We are grateful to all seminar participants for comments and suggestions Any errors are our own Email fnavalseacuk pschiraldilseacuk Abstract Sales are a widespread and wellknown phenome ID: 483742
Download Pdf The PPT/PDF document "Storage Francesco NavaDiscussio..." 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.
Storageïª Francesco NavaDiscussion Paper No. TE/2011/549 October 2006 The Suntory-Toyota International Centres for Economics and Related Disciplines London School of Economics and Political Science Houghton Street London WC2A 2AE Tel.: 020-7955 6674 We are grateful to all seminar participants for comments and suggestions. Any errors are our own. Email: f.nava@lse.ac.uk, p.schiraldi@lse.ac.uk. Abstract Sales are a widespread and well-known phenomenon that has been documented in several product markets. Regularities in such periodic price reductions appear to suggest that the phenomenon cannot be entirely attributed to random variations in supply, demand, or the aggregate price level. Certain sales are traditional and so well publicized that it is difficult to justify them as devices to separate informed from uninformed consumers. This paper presents a model in which sellers want to reduce prices periodically in order to improve their ability to collude over time. In particular, the study shows that if buyers have heterogeneous storage technologies, periodic sales may facilitate collusion by magnifying intertemporal linking in consumers decisions. The stability and the profitability of different sale strategies is then explored. The optimal sales discount and timing of sales are characterized. A trade-off between cartel size and aggregate profits arises. Keywords: Storage, sales, collusion, cartel size, repeated games. JEL classification: L11, L12, L13, L41. The authors. All rights reserved. Short sections of text, not to exceed two paragraphs, may be quoted without explicit permission provided that full credit, including © notice, is given to the source. 1IntroductionTheoccurrenceofperiodicpricereductions,orsales,onavarietyofitemsisapervasiveandwell-knownmicroeconomicphenomenonthathasbeendocumentedinseveralproductmarkets.Typically,ahighpricesarechargedinmostperiods,butoccasionallypricesarecuttosupplymoreunitstoapotentiallylargergroupofconsumers.Theregularoccurrenceofsuchphenom-enonappearstosuggestthatsalescannotbeentirelyexplainedbyrandomvariationsinsupply,demand,ortheaggregatepricelevel.Moreover,certainsaleperiodsaretraditionalandsowellpublicizedthatitisdi¢culttojustifythemasdevicestoseparateinformedfromuninformedconsumers.Agrowingempiricalliteraturealso,appearstosuggestthatthemajorityofperi-odicsalestakeplaceforproductsthatarefairlystorable,andthatstoragecapacityexplainsinparttheresponsivenessofconsumerstochangesinprices(BellandHilber2006,HendelandNevo2006&2010,Erdemetal2003,Seiler2010).Suchevidencehighlightstheprimaryrolethatstorageconstraintsmayplayindeterminingconsumerspurchasingbehaviorandthus,retailerspricingdecisions.Thepresentpaperstudieshowsalestrategiesmayfostercollusioninamarketinwhichgoodscanbestored.Thustheaimofthestudyistoprovideanadditionalmotivefor rmstoengageinsalestrategiesandtoshedlightontheoptimalsalesdiscountandtiming.Whilefactorssuchasinformationaldi¤erencesandheterogeneityinwillingnesstosearchorpayhavereceivednotableattention,thetheoreticalliteratureonstorageconstraintsremainsscarcedespiteempiricalrelevancedocumentedbyagrowingliterature.NotableexceptionsareSalopandStiglitz1982,Hong,McAfeeandNayyar,2002,andDudine,HendelandLizzeri2006.Weconsideranindustryinwhichineveryperiod,n rmsproduceahomogeneousstorablegood,andsellittoamassofheterogeneousconsumerswithunitdemand.Consumersdi¤eronlybytheiraccesstostorage.Inparticular,werestrictattentiontoeconomieswithtwotypesofconsumers:thosewithoutstoragecapacity,andthosewithstoragecapacityS.1Inthiscontext,weexaminethee¤ectsofheterogeneousstoragetechnologieson rmsincentivestoholdperiodicsalestosupportagreaterdegreeofcooperationamong rmsinarepeatedcompetitionsetting.Salestrategieswillbecharacterizedbyaregularprice,byapricemark- 1Thepaperassumesthatconsumersaresmallandunabletoestablishreputations.SeeNolkeandWhite2007forthecaseofstrategicbuyerswithinterdependentdemands.2 down(ontheregularprice)andbythefrequencyofsales.Weshowthatperiodicsalestrategiessustainedbygrimtriggerpunishmentsallow rmstocolludeonsigni cantpro tlevelsevenwhenstandardno-salestrategiescannotsustaincollusionatsuchpro tlevels.Insuchanenvironmentsalestrategiesmaystrengthencollusion,asthestoragetechnologyintertemporallylinksconsumerdemandandthus,reducestheshort-rungainsfromadeviation.Inparticular,inanysuchequilibrium rmswillcharge,inanyperiodofsales,abigenoughdiscounttoinduceallconsumerswithopenstoragecapacitytostockpileaquantitysu¢cienttosatisfyalltheirdemanduntilthenextsale.Suchbehaviorreducesincentivestodeviateinregularpriceperiods,asonlyconsumerswithoutstoragepurchaseunitsinsuchperiods.Moreover,theincentivestodeviatealso,declineinperiodswithsalesbothbecausealowerpriceischargedinsuchperiods,andbecauseconsumerswithstoragereducetheirdemandwhenadeviationisobservedinthewakeoftheimminentpricewar(impliedbythegrimtriggerpunishments).Oftenwewillrefertotheincentivesnottodeviatefromanequilibriumstrategyasthestabilityofastrategy.The rstpartoftheanalysis:characterizesconsumerdemandfortheproposedenvironment;characterizesthesetEofsalestrategieswhicharemorestablethananystrategywithoutsales;andprovidesnecessaryandsu¢cientconditionsforthenon-emptinessofsuchaset.Thesecondpartoftheanalysisrestrictsattentiontothecaseofsingle-unitstorageandcharacterizesthemoststableandthemostpro tablesalestrategiesinthesetE.Withinsuchsetatrade-o¤isproventoemergebetweencollusivepro tsandstability(whichismeasuredbythemaximalnumberof rmsthatcancolludeongivenpro ts).Suchtrade-o¤isexplicitlycharacterized.Thesecondpartoftheanalysisconcludesbystudyinghowsuchtrade-o¤isa¤ectedbychangesintheenvironment.Inparticular,increasesinpatience(i.e.thefrequencyofinteraction)willleadbothtolargercartelsatanypro tlevelandtothepersistenceofthepro t-stabilitytrade-o¤foralargerrangeofpro ts.Anincreaseinthepro tabilityofamarketinstead,willincreasebothcartelsizeatintermediatepro tlevelsandtherangeofpro tsforwhichthepro t-stabilitytrade-o¤persists.Alargerfractionofconsumerswithoutstoragewillhaveambiguouse¤ectsoncartelsizeathighpro tlevelsandwillreducethemaximalcartelsizeatlowpro tlevels.Ingeneral,thee¤ectofsuchachangewillremainambiguous,asalargefractionofconsumerwithstoragecouldreduceequilibriumpro tsduetothecostofanticipatingproduction.However,therangeofpro tsforwhichthe3 pro t-stabilitytrade-o¤existswillbeproventodeclineinthefractionofconsumerswithoutstorage.Relationshipbetweentheoptimalsalemarkdownandtheenvironmentwillalsobeexplored.The nalpartoftheanalysisconsistsoftwoextensions.The rstextensionconsidersthemulti-unitstoragescenario,characterizestheoptimaltimingofsales,andhighlightsthataccesstomulti-unitstoragetechnologiescanreducethepro t-stabilitytrade-o¤.Thesecondextensionshowsthatsalesdonotneedtobesynchronizedwhen rmscompeteinmultiplemarkets.Notetheintheproposedmodelcollusionisstrengthenedattheexpenseofaggregatepro ts,sincedeviationpro tsdeclinemorethanequilibriumpro tswhensalestrategiesareemployed.Aggregatepro tsmustdeclineintheproposedsetup,sinceallconsumersarehomogeneousintheirwillingnesstopay.Notehowever,thatifconsumerswithhigherstoragecapacityhadalowerwillingnesstopay,asalestrategymayachievehigherpro tsthantheno-salestrategybypricediscriminatingamongdi¤erenttypesofconsumers,andthusfostercollusionevenfurther.Wehaveelectedtokeepvaluationshomogeneousacrossconsumersinordertodisplaymoreexplicitlythee¤ectsoftheintertemporallinkinginconsumerdemand.LiteratureReview:Oneofthe rsttheoreticalexplanationsforsalesrelatesconsumersearchbehaviortopricediscrimination.TwoprominentexamplesinthisliteratureareVarian1980andSalopandStiglitz1982.Varian1980arguesthatinthepresenceheterogeneouslyinformedconsumers,retailpricevariationscanariseasanaturaloutcomeofmixedstrategyequilibriuminwhich