. 9
( 18)


Shelanski and Klein 1995; Crocker and Masten 1996). For present purposes, two results
are of particular note. First, the empirical literature reveals a consistent preference for
integration over contracting as the specificity of investments increases. Thus, whereas
asset specificity favors contracting when the alternative is simple exchange, contracting
becomes less attractive as a way of protecting reliance or relationship-specific
investments where the alternative to contracting is integrated ownership and production.
Contracting thus appears to be only an imperfect response to the hazards posed by
relationship-specific investments. Second, the evidence indicates that uncertainty and
complexity also diminish the attractiveness of contracting relative to integration (e.g.
Anderson and Schmittlein 1984; Masten 1984). Together with the evidence that
uncertainty and complexity discourages contracting relative to simple exchange, these
findings reinforce the conclusion that contracts are a costly and inflexible way to provide
for future adaptations.
Models of vertical integration within the complete contract framework such as Crocker
(1983) differentiate contracting from integration through the deus ex machina of eliminating
information asymmetries upon integration.

On the potentially testable implications of incomplete contract theory with respect to
ownership (as opposed to integration) and their relation to the empirical literature, see
Whinston (2000).

The empirical literature on franchise contracting versus company ownership also generally
shows that the larger the required initial investment of agents the more likely are outlets to be
integrated. Depending on whether initial investment is regarded as a proxy for risk or for
agent effort, this result may or may not be consistent with the predictions of agency theory
(see Lafontaine and Slade 2000). To the extent that initial investment is correlated with the
size of specific investments, the finding could also be interpreted as supporting transaction
cost predictions (see Bercovitz 1999).

Lafontaine and Slade (2000) review these arguments and other evidence. Although
Holmström and Milgrom (1991) frame the problem in agency terms, the effects of
measurement costs on contracting and integration decisions has long been part of the
transaction-cost literature (see, e.g., Barzel 1982). Several of the relevant empirical studies
also describe the problem in transaction-cost terms.
5 Some cautionary notes
The relative contributions of agency theory and TCE in explaining observed contracting
practices derives in some degree to differences in methodology. Agency theorists, with
their emphasis on axiomatic deduction, have been hesitant to incorporate into their
models constraints, such as bounds on cognitive ability, that cannot be easily modeled
(see, e.g., Hart 1990, 1995). Transaction-cost economists working in the tradition of
Ronald Coase and Oliver Williamson, by contrast, have sought to develop and refine
theory guided more by specific phenomena or puzzles than by the susceptibility of the
theory to mathematical modeling. Although both have their place in the evolution of
knowledge, it is not surprising in light of this difference that TCE - and, to a lesser extent,
the more empirically oriented linear agency models - has turned out to have had more
success empirically than the more mathematically elegant but ethereal complete and
incomplete contract theories.

The term "success" can be applied only relatively, however; "tentative progress" would
be a more apt description. Though not specific to empirical research on contracting, a
variety of issues should temper our confidence in the findings to date. Probably chief
among those is the quality of proxies used for the explanatory variables identified by the
theory. Often, these proxies are crude and imprecise stand-ins for the variables of true
interest or are endogenous themselves. Strictly speaking, the specificity of assets and
the level of investment in those assets, which are treated as exogenous variables in
much of the research, are themselves decision variables. The location of facilities, the
adoption of specialized designs or equipment, and the scale of investments, all of which
have been treated as exogenous in the literature, should, by rights, be treated as
endogenous variables. Only a few studies have made tentative steps in that direction
(see, e.g., Lyons 1995; Saussier 1999).

Another limitation of the existing research has been its tendency to analyze the individual
provisions from complexcontracts separately. Although focusing on individual contract
terms has facilitated statistical analysis of the role of such terms, it has done so at the
expense of ignoring potentially important interactions with and qualifications by other
contract provisions that can radically alter or even negate their nominal meaning (see
Masten 1998). Given that contract provisions will have been chosen simultaneously and
are likely to interact with one another - often, as Goldberg and Erickson (1987) note, in
subtle and unexpected ways - empirical contracting studies should, ideally, estimate the
full set of contract provisions simultaneously. The econometric tools to handle such
interactions and qualifications exist; Joskow (1987) and Crocker and Masten (1991), for
example, have analyzed interactions among contract terms using standard simultaneous
equation techniques.[17] The binding constraint is not technique but data availability. As
the number of provisions analyzed increases, the number of explanatory variables and
the size of the data set needed for statistical identification multiplies. Often, sufficient
numbers of observations to analyze more than two or three provisions at a time will
simply not exist. But even where the population is sufficiently large for statistical
confidence, inadequacies in the scope and quality of the data that can be obtained on a
large scale can temper conceptual confidence in statistical results.
For these reasons, case studies are an important, indeed necessary, complement to
econometric analysis. Although case studies are often (justifiably) disparaged on the
grounds that they lack generality and invite ex post rationalization, such concerns must
be weighed against the aforementioned limitations of statistical analyses and the richer
perspective that high-quality case studies can offer. What case studies lack in generality
they often make up in depth. Data and measurement problems that would cripple
econometric analyses often yield to intensive scrutiny of a single or small number of
cases. And while a case study cannot disprove the general validity of a theory, a single,
well-documented fact can refute the applicability of a theory to a particular case.
Moreover, puzzles and anomalies encountered in case studies can and often have been
the stimulus to refinements in the theory. Finally, some cases - the contracts between
Microsoft and computer equipment manufacturers, for example - are important enough in
their own right to warrant intensive analysis.
To compensate for lack of generality, a good case study will seek to account for a more
complete range of details in addition to exploiting whatever variation exists over time and
across transactions. The transaction-cost literature contains a number of excellent case
studies that satisfy these but not the criteria for statistical confidence. Examples include
Stuckey's (1983) analysis of organizational arrangements in the aluminum industry;
Palay's (1984) work on rail-freight contracts; Gallick's (1984) analysis of the relations
between tuna harvesters and processors; Joskow's (1985) preliminary exploration of
vertical relations between coal mines and electric utilities; Goldberg and Erickson's (1987)
study of petroleum coke contracts; Masten and Snyder's (1993) analysis of United Shoe
Machinery Corp.'s lease terms; Pirrong's (1993) analysis of variations in ocean shipping
contracts; Kaufman and Lafontaine's (1994) calculation of economic rents earned by
McDonald's franchisees; and M©nard's (1996) investigation of organizational
arrangements in the French poultry industry. What distinguishes these studies is their
success in explaining the consistency among and variations in contractual details using a
limited set of simple provisions. Such thoroughly researched and carefully argued case
studies provide an important check that, in abstracting away from contract complexity to
accommodate data limitations, econometric analyses do not misconstrue the purpose
and function of particular terms.
Research on contracting has already begun to influence how courts think about
contracting and resolve contract disputes (see, for instance, PSI Energy v. Exxon Coal,
USA, 991 F.2d 1265 [1993]) and has normative implications for anti-trust policy and
business decision-making as well. It is important, therefore, that positive theories of
contracting behavior stand on as solid an empirical footing as possible. Although
tensions are likely to persist between those who value axiomatic rigor and those willing
to invoke empirical regularities to develop testable predictions, there are indications that
agency theory and transaction-cost approaches to problems of contracting and
organization are converging (cf. Tirole 1999). If that happens, the reality check provided
by empirical research on contracting is likely to have played a significant role in that

Bercovitz (1999) and Brickley (1999) also analyze multiple provisions but do not
systematically analyze possible interactions among them.
Chapter 16 was originally published as "Econometrics of Contracts: An Assessment of
Developments in the Empirical Literature on Contracting," in Revue d'Economie
Industrielle (92, 2000).
1. Reviews of the agency literature can be found in Hart and Holmström
(1987) and Furubotn and Richter (1997), among other sources. For
purposes of this chapter, we include under the heading "agency theory"
complete contract theory (in the tradition of Myerson 1982), incomplete
contract theory (such as Grossman and Hart 1986), and linear contract
theory, the latter consisting of the set of models that restrict consideration
to linear sharing rules (see, e.g., Allen and Lueck 1999 and Lafontaine
and Slade 2000). See Masten 2000 for a discussion of the relation
among these models.
2. For a more detailed discussion of the problems of identifying the
efficiency of alternative governance arrangements, see Masten, Meehan,
and Snyder (1991).
3. Potential differences in the set of attributes that affect efficiency under
alternative governance arrangements are captured in the model by the
possibility that the estimated marginal effects of particular attributes
equal zero.
4. This correspondence between the discrete choice framework and
transaction-cost hypotheses was first outlined in Masten (1982, 1984).
5. Saussier had access to the full population of the contracts written over
the period covered by his study and therefore did not face the censoring
problem present in Joskow's and Crocker and Masten's studies.
Saussier's study, however, involved a smaller number of contracts
(twenty-nine or seventy, depending on the specification).
6. Acknowledging that much of the equipment franchisees use is
redeployable, Bercovitz includes only 10 percent of equipment
expenditures in this figure.
7. Compare Goldberg and Erickson (1987, p. 398).
8. See Lafontaine and Slade (2000). Empirically, contracts often do contain
only fixed or variable payments, not both, a fact that suggests
discontinuities in how the terms operate.
9. While sympathizing with the view that individuals are not capable of
dealing with unlimited complexity, purists complain that, in the absence
of an accepted model of bounded rationality, restrictions on feasible
contract forms are unavoidably arbitrary and ad hoc (e.g. Tirole 1994, pp.
15-17; Hart and Holmström 1987, pp. 133, 148).
10. Crocker and Masten's (1988) finding that distortions in the size of take-
orpay provisions significantly reduced the willingness of parties to
engage in long-term contracting offered further support for the incentive
interpretation of take-or-pay provisions. Case studies describing the use
of minimum purchase requirements for coal (Carney 1978), petroleum
coke (Goldberg and Erickson 1987), and bauxite (Stuckey 1983), among
other products, also corroborate this finding (see Masten 1988, pp. 91-2,
for a discussion).
11. Compare, for example, Shavell's (1984) theoretical development of
efficient breach analysis with the characterization of the optimal take-or-
pay provision in Masten and Crocker (1985).
12. In particular, Brickley (1999) interprets his results as being consistent
with the multitask agency model of Holmström and Milgrom (1991, p.
13. Models of vertical integration within the complete contract framework
such as Crocker (1983) differentiate contracting from integration through
the deus ex machina of eliminating information asymmetries upon
14. On the potentially testable implications of incomplete contract theory with
respect to ownership (as opposed to integration) and their relation to the
empirical literature, see Whinston (2000).
15. The empirical literature on franchise contracting versus company
ownership also generally shows that the larger the required initial
investment of agents the more likely are outlets to be integrated.
Depending on whether initial investment is regarded as a proxy for risk or
for agent effort, this result may or may not be consistent with the
predictions of agency theory (see Lafontaine and Slade 2000). To the
extent that initial investment is correlated with the size of specific
investments, the finding could also be interpreted as supporting
transaction cost predictions (see Bercovitz 1999).
16. Lafontaine and Slade (2000) review these arguments and other evidence.
Although Holmström and Milgrom (1991) frame the problem in agency
terms, the effects of measurement costs on contracting and integration
decisions has long been part of the transaction-cost literature (see, e.g.,
Barzel 1982). Several of the relevant empirical studies also describe the
problem in transaction-cost terms.
17. Bercovitz (1999) and Brickley (1999) also analyze multiple provisions but
do not systematically analyze possible interactions among them.
Experiments on Moral Hazard and
Chapter 17:

