Linear Price Equilibria in a Non-Exclusive

Insurance Market

Frdric Lossy Gwenal Piaserz

Abstract

We consider a competitive insurance market in which agents can privately enter into
multi-contractual insurance relationships and undertake hidden actions. We study the
existence of a linear equilibrium when insurance companies have no restrictions on their
pricing rules. For CARA utility functions we show that a linear equilibrium always exists.
For DARA utility functions we provide sucient conditions under which a linear equilib-
rium exists. We show that the linear equilibrium (when it exists) is unique, actuarially
unfair and induces partial insurance coverage. Lastly, we show that the linear equilibrium
is always third-best ecient.

Keywords: Common Agency, Insurance, Moral Hazard, Perfect Competition, Linear Price
Equilibria.
JEL Classication: D41, D82, D86, G22.

1 Introduction
We consider a competitive insurance market in which agents can privately contract with several
insurance companies. One good example of non-exclusive insurance contracts is the case of the
insurance for auto. Then, an automobilist can complement his compulsory motor insurance with
deductible by a buyback deductible (auto) insurance contract.
In this context, we rst study the existence of a linear equilibrium in prices when insurance
companies do not have any restrictions on their pricing rules and agents are subject to moral
hazard. Indeed, empirical evidence conrms the existence of linear price equilibria. For instance,

We would like to thank Andrea Attar, Johanna Etner, Pierre Picard, Casey Rothschild, Anne Marie Tauzin,
and Bertrand Villeneuve as well as seminar participants at Universit Paris-Dauphine, Ecole Polytechnique,
EGRIE (2013). All remaining errors are our own.
y
PSL, Universit Paris-Dauphine (LEDa).
z
IPAG, Business School, Paris.

1
in the United Kingdoms annuity market,1 Finkelstein and Poterba (2002) nd that pricing is
linear up to a small administrative charge.2 We then characterize the linear equilibrium we nd.
Lastly, we examine the welfare properties of the linear equilibrium.
In order to study non-exclusive competition in insurance contracts we use a common agency
model (see for instance, Bernheim and Whinston (1986)). We model insurance companies as
principals oering contracts to agents that can privately choose several contracts from within
the global set of oered insurance contracts, and that each exerts an unobservable and costly
eort that aects their probability of having an accident. We assume that agents exert an eort
which is selected from a continuous choice set. Moreover, following Peters (2001) and Martimort
and Stole (2002), we assume that insurance companies may oer menus of contracts so as to
characterize all the equilibria of the game with common agency.

The model we develop is important in at least three respects. First, it provides microeconomic
foundations for the seminal literature on competitive non-exclusive insurance markets, which
assumes linear price contracts (see Pauly (1974); Arnott and Stiglitz (1988, 1991 and 1993)). In
contrast, we allow rms to oer any type of menu of contracts (and, in particular, non-linear
contracts). We then show that a pure strategies equilibrium in which insurance companies issue
linear contracts always exist in the case of CARA utility functions.
Second, in the case of DARA utility functions, it provides general sucient conditions that
guarantee the existence of a linear price equilibrium. Providing such general conditions is only
possible because we assume that the eort exerted by an agent is continuous. This is a crucial
assumption whose importance cannot be understated. It would not possible to provide such
general conditions if eort was dichotomous (i.e., when eort can take only two values, high
or low), as assumed in Hellwig (1983) or Bisin and Guaitoli (2004).
Third, based on the provided welfare properties results of linear price equilibria, it allows
analyzing the motivation for a government to intervene in the insurance market when it is
impossible to impose exclusive contractual insurance relationships.

In the case of CARA utility functions, there always exists a linear price equilibrium. This
is due to the fact that the agents willingness to pay for more insurance is increasing (at least
weakly) in wealth. The existence of a linear price equilibrium property is very robust because
1
Pension annuity markets- markets for insurance against outliving ones resources -are also good examples
of non-exclusive insurance markets, since individuals can purchase several small annuities simultaneously from
dierent providers (see Rothschild (2013)). Although annuity markets have been much studied for adverse
selection problems (e.g., Finkelstein and Poterba (2004)), insured parties could also face a moral hazard problem,
if only through their dietary or health care behavior.
2
In the United States, posted prices are typically linear from a minimum purchase requirement on the order
of $10,000.

2
it does not rely on any assumptions related to how the eort exerted by the agent aects his
accident probability. In the case of DARA utility functions, the agents willingness to pay for
more insurance is not (necessarily) increasing in wealth. In this case, the existence of a linear
price equilibrium relies on how the accident probability faced by the agent is related to the eort
the agent exerts.

Analyzing the case in which the eort exerted by an agent signicantly changes his accident
probability (referred to as the strong moral hazard case),3 we show that there exists a unique
linear equilibrium under which agents are not fully insured, exert a level of eort that is strictly
higher than the lowest possible level, and insurance companies have strictly positive expected
prots. In the traditional Bertrand competition model, there is no strictly positive (expected)
prot in equilibrium because rms could then deviate, make aggressive oers, and take all the
markets. In our model, insurance companies cannot do this because they fear agents taking the
deviant oer, complementing it with other oers (since insurance contracts are not exclusive),
and discontinuously shifting to the lowest level of eort, which would make the deviant oer
unprotable. We thus show that strictly positive (expected) prots in a competitive setting, as
in Hellwig (1983) or Parlour and Rajan (2001), can also exist when the eort exerted by an agent
is continuous and not simply binary.4 A dichotomous eort is thus not a crucial assumption for
(expected) prots to be positive in the case of non-exclusive contracts.

Lastly, we analyze the welfare properties of linear price equilibria. We show that the linear
equilibrium is always third-best ecient. This implies that a laissez-faire policy constitutes
the best choice for a government when imposing the exclusivity of the insurance contracts is
impossible by law.

Related literature: This article is related to other papers that have studied non-exclusive
contracts in a common agency game framework. Common agency games have been a very useful
framework for the analysis of economic issues, particularly in the eld of insurance. Following an
approach close to ours, Ales and Maziero (2009) study the seminal Rothschild and Stiglitz (1976)
adverse selection model when insurance contracts are not exclusive. They show that there exists
only a linear equilibrium in which all insurance companies oer the same menu of contracts.
Rothschild (2013) also studies adverse selection in insurance markets with non-exclusive con-
tracting, but under conditions of compulsory contracting, linear pricing, and multiple indemnity
3
We refer the reader to our web Appendix for the study of the case in which moral hazard does not change
signicantly his accident probability (the weak moral hazard case).
4
Parlour and Rajan (2001) use a common agency games framework to analyze credit relationships. They
obtain strictly positive expected prots, in equilibrium, in the case of a competitive credit market with non-
exclusive contractual relationships in which agents exert a dichotomous eort that aects their credit probability
of default.

3
states. He shows that it is possible to screen dierent types into dierent contracts, despite the
conditions of linear pricing and non-exclusivity.
Our article is, however, more closely related to studies of non-exclusive contracts in insur-
ance markets under moral hazard. Hellwig (1983) also examines the existence of linear price
equilibria.5 However, he only considers isoelastic utility functions and a dichotomous eort. By
studying CARA, and DARA utility functions6 and a continuous eort we generalize Hellwigs
analysis. Kahn and Mookherjee (1998) study a very particular setting in which agents design
their own contracts (insurance companies can only either accept or reject agents oers), make
contractual decisions sequentially, and have contractual portfolios that are observable but not
contractible upon. Both their setting and results are totally dierent from ours. In particular, in
their setting, insurance companies make zero prots and agents always face fair insurance prices
in equilibrium, which is not the case in our framework. Bisin and Guaitoli (2004) consider moral
hazard in the form of a two-eort level.7 They nd that the optimal level of eort can be obtained
in equilibrium. Nevertheless, this type of equilibrium is based on latent contracts (i.e., contracts
that are oered by insurance companies but never subscribed to in equilibrium) that correspond
to negative insurance contracts (i.e., insurance contracts that pay when there is no accident).
Our analytical setting is dierent, in particular because we consider continuous levels of eort.
This leads us to dierent results. Namely, although we do not place restrictions on insurance
company oers, the equilibria in which agents exert a level of eort that is strictly higher than
the lowest level of eort are never associated with (latent) negative insurance contracts. In the
same setting as Bisin and Guaitoli (2004), Attar and Chassagnon (2009) examine the welfare
properties of equilibria. They show that market equilibria may fail to be even third-best ecient,
in contrast to Bisin and Guaitoli (2004), who nd that equilibria are third-best ecient. We
also study the welfare properties of the linear equilibria that we nd. We show that the equi-
librium is always third-best ecient. Attar, Campioni, Chassagnon, and Rajan (2006) restrict
the set of contracts that insurance companies can oer to linear or partially linear contracts.
Thus, in Attar et al, all equilibria are linear per se. In our case, we study the existence of linear
contracts equilibria, that is, situations in which insurance companies oer linear contracts in
equilibrium, even if they have no restrictions on the kind of contracts they can oer. This allows
us to give microeconomic foundations to the literature on competitive non-exclusive insurance
markets, which assumes linear price contracts. Bertola and Koeniger (2015) derive conditions
under which a competitive equilibrium can exist in an insurance market. As in our study, they
5
Hellwig shows that linear equilibria always exist if the exponent of the utility function is nonnegative (i.e., if
relative risk aversion is not greater than one).
6
We refer the reader to our web Appendix for the study of IARA utility functions.
7
Note that Kahn and Mookherjee (1998) and Bisin and Guaitoli (2004) also study credit relationships, unlike
the present paper.

