InsuranceSearchandSwitchingBehaviorJonnekeBolhaarMaartenLindeboomBasvanderKlaauwSeptember2009AbstractThis paperlooksintothesearchbehaviorofconsumersinthe marketforhealthinsurance contracts. We consider the recent healthinsurance reforminTheNetherlands,whereaprivate-publicmixofinsuranceprovisionwasreplacedbyasystembasedonmanagedcompetition. Althoughall insurersoerthesamebasic package (determinedbythe government), there is substantial premiumdispersion. Wedevelopaconsumersearchmodel containingthemainfeaturesof theDutchhealthinsurancesystem. Thismodel providesuswithanumberofhypotheses,whichwetestusingdatafromtheDutchHealthCareConsumerPanel. Wendthatthesimpleconsumersearchmodeldescribesthechoiceforinsurancecoveragewell, butfailsinexplainingindividual searchdecisions. Wearguethatsearchcostsareheterogeneousandrelatedtoknowledgeaboutthesystem. Inthis case, groupcontracts oeredbyinsurers, might cause third-degreepricediscriminationandreduceaccesstohealthcareforsomegroupsofindividuals.VUUniversityAmsterdam,TinbergenInstitute.Address: department of economics, VUUniversity Amsterdam, De Boelelaan 1105, 1081 HVAmsterdam,TheNetherlands.Email: jbolhaar@feweb.vu.nl,mlindeboom@feweb.vu.nl,bklaauw@feweb.vu.nlWe thank NIVELand especially Judith de Jong for kindly allowing us to use data fromtheHealthCareConsumerPanel. Wewouldalsoliketothankparticipantsinseminarsandconferencesin Bristol, York, EEA 2009 in Barcelona and the 18th Health Economics and Econometrics WorkshopinSardinia1 IntroductionCompetitivemarketsarewelfaremaximizingandthelawofonepriceshouldhold. Inmanymarketsthereis, however, asubstantial degreeof pricedispersion. Thismayeitherbebecauseproductsarenothomogenous, orbecauseconsumersfacecoststoobtaininformationaboutprices. Firmscanexploittheirmarketpowertosetpricesabove marginal costs. Consumer search models are often used to describe such markets.ThispaperfocusesontheDutchhealthinsurancemarket,andteststowhatdegreeasimpleconsumersearchmodelcandescribethebehaviorofconsumersinthismarket.In the empirical analyses, we exploit a major health insurance reform which took placeintheNetherlandsonJanuary1, 2006. Thereformforcedeveryonetoreassesstheirhealthinsurancecontract.Before the reform there was a mix of private and public insurance against the costsofhealthcare. Inthenewsystem, whichisoneofmanagedcompetition, all insurerscompete with each other within rules set by the government. The current Dutch systemhas many similarities with the Swiss health insurance system,and is an inspiration forthehealthinsurancereformsrecentlysuggestedbytheObamaadministration. Theseambitions have renewed international interest in incentives of competition within socialinsurances.TheDutchregulationsobligeeveryonetobuyabasicinsurancepackageofwhichthecontentis determinedbythegovernment. Insurancecompaniesare notallowedtorefuseapplicantsforthisbasicpackageandtodierentiatepremiumsbyanymeasureofrisk (age,health,etc.). A RiskEqualizationFund compensatesinsurerswhohaveadisproportionate number of high-risk individuals among their insurees. Insurance com-panies are free to set their own price for the basic insurance package and to compete forinsurees. A surveybythe Dutch Healthcare Authority indicated that consumersfocuson premiums in the decision process (Dutch Healthcare Authority, 2006). If individualsindeed search sucientlyfor the lowest premium, the system should provide incentivestoinsurerstoimprovetheireciencyandlowertheirpremiums. Consumersearchforhealth insurance therefore plays an essential role in this system. However, the monthlypremiumsforthebasiccoveragerangefrome82.50toe97.75. Byswitchinginsurersomepeoplecould, therefore, saveupto15%of theinsurancepremium, whichsug-gests that individuals do not have full information or that search costs are prohibitivelyhigh. Asasecondcontributionthispaperprovidesmoreinsightintoconsumersearchbehaviorinasystemofmanagedcompetition.We provide a simple consumer search model, which builds on Stahl (1989), Janssen1andMoraga-Gonzalez(2004)andJanssen, Moraga-GonzalezandWildebeest(2005).Individuals in our model are only heterogeneous in their health, which determines theirutility of insurance coverage. Each individual receives an oer for health insurance fromtheir current insurer, and, in addition, may receive an oer for a group contract. Thesegroup contractsaremostlyoeredviaemployersandgiveadiscounton thepremium.After havingreceivedthe oer(s), individuals decide whether or not tosearchthemarketforalowerpricedinsurancecontract.The model provides a number of testable predictions on insurance choice and searchbehavior. WeusedatafromtheDutchHealthCareConsumerPanelcollectedbytheNetherlandsInstituteforHealthServicesResearch(NIVEL).Participantsinthecon-sumerpanelcompletequestionnairesfrequently,and, therefore,thedataare extensiveonchoiceandsearchforinsurancecontracts. Thedataconrmthepredictionsonin-surance choice (i.e. there is adverse selection and a lower premium increases coverage).However, the data are not in agreement with predicted search behavior. We argue thatthelatterisduetoheterogenoussearchcosts, andthatindividuals withlowsearchcostsaremorelikelytoobtainanoerforagroupcontract. Thisgeneratesasitua-tionofpricediscriminationwhichcausesthatindividualswithoutanoerfor agroupcontract(andmostlikelyhighersearchcosts)payahigherpremium, andalsoobtainreduced insurance coverage. Stahl (1989) argues that reducing the number of informedconsumers(asisthecaseinthemarketfor individualswithoutgroup contracts)leadstomoredispersioninpremiums. Fromthisobservationonemayquestiontheuseful-nessof allowingforgroupcontracts. Afterall, withoutgroupcontractstherewouldbelessvariationbothinpremiumsandininsurancecoverage, whichmightequalizeaccesstohealthcarewithinthepopulation.Ourpapercontributestotheempiricalliteratureonconsumersearchmodels,andparticularlytothesmall literatureonsearchininsurancemarkets. Pauly, HerringandSong(2002)considerthechoiceforhealthinsurancesandBrownandGoolsbee(2002)focusonthemarketforlifeinsurances. BothpapersusedatafromtheUStoinvestigatetheconsequencesoftheintroductionofinternetsearch,whichshouldhaveloweredsearchcosts. Bothpapersshowthatempirical predictionsareinagreementwithconsumer searchmodels (e.g. Stahl, 1989). Sorensen(2000, 2001) considersthe retail market for prescriptiondrugs. Sorensen(2000) concludes that less thanone-thirdof thepricedispersioncanbeattributedtopharmacyheterogeneity. Allpapers use, however, the observeddistributionof prices toinfer the importanceofincompleteinformationandsearch. Ourdatacontaindirectmeasuresforindividualsearchbehavior. Furthermore, westudyawell-denedinstitutional settinginwhich2therulesandtimingofactionsarehighlyregulated.Theremainderofthepaperisasfollows: section2providesmorebackgroundanddetailsonthereformof thehealthinsurancesysteminTheNetherlands. Section3presents the search model. The data used for the empirical part are discussed in section4,andsection5givesresultsoftheempiricalanalyses. Section6concludes.2 TheDutchhealthinsurancereformInThe Netherlands, the healthinsurance systemis split intothree compartments.Therstcompartment, thecatastrophicinsurance, isapublicinsurancethatcoverstheentirepopulation. It insures individuals againthecosts of long-termcare(e.g.nursinghomes, andmental healthinstitutions). Thesecondcompartment includesinsurableriskandcarethatallindividualsshouldhaveaccessto. Thethirdandlastcompartmentissupplementarycoverage. TheDutchhealthinsurancereformin2006onlyaectedthesecondandthirdcompartment. Wewill rstbrieydiscusstheoldsystem. Next, wewill providedetailsonthenewsystemandonhowthereform, thetransitionfromtheoldtothenewsystem,wasexecuted.2.1 TheoldsystemBeforethereform, therewasamixof publicandprivateinsuranceprovisioninthesecondcompartment. Allbreadwinnersearninglessthansomeincomethresholdwerecompulsoryinsured, asweretheirdependents, undertheSicknessFundAct. In2005theincomethresholdwas e33,000foremployeesandbenetrecipientsande21,050forself-employed. ForpensionerseligibilitydependedonSickness