WORKING PAPER N The influence of supplementary health insurance on switching. behaviour: evidence on Swiss data

Transcription

1 WORKING PAPER N The nfluence of supplementary health nsurance on swtchng behavour: evdence on Swss data Brgtte Dormont Perre-Yves Geoffard Karne Lamraud JEL Codes: D12, I18, G22 Keywords: competton n health nsurance, swtchng behavour, premum convergence, nfluence of supplementary nsurance PARIS-JOURDAN SCIENCES ECONOMIQUES LABORATOIRE D ECONOMIE APPLIQUÉE - INRA 48, BD JOURDAN E.N.S PARIS TÉL. : 33(0) FAX : 33 (0) CENTRE NATIONAL DE LA RECHERCHE SCIENTIFIQUE ÉCOLE DES HAUTES ÉTUDES EN SCIENCES SOCIALES ÉCOLE NATIONALE DES PONTS ET CHAUSSÉES ÉCOLE NORMALE SUPÉRIEURE

2 The nfluence of supplementary health nsurance on swtchng behavour: evdence on Swss data Brgtte Dormont 12, Perre-Yves Geoffard 13, Karne Lamraud 14 October 2007 Correspondng author: Karne Lamraud, Extranef-Dorgny, Unversty of Lausanne, 1015 Lausanne, Swtzerland. Tel: Emal: We gratefully acknowledge the fnancal support of the Swss Natonal Foundaton for Scentfc Research (FNRS). 1 Insttute of Health Economcs and Management (IEMS), Unversty of Lausanne 2 Unversté Pars-Dauphne 3 Pars School of Economcs (PSE) 4 DEEP HEC, Unversty of Lausanne 1

3 The nfluence of supplementary health nsurance on swtchng behavour: evdence on Swss data Abstract Ths paper focuses on the swtchng behavour of sckness fund enrolees n the Swss health nsurance system. Even though the new Federal Law on Socal Health Insurance (LAMal) was mplemented n 1996 to promote competton among health nsurers n basc nsurance, there stll remans large premum varatons wthn cantons. Ths ndcates that competton has not been able so far to lead to a sngle prce, and reveals some nerta among consumers who seem reluctant to swtch to less expensve funds. We nvestgate one possble barrer to swtchng behavour, namely the nfluence of the supplementary nsurance. Our am s to analyse two decsons (swtchng decson n basc nsurance, subscrpton to supplementary nsurance contracts). We use survey data on health plan choce and mport some market data related to the sckness funds (number of enrollees, premums). The decson to swtch and the decson to subscrbe to a supplementary contract are estmated both separately and jontly. The results suggest that holdng a supplementary nsurance contract substantally decreases the propensty to swtch. However the mpact of supplementary nsurance s not sgnfcant when the ndvdual assesses hs/her health as "very good"; to the contrary, holdng a supplementary contract sgnfcantly reduces the propensty to swtch when the ndvual's subjectve health status deterorates. Futhermore, the swtchng decson s postvely nfluenced by the expected gan of swtchng. In comparson wth the range of the premum dfference, the lmtatons to swtch due to the supplementary nsurance s moderate, though non neglgble. As for the decson to subscrbe a supplementary contract, the results show that the ncome level has a drect postve nfluence on the propensty to buy a supplementary nsurance. Our results suggest that a major mechansm s gong on n relaton to supplementary nsurance: holdng a supplementary contract mght stop ndvduals from swtchng when the ndvdual thnks that she/he could be regarded as a bad rsk due to the selecton practces that are allowed n supplementary nsurance markets. Ths result bears major polcy mplcatons concernng the regulaton of basc and supplementary nsurance markets. Keywords: competton n health nsurance, swtchng behavour, premum convergence, nfluence of supplementary nsurance 2

4 1. Introducton Accordng to ts advocates, competton n health nsurance markets should promote effcency n nsurance and care delvery. However, t may have some potental drawbacks, lnked to the ncentves t provdes for rsk selecton. Managed competton settngs have been mplemented to deal wth these dffcultes: homogenous contracts are defned to avod competton on coverage, health funds are not allowed to turn down an enrolee and a rsk-adjustment scheme s ntroduced to elmnate ncentves for rsk selecton. Nevertheless, the effectveness of competton between sckness funds s a prerequste before any assessment about ts postve or potental harmful effects. Charactersng consumer behavour n health nsurance choce s a crucal ssue to examne whether competton s effectve. Ths paper focuses on the swtchng behavour of sckness funds enrolees n the Swss health nsurance system. Actually, Swtzerland s a wonderful case for studyng competton n health nsurance markets: n 1996, the Federal Law on Socal Health Insurance (LAMal) was adopted, whch amed at ntroducng a perfect managed competton scheme. In Swtzerland as opposed to all other countres, health nsurance cannot be provded by the employer as a frnge beneft: the premum s fully pad by the enrolee, whch should make her more reactve to dfferences n premums. The reform ntroduced by the LAMal was desgned n lne wth the managed competton scheme. The basc health nsurance was defned as a homogenous product,.e. full coverage n basc health nsurance, wth competton n prce only. In theory, premum dfferences for the basc package should then dsappear. However, emprcal results on Swss data do not support ths predcton (Beck et al., 2003 ; Colombo, 2001): snce 1996, the premum varablty has been qute large and has decreased only slghtly. Ths dsappontng result suggests several nterpretatons. It may reveal: () dfferences n servce qualty, () nerta of consumers resultng from swtchng costs, () rsk selecton practces by the nsurers, (v) the nfluence of supplementary nsurance. We focus on the fourth nterpretaton (v). Despte the fact that t s forbdden to sell basc and supplementary nsurance as a jont contract, nsurers behavour relatve to the supply of supplementary nsurance contracts may nduce some perverse effects on the basc nsurance market. Our purpose s to evaluate the nfluence of the supplementary nsurance on the choce for basc nsurance plan n Swtzerland. A survey carred out by the Federal Offce of Socal Insurance (OFAS, 2001) show that 75 % of nsurees have subscrbed to at least one supplementary nsurance contract. These contracts are usually subscrbed wth the same nsurer as for the basc contract. Ths paper studes the decsons to swtch and to subscrbe to a supplementary nsurance contract. It s structured as follows. Secton 2 provdes a lterature revew on health plan choce. Secton 3 offers some explanaton concernng how supplementary nsurance mght affect the decson to swtch. Data are descrbed n Secton 4. The related descrptve statstcs on the performance of the market and on consumer choces are provded n 3

