Pricing behaviour of nonprofit insurers in a weakly competitive social health insurance market

Journal of Health Economics 30 (2011) 439–449

Contents lists available at ScienceDirect

Journal of Health Economics journal homepage: www.elsevier.com/locate/econbase

Pricing behaviour of nonprofit insurers in a weakly competitive social health insurance market Rudy C.H.M. Douven a,∗ , Frederik T. Schut b a b

CPB, Netherlands Bureau for Economic Policy Analysis, P.O. Box 80510, 2508 GM, The Hague, The Netherlands Department of Health, Policy and Management (iBMG), Erasmus University Rotterdam, The Netherlands

a r t i c l e

i n f o

Article history: Received 20 November 2009 Received in revised form 14 December 2010 Accepted 19 December 2010 Available online 13 January 2011 JEL classification: I11 I18 L11 D41

a b s t r a c t In this paper we examine the pricing behaviour of nonprofit health insurers in the Dutch social health insurance market. Since for-profit insurers were not allowed in this market, potential spillover effects from the presence of for-profit insurers on the behaviour of nonprofit insurers were absent. Using a panel data set for all health insurers operating in the Dutch social health insurance market over the period 1996–2004, we estimate a premium model to determine which factors explain the price setting behaviour of nonprofit health insurers. We find that financial stability rather than profit maximisation offers the best explanation for health plan pricing behaviour. In the presence of weak price competition, health insurers did not set premiums to maximize profits. Nevertheless, our findings suggest that regulations on financial reserves are needed to restrict premiums. © 2011 Elsevier B.V. All rights reserved.

Keywords: Health insurance Nonprofit firms Pricing behaviour Managed competition

1. Introduction During the last fifteen years several countries (e.g. Germany, the Netherlands and Switzerland) have introduced some form of managed competition in their social health insurance schemes. The goal of these reforms is increase incentives for efficiency while maintaining universal access to affordable health insurance. An important question, that has not been addressed so far, is how competing health insurers set prices if they are not allowed to distribute profits. In other words, which factors determine nonprofit health insurers’ pricing decisions? The answer to this question may have important implications for policy decisions to impose, maintain or remove restrictions on ownership type for health insurers. In this paper we examine the pricing behaviour of health insurers in the Dutch social health insurance system where health insurers are allowed to compete for customers but – until 2006 – faced a legal non-distribution constraint. Given that prior to 2006 the market is purely nonprofit, health insurer behaviour in that

period could not have been affected by the presence of for-profit insurers. Hence, spillover effects from other ownership types were absent. This makes it possible to investigate the “pure” drivers of nonprofit insurers’ price setting behaviour. Hence, the central question is whether nonprofit health insurers set premiums to maximize profits or do they behave differently? The paper is organized as follows. Section 2 provides a brief description of the Dutch health insurance market. In Section 3, we discuss the potential impact of nonprofit ownership on insurers’ pricing decisions, and describe an empirical model to examine the price setting behaviour of health insurers in this context. Next, we discuss which variables are likely to explain pricing behaviour of nonprofit insurers and we provide descriptive statistics as well as information about the sources from which the data were obtained. In Section 5 we present the estimation results of several estimation models. Finally, we discuss the implications of our findings for the role and regulation of ownership in competitive health insurance markets. 2. Dutch social health insurance market

∗ Corresponding author. Tel.: +31 70 3383377; fax: +31 70 3383350. E-mail address: [email protected] (R.C.H.M. Douven). 0167-6296/$ – see front matter © 2011 Elsevier B.V. All rights reserved. doi:10.1016/j.jhealeco.2010.12.006

During the 1990s, the Dutch social health insurance scheme was profoundly reformed by the gradual introduction of man-

440

R.C.H.M. Douven, F.T. Schut / Journal of Health Economics 30 (2011) 439–449

aged competition among health insurers. These health insurers originated from community-based nonprofit organizations, known as sickness funds. Since 1992, sickness funds (hereafter called health insurers) were allowed to compete on price for a legally standardised basic benefits package. Coverage was financed by a combination of community-rated premiums – set by the health insurers – and uniform income-related contributions – set by the government. A system of risk-adjusted premium subsidies was introduced to compensate health insurers for enrolees with predictable high medical expenses. This risk equalization system had to counteract incentives for risk selection stemming from community rated premiums and had to guarantee a level playing field for health insurers. Freedom of choice of health insurers was introduced by requiring health insurers to accept all eligible applicants. Initially (since 1992) open enrolment periods were biennial, but since 1996 people were allowed to switch health insurers once every year. Furthermore, in 1992 the legally protected territorial monopolies of health insurers were abolished and new health insurers were permitted to enter the market. Finally, to provide health insurers with the opportunity to manage care in 1992 the government abolished the obligation for health insurers to contract with any willing provider, except for hospitals and medical specialists. In 2006 the scope of managed competition model was broadened from about two thirds to the entire population by the introduction of a new Health Insurance Act. Since then former social health insurers (sickness funds) and former private indemnity insurers (mutual as well as for-profit companies) are allowed to compete for providing basic health insurance coverage. The new Health Insurance Act also created much more opportunities for health insurers to offer preferred provider policies and to manage care (Van de Ven and Schut, 2008). The primary reason for the introduction of price competition and freedom of choice of health insurers was to increase the incentives for health insurers to improve the efficiency of health care. Prior to the reforms in the 1990s health insurers were completely retrospectively reimbursed for the medical expenses of their enrolees and consequently had no stake in a more efficient provision of medical care. Until 2006, all health insurers offering basic social health insurance (sickness funds) were required by law to be either a foundation or a mutual company, implying that they were not allowed to distribute profits. In addition they were required to have statutory provisions that guaranteed a “reasonable degree of influence” of the insured in the governance of the organization (Sickness Fund Act, Article 34). 3. Modeling pricing behaviour of nonprofit health insurers In a competitive market with profit maximizing firms pricing behaviour is expected to be determined by the inverse elasticity pricing rule: P=

mc . 1 + (1/e)

(1)

