The Role of Investment Banks in IPOs

and Incentives in Firms

Essays in Financial and Behavioral Economics
Inaugural-Dissertation
zur Erlangung des Grades
Doctor oeconomiae publicae (Dr. oec. publ.)
an der Ludwig-Maximilians-Universit at M unchen
2004
vorgelegt von
Bjorn Bartling
Referent: Prof. Dr. Klaus M. Schmidt
Korreferent: Prof. Sven Rady, Ph.D.
Promotionsabschlussberatung: 21. Juli 2004
Acknowledgements
First and foremost I would like to thank my supervisor Klaus Schmidt. This thesis
would not have been possible without his superb support, guidance, and encourage-
ment. I am also much indebted to my co-authors Andreas Park and Ferdinand von
Siemens. The joint work with both of them was and continues to be a great source of
inspiration and motivation.
Just as much I would like to thank my colleagues Brigitte Gebhard, Georg Geb-
hardt, Florian Herold, and Susanne Kremhelmer. They all contributed in many ways
to the completion of this thesis and provided an inspiring and very pleasurable envi-
ronment at the Seminar f ur Wirtschaftstheorie. I received comments and suggestions
from many other friends and colleagues. My thanks extend to all of them even more
so as they are too numerous to be all mentioned in name at this point.
Finally, I would like to thank the Faculty of Economics and Politics at Cambridge
University and Wolfson College Cambridge for hosting me as a visiting Ph.D. student in
the academic year 2000/01. Financial support from the European Commission, grant
no. HPMT-CT-2000-00056, is gratefully acknowledged.
Contents
Preface 1
1 IPO Pricing and Informational Eciency: The Role of Aftermarket
Short Covering 14
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.2 The Benchmark: Oer Prices Absent Aftermarket Short Covering . . . 18
1.2.1 The Model Ingredients and Agents Best Replies . . . . . . . . . 18
1.2.2 Derivation of the Separating Equilibrium . . . . . . . . . . . . . 24
1.2.3 An Intuitive Characterization of the Equilibrium . . . . . . . . . 27
1.3 The Impact of Aftermarket Short Covering . . . . . . . . . . . . . . . . 29
1.3.1 Overview of Short Covering and a Banks Strategy . . . . . . . 29
1.3.2 Equilibrium Analysis . . . . . . . . . . . . . . . . . . . . . . . . 30
1.3.3 How would the result change without signaling? . . . . . . . . . 34
1.4 Payo Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
1.4.1 Payo Comparison for the Investment Bank . . . . . . . . . . . 35
1.4.2 Payo Comparison for Issuer and Investors . . . . . . . . . . . . 38
1.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
1.6 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
1.6.1 Aftermarket Price Formation . . . . . . . . . . . . . . . . . . . 43
1.6.2 Threshold Prices . . . . . . . . . . . . . . . . . . . . . . . . . . 44
1.6.3 Approximate Closed Form Solutions . . . . . . . . . . . . . . . 46
1.6.4 Maximal Reputation Costs . . . . . . . . . . . . . . . . . . . . . 48
1.6.5 Omitted Proofs . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
CONTENTS ii
2 Investment Bank Compensation in Venture and Non-Venture Capital
Backed IPOs 54
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
2.2 A Stylized Model of the IPO Procedure . . . . . . . . . . . . . . . . . . 58
2.3 Investment Banks Equilibrium Price Choice . . . . . . . . . . . . . . . 64
2.3.1 Uninformative Spreads or Spreads Reecting the Issuers Inde-
pendent Signal . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
2.3.2 Spreads Reecting the Issuers Identical Signal . . . . . . . . . . 68
2.4 The Issuers Strategic Choice of the Spread . . . . . . . . . . . . . . . . 69
2.4.1 Equilibrium Spreads if the Issuer is Uninformed . . . . . . . . . 69
2.4.2 Equilibrium Spreads if the Issuer is Independently Informed . . 73
2.4.3 Equilibrium Spreads if the Issuer is Identically Informed . . . . 75
2.5 Results and Interpretation . . . . . . . . . . . . . . . . . . . . . . . . . 76
2.5.1 Positive Prots for Investment Banks . . . . . . . . . . . . . . . 77
2.5.2 VC Issuers set Lower Spreads than Non-VC Issuers . . . . . . . 78
2.5.3 Strong Commercial Banking Ties . . . . . . . . . . . . . . . . . 79
2.5.4 Underpricing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
2.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
2.7 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
2.7.1 Aftermarket Price Formation . . . . . . . . . . . . . . . . . . . 82
2.7.2 Threshold Prices . . . . . . . . . . . . . . . . . . . . . . . . . . 83
2.7.3 Approximate Closed Form Solutions . . . . . . . . . . . . . . . 85
2.7.4 Omitted Proofs . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
3 Working for Today or for Tomorrow: Incentives for Present-Biased
Agents 100
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
3.2 The Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
3.2.1 Present-Biased Preferences and Beliefs . . . . . . . . . . . . . . 103
3.2.2 Multi-Tasking with Immediate and Delayed Benets . . . . . . . 105
CONTENTS iii
3.2.3 Combining Present-Biased Preferences and Multi-Tasking . . . . 106
3.2.4 Benchmark: Incentives for Time-Consistent Agents . . . . . . . 107
3.2.5 Incentives for Sophisticated Agents . . . . . . . . . . . . . . . . 109
3.2.6 Incentives for Naive Agents . . . . . . . . . . . . . . . . . . . . 111
3.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
3.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
4 Inequity Aversion and Moral Hazard with Multiple Agents 120
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
4.2 The Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
4.2.1 Projects, Eort, and Probabilities . . . . . . . . . . . . . . . . . 124
4.2.2 Preferences: Risk- and Inequity Aversion . . . . . . . . . . . . . 125
4.3 Contracts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
4.3.1 Benchmark: The Single Agent Case . . . . . . . . . . . . . . . . 127
4.3.2 The Two Agents Case . . . . . . . . . . . . . . . . . . . . . . . 129
4.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
4.4.1 Inequity Aversion Renders Team Contracts Optimal . . . . . . . 134
4.4.2 Inequity Aversion Causes Additional Agency Costs . . . . . . . 135
4.4.3 Inequity Aversion and Eciency . . . . . . . . . . . . . . . . . . 138
4.5 The Nature and Size of the Firm . . . . . . . . . . . . . . . . . . . . . 144
4.6 Secrecy of Salaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
4.7 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
4.7.1 Rent Comparison . . . . . . . . . . . . . . . . . . . . . . . . . . 150
4.7.2 Disutility from Being Better O . . . . . . . . . . . . . . . . . . 151
4.7.3 Status Seeking . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
4.8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
4.9 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
Bibliography 159
Preface
This dissertation is comprised of two parts. Chapters 1 and 2 address the role of in-
vestment banks in initial public oerings. Chapters 3 and 4 analyze incentive provision
when agents are subject to a behavioral bias.
In Chapter 1 we model the procedure of an initial public oering (IPO) as a sig-
naling game and analyze how the possibility of potentially protable trading in the
aftermarket inuences pricing decisions by investment banks. When maximizing the
sum of both the gross spread of the oer revenue and prots from aftermarket trading,
investment banks have an incentive to distort the oer price by employing aftermar-
ket short covering and exercise of the overallotment option strategically. This results
either in informational ineciencies or, on average, exacerbated underpricing. Wealth
is redistributed in favor of investment banks.
In Chapter 2 we address two puzzles of the IPO literature: (1) Why do investment
banks earn positive prots in a competitive market? And (2) Why do banks receive
lower gross spreads in VC backed IPOs? The IPO procedure is modeled as a two-
stage signaling game. In the second stage banks set oer prices given their private
information and the level of the spread. Issuers anticipate the banks pricing decision
and set in the rst stage spreads to maximize expected revenue. Investors are aware of
this process and subscribe only if their expected prots are non-negative. As a result,
issuers oer high spreads to induce banks to set high prices, allowing them prots.
Competition may take place in additional features of the IPO contract as, for example,
the number of co-managers or analyst coverage. We show that in equilibrium superiorly
informed VC backed issuers impose smaller spreads.
PREFACE 2
In Chapter 3 we examine self-control problems modeled as time-inconsistent,
present-biased preferences in a multi-tasking environment. An agent must allocate
eort between an incentivized and immediately rewarded activity (e.g. eort at the
workplace) and a private activity that pays out only tomorrow (e.g. studying for a
degree). Present-biased agents take decisions that do not maximize their long-run wel-
fare, irrespective of the intensity of incentives. Sophisticated agents are never harmed
by incentives relative to the case where incentives are absent as they always receive
their reservation utility levels. However, naive agents are always harmed in the pres-
ence of incentives as they wrongly predict future behaviors. Furthermore, we show that
the loss to a naive agent can exceed the principals gain from providing incentives. In
this case social welfare is reduced if the principal provides incentives.
In Chapter 4 we analyze how inequity aversion interacts with incentive provision in
an otherwise standard moral hazard model with two risk averse agents. We identify the
conditions under which inequity aversion increases agency costs of providing incentives.
We show, rst, that inequity aversion can render equitable at wage contracts optimal
even though incentive contracts are optimal with selsh agents. Second, to avoid
social comparisons the principal may employ one agent only, thereby forgoing the
ecient eort provision of the second agent. We nally discuss the implications of
social preferences for the internal organization and the boundary of the rm.
The decision whether or not to conduct an initial public oering is an important
decision in the life cycle of a rm. The advantages of having shares in a rm quoted on
a stock exchange are manifold. The owner of a rm can realize part of her investments,
it includes the ability to raise additional equity nance, or even the opportunity to set
up share option plans as incentive device for employees. However, there are also costs
of going public. In this context, initial underpricing is most extensively discussed.
Ritter and Welch (2002) report for 6,249 IPOs in the U.S. between 1980 and 2001 an
average rst-day return of 18.8 percent. It is usually argued that initial underpricing
constitutes a wealth transfer from the owner of the rm to the new shareholders, and
as such can be regarded as a cost of going public.
PREFACE 3
A number of explanations have been advanced for the underpricing anomaly which
seems to violate the fundamental tenet of no arbitrage. The most prominent ones as-
sume informational asymmetries between (or among) some of the main parties involved:
the issuing rm, the investment bank, and the investors.
Rock (1986) proposes a variant of Akerlofs (1970) lemons problem. He assumes
asymmetric information between dierent types of investors. Some are perfectly in-
formed about the intrinsic value of the shares on oer whereas others are uninformed.
Given the presence of informed investors, uninformed investors face a winners curse.
Informed investors subscribe only to hot IPOs. Assuming that shares are rationed,
uninformed investors stand a greater chance of being allocated shares in cold IPOs
from which informed investors abstain. To however attract uninformed investors to
subscribe to IPOs, shares have to be underpriced on average.
Another strand of the literature assumes asymmetric information between the is-
suing rm and the investors. The signaling models by Allen and Faulhaber (1989),
Grinblatt and Hwang (1989), and Welch (1989) argue that underpricing can in anal-
ogy to Spences (1973) job market signaling be a signal for a high quality of the
rm. A single crossing property is established by assuming that subsequent to the IPO
a secondary oering is conducted. In between these two oerings new information may
arise and reveal a low quality rms true value. This rm will then be unable to recoup
the loss from underpricing its shares by way of a secondary oering. A separating equi-
librium can thus be established in which only high quality rms underprice because
they can reap the gain from doing so in the secondary oering.
Apart from missing empirical support for signaling theories of underpricing
1
there
is the question why rms would not opt for a dierent, less costly signal? Booth and
Smith (1986), for example, put forth a theory of investment bank choice. Investment
banks as repeated players have reputational capital at stake and can thus certicate
the value of a rm. Other theories stressing the role of investment banks include Ben-
veniste and Spindt (1989). They assume asymmetric information between investment
1
Helwege and Liang (1996) report for a U.S. sample of IPOs in 1983 that only 4 percent of rms
conducted a secondary oering in the subsequent 10 years.
PREFACE 4
bank and investors. The latter hold superior information about the value of the shares,
and they are assumed to subscribe repeatedly to IPOs. Benveniste and Spindt design
a mechanism in which banks use underpricing and rationing to elicit investors infor-
mation prior to an IPO. If shares are underpriced, investors can be punished by small
allocations in subsequent oerings of other rms if the post-IPO phase reveals that
material information was withheld.
The existing theoretical literature however almost completely neglects that the role
of investment banks does not end with the distribution of shares at the day of the
oering. In fact investment banks pursue supposedly price stabilizing activities in the
aftermarket of IPOs that provide potentially protable trading opportunities. This
is where the model in Chapter 1 adds to the literature. We explicitly account for
stabilizing activities by investment banks in the aftermarket of an IPO and analyze
how this inuences the oer price decision in the rst place.
The regulating authorities allow investment banks to establish a short position in
an IPO by selling more shares than initially announced. Aftermarket short covering
refers to the practice of lling these positions in the aftermarket of an IPO. This is
done if the market price falls below the oer price. The idea is that lling short
positions stabilizes prices by increasing demand. The dierence between market price
and oer price is along the way pure prot for the investment bank. If the price
instead rises, the bank is hedged by an overallotment option which grants the right
to obtain additional shares from the issuer at the oer price. The U.S. Securities and
Exchange Commission (SEC) and the Committee of European Securities Regulators
(CESR) put forward the argument that stabilizing activities ensure an orderly market
as sudden selling pressure can be countered. In their latest respective release the SEC
(1997, p. 81) opines that aftermarket price stabilization promotes the interests of
shareholders, underwriters, and issuers.
In Chapter 1 we challenge this view by showing that in the context of our model
stabilizing activities result in either informational ineciencies or, on average, exacer-
bated underpricing. Furthermore, wealth is redistributed in favor of investment banks.
PREFACE 5
We presume that these side eects will not be intended by the regulating authori-
ties. Even without trading o potential benecial eects of stabilization against our
ndings, a policy implication arising from the analysis might be the alert that current,
well meant regulation can be gamed to the disadvantage of issuers and investors.
We propose a signaling model of the IPO procedure in which both the investment
bank and investors hold private information about the intrinsic value of the shares.
The bank moves rst and sets the oer price. Besides possible trading prots in
the aftermarket, banks are directly remunerated for their services by a fraction of
the oering revenue, the gross spread. In our model banks choose the oer price
strategically to maximize their prots form both the gross spread and trading prots
in the aftermarket. A higher oer price promises a higher revenue, it however reduces
the probability that the IPO is successful. An IPO gets called o if there are not
enough investors subscribing to it, and a higher oer prices reduces the number of
investors subscribing.
2
As benchmark, in a setting without aftermarket activities we identify the conditions
for the price equilibrium to be separating. In a separating equilibrium banks with
dierent information set dierent oer prices. A bank with favorable information
about the value of the rm deems it more likely that enough investors will hold alike
information. It will thus set a higher price than a bank with less favorable information.
We call a separating equilibrium informationally ecient since the banks information
is fully revealed by the oer price. In the aftermarket prices adjust according to market
demand. In equilibrium the security can turn out to be either under- or overpriced,
but on average there is underpricing.
We then introduce stabilizing activities to the model. This augments the incentive
to set high oer prices because the potential prot from aftermarket activities is higher
at higher prices. We nd that relative to the benchmark either the oer price falls
on average or there is a pooling oer-price equilibrium. In the rst case, to uphold
a separating equilibrium, an investment bank with favorable information distorts the
2
Busaba, Benveniste, and Guo (2001) report that about 14 percent of cases in their U.S. sample
of more than 2,500 IPOs between 1984-1994 get called o.
PREFACE 6
price downwards. This increases, on average, underpricing, i.e. the cost of going public.
In the second case, a separating equilibrium cannot be upheld and investors are thus
unable to infer the investment banks signal from the oer price. This equilibrium is
informationally inecient since investors decisions are based on private information
only and not, in addition, on the information of the bank. A major objective of nancial
market regulation is market transparency. It is thus benecial if prices contain more
rather than less information and, consequently, pooling equilibria are undesirable.
The debate among nancial economists on the costs of going public has mainly
focused on initial underpricing. Other, direct costs of going public like legal expenses,
audit fees, management time, accountancy, and the gross spread as investment bank
compensation, have received relatively little attention. Among practitioners matters
are dierent. When Chen and Ritter (2000) published that in more than 90 percent of
IPOs the gross spread is exactly 7 percent, numerous lawsuits against investment banks
for price collusion and a U.S. Department of Justice investigation of alleged conspiracy
among securities underwriters to x underwriting fees were initiated.
3
Chapter 2
proposes a dierent, subtle explanation of why gross spreads are so high that investment
banks are left with prots despite market competition. Furthermore, we address the
related puzzle of why venture capital (VC) backed IPOs are associated with lower gross
spreads than non-VC backed IPOs. To the best of our knowledge, Chapter 2 oers the
rst theoretical model to explain the level of gross spreads.
In Chapter 2 we model the IPO procedure as a two-stage signaling game. As in
Chapter 1 we assume that both investment banks and investors hold private infor-
mation about the intrinsic value of the shares. While Chapter 1 is silent about the
role of issuers, in Chapter 2 two dierent types of issuers are explicitly modeled. VC
backed issuers are assumed to hold private information about the value of the rm.
In contrast, non-VC backed issuers are taken to be uninformed. In the second stage
of the signaling game the investment bank decides on the oer price, given both its
private information and the level of the spread. Issuers anticipate the banks pricing
3
See Hansen (2001) for an overview of reactions to the Chen and Ritter (2000) article.
PREFACE 7
decision. Hence, in the rst stage they set spreads strategically to maximize expected
revenue. On both stages we can have either pooling or separating equilibria in spreads
and prices, respectively. Investors are aware of this process and subscribe only if their
expected prots are non-negative.
We nd that it can be in the best interest of the issuer to oer high spreads. It is
the investment banks discretion to set the nal oer price. At high oer prices there
is the danger that the IPO gets called o, because at high prices there may not be
enough investors to subscribe to the oering. We assume that the bank then suers
a reputation loss. To nevertheless induce the bank to set a high oer price the issuer
must oer a high level of the gross spread. The investment bank then earns a rent:
Given any level of the gross spread, it could always deviate to a low price (at which
the IPO will never fail) and receive its share in the oer revenue with certainty. Hence,
to prevent banks from deviating they are oered a rent at high prices.
Our second main result addresses the dierences in spreads between VC and non-
VC backed issuers. We show that in equilibrium superiorly informed VC backed issuers
impose smaller spreads. A VC backed issuer with good news about the value of the
shares regards it as likely that the investment bank also holds favorable information.
The issuer then wants the bank to transform this information to the investors via a
separating oer price equilibrium. An issuer with bad news will however always mimic
the issuer with favorable information: Issuers receive more than 90 percent of the oer
revenue and thus have a strong interest in high prices and signaling bad news via
the level of the spread reduces investors rating of the shares. In equilibrium we thus
observe a pooling spread level and separation oer prices. In contrast, uninformed
non-VC backed issuers prefer the bank to hide its information and set a pooling price
(with uninformed issuers the level of the spread itself cannot carry information). We
show that the according spread oered by non-VC backed issuers is smaller than the
separating price equilibrium inducing spread set by VC backed issuers.
The economic agents in Chapters 1 and 2 the investors, investment banks, and
issuers are modeled in accordance with the standard paradigm of neoclassical eco-
PREFACE 8
nomics: Individuals have stable and coherent preferences, are purely self-interested,
and fully rational. The investors have no trouble in solving a rather complicated Per-
fect Bayesian Equilibrium in order to decide whether to order or to abstain. The
investment bank is interested in its own material payo only and does not care that
some actions taken may harm others. And the issuer does not waive his decision to go
public just because the day of the oering arrives.
Over the last decade, however, empirical and experimental evidence mounted against
the paradigm of homo economicus. Numerous studies have shown that even in eco-
nomically relevant environments people systematically deviate from the predictions of
the standard theory. The resulting pursuit for greater psychological realism led to the
eld of behavioral economics, which extends the scope of economics by incorporating
ndings from experimental economics, psychology, and sociology into economic the-
ory.
4
The models in Chapters 3 and 4 pay tribute to this development. We apply
recent behavioral insights, especially present-biased preferences and inequity aversion,
to contract theory in order to analyze how optimal incentive provision within rms
changes if the idea of agent is broadened.
One strand of the literature in behavioral economics analyzes the consequences
of time-inconsistent preferences. In a parsimonious way, the standard model of ex-
ponential discounting captures the fact that people have a preference for immediate
gratication. Exponential discounting however implies, in addition, that intertempo-
ral trade-os remain unaected no matter when a decision is taken. The evidence, by
contrast, shows that people exhibit present-biased preferences.
5
They show very sharp
impatience for short horizons but are much more patient at long horizons. By way of
example: When being asked to decided whether to work 8 hours on, say, Monday four
weeks from now and relax on Tuesday or, alternatively, to relax on Monday and work
9 hours on Tuesday, most people will opt for the rst choice. However, when being
asked again Sunday four weeks from now, present-biased preferences may come into
4
For an overview see, for example, Camerer, Loewenstein, and Rabin (2003).
5
In the wording of Rabin (2002): As absurd as it sounds, it is probably true to say that exactly
zero papers in all social and behavioral sciences have proposed a test of the basic exponential versus
hyperbolic discounting [...] and claimed exponential explains the generated data better. (p. 19)
PREFACE 9
play. Many people then reverse their decisions and put o the work for another day
even though this implies to work an extra hour. This kind of behavior is often referred
to as hyperbolic discounting.
The theory of hyperbolic discounting has very fruitfully been applied to savings
decisions. Laibson (1997) shows that present-biased preferences can explain why con-
sumption tracks income more tightly than predicted by the standard life-cycle model
of savings, especially in the absence of commitment devices. ODonoghue and Rabin
(1999c) apply hyperbolic discounting to retirement planning. In the U.S. many em-
ployees are eligible for a so called 401(k) retirement savings plan. Participation rates in
these savings plans are relatively low, which is surprising because they are subsidized
by the government and sometimes by the employer as well. Even though present-biased
agents will want to eventually participate (in the long run people are patient), there
is always something that hinders them to join because it promises a greater immedi-
ate reward (in the short run people are impatient). Consequently, these agents may
procrastinate indenitely. ODonoghue and Rabins theory is well supported by re-
cent evidence. Madrian and Shea (2001) nd that automatic enrollment of employees
in 401(k) plans (employees must choose to opt out of rather than opt into the plan)
exerts a strong inuence over their saving choices.
ODonoghue and Rabin (1999b) examine the implications of hyperbolic discounting
for incentive provision. They analyze a principal-agent setting in which the agent must
accomplish a single task but has discretion when to do it. A procrastinating agent is
assumed to face stochastic costs of completion. The principal has an interest in timely
completion and thus oers an incentive contract to induce the agent to nish the task
in time. With asymmetric information about costs timely completion and ecient
delay must be traded-o. The optimal contract involves deadlines and increasing
punishment for delay.
The model in Chapter 3 adds to the literature by further exploring how present-
biased preferences and incentive provision interact. Complementary to ODonoghue
and Rabin (1999b), we examine self-control problems in a multi-tasking environment.
PREFACE 10
An agent must allocate eort between an incentivized and immediately rewarded ac-
tivity eort at the workplace and a private activity that pays out tomorrow only
say, caring for ones health. The seminal contribution on multi-tasking principal-
agent theory is Holmstrom and Milgrom (1991). They analyze the implications on
optimal incentive intensity, ownership structure, and job design if some activities are
more dicult to measure than others. The focus of Chapter 3 is dierent. We explore
how a present-biased agents allocation of eort between dierent tasks is aected by
incentives if eort invested in some tasks pays out immediately while eort devoted to
other tasks pays out with some delay.
Following the literature we assume that present-biased agents can be either naive or
sophisticated. Naive agents are ignorant to the fact that they exhibit time-inconsistent
preferences. They think in the future they will act like time consistent agents. In
contrast, sophisticated agents are aware of their time-inconsistent preferences. In the
context of our model, the dierence in crucial. We consider a three period setting. In
the rst period a principal oers an incentive contract and the agent decides whether
or not to participate. In the second period the agent chooses eort levels whose cost
accrue immediately. While eort devoted to the principals purposes is remunerated
in the second period, eort devoted to the agents private benet pays out in the nal
period only. Therefore, by the time a naive agent is oered a contract he holds in
contrast to a sophisticated agent a too optimistic belief about second period behavior.
The principal takes agents beliefs into account when oering a contract, and we show
that he can exploit naivete.
The focus of the model is the comparison between a situation in which the agent
can engage in the private activity only, and a situation in which the principal adopts
an additional production opportunity (for engagement in which he oers incentives).
We show that present-biased agents take decisions that do not maximize their long-run
welfare, irrespective of the presence of incentives. Sophisticated agents are, however,
never harmed additionally by the adoption of incentive contracts relative to the case
without incentives. They always receive their reservation utility levels. In contrast,
naive agents can be harmed if the principal oers an incentive contract. The incentive
PREFACE 11
contract endows them with an additional occasion to give in to their present bias.
Due to the fact that naive agents erroneously belief to act like time-consistent agents
in the future, they do not ask for and consequently do not get a compensating
payment when accepting the contract. Furthermore, we show that the resulting loss
to the naive agent can even exceed the principals gain from providing incentives. In
this case, we nd that social welfare is reduced by the adoption of incentive contracts.
Another strand of the literature in behavioral economics is concerned with the
notions of fairness and reciprocity. The standard hypothesis on human motivation
assumes that all people are exclusively interested in their material self-interest. In
recent years results from experimental economics have challenged this hypothesis by
showing that many people are, in addition, strongly motivated by concerns for fairness
and reciprocity. The most prominent experiment in this context is the ultimatum
game. A pair of subjects has to agree on a division of xed amount of money. One
subject (the proposer) must propose a division of the amount. The other subject (the
responder) can accept or reject the division. In case of rejection both subjects receive
nothing. In case of acceptance the proposed division is implemented. Given standard
assumptions on human motivation, the subgame perfect equilibrium prescribes that the
proposer oers the smallest monetary unit (or even nothing) to the responder, and the
latter accepts because little is better than nothing. The experimental outcomes across
hundreds of replications of the ultimatum game are however signicantly dierent from
the theoretical prediction. Responders reject with probability .4 to .6 proposals oering
less than 20 percent of the available surplus, and the probability of rejection decreases
as the size of the oer increases. About 70 percent of proposers oer between 40 and
50 percent of the available surplus.
6
Early experimenters like the recent Nobel Prize winner Vernon Smith conducted
market experiments and found that experimental markets converge quickly to the com-
petitive equilibrium. These results have been interpreted as conrmation of the self-
interest hypothesis. However, recently developed models of inequity aversion by Fehr
6
See, for example, Fehr and Schmidt (2003) for an overview of the experimental evidence.
PREFACE 12
and Schmidt (1999) and Bolton and Ockenfels (2000) show that in market environments
the standard competitive equilibrium may prevail even if agents are strongly concerned
about fairness. The theory of inequity aversion assumes that some but not all agents
suer a utility loss if their own material payos dier from the payos of other agents in
their reference groups. It can be shown that the interaction of the distribution of types
(fair or selsh) with the strategic environment (market or non-market interaction) can
explain why in some situations very unequal outcome may prevail (competitive equi-
librium) while in others rather equitable outcomes are obtained (ultimatum game).
The reason is that in environments with few players only, a fair agent may be able to
enforce an equitable outcome while this is not possible in a market setting. The virtue
of theories of inequity aversion is that they can in contrast to other models of social
preferences like altruism or envy account for a large number of seemingly dierent
results in wide array of experimental settings.
The model in Chapter 4 goes a step further and applies the theory of inequity
aversion as formulated by Fehr and Schmidt (1999) to contract theory. We analyze
how fairness concerns aect incentive provision in an otherwise standard moral hazard
model with two risk-averse agents. In a classic contribution to the theory of incen-
tives Holmstrom and Milgrom (1991, p. 24) state that it remains a puzzle for this
theory that employment contracts so often specify xed wages and more generally that
incentives within rms appear to be so muted, especially compared to those of the
market. In Chapter 4 we show that inequity aversion can serve as an explanation for
the scarcity of incentive contracts within rms.
In the standard principal-agent moral hazard model the optimal contract trades o
incentive provision and agents risk bearing. Eort choices cannot be contracted upon
such that wages must condition on stochastic output realizations. In optimum the agent
bears some risk for which he must be compensated. This constitutes the agency costs
of providing incentives. We nd that behindness aversion (an agent incurs a utility loss
only when being worse o than the other agent) among agents unambiguously increases
agency costs. This holds true if agents also suer from being better o, unless they
PREFACE 13
account for eort costs when comparing to the other agent. Fairness concerns increase
agency costs because they impose an additional restriction on the design of optimal
contracts. Inequity aversion eects an utility loss in states of the world with diverging
output- and thus diverging wage realizations. The resulting, reduced utility levels could
be generated without inequity aversion as well simply by lowering the respective wage.
Since these lower utility levels (at lower wage costs) were not optimal without inequity
aversion, they cannot be optimal now.
We nd that increased agency costs can undermine eciency in two ways. First, in-
equity aversion may render equitable at wage contracts optimal even though incentive
contracts are optimal with selsh agents. Empirical studies (see, for example, Bewley
(1999)) suggest that organizations likes rms are characterized by a dense network
of social relations. Market interactions are, in contrast, rather anonymous. Taking
this into account, our rst main result oers an explanation of why incentives within
rms are muted as compared to those in the marketplace. Second, to avoid social
comparisons the principal may employ one agent only, thereby forgoing the ecient
eort provision of the second agent. We call this the reference group eect. This sec-
ond result has implications for the internal organization and the boundary of the rm.
Suppose the principal can set up dierent rms but doing so involves xed costs. The
principal now faces a trade-o. On the one hand, integrating several agents within a
single rm causes social comparisons and thus increased agency costs of providing in-
centives. On the other hand, separating agents into dierent rms involves additional
xed costs. The solution to this trade-o thus denes an optimal size of the rm.
Chapter 1
IPO Pricing and Informational
Eciency: The Role of Aftermarket
Short Covering

1.1 Introduction
Since the Securities Act of 1934, it is legal practice in the U.S. that oering syndicates
stabilize market prices of their recent public oerings. In their latest release the U.S.
Securities and Exchange Commission (SEC) states: Although stabilization is a price
inuencing activity intended to induce others to purchase the oered security, when ap-
propriately regulated it is an eective mechanism for fostering the orderly distribution
of securities and promotes the interests of shareholders, underwriters, and issuers.
8
With this paper we challenge the assertion that current regulation always serves the
interests of all involved parties. We argue that issuing investment banks can combine
two regulated stabilization tools to generate risk-free prots. Employing a model that
captures the impact of this arbitrage opportunity on the oer price, we nd that (a)
either market transparency is lower or, on average, underpricing is exacerbated, and (b)
the issuing investment banks prots are boosted at the expense of issuer and investors.

The chapter is based on joint work with Andreas Park from the University of Toronto.
8
SEC (1997), Regulation M, Release No. 34-38067, p. 81. The Committee of European Securities
Regulators (CESR (2002)) proposes rules which resemble the SEC regulations.
AFTERMARKET SHORT COVERING 15
Current regulation allows investment banks to pursue the following three types of
aftermarket activities. First, stabilizing bids can be posted at or below the oer price
during the distribution period of the securities. Second, banks can establish a short
position by selling securities in excess of the pre-announced amount. Aftermarket short
covering refers to the practice of lling these positions in the aftermarket, which is done
if the market price falls below the oer price. If the price instead rises, the bank is
hedged by an overallotment option which grants the right to obtain typically up to 15%
additional securities from the issuer at the oer price. Third, penalty bids are used to
penalize customers who immediately resell their securities in the aftermarket.
Although on average IPOs have high rst-day returns, there is a signicant number
of IPOs with negative returns. In these cold IPOs, stabilizing bids and short covering
should ensure liquidity for the security to oset potential selling pressure (in the rst
days after the oat), and thus prevent sharp drops in prices. Penalty bids are meant
to reduce selling pressure. In this paper we focus exclusively on the impact of short
covering. An investment bank intending to support the security price adheres to the
following procedure. It enters the aftermarket short. This position must be lled
eventually. Suppose that the market price exceeds the oer price. Then there is
supposedly no selling pressure and no need to provide extra liquidity. Covering the
short position in the market, however, would be expensive. This is why almost all IPO
contracts include a so-called Greenshoe or overallotment option. It allows the bank to
buy extra securities from the issuer at the oer price. In the bulk of oerings, the initial
short position is perfectly hedged by this option. Increasing prices are therefore no risk
for the bank. Suppose now that the price drops. The bank does provide liquidity,
however, by doing so it also covers the short position in the market at a price below
the oer price. The dierence between the market price and the oer price (minus
the gross spread) is pure prot. In other words, the opportunity to enter the market
short, paired with the overallotment option, provides investment banks with a second,
risk-free potential source of income.
Only recently, new data became available that allowed to analyze investment banks
activities in the aftermarket directly. Aggarwal (2000) reports that underwriters utilize
AFTERMARKET SHORT COVERING 16
a combination of aftermarket short covering, penalty bids, and exercise of the overallot-
ment option. Stabilizing bids are never observed. Ellis, Michaely, and OHara (2000)
report that the lead underwriter always becomes the dominant market maker. They
also nd that market makers take large inventory positions, but reduce their risk by
exercising the overallotment option.
There are two cases studies which support the casual observation that aftermarket
trading can be very protable. Jenkinson and Ljungqvist (2001) provide a study of
the 1995 GenCo2 IPO (U.K.) during which the price fell in the aftermarket. The as-
signed investment banks Barclays de Zoete Wedd and Kleinwort Benson repurchased
45.7 million securities at the low market price to cover their short positions that were
established at the oer price. Jenkinson and Ljungqvist (2001) conclude: It demon-
strates how valuable the over-allotment option potentially is to the syndicate of invest-
ment banks selling the issue. Since they will buy back the shares in the market only
if the price is below the issue price, in closing (partially or in full) their short position
they make prots. These prots accrue to the syndicate itself, as the holder of the
option, rather than to the [. . . ] vendors (p. 180). Boehmer and Fishe (2001) analyze
a case-study of an IPO in which the lead underwriter took a nearly perfectly hedged
short position which was then covered in the aftermarket. The prots from trading
amounted to 52% of the syndicates overall prot from the oering. In their words:
[. . . ][short covering activities] represent an economically signicant prot opportunity
for the Lead (p. 4).
9
The existing literature on the impact of price support on oer prices models stabi-
9
Aggarwal (2000) nds that short covering is not expensive for underwriters (p. 1077). In
more detail, she nds that for weak oerings investment banks make prots, for strong oerings
however they may lose money. This stems from the fact that either the overallotment option is
not fully exercised or investment banks had established a naked short prior to the oering such
that short positions had to be covered at prices above the oer price. Our model cannot explain
why investment banks sometimes establish naked shorts or do not fully exercise the overallotment
option. We merely analyze the eects aftermarket short covering can have when investment banks
utilize the possibility to make risk-free prots, i.e. when they do not establish naked shorts and
fully exercise the overallotment option when prices rise. Ellis, Michaely, and OHara (2000) nd that
aftermarket activities of the lead underwriter are protable and account for about 23% of the overall
prot of underwriting. Reported prots stem from both market making and stabilizing activities
(that is accumulating inventory positions). From the presentation in the paper it does not seem
possible to disentangle whether stabilization contributed to or reduced trading prots. The claim
that stabilization can be a protable activity is thus not rejected by the data.
AFTERMARKET SHORT COVERING 17
lization to be costly. The two seminal theoretical papers on stabilization, Benveniste,
Busaba, and Wilhelm Jr. (1996) and Chowdhry and Nanda (1996), assume that banks
post stabilizing bids to keep prices up. However, such stabilizing bids are never ob-
served. Both models imply that stabilizing activities decrease underpricing our model
predicts the opposite. This paper thus contrasts the existing literature as we model
explicitly that investment banks can earn money in the aftermarket, and to the best
of our knowledge we are the rst to do so in a theoretical framework.
We propose a stylized model of an oering procedure that is in accordance with em-
pirical ndings and perceived industry practice. We assume that both the investment
bank and investors hold private information about the intrinsic value of the oered
security. We assume this information asymmetry to arise at a point in time when all
ocial, mandatory information has been released. Thus, any further public statement
by bank or issuer will be perceived as cheap talk, and it is only actions, i.e. price-setting,
that can convey additional information. We model the procedure as a signaling game
in which the investment bank moves rst and sets the oer price. It chooses the of-
fer price strategically to maximize its prots form both the gross spread of the oer
revenue and trading prots in the aftermarket. The bank anticipates investors best
replies to the oer price.
As a benchmark, we rst analyze a setting without aftermarket activities and iden-
tify the conditions for the equilibrium to be both unique and separating (that is, a
bank with dierent information sets dierent prices). We call a separating equilibrium
informationally ecient since the banks information is fully revealed by the oer price.
After the oer is oated, prices adjust according to market demand. In equilibrium,
the security can turn out to be either under- or overpriced, but investors account for
this when ordering the security. We show that on average there is underpricing.
When introducing aftermarket short covering, relative to the benchmark one of
two outcomes transpires: either the oer price falls on average, or separation breaks
down and the oer-price equilibrium morphs into a pooling equilibrium. In the rst
case, an investment bank with favorable information distorts the price downwards
and thereby, on average, exacerbates underpricing. In the second case investors are
AFTERMARKET SHORT COVERING 18
unable to infer the investment banks signal from the oer price. This equilibrium is
informationally inecient since investors decisions are based on private signals only
and not also on the signal of the bank. A major objective of nancial market regulation
is market transparency. Without modelling an explicit payo from higher transparency
we simply assume that it is desirable if prices contain more rather than less information.
Consequently, pooling equilibria are undesirable.
Furthermore, the price distortion leads to redistribution of wealth in favor of the
investment bank. Looking at per-share prots, the issuer loses if separation prevails;
in a pooling equilibrium he is better o. The issuers losses are the investors gains
and vice versa. On the comparative statics side, an increase of the gross spread or the
amount of overalloted securities reduces the parameter-set with informational eciency.
The remainder of the paper is organized as follows. In Section 1.2 we introduce
our model of the oering procedure without aftermarket short covering and identify
necessary and sucient conditions under which the investment bank reveals its private
signals through separating oer prices. In Section 1.3 we introduce aftermarket short
covering, identify the conditions under which the investment bank pools in the oer
price and thus holds back its private information and show that, if separation is upheld,
prices fall on average. We also provide results on comparative statics. In Section 1.4 we
discuss the redistribution of prots. Section 1.5 concludes. Proofs and specications
of tools used in the equilibrium analysis are in the Appendix.
1.2 The Benchmark: Oer Prices in a Model
without Aftermarket Short Covering
1.2.1 The Model Ingredients and Agents Best Replies
Consider the following stylized model of the IPO process.
The Security. The security on oer can take values V V = 0, 1, both equally
likely. The number of securities is denoted by S.
AFTERMARKET SHORT COVERING 19
The Investors. There are N identical, risk neutral investors. N is assumed to be
strictly larger than S. They can either order one unit of the security or none. Each
investor receives a costless, private, conditionally i.i.d. signal s
i
V about the value of
the security. This information is noisy, i.e. Pr(s
i
= v[V = v) = q
i
with q
i
(
1
2
, 1). If an
investor orders, he may or may not obtain the security during the oering procedure; if
the issue is oversubscribed shares are distributed with uniform probability. If he does,
his payo is the market price minus the oer price. If the oer is not oated, his payo
is zero even if he ordered the security. An investors type is his signal. We refer to the
investor as a high-signal investor if s
i
= 1. For s
i
= 0, it is a low-signal investor.
The Issuer. We assume that the issuer has no strategic impact. He holds no private
information about the value of the security. The issuer signs a contract with an invest-
ment bank that delegates the pricing decision and constitutes the amount of securities
S to be sold.
10
It also species the gross spread of the oer revenue that remains
as remuneration at the bank. The issuers payo is thus fraction (1 ) of the oer
revenue if the oer is oated, otherwise it is zero.
The Investment Bank. The risk neutral investment bank who signed the contract
with the issuer receives a private signal s
b
V about the value of the security. This
signal is noisy and conditionally independent from investors signals. Yet it is more
informative, i.e. q
b
> q
i
, where Pr(s
b
= v[V = v) = q
b
. Signals characterize a banks
type. If s
b
= 1 we refer to the investment bank as a high-signal bank. For s
b
= 0, it
is a low-signal bank. The bank receives the signal after the contract has been signed
and then announces the oer price p.
11
If demand is too weak to match supply, i.e.
if the number of investors willing to buy is less than the number of securities to be
sold, we assume that the oer is called o.
12
In case of excess demand securities are
10
The two most widely used contracts between issuers and investment banks are rm commitment
and best eorts contracts. These contracts dier with respect to risk allocation and incentive provision
that may be necessary due to imperfectly observable distribution eort and asymmetric information
about the value of the securities. However, in this stylized model we abstract form these complications.
11
We discuss xed-price oerings vs. bookbuilding at the end of this subsection.
12
Busaba, Benveniste, and Guo (2001) report for a sample of 2,510 IPOs led with the SEC from
1984 to 1994 that 14.3% of the oerings got called o. Issuers have the option to withdraw an oer
AFTERMARKET SHORT COVERING 20
allocated at random. We assume that failure of the oering inicts xed costs C on the
investment bank.
13
These costs are external to our formulation and can be thought of
as deterioration of reputational capital. They may also capture the opportunity costs
resulting from lost market share when being associated with an unsuccessful IPO.
14
Without loss of generality, we do not specify any costs the oering procedure itself may
cause for the investment bank. Thus, if the oer is successful, the banks payo is pS;
if it fails, its payo is C.
Signaling Value of the Oer Price. An investor bases his decision on his private
information and on the information that the investment bank reveals about its own
signal through the oer price. We denote this information by (p) and write (p) = 1
if the price reects that the banks signal is s
b
= 1, (p) = 0 if the price reects that
s
b
= 0, and (p) =
1
2
to indicate that the price is uninformative. These three are the
only relevant cases in equilibrium. We refer to as the price-information about the
banks signal.
The Aftermarket Price. The equilibrium market price is determined by the ag-
gregate number of investors favorable signals. In our model this number is always
revealed, either directly through investor demand or immediately after the oat through
trading activities. Thus write p
m
(d) for the market price as a function of d 0, . . . , N,
the number of high-signal investors. Appendix 1.6.1 eshes out this argument and pro-
vides an extensive treatment of price formation.
Investors Decisions and Expected Payos. We admit only symmetric, pure
strategies; thus all investors with the same signal take identical decisions. These can
if the investment bank proposes a price that is perceived as too low. During the road show the bank
learns about investors valuations. In a rm commitment contract the bank uses this information to
propose an oer price such that it can nd enough investors to sell the entire oer; in a best eorts
contract, such that selling all securities will not be too dicult. This model abstracts from the issuers
option to withdraw, and it leaves no room to the bank to adjust the oer price to investors valuations.
13
The model could be extended to allow the bank to buy up unsold securities. Costs then result
from expensively bought inventory positions and not from failure. C would thus be smoothed. This
would, however, not alter our qualitative results but complicate the analysis considerably.
14
Dunbar (2000), for instance, provides evidence that established investment banks lose market
share when being associated with withdrawn oerings.
AFTERMARKET SHORT COVERING 21
then be aggregated so that only three cases need to be considered. First, all investors
buy, denoted B
0,1
, second, only high-signal investors subscribe, denoted B
1
, and third,
no investor buys, denoted B