Incentives-Reciprocity and Surplus-Sharing
Claudia Keser, Marc Willinger
1 Introduction
In the standard principal-agent model with moral hazard (Holmström 1979; Grossman
and Hart 1987) the principal, who cannot observe the agent's effort, generally has an
interest in proposing a contract with a variable remuneration that is a function of the
realized profit. The model is based on the assumption of a stochastic relationship
between the realized profit and the agent's effort; this relationship is common knowledge.
As the agent's effort is unobservable to the principal, the contract has to provide an
incentive for the agent to choose the effort level that is desired by the principal. In other
words, the contract has to satisfy an incentive constraint. It also has to provide the agent
with an expected utility that is as least as high as his utility without the contract. This is
called the participation constraint. If the principal offered a contract with a fixed
remuneration that is independent of the realized profit, the agent would provide the effort
level that is least costly to him, which is in general the lowest possible effort. If the
principal wants to induce a higher and more costly effort level by the agent, the contract
has to be designed such that the agent maximizes his expected utility by choosing this
effort level. Contract theory predicts that the principal keeps the entire expected surplus
of the contract for himself and makes the agent just indifferent between rejecting and
accepting the contract with the provision of the induced effort level.

This solution is based on the assumption that the stochastic relationship between the
principal's profit and the agent's effort is common knowledge. Moreover, the principal has
to know the agent's utility function in order to satisfy the incentive and participation
constraints. Owing to the complexity of real phenomena and the presence of many
sources of uncertainty the principal can, in reality, not base his contract policy on a given
stochastic relationship between profit and effort. Often, the agent himself does not know
this relationship, or, he has a different perception of it than the principal. Furthermore, in
practice, the principal does not know the agent's utility function. Thus, real contracts are
often incomplete and do not always satisfy incentive constraints. Given these difficulties,
empirical verification of the accuracy of the theoretical predictions in a textbook principal-
agent situation with moral hazard is problematic. Fehr, G¤chter and Kirchsteiger (1997)
argue that real contracts are often more equitable than in theory where the principal
keeps the entire expected surplus of the contract for himself. Owing to a norm of
reciprocity, the principal might be inclined to propose a positive share of surplus to the
agent who, in return, will provide a higher effort than would be imposed by the incentive
constraint (Akerlof 1982; Akerlof and Yellen 1990; Fehr, Kirchsteiger and Riedl 1993). A
second verification problem, thus, results: if the real contracts are different from the
predicted contracts by principal-agent theory, is this due to the incomplete information of
the stochastic relationship or to a norm of reciprocity among the contracting parties?
Experimentation in a laboratory implies the creation of an environment that allows us to
study the relationship between a principal and an agent in an artificial framework that
largely satisfies the assumptions of the theoretical model. For example, the assumption
that the stochastic relationship between the principal's profit and the agent's effort is
known to both contracting parties can easily be implemented in such an experimental
framework. Unfortunately, the non-observability of the participants' preferences remains
a problem in experimental studies. In particular, it is difficult, if not impossible, to induce
risk neutrality for those participants in an experiment who are assigned the role of
principals and risk aversion for those who are assigned the role of agents. The efforts
that Berg et al. (1992) and Epstein (1992) made in these directions lead to results that
remain debatable. In spite of this difficulty, experimentation in the laboratory presents a
great tool for testing the predictions of principal-agent theory.
Relatively few attempts have been made to date for testing principal- agent relationships
with the help of experimental methods. The first principal-agent experiments that we are
aware of are by Berg et al. (1992) and Epstein (1992). They produced results that were
more or less compatible with the theoretical predictions. However, the principal was
restricted in these experiments to the choice among an extremely limited set of
predetermined contracts. The more recent experiments by Güth et al. (1998), Anderhub,
G¤chter and Königstein (1999), and Keser and Willinger (2000, 2001) allow for a much
larger set of contracts. Their results generally contradict the predictions by contract
The participants in an experiment are assigned the role of either a principal or an agent
for the entire duration of the experiment. An experiment generally consists of several
periods in order to allow the participants to become familiar with the strategic
environment. We distinguish between two broad categories of experiments: those where
the interaction between a principal and an agent is repeated in all periods by the same
two participants and those where new pairs are (randomly) matched in each period. We
denote the first category as experiments with repeated interaction and the second
category as experiments with one-shot interaction between any two participants. Note
that in one-shot experiments with random matching the probability of the same agent
being matched with the same principal is negligibly small. We distinguish between
repeated and one-shot experiments as they generally imply different theoretical
predictions. When a contract is effectuated between a principal and an agent, each of
the two participants realizes gains (or losses) that depend on the contract terms, the
effort chosen by the agent and the result of a lottery. The gains are expressed in a fictive
unit, the experimental currency unit. The gains are cumulated during the experiment and
at the end of the experimental session converted into the monetary currency of the
country to be paid to the participants.
The results of most of the experiments discussed in this chapter show that the
participants' behavior is based on decision principles that are radically different from the
principles on which the principal-agent theory is built. In one-shot interaction experiments
(Keser and Willinger 2000) we observe that the principals propose contracts that are
generally more favorable to the agents than predicted by contract theory and that often
do not satisfy the incentive constraint. In particular, most of the contracts offer an
assurance of no loss to the agent. The agents, generally, react in the predicted way to
the incentives provided by the principal. Also in repeated interaction experiments (Güth
et al. 1998; Anderhub, G¤chter and Königstein 1999) we observe that the contracts
offered by the principals rarely satisfy the incentive constraint. The participants' behavior
seems to be guided by the principle of reciprocity: in general, participants in the role of
agents provide higher and more costly efforts in response to more favorable contracts
(see also Fehr, G¤chter and Kirchsteiger 1997). These observations are in keeping with
the concepts developed by Akerlof (1982) and Akerlof and Yellen (1990) with respect to
equity considerations in employer-employee relationships.
In section 2 of this chapter we briefly survey the first principal-agent experiments that are
characterized by the fact that their authors intended to induce specific preferences. We
also discuss the methodological difficulties with these experiments. Section 3 presents
the results of a series of experiments with one-shot interaction between a principal and
an agent. Section 4 presents experiments where the same principal and the same agent
interact repeatedly. Section 5 concludes.
2 Experiments with a mechanism to induce preferences
Experimental research on principal-agent relationship is relatively recent. The two
experiments presented in this section (Berg et al. 1992; Epstein 1992) tried to control for
the participants' risk attitude by using a mechanism of random remuneration.

2.1 Design and theoretical predictions
In the experiment by Berg et al. (1992) the principal may realize one of two profit levels,
x1 and x2 (with x2 > x1), through the contractual relationship. The probabilities of these
two profit levels depend on the agent's effort level, which can either be low, e1, or high,
e2. The higher profit, x2, is more likely with the high than with the low effort, i.e. 1 >
p(x2/e2) > p(x2/e1) > 0. The principal may propose a contract, defined as the pair (w1,w2)
where wi (with i = 1, 2) is the agent's remuneration that corresponds to a profit xi of the
principal. The principal has the choice among three contracts that exactly meet the
agent's participation constraint: a contract with fixed remuneration that would be the
optimal contract if the principal could enforce the high effort choice by the agent and two
contracts with remunerations as a function of the realized profits. Both of the latter two
contracts are designed to induce high effort, but one of them is more favorable (less
costly) to the principal than the other. In the experiment, the effect of two treatment
variables on the principals' contract choice and the agents' effort is tested. The two
treatment variables are observability of the effort and the participants' experience in the
role of an agent. In the treatments with effort observability the principal can enforce the
choice of the high effort by imposing a high penalty if the agent defects from this effort.
There is no moral-hazard problem in these treatments. In the treatments with acquisition
of experience in the role of an agent previous to the actual play, all participants,
whatever their role in the actual play will be, participate in a preliminary phase where
they interact as agents with the experimenter. In this preliminary phase, each participant
makes twelve effort choices, four for each of the three potential contracts. The idea is to
test the sensitivity of behavior to the way in which common knowledge about the
environment is conveyed to the participants in an experiment, an issue that had been
raised by Binmore, Shaked and Sutton (1988) and Neelin, Sonnenschein and Spiegel
(1988). Given a 2 — 2 treatment design, the experiment by Berg et al. (1992) consists of
four treatments, resulting from a combination of the two observability conditions (with or
without) and the two experience conditions (with or without). In all treatments each
participant kept the same role until the end of the experiment (after the preliminary phase
if there was one) and each participant knew that his interaction was always with the
same other participant. The experiment consisted of either ten periods after the
preliminary phase (with twelve effort choices in the interaction with the experimenter) or
of twenty-two periods, of which only the final ten periods were analyzed, in the
treatments without the preliminary phase.
Berg et al. (1992) aimed at inducing the participants' preferences by using a payment
procedure in probability points, which had been introduced by Roth and Malouf (1979).
This procedure is based on the conversion of the gain points accumulated over the
experiment into the probability to win the high outcome in a binary lottery at the end of
the experiment. The maximization of probability points is in this case the best strategy,
independent of the attitude towards risk. By taking a transformation function from points
into probabilities, which is linear for the principal and concave for the agent, it is claimed
that one can induce preferences which satisfy the risk neutrality assumption for the
principal and the risk aversion assumption for the agent.

