Contending with Risk Selection in Competitive Health Insurance Markets

Transcription

1 This paper is prepared for presentation at the leture, Sikness Fund Compensation and Risk Seletion at the annual meeting of the Verein für Soialpolitik, Bonn, Germany September 29, September 19, 2005 Contending with Risk Seletion in Competitive Health Insurane Markets Jaob Glazer Department of Eonomis Boston University, and Faulty of Management Tel Aviv University Thomas G. MGuire Department of Health Care Poliy Harvard Medial Shool Aknowledgements: Researh for this paper was supported by the Ageny for Healthare Researh and Quality (P01 HS10803) and the National Institute for Mental Health (R34 MH071242).

2 Introdution In many ountries, residents hoose a health plan or sikness fund through whih to reeive health insurane benefits. These hoies are regulated and at least partially paid for by governments and employers. Colletive finaning of health are redistributes the burden of ost from the sik to the healthy and from the poor to the rih, in omparison to a market system where everyone pays their own way. At the same time, soieties seek the virtues of markets: hoie, innovation, and ost and quality ompetition from their health insurane plans. Melding these desires for both a fair and ontrolled, and an effiient and innovative, health insurane setor is a entral and ommon problem faing all developed nations. As Rie and Smith (2001) point out, a ommon approah to this problem onsists of national governments olleting the funds to pay for health are, but then devolving responsibility for the purhasing of health are to a loal organization, a private insurane plan as in the federal Mediare program in the U.S., loal government in the U.K., Canada and Australia, or sikness funds as in Germany, Israel, Netherlands and Belgium. Governmental involvement intends to distribute the ost burden fairly, and ompetition among the deentralized partiipants is intended to promote effiieny. One of the major onerns with suh a poliy is adverse seletion (Enthoven, 1993; van de Ven et al., 2003). Generally, health plans or sikness funds may take ations to disourage or enourage potential enrollees from joining, and these ations may have effiieny or fairness impliations. For one thing, they may refuse some appliants, although overt ations to disourage individuals are normally prohibited. More troublesome and diffiult to monitor is that plans may distort the mix of the quality of health are they offer to disourage high-ost persons from joining the plan. As a number 1

3 of papers have observed, deisions about what are is medially neessary are fundamentally outside the sope of diret regulation (Miller and Luft, 1997; Newhouse, 1996). The purpose of this paper is to onsider how eonomi analysis an help address the problem of adverse seletion in health insurane within a entrally finaned, but ompetitively supplied, health insurane system. Most health eonomists answer to the question of how to address seletion in this ontext enters around the poliy of risk adjustment of the premiums paid to insuring organizations. Whether risk adjustment is helpful and/or neessary, and what form of risk adjustment is most useful, depends on the form of the problem raised by adverse seletion, and on the tools the regulator (or employer) has to deal with these problems. In this paper, after setting out the problems aused by seletion in health insurane, we organize our disussion about addressing these problems around five ases varying aording to what the payer knows and an do about seletion. We will onsider both ontrating and information-based poliies as well as risk adjustment. Theory will be tied, where possible, to experiene in the U.S. of payers in situations orresponding to the theoretial analysis. The paper loses with some observations about ontending with adverse seletion within the German system of health are finaning. The term health plan will be used in general disussions, and sikness fund used in the appliation to Germany. The Basi Adverse Seletion Problem One aspet of seletion is when enrollees with different osts are distributed unevenly aross health plans or sikness funds. In Mediare in the U.S., there is plenty of evidene 2

4 that lower ost benefiiaries are more likely to enroll in managed are plans paid by riskadjusted apitation, and higher ost benefiiaries remain in the fee-for-servie (FFS) setor. 1 In private health insurane, managed are plans also attrat lower ost enrollees. 2 In Germany, differenes in ontribution rates at sikness funds are evidene for seletion sine the ontribution rates are partly driven osts. 3 Seletion of this form may or may not onstitute a soial effiieny problem. In any population, there will be diversity in tolerane for ost-ontrol mehanisms, in the evaluation of different benefits, and in loational preferenes. If any of these fators or other fators are orrelated with expeted health are osts, an effiient division of the market among plans (that is, one respeting diversity in tastes) will be haraterized by seletion of higher risk individuals into some plans. As Feldman and Dowd (2000) and Pauly (1985) have effetively argued, evidene of risk segmentation of this form does not onstitute a per se violation of onditions for eonomi effiieny in insurane markets. 4 Nonetheless, seletion of this type an be assoiated with unfairness in the sense of differential premiums or ontribution rates. In the analyses referred to above, health plans are assumed to offer a fixed produt, and the effiieny issue is sorting people among plans. Analysis of adverse seletion and effiieny of health insurane is inomplete, however, without reognizing that plans take ations to affet their membership. Plans 1 Eggers and Prihoda (1982) showed that people in Mediare HMOs had signifiantly lower osts than those in traditional fee-for-servie setor. A survey by Rossiter and Wilensky (1986) indiated that most of the studies during supported the existene of seletion behavior by HMOs. Brown et al. (1993) found that the spendings of Mediare enrollees in a managed are plan would be 10% lower than average if they had been in the FFS setor. Greenwald et al. (2000) showed that there were signifiant differenes between the atual ost of managed are and FFS benefiiaries for inpatient servies. Also see Hellinger (1995). 2 Niholson et al. (2004). See also Cutler and Zekhauser (2000) for a review. 3 Buhner and Wasem (2003). We disuss this more below. 4 See Knaus and Nusheler (2004) for a parallel disussion in the German ontext. 3