According to this rule, price (P) is determined by willingness to pay and the demand elasticity (e) on the one hand, and the marginal cost (mc) of production on the other hand. If the market is perfectly competitive all competitors, irrespective of ownership type, have to pursue profit maximization in order to survive. In most markets, however, competition is imperfect and this offers the opportunity for firms to pursue other goals. Therefore, in markets in which both for-profit and nonprofit firms compete – as is typically the case in many insurance markets – both types of firms may display different pricing behaviours. For instance, in the context of the US health insurance market Barrish (2004) and Nudelman and Andrews (1996) argue that the lower operating margins reported

by nonprofit health insurers very likely reflect the organizations’ corporate missions to serve their communities by minimizing the cost of coverage and their ability to invest all gains back into the company for the future benefit of their customers. Pricing differences may also reflect differences in the quality of care of for-profit and nonprofit health plans (Schneider et al., 2005). In case of strong price competition, however, nonprofit firms are forced to adapt their pricing behaviour to that of for-profit competitors. Empirical evidence from the US hospital sector shows that spillover effects can be significant. Kessler and McClellan (2002) find that the presence of for-profit hospitals induces significant expenditure savings among nonprofit hospitals, while Silverman and Skinner (2004) find that nonprofit hospitals are more inclined to upcoding in markets with a larger for-profit hospital share. In the Dutch social health insurance market, prior to 2006 for-profit insurers were not allowed. Hence, until 2006 potential spillover effects from the presence of for-profit insurers were absent and “pure” nonprofit insurer pricing behaviour can be observed. In a market where all competitors are non-profit, pricing behaviour may not be driven by profit maximisation. If competition is strong, however, prices would be forced down to the competitive level, in which case there would presumably be little difference between pricing behaviour of nonprofit and for-profit insurers. Empirical studies show, however, that consumer price sensitivity in the Dutch social health insurance market was very low and therefore competitive pressure was rather weak (Schut and Hassink, 2002; Schut et al., 2003; Douven et al., 2007; Van Dijk et al., 2008). If competition is not sufficient to drive premiums down to the competitive level, the question becomes how nonprofit health insurers set premiums. Do they raise premiums to increase profits or do social motives play an important role and do they charge premiums that are just sufficient to cover average cost? Are there persisting differences among health insurers in costs and premiums over time? Do different health insurers follow different pricing strategies? To answer these questions we investigate what factors could have explained pricing decisions by health insurers over the period 1996–2004. Health insurers operating in the social health insurance market during this period sell both mandatory basic insurance and voluntary supplementary coverage. Since basic insurance comprised about 95 percent of insurers’ revenues, we focus on factors that explain the price setting of basic coverage. The basic benefits package is standardized by law and since health insurers were not actively involved in selective contracting or managing care, the basic insurance product is quite homogeneous (Lieverdink, 2001). Although insurers offered essentially the same basic insurance product, the cost of providing this product may vary across health insurers due to the different risk profiles of the insured population. To a large extent, the cost differences associated with different risk profiles were moderated by a system of risk equalization and by retrospective compensation payments (Van de Ven et al., 2004). Nevertheless, the remaining cost differences may have a substantial impact on insurers’ pricing decisions. In addition, other factors that may determine price setting are the (statutory) goals of the nonprofit insurer (or corporate mission), the degree of competition in the relevant market, the level of financial reserves and government regulations, such as solvency requirements. In general, the insurer’s pricing model that will be estimated can be described as: Pit = t + ˛i + Xit ˇ + εit .

(2)

In this standard panel data model out-of-pocket-premiums, of insurer i at year t is explained by a set of fixed year effects, t , and fixed insurer effects ˛i . The explanatory variables are represented by Xit , and εit represents the error term. The error terms are robust

R.C.H.M. Douven, F.T. Schut / Journal of Health Economics 30 (2011) 439–449

441

Table 1 Number and size of health insurers.

Population size (millions) Total number of health insurers Number of health insurers leaving Number of health insurers entering Health insurers with population <10,000 Health insurers with population <50,000 Health insurers with population <100,000 Health insurers with population >500,000 Health insurers with population >800,000 HHI

1996

1997

1998

1999

2000

2001

2002

2003

2004

9.8 27 0 1 3 5 6 7 2 622

9.9 29 0 2 5 7 8 7 2 622

9.9 29 2 2 6 8 10 8 3 663

9.9 29 0 0 6 8 10 8 3 663

10.3 26 3 0 2 6 7 9 4 707

10.3 24 2 0 1 5 6 9 4 711

10.2 21 4 1 0 4 6 7 6 893

10.1 21 0 0 0 4 6 7 5 886

10.2 21 0 0 0 4 6 8 5 878

for heteroskedasticity and serial correlation (i.e. the standard errors are clustered by insurer). 4. Data sources, variables and descriptive statistics In order to estimate the effect of the potential determinants of health insurer pricing behaviour on actual premiums, we constructed an unbalanced panel of 32 social health insurers over the period 1996–2004. The year 1996 was chosen as starting year because, since then, health insurers have been increasingly put at risk for the medical expenses of their enrolees and began to set different out-of-pocket premiums. The panel is unbalanced because the number of health insurers fluctuated over time due to mergers and entry of new health insurers. In Table A1 in the Appendix we give a description of the insurers and our data. Except for data on supplementary insurance, all data were obtained from the Health Care Insurance Board (CVZ), an independent government organization charged with the execution of the risk equalization system. Data on supplementary insurance were derived from a commercial database for insurance agents (ROLLS), a website comparing insurance products (www.independer.nl), or were directly obtained from (websites of) individual insurers. Basic features of the social insurance market are presented in Table 1. Over the entire period the number of health insurers decreased, particularly during the first years of the new millennium. After an initial increase, the number of small health insurers decreased as well. Almost all health insurers leaving during the sample period merged with other insurers, while new insurers entering always started with a relatively small population. The data show a clear trend of health insurers becoming larger and the market becoming more concentrated. This is confirmed in the last row by the concentration measure HHI (Herfindahl–Hirschmann index), which shows an upward trend.1 Below we discuss the variables used in our pricing model. 4.1. Dependent variable Non-profit health insurers get revenues from two different sources: (1) community rated premiums from their subscribers and (2) risk-adjusted premium subsidies, retrospective compensation payments plus a fixed administrative fee per subscriber from the Health Care Insurance Board (CVZ). Risk-adjusted premium subsidies and administrative fees are determined by the government and are financed by mandatory income-related contributions from employers and eligible employees, self-employed persons, social security beneficiaries and pensioners. The risk-adjusted premium subsidies are set at the average expected cost for each risk category (based on age, gender, region, employment status and several

1

The HHI is equal to the sum of the squared market shares of all health plans.

health status indicators) minus a fixed flat rate per subscriber, known as the “administrative premium”. If the government’s prediction of the expected cost for each risk category is perfect, an average insurer would break even if it would set its community rated premium equal to this administrative premium. Hence, the difference between community-rated and administrative premium (PDIF) provides an indication of the loading fee charged by health insurers. In our panel estimations PDIF will be the dependent variable. As shown in Table 2 the mean PDIF displays a substantial variation over time.