5 secton 5. The econometrc specfcaton and the results are presented n secton 6. Concludng remarks and a dscusson of mplcatons are ncluded n the fnal secton of the paper. 2. Health nsurance choce: present state of knowledge Outsde the US, many ndustralsed countres have a socal health nsurance where ctzens cannot choose between dfferent payers for the basc coverage. Only recently health plan choce has been ntroduced n the socal health nsurance systems of Germany and the Netherlands, whereas ths has been a tradtonal feature of the Swss health nsurance system. Israel and Belgum also have a socal health nsurance system wth health plan choce. The emprcal lterature relatve to the swtchng behavour of consumers leads to very dfferent results, dependng on natonal regulatory systems and nsttutonal context. Premum elastctes of plan choce range from 8.4 to 0.1 (Schut et al., 2003; Laske-Aldershof and Schut, 2002; Buchmueller, 2000 ; Buchmueller and Feldsten, 1997 ; Royalty and Solomon, 1999). A result common to many papers s the hgher propensty to swtch of young, healthy and hghly educated ndvduals. As noted by Strombom et al. (2002), ths fndng means that plans that ncrease ther prce relatve to compettors wll not only lose market share, but wll experence an ncrease n costs due to adverse retenton. As concerns dfferences n swtchng elastctes between countres, Schut et al. (2003) show that the propensty to swtch s much larger n Germany than n the Netherlands. There are several possble explanatons for ths observed dfference n consumer swtchng behavour 1. Frst, dfferences n premum pad by consumers are much larger n Germany than n the Netherlands. The relatve small prce dfference among Dutch sckness funds may not be suffcent to compensate the transacton costs nvolved n swtchng from one sckness fund to another. Second, ncentves for employers are dfferent n Germany and n the Netherlands: n the Netherlands, employers contrbute a unform percentage of employees ncome, ndependent of ther choce of sckness fund. By contrast, n Germany, the level of employer contrbutons strongly depends on ther employees choce of sckness fund. Hence, employers seem to play an mportant role n facltatng consumer choce and motvatng ther employees to swtch to a cheaper (company-based) sckness fund. More recently, Tamm et al. (2007) showed that short-run prce elastctes n Germany are smaller than prevously found by other studes. In the long run, however, ther estmates gve evdence of substantal prce effects. Nuscheler and Knaus (2005) analyze the determnants of swtchng behavor n Germany from 1995 to They fnd evdence that transtons are a result of swtchng costs. Selecton by funds seems to be a neglgble problem. Schut et al. (2003) and Laske-Aldershof and Schut (2002) explan the nerta of Dutch consumers by the magntude of swtchng cost n comparson to the relatvely small dfferences between premums,.e. expected 1 Note that ths analyss s based on the Dutch system before the new unversal health system was mplemented n January

6 gan of swtchng. However, ths result can be nterpreted qute dfferently and has to be examned more thoroughly. As stated by Laske-Aldershof and Schut (2002), despte the fact that t s forbdden to sell basc and supplementary nsurance as a conjunct product, there s a dscrepancy between the rules and the practce: actually, compulsory and supplementary nsurance are ted n the Dutch system. Therefore, one has to take the ssue of supplementary nsurance combned wth compulsory nsurance nto account: ths creates heterogenety between nsurance plans n terms of premums and coverage. A stable equlbrum can be reached, wth no need for swtchng, whether or not there are swtchng costs. Internatonal comparsons of fve countres n whch health plan choce s possble (Laske-Alderhof et al., 2004) ndcate that swtchng rates are larger n Germany and Swtzerland than n Belgum, the Netherlands, or Israel. The authors attrbute these hgher swtchng rates to three man features. Frst, basc nsurance contracts have optons (n Swtzerland, the deductble level may be chosen). Second, the potental benefts (monetary gans) from swtchng are larger n Germany or Swtzerland than n the three other countres. Thrd, the nsurance market seems more compettve n Germany and Swtzerland: the level of market concentraton s relatvely low, and there are no barrers to entry. One mportant ssue n managed competton settng s the possblty, offered to nsurance companes to selectvely contract wth health care provders. Then, choosng one health plan gves the consumer an access to a specfc provder network, wth a gven qualty of care. Beyond the role of prce elastctes, many papers tred recently to evaluate the mpact of qualty of care on swtchng behavor. Usng US data, Beauleu (2002) fnds that qualty nformaton has a small, but sgnfcant effect on consumer plan choces. Ths result s not supported by Abraham et al (2006), who show that nformaton about hgher qualty alternatves s not connected wth the swtchng probablty. Conversely, patents place a hgh value on ther relatonshp wth ther personal physcan whch makes swtchng more costly. In Swtzerland, swtchng behavor s lkely to be weakly nfluenced by the qualty of care. Indeed, less than 10% of enrolees have chosen an HMO opton. The others have unlmted access to all care provders. One report done by OFAS n 2001 and two studes are avalable on health nsurance choce n Swtzerland (Colombo, 2001 ; Beck et al., 2003). The OFAS report stresses that only a mnorty of households have swtched from ther basc health sckness fund snce the ntroducton of LAMal. However the results show that there s a swtch potental because 50% of households complan about the fnancal burden of basc nsurance. Both Colombo (2001) and Beck et al. (2003) underscore the lack of convergence of premums across sckness funds. However, they delver rather dfferent assessments of the functonng of the Swss health nsurance system. On the one hand, Colombo (2001) puts the stress on consumer nerta: annual swtchng percentages are very low and seem to decrease steadly from 5.4% n 1998 to 2.1% n Accordng to her, ths nerta s attrbutable to swtchng costs lnked to the costs of collectng nformaton about relatve performance of sckness funds: Swtchng s n fact a tme consumng exercse that nvolves transacton costs on the sde of ndvduals. Informaton on sckness funds performance s currently nadequate and not easly comparable. (Colombo, 2001). Analysng the responses to the OFAS survey, she shows that ndvduals are not always well nformed. 5

