so great
if the company used a sales-commission contract to provide the salesperson with
incentives.
It is worth considering the two extremes: one in which q 0; and another in
which q 1. From expression (4.12), if q gets exceedingly small, then the right-
hand side of (4.12) blows up toward positive innity. In terms of Figure 4.3, this
means that the (IC) constraint is shooting o the top of the page; hence, the
cost of providing incentives is shooting o to innityit cannot be protable
to provide incentives. This makes perfect sense: If q = 0, then the performance
measure tells us nothing about what the agent did. Therefore it cannot provide
any incentives.
As q rises toward 1, the (IC) curve falls toward, but doesnt reach the 45
line. However, at q = 1, there is a discontinuity. Now outcome is perfectly
correlated with the action, which means were in a world in which the principal
can eectively see the agents action (its no longer truly hidden). Therefore,
the principal can get the agent to undertake a = 1 for compensation C(1). The
principal just promises C(1) if the agent works hard and threatens some small
payment (or even a ne) if the agent doesnt work hard.

4.3.5 Summary
To sum up concerning hidden-action agency: With hidden-action agency, there
is a tradeo between incentives and insurance; that is, there is an additional cost
to utilizing incentive contracts because the incentive contracts force the agent
to bear risk for which he will demand compensation. If this insurance gets too

a = 1 for sure.] Then, ex post, the probability that he chose a = 1 conditional on x = xL is

(1 q)
(1 q) + 1
by Bayes Theorem. It is readily seen that the this probability is decreasing in q; moreover, it
reduces to nothing has been learnedif q = 0.
Hidden-Information Agency 49

large, then the principal will have to abandon trying to provide the agent with information rent|textsl
incentives.
The size of this compensation for risk is driven by two components, the
level of the agents risk aversion and the informativeness of the performance
measure about the agents action. The more (less) risk averse the agent, the
more (less) expensive providing incentives will be. The more (less) informative
the performance measures, the less (more) expensive providing incentives will
be. This last observation tells us that we want to base the agents compensation
on the most informative performance measures available.

4.4 Hidden-Information Agency

In this section, we consider the problem of hidden information; that is, the sit-
uation in which the agent possesses information that the principal does not and
where this information is mutually relevant for payos. Some examples include
an agent who alone knows the diculty of completing a task for the principal;
or a divisional manager who can conceal information about his divisions invest-
ment opportunities from headquarters; or a team leader with better information
than her teammates about the value of pursuing a given course of action. In
each of these situations, having private information gives the party possessing
it a potential strategic advantage in his dealings with the other party.
For example, consider a divisional manager who has better information about
his costs than corporate headquarters. By behaving as if he had high costs,
the manager can seek to induce headquarters to transfer more funds to his
division than headquarters would if it knew the managers division had low
costs. That is, the manager has an incentive to use his superior knowledge to
capture an information rent. Of course, headquarters is aware of this possibility;
so, because it has the right to set the managers contract, headquarters will
propose a contract that works to reduce this information rent. Our focus here
is on how the contract proposerthe principaldesigns contracts to mitigate
the informational disadvantage she faces.

4.4.1 The Model

To understand the ideas and concepts connected with hidden-information agency,
consider the following model. The ceo of a corporation (the principal, here)
is trying to determine the amount to have the divisional head (the agent) pro-
duce, x, and in exchange for what payment, p. For concreteness, suppose that
the division in question is an upstream division whose output is completely for
internal use in the corporation (e.g., the division makes chips that will go into
the devices that the corporation produces).
For reasons not modeled here, suppose that the managers compensation is
a fraction of divisional prots, so that the manager always seeks to maximize
 50 Agency

type!in hidden-information
agency|textsl
divisional prots.12 Assume divisional prots are
type!space|textsl
iso-profit line p Kt (x) ,

where p is the total amount transferred from (paid by) headquarters, Kt () is

the production cost function, and t {I, E} denotes whether the division is
inecient (I) or ecient (E). The variable t is know as the agents type. The
set {I, E} is the type space; that is, the range of possible types.
The critical assumption is that the manager knows whether his division is
inecient or ecient (i.e., he knows whether t = I or t = E), but the ceo
does not. All the ceo knows is the type space and the probability, f , that the
division is inecient (and, hence, also the probability that it is ecient, 1 f ).
The cost function Kt () satises the following conditions:
Kt (0) = 0 for both t.
Kt (x) > 0 for all x > 0 and both t (i.e., there is positive marginal cost for
x > 0).
Kt (x) > 0 for all x and both t (i.e., marginal cost is increasing in output,
x).
KI (x) > KE 
(x) for all x > 0 (i.e., marginal cost of production is greater
if the division is inecient than if it is ecient).
Observe the rst and last of these assumptions imply that

KI (x) > KE (x) for all x > 0 .

Assume that the value of the divisions output to the ceo is one unit of cur-
rency per unit of output; that is, x units are worth x. Hence, the corporations
prot is x p.
Finally, assume that the division cannot be made to run a decit; that is,
the divisional manager is free to refuse to produce at all if the proposed (x, p)
would lead to losses. If the divisional manager refuses to produce, divisional
prots are zero (i.e., 0 Kt (0) = 0).

4.4.2 Analysis
It is helpful to consider a graphical analysis. Figure 4.4 illustrates.
Because headquarters prots are x p, its iso-prot lines (indierence
curves) in xp space are straight lines with slope 1. Two dierent ones are
illustrated by short dashed lines going through points A and B in Figure 4.4.
Observe that because headquarters likes more x and likes less s, iso-prot lines
more to the lower right of the gure correspond to higher prots than iso-prot

12 This could be because his compensation scheme is an increasing function of divisional

prots or because his reputation and ability to attract new jobs or be promoted depends on
how protable his division is.
Hidden-Information Agency 51

type-I
p
indifference
curve

type-E
indifference
curves

C
E
B
A

x
x*I (f) xFI xFE
Figure 4.4: Ecient production should occur at xF E if the division is ecient and
xFI if it is inecient. But by distorting downward the production tar-
get should the division prove inecient, headquarters can lessen the
information rent it needs to pay if the division is ecient.

lines that are more to the upper left; for example, the iso-prot line through B
corresponds to higher prots than the one through A.
An iso-prot (indierence) curve for the division can be found by nding all
the (x, p) pairs that yield the same prot:
p Kt (x) = ,
where is some constant level of prot. Dierentiating and rearranging, we see
that the slope of an iso-prot line is
dp
= Kt (x) . (4.16)
dx
Consequently, iso-prot curves for the inecient type through a given point are
more steeply sloped than those of the ecient type. For instance, the type-
I indierence curve through point A is more steeply sloped than the type-E
indierence curve through A.
Note also from expression (4.16) that the slope of a divisional indierence
curve does not depend on p. This means that dierent indierence curves for
 52 Agency