5 an do two main things. First, if a plan or fund knows the likely ost of an identifiable potential enrollee will be more (or less) than the revenue the plan reeives for that person, the plan might take ations to disourage (or enourage) enrollment at the individual level. Suppose a government or employer pays a plan a fixed amount for eah enrollee in the plan, disregarding observable fators, suh as age, when paying the plan. Older workers ost more than younger workers. A plan has an obvious inentive to aept a young worker and deny enrollment to an older worker. Health plans are generally prohibited by regulation from denying enrollment to eligible appliants, but some overt ations to disourage ostly appliants may be hard to stop. Plans an do a seond thing, whih does not require illegal ations direted against individuals. The plan an underprovide some servies and overprovide others, attrating the low risks and deterring the high risks (Newhouse, 1996; Van de Ven and Ellis, 2000). Plan manipulation, in Cutler and Zekhauser s (2000) terminology, emerges in models of health insurane when plans ompete on the basis of servie quality (Glazer and MGuire, 2000; Nusheler and Knaus, 2004). The basi idea draws on early analyses of insurane by Rothshild and Stiglitz (1976). Demand for treatment of hroni onditions, for example, may be muh better antiipated, and more unevenly distributed in a population, than demand for aute are. In suh a ase, the health plan has a finanial inentive to distort the mix of its are away from hroni are and towards aute illness, in order to deter/attrat the high/low risks. Nearly all writers on the effiieny of health insurane markets with managed are aknowledge this effet, though they vary in the emphasis they put on it. It is in the foreground of disussion in Cao (2002), Frank et al. (2000), Glazer and MGuire (2000), Luft and Miller (1988) and Newhouse (1997; 4

6 2002a), while noted but given less prominene in Cutler and Zekhauser (2000), Feldman and Dowd (2000), Pauly (2000), and van de Ven and Ellis (2000). When a plan an set pries as well as quality, a version of this strategy is to provide low quality overall, and set a low prie, to attrat the low-risks. This quality distortion problem has reeived a good deal of attention in the literature in health eonomis. We present the basi adverse seletion model here, as a point of referene, in order to highlight some of the key assumptions behind the model. Reonsidering these assumptions will struture our analysis of ways a regulator an address problems of adverse seletion. Suppose that there are two types of individuals, L and H, who an ontrat two illnesses, a and. Illness a we all an aute illness and both types of people have the same probability of ontrating this illness, p a > 0. The two types are distinguished in their probability of ontrating the hroni illness. Let p i, i {H, L} denote the probability that a person of type i ontrats illness. Then, p > p 0. The proportion H L > of H types in the population is λ, 0 < λ < 1. Let p λp H + (1 λ)p L denote the (expeted) probability that a randomly drawn person ontrats the hroni illness. Throughout our analysis we assume that eah individual knows her type. 5 We also assume that eah individual must hoose one plan. If a person (of either type) has illness j, j { a,}, her utility from treatment will be inreased by V j (q j ), where q j > 0 denotes the quality of the servies devoted to 5 For more disussion of the assumptions in this setion, see Glazer and MGuire (2000). 5

7 treat illness j, with V j > 0 and V j 0. 6 Thus, we make the simplifying assumption that the benefits from treatment are independent of one another and the same to all individuals. If a person has both illnesses, her utility, if treated, will simply be inreased by Va (q a ) + V (q ). Treatment servies are provided by health plans. A health plan is haraterized by a quality pair (qa,q ), where q j, j { a,} is a summary indiator of the quality of servies that the plan provides, devoted to treating illness j. Thus, if a person of type i, i {H, L} joins a plan with a quality pair (q a,q ), her expeted utility will inrease by: (1) U (q,q ) = p V (q ) p V (q ) i a a a a + i Throughout the analysis we assume that eah plan gets to hoose its quality pair and a plan an offer only one quality pair. All plans have the same ost funtion. A plan s ost of treating a person with illness j, j { a,} at a quality level q j is C j (q j), where C > 0, C 0. Thus, if a person of type i, i {H, L} joins a plan that offers a quality j j > pair (qa,q ), the plan s osts are expeted to inrease by (2) C (q, q ) = p C (q ) p C (q ) i a a a a + i The soially effiient quality pair (qa,q ) equalizes marginal benefit of treatment to marginal ost, thus solving the following pair of equations: (3) V (q ) = C (q ) a a V (q ) = C (q a a ) 6 The eonomi literature models quality in two ways: as a form of quantity rationing, as here and in Pauly and Ramsey (1999), or as a shadow prie as in Keeler et al. (1998) or Frank, Glazer and MGuire (2000). When onsumers are idential in their demands given they are ill, and differ only in the probability of having an illness, the two approahes are equivalent. 6

