在中國股票市場中的價(jià)值因素

ValueFactorinChineseStockMarketAbstractOverthepasttwentyyearsorso,theHMLfactorbuiltbyFamaandFrenchgraduallybecameineffectiveintheUSstockmarketbecauseofthepopularityoffactorinvestment.ThisarticlehasmadethecorrespondingtestinChina,andattemptstoexplainthisphenomenon.BasedontheAsharesoflistedcompaniesbetween2013to2018asthesample,weverifytheeffectivenessofthethree-factormodelofinChinesestockmarketthroughaseriesofdataprocessingandconversion.Specifically,inthispaper,theportfolioperformanceconstructedbythevaluefactorssuchasbook-to-marketratioandearnings-to-priceratioarecomparedandtheFama-Macbethregressionmodelisusedfordataprocessingandregressionanalysistoverifywhethertheyhavesignificanteffectsonthestockmarketyield.Wefindthatcomparedwithbook-to-marketratio,earnings-to-priceratiohashigherpredictingpowerandcangenerateamorestablereturn.Additionally,twoconjecturesarediscussedtoexplainthisphenomenonandprovideinsightsforfuturestudy.Keywords:valuefactor,three-factormodel,book-to-marketratio,earnings-to-priceratio,Fama-Macbethmodel

IntroductionFamaandFrench(1993)hadaprofoundinfluenceinacademiaandindustry,makingtheHMLfactorconstructedbybook-to-marketratio(B/M)knowntoinvestors.Book-to-marketratio(B/M)isusuallyusedtoestimatethecheapnessofstockprices.CompanieswithlowB/Mratioaregenerally"growth"companieswithexpensivestockprices,whilecompanieswithhighBMare"value"companieswithcheapstockprices.AccordingtoKothari&Shanken(1997),theexpectedfuturereturnsofstocksofcompanieswithahighbook-to-market(B/M)ratiowillincrease,aphenomenonknownasthebook-to-marketratioeffect.However,astimegoeson,moreandmorenewfactorsareappliedtostockmarkettrading,whichmakesmanyinvestorsstarttodoubtwhetherasimpleBMindicatorcanbuildanexcellentvaluefactor.Asearlyas80yearsago,GrahamandDoddwarnedinvestorsthatBookValueshouldnotbeusedtomeasuretheintrinsicvalueofacompany.Butwiththepopularityoffactorinvestment,theactivestrategyofstockselectionbyBMandthestyleindexconstructedbyBMarestillemergingoneafteranother,andBMhasalreadybecomeextremelycrowded.ThefigurebelowshowsthetrendoftheHMLfactorbuiltwithBM.Itisnothardtoseethatsincethemid-2000s,HMLhasalmostcompletelyfailed,leavingonewonderingwhethervalueinvestingisdead.FigureSEQFigure\*ARABIC1Thelong-termtrendofstrongvalueperformancehasbrokenthiscycleSource:KennethFrench,GoldmanSachs,GlobalInvestmentResearchUnderthiscircumstance,academiaandindustryanalystsbegantore-examinethelogicbehindBManddevelopanewindextoassistfutureinvestments.E/P(earnings-to-price)ratiowasfirstintroducedanddiscussedbyNicholson(1960),whoproposedinhisarticlethatstockswithahighE/Pratiocanbringhigherearnings.Ou&Penman(1989)furtherdiscussedthepossibilityandexplanationoffutureinvestmentreturnforecastsbasedonE/Pratio.TheyaffirmedthepredictivepoweroftheE/PratioandfoundthattheE/Ppredictionofearningswasnegativelycorrelatedwiththeearningspredictionofaccountingdata.Atpresent,thereformoftheshareholderstructureinChinesestockmarkethasbeenbasicallycompleted.Inaddition,themarketsystemandtransactionstructurearebecomingmoreandmoreperfect,andthestockmarketpresentssomenewcharacteristics,soitisverynecessarytostudytheforecastingabilityofBMandEPratio.Thepointofthispaperistodiscusswhetherthebook-to-marketratioandearnings-to-priceratiocancorrectlyreflecttheinformationrelatedtothefuturedecisionunderthecurrentmarketconditionsofChinesestockmarket.Inthispaper,valuestocksandgrowthstocksinChineseAstockmarketaredividedbyB/MandE/Pindexes,andtherateofreturnbetweeneachportfolioiscompared,soastoverifytheapplicabilityofvalueinvestmenttheoryinChineseAstockmarket.Inaddition,thispapertriestoexplainfromanotherperspectivefortheapplicabilityofB/MandE/Pratio.Althoughitisdifficulttostronglyjustify,wehopethiscanprovideinsightsandpossibilitiesforfuturestudyandresearch.Therestofthispaperisorganizedasfollows.Thesecondpartisanoverviewofrelevanttheoriesandliteraturereview,introducingtheevolutionprocessofcapitalassetpricingmodel,three-factormodelandotherrelevanttheoreticalmodelsabouttheadvantagesanddisadvantagesofE/PandB/Mratiorespectively,aswellasresearchachievementsinrelatedfieldsathomeandabroad.Thethirdpartisthesampleselectionandresearchmethodofthispaper,whichintroducestheprocessofdataselectionanddataprocessingindetail.Thefourthpartmainlycarriesoutempiricalanalysisandobtainstheresults.TheempiricalanalysisincludesportfolioperformancecomparisonanddiscussionwiththemethodofPearsoncorrelationtestandGrangercausalitytesttomakesareasonableinterpretationoftheresults.Finally,sectionfiveconcludesthepaperandprovidessomeinsights.LiteratureReviewAsanimportantvaluationindexofthestockmarket,B/MratioandE/Pratiohavebeenpaidattentiontoandstudiedbytheacademicandinvestmentcirclesfordecades.Inrecentyears,withtherapiddevelopmentofChinesecapitalmarket,especiallythesecondarystockmarket,theresearchontheeffectivenessofinvestmentstrategyinChineseacademiaandtheinvestmentcommunityhasbecomeincreasinglyrich,andthediscussionontheforecastingabilityofB/MratioandE/Pratioindexhasalsobeenincreasing.Inordertosupporttheresearchofthispaper,Ihaveconsultedalargenumberofrelevantliteratureathomeandabroad,andsummarizedthemainresearchideasandbeneficialresultsasfollows.Byusingstatisticalmodelbetweenthe1860sand1890smarketdata,FamaandFrench(1992)drewthefollowingconclusion:nomatterinasinglefactorormultiplefactoranalysis,thereisasignificantpositiverelationshipbetweentheprice-to-bookratioandtheannualnetprofitoflistedcompanies,andtherelationshipismoresignificantthanthatbetweencompanysizeandprofit.Withthisfinding,theybelievethatthebook-to-marketratioeffectiswidespread.Throughfurtherresearch,FamaandFrench(1997)alsotrackedthestockmarketsof13majordevelopedcountriesfrom1975to1995,andfoundthatthereturnrateofvaluestocks(morematurestockportfolioswithhighbook-to-marketratio)in12ofthemexceededthatofgrowthstocks(portfolioswithhighgrowthrateandlowbook-to-marketratio).BasedonFeltham-OhlsonModel,Penman(1996)analyzedindetailthesimilaritiesanddifferencesofthetwoindicators,aswellastheinformationaboutthefuturereturnonequity(ROE)containedinthem.Basedontheoreticalanalysis,hefoundthattheP/Eratio(theinverseofE/P)wasdeterminedbythecurrentandfuturereturnonequity,andisalinearfunctionofthefuturegrowthrateofreturnonassets.TheB/Mratioisdeterminedbythefuturereturnonequity.Throughfurtherempiricalanalysis,theaboveinferenceisverified.HeconcludedthatundertherelevantaccountingstandardsintheUnitedStates,theB/Mratiocanbetterpredictthefutureearningsstatusandreturnonequityofenterprises.WhilestudyingB/MandE/P,scholarshaveintensifiedtheircomparativeanalysisofthetwo,hopingtofindthebestindicatortoguideinvestment.Inhisbookportfolioselection:effectivediversificationofinvestments,HarryMarkowitzconcludesthatcombinedwithaccountingattributes,themostreliableindicatorinthefinancialdataoflistedcompaniesisshareholders'equity,whichcorrespondstothenetassetvalueofthecompany.SoinvestorsshouldpaymoreattentiontoB/Mmeasures,andthecurrenttoomuchfocusonearnings-to-priceisamistake.Fairfield(1994)analyzedtheeconomicimplicationsoftheindicatorsofB/MratioandE/Pratio.Intheempiricalanalysis,heusedthedataoftheUSstockmarketfrom1970to1984,anddividedtheresearchsamplesinto9groupsaccordingtotheE/PratioandB/Mratio,analyzedthecharacteristicsofthepricefluctuationsofeachgroupofstocks,andcalculatedtheirearningsinthenextfiveyears.Itsresearchconclusionmaintainsthatintheactualinvestmentanalysis,theB/MratioandtheE/Pratioshouldbeusedjointly.Asfarasasingleindicatorisconcerned,thepredictivepowerofB/MratioismoresignificantthanE/Pratioandismorerecommended.WhileinChina,theuniquestructureofequitycirculationinChinesestockmarketaffectstheratiosofthemarket.ItcanbeseenthatmuchforeignresearchoftheE/PratioandB/Mratiocannotbesimplyappliedtoourresearchandinvestmentpractice,andthecharacteristicsofChinesestockmarketshouldbefullyconsidered.GongrongChenandLiuRan(2011)believedthatthelackofvalueinvestmentconceptintheA-sharemarketleadstothelimitationoftheE/PandB/Mratioinreflectingthedevelopmentpotentialandprofitabilityofthecompanies.AccordingtoYuxiLiuandSongWang(2011),theB/Mratiohasagreaterimpactonstockinvestmentreturnsandismorestable.Throughempiricalanalysis,ChunleBaiandChunyuBai(2008)cametotheconclusionthatthebook-to-marketratiocanmoreeffectivelyreflecttheinvestmentvalueoflistedcompaniesofsmallandmedium-sizedenterprisesthantheearnings-to-priceratio.However,somescholarshavefoundthatE/Pratiohasmoreadvantagesinobtaininginvestmentincomeinempiricalstudies.LiBo(2004)provedthatthephenomenonof"HighE/PRatio"alsoexistsinChinesestockmarketbyusingrelevantdataofChineseAshares.ThearticleholdsthattheoverallE/PratioofChinesestockmarketislow,partlybecauseoftheseriouspolarizationoftheperformanceoflistedcompanies.Therefore,investmentshouldbecarefullyscreened,andchoosethehighearnings,highE/Pratiostockswithrealinvestmentvalueofthecompetitiveadvantages.Hebelievesthatinthefutureasthevalueinvestmentconceptisacceptedandrecognized,bluechipstocks(stocksfromcompaniesthatarewellestablished)willbevaluedandsoughtafterbythemarket.Thevaluationlevelislikelytorisefurther.Moreimportantly,Liuetal.(2018)pointedoutthatduetotheuniqueshellvalueproblemofChinesestockmarket,theFamaandFrenchthree-factormodelisnoteasytouseinChina,andtheyproposedaChineseversionofthethree-factormodel.TheChineseversionthree-factormodelcanwellexplainmostoftheanomaliesfoundintheChinesemarket,whichismuchstrongerthantheFamaandFrenchthree-factormodel.Liuetal.