. Thus, the set of potential collective best replies is

B := B
0,1
, B
1
, B

.
To compute his expected payo, an investors has to account for the probability of
actually getting the security. There are two cases to consider. In the rst, all investors
buy. Thus, market demand is N and all investors receive the security with equal
probability S/N. In the second case, only high-signal investors buy. If d 1 others
buy, then an investor receives the security with probability S/(d). If overall demand
d is smaller than the number of shares on oer, d < S the IPO fails and the investor
who ordered gets it with probability 0.
Investors order the security whenever their expected payo from doing so is non-
negative. Suppose only high-signal investors buy, B
1
. After observing the oer price,
an investors information set contains both his signal s
i
and the information inferred
from the oer price, (p). Since signals are conditionally i.i.d., for every V V there is
a dierent distribution over the number of favorable signals (s
i
= 1), which we denote
f(d[V ). The investors posterior distribution over demands is given by
g(d 1[s
i
, (p)) := Pr(V = s
i
[s
i
, (p)) f(d 1[V = s
i
)
+Pr(V ,= s
i
[s
i
, (p)) f(d 1[V ,= s
i
). (1.1)
Then for a high-signal investor, at price p his rational-expectation payo from buying
has to be non-negative,
N

d=S
S
d
(p
m
(d) p) g(d 1[s
i
= 1, (p)) 0. (1.2)
Likewise for B
0,1
, in which case the summation runs from 1 to N, S/d is substituted
with S/N, and s
i
= 1 is replaced by s
i
= 0.
AFTERMARKET SHORT COVERING 22
Threshold Prices. Denote by p
s
i
,
the highest price that an investor is willing to
pay in equilibrium if all investors with signal s
i
s
i
order, given signal s
i
and price-
information . Thus p
1,1
is the highest (separating) price with B
1
, p
1,
1
2
the highest
(pooling) price with B
1
, p
0,
1
2
the highest (pooling) price with B
0,1
, and p
0,0
the highest
(separating) price with B
0,1
. Note that at all these prices investors are aware that the
security price may drop in the aftermarket and that they may not get the security.
The threshold prices are formally derived in Appendix 1.6.2.
The Investment Banks Expected Payo. First consider case B
1
. Variable d
denotes the number of buys, i.e. the number of high-signal investors. If the true value
is V = 1, we have
Pr(d S[B
1
) =
N

d=S
_
N
d
_
q
d
i
(1 q
i
)
Nd
, (1.3)
analogously for V = 0. A bank with signal s
b
assigns probability
s
b
(S) to the event
that at least S investors have the favorable signal. Since the investment bank receives
its signal with quality q
b
, for s
b
= 1,

1
(S) = q
b

d=S
_
N
d
_
q
d
i
(1 q
i
)
Nd
+ (1 q
b
)
N

d=S
_
N
d
_
(1 q
i
)
d
q
Nd
i
. (1.4)

0
(S) is dened analogously. If the bank charges a price at which only high-signal
investors buy, its expected prot is
(p[s
b
, B
1
) =
s
b
(S) pS (1
s
b
(S)) C. (1.5)
Consider now B
0,1
, the case where the oer price is low enough so that all investors
are willing to buy, irrespective of their signals. The oer never fails, thus payos are
given by (p[B
0,1
) = pS. If the price is set so high that no investor buys, as in case
B

, a loss of C results with certainty.

Simplifying Assumptions. The unconditional distribution over favorable signals
is a composite of the two conditional distribution and thus bimodal To obtain closed
AFTERMARKET SHORT COVERING 23
form solutions (or rather approximations) for success-probabilities and prices, we make
two simplifying assumptions: the rst simplies computations, since the two modes of
the distribution over favorable signals are centered around N(1 q
i
) and Nq
i
. The
results of the paper will also hold if it was not satised, as long as S < N/2, but the
assumption allows us to get closed form solutions for success-probabilities. The second
assumption ensures that we can analyze the two underlying conditional distributions
separately.
Assumption 1.1 S = (1 q
i
)N.
For every signal quality q
i
, there exists an

N(q
i
) so that for all N >

N(q
i
) the two
conditional distributions over favorable signals generated by V = 0 and V = 1 do
not overlap.
15
By standard results from statistics, sucient for

N(q
i
) is

N(q
i
) >
64q
i
(1 q
i
)/(2q
i
1)
2
.
Assumption 1.2 The number of investors N is larger than

N(q
i
).
As a consequence of the second assumption we can apply the Law of Large Numbers
and DeMoivre-Laplaces Theorem.
16
Since we assume that the IPO fails whenever
d < S, Assumption 1.1 implies
0
(S) = (2 q
b
)/2 and
1
= (1 +q
b
)/2; in what follows
we thus omit S. A consequence of the Law of Large Numbers is that p
m
(d) 0, 1
for almost all values of d.
17
Fixed Price Oerings vs. Bookbuilding. On most stock exchanges in the world
IPOs are sold through bookbuilding (for instance in the US, the UK, Germany, but not
in France), whereas our model is a xed-price oering. Current regulation allows risk-
free aftermarket short covering prots and this paper tries to capture their strategic
15
To be more precise: We need to ensure that if V = 1, the probability of demand d < S is zero.
16
For instance, the mode of a binomial distribution is generally not exactly symmetric. However,
if N is large enough, we can apply DeMoivre-LaPlace (0 < q
i
2
_
q
i
(1 q
i
)/N < 1) and employ the
normal distribution instead. Thus we can treat each mode to be symmetric. The number traders has
to large enough so that for V = 0, there are almost never more than N/2 traders with a favorable
signal and vice versa for V = 1.
17
To be more precise, for d N/2, p
m
(d) = 1, and for d N/2, p
m
(d) = 0. Thus to get
interesting equilibria, it is necessary that S is strictly smaller than N/2. If it was not, an IPO where
only s
i
= 1 investors buy, would never be at risk of being overpriced as it fails in all overpriced cases.
AFTERMARKET SHORT COVERING 24
impact. These potential prots depend primarily on price movements and thus one
should study the the oer price as the strategic decision variable. In any imaginable
framework the investment bank faces a trade-o between higher revenue and likelihood
of failure. Thus it is reasonable to assert that the oer price or, depending on the
formulation, the bookbuilding span has signaling value. A xed-price mechanism is,
arguably, the simplest possible way to capture the prices strategic dimension.
A hypothetical bookbuilding model will capture the strategic dimension in a sim-
ilar fashion, yet the analysis would become less tractable without adding insight: In
bookbuilding, the investment bank must set a bookbuilding span. This span can cer-
tainly have signaling value because it is, arguably, similar to setting a single price (a
degenerate span). Suppose bookbuilding spans have to be suciently tight so that they
are strictly in the [0, 1]-intervals interior. During the bookbuilding period, investors
submit their orders which (potentially) reveal their private information just as with
our xed price mechanism. At the end of the bookbuilding period the investment bank
will set the nal selling price somewhere in the span, distribute the shares, and reveal
overall demand. As long as the span and thus the issue price in the span is strictly in
the interior of the [0, 1]-interval, secondary market prices will adjust to a price outside
the span. Our stylized, parsimonious model is rich enough to capture the same result
that a more complicated bookbuilding model would yield.
1.2.2 Derivation of the Separating Equilibrium
The focus of this paper is the pricing decision of the investment bank given its signal.
In the following we identify the conditions under which a prot maximizing investment
bank will reveal its information through the oer price. A separating equilibrium is
dened as informationally ecient since investors can derive the banks signal from the
oer price. In a pooling equilibrium information is shaded and thus it is informationally
inecient. In this case, investors decide only on the basis of their private signals.
The Equilibrium Concept and Selection Criteria. The equilibrium concept for
this signaling game is, naturally, the Perfect Bayesian Equilibrium (PBE). A common
AFTERMARKET SHORT COVERING 25
problem with PBEs, however, is their multiplicity, stemming equilibria being sup-
ported by unreasonable out-of-equilibrium beliefs. The common way to overcome
this problem is to apply an equilibrium selection rule such as the Intuitive Criterion
(IC), introduced by Cho and Kreps (1987). We follow this line of research and consider
only equilibria that do not fail the IC. All of these PBE selection devices favour sepa-
rating over pooling equilibria. It will turn out, however, that in our framework under
certain conditions the IC cannot rule out pooling price equilibria. Moreover, from the
perspective of the investment bank the pooling equilibrium then Pareto dominates any
separating equilibrium. It would thus be unreasonable not to assume that these equi-
libria will be picked. Thus in what follows, we will only consider equilibria that satisfy
the IC and among these, we consider those that are Pareto ecient for the bank
A pooling equilibrium is specied through (i) an equilibrium oer price p

from
which investors infer (ii) price-information =
1
2
, and (iii) investors best replies given
their private signals, , and p

. A separating equilibrium is (i) a system of prices

p

, p

and price-information such that (ii) at p

= p

, the high separation price, the

price-information is that the bank has the favorable signal, = 1, at p

= p

, the low
separation price, the price-information is that the bank has the low signal, = 0, and
(iii) investors best replies given their private signals, , and p

. In both separating
and pooling equilibria, for p , p

, p

out-of-equilibrium public beliefs are chosen

appropriately. The following result is a straightforward consequence of signaling, the
proof of which is in Appendix 1.6.5.
Lemma 1.1 [The Highest Possible Low Separating Price] There exists no separating
oer price p

> p
0,0
.
In any separating equilibrium, therefore, the low price must be such that all in-
vestors buy, and the highest such separating price, given price-information = 0, is
p

= p
0,0
. In what follows we refer to p
0,0
as the low separation price.
Signaling equilibria in our setting come in one of three guises: The already men-
tioned separating equilibrium, a pooling equilibrium in which only high-signal investors
buy, and a pooling equilibrium in which all investors buy. In the following, we charac-
AFTERMARKET SHORT COVERING 26
terize the conditions guaranteeing that only separating equilibria survive our selection
criterion.
Fix a potential price p [p
0,0
, p
0,
1
2
], the interval of potential pooling prices at which
all investors would buy. Dene
1
(p) as the price at which the high-signal bank would
be indierent between charging a risky price
1
(p) at which only high-signal investors
buy, B
1
, and a safe pooling price p with B
0,1
(all investors buy). Formally,

1
(p)S (1
1
)C = pS
1
(p) =
p

1
+
1
1

1
C
S
. (1.6)
Price
0
(p) is dened analogously for the low-signal bank. Thus price
s
b
(p) is the
lowest risky price that a bank with signal s
b
is willing to deviate to from safe price
p.
18
In what follows we refer to
1
(p) as the high-signal banks deviation price, and
to
0
(p) as the low-signal banks deviation price. It is straightforward to see that

0
(p) >
1
(p) for all p [p
0,0
, p
0,
1
2
], that is, the low-signal bank requires a higher
price as compensation for risk taking. In addition,
j
(p)/p > 0, j 0, 1, so the
higher the pooling price, the higher the lowest protable deviation price. We can now
establish our rst major result.
Proposition 1.1 (Conditions for Informationally Ecient Prices)
If (i) the high-signal banks deviation price from the highest safe pooling price is not
higher than the highest separating price,
1
(p
0,
1
2
) p
1,1
, and if (ii) the low-signal
banks deviation price from the low separating price is not smaller than the highest
risky pooling price,
0
(p
0,0
) p
1,
1
2
then there exists a unique PBE that satises the
Intuitive Criterion and it is the separating equilibrium (p

= p
0,0
, = 0, B
0,1
); ( p

=
minp
1,1
,
0
(p
0,0
), = 1, B
1
); (p ,= p

, p

, = 0, B
0,1
if p p
0,0
, B
1
if p
0,0
< p
p
1,0
, B

else).
Interpretation of the Proposition. The rst condition,
1
(p
0,
1
2
) p
1,1
, together
with the IC is necessary and sucient to rule out pooling equilibria in which all in-
18
Deviation to a high, risky price can lead to increased overpricing, which is commonly perceived
to be bad for a banks reputation. Nanda and Yun (1997) analyze the impact of IPO mispricing on
the market value of investment banks. They nd that overpriced oerings result in decreased lead-
underwriter market value. In our model, however, investors fully take into account that the oer price
may drop in the aftermarket. Modelling such reputation eects would thus be contradictory in our
setting.
AFTERMARKET SHORT COVERING 27
vestors buy, irrespective of their signals. The second condition,
0
(p
0,0
) p
1,
1
2
, ensures
that there is no pooling where only high-signal investors buy, B
1
. The IC itself ensures
that the bank with s
b
= 1 always charges the highest sustainable separating price. The
high separation price p

is the minimum of p
1,1
and
0
(p
0,0
). The bank cannot charge
more than p
1,1
, and it cannot credibly charge more than
0
(p
0,0
) as otherwise the bank
with s
b
= 0 would deviate. Finally, since
1
(p
0,0
) <
1
(p
0,
1
2
) p
1,1
, the bank with
s
b
= 1 is willing to separate. The proofs details are in Appendix 1.6.5. A denition
of the IC can be found, for instance, in Fudenberg and Tirole (1991)[ p.448].
Underpricing. In the context of this model the rst-day return is the dierence
between market price and oer price. We can establish the following proposition. The
proof is in Appendix 1.6.5.
Proposition 1.2 (Underpricing)
In a separating equilibrium, on average, securities are underpriced.
Interpretation of the Result. The intuition behind the result is clear: Both types
of investors only buy if their expected payo is non-negative. At p
0,0
the low-signal
investor just breaks even in expectation but the high-signal investor expects a strictly
positive payo. At p
1,1
the high-signal investor just breaks even and the low-signal
investor abstains. Thus, ex-ante, expected payo is positive, i.e. there is underpricing.
1.2.3 An Intuitive Characterization of the Equilibrium
The concept of deviation prices
s
b
is a convenient tool to describe restrictions. We will
now reformulate the conditions from Proposition 1.1 in terms of exogenous costs C.
This allows us to derive a simple linear descriptive characterization of the equilibrium.
Consider rst condition (i),
1
(p
0,
1
2
) p
1,1
. If C is so high that

1
(p
0,
1
2
) =
p
0,
1
2

1
+
1
1

1
C
S
> p
1,1
(1.7)
then a separating equilibrium cannot be sustained. Even a high-signal bank then
prefers to sell the security at a price where all investors buy.
AFTERMARKET SHORT COVERING 28
T
oer
price
E
C C

C

C
$
$
$
$
$
$
$
$
$
$
$$
p

= p
1,
1
2
p

= p
0,0
p

= p
1,1
p

=
0
(p
0,0
)
p

= p
0,
1
2
Figure 1.1: Threshold Costs and Equilibrium Prices. For costs smaller than C,
it holds that
0
(p
0,0
) < p
1,1/2
so that the low-signal bank chooses a risky price. A pooling
equilibrium in p
1,1/2
results. If C (C,

C) a separating equilibrium results. For C (C,

C)
the high-signal bank cannot charge the highest separation price p
1,1
but must set a lower
price
0
(p
0,0
) to prevent the low-signal bank from mimicking. For C [

C,

C) the high signal
bank can charge p

= p
1,1
. Finally, if C >

C it holds that
1
(p
0,1/2
) > p
1,1
so that even the
high-type bank prefers a safe price and pooling in p
0,
1
2
results.
Consider now condition (ii),
0
(p
0,0
) p
1,
1
2
. If C is so low that

0
(p
0,0
) =
p
0,0

0
+
1
0

0
C
S
< p
1,
1
2
(1.8)
then a separating equilibrium, again, cannot be sustained (by the SIC). In this case,
even a low-signal bank is willing to choose a high, risky pooling price and the high-signal
bank can thus not credibly signal its information. If C is so high that
0
(p
0,0
) > p
1,1
then for the low-signal bank it does not even pay to deviate to the highest separating
price, p
1,1
. This bound on C is given by

C :=

0
p
1,1
p
0,0
1
0
S. (1.9)
Dene, analogously,

C and C such that (1.7) and (1.8) hold with equality. We
get C <

C <

C. The following Corollary to Proposition 1.1 summarizes the above
characterization.
Corollary 1.1 (Proposition 1.1 in Terms of Costs)
If C (C,

C) then the unique equilibrium is the separating equilibrium stated in Propo-
sition 1.1. If C (C,

C) then p

=
0
(p
0,0
), and if C [

C,

C) then p

= p
1,1
.
AFTERMARKET SHORT COVERING 29
It has often been argued that certifying agents, here the investment bank, must have
enough reputational capital at stake to make certication credible. In this context,
also too much reputation can inhibit certication (separation from a low-signal bank)
if it becomes to expensive to jeopardize ones reputation at a high, risky oer price.
Figure 1.1 plots threshold costs and corresponding equilibrium prices.
1.3 The Impact of Aftermarket Short Covering
In this section we extend the model and allow the investment bank to pursue aftermar-
ket short covering. We analyze its eect on the investment banks pricing decision and
investigate under which conditions informational eciency will be undermined. We
nd that, in general, the conditions for a separating equilibrium become more restric-
tive. Upholding separation may come at a cost thus on average the investment bank
has to distort prices down, which causes more underpricing.
1.3.1 Overview of Short Covering and a Banks Strategy
With aftermarket short covering the investment bank has the opportunity to allot a
predetermined amount of up to O securities on top of the principal volume of securities
S. This amount O is referred to as the overallotment facility. It typically constitutes
15% of the number of initial securities S. The investment bank goes short in a position
of this size. If the market price falls below the oer price, the bank lls its short
positions in the aftermarket. This practise is referred to as aftermarket short covering.
If the price is below the oer price, the bank makes a prot. If the market price rises
above the oer price, the bank exercises a so-called overallotment option, the right to
obtain up to O securities from the the issuer at the oer price. The option is only valid
if the bank had indeed established a short position. Consequently, the bank is perfectly
hedged against rising prices. We restrict attention to the case where either the entire
amount of S +O securities is sold or, if only fewer securities can be sold, the IPO fails;
the restriction merely simplies the analysis and does not aect the qualitative results.
The bank receives the gross spread only on the securities that actually remain oated.
AFTERMARKET SHORT COVERING 30
Intuitively, the size of a potential price drop and thus of prots from aftermarket
short covering is larger the higher the oer price. In the benchmark cases separating
equilibrium, a low-signal bank would not mimic a high-signal bank because it fears
costs from a potential IPO failure. With short covering expected aftermarket prots
are higher the larger the potential price drop. Moreover, a bank with a low signal
considers such a drop more likely. It is then possible, that potential losses from a failed
oering are oset by higher expected aftermarket gains. Two scenarios are possible:
In equilibrium, the high-signal bank sets a lower price to separate from a low-signal
bank. The high-signal bank, however, is only willing to do so as long as separation
pays. Thus there is a point where defending separation becomes too costly so that
the high-signal bank pools with the low-signal bank and an informationally inecient
outcome results.
1.3.2 Equilibrium Analysis
We write
1
(p

, B, s
b
) for the investment banks expected prots from their share of the
oer revenue. Let
2
(p

, B, s
b
) denote the expected second period prot from lling
the short position at lower prices. In case of a separating equilibrium these are

2
( p

, B
1
, s
b
= 1) =
N

d=S + O
O max p

(1 ) p
m
(d), 0 Pr(d[s
b
= 1) (1.10)
if the price is risky and s
b
= 1. For safe prices, the summation in
2
(p

, B
0,1
, s
b
= 0) is
from 0 to N, as the IPO never fails. The conditional distribution Pr(d[s
b
) of demand
d is the distribution derived for
s
b
(S). Note that a high-signal bank sums from S+O,
since lower demand leads to a failure of the IPO. An investment bank with s
b
= 0,
on the other hand, sums from 0 since the IPO is always successful. The bank also
accounts for the foregone gross spread when buying back in the market.
The market price after the oering p
m
(d) adjusts according to investors signals and
with respect to these signals it is informationally ecient. The bank cannot stabilize
against this ecient price, but, of course, if the price is ecient, it need not and
must not be stabilized. In our model it is, therefore, not possible to study potentially
AFTERMARKET SHORT COVERING 31
benecial eects of price stabilization. More generally, however, if one believes in
ecient markets, stabilization is undesirable and, if at all, it can have no more than a
short-term impact.
With short covering, a high separation price, p

has to be small enough so a low-

signal bank cannot protably deviate from the low, riskless price, p
0,0
. Thus the in-
vestment bank with s
b
= 1 has to determine

0
(p
0,0
) so that

1
(

0
(p
0,0
)[s
b
= 0, B
1
) +
2
(

0
(p
0,0
)[s
b
= 0) =

1
(p
0,0
[s
b
= 0, B
0,1
) +
2
(p
0,0
[s
b
= 0). (1.11)
In what follows, we make two further assumptions. The rst states that the overall
amount of shares that can be issued remains constant relative to the scenario without
aftermarket short-covering. This simplies computations and later allows us to com-
pare the relative payos in both scenarios. The second requires that together signals
of investors and bank are suciently informative. Figure 1.2 has an illustration of
Assumption 1.4.
Assumption 1.3 S + O = (1 q
i
)N.
Assumption 1.4 q
i
and q
b
are large enough so that p
1,
1
2
> 2p
0,0
.
Using Assumptions 1.3 and 1.4, we can prove the following lemma.
Lemma 1.2 (The Low-Signal Banks Deviation Price Drops)
The low-signal banks deviation price with short covering is smaller than without short
covering,
0
(p)

0
(p) p [p
0,0
, p
0,
1
2
].
In the proof we show that for any low-signal banks deviation price
0
(p), second period
prots from aftermarket short covering for the low-signal bank are higher at the high,
risky price p

. Consequently, this bank has an additional incentive to deviate. The

low-signal bank considers it more likely that the price drops, hence its potential gain
from short covering is large, in particular, relative to what it can gain by setting the low
separation price. To prevent a low-signal bank from mimicking, the high-signal bank
AFTERMARKET SHORT COVERING 32
has to reduce its oer price. The proof is in Appendix 1.6.5. In what follows, if there
is a switch from separating to pooling, we restrict attention to those switches that are
to the risky pooling price p
1,
1
2
.
19
We can now establish the main result. Analogously
to Corollary 1 we spell it out in terms of separation costs as this allows for a more
straightforward interpretation. The proof can be found in Appendix 1.6.5.
Proposition 1.3 (Equilibrium with Short Covering Relative to Benchmark)
1. There exists a lower bound threshold cost C

> C such that for all costs C

[C, C

), the only equilibrium that satises the Intuitive Criterion and Pareto ef-
ciency is a pooling equilibrium at the highest risky pooling price p
1,
1
2
.This price
is informationally inecient.
2. There exists an upper bound threshold cost

C

such that for all costs C [C

,

C

]
the unique equilibrium that satises the Intuitive Criterion and Pareto eciency
is a separating equilibrium. For the high separating price p

there exists a thresh-

old cost level

C

[

C,

C

) so that
(a) for costs C [C

,

C

) the high separation price is the low-signal banks devia-

tion price from the low separating price, p

0
(p
0,0
), p
1,
1
2
<

0
(p
0,0
) < p
1,1
,
and
(b) for costs C [

C

,

C

] the high separation price is the highest possible risky

price p

= p
1,1
.
On average, underpricing in the separating equilibrium is exacerbated.
Interpretation of the Result. The rst part of the proposition states that for all
costs smaller than C

, both types of the bank prefer to pool and hence prices are
informationally inecient. Since C

> C pooling occurs for a region of parameters

where without aftermarket short covering there was separation. That is, the cost
19
Our results on informational eciency are not aected by this restriction. On the contrary,
taking pooling in a risk-free price also into account would strengthen our ndings. In addition, if
there is a choice between the high, risky pooling price, p
1,
1
2
, and the low, safe pooling price, p
0,
1
2
, the
former will always generate more ex-ante revenue. We thus focus on high pooling prices to keep the
analysis simple.
AFTERMARKET SHORT COVERING 33
region for which we get informational eciency becomes more restrictive. The second
part of the proposition outlines the region in which separation is sustained. For all
costs smaller than threshold

C

, the investment bank with the good signal charges

0
(p
0,0
), which, by Lemma 1.2, is smaller than the price charged in that corresponding
parameter region without short covering. In other words, for costs between C

and

C

oer prices drop. By Proposition 1.2, there is underpricing in a separating equilibrium.

Thus, on average, underpricing is exacerbated when separation is sustained. At rst
glance this result is surprising since second period expected gains are larger the higher
the oer price. One might expect that agents are then more inclined to set higher
prices. In our model, this casual intuition fails.
The Impact on the Upper Threshold Level for Costs. So far we have focussed
on the relation of lower bound threshold costs C and C

and middle bound threshold

costs

C and

C

. Surely, if

C

increases relative to

C (by Lemma 1.2) and C

increases
relative to C, then also

C

should increase relative to

C. But this is not necessarily
true it may actually decrease. Furthermore, if it does increase, it is irrelevant. This
is why: Keeping N, , and O xed,

C and

C

are functions of the signal qualities q

b
and
q
i
. For low signal qualities,

C

actually decreases. For such values the high separation

price p
1,1
and the low, risk-free pooling price p
0,
1
2
are close. Expected aftermarket
prots are higher for the risk-free price and this outweighs the lower expected pooling
revenue. For high values of q
b
and q
i
, both

C and

C

exceed the natural upper bound

for costs: The worst that can happen, is that a bank loses all (discounted) future
business. This upper bound on C can be estimated. In Appendix 1.6.4 we go into the
details of this argument, but in what follows we restrict attention to C, C

,

C, and

C

.
To summarize: The rst case of a decreasing upper bound strengthens our result, the
second case does not weaken our argument.
Comparative Statics. We can express the overallotment option O as share r of S,
that is S + O = (1 + r)S. Thus, r = 0 is the benchmark case without short covering.
Potential policy variables in this setup are the banks share of the revenue, , and
AFTERMARKET SHORT COVERING 34
the size of the overallotment option, r. The proof of the following Proposition is in
Appendix 1.6.5.
Proposition 1.4 (Comparative Statics)
The conditions for informational eciency become more restrictive for the gross spread,
, or the amount of the overallotment facility, r, increasing.
Interpretation of the Proposition. A higher level of or an increased amount
of r strengthen an investment banks incentive to set higher prices. For a high-signal
bank it is thus more dicult to defend a high separation price, consequently, more
pooling results.
1.3.3 How would the result change without signaling?
In order to understand the impact of signaling, consider the case where the investment
bank gets no signal at all. This is equivalent to the case of a neutral signal q
b
= 1/2.
The conditional probability of there being at least S high-signal investors is
(S) =
N

d=S
_
N
d
_
1
2
_
q
d
i
(1 q
i
)
Nd
+ (1 q
i
)
d
q
Nd
i
_
. (1.12)
Here an oer price has no signaling value, investors learn nothing from it. If an investor
has favorable signal s
i
= 1, he buys the security if p p
1,
1
2
, if he has s
i
= 0 he buys if
p p
0,
1
2
. Thus price p
0,
1
2
is risk-free. The investment bank then sets risky price p
1,
1
2
,
if its expected payos are higher than those for the risk-free price,
(S)p
1,
1
2
S (1 (S))C p
0,
1
2
S, (1.13)
and it sets p
0,
1
2
otherwise. Thus there exists a threshold

C, such that for all costs
C

C, the investment bank would charge the high price p
1,
1
2
, and for all C >

C,
it would play safe and charge p
0,
1
2
. However, once short covering is introduced, this
second prot opportunity may enable the investment bank to charge a higher price.
Simulation of prices show that,
2
(p
1,
1
2
) >
2
(p
0,
1
2
). Thus there exists a threshold
AFTERMARKET SHORT COVERING 35
cost

C

larger than

C such that the investment bank charges the higher, riskier price
where it used to charge the low price. In this case, there would be more overpricing,
for C [

C,

C

). This contrasts our signaling model, which produces the opposite

eect: For a non trivial region of parameters we expect to observe, on average, more
underpricing.
1.4 Payo Analysis
Although the investment bank has a second source of prots, it is not immediately
obvious that it will indeed be better o if it has the high signal, it may have to
distort prices downwards. The bank will thus receive lower expected revenues that
may not be outweighed by short covering prots.
Measuring Payos. The investment banks expected payos can be measured at
two points in time: Ex-ante, that is before the bank receives its private information, and
interim, that is after the signals are realized but before investors take decisions. Issuers
have no private information, so their information is exclusively determined ex-ante. As
a convention, we compare per-share prots and costs.
Table 1.1 summarizes a banks conditional signal probabilities, the prices that are
charged for each signal, the conditional probabilities of a successful IPO and, given it
is indeed successful, the probability of short covering and its protability. For instance,
take V = 0 and s
b
= 1, which occurs with probability 1q
b
. In a separating equilibrium
the high-signal bank charges p without and p

with short covering. The IPO is successful

with probability 1/2 and, given this, there is short covering with probability 1. With
probability 1/2 the IPO fails and the bank incurs cost C. Note that if V = 1, by the
Law of Large Numbers, the IPO almost never fails.
1.4.1 Payo Comparison for the Investment Bank
We will trickle down from the the strongest to the weakest case: First we analyze the
interim type-specic payos. This is the strongest case, because we determine when
AFTERMARKET SHORT COVERING 36
Without Aftermarket Short Covering
s
b
= 1 s
b
= 0
Pr(s
b
[V ) Price Pr(success) Pr(s
b
[V ) Price Pr(success)
V = 1 q
b
p = minp
1,1
,
0
(p
0,0
) 1 1 q
b
p
0,0
1
V = 0 1 q
b
p = minp
1,1
,
0
(p
0,0
)
1
2
q
b
p
0,0
1
With Aftermarket Short Covering
s
b
= 1
Pr(s
b
[V ) Price Pr(success) Pr(short cov.) Prot p p
m
V = 1 q
b
p

= minp
1,1
,

0
(p
0,0
) 1 0 0
V = 0 1 q
b
p

= minp
1,1
,

0
(p
0,0
)
1
2
1 p

s
b
= 0
Pr(s
b
[V ) Price Pr(success) Pr(short cov.) Prot p p
m
V = 1 1 q
b
p
0,0
1 0 0
V = 0 q
b
p
0,0
1 1 p
0,0
Table 1.1: Summary of State-Prots. The table summarizes the probabilities of
signals given values, the separating prices that are charged in each case, the probabilities of
a successful IPO and, given that, the probability of short covering, and its protability. is
dened as (1 )r/(1 + r).
types gain individually. We then proceed with the ex-ante payo gains. We specify
under which conditions the bank would prefer a setting with short covering. Payos
are then averaged over signal-types because ex-ante, the signal is unknown. This is
the weaker case. We focus on the extreme scenarios, that is (a) on the costs with the
largest price drops after regime shifts and (b) on costs for which ex-ante payo with
short covering is lowest.
To derive the results, we construct the payo dierences from both settings at a
given threshold cost and then substitute in closed form approximations of the thresh-
old prices. Details of the formulae can be derived straightforwardly from Table 1.1.
Appendix 1.6.3 outlines how the risky threshold prices can be approximated. The
resulting risky threshold prices that we nd can be interpreted as
p
s,
=
expected liquidation value given the prices information content
fraction of cases where this information can be used
.
So for instance, if (p) = 1, the unconditional value of this information piece is the
q
b
, the quality of the banks signal. The fraction of cases where this information can
be used is the probability of a successful IPO, given (p) = 1: Here it is
1
. Thus
AFTERMARKET SHORT COVERING 37
p
1,1
= q
b
/
1
. By the same token p
1,
1
2
= .5/(3/4). For the statements below we
computed payo dierences for = 7% and O = .15 S, which are the empirically most
commonly observed parameters.
20
In summary:
Interim Payos for the Low-Signal Bank. Suppose that separation is main-
tained. Without short covering, per-share prots are p
0,0
. With short covering there
are additional expected aftermarket prots of q
b
p
0,0
, as can be seen from Table 1.1.
Suppose now that a pooling equilibrium results. Then, by denition of the pooling
equilibrium, the low-signal bank benets. In both cases the low-signal bank is better
o with short covering.
Interim Payos for the High-Signal Bank. If costs are lower than C or if costs
are higher than

C

, the bank always wins: In both cases expected revenue remains

constant and the bank also gets short covering prots. Suppose now that costs are in
(C,

C

) and that the price decrease is strongest, from from p

1,1
to p
1,
1
2
. Then for signal
qualities in areas A and B in the Left Panel of Figure 1.2, the bank is always better
o, despite the maximal price decrease; in areas C and D the bank loses. It may not
be better o in all cases, but the smaller the price decrease, the smaller areas C and D
become.
Ex-ante Payos There are two subcases to consider: (i) The threshold costs for
which the highest price decrease occurs, which is

C. (ii) The threshold cost for which
the ex-ante payo with short covering is lowest, which is C

.
(i) Suppose at

C, prices drop from separation in p
1,1
and p
0,0
to pooling in p
1,
1
2
.
At

C, without short covering the low type is indierent between riskless p
0,0
and risky
p
1,1
. Using the risky payos, without short covering payos are (
1

0
)p
1,1
costs.
20
Chen and Ritter (2000) report that is almost always 7 percent. Naturally, when both O is
small and is large, some of the statements below may change. Clearly, if these contract variables
are such that there is very little to be won in the aftermarket (low O) but a lot to be lost in revenue
(high ), then matters may change. However, the essence of the arguments below is that even at the
most extreme price drops there is a non-trivial parameter space where the bank is always better o.
Taking also parameters sets for and O into the description of the analysis would merely complicate
the exposition.
AFTERMARKET SHORT COVERING 38
With short covering payos are (
1

0
)p
1,
1
2
cost+short covering prots. So costs
cancel, revenues are lower, but, as it turns out, the short-covering prots always over-
compensate for the loss in revenue. Thus despite the maximum price decrease at

C,
ex-ante the investment bank is always better o with short-covering.
(ii) The bank has lowest ex-ante payos with short covering at C

, the costs where

the low bank is indierent between choosing risky separation price p
1,
1
2
=

0
(p
0,0
) and
riskless price p
0,0
. The most extreme drop in revenue happens when

C < C

, so that
without short covering, the bank plays a separation equilibrium. Note that at this
cost C

, without short covering the low type is not indierent between a risky and
a riskless price as
0
(p
0,0
) > p
1,1
. Payos without short covering are of the order