In the situation without moral hazard (i.e. with effort observability), there exists a unique
sub-game perfect equilibrium in which the principal proposes the risk-free contract with a
fixed remuneration. The agent provides the high effort level in order to avoid the penalty.
In the situation with moral hazard, the sub-game perfect equilibrium prescribes that the
principal proposes the less costly of the two contracts where remuneration is a function
of the realized profits. The agent should provide the high effort level.

2.2 Results and limitations

The observed contract and effort choices of fourteen, resp. sixteen, pairs of students
participating in each of the four treatments are in keeping with the sub-game perfect
equilibrium prediction both in the situation with and without moral hazard. The
participants' experience in the role of an agent did not modify the behavior of the
principals compared to the situation without previous experience.

These observations seem to support the major theoretical predictions of the principal-
agent model with hidden action. However, this experiment suffers from three major
weaknesses. The first weakness is that the principal's strategy space is limited to three
options, which correspond to contracts for which the agent's participation constraint is
binding in the case of the high effort. Among these three contract options are the optimal
contract without moral hazard and the optimal contract with moral hazard. The difficulty
with this particular set of contracts is that the principal is constrained to the appropriation
of the entire surplus, which is part of the sub-game perfect equilibrium prediction. Thus,
we are limited to the examination of risk-sharing between the principal and the agent,
given that the agent is just at his participation constraint. A second problem results from
the procedure to remunerate the participants in probability points. Selten, Sadrieh and
Abbink (1999) show that this type of procedure, being far away from neutralizing the
effects of risk aversion, can produce important biases in the participant's behavior. The
third weakness is that the agents don't have the opportunity to refuse the proposed
contract. This incurs problems for testing the predictions of contract theory relying on the
assumption that the agent might reject a contract offer. Although all of the potential
contracts satisfy the participation constraint, the agents who are forced to accept all
contract offers do not have the opportunity for strategic rejection with the hope of
obtaining a more favorable contract offer in the following period.

2.3Taking the agent's potential rejection into account
With respect to the experiment by Berg et al. (1992), the major modification made by
Epstein (1992) in his experiment is to introduce an explicit reservation utility by allowing
the agent to refuse the contract. Epstein considers the situation with both moral hazard
and experience in the role of an agent and compares his results to those of Berg et al.
(1992). In his experiment, he uses an explicit reservation value of the agent in the case
that he rejects the contract offer as a treatment variable: the agent's reservation value is
either low and not binding or high and binding. The principal can, again, choose among
three contracts, of which one corresponds to the unique sub-game perfect equilibrium of
the ten-fold repetition of the game. Offered this contract, the agent should choose the
high effort level. A second of the potential contracts consists of offering a fixed
remuneration to the agent. In the treatment with a low explicit reservation utility, applying
backward induction, the agent should never reject, whatever contract is offered to him. In
other words, the explicit reservation utility is not binding. The third potential contract in
this treatment is one that induces high effort but yields the principal a lower expected
utility than the sub-game perfect equilibrium contract. In the treatment where the explicit
reservation value is high, it is binding. The agent should reject the third contract if it is
offered to him. Without the opportunity for the agent to reject, however, this contract
would induce high effort and would maximize the principal's expected utility. Note that in
both treatments the presence of the option to refuse the contract offer allows the agent to
refuse an offer for strategic reasons with the expectation that the principal offers him a
more favorable contract in the following period.
Six pairs of students participated in the treatment with a low reservation utility while eight
pairs participated in the treatment with a high reservation utility. In the treatment with a
low and non-binding reservation utility, the agents never refused the contracts offered to
them. This result contradicts the hypothesis that the agents might use refusals in the
early rounds for strategic reasons. The sub-game perfect equilibrium contract is chosen
less often than in the experiment by Berg et al. (1992). Principals tend to offer the
contract that is slightly Pareto-dominated by the equilibrium contract. This contract gives,
compared to the equilibrium contract, a lower expected utility to the principal while giving
the same expected utility to the agent. In the treatment with a high and binding
reservation utility, the sub-game perfect equilibrium contract offer is less often observed
than in the experiment by Berg et al. (1992). The principals tend to choose the
equilibrium contract and the contract that is more favorable to them than the equilibrium
contract but should be rejected by the agent equally often. This is not so surprising when
we take the agents' choices into consideration. Even when offered the contract that
should induce rejection, the agents choose the high effort in almost half of the cases.
Although the experiment by Epstein improved on the experimental design introduced by
Berg et al. (1992) by introducing the agent's option to refuse the contract, it remains
open to the two shortcomings mentioned above with respect to the experiment by Berg
et al. (1992): the small number of predetermined contracts and the payment mode in
probability points. Furthermore, Epstein provided only a vague explanation of the
phenomena observed in his experiment, in particular that the sub-game perfect
equilibrium contract was chosen less often than in the experiment by Berg et al. (1992),
claiming that the adding of another dimension to the agent's action increased the
participants' confusion.
3 The principal's behavior in one-shot interactions
In Keser and Willinger (2000, 2001) we present a series of experiments designed to test
the static version of the standard principal“agent model. In these experiments the
participants are assigned a role as either a principal or an agent for the entire
experimental session. Each experimental session consists of one or two sequences of
ten periods but in each period each of the participants is randomly re-matched with one
of the participants in the other role. Thus, we consider the interaction of a principal and
an agent in each period as a one-shot interaction; it is very unlikely for the same two
participants to encounter each other again in the following periods. The parameters of
the initial experiment (Keser and Willinger 2000) have been chosen such that the sub-
game perfect equilibrium solution for a risk neutral principal and risk neutral agent is a
unique contract under the restriction of integer numbers. Contract offers are, in contrast
to the previously discussed experiments, limited only by lower and upper bounds on
each of their components. All of the contract offers observed in the experiment are
different from the sub-game perfect equilibrium prediction. The participants in the role of
a principal offer contracts that do not tend to satisfy the incentive constraint and that are
significantly more favorable to the agent than in equilibrium (i.e. the participation
constraint is satisfied but not binding). We show that these results can be explained by
behavioral principles of the principals, which are in contradiction with the behavior
assumed in the principal“agent literature.

3.1 Design and theoretical predictions
In the experiment presented in Keser and Willinger (2000), the principal may offer one of
a huge set of contracts. The model assumes two possible states of nature, which
correspond to a profit of 50 (state 1) or of 100 (state 2) experimental currency units. The
principal may propose a contract, defined as the pair (w1, w2) where wi is the
remuneration if state i is realized. The agent, if he accepts the contract, chooses one of
two activities (effort levels): activity 1 induces a uniform probability over the two states of
nature while activity 2 assigns a probability of 0.8 to state 2. Activity 2 incurs to the agent
a higher cost than activity 1. The parameters of the model are summarized in table 17.1.
If the agent rejects the principal's contract offer both players get zero payoff.

Table 17.1: Parameters of the experiment by Keser and Willinger (2000)
Probability that
principal's profit is

50 100 Agent's

Activity 1 0.5 0.5 13
Activity 2 0.2 0.8 20

In the experiment the game was repeated in a first sequence of ten periods with random
re-matching of each principal with an agent after each period. After a short break during
which the participants could not communicate with each other, a second sequence of ten
periods took place, again with random re-matching and each participant staying in the
same role as either a principal or an agent. In the second sequence we consider the
participants as experienced. Under the assumption of risk-neutrality of both the principal
and the agent, we show that there exists a unique sub-game perfect equilibrium solution
in integers in which the principal offers the contract (1,25) and the agent accepts and
chooses activity 2. Note that this contract, to satisfy the incentive constraint, lets the
agent suffer a net loss if state 1 occurs and a net gain if state 2 is realized.
Under the alternative assumption of constant absolute risk aversion of the agent (i.e. u(x)
= e , where u(x) is the agent's utility function and x denotes his payoff) two types of
equilibria are possible. The principal can induce activity 1 by proposing either (13,14) or
(14,13). Note that for each of these contracts the agent has a utility which is superior to
his reservation utility in one of the states of nature while (13,13) leaves the agent
indifferent between accepting and refusing. The principal can also induce activity 2 by
proposing a contract (w*1, w*2) such that 0 w*1 13 and w2* 20. The exact values of
w*1 and w*2 depend on the value of the coefficient in the agent's utility function. The
principal will choose to induce the activity which will maximize his expected payoff.