(2018)finallychoseEPtoconstructthevaluefactor,toreplaceBM.Tosumup,theadvantageofE/Pratioisthatitcandirectlyreflectthestockpriceandthecompany'searnings,butthedisadvantageisthattheprofithasacertainamountofroomformanipulationandhaslargefluctuations,thereforetheeffectivenessandstabilityofE/Pdataisinsufficient.Thevalueofbook-to-marketratioisrelativelystable,butthedatamaybebiasedduetothelagofthecompany'sassetvaluation.Theacademiccommunityandtheindustryhavenotreachedaunifiedconclusionontheimpactofvaluationindex,mainlyB/MratioandE/Pratio,oninvestmentreturn,whichisgreatlyinfluencedbytheresearchmethod,marketmaturityandsampletimespan.DataandMethodologyThepurposeofthispaperistomakeanempiricalanalysisoftheB/MratioandE/PratioonAshares.3.1SampleDescriptionThispaperselectsthetradingdataofChineseA-sharemarketfromJan1,2013toDec31,2018asthesample.BasedontheresearchobjectofthispaperistocomparetheverifiabilityofthetheoryofvalueinvestmenttoE/PandB/Mindex,thesampledataareexplainedandscreened.First,theselectedtimespanislongenoughtoavoidtheimpactofsuddeneventsandindustryfluctuationsontheresearchresults.Second,thestocksamplesthatweresuspendedatthebeginningoftheconstructionwereremoved.Third,becauseoftheabnormalfinancialindicatorsofSTsharesandST*shares,andthedifferentrangeofriseandfall,STandST*shareswillberemoved.ThedataarefromtheRiceQuantplatform.3.2SelectionofFactorsInordertofindoutwhichpartofB/Mrepresentstherisksfacedbythecompanyandisthereforelinkedtothefuturerateofreturnandwhichpartisjustnoiseinformation,scholarslaunchedavigorousreformofBM,andmanyofthelatestfindingswerepublishedinthetopjournalsofaccountingandfinance.Inordertomakeamoreobviouscomparison,thispaperliststhemaintransformationsofB/Mfactor,includingthreeoriginalformsandsixtransformationformsbasedonaccountingconsiderations.FortheoriginalB/M,considerthreeversions:1)theacademicversionofFamaandFrench(1993),thatis,theLastYearRatio(LYR)dataisupdatedonceayear,whichisrequiredbySEC,thuswithhighestreliability;2)theflexibleversionofAsnessandFrazzini(2013),namelyBMupdatedwiththeLastFile(LF)data,whichcancapturethelatestinformationmorequickly;3)theTrailingTwelveMonths(TTM)data,whichusesrollingfinancialdatafromthepastfourquartersforcalculationtoavoidthevagrancyoffinancialdatainonereport.InthecalculationofR&D,becauseR&Dhastwoprocessingmethods:1)capitalizationinthebalancesheet"developmentexpenditure"and"intangibleassets";2)expenseintotheincomestatementmanagementexpenses.Finally,whencalculatingR&D+SG&A,SG&AisequivalenttosalesexpensesandadministrativeexpensesinChina,andR&Dexpensesareincludedinadministrativeexpenses.ThefollowingTable1givesthespecificcalculationformulasofB/MandE/P.TableSEQTable\*ARABIC1DefinitionsoffactorsDenotationFactorFactor’sNameCalculationFormulaFac1lyrBM_LYRLastYearRatioBMequity_parent_company_lyr/market_capFac1ttmBM_TTMTrailingTwelveMonthsBMequity_parent_company_ttm/market_capFac1lfBM_LFLastFileBMequity_parent_company_lf/market_capFac4RetainedEarningsRetainedEarningtoMarketRatioRetained_earnings/market_capFac5ContributedCapitalContributedCapitaltoMarketRatioPaid_in_capital/market_capFac6NOANetOperatingAssettoNetOperatingAssetMarketValueNOA/NOA_marketNOA=total_equity+financial_liabilities-financial_asset_held_for_tradingNOA_market=market_cap+financial_liabilities-financial_asset_held_for_tradingFac7R&DBMafterR&Dadjustment(r_n_d+total_equity)/market_capFac8R&D+SG&ABMafterR&D+SG&Aadjustment(Selling_expense+ga_expense+total_equity)/market_capFac9IntrinsicValueIntrinsicValuetoMarketRatioIntrinsicvalue/market_capIntrinsicvalue=(total_equity+(du_return_on_equity-re)*total_equity/(1+re)+(du_return_on_equity-re)*(1+(1-dividend_yield)*du_return_on_equity)*total_equity/((1+re)*re))Fac10EPEarnings-to-priceRatioNet_profit_parent_company/market_capTheempiricalperiodinthispaperisfromJanuary2013toDecember2018.Intotal,indexesof3856stockshavebeencollectedatamonthbasis.Basedontheaboveconstructionmethod,thefollowingfiguregivesthecorrelationcoefficientsofallninecandidateB/MandonetypeofE/Pwithinthewholeempiricalinterval.Themonthlypriceandfinancialdataoflistedcompaniesarefromthericequantplatform.AccordingtoLiuetal.