1
p
1,1
+ p
0,0
costs, with short covering they are
1
p
1,
1
2
+ p
0,0
costs+short covering
prots. The investment bank is better o for parameters q
b
, q
i
in areas A, B, and C,
but not D (Left Panel, Figure 1.2). With respect to signal qualities, this area appears to
be large. However, taking (i) into account, this is only relevant for a strict subinterval
of [

C,

C

] and if also C

>

C. For all other costs, the bank is ex-ante always better o.
The Right Panel of Figure 1.2 illustrates this point.
To summarize, in most cases the investment bank is ex-ante and interim better o.
1.4.2 Payo Comparison for Issuer and Investors
Given our model specication we can only compare the revenue that the issuer receives
in settings with and without short covering.
21
Suppose with short covering, separation
is maintained. If the separation price decreases, p

< p, the issuer loses. Suppose

now, there is a switch from separation to pooling. The high separation price decreases
from p to p
1,
1
2
, but at the same time the low separation price rises from p
0,0
to p
1,
1
2
.
Comparison of expected payos shows that in this case the issuer is better o for
all parameter values.
22
Investors prots are directly opposed to the issuers prot.
Whenever the issuer gains (in expectation) investors lose and vice versa.
21
This is equivalent to expected prots: Prot here would be dened as the dierence between
revenue per share and the true value, which, by the LLN, is identical to the aftermarket price. We do
not take other factors such as, for e.g., costs for alternative nancing (if the IPO fails) into account.
22
Recall that we restrict the analysis to per-share prots. Taking into account that the number
of securities eventually sold to the market will be lower with short covering it can be the case that
whenever, simultaneously, q
i
is very small and q
b
is very large the issuer is worse o even with pooling.
AFTERMARKET SHORT COVERING 39

q
b
1
`
q
i
1

0.5

A
B
C
D
Left Panel

-
-
-
-
-
-
-
-
-
-
-
--

.
.
.
.
.
..-
-

.
.
.
.
.
..-
-

C

C C
1
C
2
prot without
short covering
prot with
short covering
case B & C
case D
slope:

0
+
1
2
1

-
slope:

1
1

Right Panel
Figure 1.2: Informational Eciency and Sign of Banks Prot Change.
Left Panel: Areas A, B, C, and D indicate permitted values of q
i
and q
b
, i.e. q
b
> q
i
> .5 and
p
1,
1
2
> 2p
0,0
. For C =

C, A indicates where an informational ecient separating equilibrium
is uphold with short covering; in B, C, and D a pooling equilibrium results. The high-signal
bank is better o in A and B, and worse o in C and D at C =

C. Ex-ante, the bank is
better o in A, B, and C, for all C; in D there exist C
1
[

C, C

] and C
2
[C

,

C

] such that
for C [C
1
, C
2
] it may lose. The gure is based on simulated values for = .07, r = .15,
and N = 1000.
Right Panel: The lower line indicates ex-ante prots of the bank as a function of C without
short covering. The higher lines indicate prots with aftermarket short covering. For the
values of q
i
and q
b
in areas B and C these prots are always higher; in area D it may be the
case that for C [C
1
, C
2
] these prots are lower.
Even though this section is merely concerned with redistribution, it yields an inter-
esting insight. The investment bank is nearly always better o with aftermarket short
covering, in many cases irrespective of its signal. The issuer never gains but often loses
if separation is upheld, but always wins if separation morphs into pooling; the eect
on investors payos is the opposite.
1.5 Conclusion
Investment banks legally pursue supposedly price stabilizing activities in the post-oer
market. In this paper we analyze how these aftermarket activities inuence the setting
of the oer price in the rst place. We take a dierent perspective from existing
AFTERMARKET SHORT COVERING 40
theoretical work as we build the model around the stylized fact that investment banks
can realize risk-free prots through aftermarket short covering. The current model
cannot assess why some investment banks expose themselves to risk and establish
naked shorts, or why they do not exercise the overallotment option in full even when
prices rise above the oer price. This paper only explains the strategic impact of the
possibility of risk-free prots. The investment banks behavior must not be perceived
as rogue or fraud, but as a rational response to a change in the environment. Investors
anticipate the banks behavior and react rationally to it.
We propose a stylized model of an oering procedure that is in accordance with em-
pirical ndings and perceived industry practice. We assume that both the investment
bank and investors hold private information about the intrinsic value of the oered
security. Prices are set so that rational-expectation investors only order the security if
they expect to make a prot, taking into account the behavior of the investment bank.
The market price after the oering will adjust according to investors signals. As these
are conditionally i.i.d., the price almost surely reects the fundamental value of the
security. The bank cannot stabilize against this fully ecient price, but, of course, if
the price is ecient, it need not and must not be stabilized. So in the best of worlds,
one with full transparency, the bank can make an extra prot through short cover-
ing. In the real world the IPO process is opaque; neither investors nor regulators nor
researchers know precisely the banks strategies. It is certainly reasonable to assume
that in such an imperfect world the strategic impact of the second source of prots is
rather more than less important.
There is little empirical support that stabilization is possible and has desirable,
positive eects. Indeed it is somewhat surprising that regulators allow price manip-
ulations. It is sometimes argued that investment banks will not always stabilize to
avoid a moral hazard problem with investors who believe being fully protected against
over-pricing. It is likely that this reduces the eect of potential aftermarket prots as
described in this paper. The result itself, however, obviously still holds there are
still hardly any costs involved. In fact, from the regulators perspective price distortions
can easily be ruled out if the bank is prohibited from lling the short position at prices
AFTERMARKET SHORT COVERING 41
below 1 times the oer price. As long as the banks can keep the existence of a
short position secret from investors, a moral hazard problem would not occur.
In our setting, the security may turn out to be overpriced. Investors, however, have
already taken this into account. Investment banks always set the highest feasible price
and thus acts in the issuers interest. It is important to notice that in our setting the
investment bank does not temper prices to rob issuers. The informational asymmetry
in the paper arises at a point in time when all ocial, mandatory information has been
released and any other public statement by investment bank or issuer will be perceived
as cheap talk. Only actions, that is prices, can carry a meaningful message.
The oering procedure was modelled as a signaling game. The investment bank
moves rst and strategically chooses the oer price to maximize its prots from both
the gross spread of the oer revenue and prots from short covering in the aftermarket.
We establish a benchmark by analyzing the situation without aftermarket activities,
and identify the conditions under which the equilibrium is both unique and separating.
A separating equilibrium is referred to as informationally ecient since the investment
banks information is fully revealed by the oer price. We further show that, on av-
erage, securities are underpriced in the separation equilibrium. With the introduction
of aftermarket short covering payo functions and, consequently, the strategic environ-
ment change. As a result, either the oer price falls on average, or a pooling equilibrium
results. In the rst case, an investment bank with favorable information distorts the
price downwards and thereby, on average, exacerbates underpricing. In the second case
investors are unable to infer the investment banks signal from the oer price. This
equilibrium is informationally inecient since investors decisions are based on private
signals only and not also on the signal of the investment bank.
The intuition behind the results can be best explained by relating this paper to
job-market signaling with two types of workers. In the so-called Riley-outcome, the
low type chooses education level zero, and the high type chooses his education just high
enough so that it does not pay for the low type to deviate to his level of education. In
our paper this corresponds to a low-signal bank choosing a low, risk-free price. At this
price all investors want to buy the security and consequently the oering will never
AFTERMARKET SHORT COVERING 42
fail. Nevertheless, in the aftermarket any oering can turn out to be overpriced. The
high-signal bank chooses a high, risky price just low enough so that the risky price does
not pay for the low-signal bank. A price is risky when it is so high that only high-signal
investors buy; in this case the oering will fail if there are not enough investors with
the favorable signal. When introducing prots from short covering, the eect is that of
a personal extra benet from education. Suppose this perk is higher for the low type
of worker than for the high type worker. As a result, the high type has to choose a
higher level of education to maintain separation. In our model, the low-signal bank
considers a price drop in the aftermarket more likely, thus the potential prots from
short covering are higher than for the high signal bank. And so the high-signal bank
has to distort prices downwards in order to maintain separation. At rst sight this is a
surprising result, as casual intuition suggests that potential aftermarket prots should
result in more over-pricing. There may also come a point where it does not pay for
the high signal bank to maintain separation, and so it settles for pooling. The result
is informational ineciency.
The investment bank enjoys higher payos with short covering for the vast majority
of parameter constellations. Looking at per-share prots, the issuer never gains but
often loses if separation prevails; but if there is a switch to a pooling equilibrium he is
always better o. Investors payos are directly opposed to the issuers gains or losses.
An increase in the investment banks share of the revenue or an increase in the amount
of overalloted securities reduces the parameter-set with informational eciency.
Our analysis is in accordance with recent empirical analyzes but contrasts the ex-
isting theoretical literature which argues that stabilizing activities in the aftermarket
serve eciency. We therefore challenge nancial market authorities view that cur-
rent regulations simultaneously serve the interests of issuers, investors, and investment
banks.
AFTERMARKET SHORT COVERING 43
1.6 Appendix
1.6.1 Aftermarket Price Formation
The nally prevailing market price depends on the number of positive signals about
the value of the security. In determining the price we have to distinguish between cases
B
1
and B
0,1
.
Consider rst case B
1
. Since only high-signal investors buy, aggregated demand d
indicates the number of high-signal investors. Suppose d S, i.e. the IPO is successful.
Investors are assumed to take the aggregated information about signals into account
and update their expectations accordingly. At this updated expectation all investors
irrespective of their private signals are indierent between selling and holding or buying
and abstaining, depending on whether they own a security or not, respectively. The
updated expectation thus becomes the aftermarket price, denoted by p
m
(d). We will
later show that case B
1
will occur at the high price of a separating equilibrium only,
i.e. investors know that the banks signal is s
b
= 1. Taking further into account that
the true value of the security is either 0 or 1, we can write p
m
(d[ = 1) = Pr(V =
1[d, = 1). Using Bayes rule, we can express the aftermarket price as
p
m
(d[ = 1) =
Pr(d[V = 1)Pr(s
b
= 1[V = 1)
Pr(d[V = 1)Pr(s
b
= 1[V = 1) + Pr(d[V = 0)Pr(s
b
= 1[V = 0)
. (1.14)
Due to the binomial structure of the prior distributions over signals, the conditional
distribution for demand realization d is, for V = 1,
f(d[V = 1) := Pr(d[V = 1) =
_
N
d
_
q
d
i
(1 q
i
)
Nd
, (1.15)
and for V = 0 analogously. The price-information about s
b
is unambiguous in a
separating equilibrium. We can therefore replace it with the conditional probability of
the banks signal being correct, which is q
b
or 1 q
b
. Bayes rule yields
p
m
(d[ = 1) =
q
b
q
2dN
i
q
b
q
2dN
i
+ (1 q
b
)(1 q
i
)
2dN
. (1.16)
AFTERMARKET SHORT COVERING 44
Consider now case B
0,1
in which all investors order the security, i.e. stated demand
is N and securities are allocated at random. The demand is uninformative since it does
not reveal the number of high-signal investors. Suppose that we are at the low price of
a separating equilibrium. Note that high-signal investors expect the security to be of
higher value than low-signal investors. Hence, there exists a price larger than the oer
price, p > p

at which high-signal investors who were not allocated a security would be

willing to buy the security, and low-signal investors would be willing to sell, in case they
were allocated a security. Without modelling the price-nding procedure explicitly we
assume that the following intermediate process takes place. Those high-signal investors
who did not receive the security in the oering submit a unit market-buy-order. Those
low-signal investors who obtained the security in the oering submit a unit market-
sell-order. All other investors abstain. The number of investors who want to buy or
to sell is denoted by

d and

S, respectively. Aggregate demand of high-signal investors
is then d =

d + S

S and the market price p
m
can be determined as before. The
same procedure can be applied to determine the rst period market clearing price in
the case of a pooling equilibrium. The conditional expectation which determines the
price, however, will then not contain the component about the signal of the investment
bank.
1.6.2 Threshold Prices
Denote by p
s
i
,
the maximum price at which an investor with signal s
i
and price infor-
mation buys, given all investors with s
i
s
i
buy. At this price the investors expected
return from buying the security is zero, normalizing outside investment opportunities
accordingly.
Dene (1[1, 1) := Pr(V = 1[s
i
= 1, = 1) and (0[1, 1) := Pr(V = 0[s
i
= 1, =
1). Consider now the structure of the conditional distribution f(d 1[V ). For V = 1,
this is a binomial distribution over 0, . . . , N 1 with center (N 1)q
i
, and likewise
for V = 0 with center (N 1)(1 q
i
). Since by Assumption 1.2, N is large enough
for every q
i
, f(d 1[1) = 0 for d < N/2 and f(d[0) = 0 for d > N/2. When combining
both f(d 1[1) and f(d 1[0), we obtain a bi-modal function. In g([s
i
, ), investors
AFTERMARKET SHORT COVERING 45
posterior distribution over demands, these are weighted with (1[s
i
, ) and (0[s
i
, ).
Assumption 1.2 now satises two purposes. The rst is to ensure that we pick N
large enough, so that the two modes do not overlap. The second can be seen from the
following lemma.
Lemma 1.3 For any q
i
>
1
2
, there exists a number of investors N(q
i
), such that
p
m
(d) g(d 1[s
i
, ) 0, g(d 1[s
i
, ) almost everywhere.
The lemma states that market prices are mostly 0 or 1, if they are not, then the weight
of this demand is negligible. To see this consider the following heuristic argument.
Proof: p
m
(d) is a s-shaped function in d, given by equation (1.16). For large N,
p
m
(d) 0, 1 almost everywhere. Dene I

as the interval of d around N/2 s.t. for

d I

we have p
m
(d) , 0, 1. p
m
(d) is multiplied with density g(d 1[s
i
, ), which
peaks at (N 1)(1 q
i
) and (N 1)q
i
. For N increasing I

/N 0 and the bi-modal

distribution becomes more centered around (N 1)(1 q
i
) and (N 1)q
i
. Hence, for
every q
i
there is an (N 1)(q
i
) such that for d I

, g(d[s
i
, ) p
m
(d) = 0, i.e. the
weight on p
m
(d) , 0, 1 can be made arbitrarily small.
Using Lemma 1.3 we can determine the threshold prices as follows. Consider rst p
1,1
.
0 = (1 p
1,1
)
N1

d=N/2
S
d + 1
g(d 1[1, 1) p
1,1
N/2

d=S1
S
d + 1
g(d 1[1, 1)
p
1,1
=

N1
d=N/2
S
d+1
g(d 1[1, 1)

N1
d=S1
S
d+1
g(d 1[1, 1)
. (1.17)
For d > N/2, g(d 1[s
i
, ) = (1[s
i
, )f(d 1[1) and for d < N/2, g(d 1[s
i
, ) =
(0[s
i
, )f(d 1[0). Also dene

0
:=
N/2

d=S1
f(d 1[0)
d + 1
and likewise
1
:=
N1

d=N/2
f(d 1[1)
d + 1
, and :=
0
/
1
.
Also write () := (0[1, )/(1[1, ). Thus for the combination of signal s
i
and
AFTERMARKET SHORT COVERING 46
price-information with B
1
we can write
p
1,1
= (1 + (1))
1
and likewise p
1,
1
2
= (1 + (
1
2
))
1
. (1.18)
Consider now the case for p
0,0
. At this price all agents receive the security with equal
probability and we sum from 0 to N 1. Thus
0 = (1 p
0,0
)
N1

d=N/2
S
N
g(d 1[0, 0)
p
0,0
N/2

d=0
S
N
g(d 1[0, 0) p
0,0
= (1[0, 0). (1.19)
Likewise we have
p
0,
1
2
= (1[0,
1
2
). (1.20)
1.6.3 Approximate Closed Form Solutions
We will now derive approximate closed form solutions so that we can solve our model
analytically. In this appendix we let d denotes the number of other investors with
favourable information this contrasts the exposition of the main text, but it simplies
the notation here. First consider the strategy of agent number N. There are N 1
other investors. Given that he invests and the true value is, say, V = 1, then by the law
of large numbers, demand/the number of favorable signals will always be larger than
N/2. Furthermore, the market price is almost surely p
m
(d) = 1. If d others order, then
when buying he gets the asset with probability 1/(d + 1). Thus his payo for price p
(1 p)
N1

d=(1q
i
)N1
1
d + 1
_
N 1
d
_
q
i
d
(1 q
i
)
N1d
=
(1 p)
N1

d=N/2
1
d + 1
_
N 1
d
_
q
i
d
(1 q
i
)
N1d
. (1.21)
AFTERMARKET SHORT COVERING 47
To compute the sum we proceed in a similar manner as one would to compute the
expected value of a binomial distribution: First observe that because N is large,
N1

d=N/2
1
d + 1
_
N 1
d
_
q
i
d
(1 q
i
)
N1d
=
N1

d=0
1
d + 1
_
N 1
d
_
q
i
d
(1 q
i
)
N1d
(1.22)
Then we can compute
N1

d=0
1
d + 1
_
N 1
d
_
q
i
d
(1 q
i
)
N1d
=
1
q
i
N
N1

d=0
N!
(N d)!(d + 1)!
q
i
d+1
(1 q
i
)
N1d
=
1
q
i
N
_
N

l=0
_
N
l
_
q
i
l
(1 q
i
)
Nl

_
N
0
_
q
i
0
(1 q
i
)
N0
_
=
1
q
i
N
(1 (1 q
i
)
N
). (1.23)
In the second step we made a change of variable, l = d + 1, but through this change,
we had to subtract the element of the sum for l = 0. Consequently, for large N, we
can say that
N1

d=N/2
1
d + 1
_
N 1
d
_
q
i
d
(1 q
i
)
N1d

1
q
i
N
. (1.24)
Using the same arguments, we could also show that
N1

d=0
1
d + 1
_
N 1
d
_
q
i
N1d
(1 q
i
)
d

1
(1 q
i
)N
. (1.25)
Use now familiar notation to denote the combination of private and public beliefs
s,
.
Recall that we can write p
1,1
as
p
1,1
=
1
1 + (1)

0

1
. (1.26)
What we now need to nd is a closed form for

0
=
N/2

d=N(1q
i
)1
1
d + 1
_
N 1
d
_
q
i
N1d
(1 q
i
)
d
. (1.27)
AFTERMARKET SHORT COVERING 48
For increasing N one can see that
1
d+1
_
N1
d
_
q
i
N1d
(1q
i
)
d
gets numerically symmetric
around (1 q
i
)N 1. Thus we can express

0
=
1
2
N/2

d=0
1
d + 1
_
N 1
d
_
q
i
N1d
(1 q
i
)
d
=
1
2
N

d=0
1
d + 1
_
N 1
d
_
q
i
N1d
(1 q
i
)
d

1
2
1
(1 q
i
)N
. (1.28)
Assembling, we obtain
p
1,1
=
1
1 + (1)

0

1
1 +
(1q
i
)(1q
b
)
q
i
q
b
q
i
N)
2(1q
)
N)
=
2q
b
1 + q
b

q
b

1
. (1.29)
Equivalently, we get
p
1,
1
2

1
1 +
1q
i
q
i
q
i
N
2(1q
i
)N
=
2
3
, and p
0,1

1 q
b

0
. (1.30)
The information content of a high pooling price is 1/2, and knowing this information,
the probability of the oering being successful is 3/4. Thus the interpretation of risky
prices is thus the ratio of the expected liquidation value given price-information to the
share of successful oerings given this information
p
1,
=
E[ V [ ]
Pr(IPO successful [ )
. (1.31)
1.6.4 Maximal Reputation Costs
If an IPO fails, the worst that can happen is that the investment bank loses all future
IPO business, i.e. it is out of the market. Assuming that future business takes place
in the same environment (e.g. the quality of signals remains constant), the bank can
maximally lose all discounted future prots. Assume that the bank discounts future
prots at rate . Consider the case of highest potential costs

C that can occur from
a failing IPO in a separating equilibrium. An upper bound for costs is given by the
AFTERMARKET SHORT COVERING 49
discounted lost future prots if p = p
1,1
. Then ex-ante prots of a single IPO are
(p
0,0
, p
1,1
, C) =
1
2
(S + O) (p
0,0
+
1 + q
b
2
p
1,1
)
1 q
b
4
C. (1.32)
Assuming that an investment bank would conduct one IPO each period and accounting
for the fact that in a separating equilibrium the ex-ante probability of the IPO to be
successful is (3 + q
b
)/4 we get
C
max
=

t=0
(1 )
t
((3 + q
b
)/4)
t
(p
0,0
, p
1,1
, C
max
). (1.33)
Thus maximal possible costs can be solved to be
C
max
= 2(S + O)
p
0,0
+
1+q
b
2
p
1,1
(3 + q
b
) + 2(1 q
b
)
. (1.34)
Comparing values of C
max
to those of

C shows that for q
i
and q
b
suciently large

C C
max
. Furthermore, for reasonable values of the discount rate, the reverse relation
holds true only for values of q
i
and q
b
where we get

C

<

C. That is, either

C

<

C
and informational ineciencies result, or

C is so large that it lies outside the relevant
parameter region in the context of this model.
1.6.5 Omitted Proofs
Proof of Lemma 1.1
Suppose p

> p
0,0
. At this price only high-signal investors buy. A high-signal bank will
always set a price where at least high-signal investors buy. Hence, high-signal investors
buy at both prices p

and p

. A low-signal bank can now increase its payo by setting

a higher price as
0
is not aected by this, a contradiction.
Proof of Proposition 1.1
First we will argue that the only separating equilibrium surviving the IC is the one
outlined in the proposition. Then we will argue that pooling cannot occur.
AFTERMARKET SHORT COVERING 50
Step 1 (Separating) First observe that there cannot be a separating price p

where
investors choose B
0,1
because otherwise the low-signal bank would deviate to
this price. Note that no separating price with p

>
0
(p
0,0
) can exist because
at this price, the low-signal bank would prefer to deviate. No price p

> p
1,1
can exist since not even high-signal investors would buy. Furthermore, p

1
(p
0,0
) must be satised since otherwise the high-signal bank would prefer to
deviate to p
0,0
. Finally no price p

below p
1,0
is reasonable because the high-
signal bank would then deviate to this price. Take p, with max
1
(p
0,0
), p
1,0

p minp
1,1
,
0
(p
0,0
). Note that such a p always exists as long as
1
(p
0,0
) p
1,1
and p
1,0

0
(p
0,0
). The conditions stated in Proposition 1.1 ensure this is the
case because
1
(p
0,
1
2
) >
1
(p
0,0
) and p
1,
1
2
> p
1,0
.
We analyze the candidate separating equilibrium
(p

= p
0,0
, = 0, B
0,1
); ( p

= p, = 1, B
1
);
(p

, p

, p

, = 0, B
0,1
if p p
0,0
, B
1
if p
0,0
< p p
1,0
, B

else).
By denition of
0
(p
0,0
) it holds that
p
0,0
S =
0

0
(p
0,0
)S (1
0
)C >
0
pS (1
0
)C
so that the low-signal bank would not deviate to p. Since max
1
(p
0,0
), p
1,0
p,
the high-signal bank would also not deviate. Hence this is a PBE.
Now consider the application of the IC. Suppose a high separation price p =

p
with p <

p minp
1,1
,
0
(p
0,0
) is observed. This price is equilibrium dominated
for a bank with s
b
= 0 by denition of
0
(p
0,0
). The low-signal bank can therefore
be excluded the set of potential deviators. The only remaining agent is the high-
signal bank. The best response of high-signal investors then is to buy at p =

p,
i.e. B
1
. Hence the PBE with p

= p does not survive the IC. Applying this

reasoning repeatedly, all separating prices with p < minp
1,1
,
0
(p
0,0
) can be
eliminated.
AFTERMARKET SHORT COVERING 51
Step 2a (Pooling with B
0,1
) For all investors to buy we must have p p
0,
1
2
. Suppose
there was deviation to p =
1
(p
0,
1
2
) <
0
(p
0,
1
2
). For the low-signal bank this
would not be protable by denition of
0
(p
0,
1
2
). But for some beliefs about the
signal of the bank and corresponding best responses, high-signal investors could
be better o. The best response for investors with beliefs on the remaining set
of types, i.e. = 1, however, is B
1
as we have
1
(p
0,
1
2
) < p
1,1
. Hence, applying
IC there cannot be a pooling equilibrium with B
0,1
.
Step 2b (Pooling with B
1
) We must have p p
1,
1
2
. Since
0
(p
0,0
) > p
1,
1
2
, the low-
signal bank would prefer to deviate to p
0,0
, hence this cannot be an equilibrium.
To summarize, restrictions
1
(p
0,
1
2
) < p
1,1
and
0
(p
0,0
) > p
1,
1
2
ensure that the only
equilibrium surviving the IC is the one depicted in Proposition 1.1.
Proof of Proposition 1.2
Consider the highest possible separating oer prices. The market price will by the Law
of Large Numbers resemble the true value of the security. Assumptions 1.1 and 1.2
imply that the IPO fails with probability 0.5 if the true value is V = 0 and the high
separation price is set. If the true value is V = 1 the IPO never fails. Thus, ex-ante
there is underpricing if
1
2
(1 p
0,0

1
p
1,1
) > 0. Substituting in closed form solutions
for threshold prices p
1,1
and p
1,
1
2
from Appendix 1.6.3 this can be written as
(1 q
b
)(1 q
i
)
q
b
q
i
+ (1 q
b
)(1 q
i
)
+ q
b
1 (1.35)
Recall that
1
=
1+q
b
2
. Numerically, it is straightforward to check that the inequality
holds for all q
b
, q
i
(.5, 1).
Proof of Lemma 1.2
We will analyze two cases. Firstly we will show that at C =

C, p

= p
1,1
=
0
(p
0,0
)
can no longer be a sustained as a separating equilibrium if short covering is possible.
AFTERMARKET SHORT COVERING 52
Secondly we will show that at C = C, p

= p
1,
1
2
=
0
(p
0,0
) cannot be sustained as the
separating equilibrium.
We will regard situations in which with respect to the oering price the low-signal
bank is indierent between charging p
0,0
with all investors buying, B
0,1
, and p

where
only high-signal investors buy, B
1
. If the payos from short covering are higher in
the case of deviating to price p

, then this price can no longer be sustained as a sep-

arating price and then, naturally,

0
(p
0,0
) <
0
(p
0,0
). To get this we need to show

2
( p

[B
1
, s
b
= 0) >
2
(p

[B
0,1
, s
b
= 0). Dening (p) s.t. d (p) the aftermar-
ket price is not above (1 )p this is equivalent to
( p

d=S + O
O (1 ) p

p
m
(d) Pr(d[s
b
= 0) >
(p

d=0
O (1 )p

p
m
(d) Pr(d[s
b
= 0)

(1 ) p

( p

d=S + O
Pr(d[s
b
= 0)

( p

d=S + O
p
m
(d) Pr(d[s
b
= 0)
_

_
>
_

_
(1 )p

(p

d=0
Pr(d[s
b
= 0)

(p

d=0
p
m
(d) Pr(d[s
b
= 0)

(1 ) p

q
b
2
> (1 )p
0,0
q
b
. (1.36)
The last step follows from Lemma 1.3 in Appendix 1.6.2. We can now check what
happens at the threshold points. Suppose that C = C so that p

= p
1,
1
2
. Then (1.36)
translates to p
1,
1
2
/2 > p
0,0
which is ensured by Assumption 1.4. Recall that numerically
this assumption requires that not both q
i
and q
b
are small. Suppose that C =

C so that
p

= p
1,1
. Then we need that p
1,1
/2 > p
0,0
. Informativeness of s
b
implies p
1,1
> p
1,
1
2
.
Proof of Proposition 1.3
The second step of the proof of Lemma 1.2 ensures that C

C. The model is set-

up so that all payos
1
+
2
can be dealt with as one. Hence the aforementioned
procedure can be applied here as well. The proof of the pooling outcome goes exactly
along the lines of the proof of Proposition 1.1. Take a separating equilibrium in which
both agents make less prot than in the pooling equilibrium. Pareto Ecieny rules
this equilibrium out. The existence of

C

>

C is again ensured by Lemma 1.2. By
denition, for C >

C

, the highest attainable price is p

1,1
, and it is the only one selected
by the IC.
AFTERMARKET SHORT COVERING 53
Proof of Proposition 1.4
From Proposition 1.3 we know that a pooling equilibrium results for all C < C

. C

is dened as the value of C for which equation (1.11) is fullled with

0
(p
0,0
) = p
1,
1
2
.
Solving for C

one obtains
C

(S + O)
_
2 q
b
2
p
1,
1
2
p
0,0
_
+ (1 )Oq
b
_
p
1,
1
2
2
p
0,0
_
. (1.37)
Partially dierentiating w.r.t. O we obtain
C

O
=
_
2 q
b
2
p
1,
1
2
p
0,0
_
+ (1 )q
b
_
p
1,
1
2
2
p
0,0
_
. (1.38)
Both terms in brackets are positive by Assumption 1.4 as long as q
b
< 1. Partial
dierentiation w.r.t. yields
C

O
= (S + O)
_
2 q
b
2
p
1,
1
2
p
0,0
_
q
b
_
p
1,
1
2
2
p
0,0
_

_
p
1,
1
2
2
_
2 q
b

r
1 + r
q
b
_
p
0,0
_
1
r
1 + r
q
b
__
. (1.39)
Since 2 q
b

r
1+r
q
b
> 1
r
1+r
q
b
whenever q
b
< 1, Assumption 1.4 ensures that the
term is positive.
Chapter 2
Investment Bank Compensation in
Venture and Non-Venture Capital
Backed IPOs

2.1 Introduction
This paper proposes a signaling model of the initial public oering (IPO) process that
can explain two related puzzles. First, why investment banks share of IPO revenue, the
gross spread, is so large that they are left with prots despite market competition and,
second, why these spreads are signicantly lower in venture capital (VC) backed IPOs
than in non-VC backed IPOs. Megginson and Weiss (1991) report for a U.S. sample of
640 IPOs between 1983 and 1987 gross spreads of 7.4% for VC backed IPOs and 8.2%
for non-VC backed IPOs. Data on the protability of IPOs is hard to obtain. However,
Chen and Ritter (2000) argue that there are economies of scale in underwriting IPOs.
They show that spreads do not dier in oerings that raise between $20 million and $80
million. Since investment banks at least break even in small oerings large oerings
must be protable. They report that investment bankers readily admit that the IPO
business is very protable (p. 1105).

The chapter is based on joint work with Andreas Park from the University of Toronto.
INVESTMENT BANK COMPENSATION 55
The cost of going public is subject of a long standing and lasting debate among
nancial economists. The literature focuses mainly on the underpricing of IPOs, an
empirically well-documented phenomenon. Ritter and Welch (2002), as a recent exam-
ple, report an average rst-day return of 18.8% for 6,249 IPOs in the U.S. between 1980
and 2001. However, underpricing is not the only cost of going public. Issuers leave
a fraction of the oer revenue, the gross spread, to investment banks (underwriters)
as compensation for their services. Chen and Ritter (2000) nd that the gross spread
amounts to 7% on average for a U.S. sample of 3,203 IPOs between 1985 and 1998.
Although investment bank compensation thus accounts for a substantial proportion of
the cost of going public, it has hardly attracted attention by theorists so far. In light
of the impact that the Chen and Ritter (2000) paper had, this is even more surprising.
Not only do they nd that gross spreads average 7% but that they are exactly 7% in
most of the oerings. Hansen (2001) reports that these ndings triggered 27 lawsuits
against investment banks for not competing in price and a U.S. Department of Justice
investigation of alleged conspiracy among securities underwriters to x underwriting
fees. Thus, in practice, the size of the gross spread attracts considerable attention.
Even though this paper does not explain the clustering of spreads that called attention
in the rst place, we explain the level of spreads the underlying bone of contention.
Notwithstanding the legal debate on investment bank collusion and the seemingly
obvious empirical evidence, our theoretical formulation allows a very dierent, subtle
explanation for high spreads. We nd that it can be in the best interest of the issuer to
pay seemingly inated spreads. That is, issuers would strategically pay high spreads
even if a competing bank oered its service at a lower spread. Issuers hence do not
bargain for lower spreads and, consequently, investment banks do not compete in them.
In our model, there are two reasons for this. First, it is at the banks discretion to set
the oer price. At high oer prices the IPO can fail because there may not be enough
investors willing to subscribe. Banks then have to bear reputation costs. To induce
them to take the risk and set high oer prices issuers must set suciently high spreads.
Banks earn a rent because given the spread they could set a low, risk-free price and
receive their share in revenue with certainty. Secondly, we assume that investment
INVESTMENT BANK COMPENSATION 56
banks hold private information about markets valuation of the rm on oer. In our
model the size of the spread critically aects the banks decision whether to reveal or
to hide this information which in turn aects the oer price. Thus a lower spread may
result in lower expected revenue for the issuer. However, competition could instead take
place in features of the IPO contract that are left outside the model, as for example
additional co-managers or analyst coverage. This ts the ndings of Chen and Ritter
(2000) as they report that not only the clustering of spreads at 7% has increased
over time indicating lack of competition in spreads but that the number of co-
managers in IPOs and analyst coverage has increased over time as well indicating
some competition in these features of the IPO contract.
Megginson and Weiss (1991) were the rst to show that VC backed issuers pay lower
gross spreads than non-VC backed issuers.
23
If the IPO is non-VC backed the founder of
the rm takes all relevant decisions. In VC backed IPOs the venture capitalist usually
holds all control rights, so we assume that in a VC backed IPO the venture capitalist
decides on the level of the spread. Our second main result addresses dierences between
IPOs with uninformed and privately informed issuers. Assuming that, in contrast to
the founder, the venture capitalist has private information about market sentiment,
our model predicts that in equilibrium spreads are lower in VC than in non-VC backed
IPOs.
We propose a simple stylized model of the oering procedure, cast into a signaling
game. We assume that investment banks and investors have private but noisy infor-
mation about the intrinsic value of the oered security which is either good news or
bad news. In a wider sense, this signal can also be understood as information about
market sentiment. Investment banks strategically choose the oer price to maximize
their expected prots from the oer revenue gross spread. A higher price does not nec-
essarily increase revenue: at high prices the IPO may fail as there may not be enough
investors to buy up the entire oering. Prior to the signaling game the issuer oers the
investment bank a contract that species the gross spread level. This level critically
aects banks pricing decisions. Given the contract variables, the investment bank sets
23
For more recent data see Francis and Hasan (2001)
INVESTMENT BANK COMPENSATION 57
an optimizing price that, rst, either reveals (separation) or camouages (pooling) its
private information and, second, is either low so that all investors subscribe (risk free)
or high so that only investors with good news buy (risky). If the issuer is also pri-
vately informed, the gross spread level can be either separating or pooling, too. Banks,
in turn, account for the spreads information content when deciding on the oer price.
Anticipating the banks pricing decision, the issuer sets the level of the gross spread
strategically so that the bank sets the oer price that gives the issuer highest expected
prot. Investors are aware of this process and subscribe only if their expected prof-
its are non-negative. At the equilibrium spread the investment bank makes positive
prots.
For informed issuers we consider two cases: In the rst, the issuer receives a private
signal that is conditionally independent from the banks signal (later interpreted as
VC backed IPOs). In the second case, the issuers signal is perfectly correlated with
the investment banks signal (strong banking ties). In the rst case, the issuer will
not reveal his private information and set a spread that hides his signal. Nevertheless,
the spreads is set so that the investment bank will separate in prices. In the second
case, an issuer with favorable information sets a spread that prevents its low-signal
counterpart from mimicking. Spreads are thus separating and also indicate the banks
signal. Prices can hence carry no additional information. With uninformed issuers the
spread cannot convey information. We show that spreads are then set so that both
types of investment bank pool in the oer price, causing an informationally inecient
equilibrium. The pooling spreads with independently informed issuers (VC backed) are
lower than the spread set by an uninformed issuer (non-VC backed). Furthermore, if
banks and issuers signals are perfectly correlated, spreads are, on average, the highest.
The remainder of the paper is organized as follows. Section 2.2 presents our model
of the IPO procedure. Section 2.3 derives the equilibrium prices set by the investment
bank. Section 2.4 analyzes the strategic choice of the gross spread by uninformed
and superiorly informed issuers. Section 2.5 presents the main results on levels and
dierence of gross spreads. Section 2.6 concludes.
INVESTMENT BANK COMPENSATION 58
2.2 A Stylized Model of the IPO Procedure

Consider the following stylized model of the IPO process.

