A

Adverse Selection markets; Reservation price; Self-selection;

Separating equilibrium; Signalling; Walrasian
Charles Wilson equilibria

JEL Classifications
Abstract D8
A market exhibits adverse selection when the
inability of buyers to distinguish among prod-
Adverse selection refers to a negative bias in the
ucts of different quality results in a bias
quality of goods or services offered for exchange
towards the supply of low- quality products.
when variations in the quality of individual goods
Typically, the average quality of a product
can be observed by only one side of the market.
supplied by the market depends on the price,
For instance, suppose sellers of high-quality
possibly resulting in multiple Walrasian equi-
goods have a higher reservation price than sellers
libria and even equilibria with rationing.
of low-quality goods, but that buyers cannot
Agents have an incentive to trade multi-
directly determine the quality of a specific good
dimensional contracts so that informed agents
offered for sale. Then any mix of goods offered for
can reveal their quality by the contracts they
sale at the market price must include the low-
purchase. Various mechanisms such as price
quality goods. That is, the market adversely
floors and mandatory partial insurance may
selects for low-quality products.
be used to reduce the market inefficiencies
Adverse selection may appear in any market
resulting from adverse selection.
where either the buyer or the seller has difficulties
ascertaining the quality of the product to be
exchanged. Examples include resale markets for
Keywords
durable goods where it is difficult for the buyer to
Adverse selection; Akerlof, G.; Asymmetric
identify defects known to the seller, labour mar-
information; Bertrand game; Credit rationing;
kets where the seller has a better idea of his pro-
Incentive compatibility; Insurance markets;
ductivity than his potential employer, credit
Pareto improvement; Rationing; Resale
markets where the borrower knows more about
her credit worthiness than the seller, and insurance
This chapter was originally published in The New Palgrave markets where the insured have knowledge about
Dictionary of Economics, 2nd edition, 2008. Edited by their riskiness that is unavailable to the insurer.
Steven N. Durlauf and Lawrence E. Blume

# The Author(s) 2008

Palgrave Macmillan (ed.), The New Palgrave Dictionary of Economics,
DOI 10.1057/978-1-349-95121-5_104-2
2 Adverse Selection

The theoretical study of adverse selection for sale equals the price so that their value of the
began with the seminal paper by George Akerlof, marginal car offered by sellers exceeds the price.
‘The Market for “Lemons”’ (1970). In this paper, A second source of inefficiency is that the
Akerlof demonstrated how adverse selection wrong set of cars may be exchanged. In the exam-
could eliminate all trade in otherwise efficient ple above, the net gain from trade of a car with
markets. As the title suggests, he illustrated his quality q is q2 so that the highest-quality cars should
argument in a stylized model of a market for used be exchanged first. However, if all cars are sold at
cars. Suppose there is a potential supply of ns cars the same price, lower-quality cars will always be
indexed by a quality parameter q that is uniformly supplied before higher-quality cars. In general,
distributed between 0 and 1. Assume that this inefficiency depends on our assumptions
q measures the reservation price of the owner, regarding preferences. In a dynamic model in
but that the reservation value of each of the poten- which the market for used cars arises endoge-
tial buyers is 32 q: If both buyers and sellers can nously, Hendel and Lizzeri (1999) argue that
observe the quality of each car and there are buyers of used cars generally value increases in
enough potential buyers, efficiency requires that quality less than sellers. Consequently, in their
all cars be exchanged. However, if buyers can model the sale of the lowest-quality cars is rela-
observe only the average quality of cars offered tively efficient and measures to increase the vol-
for sale at each price, there is no positive price at ume of trade may be counterproductive.
which cars will be demanded. A third source of inefficiency emerges when
The argument is as follows. If buyers cannot the preferences of buyers are heterogeneous so
observe the quality of individual cars and prices that high-quality cars should be allocated to
adjust to clear the market, then all cars must sell at quality-intensive buyers. In this case, even if the
the same price p. Since an owner offers a car of efficient set of goods were exchanged, the random
quality q for sale only if q < p, it follows that the allocation of cars among buyers implies that
supply of cars is S(p) = nsp at any price p between buyers and sellers would not be correctly
0 and 1 and the average quality of cars at that price matched.
is qa(p)¼ p2. But since a buyer’s reservation value All of these sources of inefficiency can be
of a car with expected value q is 32 q, he purchases illustrated with a slight modification to Akerlof’s
at price p only if qa (p) > 23 p. Consequently, example. Suppose we change the distribution of
demand is D(p) = 0 at each price p and the only the ns cars so that q is uniformly distributed
market clearing price is p = 0 with no trade between 1 and 5. Then, at any price p between
occurring at all. 1 and 5, the supply of cars is SðpÞ ¼ p1 4 ns and
Akerlof’s example of a zero-trade equilibrium average quality is qa ðpÞ ¼ 1þp 2 : At any price
illustrates the most extreme consequence of p > 5 , S(p) = ns and qa(p) = 3. On the
adverse selection. As demonstrated below, not demand side, we suppose there are two types of
all trade is necessarily eliminated. However, if buyers. For a car of quality q, low-intensity buyers
goods of different quality are treated as a homo- are willing to pay 32 q and high-intensity buyers are
geneous good, several sources of inefficiency may willing to pay 2q. Consequently, the demand
persist. One problem is that the marginal value of function has two steps. Low-intensity buyers are
a trade may not be equated between buyers and just indifferent to buying a car at price p = 3 where
3 a
2 q ðpÞ ¼ p: For high-intensity buyers, the point
sellers. Since sellers offer any good for exchange
that they value less than its price, the value to the of indifference is at p = 6. Consequently, if there
sellers of the average product offered for sale is are nL low intensity buyers and nH high intensity
generally lower than the price. In contrast, the buyers, demand is
uninformed buyers purchase the product to the
point where their value of the average car offered
Adverse Selection 3