Fundcoverageatage 65. The SicknessFunds covered about 65% of the population.1The SicknessFundActguaranteedanextensivecoverageagainstarelativelylowinsurancepremium. In2004, themonthlypremiumpaiddirectlytotheinsurerwasonlye25.2Mainsourceof fundingwere income-relatedcontributions made bythose coveredbythe publicinsuranceandtheiremployers.31Some civil servants (for example the police force) were covered by a compulsory insurance schemeirrespectiveoftheirincome. Thiswasabout5%ofthetotalpopulation.2In 2005 a no-claim was introduced to reduce moral hazard. Insurees who did not visit a specialistorhospitalorusedprescribedmedicationcouldreceiveacashbackuptoe225. Theintroductionoftheno-claimincreased insurancepremiumswithabout24%.3Thecontributionwas7.95% ofincome,ofwhich6.25%wastobepaidbytheemployer.3Those earningmore thanthe income thresholdhadtobuyhealthinsurance intheprivatemarket.4Individualswerefreetochoosetheirinsurerandtheextenttowhich they wishedto be covered. In practice,private insuranceplans were in coverageandqualityof care verysimilar toSickness Fundinsurance (withthe exceptionofoptional deductibles). However, the premium had to fully cover the costs and thereforepremiums werediversiedby, for example, ageandhealthrisks. Fora30-year oldwithout healthproblems theinsurancepremiumforcoveragesimilar tothat of thesicknessfundswasaboute230permonth. Theleft-handsideofFigure1summarizestheoldsystem,theright-handsidethenewsystem.2.2 ThenewsystemOn January 1, 2006 managed competition was introduced in the second compartment.ThedistinctionbetweenSickness Fundinsuranceandprivateinsurancedisappearedandtheformer providers of Sickness Fundinsurance weretransformedintoprivateinsurancecompanies. Withinthesecond-compartmentall insurersoerthesameba-sichealthinsurancepackageof whichthecontentisdeterminedbythegovernment.CoverageofthisbasicinsuranceislessextensivethanthecoverageundertheformerSicknessFundAct. ItiscompulsoryforallinhabitantsofTheNetherlandstoobtainbasic insurance from one of the insurers. Insurers are obliged to accept everyone and arenotallowedtodierentiatepremiums(communityrating). ARiskEqualizationFundwasintroducedtocompensateinsurersforaneventualdisproportionatepercentageofhighrisk insurees. Insurers primarilycompeteonthepriceof thebasicinsurancepackage, asthequalityofthedeliveredcarewasequal amonginsurers(theyall oeraccesstoallproviders).In2006, themarketconsistedof33insurancecompanies. Someoftheseoperatedundermorethanonelabel,sothatintotal43basicinsurancepackageswereoered.5In2006theaveragenominal premiumwasabout e1050per year.6However, therewas substantial dispersion in premiums. Figure 2 shows that monthly premiums rangefrome82.50toe97.75.74Chronicallyillwithahighincome,whowouldberefusedbyprivateinsurerswerecovered byaspecialinsurance.5Themajorityof theinsurers(22insurers)areincludedinoneof sixlargeholdings(seeVektis2007).6Childrenunderage18arecovered bytheirparentsinsuranceandtheirpremiumispaidbythegovernment.7ThereisnodierenceinpricelevelbetweenformerSicknessFundsandprivateinsurers.4Insurancecompaniesalso oer supplementaryinsurance,whichincludes,for exam-ple,dentalcare,alternativemedicine,extensionoftreatmentbyphysiotherapists,etc.Mostinsurersoerthreeorfourdierentsupplementaryplans, rangingfromlimitedadditional coveragetoveryextensivecoverage. Supplementaryinsuranceiselective,andboththepremiumandcompositionisdecidedbytheinsurer. Althoughinsurersareallowedtoselect forthesupplementaryinsurance, mostinsurers donot. Insur-ersthatdoselect, onlydothisfortheplanwiththemostextensivesupplementarycoverage. Therewasonlyoneinsurerthatdierentiatedpremiumsforsupplementaryplans byage. In2006intotal 137dierent supplementaryplans wereavailableonthemarket, withamonthlypremiumrangingfrome5toe77(seeDutchHealthcareAuthority, 2006). Ofthese137plans, 10plansrequireansweringquestionsabouttheinsurees health(i.e. theyselect onhealthrisks). Supplementarycoverage is verypopular,92.6%ofconsumersobtainedsomekindofsupplementaryinsurance.The basic insurance does not involve copayments, but the system allows individualsto choose for a deductible up to e500. The annual reduction in basic insurance premiumwas about e36 for every e100 additional deductible. However, this option was not verypopular, over95%of all individualsdidnottakeanyvoluntarydeductible. Insurersare allowed to oer group contracts, and to grant a premium reduction of at most 10%onthebasicandsupplementaryinsuranceinthesegroupcontracts. Themajorityofthegroupcontractswereoeredviaemployers, butalsoothergroups, suchaslaborunions, could negotiate group contracts for their members. If an individual received anoerforagroupcontract,thenalsothepartnerwaseligibleforthediscount. In2006About44%of all individualswasparticipatinginagroupcontractandtheaveragediscountwasabout7.5%(Vektis,2007).2.3 ThereformThe reformwas announcedlongbefore January1, 2006. Alargemediacampaignwassetuptoinformpeopleaboutthenewhealthinsurancesystem, andtoexplaintherules. InOctober 2005, 98.8%of therespondents inour dataknewabout thereform. InDecember2005everyinsurancecompanyhadtomakeanoertoall itsinsurees.8Theoerwasacombinationof thebasicinsuranceandasupplementaryinsuranceplan which wasclosestto the individualsold insuranceplan. Thisoer wasthedefaultoptionforanindividual. Individualscouldchangeinsurerorthelevel of8Most insurers already announced thepremium for thebasic insurance in October and November2005. However, some insurers lowered their premium after learning the premiums of their competitors.5supplementary coverage until May 1, 2006, but the insurance bought provided coveragein retrospect from January 1. In the year of the introduction (2006), insurers were alsoobliged to accept all their former insurees for anylevel of supplementary coverage untilMarch1.9ThisimpliedthatalmostallchangesininsurerorsupplementarycoverageoccurredbeforeMarch1.All health insurance contracts run from January 1 to December 31. Insurers have topost their premiums and conditions for the following year in December, and individualscan only change insurer during the month of January. So, the long period for switchingonlyappliedtotheyearoftheintroductionofthenewsystem.3 Aconsumersearchmodel forhealthinsuranceThissectiondiscussesasearchmodel forhealthinsurancesandderivesanumberofempiricallytestablepredictions. Themodel includesthebasiccharacteristicsof theDutchhealthinsurancemarket. Consumers receive adefault optionwithout costs,butcanalsolearnaboutotherinsuranceplansbymakingsearchcosts. Insurerspostpremiums both for basic insurance and one type of supplementary insurance and acceptallapplicants. Weexplicitlyallowforpremiumdiscountsduetogroupcontracts.3.1 ConsumerbehaviorEach consumer i is characterized by his health hi, which is in the population distributedaccordingtothedistributionfunctionG(h). Eachinsureroersthe sametwotypesofinsurances, abasicinsuranceandaninsurancewithhigher, supplementary, coverage.All consumers derive thesame (expected) utilityulfrombasic insurance coverage.Theexpectedutilityconsumersderivefromtheinsurancewithhigh(supplementary)coveragedependsontheconsumers healthuh(hi). Inparticular, individualsingoodhealth derive less expected utility from an insurance with high coverage than individualsinbadhealth,sou