4
consider moral hazard to take the form of a continuous level of eort. However, they assume
that the production of insurance services is exogenously costly. They show that if insurance is
actuarially unfair enough then an equilibrium with positive eort can exist and be characterized
by rst-order conditions, since (local) concavity can be guaranteed by unfair insurance pricing.
We do not assume that the production of insurance is exogenously costly. We thus have to deal
with the non-concavity of the agents objective function. The equilibrium that we nd cannot be
characterized simply by rst-order conditions.

This paper proceeds as follows. Section 2 presents the model. Section 3 studies the linear
price equilibrium with nonexclusivity in the case of CARA utility functions, after which, Section
4 studies DARA utility functions. Section 5 examines the welfare properties of the linear price
equilibrium. Section 6 presents some concluding remarks. Formal proofs are relegated to the
Appendix.

2 The model
We consider a competitive insurance economy derived from Mossins (1968) classic insurance
model. There is a measure one continuum of (ex-ante) identical agents, and a nite number of
identical insurance companies, indexed by i, where i 2 I = f1; : : : ; Ng and N 2. There is free
entry in the insurance market.

2.1 Agents
Each agent faces the possibility of having a xed-damage accident: the amount of damage
is denoted by L. The accident is distributed according to a Bernoulli distribution. There
are two states of the world: one in which there is no accident and one in which the accident
occurs. Accidents to the agents are independently distributed. Without insurance, an agents
consumption corresponds to w if there is no accident, but is w L if there is an accident.
However, each agent can instead subscribe to insurance contracts. Insurance contracts are non-
exclusive (i.e., each agent can buy several contracts from dierent insurance companies) and are
the agents private information. The insured agents consumption if there is no accident is wP ,
P
where P = N i=1 pi corresponds to the sum of all premia pi paid by the agent to the insurance
companies. However, if there is an accident, the insured agents consumption is w L + R P ,
P
where R = N i=1 ri corresponds to the sum of all repayments ri from the insurance companies.
The probability of having an accident (e) depends on the level of eort e devoted by the agent
to avoiding the accident. We assume that 0 (e) < 0 and 00 (e) > 0 and that the inverse function
01 ( ) is dierentiable. We also assume that e 2 [e; e], where e, and e denote, respectively, the
lowest and highest levels of eort that the agent can exert. Eort is unobservable by insurance
companies and costly for the agent. We take the cost function to be linear, that is C(e) = e.

5
Lastly, we assume that the agents utility function is separable in consumption and eort, and
is event-independent (i.e., the accident does not alter preferences). The expected utility of an
agent can be written as

U (w; L; R; P; e) = (e) u (w L + R P ) + (1 (e)) u (w P ) e; (1)

where u(:) is a von Neumann-Morgenstern utility function, with u0 (:) > 0, u00 (:) < 0. We take
000
u(:) to be of class C 3 , with u (:) taking nite values, for nite arguments. Without loss of
generality, we assume that u(0) = 0 and w > L.

2.2 Insurance Companies

There are no restrictions on the kind of contracts (ri ; pi ) that insurance companies can oer.
However, it is impossible to write a contract that is contingent on either an agents eort or on
that agents insurance contracts with other insurance companies because that eort and those
insurance contracts are private information to the agent. The expected prot of an insurance
company associated with an insurance contract (ri ; pi ) is equal to

i (ri ; pi ; e) = pi (e) ri : (2)

2.3 The Contracting Game

Following the common agency literature, we allow insurance companies to oer menus of
contracts. To make the problem tractable, we assume that menus are closed subsets of R2 .
Given the oered menus of contracts, we assume that each agent must choose exactly one
contract from each insurance company.8 Moreover, we assume that an agent can always choose
(ri ; pi ) = (0; 0), which corresponds to no insurance at all (i.e., to not be insured) from company
i. This means that we constrain each insurance company to propose menus that always contain
the contract (0; 0).
Insurance company is set of strategies is denoted by Mi and corresponds to the set of
all closed subsets of R2 containing (0; 0). A generic element of Mi is denoted by Mi . M
denotes a collection of menus oered by all insurance companies, M = i2I Mi , and Mi de-
notes the collection of menus oered by all insurance companies except company i: Mi =
fM1 ; : : : ; Mi1 ; Mi+1; : : : ; MN g. Hence, we have M = (Mi ; Mi ). In the same way, we will use
the notation M = i2I Mi .
For an agent, a pure strategy is a mapping (:) from the set of menus M to an element of
the set M [e; e]. By [pi (M) ; ri (M)] we denote the contract chosen by the agent from the
8
Constraining an agent to choose exactly one contract from each insurance companys menus of contracts is
not restrictive. Indeed, several contracts from the same insurance company can always be pooled into a single
contract.

6
menu oered by insurance company i when facing the collection of menus M. It follows that
(M) = i2I [pi (M ) ; ri (M )] e (M) :

The timing of the game is as follows:

1. Insurance companies simultaneously oer menus of insurance contracts.

2. Each agent privately chooses insurance contracts and the level of eort to be exerted.

3. The accident either does or does not occur and payos are paid accordingly.

We are going to use the concept of subgame perfect Nash equilibrium: given the menus of
contracts issued by the insurance companies, each agent (privately) chooses which contracts to
buy. This determines consumption in the two states of the world (no accident or accident). Each
agent also chooses the level of eort to be exerted. The insurance contracts and the exerted
level of eort are chosen by each agent so as to maximize the expected utility. Anticipating
the choices of agents as a function of the set of contracts that they oer, insurance companies
strategically choose which contracts to issue so as to maximize prots.
In formal terms, the agents equilibrium strategy is a collection of mappings (M ) =
i2I [pi (M ) ; ri (M )] e (M ) such that
! !
X X X

(M ) 2 arg max (e)u w L pi + ri + [1 (e)] u w pi e:
(ri ;pi )i2I 2M ; e2[e;e] i2I i2I i2I
(3)
Then, the equilibrium strategies of each insurance company are dened by

8i 2 I; Mi 2 arg max pi Mi ; Mi
e Mi ; Mi

ri Mi ; Mi : (4)
Mi 2Mi

Conditions (3) and (4) dene an equilibrium in our game.

Moreover, in this article, we are interested in the existence of linear equilibria (i.e., equilibria
in which the ratio of the insurance premium to the repayment in case of an accident is constant
regardless of the chosen repayment). This implies that we are going to examine the existence of
equilibria in which each insurance company i oers menus Mi such that every contract (ri ; pi )
satises pi = ri , where is a positive constant. However, we do not restrict the insurance
companies deviations to be linear. Deviations can be of any type, even nonlinear.

We are now going to study the case of CARA utility functions.

3 CARA Utility Functions

We are now going to show that in our setting with moral hazard, indierence curves are not
(globally) parallel.

7
3.1 Nonparallel Indierence Curves
Consider a given (global) insurance contract (R; P ). The agents indirect utility function, which
we denote by V (R; P ), is computed for e = e , where e corresponds to the eort that maximizes
the agents expected utility. Thus,

V (R; P ) = (e )u (w L + R P ) + (1 (e )) u (w P ) e : (5)

Any indierence curve is characterized by

V (R; P ) = ; with 2 R, (6)

and depends on the agents optimal level of eort.

An agent exerts the level of eort that maximizes his expected utility. Using Equation (1)
and computing the rst-order condition with respect to e, we obtain the optimal level of eort
exerted by an agent when the maximization program admits an interior solution. In this case,
e is dened such that

0 (e) [u (w L + R P ) u (w P )] = 1: (7)

However, there are also two possible corner solutions: e = e or e = e. For instance, if there
is full insurance (R = L), an agents consumption is the same regardless of whether there is an
accident. The agent therefore has no incentive to exert a level of eort higher than e since, for a
given accident probability, the expected utility is decreasing with the eort exerted. Moreover,
for any given P , an agent also exerts e = e when R > L. Suppose now that R < L. Let us
dene R (P ) as the implicit solution of Equality (7) for e = e. Thus, for any given P , R (P ) is
dened such that

0 (e) [u (w L + R (P ) P ) u (w P )] = 1: (8)

In the same way, for any given P , R (P ) is dened such that

0 (e) u w L + R (P ) P u (w P ) = 1: (9)

By construction, we have: R (P ) < R (P ) < L. An agent optimally chooses to exert eort

e < e < e when R (P ) < R < R (P ), while optimally exerting either the highest level of eort
for R R (P ), or the lowest level of eort for R R (P ). Moreover, 00 (e) > 0 implies that e
is unique. Finally, we show in the Appendix (See Proof of Lemma 1) that both R(P ) and R (P )
are strictly increasing in P , since u0 (:) > 0, but u00 (:) < 0.
The following lemma summarizes the previous results.

8
Lemma 1 For any given P 0, we have: e < e < e for R (P ) < R < R (P ), but either e = e
for R R (P ) or e = e for R R (P ). Moreover, both R(P ) and R (P ) are strictly increasing
in P .

Proof. See the Appendix.

Let us now consider the particular case in which an agent is not insured, R = 0 (and thus,
P = 0). Equality (7) implies that the agent optimally chooses to exert eort e < e e if and
only if

1
u (w) u (w L) > : (Condition 1)
0 (e)
Using the fact that R(P ) is strictly increasing in P , we have R (P ) 2 ]0; L] for any positive
premium P 0, when Condition 1 is satised. We are hereafter going to assume that Condition
1 is satised. This condition allows us to simplify the analysis of the model without qualitatively
altering the results.