The Security. The security on oer can take values V V = 0, 1, both equally
likely. The realization is not known to any player in the game.
The Investors. There are N identical, risk neutral investors. Each investor receives a
costless, private, conditionally i.i.d. signal s
i
V about the value of the security. This
information is noisy but (for technical reasons) suciently informative, i.e. Pr(s
i
=
v[V = v) = q, with q (0.6, 1), where v V.
24
If an investor orders, he may or
may not obtain the security during the oering procedure. Shares are distributed
with uniform probability in case the issue is oversubscribed. If an investor obtains
the security his payo is the market price minus the oer price. If he does not obtain
the security or if the oer is not oated his payo is zero. An investors type is his
signal. We refer to the investor as a high-signal investor if s
i
= 1. For s
i
= 0, it is a
low-signal investor.
The Issuer. In general the issuer can be either informed or uninformed. For the
latter we consider two subcases: in the rst, the issuer (rm) receives a private signal
s
f
0, 1, in the second, the issuer and the investment bank (see below) receive
the identical signal. Any signal is costless and conditionally independent from the
investors signals but, for simplicity, of the same quality, i.e. Pr(s
i
= v[V = v) = q. The
uninformed issuer receives no signal. We will refer to these types of issuers as privately
informed, identically informed, and uninformed. In Section 2.5 we interpret the
meaning of informative signals and relate informed and uninformed issuers to real-world
types such as VC backed and non-VC backed issuers. The issuer is risk neutral and signs
a contract with an investment bank that delegates the pricing decision and constitutes

The model presented here is in part similar to the model presented in Chapter 1. However, to
keep this chapter self-contained we accept some redundancies.
24
As turns out in the analysis, for q < .6 we have to distinguish a large number of subcases.
Avoiding these complications, we thus require suciently informative signals.
INVESTMENT BANK COMPENSATION 59
the amount of securities, S, to be sold. It also species the publicly announced gross
spread (0, 1), the share of the oer revenue that remains as remuneration at the
bank.
25
The issuer chooses this spread. If the oer is oated, his prot is fraction
(1 ) of the oer revenue, otherwise it is zero.
Investment Banks. Investment banks are risk neutral. A bank that gets oered
a contract (we will show that it always accepts) receives a costless, private signal
s
b
V about the value of the security. The signal is conditionally independent from
the investors and privately informed issuers signals but, for simplicity, of the same
quality, i.e. Pr(s
b
= v[V = v) = q. Signals characterize a banks type: If s
b
= 1 we
refer to the investment bank as a high-signal bank, for s
b
= 0, it is a low-signal
bank. After receiving the signal the bank chooses the oer price p. If there is excess
demand, securities are allocated at random; if the number of investors willing to buy
is less than the number of shares to be sold, the oer is called o. We assume that
failure of the oering inicts xed costs C on the investment bank. These costs are
external to our formulation and to be thought of as deterioration of reputational capital.
They may also capture the opportunity costs resulting from lost market share when
being associated with an unsuccessful IPO.
26
Without loss of generality, the oering
procedure itself causes no costs for the investment bank. Thus, if the oer is successful,
the banks payo is fraction of the oer revenue; if it fails, a loss of C results.
Signaling Value of the Gross Spread and the Oer Price. The level of the
gross spread and the oer price are announced rst. Then investors decide whether or
not to order, basing their decisions on their private information and on the information
25
We want each agent to have only one choice variable: issuers choose the spread, banks the price
and investors may or may not invest. Another candidate choice variable is the number of shares S,
or even the number of potential investors N that are addressed, e.g. during the road-show. However,
including these as choice variables would require a dierent, more elaborate modelling approach.
26
There is an extensive large literature on investment bank reputation. Dunbar (2000), for e.g.,
provides evidence that established investment banks lose market share when being associated with
withdrawn oerings. Booth and Smith (1986) argue that in the context of asymmetric information
between insiders and outsiders the investment bank as a repeated player in the IPO market certies
that the issue is not overpriced. Following this argument, C can be interpreted as measuring the
deterioration of the certication value of the investment banks brand name.
INVESTMENT BANK COMPENSATION 60
b
>
>
>
>
>
>

Issuer
spread-
separation
spread-
pooling
r

Bank
price-
separation
price-
pooling
r

price-
separation
price-
pooling
r

Investors
r
B
1
r
B
0,1
r
B

r
B
1
r
B
0,1
r
B

r
B
1
r
B
0,1
r
B

r
B
1
r
B
0,1
r
B

Figure 2.1: Extensive Form of the Signaling Game.

that issuer and bank reveal about their signals through the level of the gross spread
and the oer price. We denote information contained in prices by (p), information in
spreads (). In case of the uninformed issuer, the spread is uninformative and only
prices can carry information. In case of an identically informed issuer, the information
contained in is hierarchical to the information in p: Issuers with dierent signals may
set dierent levels of the gross spread which then reveals the signal of the issuer and
the bank; in this case, prices cannot carry further information. We write (p) = 1 if
the price reects that the banks signal is s
b
= 1, (p) = 0 if reveals that s
b
= 0, and
(p) =
1
2
to indicate that the price is uninformative; likewise for (). In equilibrium
these will turn out to be the only relevant cases. Thus , : [0, 1] 0, 1/2, 1. We
refer to (p) as the price-information about the banks signal and to () as the spread
information.
The Aftermarket Price. The equilibrium market price is determined by the ag-
gregate number of investors favorable signals. In our model this number is always
revealed, either directly through investor demand or immediately after the oat through
trading activities. Thus write p
m
(d) for the market price as a function of d 0, . . . , N,
the number of high-signal investors. Appendix 2.7.1 eshes out this argument and pro-
vides an extensive treatment of price formation.
Investors Decisions and Expected Payos. We admit only symmetric pure
strategies; thus all investors with the same signal take identical decisions. These can
then be aggregated so that only three cases need to be considered: First, all investors
INVESTMENT BANK COMPENSATION 61
subscribe, B
0,1
, second, only high-signal investors subscribe, B
1
, and third, no investor
subscribes, B

. Thus, the set of potential collective best replies is B := B

0,1
, B
1
, B

.
To compute his expected payo, an investor has to account for the probability of
receiving the security. There are two cases to consider. In the rst, all investors buy,
i.e. B
0,1
. Thus, market demand is N and all investors receive the security with equal
probability S/N. In the second case, only high-signal investors buy. If d investors
(including oneself) buy, then each one receives the security with probability S/d. If
overall demand d is smaller than the number of shares on oer, d < S, the IPO fails
and investors who ordered the security get it with probability 0.
Investors order the security whenever their expected payo from doing so is non-
negative. Suppose only high-signal investors buy, i.e. B
1
. After observing the oer
price, an investors information set contains both his signal s
i
and the information
inferred from the oer price and spread, (p) and (). Since signals are conditionally
i.i.d., for every V V there is a dierent distribution over the number d 1 of others
favorable signals (s
i
= 1), which we denote as f(d 1[V ). Thus investors posterior
distribution that the number of others favorable signals is d 1 is given by
g(d 1[s
i
, (p), ()) :=

V V
Pr(V [s
i
, (p), ()) f(d 1[V ). (2.1)
If only investors with favorable signals order, then for a high-signal investor, at price
p his rational-expectation payo from buying has to be non-negative,
N1

d=S1
S
d
(p
m
(d) p) g(d 1[s
i
= 1, (p), ()) 0. (2.2)
Likewise for the respective low-signal investors when all investors order, B
0,1
, in which
case the summation runs from 1 to N, and s
i
= 1 is replaced by s
i
= 0.
Threshold Prices. Denote by p
s
i
,,
the highest price that an investor with signal
s
i
, price information (p) and spread information () is willing to pay in equilibrium
if all investors with signal s
i
s
i
order. If the issuer is uninformed, is replaced
INVESTMENT BANK COMPENSATION 62
with a diamond, ; if issuer and bank get the same signal and if the issuer signals
his private information, (p) is replaced with a diamond to indicate that the price
cannot reveal further information. Suppose, the issuer reveals information . Then
p
1,1,
is the highest (price-separating) price with B
1
, p
1,
1
2
,
the highest (price-pooling)
price with B
1
, p
0,
1
2
,
the highest (price-pooling) price with B
0,1
, and p
0,0,
the highest
(price-separating) price with B
0,1
. Note that at all these prices investors are aware that
the security price may drop (or rise) in the aftermarket and that they may not get the
security. The threshold prices are formally derived in Appendix 2.7.2.
The Investment Banks Expected Payo. First consider case B
1
. Variable d
denotes the number of orders, i.e. the number of high-signal investors. If the true value
is V = 1, we have
Pr(d S[B
1
) =
N

d=S
_
N
d
_
q
d
(1 q)
Nd
, (2.3)
analogously for V = 0. Suppose the issuer gets no signal or its private signal s
f
. A
bank with signal s
b
assigns probability
s
b
,
(S) to the event that at least S investors
have the favorable signal. If the investment bank has signal s
b
and spread information
, then

1,
(S) =
N

d=S
_
N
d
_
_
Pr(V = 1[s
b
, ) q
d
(1 q)
Nd
+ Pr(V = 0[s
b
, ) (1 q)
d
q
Nd
_
. (2.4)
If the bank charges a price at which only high-signal investors buy, its expected prot
is
(p[s
b
, , B
1
) =
s
b
,
(S) pS (1
s
b

(S)) C. (2.5)
Consider now B
0,1
, the case where the oer price is low enough so that all investors
are willing to buy, irrespective of their signals. The oer never fails, thus payos are
given by (p[B
0,1
) = pS. If the price is set so high that no investor buys, as in case
B

, a loss of C results with certainty.

The unconditional distribution over favorable signals is a composite of the two con-
ditional distribution and thus bimodal. To obtain (approximate) closed form solutions
INVESTMENT BANK COMPENSATION 63
E
t t = 1 t = 2 t = 3 t = 4
The issuer oers a
contract specifying the
level of the spread.
The bank sets the
oer price.
Investors decide
whether to order
or to abstain.
Shares are oated
or oering fails.
Figure 2.2: The Timing of the Game.
for success-probabilities and prices, we make two simplifying assumptions: the rst
simplies computations, since the two modes of the distribution over favorable signals
are centered around N(1 q) and Nq. It allows us to get closed form solutions for
success-probabilities. The second assumption ensures that we can analyze the two
underlying conditional distributions separately.
Assumption 2.1 S = (1 q)N.
For every signal quality q, there exists an

N(q) so that for all N >

N(q) the two
conditional distributions over favorable signals generated by V = 0 and V = 1 do not
overlap. By standard results from statistics, a sucient condition for

N(q) is given
by

N(q) > 64q(1 q)/(2q 1)
2
.
Assumption 2.2 The number of investors N is larger than

N(q).
As a consequence of the second assumption we can apply the Law of Large Numbers and
DeMoivre-Laplaces Theorem.
27
Since we assume that the IPO fails whenever d < S,
Assumption 2.1 implies, for instance, if the spread is uninformative, i.e. = 1/2, then

0,
1
2
(S) = (2q)/2 and
1,
1
2
= (1+q)/2. In what follows we thus omit S. A consequence
of the Law of Large Numbers is that p
m
(d) 0, 1 for almost all values of d.
28
Before proceeding we summarize the timing of the model in Figure 2.2. First,
an (identically) informed issuer receives its signal and then all types of issuer oer
a contract to a bank specifying the spread level. Second, the bank sets the oer
27
For instance, the mode of a binomial distribution is generally not exactly symmetric. However,
if N is large enough, we can apply DeMoivre-LaPlace (0 < q 2
_
q(1 q)/N < 1) and employ the
normal distribution instead. Thus we can treat each mode to be symmetric.
28
To be more precise, for d N/2, p
m
(d) = 1, and for d N/2, p
m
(d) = 0. Thus to get
interesting equilibria, it is necessary that S is strictly smaller than N/2. If it was not, an IPO where
only s
i
= 1 investors buy, would never be at risk of being overpriced as it fails in all overpriced cases.
INVESTMENT BANK COMPENSATION 64
price given the spread, the information contained therein, and its own signal. Third,
investors decide whether to order given all available information. Finally, in case the
IPO takes place, the number of favorable signals is revealed in the aftermarket and the
price adjusts according to it.
In what follows proceed backwardly. In the next section we analyze the price setting
of the investment bank given the level of the gross spread. In Section 2.4 we derive
each type of issuers choice of the spread in anticipation of the banks price setting.
Section 2.5 interprets the ndings of Sections 2.3 and 2.4 and presents our main results
on investment banking prots and dierences in spreads between dierent classes of
issuers.
2.3 Investment Banks Equilibrium Price Choice
There are two cases to consider: First, spreads are uninformative or reect the issuers
independent information. In that case, the investment bank plays a signaling game
and needs to decide whether or not to reveal its private information. Second, in case
of the identically informed issuer, spreads can reveal the banks signal. Then the bank
has no strategic decision problem but merely chooses the price that is optimal given
all public information.
2.3.1 Uninformative Spreads or Spreads Reecting the Is-
suers Independent Signal
In the following we identify the conditions under which a prot maximizing investment
bank will reveal its information through the oer price. A separating equilibrium is
dened as informationally ecient since investors can derive the banks signal from the
oer price. In a pooling equilibrium information is shaded and thus it is informationally
inecient. In this case, investors decide only on the basis of their private signals. In
what follows we take the information that may be contained in spreads, (), as given.
Separation and pooling thus always refers to prices.
INVESTMENT BANK COMPENSATION 65
The Equilibrium Concept and Selection Criteria. The equilibrium concept for
this signaling game is, naturally, the Perfect Bayesian Equilibrium (PBE). A common
problem with PBEs, however, is their multiplicity, stemming equilibria being sup-
ported by unreasonable out-of-equilibrium beliefs. The common way to overcome
this problem is to apply an equilibrium selection rule such as the Intuitive Criterion
(IC), introduced by Cho and Kreps (1987). We follow this line of research and consider
only equilibria that do not fail the IC. All of these PBE selection devices favor sepa-
rating over pooling equilibria. It will turn out, however, that in our framework under
certain conditions pooling equilibria cannot be ruled out by the IC. Moreover, these
pooling equilibria then Pareto dominate any separating equilibrium.
29
It would thus be
unreasonable not to assume that these equilibria will be picked. Thus in what follows,
we will only consider equilibria that satisfy the IC and among these, we consider those
that are Pareto ecient for the agent who takes the signaling action. In this section
this agent would be the bank.
A pooling equilibrium in prices is specied through (i) an equilibrium oer price
p

from which investors infer (ii) price-information =

1
2
, and (iii) investors best
replies given their private signals, , and p

. A separating equilibrium in prices is (i)

a system of prices p

, p

and price-information such that (ii) at p

= p

, the high
separation price, the price-information is that the bank has the favorable signal, = 1,
at p

= p

, the low separation price, the price-information is that the bank has the low
signal, = 0, and (iii) investors best replies given their private signals, , and p

or
p

. In both separating and pooling equilibria, for p , p

, p

or p ,= p

, respectively,
out-of-equilibrium public beliefs are chosen appropriately. The following result is a
straightforward consequence of signaling, the proof of which is in Appendix 2.7.4.
Lemma 2.1 (The Highest Possible Low Separating Price)
There exists no PBE (price-)separating oer price p

> p
0,0,
.
In any separating equilibrium, therefore, the low price must be such that all in-
vestors buy, and the highest such separating price, given price-information = 0, is
p

= p
0,0,
. In what follows we refer to p
0,0,
as the low separation price.
29
Here Pareto domination refers to the payos of the respective banks or issuers who take the
decision rst, not to investors who react.
INVESTMENT BANK COMPENSATION 66
Price-signaling equilibria in our setting come in one of three guises: The already
mentioned separating equilibrium, a pooling equilibrium in which only high-signal in-
vestors buy, and a pooling equilibrium in which all investors buy. In the following,
we characterize the conditions guaranteeing that only separating equilibria survive our
selection criterion.
Fix a potential price p [p
0,0,
, p
0,
1
2
,
], the interval of potential pooling prices at
which all investors would buy.
30
Dene
1,
(p) as the price at which the high-signal
bank would be indierent between charging a risky price
1,
(p) at which only high-
signal investors buy, B
1
, and a safe pooling price p with B
0,1
(all investors buy).
Formally,

1,

1,
(p)S (1
1,
) C = pS
1,
(p) =
p

1,
+
1
1,

1,
C
S
. (2.6)
Price
0,
(p) is dened analogously for the low-signal bank. Thus price
s
b
,
(p) is the
lowest risky price that a bank with signal s
b
is willing to deviate to from safe price
p.
31
In what follows we refer to
1,
(p) as the high-signal banks deviation price, and
to
0,
(p) as the low-signal banks deviation price. It is straightforward to see that

0,
(p) >
1,
(p) for all p [p
0,0,
, p
0,
1
2
,
], that is, the low-signal bank requires a higher
price as compensation for risk taking. In addition,
s
b
,
(p)/p > 0, s
b
0, 1, so
the higher the pooling price, the higher the lowest protable deviation price. Taking
spread-information as given, we omit from the equilibrium specication. In what
follows we analyze equilibria depending on two conditions on primitives.
Condition 1 The high-signal banks deviation price from the highest safe pooling
price is not higher than the highest separating price,
1,
(p
0,
1
2
,
) p
1,1,
.
Condition 2 The low-signal banks deviation price from the low separating price is
not smaller than the highest risky pooling price,
0,
(p
0,0,
) p
1,
1
2
,
.
30
The order of prices is immediately obvious from Appendix 2.7.3.
31
Deviation to a high, risky price can lead to increased overpricing, which is commonly perceived
to be bad for a banks reputation. Nanda and Yun (1997) analyze the impact of IPO mispricing on
the market value of investment banks. They nd that overpriced oerings result in decreased lead-
underwriter market value. In our model, however, investors fully take into account that the oer price
may drop in the aftermarket. Modelling such reputation eects would thus be contradictory in our
setting.
INVESTMENT BANK COMPENSATION 67
Proposition 2.1 (Equilibrium Price Setting)
(a) If Condition 1 and Condition 2 are both fullled then the unique PBE that satises
the Intuitive Criterion is the separating equilibrium (p

= p
0,0,
, = 0, B
0,1
); ( p

=
minp
1,1,
,
0,
(p
0,0,
), = 1, B
1
); (p ,= p

, p

, = 0, B
0,1
if p p
0,0,
, B
1
if p
0,0,
<
p p
1,0,
, B

else).
(b) If Condition 1 is not fullled then the only PBE that satises the Intuitive Cri-
terium and Pareto eciency is the pooling equilibrium (p

= p
0,
1
2
,
, =
1
2
, B
0,1
); (p ,=
p
0,
1
2
,
, = 0, B
1
if p p
1,0,
, B

else) in which all investors buy.

(c) If Condition 2 is not fullled then the only PBE that satises the Intuitive Cri-
terium and Pareto Eciency is the pooling equilibrium (p

= p
1,
1
2
,
, =
1
2
, B
1
); (p ,=
p
1,
1
2
,
, = 0, B
1
if p p
1,0,
, B

else) in which only high-signal investors buy.

Interpretation of the Proposition. Condition 1,
1,
(p
0,
1
2
,
) < p
1,1,
, together with
the IC is necessary and sucient to rule out pooling equilibria in which all investors
buy, irrespective of their signals. Condition 2,
0,
(p
0,0,
) > p
1,
1
2
,
, ensures that there is
no pooling where only investors with good news buy, B
1
. The IC itself ensures that
the high-signal bank always charges the highest sustainable separating price. The high
separation price p

is the minimum of p
1,1,
and
0,
(p
0,0,
). The bank cannot charge
more than p
1,1,
and it cannot credibly charge more than
0,
(p
0,0,
) as otherwise the
low-signal bank would deviate. Finally, since
1,
(p
0,0,
) <
1,
(p
0,
1
2
,
) < p
1,1,
, the
high-signal bank is willing to separate. If Condition 1 is violated not even the high-
signal bank wants to take the risk of setting a price where only high-signal investors
buy. A separating price pair with all investors buying at both prices cannot be an
equilibrium. The bank charging the lower price always had an incentive to deviate to
the higher price since the success probability remains unchanged. Pareto eciency for
banks together with the IC then ensures that the highest pooling price at which all
investors buy results as unique equilibrium outcome. If Condition 2 is violated also
the low-signal bank wants to set a high price at which only high-signal investors buy.
A separating price pair with only high-signal investors buying at both prices cannot
be an equilibrium. Again, the bank charging the lower price always had an incentive
INVESTMENT BANK COMPENSATION 68
to deviate. Under eciency only the highest such pooling price survives as the unique
equilibrium. Finally, notice that it cannot be the case that both Condition 1 and
Condition 2 are violated simultaneously.
2.3.2 Spreads Reecting the Issuers Identical Signal
The moment the identically informed issuer separates, the bank has no more control
over the the prices signaling value. We signify this by including a diamond, , instead
of (p) into prices, p
s
i
,,
. The bank continues to choose the price that, given its
private information, maximizes expected prot. However, a low-signal bank can no
longer mimic a high-signal bank because investors have inferred the banks signal form
the spread.
Suppose the bank has signal s
b
= 0 and spread-information is () = 0. Then the
highest price high-signal investors are willing to pay is p
1,,0
. This price is risky as only
investors with signal s
i
= 1 are willing to buy. Price p
0,,0
is the highest safe price at
which all investors buy. However, if the spread is high enough, the risk of a failing IPO
may still be outweighed by potential gains. If the spread
s
0
is large enough so that
risky prots at p
1,,0
strictly exceed riskless prots at p
0,,0
, i.e.

,0
Sp
1,,0
(1
,0
)C > Sp
0,,0
(2.7)
then the low-signal bank will choose risky price p
1,,0
. The high-signal bank faces a
similar choice: If the spread is too low, it would rather choose a safe price. Here,
however, the highest riskless price is p
0,,1
, as at this price investors with the low signal
are willing to buy, given they believe that the banks/issuers signal is s
b
= 1. So the
high-signal bank only choose risky price p
1,,1
if spread
s
1
is high enough so that risky
prots at p
1,,1
strictly exceed riskless prots at p
0,1
, which is

,1
Sp
1,,1
(1
,1
)C > Sp
0,,1
. (2.8)
INVESTMENT BANK COMPENSATION 69
From (2.7) and (2.8) we can derive the respective threshold spreads

s
0
=
1
,0

,0
p
1,,0
p
0,,0
C
S
and
s
1
=
1
,1

,1
p
1,,1
p
0,,1
C
S
. (2.9)
It is straightforward to compute that
s
0

s
1
= (2(1 q))
1
C/S > 0. Consequently, if
spreads are separating and suciently large, then banks will set risky prices.
2.4 The Issuers Strategic Choice of the Spread
As with the investment bank, the analysis is split into two parts. In the rst, the issuer
is uninformed and thus not involved in a strategic situation. He will set prices so as
to make the bank set equilibrium prices that are revenue-maximizing. In the second
part, the issuer does have private information. The issuer then has the power to play a
signaling game. He anticipates the behavior of the investment bank, and thus he will
set spreads strategically to maximize expected revenue.
2.4.1 Equilibrium Spreads if the Issuer is Uninformed
For the investment bank, the choice of equilibrium prices critically depends on Condi-
tions 1 and 2 from Proposition 2.1. In the following we give an intuitive interpretation
of the equilibrium outcome in terms of the gross spread, demonstrating how the spread
aects these conditions. We then derive the uninformed issuers decision about the
spread level. As before we indicate that the issuer is uninformed by replacing with a
diamond.
An Intuitive Characterization of the Equilibrium. The concept of deviation
prices
s
b
,
is a convenient tool to describe restrictions. We will now reformulate
Conditions 1 and 2 from Proposition 2.1 in terms of the gross spread . This allows us
to derive a simple linear descriptive characterization of the equilibrium. Consider rst
INVESTMENT BANK COMPENSATION 70
Condition 1,
1,
(p
0,
1
2
,
) p
1,1,
. If is so low that

1,
(p
0,
1
2
,
) =
p
0,
1
2
,

1,
+
1
1,

1,
C
S
> p
1,1,
(2.10)
then the separating equilibrium cannot be sustained and the pooling equilibrium in
p
0,
1
2
,
prevails. In other words, if the gross spread is rather low then the incentive to set
a high price and take the risk of failure is reduced whereas the cost of failure remains
unchanged. The threshold value for s.t. not even the high-signal bank sets a risky
price is given by

=
1
1,

1,
p
1,1,
p
0,
1
2
,
C
S
=
1
2
1
2q 1
C
N
. (2.11)
Moreover, if is so high that

0,
(p
0,0,
) =
p
0,0,

0,
+
1
0,

0,
C
S
< p
1,
1
2
,
(2.12)
then a separating equilibrium, again, cannot be sustained and the pooling equilibrium
in p
1,
1
2
,
prevails. In this case the gross spread is so high that even the low-signal
bank is willing to take the risk of failure and set a high price at which only high-signal
investors buy. For the high-signal bank it becomes too costly to uphold separation, i.e.
it would have to lower the high separation price so much that it prefers pooling. This
threshold value for a pooling is given by

=
1
0,

0,
p
1,
1
2
,
p
0,0,
C
S
=
1
2
q/(1 q)

0
p
1,
1
2
,
p
0,0,
C
N
. (2.13)
Finally, there exists a

[
s

,
p

] such that the deviation price of the low-signal

bank is just p
1,1,
, i.e. for values of above

the high-signal bank has to lower

the high separation price in order to uphold separation. The following Corollary to
Proposition 2.1 summarizes the above characterization; Figure 2.3 oers an illustration
of the corollary.
INVESTMENT BANK COMPENSATION 71
T
0
Oer Price
E
Gross Spread
s

p
0,
1
2
,
p
0,0,
p
1,1

0
(p
0,0,
)
p
1,
1
2
,
Figure 2.3: Equilibrium Oer Prices at Dierent Levels of Uninformative Spreads.
For levels of the spread below
p

, both types of banks pool in p

= p
0,
1
2
,
. For [
s

]
there is separation: The low-signal bank always sets p

= p
0,0,
, and the high-signal bank sets
p

= p
1,1,
for [
s

] and p

=
0,
(p
0,0,
) for (

,
p

). If
p

there is pooling in
p
1,
1
2
,
.
Corollary 2.1 (Proposition 2.1 in Terms of the Gross Spread)
If spreads are uninformative and [
s

,
p

) then the unique equilibrium is the sep-

arating equilibrium stated in Proposition 2.1. If [
s

] then p

= p
1,1,
, and if
(

,
p

) then p

=
0,
(p
0,0,
). If <
s

then pooling in p
0,
1
2
,
prevails. If
p

there is pooling in p
1,
1
2
,
.
Implicitly, we assumed a tie-breaking rule specifying that at spread thresholds
p

banks set a risky pooling price and at

s

they set separating prices. At

p

both types
of banks set p
1,
1
2
,
even though the low-signal bank is indierent between p
1,
1
2
,
and
p
0,0,
. The latter, however, cannot be an equilibrium: Suppose the low-signal bank sets
p
0,0,
. Then the issuer could raise the spread by an arbitrarily small > 0 making the
low-signal bank strictly prefer p
1,
1
2
,
. However, by lowering the spread to
p

+/2, the
issuer could raise his prots and yet the low-signal bank would still set p
1,
1
2
,
, and so
on. The similar reasoning applies to price setting at
s

.
Strategic Choice of the Gross Spread. If the underlying issuer is uninformed,
his strategic choice of spreads conveys no information. For every spread, however, the
issuer knows the best response of both types of banks. Consequently, the issuer has to
INVESTMENT BANK COMPENSATION 72
choose the level of the gross spread that maximizes his overall expected payo. If he
sets the spread too low, even a bank with favorable information chooses a low, riskless
price. If spreads are high, the issuers get a smaller share of the revenue. Furthermore,
for large spreads the high-signal bank may be unable to set a separating price. Pareto
eciency for the rst mover (the issuer) ensures that out of all s triggering separation
or pooling, the issuer will always choose the smallest one. In particular, to get pooling
in the riskless price p
0,
1
2
,
, the issuer can set spread 0. The issuer then has the choice
between the following expected prots
(1
s

)

1,
p
1,1,
+ p
0,0,
2
S, p
0,
1
2
,
S, and (1
p

)

0,
+
1,
2
p
1,
1
2
,
S (2.14)
in separation, low riskless pooling, and high risky pooling, respectively. To nd the
equilibrium spreads, one has to compare the issuers payos for given equilibrium
spreads. For given parameters q, C, N, the issuer will always choose the spread with
maximal expected payos.
1. Pooling in p
1,
1
2
,
is better than separation if
(1
p

)

0,
+
1,
2
p
1,
1
2
,
S > (1

)

1,
p
1,1,
+ p
0,0,
2
S
C
N
< R
1
(q). (2.15)
2. Pooling in p
1,
1
2
,
is better than pooling in p
0,
1
2
,
if
(1
p

)

0,
+
1,
2
p
1,
1
2
,
> p
0,
1
2
,

C
N
< R
2
(q), (2.16)
where R
1
(q) and R
2
(q) are derived by reformulating the inequalities; they depend on
agents signal quality. We state their precise form at the end of Appendix 2.7.4.
The above transformations make use of the closed form expressions for prices and
success probabilities. Numerically it can easily be checked that R
1
(q) < R
2
(q) for all
q (.6, 1), that is if high risky pooling is better than separation, it is also better than
low, riskless pooling. We can now state the following proposition.
INVESTMENT BANK COMPENSATION 73
Proposition 2.2 (Gross Spreads with Uninformed Issuers)
Assume C/N < R
1
(q). There is a unique equilibrium that satises the IC and e-
ciency: The uninformed issuer oers a contract with =
p

and both types of invest-

ment banks set pooling oer price p
1,
1
2
,
.
Proof: The choice of follows directly from the comparison of the respective expected
prots. The resulting price-setting by banks follows from Proposition 2.1.
Interpretation of the Proposition. The ratio C/N measures the failure costs
that the investment bank incurs per potential investor. If failure costs per investor are
small, the level of the spread that triggers the high pooling price is most protable.
Proposition 2.2 is not a complete equilibrium analysis. However, if we impose the
restriction that maximal spreads cannot exceed 10%, the corresponding ratio C/N will
never exceed R
1
(q). Spreads above 10% are hardly observed,
32
thus for the empirically
relevant parameter it is reasonable to restrict attention to the equilibrium characterized
in the proposition.
2.4.2 Equilibrium Spreads if the Issuer is Independently In-
formed
Suppose now that the issuer gets his own, private signal, s
f
, conditionally independent
from all signals s
i
and s
b
. Then the signaling game has two stages. In the rst, the
issuer may or may not signal his information. Bank and investors incorporate this
information. In the second stage, the bank chooses its equilibrium price, which may
or may not reveal the banks private signal. There are multiple dierent constellations
imaginable:
1. The issuer pools in spreads and the bank separates in prices, pools in a risk-less
price p
0,
1
2
,
1
2
or pools in a risky price p
1,
1
2
,
1
2
.
32
See for e.g. Chen and Ritter (2000).
INVESTMENT BANK COMPENSATION 74
2. The issuer separates in spreads, and
(a) given a low-signal issuer = 0, the bank separates in p
1,1,0
or p
0,0,0
, pools
in p
1,
1
2
,0
, or pools in p
0,
1
2
,0
, and
(b) given a high-signal issuer signal, = 1, the bank separates in p
1,1,1
or p
0,0,1
,
pools in p
1,
1
2
,1
, or pools in p
0,
1
2
,1
.
Equilibrium prices for given spread-information are covered by Proposition 2.1, it re-
mains to analyze the issuers optimal spread-choice.
Analogously to Corollary 2.1 we can determine threshold levels for the gross spread
such that investment banks just set the low pooling price, separating prices, or the
high pooling price. The lowest spread that induces banks to set the low pooling price
is still = 0. The two other threshold levels are denoted by
s

for separation and

p

for risky pooling.

The issuers strategic choice of the spread follows from the comparison of the re-
spective prots. As it turns out, there are no spread-separating equilibria issuers
always pool in the spread. Furthermore, the equilibrium pooling spread induces the
bank to play a separating equilibrium in prices.
Proposition 2.3 (Gross Spreads with Independently Informed Issuers)
Assume C/N < R
1
(q). Then there exists a unique, spread-pooling equilibrium that
satises the IC: Both types of issuers oer a contract with =
s
1
2
and investment
banks separate by setting prices p
1,1,
1
2
and p
0,0,
1
2
. At p
1,1,
1
2
, investors hold price-spread
information = 1 and =
1
2
and only investors with s
i
= 1 buy. At p
0,0,
1
2
, investors
hold price-spread information = 0 and =
1
2
and all investors buy.
Interpretation of the Proposition. To prove the claim we proceed counterfactual:
We describe spreads and price-choices in a spread-separating equilibrium and show that
a spread-separating situation is not incentive compatible for the low-signal issuer. He
would always deviate and mimic the high-signal issuer. The intuition is straightforward:
With separating spreads both low- and high-signal issuer prefer to play a spread that
induces risky pooling prices; clearly a low-signal issuer would prefer the higher price
INVESTMENT BANK COMPENSATION 75
though. Furthermore, the high-signal issuer cannot defend his position by setting
dierent spreads, even when trying to play a spread that induces separation-pricing.
We then show that only spread-pooling can result. There will be two candidates for
spread-pooling: The rst spread induces price-pooling, the second price-separation.
However, only price-separation is IC-proof. Details of the proof are in Appendix 2.7.4.
2.4.3 Equilibrium Spreads if the Issuer is Identically Informed
If the issuer pools in spreads price setting by banks is as in Subsection 2.3.1. If the
spread, however, is informative the bank has no strategic considerations to take care of
in its optimal price choice: Its signal is the same as the issuers who has just revealed
his information. The high-signal bank does not have defend itself against deviation
of the low-signal bank. In Subsection 2.3.2 we have already described banks price
setting.
We derived threshold spreads which induce the banks to choose risky prices,
s

,
0, 1. The following order holds for spreads:
s
0
>
p

>
s
1
>
s
1
2
=
s

. If the
issuer signals, the banks pricing decision carries no informational value; we indicate
this by substituting with a diamond, . Empirically, the gross spread almost never
exceeds 10%, we thus restrict attention to parameter constellations so that
s
0
10%
(which implies C/N < R
1
(q)). We can now show the following proposition.
Proposition 2.4 (Gross Spreads with Identically Informed Issuers)
Assume spreads do not exceed 10%. Then there exists a unique, separating equilibrium
that satises the IC and is rst-mover ecient:
(a) The identically informed low-signal issuer oers a contract with spread
s
0
and
the investment bank sets price p
1,,0
. Investors derive information () = 0, and only
those with signal s
i
= 1 buy.
(b) The identically informed high-signal issuer sets spread
s
1
and the bank sets price
p
1,,1
. Investors derive information () = 1, and only those with signal s
i
= 1 buy.
Interpretation of the Proposition. We have derived the threshold spread in Sub-
section 2.3.2. The proof follows in three steps. First, we derive conditions under which
INVESTMENT BANK COMPENSATION 76
each issuer is satised with the bank choosing a risky price at the proposed spreads

s
0
,
s
1
. The conditions will ensure that expected payos are higher than prots from
setting zero spreads. In this step we will use that spreads must not exceed 10%. Sec-
ond, we show that these spreads are proof to derivations, so that no type of issuer wants
to mimic the other, and no type favors playing out of equilibrium spreads. Third we
argue that with identically informed issuers there can be no pooling equilibrium (under
the given restriction on ). Details are in Appendix 2.7.4.
To summarize, if the issuers have the same signal as the bank, they play a separation
equilibrium in which both low- and high-signal issuer set spreads at which the bank
sets a risky price. Notice that this is the only informationally ecient case where prices
contain all existing information. In the case with uninformed issuers, banks pool in
prices; in the case with independently informed issuers, spreads are pooling.
2.5 Results and Interpretation
We claimed to address two issues: First, why do investment banks make positive prots
in a competitive market and second, why do VC backed IPOs have lower spreads. We
deal with these issues in this section.
Even though it is hard to obtain data on banking prots, Chen and Ritter (2000)
report that investment bankers readily admit that the IPO business is very protable
(p. 1105). Chen and Ritter argue that there are economies of scale in underwriting
IPOs. They show that spreads do not dier in oerings that raise between $20 million
and $80 million. Since banks at least break even in small oerings large oerings must
be protable. Megginson and Weiss (1991) were the rst to report that spreads are
signicantly lower in VC backed IPOs than in non-VC backed IPOs. They show for
a U.S. sample of 640 IPOs between 1983 and 1987 that gross spreads for VC backed
issuers amount to 7.4% whereas they are 8.2% for non-VC backed issuers. Francis
and Hasan (2001) nd smaller but signicant dierences between spreads of VC and
non-VC backed IPOs for their U.S. sample of 843 IPOs between 1990 and 1993 as well.
In the following we show that our model can help explain both these phenomena.
INVESTMENT BANK COMPENSATION 77
In addition, we address implications of the model on the level of spreads when a
commercial bank conducts the IPO of a former client. We nally show that our model
is consistent with underpricing.
2.5.1 Positive Prots for Investment Banks
Equilibrium spreads allow investment banks positive prots. Issuers rst announce the
level of the spread and then banks choose prices at their discretion. They can always
set a low, riskless price at which all investors buy, so that they receive their revenue
share with certainty. Issuers, on the other hand, have a keen interest that banks set
high prices, receiving the bulk of the revenue (almost always more than 90%). At high
prices, however, only high-signal investors buy, making such prices risky. Spreads,
therefore, have to be suciently high so that, banks are compensated for the risk of
failure. This eect alone should leave them with zero expected prots. Moreover,
spreads must be incentive compatible so that banks set high prices and do not deviate
to a risk-free low price. An investment banks expected prot, therefore, is always at
least what they would gain by deviating to a low risk free price. Since we assume that
the oering procedure itself causes no costs for the investment bank, it follows that
investment banks earn positive prots.
Proposition 2.5 (Positive Prots for Investment Banks)
Investment banks enjoy positive prots that will not disappear in the face of competition.
Suppose a competing investment bank oered to conduct an IPO at a lower spread than
specied in the contract the issuer oered initially. The issuer would not accept: even
though he would get a higher fraction of the revenue, lower spreads trigger dierent
equilibrium prices, leading to lower payos. In our model, banks have full discretion
over the oer price. Issuers must, therefore, set incentive compatible spreads. In reality
banks do not have full discretion over prices: many oerings fail because issuer and
bank cannot agree on the oer price.
33
However, the qualitative result does not hinge
33
See Busaba, Benveniste, and Guo (2001).
INVESTMENT BANK COMPENSATION 78
upon the assumption that banks have full discretion. Banks have a good deal of power
when it comes to price setting, and this is all we need for the qualitative result to hold.
Issuers thus have no incentive to bargain for lower spreads. Competition, however,
may take place in features of the IPO contract that we do not model. Chen and Ritter
(2000), for example, report that over time the number of co-managers in IPOs and
thus analyst coverage has increased over time as well. These ndings complement our
results nicely: Chen and Ritter state that, apparently, issuers cannot negotiate the
spread; we nd assert that they do not want to.
2.5.2 VC Issuers set Lower Spreads than Non-VC Issuers
In a strict sense, signals provide information about the assets true liquidation value.
In a wider sense, signals can be seen as information about market sentiment market
prices determine an investors payo, the true liquidation value only aects market
prices through the distribution of signals. In this way it is not unreasonable to assert
that an issuer is uninformed whereas banks and investors are informed. Certainly, some
entrepreneurs have little experience with nancial markets. Venture capitalists, on the
other hand, are nancial institutions and so they should be able to assess market sen-
timent. As the venture capitalist usually holds all relevant control rights, we interpret
the independently informed issuer to be a VC-backed issuer. The uninformed issuer
we interpret to be the single non-VC backed entrepreneur. In this model investment
banks also hold private information. Before setting the oer price they closely interact
with investors, for example during the road show, and thus are informed about the
markets valuation of the rm on oer.
Proposition 2.6 (VC Backed Issuers set Lower Gross Spreads)
Assume spreads do not exceed 10%. Then VC backed issuers set lower levels of the
gross spread than non-VC backed issuers.
Proof: If spreads do not exceed 10%, C/N < R
1
(q), then Proposition 2.2 states that
uninformed issuers set spread level
p

. From Proposition 2.3, the only IC-proof and

ecient equilibrium spread with independently informed issuers is pooling spread
s
1
2
.
Numerically it is straightforward to show that
s
1
2
<
p
1
2
=
p