h(hi) < 0.Attheintroductionof thenewhealthinsurancesystem, eachconsumerreceivedanoerfromhiscurrentinsurer. Theoerischaracterizedbyapremiump0forbasicinsurance and (1+)p0 for insurance with high coverage. We impose that each insurerincreases the premium with the same fraction for obtaining supplementary insurance.Obviously,an individualpreferstheinsurancewith highcoverageifuh(hi) ul> p0.9Itwasalsoannouncedthatinlateryearsinsurerscoulddenysupplemental insurancecoveragefornewclients.6Theleft-hand sideoftheinequalityisdecreasingin(good) health,andtheright-handsideisincreasinginthepremiump0. Thisimpliesthatindividualsaremorelikelytotakeinsurancewithhighcoverageif theyareinbadhealth(adverseselection), orifthepremiump0islow.Hypothesis1: Individualswithworsehealtharemorelikelytobuyhealthinsurancewithhighcoverage(adverseselection).Hypothesis2: Alowerpremiuminducesindividualstotakemorehealthinsurancecoverage.Eachconsumerhasaprobabilityofalsoreceivinganoerforagroupcontract.The premiums of group contracts are pg and (1+)pg, for basic insurance and insurancewith high coverage respectively. Individuals prefer the group contract if pg< p0, whichalsoimpliesthatthoseindividualswhodecidedtotakethegroupcontractaremorelikelytotakeinsurancewithhighcoverage. Let pnsdenotethelowest premiumanindividualgetsoeredwithouthavingsearchedthemarket. So,withoutanoerforagroupcontractpns= p0,andwithanoerforagroupcontractpns= min{p0, pg}.After individualshavereceivedthe oer from their currentinsurerand possiblyanoerfor agroupcontract,theycandecidetosearchthemarketfor aninsureroeringalower premium. Before searchingthe market the consumer onlyknows that thedistribution of premiums in the market equalsF(p). If the consumer decides to search,hemakescostscandwillobservethepremiumsofallNinsurersinthemarket.10Obviously, theconsumerwill switchtoanotherinsurerif anyof theotherN 1insurers in the market will oer a lower premium than the current best oer pns.11Thelowest premium pmin of the other N1 insurers in the market is the rst order-statisticofN 1drawsfromthedistributionfunctionF(p),whichhasexpectedvalueE[pmin] =

FN1(p)dp10Weassumethatwhenconsumerssearch,theyobserveallpremiumsinthemarket,becausethegovernment had launched a website where consumers could compare insurance plans between insurers.Itexplicitlyaimedatloweringsearchcosts. Independentconsumerorganizationsfollowedwiththeirownwebsites. Inourdata,over60%oftheindividualswhosearchedforabetteroerindicatethattheyusedsuchwebsites.11If anindividual alsoreceivedanoerforagroupcontract, thereareinfactonlyN 2otherinsurers. Onlyiftheoerforagroup contract iswiththesameinsurertheindividualwas previouslyinsuredwith, therearestill N 1otherinsurers. Foreaseof expositionweignorethis, astakingaccountofthiscomplicatesnotationwithoutchangingourtestablepredictions.7Individualsonlysearchiftheirexpectedbenetsexceedsearchcostsc. Theexpectedbenetsareintermsofndinganinsurerwithalowerinsurancepremium. Anindi-vidualsearchesif:max {uh(hi) (1 +)pns, ul pns} < max {uh(hi) (1 +)E[pmin], ul E[pmin]} cForindividuals whoreceivedanoer for agroupcontract, pnsis thelowest of twooersratherthanjusttheinitialoer. Thisimpliesthatforaconsumerwithanoerforagroupcontracttheleft-handsidewillinexpectationsbesmaller(expectedgainsfromcontinuedsearcharesmaller). Suchanindividual isthuslesslikelytodevoteadditionaleortstosearchthemarketforabetteroer.Hypothesis3: Consumerswithoutanoerforagroupcontractaremorelikelytosearchforalowerpremium.ForeaseofexpositionweassumethatthesupportofF(p)isboundedfrom[p, p].Wecandistinguishthreetypesof individuals. First, individuals inbadhealthwhoalways choose health insurance with a high coverage. For these individuals healthhiisbelowhforwhichuh(h) ul= p. Second,thereareindividualsinsuchgoodhealththat theyalwaysonlytakebasicinsurance,sohiexceedsh for whichuh(h) ul= p.Andthird, there are individuals withhealthhibetweenhandhwhoprefer basicinsuranceincaseofhighpremium pandinsurancewithhighcoverageincaseoflowpremiump.Forindividualsinsuchbadhealththattheyalwayspreferhealthinsurancewithhighcoverage,thesearchdecisionsimpliesto(1 +)pns> (1 +)E[pmin] +c or pns> E[pmin] +c1 +Forindividuals ingoodhealththat always prefer tohaveonlybasicinsurance, thesearchdecisionispns> E[pmin] +cSince premiums do not depend on the health status, the above implies that individualsinbadhealthhavealowerpremiumthresholdforsearchingthanindividualsingoodhealth.8Hypothesis4: Individualswithworsehealtharemorelikelytosearchthemarket.Individualsinthethirdgrouponlyobtainhealthinsurancewithhighcoverageifthepremiumissucientlylow. If anindividual searchesthemarket(orreceivesanoerforagroupcontract)hemayndapremiumthatislowerthantheinitialoer.Forsomeindividualsinthethirdgroupthispremiumwillbesucientlylowtomaketheexpectedutilityofhighcoveragelargerthantheexpectedutilityfromonlybasiccoverage. Therefore, some individuals who switch insurer to get a lower premium mightalsoswitchtoahealthinsuranceplanwithhighcoverage.In the model we made three important assumptions. First, we imposed that there isdispersionofpremiumsinthemarket,i.e. F(p)isnon-atomic. Inthenextsubsectionwe sketch the behavior of insurers to argue that in equilibrium there is indeed premiumdispersion. However,iftherewouldnotbeanydispersionofpremiumsinthemarket,searchwouldnever bebenecial (recall that onestarts withanoer andsearchiscostly). Inthiscase,consumerbehaviorwouldreducetoonlychoosingbetweenbasicinsurance and health insurance with high coverage for which the model predicts adverseselection.The second key assumption is that we imposed that the premium for insurance withhigh coverage is proportional to basic health insurance. Alternatively,we could chooseanadditivespecicationimplyingthatthepremiumforhealthinsurancewithhighcoverage equals p+. Such a specication implies that consumers choose between basicinsurance and insurance with high coverage on comparing uh(hi) uland . Since thisisindependentofthepremium, individualsmaketheircoveragechoicealreadybeforelearningabouttheinitial oer. Theindividualshealthstatusaectsthedecisionforcoverage, but is no longer relevant in the choice for searching. The model thus simpliesto aconsumersearchmodelwith homogeneousproductsand homogenousindividuals.Inthisspecication, theonlypossibleequilibriumisonewherenoconsumersearchesthemarketbecauseallinsurershavethesamepremium.The nal keyassumptionis that individuals whosearchthemarket observe allpremiumsinthemarket. Thisdiersfromtheusual assumptioninconsumersearchmodelsthatwhensearching, consumersseepremiumssequentially, andmakesearchcosts for observing each additional premium. Our predictions are robust against chang-ingthesearchrule. Bothsearchrulesgeneratedispersionofpremiumsinequilibrium,andsimilarbehavioralpredictionsforconsumers.93.2 PremiumdispersioninequilibriumThe testable predictions for consumer behavior depend on existence of premium disper-sion. Inthissubsection,wearguethatthisshouldbepresentinequilibrium. SupposethereareNinsurersinthemarket, whichall havethesamemarginal costsmforin-surancewith basiccoverage and (1 +)m for insurancewith high coverage.12Insurersonlydierintheirpre-reformmarketsharej.Eachinsurer keeps its clients if these donot get anoer for agroupcontractwith a lower premium, and in addition do not search. Only the insurer with the lowestpremium in the market attracts individuals who decide to search. We assume that eachinsurer has the same market share in group contracts as their overall market share. Wefurthermoreassumethatallinsurersgivethesamediscountonthepremiumwhentheymakeanoerforagroupcontracttoapotentialclient.Fromthe behavior of consumers we knowthat there is heterogeneityinsearchbehavior. Individuals inbadhealthundertakesearchatalower expectedpremiumreductionthan individualsin good health. Insurerswith ahigh marketshare can postarelativelyhighpremium, whichwouldimplythattheymightlosesomeindividualswho get an oer for a group contract from an other insurer, and some individuals withbadhealthwhosearchthemarketforbetteroers. Thebiginsurerwouldthuslosesomemarketshare, butmakearelativelyhighprotperinsuree. Aninsurerwithalowmarketsharemightpostamuchlowerpremiumtoavoidlosingrelativelymanyinsureeswho get an oer for a group contract from an other insurer, and to induce theclients of other insurers to search the market. This means that the small insurer makesarelativelylowprotperinsuree, butgains, relativetoitsmarketshare, manynewinsurees(if itmanagestobecometheinsurerwiththelowestpremium). Obviously,thedegreeof pricedispersiondependsonthesizeof thesearchcostsc, butalsothevariation in market sharesjand the distribution of healthG(h) in the population areimportant.Premiumdispersionisnotonlyatheoretical prediction. AftertheDutchhealthinsurance reform substantial premium dispersion was observed in the market (see