Second, the cars that are sold provide the least

8
< nL þ nH for p<3 possible net benefit to buyers. If only half of the
DðpÞ ¼ nH for 3 < p < 6 cars are to be sold, efficiency requires they be the
:
0 for p>6 highest-quality cars. Third, since all buyers pur-
chase from the same pool of cars, the cars that are
At prices 3 and 6, demand is a correspondence. sold are not efficiently allocated among buyers.
Figure 1 illustrates two possible relations Since high-intensity buyers value quality more
between supply and demand depending on the than low-intensity buyers, the efficient allocation
relative number of buyers and sellers. The supply of these cars requires that the high-intensity
curve labelled S0 (p) corresponds to a case where buyers receive the cars with the highest quality.
nS < nH so that the market clears at price p = 6. At Given the asymmetry in information, there is
this price, all cars are sold to high-intensity typically no incentive-compatible mechanism that
buyers, and the corresponding allocation is Pareto achieves first-best efficiency. However, there may
efficient. The supply curve labelled S(p) corre- be instruments or mechanisms that may increase
sponds to the case where nH < n2S < nL þ nH so net surplus and in some cases even generate a
that the market clears at price p = 3. At this price, Pareto improvement. For instance, for supply
only cars of quality q < 3 are sold and every active curve S(p) a subsidy on sales would increase the
buyer receives a car of expected quality qa(p) = 2. volume of trade. However, the resulting allocation
Observe that this allocation exhibits all of the would not be completely efficient since low-
sources of inefficiency that were identified above. quality cars are still sold before high-quality cars
First, not all potential buyers purchase a car even and both types of buyers still purchase from the
though half of the cars remain unsold, all of which same pool of cars. We explore below some other
are more valuable to buyers than to owners. mechanisms that may be used to further improve
efficiency.

p
2q a
q ( p)

S'( p) S ( p)

