You are on page 1of 12

BEE2025 Microeconomics 2 Dieter Balkenborg

Departments of Economics
Adverse Selection University of Exeter
1 Adverse Selection
In this section we return to the insurance model, which diers from the employment
model only in respect to what happens when the agent does not accept the contract.
In the employment model he gets then an exogenously xed reservation utility l
0
0
from
being employed elsewhere (which we should now allow to be type dependent) whereas
he gains in the insurance model the expected utility
l
0
= (1 j) r
0
+ j
0
= (1 j) n
_
n
+
_
+ jn
_
n

_
from being uninsured.
To recall, a monopolistic and risk-neutral insurer (the principal) seeks to insure in a
prot-maximizing way a risk-averse agent against the possibility of an accident. He knows
that the agent is risk averse, in fact he knows that the agents von Neumann-Morgenstern
utility function in wealth is r = n(n).
1
We assume that this utility function n(r) is twice
dierentiable, strictly increasing (n
0
0) strictly concave (n
00
< 0) with a domain of the
form c n < +1 or c n < +1 where c 1.
2
In more specic examples we work
with the utility function r = n(n) =
p
n. Let n = , (r) denote the inverse function of
n(r), so n = , (r) ,r = n(n). (In particular , (r) = r
2
when n(r) =
p
n).
It is also known that the agents initial wealth is n
+
and that the accident would reduce
the agents wealth from n
+
to n

where n
+
n

.
3
In particular, when n(n) =
p
n
they must be both non-negative. In the numerical examples we set n
+
= $1600 and
n

= $100.
In this handout we study the case where the probability j of an accident is the private
knowledge of the agent. The principal is, in contrast, uninformed and does not know
the probability of an accident. For simplicity we assume that he considers only two
possibilities: He thinks that the probability of an accident can be either high (j = j
H
) or
or low (j = j
L
). To be specic, let j
H
=
2
3
and j
L
=
1
3
.
An agent knowing that his accident probability is low will behave dierently from
an agent who knows that his accident probability is high. The principal may hence be
facing two types of agents: A low-risk type or a high-risk type. For expected
utility maximization the principal has to have some assessment about the probabilities
with which he is facing which type. We assume that he believes to be facing the high-risk
type with probability 0 j 1 and the low-risk type with probability 1 j.
1
In game theoretic terms the model is assumed to be common knowledge.
2
This includes the cases where the utility function is dened for all numbers on the number line, as
for u(w) = e
w
, on all non-negative numbers, as for u(w) =
p
w , or on all positive numbers, as for
u(w) lnw.
3
Of course, both levels of wealth must be in the domain of the utility function since otherwise the
expected utility of the agent when he is uninsured would not be dened.
A simple contract between the insure can be describe as a pair (1. C) where 1 is the
insurance premium the agent pays and C the compensation he receives if an accident
occurs. Thus the agents utility is
r = n
_
n
+
1
_
if no accident occurs and
= n
_
n

1 + C
_
if an accident occurs. We nd it easier for the calculations to describe the contract
indirectly by the pair (r. ) of net utilities in the no accident / accident states. We will
do this from now on. From (r. ) one can calculate the premium and the compensation
payment via
1 = n
+
, (r) and C = 1 n

+ , () .
If the principal would know the agents type, matters would be easy. If he knew that
he is facing a low-risk type he would oer him the full insurance contract (r. ) with
r = = n
0
L
= (1 j
L
) r
0
+ j
L

0
where n
0
L
is the expected utility of the agent if he remains uninsured and, as before,
r
0
= , (n
+
) = 40,
0
= , (n

) = 10. In our numerical example, the expected utility of


the low-risk type if he does not insure himself is n
0
L
=
2
3
40 +
1
3
10 = 30. The low
risk type is hence indierent between having the wealth (n
0
L
)
2
= 30
2
= $900 for certain
and being uninsured. In the principals optimal insurance contract the agent pays hence
a fee of 1 = $700 and receives the compensation C = $1500 if the accident occurs. Still
assuming an agent with a low accident probability, the principals expected prot would
be
2
3
$700 +
1
3
($700 $1500) = $200
The point is, of course, that the agent could be a high-risk type. If the high-risk type
accepts the above contract, he ends up for certain with the wealth $900 and hence utility
30 whereas his expected utility is
n
0
H
= (1 j
L
) r
0
+ j
L