7 Some consumers may thnk that sckness funds can rsk adjust and cream skm, because they do not know that the health nsurance reform ntroduced a clear-cut regulatory separaton between basc compulsory nsurance (regulated by the LAMal) and the supplementary voluntary health nsurance (regulated by the Insurance Contract Law (LCA)). In addton, some consumers thnk that they cannot splt supplementary from basc nsurance and purchase them from two dfferent funds. Colombo (2001) concludes that the goals of the LAMal are not fulflled: the lack of senstvty of swtchng behavour to sckness fund performance leads to an unhealthy competton. It does not create cost-contanment ncentves but ncentves to rsk selecton. She suggests to mprove the ndvdual choce mechansm by provdng a better access to nformaton and ntroducng a clear separaton of LAMal coverage from supplementary nsurance. On the other hand, Beck et al. (2003) adopt an econometrc approach whch leads them to an opposte assessment of the Swss health nsurance market. On the bass of models close to the specfcatons used by Schut et al. (2003) (wth addtonal fxed effects for takng heterogenety between cantons nto account), they estmate premum elastctes whch range between 2.1 and 1.0. These results are obtaned on cantonal data whch were provded by OFAS or on mcro data from a sckness fund. The magntude of ther estmates s far from neglgble: they range between the elastctes reported for the Netherlands by Schut et al. (2003) and the elastctes reported for Germany. The problem of Beck et al. (2003) s then to explan hgh and permanent premum dfferentals, together wth a non neglgble moblty of consumers between sckness funds, charactersed by hgh elastctes. They use an nterpretaton n terms of conglomerate (cartel): a subgroup of some sckness funds may work together and move ther members from one nsurer to the other wthn the conglomerate, n order to create a homogenous rsk profle wthn each fund. In ths case, the flat rate premum of each fund of the conglomerate can evolve toward a rsk-rated premum. Beck et al. (2003) gve an emprcal support to ths nterpretaton: whereas the overall premum elastcty estmated on data relatve to all cantons and funds s qute large, the premum elastcty estmated on the sub-sample of nsurers suspected to apply ths strategy of conglomerate s small and not sgnfcant. As opposed to Colombo, Beck et al. (2003) consder that swtchng costs are neglgble: the basc nsurance coverage s dentcal from one health nsurer to the other, and both basc and supplementary premum nformaton s freely and easly avalable on the web or from the federal offce, whch keeps the cost and amount of effort needed to retreve the relevant nformaton low. In addton, the actual swtchng procedure s qute smple: t s entrely suffcent to wrte a letter to one s health nsurer. (Beck et al., 2003). Fnally, Beck et al. (2003) mantan that the present Swss health nsurance system gves nsurers a strong ncentve for rsk-based selecton. Ths s due to the rule of flat rates for premums, together wth an neffcent rsk compensaton mechansm. The latter beng based on a too smple formula, leads n fact to an unfar compensaton for all hgh-rsk profles. Accordng to Beck et al. (2003), the legslator should authorse rsk rated premums or mprove the rsk adjustment formula. The representatons of the Swss health nsurance system gven by Colombo (2001) and Beck et al. (2003) are very contrasted. Both notce the lack of premum convergence and healthy competton mechansms. But Colombo underlnes the consumer nerta due to swtchng costs, whereas Beck et al. (2003) gve evdence of 6

8 consumer moblty and put the stress on rsk selecton practces. Both suggest that supplementary nsurance may have some nfluence on swtchng behavour but do not make a thorough analyss of ths ssue. The common vew s that n Swtzerland supplementary nsurance contracts only cover some rare events. However, the proporton of health expendtures fnanced by supplementary nsurance s not neglgble. The supplementary nsurance accounts for 30.9 % of health nsurance fnancng (mandatory basc nsurance covers 66.7 %). Ths fnancng s manly operated by sckness funds (23 %) followed by prvate nsurers (7.9 %). The OFAS survey shows that only 7% of Swss resdents who have subscrbed to a supplementary nsurance contract dd so wth an nsurer dfferent from the sckness fund provdng ther basc coverage. As regards supply behavour, sckness funds can rsk adjust ndrectly f they te the condtons of a supplementary health nsurance contract to the possesson of a basc health nsurance contract at the same fund. Colombo (2001) reports that many people complaned that rembursement delays deterorated after they separated basc and supplementary health nsurance n two dfferent sckness funds. Such separaton s n addton very mpractcal because doctors and hosptals do not separate blls for servces ncluded under the two dfferent covers. Fnally, premums for supplementary health nsurance may be more expensve f people are not nsured by the same fund for basc health nsurance. These practces suggest sckness funds try to te basc and supplementary health nsurance. Paolucc et al. (2007) examne the role of supplementary health nsurance as a potental tool for rsk-selecton n fve countres. Ther approach s manly nsttutonal. Comparng the regulatory settngs concernng health nsurance markets, they conclude that supplementary health nsurance may be a selecton tool n Swtzerland. In addton, ncentves to use t for rsk selecton are ncreasng n Netherlands. Ths pont s analysed from a normatve perspectve by Kfmann (2006), who studes the relatve socal costs and benefts of allowng the same nsurance funds to be actve on both markets (for basc and supplementary nsurance). Under the assumpton that subscrpton to a supplementary nsurance contract s exogenous and random, Kfmann (2006) shows that nsurers cream-skm on the basc nsurance market by sellng supplementary nsurance contracts below margnal cost. However, t s more effcent to separate the two markets f ntegratng both actvtes leads to large savngs (admnstratve costs). These features are lkely to exert a great nfluence on the choce of sckness fund for basc nsurance, swtchng behavour and more generally the health nsurance market. Our purpose s to evaluate the nfluence of supplementary health nsurance on swtchng behavour n Swtzerland. In partcular, we examne the relatve mportance of two man elements n the swtchng decson: premums for basc nsurance contracts, and supplementary nsurance. 3. How supplementary nsurance may affect the decson to swtch Certan features of the Swss health nsurance market should be kept n mnd when analysng the nteracton between basc and supplementary nsurance. Frst, nformaton about dfferences n premums for basc nsurance 7