8 High and low risk types have different probabilities of beoming ill, but one ill, reeive the same utility from treatment. Thus, the effiient level of quality is independent of the probability of beoming ill and is the same for both types. We postulate the existene of a publi regulator or payer whose objetive is to implement the soially desired quality. The fous of our analysis will be on the tools that the regulator an use in order to ahieve his goal and on the onditions under whih these tools an be applied. We assume the regulator an enfore an open enrollment poliy. The following five assumptions are made in order to present the basi adverse seletion result. In the analysis that follows this result the signifiane of these assumptions and the onsequenes of relaxing them will be disussed. Assumptions: 1) Quality is not ontratible. The publi regulator annot ondition payments to plans (either by the regulator or by the individuals) on the basis of their delivered quality. 2) Cost is not ontratible. The publi regulator annot ondition payments to plan on the basis of osts per person. 3) Plans an freely enter and exit the market. The publi regulator annot ondition a plan's partiipation on the basis of its quality. However, one a plan partiipates, the regulator an require that it aepts every appliant. 4) Consumers an observe the quality of eah plan and an freely move from one plan to the other. 5) There is no risk adjustment. More speifially, we assume that the premium is set suh that the plan is expeted to break even if it offers the soially effiient 7

9 quality pair (q a,q ) and attrats randomly drawn individuals from the entire population. Thus, the premium is set at r, where r paca (q a ) + pc (q ) =. 7 The order of moves in our model is as follows: first plans (simultaneously) hoose their quality pair, (qa,q), then individuals hoose plans and plans ollet a revenue of per enrollee, finally eah individual's health state (whether she has illness a and/or ) is realized and plans pay the osts of treatment. Our definition of a ompetitive equilibrium in this ase is similar to that of Rothshild and Stiglitz (1976). A ompetitive equilibrium in this market is a set of quality pairs suh that, when individuals hoose plan to maximize expeted utility, (i) no quality pair in the equilibrium set makes negative expeted profit, and (ii) there is no quality pair outside the equilibrium set that if offered, will make a positive profit. The following proposition builds on Rothshild and Stiglitz (1976) and Glazer and MGuire (2000): Proposition: If λ (the proportion of the H types in the population) is suffiiently large, then a ompetitive equilibrium exists and is haraterized by two quality pairs. 8 H types hoose the plan(s) that offer the quality pair: r (4) (q H a,q H ) = argmax U s.t. C H H (q,q ) (q,q ) = r a a 7 The assumption that plans annot ompete on premium is not essential to the analysis. The starting point of our analysis is that when quality is multidimensional, adverse seletion inentives indue plans to distort quality. If plans ould also hoose premium, premium would be just another instrument plans ould use to "selet" patients, but this instrument by itself would not generally be enough and plans would still have the inentives to distort quality in order to affet the mixture of enrollees. Even if we allowed for the possibility that plans hoose opayments, the results would not hange muh, unless the plans ould hoose different levels of opayments for the different servies they provide. 8 The exat ondition is that λ should be suffiiently large so that U L (q L a,q L ) U L (q a,q ) for every (q a,q ) for whih p a C a (q a )+p C (q ) = r. 8

10 and L types hoose the plan(s) that offer the quality pair: (5) (q L a,q L ) = argmax U s.t. and C U L (q,q ) (q,q ) = r L L a (q,q ) = U (q a a H H a,q H ) The proof of the proposition above is well established (see Glazer and MGuire, 2000). The reason that the soially effiient quality pair (q a,q ) annot be an equilibrium pair is that if all plans offer this quality profile, eah (single) plan will have an inentive to deviate to a different pair, say q,q ) with a better aute are (i.e., ( a q > q ) ) and a worse hroni are (i.e., q < q ) ), attrating only the L types and ( a a ( making a stritly positive profit. The equilibrium is desribed in Figure 1. The urves r i,i = H, L represent all plans, i.e., pairs of (qa,q ), that break even if the plan attrats only individuals of type i, when the premium is r. The points denoted by depit the plan hosen by type i in equilibrium, i.e., q (q,q ) i = i a i. Curves u i, represent type i's indifferene urves that goes through the point q i. The urve q i, i = H or L, i = H, L, represents all plans that break even if the plan attrats a random sample of the population, and the point q depits the soially effiient levels of quality. We an see, therefore, that in a market where the five assumptions above hold, plans will not offer the soially effiient quality profile in equilibrium. Figure 2 depits the arrangements among onsumers, a entral government and health plans orresponding to the set up of the model just desribed. Consumers are represented by sik and healthy types, with siker individuals being more ostly. Siker types need more of both hospital bed days and more physiian visits than the healthy r 9

11 types. Their omposition of demand is also different: they need and value hospital are relatively more than the healthy. All onsumers pay in funds to the entral government (employer) in the form of taxes or premiums. These onsumers hoose among plans, in the figure represented by Plans 1, 2 and 3. Plans reeive funds from the entral government, set enrollment poliies, and deide about the nature of the produt they offer onsumers. Plans an do two things, illustrated by ations Plans 1 and 2. Plan 1 distorts its produt, de-emphasizing hospital are and putting resoures in to dotors, in order to attrat the healthy and deter the sik. Plan 2 puts barriers in enrollment in front of the siker potential enrollees. The poliy tools available to a regulator orrespond to the five assumptions laid out above. Relaxing eah assumption in turn an be used as a basis for disussion of different approahes to dealing with the basi adverse seletion problem. For eah of these ases we will disuss the theoretial onepts and will present, where possible, poliy settings that resemble the ase. The ases that will be disussed are the following: 1. Paying on quality-quality is observable and verifiable by a third party so the regulator an make payments to plans based on quality of the servies the plans provide. 2. Cost sharing- ost per individual is observable and verifiable by the regulator. Payment per individual to the plan an be based on the plan's spending on that individual. 3. Restriting entry- quality is observable by the regulator but it is not verifiable. Even though the regulator annot make payments to plans dependent on the 10