3/2q 6

D( p)
1

q 3 2 1 nH n* nH + nL

Adverse Selection, Fig. 1 An inefficient Walrasian allocation

4 Adverse Selection

S(p)
a p
q ( p)
3/2q

D(p)
9/4
p*
2

3/2

1
D '(p)

q 3/2 1 nS/2m B nS nB
4/3
(3/4)nS

Adverse Selection, Fig. 2 Multiple Walrasian equilibria

Multiple Walrasian Equilibria buyers again enter the market until p rises to 94 ,
after which price exceeds the buyers’ reservation
The examples above have a unique Walrasian value and D(p) falls back to zero. The result is a
equilibrium. However, since average quality non-monotonic demand function and conse-
increases with price, it is possible that over some quently it is possible that there is more than one
range of prices demand may also increase with market clearing price.
price. As a consequence, there may be multiple In this example, multiple Walrasian equilibria
market clearing prices, which can be Pareto arise whenever the number of buyers exceeds the
ranked. We can illustrate this possibility in an total number of cars. Such a case is illustrated in
example with one type of buyer and just two Fig. 2, where demand D(p), indicated by the
types of sellers. heavy dotted line, intersects S(p) at prices 32 , 2,
Suppose half of the ns sellers own cars of and 94. All cars are sold at price p ¼ 94, while only
quality q = 1 and half own a car of quality low-quality cars are sold at price p ¼ 32 . In both
q = 2. Since low-quality sellers supply cars at cases, p ¼ 32 qa ðpÞ so that buyers are just indiffer-
any price p at or above p = 1, and high-quality ent to purchasing a car. There is also a Walrasian
sellers supply cars at any price p at or above p = 2, equilibrium at price p = 2, but to clear the market
it follows that average quality jumps from 1 to 32 at only half of the owners of high-quality cars supply
price p = 2. As above, suppose that each of the nB their cars. As a result, average quality is reduced
buyers is willing to pay 32 q for a car of quality q. to 43 so that buyers are again just indifferent to
Then D(p) = nB for p < 32, but then falls to zero purchasing at that price.
until the high-quality sellers enter the market at Observe that the allocations at these three
price p = 2. At this price, qa(p) rises to 32 and all prices may be Pareto ranked. Although buyers
Adverse Selection 5

are indifferent to each of the prices, some or all opportunity for market participants to exploit the
sellers strictly benefit from selling at a higher relation between quality and price or to indirectly
price. In a more general model with heteroge- identify products of different quality may lead to
neous buyers, Wilson (1980) shows that buyers different market behaviour. To study these effects,
also benefit from buying at a higher price. we need to be more explicit in specifying the
mechanism by which trade takes place.
Consider a market mechanism in which each
buyer fixes a price at which he is willing to buy. To
Pareto Improving Price Floors
sell their cars, sellers first queue at the highest
announced price. Any excess supply then spills
Because of the dependence of average quality on
over to successively lower-price offers until the
price, it is sometimes possible to achieve an addi-
supply of cars is exhausted or there are no more
tional Pareto improvement by setting a price floor
offers to buy. Buyers who announce a price below
and rationing the excess supply of cars. Consider
the point at which supply is exhausted do not
again the example illustrated in Fig. 2. If we
obtain a car.
reduce the number of buyers to mB where n2S < mB
Suppose that all buyers value a car of quality
< nS , then we obtain a demand curve like D0 (p),
q at 32 q. Then, without regard to market condi-
illustrated by a heavy solid line. In this case, there
tions, each buyer prefers the price p that maxi-
is only one Walrasian equilibrium at price p ¼ 32 .
mizes his or her net benefit 32 qa ðpÞ  p: However,
At this price, only low-quality cars are offered for
such a price p is an equilibrium only if there is no
sale and buyers gain no net benefit.
excess demand at that price. As in a standard
Now suppose that we impose a floor ceiling at
Bertrand game, rather than face rationing, buyers
some price p* between 2 and 94 :Since high-quality
prefer a small increase in the price so that they can
cars are also supplied at this price, average quality
buy a car with certainty. Consequently, the equi-
rises to qa ðp Þ ¼ 32 which provides any buyer with
librium strategy for buyers is to set the price that
a positive net benefit. Since there are more sellers
maximizes net benefit 32 q – p subject to the
than buyers at this price, sales must be rationed.
constraint D(p)  S(p).
Nevertheless, owners of both low-quality and
Figure 2 illustrate two types of solution to this
high-quality cars benefit from selling at this
problem. For the case where the number of buyers
price. Owners of high-quality cars benefit because
is nB > nS, represented by the heavy dotted
the Walrasian price is below their reservation
demand curve D(p), the equilibrium price is p ¼ 94
value. And since more than half of the cars are
, which is the highest Walrasian price. At this
sold at this price, the expected return to low-
price, all cars are sold to buyers who are just
quality sellers is also higher at price p*. At the
indifferent to purchasing a car. For the case
Walrasian price p ¼ 32 , their net benefit from a sale
where the number of buyers mB satisfies n2S < mB
is 12, while at the price floor p* > 2, their net benefit
< nS , the equilibrium price is p = 2 (or slightly
from a sale is at least 1.
above to ensure that all owners supply their cars).
All buyers demand a car and all owners supply a
car. But since there are more sellers than buyers,
Uninformed Price Stters and Rationing the sellers must be rationed. With heterogeneous
buyers, Wilson (1980) shows that more than one
Our analysis so far has focused on primarily on price may be announced in equilibrium. In this
Walrasian allocations. In a frictionless economy case, sellers are rationed at all but possibly the
with perfect information and a large number of lowest announced price.
competing agents, this solution is generally A mechanism in which uninformed agents set
robustly independent of the mechanism or con- the price may not be applicable for most resale
ventions by which the price is set. However, once markets for durable goods. However, it may
we introduce asymmetries in information, the
6 Adverse Selection