Lemma 1 implies that the probability of an accident is strictly less than (e) for R < R (P )
but equal to (e) for R R (P ). Consider now a decrease in the premium to pay P , which is
equivalent to an increase in wealth. For a given (R; P ), the dierence in utility when the accident
does or does not occur (u (w L + R P ) u (w P )) decreases when w increases because the
utility function u(:) is increasing and concave. Therefore, using Equality (7), we can deduce that
for a given R, with R < R (P ) (that is, for e < e e), the agents optimal level of eort
decreases when w increases because the marginal gain of eort decreases while the marginal cost
of eort does not change. This implies that for a given R, with R (P ) < R < R (P ), the accident
probability strictly increases when the premium P decreases. Of course, for a given repayment
R R (P ), the agent already exerts the lowest level of eort and the probability of an accident,
which is equal to (e), does not change when P decreases.

Applying the implicit function theorem to Equality (6) (we show in the Appendix- see proof
of Lemma 2 -that any indierence curve is always dierentiable in R), for any R 0, the agents
marginal rate of substitution between R and P is: dP
dR
= VV m
mP
R
with V mR = @V (R; P ) =@R and
V mP = @V (R; P ) =@P . Making computations, we nd that
dP (e )u0 (w L + R P )
= 0; (10)
dR (e )u0 (w L + R P ) + [1 (e )] u0 (w P )
since an agents utility is increasing in R, but decreasing in P .
Lemma 1 implies that for a given repayment R, the accident probability increases, at least
dP 1
weakly, when the premium decreases. Rewriting Equality (10) as dR
= 1 (e ) u0 (wP )
1+ (e ) u0 (wL+RP )

implies that for a given R, with R (P ) < R < R (P ), the agents marginal rate of substitution

9
between R and P , strictly increases when P decreases. Indeed, for a given R, with R (P ) <
1 (e )
R < R (P ), (e )
strictly decreases (since the accident probability strictly increases), while
u0 (wP )
u0 (wL+RP )
does not change for CARA utility functions (since there is no direct wealth eect
through the utility function), when P decreases. This implies that the standard property of
parallel indierence curves associated with CARA utility functions is only valid when either
1 (e )
R R (P ) or R R (P ) (then (e )
is a constant and does not change, for a given R, when
P decreases). This property is no longer true when R (P ) < R < R (P ). Then, the indierence
curves become steeper, ceteris paribus, when the premium P decreases. In other words, the
agents willingness to pay for more insurance is strictly increasing in wealth. Indeed, when P
decreases while R does not change, the agent gets richer. He is then willing to pay more for
insurance, since he anticipates that he his going to make less eort to prevent the occurrence of
the accident. There is thus an indirect wealth eect due to the fact that when the agent becomes
richer he exerts less eort, which modicates his willingness to pay for (more) insurance. This
implies that any indierence curve is located strictly below the translation of an indierence
curve associated to a lower level of utility for low R, and is confounded with it for higher R, the
cuto occurring for R = R (P ), that is, at the boundary between e > e and e = e.
This leads us to the following lemma.

Lemma 2 Consider two points (R1; P1) and (R2 ; P2 ) with R1 < R2 and such that V (R1 ; P1 ) =
V (R2 ; P2 ). Then, if R1 < R (P1) (and either R2 < R (P2 ) or R2 R (P2)), we have V (R1; P1 ) <
V (R2 ; P2 ) for any > 0, which is such that R1 < R (P1 ). However, if R1 R (P1 ), we
have V (R1 ; P1 ) = V (R2 ; P2 ), for any > 0.

Proof. See the Appendix.

Lemma 2 is going to be very useful for studying the existence of linear equilibria. In partic-
ular, it will help us studying an agents optimal choice of insurance, at the deviation stage, that
is, if an insurance company deviates from the equilibrium oers.
Figure 1 highlights rst that both IC1 and IC2 are not concave. Indeed, when R (P ) <
R < R (P ) the optimal level of eort depends on (R; P ) and moral hazard can give rise to non-
concave indierence curves (see Helpman and Laont (1975)). However, when either R R (P )
or R R (P ), e = e (i.e., the optimal level of eort is xed) and the indierence curves are
well-shaped since they are only dened by the properties of the utility function. Moreover,
Figure 1 highlights that IC2 , which is located strictly below IC1 (and which is thus associated
to a strictly higher level of utility than the one associated to IC1 ) is rst strictly below IC1T -
which corresponds to the translation of IC1 -for R < R (P ) and then confounded with it for
R R (P ).

10
 R (P) IC1
P R(P)

IC2

IC1T

L R

Figure 1: Indifference Curves in the

Case of CARA Utility Functions

We can now study the linear equilibrium.

3.2 Linear Equilibrium

Let each insurance company issue a continuum of linear contracts Mi = (ri ; pi ), where pi =
(e) ri . Each agent then faces a continuum of insurance contracts which are located along the
straight half-line P = (e) R in the axes graph (R; P ), starting from (0; 0). Using Equation
10 we nd that MRSagent (R = L) = (e). This implies that there exists one indierence
curve which is tangent to P = (e) R, for R = L. Let us denote by IC this indierence
curve. IC corresponds to all the pairs (R; P ) which are such that the indirect utility function
V (R; P ) = u (w (e) L) e u. Two possibilities should be distinguished due to the non-
concavity of the indierence curves. Either IC does not intersect P = (e) R for R < R (P ).
We dene this case as the Weak Moral Hazard case. Or IC has at least one other intersection
with P = (e) R for R < R (P ). We dene this case as the Strong Moral Hazard case.

Consider rst the Weak Moral Hazard Case. Then we show in a web Appendix that there
always exists a unique linear equilibrium. This equilibrium is actuarially fair and there is full
insurance.9
Consider now the Strong Moral Hazard case. Then, IC has at least one other intersection
with P = (e) R for R < R (P ). This implies that the full insurance actuarially fair insurance
equilibrium no longer exists. Indeed, an agent that faces a continuum of linear contracts P =
(e) R no longer chooses to be fully insured. There is another repayment with R < R (P ) ( L)
9
Note that this result could be interpreted as a result by continuity of the case in which there is no moral
hazard. Then, a risk averse agent chooses to be fully insured in equilibrium when facing actuarially fair insurance
prices.

11
that the agent strictly prefers to R = L, as is illustrated in Figure 2.

P (e) = p (e) R
R (P)
R(P )

P(e ) = p (e ) R

Figure 2: Contract C' is preferred

to (R = L; P = p (e)L)

One question is: When does the Strong Moral Hazard Case exist? Intuitively, the answer is:
Consider that (e) is signicantly strictly lower than (e). Then, a risk averse agent can choose
to exert a level of eort that is strictly higher than e in order to decrease signicantly the accident
probability. Moreover, IC is so badly shaped that it intersects P = (e) R. Formally, we obtain
the following result (Note that the threshold value T (e) which is considered in Lemma 3 is
formally dened in the Appendix).

Lemma 3 Let T (e) 2 [0; 1] be a threshold value that is dened for any level of eort e 2 [e; e].
Then, IC intersects P = (e) R for R < R (P ), if and only if there is at least one level of eort
e 2 ]e; e] for which (e) < T (e).
Moreover, when IC intersects P = (e) R for R < R (P ), there is a unique value, denoted
, for which the straight half-line P = b
b R admits one indierence curve tangent to it both for
R < R (P ) and for R > L.

Proof. See the Appendix.

Consider now the particular value b

for which there exists one indierence curve tangent to
bR both for R = R1 < R (P ) and for R = R2 > L. This implies that an agent optimally
P =
chooses R 2 fR1 ; R2 g when each insurance company oers a continuum of linear contracts
Mi = (ri ; pi ), with pi = b
ri . Could the situation in which each insurance company oers a
continuum of linear contracts Mi = (ri ; pi ), where pi = b
ri and the agent chooses to be only
10
partially insured (R1 < R (P )) be an equilibrium?
10
Note that being over insured (i.e., R2 > L) cannot be an equilibrium since the expected prot of an insurance
company would then be strictly negative. Indeed, with a global insurance coverage R2 > L the agent would exert
the lowest level of eort (e) and the probability of the accident ( (e)) would be strictly higher than the unit
b.
price of insurance

12
 Observe rst that the expected prot of each insurance company is strictly positive. In
order to see that, let us constraint the agent to choose a level of eort e 1 (b
). Given
b. Then, the agents optimal choice of
this constraint, the accident probability is (e)
insurance is characterized by R L. Hence, we can state that the agent exerts a level of eort
e e > 1 (b
) when only partially insured. This implies that the expected prot of each
insurance company is strictly positive.
Now, in order to determine the existence of a linear equilibrium, knowing an agents optimal
insurance choices (R = R1 < R (P )), we have to show that, taking into account any kind of
deviation (that is, not only linear deviations), there are no protable deviations for any insurance
company. However, it is (almost) impossible to do so since the set of all insurance companies
strategies, Mi 2 Mi , is a set of closed subsets of R2. The following lemma shows that it is not
restrictive to consider only single oers at the deviation stage.

Lemma 4 If an insurance company has no protable deviations toward a single oer, then no
deviations toward any menus of contracts are protable.

Proof. Suppose the contrary. Company i has no protable single oer deviations, but there is
a more sophisticated menu of contracts, denoted by Mi , which gives it a higher payo. Then, if
the agent buys contract (pi ; ri ) in Mi , oering Mi or simply the menu f(pi ; ri )g are equivalent
for company i. Indeed, in both cases, the agent chooses contract (pi ; ri ) eventually. Finally, if
the agent does not buy any contract in Mi , the single contract (0; 0) is equivalent to Mi for
company i. Therefore, considering only single oers at the deviation stage is not restrictive.