rmspricediscriminateconsumerswithdi¤erentinformation.SalopandStiglitz1982instead,considersamodelwithsearchcostsinwhichconsumersareimperfectlyinformedaboutthepriceschargedbystoresanddi¤erintheirabilitytostockpile.Insuchframeworktheauthorsshowthatstoreshaveincentivestoholdunannouncedsalestoinduceconsumerstopurchasefutureconsumption.Bothmodelshowever,areessentiallystaticmodelsandcannotaccountforcorrelationinprices.Eventhoughtherandomsalesfeatureremainsacompellingexplanationforsomeerraticpricebehavior,itappearslesssuitedtoaccountformanyofthedocumentedretailmarkdownsthatarepredictable,publiclyknow,andtakeplaceinmoststoressimultaneously(Pesendofer2002andWarnerandBarsky1995).Theappealingfashion/clearanceparadigmforsales(Lazear1986,Pashigian1988,andPashigianandBowen1991)canalso,hardlybeappliedtoawidevarietyofretailitemsfor4 whichthefashionhypothesisappearsapriorilessappropriate(eitherbecausetheitemsarehomogeneous,orbecausestyleschangelittleovertime).A nalrelevantliteraturehasmotivatedsalesasaformofintertemporalpricediscrimi-nation(Conlisk,GerstnerandSobel1984,HendelandNevo2010,Hong,McAfeeandNayyar2002,NarasimhanandJeuland1985,Sobel1984).Conlisk,GerstnerandSobel1984,andinparticularSobel1984studytheincentivestoholdcyclicalsimultaneoussalesasameansofpricediscriminationinadurable-goodenvironment.Inmostperiods,pricesarekepthightoextractsurplusfromhighvalueconsumers,butperiodicallypricesaredecreasedinordertoselltoalargergroupofconsumerswithlowerreservationvalues.Akeyassumptiontogen-eratesuchpricecyclesistheconstantinowofnewheterogeneousconsumersinthemarket.HendelandNevo2010,Hong,McAfeeandNayyar2002,NarasimhanandJeuland1985,studytheincentivestoholdperiodicsalesinamarketwithstorablegoodsandheterogeneouscon-sumers.Inthissetup,theincentivestopricediscriminateconsumersovertimewithsalesfullyexplainedbythepositivecorrelationbetweenstoragecostsandconsumerswillingnesstopay.Wecomplementthesepapersbyo¤eringanewexplanationfortheexistenceofsalestrategiesinadynamicstorablegoodsmodelinwhichtheincentivestoholdperiodicsalesarisesevenintheabsenceofsuchacorrelation.Moreover,asinSobel1984,wecharacterizetheoptimaltimingforsales.Thepaperisalso,closelyrelatedtoseveralstudieswhichpointedoutimportanceoftheintertemporallinkingindecisionstoexplaincollusivebehavior(AusubelandDeneckere1987,DanaandFong2010,Gul1987,SchiraldiandNava2010).Roadmap:Section2introducesthemodel,de nestherelevantclassofsalestrategies,andpresentsseveralpreliminaryresultscomparingdi¤erentclassesofsalestrategiesintermsoftheirstability.Section3restrictsattentiontoamodelwithunit-storage.Forthatsetuptworelevantsalestrategiesarecharacterized.Namely,themoststablesalestrategyandthemostpro tablesalestrategyinE.Thesectionconcludeswiththefullcharacterizationofthepro t-stabilitytrade-o¤thatdi¤erentsalestrategiesentail.Comparativestaticresultsshowhowthetrade-o¤isa¤ectedbychangesinpatience,inpro tabilityandinthefractionofconsumerswithaccesstostorage.Section4extendsthebaselinemodelintwodirections.The rstconsidersthegeneralmodelwithmulti-unitstorageandagaincharacterizesthemoststableandmost5 pro tablesalestrategiesinsuchamoregeneralframework.Itwillbeshownthatsalestrategieswithinfrequentsalesmaybeincreasebothstabilityandpro ts,andmaythus,alleviatethetrade-o¤.Thesecondextensionshowswhysynchronizationinsalesisnotnecessarywhen rmsoperateinseveralmarkets.Section5concludes.AllproofsarerelegatedtoappendixA.AppendixBcontainsseveralusefulderivationsomittedfromthemaintextforclarity.2AModelwithStorableGoodsThissection rstintroducesasimpleeconomywithstorage,de nestheclassofsalestrategiesthatwillbeanalyzedthroughoutthepaper,anddevelopsseveralpreliminaryresultsonstability.TheSimpleEconomywithStorageConsideranin nite-horizondiscrete-timemodelwithin nitelylivedproducesandconsumers.Supposethattwogoodsaretradedinthemodelwhichweshallrefertoasconsumptionqandmoneym.Ineachperiod,allconsumersareendowedwithalargeamountofmoneyMandwithnoconsumption.Thepreferencesofaconsumerareseparableoverthetwogoodsandsatisfy:u(q;m)=8:v+mifq1mifq1Hence,themarginalvalueofconsumptionisvforthe rstunitconsumedand0foranyadditionalunitconsumed.Thebudgetconstraintfacedbyeachconsumerineveryperiodrequiresthat:m=Mpwherepdenotestheamountofmoneyspentonconsumptiongood.Allconsumersdiscountthefutureatacommonfactor,andtheirtime-preferencesoverutilitysequencesfutg1t=0satisfy:(1)X1t=0tutThereisaunitmeasureofconsumers.Consumersdi¤eronlyintheirabilitytostoretheconsumptiongood.Inparticular,assumethatafraction0oftheconsumersisunabletostoregoods,whileafractionScanstoreuptoSadditionalunitsofconsumption.Suchunitsdo6 notdepreciate,whenstoredandcanbeconsumedinanyfutureperiod.A nitesetof rms,Nwithcardinalityn,suppliesconsumptiongoodtothismarket.All rmshaveacommonconstantmarginalcostofproducingconsumptiongood,c.Ifp=(p1;:::;pn)arethepricessetbyeach rmontheunitsofconsumptionsold,andifd(p)denotestheaggregatedemandatsuchprices,theindividualdemandfacedby rmisatis es:di(p)=8:1 jargminj2Npjjd(p)ifpiminj2Npj0ifpiminj2NpjThestagepro tof rmi2Ngivenapricevectorpsatis es:vi(p)=(pic)di(p)All rmsdiscountthefutureatacommonfactor,andtheirtime-preferencesoverpro tsequencesfvtg1t=0satisfy:(1)X1t=0tvtSaleStrategiesandEquilibriumFirmsandconsumersobserveallthepricesquotedinthemarketinallpreviousperiods.Thus,thesetofpossiblehistoriesintheinducedgamecanbede nedas:H=f;g[f[1t=1ts=1Rn+gA rmsstrategymapshistoriesintoapricequotedatagivendate.Consumersusetheinformationaboutpastquotedpricesandtheequilibriumstrategyofthe rmstoformbeliefsaboutfuturepricesintheeconomy.Sinceconsumersaresmallweshallassumethattheirindividualdecisionsareunobservabletoanyotherindividual.Consumerswillthus,decidehowmanyunitstopurchasefrom rmsquotingthelowestprice,justsotomaximizetheirindividualpayo¤.Inparticular,consideranysequenceoffuturepricespt=fpzg1z=t.Let7 pz=mini2Nfpizgdenotethemarketpriceinperiodz,andlet:T(pt)=minzsubjectto ptz pt+zdenotehowlongaconsumerhastowaitbeforethecurrentmarketpriceexceedsthefuturediscountedprice.Thenextpropositionpinsdownconsumerdemandfortheproposedenviron-ment.Remark1If pzvinanyperiodz0,thedemandforconsumptiongoodattimet:(1)byconsumerswithoutstoragetechnologysatis esd0(pt)=1;(2)byconsumerswithstoragetechnologyandwithsunitsalreadyinstoragesatis es:dS(s;pt)=maxminT(pt);S+1 s;0 Thisisthecasesinceconsumerswithaccesstostoragepurchasemultipleunitsonlyiftheyperceivethestoragecosttobesmallerthanthecostoffuturepriceincreases.2Sincebuyerscannotbuildreputations,allconsumersofthesametypepurchasethesamenumberofunitsineachperiod.Hence,theaggregatedemandinaperiodinwhichallconsumerswithstoragehavethesamenumberofunitsssatis es:d(s;pt)=0+SdS(s;pt)Theequilibriumstrategiesthatwillbeanalyzedthroughoutthepaperdisciplinedeviationsastrigger-strategieswould.However,equilibriumpriceswillvaryalongtheequilibriumpath.Inparticular,consideranystrategyinwhichall rmssetpricesalongtheequilibriumpathsothatforsome{2f2;3;:::g: pt=8:(1+)cifmod(t;{)6=0(1+)cifmod(t;{)=0wheremod(t;{)6=0denotesthe{moduloofthetimeperiodt.Suchstrategymaybeinterpretedasacyclicalsalespolicyinwhichall rmsjointlyreducepricesevery{periods,wheredenotesthemarkupinperiodswithoutsales,andwhere2[0;1]denotesthefractionof 2Thecostofstoragecoincideswiththediscountfactor,astherateoftimepreferencesrepresentstheoppor-tunitycostofspendingmoneysoonertostockunits.8 suchamarkupchargedduringsaleperiods.Deviationsfromtheequilibriumpatharepunishedviareversiontocompetitivepricingineachfuturetimeperiod.Inparticular,foranyhistoryht2Hoflengtht,allthesalestrategiesthatwillbeconsideredwillsatisfy:(ht)=8: ptifpiz= pzforanyiandanyztcifotherwiseTheequilibriumpunishmentstrategyisNashinanysubgameinwhichadeviationhasalreadyoccurred,sinceno rmcanbene tfromadeviationwhenalltheother rmsarepricingcom-petitively.Thus,theincentivestocomplywithasalestrategywillbepinneddownbylookingonlyatdeviationsfromtheequilibriumpath.Lett()denotetheaggregatepayo¤ontheequilibriumpath,lett()denotethepayo¤ofthemostpro