0
=
1
3
40 +
2
3
10 = 20
if he remained uninsured. So he would certainly accept the above contract if it were
oered. However, if the above contract is sold to a high-risk type the principals expected
prot would be
1
3
$700 +
2
3
($700 $1500) = $300
Since the principal believes to face the high risk with probability j his ex-ante expected
prot if he oered the above contract would be
(1 j) $200 j $300 = $(200 500j)
Thus, if the principal believes that he is facing the high-risk type of the agent with a
probability j of more than 2,5 = 40%, he would never oer the above contract because
2
he would expect to make a loss. This is the adverse selection problem. The contract which
would (from the monopolists perspective) optimally insure the low-risk type would be
bought from the types with the high accident risk.
Of course, if the principal knew the type of the agent, he could price discriminate.
Then he would fully insure the high-risk type for a premium of 1200 and the low-risk
type for a premium of 700. (Why?)
Because there is asymmetric information the insurance contract cannot condition on
the type of the agent. However, the principal can oer a menu of contracts and let the
agent choose from it. This is what typically happens in practice. You may, e.g., be able
to buy partial car insurance where you pay a certain part of the damage or obtain full
insurance at a higher price. Having the option to oer a menu of contracts rather than
a single contract can only make the principal better o because he can always oer a
menu consisting only of identical contracts and this is the same as oering a single
contract. One can show for example that it is sucient to consider only menus with two
contracts, since there are only two types. If there would be three dierent types of the
agent we would have to allow for menus with three contracts etc.
4
We will denote a menu of two contracts by ((r
L
.
L
) . (r
H
.
H
)) and consider (r
L
.
L
)
as the contract designed for the low-risk type to pick and (r
H
.
H
) as the contract for the
high risk type to pick. At the moment this is labelling only, a more neutral notation like
((r
1
.
1
) . (r
2
.
2
)) would work as well.
We also use the following bits of notation: If the agent does not insure himself he get
the utility r
0
= , (n)
+
if no accident occurs and
0
= , (n

) if an accident occurs. The


expected utility of the uninsured agent is therefore n
L
= (1 j
L
) r
0
+j
L

0
if he is of the
low-risk type and n
H
= (1 j
H
) r
0
+j
H

0
if he is of the high-risk type. Clearly, n
L
n
H
.
We model the monopoly power of the principal by assuming that he can make a take-
it-or-leave-it oer to the agent. To obtain a perfect-information game we assume that the
agents type is determined after the principal has oered a menu of contracts. This allows
us to solve the model in essence by backward-induction. (Alternatively, one can work
with Bayesian perfect equilibrium.) Imagine the principal oers his menu of insurance
contracts on the internet. Then next customer visiting his web site and buying insurance
is with probability j of the high-risk type and with probability 1j of the low-risk type.
5
We can now x the time-structure and the extensive form game as follows:
1. The principle oers a menu of contracts ((r
L
.
L
) . (r
H
.
H
)).
2. Nature determines the type of the agent. With probability j
H
he is chosen to be a
high-risk type and with probability j
L
he is chosen to be a low-risk type.
3. The agent either chooses an insurance contract form the menu oered or he chooses
not to insure himself. (If he accepts one he has to pay the appropriate fee.)
4. An accident occurs with the relevant probability j
L
or j
H
.
4
Instead of working with menus of contract one can equivalently work with option contracts. These
are contracts which allow the agent to choose between dierent options once the contract is signed.
5
In this interpretation you may think of the dierent types of the agent as dierent players.
3
5. If required by contract, the principal has to pay the appropriate compensation.
Final payos are determined.
The extensive form game is indicated in Figure 1.
principal
chance chance
((x

