AsecondapproachdeemphasizesthemarginalsearchcostasasourceforpricedispersionInsteadconsumersaccesspriceinformationbyconsultingan147informationclearinghouse148eganewspaperoranInternetpriceco ID: 818628
Download Pdf The PPT/PDF document "1IntroductionSimpletextbookmodelsofcompe..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
1IntroductionSimpletextbookmodelsofcompe
1IntroductionSimpletextbookmodelsofcompetitivemarketsforhomogeneousproductssuggestthatall-outcompetitionamong rmswillleadtotheso-calledlawofoneprice.Yet,empiricalstudiesspanningmorethanfourdecades(seeTables1aand1b)revealthatpricedispersionistheruleratherthantheexceptioninmanyhomogeneousproductmarkets.Theobservationthatthepricesdi¤erent rmschargeforthesameproductoftendi¤erby30percentormoreledHalVariantosuggestthatthelawofonepriceisnolawatall(Varian,1980,p.651).Thischapterprovidesauni edtreatmentofseveraltheoreticalmodelsthathavebeendevelopedtoexplainthepricedispersionobservedinhomogeneousproductmarkets,andsurveystheburgeoningempiricalliterature(includingthestudiessummarizedinTables1aand1b)whichdocumentsubiquitouspricedispersion.AkeymotivationforthischapteristodispeltheerroneousviewthattheInternetthroughitsfacilitationofdramaticdeclinesinconsumersearchcostswillultimatelyleadtothelawofoneprice.Whenconfrontedwithevidenceofpricedispersion,manyarequicktopointoutthateveninmarketsforseeminglyhomogeneousproducts,subtledi¤erencesamongtheserviceso¤eredbycompeting rmsmightleadthemtochargedi¤erentpricesforthesameproduct.NobelLaureateGeorgeStiglersinitialresponsetowagsmakingthispointwasphilosophical:...[While]aportionoftheobserveddispersionispresumablyattributabletosuchdi¤erence[s]...itwouldbemetaphys-ical,andfruitless,toassertthatalldispersionisduetoheterogeneity(Stigler,1961,p.215).Thirty- veyearslater,theliteraturehasamassedconsiderablesupportforStiglersposition.AsweshallseeinSections2and3,thereisstrongtheoreticalandempiricalevidencethatmuch(andinsomemarkets,most)oftheobserveddispersionstemsfrominformationcostsconsumerscostsofacquiringinformationabout rms,and/or rmscostsoftransmittinginformationtoconsumers.AsFigure1reveals,researchoninformation,search,andpricedispersionhasbecomeincreas-inglyimportantsincethepublicationofStiglersseminalarticleontheEconomicsofInformation.Untilabout1998,moststudiesfocusedonenvironmentswhereconsumersincurapositivecostofobtainingeachadditionalpricequote.Searchcostsinthesestudiesconsistofconsumersoppor-tunitycostoftimeinsearchingforlowerprices(so-calledshoe-leathercosts),plusothercostsassociatedwithobtainingpricequotesfromcompeting rms(suchastheincrementalcostofthepostagestampsorphonecallsusedinacquiringpriceinformationfrom rms).Consumersintheseenvironmentsweighthecostofobtaininganadditionalpricequoteagainsttheexpectedbene tsofsearchinganadditional rm.AswediscussinSection2.1,equilibriumpricedispersioncanarisein2Asecondapproachdeemphasizesthemarginalsearchcostasasourceforpricedispersion.Instead,consumersaccesspriceinformationbyconsultinganinfor
mationclearinghouse(e.g.,anewspape
mationclearinghouse(e.g.,anewspaperoranInternetpricecomparisonsite);e.g.SalopandStiglitz(1977),Shilony(1977),Rosenthal(1980),Varian(1980),Narasimhan(1988),Spulber(1995),BayeandMorgan(2001),Baye,Morgan,andScholten(2004a).1Thedistinguishingfeatureofclearinghousemodelsisthatasubsetofconsumersgainaccesstoalistofpriceschargedbyall rmsandpurchaseatthelowestlistedprice.Intheearliestofthesemodels,equilibriumpricedispersionstemsfromexanteheterogeneitiesinconsumersor rms.Forexample,intheVarianandSalop-Stiglitzmodels,someconsumerschoosetoaccesstheclearinghousetoobtainpriceinformation,whileothersdonot.InShilony,Rosenthal,andNarasimhan,someconsumersareloyaltoaparticular rm(andthuswillbuyfromitevenifitdoesnotchargethelowestprice),whileotherconsumersareshoppersandonlypurchasefromthe rmchargingthelowestprice.Spulber(1995)showsthatequilibriumpricedispersionarisesevenwhenallconsumerscancostlesslyaccesstheclearinghouseprovidedeach rmisprivatelyinformedaboutitsmarginalcost.BayeandMorgan(2001)o¤eraclearinghousemodelthatendogenizesnotonlythedecisionsof rmsandconsumerstoutilizetheinformationclearinghouse(inthepreviousclearinghousemodels, rmslistingdecisionsareexogenous),butalsothefeeschargedbytheowneroftheclearinghouse(theinformationgatekeeper)toconsumersand rmswhowishtoaccessortransmitpriceinformation.Theyshowthatadispersedpriceequilibriumexistsevenintheabsenceofanyexanteheterogeneitiesinconsumersor rms.Inthissection,weprovideanoverviewofthekeyfeaturesandideasunderlyingtheseliteratures.2.1Search-TheoreticModelsofPriceDispersionWebeginwithanoverviewofsearch-theoreticapproachestoequilibriumpricedispersion.Theearlyliteraturestressestheideathat,whenconsumerssearchforpriceinformationandsearchiscostly, rmswillchargedi¤erentpricesinthemarket.Therearetwobasicsortsofmodelsused:Modelswith xedsamplesizesearchandmodelswheresearchissequential.Wewilldiscusseachoftheseinturn.1Athirdapproachdeemphasizesconsumersearchandmainlyfocusesonwhetherpricedispersioncanarisewhenconsumerspassivelyobtainpriceinformationdirectlyfrom rms(asindirectmailadvertisements);cf.Butters(1977),GrossmanandShapiro(1984),Stegeman(1991),RobertandStahl(1993),McAfee(1994),andStahl(1994).Arelatedmarketingliteratureexaminessimilarissues,rangingfromloyaltyandpricepromotionstrategiestochannelconictsandtheInternet;seeLalandVillas-Boas(1998),LalandSarvary(1999),Raju,Srinivasan,andLal(1990),andRao,ArjunjiandMurthi(1995).4the rmo¤eringthelowestprice;and3.Thedistributionof rmspricesisgivenbyanexogenousnon-degeneratecdfF(p)onp;p.Stiglerassumesthataconsumerchoosesa xedsamplesize,n,tominimizetheexpectedtotalcost(expectedpurchasecostplussearchcost)ofpurchasingKunitsoftheproduct
:E[C]=KEhp(n)mini+cnwhereEhp(n)mini=E[mi
:E[C]=KEhp(n)mini+cnwhereEhp(n)mini=E[minfp1;p2;:::;png];thatis,theexpectedlowestpricequoteobtainedfromndrawsfromF:SincethedistributionofthelowestofndrawsisF(n)min(p)=1[1F(p)]n;E[C]=KZpppdF(n)min(p)+cn=K"p+Zpp[1F(p)]ndp#+cnwherethesecondequalityobtainsfromintegrationbyparts.Noticethattheterminsquarebracketsreectstheexpectedpurchaseprice,whichisadecreasingfunctionofthesamplesize,n.However,sinceeachadditionalpriceobservationcostsc0toobtain,anoptimizingconsumerwillchoosetosearcha nitenumberoftimes,n,andthuswillgenerallystopshortofobtainingthebestpricepinthemarket.ThedistributionoftransactionpricesisthedistributionofthelowestofndrawsfromF;thatis,F(n)min(p)=1(1F(p))nFromthis,Stiglerconcludesthatdispersioninbothpostedpricesandtransactionspricesarisesasaconsequenceofcostlysearch.Howdotransactionspricesandsearchintensityrelatetothequantityoftheitembeingpur-chased(orequivalently,tothefrequencyofpurchases)?4Stiglersmodelo¤erssharppredictionsinthisdimension.Notethattheexpectedbene ttoaconsumerwhoincreaseshersamplesizefromn1tonisEhB(n)i=Ehp(n1)miniEhp(n)miniK;(1)4Kmayberelatedtopurchasefrequencyasfollows.SupposepricesarevalidforTperiods,andtheconsumerwishestobuyoneuniteverytTperiods;thatis,trepresentsaconsumerspurchasefrequency.ThenthetotalnumberofunitspurchasedduringtheTperiodsisKT=t:Thus,anincreaseinpurchasefrequency(t)isformallyequivalenttoanincreaseinKinthemodelabove.6SinceGisameanpreservingspreadofF;thereexistsauniqueinteriorpointu=F(EF[P])suchthatF1(u)=G1(u):Further,foralluu;F1(u)G1(u)-427;0andforallu-427;u;F1(u)G1(u)0:Thus=n Zu0F1(u)G1(u)(1u)n1du+Z1uF1(u)G1(u)(1u)n1du!Next,noticethat(1u)n1isstrictlydecreasinginu;hence]TJ/;༕ ;.9; T; 11;.515; 0 T; [00;n Zu0F1(u)G1(u)(1u)n1du+Z1uF1(u)G1(u)(1u)n1du!=n(1u)n1Z10F1(u)G1(u)du=0wherethelastequalityfollowsfromthefactthatFandGhavethesamemean.Proposition2SupposethatanoptimizingconsumerobtainsmorethanonepricequotewhenpricesaredistributedaccordingtoF,andthatpricedistributionGisameanpreservingspreadofF.ThentheconsumersexpectedtotalcostsunderGarestrictlylessthanthoseunderF:Proof.Supposethat,underF;theoptimalnumberofsearchesisn:ThentheconsumersexpectedtotalcostunderFisE[CF]=EFhp(n)miniKcnEGhp(n)miniKcnE[CG]wherethestrictinequalityfollowsfromProposition1,andtheweakinequalityfollowsfromthefactthatnsearchesmaynotbeoptimalunderthedistributionG:At rstblush,itmightse
emsurprisingthatconsumersengagedin
emsurprisingthatconsumersengagedin xedsamplesearchpayloweraveragepricesandhavelowerexpectedtotalcostsinenvironmentswherepricesaremoredispersed.Theintuition,however,isclear:Inenvironmentswherepricesaremoredispersed,theprospectsforpriceimprovementfromsearcharehigherbecausethelefttailofthepricedistributionthepartofthedistributionwherebargainsaretobefoundbecomesthickeraspricesbecomemoredispersed.84.Aconsumerwhoischargedthemonopolypriceearnssurplussu¢cienttocoverthecostofobtainingasinglepricequote;thatisv(p)c:Inthisenvironment,all rmspostthemonopolypriceandconsumersvisitonlyonestore,purchaseatthepostedpricep;andobtainsurplusv(p)c0.Giventhestoppingruleofconsumers,each rmsbestresponseistochargethemonopolyprice;giventhatall rmschargep,itisoptimalforeachconsumertosearchonlyonce.Toseethatthisistheuniqueequilibriuminundominatedstrategies,supposetothecontrarythatthereisanequilibriuminwhichsome rmpostedapricebelowthemonopolyprice(clearly,pricingabovethemonopolypriceisadominatedstrategy).Letp0bethelowestsuchpostedprice.A rmpostingthelowestpricecouldpro tablydeviatebyraisingitspricetothelowerofporp0+c:Anyconsumervisitingthat rmwouldstillrationallybuyfromitsincethemarginalbene tofanadditionalsearchissmallerthancthemarginalcostofanadditionalsearch.Thus,sucha rmwillnotloseanycustomersbythisstrategyandwillraiseitsearningsoneachofthesecustomers.TheDiamondparadoxisstriking:eventhoughthereisacontinuumofidentical rmscompetinginthemodelatextbookconditionforperfectcompetitioninthepresenceofanysearchfrictionswhatsoeverthemonopolypriceistheequilibrium.RothschildscriticismoftheStiglermodel,alongwiththeDiamondparadox,spawnedseveraldecadesofresearchintowhethercostlysearchcouldpossiblygenerateequilibriumpricedispersionasituationwhereconsumersareoptimallygatheringinformationgivenadistributionofprices,andwherethedistributionofpricesoverwhichconsumersaresearchingisgeneratedbyoptimal(pro t-maximizing)decisionsof rms.2.1.3TheReinganumModelandOptimalSequentialSearchReinganum(1979)wasamongthe rsttoshowthatequilibriumpricedispersioncanariseinasequentialsearchsettingwithoptimizingconsumersand rms.Reinganumsresultmaybeseeninthefollowingspecialcaseofourenvironmentwhere:1.Consumershaveidenticaldemandsgivenbyv0(p)=q(p)=Kp",where"1andK]TJ/;༣ ;.9; T; 19;.633; 0 T; [00;0;2.Consumersengageinoptimalsequentialsearch;3.FirmshaveheterogeneousmarginalcostsdescribedbytheatomlessdistributionG(m)on[m;m];10Case3.h(p)0:Thentheconsumersoptimalstrategyistosearchuntilsheobtainsapricequoteatorbelowthereservationprice,r;wherersolvesh(r)=Zrp(v(p)v(r))dF(p)c=0(4)Equation(4)representsa
priceatwhichaconsumerisexactlyindi¤eren
priceatwhichaconsumerisexactlyindi¤erentbetweenbuyingandmakinganadditionalsearch.Toseethatsuchapriceisuniquelyde nedbythisequation,noticethathp=c0,h(p)0,andh0(z)=B0(z)0:Aconsumerwhoobservesapricethatexceedsrwilloptimallyrejectthatpriceinfavorofcontinuedsearch,whileaconsumerwhoobservesapricebelowrwilloptimallyacceptthatpriceandstopsearching.Case1isclearlynoteconomicallyinterestingasitleadstotheabsenceofanymarketfortheproductinthe rstplace.Case2ariseswhentheexpectedutilityofpurchasingtheproductexceedsthecostofaninitialsearch,butthedistributionofpricesissu¢cientlytightrelativetosearchcoststomakeadditionalsearchessuboptimal.Mostoftheexistingsearchliterature,includingReinganum,restrictsattentiontoCase3,asweshalldohereafter.Thereservationpricede nedinequation(4)hasseveralinterestingcomparativestaticprop-erties.Totallydi¤erentiatingequation(4)withrespecttorandc;andusingequation(3)revealsthatdrdc=1q(r)F(r)=1Kr"F(r)0Thus,anincreaseinsearchcostsleadstoahigherreservationprice:Otherthingsequal,therangeofacceptablepricesisgreaterforproductswithhighersearchcosts.Notethat,forthespecialcasewhenq(r)=1;dr=dc=1=F(r)1:Inthiscase,aoneunitincreaseinsearchcostsincreasestherangeofacceptablepricesbymorethanoneunitthatis,thereisamagni catione¤ectofincreasesinsearchcosts.6ReinganumavoidsRothschildscriticismandtheDiamondparadoxbyintroducing rmcostheterogeneities.Sinceeach rmjdi¤ersinitsmarginalcosts,mj;priceswilldi¤eracross rmsevenwhentheypriceasmonopolists.Supposethatafraction01of rmspriceaboverandrecallthatthereareconsumersper rm.Arepresentative rmsexpectedpro twhenitpricesatpjis:Ej=8:(pjmj)q(pj)1ifpjr0ifpjr6Ingeneral,theremaybeeitheramagni cationoranattenuatione¤ectofaoneunitincreaseinthecostofsearch.12wherethelastequalityfollowsfromthefactthatristheoptimalreservationpricewhenconsumersfacethepricedistribution^F:Inshort,Reinganumsassumptionsofdownwardslopingdemandandcostheterogeneitygiverisetoanequilibriumofpricedispersionwithoptimizingconsumersand rms.Notethatdownwardslopingdemandandcostheterogeneitiestogetherplayacriticalroleingeneratingequilibriumpricedispersioninthisenvironment.Toseethatbothassumptionsarerequired,suppose rstthatcostsareheterogeneousbutthateachconsumerwishedtopurchaseoneunitoftheproduct,valuedatv.Inthiscase,givenareservationpriceofrv,all rmswould nditoptimaltopriceatr;andthedistributionofpriceswouldbedegenerate.Ofcourse,areservationpriceofrvisinconsistentwithoptimizingbehavioronthepartofconsumers.Toseethis,supposethataconsumerwasunexpectedlypresentedwithapricep0=r+;where
c:Accordingtothesearchstrategy,such
c:Accordingtothesearchstrategy,suchaconsumerissupposedtorejectthispriceandcontinuesearching;however,thebene tfromthisadditionalsearchislessthanthecost.Thus,aconsumershouldoptimallyacceptapricep0ratherthancontinuingtosearch.Theupshotofthisisthattheonlyequilibriumreservationpriceisr=v:However,thesearepreciselytheconditionsgiveninCase1;hencetheonlyequilibriumiswherenoconsumersshopatall.8Ifdemandweredownwardslopingbut rmshadidenticalmarginalcostsofm,each rmwouldhaveanincentivetosetthesameprice,p=minfr;m"=(1+")g;giventhereservationprice.ThisleadsbacktoCase2andoneobtainstheDiamondparadox:All rmschargethemonopolyprice,p=m"=(1+").Indeed,intheenvironmentabove,alimitingcasewherethedistributionofmarginalcostsconvergestoapointisexactlytheDiamondmodel.Finally,weexaminehowthevarianceinthedistributionofposted(andtransactions)pricesvarieswithsearchcosts.Notethat,inequilibrium,thevarianceinpricesisgivenby2=Ep2(E[p])2=Zrpp2dF(p) ZrppdF(p)!2=Zrpp2^f(p)dp+1^F(r)r2 Zrpp^f(p)dp+1^F(r)r!28CarlsonandMcAfee(1983)showthatifoneintroducesheterogeneitiesinconsumersearchcosts,adispersedpriceequilibriummayexistprovidedthatindividualconsumershaveperfectlyinelastic(incontrasttodownwardsloping)demand.142.1.5TheMacMinnModelInlightofthefactthatthereareinstancesinwhich xedsamplesizesearchisoptimal,onemaywonderwhetherequilibriumpricedispersioncanariseinsuchasetting.MacMinn(1980)providesana¢rmativeanswertothisquestion.MacMinnsresultmaybeseeninthefollowingspecialcaseofourenvironmentwhere:1.Consumershaveunitdemandwithvaluationv;2.Consumersengageinoptimal xedsamplesearch;and93.FirmshaveprivatelyobservedmarginalcostsdescribedbytheatomlessdistributionG(m)on[m;m],wheremv:Atthetime,MacMinnderivedequilibriumpricingbysolvingasetofdi¤erentialequationsunderthespecialcasewhereGisuniformlydistributed.However,subsequenttohispaper,akey ndingofauctiontheory,theRevenueEquivalenceTheorem(Myerson,1981)wasdeveloped.10Usingtherevenueequivalencetheorem,wecangeneralizeMacMinnsresultstoarbitrarycostdistributions.Toseethis,noticethatwhenconsumersoptimallyengageina xedsamplesearchconsistingofn rms,each rme¤ectivelycompeteswithn1other rmstoselloneunitoftheproduct.Ofthesen rms,the rmpostingthelowestpricewinstheauction.Usingtherevenueequivalencetheorem,onecanshowthattheexpectedrevenuestoa rmwithmarginalcostminanyauctionwherethe rmchargingthelowestpricealwayswinsandthe rmwiththehighestmarginalcostearnszerosurplusisR(m)=m(1G(m))n1+Zmm(1G(t))n1dt(5)IntheMacMinnmodel,expectedrevenuesaresimplya rmspostedprice,p(m),multipliedbytheprobabilityitchargesthelowestprice,which,inequilibrium,is(1G
(m))n1:UsingthefactthatR(m)=p(m)
(m))n1:UsingthefactthatR(m)=p(m)(1G(m))n1,substitutingintoequation(5);andsolvingforp(m)yieldstheequilibriumpricingstrategyofa rmwithmarginalcostmwhenconsumerssamplen rms:p(m)=m+Zmm1G(t)1G(m)n1dt(6)9MacMinnalsoprovidesaversionofthemodelthatisvalidforoptimalsequentialsearch.10SeeKlemperer(1999)foranon-technicalsurveyofauctiontheoryincludingtherevenueequivalencetheorem.McAfeeandMcMillan(1988)establishesanequivalencebetweensearchandauctionsinamechanismdesigncontext.16notethatasthesamplesizeincreases,thevarianceinequilibriumpricesincreases.Thisimpliesthat,takingintoaccounttheinteractionbetweenconsumersand rmsinthis xed-samplesizesearchmodel,dispersionvariesinverselywithsearchcosts.Conclusion2IntheMacMinnmodel,areductioninsearchcostsincreasesthevarianceofequi-libriumprices.ThisconclusionisincontrasttoConclusion1,wherepreciselytheoppositeimplicationisobtainedintheReinganumsequentialsearchmodel.Thishighlightsanimportantfeatureofsearch-theoreticmodelsofpricedispersion:Dependingonthemodel,areductioninsearchcostsmaybeassociatedwithhigherorlowerlevelsofpricedispersion.IntheReinganummodel,areductioninsearchcostsreducesthereservationpriceofconsumersandthusinducesmarginalhigh-cost rmstoreducetheirpricesfromtheirmonopolypricetothereservationprice.Sincethemonopolypricesoflow-cost rmsarebelowthereservationprice,theirpricesremainunchanged;lowersearchcoststhusreducetherangeofprices.IntheMacMinnmodel,lowersearchcostsinduceconsumerstosamplemore rmsbeforepurchasingine¤ect,each rmcompeteswithmorerivals.Asaconsequence,theoptimalamountofbidshading(pricingabovemarginalcost)isreduced,thusincreasingthelevelofpricedispersion.2.1.6TheBurdettandJuddModelBurdettandJudd(1983)werethe rsttoshowthatequilibriumpricedispersioncanariseinasearch-theoreticmodelwithexanteidenticalconsumersand rms.11BurdettandJuddsmainresultmaybeseeninthefollowingspecialcaseofourenvironmentwhere:1.Consumershaveunitdemanduptoapricev;2.Consumersengageinoptimal xedsamplesearch;123.Each rmhasconstantmarginalcost,m;andwouldoptimallychargeallconsumerstheuniquemonopolyprice,p=v;and4.Aconsumerwhoischargedthemonopolypriceearnssurplussu¢cienttocoverthecostof11JanssenandMoraga-González(2004)provideanoligopolisticversionoftheBurdettandJuddmodel.12BurdettandJuddalsoprovideaversionofthemodelthatisvalidunderoptimalsequentialsearch.18followsthataconsumermustbeindi¤erentbetweenobtainingonepricequoteandobtainingtwopricequotes.Thatis,inanydispersedpriceequilibriumEhB(1)iEhB(2)i=cEhB(3)i:::EhB(n)i:Thus,inanydispersedpriceequilibrium,1;20whilei=0foralli2:Let1=and2=1:Wearenowinapositiontocharacterizea
natomlessdispersedpriceequilibrium.First