Using Lemma 4 consider an insurance company that deviates and issues a single contract
pd
Cd = (rd ; pd ) with rd
<b
and rd < R (P ). The agent faces a new continuum of linear contracts
that belongs to the straight half-line P 0 (R), starting from Cd and parallel to but strictly below
bR. For R R(P ) all indierence curves are increasing at a decreasing rate. Moreover,
P =
MRSagent (R = L) = (e) > b
implies that there exists only one indierence curve which is
tangent to P 0 (R) for R > L. We denote this indierence curve by IC (P 0 (R)). There are
two possibilities. Consider rst that IC (P 0 (R)) does not intersect R (P ). Then, the agent
optimally chooses R > L. Consider now that IC (P 0 (R)) intersects R (P ). Let ICbT be the
translation of ICb tangent to P 0 (R) at both points (R1; P1 ) and (R2; P2 ). Lemma 2
implies that ICbT and IC (P 0 (R)) are confounded one with the other for R R (P ). However,
for R < R (P ) the indierence curves become steeper when P decreases which implies that
IC (P 0 (R)) is located strictly below ICbT . Therefore, IC (P 0 (R)), which is the only indierence
curve tangent to P 0 (R) for R > L (at (R2 ; P2 )), neither intersects nor is tangent to P 0 (R)
for R L. Therefore, if an insurance company deviates and issues a contract Cd , the agent buys
this contract and completes his insurance coverage until reaching a global insurance contract C2

13
under which R > L. The agent thus exerts the lowest level of eort e = e and the expected
prot of the deviating insurance company is negative.
Therefore, there exists a linear pricing equilibrium which is actuarially unfair and induces
partial insurance coverage. Moreover, we show in the Appendix that this equilibrium is the
unique equilibrium which exists.
We thus have the following proposition.

Proposition 1 There exists a unique linear equilibrium. This equilibrium is actuarially unfair
and induces partial insurance coverage. Each insurance company oers a continuum of linear
contracts Mi = (ri ; pi ), where pi = b
ri corresponds to the unique unit price of insurance for
which the agent optimally chooses to be either partially insured or over insured. In equilibrium,
each agent chooses to be only partially insured and exerts a level of eort e , which is such that
(e ) <
b.

Proof. See the Appendix.

The intuition of the previous proposition is the following. In the partial insurance equilibrium,
the agent is just indierent between exerting the lowest level of eort and a strictly higher level
of eort. If an insurance company deviates and makes a "more competitive" oer, the agent
then completes the deviation oer (since contracts are not exclusive) and exerts the lowest level
of eort. This makes Bertrand competition ineective.
Proposition 1 is illustrated in Figure 3. The global insurance contract C2 that the agent would
buy if an insurance company deviated and issued contract Cd is also represented in Figure 3.

P = p (e ) R
R (P )

R (P)
P = g R

P' ( R)

C2

Cd P = p (e ) R

Figure 3: Linear equilibrium and global insurance

contract C2

We are now going to illustrate the equilibrium we have just found.

3.3 Illustration: IC intersects P = (e) R

Assume that the agents consumption if there is no accident is w = 20 and that the xed-damage
1
accident is L = 18. Moreover, assume that (e) = 1+Be
, with B 2 R+ and e 2 [0; 3]. We

14
thus have 0 (e) < 0 and " (e) > 0. Lastly, assume that the agents von Neumann-Morgenstern
utility function is u(A) = 9:000016 exp (A), where A 2 R+ . The expected utility of an agent
is thus:
!
1
1+Be
[9:000016 exp ( (20 18 + R P ))]
U (20; 18; R; P; e) = 1
:
+ 1 1+Be [9:000016 exp ( (20 P ))] e
Equation 7 leads us to
p
1
e = min 3; max B (exp ( (20 18 + R P )) exp ( (20 P ))) 1 ; 0 :
B
Moreover, (0) = 1. This implies that the continuum of actuarially fair linear contracts associ-
ated to the lowest level of eort e = 0 is P = R.
Consider that B = 10; 000. Then, the agents optimal eort is represented in Figure 4.
Figure 5, represents IC (that is, V (R; P ) = 8:86468).

P P IC P=R

R (P)

e* R(P)

R
R
Figure 4: An agent's optimal effort, when B = 10,000
Figure 5: IC does intersect P = R for R < 18

IC clearly intersects P = R for R < 18. In equilibrium, an agent is going to be only partially
insured and insurance (linear) prices are going to be actuarially unfair. Figure 6 shows the
intersection between IC and P = R.
P P=R

IC

R (P)

R
Figure 6: Intersection between IC and P = R

15
Moreover, Figure 6 shows that IC is not concave (and even is locally convex). Note that for
B = 10; 000, the marginal gain of eort is (very) important. Therefore, the agent exerts an eort
e > 0 even for a repayment close to R = 18. This explains why R(P ) is located very close to
the vertical line R = 18 and IC intersects P = R for a repayment very close to R = 18.11

We are now going to consider the case of DARA utility functions. Note that Lemma 3 also
applies in the case of DARA utility functions.

4 DARA Utility Functions

For DARA utility functions (as for CARA utility functions), for a given repayment R, the
accident probability increases, at least weakly, when the premium decreases.12 However, this
does not directly imply that the agents marginal rate of substitution between R and P increases,
at least weakly, when P decreases (or equivalently that the agents willingness to pay for more
insurance is strictly increasing in wealth), since there is a direct wealth eect trough the utility
function, which must be taken into account. In other words, Lemma 2 does not necessarily hold
for DARA utility functions.

Consider rst that R R (P ). Then, the agent exerts the lowest level of eort e, and
the probability of accident is xed and equal to (e). The indierence curves thus exhibit
the standard properties of DARA utility functions: for a given R, the agents marginal rate of
substitution for more insurance decreases when P decreases, since a decrease in P is equivalent to
an increase in w, and the agent becomes less risk averse with increasing wealth. This implies that
for R > L, indierence curves become steeper, ceteris paribus, when the premium P decreases,
since for R > L, the agent increases the insurance coverage by choosing to reduce (and not
to increase) the repayment in case of accident. Moreover, for R (P ) R < L, indierence
curves become atter, ceteris paribus, when the premium P decreases, since for R < L the
agent increases his insurance coverage by choosing to increase the repayment in case of accident.
However, when the accident probability is suciently decreasing with the level of eort for e = e
11
Note that even when B = 100, IC intersects P = R. However, for B = 100, ceteris paribus, the agent exerts
a lower level of eort than for B = 10; 000, which implies that the intersection of IC with P = R is for a lower
repayment when B = 100 than when B = 10; 000.
Note however, that for B = 10, IC does not intersect P = R. This corresponds to a case of Weak Moral
Hazard (See the Web Appendix).
12
Indeed, for a given (R; P ), the dierence in utility when the accident does or does not occur
(u (w L + R P )u (w P )) also decreases with DARA utility functions, when w increases because the utility
function u(:) is increasing and concave. Therefore, using Equality (7), for a given R, with R (P ) < R < R (P ),
the agents optimal level of eort strictly decreases when w increases because the marginal gain of eort decreases
while the marginal cost of eort does not change. This implies that for a given R, with R (P ) < R < R (P ), the
accident probability strictly increases when the premium P decreases.

16
(i.e., when j 0 (e)j is high enough), the marginal gain of eort is very important and the agent
exerts a level of eort that is strictly higher than e even when R is close to L. In the limit, when
lim 0 (e) = 1, then R (P ) = L; 8P . Thus, when j 0 (e)j increases, the interval R (P ) R < L
e!e
shrinks and for j 0 (e)j high enough it almost disappears.
For R R (P ), the agent optimally chooses to exert eort e = e. Thus, for R R (P ), the
probability of accident is xed and equal to (e). The indierence curves exhibit the standard
properties of DARA utility functions, and become atter, ceteris paribus, when the premium P
decreases. However, when the marginal gain of eort is very low when the agent exerts a level
of eort close to e (i.e., when lim 0 (e) = 0) the agent always exerts a level of eort e < e since
e!e
we always have: 0 (e) [u (w L + R P ) u (w P )] < 1 8R 0.
Finally, consider that R (P ) < R < R (P ). The agent then exerts a level of eort e < e < e.
This implies that when P decreases while R remains constant, there are two countervailing eects
which aect the marginal rate of substitution. The direct wealth eect goes for a decrease in the
marginal rate of substitution. However, the agent exerts a lower level of eort, which increases
the probability of an accident. This indirect wealth eect goes for an increase in the marginal rate
@M RSagent (R;P )
of substitution. Using the fact that dP
= VV mR
, we nd that < 0, or, equivalently,
dR mP @P
V mR Vm
@ V m @ V mR
@M RSagent (R;P ) @e
@w
> 0, if and only if @w
P
[i; e; : direct wealth effect]+ @e
P
@w
[i; e; :indirect
@e
wealth eff ect] > 0. Making computations (in particular, computing @w
), and rearranging
terms, we obtain that the previous inequality is necessarily satised if and only if, 8(w; L; R; P ):

[( 0 (e ))3] 1
> k; (Regularity Condition)
00 (e ) (e ) [1 (e )]

where k 2 R+ and takes a nite value. The intuition of Regularity Condition is the following.
First, the indirect wealth eect is going to dominate the direct wealth eect if the eort has
a signicant eect on the accident probability, that is if j 0 (e)j is high enough. Moreover, an
increase in wealth (a decrease in P ) signicantly aects the eort the agent exerts (and makes
the indirect wealth eect important) if the marginal eect of eort on the accident probability
is rather constant whatever the eort exerted, that is, if 00 (e) is low. Indeed, ceteris paribus,
de
is going to be high if 00 (e) is low.
dw

Therefore, when Regularity Condition is satised, lim 0 (e) = 1 and lim 0 (e) = 0,
e!e e!e
for a given R, when P decreases, the marginal rate of substitution increases for any R 0
(or equivalently, the agents willingness to pay for more insurance strictly increases in wealth).13
This implies that any indierence curve is located strictly below the translation of an indierence
curve associated to a lower level of utility. This leads us to the following lemma.
13
Note that the two conditions: lim 0 (e) = 1 and lim 0 (e) = 0, can be interpreted as the (so-called)
e!e e!e
Inada conditions in our setting.