4.2. Explanatory variables We distinguish several factors that may explain health insurers’ pricing behaviour.

4.2.1. Uncompensated medical expenses In health insurance markets expected medical expenses are likely to play an important role in insurers’ price setting. Without risk equalization the community rated premium of an insurer would reflect the risk profile of its insured population. By contrast, in case of perfect risk equalization, differences in risk profile would not affect premiums. Since the Dutch risk equalization is not perfect, differences in risk profiles of the population may yield different expenses, insofar these different medical expenses are not adequately predicted by the risk-adjusted capitation payments (Douven, 2004). Differences in medical expenses among insurers may also reflect differences in efficiency, quality or unexpected events in the provision of health care. To some extent these differences in medical expenses are retrospectively compensated. These ex post compensation payments are based on the medical losses specific to each insurer. For instance, insurers receive a 90% compensation of losses if the annual medical expenses of an enrolee exceed a certain threshold (Douven, 2010). The remaining difference is defined as uncompensated medical expenses. We expect that the average level of uncompensated medical expenses per adult enrolee (UMEX) is positively correlated with the premium charged.2 Table 2 shows that UMEX has a positive mean in most of the years, implying that on average insurers’ actual medical expenses were higher than predicted by the government, which was particularly the case in 2002 and 2003. Since health insurers may respond asymmetrically to gains and losses we have split up UMEX in UMEX+ , in which case negative values are set to zero, and UMEX− , in which case positive values are set to zero.

2 Children under the age of eighteen are not required to pay out-of-pocket premiums. Since only adults pay an out pocket premium we calculated medical expenses, administrative expenses and financial reserves also per “premium paying” adult enrolee.

442

R.C.H.M. Douven, F.T. Schut / Journal of Health Economics 30 (2011) 439–449

Table 2 Descriptive statistics. 1996

1997

1998

1999

2000

Dependent variable PDIFi : community rated premium of insurer i minus the fixed administrative premium (in euros) Mean 35.0 27.2 27.1 45.0 48.6 Standard deviation 4.0 7.8 8.1 12.6 16.5 Maximum 42.9 37.4 37.4 66.1 82.3 Minimum 26.6 −5.7 −5.7 22.6 15.6 Explanatory variables UMEXi : uncompensated medical expenses per enrolee for insurer i (in euros) Mean 13.6 −3.4 24.9 10.6 3.2 Standard deviation 9.4 21.3 17.8 18.7 45.5 Maximum 28.0 25.5 46.0 40.4 98.1 Minimum −9.9 −4.9 −2.4 −4.5 −116.6 UADMi : uncompensated administrative expenses per enrolee for insurer i (in euros) Mean 8.1 14.5 16.5 17.0 31.6 Standard deviation 12.8 13.9 15.9 16.0 35.3 Maximum 46.9 46.2 57.3 53.2 165.1 Minimum −2.8 0.2 −0.5 1.9 5.6 RMINi : reserves in excess of minimum solvency margin per enrolee for insurer i (in euros) Mean 64.9 79.9 55.3 65.6 81.1 Standard deviation 41.0 71.1 43.1 56.7 61.9 Maximum 191.2 362.5 134.8 159.4 218.6 Minimum 4.8 7.1 −31.2 −47.4 −36.4 RMAXi : maximum minus actual financial reserves per enrolee for insurer i (in euros) Mean 101.2 Standard deviation 108.1 Maximum 376.6 Minimum −49.2 MSHAi : market share of health insurer i (as % of the insured population) Mean 4.17 4.17 4.35 4.35 4.00 Standard deviation 2.99 3.00 3.24 3.24 3.58 Maximum 11.79 12.15 12.29 12.40 12.31 Minimum 0.11 0.13 0.15 0.18 0.09 PSUPi : annual premium for supplementary insurance charged by insurer i (in euros) Mean 59.7 64.3 66.8 74.5 83.4 Standard deviation 21.5 22.0 21.7 18.4 26.7 Maximum 98.0 98.0 108.0 117.1 179.7 Minimum 27.2 27.2 27.2 39.5 49.8

4.2.2. Uncompensated administrative expenses A second potential source of premium variation is differences in administrative expenses. Administrative expenses may vary because of differences in service level, marketing expenses or administrative efficiency. We define the explanatory variable UADM as the difference between the actual administrative expenses per enrolee and the fixed administrative fee insurers receive from Health Care Insurance Board (CVZ). Table 2 shows that for almost all health insurers administrative expenses exceed the administrative budget determined by the government.3 For some health insurers in some years the losses exceed 100 euros per adult enrolee, despite that administrative expenses typically are less volatile, and therefore better predictable, than medical expenses. A caveat here is that the reported administrative expenses of providing basic insurance may not fully reflect true cost. Although basic and supplementary health insurance are formally separated and offered by different legal entities, both entities were always part of the same company. More than 90% of the compulsorily insured population bought supplementary health insurance coverage and

3 Since UADM is positive for almost all health insurers we did not split this variable to account for possible asymmetric pricing behaviour, as in case of UMEX. One of the reasons for all health insurers exceeding the administrative budget was that they were also responsible for providing mandatory long term care insurance. The separate administrative budget for providing long term care was typically not sufficient to cover the incurred administrative expenses. Since health insurers had to finance any deficit on administering long term care out of the revenues from health insurance, this was likely to result in higher health insurance premiums (CTZ, 2000). Therefore we included the administrative expenses for carrying out long term care in the figures.

2001

2002

2003

2004

16.6 26.2 76.3 −14.9

27.6 25.1 83.8 −41.0

87.7 32.8 133.0 −17.6

82.6 30.9 136.2 −6.6

−0.6 43.9 62.8 −124.8

37.1 33.9 78.4 −66.3

38.2 40.0 76.1 −98.7

41.5 26.1 98.9 4.8

39.9 28.3 133.6 8.8

26.2 28.9 129.9 −11.8

77.3 62.7 239.1 1.2

49.1 58.3 177.2 −38.1

44.8 54.0 177.7 −27.8

132.6 100.9 358.1 12.2

175.3 97.0 391.8 −7.0

197.6 109.6 480.0 −12.4

4.17 3.58 12.56 0.10 89.0 29.6 196.0 54.2

4.76 4.57 15.12 0.10 105.5 23.0 186.0 67.2

4.76 4.53 14.77 0.19 116.0 32.7 204.0 69.6

172.5 46.8 288.0 99.0

almost all of them (98%) from the same company. Due to the joint marketing and administration of basic and supplementary insurance, companies might have been able to shift administrative costs from basic to supplementary health insurance (and vice versa).4 4.2.3. Financial reserves The magnitude of an insurer’s financial reserves constitutes another potential determinant of its pricing behaviour. Differences in accumulated reserves may lead to different pricing strategies. For instance, health insurers with large financial reserves may lower premiums to gain market share at the expense of profitability, while health insurers with limited financial reserves may opt for high premiums to regain a sound solvency position. Grossman and Ginsburg (2004) argue that in the US the accumulation and subsequent spending down of financial reserves by the non-profit Blue Cross and Blue Shield plans – as opposed to commercial insurers who had to distribute excess reserves to shareholders – helped to drive the underwriting cycle (periodic up and downward fluctuations in premiums and profitability). In addition, the regulations with respect to health insurers’ financial reserves may have an impact on price setting behaviour.

4 The problem of cost shifting has been recognized by the Health Insurance Authority (CTZ), which at that time was charged with the supervision of the social health insurance schemes (CTZ, 2002). The CTZ found evidence of a large variation between companies in the ratio of administrative costs of basic and supplementary health insurance and concluded that accounting rules to allocate administrative costs were often unclear, particularly in case health plans belonged to a large holding company.