H
, y

H
), (x

L
, y

L
))
((x

H
, y

H
), (x

L
, y

L
))
((x
H
, y
H
) , (x
L
, y
L
))
chance chance

agent
1-
p
L
reject accept
(x
H
, y
H
)
chance
w
+
f(x
H
)
x
H
p
L
w
-
- f(y
H
)
y
H
chance
1-
p
L
p
L
1
0
x
0
0
y
0
chance
p
L
1-
p
L
w
+
f(x
L
)
x
L
w
-
- f(y
L
)
y
L
accept
(x
L
, y
L
)
1-p
H
reject accept
(x
H
, y
H
)
chance
w
+
f(x
H
)
x
H
p
H
w
-
- f(y
H
)
y
H
chance
1-p
H
p
H
0
x
0
0
y
0
chance
p
H
1-p
H
w
+
f(x
L
)
x
L
w
-
- f(y
L
)
y
L
accept
(x
L
, y
L
)
Figure 1
Following the backward induction procedure we obtain Figure 2 by eliminating all
decision nodes of nature at the end of the game and replacing them by terminal nodes
with the appropriate expected payos.
(1-p
L
)(w
+
f(x
H
))+p
L
(w
-
- f(y
H
))
(1-p
L
)x
H
+p
L
y
H
principal
chance chance
((x

H
, y

H
), (x

L
, y

L
))
((x

H
, y

H
), (x

L
, y

L
))
((x
H
, y
H
) , (x
L
, y
L
))
chance chance

agent
reject accept
(x
H
, y
H
)
1
0
(1-p
L
)x
0
+p
L
y
0
accept
(x
L
, y
L
)
reject accept
(x
H
, y
H
)
accept
(x
L
, y
L
)
(1-p
L
)(w
+
f(x
L
))+p
L
(w
-
- f(y
L
))
(1-p
L
)x
L
+p
L
y
L
(1-p
H
)(w
+
f(x
H
))+p
H
(w
-
- f(y
H
))
(1-p
H
)x
H
+p
H
y
H
0
(1-p
H
)x
0
+p
H
y
0
(1-p
H
)(w
+
f(x
L
))+p
H
(w
-
- f(y
L
))
(1-p
H
)x
L
+p
H
y
L
Figure 2
2 The participation- and incentive constraints
Fix a a menu of contracts ((r
L
.
L
) . (r
H
.
H
)) and consider the subgame after this menu
has been oered. If the inequalities
(1 j
L
) r
L
+ j
L

L
(1 j
L
) r
0
+ j
L

0
= n
L
(PCL)
(1 j
L
) r
L
+ j
L

L
(1 j
L
) r
H
+ j
L

H
(ICL)
(1 j
H
) r
H
+ j
H

H
(1 j
H
) r
0
+ j
H

0
= n
H
(PCH)
(1 j
H
) r
H
+ j
H

H
(1 j
H
) r
L
+ j
H

L
(ICH)
4
are satised then it will be a optimal for each type of agent to accept the contract (r
L
.
L
)
or, respectively, (r
H
.
H
) designed for him. However, unless the four constraints are
satised with strict inequality (<) other behavior can also be optimal. For instance, if
(PCL) holds with equality the low-risk type of the agent can randomize between accepting
(r
L
.
L
) and remaining uninsured. Needless to say, if one of the constraints is violated it
cannot be optimal for both types of the agent to accept the contract designed for them.
Let us consider these four constraints in more detail. In our graphs, the contract
(r
L
.
L
) will be indicated by a square, the contract (r
H
.
H
) by a circle.
Inequality (PCL) is the participation constraint for the low-risk agent, expressing that
he is at least as well o by accepting (r
L
.
L
) than by remaining uninsured. The inequality
(ICL) is the incentive constraint for the low-risk type expressing that he is at least as well
o by accepting the contract (r
L
.
L
) designed for him rather than the contract (r
H
.
H
)
designed for the high-risk type.
The conditions can be illustrated in a coordinate system (r. ) as follows: The high
risk type is indierent between two contracts (r. ) and (r
0
.
0
) if
(1 j
H
) r
0
+ j
H