17
Lemma 5 Consider two points (R1; P1) and (R2 ; P2 ) with R1 < R2 and such that V (R1 ; P1 ) =
V (R2 ; P2 ). When Regularity Condition is satised, j 0 (e)j is high enough and j 0 (e)j is low
enough, we have V (R1 ; P1 ) < V (R2 ; P2 ) for any > 0.

Proof. See the Appendix.

Let = b
denote the unique unit price of insurance for which an agent facing a continuum
of linear contracts P = R optimally chooses R 2 fR1 ; R2 g with R1 < R (P ) and R2 > L.
(R1 ; P1 ) corresponds to the candidate equilibrium. The expected prot of an insurance company
b > (e (R1 ; P1 )). Consider that
associated to (R1; P1) is strictly positive, since we have
pd
an insurance company deviates and issues a single contract Cd = (rd ; pd ), where rd
< b
and
rd < R (P ). An agent thus faces a new continuum of linear contracts that belong to the straight
half-line, P 0 (R), starting from Cd and parallel to but strictly below P = b
R. For R R (P ) all
indierence curves are increasing at a decreasing rate. Moreover, MRSagent (R = L) = (e) > b

implies that there exists one and only one indierence curve tangent to P 0 (R) for a repayment
R > L. We denote this indierence curve by IC (P 0 (R)). Let ICbT denote the translation of the
indierence curve which is tangent to P = b
R for a repayment in case of accident R 2 fR1; R2 g.
ICbT is thus tangent to P 0 (R) at both points (R1 ; P1 ) and (R2 ; P2 ). Lemma 5 implies
that the indierence curve that goes through (R2 ; P2 ) is necessarily steeper than ICbT and
intersects P 0 (R). This implies that the indierence curve tangent to P 0 (R) is tangent for a
repayment R > R2 (i.e., the tangent point is located to the right of (R2; P2 )). Thus, the
indierence curve tangent to P 0 (R) is located strictly below ICbT for a repayment R = L.
Moreover, if Regularity Condition is satised, j 0 (e)j is high enough and j 0 (e)j is low enough,
Lemma 5 implies that the indierence curve tangent to P 0 (R) for R > L is located strictly below
ICbT for R L. Therefore, if an insurance company deviates and issues a contract Cd , the agent
buys this contract and completes his insurance coverage until being overinsured (R > R2 > L).
The agent thus exerts the lowest level of eort, which implies that the expected prot of the
deviating insurance company is negative.
This leads us to the following proposition.

Proposition 2 For any DARA utility function satisfying Regularity Condition, when j 0 (e)j is
high enough and j 0 (e)j is low enough, there exists a unique linear equilibrium. In equilibrium,
the expected prot of any insurance insurance company is strictly positive and each agent is only
partially insured.

Finally and to sum up, in the case of DARA utility functions, Regularity Condition associated
to both j 0 (e)j high enough and j 0 (e)j low enough all together constitute sucient conditions
to guarantee that a linear equilibrium exists.

18
 We are now going to illustrate the sucient conditions when U (w; L; R; P; e) = (e) ln (w L + R P )
(1 (e)) ln (w P ) e.

4.1 Illustration: The sucient conditions with U (w; L; R; P; e) =

(e) ln (w L + R P ) + (1 (e)) ln (w P ) e
First of all, note that with U (w; L; R; P; e) = (e) ln (w L + R P ) +(1 (e)) ln (w P )
[ 0 (e )]3
e, Regularity Condition simply becomes 00 (e )
> 1 (See Proof of Lemma 5). Moreover, let us
[ 0 (e)]3 1
dene the function f (e) 00 (e) (e)[1 (e)]
. We are now going to consider three dierent cases.
1

Case 1: Assume that (e) = ln 100000000e
+ 1 . Then, e 2 [5:81977 109 ; +1[. Indeed,
(e = 5:81977109 ) = 1 and lim (e) = 0. Moreover, 0 (e = 5:81977109 ) = 1:62378109
e!+1
(very high) and lim 0 (e) = 0. Plotting f(e), we obtain:
e!+1

f(e)

Figure 7 : Plotting f(e) for p(e) = ln((1/(100000000e))+1)

And, f(e) < 1 for e > 0:0000707107. Therefore, in order for Regularity Condition to be satised,
we must have (using the FOC for eort):

0 (e = 0:0000707107) [ln (w L + R P ) ln (w P )] 1
RL
,w 1 + P with R L, since e = e, for R L.
exp 1:9972 1
In other words, if f (e) is not such that: f(e) k 8e 2 [e; e] (with k = 1, for U (w; L; R; P; e) =
(e) ln (w L + R P )+(1 (e)) ln (w P )e), Regularity Condition is going to be veried
if and only if w is high enough. Indeed, when w increases the incentives to exert eort decreases
(ceteris paribus, u (w L + R P ) u (w P ) ! 0 when w ! +1). Thus, for w high enough,
we necessarily have e < eT where eT corresponds to the threshold value of eort dened as
f(eT ) = 1, and Regularity Condition is necessarily satised.
RL
The sucient conditions are thus veried for w 1
exp( 1:9972
+P , since Regularity Condition
)1
then holds, 0 (e) is high enough and lim 0 (e) = 0.
e!+1

19
 1
Case 2: Assume that (e) = 100000000e
. Then, e 2 [1 108 ; +1[. Indeed, (e = 1
108 ) = 1 and lim (e) = 0. Moreover, 0 (e = 1 108 ) = 1 108 (very high) and
e!+1
lim 0 (e) = 0. Plotting f(e), we obtain:
e!+1

f(e)

Figure 8 : Plotting f(e) for p(e) = 1/(100000000e)

And, f (e) < 1 for e > 0:0000707157. In this case also, we dont have f (e) 1 8e 2 [e; e].
RL
However, Regularity Condition is going to be veried if and only if: w 1
exp( 0:9999
+P with
)1
RL
R L. This implies that the sucient conditions are veried for w 1
exp( 0:9999
+ P, since
)1
0 0
Regularity Condition then holds, (e) is high enough and lim (e) = 0.
e!+1

1
Case 3: Assume that (e) = 100000
exp[100000(e2:75)]. Then, e 2 [2:7498848707453503; +1[.
Indeed, (e = 2:7498848707453503) = 1 and lim (e) = 0. Moreover, 0 (e = 2:7498848707453503) =
e!+1
100 000 (high enough) and lim 0 (e) = 0. Plotting f(e), we obtain:
e!+1

f(e)

Figure 9 : Plotting f(e) for p(e) = (1/(100000)) exp[-100000(e - 2.75)]

And, f(e) < 1 for e > 2:75. Thus, here also, we dont have f(e) 1 8e 2 [e; e]. However,
RL
Regularity Condition is going to be veried if and only if: w exp(100000)1
+ P with R L.

20
 RL
This implies that the sucient conditions are veried for w exp(100000)1
+ P , since Regularity
0 0
Condition then holds, (e) is high enough and lim (e) = 0.
e!+1

We are now going to examine the Third-Best eciency of the linear equilibrium in the case
of strong moral hazard. We refer the reader to the web Appendix for the study of the Third-Best
Eciency of linear equilibrium in the case of weak moral hazard.

5 Third-Best Eciency of Linear Equilibria

Looking at Third-Best eciency amounts to consider that a social planner cannot neither control
the agents eort, nor restrict insurance companies from oering further contracts. This seems
to be a natural scenario to investigate in all those situations in which a social planner is in the
impossibility of enforcing exclusivity clauses.
Consider the unique linear equilibrium, which is actuarially unfair and induces partial insur-
ance coverage. Assume that this equilibrium is not Second-Best ecient.14 This implies that
a social-planner who cannot control agents eort choices may perform better than markets as
long as he retains the power of controlling insurance contracts trades (i.e., restore the exclusivity
of the insurance contracts). This case is illustrated in Figure 10: Any point in area A is strictly
preferred to the partial insurance equilibrium, both by the agent and by the insurance company.

Iso-expected-profit P =p R
curve
P R (P) R(P )
IC

Partial Insurance Equilibrium:

R < L; P = R.

L
R

Figure 10: The Partial Insurance Equilibrium

is not Second-Best Pareto Optimal

Let us now show that this partial insurance equilibrium is necessarily Third-Best ecient.
Consider that the social planner oers a contract CSP , which belongs to area A. The agent
can complement this contract since the social planner cannot control insurance contracts trades.
The agent faces a new continuum of linear contracts that belongs to the straight half-line P 0 (R),
starting from CSP and parallel to but strictly below P = R. Using either Lemma 2, in the case
of CARA utility functions, or the sucient conditions, in the case of DARA utility functions,
14
We refer the reader to the web Appendix, in which we also study the Second-Best eciency of the linear
equilibria.