12 quality of their servies, it an deide whih plans may partiipate in the market after observing their quality. 4. Quality reporting quality is not verifiable, it is observable by the regulator but not by the onsumers. The regulator an hoose how muh information about quality to provide to onsumers. 5. Risk adjustment-quality is not verifiable and not observable by the regulator. Consumers, on the other hand, an observe (but not pay on the basis of) quality. CASE 1: Quality is Verifiable- Pay for Performane If plans' quality is verifiable payments an be made on the basis of quality. The regulator an onstrut a ontrat by whih a plan reeives a premium per person that is a funtion of the plan's quality. If the premium funtion is suh that when all plans offer the soially desired quality eah plan makes zero expeted profit and no plan wishes to deviate to another quality profile (possibly attrating only one of the two types), then the regulator's objetive is obtained. One suh premium funtion would be the one that pays plans nothing if their quality is different than the soially desired one and pays them exatly their expeted osts, if they offer the soially desired quality. Notie that in ase disussed above there is no need for ompetition in order to implement the desired quality. However, what is needed is for the regulator to "know" exatly the soially desired quality, in order for the premium to be dependent on any gap between the plan's atual quality and the desired one. One ould assume a somewhat more realisti senario where the quality is observable and verifiable but the regulator does not have enough information (e.g., about plans' osts or enrollee s health are needs) 11

13 to be ertain about the level of quality of eah servie that defines the soially desired. In suh a ase the regulator may enhane quality by having many plans and paying higher premiums to plans with higher quality. One should note, however, that there are some major problems with paying on quality, and this ase is noted here for oneptual ompleteness rather than as a basis for pratial disussion. First of all, it is very hard to measure quality preisely, espeially sine a good quality measure needs to take into aount illness severity of eah and every patient. Adjusting for severity is quite omplex and involves many of the issues related to risk adjustment to be disussed later. Furthermore, even if one an ome up with a good quality measure for some servies, it is pratially impossible to ome up with a good measure of all servies. However, paying on quality only for some servies and not the other, may introdue a major multitasking problem. 9 In sum, while experimenting with paying for performane is beoming more and more popular, it is a small part of ontrats, and with urrent information about quality health are, is unlikely to be able to ontend with seletion related quality onerns. CASE 2: Cost Per Individual is Verifiable When plans' osts are verifiable the regulator has the hoie of several different payment shemes, onditioning payments to plans on their atual expenditures. One simple and ommonly used strategy is ost sharing. Under ost sharing the payer, in addition to paying the plan a fixed premium, overs a prespeified share of the plan's osts. (see Ellis and MGuire 1986 and Newhouse 1996). One an easily see that in the model studied 9 Rosenthal and Frank (2004) ontains a survey of the inentive issues and empirial results of pay-forperformane studies. One onlusion is that, Despite the assertions of its proponents, the empirial foundations of pay-for-performane in health are are rather weak. 12

14 above, suh a strategy will not solve the adverse seletion problem. In Figure 1, ost sharing will simply mean a (parallel) shift in the zero profit urves and the resulting equilibrium will still be a separating one. In a more general framework, however, ost sharing ould redue plans inentives to selet enrollees, as it makes the ost differential between a high ost individual and a low ost individual smaller. These weaker inentives to selet enrollees might result in a more effiient equilibrium outome, at least as far as adverse seletion is onerned. Notie that in order to implement the ost sharing payment sheme, all the payer needs to be able to verify is a plans' total osts and not the ost per individual. If the regulator an verify a plan's ost per individual, other payment shemes are feasible, the most ommonly used and widely disussed one is outlier ost reimbursement. Under outlier ost reimbursement the payer overs a share of the plan's expenditures on those individuals whose osts exeed a ertain prespeified threshold level (van de Ven and Ellis, 2000). When publi funds are limited, or inentives are an issue, other, more general approahes an be onsidered (Kifmann and Lorenz, 2005). In our stylized model disussed above, one an see that under suh a payment poliy, if only a share of the plan's ost beyond the threshold level, are overed, the equilibrium will still be a separating ineffiient one, and if all plan's osts, beyond the threshold level, are overed, multiple equlibria will emerge with only one of them being the effiient equilibrium. In a more general setting, however, one should expet the outlier ost reimbursement sheme to lessen plans' inentives to disourage high risk enrollees as they are no longer that more expensive to the plan than the low risk ones. 13