.
INVESTMENT BANK COMPENSATION 79
A VC backed issuer holds private information before setting the spread level. An
issuer with good news regards it as likely that the bank will also receive good news,
and he wants the bank to transform this information to investors via separating prices.
The high-signal issuer also considers it likely that there are enough high-signal investors
such that the IPO will not fail at the risky separation price. An issuer with bad news
will always mimic the high-signal issuer. Issuers receive almost always more than 90%
of the oer revenue and thus have a strong interest in high prices. The reduction in
oer price from signaling bad news is thus too costly for the low-signal issuer even
if he is forced to set the price-separation inducing spread as well.
2.5.3 Strong Commercial Banking Ties
Before going public many companies have strong, long-lasting ties with commercial
banks, for instance through credit-nancing. Thus if a commercial bank organizes
a long-term clients IPO, it is reasonable to believe that they truly have identical
information. Only recently US regulators allowed commercial banks to oer investment
banking services, including IPO underwriting. Our model predicts that bank spreads
in such IPOs will be, on average, higher than in uninformed (non-VC backed) issuers
or VC-backed IPOs. In particular, if banks and issuers signal are unfavorable, the
issuer is willing to set a high spread so that the bank still chooses the risky price.
Proposition 2.7 (Identically Informed Issuers set Higher Average Spreads)
Assume spreads do not exceed 10%. Then on average, identically informed issuers set
higher levels of the spread than uninformed (non-VC backed) or VC backed issuers.
When restricting the analysis to the empirically relevant parameter space where
spreads do not exceed 10% the uninformed issuer sets the spread such that invest-
ment banks with dierent signals pool in a high, risky oer price. If issuers receive
the same signal as the bank, they separate in spreads. In both cases, however, spreads
are so high that banks set the high risky price irrespective of their signals. If investors
observe the low separation spread they infer that the issuers inside information is bad.
INVESTMENT BANK COMPENSATION 80
But then even the high risky price at which only high-signal investors buy is relatively
low. The issuer thus has to set a relatively high spread to make the bank set the risky
price. The opposite eect occurs when the high separation spread is set. However, the
rst eect dominates so that, on average, spreads are higher with identically informed
issuers than in non-VC or VC backed oerings.
One may then conjecture that the low-signal issuer contemplates abandoning its
commercial bank to look for an independent third-party bank. In equilibrium, it turns
out, however, that this deviation is not protable. Let a deviation be common knowl-
edge. The resulting beliefs will render this deviation unprotable. It is numerically
straightforward to show that the high signal issuer would not be interested in this
move: The best that can happen to him is that he is perceived as a high type issuer.
But then even the highest expected payo hell get from working with an independent
bank is, in expectation, lower than what he gets from his commercial bank. The reason
is that with a third party, there is a risk that the bank gets an unfavorable signal and
charges the low price.
Thus if the high type would not change, any change of banking-partner would be
perceived as coming from a low type bank. It is straightforward to check numerically
that the low-type bank then would not want to deviate either.
2.5.4 Underpricing
Even though this paper is not mainly concerned with explaining underpricing, in equi-
librium the model is consistent with the empirical ndings on rst-day returns. In
the context of this model underpricing is the dierence between oer price and market
price. We can establish the following proposition.
Proposition 2.8 (Underpricing)
(a) If spreads are uninformative but prices are separating then, on average, securities
are underpriced. (b) If spreads are informative and all equilibrium prices risky then,
on average, ordering the security yields zero prots.
The intuition behind the result is simple. Both types of investors only buy if their
expected payo is non-negative. At p
0,0
the low-signal investor just breaks even in
INVESTMENT BANK COMPENSATION 81
expectation but the high-signal investor expects a strictly positive payo. At p
1,1
the
high-signal investor just breaks even and the low-signal investor abstains. Ex-ante
expected payos are positive, hence underpricing. If, however, spreads are separating
and prices risky, then only investors with the favorable signal buy. With them buying,
prices are dened so that they yield zero prot.
2.6 Conclusion
We addressed and answered two puzzles of the IPO literature. First, why do investment
banks earn positive prots in a competitive market as argued, for example, by Chen
and Ritter (2000)? Second, why do banks receive lower gross spreads in VC backed
than in non-VC backed IPOs as argued, for example, by Megginson and Weiss (1991)?
Although investment bank compensation accounts for a substantial proportion of the
cost of going public, it has hardly attracted theorists attention. Our model is, to the
best of our knowledge, the rst to explain the level of the gross spread. We model the
IPO procedure as a two-stage signaling game and nd, rst, that investment banks are
left with prots and, second, that signaling considerations under dierent information
constellations cause spreads to dier between classes of issuers.
In the rst signaling stage, the issuer decides on the spread. Assuming that ven-
ture capitalists decide on spreads in VC backed oerings and that they have private
information about market sentiment, they set dierent spreads than a non-VC backed
issuer without much experience with nancial markets. In the second stage, investment
banks set oer prices given their private information, the spread, and the information
contained in the spread. Issuers anticipate the banks pricing decision and set the
gross spread to maximize their expected revenue. Finally, investors decide whether to
subscribe or refrain. They are aware of the IPO process details and order only if their
expected prot is positive. In equilibrium, VC backed issuers oer lower spreads than
non-VC backed issuers. Furthermore, issuers must oer high enough spreads to ensure
that banks set high prices allowing them substantial prots. Issuers act rationally,
they do not want to cut spreads since lower spreads induce dierent, less favorable
equilibrium oering prices.
INVESTMENT BANK COMPENSATION 82
2.7 Appendix
2.7.1 Aftermarket Price Formation
The nally prevailing market price depends on the number of positive signals about
the value of the security. In determining the price we have to distinguish between cases
B
1
and B
0,1
.
Consider rst case B
1
. Since only high-signal investors buy, aggregated demand d
indicates the number of high-signal investors. Suppose d S, i.e. the IPO is successful.
Investors are assumed to take the aggregated information about signals into account
and update their expectations accordingly. At this updated expectation all investors
irrespective of their private signals are indierent between selling and holding or buying
and abstaining, depending on whether they own a security or not, respectively. The
updated expectation thus becomes the aftermarket price, denoted by p
m
(d). We will
later show that case B
1
will occur at the high price of a separating equilibrium only,
i.e. investors know that the banks signal is s
b
= 1. Taking further into account that
the true value of the security is either 0 or 1, we can write p
m
(d[ = 1) = Pr(V =
1[d, = 1). Using Bayes rule, we can express the aftermarket price as
p
m
(d[ = 1) =
Pr(d[V = 1)Pr(s
b
= 1[V = 1)
Pr(d[V = 1)Pr(s
b
= 1[V = 1) + Pr(d[V = 0)Pr(s
b
= 1[V = 0)
. (2.17)
Due to the binomial structure of the prior distributions over signals, the conditional
distribution for demand realization d is, for V = 1,
f(d[V = 1) := Pr(d[V = 1) =
_
N
d
_
q
d
(1 q)
Nd
, (2.18)
and for V = 0 analogously. The price-information about s
b
is unambiguous in a
separating equilibrium. We can therefore replace it with the conditional probability of
the banks signal being correct, which is q or 1 q. Bayes rule yields
p
m
(d[ = 1) =
qq
2dN
qq
2dN
+ (1 q)(1 q)
2dN
. (2.19)
INVESTMENT BANK COMPENSATION 83
Consider now case B
0,1
in which all investors order the security, i.e. stated demand
is N and securities are allocated at random. The demand is uninformative since it does
not reveal the number of high-signal investors. Suppose that we are at the low price of
a separating equilibrium. Note that high-signal investors expect the security to be of
higher value than low-signal investors. Hence, there exists a price larger than the oer
price, p > p

at which high-signal investors who were not allocated a security would be

willing to buy the security, and low-signal investors would be willing to sell, in case they
were allocated a security. Without modelling the price-nding procedure explicitly we
assume that the following intermediate process takes place. Those high-signal investors
who did not receive the security in the oering submit a unit market-buy-order. Those
low-signal investors who obtained the security in the oering submit a unit market-
sell-order. All other investors abstain. The number of investors who want to buy or
to sell is denoted by

d and

S, respectively. Aggregate demand of high-signal investors
is then d =

d + S

S and the market price p
m
can be determined as before. The
same procedure can be applied to determine the rst period market clearing price in
the case of a pooling equilibrium. The conditional expectation which determines the
price, however, will then not contain the component about the signal of the investment
bank.
2.7.2 Threshold Prices
Denote by p
s
i
,,
the maximum price at which an investor with signal s
i
and price-
information and spread information buys, given all investors with s
i
s
i
buy. At
this price the investors expected return from buying the security is zero, normalizing
outside investment opportunities accordingly.
Dene (1[1, 1, ) := Pr(V = 1[s
i
= 1, = 1, ) and (0[1, 1, ) := Pr(V = 0[s
i
=
1, = 1, ). Consider now the structure of the conditional distribution f(d 1[V ).
For V = 1, this is a binomial distribution over 0, . . . , N 1 with center (N 1)q,
and likewise for V = 0 with center (N 1)(1 q). Since by Assumption 2.2, N is
large enough for every q, f(d 1[1) = 0 for d < N/2 and f(d[0) = 0 for d > N/2.
INVESTMENT BANK COMPENSATION 84
When combining both f(d 1[1) and f(d 1[0), we obtain a bi-modal function. In
g([s
i
, , ), investors posterior distribution over demands, these are weighted with
(1[s
i
, , ) and (0[s
i
, , ). Assumption 2.2 now satises two purposes. The rst is
to ensure that we pick N large enough, so that the two modes do not overlap. The
second can be seen from the following lemma.
Lemma 2.2 For any q >
1
2
, there exists a number of investors N(q), such that p
m
(d)
g(d[s
i
, , ) 0, g(d[s
i
, , ) almost everywhere.
The lemma states that market prices are mostly 0 or 1, if they are not, then the weight
of this demand is negligible. To see this consider the following heuristic argument.
Proof: p
m
(d) is a s-shaped function in d, given by equation (2.19). For large N,
p
m
(d) 0, 1 almost everywhere. Dene I

as the interval of d around N/2 s.t. for

d I

we have p
m
(d) , 0, 1. p
m
(d) is multiplied with density g(d[s
i
, , ), which
peaks at (N 1)(1 q) and (N 1)q. For N increasing I

/N 0 and the bi-modal

distribution becomes more centered around (N 1)(1 q) and (N 1)q. Hence, for
every q there is an (N 1)(q) such that for d I

, g(d[s
i
, , ) p
m
(d) = 0, i.e. the
weight on p
m
(d) , 0, 1 can be made arbitrarily small.
Using Lemma 2.2 we can determine the threshold prices as follows. Consider rst p
1,1,
.
0 = (1 p
1,1,
)
N1

d=N/2
S
d + 1
g(d 1[1, 1, ) p
1,1,
N/2

d=S1
S
d + 1
g(d 1[1, 1, )
p
1,1,
=

N1
d=N/2
S
d+1
g(d 1[1, 1, )

N1
d=S1
S
d+1
g(d 1[1, 1, )
. (2.20)
For d > N/2, g(d 1[s
i
, , ) = (1[s
i
, , )f(d 1[1) and for d < N/2, g(d
1[s
i
, , ) = (0[s
i
, , )f(d 1[0). Dene

0
:=
N/2

d=S1
f(d 1[0)
d + 1
and likewise
1
:=
N1

d=N/2
f(d 1[1)
d + 1
, and :=
0
/
1
.
Also write (, ) := (0[1, , )/(1[1, , ). Thus for the combination of signal s
i
,
INVESTMENT BANK COMPENSATION 85
price-information and spread information with B
1
we can write
p
1,1,
= (1 + (1, ))
1
and likewise p
1,
1
2
,
= (1 + (
1
2
, ))
1
. (2.21)
Consider now the case for p
0,0,
. At this price all agents receive the security with equal
probability and we sum from 0 to N 1. Thus
0 = (1 p
0,0,
)
N1

d=N/2
S
N
g(d 1[0, 0, ) p
0,0,
N/2

d=0
S
N
g(d 1[0, 0, ) p
0,0,
= (1[0, 0, ). (2.22)
Likewise we have
p
0,
1
2
,
= (1[0,
1
2
, ). (2.23)
2.7.3 Approximate Closed Form Solutions
We will now derive approximate closed form solutions so that we can solve our model
analytically. In this appendix we let d denotes the number of other investors with
favorable information this contrasts the exposition of the main text, but it simplies
the notation here. First consider the strategy of agent number N. There are N 1
other investors. Given that he invests and the true value is, say, V = 1, then by the law
of large numbers, demand/the number of favorable signals will always be larger than
N/2. Furthermore, the market price is almost surely p
m
(d) = 1. If d others order, then
when buying he gets the asset with probability 1/(d + 1). Thus his payo for price p
(1 p)
N1

d=(1q)N1
1
d + 1
_
N 1
d
_
q
d
(1 q)
N1d
=
(1 p)
N1

d=N/2
1
d + 1
_
N 1
d
_
q
d
(1 q)
N1d
. (2.24)
To compute the sum we proceed in a similar manner as one would to compute the
expected value of a binomial distribution: First observe that because N is large,
N1

d=N/2
1
d + 1
_
N 1
d
_
q
d
(1 q)
N1d
=
N1

d=0
1
d + 1
_
N 1
d
_
q
d
(1 q)
N1d
(2.25)
INVESTMENT BANK COMPENSATION 86
Then we can compute
N1

d=0
1
d + 1
_
N 1
d
_
q
d
(1 q)
N1d
=
1
qN
N1

d=0
N!
(N d)!(d + 1)!
q
d+1
(1 q)
N1d
=
1
qN
_
N

l=0
_
N
l
_
q
l
(1 q)
Nl

_
N
0
_
q
0
(1 q)
N0
_
=
1
qN
(1 (1 q)
N
). (2.26)
In the second step we made a change of variable, l = d + 1, but through this change,
we had to subtract the element of the sum for l = 0. Consequently, for large N, we
can say that
N1

d=N/2
1
d + 1
_
N 1
d
_
q
d
(1 q)
N1d

1
q
i
N
. (2.27)
Using the same arguments, we could also show that
N1

d=0
1
d + 1
_
N 1
d
_
q
N1d
(1 q)
d

1
(1 q)N
. (2.28)
Use now familiar notation to denote the combination of private and public beliefs

s,
. For the time being, assume the issuer is uninformed so that is replaced with a
diamond. Recall that we can write p
1,1,
as
p
1,1,
=
_
1 + (1, )

0

1
_
1
. (2.29)
What we now need to nd is a closed form for

0
=
N/2

d=N(1q)1
1
d + 1
_
N 1
d
_
q
N1d
(1 q)
d
. (2.30)
For increasing N one can see that
1
d+1
_
N1
d
_
q
N1d
(1 q)
d
gets numerically symmetric
around (1 q)N 1. Thus we can express
INVESTMENT BANK COMPENSATION 87

0
=
1
2
N/2

d=0
1
d + 1
_
N 1
d
_
q
N1d
(1 q)
d
=
1
2
N

d=0
1
d + 1
_
N 1
d
_
q
N1d
(1 q)
d

1
2
1
(1 q)N
. (2.31)
Assembling, we obtain
p
1,1
=
_
1 + (1, )

0

1
_
1

_
1 +
(1 q)
2
q
2
qN
2(1 q)N
_
1
=
2q
1 + q

q

1,
. (2.32)
Equivalently, we get
p
1,
1
2
,

_
1 +
1 q
q
qN
2(1 q)N
_
1
=
2
3
, and p
0,1,

1 q

0,
. (2.33)
The information content of a high pooling price is 1/2, and knowing this informa-
tion, the probability of the oering being successful is 3/4. Thus the interpretation of
risky prices is thus the ratio of the expected liquidation value given price- and spread-
information to the share of successful oerings given this information
p
1,,
=
Pr(V = 1 [ , )
Pr(IPO successful [ , )
. (2.34)
2.7.4 Omitted Proofs
Proof of Lemma 2.1
Suppose p

> p
0,0,
. At this price only high-signal investors buy. A high-signal bank
will always set a price where at least investors with signal s
i
= 1 buy. Hence, investors
with signal s
i
= 1 buy at both prices p

and p

. A low-signal bank can now increase

its payo by setting a higher price as
0,
is not aected by this, a contradiction.
Proof of Proposition 2.1
(a) First we will argue that given Conditions 1 and 2 the only separating equilibrium
surviving the Intuitive Criterion (IC) is the one outlined in Proposition 2.1(a). Then
we will argue that pooling cannot occur.
INVESTMENT BANK COMPENSATION 88
Step 1 (Separating) First observe that there cannot be a separating price p

where
investors choose B
0,1
because otherwise the low-signal bank would deviate to this
price. Note that no separating price with p

>
0,
(p
0,0,
) can exist because at this
price, the low-signal bank would prefer to deviate. No price p

> p
1,1,
can exist
since not even investors with s
i
= 1 would buy. Furthermore, p

1,
(p
0,0,
)
must be satised since otherwise the high-signal bank would prefer to deviate to
p
0,0,
. Finally no price p

below p
1,0,
is reasonable because the high-signal bank
would then deviate to this price. Take p, with max
1,
(p
0,0,
), p
1,0,
p
minp
1,1,
,
0,
(p
0,0,
). Note that such a p always exists as long as
1,
(p
0,0,
)
p
1,1,
and p
1,0,

0,
(p
0,0,
). The conditions stated in Proposition 2.1 ensure
this is the case because
1,
(p
0,
1
2
,
) >
1,
(p
0,0,
) and p
1,
1
2
,
> p
1,0,
.
We analyze the candidate separating equilibrium
(p

= p
0,0,
, = 0, B
0,1
); ( p

= p, = 1, B
1
);
(p

, p

, p

, = 0, B
0,1
if p p
0,0,
, B
1
if p
0,0,
< p p
1,0,
, B

else).
By denition of
0,
(p
0,0,
) it holds that
p
0,0,
S =
0,

0,
(p
0,0,
)S (1
0,
)C >
0,
pS (1
0,
)C (2.35)
so that the low-signal bank would not deviate to p. Since max
1,
(p
0,0,
), p
1,0,

p, the high-signal bank would also not deviate. Hence this is a PBE.
Now consider the application of the IC. Suppose a high separation price p =

p
with p <

p minp
1,1,
,
0,
(p
0,0,
) is observed. This price is equilibrium
dominated for a bank with s
b
= 0 by denition of
0,
(p
0,0,
). The low-signal
bank can therefore be excluded the set of potential deviators. The only remain-
ing agent is the high-signal bank. The best response of investors with signal
s
i
= 1 then is to buy at p =

p, i.e. B
1
. Hence the PBE with p

= p does
not survive the IC. Applying this reasoning repeatedly, all separating prices with
p < minp
1,1,
,
0,
(p
0,0,
) can be eliminated.
INVESTMENT BANK COMPENSATION 89
Step 2 (Pooling with B
0,1
) For all investors to buy we must have p p
0,
1
2
,
. Suppose
there was deviation to p =
1,
(p
0,
1
2
,
) <
0,
(p
0,
1
2
,
). For the low-signal bank this
would not be protable by denition of
0,
(p
0,
1
2
,
). But for some beliefs about
the signal of the bank and corresponding best responses, investors with s
b
= 1
could be better o. The best response for investors with beliefs on the remaining
set of types, i.e. = 1, however, is B
1
as we have
1,
(p
0,
1
2
,
) < p
1,1,
. Hence,
applying the IC, there cannot be a pooling equilibrium with B
0,1
.
Step 3 (Pooling with B
1
) We must have p p
1,
1
2
,
. Since
0,
(p
0,0,
) > p
1,
1
2
,
, the low-
signal bank would prefer to deviate to p
0,0,
, hence this cannot be an equilibrium.
(b) We will rst argue that if Condition 1 is not fullled each separating equilibrium
is Pareto dominated by pooling in the risk-less price. Then we will show that also a
pooling price at which only high-signal investors buy is Pareto dominated. We will
nally argue that among all PBE pooling equilibrium prices at which all investors buy
only the one outlined in Proposition 2.1 is Pareto ecient.
Step 1 (Separating) If Condition 1 is not fullled we have
p
0,
1
2
,
S =
1,

1,
(p
0,
1
2
,
)S (1
1,
)C >
1,
p
1,1,
S (1
1,
)C (2.36)
so the high-signal bank prefers pooling in p
0,
1
2
,
to the highest possible separation
price p
1,1,
. Likewise, since p
0,
1
2
,
> p
0,0,
the risk-free pooling price is Pareto
dominating for the s
b
= 0 bank. Thus separation is always Pareto dominated
and deselected.
Step 2 (Pooling with B
1
) Since the high-signal bank can protably deviate from p
1,1,
it will and can do so from p
1,
1
2
,
< p
1,1,
. Pooling with B
1
can thus be no
equilibrium.
Step 3 (Pooling with B
0,1
) Not even the high-signal bank wants to set a price where
only high-signal investors buy. Candidate prices for an equilibrium are thus only
prices with B
0,1
. Consider p = p < p
0,
1
2
,
. Since both types of banks would
INVESTMENT BANK COMPENSATION 90
prefer p =

p with p <

p < p
0,
1
2
,
Pareto eciency prescribes that investors must
hold =
1
2
and thus all investors will buy at p =

p. Applying this reasoning
repeatedly, all prices with p < p
0,
1
2
,
can be eliminated.
(c) We will rst argue that if Condition 2 is not fullled every separating equilibrium
is Pareto dominated. We will then argue that the only pooling equilibrium in which
only high-signal investors buy is the one outlined in Proposition 2.1. We nally show
that pooling in a price where all investors buy cannot be an equilibrium.
Step 1 (Separating) Since Condition 2 does not hold we have
p
0,0,
S =
0,

0,
(p
0,0,
)S (1
0,
)C <
0,
p
1,1,
S (1
0,
)C (2.37)
so the low-signal bank will mimic the high-signal bank at any price p
0,
(p
0,0,
).
To uphold separation the high-signal bank must lower its price below
0,
(p
0,0,
) <
p
1,
1
2
,
. However, a high separation price below p
1,
1
2
,
cannot be an ecient equi-
librium since both types of banks would prefer pooling price p
1,
1
2
,
. There can
thus be no separating equilibrium.
Step 2 (Pooling with B
1
) From Step 1 we know that both types of banks prefer pooling
in p [
0,
(p
0,0,
), p
1,
1
2
,
] even to the separating equilibrium with the highest
possible p. Consider the candidate pooling price

p with p <

p < p
1,
1
2
,
. Since
both types prefer

p to p eciency prescribes = 0.5 and thus p cannot be an
equilibrium. Applying this reasoning repeatedly, all prices with p < p
1,
1
2
,
can be
eliminated. The only pooling equilibrium surviving is thus the one depicted in
Proposition 2.1.
Step 3 (Pooling with B
0,1
) Suppose that p
0,
1
2
,
was an equilibrium, supported by out-
of-equilibriums belief that any deviation is by a low-signal bank. Then consider
a deviation to
1,
(p
0,
1
2
,
). Naturally,
1,
(p
0,
1
2
,
) <
0,
(p
0,
1
2
,
), and thus, ap-
plying the IC, this deviation can only be triggered by a high-signal bank. It is
straightforward to check that, numerically, a violation of Condition 2 implies that
INVESTMENT BANK COMPENSATION 91
Condition 1 holds, i.e.
1,
(p
0,
1
2
,
) < p
1,1,
. Furthermore,

1,
(p
0,
1
2
,
) =
p
0,
1
2
,

1,
+
1
1,

1,
C
S
, (2.38)
which is increasing in costs C. The largest C so that Condition 2 just holds
C = S(
0,
p
0,
1
2
,
p
0,0,
/(1
0,
). Any C violating Condition is smaller than
C. Numerically, then
1,
(p
0,
1
2
,
) < p
1,
1
2
,
, thus eciency holds.
Proof of Proposition 2.3
To prove this result, we proceed in ve steps: In the rst we derive the issuers optimal
spread choice under the assumption that spreads are separating. The issuer then
chooses the spread that maximizes his payo; the spread will induce the bank to set
either a separating or a pooling price. This step serves as benchmark for comparing
deviation payos. The rst-mover Pareto eciency requirement ensures, that in any
spread-separating equilibrium, the low-signal issuer will always set his preferred spread,
irrespective of the high-signal issuers choice. In the second step, we argue that the low-
signal issuer will always mimic the high-signal issuers optimal choice. In the third step
we show that the high-signal issuer cannot defend separation in spreads by choosing a
dierent level of the spread. This step consists of three sub-steps in which we show that
neither constellation (price-separation inducing or price-pooling inducing spreads) can
be upheld. In the fourth step we show that pooling in spreads is indeed an equilibrium,
but we also show that there can be two equilibria. In the fth step we argue that only
the price-separation inducing spread satises the Intuitive Criterion (IC).
The results can only be obtained numerically: When comparing dierent payos,
the decisive equations are complicated polynomials, that cannot be expressed in an
appealing simple form. Explicit solutions, however, can be obtained from the authors
upon request. Furthermore, throughout the proof we use the restriction that < 10%.
Table 2.1 describes how an issuer computes his expected payos. In this proof we
let
s

denote the spread that yields separation given spread information .

34
34
We emphasize that this is not the same as the spreads dened in Subsection 2.3.2. Nevertheless,
INVESTMENT BANK COMPENSATION 92
Pr(V [s
f
= 1) q 1 q
V = 1 V = 0
Pr(s
b
[V ) Price Pr(IPO successful[V ) Pr(s
b
[V ) Price Pr(IPO successful[V )
s
b
= 1 q p
1,1,1
1 1 q p
1,1,1
1
2
s
b
= 0 1 q p
0,0,1
1 q p
0,0,1
1
Table 2.1: Probabilities Summary. Equilibrium price choice, signal probabilities, and
success-probabilities in price-separating equilibria. Issuers expected prot from charging,
e.g.,
s
1
is (1
s
1
) (q (qp
1,1,1
+(1 q)p
0,0,1
) +(1 q) ((1 q)p
1,1,1
1
2
+qp
0,0,1
)) = (q
2
+(1
q)
2
/2)p
1,1,1
+ 2q(1 q)p
0,0,1
.
Step 1: Suppose rst that the spread is separating and indicates s
f
= 1, so that
= 1. The issuer has the choice between expected prots in separation, low
riskless pooling, and high risky pooling. For given parameters q, C, N, the issuer
will always choose the spread with maximal expected payos. The following two
inequalities always holds when C/N < R
1
(q).
1. Pooling in p
1,
1
2
,1
is always better than separation in p
1,1,1
and p
0,0,1
as
(1
s
1
)
_
(q
2
+
(1 q)
2
2
)p
1,1,1
+ 2q(1 q)p
0,0,1
_
> (1
p
1
)
1
p
1,
1
2
,1
(2.39)
2. Pooling in p
1,
1
2
,1
is better than pooling in p
0,
1
2
,1
if
(1
p
1
)
1,1
p
1,
1
2
,1
> p
0,
1
2
,1
(2.40)
Suppose now that the spread triggers = 0. Again, we have to compare expected
prots. All the inequalities hold if we restrict C/N < R
1
(q).
1. Pooling in p
1,
1
2
,0
is better than separation in p
1,1,0
and p
0,0,0
if
(1
p
0
)
_
(1 q) +
q
2
_
p
1,
1
2
,0
> (1
s
0
)
_
3
2
q(1 q)p
1,1,0
+ (q
2
+ (1 q)
2
)p
0,0,0
_
(2.41)
2. Pooling in p
1,
1
2
,0
is better than pooling in p
0,
1
2
,0
if
(1
p
0
)
0,0
p
1,
1
2
,0
> p
0,
1
2
,0
(2.42)
for the purposes of exposition in the proof this notation is best; since the spreads
s

as dened here
are no equilibria, there should be no confusion. Details of the s used in this proof are placed after
the proof.
INVESTMENT BANK COMPENSATION 93
Thus if spreads are separating, irrespective of the spread-information inducing
risky, high price-pooling is better than both price-separation and low price-
pooling.
Step 2: We now show that the low-signal issuer will always mimic the high-signal
issuer, and that defending separation is too costly. For the low-signal issuer it is
protable to mimic the high-signal issuer in
p
1
if
(1
p
1
)
_
(1 q) +
q
2
_
p
1,
1
2
,1
> (1
p
0
)
_
(1 q) +
q
2
_
p
1,
1
2
,0
. (2.43)
Numerically the deviation prot is always higher, thus spread-separating in
p
1
,
p
0
cannot be an equilibrium.
Step 3: The high-signal issuers defenses against mimicking have to be analyzed for
any of the three candidate equilibrium spreads. Price-pooling inducing
p
1
, price-
separating inducing
s
1
, and risk-less pooling inducing = 0 would be defended
by setting a higher s. However, none of these defenses turn out to be feasible.
(a) Defending Price-Separation. The lowest spread

for which the low-signal
issuer will not mimic the price-separation inducing spread any longer, is given by
(1

)
_
3
2
q(1 q)p
1,1,1
+ (q
2
+ (1 q)
2
)p
0,0,1
_
=
(1
p
0
)
_
(1 q) +
q
2
_
p
1,
1
2
,0
. (2.44)
Solving for

, numerically

exceeds by far 10% (and thus lies outside the relevant
parameter region). It also exceeds
p
1
, which brings us to the next case.
(b) Defending Risky Price-Pooling. If the high-signal issuer sets

>
p
1
the
low-signal issuer will no longer mimic if
(1

)
_
(1 q) +
q
2
_
p
1,
1
2
,1
= (1
p
0
)
_
(1 q) +
q
2
_
p
1,
1
2
,0
. (2.45)
Solving for

, numerically

exceeds by far 10% (and thus lies outside the relevant
parameter region).
INVESTMENT BANK COMPENSATION 94
(c) Defending Riskless Price-Pooling. If the high-signal issuer sets

(0,
s
1
) the
low-signal issuer will no longer mimic if
(1

) p
0
1
2
,1
= (1
p
0
)
_
(1 q) +
q
2
_
p
1,
1
2
,0
. (2.46)
Solving for

, numerically it exceeds by far 10% (and thus lies outside the relevant
parameter region).
Thus, there is no spread-separating equilibrium.
Step 4: Consider now the spread-pooling equilibria. As usual, there are three can-
didate spreads: = 0,
s
1
2
and
p
1
2
. It turns out that prots under = 0 are
dominated by prots under the other two spreads. Moreover,
s
1
2
is preferred by
the high-signal issuer,
p
1
2
from the low-signal issuer.
(a) Low-Signal Issuer. Price-separation is better than riskless price-pooling if
(1
s
1
2
)
_
3
2
q(1 q)p
1,1,
1
2
+ ((1 q)
2
+ q
2
)p
0,0,
1
2
_
> p
0,
1
2
,
1
2
. (2.47)
Numerically, given C/N < R
1
(q), this inequality always holds. Risky price-
pooling is better than risk-less price-pooling if
(1
p
1
2
)
_
(1 q)
1 + q
2
+ q
2 q
2
_
p
1,
1
2
,
1
2
> p
0,
1
2
,
1
2
. (2.48)
Numerically, given C/N < R
1
(q), this inequality also always holds. However, the
high type prefers risky price-pooling to price-separation as
(1
p
1
2
)
_
(1 q)
1 + q
2
+ q
2 q
2
_
p
1,
1
2
,
1
2
>
(1
s
1
2
)
_
3
2
q(1 q)p
1,1,
1
2
+ ((1 q)
2
+ q
2
)p
0,0,
1
2
_
(2.49)
holds numerically, given C/N < R
1
(q).
(b) High-Signal Issuer. Price-separation is better than risk-less price-pooling if
(1
s
1
2
)
_
(q
2
+ (1 q)
2
/2)p
1,1,
1
2
+ 2q(1 q)p
0,0,
1
2
_
> p
0,
1
2
,
1
2
. (2.50)
INVESTMENT BANK COMPENSATION 95
Numerically, given C/N < R
1
(q), this inequality always holds. Risky price-
pooling is better than risk-less price-pooling if
(1
p
1
2
)
_
q
1 + q
2
+ (1 q)
2 q
2
_
p
1,
1
2
,
1
2
> p
0,
1
2
,
1
2
. (2.51)
Numerically, given C/N < R
1
(q), this inequality always holds. However, price-
separation is also almost always preferred to risky price-pooling as
(1
s
1
2
)
_
(q
2
+ (1 q)
2
/2)p
1,1,
1
2
+ 2q(1 q)p
0,0,
1
2
_
>
(1
p
1
2
)
_
q
1 + q
2
+ (1 q)
2 q
2
_
p
1,
1
2
,
1
2
(2.52)
holds numerically, given C/N < R
1
(q).
Step 5: Thus there are two spread-equilibria that can be constructed to be PBEs.
Conveniently, however, spread
p
1
2
fails the Intuitive Criterion. To see this, dene

(q) to be the spread for given q that makes the low type not wanting to deviate
from
p
1
2
, even if he was perceived to be the highest type. For simplicity, assume
that at the deviation payo spreads are set to be price-separating. Then

(q)
solves
(1

(q))
_
3
2
q(1 q)p
1,1,1
+ (q
2
+ (1 q)
2
)p
0,0,1
_
=
(1
p
1
2
)
_
(1 q)
1 + q
2
+ q
2 q
2
_
p
1,
1
2
,
1
2
. (2.53)
Numerically, for q > 0.72,

(q) can be set to
s
1
, for smaller q, it has to be
larger. However, numerically it also holds that for all q,

<

s
1
= C/S(1

0,1
)/(
0,1
p
1,1,1
p
0,0,1
), where

s
1
is the spread so that the low type bank is
indierent between choosing p
1,1,1
and p
0,0,1
. (Recall that for higher spread the
bank lowers the price to
0,1
(p
0,0,1
)). Consequently at every

(q) the bank charges
a separation price. Furthermore, numerically at for all q, the high type prefers to
deviate to

(q) if he is perceived to be the high type, whereas the low type prefers
the current equilibrium. Hence there is a deviation that, in the best of all worlds
INVESTMENT BANK COMPENSATION 96
for beliefs, is only protable for the high type issuer and so the equilibrium
p
1
2
fails the IC.
Consider now the price-separation-inducing spread and construct the same de-
viation

(q) as above. It turns out, however that for every q and any for every

< 10%,
(1

)
_
p
1,1,1
3q(1 q)
2
+ p
0,0,1
(q
2
+ (1 q)
2
)
_
>
(1
s
1
2
)
_
p
1,1,
1
2
3q(1 q)
2
+ p
0,0,
1
2
((1 q)
2
+ q
2
)
_
. (2.54)
Any

satisfying this equation with equality could be taken as a benchmark for
deviation-considerations. However, since theres no feasible

that satises our
restriction and equation (2.54) with equality, the out of equilibrium belief of low
type deviation is IC-proof.
As a consequence of all this, the only IC-proof and issuer-ecient equilibrium is
pooling in spreads
s
1
2
which induce price-separation.
In the proof we used the following threshold values for spreads. They are computed in
the same way as demonstrated in Subsection 2.4.1. Note that
s

is not the same as

the ones used in the next proof.

s
1
=
1
1,1

1,1
p
1,1,1
p
0,
1
2
,1
C
S
,
p
1
=
1
1,0

1,0
p
1,
1
2
,1
p
0,0,1
C
S
,
s
0
=
1
0,1

0,1
p
1,1,0
p
0,
1
2
,0
C
S
,

p
0
=
1
0,0

0,0
p
1,
1
2
,0
p
0,0,0
C
S
,
s
1
2
=
s

,
p
1
2
=
p

.
Proof of Proposition 2.4
If the issuer signals his information, the banks price choice carries no extra value. Thus
in prices, is substituted with a diamond. For the banks probability of a successful
IPO, spreads do not carry information, thus in
j,
, j = 0, 1, spread information
is substituted with a diamond. In Subsection 2.3.2, Equation (2.9) we have already
described the spreads which induce banks to choose risky prices: (1) The high-signal
bank chooses risky p
1,,1
with B
1
if it is oered at least
s
1
. (2) The low-signal bank
INVESTMENT BANK COMPENSATION 97
chooses risky p
1,,0
if it is oered at least
s
0
, where

s
1
=
1
1,

1,
p
1,,1
p
0,,1
C
S
, and
s
0
=
1
0,

0,
p
1,,0
p
0,,0
C
S
. (2.55)
First, we have to show that both types of issuer actually do want the respective bank to
set those risky prices. (1) The high-signal issuer prefers the high-signal bank to set p
1,,1
and not p
0,,1
if its expected revenue is higher at the risky price,
1
(1)p
1,,1
S p
0,,1
S.
(Note that = 0 is sucient for the bank to set the risk-free price.) Solving for
yields that the spread minimal separating has to satisfy

s
1
1
p
0,,1

1
p
1,,1
C/N
(2q 1)
2
2q
=: R
3
(q). (2.56)
Applying the same reasoning to the low-signal issuer, he prefers the low-signal bank to
set risky p
1,,0
and not p
0,,0
if
0
(1 )p
1,,0
S p
0,,0
S. Thus the separating threshold

s
0
has to satisfy
35

s
0
< 1
p
0,,0

0
p
1,,0
C/N
2q(2q 1)
2
(1 q)
2
p
0,,0
2
=: R
4
(q). (2.57)
We restrict the analysis to the empirically relevant parameter space where spreads
do not exceed 10%. Since we know
s
0
>
p
>
s
1
we impose
s
0
< 10%. This translates
into any has to be smaller than (4q 1 5q
2
+ 2q
3
)/5(1 2q + 2q
2
) =: R
5
(q).
Numerically it is easy to check that R
5
< minR
1
, R
2
, R
3
, R
4
, that is, requiring spreads
not to exceed 10% is sucient for all other restrictions to hold.
Second, we have to show that there is no protable deviation for either issuer.
(1) Consider the low-signal issuer. Notice that
s
1
<
s
0
and p
1,,1
> p
1,,0
, i.e. the
high-signal issuers spread is lower and the oer price is higher so the low-signal issuer
had the incentive to deviate if the low-signal bank sets p
1,,1
when being oered
s
1
.
However, at
s
1
the high-signal bank is just indierent between risky p
1,,1
and risk-
free p
0,,1
. Since the low-signal bank holds less favorable prospects about investors
valuations it will not set p
1,,1
and thus investors learn that the issuers/banks signal
35
If any of these restrictions on C/N is satised strictly, the necessary spreads can be set lower.
INVESTMENT BANK COMPENSATION 98
is s = 0. But in this case
s
0
is the best choice for the low-signal issuer. (2) Consider
now the high-signal issuer. Since
s
1
<
s
0
and p
1,,1
> p
1,,0
the high-signal issuer will
never mimic the low-signal issuer.
Third, we have to check if pooling in spreads can be an equilibrium. If there is
pooling in spreads banks set prices as in Section 2.3. Since we assume C/N < R
5
we
also have C/N < R
1
, but then the best to do is pooling in
p
1
2
and banks setting p
1,
1
2
,
1
2
.
However, since
p
1
2
>
s
1
and p
1,
1
2
,
1
2
< p
1,,1
the high-signal issuer will want to deviate,
and he is the only one who can do so protably under the conditions set by the IC, so
this cannot be an IC-proof equilibrium.
Proof of Proposition 2.5
We will show that the bank earns non-negative prots at all three possible spread
levels.
(a) With informed issuers, if s
b
= 0, the spread is
s
0
and the low-signal bank sets
p
1,,0
and incurs the risk of losing C. Instead of taking this risk, the bank may choose
a risk-free price p
0,,0
> 0. Being compensated for the risk means that the low-signal
banks gets more than
s
0
Sp
0,,0
> 0.
(b) Likewise, if s
b
= 1, the issuer sets
s
1
and the bank sets p
1,,1
. Instead, the bank
could set price p
0,,1
> 0 and realize risk-free prots. To make the bank set the risky
price, the issuer has to pick a compensation which gives the bank at least
s
0
Sp
0,,1
> 0.
(c) With uninformed issuers the spread is
p

and the high pooling price p

1,
1
2
,
results.
In this case expected prots are positive as long as we have

0
+
1
2

p
p
1,
1
2
S > (1

0
+
1
2
) C
p
>
2
0

1
(
0
+
1
)p
1,
1
2
C
S
. (2.58)
Numerical simulations show that holds true for all q (.6, 1).
Proof of Proposition 2.7
From Propositions 2.2 and 2.4 we know that an uninformed issuer always sets
p