information rent a division of a given type are all parallel shifts up and down of each other.
Consider, for example, the type-E indierence curves in Figure 4.4.
Because the divisional head likes more s and likes less x, indierence curves
more to the upper left correspond to higher prot than those to the lower left.
For instance, the top indierence for the ecient type (the one through point
C) corresponds to higher prots than the lowest one (the one through point B).
If the ceo knew how ecient the division was, she would maximize corporate
prots subject to the constraint that the division not make negative prots.
The zero-prot indierence curves for the two types are shown in Figure 4.4;
theyre the indierence curves passing through the origin (since 0 Kt (0) = 0).
Corporate prot maximization then corresponds to the ceo selecting a point
on the relevant zero-prot indierence curve that corresponds to the lowest
possible corporate iso-prot line (remember prots are increasing as we move to
the lower right). Point A illustrates that point if the ceo knew she were dealing
with an inecient division and point B is that point if the ceo knew she were
dealing with an ecient division. The amount that an inecient division would,
thus, produce is shown as xF I and the amount that an ecient division would
produce is shown as xF E . The superscript F helps remind us that this is the
optimal solution only if the ceo has f ull information about the divisions type.
Of course, the problem is that the ceo does not have this information.
Clearly, she cannot hope to get away with the contracts corresponding to points
A and B when she doesnt know the divisions type. If she oered these con-
tracts, then the manager would choose A both if his division is inecient and
if it is ecient. The reason the manager would choose A over B when his divi-
sion is ecient is that A is on a higher indierence curve for him than B (see
Figure 4.4); that is, his division would earn greater prots pretending to be
inecient than ecient.
A possible solution to this is for the ceo to simply raise the payment to the
division if it reveals it is ecient. When the division is ecient, the manager
will be willing to reveal his division is ecient rather than pretend to be ine-
cient by accepting contract A provided he is oered a contract on or above the
indierence curve through A and C. Because the ecient level of production is
unaected by the amount of the payment, the optimal contract for the ceo to
oer conditional on being on that indierence curve is C.
While production is ecient by both types of division if the contracts oered
are A and C, observe that the manager is earning a huge information rent if his
division turns out to be ecient (the information rent is the vertical distance
from B to C).
The ceo can reduce that information rent by making claiming to be inef-
cient less desirable. She does this by sliding down the zero-prot indierence
curve of the inecient type. That is, she distorts downward the output required
of an inecient division. By doing this, she puts an ecient division on a lower
indierence curve, thereby lessen the information rent it can capture.
Is it worth it to the ceo to reduce the information rent in this way? Observe
that because xF I is ecient, moving slightly away from it is only a second-
Hidden-Information Agency 53

order loss.13 On the other hand she gets a rst-order gain in terms of lowering
information rent. For observe, because xF I is ecient, it maximizes surplus,
x KI (x). So the eect of a small change, x, in output changes prots by
approximately
d  
x (x KI (x))|x=xF = x 1 KI (xF
I )
dx I

= x 0 = 0

because xF I maximizes surplus (so the rst derivative of surplus evaluated at

xF
I is zero). What is the change in the information rent? Because pI = KI (xF
I )
(there is no prot for an inecient division at A), the prot of an ecient
division pretending to be inecient is

pI KE (xF F F
I ) = KI (xI ) KE (xI ) .

This, then, is the additional payment (or rent) that the ceo must pay an ecient
division to admit it is ecient. Observe its change given a small change, x,
in output required of an inecient division is approximately
d  
x (KI (x) KE (x))|x=xF = x KI (xF
I

) KE (xF
I )
dx I

<0

where the last line follows because, recall, KI (x) > KE

(x) for all x > 0. Thus,
F
a small reduction in output target from xI causes approximately no loss if the
division proves to be inecient, but a clear reduction in the information rent if
the division proves to be ecient.
How far should the ceo reduce the output target for an inecient division?
Observe that the marginal loss in prot should the division prove inecient is
d  
x KI (x) = 1 KI (x) .
dx
But she bears this loss only when the division is inecient, which occurs with
probability f . Hence, her expected marginal loss is
 
f 1 KI (x) . (4.17)

Her gain is the reduction in information rent. The marginal rent reduction is
d  
KI (x) KE (x) = KI (x) KE

(x), .
dx
She only gains this reduction when the division proves to be ecient, which
occurs with probability 1 f . Hence, her expected marginal reduction is
 
(1 f ) KI (x) KE

(x) . (4.18)

13 When youre at the top of a hill, moving a little ways from the top lowers your elevation

very slightly.
54 Agency

The optimum occurs when the expected marginal loss, expression (4.17), just
equals the expected marginal reduction, expression (4.18):
   
f 1 KI (x) = (1 f ) KI (x) KE

(x)

or, rearranging,

1f  
1 KI (x) = 
KI (x) KE (x) . (4.19)
f

The xI (f ) in Figure 4.4 is the solution to equation (4.19).

From Figure 4.4, the solution to the hidden-information agency problem
is thus for the ceo to announce that the contract will be D if the divisional
manager claims his division is inecient and will be E if the divisional manager
claims his division is ecient.
From equation (4.19), we can see how the solution varies with the ceos
prior beliefs about how likely the division is to be ecient and inecient. If
she thinks it is very likely that the division will be inecient, then f is close
to one, the right-hand side of (4.19) is close to zero, and the solution will be
close to the full-information solution. If, however, she thinks it very likely that
the division will be ecient, so f is small, then the right-hand side is very
large and she will want to distort downward the output target of an inecient
division by a lot. Indeed, at some point, f gets so small that the expression
on the right-hand side would exceed one (the largest possible value of the left-
hand side) for any x > 0. In this case, the ceo has to essentially shut down
the inecient division; announcing that, if the divisional manager claims his
division is inecient, then it is not to produce at all. Note, from Figure 4.4, if
xI (f ) = 0, then there is no information rent paid to the ecient division.

4.4.3 Summary
The problem with hidden-information agency is that the principal has to weigh
the benets of ecient production against the level of information rent that
more ecient agents will collect. This leads to a solution (for intermediate
values of f ) with the properties:

When the agent is the ecient type, he always produces eciently.

When the agent is the inecient type, he produces less than the ecient
amount; thereby reducing the information rent that would have been paid
had the agent proved to be the ecient type.

Despite the reduction in information rent, the agent still earns some in-
formation rent when he is the ecient type.
The agent earns no information rent when he is the inecient type.
Lecture 5

Monitoring

M onitoring refers to the act of gathering information. In the analysis

of hidden-action agency or costly state verication, it is clear that more
information about what is going on can be quite valuable. Better information
leads to better decisions and tighter control. Hence, the benets of monitoring
are clear.
It is, however, also important that monitoring comes at a cost. If, for ex-
ample, to avoid a hidden-action agency problem, the monitor has to follow the
agent everywhere and scrutinize is every action, then the cost is quite high.
Moreover, this is clearly an onerous task for the monitor and one, therefore,
must why the monitor does it? What are the monitors incentives? And who
monitors the monitor?

5.1 A Simple Model

To begin, lets consider an idealand highly heuristiccase. Let e represent
total expenditures on monitoring and let B(e) represent the total benets. Total
benets can possibly be in expected value terms.
The ecient level of monitoring eort maximizes total surplus,

B(e) e .

For illustrative purposes, assume that B() is twice-dierentiable, strictly con-

cave (i.e., B  () < 0), and that B  (0) > 1; hence, there is a unique level, e > 0,
that maximizes total surplus.1
As we will see, there a number of reasons why this ideal is not typically
attained.

1 That is, e is the solution to B
(e) 1 = 0.

55
 56 Monitoring

blunted incentives|textsl 5.1.1 Blunted Incentives

In may circumstances, the party that bears the cost of monitoring (e.g., e in our
simple model) does not receive 100% of the benets. For example, because the
total benet is shared with all the shareholders, the board of directorsthat is,
the monitorsget only a small fraction of the benets, while incurring all the
cost. Or a large shareholder or dissent group of shareholders could incur all the
cost, while having to share the benet with the remaining shareholders.
To capture the consequences of having to share the benets, while fully
bearing the costs, suppose that the monitor gets , 0 < 1, of the benets;
that is, the monitors payo is

B(e) e .

Because < 1, maximizing that last expression with respect to e yields a

solution, e, that is less than e ; that is, the monitors blunted incentives cause
the monitor to monitor less than the ecient amount.2
For example, if B(e) = ln(e), then e = 1 and e = . Hence, for this example,
the percentage decrease in monitoring eort or expenditure is (1 ) 100.
This illustrates a general point, namely that one has to be concerned with
the incentives of the monitor to monitor.
In some circumstances it could be feasible for the monitors overseers to
provide stronger incentives. For instance, suppose the shareholders gave the
directors the incentive scheme:

w , if B B(e )
w(B) = . (5.1)
0 , if B < B(e )

Under this scheme, either the directors choose e and have utility

B(e) e .

Or, they choose e = e and enjoy utility3

B(e ) + w e .