9 s perfect, wth publcatons n all newspapers, and webstes comparng premums: ths rules out any explanaton n terms of swtchng costs due to dffcultes to gather nformaton about prce dfferences (remnd that contracts for basc nsurance are standardsed). Second, subscrbng basc and supplementary contracts wth two dfferent nsurers may nduce some admnstratve costs for the subscrber, such as sendng separate blls, etc. Thrd, supplementary nsurance s regulated by the Insurance Contract Law, whch allows selecton and does not mpose any constrant on the suppled coverage. Therefore, swtchng can be dffcult for supplementary contracts: the new nsurer may offer a dfferent contract or requre some medcal examnaton before acceptng a new customer. Gven ths framework, supplementary nsurance may affect the decson to swtch n three ways. (a) The frst mechansm s very smple and could be nterpreted as a pure swtchng cost effect. Gven that subscrbng basc and supplementary contracts wth two dfferent nsurers nduces admnstratve costs, subscrbers plannng to swtch to a new fund may have to consder movng both the basc and supplementary contracts: ths s more burdensome than a sngle swtch. (b) The second mechansm reles on selecton practces that are allowed n supplementary nsurance markets. If enrolees beleve that nsurers reject applcatons from ndvduals consdered as bad rsks, holdng a supplementary nsurance would act as a brake upon swtchng only when the subscrber thnks he or she could be consdered as a bad rsk. In ths case, fndng that supplementary nsurance has no effect on swtchng when the subscrber self-assesses hs or her health at the hghest level (very good health) would provde an emprcal support to such an nterpretaton, assumng that poor self-assessed health s assocated wth hgher expected expendtures covered by supplementary nsurance contracts. (c) The thrd mechansm refers to potental selecton practces by nsurers n basc health nsurance markets. They would retan those who hold supplementary contracts and push away the others. Two assumptons mght underle a selecton behavour based on supplementary nsurance. Suppose that holdng a supplementary nsurance contract reveals that the nsured s a good rsk for the basc nsurance,.e. that she/he has a lower basc health care consumpton for a gven llness. Ths assumpton s certanly relevant as concerns supplementary nsurance coverng alternatve medcne. It may be true for other knd of supplementary contracts, revealng a greater attenton to health and preventon. If ths assumpton s true, then fndng that ndvduals wth supplementary nsurance are less lkely to swtch would reveal that sckness funds try to, and succeed n retanng good rsks 2. Another very smple explanaton s based on the fact that the regulaton for supplementary nsurance s less constranng, and competton on ths market may be less aggressve. If sellng supplementary nsurance contracts s overall proftable, then proft-maxmsng nsurers would have an ncentve to retan supplementary contracts purchasers. As sad above, dfferent arguments ndcate that the proft per contract could be larger or smaller than average for ndvduals n self-assessed good health. However, f supplementary contracts are 2 In Swtzerland, the current rsk adjustment scheme s based on age and gender only. There reman strong ncentves to rsk selecton. 8

10 proftable for any level of health, then the effect of supplementary contracts on swtchng rates should be ndependent of the ndvdual self-assessed health. 4. Data We use two sources of data, the OFAS survey (2001) and nformaton extracted from the cantonal data base that we have constructed. The OFAS survey To study the swtchng behavour of the Swss health enrolees, we used the data collected by OFAS (2001) as part of a general assessment of the Law on Sckness Insurance (Art. 32 OAMal). As prevously mentoned a detaled descrptve analyss of ths dataset s avalable n the OFAS report (2001) and s also provded by Colombo (2002). However, as far as we know, no further analyss was performed. Ths database was obtaned from the Swss Informaton and Data Archve Servce (SIDOS). It dsplays nformaton on 2,152 ndvduals who were surveyed by telephone durng the early summer Selectng ndvduals older than 26, and allowng for mssng nformaton relatve to ncome leads to a sample of 1, 943 ndvduals 3. Extensve nformaton s collected at the mcro level (ndvdual or household, dependng on the queston) concernng health plan choce. People were requested to name ther nsurance funds for the basc and supplementary health nsurance package separately and to defne ther crtera for the choce of nsurers. The optons they had opted for (deductbles, HMO) and the composton of the supplementary beneft package were also reported. Respondents were asked whether they had moved from one sckness fund to another one durng the four prevous years ( ) and whether they had changed any of ther health nsurance contracts. Intenton to swtch n the future as well as general satsfacton towards nsurance coverage was collected. Knowledge, belefs and atttudes towards LAMal were nvestgated. The nformaton provded by the OFAS survey s somewhat lmted. It does not provde ndvdual nformaton about health expendtures, nor about the premums pad for supplementary health nsurance contracts. Moreover, t s a cross-secton: t wll not be possble for us to perform an accurate treatment of the unobserved heterogenety. The cantonal database We have constructed a second dataset ncludng nformaton relatve to each nsurer company for the 26 Swss cantons over the years 1996 to Insurer-related nformaton ncludes the number of nsured people per sckness fund wthn each canton, and the premums (per person and month) for each fund (for each premum regon) durng the years 1996 to The number of nsured people n each plan was drectly provded by the Federal Offce for Publc Health (OFSP) upon request; yearly premums are avalable on the OFSP webste n the yearly reports enttled Statstcs n 3 Accordng to the LAMal, premums are set dfferently for people aged