R.C.H.M. Douven, F.T. Schut / Journal of Health Economics 30 (2011) 439–449

The government not only requires that financial reserves meet a certain minimum solvency margin, as usual in many countries, but since 2001 also that financial reserves do not exceed a certain maximum level. Health insurers close to the minimum level may be forced to raise premiums, while the opposite holds for health insurers close to the maximum level. Obviously, if the maximum reserve requirement would be a binding constraint for a substantial number of health insurers, this would be a strong indication that price competition is not very effective. The required minimum solvency margin equals about 8% of the annual medical expenses for which the health insurer is at risk.5 As an indicator of a health insurer’s wealth we use the financial reserves in excess of the minimum solvency margin (RMIN), which is defined as the financial reserves minus the required solvency margin per adult enrolee. Table 2 shows that the mean RMIN sharply decreased in 2002 and 2003. This was caused by losses incurred on medical and administrative expenses (see also Table 2) and by the crash in the stock market. Table 2 also shows that in several years of the study period some health insurers were in financial trouble since the minimum of RMIN was below zero. In 2003, four health insurers had to borrow money to meet the solvency requirements. We expect that a health insurer with RMIN that is close to or even below zero will have to increase next year’s premium relative to its competitors. Since 2001, the law imposes also a maximum limit to the financial reserves of a health insurer. This maximum limit is set at about 2.5 times the required minimum solvency margin. A health insurer exceeding the maximum limit has to return the surplus to the Health Care Insurance Board (CVZ) or has to reduce its premium. For health insurers the latter strategy is of course much more attractive, and in practice no health insurer had to refund any surplus. In our panel we use the variable RMAX, which is defined as the residual reserve capital per adult enrolee that a health insurer can accumulate before the maximum limit is reached. Although the maximum limit to financial reserves was imposed in 2001, health insurers may well have anticipated to this regulation in the preceding year, since the first proposal to put a limit to health insurers’ financial reserves dates back to 1999. Therefore, in our panel we chose 2000 as starting year. We expect that if a health insurer’s RMAX is close to zero – implying that its financial reserves approach the maximum limit – it will reduce next year’s premium relative to its competitors. Table 2 shows that RMAX increases over time, but that there is a large variation among health insurers. Each year some health insurers are close to the maximum limit. Since RMIN and RMAX are strongly (negatively) correlated and since we expect different pricing behaviours of health insurers close to the minimum and maximum reserve limits, we distinguish three explanatory variables (RLOW, RMED, and RHIGH) which are defined as: – RLOW = RMIN if RMIN < 50, else RLOW = 50, – RHIGH = RMAX if RMAX < 50, else RHIGH = 50, – RMED = RMIN − 50 if RLOW = 50 and RHIGH = 50, and RMED = 0 in all other cases. Our hypothesis is that in case an insurer’s financial reserves are close to the required solvency margin (RLOW), or close to the maximum limit (RHIGH), the insurer sets premiums differently than

5 Parts of the medical expenses (e.g. fixed hospital costs and individual costs exceeding a certain threshold) are largely retrospectively compensated. Since this proportion has been steadily reduced over time, health plans’ financial risk increased from 13% in 1996 to about 53% in 2004 (Van de Ven et al., 2004). Hence, the required solvency margin increased simultaneously.

443

in case its financial reserve level is safely away from both limits (RMED). We expect that if an insurer’s financial reserves are less than 50 euros above the minimum level, the insurer will have a stronger incentive to raise its premiums, whereas the minimum reserve level does not have an impact on the pricing behaviour of other insurers.6 Hence, we expect that RLOW will have a negative sign. Equivalently, we expect that the coefficient of RHIGH will have a positive sign, that is the lower RHIGH (i.e. the closer to the maximum limit) the stronger the incentives for a health insurer to reduce its premium. We expect that RMED will not have a significant impact on insurer’s pricing behaviour, since financial reserves that are safely away from the legal minimum and maximum levels are not likely to generate incentives to adjust premiums. 4.2.4. Prices of closest competitors Prior to 1992 each health insurer had a separate legally defined regional catchment area. These territorial monopolies were abandoned by a change in legislation in 1992, and since then health insurers have been allowed to compete all over the country. Consequently, all health insurers decided to expand their statutory catchment area to the national level. Because of the open enrolment requirement, which was also introduced in 1992, health insurers could easily penetrate any former regional catchment area of another insurer. Therefore the Netherlands can be considered to be the relevant geographic market for health insurance. As argued by Moriya et al. (2010) the geographic market for health insurance may be smaller than nationwide (or statewide, in case of the US) if managed care ties insurance to a particular set of regionally based providers. During our study period, however, health insurers did not engage in selective contracting or in managed care activities, so this potential market restriction does not apply. On the other hand, until 1992 Dutch citizens were not used to switch to another health insurer, so they were probably only familiar with few neighbouring sickness funds in their area of residence. Therefore, they may well have a relatively strong preference for one of these original regional health insurers. Indeed, surveys among a representative panel of individuals with social insurance show that such preferences were significant (Laske-Aldershof and Schut, 2003). This implies that health insurers, despite competing for customers nationwide, might have been particularly sensitive to the pricing behaviour of the insurers with the largest market shares in their principal catchment areas (defined as the geographically closest competitors).7 If a sufficient number of consumers are primarily sensitive to prices of the largest regional insurers, past premiums set by these closest competitors may have a relatively strong impact on insurers’ pricing behaviour. We tested this hypothesis by including an explanatory variable reflecting the level of an insurer’s past premium relative to those of its closest competitors. We constructed the explanatory variable PCOMP by first defining for each health insurer a set of its regionally closest competitors and then comparing the premium for each insurer with the average of its closest competitors (for each insurer the set of closest competitors is specified in Table A1 of the Appendix). This dummy variable (PCOMP) is one if in a certain year the premium of this

6 A cut-off point of 50 euros is plausible in view of the prevailing premium levels and financial reserve requirements. Nevertheless, since the chosen of cut-off point is rather arbitrary, we used other cut-off points as well. Changing cut-off points to 25 euros or 75 euros, however, yielded similar results but with lower explanatory power (R2 ). 7 Usually, by charging lower prices sellers can increase sales because buyers will buy more and new buyers will be attracted to the market. Dutch health insurers did not have this option, however, since coverage is standardised and mandatory for a legally defined population. Therefore the aggregate price elasticity of demand for social insurance is zero by definition.

