Active Screening in Insurance Markets

Shinichi Kamiya Nanyang Business School Nanyang Technological University Nanyang Avenue, Singapore 639798 Mark J. Browne Wisconsin School of Business University of Wisconsin-Madison 975 University Avenue Madison, Wisconsin 53706-1323

Submitted to Journal of Public Economics March 21, 2011

Corresponding author, Email : [email protected], Tel.: +65 6790 5718, Fax : +65 6791 3697

Active Screening in Insurance Markets

Abstract In a market characterized by asymmetric information, a party with private information may reveal the information through its self-selection, the mechanism of which has received considerable attention in the academic literature. The party might also reveal the private information through a revelation test oered by an uninformed party. We consider competitive insurance markets in which insurers induce information revelation with self-selection mechanisms and risk classication tests. We demonstrate that conditional and unconditional contracts may coexist in equilibrium when the conditional contract oered by an insurer has an accuracy level that is superior to that of other insurers. We also show that when multiple rms use tests that are similar in terms of their accuracy, conditional contracts do not hold in a Nash equilibrium. Keywords: test, screening, adverse selection, insurance JEL Codes: D81, D82, G22

Introduction

Self-selection models in markets characterized by asymmetric information have received considerable attention in the literature. Sorting of informed parties in these models occurs with dierent types choosing dierent contract from menus oered by competing uninformed parties. In the context of insurance markets, uninformed insurance companies compete against each other in terms of the premium and coverage that they oer to informed individuals who then self-select. Self-selection results in a decrease in information asymmetry. In contrast to the attention paid to self-selecting models, the literature has paid little attention to another widely observed market response to information asymmetry, risk classication through 1

testing, with which uninformed parties actively screen informed parties. Underwriting in nancial markets, interviewing in job markets, and dating in the marriage market are all examples. In the current study, we analyze equilibrium in a market in which asymmetric information is reduced through self-selection and through testing. In practice, testing tends to be imperfect, and noisy tests may entail pooling contracts. We consider when contracts conditional on a test result are oered in insurance transactions and whether such conditional contracts hold in equilibrium. In our construct, insurers have a choice to oer either a single unconditional contract that allows individuals to self-select or a single conditional contract that requires individuals to take a test rst that reveals whether they are high risks or low risks. Individuals have a choice between self-selecting the unconditional contract and taking a test to purchase a conditional contract. In the absence of the risk classication associated with a conditional contract, the market is characterized by a Rothschild-Stiglitz (RS, hereafter) Nash equilibrium (Rothschild and Stiglitz, 1976). We demonstrate that individuals are willing to deviate from a self-selection contract to a conditional contract if the test used for the conditional contract is relatively accurate. Equilibrium conditions are investigated under several dierent market assumptions, and we show that both conditional and unconditional contracts can coexist in several circumstances. In contrast, it is also shown that conditional contracts are not sustainable as a Nash equilibrium if multiple insurance companies employ a similar test in the market.1 The remainder of this article is organized as follows. In Section 2, a brief overview of the literature on screening is presented. Competitive insurance markets and active-screening are characterized in Section 3. In Section 4, we investigate equilibrium in several market settings where insurance companies use symmetric tests which reveal both high-risk and low-risk individuals risk type with the same probability. To establish robustness and to provide explicit conditions, the case of asymmetric tests which perfectly identify low-risks
Our observation is that while insurers employ similar tests, for instance most use gender in personal lines of coverages, the classication schemes are unique across insurers.
1

is discussed in Section 5. A summary of our ndings and a discussion of limitations in our work are contained in Section 6.

