您的当前位置：首页正文

Accrual reliability, earnings persistence and stock prices

来源：好走旅游网

JournalofAccountingandEconomics39(2005)437–485

www.elsevier.com/locate/jae

Accrualreliability,earningspersistence

andstockprices$ScottA.Richardsona,RichardG.Sloanb,Ã,

˙remTunaaMarkT.Solimanc,I

abTheWhartonSchool,UniversityofPennsylvania,Philadelphia,PA19104,USA

RossSchoolofBusiness,UniversityofMichigan,AnnArbor,MI48109,USAcGraduateSchoolofBusiness,StanfordUniversity,Stanford,CA94305,USA

Availableonline16June2005

Abstract

ThispaperextendstheworkofSloan(1996.TheAccountingReview71,289)bylinkingaccrualreliabilitytoearningspersistence.Weconstructamodelshowingthatlessreliableaccrualsleadtolowerearningspersistence.Wethendevelopacomprehensivebalancesheetcategorizationofaccrualsandrateeachcategoryaccordingtothereliabilityoftheunderlyingaccruals.Empiricaltestsgenerallyconﬁrmthatlessreliableaccrualsleadtolowerearningspersistenceandthatinvestorsdonotfullyanticipatethelowerearningspersistence,leadingto

WewouldliketothankMaryBarth,BillBeaver,PatriciaDechow,PaulHealy,RonKahn,S.P.Kothari(theeditor),KatherineSchipper,StephenTaylor,RossWatts,therefereeandworkshopparticipantsattheAAANZ2002AnnualMeetings,BarclaysGlobalInvestors,ChicagoQuantitativeAlliance2003Conference,UniversityofColorado,UniversityofIllinois,InstituteforQuantitativeResearchinFinance2003Conference,MassachusettsInstituteofTechnology,UniversityofMelbourne,PrudentialSecurities2002QuantitativeConference,StanfordUniversity2001AccountingSummerCamp,UniversityofTechnologySydneyandUniversityofUtahfortheircommentsandsuggestions.Earlierversionsofthispaperweretitled‘‘InformationinAccrualsabouttheQualityofEarnings’’.ÃCorrespondingauthor.Tel.:+17347642325;fax:+17349360282.E-mailaddress:sloanr@umich.edu(R.G.Sloan).0165-4101/$-seefrontmatterr2005ElsevierB.V.Allrightsreserved.doi:10.1016/j.jacceco.2005.04.005

$438

S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485

signiﬁcantsecuritymispricing.Theseresultssuggestthattherearesigniﬁcantcostsassociatedwithincorporatinglessreliableaccrualinformationinﬁnancialstatements.r2005ElsevierB.V.Allrightsreserved.

JELclassiﬁcation:M4

Keywords:Accruals;Reliability;Earningspersistence;Capitalmarkets;Marketefﬁciency

1.Introduction

Thispaperinvestigatestherelationbetweenaccrualreliabilityandearningspersistence.ThepaperbuildsontheworkofSloan(1996),whoshowsthattheaccrualcomponentofearningsislesspersistentthanthecashﬂowcomponentofearningsandattributesthisdifferencetothegreatersubjectivityofaccruals.WedrawanaturallinkbetweenSloan’snotionofsubjectivityandthewell-knownaccountingconceptofreliability.Weformallymodeltheimplicationsofreliabilityforearningspersistence,withourmodelpredictingthatlessreliableaccrualsresultinlowerearningspersistence.Ourempiricaltestsemployacomprehensivecategoriza-tionofaccountingaccrualsinwhicheachaccrualcategoryisratedaccordingtoitsreliability.Consistentwiththepredictionsofourmodel,theempiricaltestsgenerallyconﬁrmthatlessreliableaccrualsleadtolowerearningspersistence.Wealsoshowthatstockpricesactasifinvestorsdonotanticipatethelowerpersistenceoflessreliableaccruals,leadingtosigniﬁcantsecuritymispricing.

Ourpapermakesthreecontributionstotheexistingliterature.First,wedirectlylinkreliabilitytoempiricallyobservablepropertiesofaccountingnumbers.Reliability,alongwithrelevance,isconsideredtobeoneofthetwoprimaryqualitiesthatmakeaccountinginformationusefulfordecision-making.1Whilealargebodyofresearchexaminesthevaluerelevanceofaccountingnumbers,thereisrelativelylittleresearchonreliability.Oneconsequenceoftheemphasisonrelevancehasbeencallsfortherecognitionofmorerelevantandlessreliableinformationinaccountingnumbers(e.g.,LevandSougiannis,1996).However,asrecentlyarticulatedbyWatts(2003),allowinglessveriﬁableandhencelessreliableestimatesintoaccountingnumberscanseriouslycompromisetheirusefulness.Ourresearchhighlightsthecrucialtrade-offbetweenrelevanceandreliability.Ouranalysissuggeststhattherecognitionoflessreliableaccrualestimatesintroducesmeasure-menterrorthatreducesearnings’persistenceandleadstosigniﬁcantsecuritymispricing.

Thesecondmajorcontributionofourworkistoprovideacomprehensivedeﬁnitionandcategorizationofaccruals.FollowingHealy(1985),alargebodyofresearchhasemployedanarrowdeﬁnitionofaccrualsthatfocusesonworkingcapitalaccruals.Non-currentaccruals,suchascapitalizedsoftwaredevelopment

Forexample,seeFig.1inStatementofFinancialAccountingConcepts2,‘‘QualitativeCharacteristicsofAccountingInformation’’(FinancialAccountingStandardsBoard,2002).

1S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485

439

costsandcapitalizedexpendituresonplantandequipment,areomittedfromHealy’sdeﬁnition.Ignoringsuchaccrualsresultsinnoisymeasuresofbothaccrualsandcashﬂows(becausecashﬂowsaretypicallycomputedasthedifferencebetweenearningsandaccruals).OurevidenceindicatesthatmanyoftheaccrualsthatareomittedfromHealy’sdeﬁnitionareoflowreliability,andweencouragesubsequentresearchtoincorporatesuchaccruals.

ThethirdmajorcontributionofourworkistocorroborateandextendSloan’sﬁndingsregardingmarketefﬁciencywithrespecttoinformationinaccruals.Sloanshowsthatinvestorsactasiftheydonotanticipatethelowerpersistenceoftheaccrualcomponentofearnings,resultinginsigniﬁcantsecuritymispricing.WecorroborateandextendSloan’sresultsintwoways.First,wedevelopamorecomprehensivedeﬁnitionofaccrualsthanSloanandshowthatitisassociatedwithevengreatermispricing.Second,weshowthatboththepersistenceofearningsandtheextentoftheassociatedmispricingaredirectlyrelatedtothereliabilityoftheunderlyingaccruals.

Whilethisstudyexaminestheimpactofaccrualreliabilityonearningspersistence,weemphasizethatthereareotherexplanationsforourresults.Inparticular,Fairﬁeldetal.(2003a)contendthatSloan’s(1996)ﬁndingsareaspecialcaseofamoregeneralgrowtheffectthatisattributabletodiminishingmarginalreturnstoinvestmentand/orconservativeaccounting.Anotherexplanationisthatextremeaccrualsmaybecausedbyextremechangesinsalesthathaveatransitoryeconomicimpactonearnings.Supportingouraccrualreliabilityexplanation,Xie(2001)andRichardsonetal.(2004)ﬁndthatthelowerpersistenceofaccrualspersistsevenaftercontrollingforsalesgrowth.Nevertheless,itispossiblethatotherexplanationscontributetoourresultsandadditionalresearchisrequiredtodiscriminatebetweencompetingexplanations.