444

R.C.H.M. Douven, F.T. Schut / Journal of Health Economics 30 (2011) 439–449

20 10 0

1996

1997

1998

1999

2000

2001

2002

2003

euros

-10 -20 -30 -40 government

-50

insurers' association

-60 * Difference between predicted and actual health expenditure per enrolee

Fig. 1. Health expenditure forecasting errors* per insured by the government and insurers’ association.

health insurer is higher than the average premium of its closest competitors, and zero otherwise. 4.2.5. Market share Given that insurers compete for customers nationwide, we also included an insurer’s national market share as a determinant of health insurers’ pricing behaviour. Large health insurers may have competitive advantages due to economies of scale in administration, in purchasing care, or in reducing insurance risk. Therefore, large health insurers may be inclined to charge lower premiums. On the other hand, large health insurers may have more market power, which may enable them to set higher prices. Thus the expected effect of market share on premium is ambiguous. We define the market share of a health insurer (MSHA) as the percentage of the total number of people covered by social health insurance that is insured with that insurer. Table 2 shows that during the study period the largest health insurer had a market share of about 15%, whereas the smallest health insurer had about only 0.1% of the market. 4.2.6. Supplementary insurance premium Since health insurers typically sell basic and supplementary health insurance as a combined product, the extent of competition in the supplementary health insurance market may have an impact on pricing behaviour in the basic health insurance market. Since supplementary health insurance products are much more heterogeneous than the standardised basic benefits package, consumer search costs are higher. This may reduce consumer price sensitivity for supplementary insurance and thus may raise the opportunity for health insurers to set supracompetitive prices. Health insurers might have been inclined to use these profits to cross-subsidise basic insurance. First, they might have used the mark-ups on supplementary insurance to compensate losses on underpriced basic health insurance. Second, health insurers might also be tempted to shift administrative expenses from basic to supplementary insurance. If some health insurers exploit these opportunities for cross-subsidisation, we expect that supplementary health insurance premiums are inversely related to basic insurance premiums (all other things equal). Most health insurers offered a choice between at least three or four different supplementary benefit packages. Premiums of the various supplementary benefit packages were highly correlated. For the empirical analysis we therefore used the premium

charged for the most popular supplementary benefits package (PSUP), including at least regular dental care and prolonged physiotherapy. Although health insurers are allowed to risk rate, most of them still charge community-rated premiums for supplementary coverage and almost all health insurers do not charge premiums for children under the age of 18. This implies that observed variation in supplementary premiums may reflect differences in benefits package or differences in risk profile rather than different levels of cross-subsidisation. Hence, the estimated effect of this variable (PSUP) on price should be interpreted with caution. Table 2 shows that supplementary premiums have increased substantially over time. The exceptionally strong increase in 2004 can be largely explained by a reduction of the mandatory basic benefit package and the concurrent expansion of supplementary insurance coverage. 4.2.7. Health-insurer objectives Pricing decisions by non-profit health insurers are likely to be influenced by the goals they pursue. During the study period all health insurers were foundations or mutual companies and were legally prohibited to distribute profits to shareholders. Because there were no residual claimants, health insurers could choose and pursue different objectives. For instance, small independent regionally based health insurers often propagated their adherence to “social objectives”, while large health insurers that were part of a holding company might have been more profit-oriented, reflecting the overall goals and strategies of the holding company. Furthermore, new entrants may be primarily interested in gaining market share rather than making profit. Four categories of health insurers can be distinguished that may pursue different goals. The first category is small independent regionally oriented health insurers, which strongly communicate their adherence to “social objectives”. These health insurers might not be interested to engage in price competition to gain market share. The second category consists of large traditional sickness funds that operate nationwide and are predominantly focused on providing health insurance. These health insurers might be primarily service-oriented and more interested in securing market share and keeping customers satisfied than in making profit. The third category we discern consists of health insurers that are part of a large multi-line insurance company. We expect that these health insurers might be most profit-oriented and therefore most inclined to charge relatively high premiums in a weakly competitive market. Finally, we identify

R.C.H.M. Douven, F.T. Schut / Journal of Health Economics 30 (2011) 439–449

new entrants, who are expected to follow the most aggressive pricing strategy in order to realise a rapid increase in market share (for the categorisation of each insurer, see Table A1 of the Appendix). 4.2.8. Price signals When setting next year’s premiums, health insurers have to deal with uncertainties such as the growth of health care expenditure, government interventions in health care prices and supply, and the price setting behaviour of competitors. Price signals that can serve as focal points may be useful for health insurers to reduce forecasting errors in predicting health care cost inflation and to reduce uncertainty about government and rivals’ behaviour. Both unanticipated fluctuations in health care expenses – often associated with shocks caused by health policy changes – and unpredictable pricing strategies by competitors appear to be the driving forces behind the persistent underwriting cycles observed in health insurance markets (Grossman and Ginsburg, 2004). Each year, the Dutch government makes a forecast of next year’s health care expenditure in order to calculate the administrative premium which would be sufficient for health insurers to break even (after receiving risk adjusted premium subsidies). This administrative premium may therefore function as an indicator of a minimum premium level. During the study period, the Netherlands Health insurers Association (ZN) each year also published a forecast of next year’s health care expenditure. Except for 2001, the forecast by the insurers’ association was substantially higher than the forecast by the government (see Fig. 1). This forecast might have been used by health insurers as an indicator about the extent to which they should set their premiums higher than the administrative premium calculated by the government. We tested whether health insurers used the price signal by their interest association in setting premiums. Therefore we define the explanatory variable PSIG as the difference between the break-even premium predicted by the insurers’ interest association and the administrative premium calculated by the government. Since the price signal is the same for all health insurers, its effect can only be tested over the years in the absence of year dummies. 5. Estimation results We estimated Eq. (2) by five different types of (unbalanced) panel data models. Since health insurers have to set next year’s premium at the end of the current year, we lag the explanatory variables UMEX+ , UMEX− , UADM, RLOW, RMED, RHIGH, PCOMP and UADM by one year in our panel data estimations. The correlation matrix as displayed in Table A2 of the Appendix shows that there is no strong correlation among the explanatory variables. This indicates that multicollinearity does not play a major role. The first model (A) is a fixed effects model including 8 year and 27 health insurer dummies. The year dummy captures time-effects that are constant across health insurers. The insurer dummies capture individual health insurer effects that are constant over time. Including year dummies implies that the effect of any cross-section invariant variables, such as the annual price signal from the insurers’ association (PSIG) cannot be estimated. In the second model (B) we apply only the fixed insurer effects and therefore can include price signals from the insurers association (PSIG) in our estimations. In model C we substitute random insurer effects for the fixed insurer effects in model B. In model D we include additionally four fixed group effects, each group representing a different type of health insurers. This makes it possible to estimate the effect of the objectives of different types of health insurers. Finally, in model E we estimated a simple OLS-model. The idea behind these five estimation models is not to obtain the “correct” model but to show