3.2 Results and interpretation

One hundred students of various disciplines at the University of Karlsruhe participated in
the experiment. Note that, as the randomized pairing of principals and agents was
effectuated within populations of five participants in the role of principals and five
participants in the role of agents, we obtained ten independent observations. We do not
observe any significant difference between inexperienced and experienced play. Among
the total of 1,000 observed contracts, we never observed the sub-game perfect
equilibrium solution under the assumption of risk neutrality of both the principal and the
agent. The sub-game perfect equilibrium prediction with a risk averse agent was
observed in only one of the contracts. The mean contract observed was (24,46) in the
inexperienced play and (23,45) in the experienced play, and is, thus, in both cases and
in both components far away from the predicted contract. About half of all contract offers
satisfied the incentive constraint for activity 2; 95 percent of these contracts were
accepted and 70 percent of those accepted led to the choice of activity 2 by the agent.
Contract offers to which the agent's best reply would be the choice of activity 1, were
accepted in 80 percent of the cases and then generally led to the choice of activity 1 by
the agent.
To explain these results, in Keser and Willinger (2000) we identified three principles that
seem to have guided the principals' decision-making: appropriateness, loss avoidance,
and sharing power. Appropriateness implies that the principal offers a higher
remuneration for a higher profit. Otherwise, the agent had no incentive to provide a high
effort. The loss avoidance principle implies that the principal proposes only contracts
such that, whatever the state of nature, the agent incurs no loss. This principle
prescribes a lower bound to the remuneration in each state, which conflicts with the
theoretical prediction that the agent's participation constraint should be binding. Many of
the contracts that satisfy this principle also violate the incentive constraint. The sharing-
power principle implies that the principal shares the surplus with the agent but in a way
that it is not less favorable to himself than to the agent. This principle prescribes an
upper bound to the remuneration in each state. It allows for many contracts where the
participation constraint is satisfied but not binding. Thus, this principle conflicts with the
theoretical prediction that the principal should keep the entire expected surplus for
himself by just satisfying the agent's participation constraint. If we denote the agent's
cost for activity i by ci, the three principles can then be defined:
1. Appropriateness: w1 w
2. Loss avoidance: w1 ci and w2 ci (i = 1, 2)
3. Sharing power: w1 ci + (50 ci)/2 and w2 ci + (100 ci)/2 (i = 1, 2)
Note that for the second and the third principles several variants are possible, depending
on the costs (c1 or c2) to be considered. The same costs might be considered in the two
different states of the world. Thus, for the second and third principle we have four
possible variants each. The combination of the three principals (in whatever variant)
describes a specific area of fair offers in the (w1, w2) contract space. Under the
hypothesis that all of the contracts are equally likely to be proposed by a principal who
chooses randomly within the strategy space, we may associate a probability to this fair
offers area, called the area rate. The area rate is defined by the number of potential
contracts in this area divided by the overall number of potential contracts in the contract
space. We then define the hit rate of a combination of the three principals as the relative
frequency with which the observed contracts fall into its predicted area. The difference
between the hit rate and the area rate defines a measure of predictive success, S, for the
considered combination of principles (see Selten 1991). The measure of predictive
success, thus, considers the frequency of contracts satisfying a combination of principles,
beyond the frequency that would be just due to chance. In the experiment, the
combination of principle variants with the highest measure of predictive success
corresponds to contracts that are restricted by the following inequalities: w1 w2, 13
w1 35 and 13 w2 60. This combination corresponds to costs of c1 = 13 for the
principle of loss avoidance in both states of nature (presenting lower boundaries of the
area) and costs of c2 = 20 for the principle of sharing power in both states of nature
(presenting upper boundaries of the area).

These results show that whatever his profit level, the principal tends to offer to the agent
a remuneration which is superior to the one predicted by the sub-game perfect
equilibrium solution. Thus, the principal gives away part of the created surplus to the
agent. Obviously, the principal avoids offering contracts where the agent might incur a
loss. Thus, the agent does not have to bear the risk of a loss and receives a strictly
positive share of the net surplus of the contract. The three principles, appropriateness,
loss avoidance, and sharing power, which we have elaborated in an explorative way
from the behavior of the participants in the experiment, are in conflict with the predictions
of contract theory. Note that these experimental results do not necessarily imply that the
principal wants to be equitable but rather tries to prevent rejection by the agent, which
would eliminate an opportunity for the principal to make a certain gain. Thus, he offers a
contract that yields a certain gain to the agent as well.

3.3 Validation of these principles
To test the robustness of the three principles of the principals' behavior, we ran an
additional series of experiments (Keser and Willinger 2001). In these experiments we
increase the agent's activity costs keeping everything else the same as in the previous
study. We consider three different cost levels maintaining, however, the difference
between the costs of the two activities constant at seven (see table 17.2). A total of 224
students (128 at the University of Strasbourg and 96 at the University of Karlsruhe)
participated in these experiments; we obtained four independent observations for each
cost situation in each of the two countries (additional to the ten independent observations
in Germany for the lowest cost level).

Table 17.2: Activity costs in the experiments by Keser and Willinger (2001)
Cost level for the agent Activity Activity
1 2

Low (Keser and Willinger 2000) 13 20
Medium 27 34
High 34 41
Very high 41 48

By increasing the costs we reduce the principal's expected surplus from the contract in
the sub-game perfect equilibrium solution. Furthermore, for these higher cost levels,
there exist multiple equilibria (in integers) where both the agent and the principal are risk
neutral. This renders the comparison of the results with the theoretical prediction less
straight-forward. The analysis of the average Euclidean distance of the observed
contracts from the respective closest equilibrium contract shows that this distance
becomes the smaller as activity costs increase. However, the contracts offered by the
principals fail to satisfy the incentive constraint more frequently as activity costs increase.
As effort costs increase, the remunerations increase that the principal must offer to the
agent in order to satisfy the loss avoidance principle and to make him accept the contract.
The area that corresponds to the combination of the three principles, appropriateness,
loss avoidance, and sharing power, that define the fair offers prediction, becomes
smaller as costs increase. These experiments, thus, allow us to test the robustness of
the fair offers predictions and the three principles on which it is based by having smaller
and smaller areas described by the principles.
In the experiments with higher activity costs, the combination of the same variants of the
three principles as in the previous experiment still yields the (second) highest measure of
predictive success among all possible combinations. We observe a reduction in the
measure of predictive success, though, when we increase the cost level. From the
experiment with the lowest cost level to the experiment with the highest cost level the
measure of predictive success of the fair offers prediction is reduced by 50 percent. This
decrease reflects a reduction of the hit rate that is more important than the reduction of
the area rate. Although the measure of predictive success remains at around 45 percent,
we do observe many contract offers not satisfying the loss avoidance principle when the
costs are very high. If we restrict, however, our analysis to those contracts that are
accepted by the agents, the hit rate of the loss avoidance principle per se (such as the
measure of predictive success of the fair offer prediction) is much higher than without
this restriction.

In Keser and Willinger (2001) we determine the set of all possible sub-game perfect
equilibrium contracts under the assumption of a risk averse agent whose utility function
belongs to the class of strictly increasing and concave functions. As neither the
experimenter nor the participant in the role of the principal can know the agent's utility
function, we predict that the solution should lie in a specific sub-set of the principal's
strategy space. Comparing this equilibrium prediction under the assumption of a risk
averse agent to the fair offers prediction, we observe that the latter yields significantly
higher measures of predictive success as long as the effort cost is not very high. In other
words, the fair offers prediction does better than the equilibrium prediction for a risk
averse agent as long as the surplus of the contract to be allocated among the principal
and the agent is not too small.

To summarize, the three principles remain a good predictor for the observed contracts as
long as there is a more or less important surplus of the contract to be allocated between
the principal and the agent. The loss avoidance principle appears to be the least robust
one among the three decision principles when the surplus becomes unimportant. It
appears that the principals want to keep the same share of the expected surplus,
whatever the size of this surplus. Interestingly, principals in Germany offer more
generous contracts to the agents than those in France.
4 Contract offers and effort in repeated interaction
The first experiments by Berg et al. (1992) and Epstein (1992) were based on repeated
interaction between the same principal and the same agent. In this section we will report
on two more recent experiments, which were designed to test the behavior of both the
principal and the agent in repeated interactions. In these experiments the participants
know that they will interact with the same other person during the entire experiment. The
results of these experiments show the importance of reciprocity in the repeated
interaction of a principal and an agent.