Existing studies

Screening models in a competitive environment have been considered by many others since the seminal work of Rothschild and Stiglitz (1976). A rich literature of competitive screening models explores possible equilibria under various assumptions (see, for instance, Miyazaki, 1977; Riley, 1979; Spence, 1978; Wilson, 1977). Recent literature has considered renement of the RS model and resulting equilibria (e.g., Dubey and Geanakoplos, 2002; Martin, 2007; Netzer and Scheuer, 2010). For instance, Netzer and Scheuer (2010) consider a model where individuals are heterogeneous in their wealth and argue that equilibria in their model are more consistent with empirical ndings. Most studies extending the RS model have relied on the self-selection mechanism for risk sorting. Typically eorts made by uninformed parties to acquire knowledge regarding unobserved information are not considered. Prior studies on information acquisition have primarily dealt with constructs where consumers are unaware of their risk type ex ante and invest in learning their risk type (see, for instance, Crocker and Snow, 1992; Doherty and Thistle, 1996; Hoy and Polborn, 2000; Polborn, Hoy, and Sadanand, 2006). In contrast, consistent with the RS model, our analysis assumes that individuals know their risk type, but insurance companies do not. The assumption that test results are public information also sets this study apart from earlier work on information acquisition by Doherty and Thistle (1996) and Hoy and Polborn (2000). Their work assumes that tests such as genetic tests are independent from the purchase of an insurance contract and that insurers do not have access to the test results. This article follows research by Browne and Kamiya (2010) which investigates the market outcome when a noisy underwriting test is oered by the insurer. Their study focuses on identifying equilibrium conditions when insurance companies are non-myopic and all insurers

have access to an identical underwriting test. This study diers from theirs by focusing on Nash competition and examining the case where each insurers test is unique.

Competitive Insurance Market

Homogeneous risk-neutral insurers oer one type of contract dened by coverage quantity, q, and premium, p. Individuals live in one period with initial endowment W > 0. Terminal wealth is uncertain with two states: W1 with a xed loss D > 0 and W2 with no loss. Individuals dier in their likelihood of loss, which is private information known only to the individual. The probability that loss state W1 occurs for a high risk (H-type) is H and that for a low risk (L-type) is L , where 0 < L < H < 1. Individuals also dier in their observable attributes, which can be used by insurance companies to predict their risk-type. The risk-averse individual, who has a standard risk-averse utility function (i.e., U > 0, U < 0), applies for one contract. The L-type individuals expected utility with insurance contract C = (p, q) is V L (C) = L U(W p + q D) + (1 L )U(W p). The proportion of H-type individuals and that of L-type individuals in the market are public information and are denoted by and 1 , respectively. The classic screening models well describe the market outcomes where there is no public information expected to be associated with private information. However, in market transactions an uninformed party tends to observe informed parties characteristics. In insurance transactions, individual characteristics such as age, gender, and marital status may be relatively easily obtained by uninformed insurance companies. It is natural in such a market that competing insurance companies immediately attempt to process a set of observed information to predict individual risk type. If one rm successfully develops an algorithm to predict risk type more accurately than other rms, the rm will gain a market advantage. Consider such an algorithm as a function f : X {H-type, L-type} where X represents a set of observable attributes. The output correctly predicts the individuals risk type with

probability y (0, 1) but fails with probability 1 y. If a rm invents a test, it has a choice to oer either an unconditional contract or a conditional contract that may be purchased by individuals depending on their test outcome. Thus, a conditional contract requires individuals to take a test rst and the test result determines if they can purchase the contract. Consider a sequential game in which a new rm decides to enter the existing insurance market by oering contracts. When individuals maximize their expected utility, CournotNash equilibrium identied in this game can be characterized by no contract in the equilibrium making a loss and no contract outside the equilibrium making a non-negative prot if oered. In the absence of any test, the well-known RS separating equilibrium identied as a set of contracts (H, L) holds (see Figure 1) if and only if the population of H-type individuals is suciently large relative to that of L-type individuals as pointed out by Rothschild and Stiglitz (1976). We follow Crocker and Snow (1985) and denote the RS critical value of the proportion of H-type individuals by RS , with which the RS equilibrium holds if and only if the actual proportion of H-type individuals is greater than or equal to the critical value (i.e., RS ). In other words, the critical value is dened where L-type individuals are indierent between a pooling contract with RS and the RS separating contract L. Let M be the pooling contract which satises the equality, V L (M) = V L (L), where the fair premium rate for the pooling contract is: M = L + RS (H L )