15 While ost sharing mehanisms seem to be helpful in addressing adverse seletion problems, they may also reate some new problems. If plans osts are, at least partially overed, plans will have less of an inentive to save on osts on more of an inentive to over provide some servies (See Newhouse 1996). The tradeoff between these two fores must be arefully examined before one is to apply a ost sharing mehanism. CASE 3: Quality is Observable and the Regulator Can Selet Plans on the Basis of Quality Assume that the regulator an observe eah plan's quality. Further assume that the regulator, after learning a plan's quality, an deide whether or not the plan is eligible to offer its servies to the onsumers. In suh a ase the regulator an easily implement the soially desired ontrat by announing a premium per person that exatly overs a plan's expeted ost at the soially desired ontrat and announing that only the plans with the desired quality will be allowed to partiipate. It is easy to see that in the model presented above, suh a mehanism will implement an equilibrium in whih all plans offer the desired quality. This approah requires only that the payer be able to observe quality, and be prepared to not ontrat with health plans that offer unsatisfatory quality. Case 3 orresponds to the private health insurane market in the U.S. It is ommon in the U.S. for workers and their families to obtain health insurane offered through their employer. Private health insurane is regulated in terms of priing, overage, and who must be offered overage in omplex ways by both the federal and state governments. As a general matter, however, there is no standard benefit, and no universal requirement that employers offer overage. Nonetheless, beause of favorable 14

16 tax treatment of health insurane inluded as part of employee benefits, approximately 71% of workers in the U.S. are overed through employer-based health insurane (Gruber and MKnight, 2002). In order to minimize osts of worker ompensation, employers have an inentive to offer an attrative health insurane benefit to workers, whih, beause of worker heterogeneity, may inlude hoie of forms of health insurane. The U.S. is haraterized by a private market in health insurane with a wide variety of produts, varying aording to the servies overed, the degree of provider hoie permitted, and the stringeny of managed are arrangements. It an be expeted that in the presene of some hoie of different types of health insurane, workers who antiipate being more healthy and low ost, will sort themselves into plans in systemati ways. There is ample evidene that risk seletion among types of plans takes plae in the private health insurane market in the U.S. (Niholson et al., 2004). How have private employers ontended with issues raised by seletion in the market for health insurane? To begin, it is notable what private employers do not do. Private employers do not use formal risk adjustment. By formal risk adjustment we mean setting a prie for individual or family enrollees to pay to health plans based on personal harateristis suh as age, gender or past health are use. Employers adjust their premiums for risk in several different ways, but formal risk adjustment is rarely used. While publi purhasers rely heavily on formal risk adjustment, very few private purhasers use formal risk adjustment to pay health plans. Indeed, among private employers, whih aount for the majority of enrollees in health plans, formal risk adjustment is virtually nonexistent, with 15

17 the exeption of a few purhasing oalitions. While the data we present are for 1998, we know of no evidene that use of formal risk adjustment among private employers is inreasing. In light of the importane of the potential impliations of this market test of risk adjustment, it is worth onsidering praties of the private setor in more detail. Relative to publi purhasers, large employers may hoose from additional praties that effetively address risk seletion problems (Glazer and MGuire 2001a disuss these strategies). First, larger private purhasers (usually more than 200 employees) may deide to self-insure, meaning that they pay the laims inurred by their employees or benefiiaries plus a fee to the health plan administering the benefits. Suh firms health are osts thus trak the atual use of medial are by their employees. This pratie has beome inreasingly ommon, even among firms with as few as 100 employees, in part beause self-insured firms are exempt from state benefit mandates. An estimated 48 million employees and dependents are enrolled in self-insured plans (MDonnell and Fronstin, 1999; InterStudy, 2000). Sine health plans are not at risk for medial are, the inentive for risk seletion at the plan level is removed. Large employers make additional deisions that influene the potential for risk seletion. In partiular, although many large employers offer a hoie of produts (a health maintenane organization [HMO], a preferred provider organization [PPO], and a point-of-servie [POS] plan, for example), employers often ontrat with one arrier to provide all produts. Beause the entire risk pool of employees remains with the one arrier, there are no inentives at the plan level to influene risk seletion among the various produts. In fat, large employers offer a hoie of arrier relatively infrequently: in 1997, 15% of firms with 500 or more employees offered a hoie of arrier (32% of 16

18 employees with hoie), while 13% of firms with employees (18% of employees) did so (Marquis and Long (1999); see also Frank and Rosenthal (2001) for disussion of impliations for employers adoption of formal risk adjustment). The payment methods in plae for large employers may ahieve goals similar to formal risk adjustment. In the private setor payment rates often are established by health plans, in negotiation with employers. Here we disuss two approahes to setting rates: experiene rating and ommunity rating by lass. In experiene rating, the prior year(s) health plan expenditures are used as a basis for the next year s premiums. Beause individual-level variations tend to average out as the group size inreases, health plans regard experiene data for large firms as highly redible (Sturm, 2001). Rates for very large firms (more than 1,000 employees) may be based entirely on past experiene (Sturm, 2001). If the average risk harateristis of a private employer s enrollees remain roughly onstant in a plan over time, experiene-related rates are useful in aligning premium payments with expeted osts one of the objetives of formal risk adjustment. Figure 3 draws on data from a number of soures (see Keenan et al., 2001 for methodology) to make a dramati ontrast between the way health plans are paid by the publi setor in the U.S. (Mediare and Mediaid) and by private employers. Private buyers aounting for about 80% of enrollees (60.6 million) virtually never use formal risk adjustment. Publi buyers aounting for about 20% of the enrollees (15.3 million) virtually always use formal risk adjustment. The ontrast between what the regulators hoose to do and what the private buyers hoose to do ould not be more stark. The ontrating praties of private employers are nearly the opposite of Mediare s in three respets. Employers do not set pries paid to plans. They either 17