The latter will exceed the formerand, hence, the contract in expression (5.1)
will induce rst-best eortif
 
w e e B(e ) B(e) . (5.2)

2 Proof: The monitors rst-order condition is B
(e) 1 = 0, which is equivalent to

B
(e) = 1/. In contrast, the ecient solution has B
(e) = 1. Because the right-hand side of
the former is greater than that of the latter and B() is concave, it follows that e < e .
3 Because e maximizes B(e) e, we know that the derivative of w + B(e) e is negative

for all e > e and, thus, for all e e . Hence, the constraint that e e if one is to get w
must be binding.
A Simple Model 57

If w is set so that it is not too much larger than the right-hand side of (5.2), free riding
then the (remaining) shareholders will be quite happy to put such a scheme in
place.4
In other cases, however, the monitors incentives are set exogenously. For
instance, the gains to a large shareholder or dissent shareholder group of mon-
itoring is simply proportional to their shareholdings. In other words, in some
circumstances, blunted incentives are unavoidable.
Moreover, recall the lesson of the last lecture: Even when explicit contracts
can be employed (e.g., such as the one in (5.1)), it can be infeasible in complex
environments (e.g., unlike the one of (5.1)) to design contracts that yield the
rst-best action at an acceptable cost.

5.1.2 Free Riding

Another reason the rst-best level of monitoring might not be achievable is due
to free riding among the potential monitors.
We saw one free-riding example in Section 2.2. It can readily be adapted to
monitoring. Now, instead of nding the right course of action, each of the ten
investors can expend eort on monitoring at a personal cost of 10. If the rms
management is monitored, the rm returns 110. If the rms management is not
monitored, the rm returns 0. As before, the conclusion is that the probability
of no monitoring at all is approximately 90%.
As another example, suppose there are N investors, each of whom owns 1/N
of the rm. As in our idealized model, the total benet of monitoring is B(e),
where, now,
e = e1 + + eN ,
where en is the eort expenditure of the nth investor. Dene

en = (e1 + + eN ) en ;

that is en is the sum of the monitoring eorts of all investors except the nth.
Each investor n maximizes
1
B(en + en ) en , (5.3)
N
given his anticipation of the sum of the other investors eorts (in equilibrium
he must anticipate correctly). It is readily shown that there is no equilibrium

4 Proof: Suppose w satises expression (5.2) as an equality, then the net gain to the other

shareholders is
     
(1 )B(e ) w (1 )B(e) = B(e ) e B(e) e
> 0,

where the last line follows because eis the unique maximizer of B(e) e. Hence, provided
w is not too much bigger than the right-hand side of (5.2), the other shareholders net gain
will be positive.
58 Monitoring

Monitor
Monitor No Monitor
Agent Behave 6,4 6,6
Dont behave 2,5 8,0

Figure 5.1: The monitoring game. Observe this game has no pure-strategy equi-
librium; hence, there is a positive probability that the principal does
not monitor and the agent doesnt behave.

of this game in which e (total eort) is as great as e .5 Furthermore, it can

be shown that total eort is falling in N .6 In other words, not only is there a
free-riding problem with monitoring, but it gets worse the more investors there
are.
This last observation, that the free-riding problem gets worse as the number
of potential monitors increases, helps to explain why there is evidence that rms
in which large blocks of stock are held by a single shareholder will outperform
rms in which stock ownership is more diversied.

5.2 Monitoring is not Always Credible

Even if there is a single monitor, so no free-riding problems, and the monitor
has strong incentives (e.g., gets 100% of the benet), monitoring can still fail
to achieve the rst best. The reason is that eective monitoring can lead to
reactions by the monitored that need to be taken into account.
To understand this point, consider the game shown in Figure 5.1. One
can interpret this game as follows: Monitoring is a better strategy than not
monitoring if the agent wont behave, but is a worse strategy if he will. In the
latter case, monitoring eort is unnecessary, hence wasted. Because misbehavior
is punished if caught, misbehaving is less desirable for the agent if the monitor
will monitor. On the other hand, behaving is less desirable than misbehaving if
not monitored; that is, if not monitored, the agent gets away with misbehaving.
This game has no pure-strategy equilibrium: The monitors best response
to behave, is not to monitor; but the agents best response to dont monitor
is to misbehave. Similarly, the monitors best response to dont behave is
to monitor; but the agents best response to monitor, is to behave. It does,
however, have a mixed-strategy equilibrium; that is, one in which the parties
randomize over which strategy to play. Remember, to be willing to randomize,

5 Proof: Suppose there were. Consider investor n. He chooses e to maximize B(e +

n n
en )/N en , where, by supposition, the solution will be such that en + en e . The
rst-order condition requires that
B
(en + en )/N 1 = 0 or, equivalently, B
(en + en ) = 1/N .
But the concavity of B() entails, then, that en + en < e , a contradiction.
6 The argument is the same as in the previous footnote.
Monitoring and Contracts 59

you must like your alternatives equally well. Hence, if p is the probability that
the monitor monitors and q is the probability that the agent behaves, then we
know:

4q + 5(1 q) = 6q + 0(1 q) (5.4)


 

expected payo if monitor expected payo don
t monitor

and

6p + 6(1 p) = 2p + 8(1 p) (5.5)


 

expected payo if behave expected payo if misbehave

Solving equations (5.4) and (5.5) reveals q = 5/7 and p = 1/3; that is, the agent
behaves with probability 5/7 and the monitor monitors with probability 1/3.
In other words, the monitor engages in surprise inspections or audits, while the
agent misbehaves roughly 29% of the time on average.
In this sense, we see that monitoring can be imperfect. The monitors ex-
pected payo is 30/7 ( 4.29), which is less than 6, the payo the monitor could
receive if the agent were always to behave.
This is conclusion is not an artice of considering only a single-play game.
Because the worse the agent can ever do is 6 per period, which is also his per-
period payo if he always behaves, there can be no equilibrium of a repeated
game other than repetition of the equilibrium of the single-play game.

5.3 Monitoring and Contracts

So far, we have ignored the possibility that the rm also uses incentive contracts.
In addition to possibly replacing the need for incentive contracts, monitoring
can help reduce the cost of employing incentive contracts.

5.3.1 Monitoring and Hidden-Action Agency Contracts

As noted in Lecture 4, if the principal could observe the agents action, then
the rst-best outcome could be achieved. Consequently, if monitoring permits
the principal to observe the agents action, then the benet of monitoring is
two-fold:

1. The additional compensation that must be paid on average to compensate

the agent for bearing risk is saved; and

2. To the extent that the hidden action distorted the action the principal
sought to induce, a more ecient action is induced.

Against this benet, the cost of monitoring must be weighed.

Even if monitoring is either not good enough to permit observation of the
agents action or would only be that good at too high a price, monitoring can
 60 Monitoring

Hermalin,
Benjamin E. still improve the situation. Recall, from the discussion of comparative statics
Weisbach,
Michael
S.
in Section 4.3.4, that the cost of using an incentive scheme falls as the infor-
mativeness of the signal (performance measure) increases. Consequently, to the
extent that monitoring yields more informative signals, it reduces the cost of
employing incentive contracts.

5.3.2 Monitoring and Hidden-Information

Agency Contracts
If the principal knows the agents type, then the rst-best outcome is attainable
in a hidden-information agency problem; because there would be no need to pay
an information rent, there would be no motive for the principal to distort the
level of output (action) asked of any type.
In fact, in many hidden-information agency models, the principal wouldnt
even have to have a perfect signal of the agents type to avoid paying information
rent. For instance, in the model of Section 4.4.1, if there were no restrictions on
nes that could be imposed on the divisional manager (i.e., he had no personal
limited liability), then the principals observation of any signal correlated with
the agents type would be sucient to allow her to achieve the rst best. In
other words, even a little monitoring could prove extremely useful in a hidden-
information setting.
Even if monitoring is not sucient to achieve the rst best, it can often
reduce information rent. The reduction in information rent improves the princi-
pals prots. Moreover, the less information rent she potentially pays, the lower
her motive to distort the agents output (action).

5.4 Monitoring for Ability

Our discussion of monitoring in Sections 5.2 and 5.3 focussed primarily on moni-
toring of actions (although we did consider monitoring for types in our discussion
of the hidden-information agency model). In addition to monitoring for actions,
there is monitoring to assess ability. One of the roles of monitors, such as the
board of directors, is to assess the ability of the monitored party. Although pub-
lic aspects of performance allow for reassessment of ability, it is often desirable
that monitors uncover additional evidence upon which to assess ability.
Here, lets consider a model based on the model in Hermalin and Weisbach
(1998). To make the model more concrete, lets call the monitor the board of
directors and lets call the monitored party the ceo.