(1)

[Insert Figure 1 Here]

Insurance Market with Active Screening

To investigate the possibility that individuals demand a conditional contract which upsets the RS separating equilibrium in the case of RS , we rst consider H-type individuals demand for a conditional contract followed by L-type individuals demand. H-type individuals are willing to deviate from their unconditional contract H to a conditional contract, A, if and only if EV H (A, H) V H (H), assuming that individuals choose a conditional contract when they are indierent between these two contracts. The left hand side of the inequality represents the H-type individuals expected utility attained by taking the test. If the test misclassies one into a L-type, contract A, dened by the L-type individuals optimal pooling contract at its actuarially fair premium pA , can be purchased. Otherwise, one takes the unconditional contract H. The light hand side of the condition is the utility corresponding to the RS separating contract H. This condition is explicitly expressed as: yU(W H D) + (1 y)[H U(W pA + qA D) + (1 H )U(W pA )] U(W H D)

(2)

which can be reduced to V H (A) V H (H). This inequality holds when L-type individuals prefer the conditional contract A to their separating contract L given the conditional contract is oered. That is, whenever L-type individuals deviate from the unconditional contract L to the conditional contract A, H-type individuals also deviate from the unconditional contract H to conditional contract A. This result connes our attention to the L-types demand. Lemma 4.1. L-type individuals prefer the conditional contract A to their unconditional contract L in the case of RS if and only if the accuracy of a test employed for the conditional contract satises: (1 y) RS (1 y) + y(1 )

(3)

Proof. The conditional contract A requires a subsidy s dened by the rms resource con6

straint where its income is equal to the expected claim payment: y(1 )A qA + (1 y)A qA = y(1 )L qA + (1 y)H qA

(4)

where A = L + s. The subsidy rate is proportional to coverage because the subsidy paid by L-type individuals to misclassied individuals depends on the optimal coverage determined by maximizing the L-types utility for contract A. Then the subsidy rate can be rewritten as: s= H L 1+
y(1) (1y)

(5)

Note that the RS critical value RS is determined where L-type individuals are indierent between a pooling contract with RS and its separating contract L (i.e. V L (M) = V L (L)). Hence the L-types demand condition, V L (A) V L (L), can be restated as V L (A) V L (M). Since the subsidy rate of A and that of M are s and RS (H L ), respectively, the condition is equivalent to: H L 1+
y(1) (1y)

RS (H L ).

(6)

which is rearranged, in (3), by the relationship between the fraction of H-type individuals in the pooling contract A and the RS critical value. The minimum required accuracy of a test decreases to 0.5 as RS . Thus both L-type and H-type individuals no longer choose their separating contracts that fully rely on their self-selection when a test can reduce the fraction of H-type individuals in a pooling contract to at least the RS critical value. They instead prefer a conditional contract that utilizes a test. When < RS , the underlying market fully relying on self-selection is characterized by non-existence of Nash equilibrium. Individuals prefer a pooling contract W (a Wilson pooling contract) at the average fair premium pW , to their separating contracts. It is straightforward to show that individuals are better o with the conditional contract, A, than the uncondi7

tional pooling contract, W , if the test accuracy is greater than 0.5. However, this accuracy does not necessarily guarantee that insurance companies cannot oer non-equilibrium unconditional contracts that attract only L-type individuals given the conditional contract A is oered. In the next section, we identify a robust condition as a part of our equilibrium argument in the case of < RS .