19 aept pries set by plans, or negotiate with plans about pries. Employers do not aept any qualified plan but hoose the set of plans to ontrat with. The prie paid by the enrollee at a plan is set by the plan in the ase of Mediare, and by the employer in the private market. Table 1 summarizes these differenes. Employers buy diretly from health plans. Pries paid by an employer to a plan reflet bargaining, and the different risk distributions plans expet from different employers. Table 2 shows pries paid by several employers to several health plans in the Boston area in Two things are evident: first, employers pay different amounts to different plans for the same person. Harvard, for example, pays different pries to plans for enrolling a single employee. Seond, plans harge different pries to different employers for membership in the same plan. Harvard Pilgrim Health Care (a PPO-type plan) gets US$ 237 for a single Harvard enrollee, and US$ 295 for a single Boston College enrollee. While one an readily ome up with various reasons why pries should differ in these two dimensions, we simply make the point here that a prie paid is negotiated between the plan and the employers. Private employers do not ontrat with every plan in a market. More than half of private employees have no hoie of plans (Marquis and Long, 1999). Most state governments at like private employers when buying health insurane for employees, though a few states, suh as California, are more open with respet to plan ontrating. The federal government, as an employer, is a hybrid. It is lose to open with respet to plans seeking ontrats, but negotiates pries (based on a formula) with potential ontrators (Merk, 1999). The Federal Employees Health Benefit Plan (FEHBP) has the 18

20 authority to deny a ontrat to a potential health plan, but appears to rarely exerise this authority. We regard private employers authority to not ontrat with a plan or provider as the fundamental differene between Mediare and private buyers. The differene may be at root of explanations for why Mediare and private buyers ontrat in suh different ways. The employer s deision about whih provider or providers to offer to employees is made simultaneously with any negotiation about prie. Employers deide to offer a provider based on the prie they reeive or an negotiate, and based on the harateristis of the provider. Private employers, not the health plan, deide how muh of the prie they pay to health plans will be ontributed by workers. Numerous onsiderations ome into plan in an employer s deision about priing. Sine employer ontributions reeive favorable tax treatment in omparison to wages, employers have a reason to minimize employer ontribution. When more than one health plan is offered, an employer has an interest in making the sorting among plans be effiient and priing to workers an help lead to sorting aording to worker tastes for health insurane. One frequently advoated approah is for employers to pay the full ost of the lowest ost plan and pass on any premium above this floor to the employers (Feldman, Dowd and Coulam, 1999). This may improve sorting but it is not fully effiient sine the ost differene among plans itself differs by worker type. 10 Risk adjusting the premium harged to workers (e.g. 10 With information about the distribution of osts and tastes, a payer an hoose a prie differential with whih to fae onsumers to sort the population between plans. See Cutler and Zekhauser (2000) and Feldman and Dowd (2000) for reent treatments. In speial ases, an effiient division of a population between plans an be ahieved when the prie to onsumers for joining the more expensive plan is set so as to just ration the appropriate marginal onsumer. As the papers noted above show, this prie is not in general the differene in ost for the average onsumer in plans. In general however, a payer annot sort 19

21 harging more to older workers) ould in priniple improve effiieny of sorting, but this, or any other form of risk adjusted harge) is infrequently observed in the US ontext. It may be that any inremental gains in effiieny over average ost priing may not be worth the loss of fairness aross types of workers at a firm. Miller (2005) points out that employer profit maximization (as opposed to the riterion of ompensation ost minimization) implies that if an employer has some monopoly power in selling health insurane to its workers, the prie it harges for a more extensive overage plan should also reflet a markup on the extra ost. In sum, private employers address seletion by struturing the hoies faing their employees, both in terms of the plans the employees may hoose among, and in terms of the pries the employees pay to join a plan. Suess of this payment poliy an be gauged by two sets of figures. First, employer buying poliy has transformed the health insurane market from one in whih most employees were in an unmanaged fee-for-servie plan to one in whih this style of health insurane has virtually disappeared. Figure 4 shows that newer forms of health insurane, frequently paid by (non-risk adjusted) negotiated per apita payments are now predominant in the U.S. private market. Seond, over the period of introdution of these new forms of insurane basially the 1990s -- premium osts in the U.S. grew at historially low rates. See Figure 5 for illustration of data from California over that period. 11 enrollees effiiently by using demand-side pries for plans sine these pries would need to be personspeifi. 11 Data in Figure 5 are for the Offie of Personnel Management (OPM) of the U.S. federal government. Beause of OPM purhasing rules, the OPM premiums trak the private market premiums very losely. 20