0
= (1 j
H
) r + j
H
,

0

r
0
r
=
1 j
H
j
H
i.e. both contracts lie on a line with slope
1p
H
p
H
. Therefore the indierence curve of the
high-risk type through the point (r. ) is the line with slope
1p
H
p
H
through this point.
All contracts on this line are as good as (r. ) for the high-risk type, contracts above
are strictly preferred and contracts below are worse than (r. ). Consider, for instance,
the indierence curve with slope
1p
H
p
H
through the initial endowment point (r
0
.
0
) of
the agent. The participation constraint (PCH) states that the contract (r
L
.
L
) designed
for the high-risk type must be on or above this line to be acceptable by the agent. In
our specic example, both contracts indicated by a circle and square would be
accepted by the high-risk type because they are above the atter line pch.
0
10
20
30
40
50
60
10 20 30 40 50 60
H
L
(x
0
, y
0
)
x
y
pch
pcl
Figure 3
Similarly, since
(1 j
L
) r
0
+ j
L

0
= (1 j
L
) r + j
L
,

0

r
0
r
=
1 j
L
j
L
5
the indierence curves of the low-risk type are the lines with slope
1p
L
p
L
. Because
j
L
< j
H
,j
L
j
L
j
H
< j
H
j
H
j
L
,j
L
(1 j
H
) < j
H
(1 j
L
)
,
(1 j
H
)
j
H
<
(1 j
L
)
j
L
,
(1 j
L
)
j
L
<
(1 j
H
)
j
H
the indierence curves of the low-risk type fall more steeply than that of the high-risk
type. (Intuitively, the high-risk type values utility in the accident state realtively more
than the low-risk type). The line pcl where participation constraint for the low-risk type
is binding also goes through the point (r
0
.
0
) . but is steeper. In the graphic, the low-risk
type would accept the contract indicated by a square above the steeper line, but not the
one indicated by a circle.
The two incentive constraints inform us about the relative position of the two contracts
(r
L
.
L
) and (r
H
.
H
). Draw the indierence curves for both types through (r
L
.
L
). Then
(ICH) requires that (r
H
.
H
) is on or above the indierence curve for the high risk-type
while (ICL) requires that (r
H
.
H
) is on or below the indierence curve for the low-risk
type. (r
H
.
H
) must hence be in the shaded region of the following Figure 4. Alternatively,
draw the indierence curves for the two types through (r
H
.
H
). Then (ICL) requires that
(r
L
.
L
) is on or above the indierence curve for the low-risk type through (r
H
.
H
) while
(ICH) requires that (r
L
.
L
) is on or below the indierence curve for the high risk type
through (r
H
.
H
). (r
L
.
L
) must hence be in the shaded region of the following Figure 5.
0
10
20
30
40
50
60
10 20 30 40 50 60
x
icl
ich
L
H
0
10
20
30
40
50
60
10 20 30 40 50 60
x
icl
ich
L
H
Figure 4 Figure 5
3 Basic structure of optimal menues
Suppose the principal oers the menu ((r
L
.
L
) . (r
H
.
H
)), the low-risk type of the agent
accepts the contract (r
L
.
L
) and the high-risk type (r
H
.
H
). Recall that , (r) is the
amount of money the agent must have such that his net utility is r. The expected prot
to the principal is therefore
((r
L
.
L
) . (r
H
.
H
))
= (1 j)
_
(1 j
L
)
_
n
+
, (r
L
)
_
+ j
L
_
n

, (r
L
)
__
+j
_
(1 j
H
)
_
n
+
, (r
H
)
_
+ j
H
_
n

, (r
H
)
__
= (1 j)
_
(1 j
L
) n
+
+ j
L
n

_
+ j
_
(1 j
H
) n
+
+ j
H
n

_
C ((r
L
.
L
) . (r
H
.
H
))
6
where the term (1 j) ((1 j
L
) n
+
+ j
L
n

) + j ((1 j
L
) n
+
+ j
L
n

) is a constant in-
dependent of the choice variables r
L
.
L
. r
H
.
H
of the principal and where
C ((r
L
.
L
) . (r
H
.
H
))
= (1 j) ((1 j
L
) , (r
L
) + j
L
, (
L
)) + j ((1 j
H
) , (r
H
) + j
H
, (
H
))
is the expected cost to be payed to the agent.
The principal seeks to oer a menu that will maximize his prots. He always has the
option to oer the contract (r
0
.
0
) in his menu, which is basically the same as oering
no insurance. Instead of rejecting, the agent could always accept this contract. It hence
intuitive that when seeking an optimal menu for the principal we can restrict ourselves
to menus such that both types of agents would accepts the contracts designed for them.
Hence we are let to the following constrained optimization problem:
(*) Minimize the function C ((r
L
.
L
) . (r
H
.
H
)) subject to the four linear
constraints (PCL), (PCH), (ICL), (ICH).
Lemma 1 The function C ((r
L
.
L
) . (r
H
.
H
)) is strictly convex.
Proof: The matrix of second partial derivatives of C is
_