5.4.1 Timing
The model has the following timing.

Stage 1. At the start, a board of directors hires a new ceo for the rm. There is
a commonly held prior distribution about the ability, , of the new hire.
 Monitoring for Ability 61

Specically, is distributed normally with mean and variance 1/ ( is precision|textsl

Holmstrom,
Bengt
R.
the precision of the distribution).

Stage 2. The board may acquire a private signal, y, about the ceo it has hired. The
probability that the board acquires this signal depends on the intensity
with which it monitors the ceo. The signal is distributed normally with
a mean equal to the ceos ability, , and a variance equal to 1/s. The
precision s is the same regardless of which candidate became ceo.

Stage 3. If the board obtains the signal, it updates its estimate of the ceos ability.
Based on this posterior estimate, the board may decide to re the ceo
and hire a replacement. A replacement ceos ability, R , represents a
random draw from a normal distribution with mean R = 0 and variance
1/R . Setting R to zero is a convenient normalization without loss of
generality. Because what is of the interest is the boards replacing the
incumbent ceo in response to a bad signal, assume 0 (were <
0, then the incumbent ceo would lose his job both when a bad signal
is obtained and when no signal is obtained). The assumption that the
mean ability of a replacement ceo is lower than the (unconditional) mean
ability of an incumbent can be justied as follows: Firing the ceo early
triggers a succession before the normal transition process will have run;
hence, the pool of candidate successors is likely weaker than it would be
in a normal succession process. A complementary justication is that R
is the expected value of the rm under a caretaker administration that
is not (fully) able to pursue new initiatives or respond aggressively to
changes in the strategic environment.7 Along these same lines, a further
justication is that R represents the expected ability of a new ceo minus
such disruption costs.8

Stage 4. Earnings, x, are realized. Earnings are distributed normally with a mean
equal to the ability of the ceo in place (the one hired in stage 1 or his
replacement if hes replaced in stage 3). The random variables y and
x are independently distributed.

5.4.2 Preferences and Ability

A ceos ability is xed throughout his career. I follow Holmstrom (1999) by
assuming that ceos and boards are symmetrically informed; in particular, both
parties know only that the ceos ability is drawn from a normal distribution
with mean and precision . This assumption can by justied by noting that
a critical component about a ceos ability is the match between him and the
job of being ceo, about which both parties are likely to be equally uncertain.

7 Eldenburg et al. (2003) nd evidence that, in hospital ceo transitions, caretaker admin-

istrations are appointed with some frequency.

8 See Vancil (1987) for a discussion of the benets of an orderly transition process.
 62 Monitoring

Hermalin,
Benjamin E.|nn Moreover, both parties are likely to have similar knowledge of those aspects of
Weisbach,
Michael
S.
DeGroot,
Morris H. ability revealed by prior work experience.
Assume that individual directors like higher earnings, but nd monitoring
to be costly; where monitoring is dened as the eorts made to acquire the
signal y.9 Following Hermalin and Weisbach (1998), I assume the preferences of
the individual directors can be aggregated in such a way that the board acts as
if it has a single utility function that positively weights earnings, but negatively
weights eorts to monitor. Such monitoring eorts can be translated, without
loss of generality, into the probabilitydenoted by pthat the board acquires
the signal y. Assume further, as in Hermalin and Weisbach (1998), that the
boards utility function is additively separable:
 
U (x, p) = x + (1 ) c(p) , (5.6)

where c(p) denotes the cost or disutility incurred by the board and and 1
are the weights on the two components. Because utilities are dened up to an
ane transformation only, there is no further loss of generality in dividing (5.6)
by 1 and reexpressing it as

x c(p) ,

where > 0 is a measure of the boards diligence or independence. Assume that

c() is strictly increasing for p > 0, strictly convex, and twice-dierentiable. The
factor , which is determined exogenously, is meant to capture those aspects of
the board that aect how it weighs the cost of monitoring versus the higher
prots that such monitoring can provide. This factor would, for instance, be
expected to vary inversely with (i) the proportion of inside directors on the
boardinsiders presumably have reasons to dislike monitoring themselves or
their boss, the ceo; (ii) the opportunity cost of the directors time; (iii) di-
rectors incentives not to rock the boat (e.g., to increase their chances for
additional directorships with other rms); and (iv) the strengths of the per-
sonal ties between directors and the ceo.

5.4.3 Updating Beliefs and Optimal Monitoring

If the board obtains the signal, y, the posterior distribution of the ceos ability
is normal with mean and precision , where
+ sy
= and = + s
+s
(see, e.g., DeGroot, 1970, p. 167).
Observe that the expected value of earnings, x, is the expected value of the
ceos ability. Therefore, it is if a signal is obtained and the incumbent ceo
9 This monitoring need not be particularly active; one could as easily interpret it as taking

the eort to be attentive to the performance of the rm and making an eort to seek from the
data made available to the board information relevant for estimating the ceos ability. See
Hermalin (2004) for other interpretations.
Monitoring for Ability 63

is retained; if no signal is obtained and the incumbent ceo is retained; and Phi@$\Phi(\cdot)$
(\textsccdf
of
standard
normal)
R = 0 if a replacement ceo is hired. By assumption 0, so, conditional CDF@\textsccdf
(cumulative
on no signal being obtained, the board maximizes rm expected earnings by distribution
function)
phi@$\phi(\cdot)$
(density
functi
retaining the incumbent ceo. If a signal is obtained, then expected earnings standard
normal)
are maximized by ring the incumbent ceo and hiring a replacement if and only
if < 0 = R . Hence, the rule for replacing the incumbent ceo is to replace
him if and only if the signal y satises

y< Y . (5.7)
s
Note Y is the cuto value for the signal, below which the incumbent loses his
job.
The distribution of the signal y given the ceos true ability, , is normal with
mean and variance 1/s; hence, the distribution of y given the prior estimate
of the ceos ability, , is normal with mean and variance 1/s + 1/ .10 Dene
s
H=
s+

to be the precision of y given .

The rms expected earnings if it will learn y are
  
+ sy H H (y)2
V max 0, e 2 dy .
+ s 2

The option to re the incumbent ceo is a valuable one, hence V > for all .
A change of variables from y to z H(y ) reveals that V can be written
as

  H  
V = 1 (Y ) H + (Y ) H

  H  
= (Y ) H + (Y ) H ,

where () is the distribution function (cdf) of a standard normal random
variable (i.e., with mean zero and variance one), () is its corresponding density
function, and the second line follows from the rst because the standard normal
is symmetric about zero. Note that
 
(Y ) H (5.8)

is the probability that the ceo will be retained if a signal is obtained.

10 The random variable y is the sum of two independently distributed normal variables

y and ; hence, y is also normally distributed. The means of these two random
variables are both zero, so the mean of y given is, thus, . The variance of the two variables
are 1/s and 1/ respectively, so the variance of y and, therefore, y given is 1/s + 1/ .
64 Monitoring

5.4.4 Analysis
In deciding how intensely to monitor the ceo (i.e., what p to choose), the board
solves  
max pV + (1 p) c(p) . (5.9)
p[0,1]

This expression is globally concave in p. For convenience, attention will be lim-

ited to cases in which (5.9) has an interior solution.11 The rst-order condition
for (5.9),
(V ) c (p) = 0 , (5.10)
is sucient, as well as necessary, and admits a unique solution. Let P be the
solution to (5.10). Properties of P are:
Proposition 4 The intensity with which the board monitors the ceo, P , is
(i) decreasing with the prior estimate of his ability, ;
(ii) decreasing with the precision of the prior estimate, ; but
(iii) increasing with the boards diligence or independence, .
Proof: Some preliminaries. First, recall that  (z) = z(z). Second, observe
that
V   H  
= (Y ) H H (Y ) H [Y ] H
Y 
  s  
= (Y ) H H + +
s+ s
= 0. (5.11)
Third, observe that (using expression (5.11) to eliminate V /Y Y / terms):

V H H
= + H + (Y )

 
s 
=+ H +
s+ s
= . (5.12)
Turning to claims (i)(iii): Let be the objective function in expression
(5.9). Consider claim (i):
2 [V ]
=
p
=1 (from expression (5.12))
< 0,

11 The analysis is readily extended to allow for corner solutions, but little is gained by

considering them and excluding them simplies the analysis. A condition that would insure
an interior solution for all parameter values is c
(0) = 0 and limp1 c
(p) = .
Who Monitors? 65

so, by the usual comparative statics, P / < 0. Similarly, who monitors the
monitors|textsl
monitors!auditors
2
= V > 0 ; and
p
2 V
=
p
 
1 s H
= 1 + <0 (5.13)
2s+ 2
(where (5.13) relies on (5.11)). Hence, P / > 0 and P / < 0.