tabledeviationfromtheequi-libriumpath,andlettheratioofequilibriumtodeviationpro ts,t=t,bedenotedbyRt().Thefollowingresultcharacterizestheupper-boundoncartelsizeforwhichtheproposedsalestrategyconstitutesaSubgamePerfectequilibrium.Remark2AsalestrategyisaSPEofthein niterepetitionofthegameifandonlyif:n1 1Rt()foranyt0(1)Theupper-boundonthepopulationsizedisciplinesthelargestnumberof rmsthatcansustainstrategyasasubgameperfectequilibrium.Notethatforanystrategywithoutsales(andthuswithoutstorage)suchconditionwouldsimplifytothecommonrequirementn1 1.Throughout,wewillrefertothestrategiesde nedinthissectionassalestrategies.Anyoneofthesestrategieswillbecompletelypinneddownbythethreeparameters:thesalesdiscount,theregularmarkup,andthefrequencyofsales{.ThesetofpossiblesalestrategieswillbedenotedbyS=0;vc c[0;1]f2;3;:::g.Theupper-boundimposedonthemarkuprequiresconsumerswithoutstoragetobewillingtotradeinperiodswithoutsales.Suchrestrictionisimposed,sinceanystrategywithahighermarkupcouldnotbeoptimalforpro tsandforstability.SimilarconsiderationsshowwhyrelaxinganyoneoftheotherboundsonSwouldnota¤ectanyofresultsdevelopedinwhatfollows.9 PreliminaryResultsonStability&Pro tsThissub-sectiondevelopsseveralpreliminaryresultsthatcomparedi¤erentsalestrategiesinSintermsofpro tsandstability.Necessaryandsu¢cientconditionforastrategywithsalestobemorestablethanastrategywithoutsalesarepresented.Necessaryandsu¢cientfortheexistenceofsuchsalestrategiesarealsopresented.Tothisend,letusbeginwithatwode nitionsclarifyingtheintentofourcomparison.De nition1Asalestrategy2Sissaidtobemorestablethanstrategy0if:mint0Rt()mint0Rt(0)De nition2Asalestrategy2Sissaidtoraisemorepro tsthanstrategy0if:0()0(0)Notethatfora xeddiscountfactor,thede nitionofstabilitycomparesanytwostrategiesbylookingatthelargestnumberof rmsn()thatcancolludeonstrategy:n()=mint0Rt() 1Analternative,butsimilar,de nitionofstabilitymayinvolvetheoorofthemapn().Butsimilarconclusionswouldhold.Alternatively,onecouldfocusonthethelowestpossibledis-countfactor()neededtocolludeonstrategyinapopulationof xedsizen.Todosoa xedpointargumentwouldhavetobeemployedtosolvefor:()=1mint0Rt(j()) nWerefrainfromdoingsoandinstead,focusonn(),sincetheanalysissimpli esconsiderablywhilethequalitativeresultscoincide.Inordertocomparethestabilityoftwosalestrategies,theequilibriumanddeviationpro tsmustbeexpressedintermsoftheparametersofthestrategy.Recursivelyde netheequilibrium10 demanddtandstoragestineachperiodt0asfollows:d0=d(0;p0)&dS0=dS(0;p0)&s1=dS01dt=d(st;pt)&dSt=dS(st;pt)&st+1=st+dSt1Thenextremarkshowsthatanysalestrategyiscyclical,andthatitiswithoutlosstoconsideronlythe rst{periodstocharacterizetheentirestreamofpayo¤s.Remark3Ifmod(t;{)=0,st=0.Ifmod(t;{)=mod(z;{),dt=dz.TheclaimfollowsfromthepropertiesoftheequilibriumpricingpathpandofthemapdS(st;pt).Itrequiresconsumersnottohaveanyunitsstoredinperiodsofsaleandthus,tohavethesamedemandatcongruentdatesinthecycle.Thus,for:S(t)=8:0ifmod(t;{)=0{mod(t;{)ifmod(t;{)6=0equilibriumpayo¤smustsatisfyfort=f1;:::;{1g:t()=(1) 1{P{1z=0zvt+z(pt+z)==(1) 1{hP{1z=0zdt+z(1)S(t)d0icSinceadeviationtopricey6=ptatstagetleadstoaconjecturedpricepathyt=(y;pit);c;c;c;:::,foryptthedeviationpayo¤sateachstagemustsatisfy:t(y;)=(yc)d(st;yt)Forconvenience,letusde nefourclassesofsalestrategiesforwhichallpreliminaryresultswillbedeveloped:De nition3LetNSdenotethosestrategiessuchthat=1.LetVSdenotethosestrategiesforwhichdt-277;0andmod(t;{)6=0forsomet.LetCSdenotethosestrategiesforwhich{S+1and(1+){1(1+).LetECdenotethosestrategiesforwhich0 {S+0.11 ThesetNconsistsofallthosesalestrategiesforwhichnodiscountisevero¤eredalongtheequilibriumpath.ThesetVcomprisesallthosestrategiesforwhichthereisaperiodalongtheequilibriumpathwithnosalesandunitdemand.ThesetCinstead,willbeproventoconsistsofallthosestrategiesinwhichconsumerswithaccesstostoragepurchaseonlyinperiodswithsales.TherestofthesectionprovesthatEcomprisesallthesalestrategieswhicharemorestablethanastrategywithoutsales,andprovidesnecessaryandsu¢cientconditionsforthenon-emptinessofE.Thenextpropositionpresentsseveralintroductoryresultsonthestabilityandtheprof-itabilityofdi¤erentsalestrategies.Inparticular,itshowsthatVandCpartitionSandthatanysalestrategyinwhichconsumerswithstoragepurchaseunitsduringano-salesperiodisdominatedbothintermsofpro tsandintermsofstabilitybyapolicyinwhichnosalesevertakeplace.Sucharesultconsiderablysimpli estheanalysisofthepro t-stabilitytrade-o¤.Thelatterobservationwillbeexploitedtocharacterizethesetofsalestrategiesthatcanbemorestablethantherevenuemaximizingno-salespolicy,=vc cand=1.Proposition4Thefollowingclaimsmusthold:(1)SnC=V(2)NV(3)anystrategyinNismorestablethananystrategyinV;(4)anystrategythatsets=vc cand=1ispro tmaximizingwithinS;(5)foranystrategyinC,st=S(t)and:dt=8:0+{Sifmod(t;{)=00ifmod(t;{)6=0ThepropositionprovesthatVandCpartitionthesetofsalestrategies.Moreover,itimpliesthatallstrategieswithnosales(i.e.strategiesinN)areequallystableandmorestablethananyothersalestrategyinwhichconsumerswithstoragepurchaseunitsinperiodsofsales(i.e.strategiesinV).Thisshouldalso,clarifywhysaleswhereassumedtotakeplaceintheinitialperiod,sinceanysalestrategyviolatingsucharequirementwouldbelongtoV.Thelastpartofthepropositionconsiderablysimpli estheexpressionoftheequilibriumpayo¤sforanystrategy12 inC.Inparticular,forany2Cweget:t()=0+1 1{S(t)[(xS+0)0]cThelemmaalso,pinsdowntherevenuemaximizingdeviationsforsuchcandidatestrategies:t()=maxyt(y;)=8:cifmod(t;{)=00cifmod(t;{)6=0andthecorrespondingpro tratios:Rt()=8:0 +1 1x({S+0)0 ifmod(t;{)=01+1 1{S(t)h({S+0) 01iifmod(t;{)6=0Thenexttabledisplaysthee¤ectsofmarginalchangesinthesalestrategyonequilibriumanddeviationpro tsandtheirratio: ddd{ dt() 00?dt() 000dRt() 0??ThesignofdRt()=disunspeci ed,sinceitisnegativeifmod(t;{)=0andpositiveoth-erwise.Thesignsofd1()=d{anddR1()=d{coincide.Inparticular,suchderivativesarepositiveeitherifmod(t;{)=0,orifmod(t;{)6=0andthefractionofconsumerswithstor-ageissu¢cientlysmall,andnegativeotherwise.AppendixBcontainsthederivationofthesederivativesandtheirsigns.Asproveninthenextproposition,wheneverastrategybelongsCitiswithoutlosstoignoreall,butthe rsttwoperiods,inordertocharacterizeitsstability.SuchconclusioncoupledwiththeobservationthatmintRt()isindependentofandsinglepeakedinforanygiven{,impliesallstrategiesinEmustbemorestablethanastrategywithnosales.Proposition5AnystrategyinEismorestablethananystrategyinSnE.Therefore,wheneverthesetEisnon-emptysalestrategiesexistthataremorestablethan13 strategieswithoutsales.ItiseasytoobservethatastrategyinEwillbestrictlymorestablethanastrategywithoutsalesifandonlyif:20 S{+0;+1 {11 Suchexpressionandthepreviousresultcanbeexploitedtoderivenecessaryandsu¢cientconditionsforthesetEtobenon-empty.Corollary6Econtainsastrategywithacycleoflength{2f2;:::;S+1gifandonlyif:{1vc v0 {0({1)+c vThus,ifvc v0 20+c v,thenEisnon-empty.Suchconditionsjointlydisciplinetheallfreeparametersofthemodel,namely:thefractionofconsumerswithstorage0,thepro tabilityofthemarketvc,andthediscountfactor.Thecomparativestaticsresults,developedbelow,discussindetailhowthesizeofEthedependsonsuchfreeparameters.Thenewboundimposedonthediscountratearisesfromtheconsumersdemand.Thus,suchconditionwouldonlydisciplinethetimepreferencesoftheconsumers,ifthosecoulddi¤erfromtimepreferencesofthe rms.However,evenwhenthetwocoincide,therestrictionimposedonremainsindependentofthenumberof rmsinthemarket.Beforeproceedingtothenextsection,fornotationalconvenience,de ne({)astheuniquepositiverootofthefollowingquadraticequation:R1(({);{)=R0(({);{)ifsuchasolutionexistsin[0;1],andset({)=1otherwise.Thedetailsofthederivationof({)andtheproofofuniquenessaredeferredtoappendixB.Further,de ne({)asthesmallestdiscountforwhichconsumerswithaccesstostoragewouldpurchase{unitsinperiodsofsaleswhentheregularmarkupissetatthemonopolylevel=(vc)=c:({)=v vc{1c vc14 3Single-UnitStorageandthePro t-StabilityTrade-O¤Forsakeoftractabilitythissectionconsidersenvironmentsinwhichconsumerscanstoreatmostasingleunit,S=1.Wedoso,sincemostqualitativeresultsareuna¤ectedbythisassumption.Partoftheextensionssectionisdevotedtogeneralizingtoarbitrarystoragecapacities.ThesectioncomparessalestrategiesinSintermsofpro