; an
identically informed issuer with signal s
b
= 0 sets
s
0
, if he has signal s
b
= 1 he sets
s
1
.
INVESTMENT BANK COMPENSATION 99
Ex ante, the identically informed issuer gets either signal with equal probability. Thus
for the claim to be true it must hold that
1
2

s
1
+
1
2

s
0
>
p

1
2
1
1

1
p
1,1
p
0,1
+
1
2
1
0

0
p
1,0
p
0,0
>
1
0

0
p
1,
1
2
p
0,0
. (2.59)
Checking this numerically, the inequality holds if C/N < R
1
(q). By Proposition 2.6,
the VC-backed issuer sets even lower spreads.
Instead of spread deviations suppose that an issuer abandons his commercial bank
and seeks investment banking services from a third party, and let this be common
knowledge. It is numerically straightforward to show that the high-signal issuer would
not be interested in this move: The best to happen is that he would be perceived as a
high-signal issuer. But then the highest expected payo he could get from an indepen-
dent bank is lower than what he gets from his commercial bank. The reason is that
with a third party, there is a risk that the bank gets an unfavorable signal and charges
the low price. Thus if the high-signal bank would not change, any change of banking-
partner would be perceived as coming from a low-signal bank. It is straightforward to
check numerically that the low-signal bank then would not want to deviate either.
List of Restrictions:
R
1
= 2q(2q 1)
2
(q 1)
2
(1 q
2
+ q)/(4q 9q
2
+ 19q
3
25q
4
+ 17q
5
2q
6
1)
R
2
= 2(q 1)(1 q 3q
2
+ 2q
3
)(2q 1)/3q(1 2q + 2q
2
)
R
3
= (2q 1)
2
/2q,
R
4
= 2q(2q 1)
2
(q 1)
2
/(1 2q + 2q
2
)
2
R
5
= (4q 1 5q
2
+ 2q
3
)/5(1 2q + 2q
2
)
Chapter 3
Working for Today or for
Tomorrow: Incentives for
Present-Biased Agents
3.1 Introduction
We examine self-control problems modeled as time-inconsistent, present-biased pref-
erences in a multi-tasking environment. An agent must allocate eort between an
incentivized and immediately rewarded activity and a private activity that pays out
with some delay. Eort costs accrue immediately. As an example think of a situation in
with an agent has to decide how much to work on her job and how much to care about
her health. Eort on the job is assumed to pay out immediately (e.g. wage payment
at the end of the month, piecework rate, job promotion, etc.) while time dedicated to
care for ones health (free weekends, workouts, balanced diet instead of fast-food, etc.)
pays out in the long-run only. Alternatively, one can think of a students decision how
much to study for a degree and how much to work for money while being a student.
The existing literature on present-biased preferences (ODonoghue and Rabin (1999a,
1999b, 2001)) analyzes environments in which an agent must accomplish a single task
only but has discretion when to do it. In contrast, we model situations in which agents
must allocate eort between tasks and has to decide how much of each activity to
accomplish with the complication that some tasks pay out earlier than others.
PRESENT-BIASED AGENTS 101
We consider three types of agents. Time-consistent agents, sophisticated agents,
and naive agents. Throughout the paper we abstract from complications of adverse
selection. The principal is thus assumed to know the type of agent he is contracting
with. To capture the basic eects we start out with the most simple set-up. The
agent has to allocate eort between two tasks only. Eort allocated to the rst task
is incentivized by a principal by way of a linear contract and rewarded immediately.
Eort devoted to the second task serves the agents private benet but pays out with
some delay. Eort is taken to be perfectly observable and contractable. There is no
risk involved. We consider a three-period setting. In a rst period the principal oers
a contract and the agent decides whether to accept or to refrain. In the next period
the agent chooses eort levels. According to the incentive scheme, eort devoted to the
principals purposes pays out immediately, and all eort costs are borne in immediately.
In the nal period the agents private benet is realized.
We show that present-biased agents take decisions that do not maximize their
long-run interest, irrespective of the intensity of incentives. Sophisticated agents are,
however, never harmed by incentives relative to a situation without incentives as they
always receive their reservation utility levels. With naive agents there are two eects.
On the one hand, they wrongly belief to have a high reservation utility because they
think they will not give in to a present bias in a situation without incentives. They
thus only participate if they are paid high enough wages. On the other hand, they
wrongly predict tomorrows eort choices (naive agents think to act like time-consistent
agents but in fact they will always give in to their present bias) which is exploited by
the principal. We show that the second eect always dominates. Naive agents are
thus harmed by incentives relative to the situation without incentives. Furthermore,
we show that social welfare can decrease in the presence of incentives. With naive
agents it may happen that their additional loss due to present-biased eort choices in
the situation with incentives exceeds the principals gain from oering the incentive
contract. The model thus oers a new theoretical possibility of detrimental eects
of incentives, complementary to existing arguments like the crowding out of intrinsic
motivation by extrinsic rewards.
PRESENT-BIASED AGENTS 102
While the current paper is (to the best of our knowledge) the rst analysis of
present-biased preferences in a multi-tasking environment, there is an extensive lit-
erature that analyzes choice problems with time-inconsistent preferences in situations
where agents have discretion when do complete a task. ODonoghue and Rabin (1999a)
derive a present-biased agents decision with exogenously given levels for costs and re-
wards one of those being immediate, the other delayed. ODonoghue and Rabin
(1999b) analyze a principal-agent setting. A procrastinating agent faces stochastic
costs of completing a task. The principal oers an incentive contract to induce the
agent to complete that task in time. If the principal knows the agents cost distribu-
tion he can always achieve the rst-best. With asymmetric information about costs
the rst-best is achieved with time-consistent agents only. With time-inconsistent
agents incentives for timely completion and ecient delay in case of high costs must
be traded-o. The second-best contract involves an increasing punishment for delay.
ODonoghue and Rabin (2001) show that providing a present-biased agent with addi-
tional choice options can harm those agents. Agents may refrain from completing an
activity because they change to a better but never-to-completed alternative. DellaV-
igna and Malmendier (2003) analyze the health club industry and provide evidence
for both, time-inconsistent behavior and naivete. DellaVigna and Malmendier (2004)
go a step further and derive the optimal contract design of rms if consumers have
time-inconsistent preferences. They show that optimal contracts for naive agents have
observed features in some industries like, among others, the health club and credit
card industry suggesting that people have self-control problems and that they are
not fully aware of it.
The remainder of the paper is organized as follows. Section 3.2 presents the multi-
tasking model with time-inconsistent agents. Section 3.3 presents our main results.
Section 3.4 concludes.
PRESENT-BIASED AGENTS 103
3.2 The Model
In this section we lay out our model of time-inconsistent preferences in a multi-tasking
environment. First, present-biased preferences and possible believes about future be-
havior are introduced. Second, a multi-tasking principal-agent environment is set up.
In contrast to the seminal contribution by Holmstrom and Milgrom (1991), we add
the complication that eort devoted to some tasks pays out immediately while eort
devoted to other tasks pays out with delay. We assume that eort costs accrue imme-
diately. Finally, we combine both approaches in a simple, three-period model with two
tasks only.
3.2.1 Present-Biased Preferences and Beliefs
Let u
t
be an agents instantaneous utility in period t. In each period an agent does
not only care about her instantaneous utility, but also about her discounted future
instantaneous utilities. Let U
t
(u
t
, ..., u
T
) denote an agents intertemporal preference
from the point of view of period t. The standard model employed by economists is
exponential discounting, that is U
t
(u
t
, ..., u
T
) =

T
=t

, where (0, 1] denotes the

discount factor. In a parsimonious way, exponential discounting captures the fact that
agents are impatient. In addition, it implies that agents decisions are time-consistent.
When considering trade-os between two periods in time it does not matter when the
agent is asked to take a decision. However, people tend to exhibit time-inconsistent
preferences (Benzion, Rapoport, and Yagil (1989), Kirby (1997), Kirby and Herrnstein
(1995)). By way of example: When being asked, most people will prefer to receive
$100 in 6 weeks over $90 in 5 weeks from now. However, when being asked again for
their preference 5 weeks from now, some people will reverse their decisions to wait for
the higher payment and opt for the immediate payment of $90. Such present-biased
preferences have been modelled by Phelps and Pollack (1968) and later, among others,
by Laibson (1997) and ODonoghue and Rabin (1999a, 1999b, 2001).
36
We follow this
36
See Frederick, Loewenstein, and ODonoghue (2002) for a comprehensive overview of the litera-
ture.
PRESENT-BIASED AGENTS 104
literature and apply a two-parameter model that can capture present-biased preferences
by a simple modication of exponential discounting.
Denition 3.1 (, )-preferences are time-inconsistent preferences that are repre-
sented as follows: For all t, U
t
(u
t
, u
t+1
, ..., u
T
) =
t
u
t
+

T
=t+1

, where 0 < ,
1.
In this formulation, represents the long-run, time-consistent discount parameter.
The parameter represents the the bias for the present. If = 1, (, )-preferences
coincide with standard exponential discounting. But if < 1, an agent places more
relative weight to period in period than she did in any other period prior to .
Applied to the above example: Be = .95 the weekly discount factor and = .9
the present-bias. The agent then prefers $100 in week 6 over $90 in week 5 in every
week before week 5 as $90 < .95 $100. But in week 5 her preference reverses as
$90 > .9 .95 $100.
Most researchers have modeled time-inconsistent preferences by interpreting an
agent at each point in time as a separate agent.
37
An agent thus consists of multiple
selves, where each self is choosing current behavior to maximize current preferences.
The current self knows that her future selves control future behaviors and thus holds
believes about her future selves. Strotz (1956) and Pollack (1968) applied two extreme
assumptions and established the following labels:
Denition 3.2 (i) A sophisticated agent is fully aware of her future selves. Such an
agent takes into account that future selves may exhibit time-inconsistent preferences.
(ii) A naive agent thinks that future selves will take time-consistent decisions. Such an
agent does not take into account that future selves in fact take present-biased decisions.
There is a long standing and lasting debate over whether people are naive or sophis-
ticated. On the one hand, ODonoghue and Rabin (1999a) report on self-commitment
devices such as alcohol clinics, Christmas clubs, or fat farms, indicating that people
are (at least partially) aware of their time-inconsistent behaviors. On the other hand,
37
For an alternative approach see Gul and Pesendorfer (2001).
PRESENT-BIASED AGENTS 105
DellaVigna and Malmendier (2003, 2004) report on evidence from, among others, the
health club and credit card industry that suggests that people are not (fully) aware of
their future self-control problems. Apart from the two extreme assumptions, in prin-
cipal, any degree of sophistication could be modeled. However, in this paper we follow
the approach in ODonoghue and Rabin (1999a) and analyze fully sophisticated and
fully naive agents only.
3.2.2 Multi-Tasking with Immediate and Delayed Benets
We analyze a situation in which an agent must allocate eort between dierent tasks.
We further assume that all eort costs are immediate while some but not all tasks pay
out with a delay only. Eort on the job is assumed to pay out immediately (wage
payment) while time dedicated to care for, say, ones health pays out in the long-run
only (absence of health problems). As another example, consider a students decision
to work a couple of hours per week for a rm or to fully concentrate on ones studies.
While the wages from working for the rm are paid out immediately, higher wages that
come along with good grades are realized only in the future.
In a classic paper Holmstrom and Milgrom (1991) derive optimal linear incentive
contracts in a principal-agent setting with non-veriable eort such that wages must
condition on noisy signals.
38
Without loss of generality we employ the linear incentive
model in this paper as well. However, the focus of our model very dierent. Holm-
strom and Milgrom are interested in the implications on optimal incentive provision
if performance measure are of diverging quality. They show that it can be optimal to
refrain from providing explicit incentives if, for example, only one of two tasks can be
measured, but some engagement in both tasks is desirable. They further analyze asset
ownership and job design. In the current model we abstract from problems of measure-
ment and risk allocation. We are interested in the implications for incentive provision
if the incentivized task pays out immediate while the private activity pays out only
38
The underlying assumption in their model is that the agent chooses eort levels continuously
over the time interval [0, 1] to control the drift vector of a Brownian motion. At each point in time
the agent can observe his accumulated performance before acting. Holmstrom and Milgrom (1987)
show that in such a setting the optimal incentive contract is indeed linear.
PRESENT-BIASED AGENTS 106
with delay. While existing models of time-inconsistent behavior analyze environments
in which an agent must accomplish a single task but has discretion when to do it, our
focus is on eort allocation between tasks.
3.2.3 Combining Present-Biased Preferences and Multi-Tasking
We now combine present-biased preferences with multi-tasking. To capture the basic
eects we start out with the most simple set-up. The agent has to allocate eort
between two tasks only. Eort allocated to the rst task, e
1
, is incentivized by a
principal and rewarded by way of a linear incentive scheme (e.g. eort in the work
place). The principals benet from e
1
is captured by B(e
1
), with B

(e
1
) > 0 and
B

(e
1
) < 0. Function B(e
1
) is assumed to measure the principals benet in monetary
terms. The reward is based on a signal that is produced by e
1
. In this basic setting
we abstract from issues of risk-allocation and assume that the signal is deterministic,
i.e. (e
1
) = e
1
. The incentive scheme can thus be written as w(e
1
) = e
1
+ . The
principals prot is then given by = B(e
1
)e
1
. Eort devoted to the second task,
e
2
, serves the agents private benet V (e
2
), with V

(e
2
) > 0 and V

(e
2
) < 0 (e.g. caring
for ones health). We assume that V (e
2
) accrues with one period delay only. Eort
costs, C(e
1
, e
2
), are however immediate. We assume C
i
(e
1
, e
2
) > 0, C
ii
(e
1
, e
2
) > 0,
and C
ij
(e
1
, e
2
) > 0, with i, j 1, 2, subscripts denoting partial derivatives. We thus
assume that eort levels are substitutes at the margin. This will be the case if, for
example, eort is interpreted as measuring the time devoted to a certain activity. Both,
V (e
2
) and C(e
1
, e
2
) are assumed to represent the agents benet and cost in monetary
terms.
We consider a setting with three periods only. In period t = 0 the principal oers a
contract, i.e. values for and , and the agent decides whether to accept or to refrain.
In the next period, t = 1, the agent chooses eort levels. According to the incentive
scheme, the eort devoted to the principals purposes, e
1
, pays out immediately. The
agent also decides on eort devoted to the own private benet. All eort costs are
borne in immediately in t = 1. In the nal period, t = 2, the agents private benet is
realized. The timing of the game is summarized in Figure 3.1.
PRESENT-BIASED AGENTS 107
E
t t = 0 t = 1 t = 2
The principal oers a
contract and the agent
decides whether to
accept or to refrain.
Eort levels are
chosen. Eort
costs accrue and
wages are paid.
Benets of
private activity
are realized.
Figure 3.1: The Timing of the Game.
3.2.4 Benchmark: Incentives for Time-Consistent Agents
As a benchmark we rst derive the optimal contract for time-consistent agents (TCs).
Throughout the paper we will assume explicit functional forms in order to obtain closed
form solutions. In Section 3.4 we discuss the generality of our results.
Assumption 3.1 B(e
1
) = b ln(e
1
), V (e
2
) = v ln(e
2
), and C(e
1
, e
2
) = .05(e
1
+ e
2
)
2
.
In addition, but without loss of generality we normalize to unity. We rst look at
eort levels without incentives. In this case no action is taken in period t = 0. Without
incentives eort devoted to the principals benet, e
1
, will always be set to zero. In
period t = 1 a TC thus chooses e
2
to maximize her intertemporal utility which is given
by
max
e
2
V (e
2
) C(e
2
) = v ln(e
2
) .05(e
2
)
2
. (3.1)
Solving the rst-order conditions yields

e
TC
2
=

10v, where superscript TC stands

for time consistency and the tilde indicates the situation without incentives. A TC
then realizes an intertemporal (reservation) utility level of U
TC
= V (

e
TC
2
) C(

e
TC
2
) =
.5v(ln(10v) 1). This is also a TCs welfare, dening an agents welfare as follows.
Denition 3.3 The welfare of an agent is the sum of her instantaneous utility levels
from a long-run perspective, i.e. U
t
(u
t
, u
t+1
, ..., u
T
) =

T
=t

.
With TCs there is no dierence between long-run and short-run perspective. The
dierence will however be important with time-inconsistent agents.
PRESENT-BIASED AGENTS 108
Consider now a principal who oers a linear incentive contract w(e
1
) = e
1
+ in
period t = 0. To determine the parameters of the contract the principal maximizes
max
,
B(e
1
()) e
1
() (3.2)
subject to the participation constraint (PC)
e
TC
1
() + + V (e
TC
2
()) C(e
TC
1
(), e
TC
2
()) U
TC
, (3.3)
where e
TC
i
() with i 1, 2 are the eort levels that a TC chooses given incentive
intensity . In optimum (3.3) must hold with equality. Substitution of the PC yields
max

B(e
TC
1
()) + V (e
TC
2
()) C(e
TC
1
(), e
TC
2
()) U
TC
. (3.4)
The principal thus maximizes the intertemporal social welfare by choice of incentive
intensity , where social welfare is dened as follows.
Denition 3.4 Social welfare is the sum the agents welfare and the principals prot.
Recall that we assumed B(e
1
), V (e
2
), and C(e
1
, e
2
) to be measured in monetary terms.
As the incentivized activity can be measured without error, the linear incentive scheme
allows the principal to implement any level of e
1
at rst-best costs.
If the TC accepts the contract she will choose eort levels in period t = 1 by
maximizing
max
e
1
,e
2
e
1
+ + v ln(e
2
) .05(e
1
+ e
2
)
2
. (3.5)
Solving the rst-order conditions we get
e
TC
1
() = 10
v

and e
TC
2
() =
v

. (3.6)
PRESENT-BIASED AGENTS 109
Maximization of (3.4) then yields

TC
=
_
(b + v)/10, (3.7)
such that eort levels are given by
e
TC
1
(
TC
) =

10 b

b + v
and e
TC
2
(
TC
) =

10 v

b + v
, (3.8)
the rst-best levels of e
1
and e
2
that maximize social welfare.
3.2.5 Incentives for Sophisticated Agents
The optimal incentive contract for sophisticated agents (sophisticates) is derived anal-
ogously. Absent incentives a sophisticate chooses e
2
in t = 1 to maximize
max
e
2
V (e
2
) C(e
2
) = v ln(e
2
) .05(e
2
)
2
, (3.9)
which diers from (3.1) in the time-inconsistency parameter only. Her optimal choice
of e
2
is given by

e
PB
2
=

10v, where superscript PB indicates present bias. Ab-

sent incentives a sophisticate thus realizes a welfare level U
PB
= V (

t
PB
2
) C(

t
PB
2
) =
.5v(ln(10v) ). Notice that from a long-run perspective the discount factor between
periods one and two is given by 1 and not by . The dierence of U
TC
and U
PB
is
thus an agents welfare loss in monetary terms due to time-inconsistent preferences.
Subtracting U
PB
from U
TC
we get .5v(1ln()), which is positive whenever < 1,
and increasing with decreasing.
The principals objective function (3.2) is now subject to a sophisticates PC
e
PB
1
() + + V (e
PB
2
()) C(e
PB
1
(), e
PB
2
()) U
PB
. (3.10)
Notice that the agents benet from e
2
, V , is not discounted by . A sophisticate
decides in period t = 0 whether or not to accept the incentive contract. Both sides
of (3.10) are thus discounted by which therefore cancels. In optimum the PC must
PRESENT-BIASED AGENTS 110
hold with equality. Substituting (3.10) into (3.2) yields
max

B(e
PB
1
()) + V (e
PB
2
()) C(e
PB
1
(), e
PB
2
()) U
PB
. (3.11)
When choosing incentive intensity the principal now has to take into account that
the agent will give in to her present-biased preference when she decides on eort levels
in period t = 1. If the contract is accepted a sophisticate maximizes
max
e
1
,e
2
e
1
+ + v ln(e
2
) .05(e
1
+ e
2
)
2
. (3.12)
Solving the rst-order conditions gives eort choices
e
PB
1
() = 10
v

and e
PB
2
() =
v

, (3.13)
which dier from (3.6) only in the time-inconsistency parameter . Maximizing now
the principals objective function (3.11) with respect to yields

S
=
_
(b + v)/10, (3.14)
such that eort levels are given by
e
TC
1
(
TC
) =

10 b

b + v
and e
TC
2
(
TC
) =

10 v

b + v
, (3.15)
Again, for = 1 both eort and incentive levels of TCs and sophisticates coincide.
By comparison of (3.7) and (3.14) it can be seen that
TC
>
S
whenever < 1.
Comparing (3.8) and (3.15) shows that < 1 ensures e
S
1
> e
TC
1
but e
TC
2
> e
S
2
. That
is, given Assumption 3.1, TCs receive stronger incentives to work for the principal but
work less hard than sophisticates. TCs choose a higher eort level in their private
activities. This higher eort level increases eort costs such that it becomes relatively
more expensive for the principal to compensate for the incentivized activity. Figures
3.2, 3.3, and 3.4 provide an illustration.
PRESENT-BIASED AGENTS 111
0.3
0.4
0.5
0.6
0.7
0.2 0.4 0.6 0.8 1
beta
Figure 3.2: Alpha levels. The gure shows alpha levels as functions of , given b = 2 and
v = 4. The top straight line depicts incentive intensity with TCs. The curve in the middle is
alpha for sophisticates, the lowest curve for naifs. For = 1 incentive intensities coincide.
Sophisticates harm themselves due to time-inconsistent preferences. Their welfare
is maximized at eort levels (3.8) which dier from (3.15) whenever < 1. However,
there is no additional loss caused by the introduction of explicit incentives for an
immediately rewarded task. The participation constraint ensures that a sophisticate
always receives her reservation welfare level of U
S
.
3.2.6 Incentives for Naive Agents
We now turn to naive agents (naifs). Absent incentives there is no dierence between
sophisticates and naifs. No action is taken in period t = 0, and it is only then that
beliefs about preferences in period t = 1 can dier. In the case with incentives it
appears - at rst sight - unclear whether naifs or sophisticates will be better o. There
are two opposing eects. On the one hand, in period t = 0 naifs wrongly believe that
they will take a time-consistent, welfare maximizing decision in period t = 1. Hence,
to ensure a naifs participation she must be given a perceived reservation utility of U
TC
which exceeds her true reservation utility of U
PB
. On the other hand, a naif wrongly
predicts her eort choices in t = 1 given incentive intensity . She expects to act like a
TC according to (3.6), but will indeed give in to her time-inconsistent preferences and
PRESENT-BIASED AGENTS 112
2
2.5
3
3.5
4
0.2 0.4 0.6 0.8 1
beta
Figure 3.3: Levels of e
1
. The gure depicts
levels of e
1
as functions of , given b = 2 and
v = 4. The straight line in the middle is
e
TC
1
, the highest, decreasing line is e
S
1
, and
the lowest, increasing line depicts e
N
1
. For
= 1 levels of e
1
coincide.
1
2
3
4
5
0.2 0.4 0.6 0.8 1
beta
Figure 3.4: Levels of e
2
. The gure depicts
levels of e
2
as functions of , given b = 2
and v = 4. The top, straight line is e
TC
2
,
the line in the middle is e
N
2
, and the lowest,
steepest line depicts e
S
2
. For = 1 levels of
e
2
coincide.
act like a present-biased agent according to (3.13). Her perceived utility must not fall
short of U
TC
, but as she will make dierent eort choices her realized utility level will
fall short of U
TC
. In Section 3.3 we show that, given Assumption 3.1, it will always
even fall short of U
PB
. That is, the eect due to wrong beliefs about future selves
dominates and naifs are thus worse o than sophisticates. Since a sophisticate receives
her reservation utility level which is identical to a naifs welfare without incentives, a
naif is harmed by the principals incentive contract.
In the following we derive the optimal incentive contract from a principals point
of view. The principals objective function (3.2) is now subject to a naifs PC which is
identical to a TCs PC, which is given in equation (3.3). Substituting (3.3) into (3.2)
now yields
max

B(e
PB
1
()) e
PB
1
() U
TC
+ e
TC
1
() + V (e
TC
2
()) C(e
TC
1
(), e
TC
2
()). (3.16)
Recall that the principal is assumed to know both an agents time preference and her
belief about future selves. The principal thus takes into account that the naif will give
PRESENT-BIASED AGENTS 113
in to her time-inconsistent preferences in period t = 1 and choose eort levels according
to equation (3.13). The rst-order condition is now given by
(B

(e
PB
1
) )
e
PB
1

e
PB
1
+ e
TC
1
+ ( C
1
)
. .
=0
e
TC
1

+ (V

(e
TC
2
) C
2
)
. .
=0
e
TC
2

= 0(3.17)

_
b
10
v

_
_
10 +
v

2
_

_
10
v

_
+
_
10
v

_
= 0. (3.18)
Solving for we get

S
=

5
10
_
k + b v(1 ), (3.19)
with k =
_

2
v
2
+ (2v
2
+ 6bv) + (v b)
2
. It is straightforward to show that
N
coincides with
TC
for = 1. Given the optimized level of alpha naifs choose eort
levels
e
S
1
(
S
) =

5 (k + b v(1 + ))
_
k + b v(1 )
and e
S
2
(
S
) =

5 (2v)
_
k + b v(1 )
. (3.20)
Figures 3.2, 3.3, and 3.4 illustrate the relative size of incentive and resulting eort
levels for the three dierent types of agents.
3.3 Results
Since general functions only implicitly dene the solutions to the maximization prob-
lems in Sections 3.2.4 to 3.2.6 we have assumed explicit functional forms. But even
these simple functional forms do not always allow for closed form solutions. If neces-
sary we will therefore stick to numerical examples to show the results of the paper. For
completeness we rst establish the following proposition.
Proposition 3.1 A time-consistent agents welfare is always higher than a sophisti-
cated or naive agents welfare.
PRESENT-BIASED AGENTS 114
Proof: An agents welfare is dened as the sum of her instantaneous utility levels set-
ting = 1. Without incentives only TCs maximize welfare. Both, a sophisticates and
a naifs objective functions at time of eort choice dier from their welfare maximizing
objective functions. Their choices must thus be suboptimal. The optimal incentive
contracts oer both TCs and sophisticates their reservation utility levels. By the rst
part of this proof the rst exceeds the latter. A naif requests the reservation utility
level of a TC. She is thus oered a value for such that she received U
TC
if she indeed
chose like a TC. But in period t = 1 she gives in to her present-bias and deviates from
her planned eort choice. Her realized utility level must thus lie below U
TC
. q.e.d.
The focus of the paper is the comparison of sophisticates and naifs. Naifs wrongly
predict their future behaviors: In period t = 0 naifs think that they will behave like
TCs in period t = 1. To accept the principals incentive contract in t = 0 they must
be given an perceived utility level that matches the reservation utility level of a TC.
This commitment eect works in favor of a naif. However, once in period t = 1 a naif
deviates from her perceived eort choices and gives in to her present-bias. A naif is
thus harmed by this second eect. In contrast, a sophisticate anticipates that she will
act according to her present-biased preferences in t = 1 and thus requires an perceived
utility level that matches the reservation utility of an time-inconsistent agent only.
However, even though naifs receive a higher perceived utility level, they realize a lower
actual utility level. We show that, given Assumption 3.1, this second eect always
dominates. Naifs thus realize a lower welfare than sophisticates, and this welfare is
even lower than the welfare level naifs realize without incentives. This is summarized
in the following proposition.
Proposition 3.2 Given Assumption 3.1, in the presence of incentives a naive agents
welfare can be lower than a sophisticated agents welfare. A naive agents welfare can
thus be reduced by accepting the incentive contract.
Proof: The existence of a non-empty parameter space for which the result holds true
is shown by numerical example. Figure 3.5 plots welfare levels as functions of , given
b = 2 and v = 4. The top, straight line depicts a TCs welfare. The curve below
PRESENT-BIASED AGENTS 115
2.5
3
3.5
4
4.5
5
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
beta
Figure 3.5: Agents Welfare Levels with Incentives. The gure depicts welfare levels
as a functions of , given v = 4. The top, straight line depicts a TCs welfare. The curve
below depicts a sophisticates welfare, the lowest curve depicts a naifs welfare. For = 1
welfare levels coincide.
depicts a sophisticates welfare, the lowest curve depicts a naifs welfare. For = 1
welfare levels coincide. q.e.d.
From Proposition 3.1 we already know that sophisticates and naifs are always worse
o than TCs. We now show that, given Assumption 3.1, even though social welfare re-
alized with sophisticates and naifs is always below social welfare realized with TCs, the
principals prot can be higher if the agent is naive; it can be lower with sophisticates.
Proposition 3.3 Given Assumption 3.1, the principals prot from contracting with
naive agents can be higher than prot from contracting with TCs, even though social
welfare is always lower if agents have present-biased preferences. With sophisticates
the principals prot can be lower than with TCs.
Proof: The existence of a non-empty parameter space for which the result holds true
is shown by numerical example. Figure 3.6 plots levels of social welfare against , given
b = 6 and v = 4. The top straight line is social welfare with TCs. The curve below
depicts social welfare with naifs, and the lowest curve with sophisticates. Figure 3.7
depicts prot levels. The straight line in the middle are prots in case of a TC. The
top, decreasing line depicts the principals prots with naifs. Prots with sophisticates
PRESENT-BIASED AGENTS 116
10.7
10.8
10.9
11
11.1
11.2
11.3
0.5 0.6 0.7 0.8 0.9 1
beta
Figure 3.6: Social Welfare Levels. The
gure depicts social welfare as function of
for b = 6 and v = 4. The top, straight line
depicts social welfare with TCs, the curve in
the middle with naifs, the lowest with sophis-
ticates. For = 1 values coincide.
5
5.5
6
6.5
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
beta
Figure 3.7: Prot Levels. The gure de-
picts prot levels as function of , given b = 6
and v = 4. The straight line depicts prof-
its with TCs, the top, decreasing curve with
naifs, the lowest curve with sophisticates.
For = 1 values coincide.
are the lowest, increasing line. q.e.d.
In the following we are interested in the change of social welfare when the principal
oers incentive contracts as compared to the situation without incentives. From Propo-
sition 3.2 we know that naifs are harmed by incentives. The principal, on the contrary,
gains when oering incentive contracts. From Proposition 3.3 we know that his prot
when contracting with naifs can increase, the more severe the time-inconsistency prob-
lem gets, i.e. the lower . Furthermore, in the following we show that the agents loss
can exceed the principals gain. That is, social welfare may decrease if the principal
provides incentives relative to the situation without incentives. With TCs or sophis-
ticates this can never happen. Those agents always receive their reservation welfare
levels and the principal extracts the complete surplus from the additional, ecient
activity. This nding is summarized in the following proposition.
Proposition 3.4 Given Assumption 3.1, with naive agents social welfare may de-
crease if the principal oers incentive contracts as compared to the situation without
incentives.
PRESENT-BIASED AGENTS 117
4.2
4.4
4.6
4.8
5
5.2
5.4
0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
beta
Figure 3.8: Welfare Comparison. The gure shows welfare levels as function of , given
b = 2 and v = 4. The top straight line depicts social welfare with TCs. The curve that
coincides with social welfare with TCs at = 1 is social welfare with naif agents. The curve
that lies below at = 1 is the welfare of a time-inconsistent agent without incentives. Social
surplus with a naif decreases faster than a naifs welfare without incentives and eventually
falls short of it as decreases.
Proof: The existence of the eect is shown by numerical example. Figure 3.8 plots
levels of social welfare against , given b = 2 and v = 4. The top straight line depicts
social welfare with TCs and is included as benchmark only. The curve that coincides
with social welfare with TCs at = 1 is social welfare with time-inconsistent agents.
The curve that lies below at = 1 is the welfare of a time-inconsistent agent without
incentives. Given the parameter, social surplus with a naif decreases faster than a
naifs welfare without incentives and eventually falls short of a naifs welfare without
incentives. Social welfare is then lower if the principal oers an incentive contract as
compared to the situation without incentives. With a sophisticate this cannot happen,
even though both a sophisticates welfare without incentives and social welfare with
incentives coincide with the respective curves for a naif. With naifs social welfare falls
short of a time-inconsistent agents welfare without incentives exactly at the value of
where the principals prot with sophisticates falls negative. For such low values of
the principal would thus not oer not oer an incentive contract to a sophisticate.
With naifs her rises as decreases. q.e.d.
PRESENT-BIASED AGENTS 118
Proposition 3.4 shows that incentives can have detrimental eects. From a very dif-
ferent perspective, beginning with Titmuss (1970), there exists a literature discussing
negative eects of incentives. The main argument is motivation crowding out. Ac-
cording to this theory extrinsic rewards can be harmful because they may destroy in-
trinsic motivation. See Frey and Jegen (2001) for an overview of both crowing theory
and empirical evidence. For recent experimental evidence see Gneezy and Rustichini
(2000a, 2000b) and Fehr and Gachter (2002). In the context of our model the cause of
the detrimental eect of incentives is very dierent. We show that naifs are harmed by
incentives because the presence of incentives increases the mistake they make due to
their present-biased preferences. By denition of naivete, the agent does not anticipate
this behavior and thus does not get compensated for this mistake. Furthermore, we
have shown that the principals prot from providing incentives can be smaller than
the loss that accrues to a naive agent. In this case, social welfare is reduced by the
presence of incentives.
3.4 Conclusion
In this paper we have analyzed self-control problems in a multi-tasking environment.
While the existing literature analyzed environments in which a present-biased agent
must decide when to accomplish a single task, in this model we look at situations in
which an agent must allocate eort between multiple tasks and decide how much eort
to exert. We furthermore assumed that eort devoted to dierent activities pays out
at dierent points in time.
More specically, a principal oers a linear incentive contract for an immediately
rewarded task. Agents must allocate eort between this task and a private activity
that pays out only tomorrow. Such an activity could be, for example, caring for ones
health or continuing to go to school. Eort costs accrue immediately. There are
three dierent types of agents. Time-consistent agents, sophisticated agents, and naive
agents. Throughout the paper we assumed away complications of adverse selection.
It was thus assumed that the principal knows the type of agent he is contracting
PRESENT-BIASED AGENTS 119
with. We nd that present-biased agents take decisions that do not maximize their
long-run welfare, irrespective of the intensity of incentives. Sophisticated agents are
never harmed by incentives relative to the case without incentives as they always
receive their reservation utility levels. However, naive agents can be worse o in the
presence of incentive contracts as compared to the case without incentives. On the one
hand, they wrongly expect a high reservation utility and participate only if they are
paid high enough wages. On the other hand, they wrongly predict tomorrows eort
choices which is exploited by the principal. Furthermore, even though agents with
time-inconsistent preferences are always harmed by their present bias, the principals
prot with naifs can increase as the present-bias becomes more severe. Finally we show
that social welfare can decrease in the presence of incentives even though the principal
oers an ecient additional production opportunity. With naive agents it can happen
that the additional welfare loss due to present-biased eort choices in the presence of
incentive contracts exceeds the principals gain from oering the incentive contract.
The model thus oers a new theoretical possibility of detrimental eects of incentives,
complementary to existing arguments like the crowding out of intrinsic motivation by
extrinsic rewards.
The results of the paper were shown assuming explicit utility functions. The exact
conditions under which the eects highlighted in this paper hold true remain to be
identied in future research activity. However, their existence could be established.
An possible extension of the model will be to drop the assumption that the prin-
cipal knows the type of agents the is contracting with and analyze possible screening
contracts. Another possible extension will be the analysis of a setting with immediate
private benets and delayed wage payment. There are plenty natural situations imag-
inable where this constellation is of relevance. While continuing to work on ones Ph.D.
thesis after 5 p.m. pays out only with delay, the private benet of a relaxed evening
however accrues immediately.
Chapter 4
Inequity Aversion and Moral
Hazard with Multiple Agents

4.1 Introduction
We analyze how inequity aversion (Fehr and Schmidt (1999), Bolton and Ockenfels
(2000)) interacts with incentive provision in an otherwise standard moral hazard model
with multiple agents.
39
The theory of inequity aversion assumes that some but not all
agents suer a utility loss if their own material payos dier from the payos of other
agents in their reference groups. The approach can explain a large variety of seemingly
diverging experimental ndings that often conict with the standard assumption of
pure selshness.
40
This paper goes a step further and applies the theory of inequity
aversion to the theory of incentives. If agents do not simply maximize their own
material payos but also care for other agents payos they will respond dierently
to incentives than predicted under the assumption of pure selshness. Incorporating
social preferences into the theory of incentives thereby either exploiting them or
paying tribute to an additional constraint may help to understand why real world
contracts often dier from those contracts found optimal by the standard theory.