21
it is easy to show that if the social planner oers contract CSP the agent buys this contract
and completes his insurance coverage until reaching a global insurance contract characterized by
R > L (See Figure 11). This implies that following the oer CSP , the agents expected utility
belongs to an indierence curve strictly higher than the one associated to the partial insurance
equilibrium. However, the expected prot of the insurance companies (globally) decreases. This
implies that the partial insurance equilibrium is Third-Best ecient. Indeed, any oer by the
social planner which would be preferred both by the agent and by the insurance companies,
leads to a situation (since insurance contracts are not exclusive) in which the agent is strictly
better-o, but the insurance companies are strictly worse-o.

Iso-expected-profit P =p R
curve
P (R)
P R (P) R(P ) R(P )
IC

Optimal choice of the agent if

the social planner offers CSP
Partial Insurance Equilibrium:
R < L; P = R.

CSP

L L
R

Figure 10: The Partial Insurance Equilibrium

is not Second-Best Pareto Optimal

Finally, if the partial insurance equilibrium is Second-Best ecient (which we cannot rule out,
see the web Appendix), it is also necessarily Third-Best ecient. We thus obtain the following
proposition.

Proposition 3 The partial insurance equilibrium is Third-Best ecient.

Proposition 3 implies that a laissez-faire policy is the best choice for a government, when it
is impossible to impose the exclusivity of insurance contracts by law.

6 Conclusion
In this article, we use a common agency game framework to study linear price equilibria in a
competitive insurance market in which contracts are non-exclusive, agents are subject to moral
hazard, and insurance companies do not face any restrictions on their pricing rules.
We show that a linear equilibrium always exists and is unique in the case of CARA utility
functions, because the agents marginal willingness to pay for more insurance is increasing in
wealth. For DARA utility functions, we characterize sucient conditions under which a linear
equilibrium exists and is then unique. We show that these sucient conditions rely on properties
of the function that links the probability of an accident to the eort exerted by the agent.

22
 We also show that the unique linear equilibrium is such that each agent is only partially
insured, exerts a level of eort that is strictly higher than the lowest level of eort and each
insurance companys expected prot is strictly positive.
These results provide microeconomic foundations for any study that restricts insurance com-
panies to oer only linear contracts. For CARA utility functions, the linear equilibrium existence
property is very robust. A linear equilibrium can thus always be taken as granted. The same
holds for DARA utility functions only when sucient conditions are satised.
We also provide a welfare analysis of the linear equilibrium. We show that the linear equi-
librium always is third-best ecient. This implies that a laissez-faire policy is the best choice
for a government, when it is impossible to impose the exclusivity of insurance contracts by law.

A Proof of Lemma 1
Applying the implicit function theorem to Equalities (8) and (9), we obtain

dR (P ) u0 (w L + R (P ) P ) u0 (w P ) dR (P ) u0 w L + R (P ) P u0 (w P )
= > 0; = >0
dP u0 (w L + R P ) dP u0 (w L + R P )
(11)
since u0 (:) > 0, but u00 (:) < 0. This implies that both R (P ) and R (P ) are strictly increasing in
P.

B Proof of Lemma 2
Dierentiability of the indierence curves in R
Consider rst that R > R (P ). Any indierence curve is increasing at a decreasing rate be-
cause the level of eort is xed (e = e). The indirect utility function V (R; P ) is thus dieren-
tiable because the utility function U (w; L; R; P; e) is dierentiable in R and P . Any indierence
curve is thus dierentiable in R for R > R (P ).
Consider now that R (P ) < R < R (P ). For a given fw; L; R; P g, the rst-order condition
(i.e., Equality 7) admits a unique solution e , with e > e, since 0 (e) is strictly increasing.
Moreover, (e), 01 ( ) and U (w; L; R; P; e) are dierentiable functions. V (R; P ) is thus dif-
ferentiable. Any indierence curve is thus dierentiable in R for R (P ) < R < R (P ).
Finally, consider that R = R (P ). The indirect utility function is left- and right-hand side
dierentiable at R = R (P ). Using Equality 6 and applying the implicit function theorem to the
left and to the right of R = R (P ) we obtain
2 3
@U(w;L;R;P;e ) @U(w;L;R;P;e ) @e
dP @R
+ @e
1 @R
lim = lim 4 5;
R!R(P ) dR R!R(P ) @U(w;L;R;P;e )
+ @U(w;L;R;P;e ) @e
1 @P
@P @e

23
and " @U(w;L;R;P;e) #
dP @R
lim + = lim + @U(w;L;R;P;e) :
R!R(P ) dR R!R(P )
@P
@U(w;L;R;P;e )
The FOC implies that @e
1 = 0 for e = e (cf. Equality 7) and lim e = e. Thus,
R!R(P )

@U (w;L;R(P );P;e)
dP dP @R
lim
dR
= lim + dR
= @U (w;L;R(P );P;e)
:
R!R(P ) R!R(P )
@P

Thus, any indierence curve is also dierentiable in R at R = R (P ).

Using the same method for R < R (P ) and for R = R (P ), we can show that any indierence
curve is dierentiable both for R < R (P ) and at R = R (P ), where R (P ) is the implicit solution
of Equality (7) for e = e.
By combining the results of the previous cases we obtain that any indierence curve is
dierentiable in R for R 0.

Nonparallel Indierence Curves

For a given R, the marginal rate of substitution is decreasing in P if and only if @M RSagent (R;P )
@P
<
@M RSagent (R;P )
0 or equivalently if @w
> 0, since P and w play the same role in the agents utility
@M RSagent (R;P )
function. Using Equality (10), we have @w
> 0, if and only if,

u00 (w P ) 0 (e ) de u00 (w L + R P )
+ > : (12)
u0 (w P ) (e ) [1 (e )] dw u0 (w L + R P )
de
Consider any (R; P ) with R (P ) < R < R (P ) (< L), using Equation (7) we have: dw
=
0 0 (wL+RP )u0 (wP )] 0 (e ) de
"(e(e)[u
)[u(wL+RP )u(wP )]
< 0. This implies that (e )[1 (e )] dw
> 0. Thus, Inequality 12 is
satised for CARA utility functions for which the Arrow-Pratt measure of absolute risk-aversion
00 00
is constant, implying that uu0 (wP
(wP )
)
= uu0 (wL+RP
(wL+RP )
)
. Thus, for a given repayment R with
R (P ) < R < R (P ), the indierence curves becomes (strictly) steeper when P decreases.
de
Consider now any (R; P ) with R R (P ). Then, e = e and dw
= 0. Inequality 12 is not
@M RSagent (R;P ) 00 00
satised. We then have @w
= 0 since uu0 (wP
(wP )
)
= uu0 (wL+RP
(wL+RP )
)
. The indierence
curves are then parallel with each other. The same holds for R R (P ).

Consider now two points (R1; P1) and (R2; P2) with R1 < R2 and V (R1; P1) = V (R2; P2).
@M RSagent (R;P )
Suppose that R1 < R (P1). For a given R, we have @P
< 0 when R (P ) < R <
@M RSagent (R;P )
R (P ), but @P
= 0 when either R R (P ) or R R (P ). Thus, for any given R,
MRSagent increases (at least weakly) when P decreases. Let IC (R2 ; P2 ) denote the indier-
ence curve going through (R2 ; P2 ) (and (R1; P1)) and IC T (R2 ; P2 ) denote the translation
of IC (R2 ; P2 ) going through (R2 ; P2 ). By construction, IC T (R2; P2 ) also goes through
(R1 ; P1 ). Let now IC (R2 ; P2 ) denote the indierence curve going through (R2 ; P2 ).

24
 @M RSagent (R;P )
For R R (P ), @P
= 0 implies that IC (R2 ; P2 ) and IC T (R2 ; P2 ) are con-
founded. However, for R < R (P ), IC (R2 ; P2 ) is located strictly below IC T (R2 ; P2 )
@M RSagent (R;P )
since @P
< 0 for R (P ) < R < R (P ). Thus, for any > 0 that satises R1 <
R (P1 ), (R1 ; P1 ) and (R2 ; P2 ) dont belong to the same indierence curve. (R1 ; P1 )
belongs to an indierence curve associated to a strictly lower level of expected utility than the
one (R2 ; P2 ) belongs to, since utility is strictly increasing in R but strictly decreasing in P .
Thus, V (R1 ; P1 ) < V (R2 ; P2 ).
Suppose now that R1 R (P1 ). Then V (R1 ; P1 ) = V (R2; P2 ) since indierence
curves are parallel for any R R (P ).

C Proof of Lemma 3
Part I: When does IC intersect P = (e) R for R < R(P )?

Denition of T (e)
Let us constraint the agent to exert a particular level of eort, which we denote ep , with
ep 2 ]e; e[.
Consider now the constrained indierence curve for which the agent is forced to exert the
level of eort ep and that is associated with the level of utility u u (w (e) L) e. This
indierence curve, that we denote by CIC (ep; (ep) ; u), is dened such that

(ep ) u (w L + R P ) + (1 (ep )) u (w P ) ep = u: (13)

CIC (ep ; (ep ) ; u) is well-shaped and, depending on the value (ep ), may or may not intersect
the straight half-line P = (e) R. There is thus necessarily a particular value (ep ) T for

which there exists a reimbursement value RT , with RT < L, such that CIC ep ; (ep ) = T ; u
is tangent to P = (e) R. Characterizing the tangent point to P = (e) R amounts to nding

the pair RT ; T that satises:

(
T u w L + RT (e) RT + 1 T u w (e) RT ep = u;
(System 1)
M RSagent CIC ep ; T ; u computed at R = RT = (e) :

Doing the same for all e 2 ]e; e], dening T = (e) = , where 2 ]0; 1[ for e = e, we have
built a function T (e) (as well as RT (e), with by construction, RT (e) R(P )) for any e 2 [e; e],
since for e = e, CIC (e; (e) ; u) is tangent to P = (e) R for RT = L.15
15
Note that T (e) is decreasing in e. Indeed, when the eort increases the rst equality of System 1 implies
that T should decrease so as to have the LHD equals the RHS of the equality because at the tangent point we
necessarily have R < L for any e 2 ]e; e].