4.1

Monopoly on Conditional Contracts

We start with the case where a test f is invented by one of the competing insurance companies and only the insurance company can oer a conditional contract based on that test. In contrast to the market discussed by Stiglitz (1977), the market analyzed here is still competitive in that other insurance companies can oer an unconditional contract. Therefore, in the market where RS , H-type and L-type individuals can at least attain the RS separating contracts, H and L, respectively. 4.1.1 Equilibrium

When individuals choose a monopolists conditional contract A, L-type individuals who are misclassied by the test and H-type individuals who are correctly identied by the test take unconditional contracts. Therefore, a set of contracts consisting of a conditional contract and two unconditional RS separating contracts, (A, L, H), may hold in equilibrium when RS . In contrast, when < RS , unconditional contract W never holds in equilibrium, while the conditional contract A may not be upset even in that case. Equilibrium holds if no new unconditional contract can earn a non-negative prot given the conditional contract is oered. Although the conditional contract A is also a pooling contract, the contract can be sustained in equilibrium. To understand this possibility, it is important to note that an individuals decision whether to take a conditional contract is based on its ex-ante utility for the conditional contract A, which could result in either of two contracts: A and L for 8

L-type individuals and A and H for H-type individuals when RS . This is because L-type individuals take their unconditional separating contract L if they are misclassied as H-types by a noisy test and H-type individuals demand their unconditional separating contract H if a noisy test correctly identies H-type individuals as H-types. The H-types expected utility is strictly less than the level of utility gained solely by the contract A, EV H (A, H) < V H (A). Analogously, L-type individuals expected utility is strictly less than the level of utility gained solely by the contract A, EV L (A, L) < V L (A). Thus, the individuals ex-ante utility is strictly lower than its utility gained solely from the conditional contract A. It can be shown that the deviation of the H-type individuals expected utility from V H (A) is positively associated with the accuracy of a test and makes it possible that no unconditional contract that attracts only L-type individuals can be oered. It is clear that EV H (A, H) V H (H) while EV L (A, L) V L (A) as y 1 (see Figure 1). Lemma 4.2. A set of contracts (A, L, H) holds in equilibrium when RS if there exists an unconditional contract L dened by V H (L ) = EV H (A, H) which satises: EV L (A, L) V L (L )

(7)

where the contract L is oered at L-types fair premium pL . In equilibrium, a conditional contract and unconditional contracts coexist when the test employed by the conditional contract is relatively accurate. Lemma 4.3. Conditional contract A holds in equilibrium when < RS if there exists an unconditional contract L dened by V H (L ) = EV H (A, W ) which satises EV L (A, W ) V L (L )

(8)

These incentive conditions may be thought problematic because unconditional pooling contract W does not hold in equilibrium. However, the Wilson pooling contract W can be 9

used to dene the constraint. This is because the contract W yields lower utility for L-type and greater utility for H-type individuals than any new potential unconditional contract oer that attracts only L-type individuals (see contract B in Figure 2). Thus, the incentive constraints are robust against the concern of non-existence of equilibrium for the pooling contract W , though it is still true that the unconditional pooling contract itself does not hold in equilibrium. [Insert Figure 2 Here] Proposition 4.1 (Active Screening Equilibrium). A conditional pooling contract may exist in equilibrium regardless of the proportion of H-type individuals, if an insurer with a unique underwriting test that is suciently accurate oers a conditional contract. This argument can be established by the lemmas discussed above. 4.1.2 Optimal Contract

Given that an equilibrium exists, we next consider the rent earned by a monopolist, the single best underwriter in the market, in the case where RS . We denote the protmaximizing contract as . Clearly, the optimal contract is obtained where the L-types incentive constraint holds with the equality, EV L (, L) = V L (L ). Otherwise, there exists a contract (p , q ) such that p + (1 )(p q ) > p + (1 )(p q ) (9)