(2018),processEPwithnon-negativetreatment,whichmeanstheindexvalueofacompanywithpositiveEPisEPitself,whiletheindexvaluewithnegativeEPiszero.Ingeneral,sinceallofthemarevaluefactorsandB/McandidatesarebasedontheoriginalBMtransformation,themodifiedB/MhasahighcorrelationwiththeoriginalBM,especiallyRetainedEarnings,whiletheIntrinsicValueandE/PhasthelowestcorrelationwiththeoriginalBM.Itisworthmentioningthatbydefinition,BMindicatorsmodifiedwithIntrinsicValueandEPonlyuseannualreportdata.Theotherversionsusedquarterlyreportingdata,whichpartlyexplainsthedifferenceincorrelation.FigureSEQFigure\*ARABIC2CorrelationforcandidatefactorsTherefore,wefinallyselectedtheB/MfactorswiththelowestcorrelationwithE/P,butwithhighcorrelationwithotherB/Mfactors,whichisfac1lf,tocomparewithE/P.3.3DescriptiveStatistics1)DistributionofSampleE/PSincetheE/Pvalueistoosmall,wetaketheinverseofit(P/E)toseeifthesampleisrepresentative.FromTable2,wecanseethatamongthesampleP/Eratios,23.07%ofthecompanieshaveP/Eratiosbetween0and30,63.98%haveP/Eratiosbetween30and100,and12.95%haveP/Eratiosabove100.ItisgenerallybelievedthatP/Eratioisbetween0to50inmaturemarkets.AccordingtoChineseone-yeardeposit(2.5%),thereasonableP/Eratioisabout50inChinesestockmarket,butgiventherapidgrowthoftheeconomyinourcountry,theP/Eratioisgenerallyhigherthanthatofmaturemarkets,thatassumeareasonablerangeof0-100.Thenall87%ofthesampleP/Eratioisinareasonablescope,andthedescriptionoftheselectedsampleisrepresentative.TableSEQTable\*ARABIC2P/E(1/fac10)distribution2)DistributionofSampleB/MBasedonthesameconsiderations,B/Misalsotakenasaninverse(P/B).FromTable3,wecanseethatamongthesampleP/Bratios,nearly76%ofthecompanieshaveP/Eratiosbetween0and10,whichisconsistentwiththenormalconditioninA-sharemarket.Inaddition,duetotheinfluenceofsomemissingvalues,thetotalsamplenumberofE/PandB/Mwillbeslightlydeviated.However,sincewemainlyconsidertheoverallportfoliointhecomparison,itcanbeconsideredthatithaslittleimpactontheresultTableSEQTable\*ARABIC3P/B(1/fac1lf)distribution FromTable4,itisclearthatthestandarddeviationofwhetherP/EorP/Bisrelativelyhigh,69.14and179.13respectively.ItshowsthatthestockmarketP/EandP/Bratiohasalargedistributionrangeundertheinfluenceofextremevalues,soitisnecessarytodefineitsrationalityandstudyitseffectivenessininvestmentdecisionanalysis.TableSEQTable\*ARABIC4DescriptiveStatistics3.4DataProcessionThen,inordertoeliminatethevaluerangedifferencebetweendifferentfactors,twostepsareconducted,whichiswinsorizationandstandardization.Winsorizingwith3σmethodCalculatethestandarddeviationσaccordingtothesampleandadjusttheoriginalfactorvaluetotherangeof