25
 Now, consider the unconstrained indierence curve associated to T (e) and u. This particular

indierence curve, denoted IC T (e) , is, by construction, confounded with a segment of P =
(e) R for e 2 [e; e].

IC intersecting P = (e) R for R < R(P )

The proof is conducted in three steps.

Step 1: Constrained indierence curves

Consider that (e) > T (e) 8e 2 ]e; e[. For R = RT (ep ), the rst equality of System
1 can only be satised if P decreases. This implies that CIC (ep ; (ep ) ; u) is strictly below
CIC(ep ; T (ep ) ; u) and does not intersect P = (e) R. Therefore, when (e) > T (e), 8e 2
]e; e] none of the constrained indierence curves intersect P = (e) R.
Consider now that for ep 2 ]e; e[ (ep ) < T (ep ). For R = RT (ep ), the rst equality of
System 1 can only be satised if P strictly increases. CIC (ep; (ep) ; u) is thus located strictly
above CIC(ep ; T (ep ) ; u) and intersects P = (e) R. Therefore, (at least) one constrained
indierence curve CIC (ep ; (ep ) ; u) intersects P = (e) R, if (ep ) < T (ep ) for a level of
eort ep 2 ]e; e[.

Step 2: Comparing the unconstrained indierence curve IC to CIC (ep ; (ep ) ; u)

Suppose that IC intersects one constrained indierence curve CIC (ep ; (ep ) ; u) at the point
(Rinter ; Pinter ). This implies that IC goes through a point- denoted by (Rbelow ; Pbelow ) -that be-
longs to a constrained indierence curve associated with the same level of eort but located
strictly below CIC (ep; (ep) ; u). We thus have a contradiction. Indeed, any constrained indif-
ference curve located strictly below CIC (ep ; (ep ) ; u) and associated to the same level of eort
is necessarily associated to a level of utility strictly higher than u. Thus, (Rinter ; Pinter ) and
(Rbelow ; Pbelow ) cannot both belong to IC. Following the same reasoning for any level of eort
e 2 [e; e], we see that IC cannot intersect any constrained indierence curve associated with a
level of utility u.

Step 3: Using Steps 2 and 3

If, 8e 2 [e; e], (e) > T (e), none of the constrained indierence curves CIC (ep ; (ep ) ; u)
intersects P = (e) R (cf. Step 1). Thus, IC, which is (by construction) built from points
that belong to the constrained indierence curves CIC(ep ; (ep ) ; u), does not intersect P =
(e) R. However, if there is at least one level of eort ep 2 ]e; e[ for which (ep ) < T (ep )
( (ep ) = T (ep )), then, there is at least one constrained indierence curve that intersects (is
tangent to) P = (e) R for the level of eort ep (cf. Step 1). Moreover, IC never intersects a
constrained indierence curve CIC (ep; (ep) ; u) (cf. Step 2). Thus, IC intersects (is tangent
to) P = (e) R for the level of eort ep 2 ]e; e[, that is for R < R(P ).

26
 Part II: Uniqueness of the value , for which the straight half-line P = R admits
one indierence curve tangent to it both for R < R (P ) and for R > L.

Consider now that IC intersects P = (e) R for R < R(P ). Let us rst show that there is
no indierence curve that intersects P = (e)R for R < R (P ).
No indierence curve intersects P = (e)R for R < R (P ).
Suppose the contrary and consider the point- denoted Point A -which intersects P = (e)R
from above to below, that is, such that MRSagent jcomputed at P oint A < (e). Consider the con-
strained indierence curve going through Point A and for which the agent is constrained to
exert a level of eort equal to eA , where eA e (RP oint A ; PP oint A ). The slope of this con-
strained indierence curve is equal to (eA ) for R = L and is strictly higher than (eA )
for R < L. Moreover, the slope of the non-constrained indierence curve going through
Point A is equal to the slope of the constrained indierence curve going through Point A
only at Point A. This implies that M RSagent jcomputed at P oint A > (eA ). Moreover, and by
construction, we have MRSagent jcomputed at P oint A < (e). Therefore, we must have: (e) >
MRSagent jcomputed at P oint A > (eA ), which is impossible because, by denition, e eA ; and
thus, (e) (eA ). This is a contradiction.16
We are now going to show that there is a unique value , with 2 ]e; e[, for which P = R
admits an indierence curve tangent to it for both R < R (P (e)) and R L.

Uniqueness of , for which the straight half-line P = R admits one indierence

curve tangent to it both for R < R (P ) and for R > L.
The dierentiability of the indierence curves (cf. Lemma 2) implies, by continuity, that
there must exist at least one value , with (e) < < (e), such that the straight half-line
P = R admits an indierence curve tangent to it for both R < R (P (e)) and R L.
Suppose now that there are two values of - denoted 1 and 2, with 1 > 2 -for which
P = i R with i = 1; 2; admits an indierence curve tangent to it both for R < R (P (e)) and
R L. Let us denote by IC1 the indierence curves tangent to P = 1 R for R 2 fR1; R2 g,
with R1 < R (P ) and R2 > L. Consider the straight half-line P = 1 R intersecting P = 2 R
for R = R2. This straight half-line, which we denote P 0 (R), is parallels to P = 1 R and located
strictly below P = 2 R for any R < R2 . Consider now ICT1 , the translation of IC1 going
through (P = 2 R2 ; R2 ). ICT1 is (by construction) tangent to P 0 (R) at point (P = 2 R2 ; R2) and
do not intersect P = 2 R for any R < R2 . Moreover, the slope of ICT1 at point (P = 2 R2 ; R2 )
is equal to 1 > 2 . This implies that the indierence curve tangent to P 0 (R) is tangent for a
16
Note that there is also
noindierence curve that is tangent to P = (e) R for R < R(P ). Suppose
the
e e e
contrary and denote by P ; R the tangent point to P = (e) R, where R < R (P ) and e e e
e e P ; R . Doing
the same reasoning as before, we would then have: (e) = M RSagent jcomputed (Pe;Re) > (e
at P oint e). This is
impossible, since by denition e e
e and thus (e) (e
e). We thus obtain a contradiction.

27
repayment R > R2 (i.e., the tangent point is located to the right of (P = 2 R2 ; R2 )). Moreover,
the indierence curve which is tangent to P 0 (R) is located strictly below ICT1 (since indierence
curves are parallel for R R(P ) but steeper for R < R(P )) and thus never intersects P = 2 R.
We thus obtain a contradiction.
Therefore, there is a unique value of - denoted b
-for which the straight half-line P = R
admits one indierence curve tangent to it both for R < R (P ) and for R > L.

D Proof of Proposition 1
We showed in the text that there exist an equilibrium in which insurance companies oer a
continuum of linear contracts Mi = (ri ; pi ), where pi = b
ri , the agent chooses to be only partially
insured (R1 < R (P )) while he is indierent between two optimal choices R 2 fR1; R2 g with
R1 < R (P ) and R2 > L. We are now going to show that this equilibrium is the unique
equilibrium which exists

Unicity of the linear equilibrium

Consider any continuum of linear contracts P = R, for which the agents optimally choice
is unique and such that: R < R (P ). Let P 0 (R) be the straight half-line parallel to but strictly
below P = R going through (R; R ). The dierentiability of indierence curves and
Lemma 2 imply that there exists a unique value of , denoted by DoubleT angency , for which the
agents optimal choice of insurance when facing the set of insurance contracts P 0 (R) is R 2
fR1; R2 g with R1 < R(P ) and R2 L. Thus, when facing P 0 (R) going through (R ; R )
with > DoubleT angency , the agent optimally chooses R L. However, when facing P 0 (R)
passing through (R ; R ) with < DoubleT angency , the agent optimally chooses Rd < R (P ).
Therefore, if an insurance company deviates and oers the contract Cd = (Rd ; P 0 (Rd )) any agent
buys this contract and exerts an eort which is such that (e (Rd ; P 0 (Rd ))) Rd < P 0 (Rd ).
The deviation is thus strictly protable for the insurance company.
Consider now any continuum of linear contracts P = R, for which the agents optimally
choice is unique and such that: R R (P ). This cannot be an equilibrium since the expected
prot of an insurance company would then be strictly negative. Indeed, with a global insurance
coverage R > L the agent would exert the lowest level of eort (e) and the probability of the
accident ( (e)) would be strictly higher than the unit price of insurance, since by denition of
the Strong Moral Hazard Case R R (P ) can only corresponds to the agents unique optimal
choice of insurance, if < (e).
Therefore, there exists no linear equilibrium in which each insurance company oers a con-
tinuum of linear contracts Mi = (ri ; pi ) with pi = ri and for which the agent optimally choice,
R , is unique and such that either R < R (P ) or R R (P ).