4.1 The experiment by Güth et al. (1998)
4.1.1 Design and theoretical predictions
Güth et al. (1998) study the behavior of a principal and an agent in a rather complex
dynamic game context with both hidden action and hidden information. Their primary aim
is not to test theoretical predictions as such but to provide empirical facts showing how
incentives compete with trust and reciprocity. In the experimental game situation, the
principal might be the owner of a firm whose management is conferred on an agent. The
principal's interest is to accumulate capital to be liquidated at the end of the game. The
agent plays a particular role in this game. On the one hand, as the manager of the
capital he can let it grow by providing an effort. On the other hand, he has to decide in
each period on how much of the profit he wants to distribute as a dividend. The
distributed profit is allocated between the principal and the agent in a proportion that is
specified in the contract.
In the beginning of the game the principal proposes at least one but not more than three
contracts to the agent. A contract has two components (f1, s1) where f1 (with 0 f1 4)
is a fixed salary and s1 (with s1 {0, 0.1, 0.2, 0.3, 0.4}) is the agent's share of the
dividend. The principal's share of the dividend is (1 s1). The agent has to accept one of
the contracts. The interaction between the principal and the agent lasts between three
and six periods in total. The principal may in each period, t (with t > 1), revise the
accepted contract upward. Concretely, he may increase the fixed salary component
and/or the agent's share of the dividend, such that ft ft 1 and st st 1.
At the beginning of the game neither the principal nor the agent knows the true value of
the firm, W1. They know that the firm can have one of two values, W1 {3, 12}, where
each value has a probability of one-half. After having chosen the contract in the first
period, the true value of the firm is communicated to the agent. The agent then chooses
an effort et ={0, 1, 2, 3} the cost of which is given by c(et) = et/2. The agent's effort
determines the random return of the firm Rt = at t, where t is a random variable with a
uniform distribution over {1, 2, 3}. Only the agent can observe the realization of the
random variable, after the choice of his effort. The firm's profit in period t is defined by t

= Rt ft. Once the agent knows the firm's profit, he decides on the amount of the
dividend Dt for the current period. The dividend must not exceed the sum of the firm's
value (Wt) plus the profit of the current period. More concretely, 0 Dt Wt Wt + if
Wt + t > 0, and Dt = 0 otherwise, with Wt = Wt 1 + t 1 Dt 1 for t > 1. Note that the
principal is informed of the dividend (Dt) only, while the agent knows in each period also
the firm's value (Wt) and the realized profit ( t). The payoffs of the agent ( t) and the
principal ( t) in period t are defined, respectively, as = ft + st Dt c(et) and = (1
t t


From the second period on, the principal's decision is whether or not to increase the
fixed pay and/or the agent's dividend share. In periods 4-5 he can end the game
unilaterally. Otherwise the game ends after period 6. The agent's total payoff
corresponds to the sum of his payoffs in each period, while the principal's payoff is
determined by the sum of his dividend payoffs plus the residual value of the firm.
Under the assumption of risk neutrality of both the principal and the agent this game has
multiple equilibria. However the authors propose a reference equilibrium to which they
compare the experimental data. The reference equilibrium is a stationary one which is
characterized by the fact that the dividend in each period is equal to the value of the firm
plus the profit of the period (Dt = Wt + t), the principal never ends the game before the
final period, and induces the maximum effort by offering a contract with a zero fixed
salary component and a 30 percent share of the dividend for the agent (ft = 0, st = 0.3).
Note that the agent provides the maximum effort if st 0.3 (incentive constraint); he
provides the lowest effort if st < 0.3. Intermediate effort levels are never the agent's
optimal choice.

Sixty-four participants (students of economics or business administration at the
Humboldt University of Berlin) participated in this experiment. The multiperiod game was
played twice in each session, with re-matching of pairs of a principal and an agent in the
beginning of the second play. Each participant remained, however, in the same role as
either a principal or an agent.
4.1.2 Results
The principal's behavior The principals generally offer several contracts in the first period.
More than 90 percent of the proposed contracts offer a positive fixed salary component,
which is costly to the principal but should have no influence on the agent's effort choice.
Almost all of the contracts offer positive dividend shares that are, however, not satisfying
the incentive constraint in the first periods. Both the dividend shares and the fixed salary
components increase over time. Note, however, that the increase of the two components
is not so surprising since the rules of the game do not allow for a decrease. The average
dividend share is around 30 percent toward the end of the game. Shares below and
above the incentive-compatible share are observed. In about half of all cases the
principals finish the game before period 6. The observed probability to finish the game
early depends on the dividends paid: early termination is more likely after a zero dividend
than after a positive one. These results hold for both inexperienced and experienced play.
The agent's behavior The agents, when they have the choice among several contracts
proposed by the principal in the first period, generally choose the one that offers the
highest fixed salary. In only 19 percent of all cases did the agents choose the contract
with the highest dividend share. The authors interpret this as self-selection by the agents.
They observe that the agents who choose the contract with the highest dividend share
tend to choose higher effort levels in the first period. About two-thirds of the efforts
chosen over the two repetitions of the multiperiod game are different from the effort to be
induced. We observe many intermediate effort levels (neither zero nor maximum effort)
and also that the agents provide strictly positive efforts when their dividend share is
below 30 percent. Furthermore, in each period the distributed dividends are inferior to
the sum of the value of the firm plus the profit in that period. However, the value of the
firm decreases over time and becomes zero in many cases in the final period.

The authors show in a regression analysis that the effort level is positively correlated to
the dividend share proposed in the contract and to the fixed salary component. In other
words, agents respond to more favorable offers with higher effort levels. According to the
authors' interpretation, the size of the fixed salary measures trust on the part of the
principals and the positive correlation between the effort and the dividend share reveals
reciprocal behavior on the part of the agents. Reciprocity becomes stronger in the course
of an experimental game but is not carried over from one game (inexperienced game) to
the next (experienced game, with a different partner). Fehr, G¤chter and Kirchsterger
(1997) show that reciprocity can be a powerful contract enforcement device.
If the objective is to test the predictions of principal-agency theory, this experiment
implies several problems, in particular the one of the complexity of the game. The
authors justify this choice by a concern for realism. They see the characteristics such as
the repeated nature of the interaction, the downward rigidity of the remuneration, etc., as
important elements of real contracts. Given its complexity, the experimental environment
examined by Güth et al. leaves little chance for the theory to do well; in particular owing
to the fact that the fixed salary component cannot be downward adjusted. Furthermore,
the agent cannot influence the principal through strategic refusals. The conditions on
which the theory is based are thus not met in this experiment. Some of these problems
have been dealt with in the experiment by Anderhub, G¤chter and Königstein (1999).
4.2 The experiment by Anderhub, G¤chter and Königstein (1999)
4.2.1 Design and theoretical prediction
The experiment by Anderhub, G¤chter and Königstein (1999) is based on the same type
of contract as is the experiment by Güth et al. (1998). This means that the contracts are
of type (ft, st) where ft is a fixed payment and st is a share of the profit realized by the
principal. The principal can announce a desired effort level without the agent's obligation
to satisfy it. The game is repeated over six periods by the same agents and then after a
random re-matching repeated once again over six periods. Note several major
differences with respect to the experiment by Güth et al.: the principal may propose a
new contract in each period, the agent may refuse any contract, the agent's effort is ex
post observable to the principal, and the relationship between profit and effort is not
stochastic. Given the absence of risk the agent's behavior is not affected by his risk
attitude. The fact that the effort will be observable to the principal ex post may affect the
agent's behavior although the agent cannot be directly punished by the principal for
deviation from the required contract. The agent might be afraid, however, of an
unfavorable reaction of the principal in periods following little cooperative behavior by the
agent. At the same time, the observation of reciprocity by the agent might be important
for the building up of a cooperative relationship between the principal and the agent.
In period 1 the principal may offer a maximum of two contracts and only a single one in
all subsequent periods. The agent has the possibility to accept or to refuse the contract(s)
in each period t. In case of a rejection both players' payoffs are zero. If the agent accepts
the contract he chooses an effort level et and then the principal is informed of his
decision. The return of the effort is given by R(et) = 35et and the effort costs are given by
a piecewise linear function, c(et), which is convex and increases with et. The payoffs to
the agent and the principal in period t are t = ft + st Rt c(et) and t = (1 st)Rt ft,
The parameters of the experiment impose the following constraints on the decision
variables: ft { 700, 699, , +700}, st {0, 0.01, , 1}, et {0, 1, , 20}. The
baseline game has multiple sub-game perfect equilibria but a unique "trembling hand"
perfect equilibrium (Selten 1975) in which the principal offers the contract ( 400, 100%)
to the agent who provides maximum effort. This solution prescribes that the principal
sells the firm to the agent at a price corresponding to the revenue in case of maximal
effort. Note that all sub-game perfect equilibria induce a relationship between ft and st
which is given by ft = 300 700st for all st 5 / 7. Thus, ft 200.
4.2.2 Results
The principal's behavior In 70 percent of the 564 observed contracts (47 principals — 6
periods — 2 sessions, participants were students at the University of Zürich with various
backgrounds other than economics) the principals have proposed negative fixed salaries.
At the same time, about two-thirds of all contracts offered a profit share greater than 71
percent. In 82 percent of all cases the contracts aimed at inducing an efficient effort by
the agent (suggested effort level). According to the authors these results show that the
participants in the role of a principal have recognized the necessity to give an incentive
to the agent for providing the desired effort level.
They observe a negligible number of contracts with purely a fixed salary component.
Most of the contracts are of the mixed type (s > 0, f + 0). Contracts where both
components are strictly positive are not incentive compatible. The "trembling hand"
perfect equilibrium (selling of the firm to the agent) is observed in about 30 percent of all
cases (s = 1, f < 0). The participation constraint is almost always satisfied (97 percent of
all cases). However, we observe that the contracts offered imply a more equitable share
of the surplus than predicted by the equilibrium solution according to which the principal
keeps the entire surplus.
The agent's behavior Although 97 percent of the contracts satisfy the agent's
participation constraint, 112 of the 551 contracts that satisfy this constraint are rejected
by the agents (about 20 percent). To examine the questions why agents rejected
contracts, the authors pick up the hypothesis by Slonim and Roth (1998) for ultimatum
bargaining games according to which the acceptance rate is positively correlated to the
relative payment to the agent. The idea is presented in Anderhub, G¤chter and
Königstein (1999) by an equity assumption, according to which the acceptance
probability for a contract is a function of the surplus share offered to the agent. The
authors show the validity of this hypothesis in a logit regression that is run only for those
contracts that satisfy the participation constraint and under the hypothesis that the agent
chooses an effort level that maximizes his payoff. The results show that the surplus
share significantly affects the acceptance probability. This implies that the acceptance
depends not only on the absolute payment but also on the relative payment to the agent.
In other words, equity considerations play a role in the decision-making.

The effort levels chosen by the agents, given an equitable contract, are the rational ones.
More than 60 percent of all effort choices are rational. However, this also implies that a
great number of the effort choices (almost 40 percent) are different from the rational
choice. The authors argue that this may be explained by reciprocal fairness
considerations. They observe that the deviations from the rational effort level (conditional
on the contract offered) are positively correlated with the agent's surplus share.
Contracts that are favorable for the agent trigger effort levels that are above the
individually optimal effort level, and vice versa.