natomlessdispersedpriceequilibrium.First,notethatsince2(0;1);thereisapositiveprobabilitythata rmfacesnocompetitionwhenitsetsitsprice.Thus,if rmichargesthemonopolyprice,itearnsexpectedpro tsofE[ijpi=v]=(vm)Incontrast,a rmchoosingsomelowerpricewinswhenitspriceisbelowthatoftheother rmaconsumerhassampled.Thus,if rmichargesapricepiv;itearnsexpectedpro tsofE[ijpiv]=(pim)(+(1)(1F(pi)))Thus,foragivendistributionofsearches,equilibriumpricedispersionrequiresthatthedistributionof rmprices,F();satis es+(1)(1F(p))=(vm)(pm)orF(p)=1(vp)(pm)1(10)whichisawell-behavedatomlesscumulativedistributionhavingsupport[m+(vm);v].Finally,itremainstodetermineanequilibriumvalueof:Sinceeachconsumermustbeindif-ferentbetweensearchingoneortwo rms,EhB(2)i=cNoticethat,when=0or=1;EB(2)=0whileEB(2)0forall2(0;1):BurdettandJuddshowthatEB(2)isquasi-concave;thus,whencissu¢cientlylow,therearegenericallytwodispersedpriceequilibriaoneinvolvingarelativelyhighfractionofconsumersmakingtwosearches,theotherwitharelativelylowfractionofconsumers.1414Thereisanon-dispersedpriceequilibriumwhereallconsumerssearchonceandall rmschargethemonopolyprice.20withamaximalwillingnesstopayofvm:15Ofthese,amass,S0,oftheconsumersareprice-sensitiveshoppers.Theseconsumers rstconsulttheclearinghouseandbuyatthelowestpricelistedthereprovidedthispricedoesnotexceedv.Ifnopricesareadvertisedattheclearinghouseoralllistedpricesexceedv,thenashoppervisitsoneofthe rmsatrandomandpurchasesifitspricedoesnotexceedv.AmassL0ofconsumersper rmpurchasefromthat rmifitspricedoesnotexceedv.Otherwise,theydonotbuytheproductatall.ItcanbeshownthatifL0or0,equilibriumpricedispersionarisesinthegeneralmodelprovidedofcoursethatisnotsolargethat rmsrefusetolistpricesattheclearinghouse.Moreprecisely,Proposition3Let0n1n(vm)S.Then,inasymmetricequilibriumofthegeneralclearinghousemodel:1.Each rmlistsitspriceattheclearinghousewithprobability=1nn1(vm)S1n1:2.Ifa rmlistsitspriceattheclearinghouse,itchargesapricedrawnfromthedistributionF(p)=10@1nn1+(vp)L(pm)S1n11Aon[p0;v];wherep0=m+(vm)LL+S+nn1L+S:3.Ifa rmdoesnotlistitspriceattheclearinghouse,itchargesapriceequaltov:4.Each rmearnsequilibriumexpectedpro tsequaltoE=(vm)L+1n1Proof.First,observethatifa rmdoesnotlistitspriceattheclearinghouse,itisadominantstrategytochargeapriceofv:Next,noticethat
11;2(0;1]whenevern(n1)(vm)
11;2(0;1]whenevern(n1)(vm)S1:15BayeandMorgan(2001)consideranenvironmentwithdownwardslopingdemand.221.Itiscostlessfor rmstolistpricesontheclearinghouse:=0and;2.Each rmhasapositivemassofloyalconsumers:L0:Since=0;itfollowsfromProposition3that=1;thatis,allofthen rmsadvertisetheirpriceswithprobabilityone.UsingthisfactandProposition3,theequilibriumdistributionofpricesisF(p)=1(vp)(pm)LS1n1on[p0;v](11)wherep0=m+(vm)LL+SThepricedispersionarisingintheRosenthalmodelstemsfromexogenousdi¤erencesintheprefer-encesofconsumers.Whileshoppersviewallproductsasidenticalandpurchaseatthelowestlistedprice,each rmisendowedwithastockofLloyals.Theequilibriumpricedispersionarisesoutofthetensioncreatedbythesetwotypesofconsumers.Firmswishtochargevtoextractmaximalpro tsfromtheloyalsegment,butifall rmsdidsoa rmcouldslightlyundercutthispriceandgainalloftheshoppers.OnemightimaginethatthisundercuttingargumentwouldleadtotheBertrandoutcome.However,oncepricesgetsu¢cientlylow,a rmisbettero¤simplychargingvandgivinguponattractingshoppers.Thus,theonlyequilibriumisinmixedstrategies rmsrandomizetheirprices,sometimespricingrelativelylowtoattractshoppersandothertimespricingfairlyhightomaintainmarginsonloyals.Itisinterestingtoexaminetheequilibriumtransactionspricesinthemarket.Loyalcustomersexpecttopaytheaveragepricechargedby rms:E[p]=Zvp0pdF(p)whileshoppersexpecttopaythelowestofndrawsfromF(p);thatis,theexpectedtransactionpricepaidbyshoppersisEhp(n)mini=Zvp0pdF(n)min(p)whereF(n)min(p)isthecdfassociatedwiththelowestofndrawsfromF:Howdotransactionspricesvarywiththenumberofcompeting rms?Rosenthalsstrikingresultisthat,asthenumberofcompeting rmsincreases,theexpectedtransactionspricespaidbyallconsumersgoup.Asweshallseebelow,theresulthingesonRosenthalsassumptionthatentry242.ThetotalmeasureofuninformedconsumerslackingaccesstotheclearinghouseisU0;hence,each rmisvisitedbyL=Unoftheseconsumers.Again,since=0;itfollowsthat=1andhencealln rmsadvertisetheirpricesattheclearinghouse:UsingthisfactandsettingL=U=ninProposition3,theequilibriumdistributionofpricesisF(p)=1 (vp)(pm)UnS!1n1on[p0;v]wherep0=m+(vm)UnUn+SThefactthatthisatomlessdistributionofpricesexistswheneverthereisanexogenousfractionofconsumerswhodonotutilizetheclearinghouseraisestheobviousquestion:Canthisequilibriumpersistwhenconsumersaremakingoptimaldecisions?Varianshowsthattheanswertothisquestionisyesprovideddi¤erentconsumershavedi¤erentcostsofaccessingtheclearinghouse.Theeasiestwaytoseethisistonotethatthevalueofinformationprovidedbytheclearinghouseisthedi¤erenceintheexpectedpricepaidbythoseaccessingtheclearinghouse,Ehp(n)mini;a
ndthosenot,E[p];thatis;VOI(n)=E[p]Eh
ndthosenot,E[p];thatis;VOI(n)=E[p]Ehp(n)mini(12)whereVOIdenotesthevalueof(price)informationcontainedattheclearinghouse.Supposeconsumersfaceacostofaccessingtheinformationprovidedbytheclearinghouse.Notethatthiscostisessentiallya xedcostofgainingaccesstotheentirelistofprices,notaperpricecostasinthesearch-theoreticmodelsconsideredabove.VarianassumesthatthecosttotypeSandLconsumersofaccessingtheclearinghouseisSandL,withSL.ThenprovidedSVOI(n)LtypeSconsumerswilloptimallyutilizetheclearinghousewhilethetypeLconsumerswillnot.Inshort,ifdi¤erentconsumershavedi¤erentcostsofaccessingtheclearinghouse,thereexistsanequilibriumofpricedispersionwithoptimizingconsumersand rms.Insuchanequilibrium,informedconsumerspayloweraveragepricesthanuninformedconsumers.Itisimportanttoemphasizethat,whenoneendogenizesconsumersdecisionstobecomein-formedintheVarianmodel,thelevelofpricedispersionisnotamonotonicfunctionofconsumersinformationcosts.Wheninformationcostsaresu¢cientlyhigh,noconsumerschoosetobecomeinformed,andall rmschargethemonopolyprice,v.Whenconsumersinformationcostsarezero,allconsumerschoosetobecomeinformed,andall rmspriceatmarginalcostinasymmetric26advertisetheirpricesandforconsumerstogainaccesstothelistofpricespostedattheclear-inghouse.Forexample,newspaperscharge rmsfeestoadvertisetheirpricesandmaychoosetochargeconsumerssubscriptionfeestoaccessanypostedinformation.Thesameistrueofmanyonlineenvironments.Moreover,theclearinghouseisitselfaneconomicagent,andpresumablyhasanincentivetoendogenouslychooseadvertisingandsubscriptionfeestomaximizeitsownexpectedpro ts.Thus,BayeandMorganexaminetheexistenceofdispersedpriceequilibriainanenviron-mentwithoptimizingconsumers, rms,andamonopolygatekeeperwhocontrolsaccesstotheclearinghouse.Speci cally,BayeandMorganconsiderahomogeneousproductenvironmentwherenidentical,butgeographicallydistinct,marketsareeachservedbya(single)local rm.Distanceorothertransactioncostscreatebarrierssu¢cienttoprecludeconsumersinonemarketfrombuyingthisproductinanothermarket;thuseach rminalocalmarketisamonopolist.Nowimaginethatanentrepreneurcreatesaclearinghousetoserveallmarkets.IntheInternetage,onecanviewtheclearinghouseasavirtualmarketplacethroughitscreation,thegatekeeperexpandsbothconsumersand rmsopportunitiesforcommerce.Eachlocal rmnowhastheoptiontopaythegatekeeperanamounttopostapriceontheclearinghouseinordertogainaccesstogeographicallydisparateconsumers.Eachconsumernowhastheoptiontopaythegatekeeperanamounttoshopattheclearinghouseandtherebypurchasefrom rmsoutsidethelocalmarket.Themonopolygatekeeper rstsetsandtomaximizeitsownexpectedpro ts.Giventhesefees,pro tmaximizing rms
makepricingdecisionsanddeterminewhethero
makepricingdecisionsanddeterminewhetherornottoadvertisethemattheclearinghouse.Similarly,consumersoptimallydecidewhethertopaytoaccesstheclearinghouse.Followingthis,aconsumercansimplyclickhermousetoresearchpricesattheclearinghouse(ifsheisasubscriber),visitthelocal rm,orboth.Withthisinformationinhand,aconsumerdecideswhetherandfromwhomtopurchasethegood.BayeandMorganshowthatthegatekeepermaximizesitsexpectedpro tsbysettingsu¢-cientlylowthatallconsumerssubscribe,andcharging rmsstrictlypositivefeestoadvertisetheirprices.Thus,BayeandMorgansmainresultsmaybeseeninthefollowingspecialcaseofthegeneralclearinghousemodel:1.Thegatekeeperoptimallysetspositiveadvertisingfees:0and;2.Thegatekeeperoptimallysetssubscriptionfeessu¢cientlylowsuchthatallconsumersaccesstheclearinghouse;thatis,L=0:28Whydoesthegatekeeper nditoptimaltosetlow(possiblyzero)feesforconsumerswishingtoaccessinformation,butstrictlypositivefeesto rmswhowishtotransmitpriceinformation?BayeandMorganpointoutthatthisresultstemsfromafreeriderproblemontheconsumersideofthemarketthatisnotpresentonthe rmside.Recallthatthegatekeepercanonlyextractrentsequaltothevalueoftheoutsideoptionof rmsandconsumers.Foreachsideofthemarket,theoutsideoptionconsistsofthesurplusobtainablebynotutilizingtheclearinghouse.Asmoreconsumersaccessthesite,thenumberofconsumersstillshoppinglocallydwindlesandtheoutsideoptionfor rmsiseroded.Incontrast,asmore rmsutilizetheclearinghouse,vigorouspricecompetitionamongthese rmsreduceslistedpricesandleadstoamorevaluableoutsideoptiontoconsumersnotusingtheclearinghouse.Thus,tomaximizepro ts,thegatekeeperoptimallysubsidizesconsumerstoovercomethisfreeriderproblemwhilecapturingrentsfromthe rmsideofthemarket.Noanalogousfreeriderproblemarisesonthe rmside;indeedgreaterconsumerparticipationattheclearinghouseincreasesthefrequencywithwhich rmsparticipate(increases)andhencepermitsgreaterrentextractionfrom rms.2.2.4ModelswithAsymmetricConsumersIngeneral,littleisknownaboutthegeneralclearinghousemodelwithasymmetricconsumers.17However,forthespecialcaseoftwo rms,resultsareavailable.Hereweshowhowonecanadaptthegeneralclearinghousemodeltoaccountforasymmetriesinduopolymarkets.Supposetherearetwo rms(i=1;2)competinginthemarket.AmassL1ofcustomersareloyalto rm1whileL2customersareloyalto rm2whereL1L2.Proposition4Let012(vm)S.Then,inanasymmetricdispersedpriceequilibrium:1.Each rmlistsitspriceattheclearinghousewithprobability=12S(vm):2.Ifa rmlistsitspriceattheclearinghouse,itchargesapricedrawnfromthedistributionFi(p)=112+(vp)Lj(pm)Son[p0;1;v];wherep0;1=m+(vm)L1L1+S+2L1+S:17F
orspeci cclearinghousemodels,somere
orspeci cclearinghousemodels,someresultsareavailable.Forinstance,Baye,Kovenock,anddeVries(1992)characterizeallequilibriainaversionoftheVarianmodelinwhich rmshaveasymmetricnumbersofconsumers.30forintoequation(13)andsolvingforFj;oneobtainstheexpressionforthedistributionsofadvertisedpricesgivenintheproposition.Itisstraightforwardtoverifythatthisisawell-de nedcdfon[p0;1;v];andthatneither rmcangainbychargingapricepi=2[p0;1;v]:ThereareseveralnoteworthyfeaturesoftheequilibriumpricingandadvertisingstrategiesgiveninProposition4.First,when=0;both rmsadvertisepricesontheclearinghousewithprobabilityone.Narasimhan(1988)analyzesduopolycompetitionwhere rmshaveanasymmetricnumberofloyalcustomersundertheassumptionthatboth rmslistpricesattheclearinghousewithcertainty.ThemodelpresentedabovethustakesthemainpartofNarasimhansanalysisasaspecialcase.When0;itisinterestingtonotethatthepropensitytoadvertiseislessthanunityand,moresurprisingly,itisexactlythesameforboth rms.Thus,asymmetriesinthecustomerbaseof rmsneednotleadtoasymmetriesin rmpropensitiestoadvertise.Incontrast,the rmsdistributionsofadvertisedpricesdodependontheircustomerbases.Comparingthedistributionsofpricesforthetwo rms,one ndsthat:F1(p)F2(p)=1vp(pm)S[L1L2]0Thatis,the rmwithfewerloyalsisactuallylessaggressiveinitspricingstrategythanthe rmwithmoreloyals.Thelargertheasymmetryinloyals,thelargerthedi¤erenceintheaveragepriceschargedbythetwo rmsbutinanunexpecteddirection.Indeed,the rmwithmoreloyals( rm1)o¤ersthelowestprice,p0;1inthemarketwithstrictlypositiveprobability.18AsinBayeandMorgan(2001),onemayendogenizetheadvertisingfeebyallowingapro t-maximizinggatekeepertodeterminethelevelofthatmaximizesitsexpectedpro ts.19Forexample,iftheonlycostsare xedcosts(FC);theexpectedpro tsoftheclearinghousearesimplyitsexpectedadvertisingrevenuesminuscosts:E[]=2FC=212S(vm)FCEquatingthegatekeepersexpectedmarginalpro tstozeroandsolvingfortheoptimaladvertisingfeeyields=S(vm)4(14)18Since rm2sdistributionisatomless,atieatpricep0;1isazeroprobabilityevent.19Forsimplicity,weassumetheclearinghousemustsetthefeechargedtoconsumersforaccessat=0toinducethemalltoparticipate.Thisistypicallythecase,forexample,atonlinepricecomparisonsites.32ItisinterestingtocomparetheSpulbermodel,whichoccursintheclearinghouseframework,withthesearch-theoreticframeworkofMacMinn.Noticethat,whenthenumberofcompeting rmsinSpulber,n,isequaltotheoptimal xedsamplesizeforconsumersintheMacMinnmodel,n;theequilibriumdistributionofprices,equations(15)and(7),areidenticalinthetwomodels.Thatis,costhete
rogeneitiesaresu¢cienttogeneratepricedi
rogeneitiesaresu¢cienttogeneratepricedispersioninoligopolymodelswhereallconsumersobtaincompletepriceinformation,aswellasinmodelswhereacontinuumof rmscompetebuteachconsumeronlyobtainspricequotesfroma nitenumbernofthese rms.2.3BoundedRationalityModelsofPriceDispersionSeveralrecentpapershaveemphasizedthatboundedrationalitycanalsoleadtopricedispersion.TheideaistorelaxtheNashequilibriumassumptionwhichrequiresthateachdecisionmakerinthemarketischoosinganaction(beitapriceorasearchstrategy)thatisabestresponsetogivenactionsofothermarketparticipants.Twoequilibriumconceptsquantalresponseequilib-rium(McKelveyandPalfrey,1995)andepsilonequilibrium(Radner,1980)areparticularlyusefulbecausetheynestthestandardNashequilibriumconceptasaspecialcase.Inaquantalresponseequilibrium(QRE),thelikelihoodthataparticular rmsetsaspeci cpricedependsontheexpectedpro tsarisingfromthatprice(seeLopez-Acevedo,1997).A rmspriceisdeterminedbyastochasticdecisionrule,butpricesleadingtohigherexpectedpro tsaremorelikelytobecharged.Ofcourse,each rmsexpectedpro tsfromdi¤erentpricingdecisionsdependontheprobabilitydistributionsofotherplayersprices.AQRErequiresthatall rmsholdcorrectbeliefsabouttheprobabilitydistributionsofotherplayersactions.ThenondegeneratedistributionsofpricesresultinginaQREmaybeviewedasshocksto rmspro tfunctions.Alternatively,nondegeneratepricedistributionsmightstemfromdecisionerrorsby rms.SucherrorsmayarisefromlimitationsinmanagerscognitiveprocessingabilitiesorbugsindynamicpricingalgorithmsusedbyInternetretailers.Inan"-equilibrium,thepriceschargedbyeach rmaresuchthatno rmcangainmorethan"inadditionalpro tsbychangingitsprice.Suchanequilibriummayarisebecauseofcognitiveormotivationalconstraintsonthepartof rms.Forexample,ifitiscostlytoreprogramdynamicpricingalgorithms,managersmaynotbewillingtoincurtheseeconomicorpsychiccostswhentheresultinggainissmall(lessthan").Recently,BayeandMorgan(2004)appliedQREand"-equilibriumconceptstopricinggamesandshowedthatonlyalittleboundedrationalityisneededtogeneratethepatternsofprice34inge¤orts,di¤erentialservicequalities,orreputations)cancontributetoequilibriumpricedispersion,suchdi¤erencesarenotnecessaryforequilibriumpricedispersion.6.ThankstotheInternet,informationgatekeepersareplayinganincreasinglyimportantroleintheeconomy.Intheirattempttomaximizepro tsandenhancethevalueofinformationprovidedbytheirsites,informationgatekeepershaveanincentivetochargefeesfortheirservicesthatinduceequilibriumpricedispersion.7.Alittleboundedrationalitygoesalongwayinexplainingpricedispersion.3EmpiricalAnalysisofPriceDispersionWenowturntotheempiricalliteratureonpricedispersion.InSection3.1,wediscusssomeofthestrengthsandwe
aknessesofcommonlyusedmetricsformeasurin
aknessesofcommonlyusedmetricsformeasuringpricedispersioninonlineandoinemarkets.Section3.2providesanoverviewoftheempiricalliterature,andhighlightsempiricalevidencesuggestingthatinformationcosts(eitherontheconsumeror rmsideofthemarket)contributetopricedispersion;thatis,dispersionisnotpurelyanartifactofsubtleproductheterogeneities.3.1MeasuringPriceDispersionTheequilibriummodelsofpricedispersionpresentedaboveeachimplynon-degeneratedistributionsofprices,F(p);onsomeintervalp;p.Givensuchadistribution,astandardmeasureofdispersionisthevarianceinprices.Foreachmodelofequilibriumpricedispersion,thismeasurecanbedirectlycomputed.Forinstance,intheMacMinnmodel,if rmshaveuniformlydistributedmarginalcosts,thevarianceinpricesis2p=n1n2(mm)212Noticethatoneistheninapositiontotestcomparativestaticpredictionsofthemodelusingthismeasure.Inasimilarmanner,expressionsforthevarianceinpricesmaybederivedfromtheothermodelspreviouslypresented.Anumberofauthorsusethesamplevariancetomeasurepricedispersion(e.g.,Pratt,Wise,andZeckhauser(1979)andAncaraniandShankar(2004)).Theobviousadvantageisthatitusesallavailabledata.Adrawbackofthismeasureisapparentwhencomparingdispersionacrossproductsorovertime.Forinstance,supposethat,duringaninationaryperiod,themarginal36Understandardconditionstherewillexistauniquesymmetricequilibriumwhereall rmspriceatmarginalcost.Butinaddition,thereareasymmetricequilibriawheretwo rmspriceatmarginalcostandtheremainingn2 rmspricestrictlyabovemarginalcost.Thus,pricedispersioncanariseinaclassicalBertrandenvironment.Yet,theapparentpricedispersionisarguablynoteconomicallyrelevantbecausetheuniquetransactionspriceismarginalcost.Toremedythistheoreticaldefect,Baye,MorganandScholten(2004a)proposeameasurecalledthegap,whichtheyde netobethedi¤erencebetweenthetwolowestpricesinthemarket.Lettingp(n)2denotethesecondlowestpricerealizationfromndrawsfromF;the(sample)gapisde nedas21G(n)=p(n)2p(n)minTheclassicalBertrandmodel(aswellastextbookmodelsofperfectcompetition)impliesthatthegapbetweenthetwolowestpricesiszeroinanyequilibrium(symmetricorotherwise).Alloftheoligopolymodelsofpricedispersiondiscussedabove,incontrast,implyapositivegap.Anadditionalpropertyofthegapisthatitgivesgreaterweighttolowprices,which,intheabsenceofquantitydata,onemightreasonablyassumeleadtomoresalesthanhigherprices.Thekeydisadvantage,sharedbytherange,isthatitreliespurelyonextremevaluesofthedata.Hence,therangeandgaparemoresensitivetooutliersandotherformsofnoisethanmeasuresthatusealltheavailabledata,suchasthesamplevarianceandcoe¢cientofvariation.Inadditiontothesemeasures,thevalueofinformation(VOI)de nedearlierinequation(12)canalsobeusedasagaugeofdispersion.Thismeasure,whichissimplythedi¤erencebetweentheave
rageobservedpriceandthelowestobservedpri