explain some pricing strategies in financial mar- buyer is indifferent between buying and not buying,
kets where the uninformed agents are large insti- and seller behaviour is optimal if buyers believe that
tutions such as banks. Stiglitz and Weiss (1981) average quality will not increase at higher prices.
implicitly use this price-setting mechanism in A second possibility is a separating equilib-
their study of credit rationing. In their model, rium that involves rationing at some prices. In
banks supplying loans correspond to the this equilibrium, low-quality sellers announce
uninformed buyers of the used car market, and price pL = 32 and high-quality sellers announce
the creditors, who know better their idiosyncratic price pH = 3. Exactly n2S buyers bid at price pL, so
riskiness, correspond to the car owners. Because that demand exactly matches supply and low-
creditors have only limited liability in the case of quality sellers sell with probability 1. However,
default, risky borrowers demand loans at higher at price pH, only n8S (or fewer) buyers bid so that
interest rates than do less risky borrowers. So, if high-quality sellers sell with probability at most 14 .
the demand for loans is sufficiently large, only Observe that at each price buyers are just indiffer-
risky borrowers are served at the Walrasian rate ent between purchasing and not purchasing. Each
of interest. In such a case, it may be more profit- seller is also acting optimally, since high-quality
able for banks to lower their interest rate to attract sellers would suffer a loss by selling at pL, while
low- risk borrowers, even though they must ration low-quality sellers prefer to earn a net gain of 12
their limited supply of funds among the resulting with certainty at price pL rather than a net gain of
increased demand. 2 with probability less than or equal to 14 at price
pH. A general analysis with heterogeneous buyers
is provided in Wilson (1980).
It is not obvious how expectations and prices
Informed Price Setters
would adjust to sustain the separating equilibrium
in this example. However, the example does illus-
In markets for products such as used cars, a mech-
trate how market participants may use another
anism in which sellers are responsible for setting
dimension, in this case the probability of selling,
the price may be of more interest. For example,
to identify products of different quality, albeit at
consider the price-setting convention in which all
some cost. The key ingredient is that sellers of
sellers simultaneously announce prices for their
different-quality cars face a different tradeoff
cars, after which each buyer submits a bid at one
between price and the probability of selling. In
of these prices. If demand does not equal supply at
general, there may be other dimensions in which
any price, the long side of the market is rationed.
the preferences of informed agents differ. In such
Since the informed agents act first, this mecha-
a case the market may exploit multidimensional
nism is essentially a signalling game, first intro-
contracts to identify product quality. A market for
duced by Spence (1973) and later formalized by
insurance provides a good example.
Cho and Kreps (1987) and others.
Consider again the example above with two
types of sellers, half with cars of quality q = 1 and
half with cars of quality q = 2, and one type of buyer Self-selection in Insurance Markets
who is willing to pay 32 q for a car of quality q.
Assume also that there are more potential buyers In its most primitive form, an insurance policy
than sellers. As in many signalling models, there is a consists of two elements, the price of coverage
continuum of sequential equilibria for this game. We and the level of coverage. Although all consumers
focus here on two possible outcomes. One possibil- prefer a lower price to a higher price and prefer
ity is a pooling equilibrium in which each seller more coverage to less coverage, their tradeoff
announces price p ¼ 94 , and exactly nS buyers bid between price and quantity depends on the prob-
to purchase at that price, resulting in a Walrasian ability of a payout. Consequently, by offering
allocation. Buyer behaviour is optimal since each contracts which differ in both price and the level
Adverse Selection 7