22 CASE 4: The Regulator Can Observe Quality but Consumers Cannot-Choosing What to Report The basi adverse seletion result presented above requires that onsumers an observe all plans' quality before hoosing a plan. The fat that onsumers are able to observe the quality of eah servie eah plan provides enables the plans to distort quality in a way that will attrat only one type of onsumers but not the other. In reality, onsumers annot observe (many important aspets of) plans' quality and they often rely on information provided to them either by the plan itself or by some third party, before making their hoie. In suh a ase, the general onsensus among health poliy makers in the U.S. is that by improving what onsumers know, health are markets will funtion better. Better-informed onsumers may hoose providers more appropriately. Furthermore, onsumers hoosing on the basis of quality onveys inentives to providers to improve quality in the first plae. These arguments are why publi regulators and publi groups suh as business oalitions are making an effort to disover and reveal harateristis of providers' quality of are. The point we would like to make, however, is that revealing more information is a double-edged sword for hroni and other illnesses in poliy ontexts in whih fores of adverse seletion are also affeting quality. From the standpoint of inentives to providers and plans, revealing more information about the quality of are for onditions like mental illness an exaerbate inentives to redue quality. Put bluntly, attrating onsumers who value the quality of hroni and mental health are may be exatly what health plans may not want to do. 21

23 The following simple but realisti senario illustrates the poliy problem. Mediaid eligibles in the Boston area an hoose among several managed are plans. 12 If an eligible joins a plan, the plan gets a apitation payment. Suppose the plans differ in the quality of mental health are they offer. For example, a better plan might have a larger network of more experiened therapists and/or wider overage of drugs in its formulary. Suppose this information is known imperfetly by eligibles. What are the effets of reporting more omplete information about the relative qualities of mental health are of the plans? One effet the good effet is that Mediaid eligible needing mental health treatment an join the plan with better overage. However, there is a seond effet: plans inentives hange when the information is more aurately known. A higher quality mental health are now means that the plan is more likely to attrat those who value mental health are. If attrating these eligibles hurts the plan finanially, and the evidene is lear that it does, the plan has an inentive to redue the quality of its mental health are, when more information about the quality of this servie is reported to benefiiaries. A similar story an be told about Mediare benefiiaries and persons with employerbased overage, adding up to a market in whih foring plans to dislose the quality of their mental health departments might undermine the quality of are offered to enrollees. The general point is that the poliy hoies are not simply to report or not to report. The question that the regulator should ask itself is what is the best way to struture the reports to onsumers to give them what they need to make hoies but at the same time avoiding the danger of reating inentives to redue quality of are for hroni and other onditions. One suh a diretion is the following: instead of providing onsumers with a separate rating of the quality of every servie the health plan provides, the 12 Mediaid is a state-run program for low-inome individuals. 22

24 regulator would group servies together and provide onsumers only with the average rating of the quality of the different servies in the group. For example, instead of getting a separate rating of the network of mental health are from primary are, onsumers would get one network rating averaging the harateristis of both. While some information is obviously "lost" by this average in what is transmitted to onsumers, from the stand point of inentives to the plan to maintain quality of mental health, there has been a gain. In the presene of an averaged report, if the plan were to redue the quality of its mental health network, it would redue the overall network rating, and the plan would loose enrollees who value primary are (the "winners") as well as enrollees who valued mental health are (the "losers"). Tying qualities for various servies together in an averaged report forges a positive link between plan's quality hoie in a servie subjet to adverse seletion to overall plan profitability. Our idea of averaged quality reporting an be easily demonstrated with the model disussed above. For a given quality pair (qa,q), hosen by a plan, and 0 < α < 1, let (6) qα = αqa + (1 α)q be the (weighted) average quality of this plan. Assume that individuals annot observe (qa,q). Assume that the regulator, who an observe, (qa,q), hooses to inform onsumers only about the average quality, q α, of eah plan. That is, individuals annot observe the quality of eah of the servies a plan offers, but they an observe some summary indiator of the plan s quality profile, reported to them by the regulator. The fat that individuals an only observe the "average" quality of eah plan, and not the quality of eah servie a plan offers, will affet the market equilibrium. The 23

25 profitability of any quality pair ( q a, q ), offered by a plan, depends on individuals beliefs about the quality of eah of the two servies, given that they an only observe the average quality of the servies. Therefore, in order to analyze ompetitive equilibrium in the market where individuals only observe average quality of eah plan, one needs to inorporate individuals' beliefs about quality in the definition of equilibrium. We apply the following definition. A ompetitive equilibrium is a set of quality pairs offered by plans and a set of individuals belief funtions that speify for eah individual her beliefs about the quality pair of eah plan, for every possible average quality q α of that plan, 13 suh that: (i) eah plan maximizes its profit given all the other pairs offered and given individuals beliefs, (ii) eah individual hooses a plan that offers her the highest expeted utility given her information and given her beliefs, (iii) there is no quality pair outside the equilibrium set that if offered will make a positive profit and (iv) in equilibrium, individuals beliefs are onfirmed. Optimal Quality Reporting The following result (studied in Glazer and MGuire, forthoming) portrays the theoretial potential of averaged quality reporting: Proposition: Suppose that all individuals an only observe the weighted average quality of eah plan and α = α, where α is given by 13 Formally, the assumption is that for eah onsumer k there is a belief funtion B k :R + R + x R +, suh that for every average quality q α, the funtion speifies beliefs about a plan s quality profile (q a,q ) given that average quality. In order to simplify the analysis, we assume that a onsumer s beliefs depend only on the plan s average quality and not on the plan s identity or the average quality of the other plans. It an be shown that all our results will hold if we allow for more general belief funtions. 24