11 Health Insurance. Fnally, the database s made of 12,423 observatons (one lne per nsurance company, per canton and per year). Ths second dataset wll be referred to as cantonal database. Ths dataset wll be used to defne the gans assocated wth swtchng by makng the connecton between the nformaton about the sckness fund chosen n the OFAS survey and the premums of all compettors avalable n the cantonal database. Notce that premums cannot be regarded as exogenous at ths macroeconomc level. It s more approprate to use ndvdual data to study the demand for nsurance and swtchng behavour. 5. Descrptve analyss 5.1. Dd competton nduce some decrease n premum varablty? In prncple, competton n prce wth homogenous contracts should nduce a reducton n premum dfferences. The emprcal results obtaned from the cantonal database do not support ths predcton. In the cantonal database, nformaton relatve to premums s recorded at three levels: sckness fund, cantons and year. Each year, the double dmenson of the data allows us to dentfy, n the changes over tme of the premum varablty, what n due to changes n the between-canton varablty and what can be attrbuted to changes n the wthn-canton varablty. Denotng by ( p j, c, t p j c, t where ), the premum pad for the basc nsurance n sckness fund j n canton c n year t, one has: V ( p j, c, t., c, t j, c, t., c, t ) = V ( p ) + V ( p p ), V s the overall varance of premums n year t and p ) V s the between-canton varance of (., c, t V p j p s the (, c, t., c, t premums (t s equal for each year to the varance of the average premum per canton). ) wthn-canton varance of premums. Gven that competton takes place wthn each canton, any assessment on competton effectveness should be based on ths second term only. Graph 1 dsplays the annual values of the total, between-canton and wthn-canton standard devatons, computed Log ( p j, c, t for ). We used the log transformaton for changes n premum varablty not to be affected by premum ncrease over tme. We computed weghted ndcators to take nto account the number of enrolees per sckness funds. The value of the overall standard devaton s decreasng very slghtly over tme, from 22 % n 1996 to 18 % n Ths small decrease s manly due to a decrease n the between-canton standard devaton. As stated above, the competton takes place wthn the framework of each canton: the wthn-canton standard devaton only can be affected by competton pressure. Graph 1 shows how stable t s over tme, suggestng there s no premum convergence wthn cantons. Graph 2 shows that the proporton of overall varablty due to average dfferences n premums between cantons s szeable: more than 80% s due to average dfferences between cantons. Adjustng premums for dfferences n the gender and age composton of the enrolees does not change substantally ths result. The hgh porporton of between cantons varablty may derve from prcng strategy on the supply sde. It questons the relevance of lmtng competton to the canton level. 10

12 Graph 3 dsplays the annual values of the rato of the nnth to frst decle of the premum dstrbuton. The rato s computed for each canton, then an average s calculated, weghtng the cantonal ratos by the number of enrolees n each canton. The same computaton s mplemented for the rato of the thrd to frst quartle of the premum. The fgures of Graph 3 are more easy to nterpret than the standard devatons computed above: there s a dfference of about 20% between the nnth and the frst decle of the premum, and a dfference of around 10% between the thrd and the frst quartle. The decrease over tme of Q3/Q1 s very tny. It s slghtly more marked for D9/D1. Graph 4 makes t possble to examne more closely ths small reducton. Followng the same procedure as above, we computed the annual values of D9/D5 and D5/D1, where D5 stands for the medan. Graph 4 shows that, at the begnnng of the perod, the dsperson s at the same level at the top and at the bottom of the dstrbuton, and then falls at the top of the dstrbuton (D9/D5), suggestng there s some competton pressure on the hghest premums. However, the observed changes are very small. On the whole, there s no szeable decrease n premum varablty; the competton does not seem to be effectve. 5.2 The ndvdual level: basc features of the data The OFAS survey dsplays nformaton at the mcro level. Basc features of ths nformaton are provded n Table 1. Durng perod , 14.3% of ndvduals swtched from one sckness fund to another. In the year 2000, 9.5 % were consderng to swtch. A supplementary nsurance contract of any knd was subscrbed to by 74.8% of enrolees. In Swtzerland, an ndvdual may subscrbe to several potental contracts for supplementary coverage: dental care, frst and second class treatments n hosptals, cross-border care, alternatve medcne, sck-leave payments, etc. There s also a very smple contract, the Dvson commune Susse entère, whch extends the basc coverage to any Swss hosptal, not only those n the home Canton. Ths contract s offered by most nsurers at low cost and wthout any medcal examnaton. When excludng the supplementary nsurance relatve to Dvson commune Susse entère, the proporton of enrolees covered by a supplementary nsurance falls to 64.6%. Table 2 shows that the subscrpton to a supplementary nsurance s not sgnfcantly lnked to gender. Conversely, age s sgnfcantly connected to supplementary coverage: people aged are more lkely to subscrbe to a supplementary nsurance contract (except for Dvson commune Susse entère ). The survey records the household s ncome as a categorcal varable wth 11 categores. We aggregated ths nformaton nto three categores: ncome lower than 5,000 CHF (.e. 3,300 ), ncome between 5,000 and 8,000 CHF, ncome hgher than 8,000 CHF (.e. 5,280 ). The threshold for the lowest category mght appear to be rather hgh to the basc European ctzen. Actually, t s representatve of the Swss ncome dstrbuton. Informaton about household s ncome was mssng for 367 ndvduals. We checked that the recordng of ncome s not sgnfcantly connected wth the probablty to swtch (sgnfcance level p = 151). To avod losng too many observatons, we mplemented an ordered probt estmaton to predct the mssng values of ncome. Ths was possble for 264 observatons, usng the followng explanatory varables: age, gender, employment status, educaton level, famly sze, locaton, health status and cantonal fxed effects. Fgures n Table 1 show that more than one thrd of ndvduals belong to the lowest ncome category, the hghest category coverng less than one fourth of ndvduals. Wthn one nsurance compagny, n a gven canton, premums are communty rated. However, a state subsdy set at the canton level help people wth a low level of 11