H
L
H

a
H

a
a L

L
c

L
L

0 1 t

Adverse Selection, Fig. 3 Equilibrium in an insurance market

of indemnity, sellers may be able to indirectly can identify his own risk type but that firms know
identify different risk classes of consumers who only the proportion a of low-risk types. Let pa =
otherwise appear to be homogeneous population. apL + (1  a)pH denote average probability of
Some of the implications of competition in these an accident among both types of consumers. To
kinds of contract can be illustrated in a simple allocate the policies, we suppose that the
model first studied by Rothschild and Stiglitz uninformed firms are Bertrand price setters that
(1996) and Wilson (1977). earn zero profit for any policy that is
Suppose there are two types of insurance con- actuarially fair.
sumers. Each consumer has the same risk-averse The model is illustrated in Fig. 3, where the
von Neumann-Morgenstern utility u, the same vertical axis represents the premium and the hor-
initial wealth W and the same reduction in wealth izontal axis represents the level of coverage. The
to W  1 in case of an accident. Low-risk types vertical line at t = 1 represents the set of policies
have an accident with probability pL and high-risk that provide full indemnity. The lines labelled pL
types have an accident with probability pH where and pH represent the set of actuarially fair policies
pL < pH. An insurance policy may be represented for the low- and high-risk types respectively. The
as pair (p, t), where t is the indemnity in case of an line labeled pa represents the set of policies that
accident and p is the premium. Therefore, a con- break even if both types purchase it. The curves
sumer who purchases policy (p, t) is left with labelled vL and vH represent typical indifference
wealth W – 1 – p + t if he has an accident and curves for the two risk types. Although both risk
W  p if he does not. Suppose that each individual types prefer more coverage and a smaller
8 Adverse Selection

premium, high-risk types have a higher marginal If the aggregate zero profit line pa lies above
rate of substitution (MRS) of premium for indem- the low-risk indifference curve that passes
nity than do low-risk types at any policy. At any through the low-risk policy bL, as illustrated in
full insurance policy, the MRS of each type is Fig. 3, then equilibrium exists. Both policies lie on
equal to their probability of an accident. their respective zero-profit lines and each con-
Suppose first that firms may offer only policies sumer selects his optimal policy from the avail-
that provide full coverage so that t = 1. In this able set. If any firm deviates with a new policy
case, the model is exactly analogous to the used- offer that attracts only the high-risk types, it must
car example above when the uninformed buyers lie below the pH line and consequently earn neg-
are price setters and there are more buyers than ative profits. However, any new policy that
sellers. Consumers demand insurance policy (p, 1) attracts the low- risk types cannot earn positive
only if their expected utility from purchasing profits unless it also attracts the high-risk types.
exceeds their expected utility from remaining But any such policy earns positive profit only if it
uninsured. The policy bH = (pH, 1) represents lies above the pa line, which in turn attracts only
the full insurance policy that just breaks even for the high-risk types.
the high- risk types. For the case illustrated here, If the aggregate zero-profit line intersects the
the low-risk types would also demand insurance at low-risk indifference curve passing through pL, as
this price. Consequently, the unique Bertrand illustrated by the dotted line labelled pc in Fig. 3,
equilibrium is the policy ba = (pa, 1) , which then there is no equilibrium for this game. In this
just breaks even when purchased by both risk case, a firm may offer a policy just above bc that
types. In effect, low-risk types are subsidizing attracts both types of consumers and still makes
the high-risk types. positive profits in the aggregate. If we permit
Now suppose that firms may also compete in individual firms to offer a menu of contracts as
the indemnity dimension. To begin, we also sup- in Miyazaki (1977), then equilibrium fails to exist
pose that each firm may offer only one insurance under an even wider range of parameters.
policy to its customers. Observe that the equilib- A number of authors have suggested alternative
rium policy under mandatory full coverage is not solution concepts, incorporating non-Nash behav-
an equilibrium for this game. The reason is that, if iour, that generate an equilibrium for this case.
some firm deviates and offers a policy near bL, Wilson (1977) defines a solution concept in
above the pL line and behind the vH curve, it which both types purchase a policy like bc. Riley
attracts only low-risk types and earns a positive (1975) proposes an alternative solution concept
profit. But if low-risk types are attracted away for which the separating allocation bL and bH is an
from policy ba, it earns negative profits. equilibrium.
The only possible equilibrium is a separating
allocation in which some firms offer policy bH,
which is purchased by high-risk types, and some Efficient Public Provision of Insurance
firms offer policy bL, which is purchased by the
low-risk types. Equilibrium requires that the pol- Consider the case where (bL, bH) is an equilib-
icy purchased by each risk type lie on its own zero rium. The low-risk types are made better off than
profit line. Otherwise, firms may exploit the dif- under the equilibrium with mandatory full cover-
ferences in the preferences of the two risk types to age by lowering their indemnity to segregate
offer a policy that attracts only the risk class that themselves from the high-risk types. But high-
earns positive profits. Competition among firms risk types are worse off since they must now pay
must then lead to the best zero-profit policy for the the actuarially fair value of their coverage.
high-risk types and the best zero-profit policy for Clearly, this allocation is not first-best efficient
the low-risk types, subject to the self-selection since an increase in the coverage of the low-risk
constraint for high-risk types to choose policy bH. types at an actuarially fair rate makes them better
off. Consequently, it may be possible to increase
Adverse Selection 9

