Vol.58,No.3,March2012,pp.570–586
ISSN0025-1909(print) ISSN1526-5501(online)
MANAGEMENT SCIENCE
doc.guandang.net
©2012INFORMS
BayesianDynamicPricingPolicies:Learningand
EarningUnderaBinaryPriorDistribution
J.MichaelHarrison,N.BoraKeskin
GraduateSchoolofBusiness,StanfordUniversity,Stanford,California94305
{harrison_michael@gsb.stanford.edu,keskin_bora@gsb.stanford.edu}
AssafZeevi
GraduateSchoolofBusiness,ColumbiaUniversity,NewYork,NewYork10027,assaf@gsb.columbia.edu
M
otivatedbyapplicationsin nancialservices,weconsiderasellerwhoofferspricessequentiallytoastreamofpotentialcustomers,observingeithersuccessorfailureineachsalesattempt.Theparametersoftheunderlyingdemandmodelareinitiallyunknown,soeachpricedecisioninvolvesatrade-offbetweenlearningandearning.Attentionisrestrictedtothesimplestkindofmodeluncertainty,whereoneoftwodemandmodelsisknowntoapply,andwefocusinitiallyonperformanceofthemyopicBayesianpolicy(MBP),variantsofwhicharecommonlyusedinpractice.BecauselearningispassiveundertheMBP(thatis,learningonlytakesplaceasaby-productofactionsthathaveadifferentpurpose),itcanleadtoincompletelearningandpoorpro tperformance.However,underoneadditionalassumption,aconstrainedvariantofthemyopicpolicyisshowntohavethefollowingstrongtheoreticalvirtue:theexpectedperformancegaprelativetoaclairvoyantwhoknowstheunderlyingdemandmodelisboundedbyaconstantasthenumberofsalesattemptsbecomeslarge.
Keywords:revenuemanagement;pricing;estimation;Bayesianlearning;exploration–exploitation
History:ReceivedJanuary15,2010;acceptedJune17,2011,byGérardP.Cachon,stochasticmodelsandsimulation.PublishedonlineinArticlesinAdvanceOctober14,2011.
1.Introduction
Weconsiderinthispaperaproblemthatwas rsttargetedforstudyatleast35yearsago,isrelativelysimpleinstructure,andiswidelyconsideredfun-damental,butisalsounsolved.Brie ystated,theproblemisthatofsequentialpricingwhentheunder-lyingdemandmodelisunknownandthemarketresponsetoanygivenpriceisconfoundedbystatisti-calnoise.Inthissituationthesellerconfrontsatrade-offbetweenexplorationofthedemandenvironment(learning)andexpectedimmediatepro t(earning).1.1.OverviewoftheProblemandApproach
Theparticularvariantofthelearning-and-earningproblemthatweconsiderherehastwosalientfea-tures:(a)Thesellerofferspricessequentiallytoindi-vidualcustomers,observingeithersuccessorfailureineachsalesattempt.Thereisanunderlyingdemandmodel,alsocalledademandfunctionordemandcurve,thatgivestheprobabilityofsuccessasafunctionofthepricethatisoffered.(b)Theparametersoftheunderlyingdemandmodelare xedbutinitiallyuncertain,asopposedtoproblemswherethedemandmodelitselfischangingovertime.
Modelfeature(a)correspondstowhatPhillips(2005)callscustomizedpricing;Phillips(2005,Chap.11)
570
listsanumberofimportantapplicationareasforcus-tomizedpricing,includingbothbusiness-to-businessandbusiness-to-consumerapplications.Oneofthoseisthepricingofconsumercredit,suchasautoloansandcreditcardlending,wherepotentialborrowersapplyindividuallyforloans,allowingbankstoofferdifferentpricestodifferentapplicantsandcompileuncensoreddataonsuccessesandfailures.
Thissituationoffersobviousopportunitiesforpriceexperimentation,butexperimentationissurprisinglyrareinpractice(Phillips2010);itiscommonto ndbanksthat xformonthsorevenyearstheinterestratetheychargeforloansofagiventype.Salesdataallowthebanktoestimatewithreasonableaccuracytheaveragedemandrateunderthatparticularinter-estrate,referredtohereafterastheincumbentprice,butthebankremainsuncertainabouttheaveragedemandratesachievablewithalternativeprices.
Followingcommonpracticeintherevenueman-agementindustry,wethinkintermsofaparametricmodelclass(suchasthelogitfamilyofdemandfunc-tions,ortheprobitfamily,orthelinearfamily)andadoptaBayesianformulation,usingapriordistri-butionovermodelparameterstoexpressuncertaintyaboutthedemandenvironment.Alsoinconformitywithstandardpractice,wetakeasourpointof
Dynamic Asset Pricing Theory_英语考试_外语学习_教育专区 暂无评价0人阅读0次下载举报文档 Dynamic Asset Pricing Theory_英语考试_外语学习_教育专区。Dynamic ...
Dynamic Restart Policies Henry Kautz University of Washington kautz@cs....We describe background work on the application of Bayesian methods to build ...
Dynamic Bayesian Network (DBN) with Structure Expectation Maximization (SEM) for Modeling o_专业资料。Exploring gene regulatory network is a key topic in...
Bayesian Dynamic Factor Models and Portfolio Allocation Omar AGUILAR and Mike WEST We discuss the development of dynamic factor models for multivariate ?...
Bayesian Dynamic Factor Models and Portfolio Allocation We discuss the development of dynamic factor models for multivariate financial time series, and the ...
Dynamic Bayesian estimation of displacement parameters of continuous curve box based on Novozhi_专业资料。维普资讯 http://www.cqvip.com Api tetsn cai(...
免费 A Bayesian Dynamic Model... 暂无评价 28页 免费 A Dynamic Algorithm for ... 暂无评价 8页 免费 Meta-modelling and graph... 暂无评价 16页 免费...
A dynamic Bayesian network is described by a directed acyclic graph in which nodes represent stochastic variables and arrows represent temporal dependen- 6 A...
DYNAMIC BAYESIAN NETWORKS FOR MEETING STRUCTURING This paper is about the automatic structuring of multiparty meetings using audio information. We have used ...
热门文档