26 p ac a (q a ) p C (q ) (7) =, α 1 α then all plans offer the soially effiient quality pair (qa,q) in the ompetitive equilibrium. A detailed proof of this result an be found in Glazer and MGuire (forthoming). Here we provide a sketh of the proof. The first observation to make is that for every α, 0 < α < 1, if individuals an only observe the average quality q α of eah plan and if, in equilibrium, a plan offers the quality pair q,q ) and a share λ, 0 λ 1, of the ( a individuals that join this plan are of type H, then it must be that (8) p C (q ) a a α a = p C (q ) 1 α where (9) p = λ p H + (1 λ )p L. The intuition for this first result is quite simple and very general. Sine individuals an only observe the average quality of all the servies a plan offers, the plan has no inentive to provide a quality profile that yields the same average as another quality but osts more. Condition (8) above (later referred to as the inentive ompatible (IC) ondition) speifies the quality pair that minimizes the plan's osts given a prespeified level of quality. For a given α, 0 < α < 1, the urve IC(α, λ) in Figure 6 represents all quality profiles that satisfy equation (8) above, for the (pooling) ase where λ = λ, i.e., the ase 25

27 where the plan attrats a random sample of the population. The urve depits all quality profiles that satisfy the zero profit (pooling) ondition: r in that figure (10) p C (q ) + p C (q ) = r a a a The seond part of the proof of the Proposition is to show that, for every α, the quality pair that is loated at the intersetion of these two urves, denoted by (q,q ), is the unique quality pair offered by the plans in the ompetitive equilibrium in the ase where individuals an only observe the average quality of eah plan. α α In order to show that (qa,q ) is indeed an equilibrium, assume that (all) individuals beliefs are suh that for every average quality ( q α ) they observe, and for every plan, individuals believe that the plan has hosen the quality profile that satisfies α a α the IC ondition, with λ = λ, that yields q α. Thus, individuals believe that among all quality profiles that yield an (observed) average quality, the plan offers the one loated on the urve IC(α, λ) in Figure 6. Given these beliefs, one an see that if all plans offer α a α the quality pair (q,q ), individuals' beliefs will be onfirmed. Given the beliefs above, one an also see that if all plans offer the quality profile (q,q ), no plan has an inentive to deviate and, hene, it is an equilibrium. α α In order to show that (qa,q ) is the unique equilibrium, notie first that there annot be any other (pooling) equilibrium in whih eah plan attrats a random sample of the population. Using the single rossing property one an also show that in the ase where onsumers observe only an averaged quality, there annot exist a separating equilibrium. α a α 26

28 The last part of the proof is to show that when α is given by the ondition in (7) the equilibrium quality is the soially desired one. This, however, is straightforward given the way α is defined. Returning to equation (7), one an see that the relative weights on quality of servies a and depend on the probabilities of the two illnesses and the marginal osts of quality at the effiient quality. The quality weights are like relative pries equal to marginal ost at the effiient level of prodution. Casting quality reports as a poliy instrument turns up an important and simple onlusion: an averaged quality report an remedy adverse seletion inentives in markets for health plans. The reasoning is straightforward. Averaging quality aross its many dimensions and reporting only the average enfores pooling in health insurane. Choosing the weights in the average to reflet relative marginal benefits at the effiient quality mix ensures health plans alloate resoures to elements qualities in the right way. The power of quality reporting to orret seletion-related inentives seems not to have been appreiated previously. For purposes of omparison, as we have shown here, in the basi adverse seletion model, an averaged quality report mathes the performane of optimal risk adjustment. CASE 5: Consumers Can Observe Quality and the Regulator Can Observe "Signals" about Consumers' Types-Risk Adjustment When a plan is paid using risk adjustment, the premium paid to the plan (often referred to as "apitation") is onditioned on observable harateristis of the enrollee. The apitation payment might be based, for example, on the enrollee s age, with older 27

29 enrollees having higher payments assoiated with them beause they are expeted to ost more. Methods of risk adjustment are onerned with how muh more to pay the plan for an older enrollee than for a younger enrollee (see Cutler and Zekhauser, 2000 and van de Ven and Ellis, 2000). Two approahes to addressing this question have been taken in the eonomi literature. Conventional risk adjustment sees the goal of risk adjustment as to pay plans as lose as possible to the amount the enrollee is expeted to ost. If an older enrollee is expeted to be twie as expensive as a younger enrollee, onventional risk adjustment would pay twie as muh for the older enrollee. Many fators other than age matter for expeted osts. Researh on onventional risk adjustment is statistial and data oriented. Researhers seek to find the right ombination of variables (referred to as risk adjustors) to inlude in regression models so that the explained variation in health are osts is high, without relying on risk adjustors that are diffiult to ollet in pratie or an be manipulated by providers seeking to inrease revenue. The premise behind this researh sometimes regarded to be so obvious as to not require justifiation or analysis -- is that the health are market in question will funtion better the better job the regression model an do in prediting health are osts of enrollees. Optimal risk adjustment methods also yield an answer to how muh more to pay for an older enrollee but by a different method. Optimal risk adjustment views risk adjustment as a set of inentives aimed to indue providers to behave in aordane with some well defined objetive. Calulating the optimal risk adjustment begins with an expliit assumption about the funtioning of prie in the relevant market and a model that relates the terms of that prie (e.g., the payment for young and old) to the behavior of 28