The major conclusion that the authors draw is that these results are not in contradiction
to the maximization hypothesis but show the presence of other motivations such as
equity and reciprocity that might influence the choices made by the principals and the
5 Conclusion
In this chapter we have surveyed several contributions of the experimental literature to
the understanding of the relationship between a principal and an agent. We distinguish
between experiments that are based on repeated interactions and those that examine
one-shot interactions.

In the experiments with one-shot interaction the contracts are almost always different
from the predicted ones. Not even the repetition of the interaction with varying other
participants brings a significant convergence to the predicted contracts. In the observed
contract offers, the participation constraint is not binding and in about half of the cases,
the incentive constraint is not satisfied. The principals are motivated by considerations
other than those that correspond to the participation and incentive constraints when they
design contracts. Most of the contracts avoid potential losses to the agent and imply a
more equitable share of the surplus between the principal and the agent than the share
predicted by principal-agent theory. The fair offer prediction proposed by Keser and
Willinger (2000) is to a large extent robust to changes in the size of the surplus created
through the contract.

The observed effort choices by the agents and their decisions whether or not to accept a
contract tend to be compatible with the theoretical predictions. Note however, that the
contract offers by the principal already reflect the principals' fear of rejection (loss
avoidance principle) so that we cannot observe what the agents' reaction would have
been to the theoretically predicted contract offers that are much more unfavorable to

When the interaction between the same principal and the same agent is repeated and
the agent may choose among a set of effort levels larger than just a low and a high one,
the agent's behavior also shows significant differences from the theoretical prediction.
We observe here that the respect of the participation constraint in a proposed contract
does not guarantee its acceptance by the agent but that the probability that the agent
accepts a contract increases with its (absolute and relative) remuneration level. Also the
agent's effort level is positively correlated with the remuneration level proposed by the
principal. The agent's effort level is in many cases even higher than the effort level to be
induced by the contract.

The question now presents itself as to whether the principal consciously proposes a
contract that is favorable to the agent anticipating the reciprocity of the latter. This would
suppose that the principal assumes that the agent is of a reciprocal type and that the
agent would sanction an inequitable proposal by a rejection and recompense a favorable
offer by a high effort level. An alternative explanation could be that the principal plays an
incomplete information game where he does not know, for example the acceptance
benchmark of the agent. Future experimental studies will likely consider the predictive
power of these explanations in principal-agent situations. Note, however, that
experimental results on other forms of repeated interactions (e.g. in ultimatum bargaining,
investment games, voluntary contributions to a public good) also reveal the importance
of reciprocity considerations on human behavior.

To conclude this chapter, recall that the principal aim of these experiments is not to
understand the formation of real-life contracts but to test the predictions of the principal-
agent model and the influence of considerations such as trust and reciprocity. We
therefore need to construct frameworks that are appropriate to observe interactions such
as they are described by the theory. The link between contract characteristics observed
in the laboratory environment and in a real-life environment remains an open question.
This question poses itself more generally with respect to the comparison of experimental
results to observations in a real-life context.
Chapter 17 was originally published as "La th©orie des contrats daus une contexte
exp©rimental: un survol des exp©riences sur les relations ˜Principal-agent,' " Revue
d'Economie Industrielle (92, 2000).
Applied Issues-Contributions to
Part VI:

Industrial Organization
Chapter 18: Residual Claims and Self-Enforcement as Incentive Mechanisms in
Franchise Contracts-Substitutes or Complements?
Chapter 19: The Quasi-Judicial role of Large Retailers-An Efficiency Hypothesis of their
Relation with Suppliers
Chapter 20: Interconnection Agreements in Telecommunications Networks-From
Strategic Behaviors to Property Rights
Chapter 21: Licensing in the Chemical Industry
Residual Claims and Self-enforcement
Chapter 18:

as Incentive Mechanisms in Franchise
Contracts”Substitutes or Complements?
Francine Lafontaine, Emmanuel Raynaud
1 Introduction
Franchising is a contractual relationship that has received a significant amount of
attention in the empirical literature on contracting. In large part, this is because
franchising is one of the few types of contractual relationships about which significant
amounts of data are available from public sources. But franchising is also, as noted by
Williamson (1991), a hybrid organizational form, which lies somewhere between
complete vertical integration and spot markets. Thus insights gleaned from the study of
franchise contracting have allowed researchers to develop a better understanding not
only of this organizational form, but also of how firms organize their activities much more
generally, both within and across firms.

Much of the literature on franchising has specifically been concerned with incentive
issues and how these are managed in these contracts. This literature has identified two
main categories of incentive mechanisms relevant to the franchise relationship: residual
claims and self-enforcement. The former relates to the fact that franchisees get to keep
their outlet's profits net of the fees they pay to their franchisors, giving them incentives to
maximize those residual profits. The second relies on the presence of on-going rent at
the outlet level, rent that the franchisee forgoes if his contract is terminated. Such rent is
simply the difference between the (net present value of) returns that the franchisee earns
as a result of being associated with the franchise network and the returns he could
garner in his next best alternative. If the rent is positive, and franchisors can terminate
franchisees, franchisees will have incentives to perform according to the standards set
by the franchisor to reduce their chances of termination and protect their access to the
In this chapter, we describe how these two types of incentive mechanisms work in theory
and in practice in franchise contracting, and then explore the relationship between them.
Our contention is that rather than being alternative approaches to aligning the incentives
of contracting parties, as suggested by much of the current literature, these approaches
are in fact complementary. Specifically, we argue that residual claimancy rights motivate
individual parties to a contract to invest greater effort as per standard agency arguments.
At the same time, the existence of self-enforcement mechanisms prevents parties from
engaging in individually profitable activities that can have a negative impact on the whole,
or on other members of the franchise system. In fact, it is precisely the franchisee's
status as a residual claimant that brings about the need to use a self-enforcement
mechanism to curb his tendency to maximize his own profit at the expense of the brand
or the rest of the chain. We argue that optimal contract design in this context must
effectively balance the provision of high effort incentives for individuals with coordination
incentives that preserve the value of group membership.
The chapter is organized as follows. To fix ideas, we begin in section 2 with a definition
of franchising. In section 3, we describe the theory behind the two types of incentive
mechanisms mentioned above. In section 4 we develop our main argument on the
complementarity between residual claims and self-enforcement in these contracts.
Section 5 describes more specifically how, in practice, various aspects of the franchise
relationship or contract clauses support the two types of incentive mechanisms
discussed herein. Finally, in section 6, we extend our analysis to non-franchised
production and/or retail networks operating under common reputation concerns, such as,
for example, production cooperatives in agro-food industries. Concluding remarks are
found in section 7.
2 Defining franchising
From a legal standpoint, a contract is a franchise contract in the United States if three
main conditions are met: (1) the franchisee operates under the franchisor's brand name
and trademarks, (2) the franchisor provides on-going support and exerts, or can exert,
significant control over the franchisee's operations, and (3) the franchisee is required, as
a condition to obtain the franchise or to commence operation, to pay more than $500 to
the franchisor before the end of the first six months in operation. The legal definition of a
franchise in the European Union (EU) is similar except that it is more specific about the
requirement that the franchisor transfer know-how to the franchisee.[1]

Within franchising, the US Department of Commerce further categorizes relationships
either as "product and trade name," also called "traditional," franchises, and "business
format" franchises. In a product and trade name franchise, the franchisor mostly sells a
finished product to the franchisee who then resells it. Examples include dealer-owned
gasoline stations and car dealerships. In such relationships, the franchisor's profits arise
from the markups charged on products sold to the franchisee. In business format
franchising, by contrast, the franchisor mostly sells the right to use her tradename and
business methods to the franchisee. In this case, the franchisee is responsible for both
the production and sale of the finished product, as in the fast-food or hotel industries. In
exchange for the use of the trade name and business methods, the franchisee pays a
combination of fees to the franchisor. These most often include a franchise fee, payable
once, at the beginning of the period covered by the contract, as well as royalties and
advertising fees which are usually defined as a percentage of the outlet's sales or
revenues.[2] These fees are typically the same for all franchisees joining a chain at a
point in time.[3]

As a result of the emphasis on franchisor know-how in the EU definition of a franchise,
the EU version corresponds more closely to the US definition of a business format
franchise. Within Europe, further slight differences in definition also arise across
countries. Such definitional differences make it difficult or inappropriate to directly
compare statistics on the extent of franchising across countries and jurisdictions.
However, from an incentive perspective, franchisors involved in business format and/or
traditional franchising face very similar sets of challenges. Consequently, the analyses
below apply to both types of franchised relationships, except as specifically noted.

See FTC, "Disclosure Requirements and Prohibitions Concerning Franchising and Business
Opportunity Ventures" (16 CFR § 436.1 et seq.), and EU rule 4087/88; 1988.

According to the US Department of Commerce (DOC) (1988), franchise contracts in the
United States can last anywhere from five years to perpetuity, with an average of about fifteen

See Lafontaine (1992). Also, these fees are fairly stable over time. See Lafontaine and
Shaw (1999) for empirical evidence on this.
3 Residual claims and on-going rent as incentive
Franchising fundamentally involves franchisors granting franchisees the right to operate
under their trade marks and business processes. But as these intangible assets remain
the property of the franchisor, the granting of these rights gives rise to incentive
problems and agency costs. As noted above, two main types of incentive mechanisms
have been identified in the literature on franchise contracting as ways to mitigate these
problems: the granting of residual claimancy rights, as emphasized in the principal“agent
literature (see, e.g., Rubin 1978; Mathewson and Winter 1985) and the reliance on self-
enforcement, which involves the provision of a stream of on-going rent downstream that
the franchisee forgoes in the event of contract termination (see in particular Klein and
Leffler 1981; Klein 1995). In what follows, we discuss the theory behind the functioning of
these two types of mechanisms in the context of franchising.