rageobservedpriceandthelowestobservedprice,iszerointheabsenceofanypricedispersionbutotherwisepositive.Theprincipaladvantageofthismeasureofdispersionisthatithasaveryintuitiveinterpretation:Itsvalueindicatestheamountofmoneyaconsumersavesbypurchasingatthebestpriceratherthanfromarandomlyselected rminthemarket.3.2PriceDispersionintheFieldIfpricedispersionstemsfromfrictionsrelatedtotheacquisitionandtransmissionofinformation(asimpliedbythemodelsinSection2)ratherthansubtledi¤erencesin rmsservicelevels,observedlevelsofdispersionshouldsystematicallydependonenvironmentalfactorspresentinthemodels.21Aswiththerange,onecanperformcomparativestaticanalysesforanyofthetheoreticalmodelsusingtheexpectedgap,anditissometimesusefultonormalizethegapbydividingbythelowestprice.Inthisformulation,thegaprepresentsthedi¤erencebetweenthetwolowestpricesexpressedasapercentageofthelowestpricerealization.38iteminaconsumersoverallbudget,andthefrequencywithwhichanitemispurchased,aregoodproxiesforK.DispersionforCheapversusExpensiveItemsStigler(1961)providescasualevidenceinsupportofhis rsthypothesisthatdispersionislowerforitemsthataccountforalargeexpenditureshareofasearchersconsumptionbundle(expensiveitems)thanthosethataccountforasmallerexpenditureshare(cheapitems).Governmentcoalpurchasesareasmallpercentageoftheoverallgovernmentbudget,whileahouseholdsexpendituresonanautomobilecomprise(in1961aswellastoday)amuchlargerpercentageofitsoverallbudget.Stiglerobtaineddi¤erentsellerspricesfortwohomogeneousproductsanthracite-gradecoaltobesoldtothegovernment,andanautomobiletobesoldtoahousehold.Pricesforanthracitecoalrangedfrom$15.46to$18.92,withanaveragepriceof$16.90andastandarddeviationof$1.15.Pricesfortheautomobile(basedonwhatStiglercalledanaverageamountofhiggling)rangedfrom$2,350to$2,515,withanaveragepriceof$2,436andstandarddeviationof$42.Stiglersdatathustendtosupporthis rstconjecture:Ifonecalculatestheimpliedcoe¢cientofvariationbasedonStiglers gures,thecoe¢cientofvariationforcoal(whichmakesupasmallpercentageofthegovernmentsbudget)is14.7percent,whilethatforanautomobile(whichmakesupalargepercentageofahouseholdsbudget)is1.7percent.Pratt,WiseandZeckhauser(1979)observeasimilarpatterninacross-sectionofconsumerproductssoldinBostoninthe1970s.Theyobtainthefollowingregressionresultregressingthesample(log)standarddeviationofpricesforagivenitemonthesample(log)meanpriceforthesameitem.ln=1:517+0:892lnE[p](16)Straightforwardmanipulationofequation(16)revealsthata1percentincreaseinthemeanpriceofanitemdecreasesthecoe¢cientofvariationby10.8percent.Thus,thePratt,Wise,andZeckhauserdataalsosuggestthat,empirically,thecoe¢cientofvariationislowerformoreexp
ensiveitemsthancheaperitems.However,equa
ensiveitemsthancheaperitems.However,equation(16)alsohighlightsthattherelationshipdependscruciallyonthemeasureofpricedispersionused:Ifoneweretousethestandarddeviationtomeasurepricedispersion,equation(16)impliesthataonepercentincreaseinthemeanpriceofaproductleadstoa0.892percentincreaseindispersion,asmeasuredbythestandarddeviation.Anumberofotherauthorshavereportedsimilarpatternsinonlineandoinemarkets,bothintheUSandinEuropeforproductsrangingfromconsumersundries,electronicproducts,andgasoline;cf.Marvel(1976),CarlsonandPescatrice(1980),ClayandTay(2001),Scholtenand40dealingwiththeseproblems.OneimportantexampleisBrownandGoolsbee(2002).TheirstartingpointistheStahl(1989)modelofequilibriumpricedispersion,whichaswenotedinSection2,predictsthatpricedispersionisinitiallyanincreasingfunctionofthefractionofshopperswhoenjoyzerosearchcosts,butafterathreshold,isadecreasingfunctionofthefractionofshoppers.BrownandGoolsbeepointoutthattheStahlmodelcloselymatchesthemarketforterm-lifeinsuranceduringthe1992-1997period.ConsumerswhodidnothaveanInternetconnectionarguablyhadtosearchsequentiallytoobtainpricequotesfromdi¤erentinsuranceagents,whilethosewithInternetaccesscouldusewebsitessuchasQuickquote.comtocostlesslyidentifythecompanyo¤eringthelowestannualpremium.Intheirdata,variationinthefractionofshoppers(thosewhoresearchinsuranceonline)stemsnotonlyfromthegeneralriseinInternetpenetrationduringthe1990s,butmoreimportantly,fromvariationinthegrowthratesinInternetusageacrossdi¤erentgroupsofpolicyholders.BrownandGoolsbeeregressthestandarddeviationinresiduals(obtainedfromapriceregressionthatcontrolsforobservablecharacteristicsofpeopleandpolicytypes)onacubicfunctionoftheirproxyforthefractionofshoppers.ConsistentwiththepredictionoftheStahlmodel,pricedispersioninitiallyrisesasthefractionofshoppersincreases,butstartstodeclineoncethefractionofconsumersresearchinginsuranceonlineexceedsabout5percent.Asimilarapproachisimplicitinanumberofpapersthathavecomparedlevelsofdispersioninonlineversusoinemarkets(cf.BrynjolfssonandSmith,2000;CarltonandChevalier,2001;AncaraniandShankar,2004;andScholtenandSmith,2002.)Thebasicpremiseisthatsearchcostsarelowerinonline(searchentailsclicks)versusoinemarkets(searchentailstravelcosts).22Ingeneral,sincedi¤erentsearchmodelsmakedi¤erentpredictionsabouttheimpactofreductionsinsearchcostsonlevelsofpricedispersion,itisnottoosurprisingthatthe ndingsofthisliteraturearedecidedlymixed;forsomeproducts,dispersionislowerinonlinemarkets;forotherproducts,dispersionisactuallyhigheronline.2322Theviewthatonlinesearchiseithermoreprevalentorcheaperthanoinesearchisamatterofsomedebate;see,forinstance,AdamicandHuberman(2001),Johnson,Moe,Fader,Bellman,andLohse(2004).Bakos(1997)wasamongthe rsttoadvanceatheoreticalargumentthatwhenth
ecostofpriceinformationisclosetozero,equ
ecostofpriceinformationisclosetozero,equilibriumpriceisclosetomarginalcost.Morerecently,however,Harrington(2001)hasarguedthatBakosresultsareawed.Finally,theInternetitselfalsoo¤ersopportunitiesforobfuscation(seeEllisonandEllison(2004))orunobservedlackofinventories(seeArnold&Saliba(2002))thatcanraisesearchand/ortransactionscostsrelativetooinemarkets.23Onemayspeculatethatonceshippingcostsareaccountedfor,pricedispersiononlinevanishes.Thisisnotthecase;cf.Pan,RatchfordandShankar(2002);AncaraniandShankar(2004);Brynjolfsson,DickandSmith(2004);BrynjolfssonandSmith(2000);SmithandBrynjolfsson(2001);DinlersozandLi(2005).42aswesawintheVarianandRosenthalmodels.Thus,examiningtherelationshipbetweenthepricedispersionandthenumberofcompetingsellersnotonlyprovidesatestofwhetherinformationalfactorsplayaroleingeneratingobservedpricedispersion,butalsoinmakingdistinctionsamongthevarioustheorymodels.Forinstance,Baye,MorganandScholten(2004a)examinethetheoreticalandempiricalrela-tionshipbetweenthenumberofcompetitorsandlevelsofpricedispersioninclearinghousemodels.Theyshowthatthetheoreticalrelationshipbetweennumberofcompetitorsandthelevelofpricedispersioninclearinghousemodelsis,ingeneral,ambiguous,duetocompetingorderstatisticandstrategice¤ects.ThroughacalibrationdisplayedinFigure2,theyshowthattheimpactofthenumberofsellersonpricedispersiondependsonthevariantofthemodel.Asthe gureshows,intheVarianmodel(where rmsinformationtransmissioncostsdonotdrivepricedispersion),theexpectedgapbetweenthetwolowestpricesisinitiallyincreasinginthenumberofsellers,andthendeclines.Incontrast,intheBayeandMorganmodel(where rmsinformationtransmissioncostsarethemaindriverofpricedispersion),theexpectedgapismonotonicallydecreasinginthenumberof rms.Basedononlinedatafromapopularpricecomparisonsiteforconsumerelec-tronicsproducts,andcontrollingforotherfactorscontributingtopricedispersion,they ndaninverserelationbetweenthegapandthenumberofonlinesellers.ThisrelationshipisdepictedasthedottedobservedlineinFigure2.Asthe gurereveals,thenon-monotonicitypredictedbytheVarianmodel,aswellastherelativelyatrelationshipbetweenthegapandnumberof rmspredictedinthecalibratedversionoftheRosenthalmodel,isabsentinthedata.Speci cally,inmarketsservedbybetweentwoandfour rms,theaveragegap(asapercentageofthelowestprice)isabout14percent.Theaveragepercentagegapfallstoabout3percentinmarketswith vetoten rms,andislessthanonepercentinmarketswithmorethan10 rms.Morebroadly,severalempiricalpapershavesuggestedthattheamountofpricedispersionobservedinthemarketdependsonvariousmeasuresofthenumbersofcompetitors.Marvel(1976)reportsthatanincreaseinthenumberofcompetitors(measuredbytheln(HHI))reducestherangeinthepriceofgasoline.Barron,Tayl
orandUmbeck(2004)studythestructuraldeter
orandUmbeck(2004)studythestructuraldeterminantsofpricedispersionintheretailgasolineindustryinfourgeographiclocations,andprovideempiricalevidencethat,controllingforstation-levelcharacteristics,anincreaseinstationdensitydecreasesbothpricelevelsandpricedispersion.24BorensteinandRose(1994)investigatetherelationshipbetweendispersionamongairfaresandthenumberofcompetitorsorightdensity.They ndthat24SeealsoPngandReitman(1994).44alsosuggestthatheterogeneitieseitheracross rmsoracrossmarketsimpactpricedispersioninonlinemarkets(cf.Smith,BaileyandBrynjolfsson(1999);Clay,KrishnanandWol¤(2001);SmithandBrynjolfsson(2001);ChenandHitt(2002);ResnickandZeckhauser(2002);andBrynjolfsson,DickandSmith(2004)).Inallcases,however,evenaftercontrollingforvariousheterogeneities,economicallysigni cantlevelsofpricedispersionremain.Thereisalsoevidencethatonlinepricesexhibittemporalpricedispersion.Forinstance,Baye,MorganandScholten(2004b)examineturnoveroftheidentityofthelow-priceandhigh-price rmsusingadatasetconsistingof36popularconsumerelectronicsproductssoldovera19-monthperiod.They ndconsiderableevidenceformonth-to-monthchangesintheidentityofthelow-price rms,butsomeevidenceofpersistenceintheidentityofhigh-priced rms.Similarly,IyerandPazgal(2003)collectbi-weeklypricedataonmusicCDs,movievideosandbooksfrom vepricecomparisonsites:MySimon,BottomDollar,EvenBetter,BsillyandPricescanduringtheperiodApril-October2000and ndempiricalresultssuggestingthatnosingle rmconsistentlychargesthelowprice.Finally,Baye,MorganandScholten(2004a)examinetheconvergencehypothesisofpricedis-persionusingadatasetconsistingofoverfourmilliondailypriceobservationsforoveronethousandconsumerelectronicsproductssoldonapopularInternetpricecomparisonsiteoveraneightmonthperiod.Evenallowingforanonlinearrelationshipbetweenobservedpricedispersionandtime,they ndnoevidencefortheconvergencehypothesisinthismarketthelevelofpricedispersionre-mainedstableovertheperiod.3.3ConcludingRemarks:EmpiricsWeconcludewithfoursimpleobservations.1.AsisevidentfromthestudieshighlightedinTable1,pricedispersionisubiquitousandpersistent.Regardlessoftheparticularproduct(tinplatecansorPDAs),thevenueinwhichtheyaresold(onlineoroine,intheUSorabroad),orthetimeperiod(1901or2005),theinescapableconclusionfromtheempiricalliteratureisavalidationofStiglersandVariansinitialobservations:Informationremainsavaluableresource,andthelawofonepriceisstillnolawatall.2.Theoryisusefulforunderstandingdispersiondata,anddispersiondataisusefulfordiscrim-inatingamongalternativetheoreticalmodels.46ReferencesAdamic,L.A.andB.A.Huberman.2001.TheWebsHiddenOrder.Comm.ACM,44(9),55-59.Ancarani,F.andV.Shankar.2004.PriceLevelsandPriceDispersionWithinandAcrossMultipleRetailerTypes:FurtherEvidence
andExtension.JournaloftheAcademyof
andExtension.JournaloftheAcademyofMarketingScience,32(2),176-187.Arbatskaya,M.andM.R.Baye.Forthcoming.ArePricesStickyOnline?MarketStructureE¤ectsandAsymmetricResponsestoCostShocksinOnlineMortgageMarkets.InternationalJournalofIndustrialOrganization.Aalto-Setälä,V.2003.ExplainingPriceDispersionforHomogeneousGroceryProducts.JournalofAgriculturalandFoodIndustrialOrganization.1,Article9,1-14.Arnold,M.A.andC.Saliba.2002.PriceDispersioninOnlineMarkets:TheCaseofCollegeTextbooks.WorkingPaper.Axell,B.1977.SearchMarketEquilibrium.ScandinavianJournalofEconomics,79(1),20-40.Bailey,J.1998.ElectronicCommerce:PricesandConsumerIssuesforThreeProducts:Books,CompactDiscs,andSoftware.OrganizationEconomicsCo-OperationDevelopment,98(4).Bakos,Y.1997.ReducingBuyerSearchCosts:ImplicationsforElectronicMarketplaces.Man-agementScience,43(2),1676-1692.Barron,J.M.,B.A.TaylorandJ.R.Umbeck.2004.NumberofSellers,AveragePrices,andPriceDispersion.InternationalJournalofIndustrialOrganization,22,1041-1066.ReferencesBaye,M.R.,R.Gatti,P.Kattuman,andJ.Morgan.2005.EstimatingFirm-LevelDemandataPriceComparisonSite:AccountingforShoppersandtheNumberofCompetitors.WorkingPaper.Baye,M.R.andD.Kovenock.1994.HowtoSellaPickupTruck:BeatorPayAdvertisementsasFacilitatingDevices.InternationalJournalofIndustrialOrganization,12(1),21-33.Baye,M.R.,D.KovenockandC.deVries.1992.ItTakesTwotoTango:EquilibriainaModelofSales.GamesandEconomicBehavior,4,493-510.48Butters,G.R.1977.EquilibriumDistributionsofSalesandAdvertisingPrices.ReviewofEco-nomicStudies,44,465-491.Carlson,J.A.andR.P.McAfee.1983.DiscreteEquilibriumPriceDispersion.JournalofPoliticalEconomy,91(3),480-493.Carlson,J.A.andD.R.Pescatrice.1980.PersistentPriceDistributions.JournalofEconomicandBusiness,33,21-27.Carlton,D.andJ.Chevalier.2001.FreeRidingandSalesStrategiesfortheInternet.JournalofIndustrialEconomics,49(4),441-461.Chen,S.andL.M.Hitt.2002.MeasuringSwitchingCostsandtheDeterminantsofCustomerRetentioninInternet-EnabledBusiness:AStudyoftheOnlineBrokerageIndustry.InformationSystemsResearch,13(3).255-274.Chevalier,J.andA.Goolsbee.2003.MeasuringPricesandPriceCompetitionOnline:Ama-zon.comandBarnesandNoble.com.QuantitativeMarketingandEconomics,1,203-222.Clay,K.andC.H.Tay.2001.TheCross-CountryPriceDi¤erentialsintheOnlineTextbookMar-ket.WorkingPaper.Clay,K.,R.KrishnanandE.Wol¤.2001.PricesandPriceDispersionontheWeb:EvidencefromtheOnlineBookIndustry.JournalofIndustrialEconomics,49,521-539.Clay,K.,R.Krishnan,E.Wol¤andD.Fernandes.2003.RetailStrategiesontheWeb:PriceandNon-PriceCompetitionintheOnlineBookIndustry.JournalofIndustria
lEconomics,50(3),351-367.Clemons,E.I.Han
lEconomics,50(3),351-367.Clemons,E.I.HannandL.Hitt.2002.PriceDispersionandDi¤erentiationinOnlineTravel:AnEmpiricalInvestigation.ManagementScience,48,534-549.Dana,J.D.1994.LearninginanEquilibriumSearchModel.InternationalEconomicReview,35,745-771.Daughety,A.1992.AModelofSearchandShoppingbyHomogeneousConsumerswithoutPricePrecommitmentbyFirms,JournalofEconomicsandManagementStrategy1(3),455-473.Diamond,P.1971.AModelofPriceAdjustment.JournalofEconomicTheory,3,156-168.50Johnson,E.J.,W.Moe,P.Fader,S.BellmanandJ.Lohse.2004.OntheDepthandDynamicsofWorldWideWebShoppingBehavior.ManagementScience,50(3),299-308.Johnson,R.N.2002.SearchCosts,LagsandPricesatthePump.ReviewofIndustrialOrganiza-tion,20,33-50.Klemperer,P.1999.AuctionTheory:AGuidetotheLiterature.JournalofEconomicSurveys,13,227-286.Lach,S.2002.ExistenceandPersistenceofPriceDispersion:AnEmpiricalAnalysis.ReviewofEconomicsandStatistics,84(3),433-444.Lal,R.andM.Villas-Boas.1998.PricePromotionsandTradeDealswithMultiproductRetailers.ManagementScience,44(7),935-949.Lal,R.andM.Sarvary.1999.WhenandHowistheInternetLikelytoDecreasePriceCompeti-tion.MarketingScience,18(4),485-503.Lopez-Acevedo,G.1997.QuantalResponseEquilibriaforPostedO¤erMarkets.EstudiosEconómicos,12,95-131.MacMinn,R.D.1980.SearchandMarketEquilibrium.JournalofPoliticalEconomy,88(2),308-327.Marvel,H.P.1976.TheEconomicsofInformationandRetailGasolinePriceBehavior:AnEm-piricalAnalysis.JournalofPoliticalEconomy,84,1033-1080.McAfee,R.P.1994.EndogenousAvailability,Cartels,andMergerinanEquilibriumPriceDisper-sion.JournalofEconomicTheory,62,24-47.McAfee,R.P.1995.MultiproductEquilibriumPriceDispersion.JournalofEconomicTheory,67,83-105.McAfee,R.P.andJ.McMillan.1988.SearchMechanisms.JournalofEconomicTheory,44,99-123.McKelvey,R.andT.Palfrey1995.QuantalResponseEquilibriaforNormalFormGames.GamesandEconomicBehavior,10,6-38.52Reinganum,J.F.1979.ASimpleModelofEquilibriumPriceDispersion.JournalofPoliticalEconomy,87,851-858.Resnick,P.andR.Zeckhauser.2002.TrustAmongStrangersinInternetTransactions:EmpiricalAnalysisofeBaysReputationSystem.AdvancesinAppliedMicroeconomics,11,127-158.Robert,J.andD.O.Stahl.1993.InformativePriceAdvertisinginaSequentialSearchModel.Econometrica,61,657-686.Roberts,M.J.andD.Supina.2000.OutputPriceandMarkupDispersioninMicroData:TheRolesofProducerHeterogeneityandNoise.AdvancesinAppliedMicroeconomics,9,1-36..Rosenthal,R.W.1980.AModelinWhichanIncreaseintheNumberofSellersLeadstoaHigherPrice.Econometrica,48(6),1575-1580.Rothschild,M.1973.ModelsofMarketOrganizationwithImperfectInformation:ASu
rvey.JournalofPoliticalEconomy,81(
rvey.JournalofPoliticalEconomy,81(6)1283-1308.Rothschild,M.1974.SearchingfortheLowestPriceWhentheDistributionofPricesisUnknown.JournalofPoliticalEconomy,82(4),689-711Salop,S.C.andJ.E.Stiglitz.1977.BargainsandRipo¤s:AModelofMonopolisticallyCompetitivePriceDispersion,ReviewofEconomicStudies,44,493-510.Scholten,P.andA.Smith.2002.PriceDispersionThenandNow:EvidencefromRetailandE-tailMarkets.inMichaelR.Baye(ed.)inTheEconomicsoftheInternetandE-commerce,AdvancesinAppliedMicroeconomics,11,2002.Scott-Morton,F.,F.ZettelmeyerandJ.Silva-Risso.2001.InternetCarRetailing.JournalofIndustrialEconomics,49(4),501-519.Shilony,Y.1977.MixedPricinginOligopoly.JournalofEconomicTheory,14,373-388.Smith,M.D.,J.BaileyandE.Brynjolfsson.1999.UnderstandingDigitalMarkets:ReviewsandAssessmentsinUnderstandingtheDigitalEconomy,BrynjolfssonandKahin,eds.,MITPress,Cambridge,MA.Smith,M.D.andE.Brynjol¤son.2001.ConsumerDecision-MakingatanInternetShopbot:BrandStillMatters.JournalofIndustrialEconomics,49(4),541-558.54*2005 data through third quarter.Figure 1: Percentage of Articles Published in the American Economic Review, Journal of Political Economy, and Econometrica on Information, Search or Price DispersionSource: Social Science Citation Index, Keyword search for "Information OR Price Dispersion OR Search," and Authors' Calculations.5.00%10.00%1961-19651966-19701971-19751976-19801981-19851986-19901991-19951996-20002001-2005*0102030405060Percentage GapVarianRosenthalBaye-MorganTable 1a: Measures of Price Dispersion RepStudyData PeriodProduct MarketBailey (1998)1997Books 13.2%Standard DeviationBooks 10.4%Standard DeviationCompact Discs 17.6%Standard DeviationCompact Discs 11.0%Standard DeviationSoftware 7.1%Standard DeviationSoftware 8.1%Standard DeviationBorenstein and Rose (1994)1986U.S. Airline 0.018 - 0.416Gini coefficientCarlson and Pescatrice (1980)1976Consumer Sundries3.3% - 41.4%Coefficient of VariationEckard (2004)1901 - 2001Baking Powder, Sugar, Salt -- 1901 3.1% - 10.1%Coefficient of VariationBaking Powder, Sugar, Salt -- 2001 0.0% - 13.4%Coefficient of Variation1999Books $54.00 - $122.00RangeBooks $21.94 - $76.20Standard DeviationCompact Discs $20.00 - $40.00RangeCompact Discs $12.91 - $23.86Standard DeviationBooks (Sweden)$19.00 - $58.00RangeCompact Discs (Sweden)$21.00 - $46.00RangeLach (2002)1993 - 1996Refrigerator (Israel)4.9%Coefficient of VariationChicken, Flour, Coffee (Israel)11.4% - 19.7%Coefficient of VariationMarvel (1976)1964 - 1971Regular Gasoline $0.048RangeRegular Gasoline $0.015Standard DeviationPremium Gasoline $0.048RangePremium Gasoline $0.017Standard Deviation1975Various Products and Services 4.4% - 71.4%Coefficient of VariationVarious Products and Services 11.0% - 567.
