Transcription of Lecture Notes - Hou, Mo, Xue, and Zhang (2018, Review of ...
1 LectureNotesHou,Mo,Xue,andZhang (2018, Rev iewofFinance,WhichFactors?)LuZhang11 OhioStateandNBERBUSFIN8250 Autumn2018,OhioStateIntro ductionThemeManyrecentlyprop osed,seeminglydi erentfactormo delsarecloselyrelatedtotheq-factormo delIntro ductionOverviewofresultsInspanningregres sions,theq-factormo dellargelysubsumestheFama-French5-and6-f actormo delsTheStambaugh-Yuanfactorssensitivetot heirconstruction,oncereplicatedviathetra ditionalapproach,areclosetotheq-factors, ,oncereplicatedviathetraditionalapproach ,areclosetotheq-factors, ositiverelationb etweentheexp ectedinvestmentandtheexp ectedreturnOutline1 ThePlayingField2 SpanningTests3 ValuationTheory.
2 AssetPricingImplicationsOutline1 ThePlayingField2 SpanningTests3 ValuationTheory:AssetPricingImplications ThePlayingField8comp etingfactormo delsTheq-factormo del,theq5mo delTheFama-French5-factormo del,the6-factormo del,thealternative6-factormo delwithRMWcTheStambaugh-Yuan4-factormo delTheBarillas-Shanken6-factormo del,includingMKT,SMB,RI~A,RRo e,theAsness-FrazzinimonthlyformedHML,UMD TheDaniel-Hirshleifer-Sun3-factormo delThePlayingFieldTheq-factormo del,Hou,Xue,andZhang(2015)E[Ri Rf]= iMKTE[MKT]+ iMeE[RMe]+ iI~AE[RI~A]+ iRo eE[RRo e]MKT,RMe,RI~A,andRRo earethemarket,size,investment,andpro tability(returnonequity,Ro e)
3 Factors,resp ectively iMKT, iMe, iI~A,and iRo earefactorloadingsThePlayingFieldConstru ctionoftheq-factorsRME,RI~A,andRRo efromindep endent,triple2 3 3sortsonsize,investment-to-assets,andRo eVariablede nitions:Size:Sto ckpricetimessharesoutstandingfromCRSPI nvestment-to-assets,I/A:Annualchangesint otalassets(itemAT)dividedbylaggedtotalas setsRo e:Incomeb eforeextraordinaryitems(itemIBQ)dividedb yone-quarter-laggedb o okequityThePlayingFieldConstructionofthe q-factorsNYSE breakp oints:50-50forsize,30-40-30forI/A,and30- 40-30forRo e;value-weightedreturnsTiming:Annualsort inJuneonthemarketequityattheJuneendAnnua lsortinJuneofyeartonI/Aforthe scalyearendingincalendaryeart 1 Monthlysortattheb eginningofeachmonthonRo ewiththemostrecentlyannouncedquarterlyea rningsResultsrobusttoallmonthlysortsonsi ze,I/A,andRo eThePlayingFieldExtendingtheq-factorsbac kwardtoJanuary1967 Hou,Xue,andZhang(2015)
4 StartinJanuary1972,restrictedbyearningsa nnouncementdatesandquarterlyb o okequitydataPriortoJanuary1972,usethemos trecentearningsfromthe scalquarterendingatleast4monthspriortoth ep ortfolioformationMaximizethecoverageofqu arterlyb o okequityThePlayingFieldBackwardextending theq-factors,maximizethecoverageofquarte rlyb o okequityUsequarterlyb o okequitywheneveravailableSupplementtheco veragefor scalquarter4withb o okequityfromCompustatannual lesIfavailable,backwardimputeb eginning-of-quarterb o okequityasend-of-quarterb o okequityminusquarterlyearningsplusquarte rlydividendsFinally,forwardimputeBEQt=BE Qt j+IBQt j+1,t DVQt j+1,t,inwhichBEQt jisthelatestavailablequarterlyb o okequityasofquartert,IBQt j+1,tandDVQt j+1,tthesumofquarterlyearningsandthesumo fquarterlydividendsfromquartert j+1toquartert,resp ectively,and1 j 4 ThePlayingFieldHou,Mo,Xue,andZhang(2018, q5)Augmenttheq-factormo delwiththeexp ectedgrowthfactortoformtheq5mo del.
5 E[Ri Rf]= iMKTE[MKT]+ iMeE[RMe]+ iI~AE[RI~A]+ iRo eE[RRo e]+ iEgE[REg]Stress-testsfromalargesetof158a nomaliesshowthattheq5mo delimprovesontheq-factormo delsubstantiallyThePlayingFieldConstruct ingtheexp ectedgrowthfactorForecastd I/A, -yearaheadinvestment-to-assetschanges,vi amonthlycross-sectionalregressionsMotiva tingpredictorsbasedonaprioriconceptualar guments(internalfundsavailableforinvestm ents,accountingconservatism,short-termdy namicsofinvestmentgrowth):Tobin'sq:Erick sonandWhited(2000)Cash ows:Fazzari,Hubbard,andPetersen(1988)Cha ngeinreturnonequity:Liu,Whited,andZhang( 2009)REgfrommonthly,indep endent2 3sortsonsizeandEt[d1I~A]ThePlayingFieldT heFama-French(2015,2018)5-and6-factormo delsTheFama-French5-factormo del:E[Ri Rf]=biE[MKT]+siE[SMB]+hiE[HML]+riE[RMW]+ ciE[CMA]MKT,SMB,HML,RMW ,andCMAarethemarket,size,value,pro tability,andinvestmentfactors,resp ectivelybi,si,hi,ri,andciarefactorloadin gsFamaandFrench(2018)addUMDtoformthe6-fa ctormo del,alsoprop oseanalternative6-factormo delwithRMWcThePlayingFieldHistoricaltime line.