μ?3σ≤whereμisthemeanofthesample,σisthestandarddeviationofthesample.StandardizationWeuniformthecaliberoffactorvalueinordertocombinefactorsusingtheformula, FactorValue?whereμisthemeanofthesample,σisthestandarddeviationofthesample.NeutralizationTheperformanceoffactorsisaffectedbymultipleindicators.Factorsfromdifferentsectorswithdifferentsizehavedifferentperformance.Forinstance,valuationfactorsperformbetterinlarge-capstocks,butoperationfactorsperformbetterinsmall-capstocks.Indifferentsectors,thedifferenceinfundamentalswillleadtodifferentlevelsofvaluation.Valuestocksandgrowthstocksalsohavedifferentvaluationsystems.Underthiscircumstance,testingfactorsintheglobalmarketoftenleadstowrongconclusions.Tomakestockswithdifferentsizeandsectorcomparable,weneutralizethefactorsusingtheformula, FactorValuei=C+βwhereMViisthetotalmarketvalueofstocki,Secj,iMethodologyTheresearchmethodsadoptedinthispaperincludeempiricalanalysis,quantitativeanalysisandqualitativeanalysis.Theformertwoarethemainresearchmethods,andqualitativeanalysis,asanauxiliaryresearchmethod,realizesthecombinationoftheoryandempiricalanalysis.Thispapermainlyconductsempiricalanalysisbyconstructingaportfoliotocomparethesingle-factormodelandbymeansofFama-Macbethregression.Onthebasisofempiricalanalysis,combiningwiththecurrentsituationoftheoreticalresearchinrelatedfieldsandtheparticularityofChinesestockmarket,acomprehensiveanalysisiscarriedouttorealizethecombinationofquantitativeandqualitativeanalysis.AnalysisoftheinfluenceofBook-to-marketratioonstockreturnrate,theregressionequationis: Ri,t=awheretrepresentsthetransactionmonth,nisthenumberofstocksattimet,Ri,trepresentstherateofreturnattimetfortheit?stock,BMi,tisthebook-to-marketratioattimetforit?stock,βtisthecoefficientforBMratio.AfterknowingtherateofreturnforeachstockRi,tandtheirbook-to-marketratio,wecouldestimateatand β=1m stdβ=1 tβ=mReferringtotheanalysisoftheinfluenceofB/Mfactoronthestockmarketreturnrate,theregressionequationofE/Pfactoris: Ri,t=atAfterestimatingwhethereachfactorhasasignificantimpactontherateofreturn,let’slookattheimpactofthesesinglefactorsthatappeartohaveastrongimpactonthemarketrateofreturn.Regressionequationcanbeused: Ri,ttogettheβ1t,ResultsandDiscussionThewholestockaredividedintothreedeciles(30%,40%,30%)asFamaandFrench(1993).AfterdivingintothreedecilesaccordingtotheirE/P,B/Mrespectively,wecanfindfromFigure3thatthelargestrangeofchangeisinthetop30%ofthetwofactors,andthevalueofthemiddle40%isclosertothelower30%.ThetrendofE/PisroughlythesameasthatofB/M,whichdeclinesfirstandthenrises,buttheoverallvolatilityislargerthanthatofB/M,whichisalsoaffectedbyfrequentchangesinearningsdata.FigureSEQFigure\*ARABIC3RawfactorDistributionAndaccordingtoLiuetal.(2018),inChinesestockmarket,theminimumof30%ofthelistedcompanieswillbeseriouslyaffectedbytheshellvalueofpollution,causingtheassetpricingmodelthatcannotcorrectlyreflectthestockcross-sectionexpectedreturndifference.InordertobetterstudytheA-sharemarketpricingmechanism,thispaperalsoabandonsthevalueoftheminimumof30%ofalllistedcompanies.ThispapercomparesthetwofactorsmainlybycomparingtheirrateofreturnbetweeneachportfolioandtheirabnormalreturnwhenregressingwithFama-Frenchmodel.4.1PortfolioPerformanceForasinglefactor,itisnecessarytotestitsabilitytodistinguishbetweenstocks.ThestocksamplesweredividedfromhightolowaccordingtotheratioofB/MandE/Prespectively.Thestockportfolioswithhighbook-to-marketratio(B/M)inthefirst30%wereregardedasvaluestockportfolios,andthestockportfolioswithlowbook-to-marketratio(B/M)inthelast30%wereregardedasgrowthstockportfolios.Thestockreturnrateiscalculatedbythepriceattheendofeachmonthandthebeginningpriceatthebeginningofeachmonth.Thecalculationformulaisasfollows: Rx,nwherenreferstothetransactionmonth,xtothestocksymbol,Rtotherateofreturnofstockxheldatthebeginningofnt?month,andPreferstothepriceofstockxinnWithregardtotheportfoliogroupreturn,itmeanstoobservethetrendoftheaccumulatednetvalueofthe3groupsatthesametime,andjudgewhetherthenetvaluecurveofthe5groupsincreasesordecreasesinsequenceaccordingtothegroups.FigureSEQFigure\*ARABIC4FactorgroupnetreturnAccordingtoFigure4,whichshowstheaccumulativereturnofeachportfoliowhenadjustingeverysinglemonthfrom2013to2018,thegroupreturnderivedfromE/PrankingsisrelativelymoremonotonouscomparedwiththeonefromB/Mratio.Itisclearthatthemedian40%groupperformbetterthanbothhigh30%groupandlow30%groupwhenadjustedwithB/Mratio.However,itisobviousfromtheFigure4thatthehighE/Pratioexistsinthesampleperiod.ThediversifyingabilityofE/Pratioishigh,causingahugegapbetweenthehigh30%groupandlow30%group.Thegroupyieldcurveincreasesfromthefirstgrouptothelastgroup.Therefore,ahighandstablelong-shortspreadreturncanbeanticipatedwithportfoliosrankedbyE/P,whileportfolioscausedbyB/Mratioishardtoobtainahighrateofreturn.Toverifyouranticipation,thelong-shortportfolioisconstructedbybuyingthestocksinthetop30%(group1)rankedbyfactors,sellingthestocksinthetop30%(group3)rankedbyfactors,andadjustingthepositionaccordingtotheupdatedstockpooleverymonth.Theaccumulativelong-shortreturn,andannualizedrateofreturninthesampleperiodarecalculated,andthemonotonicityandstabilityarejudgedaccordingtothelong-shortyieldcurve.Observethetrendofyieldcurveinthesampleperiod.Iftheyieldcurveofthelong-shortportfoliocanincreaseordecreasesteadilyforalongtime,thenthisfactorisaneffectivefactor.FigureSEQFigure\*ARABIC5Long-shortreturnof3deciles,with30%eliminationTheyieldcurveinFigure5confirmsourexpectations.TheyieldcurveofE/Pratio,althoughdropsinthemiddleof2014,showsamorestabletrendandreachedover2.0attheendof2018,whiletheyieldcurveofB/Monlyreachedaround1.2.TableSEQTable\*ARABIC5Back-testingResultTable5summarizestheportfolioperformanceofBook-to-marketratioandEarningtoPriceratioofallsamplestocks.BenchmarkreferstothemonthlyrateofreturnofShanghaiCompositeIndexandH-LmeansthelongshortrateofreturnwithregardtoB/MratioandE/Pratiorespectively.Ascanbeseenfromthetable,althoughneitherE/PnorB/Mcanbringhighreturnstoinvestorsinthesampleperiod,incomparison,E/Pstillhasahigherannualreturnandsharperatiointhelong-termshortstrategy.4.2Fama-MacbethRegressionAccordingtoFama-Mecbeth,undercertainassumptions,Tstatisticscanbeusedtotestwhetherfactorshaveasignificantimpactonreturns.Table6showstheregressioncoefficientsofstockreturnsforthesefactors.TableSEQTable\*ARABIC6Fama-MacbethregressionresultNote:eachrowisrespectivelythemeanvalueoftheregressioncoefficientβandthet-statisticvaluetβThispaperfirsthascarriedoutthedataprocessingintheprocessofregressionbycalculatingthelogarithmofeachfactor,whichmainlyrefertotheLongzhenFan(2002)inChinesestockmarketthreefactormodelfordataprocessingprocedure.Thereasonforthiskindofdataprocessing,mainlyconsideringtheriskthatthevalueofeachfactordifferencesbetweeneachstockistoobig,andwillhavelargerfluctuationinthetimeseriesregression.Afterprocessingthe