445

possible effects of potential explanatory variables. The “true” effect of an explanatory variable on the dependent variable is difficult to measure when it is correlated with the fixed or random effects. Therefore, the five models present for each explanatory variable a range of possible effects. Model selection criteria indicate that the price signals from the insurers’ association in models B–E capture for a large part the variation of the year dummies in model A.8 A Hausman test indicates that random effects in models C and D are slightly preferred over fixed effects.9 Furthermore, a Lagrange multiplier (LM) test indicated that the random effects models C and D are favoured over OLS models, like model E.10 We performed various sensitivity and robustness analyses by constructing other types of variables, considering different time periods and different types of estimators. For example, we tested for the possibility of selection bias because some insurers have entered the market, finished business or merged during the study period, by estimating models A–E also for the largest balanced subpanel. The parameters turned out to be robust. In addition, we tested for the presence of autocorrelation by allowing for a first order autoregressive error structure in our panel data models. The Balthagi-Wu test did not indicate any presence of autocorrelation while the Bhargava Durbin Watson test indicated some weak form of autocorrelation. More important, however, also during these tests the estimated parameters proved to be robust. To conclude, our sensitivity analysis, of course, yielded different estimators and standard errors but these never changed drastically. We therefore believe that our general results are well captured in the estimation results presented in Table 3. As shown in Table 3, we find positive values for UMEX+ . The estimated coefficients in all models imply that a health insurer that lost money on medical expenses raised its premium, for every lost euro, between 0.15 and 0.50 euros in the next year. The values are lower than one, implying that not all uncompensated medical expenses of the previous year feed through to the premium. The positive coefficient of UMEX+ increases from estimation models A to E, suggesting that UMEX+ is positively correlated with the fixed effects in models A and B and the random effects in models C and D. A likely explanation may be that imperfect risk adjustment leads to structural differences in uncompensated medical expenses across health insurers. These structural differences persist over the years and are partly captured by the fixed or random cross-section effects.11 We observe a similar pattern for UMEX− , but we find much smaller coefficients than for UMEX+ , which are not statistically significant. This suggests that health insurers indeed react asymmetrically to gains and losses, and losses lead more quickly to premium adjustments than gains.12 The estimated coefficients for UADM, ranging from 0.08 to 0.26, imply that uncompensated administrative expenses are not fully passed on to next year’s premium. The low coefficients, however, may partly be explained by our bias in measuring administrative expenses since, as we explained earlier, health insurers could have shifted administrative expenses from basic to supplementary

8 For example, the Schwarz information criterion prefers model B over model A (1310 versus 1326), while Akaike’s information criterion only slightly favours model A over model B (1279 versus 1276). 9 For both models C and D, the Prob > 2 = 0.07 indicating a slight preference for the random effects model. 10 For models C and D the LM test yielded Prob > 2 = 0.002 and Prob > 2 = 0.004 indicating that the random effects model is appropriate. 11 In Douven (2004) it is also shown that during the period 1993–2001 structural differences across health plans exist. 12 For all models coefficient tests reject (at a 5% level) the hypothesis that the coefficient of UMEX+ equals the coefficient of UMEX− .

446

R.C.H.M. Douven, F.T. Schut / Journal of Health Economics 30 (2011) 439–449

Table 3 Estimation results for equation 2 (standard errors in parentheses). Model A Panel: fixed effects, year effects robust s.e.

Model B Panel: fixed effects robust s.e.

Model C Panel: random effects robust s.e.

Model D Panel: random effects robust s.e.

Model E Panel: OLS, robust s.e.

PSIG

0.15 (0.13) −0.04 (0.12) 0.10 (0.09) 0.39* (0.12) −0.39* (0.12) −0.06 (0.04) 5.2 (1.8) 1.6 (1.6) −1.8 (2.1) –

New entrant (Y/N)

–

0.29* (0.09) 0.04 (0.09) 0.08 (0.07) 0.28* (0.09) −0.42* (0.11) −0.04 (0.05) 4.9* (2.1) 2.8* (0.8) −1.4 (2.6) 0.77* (0.08) –

0.38* (0.10) 0.11 (0.11) 0.15 (0.08) 0.25* (0.06) −0.34* (0.13) −0.03 (0.03) 6.8* (2.3) 1.7* (0.6) 0.3 (0.6) 0.78* (0.13) –

0.50* (0.10) 0.19 (0.15) 0.26* (0.10) 0.16 (0.10) −0.20* (0.13) −0.05 (0.04) 10.4* (2.9) 0.4 (0.6) −0.2 (0.4) 0.83* (0.13) –

Regional health insurer (Y/N)

–

–

–

National health insurer (Y/N)

–

–

–

Multiline insurer (Y/N)

–

–

–

0.86 0.25 0.59 164 27

0.84 0.30 0.59 164 27

0.83 0.53 0.71 164 27

0.36* (0.09) 0.03 (0.11) 0.23* (0.11) 0.18* (0.08) −0.34* (0.13) −0.02 (0.03) 7.2* (2.1) 2.0* (0.7) −1.0 (0.6) 0.82* (0.12) −26.9* (12.6) 4.3 (4.4) 8.5 (4.5) 3.9 (9.5) 0.83 0.69 0.76 164 27

Explanatory variables UMEXi + UMEXi − UADMi RHIGHi RLOWi RMEDi PCOMPi PSUPi MSHAi

2

R (within) R2 (between) R2 (overall) Number of observations Cross-sections included *

– – – R2 = 0.75

164 27

Significant at a 5% level.

insurance. The increasing coefficients from model A to model E suggests that UADM is also correlated with the fixed and random effects. Our findings suggest that a sound solvency position is an important long term health insurer objective. We find significant (on a 5% level) negative coefficients for RLOW in all five models, implying that health insurers with a critical solvency position charge higher premiums in the next year to safeguard a sound solvency position. For example, a health insurer with a solvency position equal to the required solvency margin sets its premium about 10–20 euros higher than health insurers without a critical solvency position. We also find positive and often significant coefficients for RHIGH, implying that health insurers with a reserve capital close to the maximum limit charge lower premiums next year than other health insurers. For a health insurer at the maximum reserve limit next year’s premium is about 5–20 euros lower than the premium of other insurers (all other things equal). In all five models coefficients of RMED are negative but small and not significant. This implies that financial reserves do not play a significant role in insurers’ pricing behaviour if these reserves are well above the minimum limit and well below the maximum limit. The positive coefficients for PCOMP show that health insurers charging higher premiums than their closest competitors do not charge lower premiums in the following year. On the contrary, we find that most health insurers charging a higher premium than their competitors do so every year. This finding suggests that premium differences across health insurers persist over time, which is sup-

ported by a positive and high Spearman correlation coefficient for premiums (Van Dijk et al., 2008).13 We find no evidence of cross subsidisation from supplementary to basic insurance, since the estimated coefficients for PSUP are positive and significant (except for models A and E). However, as explained earlier, since differences in PSUP may also reflect differences in benefits package and differences in risk profile, this finding should be interpreted with caution. The estimation results of model D provide some evidence of the impact of the objectives of different types of health insurers on pricing behaviour. We find that regional, national and multiline health insurers tend to charge relatively high premiums, which may be explained by a more service than price oriented business strategy. New entrants charged on average about 27 euros lower premiums than health insurers that operated already on the market before 1996.14 New health insurers could follow such a price

13 The competition variable may bias our other estimators because it is constructed as a function of lagged dependent variables. However, dropping PCOMP as explanatory variable did not really alter other estimated parameters. We also tested for other possible competition variables. One variable was an insurers’ loss of its own enrolees in the previous year and the other one was the inclusion of region by year random effects. Only for the latter one we found sometimes a small but significant positive effect in the middle and western part of the country, indicating that in those regions insurers tend to set somewhat higher premiums. 14 We tested the hypothesis that the coefficient for the new insurer dummy (−26.9) equals the coefficient for the regional (4.3), national (8.5) or multiline insurer dummy (3.9). In all three cases this hypothesis is rejected at a 5% level.