The chapter is based on joint work with Ferdinand von Siemens from the University of Munich.
39
See Grossman and Hart (1983) and Mookherjee (1984)
40
For an overview of the literature see, for example, Fehr and Schmidt (2003) and Camerer (2003).
INEQUITY AVERSION AND MORAL HAZARD 121
In a classic contribution to the theory of incentives Holmstrom and Milgrom (1991,
p. 24) state that it remains a puzzle for this theory that employment contracts so often
specify xed wages and more generally that incentives within rms appear to be so
muted, especially compared to those of the market. The authors oer an explanation
for the paucity of incentives based on the assumption that agents conduct multiple
tasks, and that tasks are measured with varying degrees of precision.
We oer an alternative, behavioral explanation to account for the observation that
incentives oered to employees within rms are generally low-powered compared to
high-powered incentives oered to independent contractors. We assume that within
rms social comparisons are pronounced whereas in the marketplace they are negligi-
ble.
41
We further assume that an agent suers a utility loss if another agent conducting
a similar task within the same rm receives a higher wage. We nd that behindness
aversion (suering only when being worse o) unambiguously increases agency costs
of providing incentives. As a consequence, behindness aversion may render equitable
at wage contracts optimal even though incentive contracts are optimal with selsh
agents. Hence, within rms where social comparisons are signicant we nd low pow-
ered at wage contracts to be optimal, whereas high powered incentive contracts will
be given to unrelated agents in the marketplace.
Furthermore, we argue that our analysis can contribute to the question of the
optimal size of a rm. Suppose the principal can set up dierent rms, but setting
up a rm involves xed costs. The principal now faces a trade-o. On the one hand,
integration of several agents within a single rm causes social comparisons and, as
shown in this paper, increased agency costs of providing incentives. On the other hand,
separation of agents into dierent rms involves additional xed costs. The solution
to this trade-o denes, in the context of this model, the optimal degree of integration.
More specically, in this paper we derive optimal moral hazard contracts assum-
ing risk- and inequity averse agents that constitute each others reference group. The
agents however do not compare themselves to the principal. Agents carry out iden-
tical tasks and regard it as unfair if their wage payments dier. We further assume
41
See, for example, Bewley (1999) for supporting evidence.
INEQUITY AVERSION AND MORAL HAZARD 122
that the principal is both risk neutral and selsh. To keep the analysis tractable we
consider the most simple set-up with two agents, two eort levels and two possible
output realizations; to receive closed form solutions we assume an explicit utility func-
tion and a linear inequity term as in Fehr and Schmidt (1999). In the appendix we
however show that our results hold true (1) for any concave utility function and (2)
irrespective of the functional form of disutility from inequity. Eort is taken to be
non-contractible such that incentive compatible wages must condition on stochastic
output realizations. Hence, agents suer if output realizations and thus wages dier.
We show that behindness aversion among agents unambiguously increases agency costs
of providing incentives. This also holds true if agents, in addition, suer from being
better o unless they account for eort costs in their comparisons.
The intuition behind this nding can be seen as follows. Inequity aversion eects
an utility loss if output realizations diverge. The resulting, reduced utility levels could
be implemented without inequity aversion as well, simply by lowering the wages. Since
these lower utility levels were not optimal without inequity aversion, they cannot be
optimal now.
Increased agency costs can undermine eciency in two ways. First, equitable at
wage contracts may become optimal even though incentive contracts are optimal with
selsh agents. Second, to avoid social comparisons the principal may employ one agent
only, thereby forgoing the ecient eort provision of the other agent. This second eect
of inequity aversion is qualitatively dierent from the impact of risk aversion on optimal
contracts. The principal can respond to high degrees of risk aversion only by waiving
incentives and oering at wages, whereas with inequity aversion or more generally
with social preferences he has an additional instrument at hand if he can control an
agents reference group. It is possible to eliminate inequity and still provide incentives
to at least one agent. We call this the reference group eect. Third, endowing the
principal with the option to set up a second rm at a xed cost allows to analyze
whether integration or separation is optimal.
Further results are derived. Since optimal wages condition on the output realization
of the respective other agent as well, the sucient statistics result due to Holmstrom
INEQUITY AVERSION AND MORAL HAZARD 123
(1979) does not apply. We nd that inequity aversion renders team contracts optimal
even if output is uncorrelated. Analyzing the interaction between risk and inequity
aversion, we nd that the additional agency costs due to inequity aversion are higher,
the higher the degree of risk aversion. With risk neutral agents inequity aversion does
not impact equilibrium agency costs as long as no limited liability constraint binds.
Finally, labor contacts often encompass a clause prohibiting employees to communi-
cate their salary. At rst sight, inequity aversion could serve as an explanation for
this observation. We however show that secrecy of salaries only further increases the
additional agency costs due to inequity aversion.
Related Literature
Itoh (2003) and Demougin and Fluet (2003) are most related to our paper. Itoh (2003)
analyzes how inequity aversion among risk neutral agents changes optimal incentive
contracts, assuming limited liability to be the source of moral hazard. In contrast
to our results, Itoh nds that inequity aversion can never harm the principal. With
risk neutrality the principal can always choose a fully equitable contract out of the set
of contracts that are optimal without inequity aversion. Moreover, inequity aversion
can even increase the principals prot. With limited liability the principal may be
forced to pay the agents rents to provide incentives because there is a lower bound
on agents wage payments. However, inequity aversion enables the principal to punish
an agent harsher than paying the lowest possible wage level, simply by paying other
agents more, thereby reducing agents rents. Demougin and Fluet (2003) also analyze
a two agents moral hazard problem assuming risk neutrality and limited liability. They
compare group and individual bonus schemes for behindness-averse agents and derive
conditions under which either scheme implements a given eort level at least costs.
Inequity aversion between multiple agents is also analyzed by Rey Biel (2003) and
Neilson and Stowe (2003). Rey Biel (2003) analyzes a setting with two inequity averse
agents and a principal in which agents eort choices deterministically translate into
output. He exogenously assumes the participation constraint to be slack and nds
that the principal can always exploit inequity aversion to extract more rents from his
INEQUITY AVERSION AND MORAL HAZARD 124
agents. Neilson and Stowe (2003) restrict their analysis to linear piece-rate contracts
and identify the conditions under which other-regarding preferences lead workers to
exert more or less eort than selsh agents, and whether the optimal piece rate is
higher or lower for inequity averse agents.
Englmaier and Wambach (2003) and Dur and Glaser (2004) consider comparisons
between agents and principal. Englmaier and Wambach (2003) nd that the sucient
statistics result does not apply and that inequity aversion causes a strong tendency
towards linear sharing rules. Dur and Glaser (2004) show that inequity aversion can
be a reason for high incentives, even for prot sharing, as this reduces inequity.
In Bartling and von Siemens (2004) we analyze how incentive provision in team
production is aected if agents are inequity averse. In contrast to the classic result by
Holmstrom (1982) we nd that ecient eort choices can be implemented by simple
budget-balancing sharing rules if agents are suciently inequity averse. Conditions for
eciency become less restrictive the smaller the team. This ts common observation
that small teams often work well whereas larger ones suer from free-riding.
The remainder of the paper is organized as follows. Section 4.2 presents the basic
model. In Section 4.3 we derive the optimal incentive contracts for inequity averse
agents. Section 4.4 presents our main results. Section 4.5 explores the implications of
our results for the optimal rm size. Section 4.6 analyzes the case with secret salaries.
In Section 4.7 we discuss comparison of rents, disutility from being better o, and
status preferences. Section 4.8 concludes. In the Appendix we discuss the generality
of our results.
4.2 The Model
4.2.1 Projects, Eort, and Probabilities
Suppose a principal can employ two risk averse agents. If employed, each agent manages
a project with stochastic output x x
l
, x
h
, where x
h
> x
l
and x := x
h
x
l
. Each
agent faces a binary eort choice. He either exerts eort, e = 1, or he shirks, e = 0.
INEQUITY AVERSION AND MORAL HAZARD 125
Eort costs are denoted by (1) = > 0 while shirking is assumed to be costless,
(0) = 0. If an agent exerts eort, the output of his project is x
h
with probability
and x
l
with probability 1 , where ]0, 1[. If an agent shirks, the output of his
project is always x
l
. Eort is assumed not to be contractible. The agents projects are
independent, their production outcomes are uncorrelated.
4.2.2 Preferences: Risk- and Inequity Aversion
We depart from the standard literature by assuming that agents are inequity averse in
the sense of Fehr and Schmidt (1999).
42
We assume that an agents utility is additively
separable in the following three components. First, each agent enjoys utility u(w) from
his wage payment w by the principal. To derive explicit results we assume this utility
function to take on the specic form
43
u(w) = (1 +

1 + 2rw)/r. (4.1)
This function is strictly increasing and convex for all w > 1/2r. Thus, the agent
is risk averse with respect to his income. The corresponding inverse function h(x) :=
u
1
(x) = x + rx
2
/2 is well dened for all x > 1/r. For small w, r can be considered
as the agents approximated degree of absolute risk aversion. This approximation is
correct at a zero wage: u

(w)/u

(w)[
w=0
= r. Second, an agent incurs eort costs
if he works; shirking is costless. Finally, an agent suers from inequity. We assume
an agents reference group to be conned to the other agent, thus the agents do not
compare themselves to the principal. Agents carry out an identical task and regard
it as unfair if wage payments dier. Since the principal conducts a dierent task his
payo is not taken to be a point of reference. The identication of an agents relevant
reference group will, however, ultimately be an empirical question.
In the body of the paper we restrict attention to behindness aversion. Whenever
42
See Bolton and Ockenfels (2000) for a related formulation of inequity aversion.
43
In the appendix we show that our results neither hinge upon this explicit utility function nor
on the assumed linear formulation of inequity aversion by Fehr and Schmidt (1999). The chosen
functional forms however allow to derive closed from solutions.
INEQUITY AVERSION AND MORAL HAZARD 126
an agent receives a lower payo than the other agent he suers a utility loss, but agents
do not suer if they are better o than the other agent. More formally, suppose agent
i 1, 2 receives wage w
i
, whereas agent j ,= i receives wage w
j
. Agent i

s utility
function can then be written as
v
i
(w
i
, w
j
) = u(w
i
) (e) max[u(w
j
) u(w
i
), 0]. (4.2)
The parameter 0 is a measure of behindness aversion. The higher the more
an agent suers from inequity. Notice that the above formulation does not imply
that agents compare utilities interpersonally, but rather that agent i suers from the
inequity between the utility he obtains from wage w
i
and the utility he would enjoy
when receiving the higher wage w
j
himself. Both agents maximize expected utility.
Despite the evident experimental evidence on inequity aversion it is still an open
question what exactly people compare; whether they focus, for example, on wage pay-
ments or utility from wage payments, and whether they account for dierences in
eort costs or not.
44
In this paper, we assume that agents compare utility levels as
this renders the principals maximization problem well behaved.
45
To avoid tedious
case distinctions we neglect the possibility that agents account for eort costs in their
comparisons. In Section 4.7.1 we however show that accounting for eort costs in the
inequity term does not conict with but rather reinforces the qualitative results of this
paper. In Section 4.7.2 we show that introducing suering from being better o, again,
only reinforces our qualitative results unless agents account for eort costs in their
comparisons.
The principal is both risk-neutral and unaected by inequity concerns. He maxi-
mizes expected output minus expected wage payments.
44
For a more detailed discussion of inequity aversion see Fehr and Schmidt (1999, 2003).
45
Otherwise constraints are not linear, the maximization problem not concave, and the solution
not straightforwardly characterized by rst-order conditions.
INEQUITY AVERSION AND MORAL HAZARD 127
4.3 Contracts
We focus on symmetric contracting such that the principal oers identical contracts
when employing both agents. The principal has three options. He can either employ
both agents and implement eort or shirking, or he can decide to employ one agent
only to avoid social comparisons.
46
In the following section we derive optimal contracts
implementing these eort choices.
4.3.1 Benchmark: The Single Agent Case
The principal can avoid social comparisons by employing one agent only. Recall that
we have conned an agents reference group to the respective other agent working
with the same principal. With a single agent inequity aversion is thus irrelevant. The
optimal contract for the employed agent (incentive or at wage contract) then depends
on the standard parameters of the model via the participation and incentive constraint.
Suppose rst the principal wants to implement high eort. Since eort is not veriable
wages must condition on stochastic output realizations and the classic risk-incentive
trade-o arises. Dene w
i
as the agents wage if his output is i h, l, and dene
u
i
:= u(w
i
). To render the principals maximization problem concave, we rewrite the
principals objective function and the constraints in terms of u
h
and u
l
. An agents
outside option is normalized to zero. The resulting rst-order conditions then yield
u

h
=

and u

l
= 0 (4.3)
as the optimal contract, and prot can be written as
P
i
1
= x
h
+ (1 )x
l

_
h() +
r
2
(1 )
2
_
(4.4)
46
In principle, he could also oer a hybrid contract: an incentive contract to one agent and a
non-incentive contract to the other agent. Note that due to inequity aversion such a non-incentive
contract would not be a at wage contract. It can be shown that considering the hybrid contract
would not change the qualitative results of this paper.
INEQUITY AVERSION AND MORAL HAZARD 128
where superscript i denotes incentive contract and the subscript shows the number of
agents employed. Dene
RAC :=
r
2
(1 )
2
as the risk-agency-costs that have to be payed on top of the rst-best cost of eort
implementation h() due to risk aversion.
Suppose now the principal oers a at wage contract. The agent then never exerts
eort and the participation constraint is satised at at wage w
f
= 0. The principals
prot in this case is P
f
1
= x
l
. The dierence in expected prot from implementing
eort as compared to paying a at wage is given by
B := x h() RAC.
Thus, it is optimal for the principal to implement high eort if and only if
B 0 x h() + RAC. (4.5)
The principal oers an incentive contract whenever the expected output increase is
suciently large relative to the rst best cost of implementing eort and the RAC.
The condition is more likely to be met if eort cost and risk aversion are small and
the information content of the project outcomes is high. Exerting eort is ecient if
x h() but risk aversion leads to a trade-o between insurance and eciency and
causes additional RAC. This leads to inecient eort choices if h() +RAC x
h(). If x h() + RAC the ecient eort level is implemented but risk aversion
reduces the principals expected prot.
In the next section we show that inequity aversion amplies these eects. Inequity
aversion causes additional agency costs which unambiguously rise as the level of in-
equity aversion rises. This further reduces the principals expected prot, and it can
lead to additional ineciencies. Throughout the paper we therefore assume incentive
condition (4.5) to be fullled. B < 0 is the uninteresting case since at wage contracts
would then always be optimal even without inequity aversion.
INEQUITY AVERSION AND MORAL HAZARD 129
4.3.2 The Two Agents Case
In this section we consider the two agents case. Both agents work within the same rm
and we thus assume that they compare their wage levels. An agent suers a utility
loss in case he is behind. In contrast, we assume that an agent would not compare his
wage to the wage of an agent with whom he only interacts in the market, i.e. an agent
that works for another principal.
With incentive contracts inequity arises naturally as output is stochastic and in-
centive compatible wages must condition on output realizations. At rst glance the
eect of inequity aversion on agency costs is ambiguous. Behindness aversion increases
incentives because exerting eort reduces the probability of being behind. At the same
time agents anticipate that even if they exert high eort with positive probability they
will be behind. Ex ante agents have to be compensated for this expected utility loss
to ensure participation.
We show that the positive eect on incentives is always dominated by the negative
eect on participation and, therefore, behindness aversion unambiguously increases the
agency costs of providing incentives. The intuition can be seen as follows. Without
inequity aversion the second-best optimal incentive contract assigns wage levels to each
possible output realization such that both IC and PC are fullled and binding. For
some output realizations (i.e. agent one is successful, agent two is not) the contract
assigns diverging wage levels to the agents (agent one receives a higher wage than
agent two, assuming the monotone likelihood ration to hold). If now inequity aversion
is considered, the utility of agents receiving less than others (agent two) is reduced
by the amount of suering from being behind. However, this lower utility level could
have been achieved without inequity aversion as well simply by lowering the respective
wage level, which reduces the principals cost. As this was not optimal without inequity
aversion it cannot be optimal now.
In the appendix we show that this intuition holds generally. Assuming only con-
cavity of the utility function we show that inequity aversion renders it weakly more
expensive to implement each possible eort level. However, our arguing does not hold
INEQUITY AVERSION AND MORAL HAZARD 130
when there is limited liability. With limited liability the lowest possible wage pay-
ment and thus the lowest possible utility level for an agent is bounded from below. To
provide incentives the principle may thus be forced to leave the agents rents. In this
case inequity aversion provides the principal with the possibility to reduce the lowest
possible utility level. An agent can now not only be punished by paying out the lowest
wage level but in addition by paying other agents a higher wage. The lowest possible
utility level for an agent can thus be reduced without violating the limited liability
constraint. This in turn enables the principal to reduce the agents rents.
Suppose rst the principal does not want to implement eort. He then oers two
at wage contracts. Since there is never inequity, inequity aversion is irrelevant and
the principals prot is simply
P
f
2
= 2 P
f
1
= 2 x
l
. (4.6)
Suppose now the principal wants to implement eort. We show that the principals
expected prot is not just twice the expected prot in the single agent case but P
i
2

2 P
i
1
. As both agents are symmetric, we assume that optimal wages are symmetric
in the sense that they condition on the output realizations of both projects but not
on the identity of the agent. Denote by w
ij
the wage of an agent with output i if
the other agents output is j. Dene u
ij
:= u(w
ij
) as an agents utility from wage
w
ij
. As there are four possible states of the world, a contract determines four wage
levels: w
ll
, w
hh
, w
lh
, and w
hl
, where h stands for high and l for low output. To render
the principals maximization problem concave with linear constraints, we rewrite the
principals objective function and the constraints in terms of u
hh
, u
hl
, u
lh
, and u
ll
.
Recall that the maximum functions in the agents utility functions in (4.2) create
potential kinks. At these points, the utility functions and thus the PC and IC are not
dierentiable, potentially rendering it impossible to characterize optimal contracts by
rst-order conditions. However, the following lemma allows to avoid this problem.
Lemma 4.1 The optimal incentive compatible contract for two inequity averse agents
satises u

hl
u

lh
.
INEQUITY AVERSION AND MORAL HAZARD 131
Proof: Suppose this was not the case, that is u
hl
< u
lh
at the optimum. Then the IC
and PC are given by
(I C)
2
u
hh
+ (1 )[u
hl
(u
lh
u
hl
)]
2
u
lh
(1 )u
ll
0
(PC)
2
u
hh
+ (1 )[u
hl
(u
lh
u
hl
)] + (1 )u
lh
(1 )
2
u
ll
0
Consider changes du
lh
< 0 and du
hl
= du
lh
(1 )/(1 + ). This leaves (PC)
unaected but improves (IC). The principals prot increases by
dP
i
2
= 2(1 )
_
h

(u
hl
)
1
1 +
h

(u
lh
)
_
du
lh
,
which is strictly larger than zero as (1 )/(1 + ) 1, du
lh
< 0, u
hl
< u
lh
, and
h

(u) > 0. q.e.d.

Notice that du
lh
< 0 has a twofold eect on (PC). On the one hand, this decreases
the agents utility if his own project fails whereas the other agents project is successful.
On the other hand, unfavorable inequity decreases if the agent himself is successful
whereas the other agent is unfortunate. In the latter case the agents utility increases.
If the inequity reducing eect dominates, > 1, the principal may decrease both u
hl
and u
lh
while keeping (PC) unaected and not impairing (IC). In either case, the
principal can increase his expected prot without violating a constraint, and u
hl
< u
lh
cannot be optimal.
By Lemma 4.1 we can introduce an additional constraint, u
hl
u
lh
0, without
restricting the attainable maximum. We call this constraint the Order Constraint
(OC). The maximum functions in the agents utility functions are thus removed and
the principal maximizes
P
i
2
= 2
_
x
l
+
2
[x h(u
hh
)] + (1 )[x h(u
lh
) h(u
hl
)] (1 )
2
h(u
ll
)

(4.7)
INEQUITY AVERSION AND MORAL HAZARD 132
with respect to u
hh
, u
hl
, u
lh
, and u
ll
, where
(IC)
2
u
hh
+ (1 )u
hl

2
[u
lh
(u
hl
u
lh
)] (1 )u
ll
0
(PC)
2
u
hh
+ (1 )u
hl
+ (1 )[u
lh
(u
hl
u
lh
)] + (1 )
2
u
ll
0
(OC) u
hl
u
lh
0.
are the constraints characterizing the principals choice set.
47
We begin by assuming that the OC is not, whereas the IC and PC are binding.
Solving the resulting rst-order conditions then yields
u

hh
=

+
(1 )(1 + ( + r))
rk
(4.8)
u

hl
=

(1 + ( + r))
rk
(4.9)
u

lh
=
(1 (1 + ) + (1 + )r)
rk
(4.10)
u

ll
=
((1 + (1 + )) r)
rk
(4.11)
where k = 1+(2+(1+)). The Lagrange multipliers for the PC and IC are given
by = 2(1 + r[1 + (1 + ) + 2
2
] + 2(1 + ))/k and = 2(1 )(r[1 +
+ 2
2
] + (1 + 2))/k. Since these are strictly positive, both the PC and
IC are indeed binding as initially assumed. We also have to check whether it holds
true that the OC is slack, i.e. whether we have u

hl
u

lh
. The dierence is given by
(r + (r 1))/rk. Thus, the OC is indeed slack and the solutions (4.8) - (4.11)
are valid if and only if either r 1 or r < 1 and < where
:= r/((1 r)). (4.12)
47
We do not consider dominant strategy implementation in this paper, i.e. we only look at contracts
such that the constraints are satised for one agent given that the other agent behaves as expected.
Even though both agents participating and exerting eort then forms a Nash equilibrium it is possibly
not unique.
INEQUITY AVERSION AND MORAL HAZARD 133
Finally, since h(u) is dened for u 1/r only, we have to verify that this always
holds. Algebraic manipulations show that u

lh
u

ll
. The solution is thus valid if
u

ll
1/r, or ru

ll
1. This condition holds with equality if r = r = (1+)/().
Dierentiating ru

ll
with respect to r yields /k 0. Hence, ru

ll
rises in r. Since we
must have r > 0, r always exceeds r, and u

ll
never falls short of 1/r.
Suppose now that all the constraints PC, IC, and OC are binding. The binding OC
forces the principal to set u
lh
= u
hl
. This restriction on the contract design eliminates
inequity but comes at a cost. Solving the corresponding rst-order conditions we get
u

hh
=

+
(1 )

(4.13)
u

hl
= u

lh
=

=
(1 )

(4.14)
u

ll
= . (4.15)
The Lagrange multipliers of the PC and IC are = 2r + 2 and = 4r(1 )/.
Since both are strictly positive, the PC and IC are indeed binding as initially assumed.
As h(u) is dened for u 1/r only, the above solution is valid only if r < 1.
The overall optimal solution depends on whether the OC is binding or not, which
in turn depends on r, and . This is summarized in the following proposition.
Proposition 4.1 (Optimal Contracts For Inequity-Averse Agents)
i) Suppose r 1. The optimal incentive compatible contract for two inequity
averse agents is given by (4.8) - (4.11).
ii) Suppose r < 1. If < , the optimal incentive compatible contract for two
inequity averse agents is given by (4.8) - (4.11). If , it is given by (4.13) -
(4.15).
Proof: There are two cases. First, suppose r 1. Then solution (4.13) - (4.15) is
not valid as u

ll
< 1/r, whereas solution (4.8) - (4.11) is valid for all as we always
INEQUITY AVERSION AND MORAL HAZARD 134
get (r + (r 1))/rk > 0. Second, suppose r < 1. Then for all < both
extreme points are candidates for the overall solution, but (4.8) - (4.11) dominates as
the maximum is not restricted by the OC. For all , only solution (4.13) - (4.15)
is valid. q.e.d.
4.4 Results
4.4.1 Inequity Aversion Renders Team Contracts Optimal
Since we assume output to be uncorrelated an agents output realization does not
contain information about the other agents eort choice. According to the classic re-
sult by Holmstrom (1979) optimal wages should only condition on sucient statistics
for eort choices. In our model wages should thus only condition on the own output
realization. Nonetheless, since agents compare the utility levels from their wages opti-
mal contracts also condition on the other agents output realization in order to reduce
inequity. Therefore, the sucient statistics result does not apply.
48
Dene a team
contract as a compensation scheme such that an agents wage depends positively on
the other agents success. Thus, in a team contract we have w
hh
> w
hl
and w
lh
> w
ll
.
As summarized in the following proposition inequity aversion renders team contracts
optimal.
Proposition 4.2 (Team Contracts)
The sucient statistics result does not apply: Inequity aversion renders team contracts
optimal even if output is uncorrelated.
Proof: Comparison of the relevant utility levels in Proposition 4.1 yields u

hh
u

hl
=
(1 + ( + r))/rk 0 and u

lh
u

ll
= (1 + ( + r))/rk 0. q.e.d.
Since output is stochastic, agents obtain dierent output realizations with positive
probability even though both agents exert high eort. The unfortunate agent then
48
In the context of interdependent preferences this result naturally arises. It was rst shown in
Englmaier and Wambach (2003).
INEQUITY AVERSION AND MORAL HAZARD 135
suers from obtaining a lower wage than the fortunate agent. The optimal contract
accounts for this eect and adjusts wage levels accordingly.
4.4.2 Inequity Aversion Causes Additional Agency Costs
In the benchmark case of a single agent inequity aversion is irrelevant and does not
inuence the principals prot. This is also the case with at wage contracts for two
agents as there is never inequity. However, with incentive contracts for two inequity
averse agents additional agency costs arise. Suppose r 1 or r < 1 but <
such that the optimal contract is characterized by (4.8) - (4.11). Substituting optimal
utility levels, the principals maximum prot is then given by
P
i
2
= 2 P
i
1
IAC, (4.16)
where
IAC :=
(1 )(2r + (r 1) + r
2

2
)
rk
. (4.17)
denotes the inequity agency costs, the additional agency cost due to inequity aversion.
Inequity aversion has a negative eect on the principals maximum prot as the above
solution is only valid if either r 1 or holds, and this ensures that IAC are
positive. Equivalently, suppose r < 1 and < such that the optimal contract
is characterized by (4.13) - (4.15). Substituting optimal utility levels, the principals
maximum prot is then given by
P
i
2
= 2 P
i
1
IAC, (4.18)
where
IAC :=
r
2
(1 )

(4.19)
denotes the inequity agency costs in this case. Again, the principals prot with two
hard working agents is strictly less than twice the prot with only one hard working
INEQUITY AVERSION AND MORAL HAZARD 136
agent as the IAC are always positive. Note that in the latter case the IAC do not depend
on as the above solution is subject to the OC binding and inequity is completely
eliminated. However, inequity aversion reduces the principals prot as it forces him
to set u

hl
= u

lh
via the binding OC. We can now derive the following result.
Proposition 4.3 (Additional Agency Costs)
Inequity aversion among agents causes additional agency costs of implementing eort.
These agency costs weakly increase and converge as the level of inequity aversion rises.
Proof: Suppose r 1. The IAC are then given by (4.17) and r 1 ensures
(4.17) > 0. Dierentiating (4.17) with respect to yields
IAC

=
2(1 )(1 + (r + ))(r(1 + ) )
rk
2
, (4.20)
which is strictly positive as r 1. The limit of IAC is given by
lim

IAC =
1
1 +
_
r(2 + r) 1
_
, (4.21)
where r 1 again ensures the expression to be positive. Suppose now r < 1. In
case the above arguments on sign of IAC and their derivative w.r.t. apply. In
case > the IAC are given by (4.19) which is positive as we have r > 0, > 0, and
]0, 1[, does not change in , and is thus equal to the limit as . q.e.d.
Proposition 4.3 proves that the negative eect of inequity aversion on the PC always
dominates the positive eect on the IC. The negative eect of inequity is however
bounded because the principal can always equate w
hl
and w
lh
if becomes too large.
The optimal contract then remains unchanged as further increases. The intuition for
the dominance of the eect on the PC can best be seen when approaching the problem
from a dierent angle. Inequity aversion eects a utility loss in certain states of the
world. If the resulting reduced utility level were second-best optimal, then they could
be realized without inequity aversion as well simply by lowering wage payments.
As lower utility levels are not second-best optimal without inequity aversion, they
INEQUITY AVERSION AND MORAL HAZARD 137
cannot be optimal now. In the appendix we show that this intuition straightforwardly
generalizes to less restrictive settings.
As an alternative intuition for the result consider the following. Suppose the OC
is binding. To eliminate suering from inequity aversion utility levels in case of di-
verging output realizations are equated. This clearly impairs incentives to exert eort.
Hence, in cases with identical output realizations wage payments must become more
extreme. Agents then have to bear more risk for which they must be compensated. The
same reasoning holds true if the OC is not binding. In addition to the increased risk,
agents then also have to be compensated for the inequity they bear despite the wage
compression in case output realizations diverge. This leads to the next proposition.
Proposition 4.4 (Complementarity)
The more risk averse the agents, the higher the additional agency costs due to inequity
aversion.
Proof: In case the OC does not bind the IAC are given by (4.17). Dierentiating
(4.17) with respect to r yields
IAC
r
=
(1 )
2
( + r
2

2
)
rk
2
(4.22)
which is unambiguously positive. In case the OC binds and the IAC are given by (4.19)
the respective partial derivative is clearly positive. q.e.d.
Since contracts that account for inequity aversion lead to more risk bearing, the
higher the degree of risk aversion, the higher the additional agency costs caused by
inequity aversion. Risk aversion and inequity aversion thus have complementary eects.
Consider the extreme case of risk neutral agents, i.e. u(w) = w. The principals ex-
pected incentive compatible wage payment per agent is then +(1)(+)(w
hl

w
lh
), the sum of the rst-best costs of implementing eort and compensation for in-
equity bearing. Notice that in the context of this model a limited liability constraint
will never bind as we have normalized the success probability when shirking to zero.
There is thus no rent that has to be given to the agent, i.e. the PC is binding. Since
INEQUITY AVERSION AND MORAL HAZARD 138
inequity aversion has an unambiguously negative eect on the PC, any amount of in-
equity decreases the principals expected prot. A possible positive eect of inequity
aversion on incentive provision cannot be realized since incentives can be provided at
rst-best costs already. With risk neutral agents (and no limited liability constraint
binding) a large set of optimal contracts can implement ecient eort choices at rst-
best costs. With inequity aversion only a subset of these optimal contracts remains
optimal, namely those contracts with w
hl
= w
lh
. The remaining subset of optimal
contracts is however non-empty. For example, the contract with w
hh
= /
2
and
w
hl
= w
lh
= w
ll
= 0 is always possible. It provides incentives at rst-best costs and
eliminates all inequity. We summarize our ndings in the following proposition.
Proposition 4.5 (Risk Neutrality)
With risk-neutral agents and no limited liability constraint binding, inequity aversion
reduces the set of optimal contracts but does not impact the equilibrium outcome.
Itoh (2003) also analyzes a moral hazard setting with risk-neutral agents but assumes
limited liability constraints to bind. In this case agents receive a rent. Inequity aversion
provides the principal with the possibility to reduce an agents utility below the level
that arises from paying the lowest possible wage level, simply by paying other agents
more. Inequity aversion can thus reduce the principals rent payments in case of eort
implementation, and inequity aversion can then have an impact on the equilibrium
outcome.
4.4.3 Inequity Aversion and Eciency
In this section we derive the conditions under which inequity aversion causes an e-
ciency loss similar to the eciency loss that arises if risk aversion renders at wage
contracts optimal. There are however two qualitative dierences between risk agency
costs, RAC, and inequity agency costs, IAC. First, the RAC are unbounded. There-
fore, an eciency loss due to underprovision of eort always occurs if only risk aversion
is suciently large. In contrast, the IAC are bounded. It can be that no ineciency
arises even if the degree of inequity aversion goes to innity. The reason is that the
INEQUITY AVERSION AND MORAL HAZARD 139
principal can always equate wage levels in case of diverging output realizations, thereby
eliminating inequity while still providing incentives. It is however not possible to pro-
vide incentives and eliminate agents risk. Second, if the RAC are large the principal
can only oer at wage contracts to avoid the agents risk exposure. In contrast, there
are two means by which inequity can be avoided. As with risk aversion, the principal
can either oer at wage contracts thereby forgoing prots from eort implementa-
tion. We call this case underprovision of eort. Or he can employ a single agent only.
Then there is no reference group and thus no social comparisons and no suering from
inequity. The principal will then provide incentives to a single agent thereby forgoing
the prot from employing the second agent. We call this the reference group eect.
In the following we identify the conditions under which either case arises.
Underprovision of Eort
Two conditions have to be met such that inequity aversion renders at wage contracts
more protable than incentive contracts. First, the expected prot from two at wage
contracts must exceed expected prots from a single incentive contract. This condition
ensures that oering two at wage contracts is the best alternative to oering two
incentive contracts. Second, for suciently high levels of the IAC must exceed the
dierence in expected prots from two incentive contracts (without inequity aversion)
and two at wage contracts. With at wage contracts wages never diverge and inequity
aversion is irrelevant. This is summarized in the following proposition.
Proposition 4.6 (Underprovision of Eort)
If and only if x
l
B and 2 B < lim

IAC, there exists a threshold level of inequity

aversion such that for all at wage contracts maximize the principals expected
prot, even though incentive contracts are prot maximizing with selsh or unrelated
agents.
Proof: The rst condition ensures that expected prot from two at wage contracts
exceed expected prots from a single incentive contract. Formally, P
f
2
= 2x
l
x
l
+B =
P
i
1
x
l
B. Consider now the second condition. P
i
2
denotes the principals expected
INEQUITY AVERSION AND MORAL HAZARD 140
prot when oering two incentive contracts. If = 0 we have P
i
2
( = 0) = 2 P
i
1
.
By assumption, 2 P
i
1
P
f
2
= 2 B > 0. Without inequity aversion the principal thus
employs both agents and implements high eort. By Proposition 4.3, P
i
2
decreases in
and converges to
lim

P
i
2
() = 2 P
i
1
lim

IAC. (4.23)
We thus have P
f
2
> lim

P
i
2
if and only if
2 B < lim

IAC. (4.24)
From (4.17) and (4.19) we know that lim

IAC > 0. The parameter space for

which (4.24) holds is thus non-empty. If 2 B, the gain of providing incentives to two
agents, falls short of lim

IAC, the limit of the inequity agency costs of providing

incentives, there exists a unique threshold level of inequity aversion such that for
< two incentive contracts, and for two at wage contracts maximize the
principals expected prot. Existence and uniqueness of threshold is ensured since
P
i
2
is continuous and strictly decreasing in . q.e.d.
The left panel of Figure 4.1 provides an illustration of Proposition 4.6. Without
inequity aversion, = 0, expected prots from two incentive contracts exceed expected
prots from both two at wage contracts and a single incentive contract. Condition
x
l
B ensures that the principals best alternative to oering two incentive contracts
is oering two at wage contracts. As increases, the IAC increase and reduce the
principals expected prot from two incentive contracts. At the IAC equal the
dierence in expected prots between two incentive and two at wage contracts, 2B.
Therefore, for levels of inequity aversion exceeding , two at wage contracts maximize
the principals expected prot.
Proposition 4.6 is the central nding our this paper: inequity aversion can render at
wage contracts optimal even though incentive contracts are optimal with selsh agents.
We interpret this as an explanation for the observed low powered incentives within
INEQUITY AVERSION AND MORAL HAZARD 141

0
`
exp. prots
2(B+x
l
)
prot from two
incentive contracts

-
2x
l
prot from two
at wage contracts

-
B+x
l
prot from one
incentive contract
`

2B
`

IAC

Left Panel

`
0
exp. prots
2(B+x
l
)
prot from two
incentive contracts

-
B+x
l
prot from one
incentive contract

-
2x
l
prot from two
at wage contracts
`

B+x
l
`

IAC

Right Panel
Figure 4.1: Underprovision of Eort and the Reference Group Eect.
Left Panel: Expected prot levels for B < x
l
. In this case expected prots from two at
wage contracts exceed prots from one incentive contract. If the additional agency costs due
to inequity aversion, IAC, exceed the dierence in expected prots between at wage and
incentive contracts, 2B, as increases, then there exists a threshold level such that two
at wage contracts maximize the principals expected prot for .
Right Panel: Expected prot levels for x
l
< B. In this case expected prots from one
incentive contract exceed prots from two at wage contracts. If the additional agency costs
due to inequity aversion, IAC, exceed the expected prot from an additional incentive contract
absent inequity aversion, B + x
l
, as increases, then there exists a threshold level such
that a single incentive contract maximizes the principals expected prot for .
rms as compared to high powered incentives in the market. This interpretation
hinges upon the assumption that agents compare their wage payments within rms but
not within the market. Although the determinants of an agents reference group will
ultimately be an empirical question, co-workers within a rm are a natural candidate
for a reference group. However, crucial to our analysis is that there are two agents who
compare their wages and dislike inequity. Our results though not our interpretation
would hold if we assumed two principals, each of them oering an incentive contract
to a single agent, and these two agents comparing wages.
The Reference Group Eect
Suppose now that the principal can inuence an agents reference group. Two con-
ditions have to be met such that inequity aversion renders it more protable for the
principal to oer an incentive contract to a single agent than oering incentive con-
INEQUITY AVERSION AND MORAL HAZARD 142
tracts to two agents. When two inequity averse agents work for the principal they
compare their wage levels and suer from inequity. In contrast, with a single agent no
comparisons take place, and thus no IAC arise. In Section 4.5 we further explore this
reference group or rm size eect in a slightly enriched setting; for completeness we
now derive the conditions that have to be met in this basic set-up. First, the expected
prot from a single incentive contract must exceed expected prots from two at wage
contracts. This condition ensures that oering a single incentive contract is the best
alternative to oering two incentive contracts. In contrast to the previous section,
here it must hold that x
l
< B. Second, for suciently high levels of the IAC must
exceed the dierence in expected prots from oering two incentive contracts (without
inequity aversion) and expected prots from oering a single incentive contract. With
a single incentive contract inequity aversion is irrelevant as there is no reference group.
This is summarized in the following proposition.
Proposition 4.7 (Reference Group Eect)
If and only if x
l
< B and B + x
l
< lim

IAC, there exists a threshold level of

inequity aversion such that for > the principal employs a single agents only to
avoid social comparisons, even though employing both agents maximizes the principals
expected prot without inequity aversion.
Proof: As before, P
i
2
( = 0) = 2 P
i
1
such that without inequity aversion it maximizes
the principals expected prot to employ both agents and implement high eort. How-
ever, P
i
2
decreases as rises, and it may eventually fall short of P
i
1
. As P
i
1
= P
f
1
+ B
and P
f
1
= x
l
, it holds that lim

P
i
2
< P
i
1
if and only if
B + x
l
< lim

IAC. (4.25)
From (4.17) and (4.19) we know that lim

IAC > 0, so the parameter space for

which (4.25) holds true is non-empty. Whenever the base output, x
l
, and the benet
from giving incentives, B, are suciently small, there exists a unique level of inequity
aversion such that for < two incentive contracts, whereas for a single
INEQUITY AVERSION AND MORAL HAZARD 143
incentive contract maximizes the principals expected prot. Existence and uniqueness
of threshold is ensured since P
i
2
is continuous and strictly decreasing in . q.e.d.
The right panel of Figure 4.1 provides an illustration of Proposition 4.7. Without
inequity aversion, = 0, expected prots from two incentive contracts exceed expected
prots from both two at wage contracts and a single incentive contract. Condition
x
l
< B ensures that the principals best alternative to oering two incentive contracts
is oering a single incentive contract. As increases, the IAC increase and reduce
the principals expected prot from two incentive contracts but not the expected
prot from a single incentive contract as in this case no social comparisons take place.
At the IAC equal the expected prot from an additional incentive contract without
inequity aversion, x
l
+B. Therefore, for levels of inequity aversion exceeding , a single
incentive contract maximizes the principals expected prot.
In case neither x
l
B and 2 B < lim