tsandstability,andshowsthatatrade-o¤canemergebetweenthetwo.ThenexttwopropositionscharacterizetwostrategiesparticularstrategiesinE.Theformerwillbethemoststablesalestrategy,whilethelatterwillbethemostpro tableofallthesalestrategiesinE.Wheneverthetwostrategiesdonotcoincide,atrade-o¤betweencartelpro tsandstabilitywillemerge.Thesectionproceedswithtoexplicitcharacterizationofthepro t-stabilitytrade-o¤andtoseveralcomparativestatics.Tosimplifynotationinthissectionlet=0and=c=(vc).TriviallyobservethatS=1impliesthat{=2foranypolicyinE.Notethatgiventhestatedassumptionsandde nitions,(2)and(2)respectivelysatisfy:(2)=minf1;g(2)=(1+)whereistheuniquepositiverootofthefollowingquadraticequation:2(2)(1)2=0IfEisnon-empty,themoststablesalestrategyischaracterizedbythefollowingresult.Proposition7IfE6=;,nostrategyinSisstrictlymorestablethanstrategy2E: { 1= minf(2);(2)g 2 Moreover,isthemostpro tableofallthestrategiesinSwithequalstability.Suchastrategyrequires rmstosetcollusivemarkupsinperiodswithoutsalesanduniquelypinsdowntheoptimaldiscountfortheremainingperiods.Themonopolymarkupcanbecharged15 inperiodswithoutsalessincehasnoe¤ectonthestability.TheoptimalsalesdiscountischosentominimizeminfR0();R1()gwithinthefeasibleset2[=(2);(1+)]andcruciallydependsonthefractionofconsumerswithstorageandonthemonopolymarkupintheeconomy.Thecomparativestaticssectiondiscussessuchdependenceindetail.Noticethattheoptimalsalestrategymaydependonalltheparametersofthemodelexceptforn,sinceboth(2)and(2)areindependentofn.Thus,thelargestnumberof rmsn()willingtocolludeonanystrategyinScanbefoundbylookingat:n()=R1() 1Eventhoughsuchstrategyisoptimalintermsofstability,morepro tablepoliciesexistinE.Thenextpropositioncharacterizes,themostpro tablesalestrategyinE.Proposition8IfE6=;,nostrategyinEisstrictlymorepro tablethanstrategy+2E: + + {+ 1= (2) 2 Asinthepreviouspropositionthestrategyrequires rmstosetcollusivemarkupsinperiodswithoutsales.However,thepro tmaximizingsalediscountistheuniquelypinneddownbytheconsumersstorageconstraint.Obviously,suchadiscountmaybesmallerthanthatofthemoststablepolicy,andnolongerdependsonthefractionofconsumerswithstorageintheeconomy.Hence,theupper-boundonthenumberof rmsneededtoacolludeon+maybesmallerthanfor:n(+)=minfR1(+);R0(+)g 1n()Theprevioustwopropositionsweremeanttohighlightthetrade-o¤thatmayarisebetweenpro tsandstabilityinsuchenvironments.The rstresultinfact,showedthatsalestrategiescouldbeusedimprovestabilityattheexpenseofpro ts.Strategywasproventobemorestablethananyotherstrategywithoutsales,butlesspro tablethanfullcollusion.Similarlywasclearlymorepro tableandlessstablethanthecompetitiveoutcome(i.e.theNashequilibriumofthestagegame).Thesecondresultinstead,showedthatevenwithinEpro t-stabilitytrade-o¤swouldarisewhenever6=+.16 SinceSwas xedto1,amorestringentcharacterizationofthepro t-stabilitytrade-o¤withinScanbederived.Inparticular,notethatthepreviouspropositionsrequirethat:(1)increasingcanonlybene tpro tsandcannotharmthestability;(2)increasingcanonlybene tpro ts;(3)increasingcanharmstabilityifandonlyif2[(2);(2)].Then,for xedvaluesof,,and,considerastrategy(1=;;2).LetR()=mintRt(1=;;2)denoteitsstability,let()=0(1=;;2)denoteitspro tsandlet:n()=1 1R()denotethemaximalnumberof rmsthatcansustainsuchastrategyinequilibrium.Notethat()isstrictlyincreasing,andthusinvertible,in.Finally,foranypro tlevel2R+letN()=n(1())denotethemaximalnumberof rmsthatcancolludeonsuchapro tlevelwhileemployingasalestrategy.Ifso,notethatatrade-o¤emergesbetweenpro tsandstabilityfor2[(2);(2)]:dN() d=dn=d d=d(1())=1 1dR0=d d0=d(1=;1();2)0sincethedenominatorispositive,whilethenumeratorisnegative(detailsinAppendixB).Inparticular,thelastexpressionimpliesthatmaximalnumberof rmsthatcansustainacollusivesalestrategy(1=;;2)mustdeclineaspro tsincrease.Furthernoticethatanysalestrategythatraisesmorepro tsbysetting(2)cannotbeoptimalintermsofstability,sinceapolicyinNexiststhatisbothmorestableandmorepro table,as:1 1n()Similarly,nopolicysetting(2),couldeverbeoptimal,sinceitwouldsimultaneouslyreducestabilityandpro ts.Infact,bycontinuityasalestrategy(;(2);2)couldbeproventoexistthatisequallypro tableas(1=;;2),butmorestable.Forconvenienceletm=vcdenotethemonopolypro tandletand+denotethepro tsrespectivelyofthemoststable17 andthemostpro tablestrategiesinE:= 1++2 1+(vc)+= 1++2 1++(vc)Noteimmediatelythat+.Thenextpropositionhighlightsthespeci cnatureofthepro t-stabilitytrade-o¤foreconomiesinwhichS=1.Proposition9IfE6=;,foranypro tlevel2(0;m]themaximalnumber rmsthatcancolludeonwhileemployingasalestrategysatis es:N()=8]TJ ; -2;.51; Td; [00;]TJ ; -2;.51; Td; [00;]TJ ; -2;.51; Td; [00;:1 1if2(+;m]1 1(2) (1+)(vc)if2(;+]n()if2(0;]Intheinterval(;+],N()isdecreasingandconvex,andbyconstructionsatis esN()1=(1).Thus,propositionexactlyquanti esthetrade-o¤betweenpro tsandcartelsize(sta-bility)thatdi¤erentsalestrategiesimply.Suchatrade-o¤emergessincelargersalesdiscountsmayfavorstability,butcertainlyhurtpro ts. Figure1:TheleftplotdepictsR1(;{)andR0(;{)asfunctionsof,where({)=0=(S{+0).Therightplotdepictsthepro t-stabilitytrade-o¤N().Theleftplotof gure(1)depictsR1(;{)andR0(;{)andshowsthatforanyvalueof{18 auniquediscount({)existswhichmaximizesthesizeofthecartel.Therightplotinstead,depictspro t-stabilitytrade-o¤N()derivedinthepreviousproposition.Tohighlightthetrade-o¤moreexplicitlyconsideraneconomyinwhich=0:95,=0:15,v=10,andc=1.Noticethatthemaximalcartelsizegrowsfrom20to29when rmscolludeonthemoststablesalestrategyinsteadofthemonopolystrategym.Pro tshowever,declinesigni cantlyfrom9to2as rmsneedtopriceveryaggressivelyduringsalesinordertosustaincollusion.Themostpro tablesalestrategyinstead,marginallyimprovesthecartelsize,butraisesalmostasmuchpro tasthemonopolypolicy.Thefollowingtablereportsalltherelevantvariablesfortheexamplediscussed: n{ m 20:09:001:0098+ 20:58:720:9492 28:61:960:1592Thenextsectiondevelopscomparativestaticstohighlighthowchangesintheenvironmentmaya¤ectsuchatrade-o¤.ComparativeStaticsAllthecomparativestaticsaredevelopedforthefourrelevantfreeparametersinthemodel,,,vandc.The rstpreliminaryresultcharacterizeshowthesizeofthesetEchangesassuchparametersvary.Asexpected,moresalestrategiesarestablebothwhenmanyconsumerscanstoreunits,andwhenagentsarepatient.Increasesinthepro tabilityofthemarketfurtherimprovetheabilitytocolludeonagivensalestrategy.Proposition10ThesizeofthesetEdecreaseswithcand,andincreaseswithvand.TheresultisprovenbystudyinghowtheboundscharacterizingthestablestrategysetEvarywiththefreeparameters.AlargerfractionofconsumerswithstorageincreasesthesizeofE,sincemoresalediscountsarestableatanyfrequency{.Similarlypatience,andpro tabilityvc,increasethesizeofE,sincetheconsumerdemandconstraint({)isrelaxedwhensuchvariablesgrow.19 Amorecompellingresultcharacterizeshowthepro t-stabilitytrade-o¤isa¤ectedbychangesinthefreeparameters.Thenextpropositionexplicitlycharacterizesthisdependence,andshowshowequilibriumstrategiesarea¤ectedbychangesintheenvironment.Proposition11IfE6=;,foranypro tlevel2(0;m]themaximalnumber rmsthatcancolludeonwhileemployingasalestrategysatis es:dN() dddvdc 2(+;m] +0002(;+] +?