R.C.H.M. Douven, F.T. Schut / Journal of Health Economics 30 (2011) 439–449

strategy because their medical expenses were substantially lower than of the incumbent health insurers.15 Finally, if we leave out the year dummies (as in models B–E), the positive and significant coefficient of PSIG suggests that the price signal by the health insurers’ association plays an important role in insurers’ pricing decisions. We find a coefficient close to one in all four models without year dummies. This suggests that health insurers incorporate the “mark-up” on the government’s administrative premium as projected by the insurers’ association almost fully in their premiums. If we neglect mergers or new entrants in our sample, and consider only health insurers that were on the market during the whole sample period, we find that for those insurers the “mark-up” is not significantly different from one. This suggests that health insurers “usually” fully incorporate the price signal in their premium setting and that only under special market conditions they deviate from this behaviour. This implies that the forecast by the health insurance association is an important focal point for individual health insurers to set their premium. These price signals might well have weakened competition, as they might have facilitated implicit price agreements among health insurers (Ivaldi et al., 2003). 6. Conclusion and discussion How do nonprofit health insurers set premiums in a social health insurance market with managed competition? The Dutch social health insurance market prior to 2006 offers a unique setting to answer this question. During our study period (1996–2004) health insurers were allowed to compete for customers but were not allowed to distribute profits. In addition, all health insurers were legally obliged to offer a standardized benefits package, so insurance products were quite homogeneous and primarily differed in price. Since potential spillover effects from the presence of for-profit insurers were absent, ‘pure’ nonprofit insurer pricing behaviour could be observed using a rich panel data set including all health insurers over an eight years period. Given the low price sensitivity of consumers found in other studies (Schut and Hassink, 2002; Schut et al., 2003; Douven et al., 2007; Van Dijk et al., 2008), one would expect that health insurers displaying profit-maximizing behaviour would charge high premiums and

15 These large differences in medical expenses were partly caused by inadequacies in the risk adjustment system. For example, in the year 2000 the privately insured lower-income self employed were legally obliged to obtain social health insurance coverage and most self employed opted for the cheaper entrants (Schut et al., 2003). Self employed appeared to have much lower medical expenses than average and therefore an indicator on employment status was subsequently included in the risk adjustment system in 2003.

447

realize huge profits. Indeed, Schut and Hassink (2002) show that raising premiums would be a very profitable strategy for health insurers. Instead of premiums converging towards a monopoly level, however, we find evidence of substantial and persistent premium variation. Furthermore, in most years the average mark-up was rather small. In 2001 and 2002 health insurers even incurred substantial losses, resulting in decreasing financial reserves (see Table 2). These losses seem to be caused by a general underestimation of health care cost inflation (particularly in 2002, see Fig. 1) as a result of a temporary release of government control on hospital cost in an attempt to reduce waiting lists. The incurred losses may also explain why all health insurers substantially raised their premiums in the two subsequent years. Rather than maximizing profits we find that most important drivers for health insurers’ pricing behaviour was the solvency regulation imposed by the government and the price signal by the insurers’ association. In addition, our findings suggest that setting a maximum limit to insurers’ financial reserves had a significant downward effect on insurers’ premiums. Hence, although nonprofit ownership may have prevented charging excessively high prices, regulation was needed to enforce lower premiums. Instead of by restricting ownership and imposing caps on insurers’ financial reserves, governments may also restrain premiums by strengthening the incentives for competition, which were rather weak in the Dutch health insurance market throughout our study period. As a matter of fact, strengthening competition among health insurers was a major goal of the profound health insurance reform that was implemented in 2006 (Van de Ven and Schut, 2008). A first empirical study shows that the reforms indeed induced strong price competition among health insurers and was effective in constraining health insurance premiums in the first few years after the reform (Douven et al., 2007). Acknowledgements We would like to thank the Dutch Health Care Insurance Board (CVZ) for providing most of the data for this study. We would also like to thank Riemer Faber and Pierre Koning for helping us with the econometric part of the paper and three anonymous referees for their helpful suggestions.

448

R.C.H.M. Douven, F.T. Schut / Journal of Health Economics 30 (2011) 439–449

Appendix A.

Table A1 Health insurers, data description, closest competitors and type of insurer. Health Insurers

Data informationa

Closest competitorsb

Type of insurer

1. Unive 2. Ohra 3. Anova 4. ZAO 5. OZ 6. DSW 7. ANOZ 8. AGIS 9. Salland 10. NZC 11. Topzorg 12. Pro Life 13. SR Rotterdam 14. ONVZ 15. Anderzorg 16. OZB 17. Nederzorg 18. Trias 19. Nuts 20. Azivo 21. ZK Spaarneland 22. ZK Noordwijk 23. Geove 24. ZON 25. PWZ 26. De Friesland 27. Zorg and Zekerheid 28. Groene Land 29. VGZ 30. ZK 31. CZ Groep 32. Amicon

1996–2003, large insurer, all 8 obs. in sample 1996–2000, medium size insurer, merged with Nuts, 4 obs. in sample 1996–2001, large insurer, merged in 2002 into AGIS, 6 obs. in sample 1996–2001, large insurer, merged in 2002 into AGIS, 5 obs. (1 missing) 1996–2003, large insurer, all 8 obs. in sample 1996–2003, large insurer, all 8 obs. in sample 1996–2001, large insurer, merged in 2002 into AGIS, 6 obs. in sample 2002–2003, large insurer, merger of Anova, ZAO, ANOZ, 2 observations 1996–2003, medium size insurer, 7 obs. (1 missing) 1997–1999, new very small insurer, all obs. missing 1996–1999, medium size insurer, merged with ZAO, 3 obs. (1 missing) 1996–2000, very small insurer, all obs. missing 1996–2003, medium size insurer, 4 observations in sample (4 missing) 1997–2003, medium size insurer, 4 observations in sample (3 missing) 1996–2003, small insurer, 4 observations in sample (4 missing) 1998–2003, new small insurer, 5 observations in sample (1 missing) 1998–2003, new small insurer, 4 observations in sample (2 missing) 1996–2003, medium size insurer, 3 observations in sample (5 missing) 1996–2003, large insurer, all 8 obs. in sample 1996–2003, large insurer, 7 obs. (1 missing) 1996–1997, large insurer, was financially part of ZK and merged in 1998 1996–1997, large insurer, was financially part of ZK and merged in 1998 1996–2003, large insurer, all 8 obs. in sample 1996–1999, large insurer, was financially part of Amicon 1996–2001, large insurer, 5 obs. (1 missing) 1996–2003, large insurer, 7 observations in sample (1 missing) 1996–2003, large insurer, all 8 obs. in sample 1996–2003, large insurer, all 8 obs. in sample 1996–2003, large insurer, all 8 obs. in sample 1996–2003, large insurer, all 8 obs. in sample 1996–2003, large insurer, all 8 obs. in sample 1996–2003, large insurer, all 8 obs. in sample

