简介:Theaimofthispaperistodevelopamulti-periodeconomicmodeltointerprethowthepeoplebecomeoverconfidentbyabiasedlearningthatpeopletendtoattributethesuccesstotheirabilitiesandfailurestootherfactors.TheauthorssupposethattheinformedtraderdoesnotknowthedistributionoftheprecisionofhisprivatesignalandupdateshisbeliefonthedistributionoftheprecisionofhisknowledgebyBayer'srule.Theinformedtradercaneventuallyrecognizethevalueoftheprecisionofhisknowledgeafteranenoughlongtimebiasedlearning,butthevalueisoverestimatedwhichleadshimtobeoverconfident.Furthermore,basedonthedefinitionontheluckiertraderwhosucceedsthesametimesbuthasthelargervarianceoftheknowledge,theauthorsfindthattheluckiertheinformedtraderis,themoreoverconfidenthewillbe;thesmallerthebiasedlearningfactoris,themoreoverconfidenttheinformedtraderis.Theauthorsalsoobtainalinearequilibriumwhichcanexplainsomeanomaliesinfinancialmarkets,suchasthehighobservedtradingvolumeandexcessvolatility.
简介:Theslowconvergencerateofreinforcementlearningalgorithmslimitstheirwiderapplication.Inengineeringdomains,hierarchicalreinforcementlearningisdevelopedtoperformactionstemporallyaccordingtopriorknowledge.Thissystemcanconvergefastduetoreducedstatespace.Thereisatestofelevatorgroupcontroltoshowthepowerofthenewsystem.Twoconventionalgroupcontrolalgorithmsareadoptedaspriorknowledge.Performanceindicatesthathierarchicalreinforcementlearningcanreducethelearningtimedramatically.
简介:AninformationtheorymethodisproposedtotesttheGrangercausalityandcontemporaneousconditionalindependenceinGrangercausalitygraphmodels.Inthegraphs,thevertexsetdenotesthecomponentseriesofthemultivariatetimeseries,andthedirectededgesdenotecausaldependence,whiletheundirectededgesreflecttheinstantaneousdependence.Thepresenceoftheedgesismeasuredbyastatisticsbasedonconditionalmutualinformationandtestedbyapermutationprocedure.Furthermore,fortheexistedrelations,astatisticsbasedonthedifferencebetweengeneralconditionalmutualinformationandlinearconditionalmutualinformationisproposedtotestthenonlinearity.Thesignificanceofthenonlinearteststatisticsisdeterminedbyabootstrapmethodbasedonsurrogatedata.Weinvestigatethefinitesamplebehavioroftheprocedurethroughsimulationtimeserieswithdifferentdependencestructures,includinglinearandnonlinearrelations.
简介:Inthispaper,aKELM-basedensemblelearningapproach,integratingGrangercausalitytest,greyrelationalanalysisandKELM(KernelExtremeLearningMachine),isproposedfortheexchangerateforecasting.Thestudyusesasetofsixteenmacroeconomicvariablesincluding,import,export,foreignexchangereserves,etc.Furthermore,theselectedvariablesarerankedandthenthreeofthem,whichhavethehighestdegreesofrelevancewiththeexchangerate,arefilteredoutbyGrangercausalitytestandthegreyrelationalanalysis,torepresentthedomesticsituation.Then,basedonthedomesticsituation,KELMisutilizedformedium-termRMB/USDforecasting.TheempiricalresultsshowthattheproposedKELM-basedensemblelearningapproachoutperformsallotherbenchmarkmodelsindifferentforecastinghorizons,whichimpliesthattheKELM-basedensemblelearningapproachisapowerfullearningapproachforexchangeratesforecasting.
简介:Detectionandclarificationofcause-effectrelationshipsamongvariablesisanimportantproblemintimeseriesanalysis.Thispaperprovidesamethodthatemploysbothmutualinformationandconditionalmutualinformationtoidentifythecausalstructureofmultivariatetimeseriescausalgraphicalmodels.Athree-stepprocedureisdevelopedtolearnthecontemporaneousandthelaggedcausalrelationshipsoftimeseriescausalgraphs.Contrarytoconventionalconstraint-basedalgorithm,theproposedalgorithmdoesnotinvolveanyspecialkindsofdistributionandisnonparametric.Thesepropertiesareespeciallyappealingforinferenceoftimeseriescausalgraphswhenthepriorknowledgeaboutthedatamodelisnotavailable.Simulationsandcaseanalysisdemonstratetheeffectivenessofthemethod.
简介:Thisresearchexaminesthemicro-levelcorrelationbetweentraditionalmarketingactions(TVadsandpublicrelations)andpre-releaseconsumers’sociallearningaboutvideogameconsoles(WiiandPS3,launchedin2006).Weevaluateconsumers’learningprocessesviatheperusalofinformationinonlinecommunitiesusing“pageview”dataformultiplewebsitesfromaclickstreampanelasindicators.WeproposeabivariateBayesianlearningmodelcombinedwithcomplementarypurchasechoices.Theproposedmodelenablessimplerestimationofparametersandallowstoaccommodatedetailedinformationaboutinteractionsbetweensocialandpersonallearningprocesses.Fromtheresults,wefindempiricalevidencethatcompanies’traditionalmarketingactionshaveagreaterimpactonsociallearningthanonregularpersonallearningduringthepre-launchperiod.Whenconsumersmakepurchasedecisions,theirsocialbeliefsaboutproductqualityareweighedatleastthreetimesmoreheavilythantheirpersonalbeliefs.Counterfactualsimulationssuggestthatbyoptimizingmarketingactions,firmscanstimulateconsumers’learningandpromoteincreasedproductengagement.