Theremainderofthepaperisorganizedasfollows.Section2developsourresearchdesign.Section3describesourdata,Section4presentsourresultsandSection5concludes.

2.Researchdesign

OurresearchdesignbuildsonSloan(1996),whoarguesthatthekeydifferencebetweentheaccrualandcashﬂowcomponentsofearningsisthattheaccrualcomponentinvolvesagreaterdegreeofsubjectivity.Theaccrualcomponentofearningstypicallyincorporatesestimatesoffuturecashﬂows,deferralsofpastcashﬂows,allocationsandvaluations,allofwhichinvolvehighersubjectivitythansimplymeasuringperiodiccashﬂows.ThisleadsSloantoreasonthatwhentheaccrualcomponentofearningsisunusuallyhighorlow,earningswillbelesspersistent.

WebegininSection2.1bypresentingasimplemodelformalizingthelinkbetweenaccrualsubjectivityandearningspersistence.Section2.2alertsreaderstothekeylimitationsofthismodelandsuggestssomepossibleextensions.Section2.3reconsidersthestandarddeﬁnitionofaccrualsandproposesanewandmore

440

S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485

comprehensivedeﬁnition.inally,Section2.4providesourcategorizationofaccrualsandourassessmentoftherelativereliabilityofeachaccrualcategory.2.1.Asimplemodelofaccrualreliabilityandearningspersistence

Sloanarguesthatthekeyfactordrivingthedifferentpropertiesoftheaccrualandcashﬂowcomponentsofearningsisthegreatersubjectivityinvolvedintheestimationofaccruals.Accountantsmoretypicallyrefertothedegreeofsubjectivityinvolvedinanaccountingmeasurementintermsoftheveriﬁabilityandreliabilityofthemeasurement.Forexample,theFinancialAccountingStan-dardBoard’s(FASB)StatementsofFinancialAccountingConcepts(SFAC)deﬁnesveriﬁabilityas:

Theabilitythroughconsensusamongmeasurerstoensurethatinformationrepresentswhatitpurportstorepresentorthatthechosenmethodofmeasurementhasbeenusedwithouterrororbias.[SFAC2,GlossaryofTerms]Veriﬁability,alongwithrepresentationalfaithfulness,providesthebasisforthekeyaccountingqualityofreliability,deﬁnedbytheFASBas:

Thequalityofinformationthatassuresthatinformationisreasonablyfreefromerrorandbiasandfaithfullyrepresentswhatitpurportstorepresent.[SFAC2,GlossaryofTerms]

Maximizingtheusefulnessofaccrualaccountinginformationinvolvesatrade-offbetweenrelevanceandreliability(seeSFAC2,paragraph42).Alargebodyofaccountingliteraturehasevaluatedaccountinginformationonthebasisoftherelevancecriterion(seeHolthausenandWatts,2001).Incontrast,relativelylittleattentionhasbeengiventoevaluatingaccountinginformationonthebasisofthereliabilitycriterion.Giventheone-sidednatureofresearchinthisarea,itisperhapsnotsurprisingthatsomeresearchersandregulatorshavecalledformorevalue-relevantandlessreliableestimatestobeincorporatedinﬁnan-cialstatements(seeWatts,2003,forexamples).However,asarticulatedbyWatts(2003),researchersandregulatorswhoproposetherecognitionofrelativelyunreliableestimatesinﬁnancialreportsshouldconsiderthecostsgeneratedbytheirproposals.OnesuchcostthatfeaturesprominentlyintheFASB’sdeﬁni-tionsofreliabilityandveriﬁabilityisthegreaterpotentialformeasurementerrorinlessreliableandveriﬁableaccountingnumbers.Accordingly,wemodeltheeffectsofmeasurementerrorinaccountingaccrualsonthepersistenceofearnings.

Theprimaryroleofearningsistomeasuretheperiodicﬁnancialperformanceofanenterprise(seeSFAC1,paragraph43).WedenotetheunderlyingperiodicﬁnancialperformanceoftheenterpriseasEÃ.Itiswellestablishedthatcompetitiveforcesresultinthedissipationofeconomicrents,andthatthisisexpectedtoresultinthemeanreversionofﬁnancialperformance(e.g.,Palepuetal.,2000).Thus,evenintheabsenceofreliabilityissues,weexpectﬁnancialperformance(deﬂatedbyanappropriatemeasureofinvestedcapital)tofollowa

S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485

441

meanrevertingprocess:2Ã

EÃtþ1¼gEtþ󰀁tþ1,(1)

where0ogo1andgreatercompetitionimplieslowervaluesofg.

Toillustratethedifferencebetweenthetrueunderlyingﬁnancialperformance,EÃ,andreportedearnings,E,weusesomesimpleexamples.Webeginwiththespecialcaseinwhichthecashconsequencesofabusiness’stransactionsandeventsalloccurduringasingleperformancemeasurementinterval.Suchasituationwouldarisewhentheoperatingandinvestingcyclesofthebusinessareveryshort.Inthiscase,thenetcashreceiptsofthebusinessprovideacompletelyreliablerepresentationofperiodicperformance.DenotingthenetcashreceiptsasC,thisspecialcasethereforeresultsin

EÃ¼C.

(2)

Next,weintroducethemorerealisticsituationinwhichtheactualcashreceiptsanddisbursementsoccurindifferentperiodsfromtheunderlyingtransactionsandeventsthatgeneratethem.InthiscaseEÃdoesnotnecessarilyequalCandaccountingaccrualsareintroducedinordertoprovideamorerepresentativemeasureofperiodicperformancethanperiodicnetcashreceipts.Accrualaccountinginvolvestheaccrualanddeferralofpast,currentandanticipatedfuturecashreceiptsanddisbursements.However,earningsmeasuredusingaccrualaccountingarenotexpectedtoperfectlycoincidewithEÃfortworeasons.First,certainaccrualsanddeferralsrelatingtofutureeconomicbeneﬁtsthatcannotbemeasuredwithsufﬁcientreliabilityaredisallowedunderGAAP(e.g.,immediateexpensingofR&Dcosts).Thisresultsinanomittedvariableproblemthatisnotexploredinthispaper.3Second,mostaccrualsrequiredbyGAAPmeasurethefutureeconomicbeneﬁtsofcurrentperiodeventsandtransactionswitherror.Itisthissecondscenariothatservesasthefocusofourpaper.

Toillustratethissecondscenario,considertherecognitionofsalesrevenue.Ifcashsalesaremadefor$100(withnoreturnsallowedandnoaftersalesservice),thenitisstraightforwardtobook$100ofrevenueincurrentperiodearnings.Butifthesalesarecreditsalesfor$110,ofwhich$100ultimatelyendsupbeingcollectedascashinthenextperiod,uncertaintyexistsabouttheamounttorecordincurrentperiodearnings.Anaggressivemanagercouldbooksalesof$110,representinganerrorof+10incurrentperiodaccrualsandearnings.Conversely,aconservativemanager

Forexpositionalease,ourmodelrevertsEÃtowardameanofzero.Theanalysisiseasilyadaptedtonon-zeromeanbyredeﬁningEÃasthedeviationfromthelongrunmeanperformance.Ourempiricalanalysisallowsforanon-zeromeanthroughtheinclusionofaninterceptintheregression.3Theomissionofarelevantvariableresultsinbiasedcoefﬁcientestimatesontheincludedvariableswhentheomittedvariableiscorrelatedwiththeincludedvariables.However,wehavenoreasontobelievethatanyomittedaccrualswoulddifferentiallybiasthecashﬂowandincludedaccrualcomponentsofearnings.PenmanandZhang(2002)provideevidenceofpredictablechangesinearningsandstockreturnsassociatedwithomittedaccrualsforresearchanddevelopmentandadvertisingexpenditures.Inordertoverifytherobustnessofourresultswithrespecttotheseomittedaccruals,wereconstructedtheomittedaccrualsfollowingPenmanandZhang’sprocedures(the‘C’score)andincludedtheminourearningsandstockreturnregressions.Theirinclusionhadnosigniﬁcantimpactonourresults.