0%RangeVarious Products and Services 7.2
0%RangeVarious Products and Services 7.2% - 200.0%Value of InformationRoberts and Supina (2000)1963 - 1987Wood Products 13.8% - 90.2%Coefficient of VariationFabrics 18.8% - 78.1%Coefficient of VariationCoffee 14.3% - 25.1%Coefficient of VariationReady-Mixed Concrete 13.2% - 37.2%Coefficient of VariationNewsprint 4.5% - 8.2%Coefficient of VariationGasoline 6.2% - 11.8%Coefficient of VariationTinplate Steel Cans 25.0% - 31.0%Coefficient of VariationPan Bread 26.0% - 49.6%Coefficient of VariationCorrugated Shipping Containers 21.8% - 39.6%Coefficient of VariationScholten and Smith (2002)1976 - 2000Consumer Sundries -- 1976 3.3% - 41.4%Coefficient of VariationConsumer Sundries -- 2000 1.6% - 42.0%Coefficient of VariationConsumer Sundries -- 2000 5.7% - 28.4%Coefficient of VariationSorensen (2000)1998Prescription Drugs $13.17RangePrescription Drugs 22.0%Coefficient of VariationStigler (1961)1953Anthracite Coal $3.46RangeAnthracite Coal $1.15Standard Deviation1959Identical Automobiles $165.00RangeIdentical Automobiles $42.00Standard DeviationVillas-Boas (1995)1985 - 1987Coffee 21.5%Coefficient of VariationConsumer Electronics (UK)7.0% - 54.9%RangeConsumer Electronics (Denmark)12.8% - 42.9%RangeConsumer Electronics (France)1.6% - 16.1%GapConsumer Electronics (Italy)3.6% - 13.7%GapConsumer Electronics (Netherlands)8.9% - 34.6%GapConsumer Electronics (Spain)3.7% - 18.0%GapConsumer Electronics (Sweden)5.9% - 15.6%GapConsumer Electronics (UK)2.5% - 14.5%GapConsumer Electronics (Denmark)3.6% - 31.9%GapHong and Shum (Forthcoming)2002Books $8.19 - 27.05RangeBooks 6.2% - 8.5%Coefficient of Variation2004Market for Keyboards $6.50 - $91.67RangeMarket for Keyboards 8.0% - 52.0%Coefficient of Variation2000Books 15.0%Coefficient of VariationCompact Discs 15.4%Coefficient of VariationDVDs 12.7%Coefficient of VariationPDAs 11.8%Coefficient of VariationSoftware 11.7%Coefficient of VariationConsumer Electronics 9.6%Coefficient of Variation2000 - 2003Consumer Electronics and Books 9.8% - 11.7%Coefficient of VariationBooks 33.3% - 48.9%RangeCompact Discs 22.2% - 51.0%RangeDVDs 30.7% - 43.7%RangeComputers 15.0% - 34.4%RangeSoftware 19.0% - 35.6%RangeConsumer Electronics 22.1% - 45.7%RangeSmith and Brynjolfsson (2001)1999Books 28.0% - 33.0%Value of InformationBooks $6.29 - $10.51Standard DeviationInformation,Search,andPriceDispersionMichaelR.BayeKelleySchoolofBusinessIndianaUniversityJohnMorganHaasSchoolofBusinessandDepartmentofEconomicsUniversityofCalifornia,BerkeleyPatrickScholtenBentleyCollegeThisDraft:October31,2005.Forthcoming,HandbookonEconomicsandInformationSystems(Elsevier,T.Hendershott,ed.).AbstractWeprovideauni edtreatmentofalternativemodelsofinformationacquisition/transmissionthathavebeenadvancedtorationalizepricedispersioninonlineandoinemarketsforhomo-geneousprodu
cts.Thesedi¤erentframeworkswhichi
cts.Thesedi¤erentframeworkswhichincludesequentialsearch, xedsam-plesearch,andclearinghousemodelsrevealthatreductionsin(ortheeliminationof)con-sumersearchcostsneednotreduce(oreliminate)pricedispersion.Ourtreatmenthighlightsadualitybetweensearch-theoreticandclearinghousemodelsofdispersion,andshowshowauction-theoretictoolsmaybeusedtosimplify(andevengeneralize)existingtheoreticalre-sults.Weconcludewithanoverviewoftheburgeoningempiricalliterature.Theempiricalevidencesuggeststhatpricedispersioninbothonlineandoinemarketsissizeable,pervasive,andpersistentanddoesnotpurelystemfromsubtledi¤erencesin rmsproductsorservices.WeoweaspecialdebttoMichaelRauhandFelixVárdyforprovidinguswithdetailedhand-writtencommentsonearlierdrafts.WealsothankVilleAalto-Setälä,FabioAncarani,MariaArbatskaya,JudyChevalier,KarenClay,WoodyEckard,SaraFisherEllison,XianjunGeng,RupertGatti,JoseMoragaGonzalez,JoeHarrington,TerryHendershott,GaneshIyer,MaartenJanssen,RonaldJohnson,KenJudd,RamayyaKrishnan,SolLach,RajivLal,PrestonMcAfee,XingPan,Je¤Perlo¤,IvanPng,RamRao,JenniferReinganum,NancyRose,VenkyShankar,JorgeSilva-Risso,MichaelSmith,AlanSorensen,DanSpulber,MarkStegeman,BeckTaylor,MiguelVillas-Boas,XiaolinXing,andRichardZeckhauserforencouragement,helpfulcomments,andsuggestions.Ofcourse,weareresponsibleforanyshortcomingsthatremaininthiso¤ering.1Information,Search,andPriceDispersionMichaelR.BayeKelleySchoolofBusinessIndianaUniversityJohnMorganHaasSchoolofBusinessandDepartmentofEconomicsUniversityofCalifornia,BerkeleyPatrickScholtenBentleyCollegeThisDraft:October31,2005.Forthcoming,HandbookonEconomicsandInformationSystems(Elsevier,T.Hendershott,ed.).AbstractWeprovideauni edtreatmentofalternativemodelsofinformationacquisition/transmissionthathavebeenadvancedtorationalizepricedispersioninonlineandoinemarketsforhomo-geneousproducts.Thesedi¤erentframeworkswhichincludesequentialsearch, xedsam-plesearch,andclearinghousemodelsrevealthatreductionsin(ortheeliminationof)con-sumersearchcostsneednotreduce(oreliminate)pricedispersion.Ourtreatmenthighlightsadualitybetweensearch-theoreticandclearinghousemodelsofdispersion,andshowshowauction-theoretictoolsmaybeusedtosimplify(andevengeneralize)existingtheoreticalre-sults.Weconcludewithanoverviewoftheburgeoningempiricalliterature.Theempiricalevidencesuggeststhatpricedispersioninbothonlineandoinemarketsissizeable,pervasive,andpersistentanddoesnotpurelystemfromsubtledi¤erencesin rmsproductsorservices.WeoweaspecialdebttoMichaelRauhandFelixVárdyforprovidinguswithdetailedhand-writtencommentsonearlierdrafts.WealsothankVilleAalto-Setälä,FabioAncarani,MariaArbatskaya,JudyChevalier,KarenClay,WoodyEckard,SaraFisherEllison,XianjunGeng,RupertGatti,JoseMoragaGonzalez
,JoeHarrington,TerryHendershott,GaneshIy
,JoeHarrington,TerryHendershott,GaneshIyer,MaartenJanssen,RonaldJohnson,KenJudd,RamayyaKrishnan,SolLach,RajivLal,PrestonMcAfee,XingPan,Je¤Perlo¤,IvanPng,RamRao,JenniferReinganum,NancyRose,VenkyShankar,JorgeSilva-Risso,MichaelSmith,AlanSorensen,DanSpulber,MarkStegeman,BeckTaylor,MiguelVillas-Boas,XiaolinXing,andRichardZeckhauserforencouragement,helpfulcomments,andsuggestions.Ofcourse,weareresponsibleforanyshortcomingsthatremaininthiso¤ering.theseenvironmentsunderavarietyofmarketconditionsandsearchstrategies(includingsequentialand xedsamplesearch).Whilemarginalsearchcostsareusefulinexplainingpricedispersioninsomemarkets,inmanyonlinemarketsincrementalsearchcostsareverylowandinsomecases,zero.Forexample,pricecomparisonsitesandshopbottechnologiescreateenvironmentswhereconsumersmayobtainalistofthepricesthatdi¤erentsellerschargeforthesameproduct.Despitethefactthatthisinformationisavailabletoconsumersinseconds,ultimatelyatthecostofasinglemouseclick,theoverwhelmingempirical ndingisthatevenintheseenvironments,pricedispersionispervasiveandsigni cantthelawofonepriceisegregiouslyviolatedonline.InSection2.2,weexamineanalternativelineoftheoreticalresearchwheremarginalsearchcostsarenotthekeydriverforpricedispersion.OurtheoreticalanalysisconcludesinSection2.3withadiscussionofalternativebehavioralrationalesforpricedispersion(includingboundedrationalityonthepartof rmsand/orconsumers).Section3providesamoredetailedoverviewofthegrowingempiricalliterature.AsonemightsuspectbasedonthetrendinFigure1andtheresearchsummarizedinTables1aand1b,mostempiricalstudiesofpricedispersionpost-datetheInternetandrelyononlinedata.Ourviewisthatthisismoreanartifactoftherelativeeasewithwhichdatamaybecollectedinonlinemarketsnotanindicationthatpricedispersionismoreimportant(ormoreprevalent)inonlinethanoinemarkets.Forthisreason,wehaveattemptedtoprovideabalancedtreatmentoftheliteraturesononlineandoinepricedispersion.Asweshallargue,theoverwhelmingconclusionofbothliteraturesisthatpricedispersionisnotpurelyanartifactofproductheterogeneities.2TheoreticalModelsofPriceDispersionThissectionpresentsalternativemodelsthathavebeenusedtorationalizethepricedispersionob-servedinbothoineandonlinemarkets.Oneapproachistoassumethatitiscostlyforconsumerstogatherinformationaboutprices.Inthesesearch-theoreticmodels,consumerssearchingforthebestpriceincurapositivecostofobtainingeachadditionalpricequote.RepresentativeexamplesincludeStigler(1961),Rothschild(1973),Reinganum(1979),MacMinn(1980),Braverman(1980),BurdettandJudd(1983),CarlsonandMcAfee(1983),Rob(1985),Stahl(1989,1996),Dana(1994),McAfee(1995),JanssenandMoraga-González(2004),aswellasJanssen,Moraga-González,andWildenbeest(2005).3theseenvironmentsunderavarietyofmarketconditionsandsearchstrategies(including
sequentialand xedsamplesearch).Whil
sequentialand xedsamplesearch).Whilemarginalsearchcostsareusefulinexplainingpricedispersioninsomemarkets,inmanyonlinemarketsincrementalsearchcostsareverylowandinsomecases,zero.Forexample,pricecomparisonsitesandshopbottechnologiescreateenvironmentswhereconsumersmayobtainalistofthepricesthatdi¤erentsellerschargeforthesameproduct.Despitethefactthatthisinformationisavailabletoconsumersinseconds,ultimatelyatthecostofasinglemouseclick,theoverwhelmingempirical ndingisthatevenintheseenvironments,pricedispersionispervasiveandsigni cantthelawofonepriceisegregiouslyviolatedonline.InSection2.2,weexamineanalternativelineoftheoreticalresearchwheremarginalsearchcostsarenotthekeydriverforpricedispersion.OurtheoreticalanalysisconcludesinSection2.3withadiscussionofalternativebehavioralrationalesforpricedispersion(includingboundedrationalityonthepartof rmsand/orconsumers).Section3providesamoredetailedoverviewofthegrowingempiricalliterature.AsonemightsuspectbasedonthetrendinFigure1andtheresearchsummarizedinTables1aand1b,mostempiricalstudiesofpricedispersionpost-datetheInternetandrelyononlinedata.Ourviewisthatthisismoreanartifactoftherelativeeasewithwhichdatamaybecollectedinonlinemarketsnotanindicationthatpricedispersionismoreimportant(ormoreprevalent)inonlinethanoinemarkets.Forthisreason,wehaveattemptedtoprovideabalancedtreatmentoftheliteraturesononlineandoinepricedispersion.Asweshallargue,theoverwhelmingconclusionofbothliteraturesisthatpricedispersionisnotpurelyanartifactofproductheterogeneities.2TheoreticalModelsofPriceDispersionThissectionpresentsalternativemodelsthathavebeenusedtorationalizethepricedispersionob-servedinbothoineandonlinemarkets.Oneapproachistoassumethatitiscostlyforconsumerstogatherinformationaboutprices.Inthesesearch-theoreticmodels,consumerssearchingforthebestpriceincurapositivecostofobtainingeachadditionalpricequote.RepresentativeexamplesincludeStigler(1961),Rothschild(1973),Reinganum(1979),MacMinn(1980),Braverman(1980),BurdettandJudd(1983),CarlsonandMcAfee(1983),Rob(1985),Stahl(1989,1996),Dana(1994),McAfee(1995),JanssenandMoraga-González(2004),aswellasJanssen,Moraga-González,andWildenbeest(2005).Thesearchmodelsconsideredinthissubsectionareallbasedonthefollowinggeneralenvi-ronment.Acontinuumofprice-setting rms(withunitmeasure)competeinamarketsellinganidentical(homogeneous)product.Firmshaveunlimitedcapacitytosupplythisproductatacon-stantmarginalcost,m:Acontinuumofconsumersisinterestedinpurchasingtheproduct.Letthemassofconsumersinthemarketbe;sothatthenumberofcustomersper rmis:Eachconsumerhasaquasi-linearutilityfunction,u(q)+y,whereqisthequantityofthehomogeneousproductandyisthequantityofsomenumerairegoodwhosepriceisnormalizedtobeunity.Thisimpliesthattheindirectutilityofac
onsumerwhopaysapricepperunitoftheproduct
onsumerwhopaysapricepperunitoftheproductandwhohasanincomeofMisV(p;M)=v(p)+Mwherev()isnonincreasinginp:ByRoysidentity,notethatthedemandfortheproductofrelevanceisq(p)v0(p).Toacquiretheproduct,aconsumermust rstobtainapricequotefromastoreo¤eringtheproductforsale.Supposethatthereisasearchcost,c,perpricequote.2If,afterobtainingnpricequotes,aconsumerpurchasesq(p)unitsoftheproductfromoneofthe rmsatpricepperunit,theconsumers(indirect)utilityisV=v(p)+McnTheanalysisthatfollowsfocusesonpostedpricemarketswhereconsumersknowthedistributionofpricesbutdonotknowthepriceschargedbyparticularstores.32.1.1TheStiglerModelStigler(1961)considersthespecialcaseofthisenvironmentwhere:1.EachconsumerwishestopurchaseK1unitsoftheproduct;thatis,q(p)=v0(p)=K;2.Theconsumerssearchprocessis xedsamplesearchpriortosearching,consumersdeter-minea xedsamplesize,n;of rmsfromwhomtoobtainpricequotesandthenbuyfrom2Inwhatfollows,weassumethatconsumershaveidenticalsearchcosts.Axell(1977)o¤ersamodelofpricedispersionwithheterogeneoussearchcosts.3ThisassumptionisrelaxedinRothschild(1974),BenabouandGertner(1993)andDana(1994),wherebuyerslearnaboutthedistributionofpricesastheysearch,andinRauh(1997),wherebuyerssearchstrategiesdependononly nitelymanymomentsofthedistributionofprices.Daughety(1992)o¤ersanalternativesearch-theoreticmodelofequilibriumpricedispersionthatresultsfrominformationalasymmetriesandalackofpriceprecommitmentonthepartof rms.5Thesearchmodelsconsideredinthissubsectionareallbasedonthefollowinggeneralenvi-ronment.Acontinuumofprice-setting rms(withunitmeasure)competeinamarketsellinganidentical(homogeneous)product.Firmshaveunlimitedcapacitytosupplythisproductatacon-stantmarginalcost,Acontinuumofconsumersisinterestedinpurchasingtheproduct.Letthemassofconsumersinthemarketbesothatthenumberofcustomersper rmisEachconsumerhasaquasi-linearutilityfunction,)+,whereisthequantityofthehomogeneousproductandisthequantityofsomenumerairegoodwhosepriceisnormalizedtobeunity.Thisimpliesthattheindirectutilityofaconsumerwhopaysapriceperunitoftheproductandwhohasanincomeofp;M)=)+whereisnonincreasinginByRoysidentity,notethatthedemandfortheproductofrelevanceisToacquiretheproduct,aconsumermust rstobtainapricequotefromastoreo¤eringtheproductforsale.Supposethatthereisasearchcost,,perpricequote.If,afterobtainingquotes,aconsumerpurchasesunitsoftheproductfromoneofthe rmsatpriceperunit,theconsumers(indirect)utilityis)+Theanalysisthatfollowsfocusesonpostedpricemarketswhereconsumersknowthedistributionofpricesbutdonotknowthepriceschargedbyparticularstores.2.1.1TheStiglerModelStigler(1961)considersthespecialcaseofthisenvironmentwhere:1.Eachconsumerwishestopurchaseunitsoftheproduct;thatis,)=)=2.Theconsumer
ssearchprocessis xedsamplesea
ssearchprocessis xedsamplesearchpriortosearching,consumersdeter-minea xedsamplesize,of rmsfromwhomtoobtainpricequotesandthenbuyfromInwhatfollows,weassumethatconsumershaveidenticalsearchcosts.Axell(1977)o¤ersamodelofpricedispersionwithheterogeneoussearchcosts.ThisassumptionisrelaxedinRothschild(1974),BenabouandGertner(1993)andDana(1994),wherebuyerslearnaboutthedistributionofpricesastheysearch,andinRauh(1997),wherebuyerssearchstrategiesdependononly nitelymanymomentsofthedistributionofprices.Daughety(1992)o¤ersanalternativesearch-theoreticmodelofequilibriumpricedispersionthatresultsfrominformationalasymmetriesandalackofpriceprecommitmentonthepartof rms.whichisdecreasinginn.Furthermore,theexpectedbene tfromsearcharegreaterforproductsboughtingreaterquantitiesormorefrequently;thatis,equation(1)isincreasinginK:SincethecostofthenthsearchisindependentofKwhiletheexpectedbene tisincreasinginK;itimmediatelyfollowsthattheequilibriumsearchintensity,n;isincreasinginK:Thatis,consumersobtainmorepricequotesforproductstheybuyingreaterquantities(orfrequencies).DespitethefactthattheStiglermodelassumeseachindividualinelasticallypurchasesKunitsoftheproduct,aversionofthelawofdemandholds:Each rmsexpecteddemandisanonincreasingfunctionofitsprice.Toseethis,notethata rmchargingpricepisvisitedbynconsumersando¤ersthelowestpricewithprobability(1F(p))n1:Thus,arepresentative rmsexpecteddemandwhenitchargesapriceofpisQ(p)=nK(1F(p))n1(2)whichisdecreasinginp:TheStiglermodelimpliesthatboththeexpectedtransactionsprice(Proposition1)aswellastheexpectedtotalcostsinclusiveofsearchcosts(Proposition2)arelowerwhenpricesaremoredispersed(inthesenseofameanpreservingspread).5Proposition1SupposethatapricedistributionGisameanpreservingspreadofapricedistri-butionF.Thentheexpectedtransactionspriceofaconsumerwhoobtainsn1pricequotesisstrictlylowerunderpricedistributionGthanunderF:Proof.Let=EFhp(n)miniEGhp(n)minibethedi¤erenceintheexpectedtransactionspriceunderFcomparedtoG:Wewillshowthatforalln1;0:Usingthede nitionofEhp(n)mini;=Z11pn(1F(p))n1dF(p)Z11tn(1G(t))n1dG(t)Letu=F(p)andv=G(p);sothatdu=dF(p),dv=dG(p);p=F1(u);andt=G1(v):Then=nZ10F1(u)(1u)n1dunZ10G1(v)(1v)n1dv=nZ10F1(u)G1(u)(1u)n1du5GisameanpreservingspreadofFif(a)R11[G(p)F(p)]dp=0and(b)Rz1[G(p)F(p)]dp0;withstrictinequalityforsomez:Notethat(a)isequivalenttothefactthatthemeansofFandGareequal.Together,thetwoconditionsimplythatFandGcrossexactlyonce(atthemean)ontheinteriorofthesupport.7whichisdecreasingin.Furthermore,theexpectedbene tfromsearcharegreaterforprod
uctsboughtingreaterquantitiesormorefrequ
uctsboughtingreaterquantitiesormorefrequently;thatis,equationisincreasinginK:Sincethecostofthethsearchisindependentofwhiletheexpectedbene tisincreasinginK;immediatelyfollowsthattheequilibriumsearchintensity,isincreasinginK:Thatis,consumersobtainmorepricequotesforproductstheybuyingreaterquantities(orfrequencies).DespitethefactthattheStiglermodelassumeseachindividualinelasticallypurchasesunitsoftheproduct,aversionofthelawofdemandholds:Each rmsexpecteddemandisanonincreasingfunctionofitsprice.Toseethis,notethata rmchargingpriceisvisitedbyconsumersando¤ersthelowestpricewithprobabilityThus,arepresentative rmsexpecteddemandwhenitchargesapriceof)=whichisdecreasinginTheStiglermodelimpliesthatboththeexpectedtransactionsprice(Proposition1)aswellastheexpectedtotalcostsinclusiveofsearchcosts(Proposition2)arelowerwhenpricesaremoredispersed(inthesenseofameanpreservingspread).Proposition1Supposethatapricedistributionisameanpreservingspreadofapricedistri-bution.ThentheexpectedtransactionspriceofaconsumerwhoobtainsnpricequotesisstrictlylowerunderpricedistributionthanunderF:Proof.=minminbethedi¤erenceintheexpectedtransactionspriceundercomparedtoWewillshowthatforallnUsingthede nitionofmin=dFsothatdudFdG;pThen=)(1)(1dvisameanpreservingspreadofiffG(p)F(p)]dp=0anddG(p)F(p)]dp0;withstrictinequalityforsomez:Notethatisequivalenttothefactthatthemeansofandareequal.Together,thetwoconditionsimplythatandcrossexactlyonce(atthemean)ontheinteriorofthesupport.2.1.2TheRothschildCritiqueandDiamondsParadoxWhileStiglero¤eredthe rstsearch-theoreticrationaleforpricedispersion,themodelhasbeencriticizedfortworeasons.First,aspointedoutinRothschild(1973),thesearchprocedureassumedinStiglersmodelmaynotbeoptimal.In xedsamplesearch,consumerscommittoa xednumber,n,ofstorestosearchandthenbuyatthelowestpriceattheconclusionofthatsearch.Acleardrawbacktosuchastrategyisthatitfailstoincorporatenewinformationobtainedduringsearch,suchasanexceptionallylowpricefromanearlysearch.Indeed,oncethebestpricequoteobtainedissu¢cientlylow,thebene tintheformofpriceimprovementdropsbelowthemarginalcostoftheadditionalsearch.Aswewillseebelow,sequentialsearchresultsinanoptimalstoppingrulesuchthataconsumersearchesuntilshelocatesapricebelowsomethreshold,calledthereservationprice.Second,thedistributionofprices,F;isexogenouslyspeci edandisnotbasedonoptimizing rmbehavior.Infact,inlightofequation(2),arepresentative rmwithconstantmarginalcostofmenjoysexpectedpro tsof(p)=(pm)Q(p):Thatis,absentanycostheterogeneities,each rmfacesexactlythesameexpectedpro tfunc-tion.Whythen,would rmsnotchoosethesamepro t-maximizingpriceor,moregenerally,howcouldthedistributionofpricesgenerated
bypro t-maximizing rmsbeconsis
bypro t-maximizing rmsbeconsistentwiththepricedistributionoverwhichconsumersweresearching?Inshort,Rothschildpointedoutthatitisfarfromclearthatinformationcostsgiverisetoanequilibriumofpricedispersionwithoptimizingconsumersand rms;inStiglersmodel,onlyonesideofonemarket,theconsumers,areactinginanoptimizingfashionconsistentwithequilibrium.Forthisreason,Rothschildcriticizedtheearlyliteratureforitspartial-partialequilibriumapproach.Diamond(1971)advancedthisargumentevenfurtherheessentiallyidenti edconditionsundercostlysearchwheretheuniqueequilibriuminundominatedstrategiesinvolvesall rmschargingthesamepricethemonopolyprice.Diamondsresultmaybereadilyseeninthefollowingspecialcaseofourenvironmentwhere:1.Consumershaveidenticaldownwardslopingdemand,i.e.v00(p)=q0(p)0;2.Consumersengageinoptimalsequentialsearch;3.A rmactingasamonopolywouldoptimallychargeallconsumerstheuniquemonopolyprice,p;and92.1.2TheRothschildCritiqueandDiamondsParadoxWhileStiglero¤eredthe rstsearch-theoreticrationaleforpricedispersion,themodelhasbeencriticizedfortworeasons.First,aspointedoutinRothschild(1973),thesearchprocedureassumedinStiglersmodelmaynotbeoptimal.In xedsamplesearch,consumerscommittoa xednumber,,ofstorestosearchandthenbuyatthelowestpriceattheconclusionofthatsearch.Acleardrawbacktosuchastrategyisthatitfailstoincorporatenewinformationobtainedduringsearch,suchasanexceptionallylowpricefromanearlysearch.Indeed,oncethebestpricequoteobtainedissu¢cientlylow,thebene tintheformofpriceimprovementdropsbelowthemarginalcostoftheadditionalsearch.Aswewillseebelow,sequentialsearchresultsinanoptimalstoppingrulesuchthataconsumersearchesuntilshelocatesapricebelowsomethreshold,calledthereservationprice.Second,thedistributionofprices,F;isexogenouslyspeci edandisnotbasedonoptimizing rmbehavior.Infact,inlightofequation,arepresentative rmwithconstantmarginalcostenjoysexpectedpro tsof)=(Thatis,absentanycostheterogeneities,each rmfacesexactlythesameexpectedpro tfunc-tion.Whythen,would rmsnotchoosethesamepro t-maximizingpriceor,moregenerally,howcouldthedistributionofpricesgeneratedbypro t-maximizing rmsbeconsistentwiththepricedistributionoverwhichconsumersweresearching?Inshort,Rothschildpointedoutthatitisfarfromclearthatinformationcostsgiverisetoanequilibriumofpricedispersionwithoptimizingconsumersand rms;inStiglersmodel,onlyonesideofonemarket,theconsumers,areactinginanoptimizingfashionconsistentwithequilibrium.Forthisreason,Rothschildcriticizedtheearlyliteratureforitspartial-partialequilibriumapproach.Diamond(1971)advancedthisargumentevenfurtherheessentiallyidenti edconditionsundercostlysearchwheretheuniqueequilibriuminundominat