+2(0;]\6=+ +002(0;]\=+ ++wheredN()=d0ifandonlyif=(2m).Moreover,thecut-o¤pro tlevelsand+andthemapsandevaluatedat{=2and(2)(2)furthersatisfy: dddvdc d+ +++d ++d +00d +0+Thepropositionshowsthatincreasesinpatience(i.e.thefrequencyofinteraction)mayleadbothtolargerequilibriumcartelsatanypro tlevelandtothepersistenceofthepro t-stabilitytrade-o¤onalargerrangeofpro ts.Increaseinthepro tabilityofamarket(i.e.vc)instead,wereshowntoincreasethemaximalequilibriumcartelsize,butonlyforintermediatepro tlevels,asthestabilitybothofthemoststablestrategyandofthemonopolystrategymwereproventobeindependentofvaluesandcosts.Increasesinthepro tabilityofamarketwerealsoproventoincreasetherangeofpro tsforwhichthepro t-stabilitytrade-o¤persists.Finally,increasingthefractionofconsumerswithoutstorage(i.e.)wasproventoreducethestabilityofthemoststablesalestrategyasintertemporallinkingbetweendecisionswoulddecline.Thee¤ectofsuchachangeonthemaximalcartelsizeatintermediatepro tvalueswasinstead,provenambiguous,asalargefractionofconsumerwithstoragecouldleadtoa20 declineinequilibriumpro tsduetothecostofanticipatingproduction.Clearlysuchachangewouldhavenoe¤ectonthestabilityofstrategieswithoutsales.However,therangeofpro tsforwhichthepro t-stabilitytrade-o¤existswoulddeclineinandwouldeventuallyvanishatsomevalue1.Notethatthediscounto¤eredduringaperiodwithsales(i.e.1)inthemostpro tablepolicy+declineswithpatienceandpro tabilityandisuna¤ectedbythefractionofconsumerswithstorage.Thediscounto¤eredinthemoststablestrategy(whensuchstrategydoesnotcoincidewith+)instead,growswithpatienceandthefractionofconsumerswithstorageandisuna¤ectedbypro tability.Figure2belowprovidesavisualcharacterizationofthecomparativestaticsresultspresentedinthepreviousproposition. Figure2:Comparativestaticsonthetrade-o¤N()withrespecttoanincrease:in(topleft),in(topright),inv(bottomleft),andinc(bottomright).21 4Multi-UnitStorageandAsynchronizedSalesThissectionconsiderstwoextensionsofthebaselinemodel.Inthe rstconsumershaveaccesstomulti-unitstoragetechnologies,S1,whileinthesecond rmscompeteinmultiplemarkets.GeneralStorageS1The rstextensioncharacterizestheoptimaltimingofsales,andhighlightsthataccesstomulti-unitstoragetechnologiescanreducethepro t-stabilitytrade-o¤.Thiswillbethecase:sincethemoststablesalestrategymaybothdisplayinfrequentsalesandbebothmorepro table,thanthemoststablepolicywithfrequency2;andbecausethemostpro tablesalestrategywillnotbea¤ectedbythechangeinstorageconstraints.Asintheprevioussection,thenexttwopropositionscharacterizethemoststableandthemostpro tablesalestrategiesinE.Forconvenience,letusidentifytwoparticularsalefrequencieswhichwillbeemployedinthecharacterizationofthemoststablesalestrategies:{=argmax{2f2;:::;S+1gR0(({);{)s.t.({)({){=argmin{2f2;:::;S+1g{s.t.({)({)Proposition12AssumethatE6=;.If{isde nedandifR0(({);{)R1(({);{)when-ever{isalsode ned,thennostrategyinSisstrictlymorestablethanstrategy2E: { 1= ({) { otherwise,nostrategyinSisstrictlymorestablethanstrategy2E: { 1= ({) { Moreover,isthemostpro tableofallthestrategiesinSwithequalstability.Suchastrategyrequires rmstosetcollusivemarkupsinperiodswithoutsalesanduniquelypinsdowntheoptimaldiscountfortheremainingperiods.Comparedtothepolicyfoundintheprevioussection,themoststablestrategymaydisplaylessfrequentsales,{2.Inparticular,22 anecessaryconditionforthisphenomenontooccuristhatthecriticalratio,R0(({);{),beincreasingwith{at{=2.Theoptimalsalesdiscountandtheoptimalfrequencyofsalescruciallydependonthefractionofconsumerswithstorage,onthediscountrate,andonthemonopolymarkupintheeconomy.Inmanycommonscenarios,thesalestrategycanbesimpli edto(1=;({1);{1)if{2,andto(1=;(2);2)otherwise.Themostpro tablesalestrategyinEremainsuna¤ectedwhenconsumerscanstoremultipleunitsandthus,thepropositioncharacterizingsuchstrategycoincideswiththeonepresentedintheprevioussection.Proposition13IfE6=;,nostrategyinEisstrictlymorepro tablethanstrategy+2E: + + {+ 1= (2) 2 Thepro tmaximizingsalestrategycannotchangeevenifconsumerscanstoremultipleunits,sincealowersalesfrequencywouldleadtoabiggerdiscountandthustolowerpro ts.Again,theoptimalsalesdiscountistheuniquelypinneddownbytheconsumersstorageconstraint.Asintheprevioussection,suchpropositionswoulddisplaythepro t-stabilitytrade-o¤forthismoregeneralenvironment.The rstresultinfact,showedthatsalestrategieswithinfrequentsalescouldbeusedtofurtherimprovestabilityattheexpenseofpro ts.Infact,strategywasproventobemorestablethananyotherstrategywithoutsales,butlesspro tablethanfullcollusion.Similarlywasclearlymorepro tableandlessstablethanthecompetitiveoutcome(i.e.theNashequilibriumofthestagegame).Thesecondresultinstead,showedthatthepro t-stabilitytrade-o¤wouldpersistwhenwouldentailinfrequentsales.ThecomparativestaticresultsonthesizeofthesetEdevelopedintheprevioussectionalsoholdwithoutfurthermodi cationsinthismorecomplexenvironment.Asexpected,moresalestrategieswillbestable:whenconsumersaremorepatient;whenmoreconsumershaveaccesstostorage;orwhenthemarketbecomesmorepro table.Proposition14ThesizeofthesetEdecreaseswithcand,andincreaseswithvand.Obviously,allthecomparativestaticsdevelopedintheprevioussectiononthemostpro tablepolicy+holdwithoutanyfurthermodi cation.Asforthemoststablepolicy,resultswould23 havetobeadjustedtoallowoptimalfrequency{torespondtochangesintheenvironment.However,resultswithaavorsimilartotheonesdevelopedintheprevioussectionwouldholdatanygivenfrequency{.Werefrainfromdevelopingsuchcomparativestaticsinfullgeneralityastheadditionalintuitiongainedislimited.Toconcludethispartoftheanalysisconsideragaintheeconomyinwhich=0:95,=0:15,v=10,andc=1.Supposethatconsumerswithaccesstostoragecanstoreuptothirtyunits,S=30.Noticethatthemaximalcartelsizeisachievedwithinfrequentsaleswhichtakeplaceevery21periods.Maximalcartelsizegrowsfromto37when rmscolludeonthemoststablesalestrategy(S).Sincesalesoccurlessfrequentlysmallerdiscountsarenecessarytosustainthemaximalcartelsize.Hence,inthisenvironmentthepro tsofthemoststablestrategy(S)canbelargerthanthoseassociatedtothemoststablestrategy(1)ofaneconomyinwhichatmostasingleunitcanbestoredS=1,asisthecaseintheexamplereportedbelow.Thus,thepro t-stabilitytrade-o¤candecreasewhenconsumersgainaccesstomoree¢cientstoragetechnologies.Thefollowingtablereportsalltherelevantvariablesfortheexamplediscussed: n{ m 20:09:001:0098+ 20:58:720:9492(1) 28:61:960:1592(S) 37:44:620:27921MultipleMarketsandAsynchronizedSalesFromthepreviousdiscussion,itmayappearthatcoordinationinsalesisnecessarytoachieveanystabilitygain.Incontrast,weprovideasimpleexampletoarguethatsalestrategiesdonotneedtosynchronized.Inparticular,wewillarguethatwhen rmsoperateinmultiplemarkets,salesdonotneedtobesimultaneousandsymmetriceitherwithinoracrossmarkets.ConsideravariationonthepreviouslydescribedeconomyinwhichtherearetwoidenticalmarketsAandB,eachwithamass1=2ofconsumers,andanevennumbern4of rmsoperatinginbothmarkets.3Corollary9providessu¢cientconditionsfortheexistenceofastablesalestrategy 3Noticethatsincemarketsaresymmetric,since rmsobjectivefunctionsarenotstrictlyconcave,andsincereturnstoscaleareconstant,therearenostabilitygainsduetothemulti-marketsetup(BernheimandWhinston24 k2Eineachmarketk2fA;Bg.Eachofthesestrategiesstillrequires rmstochargethe xedmarkup,k,inalmosteveryperiod,andtoperiodicallyholdsalesbyreducingthemarkuptokkevery{kperiods.Themoststablesalestrategyinsuchanenvironmentstillprescribessetk=ineachmarketk2fA;Bg.Suchasalestrategysustainscollusionifinanyperiodt2f0;1g:n1 1t(A)+t(B) t(A)+t(B)=1 1Rt()(2)sincet(k)=t()andt(k)=t()foranyk2fA;Bg.Nowconsiderastrategyinwhichthemarkupineachmarketis xedtok=,butdi¤erent rmsholdsalesindi¤erentmarketsevery{periods.Inparticular,considerastrategyinwhichsalesoccurringalongtheequilibriumpathsatisfyineveryperiodt:(1)ifmod(t;2{)=0, rmsf1;2;:::;n=2gsetadiscountA=inmarketAandBinmarketB,whilealltheremaining rmssetB=inmarketBandAinmarketA;(2)ifmod(t;2{)={, rmsfn=2+1;:::;ngsetadiscountA=inmarketAandBinmarketB,whilealltheremaining rmssetB=inmarketBandAinmarketA;(3)ifmod(t;2{)6=0;{,A=B=1forevery rmineverymarket.Notethatany rmchargingkinmarketkdoesnotcollectpro tsinthatmarketduringasalesperiod.Also,observethatthetotalpro tacrossmarketsisconstantforeach rmandequaltothetotalpro tachievedincaseofsimultaneoussales.Withoutlossconsiderperiod0anda rmi2f1;2;:::;n=2gandnotethat:0i(A)=1 2x 1x0+1 12x[2({S+0)]c0i(B)=1 2x 1x0+xx+1 12x[2({S+0)]c)0i(A)+0i(B)=0(A)+0(B)Moreover,notethatinanyperiodtthedeviationpro tsofeachplayercoincideineachmarketkwiththoseofthemoststablesalestrategysincet(k)=t(k).Thefewlastobservationsinturnimplythatstrategyisasstableasmoststablestrategy.Withoutlossofgeneralityconsidertheincentivestodeviateinperiod0ofafora rmi2f1;2;:::;n=2gholdingasales 1990,andSpagnolo1999).25 inmarketA:n1 10(A)+0(B) 0(A)+0(B)=1 