The optimal contract may be seen diagrammatically in Figure 3. Due to the test, the monopolists breakeven point is at A = L + s as discussed above. The contract that maximizes prots must lie on the indierence curve V L () and on the line parallel to the line pA . The reason is that the vertical distance between the indierence curve V L () and the line pA represents the prot which is maximized where a line shifted upward from pA 10

is tangent to the indierence curve. Let p = (L + s)q + be the tangent line where represents a prot per insured. Thus, the prot-maximizing contract is identied as the tangent point of the line p to the indierence curve V L (). Note that p implies that q = qA . Therefore, the conditional contract oered to those classied as a L-type is = (p , qA ) where p = (L + s)qA + . [Insert Figure 3 Here] The prot also represents the value of a noisy test, which may be sold to other rms. A conditional contract, however, can hold in equilibrium only when the monopolist alone oers the conditional contract. We show later that there is no equilibrium for conditional contracts if the same test is used by other competing insurance companies.

4.2
4.2.1

Competitive Market with Multiple Tests

We investigate the impact of rm competition in developing a better test to gain a competitive advantage. Our particular interest here is whether a conditional contract at the actuarially fair premium still holds in equilibrium. Consider a simple case where rm i invents a test fi : X {H-type, L-type} with probability yi where i = 1, 2. Assume that y1 > y2 > 0. Thus, one rm has a more accurate test than the other. Firm i can oer a conditional contract Ai for those identied as a L-type by its test at the fair premium. Hence, rm 1 oers a conditional contract A1 = (pA1 , qA1 ) where: pA1 = qA1 L + H L 1+
y1 (1) (1y1 )

(10)

A question is whether the conditional contract A1 can be sustained when both another conditional contract and unconditional contracts are oered. Our primary concern here is another conditional contract, A2 , potentially oered by rm 2 (see Figure 4). 11

[Insert Figure 4 Here] For the conditional contract A1 to be sustained against any new conditional contract A2 , the new contract must result in an immediate loss if it is oered. This condition is rephrased as: if a conditional contract A2 attracts L-type individuals, it must attract H-type individuals as well. That is, the following two inequalities must be satised simultaneously. EV L (A2 , L) EV L (A1 , L) EV H (A2 , H) EV H (A1 , H)

(11) (12)

To analyze these conditions, it should be noted that by denition the test for contract A2 is less accurate than the test for existing contract A1 . Therefore, if rm 2 oers a conditional contract A1 with test accuracy y2 , only H-type individuals deviate from A1 to A2 . Thus, in order to avoid attracting H-type individuals, rm 2 may oer a combination of coverage and premium which at least osets the H-types benet gained by the less accurate test, but still attract L-type individuals. Proposition 4.2. A set of contracts (A1 , L, H) holds in equilibrium in a market where rms oer conditional pooling contracts if there is no new contract A2 that satises both: EV L (A2 , L) EV L (A1 , L) EV H (A2 , H) < EV H (A1 , H)

(13) (14)

As long as the accuracy of the second best test satises (13) and (14), a conditional contract A1 cannot be upset. The accuracy of a test required for a conditional contract to survive in equilibrium cannot be explicitly derived in this case. The explicit form under the assumption of an asymmetric test is derived in the next section. This argument can be extended to the case where more than two insurance companies oer tests with dierent degrees of accuracy.

12

4.2.2

Homogeneous Tests

In contrast to the previous case, we now consider the case in which multiple rms employ the same test. Each rm oers a conditional contract A to those who are identied as a L-type. We nd that the conditional contract does not hold as a part of equilibrium. It can be shown that there exist possible protable deviations against a new conditional pooling contract A. Given that contract A is oered, insurance companies may oer a conditional contract that attracts only L-type individuals but not misclassied H-type individuals (a contract such as A2 in Figure 4). Such a conditional contract always exists as long as L-types individuals subsidize misclassied H-type individuals. The non-existence of equilibrium for a conditional contract can be explained just as a pooling contract cannot be sustained in the absence of any tests. Browne and Kamiya (2010) show that the conditional contract holds in a Wilson equilibrium where insurance companies are assumed to be non-myopic. Thus, equilibrium in a market where rms use the same test reverts to the RS equilibrium argument. A set of unconditional separating contracts (H, L) holds in equilibrium if RS . Otherwise, there is no equilibrium in the presence of conditional contracts. Eciency of equilibrium is another major issue when contracts fully rely on individuals self-selection. Since we discuss conditional contracts that could be introduced to the market where self-selection contracts are oered, by denition both H-type and L-type individuals are better o ex ante when conditional contracts exist in equilibrium. The allocation of the utility gain between risk types is determined by the accuracy of a test in equilibrium. The utility gain of H-types is maximized at the minimum required accuracy of the test that sustains equilibrium and decreases as the accuracy increases. In contrast, L-types expected utility monotonically increases with test accuracy.