3.1 Residual claims in franchising

Franchisee-owned businesses are legally independent from the business of their
franchisor. Franchisees can own one or several franchised outlets in a chain, and as
owners, they have a claim on the profits generated by their outlet(s). Franchisees claim
these profits net of the usual sales-based royalties and advertising fees they pay to their
franchisors. As these payments normally represent 6“10 percent of revenues,
franchisees obtain the bulk of every additional dollar of sales generated within their
outlet(s). Also, since royalty and advertising fee payments are based on revenues and
not profits, franchisees reap the full benefit from every additional dollar decrease in
operating costs.[4]
When franchisee effort is not observable, and so cannot be contracted on directly, it is
optimal for the franchisor to sell the outlet to the franchisee for a fixed price (assuming
that the franchisee's effort is central to production). This outright sale makes the
franchisee a full residual claimant, thereby giving him incentives to put forth the optimal
level of effort (see, e.g., Rubin 1978; Mathewson and Winter, 1985). Specifically, assume
that sales S depend on franchisee effort in the following way:

where e is franchisee effort and a measures the importance of the franchisee's effort in
the sales generation process. is a random variable with mean 0 and variance 2 that
prevents the franchisor from inferring e from observed S. If the cost of effort for the
franchisee is C(e) = e2/2 and F is the price at which the franchisor sells the outlet to the
franchisee, the risk-neutral franchisee will want to maximize expected profits, namely

The first-order condition for this maximization problem gives e** = a, which corresponds
to the first-best level of effort and, thus, to the level of effort that the franchisor would
have chosen if she had control over it. In that sense, selling the outlet at a fixed price to
the franchisee completely resolves franchisee incentive issues. The franchisor can
extract all the profits from the outlet operations by appropriately setting the price, F, at F
= a2/2.[5]
In practice, franchisees usually do not acquire outlets at a fixed price. Instead, they pay a
nominal fixed fee, plus a proportion of their revenues every period over the whole
duration of the contract. The typical franchise contract thus involves sharing.[6] Yet, under
our assumptions, sharing is counter-productive “ it prevents the realization of the first-
best outcome. In particular, the franchisee who must pay a portion of his revenues to
the franchisor (where 0 < < 1 represents the sum of all revenue-based fees such as
royalty rates and advertising fees), maximizes

by setting effort e* = (1 )a. As this effort level is lower than the firstbest level (= a),
which is readily achievable under a fixed-price sale contract, one would not expect
sharing to occur in this setting.
The principal“agent literature provides two alternative amendments to the model above
to account for the use, in practice, of sharing arrangements. The first amendment, which
is the most traditional, involves introducing the assumption that the franchisee is risk
averse rather than risk neutral (see for example Stiglitz 1974 for the first such model,
applied to sharecropping). In this case, the franchisee no longer maximizes expected
profits, but rather expected utility. Sharing in this model then becomes a means of
shifting risk from risk averse agents (franchisees) to risk neutral principals (franchisors).
The second amendment relies instead on the assumption that the principal (franchisor)
provides some valuable input in the production process and that her behavior, like that of
the agent (franchisee), is difficult to monitor. In this model, called a double-sided moral-
hazard model, sharing arises from the need to provide incentives to the franchisor as
well as the franchisee (see notably Rubin 1978; Eswaran and Kotwal 1985; Lal 1990;
and Bhattacharyya and Lafontaine 1995).
Where the franchisee is risk averse rather than risk neutral, he maximizes his expected
utility from outlet profits or his certainty-equivalent income. Assuming that is normally
distributed, and that the franchisee has a constant absolute risk aversion parameter of ,
his certainty-equivalent income is given by

where E(y) are his expected revenues. The first-order condition for this maximization
problem again yields e* = (1 )a. Once substituted back into the franchisor's
maximization problem, who chooses to maximize total surplus, we have

2 2 2
The first-order condition for the franchisor's problem implies that * = [ /(a + )] > 0.
In words, the "best" contract from the franchisor's perspective now involves sharing. This
sharing arises as a way to balance the need to motivate franchisee effort (which leads
toward = 0) while providing insurance to now risk averse agents (which leads toward
= 1). Of course, while this solution is optimal from the franchisor's perspective, it does
not give rise to the first-best level of effort and output as e* = (1 )a < a = e**.

Assuming instead that the franchisor also provides some non-observable input that
contributes to the franchised outlet production or sales process, then even under risk
neutrality for both franchisor and franchisee the optimal contract will involve sharing.[7]
The share parameter now trades off franchisor and franchisee incentives. To illustrate,
assume that outlet sales are given by

where r is the franchisor's effort level. Assume further that the franchisor's cost of effort is
given by C(r) = r /2. The franchisee maximizing his profits given will again choose e* =
(1 )a. The franchisor who gets a fraction of outlet sales will set r * = b.
Substituting these two effort levels into the franchisor's maximization problem yields

The first-order condition for this maximization problem gives * = b2/(a2 + b2) > 0, which
once again implies sharing.
Three main testable implications arise from these principal“agent models. The share
parameter (here, the sum of royalties and sales-based advertising fees) will be higher:
1. the lower the importance of franchisee effort, as captured by a above
2. the higher the level of risk involved ( ) (assuming the franchisee is
more risk averse than the franchisor)
3. the more important franchisor effort is, as captured by b above
(assuming this effort is non-observable).
The empirical literature on franchising has found support for (1) and (3), but not for (2).

Our discussion thus far has focused on the incentives embedded in franchise contracts
via residual claims. It is important to note, however, that employment contracts can also
accord residual claimant status to employees. And an employee“manager whose
compensation was directly tied to the profits of the outlet he manages would choose the
same effort level as a franchisee as long as his contract entailed the same "level" of
residual claims. In practice, however, franchise contracts normally entail residual
claimant status for franchisees whereas the compensation of managers of company
units in franchised chains usually is not tied very closely to outlet profits (see Bradach
1997 for evidence.)

3.2 Self-enforcement
In this section, we turn our attention to the role of hostages (Williamson 1985), efficiency
wages (Shapiro and Stiglitz 1984; Akerlof and Yellen 1986) and self-enforcement more
generally(Klein 1980; Klein and Leffler 1981; Klein and Saft 1985) in franchise
contracting. The common thread across all these analyses is the notion that parties to a
contract can be given incentives to put forth effort by making sure that they derive a
benefit from the relationship that is at risk if they do not behave as requested. The
incentives embedded in a franchise contract in this context do not stem from residual
claims, but rather from the combined effect of three elements: (1) an ongoing stream of
rent that the franchisee earns within the relationship but forgoes if he "leaves" the
franchised chain, (2) franchisee monitoring by the franchisor, and (3) franchisor ability to
terminate the franchise contract. Since the ease or cost of termination is largely
determined by the applicable legal system, the franchisor is left with the tasks of
choosing the level of ongoing rent to be left with franchisees and selecting the frequency
of monitoring so as to minimize the ex post cost of enforcing the desired level of effort.[10]
Specifically, let W t represent the (expected) gain that the franchisee can obtain when
deviating from the franchisor's requested behavior, and W t be the present value of the
ongoing rent that the franchisee can earn within the relationship. Then a franchise
2 1
contract is self-enforcing iff W t > W t at every time t. In other words, for the contract to
be continuously-self-enforcing, the franchisee must have a minimum amount of rent to
look forward to each period. W2t must therefore include not only the rent expected over
the remainder of the contract, which by definition decreases as the franchise gets closer
to expiration, but also rent associated with future additional outlets and with the
probability of contract renewal.
In this framework, specific contract terms (described in more detail in section 4) play
different roles (Klein 1995), influencing either W1t or W2t:
Contract terms affecting W1t: Some contract terms specify certain
franchisee obligations, for example the mandatory level of input
purchases from the franchisor, other procurement requirements,
minimum local advertising expenditures, or staffing levels. These
contract terms limit W as they make it easier for the franchisor to
detect non-conformance and quickly intervene to limit the associated
benefit to the franchisee. They also make it less costly for the
franchisor to rely on third-party or court enforcement as they provide
more objective bases from which to establish non-conformance.
ii. Contract terms affecting W : Other contract terms serve to ensure the
existence of the stream of ongoing rent whose potential loss gives
incentives to the franchisee. Although Klein (1980, 1995) does not
specify exactly how the stream of rent is created, he suggests that
clauses such as exclusive territories limit intra-brand competition and
thus contribute to the franchisee's profitability. As noted above,
guarantees about future expansion opportunities and likelihood of
contract renewal could further affect the level of expected rent
Because of uncertainty, complexity and lack of perfect monitoring, all aspects of the
behavior desired of franchisees cannot be specified by the franchisor in the contract a
priori. Hence the franchisee always has some leeway, and W1 is never zero. As a
consequence, the contract must always give rise to positive rent W if the incentive
constraint above is to be continuously satisfied. At the same time, there exists a
maximum amount of rent to which the franchisor can credibly commit. If the franchisor
prefers franchising to company-managed stores, it is presumably because vertical
integration (company management) is less profitable. This implies that > 0, the
difference in profit from operating a unit under franchising versus vertical integration, is
positive. Then the franchisor's promise of rent to the franchisee is credible if the value of
the rent is less than the discounted profit difference at every time t, namely

If this condition is met, then it is in the best interest of the franchisor to pay the rent.
Otherwise, it is more economical for the franchisor to vertically integrate and appropriate
the rent.