One way to interpret Proposition 4 is in terms of the gain from obtaining a

signal about the ceo, which is V (this is, essentially, the marginal return to
monitoring, p, see (5.10)). The marginal value of the signal increases the greater
is the likelihood it will be decisive with regard to whether the ceo is retained
or red. If the prior about the ceos ability is high, then the signal is less likely
to be decisive. Similarly, if the precision of that prior estimate is high, then the
signal is less likely to be decisive. Because the board will rationally monitor less
the lower is the value of the signal, these insights explain results (i) and (ii) of
Proposition 4.

5.4.5 Discussion
Observe that the board monitors more intensely (chooses a higher P ) the
greater is . That is, the stronger the incentives of the board to monitor,
the more monitoring it does. This is consistent with our insights about blunted
incentives in Section 5.1.1. As in that section, a lesson of this analysis is that it
matters what incentives the monitors have to monitor; or, as it sometimes put,
who monitors the monitors?
From our analysis of free-riding (Section 5.1.2), we might also expect the
larger is the board, the less its ability to monitor eectively. We can interpret
that in this context as falling in board size.

5.5 Who Monitors?

In this section, we consider who carries out the monitoring of management in
corporations and consider their eectiveness, as well as related issues.

5.5.1 Auditors
All corporations are required for tax and security law reasons to have their
books audited; that is, an outside partythe auditing rmmust review the
corporations accounting and certify it as accurate.
Auditing serves chiey to monitor management against stealing corporate
resources or misleading investors about how corporate resources are being em-
ployed. While this is a valuable service, it must be noted that it says little about
 66 Monitoring

Enron most of managements activities or abilities. Moreover, as seen with Enron and
Sarbanes-Oxley
Act
Arthur
Andersen other recent scandals, outside auditing does not always do a perfect job in terms
Enron of the limited scope it does have.
monitors!the
state
as One reason that has been alleged for why auditing rms have fallen short
of the mark is that they suer from conict of interests. In other words, the
issue of what incentives the monitors have is quite relevant here. First, prior
to the Sarbanes-Oxley Act, auditing rms could also sell other services to the
rms that they audited. To preserve these other services, the auditing could
have been less than ideally diligent: Presumably, a management team that has
been given a hard time by its auditor is less inclined to continue or initiate the
employment of the auditors other services. Second, and related to the rst
point, top management has a lot of inuence over whether the auditing rm
audits the corporations books next year or not. Again, too tough an audit
could jeopardize keeping the auditing business next year.
Yet there are still incentives for auditing rms not to do too lax a job.
First, the government oversees auditing and can bring charges against rms
that were derelict in their duties. The actions of the us government against
Arthur Andersen after Enron served to destroy the company. Second, it is a
bad sign (signal) if a corporation switches auditors or hires auditors who are
suspected of doing less than a diligent job; that is, the corporation risks taking
a hit on its stock price if it looks like it is retaliating against its auditor or hiring
less reputable auditors. Third, the shareholders and other aected parties can
sue the auditing rm if they suer losses as the result of poor audits by the
auditing rm (e.g., they fail to detect fraudulent accounting whose subsequent
detection leads to a sharp drop in the stock price).

5.5.2 Government Oversight

The state often provides an amount of oversight. The ability to investigate
corporations for failures to abide by the tax code or securities law means that
the state is engaged in a form of monitoring.
Similar to auditing, the range of activities that the state monitors is fairly
limited. Hence, the impact of this monitoring on managerial behavior is simi-
larly limited.
In addition, the political economy of state oversight often limits oversight.
Executives are wealthy and hold positions of inuence. Hence, they are typically
better situated than a bunch of small shareholders to inuence how the state
exercises its oversight function. In the us, where campaign contributions play
an important role in the electoral process, the ability of executives to make
corporate donations and to bundle individual donations gives them additional
inuence.
One area where government oversight can be more inuential is when the
government is a customer of the corporation. In many procurement contracts,
the states payment to the corporation is on a cost-plus basis or other formula
that requires periodic auditing. In addition, because of the potential for fraud in
government procurement, there are often agencies of the state whose purpose is
Who Monitors? 67

to oversee procurement and be on the outlook for bad behavior. Of course, what monitors!third-party
market for corporate control
could be considered bad behavior from the governments perspective could
be considered good behavior on the part of shareholders: If, for instance,
management takes actions that cause the state more than it otherwise would,
then shareholders benets at the governments expense.
All in all, while government oversight can provide some monitoring of man-
agement, its value to the shareholders will be limited.

5.5.3 Other Third Parties

In addition to auditors and the government, third-party monitors include em-
ployees (especially if unionized), stock exchanges, and public-interest groups.
Non-management employees or subordinate managers will sometimes blow
the whistle on wrong doing by management or superiors. Because the repercus-
sions suered by whistle-blowers are often severe (e.g., demotion or dismissal),
the incentives are very much against such actions. Nonetheless, a sense of jus-
tice or fairness can propel such people to blow the whistle if the wrong doing is
especially egregious.
Because unions can often protect whistle-blowers and have their own reasons
for monitoring management, the monitoring of management by employees can
be more pronounced in unionized companies. Because the interests of unions
and shareholders are not always aligned, the value of such monitoring for share-
holders can be limited, however.
Stock exchanges have rules by which listed companies must abide. To a very
limited degree stock exchanges may intervene if evidence suggests that these
rules have been violated. The impact of this monitoring is quite small, however.
Public-interest groups (e.g., green groups) also monitor corporations that
have an eect on matters of interest to them. To a degree, such public-interest
monitoring can limit managerial misconduct. However, what is misconduct
according to the public interest is not always misconduct with respect to the
shareholders interests and vice versa.

5.5.4 Other Monitors

Board of Directors
The board of directors have the ultimate responsibility for monitoring manage-
ment. We will, however, take up the role of the board of directors in a later
lecture.

The Financial Markets

The nancial markets play a monitoring role insofar as they react to observa-
tions about what the corporation is doing and assessments of the corporations
management.
Such nancial-market monitoring has two avors. One is the market for
corporate control: If certain market observers decide that the corporation is
68 Monitoring

poorly managed, then they can seek to takeover the corporation and improve
its operations. The market for corporate control will be considered in a future
lecture.
The second avor is that the prices of the corporations securities move
in response to information learned by the nancial markets. A drop in the
price of securities, especially the stock price, is typically an adverse signal;
evidence that something bad has been discovered. This can have a direct eect
on management, with the board of directors or shareholders responding by ring
executives. It can also have an indirect eect: To the extent the executives have
compensation tied to the performance of the corporations securities, they stand
to lose if the market concludes that they are taking poor actions. We take this
second avor up below.

Large Security Holders

The holders of large blocks of a security (debt or equity) can have a strong
incentives to monitor. We consider their monitoring below.