againFigure2). Indeed, thelowestpremiuminthemarketwaspostedbyasmall insurer(named AnderZorg). Relating premiumsto marketshares is dicult,because insurersareverycautiousinprovidinginformationonmarketshares. However, inthenews-papersoneoftheveinsurerswithoveronemillioninsurees(Agis, whichpostedthe12Obviously, marginal costsshoulddependonthehealthstatusof theinsuree. However, recallthattheRiskEqualizationFundcompensatesinsurersforinsuringindividualsinbadhealthinsuchwaythattheexpectedcostsofallinsureesarethesame.10highestpremiumamongthem)wasconsideredtobethebiggestloserof thereform.Theotherfourinsurerswithoveramillioninsureesmainlymaintainedtheirmarketsharebecauseofwritingmanygroupcontracts. Inparticular, usingourowndatatocalculatethefractionof groupcontracts, thereisasubstantial, positivecorrelation(0.40) between the premium posted by insurers and the fraction of insureescovered byagroupcontract.4 ThedataOurdataarefromtheDutchHealthCareConsumerPanel whichiscollectedbytheNetherlands Institute for Health Services Research (NIVEL). The panel contains about1500individuals andis aimedtobe representative for the overall population. Forwomentheagestructureinthepanel largelycoincides withtheDutchpopulation,for menolder individuals are somewhat overrepresented inthe panel. Individualsintheconsumerpanel completequestionnairesonhealthcare, healthinsuranceandrelatedissues betweentwoandvetimes per year. After twotothreeyears panelmembers are replaced to maintain representativeness. The content varies substantiallybetweenquestionnaires. Inthe empirical analyses we use informationfromthe 15questionnairessendoutbetween2004and2008. Thisobservationperiodcoversthetime period around the Dutch health insurance reform (2006). Most questionnaires arenotsenttoall panel members, inordernottooverwhelmthemwithquestionnaires.Usuallyaround 70% of the panelmembersis randomlyselectedtoreceiveaparticularquestionnaire. Combiningvariablesfromdierentquestionnairesthusquicklyreducesthesamplesize. Socioeconomicandotherbackgroundvariablesareonlyaskedonce,atthemomentaparticipantrstenterstheconsumerpanel.In December 2005, a month before the introduction of the new system, participantsanswereda set of questionsabout the oer they receivedfrom their current insurer. Itwasalsoaskedwhethertheywereplanningtosearchforbetterdealsoeredbyotherinsurers. The April 2006 questionnaire contains information on actual consumer searchbehavior, thechoiceoftheinsuranceplanandinsurer, aswell asinformationonthetotal number of oers for a group contract participants had received, and whether theyacceptedoneof theseoers for agroupcontract. Wethus knowthenames of thepre-reform and the post-reform insurer and hence whether the individual has switchedinsurer. We observe whether anindividual participates inagroupcontract, has avoluntary deductible,and has supplementaryinsurance coverage. We do not know the11extent of thesupplemental insurancecoverage. However, individuals wereaskedtoreportthetotalamountofpremiumtheypayforhealthinsurance. Wecombinethisinformationwithexternal informationaboutthepremiumforbasiccoverageof eachinsurer(and the reducedpremiumfor participation in a group contract),whichallowsustodeterminetheamountpaidforsupplementalcoverage.Table1providessomedescriptivestatistics. Wedistinguishbetweenindividualswithandwithout anoer for agroupcontract. More than70%of all individualsreceived an oer for a group contract. Individuals with a group contract are more oftenemployed, andlessoftenretired, sotheyarealsoonaverageyounger, haveahigherincomeandarehighereducated.13Recallthatabouttwo-thirdofthegroupcontractsarewithemployers, andone-thirdwithlaborunions, consumer organizations, etc.14Group contracts give an average reduction of about 6.5% on the basic insurance, and a8.5% reduction in premium for supplementary insurance (Dutch Healthcare Authority,2006). Couples are more likely to receive an oer for a group contract, mainly becausesuchanoercoversall familymembers. Thereare, however, nodierences inself-assessedphysical andmental healthbetweenbothgroups, neitherintheaveragenorinthedistribution. Alsoexpectedhealthcareuseisverysimilar. Self-assessedhealthwas only asked when individuals rst entered the panel, while expected health care usewasaskedinApril2006,afterindividualsmadetheirhealthinsurancechoice.Individualswithanoerforagroupcontractchangeinsurerandinsuranceplanmore often, but have a similar health insurance plan in terms of choice for a deductibleand the presenceofsupplementarycoverage. Individualswith agroup contract payintotal only 3.1% less on health insurance while they receiveabout 6.5% discount on thepremium for the basic insurance package. Comparing premiums that are paid, individ-ualswithanoerforagroupcontractspendmoreonsupplementaryinsurance, bothbefore andafter discounts. Table 2compareshealth insurancedecisionsinour sampletonationwidebehavior. Inoursamplemoreindividualshaveagroupcontract(72%)thannationwide(44%). Intermsof supplementaryinsurance, voluntarydeductiblesandinsurancepremiumoursamplematchesthenationwidestatisticsfairlywell.The questionnaire of April 2006 contained a question on consumer search behavior.Inparticular,individualswereaskedtoanswerthequestionDidyousearchforanew13Wedonotobserveincomedirectly,butratherobservetheamountofgovernment compensationan individual receives. Very low income households (less than e17,500 per year) receive the maximummonthlycompensationof e33.58forasingle, ande96.25foracouple. Partial(incomedependent)compensation was paid to low income household (belowe25,068 for singles ande40,120 for couples).14Thesearenational level gures. Inoursampleweobservethat85%of thegroupcontractsisobtainedviatheemployer12healthinsurancecontract? Table3displays thefractionof individual searchingthemarket. Intotal about46%of theindividualsreportstohavesearchedactivelyforotherhealthinsuranceplans. Searchislesscommonamongindividualswhodonothaveanoerforagroupcontract. Therawdataarethereforenotinlinewiththetheoretical predictions fromthesimpleconsumer searchmodel (Hypothesis 3). Wereturntothisissueinthenext section. Onlyabout30%of all individuals didnotreceiveanoerforagroupcontract, while33%of theindividualsreceivedmultipleoersfor agroupcontract. Thetableshowsthatsearchingispositivelyrelatedtothenumberofoersforagroupcontractreceived. Thisremainstrueafterstratifyingthesamplebylabormarketstatus.ExpectedhealthcareusewasaskedintheApril2006questionnaire. Respondentshad6options,answers1to5formedacategoricalscalefrom verylittleto verymuch.Thesixthanswerwasdont know. The9%of individualsthatanswereddont knowwereremovedfromthesampleforall analysesthatinvolvedtheuseof thevariableexpectedhealthcare use. Furthermore, the categories much andvery much weremerged,becauseonlyveryfewindividualsexpectedtouseverymuchhealthcare.5 Empirical resultsThis section provides insight in how well the consumer search model describes observedbehavioratthetimeoftheDutchhealthinsurancereform. Morespecically, weusethedatatotestthehypothesesderivedfromthetheoreticalmodel.5.1 TestingthehypothesesHypothesis1: Individualswithworsehealtharemorelikelytobuyhealthinsurancewithhighcoverage(adverseselection).Adverse selectionimplies that individuals withhighexpectedhealthcareneeds(thoseinbadhealth) takeahigher level of insurancecoverage, i.e. buymoresup-plementaryinsurance. Wetestforadverseselectionbyinvestigatinghowthedegreeof supplementarycoveragedependsonexpectedhealthcareuseandonself-assessedhealth. Taking the premium of the supplementaryinsurance as a measure for coverageisnotappealing,becauseofpremiumdiscountsingroupcontractsandthelargevari-ationininsurancepremiumsbetweeninsurers. Alternatively, weconstructameasurethatrelatestheadditional expenditures onsupplementaryinsurancetothepriceof13thebasicpackage. Wedenethedegreeofsupplementarycoverageastheratioofthepremiumforsupplementaryinsurance(beforediscounts)overthepremiumforbasicinsurance (before discounts). This gives the degree of supplemental insurance