edstrategiesinvolvesall rmscharging
edstrategiesinvolvesall rmschargingthesamepricethemonopolyprice.Diamondsresultmaybereadilyseeninthefollowingspecialcaseofourenvironmentwhere:1.Consumershaveidenticaldownwardslopingdemand,i.e.)=2.Consumersengageinoptimalsequentialsearch;3.A rmactingasamonopolywouldoptimallychargeallconsumerstheuniquemonopoly;and4.Aconsumerwhoischargedthemonopolypricebya rmwiththehighestmarginalcost,m;earnssurplussu¢cienttocoverthecostofobtainingasinglepricequote;thatisv"1+"mc:Reinganumshowsthat,undertheseassumptions,thereexistsadispersedpriceequilibriuminwhich rmsoptimallysetpricesandeachconsumerengagesinoptimalsequentialsearch.Toestablishthis,we rstshowhowonederivestheoptimalreservationpriceinasequentialsearchsetting.SupposeconsumersareconfrontedwithanondegeneratedistributionofpricesF(p)onp;pthatisatomless,exceptpossiblyatp:Consumersengageinoptimalsequentialsearchwithfreerecall.If,followingthenthsearch,aconsumerhasalreadyfoundabestpricezmin(p1;p2;:::;pn);then,bymakinganadditionalsearch,suchaconsumerexpectstogainbene tsofB(z)=Zzp(v(p)v(z))dF(p)=Zzpv0(p)F(p)dp;wherethesecondequalityobtainsthroughintegrationbyparts.UsingLeibnitzrule,wehaveB0(z)=v0(z)F(z)=Kz"F(z)0(3)Thus,theexpectedbene tsfromanadditionalsearcharelowerwhentheconsumerhasalreadyidenti edarelativelylowprice.Sincesearchiscostly(c0),consumersmustweightheexpectedbene tsagainstthecostofanadditionalsearch.Theexpectednetbene tsofanadditionalsearchareh(z)B(z)cIftheexpectedbene tsfromanadditionalsearchexceedtheadditionalcost,h(z)0;itisoptimalfortheconsumertoobtainanadditionalpricequote.Ifh(z)0,theconsumerisbettero¤purchasingatthepricezthanobtaininganadditionalpricequote.Aconsumersoptimalsequentialsearchstrategymaybesummarizedasfollows:Case1.h(p)0andRppv(p)dF(p)c:Thentheconsumersoptimalstrategyistonotsearch.Case2.h(p)0andRppv(p)dF(p)dpc:Thentheconsumersoptimalstrategyistosearchuntilsheobtainsapricequoteatorbelowthereservationprice,r=p:114.Aconsumerwhoischargedthemonopolypricebya rmwiththehighestmarginalcost,earnssurplussu¢cienttocoverthecostofobtainingasinglepricequote;thatis1+"Reinganumshowsthat,undertheseassumptions,thereexistsadispersedpriceequilibriuminwhich rmsoptimallysetpricesandeachconsumerengagesinoptimalsequentialsearch.Toestablishthis,we rstshowhowonederivestheoptimalreservationpriceinasequentialsearchsetting.Supposeconsumersareconfrontedwithanondegeneratedistributionofprices;thatisatomless,exceptpossiblyatConsumersengageinoptimalsequentialsearchwithfreerecall.If,followingthethsearch,aconsumerhasalreadyfoundabestpricemin(;p;:::;pthen,bymakinganadditionalsearch,suchaconsumerexpectstogainbene tsof)=(v(p)
v(z))dF(p)=Zzpdp;wherethesecondequa
v(z))dF(p)=Zzpdp;wherethesecondequalityobtainsthroughintegrationbyparts.UsingLeibnitzrule,wehave)=KzThus,theexpectedbene tsfromanadditionalsearcharelowerwhentheconsumerhasalreadyidenti edarelativelylowprice.Sincesearchiscostlyc,consumersmustweightheexpectedbene tsagainstthecostofanadditionalsearch.Theexpectednetbene tsofanadditionalsearchareIftheexpectedbene tsfromanadditionalsearchexceedtheadditionalcost,itisoptimalfortheconsumertoobtainanadditionalpricequote.If,theconsumerisbettero¤purchasingatthepricethanobtaininganadditionalpricequote.Aconsumersoptimalsequentialsearchstrategymaybesummarizedasfollows:Case1.p)0andRppdFc:Thentheconsumersoptimalstrategyistonotsearch.Case2.p)0andRppdFdpThentheconsumersoptimalstrategyistosearchuntilsheobtainsapricequoteatorbelowthereservationprice,p:11Ignoringforamomentthefactthata rmsdemandiszeroifitpricesabover;notethatpro t-maximizationimpliesthe rst-ordercondition(pjmj)q0(pj)+q(pj)1=0:Standardmanipulationofthe rst-orderconditionforpro t-maximizationimpliesthat rmjs(unconstrained)pro t-maximizingpriceisaconstantmarkupoveritscost:pj="1+"mj:Supposethat rmssimplyignoretheconsumersreservationprice,r;andpriceatthismarkup.Thiswouldimplythatconsumersfaceadistributionofpostedprices^F(p)=G(p(1+")=")ontheinterval[m"=(1+");m"=(1+")].Giventhisdistributionofprices,optimizingconsumerswouldsetareservationprice,r;suchthath(r)=Zrp(v(p)v(r))d^F(p)c=0Furthermore,ifrm"=(1+"); rmschargingpricesintheinterval(r;m"=(1+")]wouldenjoynosales.Sincetheelasticityofdemandisconstant, rmsthatwouldmaximizepro tsbypricingaboverintheabsenceofconsumersearch nditoptimaltosettheirpricesatrwhenconsumerssearch.7Thus,thedistributionofprices,^F(p);isinconsistentwithoptimizingbehavioronthepartof rms.Infact,giventhereservationpricer;optimizingbehavioronthepartof rmswouldimplyadistributionofpricesF(p)=8:^F(p)ifpr1ifp=rToestablishthatthisis,infact,anequilibriumdistributionofpricesonemustverifythatconsumersfacingthistruncateddistributionofpriceshavenoincentivetochangetheirreservationprice.Giventhistruncateddistributionofprices,thenetexpectedbene tsofsearchareh(r)=Zrp(v(p)v(r))dF(p)c=Zrp(v(p)v(r))d^F(p)+h1^F(r)i[v(r)v(r)]c=Zrp(v(p)v(r))d^F(p)c=07Reinganumassumesthatmm"=(1+"),whichguaranteesthat rmswhowouldotherwisepriceabover nditpro tabletopriceatr.13Ignoringforamomentthefactthata rmsdemandiszeroifitpricesabover;notethatpro t-maximizationimpliesthe rst-ordercondition)+=0Standardmanipulationofthe rst-orderconditionforpro t-maximizationimpliesthat rm
s(unconstrained)pro t-maximiz
s(unconstrained)pro t-maximizingpriceisaconstantmarkupoveritscost:1+Supposethat rmssimplyignoretheconsumersreservationprice,r;andpriceatthismarkup.Thiswouldimplythatconsumersfaceadistributionofpostedprices)=(1+ontheinterval(1+(1+.Giventhisdistributionofprices,optimizingconsumerswouldsetareservationprice,r;suchthat)==0Furthermore,ifr(1+ rmschargingpricesintheintervalr;(1+wouldenjoynosales.Sincetheelasticityofdemandisconstant, rmsthatwouldmaximizepro tsbypricingaboveintheabsenceofconsumersearch nditoptimaltosettheirpricesatwhenconsumerssearch.Thus,thedistributionofprices,isinconsistentwithoptimizingbehavioronthepartof rms.Infact,giventhereservationpricer;optimizingbehavioronthepartof rmswouldimplyadistributionofprices)=prToestablishthatthisis,infact,anequilibriumdistributionofpricesonemustverifythatconsumersfacingthistruncateddistributionofpriceshavenoincentivetochangetheirreservationprice.Giventhistruncateddistributionofprices,thenetexpectedbene tsofsearchare)=dF)+=0Reinganumassumesthatmm(1+,whichguaranteesthat rmswhowouldotherwisepriceabove nditpro tabletopriceatwhere^f(p)isthedensityof^F(p):Hence,d2dr=2r1^F(r)2 Zrpp^f(p)dp+1^F(r)r!1^F(r)=2h1^F(r)i(rE[p])0withstrictinequalityifrm"=(1+"):Thus,wehave:Conclusion1IntheReinganummodel,areductioninsearchcostsdecreasesthevarianceofequilibriumprices.Aswewillseebelow,however,thisisnotageneralpropertyofsearch-theoreticmodelsofpricedispersion.2.1.4RemarksonFixedversusSequentialSearchItisusefultohighlightsomekeydi¤erencesbetweensequentialand xedsamplesizesearch.Withsequentialsearch,thenumberofsearchesisarandomvariablefromageometricdistribution,andtheexpectednumberofsearches,givenadistributionofpricesF(p)andreservationpricer,isE[n]=1F(r)Incontrast,with xedsamplesizesearch,consumerscommitupfronttonsearches.Bothtypesofsearchhaveadvantagesanddisadvantages,andindeedMorganandManning(1985)haveshownthatbothtypesofsearchcanbeoptimalindi¤erentcircumstances.Thekeyadvantageofsequentialsearchisthatitallowsasearchertoeconomizeoninformationcoststhedecision-makerweighstheexpectedbene tsandcostsofgatheringadditionalpriceinformationaftereachnewpricequoteisobtained.Ifanacceptablepriceisobtainedearlyon,theexpectedgainsfromadditionalsearchesaresmallandthereisnoneedtopaythecostofadditionalsearches.Theprimaryadvantageof xed-samplesizesearchisthatitallowsonetogatherinformationquickly.Consider,forinstance,a rmthatrequiresrawmaterialsbytheendoftheweek.Ifittakesaweekforarawmaterialsvendortoprovideapricequote,sequentialsearchwouldpermitthe rmtoobtainonlyapricequotefromasinglevendor.Inthiscase, xedsamplesizesearchisoptimalthe rmcommi
tstoobtainquotesfromnvendors,wherenischo
tstoobtainquotesfromnvendors,wherenischosenbythe rmtominimizeexpectedcostsasoutlinedaboveinourdiscussionoftheStiglermodel.15whereisthedensityofHence,d=2=22p])0withstrictinequalityifr(1+Thus,wehave:Conclusion1IntheReinganummodel,areductioninsearchcostsdecreasesthevarianceofequilibriumprices.Aswewillseebelow,however,thisisnotageneralpropertyofsearch-theoreticmodelsofpricedispersion.2.1.4RemarksonFixedversusSequentialSearchItisusefultohighlightsomekeydi¤erencesbetweensequentialand xedsamplesizesearch.Withsequentialsearch,thenumberofsearchesisarandomvariablefromageometricdistribution,andtheexpectednumberofsearches,givenadistributionofpricesandreservationprice,isisn]=Incontrast,with xedsamplesizesearch,consumerscommitupfronttosearches.Bothtypesofsearchhaveadvantagesanddisadvantages,andindeedMorganandManning(1985)haveshownthatbothtypesofsearchcanbeoptimalindi¤erentcircumstances.Thekeyadvantageofsequentialsearchisthatitallowsasearchertoeconomizeoninformationcoststhedecision-makerweighstheexpectedbene tsandcostsofgatheringadditionalpriceinformationaftereachnewpricequoteisobtained.Ifanacceptablepriceisobtainedearlyon,theexpectedgainsfromadditionalsearchesaresmallandthereisnoneedtopaythecostofadditionalsearches.Theprimaryadvantageof xed-samplesizesearchisthatitallowsonetogatherinformationquickly.Consider,forinstance,a rmthatrequiresrawmaterialsbytheendoftheweek.Ifittakesaweekforarawmaterialsvendortoprovideapricequote,sequentialsearchwouldpermitthe rmtoobtainonlyapricequotefromasinglevendor.Inthiscase, xedsamplesizesearchisoptimalthe rmcommitstoobtainquotesfromvendors,whereischosenbythe rmtominimizeexpectedcostsasoutlinedaboveinourdiscussionoftheStiglermodel.Noticethat,afterintegrationbyparts,wecanrewriteequation(6)toobtainthefamiliarformulaforequilibriumbiddinginreverse rst-priceauctionsp(m)=Ehm(n1)minjm(n1)minmi(7)wherem(n1)ministhelowestofn1drawsfromthedistributionG:ForthespecialcasewhereGisuniformlydistributed,theequilibriumpricingstrategysimpli estop(m)=n1nm+1nm:(8)Noticethattheequilibriumpricingstrategygivesrisetoadistributionofpostedprices,F(p),inducedbythedistributionofcosts;thatisF(p)=G(p(m))Forthistobeanequilibriumdistributionofprices,itmustbeoptimalforconsumerstosamplen rms.Thatis,EhB(n+1)icEhB(n)iwheretheexpressionEB(n),aspreviouslyde nedinequation(1)whenK=1,istheexpectedbene tfromincreasingthenumberofpricequotesobtainedfromn1ton:AsintheStiglermodel,areductioninsearchcostsincreasestheoptimalsamplesizen(sothatconsumersoptimallysamplemore rms).Thus,MacMinnshowsthat,providedsearchcostsarelowenough,adispersedpriceequilibriumexists.Thisnotonlyleadstoexpostdi¤erencesinco
nsumersinformationsets(di¤erentco
nsumersinformationsets(di¤erentconsumerssampledi¤erent rmsandsoobservedi¤erentprices),butinducesadegreeofcompetitionamong rms(sincetheyarecompetingagainstatleastoneother rm,whosecosttheydonotknow).AsintheReinganummodel,thelevelofpricedispersiondependsonthedispersionin rmscosts.Forthespecialcasewherecostsareuniformlydistributed,thevarianceinequilibriumprices2pisgivenby2p=n1n22m(9)wherenistheoptimalnumberofsearchesbyconsumersand2misthevariancein rmscosts.Twointerestingresultsemergefromthemodel.First,thevarianceinpricesincreasesasthevariancein rmsmarginalcostsincreases.Thisresultisintuitive.Somewhatcounterintuitively,17Noticethat,afterintegrationbyparts,wecanrewriteequationtoobtainthefamiliarformulaforequilibriumbiddinginreverse rst-priceauctions)=minminwhereministhelowestofdrawsfromthedistributionForthespecialcasewhereisuniformlydistributed,theequilibriumpricingstrategysimpli es)=nm+1nNoticethattheequilibriumpricingstrategygivesrisetoadistributionofpostedprices,inducedbythedistributionofcosts;thatis)=Forthistobeanequilibriumdistributionofprices,itmustbeoptimalforconsumerstosample rms.Thatis,+1)cwheretheexpression,aspreviouslyde nedinequationwhen=1,istheexpectedbene tfromincreasingthenumberofpricequotesobtainedfromAsintheStiglermodel,areductioninsearchcostsincreasestheoptimalsamplesize(sothatconsumersoptimallysamplemore rms).Thus,MacMinnshowsthat,providedsearchcostsarelowenough,adispersedpriceequilibriumexists.Thisnotonlyleadstoexpostdi¤erencesinconsumersinformationsets(di¤erentconsumerssampledi¤erent rmsandsoobservedi¤erentprices),butinducesadegreeofcompetitionamong rms(sincetheyarecompetingagainstatleastoneother rm,whosecosttheydonotknow).AsintheReinganummodel,thelevelofpricedispersiondependsonthedispersionin rmscosts.Forthespecialcasewherecostsareuniformlydistributed,thevarianceinequilibriumpricesisgivenbywhereistheoptimalnumberofsearchesbyconsumersandisthevariancein rmscosts.Twointerestingresultsemergefromthemodel.First,thevarianceinpricesincreasesasthevariancein rmsmarginalcostsincreases.Thisresultisintuitive.Somewhatcounterintuitively,obtainingasinglepricequote:13IntheBurdettandJuddmodel,anequilibriumconsistsofapricedistributionF(p)(basedonoptimalpricingdecisionsby rms)andanoptimalsearchdistributionn-278;1n=1,wheren-278;1n=1isthedistributionofthenumberoftimesaconsumersearchesinthepopulation.Thus,iistheprobabilitythataconsumersearches(oralternatively,thefractionofconsumersthatsearch)exactlyi rms.If1=1;thenallconsumerssampleonlyone rm.If1=0;thenallconsumerssampleatleasttwo rms,andsoon.Consumerspurchas
efromthe rmsampledthato¤ersthelowe
efromthe rmsampledthato¤ersthelowestprice.WebeginbystudyingoptimalsearchonthepartofconsumersgivenapricedistributionF(p):Recallthattheexpectedbene ttoaconsumerwhoincreaseshersamplesizefromn1tonisEhB(n)i=Ehp(n1)miniEhp(n)miniasintheStiglermodel.Moreover,theexpectedbene tscheduleisstrictlydecreasinginn:Thus,anoptimalnumberofpricequotes,n;satis esEhB(n+1)icEhB(n)iFirstconsiderthecasewhereallconsumersobtaintwoormorepricequotes;thatis,where1=0:Inthiscase,theoptimalpricingstrategyonthepartof rmsistopriceatmarginalcost(theBertrandparadox)sinceeach rmisfacingpurepricecompetitionwithatleastoneother rmandall rmsareidentical.Ofcourse,ifall rmsarepricingatmarginalcost,thenitwouldbeoptimalforaconsumertosampleonlyone rm,whichcontradictsthehypothesisthat1=0.Thus,wemayconcludethat,inanyequilibrium1-278;0:Next,considerthecasewhereconsumersallobtainexactlyonepricequote.Inthatcase,each rmwouldoptimallychargethemonopolyprice,p=v:Hence,16=1inanydispersedpriceequilibrium.Fromthesetwoargumentsitfollowsthat,inanydispersedpriceequilibrium,12(0;1).Inlightofthefactthatconsumersexpectedbene tsfromsearcharedecreasinginthesamplesize,it13Theseassumptionsaresatis ed,forexample,whenq(p)=8]TJ ; -1;.13; Td;[000;]TJ ; -1;.13; Td;[000;]TJ ; -1;.13; Td;[000;:1ifpv1pvifvpv+0ifpv+andc=2:19obtainingasinglepricequote13IntheBurdettandJuddmodel,anequilibriumconsistsofapricedistribution(basedonoptimalpricingdecisionsby rms)andanoptimalsearchdistribution,whereisthedistributionofthenumberoftimesaconsumersearchesinthepopulation.Thus,istheprobabilitythataconsumersearches(oralternatively,thefractionofconsumersthatsearch)exactly rms.If=1thenallconsumerssampleonlyone rm.If=0thenallconsumerssampleatleasttwo rms,andsoon.Consumerspurchasefromthe rmsampledthato¤ersthelowestprice.WebeginbystudyingoptimalsearchonthepartofconsumersgivenapricedistributionRecallthattheexpectedbene ttoaconsumerwhoincreaseshersamplesizefromminminasintheStiglermodel.Moreover,theexpectedbene tscheduleisstrictlydecreasinginThus,anoptimalnumberofpricequotes,satis es+1)cFirstconsiderthecasewhereallconsumersobtaintwoormorepricequotes;thatis,where=0Inthiscase,theoptimalpricingstrategyonthepartof rmsistopriceatmarginalcost(theBertrandparadox)sinceeach rmisfacingpurepricecompetitionwithatleastoneother rmandall rmsareidentical.Ofcourse,ifall rmsarepricingatmarginalcost,thenitwouldbeoptimalforaconsumertosampleonly rm,whichcontradictsthehypothesisthat=0Thus,wemayconcludethat,inanyequilibriumNext,considerthecasewhereconsume
rsallobtainexactlyonepricequote.Inthatca
rsallobtainexactlyonepricequote.Inthatcase,each rmwouldoptimallychargethemonopolyprice,v:Hence,=1inanydispersedpriceequilibrium.Fromthesetwoargumentsitfollowsthat,inanydispersedpriceequilibrium,.Inlightofthefactthatconsumersexpectedbene tsfromsearcharedecreasinginthesamplesize,it13Theseassumptionsaresatis ed,forexample,when)=ifpvififpvandc=Tosummarize,BurdettandJuddshowthatequilibriumpricedispersioncanariseevenwhenall rmsandconsumersareexanteidentical.Intheequilibriumpricedistribution,all rmschargepositivemarkups.Afractionofconsumersdonotcomparisonshoptheysimplysearchatonestoreandpurchase.Theremainingfractionofconsumersareshopperstheseconsumerssearchattwostoresandbuyfromwhichevero¤ersthelowerprice.2.2ModelswithanInformationClearinghouseInsearch-theoreticmodels,consumerspayanincrementalcostforeachadditionalpricequotetheyobtain.Thesemodelsarerelevant,forexample,whenconsumersmustvisitorphonetraditionalsellersinordertogatherinformationaboutprices.Theyarealsorelevantinonlineenvironmentswhereconsumersmustsearchthewebsitesofindividualretailerstogatherinformationaboutthepricestheycharge.Analternativeclassofmodelsisrelevantwhenathirdpartyaninformationclearinghouseprovidesasubsetofconsumerswithalistofpriceschargedbydi¤erent rmsinthemarket.Examplesofthisenvironmentincludenewspaperswhichdisplaypricesdi¤erentstoreschargeforthesameproductorserviceandonlinepricecomparisonsites.Inthissectionweprovideageneraltreatmentofclearinghousemodels,andshowthatthesemodelsaresurprisinglysimilartothosethatariseunder xedsamplesizesearch.Oneofthekeymodelingdi¤erencesisthatclearinghousemodelstendtobeoligopolymodels;thus,thereisnotacontinuumof rmsinsuchsettings.Wherepossible,weshallusethesamenotationasintheprevioussection;however,forreasonsthatwillbecomeclearwhenwecompareclearinghousemodelswiththesearchmodelspresentedabove,wenowletndenotethenumberof rmsinthemarket.ThegeneraltreatmentthatfollowsreliesheavilyonBayeandMorgan(2001)andBaye,MorganandScholten(2004a).Considerthefollowinggeneralenvironment(whichwewillspecializetocoveravarietyofdi¤er-entmodels).Thereisa nitenumber,n1;ofprice-setting rmscompetinginamarketsellinganidentical(homogeneous)product.Firmshaveunlimitedcapacitytosupplythisproductataconstantmarginalcost,m:Acontinuumofconsumersisinterestedinpurchasingtheproduct.Thismarketisservedbyapriceinformationclearinghouse.Firmsmustdecidewhatpricetochargefortheproductandwhethertolistthispriceattheclearinghouse.Letpidenotethepricechargedby rmi:Itcostsa rmanamount0ifitchoosestolistitsprice.Allconsumershaveunitdemand21Tosummarize,BurdettandJuddshowthatequilibriumpricedispersioncanariseevenwhenall rmsandconsumersareexanteiden
tical.Intheequilibriumpricedistribution,
tical.Intheequilibriumpricedistribution,all rmschargepositivemarkups.Afractionofconsumersdonotcomparisonshoptheysimplysearchatonestoreandpurchase.Theremainingfractionofconsumersareshopperstheseconsumerssearchattwostoresandbuyfromwhichevero¤ersthelowerprice.2.2ModelswithanInformationClearinghouseInsearch-theoreticmodels,consumerspayanincrementalcostforeachadditionalpricequotetheyobtain.Thesemodelsarerelevant,forexample,whenconsumersmustvisitorphonetraditionalsellersinordertogatherinformationaboutprices.Theyarealsorelevantinonlineenvironmentswhereconsumersmustsearchthewebsitesofindividualretailerstogatherinformationaboutthepricestheycharge.Analternativeclassofmodelsisrelevantwhenathirdpartyaninformationclearinghouseprovidesasubsetofconsumerswithalistofpriceschargedbydi¤erent rmsinthemarket.Examplesofthisenvironmentincludenewspaperswhichdisplaypricesdi¤erentstoreschargeforthesameproductorserviceandonlinepricecomparisonsites.Inthissectionweprovideageneraltreatmentofclearinghousemodels,andshowthatthesemodelsaresurprisinglysimilartothosethatariseunder xedsamplesizesearch.Oneofthekeymodelingdi¤erencesisthatclearinghousemodelstendtobeoligopolymodels;thus,thereisnotacontinuumof rmsinsuchsettings.Wherepossible,weshallusethesamenotationasintheprevioussection;however,forreasonsthatwillbecomeclearwhenwecompareclearinghousemodelswiththesearchmodelspresentedabove,wenowletdenotethenumberof rmsinthemarket.ThegeneraltreatmentthatfollowsreliesheavilyonBayeandMorgan(2001)andBaye,MorganandScholten(2004a).Considerthefollowinggeneralenvironment(whichwewillspecializetocoveravarietyofdi¤er-entmodels).Thereisa nitenumber,nofprice-setting rmscompetinginamarketsellinganidentical(homogeneous)product.Firmshaveunlimitedcapacitytosupplythisproductataconstantmarginalcost,Acontinuumofconsumersisinterestedinpurchasingtheproduct.Thismarketisservedbyapriceinformationclearinghouse.Firmsmustdecidewhatpricetochargefortheproductandwhethertolistthispriceattheclearinghouse.Letdenotethepricechargedby rmItcostsa rmanamountifitchoosestolistitsprice.AllconsumershaveunitdemandThisconditionholds,sincen1n(vm)S:Noticethatp0m;providedthatL0or0:Inthiscase,itcanbeshownthatFisawell-de ned,atomlesscdfon[p0;v].WhenL=0and=0,noticethatp0=m.Inthiscase,thesymmetricequilibriumdistributionofpricesisdegenerate,withall rmspricingatmarginalcost(theBertrandparadoxoutcome).Next,weshowthat,conditionalonlistingaprice,a rmcandonobetterthanpricingaccordingtoF:ItisobviousthatchoosingapriceaboveorbelowthesupportofFisdominatedbychoosingapriceinthesupportofF:A rmchoosingapricepinthesupportofFearnsexpectedpro tsofE(p)=(pm) L+ n1Xi=0n1i