简介:Abankhadbetterrationitscreditifitfirstentersagraysystemofcreditinformationwhereitcannotdistinguishbetweenthelow-andhigh-riskborrowers.Asthebankkeepsalong-termrelationshipwithitsborrowersthebanklearnsabouttheborrowersthroughtime.WiththehelpoflogitmodelandBayesrule,abankcanprocesscustomer'screditinformationandbuildabettercredittermitgives.
简介:在最近的年里,电子learning系统在许多公司作为一个训练工具被采用了,当它能帮助雇员到在一短时间并且以低成本的完全的训练。在这上下文以内,自从2003,在台湾的经济事情(MOEA)的部的工业开发局(IDB)每年赞助了超过50家企业帮助他们实现设定的电子learning系统。关于电子learning的过去的研究主要集中了于一个标准并且采用了回归途径,例如检验在满足或继续的用法意愿和各种各样的先前的因素之间的关系。尽管回归方法能揭开电子learning采纳的重要因素,他们不能显示出每个因素的罐头(weighting),并且因此,公司不能分配资源的最佳的层次。克服这个问题,这研究收集了与从文学和专家意见的评论实现设定的电子learning系统的公司有关的因素,然后构造了一张层次因素桌子。基于因素的适当和独立,并且是由IDB,电子learning方案供应商和IDB的工程领导人的工程提升办公室的专家估计了赞助企业,这篇论文然后设计了一个分析层次过程(AHP)格式问询表并且送了它给这些一样的专家收集他们的回答。因素weightings基于模糊分析层次过程(FAHP)方法被导出。最后,关键因素的分析被执行,建议提供了。基于这个工作的结果,有有限资源的服务工业公司能完成合适的资源分配并且更快速并且成功地采用设定的电子Learning系统。
简介:在变量之中的原因效果关系的察觉和澄清及时是一个重要问题系列分析。传统的诱发性推理方法有模型一定是线性的突出的限制并且与Gaussian噪音。尽管添加剂模型回归能有效地推断添加剂的非线性的原因的关系非线性的时间系列,它受不了变量的同时期的原因的关系一定线性、不总是有效测试有条件的独立关系的限制。这份报纸提供采用相互的信息和有条件的相互的信息识别非线性的时间系列模型的一个班的原因的结构的一个nonparametric方法,它扩大到非线性的结构的向量autoregressive的非线性的时间系列建模的添加剂。一个算法被开发学习同时期并且变量的落后原因的关系。模拟表明建议方法的有效性。
简介:Whilelastdecadehaswitnessedarapidgrowthofdigitaleconomy,thereislimitedunderstandinginliteratureonwhethertheconventionalwisdomonpricingstrategystillholdsforinformationgoods.Ononehand,informationgoods,similartodurablegoods,aresubjecttovaluedepreciation;ontheother,theydifferfromtraditionalgoodsinnegligiblemarginalcostandthesensitivitytosocialinfluences.Thispaperdevelopsatwo-period,game-theoreticmodeltoinvestigateoptimalpricingstrategyofinformationgoods.Ononedimension,twodifferentdepreciationmechanisms(self-andtime-depreciation)areconsidered;ontheother,twoprevalentpricingschemes(perpetuallicensingandsubscription-feemodels)arestudied.Weobtainclosed-formsolutionsinallscenarios.Ourfindingssuggestthatvendorsoftime-depreciationinformationgoodsshouldadoptsubscription-feemodeltoattractearlyadoptersandexploitsocialinfluences,whilethevendorsofself-depreciationinformationgoodsshouldstrategicallybalancebetweendepreciationandsocialinfluences.Interestingly,associalinfluencesbecomestrongenough,thedifferencebetweenpricingschemesdiminishesandthetradeoffbetweencandidatestrategiesvanishes.Wealsoextendthemodeltostaticpricinginwhichthevendorcommitstofutureprice.Wediscoverthatthesuperiorityofsubscription-feemodelmightbeoverturnedunderstaticpricing.Ourresultsabovealsoimplythatbuildingconsumerfeedbackandinteractionsystemscouldbehelpfulforminimizingthepotentiallossofasuboptimalpricingscheme.
简介:High-frequencystocktrendpredictionusingmachinelearnershasraisedsubstantialinterestinliterature.Nevertheless,thereisnogoldstandardtoselecttheinputsforthelearners.Thispaperinvestigatestheapproachofadaptiveinputselection(AIS)forthetrendpredictionofhigh-frequencystockindexpriceandcomparesitwiththecommonlyuseddeterministicinputsetting(DIS)approach.TheDISapproachisimplementedthroughcomputationoftechnicalindicatorvaluesondeterministicperiodparameters.TheAISapproachselectsthemostsuitableindicatorsandtheirparametersforthetime-varyingdatasetusingfeatureselectionmethods.Twostate-of-the-artmachinelearners,supportvectormachine(SVM)andartificialneuralnetwork(ANN),areadoptedaslearningmodels.AccuracyandF-measureofSVMandANNmodelswithboththeapproachesarecomputedbasedonthehigh-frequencydataofCSI300index.TheresultssuggestthattheAISapproachusingt-statistics,informationgainandROCmethodscanachievebetterpredictionperformancethantheDISapproach.Also,theinvestmentperformanceevaluationshowsthattheAISapproachwiththesamethreefeatureselectionmethodsprovidessignificantlyhigherreturnsthantheDISapproach.
简介:ThearticleExploringthevalueofinstalledbase:PricinginformationgoodsundervaluedepreciationandconsumersociallearningpublishedinVolume22,Issue3,pages362-382byYifanDouandTianliangLiu,hasbeenretractedbyagreementbetweentheauthorsandthejournalEditor-in-Chief.Thecorecontributionsinthisarticlehavebeenusedwithoutsignificantacknowledgementandapprovalbytheprojectresearchteam.