_
(1 j) (1 j
L
) ,
00
(r
L
) 0 0 0
0 (1 j) j
L
,
00
(
L
) 0 0
0 0 j (1 j
H
) ,
00
(r
H
) 0
0 0 0 jj
H
,
00
(
L
)
_

_
which is immediately veried to be negative denite since ,
00
< 0.
Let the menu ((r

L
.

L
) . (r

H
.

H
)) be a solution to the constrained minimization prob-
lem. We will show later that it exists and is unique.
Theorem 1 a) The game has a subgame perfect equilibrium where the principal oers
the menu ((r

L
.

L
) . (r

H
.

H
)) solving the constrained optimization problem (*), the low
risk type accepts (r

L
.

L
) and the high-risk type accepts (r

H
.

H
).
b) In all other subgame perfect equilibria of the game the principal and both types of
the agent get the same payo as in the equilibrium described in a).
Theorem 2 In a solution ((r

L
.

L
) . (r

H
.

H
)) to the constrained optimization problem
(*)
1. The high-risk type is fully insured, i.e.
r

H
=

H
.
Moreover, n
H
r

H
n
L
.
2. The incentive constraint of the high-risk type is binding, i.e.
(1 j
H
) r

H
+ j
H

H
= (1 j
H
) r

L
+ j
H

L
7
3. The participation constraint of the low-risk type is binding
(1 j
L
) r

L
+ j
L

L
= (1 j
L
) r
0
+ j
L

0
= n
L
Proof: The proof is done in four steps, each time by contradiction. We make heavy use of
the two principles established in Lemma 1 of the handout on insurance: Provided it does
not interfere with the participation and incentive constraints, it increases prots to reduce
the utilities in any of the contracts. Moreover, it increases prots to move the contract
(r
L
.
L
) (respectively (r
H
.
H
)) along an indierence curve of the low-risk (respectively
high-risk) type.
Step 1: One of the two participation constraints must be binding (i.e. hold with equality)
in the optimal menu ((r

L
.

L
) . (r

H
.

H
)).
0
10
20
30
40
50
60
y
10 20 30 40 50 60 70
x
H
*
ich
*
icl
*
icl
pch
pcl
H
L
L
*
ich
0
10
20
30
40
50
60
y
10 20 30 40 50 60 70
x
L
*
H
*
H
pch
icl
*
icl
pcl
Figure 6 Figure 7
0
10
20
30
40
50
60
y
10 20 30 40 50 60 70
x
pcl
pch
H
*
L
L
*
icl
*
0
10
20
30
40
50
60
y
10 20 30 40 50 60 70
x
L
*
L
H
*
pch
pcl icl
*
ich
*
Figure 8 Figure 9
0
10
20
30
40
50
60
y
10 20 30 40 50 60 70
x
L
L
*
H
*
pch
pcl
ich
*
0
10
20
30
40
50
60
y
10 20 30 40 50 60 70
x
H
*
H
H
pcl
pch
ich
*
L
*
Figure 10 Figure 11
8
Proof: (See Figure 6) Suppose all four constraints hold with a strict inequality in the
optimal menu ((r

L
.

L
) . (r

H
.

H
)). Since all terms in the constraints depend continuously
on the variables we can nd a small number o 0 such that the menu
((r
L
.
L
) . (r
H
.
H
)) = ((r

L
o.

L
o) . (r

H
o.

H
o))
also satises all four constraints with strict inequality. However,
C ((r
L
.
L
) . (r
H
.
H
)) < C ((r

L
.

L
) . (r

H
.