2442

S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485

couldbooksalesof$90,representinganerrorofÀ10incurrentperiodaccrualsandearnings.Thisresultsinanerrors-in-variablesprobleminaccruals,becauseobservedaccrualsarenoisymeasuresofthefuturebeneﬁtsorobligationsthattheyrepresent.Notethatwhilewecharacterizetheerrorsasresultingfromaggressiveandconservativeaccounting,respectively,wedonotmeantoimplythatallerrorsresultfromintentionalearningsmanagement.ErrorscouldalsoresultfromtheneutralapplicationofGAAP(e.g.,aone-offgainfromaLIFOinventoryliquidation)andunintentionalerrors(e.g.,overestimatingthecreditworthinessofanewcustomer,oroverestimatingthefuturesalespriceofwork-in-processinventory).

Todeveloptheerrorsinvariablesrepresentationofthisproblem,weﬁrstdeﬁneAnastheunobservable‘perfectforesight’accrualthatwouldresultinanearningsperformancemeasureequaltoEÃ.

AÃ¼EÃÀC.

(3)

Referringbacktotheexampleabove,Anrepresentsthe$100amountthatisultimatelycollectedonthecreditsales.WenextdeﬁneobservedaccrualsasthesumofAnande,whereeisameanzero,independentlyandidenticallydistributederrortermthatisassumedtobeuncorrelatedwithCandAn:

A¼AÃþe.

(4)

Intheexampleabove,ewouldbe+10fortheaggressivemanagerandÀ10fortheconservativemanager.The‘true’earningspersistencerelationcanbewrittenintermsofunobservable‘perfectforesight’earnings,observablecashﬂowsandtheunobservable‘perfectforesight’accrualsas:

EÃtþ1¼gCtþgAtþ󰀁tþ1.

(5)

IfwereplaceEÃandAÃwiththeirobservablecounterparts,EandA,wehavetheclassicerrors-in-variablesmodelfortwoexplanatoryvariableswhereoneismeasuredwitherror:

Etþ1¼gCtþgAtþotþ1,(6)whereotþ1¼󰀁tþ1þetþ1Àget.

Becauseotþ1iscorrelatedwithAtthroughthemutualinclusionofet,theestimatedcoefﬁcientsonCðgCÞandAðgAÞwillprovidebiasedestimatesofg.4GrilichesandRingstad(1971)showthatthebiasesaregivenby

ðgAÀgÞ¼

varðeÞÀgvarðAÞ1Àr2C;A

,(7)

ðgCÀgÞ¼ÀrC;AðgAÀgÞ,(8)wherevar(e)andvar(A)denotethevarianceofeandA,respectively,andrC;A

denotesthecorrelationbetweenCandA.Eq.(7)indicatesthatgAisbiased

Anadditionalsourceofbiasisintroducediferrorsinaccrualsreverseinthenextperiod.Thesereversalsinduceanegativecorrelationbetweenetandetþ1,whichwillexaggeratethebiasesdescribedbelow,butnotaffecttheirdirections.

4S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485

443

downwardtowardszero,withthebiasincreasinginvar(e)/var(A),theproportionofthetotalvariationinobservedaccruals(A)thatisattributabletomeasurementerror(e).

Intermsofourearlierexample,var(e)/var(A)isincreasingintherangeofplausiblevaluesthattheerrorcantake.Moreunreliableaccrualsleadtogreaterpoten-tialerrors,resultinginanincreaseinvar(e).Wecanmodifyouroriginalexampletomaketheaccrualsmoreunreliablebyassumingthatwemakecreditsalesof$110ofwhich$90ultimatelyendsupbeingcollectedascashinthenextperiod.Furtherassumethatanaggressivemanagercouldplausiblybooksalesof$110,representinganerrorof+20incurrentperiodaccrualsandearnings,whileaconservativemanagercouldplausiblybooksalesof$70,representinganerrorofÀ20incurrentperiodaccrualsandearnings.Astherangeofplausibleerrorsincreases,var(e)/var(A)willincrease.Inotherwords,themoreuncertaintythereisabouttheamountofcashthatwillultimatelybecollected,thegreatertherangeofpossibleaccrualsthatcanbeplausiblybooked.Var(e)/Var(A)thereforeprovidesanaturalmeasureofthereliabilityofaccruals,withhighervaluesrepresentinglowerreliability.

Eq.(7)indicatesthatthebiasingAisalsoincreasinginthestrengthofthecorrelationbetweencashﬂowsandaccruals.Weknowfrompreviousresearch(seeDechow,1994)thatthiscorrelationisaroundÀ0.6forannualdata,exaggeratingthedownwardbiasbyover50%.ThebiasingCisafunctionofboththecorrelationbetweenobservedaccrualsandcashﬂowsandthebiasingA.UsingthehistoricalaverageannualcorrelationbetweenaccrualsandcashﬂowsofaroundÀ0.6,gCwillalsohaveadownwardbias,butitwillonlybe60%aslargeasthedownwardbiasingA.

Theaboveanalysisgeneratestwotestablepredictionsforourempiricalresearch:1.Ceterisparibus,themagnitudeofthedownwardbiasinthepersistencecoefﬁcientswillalwaysbegreaterfortheaccrualcomponentofearningsthanforthecashﬂowcomponentofearnings(i.e.,gA2gCo0).

2.Ceterisparibus,thedownwardbiasinthepersistencecoefﬁcientsontheaccrualcomponentsofearningsrelativetothecashﬂowcomponentofearnings,(gA2gC),isincreasingintheproportionofthevariationinaccrualsthatisattributabletomeasurementerror.Thisleadstothepredictionthat(gA2gC)willbemorenegativeforlessreliableaccruals.