5.6 Financial Markets as Monitors

Consider a simple model in which if the ceo takes action a = 0 the rm returns
1 per share for sure. If he takes action a = 1, then it returns 1 per share with
probability 1/2 and 3 per share with probability 1/2. Suppose everyone believed
the ceo was equally likely to choose a = 0 as a = 1. Then the consensus value
of a share of the rm would be
 
3 1 1 1 1
= 1+ 1+ 3 .
2 2 2 2 2
Suppose that you alone observe the ceos choice of a. Then you stand to
make a considerable prot: If a = 0, then you know the rm is overvalued and
you can make money by selling short (alternatively, by buying puts). If a = 1,
then you know the rm is undervalued and you can make money by purchasing
stock (alternatively, by buying call options).
For concreteness, suppose you can trade one million shares. Then you will
make an expected prot of 500,000. Note thats your expected prot regardless
of what you learn. Hence, the value of that information is 500,000; that is, you
would be willing to spend up to 500,000 to gain that information.12
What this simple example shows is that investors have an incentive to learn
information about how the rm is doing; that is, to engage in monitoring. The
rm can exploit such monitoring by making it the basis for managerial incentives
by basing compensation on stock price or by basing dismissal decisions on stock
price or both.
While this simple model suggests a strong incentive for investors to monitor,
reality is more complex. In the model above, the market did not react to your
12 Assuming you are risk neutral.
Large Security Holders 69

buying and selling. In reality, markets look at such activity to, in essence, judge
whether anyone has learned something and, then, to infer from the direction of
that activity whether or not the information points towards greater rm value or
lower rm value. For instance, suppose you learned that a = 1. As you start to
buy, others will observe an upsurge in buying. They will infer that someone has
learned something positive about the rm. Hence, they too will start buying
causing the stock price to rise quickly. But as the stock price increases, the
expected prot you make from each share decreases. Consequently, you wont
realize the full 500,000 in expected prot that we found above. In other words,
your incentives arent necessarily as strong as they rst appeared.
Moreover, the fact that the market will react to transactions suggests an
alternative strategy: Wait until the market begins to move, infer what the in-
formation must be, and then take the appropriate decision. That is, let someone
else incur the expense (in time or money) determining what the ceo will do.
You can freeride on that persons eorts. But, as we saw in Section 5.1.2, once
potential monitors can free ride, the eort they will expend will go down consid-
erably. In a world with millions of potential investor/monitors, the total eort
expended on such monitoring could be small.
In addition, there is the risk of duplication of eort. Only one person need
determine a. If many do, then there has been a wasteful duplication of eort.

5.7 Large Security Holders

A large security holder (i.e., a holder of a large block of a security) has poten-
tially strong incentives to monitor. Such a party has stronger incentives than a
smaller security holder in terms of having less blunted incentives (Section 5.1.1).
In addition, as the portion of the rm owned by the security holder increases in
expression (5.3) in Section 5.1.2, the amount of eort he expends increases.
On the other hand, as we will see, there can also be downsides to large block
holders.

5.7.1 Large Shareholders

Large shareholders come in two varieties: institutional investors (e.g., pension
funds, mutual funds, etc.) and private investors (individuals or families).
Despite holding large blocks, institutional investors could have weak incen-
tives to monitor. Remember that it matters what incentives the monitor has.
The people overseeing an institutions investments do not own those invest-
ments. They could enjoy a rather limited share of the return from those invest-
ments. Hence, their motive to monitor could be quite limited. For instance,
suppose that a manager of a institutional fund gets 1% of the return. Suppose
the fund holds 10% of a company. Then, in terms of the managers motive to
monitor the company, it is as if he owns 1/10 of a percent of that companys
stock. In this sense, the manager is hardly a large shareholder and his monitor-
ing incentives are correspondingly modest.
 70 Monitoring

monitors!banks Private investors who hold large blocks have stronger motives to monitor;
they, after all, get 100% of the returns from the shares they own. However, there
are three issues. First, rather than monitor actively (i.e., use the information
as the basis of changes they impose via changes in the board of directors), they
could monitor more passively; that is, simply use their information to make
better trades. While there is some benet to such passive monitoring (recall
our discussion of the market as a monitor), it could be less valuable than more
active monitoring and oversight.
A second issue is that large blockholders could use their position to exploit
minority shareholders, as discussed previously.
A third issue is that if enough stock is held in blocks, then the stock isnt very
liquid. Hence, the kind of market monitoring discussed above is harder. If the
market isnt liquid, then it is hard to execute trades that convey information.
Furthermore, there is the cost that a relatively illiquid stock imposes on those
who hold it.

5.7.2 Large Creditors

Large creditors, such as banks that have made signicant loans, have a strong
motive to monitor.
The eectiveness of that monitoring depends, again, on the incentives of
the person doing the monitoring to do the monitoring. A bank doesnt moni-
tor, people in the bank monitor. If those employees of the bank charged with
monitoring the corporation dont, themselves, have strong incentives or are not
adequately monitored, then they are unlikely to be eective monitors.
In addition, the eectiveness of bank monitoring from the perspective of
shareholders depends on the alignment of the interests of shareholders and
debtholders. If, for instance, there is very little risk of default, then the bank
will have less incentive to be a diligent monitor, whereas shareholders could
still care considerably about oversight of management. Moreover, the banks
motives are to monitor against activities that increase the likelihood of default.
As we saw from our analysis of asset substitution, activities that increase the
likelihood of default can increase the value of stock.
 Hall,
Brian J.
Liebman,
Jeffrey
B.
stock
grants
stock
grants!restricted|textsl
stock
options
strike price|textsl
phantom
stock|textsl

Lecture 6

Executive Compensation

F ew areas of corporate governance have been as controversial as exec-

utive compensation. The fact, for instance, that the average ceos total
compensation increased 209% from 1980 to 1994 (Hall and Liebman, 1998) is
seen as just one piece of evidence that executive compensation has gotten out
of hand.
Yet to evaluate whether executive compensation has, indeed, gotten out
of hand, we need to understand both the theory and practice of executive
compensation.
Much of the theory was covered in the lecture on agency; however, there is
still more theory worth considering. Before extending the theory, we need rst
to review some terminology

6.1 Executive Compensation: Terminology

Executive compensation has a number of components:

Salary: Agreed upon xed annual compensation

Bonus: Additional monetary compensation paid at the end of year either at
discretion of board or superiors or according to xed schedule
Stock-based compensation

Stock grants an outright gift of stock with no restrictions.

Restricted stock grants a grant of stock with restrictions prevent-
ing the sale of the stock granted before a specied period of time.
Stock options the right to purchase stock at a specied price (the
strike price) within some window of time in the future.
Phantom stock a plan under which the executive is paid a bonus
equal to the increase in the stock price (if any) over a specied period
times some amount (the number of phantom shares).

71
 72 Executive Compensation

tournament|textsl Stock held by pension fund the executive could be promised a pen-
sion that is paid by a pension fund. To the extent the pension fund
holds stock in the company, the amount of the executives pension
could depend on how the value of the stock.

Benets: Insurances (e.g., health in the us), pensions, memberships

In-kind compensation: Use of company resources (e.g., use of the company

jet)

6.2 Executive Compensation: Theory

6.2.1 Intrinsic Incentives
In addition to the incentives that a compensation plan can create for an execu-
tive, an executive can have intrinsic incentives. It is worth considering them.

Intrinsic Motivation
People have some amount of intrinsic motivation to do their jobs well. Part of
this is simply personal drive. People often feel happier if they believe they are
doing their job well.
In addition, people believe that it would be unfair not to work hard or do
a good job if others are expecting it of them. They may feel they owe others a
good job. Other normative pressures include fear of being shamed for doing a
poor job, being outcast for doing a poor job, or losing prestige for doing a poor
job.

Stock Holdings
Executives could own stock in the company other than those given to them as
compensation. Because the value of their stock is tied to how well they do their
job, this ownership creates incentives to do a good job.

6.2.2 Promotion Seeking

Most executives would like to be promoted (assuming they are not already at
the top). Such a promotion could be internal or could involve being hired for a
higher position at another rm.
Because such promotions typically involve competition with other aspirants
for the job, promotion seeking can create strong incentives.