IAC, the conditions stated in Proposition

4.6, nor x
l
< B and B + x
l
< lim

IAC, the conditions stated in Proposition 4.7,

there is no ineciency caused by the additional agency cost due to inequity aversion
even if the degree of inequity aversion goes to innity. The principal is nevertheless
harmed by inequity aversion since his expected prot is reduced by the amount of the
IAC. In contrast, the RAC will always lead to an ineciency if only the degree of risk
aversion becomes suciently large.
The eect of inequity aversion in the case with underprovision of eort is qual-
itatively similar to the eect of risk aversion. Providing incentives becomes more
expensive as either aversion becomes more pronounced, and this may render at wage
contracts optimal for the principal. However, the rm size eect is qualitatively dier-
ent from the ineciency that can arise due to risk aversion. The principal can respond
to risk aversion only by adopting an agents contract, whereas with inequity aversion
or more generally with social preferences he has an additional instrument at hand
as he can control the agents reference groups. Incorporating this nding into richer
models with, for example, heterogeneous agents with respect to the degree of inequity
aversion or productivity, or allowing for multi-tasking will yield deeper insights into
INEQUITY AVERSION AND MORAL HAZARD 144
the determinants of real world wage contracts, the optimal design of institutions, and
the boundary of the rm. In the following section, while keeping the assumption of
homogeneous agents, we enrich the model by allowing the principal to separate the
agents into dierent rm at a xed cost. We will argue that the interaction between
inequity aversion and moral hazard can contribute to the old question of the nature
and size of the rm.
4.5 The Nature and Size of the Firm
The property rights approach of the theory of the rm pioneered by Grossman
and Hart (1986) and Hart and Moore (1990) denes a rm as the physical assets
it consists of. In contrast to the transaction cost approach of the theory of the rm
(Coase (1937), Williamson (1975, 1985)), the property rights approach can explain
both, advantages and disadvantages (better incentives to invest for one, but worse
incentives for the other party) of integration within a unied framework. An optimal
degree of integration, that is, an optimal rm size can thus be determined. In this
section, we propose a new approach. We focus on one characteristic that distinguishes
the rm from the market. The rm is seen as an economic entity within which social
comparisons matter in contrast to the market in which they are negligible.
In this section we enrich our model by endowing the principal with the option to
separate the agents by setting up an additional rm. We assume that agents compare
payos only with agents that work within the same rm but not with agents that
work in distinct rms.
49
Additional agency costs due to inequity aversion can thus be
avoided by separating agents into dierent rms. If agents can be separated, that is,
if social comparisons can be prevented at not cost, the purpose of this paper dissolves.
The principal would then always separate the agents. However, we further assume
that setting up a rm involves the expense of xed costs, denoted by F. These xed
costs are taken to be suciently low such that the principal realizes positive prots
when oering an incentive contract to a single agent, that is F < P
i
1
. Alternatively,
49
See, again, Bewley (1999) for supporting evidence.
INEQUITY AVERSION AND MORAL HAZARD 145
complementarities in production could be assumed such that, absent inequity aversion,
it is advantageous to have agents work together.
The principal now faces a trade o. On the one hand, employing two agents within
a single rm economizes on xed costs (or enables the principal to realize complemen-
tarities in production). On the other hand, integrating the agents within a single rm
provokes social comparisons that increase agency costs of providing incentives. The
solution to this trade-o thus denes the optimal size of the rm, whether there is
integration of both agents within a single rm or separation of the agents into two
distinct rms. If the rm is integrated, we can have both, incentive and at wage con-
tracts. In case of separation, the principal will always oer incentive contracts. The
following proposition identies the conditions under which either regime is optimal.
Proposition 4.8 (Optimal Firm Size)
i) If and only if F min[lim

IAC, 2B] then there exists a threshold level of

inequity aversion

such that for

separation is optimal: The principal
bears xed costs F twice to set up two distinct rms, and she oers in each rm
a single incentive contract. For <

integration is optimal: The principal sets
up a single rm and oers two incentive contracts.
ii) If F > 2B then there is always integration, irrespective of the degree of inequity
aversion . If, in addition, lim

IAC 2B then integration with incentive

contracts is optimal for all . If, in addition, lim

IAC > 2B then there exists

a threshold level of inequity aversion

such that for

integration with at
wage contracts is optimal, whereas for <

integration with incentive contracts
is optimal.
Proof: i) If F 2B then it is always better to oer incentive contracts in two separated
rms than to oer two at wage contracts within a singe rm. Recall that the gain of
providing incentives is given by B per agent, while the cost of setting up a second rm
is F. Notice that in both cases the degree of inequity aversion is irrelevant. The best
alternative to oering two incentive contracts within a single rm is thus separating
INEQUITY AVERSION AND MORAL HAZARD 146
the agents in two rms but still oering incentive contracts. Integrating two agents
with incentive contracts saves on xed costs but provokes social comparisons, that is
additional agency costs IAC. If the latter exceed the rst, F IAC, then separation
becomes optimal. From Proposition 4.3 we know that the IAC rise with the degree of
inequity aversion ; at = 0 we have IAC = 0. If F lim

IAC there must thus

exist a threshold level

such that for <

we have F > IAC, i.e. integration, and
for

we have F IAC, i.e. separation.
ii) If F > 2B then, by the above arguing, the best alternative to oering two incen-
tive contracts within a single rm is oering two at wage contracts within a single rm.
Notice that it is never optimal to oer at wage contracts and to separate the agents.
Even if the IAC become very large there is thus never separation. Absent inequity aver-
sion the prot dierence between the two regimes is 2B. If lim

IAC 2B there
will thus always be integration with incentive contracts. If however lim

IAC > 2B
then, by the above arguing, there exists a threshold

such that for <

integra-
tion with incentive contracts is still optimal but for

integration with at wage
contracts becomes optimal. q.e.d.
Figure 4.2 oers an illustration of Proposition 4.8. In all cases, at = 0 the
expected prots from two integrated incentive contracts exceeds the expected prot
either from oering separated incentive contracts or oering two integrated at wage
contracts. Notice that separated at wage contracts can never be optimal. The left
panel of Figure 4.2 shows expected prot levels in case F < 2B. This condition
ensures that expected prots from two separated incentive contracts exceed prots
from two integrated at wage contracts. If the additional agency costs due to inequity
aversion, IAC, exceed the cost of separation, F, as goes to innity, then there exists
a threshold level

such that expected prots from two separated incentive contracts
exceed expected prots from two integrated incentive contracts for

.
The right panel of Figure 4.2 shows expected prot levels in case F > 2B. This
condition ensures that expected prots from two integrated at wage contracts exceed
prots from two separated incentive contracts. If the IAC exceed the expected prot
INEQUITY AVERSION AND MORAL HAZARD 147

0
`
exp. prots
2(B+x
l
)-F
two integrated
incentive contracts

-
2(B+x
l
-F)
two separated
incentive contracts

-
2x
l
-F
two integrated
at wage contracts
`

F
`

IAC

Left Panel

`
0
exp. prots
2(B+x
l
)-F
two integrated
incentive contracts

-
2x
l
-F
two integrated
at wage contracts

-
2(B+x
l
-F)
two separated
incentive contracts
`

2B
`

IAC

Right Panel
Figure 4.2: The Optimal Firm Size.
Left Panel: Expected prot levels with F < 2B. In this case expected prots from two
separated incentive contracts exceed prots from two integrated at wage contracts. If the
additional agency costs due to inequity aversion, IAC, exceed the cost of separation, F, as
increases, then there exists a threshold level

such that separated incentive contracts become
optimal for

.
Right Panel: Expected prot levels with F > 2B. In this case expected prots from
two integrated at wage contracts exceed prots from two separated incentive contracts. If
the IAC exceed the expected prot dierence between integrated incentive contracts absent
inequity aversion and integrated at wage contracts, 2B, as increases, then there exists a
threshold level

such that integrated at wage contracts become optimal for

.
dierence between two integrated incentive contracts absent inequity aversion and two
integrated at wage contracts, 2B, as goes to innity, then there exists a thresh-
old level

such that expected prots with two integrated at wage contracts exceed
expected prots with two integrated incentive contracts for

.
In this section we have argued that social comparisons can contribute to the old
question of the the optimal degree of integration. We do not claim that social com-
parisons can fully explain the size of the rm nor do we claim that they are the main
determinant. However, many situation are imaginable where an employer being in-
dierent otherwise wants to separate employees to prevent social comparisons. Even
though throughout the paper we did not model heterogeneity in agents productivity,
consider the observation that many rms outsource activities very often (but certainly
not exclusively) at the extreme ends of the productivity scale. There are often external
consultants that are, in comparison to customary wage levels within the rm, relatively
INEQUITY AVERSION AND MORAL HAZARD 148
well payed. By the same token, employees of external cleaning companies earn rela-
tively little. Outsourcing of these activities may thus at least partly be explained
by the intent to maintain a balanced wage structure within the core of the rm.
An employer may not necessarily separate employees into dierent rms but, in case
this suciently cuts down social comparisons, into dierent, say, departments of a rm.
In this case our model can contribute to the literature on the internal organization of
the rm. Consistent with our arguing is also the observation that within rms (or any
other organization) there are often many small rungs in the job ladder, all distinguished
by dierentiated job titles (junior analyst, senior analyst, junior consultant, senior
consultant, etc.). If employees tend to compare only to other employees on same rung
of the job ladder and accept that, for example, employees above them may earn
more, then separating agents into dierent job categories may be explained by the
employers intent to cut down social comparisons. The determinants of employees
relevant reference groups be it a rm, a department, a job category, or some other
attribute will ultimately be an empirical question. However, we claim that co-workers
within the same rm are a natural candidate.
4.6 Secrecy of Salaries
The central result of the paper states that inequity aversion among agents increases
agency costs. At rst sight our results could serve as an explanation for the fact that
many labor contracts impose a clause that prohibits employees from communicating
their salaries to their colleagues. If by way of secret salaries social comparisons can
be prevented, the increase in agency costs can be prevented as well. In this section we
show that this is not necessarily the case.
Suppose agents can be separated such that the other agents output realization is
not observable. Suppose further that wages do not get communicated because labor
contracts prohibit this but that the contracts themselves are common knowledge. We
maintain the assumption that the agents reference group is the respective other agent
that is employed with the same principal. (If agents can be separated in a way such
INEQUITY AVERSION AND MORAL HAZARD 149
that they do not compare themselves any longer the IAC can trivially be avoided.) We
now derive the optimal incentive contract for both agents. Even though the agents
cannot observe each others project outcome and wages, they know that their wages
dier in certain states of the world because an incentive contract must condition wages
on project realizations. In order not to transfer information about the other agents
project outcome, each agents wage can only depend on his own output realization.
Thus, there are two wage levels only. The principal therefore maximizes
P
s
2
= 2x
l
+ 2
2
[x h(u
h
)] + 2(1 )[x h(u
h
) h(u
l
)] 2(1 )
2
h(u
l
) (4.26)
with respect to u
h
, u
l
, and under the incentive and participation constraint
(IC) (1 + )(u
h
u
l
) 0
(PC) u
h
+ (1 )(u
l
(u
h
u
l
)) 0
Superscript s stands for secrecy contract. Solving the resulting rst-order conditions
yields
u

h
=

and u

l
=

1 +
. (4.27)
At = 0 wages and prot equal (twice) the single agent solution. With increasing
the low wage increases in order to reduce inequity, and the principals expected prot
falls. Dierentiating P
s
2
with respect to yields
P
s
2

=
2(1 )(1 + ( + r))
(1 + )
3
(4.28)
which is unambiguously negative. We can establish the following proposition.
Proposition 4.9 (Secrecy of Salaries)
Separating the agents such that project outcomes and wages are unobservable ampli-
es the negative eect of inequity aversion if the agents reference group remains the
respective other agent and contracts are common knowledge.
INEQUITY AVERSION AND MORAL HAZARD 150
Proof: Comparing (4.28) to (4.20) it can be seen that the secrecy prot falls faster in
than the prot in the two agents case. Subtracting (4.28) from (4.20) yields (2(1
)(1+(+r))(1+(((3+(3+)))+(2+(4+(1+2)))r))/(rk
2
(1+
)
3
) which is always positive. q.e.d.
The secrecy contract is therefore never optimal. Covering up the respective other
agents output realization and wage payments with intent to avoid social comparisons
does not mitigate but amplies the principals problem. In the secrecy contract wage
payments cannot depend on both agents output realizations since this would reveal
the respective other agents outcome realization and thus wage payment. In states with
diverging output realizations wages can therefore not be compressed as it was found
optimal in the previous sections. This restriction on the contract design renders the
secrecy contract too costly.
4.7 Discussion
4.7.1 Rent Comparison
Suppose now that agents compare rents, that is they explicitly account for eort cost
in their inequity term. Notice rst that in equilibrium both agents exert eort such
that eort terms cancel out in the PC. We must, however, reconsider the IC because
an agent now has to account for the dierence in eort costs in the inequity term when
considering to shirk. The IC can now be written as
(IC

)
2
u
hh
+ (1 )u
hl

2
u
lh
(1 )(u
hl
u
lh
)
+max [u
hl
u
lh
, 0] (1 )u
ll
0.
Subtracting the l.h.s. of (IC

) from the l.h.s. of (IC), the IC if eort costs are not

considered in the inequity term, yields
((2 )(u
hl
u
lh
) max [u
hl
u
lh
, 0]) (4.29)
INEQUITY AVERSION AND MORAL HAZARD 151
which is always positive. Hence, considering eort costs in the inequity term can only
increase agency costs thereby reinforcing our results. The intuition is straightforward.
The expected utility when shirking increases because suering from being behind is
now lower. The dierence in utility from wages is reduced by the amount of eort
costs, if not cancelled. The incentive to exert eort is thus reduced.
4.7.2 Disutility from Being Better O
In their original formulation of inequity aversion Fehr and Schmidt (1999) assume that
inequity averse individuals dislike both unfavorable and favorable inequity. In this
section we discuss the implications if suering from being better o is incorporated in
our model. It now makes a crucial dierence whether eort costs enter the comparison
or not.
Consider rst the case in which agents compare utility from wages only. Instead
of the simplied version of inequity aversion assumed in the previous sections, agents
utility function is now given by
v
i
(w
i
, w
j
) = u(w
i
) max[u(w
j
) u(w
i
), 0] max[u(w
i
) u(w
j
), 0].(4.30)
If > 0, agents suer from receiving a higher wage than the respective other agent.
Suppose the principal oers incentive contracts to both agents. Incorporating disu-
tility from being better o into our model has two eects. First, the agents PCs
are tightened. If agents are paid dierent wages in case their project outcomes dier,
an agent now also suers from inequity whenever he is fortunate whereas the other
agent is not. As this happens with positive probability agents have to be compensated.
Second, incentive provision is impaired because suering from being better o clearly
reduces the incentive to exert eort. Recall that the results in our model are driven by
the observation that the overall impact of inequity aversion on the principals prot is
negative even when neglecting the utility loss from being better o. Incorporating
this disutility adds an unambiguously negative eect and would thus only reinforce our
results.
INEQUITY AVERSION AND MORAL HAZARD 152
Consider now the case in which eort costs enter the inequity term. Again, the
PC is tightened if > 0. In equilibrium both agents exert eort and eort costs thus
cancel in the inequity term. However, eort cost enter the IC and suering from being
better o may now facilitate incentive provision. To see this, assume the most extreme
case, which is > u
hl
u
lh
. A shirking agent that saves on eort costs is then always
better o than the other agent (who works) even if the other agent receives the higher
wage in case of diverging output realizations. The IC can then be written as
(IC

)
2
u
hh
+ (1 )u
hl

2
u
lh
(1 )u
ll

(1 )(u
hl
u
lh
) + [ (2 )(u
hl
u
lh
)] 0.
The positive eect of on the IC

may be very strong. As long as x is suciently

large to ensure B 0, can become very large without violating our assumption that
incentive contracts are optimal without inequity aversion. Intuitively, if an agent shirks
he saves on eort costs and may thus be better o than the other agent who exerts
eort. If agents suer from being better o incentives to exert eort are increased. This
eect could, in principle, be so strong that agency costs are lowered in comparison to
the case without inequity aversion.
50
4.7.3 Status Seeking
In the previous section we have discussed the possibility that agents suer from being
better o than others. In contrast, suppose now that agents are status seekers, that
is they receive additional utility from being better o than others. In the context, of
this model this translates into < 0. Incorporating status seeking into our model
has two eects. First, the agents participation constraints are relaxed. Whenever
diverging project outcomes realize the successful agent receives additional utility from
being better o than the unsuccessful agent. Second, there is an positive eect on
incentives because on top of a high wage an agent receives status utility whenever he
is successful whereas the other agent is not. In summary, the unambiguously positive
50
This eect is analyzed in Bartling and von Siemens (2004).
INEQUITY AVERSION AND MORAL HAZARD 153
eect of status seeking on the principals prot opposes the negative eect of inequity
aversion that we have identied in this paper. Since there is no natural lower bound
on agency costs could, in principle, be reduced without bounds. Carrying this eect
to the extremes, status seeking would eventually result in contracts in which agents
actually pay the principal in order to be employed and sometimes receive status utility.
This is only reinforced if eort costs are considered in the inequity terms. However,
in this paper we focus on the more natural and more interesting case in which other-
regarding preferences provoke a trade-o between the positive eect on the incentive
and the negative eect on the participation constraint.
4.8 Conclusion
Recent insights from experimental economics have shown that many people are not
fully selsh but have some kind of social preferences. This, in turn, raises the question
of how other-regarding behavior interacts with incentive provision. In a moral hazard
model with risk averse agents we have shown that inequity aversion among agents
unambiguously increases agency costs unless agents compare rents and suer from
being better o. As a result, optimal contracts for inequity averse agents may be low
powered, equitable at wage contracts even when high powered incentive contracts
are optimal with selsh agents. Accounting for inequity aversion may thus oer an
explanation for the scarcity of incentive contracts many real world situation in which
veriable performance measures would be available but are not contracted upon.
More specically, assuming that social comparison are pronounced within rms
but less so in the market, we have argued that inequity aversion helps to understand
Williamsons (1985) observation that incentives oered to employees within rms are
generally low powered as compared to high powered incentive in markets.
Furthermore, we have argued that inequity aversion among agents and the resulting
increased agency costs contribute to the old question of the boundary of the rm. In an
enriched setting of the basic model, the principal could set up a second rm to separate
the agents with intent to avoid social comparisons. If this involves costs, the principal
INEQUITY AVERSION AND MORAL HAZARD 154
faces the trade-o to either bear increased agency costs or the cost of operating the
second rm. The solution to this trade-o denes an optimal size of the rm.
Incorporating our ndings into richer models with, for example, heterogeneous
agents with respect to the degree of inequity aversion or productivity, or allowing
for multi-tasking promises to yield further insights into the determinants of real world
wage contracts, the optimal design of institutions, and the boundary of the rm.
4.9 Appendix
Throughout the paper we have assumed an explicit utility function in order to obtain
simple closed form solutions. As in Fehr and Schmidt (1999) we also assumed a linear
inequity term. In this appendix we show that our results hold true for any concave
utility function and irrespective of the functional form the inequity term. To show
and illustrate the basic reasoning we, rstly, maintain the assumption that there are
only two possible output realizations. Later we will drop this restriction and allow for
arbitrary numbers of possible output realizations.
As benchmark, consider the single agent case. With only two possible outcome
realizations, wage levels are well-dened by the incentive and participation constraints.
Recall that the utility level arising from the wage payment in case of a high output
realization is given by u
h
, in analogy we dened u
l
. From
(

IC) u
h
+ (1 )u
l

u
h
+ (1

)u
l
(

PC) u
h
+ (1 )u
l
0
we thus get
u

h
=
(1

and u

l
=

. (4.31)
If now a second agent is introduced and wages are contingent on the respective other
agents output realization, each agent faces an additional lottery. Suppose an agents
outcome realization is high. If the other agent works, he will also receive a high output
realization with probability , and a low output realization with probability 1 .
INEQUITY AVERSION AND MORAL HAZARD 155
Recall that u
ij
was dened as an agents utility from wage w
ij
, if the agents output is
i and the other agents output is j. Absent inequity aversion we must have
u
hh
+ (1 )u
hl
= u

h
and u
lh
+ (1 )u
ll
= u

l
(4.32)
The inverse function h = u
1
species the wage payment that is necessary to generate
a certain utility level. The principal minimizes wage payments h(u
hh
) and h(u
hl
), and
h(u
lh
) and h(u
ll
) such that (4.32) holds. From the rst-order condition
h

(u
hh
) + (1 )h

(u
hl
)()/(1 ) = 0 (4.33)
and convexity of h() it follows that u

hh
= u

hl
= u

h
and, equivalently, u

lh
= u

ll
= u

l
.
The intuition is straightforward. The second-best utility levels that induce the agent
to exert eort are given buy u

h
and u

l
. If an agents wages depend on the other agents
output realization, in expectation he should nevertheless receive u

h
and u

l
. Incentives
are thus not aected but contingent wages introduce an additional lottery, and agents
must be compensated for the associated risk. Absent inequity aversion, wages will thus
be independent of the other agents output realization.
Consider now inequity averse agents. The second-best optimal utility levels in case
of high and low output realizations are still given by u

h
and u

l
, respectively. However,
in case of diverging output realizations there is now a utility loss arising from the
inequity. In analogy to (4.32) wage levels must now be such that
u
hh
+ (1 )(u
hl
max[u
lh
u
hl
, 0]) = u

h
, and (4.34)
(u
lh
max[u
hl
u
lh
, 0]) + (1 )u
ll
= u

l
. (4.35)
It can be seen that the cost of providing the second-best optimal utility level are weakly
increasing in the level of inequity aversion . Consider the following reasoning.
1. Fix u
hl
at some level.
2. Consider the set of (u
lh
, u
ll
) such that an agent with a low output realization
INEQUITY AVERSION AND MORAL HAZARD 156
T
E

45

u
hh
u
lh
u
ll
u
hl

l
(1)
u

h
(1)

.
(u

hh
, u

hl
)
/
/
/`
(u

lh
, u

ll
)
s
s
Figure 4.3: Inequity aversion increases agency costs.
The negatively sloped, parallel lines depict the constraints subject to which the principal
minimizes wages. Without inequity aversion, the lowest iso-cost curves that satisfy the re-
strictions are tangent where u
hh
= u
hl
and u
lh
= u
ll
. With inequity aversion, the constraints
become weakly more restrictive, depicted by the dashed lines. Utility combinations that
satisfy the constraints cannot lie on lower iso-cost curves.
receives an expected utility level of u

l
.
3. Given any u
lh
, the level of u
ll
to yield u

l
is given by
u
ll
=
u

l
u
lh
+ max[u
hl
u
lh
, 0]
1
(4.36)
4. Hence, the cost to implement u

l
weakly increases in .
The reasoning for u

h
is analogous.
Figure 4.3 illustrates the above reasoning and shows how inequity aversion tightens
the constraints subject to which the principal minimizes costs. The decreasing, parallel
lines depict combinations of u
hh
and u
hl
, and u
lh
and u
ll
that lead to expected utility
levels of u

h
and u

l
, respectively. The total dierential of (4.32) at constant utility levels
yields their slope with (1 )/. The iso-cost curves in the case without inequity
aversion are tangent at u

hh
= u

hl
= u

h
, and u

lh
= u

ll
= u

h
, as argued above. Consider
INEQUITY AVERSION AND MORAL HAZARD 157
now the case with inequity aversion. Algebraically, the combinations of u
hh
and u
hl
,
and u
lh
and u
ll
that lead to expected utility levels of u

h
and u

l
are now given by (4.34)
and (4.35), respectively. The total dierential of (4.34) while setting du
h
= 0 yields
du
hh
du
hl
=
(1 )(1 + )

, (4.37)
if we have u
lh
> u
hl
. For u
lh
u
hl
we get (1 )/. Graphically, with inequity
aversion, the dashed line depicting the combinations of u
hh
and u
hl
such that the agents
receives an expected utility level of u

h
is steeper for u
lh
> u
hl
and has the same slope
otherwise. Equivalently, the total dierential of (4.35) while setting du
l
= 0 yields
du
lh
du
ll
=
(1 )
(1 + )
, (4.38)
if we have u
lh
< u
hl
, and (1)/ otherwise. The dashed line depicting the combina-
tions of u
lh
and u
ll
such that the agents receives an expected utility level of u

l
is atter
for u
lh
< u
hl
and has the same slope otherwise. Hence, the constraints subject to which
the principals minimizes wage payments thus become (weakly) more restrictive.
The above reasoning generalizes straightforwardly to the case where one agent has
N and the other agent has M possible output realizations. Now, the solution to the
principals prot maximization problem is not determined by IC and PC alone any
longer. The principal rst derives the contract that implements each action at the
least cost. She then implements the action that maximizes her prot. Incentive and
participation constraint for agent i in case the principal wants to implement a
h
can
now be written as
(

IC)
N

n=1
M

m=1
U
i
(x
n
, x
m
)f(x
n
, x
m
[ a
h
, a
h
) 0
(

PC)
N

n=1
M

m=1
U
i
(x
n
, x
m
)[f(x
n
, x
m
[ a
h
, a
h
) f(x
n
, x
m
[ a
l
, a
h
)] 0
where f() denotes the conditional joint density function over output realizations, and
INEQUITY AVERSION AND MORAL HAZARD 158
U
i
(x
n
, x
m
) is given by
U
i
(x
n
, x
m
) = u
i
(x
n
, x
m
) max[u
i
(x
n
, x
m
) u
i
(x
n
, x
m
), 0]. (4.39)
u
i
(x
n
, x
m
) denotes the utility that arises from the wage payment in case the own output
realization is x
n
and the other agents output realization is x
m
. U
i
(x
n
, x
m
) denotes
the utility level in this case net of a possible utility loss due to suering from inequity
aversion. Equivalently for agent j. Denote by | the set of all U
i
(x
n
, x
m
) and U
j
(x
n
, x
m
)
such that (

IC) and (

PC) are binding,
| := U
i
(), U
i
() [ (

IC) and (

PC) binding. (4.40)
The principal chooses those U
i
(x
n
, x
m
) and U
j
(x
n
, x
m
) from | that minimize her cost.
The wage cost w() of providing the respective utility levels is given by
w(U
i
(x
n
, x
m
)) = h(u
i
(x
n
, x
m
)) = w
nm
. (4.41)
Recall that h() = u
1
. As can be seen from equation (4.39), for any strictly positive
level of , the utility from wage to attain any xed level of net utility U must be
higher whenever u
j
(x
n
, x
m
) > u
i
(x
n
, x
m
). Since h

() 0, the wage payment w

nm
must be higher. Hence, if the principal wants to implement the high eort choice a
h
her costs are weakly increased by inequity aversion. If the principal want to implement
a
l
, she will pay a xed wage and inequity aversion is thus irrelevant. With additional
expenses on notation, this reasoning generalizes to the cases with any nite number of
possible eort levels and more than two agents.
Bibliography
Aggarwal, R. (2000): Stabilization Activities by Underwriters after Initial Public
Oerings, Journal of Finance, 55, 10751103.
Akerlof, G. (1970): The Market for Lemmons: Quality Uncertaintiy and the Mar-
ket Mechanism, Quarterly Journal of Economics, 89, 488500.
Allen, F., and G. Faulhaber (1989): Signalling by Underpricing in the IPO
Market, Journal of Financial Economics, 23, 303323.
Bartling, B., and F. von Siemens (2004): Eciency in Team Production with
Inequity Averse Agents, Discussion Paper, University of Munich.
Benveniste, L., W. Busaba, and W. Wilhelm Jr. (1996): Price Stabilization
as a Bonding Mechanism in Equity Issues, Journal of Financial Economics, 42,
223255.
Benveniste, L., and P. A. Spindt (1989): How Investment Bankers Determine
the Oer Price and Allocation of Initial Public Oerings, Journal of Financial
Economics, 24, 343362.
Benzion, U., A. Rapoport, and J. Yagil (1989): Discount Rates Inferred from
Decisions: An Experimental Study, Management Science, 35, 27084.
Bewley, T. F. (1999): Why Wages Dont Fall During a Recession. Harvard University
Press, Cambridge, MA.
Boehmer, E., and R. Fishe (2001): Who ends up short from Underwriter Short
Covering A detailed Analysis of IPO Price Stabilization, working paper, NYSE.
BIBLIOGRAPHY 160
Bolton, G. E., and A. Ockenfels (2000): ERC: A Theory of Equity, Reciprocity,
and Competition, American Economic Review, 90, 166193.
Booth, J. R., and R. L. Smith (1986): Capital Raising, Underwriting and the
Certication Hypothesis, Journal of Financial Economics, 15, 261281.
Busaba, W. Y., L. M. Benveniste, and R.-J. Guo (2001): The Option to
Withdraw IPOs during the Premarket: Empirical Analysis, Journal of Financial
Economics, 60, 73102.
Camerer, C. F. (2003): Behavioral Game Theory. Princeton University Press,
Princeton and Oxford.
Camerer, C. F., G. Loewenstein, and M. Rabin (eds.) (2003): Advances in
Behavioral Economics. Princeton University Press, Princeton and Oxford.
CESR(2002): Stabilisation and Allotment, a European Supervisory Approach, Con-
sultative paper, cesr 02/020b, The Committee of European Securities Regulators.
Chen, H.-C., and J. Ritter (2000): The seven percent solution, Journal of Fi-
nance, 55, 11051131.
Cho, I.-K., and D. Kreps (1987): Signaling Games and Stable Equilibria, Quar-
terly Journal of Economics, 102, 179222.
Chowdhry, B., and V. Nanda (1996): Stabilization, Syndication, and Pricing of
IPOs, Journal of Financial and Quantitative Analysis, 31, 2542.
Coase, R. (1937): The Nature of the Firm, Economica, 4, 386405.
DellaVigna, S., and U. Malmendier (2003): Overestimating Self-Control: Evi-
dence from the Health Club Industry, mimeo, Stanford University.
(2004): Contract Design and Self-Control: Theory and Evidence, Quarterly
Journal of Economics, forthcoming.
BIBLIOGRAPHY 161
Demougin, D., and C. Fluet (2003): Group vs. Individual Performance Pay When
Workers Are Envious, CIRPE

E Working Paper 03-18.

Dunbar, C. G. (2000): Factors Aecting Investment Bank Initial Public Oering
Market Share, Journal of Financial Economics, 55, 341.
Dur, R., and A. Glaser (2004): Optimal Incentive Contracts When Workers Envy
Their Boss, Tinbergen Institute Discussion Paper 2004-046/1.
Ellis, K., R. Michaely, and M. OHara (2000): When the Underwriter Is the
Market Maker: An Examination of Trading in the IPO Aftermarket, Journal of
Finance, 55, 10391074.
Englmaier, F., and A. Wambach (2003): Contracts and Inequity Aversion,
CESifo Working Paper No. 809.
Fehr, E., and S. G

achter (2002): Do Incentive Contracts Crowd Out Voluntary

Cooperation?, Institute for Empirical Research in Economics Working Paper No.
34, University of Zurich.
Fehr, E., and K. Schmidt (1999): A Theory of Fairness, Competition, and Coop-
eration, Quarterly Journal of Economics, 114, 81768.
(2003): Theories of Fairness and Reciprocity - Evidence and Economic Ap-
plications, in Advances in Economics and Econometrics - 8th World Congress, ed.
by M. Dewatripont, L. Hansen, and S. Turnovsky, pp. 20857. Cambridge University
Press.
Francis, B. B., and I. Hasan (2001): The Underpricing of Venture and Nonventure
Capital IPOs: An Empirical Investigation, Journal of Financial Services Research,
19, 99113.
Frederick, S., G. Loewenstein, and T. ODonoghue (2002): Time Discount-
ing and Time Preference: A Critical Review, Journal of Economic Literature, 40,
351401.
BIBLIOGRAPHY 162
Frey, B. S., and R. Jegen (2001): Motivation Crowding Theory, Journal of
Economic Surveys, 15, 589611.
Fudenberg, D., and J. Tirole (1991): Game Theory. MIT-Press, Cambridge,
Massachusetts.
Gneezy, U., and A. Rustichini (2000a): A Fine is a Price, Journal of Legal
Studies, 29, 117.
(2000b): Pay Enough or Dont Pay at All, Quarterly Journal of Economics,
115, 791810.
Grinblatt, M., and C. Y. Hwang (1989): Signalling and the Pricing of New
Issues, Journal of Finance, 44, 393420.
Grossman, S., and O. Hart (1983): An Analysis of the Principal-Agent Problem,
Econometrica, 51, 745.
(1986): The Costs and Benets of Ownership: A Theory of Vertical and
Lateral Integration, Journal of Political Economy, 94, 691719.
Gul, F., and W. Pesendorfer (2001): Temptation and Self-Control, Economet-
rica, 69, 140335.
Hansen, R. S. (2001): Do Investment Banks compete in IPOs?: The Advent of the
7% Plus Contract, Journal of Financial Economics, 59, 313346.
Hart, O., and J. Moore (1990): Property Rights and the Nature of the Firm,
Journal of Political Economy, 98, 11191158.
Helwege, J., and N. Liang (1996): Is There a Pecking Order? Evidence from a
Panel of IPO Firms, Journal of Financial Economics, 40, 429458.
Holmstr

om, B. (1979): Moral Hazard and Observability, Bell Journal of Eco-

nomics, 10, 7491.
(1982): Moral Hazard in Teams, Bell Journal of Economics, 13, 32440.
BIBLIOGRAPHY 163
Holmstr

om, B., and P. Milgrom (1987): Aggregation and Linearity in the Pro-
vision of Intertemporal Incentives, Econometrica, 55, 30328.
(1991): Multitask Principal-Agent Analyses: Incentive Contracts, Asset
Ownership, and Job Design, Journal of Law, Economics, and Organization, 7, 24
52.
Itoh, H. (2003): Moral Hazard and Other-Regarding Preferences, Discussion Paper,
Hitotsubashi University.
Jenkinson, T., and A. Ljungqvist (2001): Going Public: The Theory and Evidence
on How Companies Raise Equity Finance. Oxford University Press, (2nd edition).
Kirby, K. N. (1997): Bidding on the Future: Evidence Against Normative Dis-
counting of Delayed Rewards, Journal of Experimental Psychology, 126, 5470.
Kirby, K. N., and R. J. Herrnstein (1995): Preference Reversals Due to Myopic
Discounting of Delayed Reward, Psychological Science, 6, 8389.
Laibson, D. (1997): Golden Eggs and Hyperbolic Discounting, Quarterly Journal
of Economics, 112, 44377.
Madrian, B. C., and D. F. Shea (2001): The Power of Suggestion: Inertia in
401(k) Participation and Savings Behavior, Quarterly Journal of Economics, 116,
1149 1187.
Megginson, W. L., and K. A. Weiss (1991): Venture Capitalist Certication in
Initial Public Oerings, Journal of Finance, 46, 879903.
Mookherjee, D. (1984): Optimal Incentive Schemes with Many Agents, Review
of Economic Studies, 51, 433446.
Nanda, V., and Y. Yun (1997): Reputation and Financial Intermediation: An
Empirical Investigation of the Impact of IPO Mispricing on Underwriter Market
Value, Journal of Financial Intermediation, 6, 3963.
BIBLIOGRAPHY 164
Neilson, W. S., and J. Stowe (2003): Incentive Pay for Other-Regarding Work-
ers, Discussion Paper, Texas A&M University.
ODonoghue, T., and M. Rabin (1999a): Doing It Now or Later, American
Economic Review, 89, 103124.
(1999b): Incentives for Procrastinators, Quarterly Journal of Economics,
114, 769816.
(1999c): Procrastinaton in Preparing for Retirement, in Behavioral Dimen-
sions of Retirement Economics, ed. by H. Aaron, pp. 125156. Brookings Institution
Press and Russell Sage Foundation.
(2001): Choice and Procrastination, Quarterly Journal of Economics, 116,
121160.
Phelps, E. S., and R. A. Pollack (1968): On Second-Best National Saving and
Game-Equilibrium Growth, Review of Economic Studies, 35, 18599.
Pollack, R. A. (1968): Consistent Planning, Review of Economic Studies, 35,
20108.
Rabin, M. (2002): A Perspective on Psychology and Economics, European Eco-
nomic Review, 46, 657685.
Rey Biel, P. (2003): Inequity Aversion and Team Incentives, Discussion Paper,
University College London.
Ritter, J., and I. Welch (2002): A Review of IPO Activity, Pricing, and Alloca-
tions, Journal of Finance, 57, 17951828.
Rock, K. (1986): Why New Issues Are Underpriced, Journal of Financial Eco-
nomics, 15, 187212.
SEC (1997): Regulation M, Release no. 34-38067, Securities and Exchange Com-
mission (SEC).
BIBLIOGRAPHY 165
Spence, M. (1973): Job Market Signaling, Quarterly Journal of Economics, 87,
35574.
Strotz, R. H. (1956): Myopia and Inconsistency in Dynamic Utility Maximization,
Review of Economic Studies, 23, 16580.
Titmuss, R. M. (1970): The Gift Relationship. Allen and Unwin, London.
Welch, I. (1989): Seasoned Oerings, Imitation Costs, and the Underpricing of
Initial Public Oerings, Journal of Finance, 44, 421449.
Williamson, O. (1975): The Economic Institutions of Capitalism: Firms, Markets,
and Relational Contracting. Free Press, New York.
(1985): The Economic Institutions of Capitalism. Free Press, New York.
Eidesstattliche Versicherung
Ich versichere hiermit eidesstattlich, dass ich die vorliegende Arbeit selbstandig und
ohne fremde Hilfe verfasst habe. Die aus fremden Quellen direkt oder indirekt uber-
nommenen Gedanken sowie mir gegebene Anregungen sind als solche kenntlich gemacht.
Die Arbeit wurde bisher keiner anderen Pr ufungsbehorde vorgelegt und auch noch nicht
ver oentlicht.
M unchen, den 5. Marz 2004
Lebenslauf
22. Mai 1974 Geboren in Herford
1993 Abitur am Friedrichs-Gymnasium Herford
1993/94 Zivildienst in Krankenhausern auf Norderney
und in M unchen
1994-1997 Studium der Volkswirtschaftslehre an der
Ludwig-Maximilians-Universitat M unchen
1997/98 Masters Course in Economics an der London
School of Economics and Political Science
1998-2000 Wissenschaftlicher Mitarbeiter am Seminar
f ur Wirtschaftstheorie der LMU M unchen
2000/01 Marie Curie Fellow an der Faculty of Economics
and Politics der Cambridge University
2001-2004 Wissenschaftlicher Mitarbeiter am Seminar
f ur Wirtschaftstheorie der LMU M unchen
21. Juli 2004 Promotion zum Dr. oec. publ.
2004/05 Post-Doctoral Fellow am Department of
Economics der Harvard University