2.2.Modellimitations

Aswithanymodel,themodelinSection2.1makesmanysimplifyingassumptions.Inthissection,weidentifysomeofthemoreimportantsimplifyingassumptions.Considerationofthesesimplifyingassumptionsisessentialforthreereasons.First,ifweﬁndthatthepredictionsofthemodelaresupportedbythedata,itisalwayspossiblethattheregularitiesinthedataareinsteadattributabletoalternativeexplanationsthatareassumedawaybyourmodel.Second,ifweﬁndthatthe

444

S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485

predictionsofthemodelarenotsupportedbythedata,itisalwayspossiblethattheyaremaskedbyconfoundingeffectsthatareassumedawaybyourmodel.Third,byidentifyingourkeysimplifyingassumptions,weprovideguidelinesforfutureresearchtocorroborateandextendouranalysisbyrelaxingtheseassumptions.TheﬁrstlimitationofourmodelisthatthespeciﬁcationinEq.(1)imposestheconstraintthatthepersistencecoefﬁcient,g,isconstantacrossAÃandC.Inotherwords,weimposetheconstraintthatthecoefﬁcientonaccrualsandcashﬂowsisequalintheabsenceofaccrualmeasurementerror.Thisconstraintiscommoninpriorempiricalwork(seeKothari,2001,pp.145–150forareview)andpriorresearchmodelingthepropertiesofthecashﬂowandaccrualcomponentsofearnings(seeEq.(17)inDechowetal.,1998,p.141).Nevertheless,plausibleextensionsofourmodelcouldleadtoviolationsofthisconstraintandalternativeexplanationsforourempiricalresults.Forexample,assumingthat(i)theaccrualcomponentofearningsiscloselyrelatedtosalesgrowth(e.g.,Jones,1991);and(ii)extremesalesgrowthismorelikelytobetransitory(e.g.,NissimandPenman,2001)alsogeneratesthepredictionthataccrualsarelesspersistentthancashﬂows.5Second,weassumethatmeasurementerrorinaccrualsisrepresentedbyameanzeroindependentlyandidenticallydistributederrorterm(e)thatisuncorrelatedwiththecashﬂow(C)and‘perfectforesight’accrual(AÃ)componentsofearnings.Inpractice,strategicmanagerialreactionandtheaccountingconventionofconserva-tismarelikelytoresultinviolationsofthisassumption.Thereareseveralpossibilitiesinthisrespect.Onepossibilityisthatmanagerssmoothearnings,causinganegativecorrelationbetweeneandbothCandAÃ.Asecondpossibilityisthatmanagerstakebigbathsduringperiodsofpoorperformance,causingapositivecorrelationbetweeneandbothCandAÃ.Athirdpossibilityisthatopportunisticmanagerswithlimitedtenuresandlimitedliabilityfacegreaterincentivestogeneratepositiveerrors,causingpositiveerrorstobemagniﬁed(seeWatts,2003).Afourthpossibilityisthatsinceaccountingconservatismimposesalowerveriﬁabilitythresholdforlossesversusgains,negativeerrorswillbemagniﬁed(seeWatts,2003).Thesepossibilitiesarenotmutuallyexclusiveandarepotentiallyoffsetting,buttheyarenotaddressedbyourmodel.

Third,ouranalysisassumesthatcashﬂowsaremeasuredwithouterror.Inpractice,however,itispossiblethatcashﬂowscouldbemeasuredwitherrorduetomanagerialmanipulationofcashﬂows(seeRoychowdhury,2003;Grahametal.,2005).Ifbothcashﬂowsandaccrualsaremeasuredwitherror,thebiasesinthecashﬂowandaccrualpersistencecoefﬁcientswilldependonthevariance–covariancematrixoftheerrors(seeKlepperandLeamer,1984).Weconductedsimulationsofthis‘errors-in-two-variables’scenario.Thesimulationresultsindicatethatifonevariableismeasuredwithsigniﬁcantlygreatererrorthantheother,thedownwardbiascontinuestobemuchgreaterintheestimatedcoefﬁcientonthevariablewiththegreatererror(ourprimarysimulationsuseanerrorvarianceratioof5–1).Thus,asapracticalmatter,thepredictionsofourmodelsimplyrequirethataccrualsaremeasuredwithsigniﬁcantlygreatererrorthancashﬂows.

5WethankS.P.Kothariforbringingthisalternativeexplanationtoourattention.

S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485

445

Finally,wehaveassumedthattheaccountingconstructofreliabilityisperfectlycapturedbyourmodelofmeasurementerrorinaccruals.Whilewehavearguedthatthischaracterizationseemsappropriate,itispossiblethatthereareotherelementsofreliabilitythataremissingfromourmodel.2.3.Deﬁnitionofaccruals

Sloan(1996)followstheconventioninacademicresearchofdeﬁningaccrualsasthechangeinnon-cashworkingcapitallessdepreciationexpense.ThisconventioncanbetracedbackatleastasfarasHealy(1985)andcorrespondscloselywiththedeﬁnitionofoperatingaccrualsusedintheFASB’sStatementofFinancialAccountingStandardNumber95‘‘StatementofCashlows’’.However,thisdeﬁnitionofaccrualsomitsmanyaccrualsanddeferralsrelatingtonon-currentoperatingassets,non-currentoperatingliabilities,non-cashﬁnancialassetsandﬁnancialliabilities.

Toformallyderiveacomprehensivemeasureofaccruals,webeginbynotingthatintheabsenceofaccrualaccounting,theonlyassetorliabilityaccountappearingonthebalancesheetwouldbethecashassetaccount.Allotherassetandliabilityaccountsareproductsoftheaccrualaccountingprocess.Netassetswouldthereforeequalcash,andthroughthebalancesheetidentity,owners’equitywouldalsoequalcash:

CashBasisNetAssets¼Cash¼CashBasisOwnersEquity

Earningsunderthecashbasisofaccounting,‘cashearnings’,canthenbederivedthroughthecleansurplusrelationas:

CashEarningst¼ChangeinCashBasisOwners’Equityt

þNetCashDistributionstoEquityt

¼ChangeinCashtþNetCashDistributionstoEquityt

Netcashdistributionstoequityholdersrepresentallcashpaymentsfromtheﬁrmtoequityholders(i.e.,cashdividendsplusstockrepurchaseslessequityissuances).Thisexpressiontellsusthatcashearningsrepresentthenetcashreceiptsoftheﬁrm,exclusiveoftheeffectofanycashdistributions(receipts)to(from)equityholders.Conventionalaccrual-basedearningscanbedeﬁnedthroughtheusualdeﬁnitionofnetassets:

AccrualEarningst¼ChangeinOwners’EquitytþNetCashDistributionstoEquityt

¼ChangeinNetAssetstþNetCashDistributionstoEquityt

¼ChangeinAssetstÀChangeinLiabilitiestþNetCashDistributionstoEquityt

Accrualsrepresentthedifferencebetweenaccrualearningsandcashearnings:Accrualst¼AccrualEarningstÀCashEarningst

¼ChangeinNon-CashAssetstÀChangeinLiabilitiest

446

S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485

Withoutaccrualaccounting,theonlyassetorliabilityiscash,andsoaccrualsrepresentthechangeinallnon-cashassetslessthechangeinallliabilities.

ThiscomprehensivedeﬁnitionofaccrualsdiffersfromthedeﬁnitionofaccrualsusedbyHealy(1985),Sloan(1996)andothersinseveralkeyrespects.First,thisdeﬁnitionincorporatesaccrualsrelatingtonon-currentoperatingassets,suchascapitalexpenditures.Healy’sdeﬁnitionignorestheoriginationofnon-currentoperatingassetaccrualsandonlyincorporatesthereversalofasubsetofnon-currentoperatingassetaccrualsthroughsubtractionofdepreciationexpense.Second,ourdeﬁnitionincorporatesaccrualsrelatingtonon-currentoperatingliabilities,suchaspost-retirementbeneﬁtobligations.Third,ourdeﬁnitionincorporatesaccrualsrelatingtoﬁnancialassets,suchaslong-termreceivables.Finally,ourdeﬁnitionincorporatesaccrualsrelatedtoﬁnancialliabilities,suchaslong-termdebt.Whiletheseaccrualsrepresentavarietyofdifferenttransactionsandevents,theyareneverthelessallmanifestationsoftheaccrualaccountingprocess.Wethereforebeginourempiricalanalysisusingthiscomprehensivedeﬁnitionofaccruals.2.4.Categorizationofaccruals

Thebalancesheetprovidesasystematicclassiﬁcationofaccountingaccrualsbasedonthenatureoftheunderlyingbeneﬁtsorobligationsthattheyrepresent.Somebalancesheetcategories,suchasmarketablesecuritiesandshort-termdebt,cangenerallybemeasuredwithhighreliability.Otherbalancesheetclassiﬁcations,suchasaccountsreceivableandintangibleassets,aregenerallymeasuredwithlowerreliability.Inthissection,wedecomposeaccrualsalongbroadbalancesheetcategoriesanduseourknowledgeofthemeasurementissuesunderlyingeachaccrualcategorytomakequalitativeassessmentsconcerningtherelativereliabilityofeachcategory.Theseassessmentsprovidethebasisforpredictionsconcerningtherelativemagnitudesofthepersistencecoefﬁcients.