1R0()whichisequivalenttocondition(2).Similarly,incentivestocomplywiththeequilibriumstrategyremainuna¤ectedinperiodswithoutsales.Thus,maximalcartelsizeunderwhichasalestrategysustainscollusionremainsuna¤ectedevenwithasynchronizedsales.Hence,anasynchronizedsalestrategywouldstrictlydominateasimultaneoussalestrategyforanyarbitrarilysmallmenucostincurredby rmswhilechangingprices.Thepreviousargumentrequiredthenumberof rmsoperatingineachmarkettoexceedfour.Thiswasnecessary,sincethedeviationpayo¤0(k)wouldincrease,ifasingle rmheldsalesinmarketk,askforany rmnotholdingsales.Ifso,thelargestsustainablecartelwithasynchronizedsaleswouldsmallerthanwithsynchronizedsales,asstabilityisinverselyrelatedtothelowestpricechargedbyacompeting rm.Notethatthestraightforwardextensionofthemulti-marketmodeltoasymmetricmarketswouldgeneratesalestrategieswhicharenotsynchronizedacrossmarketsaswellaswithinmarkets.Let1denotethevariantofstrategyinwhichasingle rmhassalesinmarketAinperiodsmod(t;2{)=0andinmarketBinperiodsmod(t;2{)={.AgainconsideraneconomyinwhichS=1,=0:95,=0:15,v=10,andc=1.Fixthethreatdiscountofallthe rmsnotsellinginamarketkduringasalesperiodtok=0:2.Asexpected,thestabilityofstrategy1issmallercomparedtoascartelsizedeclineswheneverdeviationpro tsgrow: n{ m 20:09:001:0098+ 20:58:720:9492 28:61:960:1592 28:61:960:15921 24:71:960:159226 5ConclusionTheanalysispresentedanovelrationaleforsalesinanindustryinwhichahomogeneousstorablegoodisproducedbyn rms,andsoldtoconsumerswithaccesstoheterogeneousstoragetechnologies.Inthiscontext,thepaperexaminedthee¤ectsofheterogeneityinstorageon rmsincentivestoholdperiodicsalestosupportagreaterdegreeofcollusion.Insuchanenvironmentsalestrategieswereproventostrengthentheincentivestocollude,asstoragewouldintertemporallylinkconsumerdemandandthus,reducetheshort-rungainsfromadeviation.Inparticular,inanystableequilibrium rmswouldchargeinanyperiodofsalesabigenoughdiscounttoinduceallconsumerswithopenstoragecapacitytostockpileaquantitysu¢cienttosatisfyalltheirdemanduntilthenextsale.Suchbehaviorwasshowntoreduceincentivestodeviatebothinregularpriceperiods(asonlyconsumerswithoutstoragewouldpurchaseunitsinsuchperiods)andinperiodswithsales(bothbecausealowerpricewouldbechargedinsuchperiods,andbecauseconsumerswithstoragewouldreducetheirdemandifadeviationwereobservedinthewakeofanimminentpricewar).The rstpartoftheanalysis:characterizedconsumerdemand;characterizedthesetEofsalestrategieswhicharemorestablethananystrategywithoutsales;andpresentednecessaryandsu¢cientconditionsforthenon-emptinessofsuchaset.Thesecondpartoftheanalysisfocusedonsingle-unitstorageandcharacterizedthemoststableandthemostpro tablesalestrategiesinthesetE.Withinsuchsetatrade-o¤wasproventoemergebetweencollusivepro tsandcartelsize.Suchtrade-o¤anditsdependenceontheenvironmentwereexplicitlycharacterized.Relationshipbetweentheoptimalsalemarkdownandtheenvironmentwasalsoexplored.The nalpartoftheanalysisextendedthebaselinemodelintwodirections.The rstextensionconsideredthemulti-unitstoragescenario,characterizedtheoptimaltimingofsales,andshowedthataccesstomulti-unitstoragetechnologiescouldreducethepro t-stabilitytrade-o¤.Thesecondextensioninstead,provedthatsynchronizationinsaleswouldnotbenecessaryinmulti-marketsetup.Intheproposedmodelcollusionwasstrengthenedattheexpenseofaggregatepro ts,sincedeviationpro tswoulddeclinemorethanequilibriumpro tsifsalestrategieswereemployed.Aggregatepro tshowever,hadtodeclinewhen rmsheldsales,asconsumerswerehomogeneousintheirwillingnesstopay.Notehowever,thatifconsumerswithhigherstoragecapacityhada27 lowerwillingnesstopay,asalestrategycouldachievehigherpro tsthantheno-salestrategybypricediscriminatingamongdi¤erenttypesofconsumers,andcouldthusfostercollusionevenfurther.Valuationswerekepthomogeneousacrossconsumersonlytodisplaymoreexplicitlythee¤ectsoftheintertemporallinkinginconsumerdemand.Finally,notethattherationalityimposedonconsumerdemandrequiredthatallbuyerswouldunderstandtheconsequencesofadeviationonfutureprices.Thisassumptioncouldeasilyberelaxedbyintroducingbehavioralbuyerswhocanonlyimperfectlyforecastfutureprices.Allresultswouldstillholdqualitatively,eventhoughthestabilityofanygivenstrategymaydecline.References[1]Ausubel,L.andR.Deneckere,Oneisalmostenoughformonopoly,RANDJournalofEconomics,1987,18:2,pp.255-274.[2]Bell,D.R.andC.A.L.Hilber,AnEmpiricalTestoftheTheoryofSales:DoHouseholdStorageConstraintsA¤ectConsumerandStoreBehavior?,QuantitativeMarketingandEconomics,2006,4(2),pp.87-117[3]BernheimD.B.andM.D.Whinston.Multimarketcontactandcollusivebehavior,RANDJournalofEconomics,1990,21,pp.1-26.[4]Dana,J.D.andY.F.Fong,Long-LivedConsumers,IntertemporalBundling,andTacitCollusion,Mimeo,2006.[5]Dudine,P.,Hendel,I.andLizzeri,A.,StorableGoodMonopoly:TheRoleofCommit-ment,AmericanEconomicReview,2006,96(5),17061719.[6]Erdem,T.,M.KeaneandS.Imai,ConsumerPriceandPromotionExpectations:Cap-turingConsumerBrandandQuantityChoiceDynamicsunderPriceUncertainty,Quan-titativeMarketingandEconomics,2003,1,pp.5-64.[7]Gul,F.,NoncooperativeCollusioninDurableGoodsOligopoly,RANDJournalofEco-nomics,1987,18:2,pp248-254.28 [8]Hendel,I.andA.Nevo,MeasuringtheImplicationsofSalesandConsumerInventoryBehavior,Econometrica,2006.[9]Hendel,I.andA.Nevo,IntertemporalPriceDiscriminationinStorableGoodsMarkets,Mimeo,NorthwesternUniversity.[10]HongP.,P.McAfee,andA.Nayyar,EquilibriumPriceDispersionwithConsumerInven-tories,JournalofEconomicTheory,2002,105(2),pp.503-517.[11]Pesendorfer,M.,RetailSales.AStudyofPricingBehaviorinSupermarkets,JournalofBusiness,2002,75(1),pp.33-66.[12]Salop,S.andJ.E.Stiglitz,Bargainsandripo¤s:Amodelofmonopolisticallycompetitivepricedispersion,ReviewofEconomicStudies,1977,44(3),pp.493-510.[13]Salop,S.andJ.E.Stiglitz,Thetheoryofsales:Asimplemodelofequilibriumpricedispersionwithidenticalagents,AmericanEconomicReview,1982,72(5),pp.1121-1130.[14]SeilerS.,TheImpactofSearchCostsonConsumerBehavior:aDynamicApproach,2010,Mimeo,LondonSchoolofEconomics.[15]Schiraldi,P.andF.Nava,ResaleandCollusioninaDynamicMarketforSemidurableGoods,Mimeo,LondonSchoolofEconomics,2010.[16]Sobel,J.Thetimingofsales,ReviewofEconomicStudies,51(3),pp.353-368(1984).[17]SpagnoloG.OnInterdependentSupergames:MultimarketContact,Concavity,andCol-lusion,JournalofEconomicTheory,1999,89,pp.127-139.[18]Varian,H.Amodelofsales,AmericanEconomicReview,1980,70(4),pp.651-659.[19]VolkerN.andL.White,DoVerticalMergersFacilitateUpstreamCollusion?,AmericanEconomicReview,2007,97(4),pp.1321-1339.6AppendixPartA:Proofs29 Remark1If pzvinanyperiodz0,thedemandforconsumptiongoodattimet:(1)byconsumerswithoutstoragetechnologysatis esd0(pt)=1;(2)byconsumerswithstoragetechnologyandwithsunitsalreadyinstoragesatis es:dS(s;pt)=maxminT(pt);S+1 s;0 Proof.Part(1)oftheclaimistrivial.Toprove(2)noticethatbyconstructiondS(s;pt)2[0;S1+s].Theupper-boundmusthold,sincenoconsumercanstoremorethanSunits.Thelower-boundmustholdbecausenoplayercanbene tbydisposingalreadypurchasedunitsgiventhat pz0foranyz.Alsonoticethatonlypro lesofdemandsuchthatguaranteeaconsumptionstreamofaunitineveryperiodcanbeoptimal,sincepricessatisfy pzv.Thus,payo¤streamcanbecomparedbylookingonlyatthetotalexpenditureonconsumptiongood.Then,considerthecaseinwhichT(pt)S+1.Bycontradictionconsiderapro leofdemandforthesuccessiveT(pt)periods,fdt+zgT(pt)z=0andsupposethatdt6=maxfT(pt)s;0g.Ifso,thereexistsapro leofdemandsd0t+z T(pt)z=0thatcostslessandthatleavestheconsumerwithexactlyasmanyunitsstoredinperiodt+T(pt).Infact,consider:d0t=maxT(pt)s;0 d0t+z=0ifz20;T(pt)d0t+T(pt)=hPT(pt)z=0dzid0tbyconstructionthepro leleavestheconsumerwithexactlyasmanyunitsstoredinperiodt+T(pt).Moreoverd0costsless,since:PT(pt)z=0z pzdz=PT(pt)1z=0z pzdz+T(pt)pt+T(pt)dt+T(pt) ptPT(pt)1z=0dz+T(pt)pt+T(pt)dt+T(pt)== ptd0t+ pthPT(pt)1z=0dzd0ti+T(pt)pt+T(pt)dt+T(pt)== ptd0t+ ptd0t+T(pt)dt+T(pt)+T(pt)pt+T(pt)dt+T(pt)==ptd0t+T(pt)pt+T(pt)d0t+T(pt)+hT(pt)pt+T(pt) ptidt+T(pt)d0t+T(pt)ptd0t+T(pt)pt+T(pt)d0t+T(pt)30 giventhat:(i)PT(pt)1z=0dzd0t&dt+T(pt)d0t+T(pt),sincethatconsumersconsumeoneunitineveryperiod;and(ii)T(pt)pt+T(pt) ptz