H
))
as one can see when comparing term-by-term because , is strictly increasing. For instance,
, (r
L
) = , (r

L
o) < , (r
L
) etc. This contradicts the assumption that ((r

L
.

L
) . (r

H
.

H
))
minimizes C subject to the constraints.
Step 2: The participation constraint must be binding for the low-risk type.
Proof: Suppose the participation constraint is binding for the high risk type (i.e. (r

H
.

H
)
is on the line (pch) in Figures 7), but not for the low risk type (i.e. (r

L
.

L
) is above the
line (pcl)). There are two possibilities. a) r

H
, as in Figure 7. Because the incentive
constraints must hold, (r

L
.

L
) must be in dark shaded area of the gure. Consider the
alternative contract ((r

L
.

L
) . (n
H
. n
H
)) where we oer full insurance to the high-risk type.
All incentive constraints and participation constraints are satised in this new contract.
The principal gets the same expected prot from the low-risk agent as before. However, his
expected prot from the high-risk type is strictly higher (movements along an indierence
curve towards full insurance increase the prots).
6
Thus the menu ((r

L
.

L
) . (r

H
.

H
))
cannot have been optimal. b) r

H
, as in Figure 8. The constraints (PCL) and
(ICH) imply that (r

L
.

L
) is in the shaded region of the gure, but not on the line (pcl).
Consider the alternative menu ((r
L
.

L
) . (r

H
.

H
)) where r
L
is obtained by reducing the
rst component in (r

L
.

L
) until (PCL) becomes binding (indicated as the contract 1 in
the graph). Again, the new contract satises all four constraints. However, the low-risk
agent is now paid less when no accident occurs, so C ((r
L
.
L
) . (r
H
.
H
)) is reduced, again
in contradiction to the optimality of ((r

L
.

L
) . (r

H
.

H
)).
Step 3: r
0
r

L
n
L
, i.e., (r

L
.

L
) is on the line segment between non-insurance (r
0
.
0
)
and full insurance (n
L
. n
L
) for the low risk agent.
Proof: First assume r

L
r
0
(see Figure 9). This means that if no accident occurs the
low-risk type has a higher utility when he is insured than when he is uninsured. The
contract requires a negative premium for the low-risk type. (r

H
.

H
) must then be in the
shaded region of Figure 9, by (PCH), and (ICL). (ICH is then automatically satised.)
Then the contract ((r

H
.

H
) . (r
0
.
0
)) where the low-risk type is fully insured, still satises
all four constraints, but yields higher prot to the principal. Contradiction. Suppose next
r
L
< n
L
and therefore
L
n
L
. (r

H
.

H
) must be in the shaded region of the Figure 10.
This time the contract ((r

H
.

H
) . (n
L
. n
L
)) also satises all constraints and brings higher
prot to the principal.
Step 4: (ICH) is binding and the low-risk type is fully insured.
6
To make sure that both types accept their contracts in the new menu, increase utility in each contract
by a small amount, but not so much that the principal loses compared to the previous oer. I will not
always point out this type of subtlety in the following.
9
Proof: (r

H
.

H
) must be in the shaded region of Figure 11. First reduce r
H
from r

H
until
(ICH) becomes binding. This increases the principals expected payo and all constraints
remain binding. Then move the contract (r
H
.
H
) along the line (ich) until it hits the
45
o
-line. Again, the principals contract increases and all constraints are satised. Thus,
unless we already started in a contract where (ICH) is binding and the high-risk type is
fully insured, the initial menu cannot have been binding.
Steps 3 and 4 imply n
H
r

H
n
L
.
4 The nal step
The previous result gave us the restriction
n
H
r

H
n
L
(1)
to be satised by one of the unknowns and the three equations

H
= r

H
r

H
= (1 j
H
) r

L
+ j
H

L
n
L
= (1 j
L
) r

L
+ j
L

L
With the three equations we can eliminate all but one variable, for instance

H
= r

H
(2)
r

L
=
j
L
r

H
j
H
n
L
j
H
j
L

L
=
(1 j
H
) n
L
(1 j
L
) r

H
j
H
j
L
We have not completely solved the problem, but we have reduced a constrained optimiza-
tion problem with four unknowns to an almost unconstrained optimization problem in
one variable.
7
Namely, we must still minimize the cost function
c (r

H
) = C ((r

L
.