13

Asymmetric Test

The analysis in Section 4 leads to several conclusions: rst, active-screening may be utilized in a market where private information prevails; second, a pooling contract may hold in equilibrium; third, a conditional contract cannot be a part of an equilibrium, if the most accurate test can be used by more than one insurance company. We next consider an asymmetric test, which imperfectly identies H-types but can perfectly identify L-types. We show that our conclusion does not depend on the assumption that both risk-types are correctly classied with the same accuracy. We also show that some equilibrium conditions are explicitly provided in terms of the accuracy of a test. Specically, we consider a test that always correctly identies L-type individuals as L-types while H-type individuals are misclassied with the probability 1 y. It is clear that L-type individuals will purchase a contract A if they take the test. With the asymmetric test, the subsidy implicit in a conditional contract A is: s = (H L ) 1y , 1 y

(15)

and the required accuracy of the test for the case when RS is dened as: y 1 (RS /) . 1 RS

(16)

As RS , the right hand side of (16) decreases to zero. When the market is characterized by < RS , it is straightforward to show that L-type individuals always prefer a conditional contract, regardless of its accuracy. Thus, the accuracy condition above is a sucient condition such that L-type individuals prefer a conditional contract A to any unconditional contract oer regardless of the underlying market. Market equilibrium when the conditional contract oered by the monopolist of a conditional contract employs an asymmetric noisy test can be identied analogously to the

14

symmetric test case. The argument is omitted here because any further simplication cannot be made. The insurers prot in this case can be expressed in terms of the L-types risk premium for contract and L , denoted as and L , respectively. With these risk premiums, the prot-maximizing condition, V L () = V L (L ), can be expressed as: U(W p A A (D qA ) ) = U(W L D L ).

(17)

This implies that the prot per insured is = (L ) sqA .

(18)

Next, we investigate a market discussed in Section 4.2.1, in which two conditional contracts are oered. A conditional contract A1 holds in equilibrium if a new conditional contract A2 that attracts L-type individuals also attracts H-type individuals (see Figure 5). That is, a contract A2 that satises V L (A2 ) V L (A1 ) must also hold EV H (A2 , H) EV H (A1 , H). [Insert Figure 5 Here] Consider a new contract oer such that H-types expected utility from a new conditional contract A2 , with which only L-type individuals are worse o, makes H-types better o by the less accurate test y2 as before. It is straightforward to show that contract L , dened by V L (A1 ) = V L (L ) at L-types fair premium, maximizes L-types utility loss but still attracts L-type individuals. Therefore, if the test employed by rm 2 is inaccurate enough to increase H-types expected utility even for contract L , the equilibrium condition is satised. Thus, we can restate the condition as EV H (L , H) EV H (A1 , H), which can also be rewritten as: y2 V H (L ) EV H (A1 , H) . V H (L ) V H (H) (19)

The equilibrium condition is identied in terms of the accuracy of the second best test. 15

A principle dierence from the symmetric test case is observed in regards to an unconditional contract oer. When L-types are perfectly classied, no unconditional contract for L-type individuals is oered. This also implies that, regardless of the fraction of H-type individuals, no unconditional pooling contract is oered. Therefore, equilibrium, if it exists, is always characterized by a set of contracts (A, H).