Adverse Selection, p
Fig. 4 The public
provision of insurance
H H

H
H

a L

0 1 t

the welfare of both types by introducing a menu of there is no need to appeal to alternative solution
policies in which the low-risk types subsidize the concepts to ensure the existence of an equilibrium.
high-risk types. Such an allocation is represented
by policies gL and gH as illustrated in Fig. 4.
To see that the policies are actuarially fair in the
See Also
aggregate, observe that they can be constructed by
decomposing each policy into a common policy
▶ Credit Rationing
ga that lies on the aggregate zero-profit line and
▶ Implicit Contracts
then supplementing the coverage of each risk type
▶ Incomplete Contracts
with an additional policy that lies on their respec-
▶ Moral Hazard
tive isoprofit line that passes through policy ga.
▶ Selection Bias and Self-Selection
One way to implement such an allocation is for
▶ Signalling and Screening
the government to provide policy ga to all con-
sumers and then let the market supply the supple-
mentary policies. Furthermore, by choosing the
appropriate policy ga this mechanism may be used Bibliography
to attain any constrained Pareto-optimal alloca-
tion (subject to the self-selection constraints and Akerlof, G. 1970. The market for ‘lemons’: Quality uncer-
tainty and the market mechanism. Quarterly Journal of
aggregate zero-profit condition). In this case, the Economics 84: 488–500.
supplementary pair of policies required to attain Cho, I., and D. Kreps. 1987. Signalling games and stable
allocation (gL, gH) is necessarily an equilibrium so equilibria. Quarterly Journal of Economics 102:
179–221.
10 Adverse Selection

Hendel, I., and A. Lizzeri. 1999. Adverse selection in Spence, M. 1973. Job market signalling. Quarterly Journal
durable goods markets. American Economic Review of Economics 87: 355–374.
89: 1097–1115. Stiglitz, J., and A. Weiss. 1981. Credit rationing in markets
Miyazaki, H. 1977. The rat race and internal labor markets. with imperfect information. American Economic
Bell Journal of Economics 8: 394–418. Review 71: 393–410.
Riley, J.G. 1975. Competitive signalling. Journal of Eco- Wilson, C. 1977. A model of insurance with incomplete
nomic Theory 10: 175–186. information. Journal of Economic Theory 16: 167–207.
Rothschild, M., and J. Stiglitz. 1996. Equilibrium in com- Wilson, C. 1980. The nature of equilibrium in markets with
petitive insurance markets: An essay on the economics adverse selection. Bell Journal of Economics 11:
imperfect information. Quarterly Journal of Econom- 108–130.
ics 90: 629–649.