;i(1)n1i(1F(p)
;i(1)n1i(1F(p))i!S!:Usingthebinomialtheorem,wecanrewritethisas:E(p)=(pm)L+(1F(p))n1S=(vm)L+n1;wherewehavesubstitutedforFtoobtainthesecondequality.Sincea rmsexpectedpro tsareconstanton[p0;v],itfollowsthatthemixedpricingstrategy,F;isabestresponsetotheothern1 rmspricingbasedonF:When=0;itisaweaklydominantstrategytolist.Itremainstoshowthatwhen0and2(0;1),a rmearnsthesameexpectedpro tsregardlessofwhetheritlistsitsprice.Buta rmthatdoesnotlistearnsexpectedpro tsofE=(vm)L+Sn(1)n1=(vm)L+n1;whichequalstheexpectedpro tsearnedbylistinganypricep2[p0;v].Wearenowinapositiontoexaminethemanywell-knownclearinghousemodelsthatemergeasspecialcasesofthisgeneralenvironment.2.2.1TheRosenthalModelRosenthal(1980)wasamongthe rsttoshowthatequilibriumpricedispersioncanariseinaclearinghouseenvironmentwhensomeconsumershaveapreferenceforaparticular rm.Underhisinterpretation,each rmenjoysamassLofloyalconsumers.Rosenthalsmainresultsmaybeseeninthefollowingspecialcaseofthegeneralclearinghousemodel:23Thisconditionholds,sinceS:Noticethatm;providedthatLInthiscase,itcanbeshownthatisawell-de ned,atomlesscdfononp0;v.When=0=0,noticethat.Inthiscase,thesymmetricequilibriumdistributionofpricesisdegenerate,withall rmspricingatmarginalcost(theBertrandparadoxoutcome).Next,weshowthat,conditionalonlistingaprice,a rmcandonobetterthanpricingaccordingF:ItisobviousthatchoosingapriceaboveorbelowthesupportofisdominatedbychoosingapriceinthesupportofF:A rmchoosingapriceinthesupportofearnsexpectedpro tsofE)=(Usingthebinomialtheorem,wecanrewritethisas:E)=(F=(wherewehavesubstitutedfortoobtainthesecondequality.Sincea rmsexpectedpro tsareconstantononp0;v,itfollowsthatthemixedpricingstrategy,F;isabestresponsetotheother rmspricingbasedonF:=0itisaweaklydominantstrategytolist.Itremainstoshowthatwhen,a rmearnsthesameexpectedpro tsregardlessofwhetheritlistsitsprice.Buta rmthatdoesnotlistearnsexpectedpro tsofE=(=(whichequalstheexpectedpro tsearnedbylistinganypriceicep0;vWearenowinapositiontoexaminethemanywell-knownclearinghousemodelsthatemergeasspecialcasesofthisgeneralenvironment.2.2.1TheRosenthalModelRosenthal(1980)wasamongthe rsttoshowthatequilibriumpricedispersioncanariseinaclearinghouseenvironmentwhensomeconsumershaveapreferenceforaparticular rm.Underhisinterpretation,each rmenjoysamassofloyalconsumers.Rosenthalsmainresultsmaybeseeninthefollowingspecialcaseofthegeneralclearinghousemodel:bringsmoreloyals
intothemarket.Indeed,thefractionofshoppe
intothemarket.Indeed,thefractionofshoppersinthemarketisS=(S+nL)anditmayreadilybeseenthatasnbecomeslarge,shoppersaccountforanincreasinglysmallfractionofthecustomerbaseof rms.Asaconsequence,theincentivestocompeteforthesecustomersisattenuatedandpricesriseasaresult.Thekeyistorecognizethatincreasesinnchangethedistributionofprices,andthise¤ectaswellasanyorderstatistice¤ectassociatedwithanincreaseinnmustbetakenintoaccount.Formally,noticethattheequilibriumdistributionofprices,F;isstochasticallyorderedinn:Thatis,thedistributionofpriceswhentherearen+1 rmscompeting rst-orderstochasticallydominatesthedistributionofpriceswheretherearen rmscompeting.Thisimpliesthatthetransactionspricespaidbyloyalsincreaseinn.Toshowthatthetransactionspricespaidbyshoppersalsoincreaseinnrequiresabitmorework;however,onecanshowthatthesamestochasticorderingobtainsforthecdfF(n)min(p):Finally,itisusefultonotethesimilaritybetweentheRosenthalversionoftheclearinghousemodelandthesearch-theoreticmodelofBurdettandJudd.InBurdettandJudd,eventhoughthereisacontinuumof rms,eachconsumeronlysamplesa nitenumberof rms(oneortwo).Further,inBurdettandJudd,a xedfractionofconsumersper rm,;sampleonlyasingle rm.Ine¤ect,theseconsumersareloyaltothesingle rmsampledwhilethefraction(1)ofcustomerssamplingtwo rmsareshopperstheychoosethelowerofthetwoprices.Forthisreason,whenn=2intheRosenthalmodel,theequilibriumpricedistributiongiveninequation(11)isidenticaltoequation(10)intheBurdettandJuddmodel(modulorelabelingthevariablesforloyalsandshoppers).2.2.2TheVarianModelVarian(1980)wasamongthe rsttoshowthatequilibriumpricedispersioncanariseinaclear-inghouseenvironmentwhenconsumershavedi¤erentexanteinformationsets.16VarianinterpretstheSconsumersasinformedconsumersandtheLconsumersasuninformedconsumers.Thusamass,S;ofconsumerschoosetoaccesstheclearinghousewhileothers,themassLper rm,donot.Variansmainresultmaybeseeninthefollowingspecialcaseofthegeneralclearinghousemodel:1.Itiscostlessfor rmstolistpricesontheclearinghouse:=0;and16PngandHirshleifer(1987),aswellasBayeandKovenock(1994),extendtheVarianmodelbyallowing rmstoalsoengageinpricematchingorbeatorpayadvertisements.25bringsmoreloyalsintothemarket.Indeed,thefractionofshoppersinthemarketisS=anditmayreadilybeseenthatasbecomeslarge,shoppersaccountforanincreasinglysmallfractionofthecustomerbaseof rms.Asaconsequence,theincentivestocompeteforthesecustomersisattenuatedandpricesriseasaresult.Thekeyistorecognizethatincreasesinchangethedistributionofprices,andthise¤ectaswellasanyorderstatistice¤ectassociatedwithanincreaseinmustbetakenintoaccount.Formally,noticethattheequilibriumdistributionofprices,F;isstochasti
callyorderedinThatis,thedistributionofpr
callyorderedinThatis,thedistributionofpriceswhenthereare+1 rmscompeting rst-orderstochasticallydominatesthedistributionofpriceswherethereare rmscompeting.Thisimpliesthatthetransactionspricespaidbyloyalsincreasein.Toshowthatthetransactionspricespaidbyshoppersalsoincreaseinrequiresabitmorework;however,onecanshowthatthesamestochasticorderingobtainsforthecdfminFinally,itisusefultonotethesimilaritybetweentheRosenthalversionoftheclearinghousemodelandthesearch-theoreticmodelofBurdettandJudd.InBurdettandJudd,eventhoughthereisacontinuumof rms,eachconsumeronlysamplesa nitenumberof rms(oneortwo).Further,inBurdettandJudd,a xedfractionofconsumersper rm,;sampleonlyasingle rm.Ine¤ect,theseconsumersareloyaltothesingle rmsampledwhilethefractionofcustomerssamplingtwo rmsareshopperstheychoosethelowerofthetwoprices.Forthisreason,when=2intheRosenthalmodel,theequilibriumpricedistributiongiveninequationidenticaltoequationintheBurdettandJuddmodel(modulorelabelingthevariablesforloyalsandshoppers).2.2.2TheVarianModelVarian(1980)wasamongthe rsttoshowthatequilibriumpricedispersioncanariseinaclear-inghouseenvironmentwhenconsumershavedi¤erentexanteinformationsets.16Varianinterpretsconsumersasinformedconsumersandtheconsumersasuninformedconsumers.Thusamass,S;ofconsumerschoosetoaccesstheclearinghousewhileothers,themassper rm,donot.Variansmainresultmaybeseeninthefollowingspecialcaseofthegeneralclearinghousemodel:1.Itiscostlessfor rmstolistpricesontheclearinghouse:=0;and16PngandHirshleifer(1987),aswellasBayeandKovenock(1994),extendtheVarianmodelbyallowing rmstoalsoengageinpricematchingorbeatorpayadvertisements.equilibriumtheBertrandparadox.Thus,forsu¢cientlyhighorlowinformationcosts,thereisnopricedispersion;formoderateinformationcosts,pricesaredispersedonthenondegenerateinterval[p0;v].AsimilarresultobtainsinStahl(1989),whichisrelatedtoVarianasfollows.Stahlassumesafractionofconsumershavezerosearchcostsand,asaconsequence,viewall rmspricesandpurchaseatthelowestpriceinthemarket.TheseconsumersplaytheroleofSinVariansmodel(informedconsumers).TheremainingfractionofconsumerscorrespondtotheLsintheVarianmodel,butratherthanremainingentirelyuninformed,theseconsumersengageinoptimalsequen-tialsearchinpresenceofpositiveincrementalsearchcosts.Stahlshowsthatwhenallconsumersareshoppers,theidentical rmspriceatmarginalcostandthereisnopricedispersion.Whennoconsumersareshoppers,Diamondsparadoxobtainsandall rmschargethemonopolyprice.Asthefractionofshoppersvariesfromzerotoone,thelevelofdispersionvariescontinuouslyfromzerotopositivelevels,andbackdowntozero.Conclusion3Ingeneral,pricedispersionisnotamonotonicfuncti
onofconsumersinformationcostsorthe
onofconsumersinformationcostsorthefractionofshoppersinthemarket.Howdoesthenumberofcompeting rmsa¤ecttransactionsprices?IntheRosenthalmodel,wesawthatincreasedcompetitionledtohigherexpectedtransactionspricesforallconsumers.IntheVarianmodel,incontrast,thee¤ectofcompetitiononconsumerwelfaredependsonwhetherornottheconsumerchoosestoaccesstheclearinghouse.Morgan,Orzen,andSefton(forthcoming)showthatasnincreases,thecompetitivee¤ectpredictablyleadstoloweraveragetransactionpricesbeingpaidbyinformedconsumers.However,theoppositeistrueforuninformedconsumersasthenumberofcompeting rmsincreases, rmsfacereducedincentivestocutpricesinhopesofattractingtheshoppersand,asaconsequence,theaveragepricechargedbya rm,whichisalsotheaveragepricepaidbyanuninformedconsumer,increases.IfoneviewstheclearinghouseasrepresentingaccesstopriceinformationontheInternet,thenonecaninterpretthepricee¤ectasoneconsequenceoftheso-calleddigitaldivide;seeBaye,Morgan,andScholten(2003).ConsumerswithInternetaccessaremadebettero¤bysharperonlinecompetitionwhilethosewithoutsuchaccessaremadeworseo¤.2.2.3TheBayeandMorganModelAlloftheabovemodelsassumethatitiscostlessfor rmstoadvertisetheirpricesattheclear-inghouse.BayeandMorgan(2001)pointoutthat,inpractice,itisgenerallycostlyfor rmsto27equilibriumtheBertrandparadox.Thus,forsu¢cientlyhighorlowinformationcosts,thereisnopricedispersion;formoderateinformationcosts,pricesaredispersedonthenondegenerateintervalalp0;v.AsimilarresultobtainsinStahl(1989),whichisrelatedtoVarianasfollows.Stahlassumesafractionofconsumershavezerosearchcostsand,asaconsequence,viewall rmspricesandpurchaseatthelowestpriceinthemarket.TheseconsumersplaytheroleofinVariansmodel(informedconsumers).TheremainingfractionofconsumerscorrespondtothesintheVarianmodel,butratherthanremainingentirelyuninformed,theseconsumersengageinoptimalsequen-tialsearchinpresenceofpositiveincrementalsearchcosts.Stahlshowsthatwhenallconsumersareshoppers,theidentical rmspriceatmarginalcostandthereisnopricedispersion.Whennoconsumersareshoppers,Diamondsparadoxobtainsandall rmschargethemonopolyprice.Asthefractionofshoppersvariesfromzerotoone,thelevelofdispersionvariescontinuouslyfromzerotopositivelevels,andbackdowntozero.Conclusion3Ingeneral,pricedispersionisnotamonotonicfunctionofconsumersinformationcostsorthefractionofshoppersinthemarket.Howdoesthenumberofcompeting rmsa¤ecttransactionsprices?IntheRosenthalmodel,wesawthatincreasedcompetitionledtohigherexpectedtransactionspricesforallconsumers.IntheVarianmodel,incontrast,thee¤ectofcompetitiononconsumerwelfaredependsonwhetherornottheconsumerchoosestoaccesstheclearinghouse.Morgan,Orzen,andSefton(forthcomin
g)showthatasincreases,thecompetitivee¤e
g)showthatasincreases,thecompetitivee¤ectpredictablyleadstoloweraveragetransactionpricesbeingpaidbyinformedconsumers.However,theoppositeistrueforuninformedconsumersasthenumberofcompeting rmsincreases, rmsfacereducedincentivestocutpricesinhopesofattractingtheshoppersand,asaconsequence,theaveragepricechargedbya rm,whichisalsotheaveragepricepaidbyanuninformedconsumer,increases.IfoneviewstheclearinghouseasrepresentingaccesstopriceinformationontheInternet,thenonecaninterpretthepricee¤ectasoneconsequenceoftheso-calleddigitaldivide;seeBaye,Morgan,andScholten(2003).ConsumerswithInternetaccessaremadebettero¤bysharperonlinecompetitionwhilethosewithoutsuchaccessaremadeworseo¤.2.2.3TheBayeandMorganModelAlloftheabovemodelsassumethatitiscostlessfor rmstoadvertisetheirpricesattheclear-inghouse.BayeandMorgan(2001)pointoutthat,inpractice,itisgenerallycostlyfor rmstoUndertheseconditions,usingProposition3,weobtainthefollowingcharacterizationofequilib-rium rmpricingandlistingdecisions:Each rmlistsitspriceattheclearinghousewithprobability=1nn1(vm)S1n12(0;1)Whena rmlistsattheclearinghouse,itchargesapricedrawnfromthedistributionF(p)=1 1nn1(pm)S1n1!on[p0;v];wherep0=m+nn1S:Whena rmdoesnotlistitsprice,itchargesapriceequaltov;andeach rmearnsequilibriumexpectedpro tsequaltoE=1n1Noticethatnrepresentstheaggregatedemandby rmsforadvertisingandisadecreasingfunctionofthefeechargedbythegatekeeper.Pricesadvertisedattheclearinghousearedispersedandstrictlylowerthanunadvertisedprices(v).Severalfeaturesofthisequilibriumareworthnoting.First,equilibriumpricedispersionariseswithfullyoptimizingconsumers, rms,andendogenousfee-settingdecisionsonthepartoftheclearinghousedespitethefactthattherearenoconsumeror rmheterogeneitiesandallconsumersarefullyinformedinthesensethat,inequilibrium,theyalwayspurchasefroma rmchargingthelowestpriceintheglobalmarket.Second,whileequilibriumpricedispersionintheVarianmodelisdrivenbythefactthatdi¤erentconsumershavedi¤erentcostsofaccessingtheclearinghouse,BayeandMorganshowthatanoptimizingclearinghousewillsetitsfeessu¢cientlylowthatallconsumerswillrationallyaccesstheclearinghouse.Equilibriumpricedispersionarisesbecauseofthegatekeepersincentivestosetstrictlypositiveadvertisingfees.Strikingly,despitethefactthatallconsumersusethegatekeeperssiteandthuspurchaseatthelowestglobalprice, rmsstillearnpositivepro tsinequilibrium.Inexpectation,thesepro tsareproportionaltothecost,;ofaccessingtheclearinghouse.Conclusion4IntheBayeandMorganmodel,equilibriumpricedispersionpersistsevenwhenitiscostlessforallconsumerstoaccesstheinfor
mationpostedatthegatekeeperssite.I
mationpostedatthegatekeeperssite.Indeed,pricedispersionexistsbecauseitiscostlyfor rmstotransmitpriceinformation(advertiseprices)atthegatekeeperssite.29Undertheseconditions,usingProposition3,weobtainthefollowingcharacterizationofequilib-rium rmpricingandlistingdecisions:Each rmlistsitspriceattheclearinghousewithprobability=1n1(vm)S1Whena rmlistsattheclearinghouse,itchargesapricedrawnfromthedistribution)= 1nn1(pm)S1p0;vwheren1Whena rmdoesnotlistitsprice,itchargesapriceequaltov;andeach rmearnsequilibriumexpectedpro tsequaltoENoticethatrepresentstheaggregatedemandby rmsforadvertisingandisadecreasingfunctionofthefeechargedbythegatekeeper.PricesadvertisedattheclearinghousearedispersedandstrictlylowerthanunadvertisedpricesSeveralfeaturesofthisequilibriumareworthnoting.First,equilibriumpricedispersionariseswithfullyoptimizingconsumers, rms,andendogenousfee-settingdecisionsonthepartoftheclearinghousedespitethefactthattherearenoconsumeror rmheterogeneitiesandallconsumersarefullyinformedinthesensethat,inequilibrium,theyalwayspurchasefroma rmchargingthelowestpriceintheglobalmarket.Second,whileequilibriumpricedispersionintheVarianmodelisdrivenbythefactthatdi¤erentconsumershavedi¤erentcostsofaccessingtheclearinghouse,BayeandMorganshowthatanoptimizingclearinghousewillsetitsfeessu¢cientlylowthatallconsumerswillrationallyaccesstheclearinghouse.Equilibriumpricedispersionarisesbecauseofthegatekeepersincentivestosetstrictlypositiveadvertisingfees.Strikingly,despitethefactthatallconsumersusethegatekeeperssiteandthuspurchaseatthelowestglobalprice, rmsstillearnpositivepro tsinequilibrium.Inexpectation,thesepro tsareproportionaltothecost,accessingtheclearinghouse.Conclusion4IntheBayeandMorganmodel,equilibriumpricedispersionpersistsevenwhenitiscostlessforallconsumerstoaccesstheinformationpostedatthegatekeeperssite.Indeed,pricedispersionexistsbecauseitiscostlyfor rmstotransmitpriceinformation(advertiseprices)atthegatekeeperssite.3.Ifa rmdoesnotlistitspriceattheclearinghouse,itchargesapriceequaltov:4.Firmiearnsequilibriumexpectedpro tsequaltoEi=(vm)Li+Proof.LetAi2f0;1gdenoteanindicatorvariableforwhether rmiadvertisesitspriceattheclearinghouse.LetidenotetheprobabilitythatAi=1:Thus, rmisexpectedpro tsfromeachofthetwolistingdecisionsgiventhestrategy(j;Fj)ofitsrivalareE[i(pjAi=0)]=(Li+(1j)S2)(pm)andE[i(pjAi=1)]=(Li+S(1jFj(p)))(pm)Fromthe rstequation,itisclearthatthedominantstrategyofa rmnotlistingitspriceistochargethemonopolyprice,v:Furthermore,i2(0;1)requiresthatE[i(vj
Ai=0)]=E[i(pjAi=1)](13)forpinthesup
Ai=0)]=E[i(pjAi=1)](13)forpinthesupportofFi:Supposethatthemonopolyprice,v;istheupperboundofthesupportofprices.Inthiscase,theaboveequalityreducesto:(Li+(1j)S2)(vm)=(Li+S(1j))(vm)Theuniquesolutiontothisequationis1=2=;whereisde nedinProposition4.Letp0;idenotethepricesuchthat,if rmichargedthelowestpriceinthemarketandattractedallshoppers,itsexpectedpro tsatpricep0;iwouldexactlyequalthepro tsgainedbysimplypricingatv.Thispricesatis es(Li+(1)S2)(vm)=(Li+S)(p0;im)Substitutingforandsolvingyieldsp0;i=m+Li(vm)+2Li+SNoticethatp0;1p0;2;thatis,thelowestsalepriceo¤eredtoattractshoppersishigherforthelarge rmthanforthesmall rm.Sinceequilibriumpricedistributionsmusthaveidenticalsupports,itfollowsthatFi(p)hassupport[p0;1;v]foralli:Finally,substitutingtheexpression313.Ifa rmdoesnotlistitspriceattheclearinghouse,itchargesapriceequaltov:4.Firmearnsequilibriumexpectedpro tsequaltoE=(Proof.2fdenoteanindicatorvariableforwhether rmadvertisesitspriceattheclearinghouse.Letdenotetheprobabilitythat=1Thus, rmsexpectedpro tsfromeachofthetwolistingdecisionsgiventhestrategy;Fofitsrivalareei(pjAi=0)]=(+(1i(pjAi=1)]=()))(Fromthe rstequation,itisclearthatthedominantstrategyofa rmnotlistingitspriceistochargethemonopolyprice,v:Furthermore,requiresthatthati(vjAi=0)]==i(pjAi=1)]inthesupportofSupposethatthemonopolyprice,v;istheupperboundofthesupportofprices.Inthiscase,theaboveequalityreducesto:+(1)=(Theuniquesolutiontothisequationis;whereisde nedinProposition4.denotethepricesuchthat,if rmchargedthelowestpriceinthemarketandattractedallshoppers,itsexpectedpro tsatpricewouldexactlyequalthepro tsgainedbysimplypricing.Thispricesatis es+(1)=()(Substitutingforandsolvingyields)+2Noticethatp;thatis,thelowestsalepriceo¤eredtoattractshoppersishigherforthelarge rmthanforthesmall rm.Sinceequilibriumpricedistributionsmusthaveidenticalsupports,itfollowsthathassupportrtp0;1;vforallFinally,substitutingtheexpressionEquation(14)showsthatthegatekeeperoptimallychargesafeeto rmsadvertisingpricesthatisproportionaltotheconsumertra¢c(S)onitssite.Thisfeeinducesequilibriumpricedispersion;indeed,as rstnotedbyBayeandMorgan,pricedispersionisnecessaryinorderforthegatekeepertopro tablymaintainaclearinghouse.2.2.5CostHeterogeneitiesandtheSpulberModelSpulber(1995)considersasituationwhereconsumershaveaccesstothecompletelistofpricesandbuyfromthe rmo¤eringthelowestprice.Ofcourse,insuchasetting,if rmswereidenticalonewouldimmediatelyobtaintheBertrandoutcome.Togeneratepricedispersion,Spulberexaminesthesituationwhere rmshavehe
terogeneouscostsandconsumershavedownward
terogeneouscostsandconsumershavedownwardslopingdemand.However,themaineconomicintuitionunderlyingthemodelmaybeseenthroughthefollowingadaptationofourgeneralclearinghouseframeworkfortheunitdemandcase:1.Allconsumersareshoppers:S0andL=0;2.Thereisnocosttoadvertisepricesontheclearinghouse:=0;and3.FirmshaveprivatelyobservedmarginalcostsdescribedbytheatomlessdistributionG(m)on[m;m]:Sincetherearenocoststoadvertiseprices,all rmslistpricesontheclearinghouse.Each rmfacescompetitionfromn1other rmswithrandommarginalcosts.Sincethe rmchargingthelowestpricewinstheentiremarket, rmsaree¤ectivelycompetinginanauctioninwhichtheirowncostsareprivateinformation.Forthespecialcaseofunitdemand,theequilibriumpricefora rmisagainthefamiliarexpressionfroma rst-priceauction:p(m)=Ehm(n1)minjm(n1)minmi(15)wherem(n1)ministhelowestofn1drawsfromthedistributionG:Thereareseveralnoteworthyfeaturesofthisequilibrium.First,equilibrium rmpricingentailspositivemarkupsdespitethefactthatallconsumersareshoppersandhaveacompletelistofprices.Intuitively,thereisatrade-o¤betweenloweringonespricetoattractshoppersandthepro tabilityofthisprice.Inequilibrium,thisresultsinamarkupwhichdependsonthenumberofcompeting rms.Asthenumberof rmsgrowslarge,theequilibriummarkupbecomessmall.Second,noticethatcostheterogeneityleadstoequilibriumpricedispersiondespitethefactconsumersareidenticalandallconsumersarepurchasingatthelowestprice.33showsthatthegatekeeperoptimallychargesafeeto rmsadvertisingpricesthatisproportionaltotheconsumertra¢conitssite.Thisfeeinducesequilibriumpricedispersion;indeed,as rstnotedbyBayeandMorgan,pricedispersionisnecessaryinorderforthegatekeepertopro tablymaintainaclearinghouse.2.2.5CostHeterogeneitiesandtheSpulberModelSpulber(1995)considersasituationwhereconsumershaveaccesstothecompletelistofpricesandbuyfromthe rmo¤eringthelowestprice.Ofcourse,insuchasetting,if rmswereidenticalonewouldimmediatelyobtaintheBertrandoutcome.Togeneratepricedispersion,Spulberexaminesthesituationwhere rmshaveheterogeneouscostsandconsumershavedownwardslopingdemand.However,themaineconomicintuitionunderlyingthemodelmaybeseenthroughthefollowingadaptationofourgeneralclearinghouseframeworkfortheunitdemandcase:1.Allconsumersareshoppers:S=02.Thereisnocosttoadvertisepricesontheclearinghouse:=0;and3.Firmshaveprivatelyobservedmarginalcostsdescribedbytheatomlessdistribution;Sincetherearenocoststoadvertiseprices,all rmslistpricesontheclearinghouse.Each rmfacescompetitionfromother rmswithrandommarginalcosts.Sincethe rmchargingthelowestpricewinstheentiremarket, rmsaree¤ectivelycompetinginanauctioninwhichtheirowncostsareprivateinformation.Forthespecialcaseofunitdemand,t