pt+zforanyz2(0;T(pt)).Thusacontradictionisestablished.AverysimilarandomittedargumentworksalsoforthecaseinwhichT(pt)S+1andestablishestheclaim. Remark2StrategyisaSPEofthein niterepetitionofthegameifandonlyif:11 nt tProof.Theproofoftheresultistrivial.Noplayerbene tsfromadeviationalongtheequilibriumpathif:t n(1)twheretdenotesthemostpro tabledeviation.Suchconditionisexploitedtopindowntherequirementonthecriticaldiscountrate.Moreovernodeviationcanbepro tableo¤theequilibriumpath,sinceallplayersmakeatmostzeropro tswhenallcompetitorsquotepricesatmarginalcost. Remark3Ifmod(t;{)=0,st=0.Ifmod(t;{)=mod(z;{),dt=dz.Proof.The rstclaimisprovenbyinduction.Notethatmod(0;{)=0ands0=0.WeshowthatiftheclaimistrueforanytTsuchthatmod(t;{)=0itistrueforanytT+{suchthatmod(t;{)=0.InfactconsiderthelargestdatetsuchthattTandmod(t;{)=0.Suchdateexistsbytheinitialconditionandtheinductionhypothesis.Atsuchdatethedemandofanindividualwithstoragesatis es:dS(0;pt)=minT(pt);S+1 MoreoverT(pt){,since(1+){(1+).Hence,dS(0;pt){andst{giventhatoneunitwillbeconsumed.Moreover,inanyperiodz2ft+1;:::;t+{g,since(1+)]TJ/;༴ ;.9;Ւ ;Tf 1;.80; 0 ;Td [;(1+)]TJ/;༴ ;.9;Ւ ;Tf 1;.80; 0 ;Td [;(1+),wehavethatT(pt)=1andconsequently:dS(st;pt)=8:0ifst]TJ ; -2;.52; Td ;[000;01ifst=0)st+1=8:st1ifst]TJ ; -2;.52; Td ;[000;00ifst=031 Whichestablishesthatst+{=0,sincest{.Thesecondclaimfollowsimmediately,sincefromthepreviouspartoftheproofitisstraightforwardtoobservethat:dS(st;pt)=8]TJ/;༴ ;.9;Ւ ;Tf 1;.69; 0 ;Td [;]TJ/;༴ ;.9;Ւ ;Tf 1;.69; 0 ;Td [;]TJ/;༴ ;.9;Ւ ;Tf 1;.69; 0 ;Td [;]TJ ; -2;.52; Td ;[000;]TJ ; -2;.52; Td ;[000;]TJ ; -2;.52; Td ;[000;:dS(0;p0)ifmod(t;{)=00ifst]TJ ; -2;.52; Td ;[000;0&mod(t;{)6=01ifst=0&mod(t;{)6=0(3) Proposition4Thefollowingclaimsmusthold:(1)SnC=V(2)NV(3)anystrategyinNismorestablethananystrategyinV;(4)anystrategythatsets=vc cand=1ispro tmaximizingwithinS;(5)foranystrategyinC,st=S(t)and:dt=8:0+{Sifmod(t;{)=00ifmod(t;{)6=0Proof.Toprovepart(1),webeginbyarguingthatSnCV.Considerastrategy2SnCforwhichtheconstraint(1+){1(1+)isviolated.Bythedemandstructureestablishedincondition(3)aperiodexistsinwhichmod(t;{)6=0anddSt]TJ ; -2;.51; Td; [00;1,sinced0S{.Therefore,everystrategyviolating(1+){1(1+)mustbelongtoV.Similarly,if{]TJ/;༴ ;.9;Ւ ;Tf 1;.44; 0 ;Td [;S+1aperiodwouldexistsinwhichdSt]TJ/;༴ ;.9;Ւ ;Tf 1;.44; 0 ;Td [;1andmod(t;{)6=0,andthestrategywouldagainbelongtoV.Thus,SnCV.Furthernotethatcondition(5)requiresC\V=;,andthusestablishes(1).Notethat(2)isimmediate,because=1impliesdS(st;pt)=1foranytas pt+1= ptrequires pt pt+1.Toprove(3), rstobservethatallstrategiesinNareequallystable.Notethat,bytheproofof(2),foranystrategy2Nequilibriumpayo¤ssimplifytot()=c.Thus,adeviatingplayercancaptureatmostsuchapro tbyundercuttingthepricemarginally.Anydeviationtoapricey2(c;(1+)c),mustsatisfyd(st;yt)1,since ptcandthereforet(y;)(yc).Hence,t()=candRt()=1foranyt2f0;1;:::gandany2N.Now,considerastrategy2Vandaperiodtinwhichdt0and pt=(1+)c.Notesuchconditionsimplythat32 st=0anddt1.Ifso,by pt=(1+)c,wegetthatt(y;)=(yc)d(st;yt)=(yc),andt()=c.Moreover,ifsuchaperiodexists,itmustbethat:dS0=minT(p0);S+1 {becauseoftheevolutionofsavingsanddemanddiscussedintheperviouslemma(condition3).Inturnthisrequiresthatt1(1+)(1+)]TJ/;༴ ;.9;Ւ ;Tf 1;.42; 0 ;Td [;t(1+)forsomet2f1;:::;{1g.Ifso,pickthesmallesttforwhich(1+)]TJ/;༴ ;.9;Ւ ;Tf 1;.42; 0 ;Td [;t(1+)andnoticethat:t()=(1) 1{hP{1z=0zdt+z(1)S(t)d0ic==(1) 1{hhPS(t)1z=0z+0P{1z=S(t)+1zi+S(t)(0+St)ic==00t()+SSt()cwherethelastinequalitymustholdsince:0t()=1S(t)+S(t)+1{+(S(t)S(t)+1) 1{ccSt()=1S(t)+t(S(t)S(t)+1) 1{c==1{t+t({t{t+1) 1{cc,1t t(1)Theinequalitybounding0t()musthold,sinceitcannotbepro tabletocutpricesoncon-sumersthatdonotaltertheirdemand.TheinequalityboundingSt()mustholdinstead,since rmsprefertodelayproductioncostsandbecauset1(1+)(1+)requires:1+ 1+t1Pt1z=0z t=1t t(1)Hence,astrategy2VcannotbemorestablethanastrategyinN,sinceRt()1.Theproofof(4)istrivial.Theproposedstrategyraisesapro tofvc,sincedt=1foranyt.Nostrategyinwhichdt=1foranytcandobetter,sincevisthehighestpricethatabuyerwillingpayforaunitofconsumption.Butanyotherstrategysuchthatdt6=1forsometmustsatisfydS01,bythepropertiesofthedemandfunctionderivedincondition(3).In33 turn,ifdS01,itmustbethat(1+)(1+).Thusby(3),wegetthatpro tscanbeexpressedasfollowsforsomedS02(1;{]:0()=(1) 1{P{1z=0zdz(1)d0c==(1) 1{hh0PdS01z=1z+P{1z=d0Szi+(0+SdS0)icAnargumentsimilartotheonedevelopedinthepreviouspartoftheproofshowsthat0()c.Inparticular,writepro tsas0()=000()+SS0()andnoticethatforthesamereasondescribedinpart(3)00()c.Thenlett=d0snoticethat:S0()=t{+t(1) 1{cc,1t t(1)wheretheinequalityboundingS0()isestablishedbyt1(1+)(1+)asinpart(3).Thisestablishes(4),sincecvcisnecessaryforpro tstobemaximalbythepropertiesofthedemandfunction.Part(5)followstriviallyfromcondition(3)andthedemandfunctionsofbothtypesofconsumersdiscussedinthetext. Proposition5AnystrategyinEismorestablethananystrategyinSnE.Proof.FirstweestablishthatastrategyEinismorestablethanastrategyinN.Considerastrategy2E.Byde nitionofE,0 {S+0,andthereforewegetthatR1()1andthatforanyt2f1;2;:::;{2g:Rt()=1+(1) 1{{t{S+0 01Rt+1()Hence,thestabilityofastrategyin2EwillbepinneddownbytheminimumbetweenR0()andR1().Moreover,R0()1sinceforanyandforany{1:R0()=0 +1 1xh({S+0)0 i0+(1){ 1xS=0+{ 1++:::+x1S134 wherethe rstinequalityholdssincedR0()=d0.WhichestablishesthatifastrategybelongstoEthenitmustbemorestablethananystrategyinN,sincemint0Rt()1.SinceanystrategyinNismorestablethananystrategyinV,whatremainstobeprovenisthatanystrategyinEismorestablethanstrategiesinCnE.Butthisisimmediatesince2CnEimplies0 {S+0,andthusR1()1. Corollary6Econtainsastrategywithacycleoflength{2f2;:::;S+1gifandonlyif:{1vc v0 {0({1)+c v(4)Thus,ifvc v0 20+c v,thenEisnon-empty.Proof.First,weestablishthat4impliestheexistenceofastrategywithacycleoflength{2f2;:::;S+1ginE.Letconstraint4holdforsome{2f2;:::;S+1g.Takeanystrategythatsets=v c1and20 S{+0;{11+1 1 (5)atthegivenvalue{.ThestrategyobviouslybelongstoE.Moreover,suchastrategyexistssincetheintervalinwhichwaschosenisnon-empty,whenever4holdsat{.Nextweestablishthenecessityof4.AnystrategyinEmustsatisfy5byconstruction.Consideranyoneofthesestrategies,andnoticethat:0 S{+0;{11+1 1 0 S{+0;{1v vcc vcSincethenon-emptinessofthebiggerintervalisequivalentto4,wegetthat4beingviolatedpreventtheexistenceofapolicywithcyclelength{inE.Thisestablishesthenecessity.Thelastobservationisatrivialcorollary. Proposition7IfE6=;,nostrategyinSisstrictlymorestablethanstrategy2E: { 1= minf(2);(2)g 2 Moreover,isthemostpro tableofallthestrategiesinSwithequalstability.35 Proof.Thisfollowsfromtheproofofproposition(12). Proposition8IfE6=;,nostrategyinEisstrictlymorepro tablethanstrategy+2E: + + {+ v=c1 (2) 2 Proof.Thisfollowsfromtheproofofproposition(13). Proposition9IfE6=;,foranypro tlevel2(0;m]themaximalnumber rmsthatcancolludeonwhileemployingasalestrategysatis es:N()=8]TJ ; -2;.52; Td ;[000;]TJ ; -2;.52; Td ;[000;]TJ ; -2;.52; Td ;[000;:1 1if2(+;m]1 1(2) (1+)(vc)if2(;+]n()if2(0;]Proof.Firstnotethatif+nostrategyinEismorepro tablethan+.Thus,nosuchpro tlevelcanbesustainedbyasalestrategybelongingtoE.Ifso,themoststablestrategyisonewithoutsales.However,allstrategiesinNareequallystablebyproposition(7)andthus,N()=1 1foranysuchstrategy.Thensupposethatandconsideranystrategy(;;2)withpro ts.Notethatimpliesthateitheror.Alsonotethatadi¤erentstrategy(;;2)existswhichraisesexactlythesamepro ts,sinceanypro tlevelcanbeobtainedbypicking2(0;].Thusobservethatstrategy(;;2)isequallystabletostrategy=(;;2)andthusmorestablethan(;;2).Finallyconsiderthecaseinwhich2(;+].Notethatforthistobethecaseitmustbethat+,whichinturnrequires=(2)(2)=+Notethatsetting=isalwaysoptimalforbothpro tsandstability.Thus,foranypro tlevel2(;+]acorrespondingsalesdiscountexists()2(;+]whichsustainspro t36 level.Suchadiscountisfoundbysolvingthefollowingequalitywithrespectto:= 1++(2)1 1+c)()=1 2 vc(1+)ThevalueofN()insuchintervalcanthenbefoundbycomputingn(()):N()=n(())=1 1R0(;();2)=1 1 ()c==1 1(2) (1+)(vc)whichestablishesthedesiredresult. Proposition10ThesizeofthesetEdecreaseswithcand,andincreaseswithvand.Proof.Firstnotethatbycorollary9thesu¢cientconditionfortheexistenceofasalestrategywithperiod2requires: 2+c v1 2=h(6)Furthernoticethatsuchconditioniseasiertosatisfywhentherighthandsideof(6)issmaller.Thisinturnimpliesthedesiredresultssince:dh d=vc v2 (2)20dh dv=c v21 20dh dc=1 v1 20The nalobservationonistrivialthelefthandsideof(6)increasesin. Proposition11IfE6=;,foranypro tlevel2(0;m]themaximalnumber rmsthatcan37 colludeonwhileemployingasalestrategysatis es:dN() dddvdc 2(+;m] +0002(;+] +?