Empirically, Kaufmann and Lafontaine (1994) have shown, through a detailed analysis of
the economics of McDonald's restaurants in the United States, that there is indeed rent
left downstream in that chain. Following a similar methodology, Michael and Moore
(1995) confirmed the existence of rent in a number of other franchised chains. Moreover,
Brickley, Dark and Weisbach (1991) have shown that the proportion of corporate units in
franchised chains is higher in US states that restrict the termination of contracts
compared to other states (see also Beales and Muris 1995). This result suggests that the
cost of termination indeed affects franchisors' decisions to franchise or vertically
integrate outlets, thereby lending support to the idea that franchisors rely on rent and
termination in their dealings with their franchisees.

Finally, it is important to note that the use of rent and self-enforcement as an incentive
mechanism is in no way limited to the franchise context. In fact, the huge literature on
"efficiency wages" in labor economics shows that the provision of "rent" and its potential
loss are used to motivate employees within firms just as they can be used to motivate
franchisees. This suggests that a franchisor could well use efficiency wages to motivate
her store managers and, in doing so, eliminate the need to give store managers residual
claims or use franchisees.

In product and trade name franchising, franchisees do not pay these salesbased royalties.
However, the markups charged by the franchisor on every unit of input can be equivalent to
sales royalties under certain conditions. (See Lafontaine and Slade 2001 for more on this.)

To simplify the algebra, we ignore issues of contract duration and discounting. This in no
way affects the generality of the result that a fixed price contract resolves all incentive issues
when franchisees are risk neutral. Note that F could be set at any level not exceeding a /2.
But any F below a /2 means that the franchisor does not make as much as he could, and the
franchisee does better than required by his participation constraint. We come back to this

Sharing occurs also, for example, in share cropping, licensing, film distribution, and
publishing contracts.

If the franchisee is risk averse while the franchisor is risk neutral, as in our previous setting,
sharing will arise as an optimal response still. In this case, the share parameter will play the
double role of providing incentives to the franchisor as well as insurance to the franchisee.

See Lafontaine and Slade (2001) for a review of the empirical literature on franchise

See Lutz (1995) for more on this.

For self-enforcement to work, the franchisor must be able to evaluate, ex post, whether or
not the franchisee's performance is satisfactory even if the desired effort is too complex to
specify in the contract.

Indeed, only high-performance franchisees can expect renewal and additional outlets within
the same chain. These decisions therefore entail rent that gives further incentives to
franchisees. See Kaufmann and Lafontaine (1994) for more on this.
This assumes that franchisees can earn profits in the long run, i.e. that they do not operate
in a perfectly competitive or mono polistically competitive market. If profits were dissipated in
the long run, there would be no rent in the long run, and thus no self-enforcement. In other
words, Klein's analysis presumes that branding allows franchisees to differentiate their
product enough that they earn positive profits in the long run (that the franchisor may or may
not extract fully-up-front “ we come back to the issue of rent extraction below).

[13] 2
We assume that W t varies over time. See Williams (1996) and Brickley (2001) for an
argument that as the market changes, the amount of rent may change in a way that makes
integration the preferred option. The franchisor who then terminates or does not renew a
franchise contract can be thought of as exercising the equivalent of a "call option."
4 Substitutes or complements?
In this section, we consider why both residual claims and self-enforcement coexist at
McDonald's and in other franchise systems. This coexistence is puzzling given that the
agency and the self-enforcement literature each suggests that its incentive mechanism is
sufficient, in itself, to resolve incentive issues.

Specifically, the self-enforcement literature, and Klein in particular, never considers the
use of residual claims as an incentive mechanism. In this literature, the combination of a
stream of rent, periodic monitoring, and the termination option are sufficient to achieve
the desired outcome. There is therefore no role for residual claims in the analysis.
Similarly, in the literature that emphasizes residual claims as a source of incentives, rent
does not enter into play at all. If, ex ante, the franchisor has designed an optimal contract,
i.e. a contract that satisfies the franchisee's incentive constraint as well as his
participation constraint, the franchisee earns no rent. In some models franchisees do
earn rent, but these arise from the need to use the right share parameter, , while also
satisfying some liquidity or selection constraint (see Mathewson and Winter 1985). The
rent serves no direct incentive purpose in these models. Furthermore, as the contract is
designed with the franchisee's optimal reaction in mind, the franchisee has no reason to
deviate ex post, and the final outcome is exactly what the franchisor expects it to be.

Why, given this, do we see residual claims and self-enforcement being used together in
franchise contracts? We believe that this coexistence arises because residual claims
give franchisees the incentive to put forth effort and not shirk, while ongoing rent gives
franchisees the incentive to maintain the value of the brand by acting in the chain's
collective interests. In fact, we would argue that it is precisely franchisees' residual
claimant status that reinforces the need to use a self-enforcement mechanism to curb
the tendency of franchisees to maximize their own profit at the expense of the overall
chain. In short, we contend that the two mechanisms work in tandem and complement
one another rather than being alternatives for one another.

But what are those behaviors that franchisees might engage in to increase their profits at
the expense of the chain? Franchisee free-riding on the value of the brand is one form of
franchisee "misbehavior" that has been discussed frequently in the franchising literature.
The issue, in essence, is one of externality: the franchisee bears the full cost of
maintaining high quality in his outlet, but the benefit of his behavior accrues not only to
him in the form of high outlet sales, but also to all others in the chain as well as to the
franchisor as high quality in each outlet leads to higher sales overall in the chain. In that
sense, the quality level that maximizes the franchisee's profits is always lower than that
desired by the franchisor (see Brickley, Dark and Weisbach, 1991; Blair and Kaserman,
1994). Similarly, the prices that maximize a franchisee's profits are higher than those that
maximize chain profits. This again stems from the fact that the franchisee does not
appropriate the positive effect of his low prices on sales at other outlets in the chain (see,
e.g., Barron and Umbeck 1984; Shepard 1993; Lafontaine 2001, for more on this).
Finally, franchisees can refuse to implement new production processes or to sell new
products that they don't expect will be profitable in their particular market even if they are
expected to be worthwhile for the chain, or they may choose to modify processes or
product offerings to better fit their particular market (see for instance Kaufmann 1987;
Lewin-Solomon 1998).
All of these franchisee profit maximization strategies correspond to "misbehaviors" from
the chain's perspective. To understand why, one need merely reflect on the franchised
chain's "raison d'être": to offer consumers a predictable, homogeneous product across a
large number of geographically dispersed establishments. In fact, homogeneity is the
goal not only for product offerings, but also for building design, ambiance, service, and
price as this is at the heart of sustaining the value of the franchised chain brand. If a
particular franchisee offers lower service or less quality, consumers may well infer that
overall chain quality is declining, and choose not to frequent any of the chain's
establishments in the future. Similarly, a franchisee's effort to satisfy his local customers
via special product offerings may affect the franchise chain negatively if consumers
become confused about what to expect, or are disappointed when other outlets do not
carry their favorite product. In short, franchisees' efforts towards individual profit
maximization can adversely affect the franchisor and other franchisees by eroding the
value of the brand on which all parties in the chain depend, and thus adversely impacting
the value of group membership itself.
The need for homogeneity in franchised chains in fact gives rise to significant restrictions
on franchisee behavior (see section 5 for more details). Such restrictions reduce the
profit that the franchisee could derive from outlet ownership. The ongoing stream of rent
earned by franchisees can be viewed as "compensation" for the profits they forgo owing
to these restrictions. This compensation should then be such that the franchisee earns at
least as much under the contract than by maximizing only his own profits. This will
automatically be satisfied if the contract is self-enforcing since it implies that W2t > W1t.
Further, W2t includes expected rent from the additional outlets that a franchisee might be
given the right to operate in the future. As the cost of having their behavior constrained
by the franchise system is likely to be larger for better, more highly motivated franchisees,
it is important that their expected rent also be larger. This occurs automatically here
since better franchisees are more likely to be given the opportunity to own several

Our argument so far however raises an important issue: if the franchisee's residual
claimant status leads him or her to behave in ways that are inconsistent with what is
optimal for the chain, there by requiring the use of supplemental incentive mechanisms,
why don't franchisors simply use self-enforcement mechanisms without residual claims
to motivate franchisees? After all, the self-enforcement literature suggests one can
obtain the desired behavior simply with an appropriate combination of on-going rent,
monitoring, and termination.

We would argue that the answer to this question lies in the different types of tasks
required of franchisees. Specifically, some of the activities that franchisees engage in,
such as all those related to day-to-day outlet operations, are very costly to monitor,
especially for geographically dispersed outlets. Moreover, individual outlet sales and
profits are fairly highly correlated to franchisee effort for these types of tasks. Residual
claims are a particularly appropriate incentive tool in such contexts, i.e. when output
measures (here sales and profits) are good proxies for effort and effort is difficult to
monitor (see for example Lafontaine and Slade 1996 for more on this).[16] By contrast, a
franchisee's decision to implement or not new production procedures or new product
offerings, or to participate in various system-level activities, and more generally to
comply with explicit contract clauses such as those that govern supplier choices and
minimum advertising levels, are all fairly easy (low-cost) to monitor. Furthermore, as
argued above, the correlation between outlet sales and compliance with all these policies
need not be high at all. If sales and/or profits do not provide a good measure of such
effort, residual claims will not provide the right incentives to implement these.
Franchisors will therefore do better using a self-enforcement mechanism to get the
franchisee to participate in these.

Brickley and Dark (1987) also point out that franchisees tend to under-invest in their outlets
as they must assume most of the investment risk. While this effect is due to risk aversion
rather than the presence of an externality, it again implies that the franchisee will not act in
the best interest of the chain as a whole.

Also, by granting franchisees several outlets that are close to one another, the franchisor
may benefit even more as the franchisee then internalizes more of the horizontal effects of his


. 9
( 18)