OurcompletebalancesheetdecompositionisillustratedinTable1andasummaryofourreliabilityassessmentsisprovidedinTable2.Ourinitialbalancesheetdecompositionisbasedonthenatureoftheunderlyingbusinessactivity.Weusethreebroadcategoriesofbusinessactivities—currentoperatingactivities,non-currentoperatingactivitiesandﬁnancingactivities.Werefertothecorrespondingaccrualcategoriesasthechangeinnon-cashworkingcapital(DWC),thechangeinnetnon-currentoperatingassets(DNCO)andthechangeinnetﬁnancialassets(DFIN),respectively:

Accruals¼DWCþDNCOþDFIN;

(9)

whereDWCrepresentsthechangeincurrentoperatingassets,netofcashandshort-terminvestments,lessthechangeincurrentoperatingliabilities,netofshort-termdebt.TheseDWCaccrualsformthecoreofthetraditionalmeasureofaccrualsusedbySloan(1996).WegenerallyagreewithSloan’soriginalassessmentthatconsiderablesubjectivityisinvolvedinthemeasurementofthisaccrualcategory.Thereare,however,signiﬁcantdifferencesbetweentheunderlyingassetandliabilitycomponents,soweconductanextendedaccrualdecompositionthatfurther

Table1

IllustrationofbalancesheetcategorizationofaccrualsNon-cashassets

Liabilities

Dataitem

InitialExtendedDataitemInitialExtendedShortterminvestments#193DFINDSTIDebtincurrentliabilities#34DFINDFINLReceivables#2DWCDCOAAccountspayable#70DWCDCOLInventory

#3DWCDCOAIncometaxespayable#71DWCDCOLOthercurrentassets

#68DWCDCOAOthercurrentliabilities#72DWCDCOLProperty,plantandequipment,net#8DNCODNCOALongtermdebt#9DFINDFINLInvestments—equitymethod#31DNCODNCOAOtherliabilities#75DNCODNCOLInvestments—other#32DFINDLTIDeferredtaxes#35DNCODNCOLIntangibles#33DNCODNCOAMinorityinterest#38DNCODNCOLOtherassets

#69

DNCO

DNCOA

Preferredstock#130

DFIN

DFINL

DataItemdenotestheCompustatannualdataitemnumber.Initialdenotestheaccrualcomponentinwhichtheitemisclassiﬁedforourinitialaccrualdecomposition.Extendeddenotestheaccrualcomponentinwhichtheitemisclassiﬁedforourextendedaccrualdecomposition.

DWCisthechangeinworkingcapitalaccruals,deﬁnedasWCt–WCtÀ1.WC¼CurrentOperatingAssets(COA)ÀCurrentOperatingLiabilities(COL)whereCOA¼CurrentAssets(CompustatItem#4)ÀCashandShortTermInvestments(STI)(CompustatItem#1).COL¼CurrentLiabilities(CompustatItem#5)ÀDebtinCurrentLiabilities(CompustatItem#34).

DFINisdeﬁnedasFINtÀFINtÀ1.FIN¼FinancialAssets(FINA)ÀFinancialLiabilities(FINL).FINA¼ShortTermInvestments(STI)(CompustatItem#193)+LongTermInvestments(LTI)(CompustatItem#32).FINL¼Longtermdebt(CompustatItem#9)+DebtinCurrentLiabilities(CompustatItem#34)+PreferredStock(CompustatItem#130).

S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485447Table2

Summaryofreliabilityassessmentsbyaccrualcategory

Accrualcategory

DecompositionReliabilitySummaryofreasoningbehindreliabilityassessment

levelassessmentDCOA

Extended

Low

Categoryisdominatedbyreceivablesandinventory.Receivablesrequiretheestimationofuncollectiblesandareacommonearningsmanagementtool(e.g.,channelstufﬁng).Inventoryaccrualsentailvariouscostﬂowassumptions/allocationsandsubjectivewrite-downs.

DCOLExtendedHighCategoryisdominatedbypayables,whichrepresentﬁnancialobligationsoftheﬁrmthatcanbemeasuredwithahighdegreeofreliability.

DWCInitialMediumCombinationofDCOA(lowreliability)andDCOL(highreliability)suggestsmediumreliability.

DNCOA

Extended

Low

CategoryisdominatedbyPP&Eandintangibles.BothPP&Eandinternallygeneratedintangibles(e.g.,capitalizedsoftwaredevelopmentcosts)involvesubjectivecapitalizationdecisions.Moreover,PP&Eandintangiblesinvolvesubjectiveamortizationandwrite-downdecisions.

DNCOLExtendedMediumCategoryincludeslong-termpayables,deferredtaxesandpostretirementbeneﬁtobligations.Best

characterizedasamixtureofaccrualswithvaryingdegreesofreliability,soclassifyasmediumreliability.DNCOInitialLow/MediumCombinationofDNCOA(lowreliability)andDNCOL(mediumreliability)suggestslow/mediumreliability.DSTIExtendedHighRepresentsmarketableﬁnancialsecuritiesthatareexpectedtobesoldwithin12months.Marketvaluesofmarketableﬁnancialsecuritiescanbemeasuredwithahighdegreeofreliability.

DLTI

Extended

Medium

Categoryincludeslong-termreceivablesandinvestmentsinmarketablesecuritiesthatareexpectedtobeheldformorethanayear.Bestcharacterizedasamixtureofaccrualswithvaryingdegreesofreliability,soclassifyasmediumreliability.

DFINLExtendedHighCategorycontainsinterest-bearingﬁnancialobligations.Typicallymeasuredwithahighdegreeofreliabilityusingeffectiveinterestrateatorigination.

DFIN

Initial

High

CombinationofDSTI(highreliability),DLTI(mediumreliability)andDFINL(highreliability)suggestshighreliability.

448S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485

449

decomposesDWCintoitsunderlyingasset(DCOA)andliability(DCOL)components:

DWC¼DCOAÀDCOL:

(10)

ThemajorunderlyingassetsdrivingDCOAaretradeaccountsreceivableandinventory(seeThomasandZhang,2002).Bothofthesecategoriesaremeasuredwithrelativelylowreliability.Accountsreceivableaccrualsinvolvethesubjectiveestimationofuncollectibleaccounts.Moreover,accountsreceivableiscommonlyusedtomanipulateearningsthroughtechniquessuchastrade-loadingandprematurerevenuerecognition(seeDechowetal.,1996).Themeasurementofinventoryallowsforanumberofdifferentcostﬂowassumptionsandinvolvessubjectivecostallocations.Forexample,LIFOliquidationscanartiﬁciallydistortreportedincome,andtheallocationofﬁxedcoststoinventorycanleadtodistortionswhenproductionlevelsareunusuallyhighorlow.Inventoryaccountingalsocallsforsubjectivewrite-downdecisionsbasedonestimatesoffairvalue.