28
E Proof of Lemma 5
Obtaining Regularity Condition
@M RSagent (R;P )
We want to demonstrate that @P
< 0 for R (P ) < R < R (P ) if and only if
dP
Regularity Condition is satised. Using the expression of the dR
given in Equality (10), we have
@M RSagent (R;P ) @M RSagent (R;P )
@P
< 0, or, equivalently, @w
> 0, if and only if

00 00
0 (e) de u (w L + R P ) u (w P )
> 0 0 : (Condition 2)
(e) [1 (e)] dw u (w L + R P ) u (w P )
| {z } | {z }
>0 >0 for DARA utility functions

For e 2 ]e; e[ (that is, for an optimal eort which is an interior solution), using Equation (7),
0 0 0 0 (e )
de de
we obtain dw
= "(e(e)[u (wL+RP )u (wP )]
)[u(wL+RP )u(wP )]
< 0. We thus have (e )[1 (e )] dw
> 0. For DARA
utility functions,
h 00the Arrow-Pratt
measure iof absolute risk-aversion is strictly decreasing. This
u (wL+RP ) u00 (wP )
implies that u0 (wL+RP ) u0 (wP ) > 0 for R < L. Replacing de dw
by its expression and
rearranging terms, we obtain that the previous inequality is necessarily satised if and only if,
8e 2 ]e; e[ the function (e) satises the following technical condition:17
00
[ 0 (e )]2 1 u (w P ) u00 (w L + R P ) u (w P ) u (w L + R P )
00
> 0
0 0 :
(e ) (e ) [1 (e )] u (w P ) u (w L + R P ) u (w L + R P ) u0 (w P )
Finally, using the fact that for an interior solution 0 (e ) [u (w L + R P ) u (w P )] = 1
(cf. Equation (7)), Condition 2 reduces to
h i
u00 (wP ) u00 (wL+RP )
0
[ (e )] 3
1 u0 (wP )
u0 (wL+RP )
> 0 : (Condition 3)
00 (e ) (e ) [1 (e )] u (w L + R P ) u0 (w P )
For R strictly dierent from P , the RHS of Equality Condition 3 is well-dened and nite.
Moreover, for R close to L, we have:
h u00 (wP ) 00 i
00 2
h i u0 (wP )
uu0 (wL+RP
(wL+RP )
)
000
u (wP )u0 (wP ) u (wP )
u00 (wP ) u00 (wL+RP )
u0 (wP )
u0 (wL+RP ) (wP )(wL+RP )
(u0 (wP ))2
= lim = :
u0 (w L + R P ) u0 (w P ) R!L 0
u (wP
0
)u (wL+RP ) u00 (w P )
(wP )(wL+RP )

Therefore, the RHS of Equality Condition 3 is also well-dened and nite for R close to L, since
000
by assumption u(:) is of class C 3 , with u0 (w P ), u00 (w P ) and u (w P ) that take nite
values, when (w P ) takes nite values, which is the case for R 2 [0; L].
Finally, let us now dene
h i
u00 (wP ) u00 (wL+RP )
d u0 (wP )
u0 (wL+RP )
k = max , for R 2 [0; L] .
u0 (w L + R P ) u0 (w P )
17
Condition 2 is necessarily satised for CARA utility functions for which the RHS is equal to zero while the
LHS is strictly positive. Thus, by continuity Condition 2 is also satised for DARA utility functions which are
u00 (wP ) u00 (wL+RP )
close to CARA utility functions and for which u0 (wP )
' u0 (wL+RP )
.

29
Then, Condition 3 simply becomes

[ 0 (e )]3 1
> k;
00 (e) (e ) [1 (e )]

where k takes a nite value for R 2 [0; L].

When U (w; L; R; P; e) = (e) ln (w L + R P ) + (1 (e)) ln (w P ) e

1 1
Making computations, we nd that u0 (w L + R P ) u0 (w P ) = wL+RP
wP
1 1

u00 (wP ) u00 (wL+RP ) (wP )2 (wL+RP )2 1 1
and u0 (wP )
u0 (wL+RP )
= 1 1 = wL+RP
wP
. This implies that
h i wP wL+RP
u00 (wP ) u00 (wL+RP )
u0 (wP )
u0 (wL+RP ) [ 0 (e )]2 1
u0 (wL+RP )u0 (wP )
= 1 and Regularity Condition reduces to 00 (e ) (e )[1 (e )]
> 1.

Nonparallel Indierence Curves

@M RSagent (R;P )
Ceteris paribus, we know that @P
< 0 either for R > L or for R (P ) < R < R (P )
when Regularity Condition is satised.
@M RSagent (R;P )
However, @P
> 0, either when 0 < R < R (P ), or when R (P ) < R < L. For any
given P , R (P ) is dened such that 0 (e) [u (w L + R (P ) P ) u (w P )] = 1. Consider
that lim 0 (e) = 1. R(P = 0) is dened such that 0 (e) [u (w L + R) u (w)] = 1. For
e!e
any R < L we necessarily have 0 (e) [u (w L + R) u (w)] > 1 (even for an R very close
to L). We thus have R(P = 0) = L. Moreover, for any P > 0, we also have R (P ) = L
because R(P = 0) = L and R (P ) is increasing in P . For any given P , R (P ) is dened such

that 0 (e) u w L + R (P ) P u (w P ) = 1. Consider now that lim 0 (e) = 0. For
e!e
any R < L, we can have 0 (e) u w L + R (P ) P u (w P ) = 1 only if R(P ) < 0.
Therefore, we have e < e < e for 0 R < L. This implies that for a given repayment R with
0 R < L, the indierence curves become steeper when P decreases when Regularity Condition
is satised and j 0 (e)j is high enough whereas j 0 (e)j is low enough, by continuity.

Consider now two points (R1 ; P1 ) and (R2 ; P2 ) with R1 < R2 and V (R1; P1) = V (R2 ; P2 ).
@M RSagent (R;P )
For any given R, with R 6= L we have @P
< 0. Let IC (R2 ; P2 ) denote the indierence
curve going through (R2; P2) (and (R1 ; P1 )). Let now IC T (R2; P2 ) denote the translation
of IC (R2 ; P2 ) going through (R2 ; P2 ). By construction, IC T (R2; P2 ) also goes through
(R1 ; P1 ). Let now IC (R2 ; P2 ) denote the indierence curve going through (R2 ; P2 ).
@M RSagent (R;P )
@P
< 0 implies that IC (R2 ; P2 ) is located strictly below IC T (R2 ; P2 ). There-
fore, (R1 ; P1 ) and (R2; P2 ) dont belong to the same indierence curve, and (R1 ; P1 )
belongs to an indierence curve associated to a strictly lower expected utility than the one
(R2 ; P2 ) belongs to, since utility is strictly increasing in R but strictly decreasing in P .
Therefore: V (R1 ; P1 ) < V (R2 ; P2 ).

30
References
[1] Ales, L., Maziero, P., 2009. Adverse selection and non-exclusive contracts, Carnegie Mellon
University, Tepper School of Business, Working Paper 574.
[2] Arnott, R., Stiglitz, J. E., 1988. The Basic Analytics of Moral Hazard, Scandinavian Journal
of Economics, 90(3), 383-413.
[3] Arnott, R., Stiglitz, J. E., 1991. Price equilibrium, eciency, and decentralizability in in-
surance markets, Working Paper NBER, N :3642.
[4] Arnott, R., Stiglitz, J. E., 1993. Equilibrium in Competitive Insurance Markets with Moral
Hazard, Boston College Economics Working Paper.
[5] Attar, A., Campioni, E., Chassagnon, A., Rajan, U., 2007. Incentives and competition
under moral hazard, Mimeo IDEI, Toulouse.
[6] Attar, A., Chassagnon, A., 2009. On moral hazard and non-exclusive contracts, Journal of
Mathematical Economics, 45(9-10), 511-525.
[7] Bernheim, B., Whinston, M., 1986. Common Agency, Econometrica, 54, 923-942.
[8] Bertola, G., Koeniger, W., 2015. Hidden Insurance in a Moral Hazard Economy, RAND
Journal of Economics, 46(4), 777-790.
[9] Bisin, A., Guaitoli, D., 2004. Moral hazard and non-exclusive contracts, RAND Journal of
Economics, 35, 306-328.
[10] Finkelstein, A., Poterba, J., 2002. Selection eects in the market for individual annuities:
New evidence from the United Kingdom, Economic Journal, 112, 2850.
[11] Finkelstein, A., Poterba, J., 2004. Adverse Selection in Insurance Markets: Policyholder
Evidence from the U.K. Annuity Market, Journal of Political Economy, 112, 183-208.
[12] Hellwig, M., 1983. On the moral hazard and non-price equilibria in competitive insurance
markets, Discussion Paper N109, Institut fur Gesellschafts.
[13] Helpman, E., Laont, J.-J., 1975. On moral hazard in general equilibrium, Journal of
Economic Theory, 10, 8-83.
[14] Kahn, C., Mookherjee, D., 1998. Competition and incentives with non-exclusive contracts,
RAND Journal of Economics, 29(3), 443-465.
[15] Martimort, D., Stole, L., 2002. The revelation and delegation principles in common agency
games, Econometrica, 70, 1659-1673.
[16] Mossin, J., 1968. Aspects of rational insurance purchasing, Journal of Political Economy,
76(4), 553-568.
[17] Parlour, C., Rajan, R., 2001. Competition in loan contracts, American Economic Review,
91(5), 1311-1328.
[18] Pauly, M., 1974. Overproduction and public provision of insurance, Quarterly Journal of
Econcomics, 88, 44-62.

31
[19] Peters, M., 2001. Common agency and the revelation principle, Econometrica, 69(5), 1349-
1372.
[20] Rothschild, C., 2013. Non-Exclusivity, Linear Pricing and Annuity Market Screening, Jour-
nal of Risk and Insurance.
[21] Rothschild, M., Stiglitz, J., 1976. Equilibrium in competitive insurance markets: An essay
on the economics of imperfect information, Quarterly Journal of Economics, 90(4), 629649.

32