mean reversion in prices returnsevidence and i

NBERWORKINGPAPERSERIESMEANREVERSIONINSTOCKPRICES:EVIDENCEANDIMPLICATIONSJamesM.PoterbaLawrenceH.SummersWorkingPaperNo.2343NATIONALBUREAUOFECONOMICRESEARCH1050MassachusettsAvenueCambridge,MA02138August1987WearegratefultoBarryPerlstein,ChangyongRhee,JeffZweibelandespeciallyDavidCutlerforexcellentresearchassistance,toBenBernanke,JohnCampbell,RobertEngle,EugeneFama,PeteKyle,GregMankiw,JulioRotemberg,WilliamSchwert,KennethSingleton,andMarkWatsonforhelpfulcomments,andtoJamesDarcel,MatthewShapiro,andIanTonksfordataassistance.ThisresearchwassupportedbytheNationalScienceFoundationandwasconductedwhilethefirstauthorwasaBatterymarchFellow.TheresearchreportedhereispartoftheNBER'sresearchprograminFinancialMarketsandMonetaryEconomics'.AnyopinionsexpressedarethoseoftheauthorsandnotthoseoftheNationalBureauofEconomicResearch.qdA­qd句La品UVEaEWnvaunrnuJn哼F-1·noiknHJFE'··­nuMH+LPbnnuFEnEnDUM川AHMeanReversioninStockPrices:EvidenceandImplicationsABSTRACTThispaperanalyzesthestatisticalevidencebear‘ingonwhether‘transitorycomponentsaccountforalar‘gefractionofthevarianceincommonstockreturns.Thefirstparttreatsmethodologicalissuesinvolvedintestingfortransitoryreturncomponents.Itdemonstratesthatvarianceratiosareamongthemostpowerfultestsfordetectingmeanreversioninstockprices,butthattheyhave1ittlepoweragainsttheprincipalinterestingalternativestotherandomwalkhypothesis.ThesecondpartappliesvarianceratioteststomarketreturnsfortheUnitedStatesoverthe1871-1986periodandforseventeenothercountriesoverthe1957-1985period,as树ellastoreturnsonindividua1firmsoverthe1926-1985period.Wefindconsistentevidencethatstockreturnsarepositive1yserial1ycorrelatedover‘shorthorizons,andnegativelyautocorrelatedover10nghorizons.Thepointestimatessuggestthatthetransitorycomponentsinstockpriceshaveastandarddeviationofbetween15and25percentandaccountformorethanhalfofthevarianceinmonthlyreturns.Thelastpartofthepaperdiscussestwopossibleexplanationsformeanrever‘sion:timevaryingrequiredreturns,andslowly-decayingpricefadsthatcausestockpricestodeviatefromfundamentalvaluesforperiodsofsevera1years.树econcludethatexplainingobservedtransitorycomponentsinstockpricesonthebasisofmove-mentsinrequiredreturnsduetoriskfactorsislike1ytobedifficult.JamesM.PoterbaDepartmentofEconomicsMassachusettsInstituteofTechnologyCambridge,MA02139(617)253-6673LawrenceH.SummersOepartmentofEconomicsHarvardUniversityCambridge,MA02138(617)495-2447Thispaperexaminestheevidenceontheextenttowhichstockpricesexhibitmean-revertingbehavior.Thequestionofwhether‘stockpricescontaintransitorycomponentsisimportantforfinancialpracticeandtheory.Forexamp1e,considerthequestionofinvestmentstrategy.Ifstockpricemovementscontain1argetransitorycomponentsthenfor10ng-horizoninvestorsthestockmarketmaybemuch1essriskythanitappearswhenthevarianceofsingle-periodreturnsisextrapolatedusingtherandomwa1kmodel.Marketfolk10rehas10ngsuggestedthatthosewhotakethe10ngviewshouldinvestmoreinequitythanthosewithashorthorizon.Althoughharshlyrejectedbymosteconomists,thisviewiscorrectifpricesexhibitmean-revert才ngbehavior.1Furthermore,thepresenceoftransitorypricecomponentssuggeststhedesirabilityofinvestmentstrategiesinvo1vingthepurchaseofsecuritiesthathaverecentlydec1inedinvalue.Importanttransitorycomponentsinstockpricescoulda1soimpartsome10gictoeconomicagents'reluctancetotiedecisionstocurrentmarketvalues.Corporatemanagersoftenassertthattheircommonstockismisvaluedandclaimthatitwouldbeunwisetobaseinvestmentdecisionsonitscurrentmarketprice.Acommonprocedureamonguniversitiesandother‘institutionsthatrelyonendowmentincomeistospendonthebasisofaweightedaverageofpastendowmentva1ues.HarvardUniversityspendsoutofendowmentaccordingtoapresettrendlineregardlessofthemarket'sva1ue.Suchrulesarehardtounder‘standifstockpricesfollowarandomwa1k,butmakesenseifpricescon­tainimportanttransitorycomponents.Asamatteroftheory,evaluatingtheextentofmean-rever‘sioninstockpricesiscrucialforassessingclaimssuchasKeynes'(1936)assertionthat-2-allsortsofconsiderationsenter‘intomarketvaluationwhichareinnowayrelevanttotheprospectiveyield(p.152).lfdivergencesbetweenmarketandfundamentalvaluesexist,butbeyondsomelimitareeliminatedbyspeculativeforces,thenstockpriceswillexhibitmeanreversion.Returnsmustbenegati­velyseriallycorrelatedatsomefrequencyiferroneousmarketmovesareeven­tuallycorrected.2As问erton(1987)notes,reasoningofthistypehasbeenusedtodrawconclusionsaboutmarketvaluationsfromfailurestorejecttheabsenceofnegativeserialcorrelationinreturns.Conversely,thepresenceofnegativeautocorrelationmaysignaldeparturesfromfundamentalvalues,thoughitcouldalsoarisefromriskfactorsthatvarythroughtime.Thepaperisorganizedasfollows.Section1beginsbyevaluatingalter­nativestatisticalproceduresfortestingfortransitorycomponentsinstockprices.训efindthatvarianceratiotestsofthetypeusedbyFamaandFrench(1986a)andLoandMacKinlay(1987)comeclosetobeingthemostpowerfultestsofthenullhypothesisofmarketefficiency旦旦旦constantrequiredreturnsagainstplausiblealternativehypothesessuchasthefadsmodelsuggestedbyShiller(1984)andSummers(1986).Never‘the