ThemajorliabilitydrivingDCOLisaccountspayable.Incontrasttoreceivablesandinventory,payablescangenerallybemeasuredwithahighdegreeofreliability.Accountspayableareﬁnancialobligationstosuppliersthatarerecordedattheirfacevalue.Ifaﬁrmistocontinueasagoingconcern,thenitwilltypicallyhavetopayitssuppliersinfull.Theonlycommonsourceofsubjectivitythatarisesforpayablesisinestimatingdiscountsforpromptpaymentthatmaybeofferedbysuppliers.Butsincetheamountofanydiscountscantypicallybeveriﬁedwiththesuppliers,thereisrelativelylittleroomforerror.

Oursecondmajorcategoryofaccrualsisnon-currentoperatingaccruals,DNCO.Thiscategoryismeasuredasthechangeinnon-currentassets,netoflong-termnon-equityinvestmentsandadvances,lessthechangeinnon-currentliabilities,netoflong-termdebt.Itcontainsaccrualsthathavegenerallybeenignoredinpreviousresearch.Nevertheless,aswewillarguebelow,thiscategorycontainssubjectiveandunreliableaccruals.Aswiththeworkingcapitalaccruals,wefurtherdecomposeDNCOintoitsunderlyingasset(DNCOA)andliability(DNCOL)components:

DNCO¼DNCOAÀDNCOL:

(11)

ThemajorunderlyingcomponentsofDNCOAareproperty,plantandequipment(PP&E)andintangibles.Considerableuncertaintyisinvolvedintheestimationoftheseaccruals.First,thereisconsiderablesubjectivityinvolvedintheinitialdecisionofwhichcoststocapitalizeforbothPP&Eandrecognizableinternallygeneratedintangibles(suchascapitalizedsoftwaredevelopmentcosts).Forexample,thewell-publicizedaccountingscandalatWorldCominvolvedbillionsofdollarsofoperatingcoststhatwereaggressivelycapitalizedasPP&E.Second,thereisconsiderablesubjectivityinvolvedinselectinganamortizationscheduleforbothPP&Eandintangibles.Thedepreciation/amortizationmethod,theusefullifeandthesalvagevalueareallsubjectivedecisionsthatimpactthisaccrualcategory.Third,bothPP&Eandintangiblesaresubjecttoperiodicwrite-downswhentheyaredeterminedtohavebeenimpaired.Theestimationoftheamountoftheseimpairmentsinvolvesgreatsubjectivity.Forexample,AOLTimeWarnerwroteoffover$50billionina

450

S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485

singlequarterinrelationtogoodwillacquiredinthemergerofAOLandTimeWarner.Becausesuchwrite-downsaretypicallymadeinlargediscreteamounts,theyinevitablyintroduceperiodicdistortionsintoearnings.

DNCOL,theliabilitycomponentofDNCO,isdrivenbyawidevarietyofdifferentliabilities.Examplesincludelong-termpayables,deferredtaxesandpostretirementbeneﬁts.Someoftheseliabilities,suchaslong-termpayables,canbemeasuredwithahighdegreeofreliability.Others,suchaspostretirementbeneﬁts,involvemanysubjectiveestimatesandsoaremeasuredwithalowdegreeofreliability.Themixofdifferentliabilitiesinthiscategorymakesitdifﬁculttoassignanunambiguousreliabilityranking.Weratethereliabilityofthiscategoryasmedium,reﬂectingtherangeofdifferentliabilitieswithvaryingdegreesofreliability.

Ourthirdandﬁnalmajorcategoryofaccrualsisthechangeinnetﬁnancialassets,DFIN.DFINismeasuredasthechangeinshort-terminvestmentsandlong-terminvestmentslessthechangeinshort-termdebt,long-termdebtandpreferredstock.Thiscategoryofaccrualshasalsobeenignoredinpastaccrualsresearch.However,aswewillarguebelow,theseaccrualscangenerallybemeasuredwithahighdegreeofreliability,andinthisrespectaresimilartocash.Infact,someoftheseaccrualsareactuallyreferredtoas‘cashequivalents’toreﬂecttheirclosecorrespondencewithcash.Therearesomesubtledifferencesbetweenthevariousﬁnancialassetsandliabilities,andtohighlightthesedifferences,wefurtherdecomposeDFINintoitsunderlyingshort-terminvestment(DSTI),long-terminvestment(DLTI)andﬁnancialliability(DFINL)components:

DFIN¼DSTIþDLTIÀDFINL:

(12)

Short-terminvestmentsandﬁnancialliabilitiescanbothbemeasuredwithahighdegreeofreliability.Short-terminvestmentsarecomprisedofsecuritieswithreadilyobservablemarketvaluesthatareexpectedtobeconvertedintocashwithinayear.Financialliabilitiesincludedebt,capitalizedleaseobligationsandpreferredstock,mostofwhichareﬁnancialobligationsthatarevaluedatthepresentvalueofthefuturecashobligationsowedbytheﬁrm,withthediscountrateﬁxedattheissuedate.Firmsarenotpermittedtobookanallowancefortheanticipatednon-paymentoftheirownﬁnancialobligations,sothereisrelativelylittlesubjectivityinvolvedinthemeasurementoftheseliabilities.Wethereforeexpectthereliabilityofshort-terminvestmentandﬁnancialliabilityaccrualstoberelativelyhigh.Reliabilityis,however,moreofanissueforlong-terminvestments.Thelong-terminvestmentscategoryincorporatesavarietyofﬁnancialassetsincludinglong-termreceivablesandlong-terminvestmentsinmarketablesecurities.Long-termreceivableshaveatleastasmuchpotentialformeasurementerrorasshort-termreceivablesandhavebeenattheheartofsomeofthemostcelebratedcasesofearningsmanipulation.6Ontheotherhand,mostlong-terminvestmentsinliquidmarketablesecuritiescanbemeasuredwithahighdegreeofreliability.Toreﬂectthemixoflowandhigh

Inawell-publicizedaccountingscandal,BostonChickenreportedsteadilygrowingearningsbyoverstatingitslong-termnotesreceivableintheyearsfrom1994to1996.

6S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485

451

reliabilityaccrualsincludedinDLTI,weassignitanoverallreliabilityratingofmedium.7Table2listsreliabilityassessmentsforboththeinitialandtheextendedbalancesheetdecompositions.Thereliabilityassessmentsfortheinitialdecompositionrepresentasynthesisofthereliabilityassessmentsfortheaccrualcategoriesfromtheextendeddecomposition.DWCisassignedareliabilityratingofmedium,becauseitcombinesthelowreliabilityaccrualsinDCOAwiththehighreliabilityaccrualsinDCOL.DNCOreceivesalow/mediumreliabilityrating,asitcombinesthelowreliabilityaccrualsinDNCOAwiththemediumreliabilityaccrualsinDCOL.Finally,DFINreceivesahighreliabilityrating,becauseitrepresentsthecombinationoftwohighreliabilitycategorieswithamediumreliabilitycategory.Wepredictthataccrualsinthehighreliabilitycategorieswillhavepersistencecoefﬁcientsslightlylowerthancashﬂows,whileaccrualsinthemediumandlowreliabilitycategorieswillhaveprogressivelylowerpersistencecoefﬁcients.3.Data

Ourempiricaltestsemploydatafromtwosources.FinancialstatementdataareobtainedfromtheCompustatannualdatabaseandstockreturndataareobtainedfromtheCRSPdailystockreturnsﬁles.Oursampleperiodcoversallﬁrm-yearswithavailabledataonCompustatandCRSPfortheperiod1962–2001.Weexcludeallﬁrm-yearobservationswithSICcodesintherange6000–6999(ﬁnancialcompanies)becausethedemarcationbetweenoperatingandﬁnancingactivitiesisnotclearintheseﬁrms.Weeliminateﬁrm-yearobservationswithinsufﬁcientdataonCompustattocomputetheprimaryﬁnancialstatementvariablesusedinourtests.8Thesecriteriayieldaﬁnalsamplesizewithnon-missingﬁnancialstatementandreturnsdataof108,617ﬁrm-yearobservations.