25, 28, 30 7, 11, 32 30 25, 27 18, 29, 30, 31 2, 20, 30 28, 32 28, 30, 32 7, 8, 28, 32

National Multiline Regional Regional National Regional Regional New Regional

2, 32

Regional

5, 6, 30 8, 27 23 28, 32 4, 8 5, 29, 30, 8 6, 20 6, 19

Regional New New New New Regional Multiline Regional

15, 28

National

1, 8, 30 7, 8, 28 8, 18 7, 8, 23, 26 5, 31, 30, 32 5, 6, 31 5, 29, 30 7, 8, 28, 29

Multiline Regional Regional Multiline National Multiline National National

a Our observation series are unbroken for single insurers but for some small health insurers we could not recover or use all financial data at the beginning of our sample period. However, the observations in our sample can be considered representative for the Dutch social insurance market since the data in our sample cover about 85% of the population in 1996, about 95% in 1997–1998, more than 99% in 1999–2000 and cover all health insurers during 2001–2003. Note that in our estimations we lag most explanatory variables with one year. For the dependent variable we have also data for the year 2004. In total we obtained 164 observations in 27 cross-sections. b For each insurer the set of closest competitors is determined by investigating which insurers have the largest market shares in regions in which the insurers’ enrolees are primarily concentrated, based on information from the Health Care Insurance Board (CVZ). Because insurers vary in size and the regional concentration of enrolees, the sets of closest competitors are not completely reciprocal (e.g. a small insurer with geographically dispersed enrolees may have many close competitors but, in turn, may not be a close competitor to many other insurers).

Table A2 Correlation matrix of dependent and explanatory variables.

PDIF UMEX+ UMEX− UADM RHIGH RLOW RMED PCOMP PSUP MSHA PSIG

PDIF

UMEX+

UMEX−

UADM

RHIGH

RLOW

RMED

PCOMP

PSUP

MSHA

PSIG

1 0.65 0.27 0.24 0.32 −0.42 −0.10 0.32 0.35 −0.02 0.69

1 0.38 −0.07 0.21 −0.17 −0.04 0.28 0.23 0.20 0.45

1 −0.36 0.26 0.01 0.13 0.10 −0.11 0.35 0.15

1 0.00 −0.30 −0.14 0.14 0.15 −0.31 0.08

1 −0.17 0.17 0.15 0.02 0.00 0.24

1 0.42 −0.14 −0.52 0.24 −0.22

1 0.00 −0.15 0.23 0.06

1 −0.07 0.00 0.02

1 −0.12 0.35

1 0.00

1

References Barrish, S.R., 2004. Nonprofit health insurers: the story Wall Street doesn’t tell. Inquiry 40 (4), 318–322. CTZ, 2000. Budget beheerskosten (Administrative budget). Publication number 7, College Toezicht Zorgverzekeringen, Amstelveen. CTZ, 2002. Doorberekening beheerskosten ziekenfondsen (Calculation of Administrative Costs Sickness Funds). Publication number 14, College Toezicht Zorgverzekeringen, Amstelveen.

Douven, R., 2004. Risk adjustment in the Netherlands: an analysis of insurers’ health care expenditures. CPB Discussion Paper 39, The Hague. Douven, R., Lieverdink, H., Ligthart, M., Vermeulen, I., 2007. Measuring annual price elasticities in Dutch health insurance: a new method. CPB Discussion Paper 90, The Hague. Douven, R., 2010. Ex-post correctiemechanismen in de Zorgverzekeringswet: hoe nu verder? (Ex Post Correction Mechanisms in the Health Insurance Act: How to Proceed?). CPB Document 212, The Hague.

R.C.H.M. Douven, F.T. Schut / Journal of Health Economics 30 (2011) 439–449 Grossman, J.M., Ginsburg, P.B., 2004. As the health insurance underwriting cycle turns: what next? Health Affairs 23 (6), 91–102. Ivaldi, M., Jullien, B., Rey, P., Seabright, P., Tirole, J., 2003. The Economics of Tacit Collusion, Final Report for DG Competition. European Commission. Kessler, D.P., McClellan, M.B., 2002. The effects of hospital ownership on medical productivity. RAND Journal of Economics 33 (3), 488–506. Laske-Aldershof, T., Schut, F.T., 2003. Effects of switching costs and consumer information on health insurer choice in the Netherlands. Paper Presented at the Third iHEA-Conference in San Francisco, Erasmus University Rotterdam. Lieverdink, H., 2001. The marginal success of regulated competition policy in the Netherlands. Social Science and Medicine 52, 1183–1194. Moriya, A.S., Vogt, W.B., Gaynor, M., 2010. Hospital prices and market structure in the hospital and insurance industries. Health Economics Policy and Law 5 (4), 459–479. Nudelman, P.M., Andrews, L.M., 1996. The “value added” of not-for-profit health plans. New England Journal Medicine 334 (16), 1057–1059.

449

Schneider, E.C., Zaslavsky, A.M., Epstein, A.M., 2005. Quality of care in for-profit and not-for-profit health plans enrolling Medicare beneficiaries. The American Journal of Medicine 118 (12), 1392–1400. Schut, F.T., Greß, S., Wasem, J., 2003. Consumer price sensitivity and social health insurer choice in Germany and the Netherlands. International Journal of Health Care Finance and Economics 3, 117–138. Schut, F.T., Hassink, W.H.J., 2002. Managed competition and consumer price sensitivity in social health insurance. Journal of Health Economics 21 (6.), 1009–1029. Silverman, E., Skinner, J., 2004. Medicare upcoding and hospital ownership. Journal of Health Economics 23 (2), 369–389. Van de Ven, W.P.M.M., van Vliet, R.C.J.A., Lamers, L.M., 2004. Health-adjusted premium subsidies in the Netherlands. Health Affairs 23 (3), 45–55. Van de Ven, W.P.M.M., Schut, F.T., 2008. Universal mandatory health insurance in the Netherlands: a model for the United States? Health Affairs 27 (3), 771–781. Van Dijk, M., Pomp, M., Douven, R., Laske-Aldershof, T., Schut, E., de Boer, W., de Boo, A., 2008. Consumer price sensitivity in Dutch health insurance. International Journal of Health care Finance and Economics 8, 225–244.