heequilibriumpricefora rmisagainthe
heequilibriumpricefora rmisagainthefamiliarexpressionfroma rst-priceauction:)=minminwhereministhelowestofdrawsfromthedistributionThereareseveralnoteworthyfeaturesofthisequilibrium.First,equilibrium rmpricingentailspositivemarkupsdespitethefactthatallconsumersareshoppersandhaveacompletelistofprices.Intuitively,thereisatrade-o¤betweenloweringonespricetoattractshoppersandthepro tabilityofthisprice.Inequilibrium,thisresultsinamarkupwhichdependsonthenumberofcompeting rms.Asthenumberof rmsgrowslarge,theequilibriummarkupbecomessmall.Second,noticethatcostheterogeneityleadstoequilibriumpricedispersiondespitethefactconsumersareidenticalandallconsumersarepurchasingatthelowestprice.dispersiondocumentedinlaboratoryexperimentsaswellasobservedonInternetpricecomparisonsites.Inasimilarvein,Rauh(2001)showsthatpricedispersioncanarisewhenmarketparticipantsmakesmallbutheterogeneousmistakesintheirbeliefsaboutthedistributionofprices.Ellison(2005)providesamoredetailedtreatmentofrecentadvancesalongtheselines.2.4ConcludingRemarks:TheoryDespiteaslowstart,therearenowavarietyofmodelsthatcanbeusedtorationalizeequilibriumpricedispersioninonlineandoinemarkets.Weconcludeourtheoreticaldiscussionwiththefollowinggeneralobservations:1.Thereisnotaone-size- ts-allmodelofequilibriumpricedispersion;di¤erentmodelsareappropriateforanalyzingdi¤erentmarketenvironments.Forinstance,search-theoreticmod-elsaremostappropriateforanalyzingenvironmentswhereconsumersmustvisitdi¤erentstoresor rmswebsitestogatherpriceinformation.Clearinghousemodelsareappropriatewhenconsumersareabletoaccessalistofprices(forexample,inanewspaperoratapricecomparisonsite).2.Thedistributionofpricesisdeterminedbytheinteractionofallmarketparticipants rms,consumersand,inthecaseofclearinghousemodels,informationgatekeepers.Asaconse-quence,thelevelofpricedispersiondependsonthestructureofthemarketthenumberofsellers,thedistributionofcosts,consumerselasticitiesofdemand,andsoon.3.Reductionsinsearchcostsmayleadtoeithermoreorlesspricedispersion,dependingonthemarketenvironment.Furthermore,theeliminationofconsumersearchcostsneednoteliminatepricedispersion.4.Dependingonthemarketenvironment,heightenedcompetition(increasesinthenumberof rms)canincreaseordecreasethelevelofdispersion.Moreover,insomemodels,heightenedcompetitionofthisformleadstohighertransactionspricespaidbyallconsumers.Inothermodels,thee¤ectofincreasedcompetitiononthewelfareofconsumersdependsonwhichsideofthedigitaldivideaconsumerresides.5.Pricedispersionisnotpurelyanartifactofexanteheterogeneitiesin rmsorconsumers.Whiledi¤erencesin rmscostsorbaseofloyalconsumers(stemmingfrom rmsbrand-35dispersiondocumentedinlaboratoryexperimentsaswellasobserve
donInternetpricecomparisonsites.Inasimil
donInternetpricecomparisonsites.Inasimilarvein,Rauh(2001)showsthatpricedispersioncanarisewhenmarketparticipantsmakesmallbutheterogeneousmistakesintheirbeliefsaboutthedistributionofprices.Ellison(2005)providesamoredetailedtreatmentofrecentadvancesalongtheselines.2.4ConcludingRemarks:TheoryDespiteaslowstart,therearenowavarietyofmodelsthatcanbeusedtorationalizeequilibriumpricedispersioninonlineandoinemarkets.Weconcludeourtheoreticaldiscussionwiththefollowinggeneralobservations:1.Thereisnotaone-size- ts-allmodelofequilibriumpricedispersion;di¤erentmodelsareappropriateforanalyzingdi¤erentmarketenvironments.Forinstance,search-theoreticmod-elsaremostappropriateforanalyzingenvironmentswhereconsumersmustvisitdi¤erentstoresor rmswebsitestogatherpriceinformation.Clearinghousemodelsareappropriatewhenconsumersareabletoaccessalistofprices(forexample,inanewspaperoratapricecomparisonsite).2.Thedistributionofpricesisdeterminedbytheinteractionofallmarketparticipants rms,consumersand,inthecaseofclearinghousemodels,informationgatekeepers.Asaconse-quence,thelevelofpricedispersiondependsonthestructureofthemarketthenumberofsellers,thedistributionofcosts,consumerselasticitiesofdemand,andsoon.3.Reductionsinsearchcostsmayleadtoeithermoreorlesspricedispersion,dependingonthemarketenvironment.Furthermore,theeliminationofconsumersearchcostsneednoteliminatepricedispersion.4.Dependingonthemarketenvironment,heightenedcompetition(increasesinthenumberof rms)canincreaseordecreasethelevelofdispersion.Moreover,insomemodels,heightenedcompetitionofthisformleadstohighertransactionspricespaidbyallconsumers.Inothermodels,thee¤ectofincreasedcompetitiononthewelfareofconsumersdependsonwhichsideofthedigitaldivideaconsumerresides.5.Pricedispersionisnotpurelyanartifactofexanteheterogeneitiesin rmsorconsumers.Whiledi¤erencesin rmscostsorbaseofloyalconsumers(stemmingfrom rmsbrand-costsofall rmsintheMacMinnmodelincreasedbyafactor 1:Inthatcase,thenewvariancewouldsimplyscaleuptheoriginalvariancebyafactor 2:Thus,thismeasureofpricedispersionwouldchangeeventhoughtheunderlyingrealeconomicsofthesituationarethesameaftertheinationaryperiod.Forthisreason,ifonewishestocomparelevelsofpricedispersioneitheracrossdi¤erentproductsoracrosstime,onemuststandardizethedatainsomefashion.Analternativeistousethecoe¢cientofvariation,CV=p=E[p](oritssampleanalogue),whichishomogenousofdegreezerointhelevelofprices.TheCVisparticularlyusefulwhencomparinglevelsofpricedispersionoverlongperiodsoftime(e.g.,ScholtenandSmith(2002)andEckard(2004))oracrossdi¤erentproducts(e.g.,CarlsonandPescatrice(1980);Sorensen(2000);Aalto-Setälä(2003);andBaye,MorganandScholten(2004a,b)).Anaddedadvantageisthat,unlikesomemethodsofstandard
ization,thecoe¢cientofvariationmayprese
ization,thecoe¢cientofvariationmaypreservethecomparativestaticpredictionsofthemodelofinterest.Forinstance,intheMacMinnmodel,equation(8)impliesthattheexpectedpriceisE[p]=n1nm+m2+mn,andthusthecoe¢cientofvariationisCV=1p3(n1)(mm)(n1)(m+m)+2mOnemayverifythatthisstatisticis,likethevariance,decreasinginsearchcosts,but,unlikethevariance,thisstatisticdoesnotchangewithamultiplicativeshiftin rmscosts.Anotherwidelyusedmeasureofpricedispersionisthe(sample)range;see,forinstance,Pratt,Wise,andZeckhauser(1979)andBrynjolfssonandSmith(2000).Lettingp(n)minandp(n)maxdenote,respectively,thelowestandhighestofnobservedpricesdrawnfromF;thentherangeisR(n)=p(n)maxp(n)minGiventheequilibriumdistributionofpricesimpliedbyaparticulartheoreticalmodel,comparativestaticpredictionsaboutchangesintherangearepossiblebasedonthebehaviorofthehighestandlowestorderstatistics.Thatis,onecanperformcomparativestaticanalysisontheexpectedrange:20EhR(n)i=Ehp(n)maxiEhp(n)miniUnfortunately,alloftheabovemeasuresofpricedispersionsu¤erfromapotentialtheoreticaldefect.Supposethatn2 rmscompeteinaclassicalhomogeneousproductBertrandsetting.20Tofacilitatecomparisonsacrossdi¤erentproductsorovertime,itissometimesusefultonormalizetherangebydividingitbytheminimumoraverageprice;seeBaye,Morgan,andScholten(2004b)andBrynjolfssonandSmith(2000).37costsofall rmsintheMacMinnmodelincreasedbyafactor\rInthatcase,thenewvariancewouldsimplyscaleuptheoriginalvariancebyafactorThus,thismeasureofpricedispersionwouldchangeeventhoughtheunderlyingrealeconomicsofthesituationarethesameaftertheinationaryperiod.Forthisreason,ifonewishestocomparelevelsofpricedispersioneitheracrossdi¤erentproductsoracrosstime,onemuststandardizethedatainsomefashion.Analternativeistousethecoe¢cientofvariation,CVVp](oritssampleanalogue),whichishomogenousofdegreezerointhelevelofprices.TheCVisparticularlyusefulwhencomparinglevelsofpricedispersionoverlongperiodsoftime(e.g.,ScholtenandSmith(2002)andEckard(2004))oracrossdi¤erentproducts(e.g.,CarlsonandPescatrice(1980);Sorensen(2000);Aalto-Setälä(2003);andBaye,MorganandScholten(2004a,b)).Anaddedadvantageisthat,unlikesomemethodsofstandardization,thecoe¢cientofvariationmaypreservethecomparativestaticpredictionsofthemodelofinterest.Forinstance,intheMacMinnmodel,equationimpliesthattheexpectedpriceisisp]=nm+m2+m,andthusthecoe¢cientofvariationisCVp1)(mm)1)(m+m)+2Onemayverifythatthisstatisticis,likethevariance,decreasinginsearchcosts,but,unlikethevariance,thisstatisticdoesnotchangewithamultiplicativeshiftin rmscosts.Anotherwidelyusedmeasureofpricedispersionisthe(sample)range;see,forinstance,Pratt,Wise,andZeckhauser(1979)andBrynjolfssonandSmith(2000).Lettingminmaxdenote,re
spectively,thelowestandhighestofobserved
spectively,thelowestandhighestofobservedpricesdrawnfromF;thentherangeismaxminGiventheequilibriumdistributionofpricesimpliedbyaparticulartheoreticalmodel,comparativestaticpredictionsaboutchangesintherangearepossiblebasedonthebehaviorofthehighestandlowestorderstatistics.Thatis,onecanperformcomparativestaticanalysisontheexpectedrange:20maxminUnfortunately,alloftheabovemeasuresofpricedispersionsu¤erfromapotentialtheoreticaldefect.Supposethatn rmscompeteinaclassicalhomogeneousproductBertrandsetting.20Tofacilitatecomparisonsacrossdi¤erentproductsorovertime,itissometimesusefultonormalizetherangebydividingitbytheminimumoraverageprice;seeBaye,Morgan,andScholten(2004b)andBrynjolfssonandSmith(2000).Forexample,inhisseminalarticleontheeconomicsofinformation,GeorgeStigler,advancedthefollowinghypotheses:...dispersionitselfisafunctionoftheaverageamountofsearch,andthisinturnisafunctionofthenatureofthecommodity:1.Thelargerthefractionofthebuyersexpendituresonthecommodity,thegreaterthesavingsfromsearchandhencethegreatertheamountofsearch.2.Thelargerthefractionofrepetitive(experienced)buyersinthemarket,thegreaterthee¤ectiveamountofsearch(withpositivecorrelationofsuccessiveprices).3.Thelargerthefractionofrepetitivesellers,thehigherthecorrelationbetweensuccessiveprices,andhence,thelargertheamountofaccumulatedsearch.4.Thecostofsearchwillbelarger,thelargerthegeographicsizeofthemarket.Stigler(1961,p.219).Stiglershypotheseso¤erausefulguideforunderstandingtheempiricalliteratureonpricedispersion.MuchofthisliteraturetestsStiglershypothesesbyexaminingwhethersearchintensity(proxiedbyvariablesthata¤ectthebene tsandcostsofsearch)iscorrelatedwithlevelsofpricedispersion.Aswehaveseen,however,whenonetakesRothschildscriticismintoaccount,anincreaseinsearchintensitycanleadtoincreasesordecreasesinthelevelofequilibriumpricedispersion,dependingonthemodel.Thus,onechallengeforempiricalresearchersischoosingamodelthatcloselyapproximatesthedatageneratingenvironment.Asecondchallengeistocontrolforfactorsoutsideofthemodelthatmightinuencelevelsofdispersion.Athirdchallengearisesbecause rmoptimizationisabsentinStiglersmodel,butisclearlypresentinthedata.Forthisreason,anumberofempiricalstudieslookbeyondStiglershypothesestotesthypothesesderivedfromspeci csearch-theoreticorclearinghousemodelsofequilibriumpricedispersion.Weprovideabroadoverviewoftheseandrelatedstrandsoftheliteraturebelow.3.2.1DispersionandtheBene tsofSearchThesearch-theoreticmodelspresentedinSection2implythatsearchintensitydepends,inpart,ontheconsumersdemandforaproduct.IntheStiglermodel,demandisrepresentedbytheparameter,K.ThegreaterisK;thegreatertheexpectedbene tsofsearchandhencethegreaterthesearchintensity.Stiglers r
sttwohypothesesarebasedonthenotionthatth
sttwohypothesesarebasedonthenotionthattheshareofan39Forexample,inhisseminalarticleontheeconomicsofinformation,GeorgeStigler,advancedthefollowinghypotheses:...dispersionitselfisafunctionoftheaverageamountofsearch,andthisinturnisafunctionofthenatureofthecommodity:1.Thelargerthefractionofthebuyersexpendituresonthecommodity,thegreaterthesavingsfromsearchandhencethegreatertheamountofsearch.2.Thelargerthefractionofrepetitive(experienced)buyersinthemarket,thegreaterthee¤ectiveamountofsearch(withpositivecorrelationofsuccessiveprices).3.Thelargerthefractionofrepetitivesellers,thehigherthecorrelationbetweensuccessiveprices,andhence,thelargertheamountofaccumulatedsearch.4.Thecostofsearchwillbelarger,thelargerthegeographicsizeofthemarket.Stigler(1961,p.219).Stiglershypotheseso¤erausefulguideforunderstandingtheempiricalliteratureonpricedispersion.MuchofthisliteraturetestsStiglershypothesesbyexaminingwhethersearchintensity(proxiedbyvariablesthata¤ectthebene tsandcostsofsearch)iscorrelatedwithlevelsofpricedispersion.Aswehaveseen,however,whenonetakesRothschildscriticismintoaccount,anincreaseinsearchintensitycanleadtoincreasesordecreasesinthelevelofequilibriumpricedispersion,dependingonthemodel.Thus,onechallengeforempiricalresearchersischoosingamodelthatcloselyapproximatesthedatageneratingenvironment.Asecondchallengeistocontrolforfactorsoutsideofthemodelthatmightinuencelevelsofdispersion.Athirdchallengearisesbecause rmoptimizationisabsentinStiglersmodel,butisclearlypresentinthedata.Forthisreason,anumberofempiricalstudieslookbeyondStiglershypothesestotesthypothesesderivedfromspeci csearch-theoreticorclearinghousemodelsofequilibriumpricedispersion.Weprovideabroadoverviewoftheseandrelatedstrandsoftheliteraturebelow.3.2.1DispersionandtheBene tsofSearchThesearch-theoreticmodelspresentedinSection2implythatsearchintensitydepends,inpart,ontheconsumersdemandforaproduct.IntheStiglermodel,demandisrepresentedbythe.ThegreaterisK;thegreatertheexpectedbene tsofsearchandhencethegreaterthesearchintensity.Stiglers rsttwohypothesesarebasedonthenotionthattheshareofanSmith(2002),Johnson(2002),GattiandKattuman(2003),andAalto-Setälä(2003).Morere-cently,Eckard(2004)comparespricedispersionforstapleproductsin1901and2001,andreportscoe¢cientsofvariationin2001thatarealmosttwicethosebasedondatafrom1901.Eckardarguesthatonereasonfortheincreaseddispersionisthathissampleconsistsofstapleitems(suchassugarandbakingpowder)thataccountedforamuchlargershareofhouseholdbudgetsin1901thanin2001.DispersionandPurchaseFrequencyInhissecondhypothesis,Stiglerarguesthatinmarketswheretherearemorerepetitiveorexperiencedbuyers,thegreateristheamountofe¤ectivesearch.Unfortunately,itisdi¢culttodirectlytestth
ishypotheses,sinceinmostmarketsthereisno
ishypotheses,sinceinmostmarketsthereisnotadirect(objective)measureofbuyerexperienceorpurchasefrequencytouseinexaminingitsimpactonlevelsofpricedispersion.Anumberofthestudiesmentionedabove,however,providecasualevidencethatpurchasefrequencyimpactsthelevelofpricedispersion(cf.CarlsonandPescatrice,1980;Pratt,Wise,andZeckhauser,1979).Sorensen(2000),however,hasprovidedaverycleanandeleganttestofStiglerssecondhypothesis.Hisanalysisisbasedondatafromthemarketforprescriptiondrugs.Theuniqueaspectofthismarketisthatpurchasefrequencythetypicaldosageanddurationoftherapyforagivenprescriptiondrugmaybeobjectivelymeasured.Aconsumersbene tpersearchisclearlyhighestforfrequentlypurchaseddrugs,and,Sorensenargues,thisshouldleadtogreatersearchandlowerpricedispersion.Hisempiricalanalysisidenti esastronginverserelationshipbetweenpurchasefrequencyandpricedispersion.Forexample,aftercontrollingotherfactors(whichtogetherexplainaboutone-thirdofthevariationinprices),Sorensen ndsthatthepricerangeforadrugthatmustbepurchasedmonthlyisabout30percentlowerthanifitwereaone-timetherapy.Importantly,Sorensenshowsthattheresultsarequalitativelysimilarwhenalternativemeasuresofpricedispersion(suchasthestandarddeviation)areused.3.2.2DispersionandtheCostofSearchResearchersstudyingtheempiricalrelationshipbetweensearchcostsandpricedispersionhavefacedobstaclessimilartothoseofresearchersfocusingonthebene tsideofthesearchequation.First,thepredictedimpactofsearchcostsonlevelsofdispersiondependsnotonlyonthemodel,butalsoonthemetricusedformeasuringdispersion.Second,searchcostsaregenerallyunobservable.Someofthemoreinuentialpapersintheareaareonesthathavedevisedinnovativemethodsof41Smith(2002),Johnson(2002),GattiandKattuman(2003),andAalto-Setälä(2003).Morere-cently,Eckard(2004)comparespricedispersionforstapleproductsin1901and2001,andreportscoe¢cientsofvariationin2001thatarealmosttwicethosebasedondatafrom1901.Eckardarguesthatonereasonfortheincreaseddispersionisthathissampleconsistsofstapleitems(suchassugarandbakingpowder)thataccountedforamuchlargershareofhouseholdbudgetsin1901thaninDispersionandPurchaseFrequencyInhissecondhypothesis,Stiglerarguesthatinmarketswheretherearemorerepetitiveorexperiencedbuyers,thegreateristheamountofe¤ectivesearch.Unfortunately,itisdi¢culttodirectlytestthishypotheses,sinceinmostmarketsthereisnotadirect(objective)measureofbuyerexperienceorpurchasefrequencytouseinexaminingitsimpactonlevelsofpricedispersion.Anumberofthestudiesmentionedabove,however,providecasualevidencethatpurchasefrequencyimpactsthelevelofpricedispersion(cf.CarlsonandPescatrice,1980;Pratt,Wise,andZeckhauser,1979).Sorensen(2000),however,hasprovidedaverycleanandeleganttestofStiglerssecondhypo
thesis.Hisanalysisisbasedondatafromthema
thesis.Hisanalysisisbasedondatafromthemarketforprescriptiondrugs.Theuniqueaspectofthismarketisthatpurchasefrequencythetypicaldosageanddurationoftherapyforagivenprescriptiondrugmaybeobjectivelymeasured.Aconsumersbene tpersearchisclearlyhighestforfrequentlypurchaseddrugs,and,Sorensenargues,thisshouldleadtogreatersearchandlowerpricedispersion.Hisempiricalanalysisidenti esastronginverserelationshipbetweenpurchasefrequencyandpricedispersion.Forexample,aftercontrollingotherfactors(whichtogetherexplainaboutone-thirdofthevariationinprices),Sorensen ndsthatthepricerangeforadrugthatmustbepurchasedmonthlyisabout30percentlowerthanifitwereaone-timetherapy.Importantly,Sorensenshowsthattheresultsarequalitativelysimilarwhenalternativemeasuresofpricedispersion(suchasthestandarddeviation)areused.3.2.2DispersionandtheCostofSearchResearchersstudyingtheempiricalrelationshipbetweensearchcostsandpricedispersionhavefacedobstaclessimilartothoseofresearchersfocusingonthebene tsideofthesearchequation.First,thepredictedimpactofsearchcostsonlevelsofdispersiondependsnotonlyonthemodel,butalsoonthemetricusedformeasuringdispersion.Second,searchcostsaregenerallyunobservable.SomeofthemoreinuentialpapersintheareaareonesthathavedevisedinnovativemethodsofAlongthesesamelines,anumberofstudiescompareaveragepricesinonlineversusoinemarkets.Theideaisthatsearchcostsareloweronline,thusa¤ectingnotonlytherangeorvarianceinprices,butalsothemeanprice(andhencethecoe¢cientofvariationthroughboththemeanandvariance).Scott-Morton,ZettelmeyerandSilva-Risso(2001) ndthatpricesarelowerinonlinemarketsforautomobiles.ConsumerswhopurchaseacarthroughtheInternetreferralserviceAutobytel.comreducetheirpurchasepricebyapproximately2.2percent.Apotentiallyconfoundingexplanationforthispricedi¤erenceisthattheconsumerswhochoosetoshoponlinemayalsobeskilledhigglers,touseStiglersphraseandthusthepricedi¤erencemightpurelyreectadi¤erenceinthenegotiatingskillsofconsumersacrossthetwochannels.Interestingly,Zettelmeyer,Scott-MortonandSilva-Risso(2004)provideevidencethatthisisnotthecase:consumerswhopurchaseautomobilesonlinearenottypicallythosewhonegotiatewellinthetraditionalchannel.Thereareanumberofotherstudies,however,that ndequalorhigherpricesonline(cf.Clemons,HannandHitt(2002);Bailey(1998);Goolsbee(2001);Clay,etal.(2003);Erevelles,RollandandSrinivasan(2001)).Furtherstudiesdistinguishpricelevelsdependingonwhethertheretailerisasolelyonlineormultichannel(cf.ChevalierandGoolsbee(2003)and;TangandXing(2001)).Analternativeapproachistorecoversearchcostsusingstructuralparametersfromapar-ticularmodelofpricedispersion.Forexample,HongandShum(forthcoming)obtainsearchcostsestimatesusingrestrictionsimposedbytheoreticalsearchmodelsandassumi
ngthatobservedpricedispersionisanequilib