6 Theq-factormo delpredatestheFama-French5-factormo delby3 6yearsNeo classicalfactorsJuly2007 Anequilibriumthree-factormo delJanuary2009 Pro duction-basedfactorsApril2009Ab etterthree-factormo delJune2009thatexplainsmoreanomaliesAnal ternativethree-factormo delApril2010,April2011 Digestinganomalies:AninvestmentapproachO ctob er2012,August2014 FamaandFrench(2013):Afour-factormo delJune2013forthesize,value,andpro tabilitypatternsinsto ckreturnsFamaandFrench(2014):Novemb er2013,A ve-factorassetpricingmo delSeptemb er2014 ThePlayingFieldHistoricaltimeline,contin uedAcomparisonofnewfactormo delsOctob er2014 ReplicatinganomaliesMay2017,July2018 MotivatingfactorsDecemb er2017q5 June2018,July2018 Whichfactors?
7 July2018,Septemb er2018 Dissectinganomalies2015,2016witha ve-factormo delWhichalpha?2015,2017 Mispricingfactors2015,2017 Comparingassetpricingmo dels2015,2018 Cho osingfactors2017,2018 Short-andlong-horizon2017,2018b ehavioralfactorsThePlayingFieldTheStamba ugh-Yuan mispricing factorsStartwithtwoclustersofanomalies:M GMT:netsto ckissues,comp ositeissues,accruals,netop eratingassets,assetgrowth,andchangeingro ssPPEandinventoryscaledbylaggedb o okassetsPERF:failureprobability,O-score, momentum,grosspro tability,andreturnonassetsFormcomp ositescoresbyequal-weightingasto ck'sp ercentilesineachcluster(realignedtoyield averageL Hreturns>0)
8 FormtheMGMTandPERF factorsfromindep endent2 3sortsbyinteractingsizewitheachcomp ositescoreThePlayingFieldStambaughandYua ndeviatefromthetraditionalconstructionin imp ortantwaysTheNYSE-Amex-NASDAQ20 80breakp oints,asopp osedtotheNYSE30 70breakp ointsThesizefactorcontainssto cksonlyinthemiddlep ortfoliosofthedoublesorts,asopp osedtofromallp ortfoliosUsetheiroriginalfactors,aswella sreplicatedfactorsviathetraditionalconst ructionResultsaresensitivetotheconstruct ionmetho dThePlayingFieldTheDaniel-Hirshleifer-Su nmo del:MKT+FIN+PEADFIN basedon1-yearnetshareissuanceand5-yearco mp ositeissuance;PEADon4-daycumulativeabnor malreturnaroundthemostrecentquarterlyear ningsannouncement,AbrFactorconstructiona lsodeviatesfromthemorecommonapproach:NYS E20 80,asopp osedtoNYSE30 70,breakp ointsAbronly,asopp osedtoAbr,Sue,andRep erChan,Jegadeesh,andLakonishok(1996)More adho c,involvedsortsonFINU serepro ducedandreplicatedfactors(NYSE30 70breakp ointsonthecomp ositescoresofFINfromcombiningnetshareand comp ositeissuancesandofPEAD fromcombiningAbr,Sue,andRebyequal-weight ingasto ck'sp ercentilerankings)Outline1 ThePlayingField2 SpanningTests3 ValuationTheory.
9 AssetPricingImplicationsSpanningTestsEmp iricaldesignRelymostlyonspanningtestsasa ninformativeandconcisewaytocomparefactor mo delsonempiricalgroundsBarillasandShanken (2017,2018):Fortradedfactors,theextentto whicheachmo delisabletopricethefactorsintheothermo delisallthatmattersforcomparison;testing assetsirrelevantIncomplementarywork,Hou, Mo,Xue,andZhang(2018)stress-testfactormo delswithalargesetof158signi cantanomalies,withresultsconsistentwitho urspanningtestsSpanningTestsTheFama-Fren ch5-and6-factormo delscannotexplaintheqandq5factors,1/1967 12/2016R MKT SMB HML RMW CMA UMD ~ delscannotexplaintheqandq5factors,1/1967 12/2016R MKT SMB HML RMW CMA UMD RMWcRRo delscannotexplaintheqandq5factors,theGib b ons-Ross-Shankentest,1/1967 12/2016 I~A, Ro e=0 I~A, Ro e, Eg= delslargelysubsumetheFama-French5-and6-f actormo dels.
10 1/1967 12/2016R MKT Me I~A Ro e delslargelysubsumetheFama-French5-and6-f actormo dels,1/1967 12/2016R MKT Me I~A Ro e delslargelysubsumetheFama-French5-and6-f actormo dels,theGRStest,1/1967 12/2016 HML, CMA, RMW, UMD HML, CMA, RMW=0 RMW, UMD=0 RMWc, UMD= ortedbyBarillasandShanken(2017,2018)HXZ: A Comparison of New Factor Models Discussion ASU Sonoran Winter Finance Conference Jay Shanken Emory University February 20, 2015 SpanningTestsResultsreplicatedbutnotrep ortedbyBarillasandShanken(2017,2018)6 Empirical Results: Barillas-Shanken (2015b) We develop a Bayesian test for comparing models q-model prob = 97%, FF5 3% Asness and Frazzini (2013) argue for a better value factor than HML FF (1993) update portfolios once a year using prices lagged 6 months.