coverageasafractionofbasicinsurancecoverage, thelatterbeingthesameforallindividualsatallinsurers.Table 4presents the results of regressions for supplemental insurance coverage.Column (1) shows the results of a base specication where only expected health care useisincluded. Individualswhoexpectverylittleuseofhealthcare(thereferencegroup)have signicantly lower supplementary insurance coverage than individuals who expectto use more health care (i.e. little, average or (very) muchexpected care use). Beyondthereferencecategoryverylittleexpectedcare supplementaryinsurancecoverageisnotincreasinginexpectedhealthcareuse. Thissuggeststhresholdbehavior,whichisconsistent with our consumer search model. Individuals with very little expected healthcare use prefer a low level of (supplementary) insurance coverage. If the expected healthcare use is more than very little, it is benecial to take higher supplementary insurancecoverage.Thequestiononexpectedhealthcareusewasaskedinthesamequestionnaireasthequestiononthehealthinsurancechoice(April 2006). Thismaycausetwoprob-lems. First,expectedhealthcareuseisaskedoverthefullcalendaryearof2006,andindividuals mightalreadyhaveapartial observationontheir healthcareuse. Thiscanpotentiallyweakenthelinkbetweenexpectationsandinsurancechoice. Further-more, individuals report their expected health care use after having decided about theirhealth insurance plan. The expected health care use may thus reect adverse selectionas well as moral hazard. To get a better idea of the importance of adverse selection, weconsiderthequestionDidyoutakeintoaccounttheamountof healthcareyouexpectto use this year in deciding upon which health insurance to purchase?. When answeringpositively, individualscouldindicateIbought extensivesupplementarycoverage, orIbought very limited supplementary coverage or I bought no supplementary coverage. Be-cause this question refers to expected health care use at the time the insurance decisionwasmade,itseparatesadverseselectionfrommoralhazard. Wegroupverylittleandno supplementary coverage and show the answers in Table 5, broken down by expectedhealthcareuse. Indeed, thehighertheexpectedhealthcareuse, themorelikelyitisthatanindividualtookmoreextensivesupplementarycoverage.Asanalternativetoexpectedhealthcareuse,wecanalsouseself-assessedhealthtoinvestigateadverseselection. Recall thatself-assessedhealthisaskedonlyatthemoment an individual rst enters the panel. For our sample it is therefore always asked14beforethereform, andthusbeforeindividualshadtodecideontheirinsuranceplan.However, forsomeindividualstheinformationonself-assessedhealthisalreadyafewyearsold.15Column(2)ofTable4showstheresultsfromregressingsupplementaryhealthinsurancecoverageonself-assessedphysical andmental health. Onlyphysicalhealthhasasignicantimpactonthehealthinsurancedecision. Recallthatahighervalueof healthindicatesworsehealth. Individualswithagoodphysical healththusobtainonaveragelessextensivesupplementaryhealthinsurancecoverage. Thisindi-catesadverseselection,which, again,conrmsthersthypothesisfromtheconsumersearchmodel.Hypothesis2: Alowerpremiuminducesindividualstotakemorehealthinsurancecoverage.Toinvestigatethissecondhypothesisweregressthesupplementaryinsurancecov-eragenot onlyonexpectedhealthcareuseor self-assessedhealth, but alsoonthepremiumforthebasichealthinsurance. Columns(3)and(4)of Table4reporttheresults. Thebasic insurancepremiumhas asignicant negative impact onsupple-mentaryinsurancecoverage. Individuals whopayalower premiumaremorelikelytoobtainmoresupplementaryinsurancecoverage(evenaftercontrollingforexpectedhealthcareuseorhealth), whichconrmsthesecondhypothesisfromtheconsumersearchmodel.To investigate the robustness of this conclusion, we add additional control variables.First,inthecolumns(5)and(6),weincludegenderandincome. Womentake,onav-erage, more health insurance coverage, which is consistent with the common belief thatwomen are more risk averse than men. Furthermore, health insurance is a normal good(i.e. healthinsurancecoverageincreasessignicantlywiththeincomeofindividuals).But moreimportant, theeect of thepremiumonsupplementaryhealthinsurancecoverageremainsnegativeandsignicant. Thisremainswhenaddingage, householdcompositionandyearsofeducationtotheregression(seecolumns(7)and(8)). Noneof these covariates has a signicant eect on supplementary health insurance coverage,andothercovariateeectsdonotchangeafterincludingtheseadditionalvariables.Hypothesis3: Consumerswithoutanoerforagroupcontractaremorelikelytosearchforalowerpremium.15As panel members are replaced after two to three years, the health information can at maximumbethreeyearsold.15Individualswhoreceivedanoer for agroupcontract,canchoose(withouthavingsearched) between two oers. Their best oer has in expectation a lower premium thanindividualswhodidnotreceiveanoerforagroupcontract. Thisimpliesthattheexpectedgainsfrom searchare lowerfor individualswith anoer for agroup contractandthatthey,therefore,arelesslikelytoengageinsearch.Recall from the previous section that individualswith an oer for a group contractindicate to search on average more often for a new health insurance contract(see Table3). A potential problem is that individuals might consider the oer for a group contractasanewhealthinsurancecontract. Theymaythenclassifythemselvesassearchersafterhavingcomparedtheinitial oerwiththeoerforagroupcontract, whichisnotconsideredassearchinginourmodel. Therefore, wealsoconsiderthefollow-upquestion: What sourcesdidyouusewhensearchingforahealthinsurancecontract?Multiple answerswere allowed. Individualsmost often report having used the internet(73%),especiallywebsitesthatcompareinsurancecontractsfromallinsurers(84%ofthosehavingusedtheinternet) andwebsites of insurers (80%of thosehavingusedinternet). Other answersincluded advicefrom a family member (23%), contact with ahealthinsurerviae-mailortelephone(21%)andadvertisements(19%).Weconsider asastricter denitionforsearchonlyusing(independent) websitesthatcompareinsurancecontractsofallinsurers. Accordingtothisdenition, 32%ofthosewithanoerforagroupcontract, andonly19%of thosewithoutanoerforagroupcontract, havesearched. Table6showstheresultsofaprobitmodel fortheeect of an oer for a group contract on search behavior,using the strict denition forsearching. Column(1)showsthatreceivinganoerforagroupcontractsignicantlyincreasesthepropensitytosearch. Column(2)showsthatthiseect remainsaftercontrollingforlabormarketstatus. Incolumn(3)wealsoaddthepremiump0oftheinitial oer. Thiscolumnshowsthat individualswith an oer for agroup contract aresignicantlymorelikelytosearchiftheinitialoerwashigh. Theoppositeistrueforindividualswithoutan oer for a group contract. Thisresult remainsafter controllingforadditional observedcharacteristics (seecolumn(4)). Inthenext subsection, weinvestigatefurther whythe data are not consistentwith the third hypothesisfrom ourconsumersearchmodel.Table7showsthepercentageofindividualsthatswitchesinsureratthemomentof thereform. Wedistinguishbetweenthosewithandwithoutanoerforagroupcontract, andthosewhodidanddidnotsearchthemarketforbetteroers. Asonemight expect individuals who have searched the market, and those who received an oer16for a group contract, are much more likely to switch insurer than their counterparts. Inthe table we used the strict denition of searching, which explains why some individualswhodidnotreceiveanoerforagroupcontract,andwhodidnotsearch,stillswitchinsurer. Thistableshowsthatsearchingactuallyincreasesthelikelihoodofswitchinginsurer,andthusmeasuresrelevantindividualbehavior.Hypothesis4: Individualswithworsehealtharemorelikelytosearchthemarket.The model predicts that individuals inbadhealthderive more expectedutilityfromahealthinsurancewithextensivesupplementarycoverage. Recallfromthersthypothesis that this adverse selection was present in the data. Individuals in bad healthshouldthusbemorelikelytobenetfromsearchingthemarket. Sinceweimposedthatall individualshavethesamesearchcosts, andthepremiumsdonotdependonhealth,individualsinworsehealthshouldsearchmoreoften.InTable8weshowagain