A Model
Consider two junior executives who both desire a promotion. Only one of them
can get the promotion (the competition is a tournament). Suppose the exec-
utives simultaneously choose eort. Eort, e, lies in [0, 1]. Let an executives
Executive Compensation: Theory 73

payo be influence activities|textsl

e2
1 if promoted; or
2
e2
if not promoted.
2
Let the probability that executive 1 gets the promotion equal
e1 e2 + 1
,
2
hence the probability that executive 2 gets the promotion is
e2 e1 + 1
.
2
Executive is expected payo (i = 1 or 2) is, therefore,
ei ej + 1 e2i
(6.1)
2 2
(where j = 3 i indexes the other executive).
Each executive i maximizes expression (6.1) with respect to ei treating ej
as a constant. The rst-order condition is
1
ei = 0 .
2
Hence, each executive expends eort 1/2 in equilibrium. Each executives ex-
pected equilibrium payo is, thus, 3/8.
Observe that this tournament is equivalent to giving each executive the in-
centive plan w(e) = e/2; that is, the tournament substitutes for direct incentives.

Drawbacks to Tournaments
Although promotion tournaments are often unavoidable and they can oer
strong incentives, it is worth briey noting that there are potential drawbacks
to tournaments:
They impose risk on the agents. In the above example, the executives
were implicitly assumed to be risk neutral, so risk didnt matter. One
could, however, imagine that they could be risk averse, in which case they
will require compensation for bearing risk.
Because the executives are competing against each other, their incentive
to cooperate could be reduced or eliminated. To the extent cooperation
between them is valuable, this is a clear drawback.
Promotion seeking can lead to influence activities; that is, the executives
divert time and energy from productive activity toward activities meant
to sway the people making promotion decision (e.g., devoting excessive
time to making their PowerPoint slides pretty).
 74 Executive Compensation

career concerns|textsl Related to inuence activities, promotion seeking can lead to bribery of
the people who control the promotion decision.

6.2.3 Job Retention

People like to keep their jobs. Hence, the fear of losing their job for poor
performance creates certain incentives for people.
Because the performance criteria for keeping or losing ones job are endo-
genousthat is, determined by those making the decision to keep or terminate
job retention can be seen as an incentive scheme, albeit one with peculiar prop-
erties.
At a simple level, job retention is no dierent than the sort of incentive
scheme used in response to a hidden-action agency problem. Based on some
performance measure there are two outcomes: keep or re (just as, in our model
in Lecture 4, there were two wages, a high one and a low one). Keep implies
a certain level of wealth for the executive, while re implies a dierent level
of wealth. But what these levels of wealth are is, to a degree, endogenous. The
principal (e.g., board) sets the salary of the ceo, which is the keep payment.
There is some amount a red executive can expect to earn, this is the re
payment. Of course, because this scheme imposes risk on the executive, he will
require compensation for bearing risk. This means an even higher salary if kept
or some additional payment (e.g., golden parachute) if red or some combination
of both.

6.2.4 Career Concerns

Executives want to retain their value on the outside market. The higher their
value on the outside market, the more their current employer will have to pay
them (i.e., the current employer has to match the market, otherwise the execu-
tive will jump ship). Executives desires to retain their outside value are known
as career concerns.
Career concerns have to do with the markets ability to infer managerial
ability.

A Simple Model
Suppose that a managers performance is x = + e + . As before, denotes
the managers ability. It is unknown by anyone, but everyone knows that is
drawn from a normal distribution with mean and variance 1/ ( is, thus, the
precision). The variable e [0, ) denotes the managers eort, which is not
observable by the market. The term is a mean-zero normal random variable
with variance 1/s. Suppose that the managers pay next period is , where
(0, 1] is some constant and is the markets estimate of the managers
ability conditional on having seen x.
Because the market wishes to estimate , the market wishes to subtract
e from x to have a statistic for estimating . The market doesnt observe e,
Executive Compensation: Theory 75

however, so such subtraction isnt possible. The market will, though, have a
guess as to what e will be, e (and, in equilibrium, this guess will be correct).
The market will thus subtract e from x in forming its estimate of . It can be
shown that the best estimate of is
+ s(x e)
= .
+s
Suppose the managers utility function is e2 /2. Then the manager
wishes to maximize
 
+ s(x e) e2 + s(E{ + } + e e) e2
E =
+s 2 +s 2

with respect to e. The rst-order condition is

s
e = 0.
+s
Solving, we see that the managers eort will be
s
e = . (6.2)
+s
Observe, from expression (6.2), that the managers eort increases in ;
that is, the more of an impact todays performance has on tomorrows wage,
the harder the manager will work.
Also observe that, as gets larger, the managers eort falls. The variable
is the precision of the prior estimate of the managers ability. The higher is
, the more precise is the prior estimate, which means there is less uncertainty
about the managers ability. The lower is , the less precise the prior estimate,
and, hence, the more uncertainty about his ability. Thus we see that the more
uncertain the market is about the managers ability, the stronger are his in-
centives to work hard. Conversely, the more certain the market is about the
managers ability, the less is his incentive to work hard.
Finally, it can be shown that the managers equilibrium eort is increasing
in s.1 Recall that 1/s is the variance of , the noise term. What this says is
that the manager works harder the less noisy a signal x is about .
To summarize:

Proposition 5 In the career concerns model, the managers equilibrium level

of eort, e is
(i) increasing in the importance the market places on previous performance
(i.e., is increasing in );
(ii) decreasing with how noisy a signal performance is about ability (i.e., is
decreasing in 1/s; equivalently, is increasing in s); and

1 de /ds
= ( +s)2
> 0.
76 Executive Compensation

(iii) decreasing with how precise the prior estimate of his ability is (i.e., is
decreasing in ).

This simple model thus illustrates that career concerns generate incentives
for managers.

Some Drawbacks to Career Concerns

While career concerns generate incentives for managers, they are not a perfect
solution to the agency problem. First, observe that the managers marginal
return to eort is
s
.
+s
That incentive can, however, be either too weak relative to what the principal
would desire or too strong. In other words, there is no guarantee that career
concerns provides the best solution to the agency problem.
Second, item (iii) in Proposition 5 shows the better the prior estimate of the
managers ability, the less career concerns drive him. Over time, a managers
ability becomes better known. Hence, at any point late in his career his prior
estimate (i.e., the estimate based on previous performance) is fairly precise.
We see, therefore, that career concerns must die out over the course of the
managers career. In other words, even if career concerns are a good incentive
initially, later in a managers career they will be inadequate.2

Further Drawbacks to Career Concerns

While career concerns can create good, if imperfect, incentives with respect to
eort, they can create bad incentives with respect to the managers choice of
projects or strategies.
This can be illustrated by a simple model. Suppose the same setup as before,
except, now, let
x = + (p) + (p) ,
where (p) is the mean return if project p is chosen and (p) is a mean-zero
random variable that represents the uncertainty about that mean return. As-
sume (p) is normally distributed with variance 1/h(p). The other change is
now to assume the manager is risk averse, with utility function u(), where
u() exhibits diminishing marginal utility.
At the point at which the manager chooses the project, is a random vari-
able. Because the manager is risk averse, he cares not only about

2 See Holmstrom (1999) for details.

Bibliography 77

Bibliography

Aghion, Philippe and Benjamin E. Hermalin, Legal Restrictions on Pri-

vate Contracts Can Enhance Eciency, Journal of Law, Economics, and
Organization, Fall 1990, 6 (2), 381409.
Baskin, Jonathan Barron and Paul J. Miranti Jr., A History of Corporate
Finance, Cambridge, England: Cambridge University Press, 1997.
Caillaud, Bernard and Benjamin E. Hermalin, Hidden Actions and
Incentives, http://faculty.haas.berkeley.edu/hermalin/#WP, 2000. Unpub-
lished manuscript.
Clark, Robert C., Corporate Law, Boston: Little, Brown and Company, 1986.
Dean, John W., Worse than Watergate, New York: Little, Brown and Com-
pany, 2004.
DeGroot, Morris H., Optimal Statistical Decisions, New York: McGraw-Hill,
1970.
Eldenburg, Leslie, Benjamin E. Hermalin, Michael S. Weisbach, and
Marta Wosinska, Governance, Performance Objectives and Organizational
Form: Evidence from Hospitals, Journal of Corporate Finance, 2003. Forth-
coming.
Gibbons, Robert, Game Theory for Applied Economists, Princeton, NJ:
Princeton University Press, 1992.
Grossman, Sanford and Oliver D. Hart, An Analysis of the Principal-
Agent Problem, Econometrica, 1983, 51, 746.
Hall, Brian J. and Jerey B. Liebman, Are CEOs Really Paid Like Bu-
reaucrats?, Quarterly Journal of Economics, August 1998, CXIII (3), 653
691.
Hansmann, Henry and Reinier Kraakman, The Essential Role of Orga-
nizational Law, Yale Law Journal, 2000, 110, 387440.
and , Organizational Law as Asset Partitioning, European Economic
Review, 2000, 44, 807817.
78 Bibliography

Hermalin, Benjamin E., The Eects of Competition on Executive Behav-

ior, RAND Journal of Economics, 1992, 23, 350365.