Conclusion

Competitive screening has been discussed in the absence of uninformed parties eorts to classify informed parties. Many dierent forms of screening are, however, employed when private information prevails. It is reasonable for uninformed parties to attempt to predict informed partys private information to obtain a competitive advantage in a market. We investigate the impact of testing on the existence of equilibrium. This paper identies the demand for a conditional contract and we show that individuals deviate from their self-selection contracts to a conditional contract, which requires them to take a test, if the test reduces the fraction of H-type individuals in a pooling contract to lower than the RS critical value of the underlying population. Perhaps most importantly, it is shown that a pooling conditional contract holds in equilibrium when the most accurate test has signicant competitive advantage and can attract all individuals. In contrast, we also nd that conditional contracts do not hold in equilibrium when multiple rms utilize tests similar in terms of their accuracy.

Acknowledgements
We thank Michael Hoy and seminar participants at the 2010 Risk Theory Seminar for helpful comments. All errors are our own.

16

References
Browne, M. J., and S. Kamiya, 2010, A theory of demand for underwriting, Working Paper, University of Wisconsin-Madison. Crocker, K. J., and A. Snow, 1985, The eciency of competitive equilibria in insurance markets with asymmetric information, Journal of Public Economics 26 (2), 207219. Crocker, K. J., and A. Snow, 1992, The social value of hidden information in adverse selection economies, Journal of Public Economics 48 (3), 317347. Doherty, N. A., and P. D. Thistle, 1996, Adverse selection with endogenous information in insurance markets, Journal of Public Economics 63 (1), 83102. Dubey, P., and J. Geanakoplos, 2002, Competitive pooling: Rothschild-stiglitz reconsidered, Quarterly Journal of Economics 117 (4), 15291570. Hoy, M., and M. Polborn, 2000, The value of genetic information in the life insurance market, Journal of Public Economics 78 (3), 235252. Martin, A., 2007, On rothschildstiglitz as competitive pooling, Economic Theory 31, 371 386. Miyazaki, H., 1977, The rat race and internal labor markets, Bell Journal of Economics 8 (2), 394418. Netzer, N., and F. Scheuer, 2010, Competitive screening in insurance markets with endogenous wealth heterogeneity, Economic Theory 44, 187211. Polborn, M. K., M. Hoy, and A. Sadanand, 2006, Advantageous eects of regulatory adverse selection in the life insurance market, The Economic Journal 116 (508), 326354. Riley, J. G., 1979, Noncooperative equilibrium and market signalling, American Economic Review 69 (2), 303307. 17

Rothschild, M., and J. Stiglitz, 1976, Equilibrium in competitive insurance markets: An essay on the economics of imperfect information, Quarterly Journal of Economics 90 (4), 629649. Spence, M., 1978, Product dierentiation and performance in insurance markets, Journal of Public Economics 10 (3), 427 447. Stiglitz, J. E., 1977, Monopoly, non-linear pricing and imperfect information: The insurance market, Review of Economic Studies 44 (3), 407430. Wilson, C., 1977, A model of insurance markets with incomplete information, Journal of Economic Theory 16 (2), 167207.

18

W2 EVL(A,L) pA EVH(A,H) VH(H) H pH L L E 0

pL

W1

Figure 1: Demand of Conditional Contract A

19

W2 EVL(A,W) EVH(A,W) pA A pL

W H pH

L E

45 degree

W1

Figure 2: Conditional Contract A in Equilibrium

p pH H EVH( ,H) H) p VL( ) A pA pL

L EVL( ,L) D q

Figure 3: Prot-maximizing Contract

20

W2 EVL(A1,L) pL VH(A1) EVH(A1,H) pA A1 A2

H pH

L L E

45 degree W1

Figure 4: Multiple Conditional Contracts

W2 VL(A1) VH(A1) EVH(A1,H) pA A1 A2 pL

L H pH E 0 45 degree W1 L

Figure 5: Conditional Contract A with Asymmetric Tests

21