estimationresultsfor aprobit modelfor thesearchdeci-sion,butweincludehealth asan explanatoryvariable. Again,weuseexpectedhealthcare utilization and self-assessed health as measures for individual health. Columns (1)and (2) indicate that both expected health care use and self-assessed health do not haveasignicantimpactonsearchbehavior(althoughthecoecientshavetheexpectedsigns). Incolumns(3)and(4)weaddthepremiumoftheinitial oer, butthisdoesnot changethe eectof health on searchbehavior. Finally, in columns(5)-(8), weaddindividual characteristics. This does not change the eect of health on search behavior.Columns(5)and(6)pointoutaneectofincomeonsearchbehavior. Ifweincludeageandeducation, incolumns(7)and(8), theeectof incomeisabsorbedinthesevariables. All resultsshowpositive, butinsignicanteects, of badhealthandhighexpectedhealthcareuseonsearching. Thisimpliesthatwecannotrejectthefourthhypothesisofthemodel. However, theeectsofhealthonsearchingareatmostverysmall(ifatallpresent).5.2 ExplainingsearchbehaviorThe consumersearch model fails in explainingthe search behavior of individuals,and,inparticular, thedierenceinsearchbehaviorbetweenindividualswithandwithoutan oer for a group contract. In this subsection, we further investigate search behavior.Inourtheoreticalmodel,wemadeanumberofsimplifyingassumptions. First,weimposedthatindividualsareonlyheterogeneousinhealth, buthavethesamesearch17costs. However,heterogeneityin searchcostswill only change the model predictionsifthesizeofthesearchcostsarerelatedtohealth. Indeed, Buchmueller, FeldsteinandStrombom (2002) show, for the US, that less healthy individuals (with higher expectedhealth care expenditures) experience higher costs of switching medical provider, and aretheleastpricesensitive. Second, weassumedthatindividualsknowthedistributionof premiums inthemarket. If individuals, however, donot knowthis distribution,receiving an oer for a group contract may be informative on the variation in premiumsinthemarket. Individualscanusethisinformationtoupdatetheirbeliefs. Receivinganoer for agroupcontract couldthenstimulatesearch. Third, weimposedthateachindividualhasthesameprobabilityofreceivinganoerforagroupcontract. In2006, thelargerpartof thegroupcontractswaswithemployers. Thissuggeststhatnot everyindividualhas thesameprobabilityofreceivingsuchanoer. Inparticular,if receiving an oer for a group contract is correlated to the size of the search costs, themodel predictionschange. Belowweprovidesomeempirical evidenceonthesethreepossibleviolationsofthemodelassumptions.Thereareanumber of reasons whythesizeof searchcosts couldberelatedtoindividual health. First, within85%of thecouplesbothpartnershavethesamein-surer,andtheymayhaveeconomiesofscalewhensearching(i.e. onepartnercollectsinformation and decides about which insuranceto take). Furthermore, within a multi-personhousehold, theprobabilityishigherthatsomeonehasabadhealthandthushigh expected health care use. Second, older people, who on average have worse healththanyoungerpeople, mayhavemoreproblemscollectinginformation. Forexample,olderpeoplemayhavemoreproblems ndinginformationontheinternet, whichisthemostusedandprobablycheapestsearchmethod. Indeed, only50%ofthepeopleaboveage 65haveaccesstointernetathomecomparedtoabout 90%oftheindividu-alsbelowage65. However, evenaftercontrollingforhouseholdcomposition, ageandotherobserveddierences, receivinganoerforagroupcontractstill hasapositiveandsignicant eect onsearch(seeTable9). Another, third, explanationwhythesizeof searchcostsmayberelatedtohealthisthatindividualsinbadhealthcouldbeafraidthatinsurerswillrejectthem. Beforethereformitwascommonpracticeinthe private market that insurers declined applicants (recall section 2, and in particularfootnote4). Afterthereform, insurersarebylawnotallowedtodeclineapplicantsforthebasicinsurancepackage, buttheycandenyclientssupplementaryinsurance.Although all insurersannouncedbeforehand that theywould accepteveryoneevenforsupplementaryinsurance(whichalsohappenedinpractice),individualsinbadhealthmaystillworryaboutbeingrejectedeitheratthisstageorlater. Thesurveycontains18aquestionaboutaboutwhypeopledidnotchangeinsurer. Lessthan1%indicatesthattheydidnotswitchbecausetheywereafraidofbeingdeniedbyanotherinsurer.AsecondpossibleexplanationisthatindividualsdonotknowthedistributionofpremiumsF(p) inthemarket. Individualsbelievingthat thevariation inpremiumsislow, arelikelytodecidenottosearch. Individualswhoreceivedanoerforagroupcontract may realize that the variation is premiums is larger than assumed, which mayinducethemtosearchactively. Althoughwedonothaveanydirectevidenceonindi-vidualbeliefs,itisrelevanttonotethatbeforethereformthegovernmentannouncedthattheaverageannualpremiumwouldbeaboute1106. Theactualpremiumswerebetweene990 and e1120, so almost all insurers had a lower premium. This may implythat after learningtheir premium,most individualsbelievedthat they receiveda goodoer. If individualsareunawareof thedistributionof premiums, thenthosewithahighpremiumoer(closetothatannouncedbythegovernment)mighthavebelievedthatthevariationinpremiumswasverylowandthuswouldnothavesearchedfur-ther. Whilethosewithalowpremiumoermayhaveoverestimatedthevariationinpremiums,andthussearchedforanevenlowerpremium.A third possible explanation is that not all individuals have the same probability ofreceivingan oer for agroup contract. Obviously,theprobabilityof receivingan oerforagroupcontractisrelatedtoemploymentstatus. However,evenamongemployedworkers there is substantial heterogeneity. Table 10 presents estimates of a probit modelfor receiving an oer for a group contract. Column (1) shows that employed individualsare, indeed, morelikelytoreceivesuchanoer. Healthdoesnothaveasignicantimpactontheprobabilityofreceivinganoerforagroupcontract. Next,weincludeasregressoravariablewhichmeasurestheknowledgeofindividualsaboutthehealthinsurance reform. This variable is based on 15 statements, included in the questionnaireofOctober2005(sobeforepremiumswereannounced,andoersforagroupcontractweremade), towhichindividualshadtoanswertrueorfalse(ortheycouldanswerdont know). The knowledge variableequals the number of correct answers minusthenumberof wronganswers. Thisguaranteesthatsomeonewhodoesntknowananswer, getsthesameexpectedscorewhenguessingaswhenansweringdontknow.The average score in our population equals 4.2 (with a maximum of 13 and a minimumof-7). Incolumn(2)weaddthisasanadditionalregressor,andndthatindividualswithmoreknowledgeaboutthehealthinsurancereformweresignicantlymorelikelytoreceiveanoerforagroupcontract. Column(3)showsthatthiseect remainslargeandsignicant after controllingfor other individual characteristics. Theonlyimportant individual characteristic is income, oers for a group contract are associated19withhighincomes.Fourth, searchcostsmayalsobedirectlyrelatedtoreceivinganoerforagroupcontract. Anindividual withsuchanoerhastocomparethisoertotheoerfromthecurrent insurer andthereforeincreases his/her knowledge of the system, whichreducesthecostsofcomparingfurtheroers.Obviously,not all employed workershave the same probability of receivingan oerforagroupcontract. Insurersmainlywritegroupcontractsfor rmswithhigherpaidemployees, andwithmoreknowledgeaboutthehealthinsurancereform. Of course,itmightbethatthisismainlydrivenbytherms,becausehigher-incomeworkersorworkers with more knowledge about the reform may push their rm harder to establishagroupcontract. Orlargerrms(withmanypreviouslyprivatelyinsuredemployees)already had a group contract with a private insurer before the reform. Knowledge aboutthehealthinsurancereformisalsopositivelycorrelatedwithsearching(acorrelationof0.167,signicantatthe1%level).Oers for agroupcontract are more oftenmade toindividuals whowere morelikelytocompareinsurers, i.e. individualswithlowsearchcosts. Formostinsurersthepremiumofagroupcontractisbelowtheregularpremiumsoftheotherinsurers.Insurers canthus set higher regular premiums, because averagesearchcosts inthemarket for individual contracts (i.e. without an oer for a group contract) are relativelyhigh. ThisargumentfollowsStahl (1989), whoshowswithina consumersearchmodelthatifthenumberofinformed(lowcosts)individualsisreduced(asisthecaseinthemarketforindividualcontracts),pricedispersionincreases.The possibility of oering a group contract facilitates insurers to apply third degreepricediscrimination, whichmaybewelfarereducing. Themainreasonforthegov-ernmenttoallowforgroup contractswasthatitcreatedthepossibilityforinsurerstoinsure most employeesof a rm and at the same time also insure the rm for the costsof for example sickness absenteeism. The government hoped that such combinations ofinsuranceswouldinduceinsurerstoputmoreeortinpreventionofhealthproblems.Our consumer search model should be modied such that it allows for heterogeneityinsearchcostsandtheprobabilityofreceivingan oer for a group contractshouldbenegatively related to search costs. If this is the case, then the model is, of course, capa-bleofexplainingthefactthatindividualswithanoerforagroupcontractaremoreengaged in search. We simulated the model with both homogeneous and heterogeneoussearchcosts(seeTable11for theparametervaluesthat wereused). ResultsarelistedinTable12, andshowthatheterogeneityinsearchcostscangeneratethatindividu-alswithanoerforagroupcontractmoreoftensearchandobtaininsuranceswitha20lowerpremiumforbasiccoverage. Becauseof thelowerpremium, theyobtainmoresupplementary coverage, and pay a higher overall premium (which is also what the datashow). However, heterogeneityinsearchcostsbetweenindividualswithandwithoutanoerforagroupcontractcannotexplainthatamongindividualswithoutanoerforagroupcontractsearchisnegativelyrelatedtotheoerfromthecurrentinsurer.AlsochoosingforthedefaultincaseofadecisionoverloadasdiscussedbyFrankandLamiraud(2008)cannotexplainthis. Itcanonlybeexplainedif insurersbasetheirpremiumontheestimatedsearchcostsamongtheirinsurees(recall thatindividualsare not randomlydistributedoverinsurers,but thatthisis theconsequenceoftheoldcombinedpublicandprivatesystem). Theearlierdiscussedalternativeexplanationisthatindividualsdonothavecorrectbeliefsaboutthedistributionofpremiums.6 DiscussionandconclusionWe presented a simple consumer search model for individual health insurance decisionsatthemomentof theDutchhealthinsurancereform. Themodel providedfourhy-potheses, whichwecouldtestempirically. Ourdataconrmbothhypothesesonthechoiceforinsuranceplan. Inparticular, thereisadverseselectioninthemarketandhealthinsurancecoverageisdecreasinginthepremium.The simpleconsumer searchmodel had more problemsexplainingboth hypothesesonindividual searchbehavior. Inparticular, thedatarejectedthehypothesis thatconsumerswithoutanoerforagroupcontractaremorelikelytosearch. Ourpre-ferred explanation is that group oers are targeted towards individualswhoare betterinformedaboutthehealthinsurancesystem. Weprovidedsomeempirical evidencethatsupportsthisexplanation.Forpublicpolicyitmightbeaseriousconcernthatbetterinformedindividuals,i.e. those withlower searchcosts, are more likelytoreceive anoer for agroupcontract. Thismightsuggestthatinsurersusegroupcontractsforcream-skimming,for example, bysettinghighpremiums, but oeringmaximumdiscounts ongroupcontractstolowhealth-risksemployees. Furthermore,thegroupcontractstakebetterinformedindividualsoutof theregularmarket, whichallowsinsurerstoexploitthehigher searchcosts of theremainingindividuals inthis segment. This will leadtolowercompetition, andmorepricedispersion. Sincewesawthatthechoiceofhealthinsurancecoverageisstronglyrelatedtothepremium, itmayalsoaectequityandaccesstohealthcarewithinthepopulation.21Inrecent years, theuseof groupcontractsbecameevenmorepopular. Insurersalso started to oer premium reductions to associations of individuals with a particularillness. Oeringpremiumreductionstoveryspecichigh-riskgroupsseemscounter-intuitive, butistheconsequenceoftheexistenceoftheRiskEqualizationFund. TheRiskEqualizationFunditselfisusefulinthecurrentsystemofmanagedcompetition,asitstimulatescompetitionbyequalizingtheexpectedhealthcarecostsof all indi-vidualsforinsurers. However,insurersstartedexploitingsmallawsinthesystembytargetinggroupcontractstowardsgroupsofindividualswithanillnessforwhichthecompensationmighthavebeensettohigh.Thesystemof managedcompetitionseemstobesuccessful inkeepingpremiumsforhealthinsurancelow. Intheyearoftheintroductioninsurersincurredsubstantiallosses, mainlybecauseof theirattemptstoattractasmanyinsureesaspossible. Atthat time experts feared that premiums would increase sharply to compensate for theselosses. However, theriseinpremiumswaslowin2007and2008andevennegativein2009. Onemightarguethatthewillingnessof individualstoswitchcreatedenoughcompetitiontowithholdinsurerstoincreasepremiums.22ReferencesBrown, J.R. and Goolsbee, A., 2002. Does the Internet Make Markets More Compet-itive?Evidencefrom the Life InsuranceIndustry,JournalofPoliticalEconomy110,p.481-507.Buchmueller, T.C., Feldstein, P.J. andStrombom, B.A., 2002. SwitchingCosts,Price Sensitivity andHealthPlanChoice, Journal of HealthEconomics 21,p.89-116.Connie,D. and ONeill,D., 2008. An EcientEstimatorfor Dealingwith MissingData on Explanatory Variables in a Probit Choice Model, NUIM Working Paper#N1960908.DutchHealthcareAuthority(NZa, NederlandseZorgautoriteit), 2006. Detussen-standopdeverzekeringsmarkt. Monitorzorgverzekeringsmarktjuni2006(Theinterimscoreonthehealthinsurancemarket. MonitorhealthinsurancemarketJune2006).Frank, R. andLamiraud, K., 2008. Choice, PriceCompetitionandComplexityinMarketsforHealthInsurance,NBERWorkingPaper13817.Janssen, M.C.W. andMoraga-Gonzalez, J.L., 2004. StrategicPricing, ConsumerSearch and the Number of Firms, Review of Economics and Statistics71, p.1089-1118.Janssen, M.C.W., Moraga-Gonzalez, J.L. and Wildenbeest, M.R., 2005. Truly costlysequential searchandoligopolisticpricing, International Journal of IndustrialOrganisation23,p.451-466.Pauly, M.V., Herring, B. and Song, D., 2006. Information Technology and ConsumerSearch for Health Insurance, International Journal of the Economics of Business13,45-63.Sorensen, A.T., 2000. EquilibriumPriceDispersioninRetailMarketsforPrescrip-tionDrugs,JournalofPoliticalEconomy108,p.833-850.Sorensen, A.T., 2001. AnEmpirical Model ofHeterogeneousConsumerSearchforRetailPrescriptionDrugs,NBERWorkingPaper8548.23Stahl,D.O.,1989. OligopolisticPricingwithSequentialConsumerSearch,Ameri-canEconomicReview79,p. 700-712.Vektis, 2007. Zorgmonitor. Jaarboek2007. Financiering vande zorgin2006(Healthcaremonitor. Yearbook2007. Financinghealthcarein2006).24CATASTROPHIC INSURANCEBASIC INSURANCESUPPLEMENTARY INSURANCE Private market Private marketPublic provision/social insurancePublic provision/social insurancecompulsorypubliclow premium(ca. 25)no selectionvoluntaryprivatemarketbasedpremium(ca. 230)selectionallowedcompulsoryregulated private provisionnominal premium (ca. 90)+ incomebased contributionno selection< 33.000 33.000BEFORE 2006 FROM 2006 = the given premium is for a 30year old healthy maleFigure1: TheDutchhealthinsurancereform0.050.100.150.200.250.300.3580 82 84 86 88 90 92 94 96 98 100MONTHLY PREMIUM BASIC INSURANCEFRACTIONFigure2: Histogrammonthlyinsurancepremium25Table1: Descriptivestatisticsoerforgroup contractno yesfraction 27.62% 72.38%age(inyears) 56.00 51.20female 59.87% 54.61%couple 75.88% 80.96%haschildren 35.37% 46.71%lowincome 38.94% 28.95%verylowincome 12.21% 6.33%yearsofeducation 11.84 12.36employed 35.29% 58.36%retired 44.12% 27.81%physicalhealth(1-5scale;1=excellent) 2.95 2.77mentalhealth(1-5scale;1=excellent) 2.41 2.39expectedhealthcareuse(1-5scale;1=verymuch) 2.76 2.66changed insurer 10.69% 24.76%changed plan,sameinsurer 12.26% 15.43%hasdeductible 7.35% 7.14%hassupplementaryinsurance 95.81% 95.44%totalinsurancepremium(ine) 108.15 105.89basicinsurancepremium(ine) 87.43 83.80supplementaryinsurancepremium(ine) 20.62 22.09basicinsurancepremiumbeforediscount(ine) 87.43 87.74supplementaryinsurancepremiumbeforediscounts(ine) 20.62 23.41healthinsurancefromsicknessfundin2005 