, Trends in Corporate Governance, Journal of Finance, 2004. Forthcoming.

and Michael S. Weisbach, Endogenously Chosen Boards of Directors and

Their Monitoring of the CEO, American Economic Review, March 1998, 88
(1), 96118.
and Nancy E. Wallace, The Determinants of Eciency and Solvency
in Savings and Loans, RAND Journal of Economics, Autumn 1994, 25 (3),
361381.

Holmstrom, Bengt, Managerial Incentive ProblemsA Dynamic Perspec-

tive, Review of Economic Studies, January 1999, 66 (226), 169182.

Jensen, Michael C., Agency Costs of Free Cash Flow, Corporate Finance,
and Takeovers, American Economic Review, May 1986, 76 (2), 323329.
Papers and Proceedings of the Ninety-Eighth Annual Meeting of the American
Economic Association.

Laont, Jean-Jacques and David Martimont, The Theory of Incentives:

The Principal-Agent Model, Princeton, NJ: Princeton University Press, 2002.

MacAvoy, Paul W. and Ira M. Millstein, The Recurrent Crisis in Corpo-

rate Governance, New York: Palgrave Macmillan, 2003.

Modigliani, Franco and Merton Miller, The Cost of Capital, Corporate

Finance, and the Theory of Corporation Finance, American Economic Re-
view, 1958, 48, 261297.
Rabin, Matthew, Psychology and Economics, Journal of Economic Liter-
ature, March 1998, 36, 1146.

Smith, Adam, An Inquiry into the Nature and Causes of the Wealth of Na-
tions, Indianapolis: Liberty Press, 1776.

Spence, A. Michael, Job Market Signaling, Quarterly Journal of Eco-

nomics, 1973, 87, 355374.

Townsend, Robert M., Optimal Contracts and Competitive Markets with

Costly State Verication, Journal of Economic Theory, 1979, 20, 265293.

van Tiel, Jan, Convex Analysis: An Introductory Text, New York: John Wiley
& Sons, 1984.

Vancil, Richard F., Passing the Baton: Managing the Process of CEO Suc-
cession, Boston: Harvard Business School Press, 1987.
Notation Index 79

Index of Abbreviations &

Notation

Note: Numbers refer to the page of rst and/or important use.

A (set of feasible actions), 36

a is an action (usually), 36

cdf (cumulative distribution func-

tion), 63
CE (certainty equivalent value), 10
ceo (chief executive ocer), 3

E (expectation operator), 7
E{X|E} (expectation of X conditional
on event E occurring), 26

max (maximum function), 9n

min (minimum function), 9n

p denotes a price (usually), 6

() (cdf of standard normal), 63
() (density function of standard
normal), 63
denotes a probability (usually), 7
Pr{E} (probability of event E), 26
P V (present value), 7

r (interest rate), 6

s&l (savings and loan), 31

sec (Securities and Exchange Com-
mission, us), 27

t is an index of time (usually), 7

U () is a utility function (usually),

10
UR (reservation utility), 37
80 End Matter

Author Index

Aghion, Philippe, 22, 23 Smith, Adam, 1, 2

Spence, A. Michael, 22n
Baskin, Jonathan Barron, 3
Townsend, Robert M., 25
Caillaud, Bernard, 36n
Clark, Robert C., 16n, 17 van Tiel, Jan, 11n
Vancil, Richard F., 61n
Dean, John W., 1n
DeGroot, Morris H., 62 Wallace, Nancy E., 32
Weisbach, Michael S., 60, 61n, 62
Eldenburg, Leslie, 61n Wosinska, Marta, 61n

Gibbons, Robert, 4n, 26n

Grossman, Sanford J., 45n

Hall, Brian J., 71

Hansmann, Henry, 17, 19, 21
Hart, Oliver D., 45n
Hermalin, Benjamin E., 22, 23, 32,
36, 37n, 60, 61, 62n
Holmstrom, Bengt R., 61, 76n

Jensen, Michael C., 28

Kraakman, Reinier, 17, 19, 21

Laont, Jean-Jacques, 4n, 36n

Liebman, Jerey B., 71

MacAvoy, Paul W., 1n

Martimont, David, 4n, 36n
Miller, Merton, 4, 9, 25
Millstein, Ira M., 1n
Miranti, Paul J., 3
Modigliani, Franco, 4, 9, 25

Rabin, Matthew, 10n

Main Index 81

Index

Note: Slanted page numbers (e.g., that look like this 1234567890) refer to main
denitions.

Adelphia, 1, 2 diversication, 12
agency, 4
agent, 4, 35 East India Company
risk aversion of, 12, 37 Dutch, 3
hidden action, 35 English, 3, 17
timing, 36 Enron, 1, 2, 27, 66
hidden information, 35 entrenchment, 5
principal, 4, 35 expected utility, 10
risk neutrality of, 12, 37 maximizers, 10
Arthur Andersen, 66 expected value, 7
asset partitioning, 21 formula for, 7
asset substitution, 3, 29
average value, 8 fair-odds line, 40
fair-odds ratio, 40
bargaining xed-wage contract, 44
take-it-or-leave-it, 37 free cash ow, 28
blunted incentives, 56 free riding, 57

California Global Crossing, 1

incorporation in, 15 governance, 1
limited liability law in, 17 list of problems, 23
career concerns, 74
certainty equivalent, 10 imputed costs, 6
common knowledge, 36 incentive compatibility constraint, 37
concavity, 11n indierence curve, 39n
contingent compensation, 5 individual rationality constraint, 37
corporate form inuence activities, 73
advantages of, 17 information rent, 49, 52
costly state verication, 25 interest
compounding of, 7
debt overhang, 32 iso-prot line, 50
diminishing marginal utility, 10
strict, 11 law of large numbers, 8n
discount factor, 6 limited liability
82 End Matter

corporations predating, 17 state-contingent commodities, 38

signaling justication for, 22 riskless, 39
liquidity premium, 22 risky, 39
theory of, 38
market for corporate control, 67 stochastic, 7
market liquidity, 21 stock grants, 71
minority shareholder, 33 restricted, 71
mixed strategy, 19n stock options, 71
Modigliani-Miller Theorem, 4, 9, 25, strike price, 71
28, 30 sunk expenditure, 6
monitoring, 4
monitors team theory, 4, 18
auditors, 65 free-riding problem, 18
banks, 70 thrift, see savings & loan
the state as, 66 tournament, 72
third-party, 67 tradeo between incentives & insur-
moral hazard, 2 ance, 47
Tyco, 1, 2
Nash equilibrium, 19n type
in a signaling model, 22
opportunity costs, 5 in hidden-information agency, 50
space, 50
Parmalat, 2
phantom stock, 71 utility, 10
precision, 60 utility function, 10
present value, 6
formula for expectation of, 8 who monitors the monitors, 65
PV , 7 WorldCom, 1
prots, 5

random variable, 7
reservation utility, 37
residual claimant, 9
revelation principle, 26
risk aversion, 11
risk neutrality, 12

Sarbanes-Oxley Act, 66
savings & loan, 31
self dealing, 33
Shell Oil (Royal Dutch/Shell Group),
2
signaling, 22
pooling equilibrium, 22n
separating equilibrium, 22n
signaling theory, 4
South Sea Bubble, 1