AsdiscussedinSection2.4,ourmeasureoftotalaccruals,TACC,isdeﬁnedasfollows:TACC¼DWC+DNCO+DFINwhere:

(i)DWC,thechangeinnetworkingcapitalisdeﬁnedasWCt–WCtÀ1.WCiscalculatedasCurrentOperatingAssets(COA)ÀCurrentOperatingLiabilities(COL),andCOA¼CurrentAssets(CompustatItem#4)ÀCashandShortTermInvestments(STI)(CompustatItem#1),andCOL¼CurrentLiabilities(CompustatItem#5)ÀDebtinCurrentLiabilities(CompustatItem#34).

Ideally,wewouldliketodecomposelong-terminvestmentsfurthertodistinguishbetweenitemssuchaslong-termreceivablesandinvestmentsinmarketablesecurities.Unfortunately,Compustatgroupstheseitemstogetherinannualdataitem#32anddoesnotprovideaﬁnerdecomposition.8Speciﬁcally,werequireavailabilityofCompustatdataitems1,4,5,6,12and181inboththecurrentandpreviousyearanddataitem178inthecurrentyearinordertokeepaﬁrm-yearinthesample.Ifdataitems9,32,34,130or193aremissing,wesetthemequaltozeroratherthaneliminatingtheobservation.Eachofthelatterdataitemsrepresentabalancesheetitemthatmaynotberelevantformanycompanies(e.g.,preferredstock),sowesetthemtozeroratherthanneedlesslydiscardingobservations.Allresultsarequalitativelysimilarifweinsteadestimateeachregressionusingonlyobservationswithalldataavailableforthatparticularregression.

7452

S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485

(ii)DNCO,thechangeinnetnon-currentoperatingassetsisdeﬁnedas

NCOt–NCOtÀ1.NCOiscalculatedasNon-CurrentOperatingAssets(NCOA)—Non-CurrentOperatingLiabilities(NCOL),andNCOA¼TotalAssets(Compustatitem#6)ÀCurrentAssets(CompustatItem#4)ÀInvest-mentsandAdvances(CompustatItem#32),andNCOL¼TotalLiabilities(CompustatItem#181)ÀCurrentLiabilities(CompustatItem#5)–Long-TermDebt(CompustatItem#9).

(iii)DIN,thechangeinnetﬁnancialassetsisdeﬁnedasINt–FINtÀ1and

FIN¼FinancialAssets(FINA)ÀFinancialLiabilities(FINL).FINA¼ShortTermInvestments(STI)(CompustatItem#193)+LongTermInvestments(LTI)(CompustatItem#32),andFINL¼LongTermDebt(CompustatItem#9)+DebtinCurrentLiabilities(CompustatItem#34)+PreferredStock(CompustatItem#130).Asinpreviousresearch,wedeﬂateeachofthesecomponentsofearningsbyaveragetotalassets.9Forouranalysisofthepersistenceofthevariouscomponentsofearnings,weusereturnonassets(ROA).ROAiscalculatedasoperatingincomeafterdepreciation(CompustatItem#178)deﬂatedbyaveragetotalassets.10Inourregressionanalyses,eachcomponentofearningsiswinsorizedat+1andÀ1inordertoeliminatetheinﬂuenceofextremeoutliers.11Ourextendedbalancesheetdecompositionfurtherdecomposeseachofthethreeaccrualcomponentsdeﬁnedaboveintotheirrespectiveassetandliabilitycomponents.DWCisdecomposedintoitscurrentoperatingasset(DCOA)andcurrentoperatingliability(DCOL)components,asdeﬁnedabove.DNCOisdecomposedintoitsnoncurrentoperatingasset(DNCOA)andnoncurrentoperatingliability(DNCOL)components,asdeﬁnedabove.Finally,DFINisdecomposedintoitsrespectiveshort-terminvestment(DSTI),long-terminvestment(DLTI)andliability(DFINL)components,asdeﬁnedabove.ThebalancesheetitemsandtheirassociatedaccrualcategorizationsaresummarizedinTable1.

OurstockreturntestsusedatafromtheCRSPdailyﬁles.Stockreturnsaremeasuredusingcompoundedbuy-holdsize-adjustedreturns,inclusiveofdividendsandotherdistributions.Thesize-adjustedreturniscalculatedbydeductingthevalue-weightedaveragereturnforallﬁrmsinthesamesize-matcheddecile,wheresizeismeasuredasthemarketcapitalizationatthebeginningofthereturncumulationperiod.Returnsarecalculatedfora12-monthperiodbeginningfour

BarthandKallapur(1996)showthatdeﬂationcanintroducebiasesintoregressioncoefﬁcientswhenthedeﬂatormeasuresthetrueunderlyingscalevariablewitherror.Suchbiasesmaybepresentinouranalysis.However,wehavenoreasontobelievethatanysuchbiaseswoulddifferentiallyaffectthecashﬂowandaccrualcomponentsofearnings.10FollowingSloan(1996),weuseameasureofincomefromcontinuingoperationsinouranalysisofearningspersistenceasitisunaffectedbyexplicitlyidentiﬁednon-recurringcomponentsofnetincome.Thisincreasestheoverallpowerofourpersistenceregressions.However,inferencesregardingtherelativepersistenceofthedifferentcomponentsofearningsarequalitativelysimilarifweinsteadusebottomlinenetincome.11Inferencesregardingtherelativepersistenceandpricingofthedifferentcomponentsofearningsarequalitativelysimilarwithoutwinsorization.Thewinsorizedresults,however,havelowerstandarderrors.

9S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485

453

monthsaftertheendoftheﬁscalyear.12Forﬁrmsthataredelistedduringourfuturereturnwindow,wecalculatetheremainingreturnbyﬁrstapplyingCRSP’sdelistingreturnandthenreinvestinganyremainingproceedsintheCRSPvalue-weightedmarketindex.13Thismitigatesconcernswithpotentialsurvivorshipbiases.

Apotentialshortcomingofouraccrualmeasurementproceduresisthattheyusebalancesheetdata.HribarandCollins(2002)pointoutthattheuseofbalancesheetdatacanintroduceerrorsintothemeasurementofaccruals,particularlyinthepresenceofmergersandacquisitions.Indeed,theempiricaldistributionsofouraccrualvariablescontainedasmallnumberofextremeoutliers,andinspectionofsometheseoutliersindicatesthattheyareattributabletomeasurementissuesassociatedwiththeuseofthebalancesheetapproach.