13 resources to pay ther premum. The generosty of the coverage, and the elgblty crtera depend on the canton: on the whole, 23 % of the respondents beneft from such a subsdy. Table 3 shows that havng subscrbed to a supplementary nsurance s negatvely connected wth the propensty to swtch: the probablty of swtchng s one thrd lower (odds rato sgnfcant and equal to 72 %) when the ndvdual has a supplementary nsurance. Conversely, the ntenton to swtch seems to be ndependent of supplementary nsurance (odds rato equal to 93 %, wth a sgnfcance level p=0.66). Other statstcs (not reported here, avalable on request) show that swtchng propensty depends on age group and supplementary nsurance status. Young people exhbt a hgh swtchng propensty, whatever ther supplementary nsurance status. Ths s not the case for older people: ther swtchng propensty decreases wth age and s reduced by a supplementary coverage. Health status s measured by the ndvdual s self assessed health, whch s recorded on a fve pont scale that we collapsed nto three categores: Bad, Good and Very good. Table 1 shows that about 17 % of ndvduals grade ther health as Bad and 35 % as Very good. The level of ncome s strongly related to self-assessed health, the rcher beng n better health (Table 3bs). The magntude of the nfluence s amazng: the probablty to be n bad health decreases by two-thrd for people earnng an ncome n the ntermedate category. For people belongng to the hgh ncome category, the probablty to be n bad health s dvded by fve, n comparson wth low-ncome ndvduals. People havng subscrbed to a supplementary nsurance are n a rather good health (odds rato equals to 0.72, see Table 3bs). But ths mpact s entrely due to the nfluence of ncome on the supplementary nsurance: when ncorporatng both supplementary nsurance and ncome n the logstc model, one fnds that the mpact of supplementary nsurance on the probablty to be n bad health s no longer sgnfcant. 6. Estmaton and results 6.1. Econometrc specfcaton Swtchng behavour We estmate a model explanng the swtchng decson n the past. Consder the bnary varable y defned by y = 1 f the ndvdual swtched and y = 0 when he/she dd not swtch. We model the beneft of swtchng as a latent (unobserved) varable where * y defned by: * y = x ' β + ( dp) η + s γ + u, (1), x ' s a vector of ndvdual characterstcs and s s a dummy varable ndcatng whether the ndvdual had subscrbed to supplementary nsurance. dstrbuton. The decson to swtch s gven by: u s a dsturbance supposed to follow a normal or a logstc y = 1 f y * 0. (2) 12

14 The varable dp s defned to capture the monetary gan of swtchng n terms of cost to be nsured for the basc nsurance. To construct ths varable, we consder the premums observed for each deductble level n the cantonal database and adjust these premums for nflaton n the basc health nsurance sector. We observe the level of the premum of the sckness fund chosen. In a gven year, f ndvdual comes from sckness fund k and swtches to fund j, the monetary gan of swtchng s equal to p p. We constructed a varable that evaluates potental swtchng gans for swtchers and non-swtchers. One dffculty s that we do not observe the fund the swtchers come from. We thus decded to compute the expected gan of swtchng. Ths s defned as follows: n (dp) = E[ p k p j k j]= (p k p j )( k n ) (3) k j k k j k j where n s the number of enrolees of sckness fund k. Ths defnton of the expected gan of swtchng s based k on the assumpton that the probablty of to come from sckness fund k s equal to the proporton observed for all ndvduals of the sample n the year precedng the swtch. For non-swtchers, the varable dp represents the potental swtchng gans, f the fund s chosen at random. It undervalues potental swtchng gans for non-swtchers, gven that a fund s actually not chosen randomly, but because t s the least expensve one. Decson to subscrbe to a supplementary nsurance contract We also estmate a model explanng the decson to subscrbe to at least one supplementary nsurance contract (except Dvson commune Susse entère ). The beneft of subscrbng s specfed as a latent varable defned by: where * s = z ' α + v, (4) z ' s a vector of ndvdual characterstcs. The decson to subscrbe s gven by: The premum dfferental s = 1 f s * 0. (5) ( dp) for basc nsurance s n prncple not lnked to the decson to subscrbe. We ncluded t nto specfcaton (4) to check that t s uncorrelated wth the choce to subscrbe a supplementary nsurance. Actually, we found that ( dp) was not sgnfcant. Therefore, ( dp) and s act as orthogonal determnant of the swtchng decson n equaton (1). * s A bvarate model of decsons to swtch and to subscrbe to a supplementary nsurance The bvarate model conssts n estmatng a smultaneous equaton model of jont decson to swtch and to subscrbe to a supplementary nsurance contract. A separate estmaton of equatons (1) and (4) would lead to asymptotcally based estmates f ther dsturbances ( u and v ) are correlated. The dsturbances u and v may be correlated for several reasons. Both are nfluenced by unobserved heterogenety relatve to ndvdual, n other words, by unobserved varables whch nfluence the decsons we 13