ngthatobservedpricedispersionisanequilibriumphenomenonarisingfromheterogeneousconsumersearchcosts.Theirestimationtechniqueisappliedtoonlinepricedataonfoureconomicsandstatisticstextbooks.Theyobtainsearchcostestimatesrangingfrom$1.31to$29.40fortheseitems.Asimilarapproachcanbeusedinclearinghousemodels.Villas-Boas(1995)usesthetheoreticaldensityfunctionim-pliedbytheVarian(1980)clearinghousemodeltoobtainestimatesofthenumberofshoppersintheoineco¤eeandsaltinecrackermarkets.Morerecently,Baye,Gatti,Kattuman,andMorgan(2005)usedatheoreticalclearinghousemodelasthebasisforestimatingthefractionofshop-persinanonlinemarketforPDAsintheUK.Theirresultssuggestthatabout13percentoftheconsumersinthismarketareshoppers.3.2.3DispersionandtheNumberofSellersTheoligopolymodelspresentedinSection2revealthatequilibriumdistributionsofprices,andhencelevelsofdispersion,varywiththenumberofsellerscompetinginthemarket.Thedirectioninwhichpricesmoveasaconsequenceofachangeinthenumberofsellersis,however,modelspeci c,43Alongthesesamelines,anumberofstudiescompareaveragepricesinonlineversusoinemarkets.Theideaisthatsearchcostsareloweronline,thusa¤ectingnotonlytherangeorvarianceinprices,butalsothemeanprice(andhencethecoe¢cientofvariationthroughboththemeanandvariance).Scott-Morton,ZettelmeyerandSilva-Risso(2001) ndthatpricesarelowerinonlinemarketsforautomobiles.ConsumerswhopurchaseacarthroughtheInternetreferralserviceAutobytel.comreducetheirpurchasepricebyapproximately2.2percent.Apotentiallyconfoundingexplanationforthispricedi¤erenceisthattheconsumerswhochoosetoshoponlinemayalsobeskilledhigglers,touseStiglersphraseandthusthepricedi¤erencemightpurelyreectadi¤erenceinthenegotiatingskillsofconsumersacrossthetwochannels.Interestingly,Zettelmeyer,Scott-MortonandSilva-Risso(2004)provideevidencethatthisisnotthecase:consumerswhopurchaseautomobilesonlinearenottypicallythosewhonegotiatewellinthetraditionalchannel.Thereareanumberofotherstudies,however,that ndequalorhigherpricesonline(cf.Clemons,HannandHitt(2002);Bailey(1998);Goolsbee(2001);Clay,etal.(2003);Erevelles,RollandandSrinivasan(2001)).Furtherstudiesdistinguishpricelevelsdependingonwhethertheretailerisasolelyonlineormultichannel(cf.ChevalierandGoolsbee(2003)and;TangandXing(2001)).Analternativeapproachistorecoversearchcostsusingstructuralparametersfromapar-ticularmodelofpricedispersion.Forexample,HongandShum(forthcoming)obtainsearchcostsestimatesusingrestrictionsimposedbytheoreticalsearchmodelsandassumingthatobservedpricedispersionisanequilibriumphenomenonarisingfromheterogeneousconsumersearchcosts.Theirestimationtechniqueisappliedtoonlinepricedataonfoureconomicsandstatisticstextbooks.Theyobtainsearchcostestimatesrangingfrom$1.31to$29.40fortheseitems.Asimilarapproachcanbeusedincl
earinghousemodels.Villas-Boas(1995)usest
earinghousemodels.Villas-Boas(1995)usesthetheoreticaldensityfunctionim-pliedbytheVarian(1980)clearinghousemodeltoobtainestimatesofthenumberofshoppersintheoineco¤eeandsaltinecrackermarkets.Morerecently,Baye,Gatti,Kattuman,andMorgan(2005)usedatheoreticalclearinghousemodelasthebasisforestimatingthefractionofshop-persinanonlinemarketforPDAsintheUK.Theirresultssuggestthatabout13percentoftheconsumersinthismarketareshoppers.3.2.3DispersionandtheNumberofSellersTheoligopolymodelspresentedinSection2revealthatequilibriumdistributionsofprices,andhencelevelsofdispersion,varywiththenumberofsellerscompetinginthemarket.Thedirectioninwhichpricesmoveasaconsequenceofachangeinthenumberofsellersis,however,modelspeci c,dispersioninfaresincreasesonrouteswithlowerightdensityormorecompetition.Thus,thereisevidencethatthenumberofsellersmattersforpricedispersion.3.2.4DispersionandPricePersistenceVarian(1980)wasthe rsttodistinguishbetweenwhathereferredtoasspatialandtemporalpricedispersion.Underspatialpricedispersion,di¤erent rmschargedi¤erentpricesatanypointintime,buta rmspositioninthedistributionofpricesdoesnotchangeovertime.Absentrandomcostshocks,spatialpricedispersionarisesintheReinganum,MacMinn,andSpulbermodels.Incontrast,withtemporalpricedispersion, rmschargedi¤erentpricesateachpointintime,buttheirpositioninthedistributionofpriceschangesovertime.Temporalpricedispersionarisesinthegeneralclearinghousemodel(andvariousspecialcases)aswellasintheBurdettandJuddmodel.Variancritiquesmodelsofspatialpricedispersion,arguingthatifconsumerscanlearnfromexperiencethatsome rmspersistentlyo¤erlowerpricesthanother rms,thenmodelsofspatialpricedispersionsuggestaconvergencehypothesis:pricedispersionshoulddiminishovertimeduetothepositivecorrelationinsuccessiveprices(touseStiglersterminology)andcumulativesearchinformation.Thishasledtoanumberofstudiesthatexaminewhetherthereisanyevidencefortheconvergencehypothesisandwhetherthetemporalpricedispersionpredictedbytheclearinghousemodelsis,infact,presentinthedata.Usingmonthlystore-levelpricedatafromIsrael,andaftercontrollingforobservedandun-observedproductheterogeneities,Lach(2002) ndssomeevidenceoftemporalpricedispersion.Lachestimatesmonth-to-monthtransitionsamongquartilesby rms;thatis,theprobabilitythata rmo¤eringapriceinagivenquartileatthestartofthemonthisstillo¤eringapriceinthesamequartileattheendofthemonth.Hisestimatessuggestthattheprobabilityofremaininginthesamequartileis78percentfor rmssellingrefrigeratorsand71percentfor rmssellingour.Theseprobabilitiesaresomewhatlowerfor rmssellingchicken(51percent)andco¤ee(43per-cent).Whenthetransitionperiodisextendedtosixmonthsinsteadofonemonth,theprobabilityofremaininginthesame
quartileisconsiderablylowerfalling
quartileisconsiderablylowerfallingtoaround30-35percent.RobertsandSupina(2000)suggestthatstructuraldi¤erencesin rmscostsaccountforaconsiderableportionofpricedispersionintheoinesectoraspredictedbyavarietyofsearch-theoreticmodels.Usingplant-levelUSCensusdata,they ndsomeevidenceforpricepersistence.Theevidenceisstrongestinthetailsofthedistribution:high-price rmstendtopersistentlychargehighprices,andlow-price rmstendstopersistentlychargelowprices.Avarietyofotherstudies45dispersioninfaresincreasesonrouteswithlowerightdensityormorecompetition.Thus,thereisevidencethatthenumberofsellersmattersforpricedispersion.3.2.4DispersionandPricePersistenceVarian(1980)wasthe rsttodistinguishbetweenwhathereferredtoasspatialandtemporalpricedispersion.Underspatialpricedispersion,di¤erent rmschargedi¤erentpricesatanypointintime,buta rmspositioninthedistributionofpricesdoesnotchangeovertime.Absentrandomcostshocks,spatialpricedispersionarisesintheReinganum,MacMinn,andSpulbermodels.Incontrast,withtemporalpricedispersion, rmschargedi¤erentpricesateachpointintime,buttheirpositioninthedistributionofpriceschangesovertime.Temporalpricedispersionarisesinthegeneralclearinghousemodel(andvariousspecialcases)aswellasintheBurdettandJuddmodel.Variancritiquesmodelsofspatialpricedispersion,arguingthatifconsumerscanlearnfromexperiencethatsome rmspersistentlyo¤erlowerpricesthanother rms,thenmodelsofspatialpricedispersionsuggestaconvergencehypothesis:pricedispersionshoulddiminishovertimeduetothepositivecorrelationinsuccessiveprices(touseStiglersterminology)andcumulativesearchinformation.Thishasledtoanumberofstudiesthatexaminewhetherthereisanyevidencefortheconvergencehypothesisandwhetherthetemporalpricedispersionpredictedbytheclearinghousemodelsis,infact,presentinthedata.Usingmonthlystore-levelpricedatafromIsrael,andaftercontrollingforobservedandun-observedproductheterogeneities,Lach(2002) ndssomeevidenceoftemporalpricedispersion.Lachestimatesmonth-to-monthtransitionsamongquartilesby rms;thatis,theprobabilitythata rmo¤eringapriceinagivenquartileatthestartofthemonthisstillo¤eringapriceinthesamequartileattheendofthemonth.Hisestimatessuggestthattheprobabilityofremaininginthesamequartileis78percentfor rmssellingrefrigeratorsand71percentfor rmssellingour.Theseprobabilitiesaresomewhatlowerfor rmssellingchicken(51percent)andco¤ee(43per-cent).Whenthetransitionperiodisextendedtosixmonthsinsteadofonemonth,theprobabilityofremaininginthesamequartileisconsiderablylowerfallingtoaround30-35percent.RobertsandSupina(2000)suggestthatstructuraldi¤erencesin rmscostsaccountforaconsiderableportionofpricedispersionintheoinesectoraspredicte
dbyavarietyofsearch-theoreticmodels.Usin
dbyavarietyofsearch-theoreticmodels.Usingplant-levelUSCensusdata,they ndsomeevidenceforpricepersistence.Theevidenceisstrongestinthetailsofthedistribution:high-price rmstendtopersistentlychargehighprices,andlow-price rmstendstopersistentlychargelowprices.Avarietyofotherstudies3.Therelationshipbetweenpricedispersionandeconomicprimitivesisoftensensitivetothemeasureofpricedispersionused.4.Despitethewidespreadadoptionofinventionssuchastheautomobile,thetelephone,televi-sion,andtheInternet,pricedispersionisstilltheruleratherthantheexceptioninhomoge-neousproductmarkets.Reductionsininformationcostsoverthepastcenturyhaveneitherreducednoreliminatedthelevelsofpricedispersionobservedinhomogeneousproductmar-kets.473.Therelationshipbetweenpricedispersionandeconomicprimitivesisoftensensitivetothemeasureofpricedispersionused.4.Despitethewidespreadadoptionofinventionssuchastheautomobile,thetelephone,televi-sion,andtheInternet,pricedispersionisstilltheruleratherthantheexceptioninhomoge-neousproductmarkets.Reductionsininformationcostsoverthepastcenturyhaveneitherreducednoreliminatedthelevelsofpricedispersionobservedinhomogeneousproductmar-kets.Baye,M.R.andJ.Morgan.2004.PriceDispersionintheLabandontheInternet:TheoryandEvidence.RandJournalofEconomics,35(3),449-466.Baye,M.R.,J.MorganandP.Scholten.2003.TheValueofInformationinanOnlineConsumerElectronicsMarket.JournalofPublicPolicyandMarketing,22(1),17-25.Baye,M.R.,J.MorganandP.Scholten.2004a.PriceDispersionintheSmallandintheLarge:EvidencefromanInternetPriceComparisonSite.JournalofIndustrialEconomics,52(4),463-496.Baye,M.R.,J.MorganandP.Scholten.2004b.TemporalPriceDispersion:EvidencefromanOnlineConsumerElectronicsMarket.JournalofInteractiveMarketing,18(4),101-115.Baylis,K.andJ.M.Perlo¤.2002.PriceDispersionontheInternet:GoodFirmsandBadFirms.ReviewofIndustrialOrganization,21,305-324.Benabou,R.andR.Gertner.1993.SearchwithLearningfromPrices:DoesIncreasedUncertaintyLeadtoHigherMarkups?ReviewofEconomicStudies,60(202),69-94.Borenstein,S.andN.L.Rose.1994.CompetitionandPriceDispersionintheU.S.AirlineIndus-try.JournalofPoliticalEconomy,102(4),653-683.Braverman,A.1980.ConsumerSearchandAlternativeMarketEquilibria.ReviewofEconomicStudies47,487-502.Brown,J.R.andA.Goolsbee.2002.DoestheInternetMakeMarketsMoreCompetitive?EvidencefromtheLifeInsuranceIndustry.JournalofPoliticalEconomy,110(3),481-507.Brynjolfsson,E.andM.D.Smith.2000.FrictionlessCommerce?AComparisonofInternetandConventionalRetailers.ManagementScience,46(4),563-585.Brynjolfsson,E.,A.A.DickandM.D.Smith.2004.SearchandProductDi¤erentiationatanInternetShopbot.WorkingPaper.Burdett,K.andK.L.Judd.1983.EquilibriumPriceDispersion.Econometri
ca,51,955-969.Butters,G.R.1977.Equ
ca,51,955-969.Butters,G.R.1977.EquilibriumDistributionsofSalesandAdvertisingPrices.ReviewofEco-nomicStudies,44,465-491.49Baye,M.R.andJ.Morgan.2004.PriceDispersionintheLabandontheInternet:TheoryandEvidence.RandJournalofEconomics,35(3),449-466.Baye,M.R.,J.MorganandP.Scholten.2003.TheValueofInformationinanOnlineConsumerElectronicsMarket.JournalofPublicPolicyandMarketing,22(1),17-25.Baye,M.R.,J.MorganandP.Scholten.2004a.PriceDispersionintheSmallandintheLarge:EvidencefromanInternetPriceComparisonSite.JournalofIndustrialEconomics,52(4),463-Baye,M.R.,J.MorganandP.Scholten.2004b.TemporalPriceDispersion:EvidencefromanOnlineConsumerElectronicsMarket.JournalofInteractiveMarketing,18(4),101-115.Baylis,K.andJ.M.Perlo¤.2002.PriceDispersionontheInternet:GoodFirmsandBadFirms.ReviewofIndustrialOrganization,21,305-324.Benabou,R.andR.Gertner.1993.SearchwithLearningfromPrices:DoesIncreasedUncertaintyLeadtoHigherMarkups?ReviewofEconomicStudies,60(202),69-94.Borenstein,S.andN.L.Rose.1994.CompetitionandPriceDispersionintheU.S.AirlineIndus-try.JournalofPoliticalEconomy,102(4),653-683.Braverman,A.1980.ConsumerSearchandAlternativeMarketEquilibria.ReviewofEconomicStudies47,487-502.Brown,J.R.andA.Goolsbee.2002.DoestheInternetMakeMarketsMoreCompetitive?EvidencefromtheLifeInsuranceIndustry.JournalofPoliticalEconomy,110(3),481-507.Brynjolfsson,E.andM.D.Smith.2000.FrictionlessCommerce?AComparisonofInternetandConventionalRetailers.ManagementScience,46(4),563-585.Brynjolfsson,E.,A.A.DickandM.D.Smith.2004.SearchandProductDi¤erentiationatanInternetShopbot.WorkingPaperBurdett,K.andK.L.Judd.1983.EquilibriumPriceDispersion.Econometrica,51,955-969.Butters,G.R.1977.EquilibriumDistributionsofSalesandAdvertisingPrices.ReviewofEco-nomicStudies,44,465-491.Eckard,E.W.2004.TheLawofOnePricein1901.EconomicInquiry,42(1),101-110.Ellison,G.2005.BoundedRationalityinIndustrialOrganization,MITWorkingPaper.Ellison,G.andS.Ellison.2004.Search,Obfuscation,andPriceElasticitiesontheInternet.MITWorkingPaper.Erevelles,S.,E.RollandandS.Srinivasan.2001.ArePricesReallyLowerontheInternet?:AnAnalysisoftheVitaminIndustry.WorkingPaper.Friberg,R.,M.GanslandtandM.Sandström.2001.PricingStrategiesinE-Commerce:Bricksvs.Clicks.WorkingPaper.Gatti,R.andP.Kattuman.2003.OnlinePriceDispersionWithinandBetweenSevenEuropeanCountries.AdvancesinAppliedMicroeconomics,12,107-143.Goolsbee,A.2001.CompetitionintheComputerIndustry:OnlineversusRetail.JournalofIndustrialEconomics,49(4),498-499.Grossman,G.M.andC.Shapiro.1984.InformativeAdvertisingwithDi¤erentiatedProducts.Review
ofEconomicStudies,51,63-84.Harrington,J.
ofEconomicStudies,51,63-84.Harrington,J.E.2001.CommentonReducingBuyerSearchCosts:ImplicationsforElectronicMarketplaces.ManagementScience,47(12),1727-1732.Hong,H.andM.Shum.Forthcoming.UsingPriceDistributionstoEstimateSearchCosts.RandJournalofEconomics.Iyer,G.andA.Pazgal.2003.InternetShoppingAgents:VirtualColocationandCompetition.MarketingScience,22(1),85-106.Janssen,M.andJ.L.Moraga-González.2004.StrategicPricing,ConsumerSearch,andtheNum-berofFirms.ReviewofEconomicStudies,71(4),1089-1118.Janssen,M.,J.L.Moraga-GonzálezandM.Wildenbeest.2005.TrulyCostlySequentialSearchandOligopolisticPricing,InternationalJournalofIndustrialOrganization23,451-466.Johnson,E.J.,W.Moe,P.Fader,S.BellmanandJ.Lohse.2004.OntheDepthandDynamicsofWorldWideWebShoppingBehavior.ManagementScience,50(3),299-308.51Eckard,E.W.2004.TheLawofOnePricein1901.EconomicInquiry,42(1),101-110.Ellison,G.2005.BoundedRationalityinIndustrialOrganization,MITWorkingPaperEllison,G.andS.Ellison.2004.Search,Obfuscation,andPriceElasticitiesontheInternet.MITWorkingPaperErevelles,S.,E.RollandandS.Srinivasan.2001.ArePricesReallyLowerontheInternet?:AnAnalysisoftheVitaminIndustry.WorkingPaperFriberg,R.,M.GanslandtandM.Sandström.2001.PricingStrategiesinE-Commerce:Bricksvs.Clicks.WorkingPaperGatti,R.andP.Kattuman.2003.OnlinePriceDispersionWithinandBetweenSevenEuropeanCountries.AdvancesinAppliedMicroeconomics,12,107-143.Goolsbee,A.2001.CompetitionintheComputerIndustry:OnlineversusRetail.JournalofIndustrialEconomics,49(4),498-499.Grossman,G.M.andC.Shapiro.1984.InformativeAdvertisingwithDi¤erentiatedProducts.ReviewofEconomicStudies,51,63-84.Harrington,J.E.2001.CommentonReducingBuyerSearchCosts:ImplicationsforElectronicMarketplaces.ManagementScience,47(12),1727-1732.Hong,H.andM.Shum.Forthcoming.UsingPriceDistributionstoEstimateSearchCosts.RandJournalofEconomicsIyer,G.andA.Pazgal.2003.InternetShoppingAgents:VirtualColocationandCompetition.MarketingScience,22(1),85-106.Janssen,M.andJ.L.Moraga-González.2004.StrategicPricing,ConsumerSearch,andtheNum-berofFirms.ReviewofEconomicStudies,71(4),1089-1118.Janssen,M.,J.L.Moraga-GonzálezandM.Wildenbeest.2005.TrulyCostlySequentialSearchandOligopolisticPricing,InternationalJournalofIndustrialOrganization23,451-466.Johnson,E.J.,W.Moe,P.Fader,S.BellmanandJ.Lohse.2004.OntheDepthandDynamicsofWorldWideWebShoppingBehavior.ManagementScience,50(3),299-308.Myerson,R.1981.OptimalAuctions.MathematicsofOperationsResearch,6,58-63.Narasimhan,C.1988.CompetitivePromotionalStrategies.JournalofBusiness,61(4),42
7-449.Pan,X.,B.T.RatchfordandV.Shankar.2
7-449.Pan,X.,B.T.RatchfordandV.Shankar.2002.CanPriceDispersioninOnlineMarketsbyEx-plainedbyDi¤erencesine-TailerServiceQuality?JournaloftheAcademyofMarketingScience,specialissueMarketingandServicingConsumersthroughtheInternet,30(4),433-445.Pan,X.,V.ShankarandB.Ratchford.2003.TheEvolutionofPriceDispersioninInternetRetailMarkets.AdvancesinAppliedMicroeconomics,12,85-105.Png,I.P.L.andD.Hirschleifer.1987.PriceDiscriminationthroughO¤erstoMatchPrice.JournalofBusiness,60(3),365-384.Png,I.P.L.andD.Reitman.1994.ServiceTimeCompetition.RandJournalofEconomics,25(4),619-634.Pratt,J.W.,D.A.WiseandR.Zeckhauser.1979.PriceDi¤erencesinAlmostCompetitiveMar-kets.QuarterlyJournalofEconomics,93(2),189-211.Radner,R.1980.CollusiveBehaviorinOligopolieswithLongButFiniteLives.JournalofEco-nomicTheory,22,136-156.Raju,J.S.,V.SrinivasanandR.Lal.1990.TheE¤ectsofBrandLoyaltyonCompetitivePricePromotionalStrategies.ManagementScience,36(3),376-404.Rao,R.C.,R.V.ArjunjiandB.P.S.Murthi.1995.GameTheoryandEmpiricalGeneralizationsConcerningCompetitivePromotions.MarketingScience,14(3),G89-G100.Rauh,M.T.1997.AModelofTemporarySearchMarketEquilibrium.JournalofEconomicTheory,77(1),128-153.Rauh,M.T.2001.HeterogeneousBeliefs,PriceDispersion,andWelfareImprovingPriceControls.EconomicTheory,18(1),577-603.Reinganum,J.F.1979.ASimpleModelofEquilibriumPriceDispersion.JournalofPoliticalEconomy,87,851-858.Resnick,P.andR.Zeckhauser.2002.TrustAmongStrangersinInternetTransactions:EmpiricalAnalysisofeBaysReputationSystem.AdvancesinAppliedMicroeconomics,11,127-158.53Myerson,R.1981.OptimalAuctions.MathematicsofOperationsResearch,6,58-63.Narasimhan,C.1988.CompetitivePromotionalStrategies.JournalofBusiness,61(4),427-449.Pan,X.,B.T.RatchfordandV.Shankar.2002.CanPriceDispersioninOnlineMarketsbyEx-plainedbyDi¤erencesine-TailerServiceQuality?JournaloftheAcademyofMarketingSciencespecialissueMarketingandServicingConsumersthroughtheInternet,30(4),433-445.Pan,X.,V.ShankarandB.Ratchford.2003.TheEvolutionofPriceDispersioninInternetRetailMarkets.AdvancesinAppliedMicroeconomics,12,85-105.Png,I.P.L.andD.Hirschleifer.1987.PriceDiscriminationthroughO¤erstoMatchPrice.JournalofBusiness,60(3),365-384.Png,I.P.L.andD.Reitman.1994.ServiceTimeCompetition.RandJournalofEconomics,25(4),619-634.Pratt,J.W.,D.A.WiseandR.Zeckhauser.1979.PriceDi¤erencesinAlmostCompetitiveMar-kets.QuarterlyJournalofEconomics,93(2),189-211.Radner,R.1980.CollusiveBehaviorinOligopolieswithLongButFiniteLives.JournalofEco-nomicTheory,22,136-156.Raju,J.S.,V.SrinivasanandR.Lal.1990.TheE¤ectsofBrandLoy
altyonCompetitivePricePromotionalStrateg
altyonCompetitivePricePromotionalStrategies.ManagementScience,36(3),376-404.Rao,R.C.,R.V.ArjunjiandB.P.S.Murthi.1995.GameTheoryandEmpiricalGeneralizationsConcerningCompetitivePromotions.MarketingScience,14(3),G89-G100.Rauh,M.T.1997.AModelofTemporarySearchMarketEquilibrium.JournalofEconomicTheory,77(1),128-153.Rauh,M.T.2001.HeterogeneousBeliefs,PriceDispersion,andWelfareImprovingPriceControls.EconomicTheory,18(1),577-603.Reinganum,J.F.1979.ASimpleModelofEquilibriumPriceDispersion.JournalofPoliticalEconomy,87,851-858.Resnick,P.andR.Zeckhauser.2002.TrustAmongStrangersinInternetTransactions:EmpiricalAnalysisofeBaysReputationSystem.AdvancesinAppliedMicroeconomics,11,127-158.Stahl,D.O.1989.OligopolisticPricingwithSequentialConsumerSearch.AmericanEconomicReview,79,700-712.Stahl,D.O.1994.OligopolisticPricingandAdvertising.JournalofEconomicTheory,64,162-177.Stahl,D.O.1996.OligopolisticPricingwithHeterogeneousConsumerSearch.InternationalJour-nalofIndustrialOrganization,14,243-268.Stegeman,M.1991.AdvertisinginCompetitiveMarkets.AmericanEconomicReview,81,210-223.Stigler,G.1961.TheEconomicsofInformation.JournalofPoliticalEconomy,69(3),213-225.Tang,F.andX.Xing.2001.WilltheGrowthofMulti-ChannelRetailingDiminishthePricingE¢ciencyoftheWeb?JournalofRetailing,77,319-333.Varian,H.R.1980.AModelofSales.AmericanEconomicReview,70,651-659.Villas-Boas,M.1995.ModelsofCompetitivePricePromotions:SomeEmpiricalEvidencefromtheCo¤eeandSaltineCrackersMarkets.JournalofEconomicsandManagementStrategy,4(1),85-107.Zettelmeyer,F.,F.Scott-MortonandJ.Silva-Risso.2004.CowboysorCowards:WhyareInternetCarPricesLower?WorkingPaper.55Stahl,D.O.1989.OligopolisticPricingwithSequentialConsumerSearch.AmericanEconomicReview,79,700-712.Stahl,D.O.1994.OligopolisticPricingandAdvertising.JournalofEconomicTheory,64,162-Stahl,D.O.1996.OligopolisticPricingwithHeterogeneousConsumerSearch.InternationalJour-nalofIndustrialOrganization,14,243-268.Stegeman,M.1991.AdvertisinginCompetitiveMarkets.AmericanEconomicReview,81,210-Stigler,G.1961.TheEconomicsofInformation.JournalofPoliticalEconomy,69(3),213-225.Tang,F.andX.Xing.2001.WilltheGrowthofMulti-ChannelRetailingDiminishthePricingE¢ciencyoftheWeb?JournalofRetailing,77,319-333.Varian,H.R.1980.AModelofSales.AmericanEconomicReview,70,651-659.Villas-Boas,M.1995.ModelsofCompetitivePricePromotions:SomeEmpiricalEvidencefromtheCo¤eeandSaltineCrackersMarkets.JournalofEconomicsandManagementStrategy,4(1),Zettelmeyer,F.,F.Scott-MortonandJ.Silva-Risso.2004.CowboysorCowards:WhyareInternetCarPricesLower?