65.84% 57.77%privatehealthinsurancein2005 30.43% 35.59%civilservanthealthinsurancein2005 3.73% 6.64%observations 322 84426Table2: Descriptivessamplevs. nationalSample NationalPercentage withinsuranceongroupcontract 72% 44%Percentage withsupplementaryinsurance 95% 93%percentageamonginsuredongroup contract 96% 94%percentageamongindividuallyinsured 94% 92%Percentage switchinsurer 21% 18%percentageamonginsuredongroup contract 25% 28%percentageamongindividuallyinsured 11% 10%Percentage withdeductible 7% 5%percentageofwhichhasdeductibleofe100 31% 38%percentageofwhichhasdeductibleofe200 33% 18%percentageofwhichhasdeductibleofe300 11% 10%percentageofwhichhasdeductibleofe400 3% 4%percentageofwhichhasdeductibleofe500 22% 31%average premiumpaidforbasicinsurance(ine) 84.56 85.42average premiumbasicinsurancebeforediscounts(ine) 87.77 88.33Source: Vektis(2007)Table 3: Percentageof searchersbyreceivednumberof oers for agroup contract andlabormarketstatus.All Employed Retired Othernooerforgroup contract 31.09 (267) 45.26 (95) 17.09 (117) 36.36 (55)1oerforgroupcontract 47.21 (377) 50.48 (208) 33.33 (114) 63.64 (55)2+oersforgroup contract 58.25 (309) 65.28 (193) 38.16 (76) 62.50 (40)average 46.27 (953) 55.24 (496) 28.34 (307) 53.33 (150)Note: numberofobservationsinbracketsNote: othercontainsunemployed,disabled,infull-timeeducationandhomeduties27Table4: Estimationresultsfromregressingsupplementaryinsurancecoverage.(1) (2) (3) (4) (5) (6) (7) (8)verylittleexpectedcare 0 0 0 0littleexpectedcare 0.0330.0340.0320.028(0.017) (0.016) (0.017) (0.017)average expectedcare 0.0350.0380.0390.038(0.017) (0.016) (0.017) (0.017)(very)muchexpectedcare 0.024 0.0320.0380.035(0.018) (0.018) (0.018) (0.018)physicalhealth 0.0110.0140.0170.016(0.006) (0.006) (0.006) (0.006)mentalhealth 0.009 0.010 0.008 0.008(0.006) (0.006) (0.006) (0.006)monthlybasicinsurance 0.3320.4640.2530.3950.2560.412premium(/100) (0.128) (0.132) (0.130) (0.138) (0.133) (0.139)female 0.0160.0180.0190.021(0.010) (0.010) (0.010) (0.010)lowincome 0.0340.0310.0310.030(0.010) (0.010) (0.011) (0.011)verylowincome 0.0620.0620.0560.058(0.016) (0.017) (0.020) (0.021)age 0.0001 0.0002(0.0004) (0.0004)single 0.001 0.001(0.015) (0.014)haschildren 0.005 0.004(0.011) (0.010)yearsofeducation 0.002 0.001(0.002) (0.002)intercept 0.233 0.252 0.510 0.637 0.447 0.571 0.421 0.563(0.014) (0.016) (0.110) (0.113) (0.110) (0.117) (0.123) (0.124)observations 858 933 858 933 823 893 817 888Note:=signicantat1%level,=signicantat5%level,=signicant at10%levelNote: Physical andmentalhealtharemeasuredona5-point scale,where1isexcellent and5ispoor28Table5: Howdidexpectedhealthcareuseaectyourinsurancechoice?Expecteduseofhealthcarenotmuch,verylittle little notlittle (very)muchboughtextensivesupplementarycoverage 6.67% 12.04% 15.99% 25.76%boughtverylimitedornosupplementarycoverage 9.16% 11.48% 5.15% 2.53%observations 120 357 369 198Table6: Probitmodelforsearching(1) (2) (3) (4)oerforgroup contract 0.4170.34214.61213.949(0.093) (0.098) (5.417) (5.729)employed 0.047 0.053 0.237(0.122) (0.124) (0.140)retired 0.6920.6870.493(0.139) (0.142) (0.183)monthlybasicinsurancepremiumoered 0.0970.097nooerforgroup contract (0.055) (0.058)monthlybasicinsurancepremiumoered 0.0730.065oerforgroup contract (0.028) (0.029)female 0.056(0.104)yearsofeducation 0.054(0.021)age 0.014(0.005)lowincome 0.090(0.116)verylowincome 0.015(0.220)single 0.064(0.136)children 0.064(0.108)intercept 0.890 0.617 7.943 8.120(0.082) (0.122) (4.820) (5.130)observations 1143 1143 948 893Note:=signicantat1%level,=signicantat5%level,=signicant at10%level.Note: thestrictdenitionforsearch,i.e. search viacomparison websites,isusedhere.Note: column(2)-(4)appliestheConnieandONeill(2008) correction formissingcovariates.29Table7: Fractionofswitchersbyoerreceivalfor groupcontractandsearchbehavioroerforgroup contract nooerforgroup contractsearch nosearch search nosearchswitchedinsurer 34.29 14.88 28.26 3.18notswitchedinsurer 65.71 85.12 71.74 96.82observations 417 410 92 220Note: thestrictdenitionforsearch,i.e. search viacomparison websites,isusedhere.Table8: Probitmodelforsearching(1) (2) (3) (4) (5) (6) (7) (8)littleexpecteduse 0.178 0.268 0.261 0.219(0.138) (0.169) (0.172) (0.176)average expecteduse 0.084 0.187 0.175 0.069(0.136) (0.167) (0.170) (0.176)(very)muchexpecteduse 0.247 0.083 0.054 0.084(0.152) (0.185) (0.190) (0.197)physicalhealth 0.061 0.002 0.037 0.096(0.050) (0.062) (0.064) (0.066)mentalhealth 0.012 0.073 0.082 0.07(0.048) (0.061) (0.063) (0.064)oeredmonthlypremium 0.033 0.031 0.028 0.024 0.029 0.029basicinsurance (0.023) (0.022) (0.023) (0.023) (0.024) (0.024)female 0.130 0.145 0.065 0.057(0.107) (0.103) (0.118) (0.113)lowincome 0.2800.3170.081 0.106(0.118) (0.113) (0.130) (0.125)verylowincome 0.4020.4250.094 0.036(0.215) (0.205) (0.253) (0.241)age 0.0140.017(0.005) (0.004)single 0.221 0.166(0.158) (0.154)children 0.049 0.025(0.123) (0.120)yearsofeducation 0.0780.071(0.024) (0.023)intercept 0.408 0.382 3.208 3.126 2.781 2.529 3.148 3.086(0.118) (0.137) (2.005) (1.995) (2.087) (2.081) (2.133) (2.125)observations 1040 1128 649 711 620 678 617 675Note:=signicantat1%level,=signicantat5%level,=signicant at10%level.Note: thestrictdenitionforsearch,i.e. search viacomparison websites,isusedhere.30Table9: Probitmodelforsearching(1) (2) (3)oerforgroup contract 0.5230.3290.489(0.096) (0.102) (0.097)oneemployedpartner 0.494(0.123)twoemployed partners 0.531(0.126)olderthan65 0.411(0.112)couple 0.202(0.113)female 0.2060.140 0.147(0.084) (0.088) (0.087)yearsofeducation 0.0600.0450.064(0.018) (0.019) (0.018)lowincome 0.060 0.041 0.007(0.097) (0.106) (0.099)verylowincome 0.272 0.128 0.034(0.172) (0.180) (0.187)expectedhealthcareuse 0.117 0.002 0.147(0.141) (0.123) (0.140)expectedhealthcareuse 0.075 0.189 0.115(0.141) (0.122) (0.140)expectedhealthcareuse 0.025 0.075 0.107(0.156) (0.143) (0.155)intercept 1.365 1.841 1.511(0.293) (0.292) (0.318)observations 973 854 973Note:=signicantat1%level,=signicantat5%level,=signicant at10%level.Note: thestrictdenitionforsearch,i.e. search viacomparison websites,isusedhere.Note: column(2)appliestheConnieandONeill(2008)correction formissingcovariates.31Table10: Probitmodelforreceivinganoerforagroupcontract(1) (2) (3) (4)employed 0.5500.5360.2980.364(0.129) (0.134) (0.150) (0.156)retired 0.036 0.022 0.068 0.041(0.129) (0.134) (0.180) (0.187)physicalhealth 0.090 0.1200.115(0.058) (0.060) (0.062)mentalhealth 0.075 0.1010.133(0.055) (0.057) (0.059)littleexpecteduse 0.015(0.178)average expecteduse 0.179(0.179)(very)muchexpecteduse 0.124(0.195)knowledgereform 0.0330.0270.038(0.015) (0.016) (0.017)female 0.143 0.142(0.108) (0.114)lowincome 0.2710.260(0.115) (0.120)verylowincome 0.4020.386(0.203) (0.215)age 0.003 0.003(0.005) (0.006)single 0.065 0.012(0.136) (0.141)haschildren 0.047 0.053(0.117) (0.120)yearseducation 0.020 0.026(0.021) (0.023)intercept 0.426 0.309 0.498 0.188(0.205) (0.225) (0.476) (0.477)observations 942 879 829 755Note:=signicantat1%level,=signicantat5%level,=signicant at10%level32Table11: Parametersusedinsimulationpremiumoeredbypreviousinsurer p0N(1080, 45)utilityfromhighcoverage uhU(2680; 3080)discountoeredbygroupcontract 6.5%markupforsupplementaryinsurance 0.40search costs c 150search costswithoutgroup oer c0150search costswithgroup oer c125utilityfromonlybasiccoverage ul2400numberofrmsinthemarket N 20probabilityreceiveoerforgroup contract 0.56Table12: SimulationresultsSearchcostsequal c,bothwithandwithoutoerforgroupcontractnooerforgroup contract oerforgroup contractpercentagechoosehighcoverage 64.8% 69.4%percentagechoosetosearch 22.2% 0.2%average pricebestoerbeforesearch 1080.0 1006.2average pricepaidforbasicinsurance 1049.2 1006.0Searchcostsequal c0withoutoerforgroupcontractandc1withoerforgroupcontractnooerforgroup contract oerforgroup contractpercentagechoosehighcoverage 64.8% 71.1%percentagechoosetosearch 22.2% 40.4%average pricebestoerbeforesearch 1080.0 1006.2average pricepaidforbasicinsurance 1049.2 988.2Note: parametersusedinthesimulationsareshown inTable1133