15 try to explan. It may be for nstance rsk averson or health status. Another source of bas s the fact that the premum of the supplementary nsurance s unobserved. Ths varable could be an explanatory varable of the decson to subscrbe to supplementary nsurance and of the decson to swtch. In ths case, t would be a common component of dsturbances u and v. However, the regulaton n the basc nsurance market, especally the contract standardsaton, s lkely to elmnate any role of rsk averson or health status n the swtchng decson: ths equaton has nothng to do wth demand for nsurance, gven that basc nsurance s mandatory. It s possble that these components of the ndvdual unobserved heterogenety affect only v. Such a two equaton model s defned by Maddala (1983) as a bvarate recursve model. If u and v are not ndependent, a two-stage method s not approprate to estmate the decson to swtch equaton. One has to use a maxmum lkelhood estmator, where the lkelhood s bult on the bass of the jont dstrbuton of ( Conversely, f u and v are not correlated, both equatons can be estmated separately. s, y ). In what follows, we present the results of separate estmatons and of the estmaton of the bvarate probt model. The jont estmaton of the bvarate model led to a non sgnfcant correlaton coeffcent ρ between u and v, justfyng a separate estmaton. Actually, the condtons for an dentfcaton of the model are met: several nstruments explanng the decson to subscrbe a supplementary nsurance are excluded from the decson to swtch (ncome (3 dgts), gender, and educaton (5 dgts)). These varables are not sgnfcant n the swtchng equaton Results Equatons (1) and (4) were estmated wth or wthout canton fxed effects to deal wth the unobserved heterogenety between cantons. The results of both specfcatons were qualtatvely the same. Therefore we only reported the results based on the fxed-canton effect models (Table 4 and Table 6). The decson to swtch was estmated by a logt estmator. The results are dsplayed n Table 4. They show that the expected gan of swtchng (dp) nfluences postvely the swtchng decson. Havng subscrbed to supplementary nsurance has a sgnfcant negatve mpact on swtchng. The mpact of the supplementary nsurance s szeable: ceters parbus, t reduces the probablty to swtch by about 30 %. Columns three and four report estmates from a model that allows the mpact of supplementary nsurance to vary when the subscrber s n very good, good or bad health. Three nteracton terms are ncorporated. One obtans the very nterestng result that supplementary nsurance has no effect on swtchng when the enrolee self-assesses hs/her health status as "very good". However, holdng a supplementary nsurance s assocated wth a monotoncally declnng lkelhood of swtchng when the subjectve health status gets worse. These results are consstent wth ratonal expectatons from enrolees that swtchng may be more dffcult for supplementary contracts due to the selecton practces that are allowed n supplementary nsurance markets. These results thus rule out the two other potental 14

16 mechansms that were hghlghted n secton 3 to descrbe how supplementary nsurance mght affect the decson to swtch. Otherwse, we fnd that the probablty to swtch s contnuously decreasng wth age. Ths result s consstent wth the emprcal lterature on swtchng behavour we have revewed above. The youngest (27-35) have a probablty to swtch whch s more than 80 % hgher than the reference age group (35-50) whereas the oldest (>65) have a much lower probablty to swtch (- 66 % to -67 %). Havng chosen a deductble level hgher than the basc deductble s lnked wth a sgnfcantly hgher propensty to swtch (+53 to +55 %). The descrptve analyss has suggested that the mpact of the supplementary nsurance should vary wth age. To allow for ths possblty, we have estmated the same model wth cross effects of age group wth the supplementary nsurance dummy. Ths specfcaton led to a loss n precson wth many non sgnfcant coeffcents. Table 5 gves the average expected gan of swtchng (varable dp), for swtchers and non-swtchers, wth or wthout a supplementary nsurance contract. Not surprsngly, we fnd that the average swtchng gan s larger for those who have ndeed swtched (CHF 15.75) than for non-swtchers (CHF 2.21). The annual swtchng gan s therefore CHF 189 for swtchers,.e Graph 5 dsplays the correspondng dstrbutons of the annual swtchng gans for swtchers and non-swtchers. The effect of supplementary contracts s more surprsng. Snce holdng a supplementary nsurance contract decreases the lkelhood to swtch, t may be nterpreted as a swtchng cost, and we may expect that ndvduals wth a supplementary contract would swtch for larger swtchng gans than those wthout supplementary nsurance. Ths s not the case: the average swtchng gan s CHF for those wth supplementary contracts, and CHF for those wthout. Ths result suggests the exstence of heterogenety wthn swtchers and deserves further nvestgaton. Actually, people who are not holdng supplementary nsurance dffer from the others as regards the level of ther premum for the basc health nsurance: ther premums appear to be hgher, especally at the top of the dstrbuton. More precsely, the average level of ther monthly premum s CHF 6.2 hgher, the frst quartle of ther monthly premum s CHF 6.5 hgher and the thrd quartle of ther monthly premum CHF 16.3 hgher. The nterpretaton of our results could be the followng. Indvduals who are not holdng supplementary nsurance dffer strongly from the others by ther unobservable characterstcs. Snce they are characterzed by a low level of educaton (as shown by the results commented on above), ther swtchng costs may be larger than for more educated people (nformaton may be more dffcult to collect and nterpret). Ths effect could be large enough to more than compensate for the negatve effect of holdng supplementary nsurance. The fact that people wthout a supplementary nsurance pay hgher premums (for a gven level of deductble), gves support to ths nterpretaton. Table 6 dsplays the results relatve to the decson to subscrbe to a supplementary nsurance contract. They show that the decson to subscrbe to a supplementary nsurance contract s not a monotonc functon of age: the probablty of subscrbng ncreases wth age untl the class Then, t decreases for people older than 65. The ncome level has a sgnfcant postve nfluence on the propensty to buy a supplementary nsurance. The 15

17 magntudes of the effects are amazng: the probablty to subscrbe s almost twce as hgh for an ntermedate ncome ndvdual n comparson wth a low ncome ndvdual, and more than doubles (OR= 2.40) for the hghest level of ncome. Ths result s opposed to the basc nsurance theory, whch predcts a lower absolute rsk averson for rch people. Therefore, they should be less lkely to buy nsurance. Our estmates suggest that the purchase of supplementary nsurance s nfluenced, not only by rsk averson, but also by the wllngness to pay for goods covered by the supplementary nsurance. Descrptve statstcs have shown that bad health has a negatve nfluence on the subscrpton to a supplementary contract. But ths nfluence s no longer sgnfcant when ncome and educaton are ntroduced as covarates. All the nformaton about health seems to be captured by the ncome level: as stated above, a low ncome s strongly correlated wth bad health status (Table 3bs). Therefore the results exhbted n Table 6 do no nclude health status as a covarate. The estmatons of the bvarate probt model are dsplayed n the appendx (together wth the results of the sngle probt equatons). In both the sngle and bvarate equaton models, two specfcatons are tested dependng on whether the supplementary varable s ncluded as a plan covarate or as an nteracton varable wth health status. The results are smlar to those of the separate equaton models. Furthermore the correlaton of the dsturbances between both equatons s not sgnfcant. Ths allows us to report the sngle equaton results. The bvarate probt results exhbt only one dfference wth the sngle equaton results: the effect of the supplementary nsurance s no longer sgnfcant n the swtchng equaton. However t s mportant to notce that the coeffcents of the bvarate probt models have the same magntude as the coeffcents of the sngle equatons. In partcular ths s true for the supplementary nsurance varable and the nteracton terms between supplementary coverage and health status. However the precson of the estmates deterorates greatly n the bvarate estmatons. Ths suggests that the nstruments that we use to explan the subscrbton to a supplementary contract lead to an mportant loss of nformaton. Are the lmtatons to swtch nduced by the supplementary nsurance large enough to explan the lack of premum convergence? We can use the estmates of equaton (1) to have a frst nsght nto ths queston. Consder the dstrbuton functon of the logstc. Equatons (1) and (2) gve the expresson of the probablty to swtch for ndvdual : ( η) ' ( y = 1) = F x β + s γ ( dp) Pr ob + Ths probablty can be estmated by : ^ ( ˆ η) ' ( y 1) = F x ˆ β + s ˆ γ ( dp) Pr ob = + (6) where βˆ,γˆ and ηˆ stand for the estmators of β, γ and η. Ths predcted probablty vares wth the observaton consdered. We want to focus on the respectve mpacts of the premum dfferental and of the supplementary nsurance. To summarze these ndvdual probabltes, we evaluate them for the average ndvdual, except for the varable dp, whch s allowed to vary over ts range. Ths 16