+2(0;]\6=+ +002(0;]\=+ ++wheredN()=d0ifandonlyif=(2m).Moreover,thecut-o¤pro tlevelsand+andthemapsandevaluatedat{=2and(2)(2)furthersatisfy: dddvdc d+ +++d ++d +00d +0+Proof.Firstnotewhen2(+;m]thesignofallthederivativesofN()=1=(1)istrivial.Next,considerthecaseinwhich2(;+].Notethatwithinsuchinterval2(;+]and:dN() d=1 1[2(vc)(1+)] ((1+)(vc))20,2(vc)(1+)dN() d=1 (1)2(2)[2((vc))+(vc)] ((1+)(vc))20dN() dv=dN() dc=1 1(2) ((1+)(vc))20Thesecondinequalityholds,since(vc)andtogetherimply(vc)(wherethe rstconditionholdssince00forthestrategytobelongtoE,andwherethesecondconditionholdssinceandsincetheonlypositiverootof(2)satis es,asexplainedinappendixB).Beforeweproceed nalscenario2(0;],letusprovealltheremainingresults.First,observethatd(2)=dv=d(2)=dc=0,sincebothR0andR1areindependentofvaluesand38 costs(seeappendixB).Furthernotethatbytheimplicitfunctiontheoremappliedtothemap(2):=d(2) d=R1R0 R1R0=(1) [2(2)(1)]=d(2) d=R1R0 R1R0=(+1)+2() [2(2)(1)]Moreover,notethatd(2)=d0,since2(2)]TJ/;༕ ;.9;Ւ ;Tf 3;.31; 0 ;Td [;(1)byde nitionof;andthatintheonlyrelevantscenario(i.e.(2)]TJ/;༕ ;.9;Ւ ;Tf 3;.31; 0 ;Td [;(2))d(2)=d]TJ/;༕ ;.9;Ւ ;Tf 3;.31; 0 ;Td [;0,since]TJ/;༕ ;.9;Ւ ;Tf 3;.31; 0 ;Td [;(2)]TJ/;༕ ;.9;Ւ ;Tf 3;.31; 0 ;Td [;(2)=.Also,notethatR1]TJ/;༕ ;.9;Ւ ;Tf 3;.31; 0 ;Td [;0,R00,R10andR00.Thesignofthederivativesofthemap(2)followtriviallyfromitsde nition.Thennotethat+anditsderivativeswithrespecttotherelevantparameterssatisfy:+= 1+(vc)+2 1+(vc)d+ d=2v+2(1)c (1+)20&d+ d=c(1) 1+0d+ dv=2 1+0&d+ dc=2(1) 1+0Tocomputethederivativesof,considerthecaseinwhich(2)]TJ/;༕ ;.9;Ւ ;Tf 1;.42; 0 ;Td [;(2).Orelse,and+andtheirrespectivederivativeswouldcoincide.Ifso:= 1++2 1+(2)(vc)d d=1 (1+)2(2)(2)+(2)(1+)d(2) d(vc)0d d=(2) 1++2 1+d(2) d(vc)0d dv=d dc= 1++2 1+(2)0wherethe rstinequalityholdssinceE6=;implies(2)(2)0,andthesecondinequalityholdssince(2)(2)implies(2).Atlast,considerthecaseinwhich39 2(0;].Supposethat6=+.Ifso,(2)(2)andtherefore:dN() d=1 (1)2[R0+(1)[R0+R0]]0(7)dN() d=1 1[R0+R0]0(8)dN() dv=dN() dc=0(9)where(7)ispositivesinceR00andbecause:R0+(1)R0= +2 (1+)2h(2) i0where(8)isnegativesince:R0+R0=R1R0R0R1 R1R0==1 R1R0 (1+)2[+(2)]0andwhere(9)holdstrivially.Finally,considerthecaseinwhich2(0;]and=+.Ifso,(2)(2)and:dN() d=1 1222+2 (1+2)+(12)0dN() d= 122 20dN() dv= 122 v0&dN() dc= 122 c0whichconcludestheproof. Proposition12AssumethatE6=;.If{isde nedandifR0(({);{)R1(({);{)when-ever{isalsode ned,thennostrategyinSisstrictlymorestablethanstrategy2E: { 1= ({) { 40 otherwise,nostrategyinSisstrictlymorestablethanstrategy2E: { 1= ({) { Moreover,isthemostpro tableofallthestrategiesinSwithequalstability.Proof.Toprovetheclaimitsu¢cestoshowthatismorestablethananyotherstrategyinE.Consideranyotherstrategy=(;;{)2S.Firstletusestablishthatif2E,thenthesalestrategy({)=(1=;minf({);({)g;{)alsobelongstoEandismorestablethan.Notethat:({)0 S{+0{11+1 1 ({)wherethe rstinequalityholdsby2E,andthesecondsince1=.Moreover,minf({);({)g2[({);({)]since({)({)giventhatR1(;{)1R0(;{)forany({).Thus,({)2E.Toprovethat({)ismorestablethan, rstnotethatthemarkupdoesnota¤ectmint0Rt()andincreases({),whichinturnimpliesthatsettingtoitsupper-boundcannotreducethestability.Thennotethatmint0Rt(;{)issinglepeakedin2[0;1],sinceR0(;{)decreasesin,sinceR1(;{)increasesin,andsinceR0(0;{)-286;R1(0;{)(seeappendixpartBfordetail).Alsonoticethatthepeakmint0Rt(;{)withrespecttoisachievedexactlyat=({).Thus,if({)({),nostrategywiththesamecyclelengthcanbemorestablethan({).Ifhowever,({)-286;({),themoststablestrategymustsatisfy=({),sincemint0Rt(;{)increasesinfor({).Nextobservethatbytheimplicitfunctiontheoremwegetthat:0({)=R1{R0{ R1R0041 Notethatthedenominatoristriviallypositive(seeappendixB),andthatthenumeratorisnegativesinceat=({):R0{R1{=1 1{S1{1 0+{1log 1{0 1 ({)10wherethe rsttermispositivesince({)0={1,andwherethesecondtermispositivesince({)minf({);0g(seeappendixBfordetails).Hence,since0({)0,thereexistsauniquevalue{suchthat({)=({).Letf({)=(({)=({))1.Notethatforasalestrategytobemorestablethanastrategywithoutsalesitmustbethatf({)0.Furthermore:@ @{R1(({);{)={1{ 1{f0({)+log 1{f({){1{ 1{f0({)1 1{f({){1{ (1{){[{f0({)f({)]0wherethe rstinequalityholdssincelog1,andthesecondsince1{{(1).Thethirdinequalityinstead,holdssince:{f0({)f({)=1 ({){0({){0({)({) ({)({)+({)==1 ({)[log{(({)+1=)+({)(1({))({)+({)]==1 ({)log{({)+1 +({)(1({))==1 ({)1+1 log{{1+({)1{11 ({)1+1 ({1){1+({)1{11 ({)1+1 (({){1)1{1042 wherethe rstinequalityholds,sincelog{{1,wherethesecondholdstrivially,andwherethelastinequalityholdssince({)({)isequivalentto:{1({)++1 (1({))({)Thelastfewobservationstogetherestablishedthatif({)({)forsome{,thenincreasingthecyclelengthwouldonlyreducethestabilityofthesalestrategy({).Inturnthisestablishesthatsetting{{cannotimprovestability.Finally,notethat,if{exits,nostrategywithperiod{{canbemorestablethan(1=;({);{)byde nitionof{.Thus,themoststablesalestrategywillbeeither(1=;({);{)or(1=;({);{)dependingontherelativestabilityofthetwo.Theobservationaboutpro tsfollowstrivially,sincechangingand{wouldnecessarilyreducestabilitybyconstructionofandbecause=v c1raisesthehighestpro tandcannotlowerstability. Proposition13IfE6=;,nostrategyinEisstrictlymorepro tablethanstrategy+2E: + + {+ v=c1 (2) 2 Proof.Bythepropertiesofthetimezeropro tfunctiondiscussedinappendixB,pro tsattime0increasein,,and{.Thus,themostpro tablestrategyinEwithacycleoflength{musttriviallysatisfy=v c1and=({),since{11+1 1 increasesin.Thus,(2)followsimmediatelysince{ischosenbyde nitionsotomaximizepro tsinEandsince:@0(;({);{) @{=0c(1)2{1(1+) (1{)2log+Sc(1) (1{)2(1{)({1(1+)1)+{1((1+))logxSc(1) (1{)2({1(1+)1)(1{)+{1((1+))logxSc(1) (1{)2({1(1+)1){1((1+))(1{)==Sc(1)2 (1{)043 wherethesecondinequalityholdslogxx1andtherestissimplealgebra. Proposition14ThesizeofthesetEdecreaseswithcand,andincreaseswithvand.Proof.Firstnotethatbycorollary9thesu¢cientconditionfortheexistenceofasalestrategywithperiod{requires:{11 1+ {({1)+ 1+=h(;)(10)Furthernoticethatsuchconditionishardertosatisfywheneitherorincreasesince:dh(;) d=1 1+{ ({({1))20dh(;) d=1 (1+)2(1){ {({1)0Thus,thesizeofthesetEdecreaseswithbothand.Toestablishthecomparativestaticsoncandv,simplynotethatd=dv0andthatd=dc]TJ/;༕ ;.9;Ւ ;Tf 4;.99; 0 ;Td [;0.The nalobservationonistrivialthelefthandsideof(10)increasesin. PartB:DerivativesandSignsRecallthatforanystrategy2Cequilibriumanddeviationpro tsinthetwocriticalperiodsrespectivelysatisfy:0()=0+1 1x[({S+0)0]c1()=0+{1{ 1{[({S+0)0]c0()=c1()=0c44 Derivativesatt=0:d0() d=0+1 1x[({S+0)0]c0d0() d=1 1{({S+0)c0d0() d{=1 1{S+log{{+1 (1{)2[({S+0)0]c0d0() d=c0&d0() d=c0&d0() d{=0Tosignd0()=d{considerhardercase,namelyS=1.Ifso:d0() d{=1 1{1+{log{ 1{c0whichispositive,since:xlog=logx11 {Similarly,derivativesatt=1,satisfy:d1() d=0+{1{ 1{[({S+0)0]c0d1() d={1{ 1{({S+0)c0d1() d{={1{ 1{S+log{1{ (1{)2[({S+0)0]cd1() d=0c&d1() d=d1() d{=0Again,tosignd1()=d{considerhardercase,namelyS=1.Ifso:d1() dx={1{ 1{1+{log1 1{c0whichisnegative,since:log{{145 Moreoverd1()=d{0,when0=1.Thus,thesignofd1()=d{dependsonthefractionofconsumerswithstorageintheeconomy.Noticethattheresultingcriticalratiosareindependentof:R0()=0 +1 1xh({S+0)0 iR1()=1+{1{ 1{({S+0) 01Derivativesatt=0:dR0() d=x 1x0 20dR1() d={1{ 1{{S+0 00dR0() d{=1 1{S+log{{+1 (1{)2h{S+00 i0dR1() d{={1{ 1{S 0+log{1{ (1{)2{S+0 01wherethesignofdR1()=d{coincideswiththatofd1()=d{.NoticethatR1()R0()=0requires:2x1S{+0 0+1{1 1[S{+0]0{ 1=0Suchconditionalwayshasuniquepositivesolutionwhichsatis es2[0;0={1],sinceitisnegativebothat=0andat=0,andpositiveat=0={1.However,thesolutioncouldinprinciplerequire1.Ifso,thesolutiontothegeneralprogrammax2[0;1]mintRt()=R1()willsatisfy({)=minf1;g.46