L
) . (r

H
.

H
))
= (1 j) ((1 j
L
) , (r

L
) + j
L
, (

L
)) + j ((1 j
H
) , (r

H
) + j
H
, (

H
))
= (1 j)
_
(1 j
L
) ,
_
j
L
r

H
j
H
n
L
j
H
j
L
_
+ j
L
,
_
(1 j
H
) n
L
(1 j
L
) r

H
j
H
j
L
__
+j, (r

H
)
c (r

H
) is convex since it is obtained from a convex function by linear substitutions of the
variables.
The results in the previous section did not depend on the probability j with which the
principal faces a high-risk type. When calculating the minimum of c (r

H
) this is crucial.
7
Except that restriction (1) must still hold. Notice that we use here the substitution rather than the
Lagrangian approach to solve the constrained optimization problem.
10
In particular, when j = 1 we have to minimize the function , (r

H
) subject to the
constraint n
H
r

H
n
L
. Since , (r) is increasing this minimum occurs when r

H
= n
H
.
The equations (2) imply then

H
= r

H
and r

L
= r
0
,

L
=
0
. This is as it should be,
when there is no low-risk type, the optimal insurance is to fully insure the high-risk type
and to oer no insurance to the low-risk type.
Notice that this solution is not an interior optimum determined by the rst-order
condition c
0
(r

H
) = 0. The rst-order condition has for j = 1 no solution in the positive
numbers.
8
The solution is determined by the constraint n
H
r

H
. Using continuity
considerations one can show that this constraint must also be binding for all j suciently
close to 1. This means that if the probability of a high-risk type is suciently close to
one, then it is optimal to fully insure the high-risks and oer no insurance to the low-risks.
The high risks drive the low risks out of the market.
For j = 0 the optimization problem boils down to nding the optimal insurance
contract for the low-risk type subject to the participation constraint of the low-risk type
holding with equality. As we know, the solution is to fully insure the low-risk type.
This could mean that a loss is made on the high-risk type. However, those occur with
probability zero.
Finally, a more detailed analysis of the model shows that r

H
is nonincreasing in j.
Rather than giving the general analysis it must suce here to do the calculation in
our example. Our restriction is 20 r

H
30. We must have

H
= r

H
r

H
=
1
3
r

L
+
2
3

L
30 =
2
3
r

L
+
1
3

L
or

H
= r

H
(3)
r

L
= 60 r

L
= 2r

H
30
The function to be minimized is
c (r

H
) = (1 j)
_
2
3
(r

L
)
2
+
1
3
(

L
)
2
_
+ j (r

H
)
2
= (1 j)
_
2
3
(60 r

H
)
2
+
1
3
(2r

H
30)
2
_
+ j (r

H
)
2
8
At least when some additional technical assumptions are satised.
11
0 10 20 30
0
500
1000
1500
2000
x
y
Figure 12: c (r

H
) for j =
1
6
.
5
6
. 1
This is a quadratic function in r

H
with a unique minimum where
Jc
Jr

H
= (1 j)
_
2
3
(60 r

H
) (1) +
1
3
(2r

H
30) (2)
_
+ 2jr

H
= (1 j) [2r

H
60] + 2jr

H
= 2r

H
60 (1 j) = 0
, r

H
= 30 (1 j)
The restriction 20 r

H
implies that this solution is only valid when 30 (1 j) 20 or
j
1
3
. Thus the nal solution is given by
r

H
=
_
30 (1 j) for j
1
3
20 for j
1
3
and the equations 3.
In particular, we obtain for j =
1
6
r

H
=

H
= 25, r

L
= 35.

L
= 20
1
H
= 1600 25
2
= 975, C
H
= 1500
1
L
= 1600 35
2
= 375, C
L
= 375 100 + 400 = 675
for j =
5
6
and j = 1
r

H
=

H
= 20, r

L
= 40.

L
= 10
1
H
= 1600 20
2
= 1200, C
H
= 1500
1
L
= 1600 40
2
= 0, C
L
= 0 100 + 100 = 0
In the latter cases the low-risk is not insured.
12