18 computaton leads to a functon g(dp ) descrbng the swtchng probablty accordng to the premum dfference dp, when the ndvdual has not subscrbed to a supplementary nsurance : ^ ' ' ' ( y = 1/ x = x ; s = 0) = F( x ˆ β + ( dp ) ˆ η) g( dp ) Pr ob = The same computaton s used to buld the functon h(dp ) descrbng the swtchng probablty accordng to the premum dfference dp, when the ndvdual has subscrbed to a supplementary nsurance : ^ ' ' ' ( y = 1/ x = x ; s = 1) = F( x ˆ β + ˆ γ + ( dp ) ˆ η) h( dp ) Pr ob = Graph 6 dsplays the curves for the functons g(dp ) and h(dp ). Vertcal lnes are drawn to represent the values of the frst and nnth decles of the expected premum dfference dp (equal to the expected gan to swtch). In comparson wth the range of dp, the lmtatons to swtch due to the supplementary nsurance s not very large, but non neglgble. 7. Concluson In ths paper we hghlghted that the ntroducton of a managed competton scheme for basc health nsurance n Swtzerland faled n reducng the premum varablty: wthn canton, the premum varablty appears to be stable between 1996 and Ths fndng rases the queston of the effectveness of competton n Swtzerland for the basc health nsurance market. Our results show that the swtchng decson s postvely nfluenced by the expected gan of swtchng and that holdng a supplementary nsurance contract substantally decreases the propensty to swtch. However, holdng a supplementary nsurance has no effect on the propensty to swtch when the ndvdual selfassesses hs/her health as "very good"; holdng a supplementary contract monotoncally reduces the probablty to swtch when the ndvdual's subjectve health status gets worse. As far as the decson to subscrbe to a supplementary contract s concerned, our results suggest that the ncome plays a major role. It has a drect postve nfluence on the propensty to buy a supplementary nsurance. Ths fndng suggests that the purchase of supplementary nsurance s nfluenced, not only by rsk averson, but also by the wllngness to pay for the goods covered by the supplementary nsurance, whch would be hgher for rch people. Our estmates allow us to compute the swtchng probablty as a functon of the premum dfference, whch corresponds to gans of swtchng. Ths functon s drawn for people wth and wthout a supplementary nsurance contract. In comparson wth the range of the premum dfference, the lmtatons to swtch due to the supplementary nsurance s moderate, though non neglgble. 17

19 A better understandng of the effectveness of competton on the basc health nsurance market s of major nterest for polcy makers. Needless to say, competton puts provders under pressure only f consumers are able to swtch from the least effcent to the most effcent funds, and potental barrers to swtchng may explan the persstence of neffcences. In ths paper we nvestgated one specfc barrer to swtchng, namely the nfluence of supplementary nsurance. As explaned, the supplementary nsurance mght affect negatvely the decson to swtch n three possble ways: (a) holdng a supplementary contract mght act as a pure swtchng cost when the enrolees subscrbe basc and supplementary contracts wth the same fund (b) holdng a supplementary contract mght stop ndvduals from swtchng when the ndvdual thnks that he/she could be regarded as a bad rsk; (c) The negatve mpact of supplementary nsurance on the swtchng propensty mght reveal rsk selecton practces by nsurers to the extent that hgh proft margns are assocated wth supplementary nsurance provson and that holdng holdng supplementary nsurance may be synonymous to beng a good rsk. Our results suggest that mechansm b s a major explanatory factor of consumers' nerta n relaton to supplementary coverage. Note that mechansm b entals antcpaton of selecton practces from the consumer. Our analyss llustrates how the consumer choce for health plan nteracts wth the decson to subscrbe to a supplementary nsurance contract. If these prvate decsons are not ndependent, then the regulaton of the supplementary health nsurance market should ntegrate the effects t may have on the basc nsurance market. Even f basc nsurance and supplementary nsurance are regulated by two dfferent laws (respectvely, LAMal and LCA) and supervsed by two dfferent nsttutons (respectvely, OFSP and OFAP), both markets turn out to be closely ted as nsurance companes are allowed to operate both on basc and supplementary nsurance markets. We may wonder whether t s relevant to mplement managed competton n basc nsurance markets and allow pure competton (no regulaton) n supplementary nsurance markets. As regards the polcy agenda, our results suggest two alternatve changes: ether more effectve separaton bewteen basc and supplementary nsurance markets or more severe regulaton n the supplementary nsurance sector. Our analyss would deserve further nvestgaton. Data ncludng the prema of supplementary contracts as well as a longer perod of observaton of consumer choces and nsurance markets would be much helpful. Sofar our results provde recommendatons for better regulaton n Swss nsurance markets. 18