Weconductthreeprocedurestoconﬁrmtherobustnessofourresultswithrespecttopotentialmeasurementissuesintroducedbythebalancesheetapproach.First,wedelete(asopposedtowinsorize)ﬁrm-yearobservationsfromourregressionanalysiswhenanyoneoftheregressionvariablesexceedsoneinabsolutevalue.Wedothisinanefforttoremoveoutliersthatmaybeattributabletomeasurementissues.Ourresultsaresimilarforthisreducedsample.Second,weconﬁrmtherobustnessofourtotalaccrualresultsusingameasureoftotalaccrualsderivedfromdatainthestatementofcashﬂowsratherthandatafromthebalancesheet.Noteinthisrespectthatcashﬂowdataisonlyavailablefrom1988,andwhilewecancomputetotalaccrualsusingstatementofcashﬂowdata,wecannotcomputemanyofourunderlyingaccrualcomponentswithoutusingbalancesheetdata.Thisalternativemeasureofaccrualsiscomputedasthedifferencebetweenincomebeforeextraordinaryitems(Compustatitem123)lesstotaloperating,investingandﬁnancingcashﬂows(items308,311and313)plussalesofcommonstock(item108)lessstockrepurchasesanddividends(items115and127).Resultsusingthiscashﬂow-basedmeasureoftotalaccrualsarequalitativelysimilartotheresultsobtainedusingourbalancesheetmeasureoftotalaccruals.Third,wealsoconductedouranalysesafterremovingﬁrmsexperiencinganincreaseinreportedgoodwill.Thisprocedureisexpectedtoeliminatesomeobservationsinvolvingsigniﬁcantmergersandacquisitions.14Ourresultsremainqualitativelysimilar.

4.Results

Wepresentourresultsinfoursections.Section4.1beginswithdescriptivestatisticsforouraccrualdecompositions.Section4.2presentstestsofourpredictions

Thisisstandardintheliterature,asﬁrmsgenerallyﬁleForm10-K’swithinfourmonthsaftertheendoftheﬁscalyear(seeAlfordetal.,1994).13Firmsthataredelistedforpoorperformance(delistingcodes500and520–584)frequentlyhavemissingdelistingreturns(seeShumway1997).WecontrolforthispotentialbiasbyapplyingdelistingreturnsofÀ100%insuchcases.Ourresultsarequalitativelysimilarifwemakenosuchadjustment.14Notallobservationswithmergersandacquisitionswillbeeliminated.Mergersandacquisitionsaccountedforusingthepoolingofinterestmethodorinvolvingthecreationofnewgoodwillthatislessthanorequaltotheamortizationofgoodwillfortheperiodwillnotbeeliminated.

12454

S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485

concerningtherelativemagnitudesofthepersistencecoefﬁcientsontheaccrualcomponentsofearnings.Section4.3providesadditionalanalysisofcertainissuesuncoveredinSection4.2.Finally,Section4.4presentsourstockreturntests,providingevidenceontheextenttowhichinvestorsappeartounderstandtherelativemagnitudesofthepersistencecoefﬁcients.

4.1.Descriptivestatistics

Webeginbypresentingunivariatestatisticsandpair-wisecorrelationsforourkeyvariables.Weorganizethesedescriptivestatisticsaroundtheaccrualdecompositionsthatweusetomotivateourempiricalanalysis.PanelAofTable3containsunivariatestatisticsfortheinitialaccrualdecomposition.ThemeanvalueofTACCis0.052,indicatingtheaverageﬁrm’saccrualsarejustover5%oftotalassets.NotethatthepositivemeanvalueforaccrualsdocumentedherediffersfromthenegativemeanvalueforaccrualsdocumentedbySloan(1996).ThekeyreasonforthisdifferenceisthatSloan’sdeﬁnitionofaccrualsincludesthereversalofcertainnon-currentoperatingassetaccruals(throughthesubtrac-tionofdepreciationandamortizationexpense),butdoesnotincludetheoriginationoftheseaccruals.Inspectionoftheworkingcapital(DWC),netnon-currentoperatingasset(DNCO)andﬁnancing(DFIN)accrualsindicatesthatDWCandDNCObothhavepositivemeans,whileDFINhasanegativemean.Thesemeansindicatethattheaverageﬁrmisgrowingitsnetoperatingassetsandreducingitsnetﬁnancialassets(e.g.,increasingdebt)toﬁnancethisgrowth.Focusingonthestandarddeviationsoftheaccrualcomponents,weseethatDWChastheloweststandarddeviation(0.108),followedbyDNCO(0.151)andDFIN(0.181).ThesestandarddeviationssuggestthatDNCOandDFIN,thenewaccrualcomponentsexaminedinthispaper,accountforarelativelylargeproportionofthevariationintotalaccruals.

PanelBofTable3reportsthepair-wisecorrelationsfortheinitialaccrualdecomposition.Thecorrelationsbetweentheaccrualcomponentsrevealseveralregularities.DWCandDNCOarepositivelycorrelated,indicatingthatﬁrmstendtogrowtheircurrentandnon-currentoperatingactivitiesintandem.Incontrast,bothDWCandDNCOareverystronglynegativelycorrelatedwithDFIN.Thenegativecorrelationsindicatethatﬁrmstendtoﬁnancegrowthintheirnetoperatingassetsbyusingupﬁnancialassetsand/orgeneratingﬁnan-cialliabilities.Whatisclearfromthesedescriptivestatisticsisthattheﬁnanc-ingaccruals(DFIN)behaveverydifferentlyfromtheoperatingaccruals(DWCandDNCO).

Table4providesdescriptivestatisticsfortheextendedaccrualdecomposition.Recallthatthemajorextensioninthisdecompositioninvolvesseparatingtheaccrualcomponentsintotheirassetandliabilitysubcomponents.ThestandarddeviationsinpanelAindicatethatmostofthevariationinDWCisattributabletoDCOA,mostofthevariationinDNCOisattributabletoDNCOAandmostofthevariationinDFINisattributabletoDINL.Thus,variationinoperatingaccrualstendstobe

S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485

455

Table3

Descriptivestatisticsandcorrelationsfortheinitialaccrualdecomposition.Thesampleconsistsof108,617ﬁrm-yearobservationsfrom1962to2001PanelA:Descriptivestatistics

Mean

TACCtDWCtDNCOtDFINtROAtROAt+1FROAt+2RETt+10.0520.0220.051À0.0210.0700.0630.0780.010

Std.Dev.0.1810.1080.1510.1810.1750.1770.1300.715

25%À0.007À0.020À0.009À0.0800.0320.0250.040À0.330

Median0.0390.0150.025À0.0060.0930.0890.091À0.078

75%0.0980.0630.0820.0410.1510.1460.1400.199

PanelB:Correlationmatrix—Pearson(abovediagonal)andSpearman(belowdiagonal)(p-valuesshowninitalicsbelowcorrelations)

TACCtTACCtDWCtDNCOtDFINtROAtROAt+1FROAt+2RETt+1—

DWCt0.392(0.0001)

DNCOt0.423(0.0001)

DFINtROAtROAt+1FROAt+2RETt+1À0.058(0.0001)À0.046(0.0001)À0.062(0.0001)0.021(0.0001)À0.003(0.3098)0.138(0.0001)0.126(0.0001)—

0.4180.2500.1250.072(0.0001)(0.0001)(0.0001)(0.0001)

0.419—(0.0001)0.416(0.0001)0.226(0.0001)0.437(0.0001)0.274(0.0001)0.164(0.0001)

0.108À0.2880.2240.1050.059(0.0001)(0.0001)(0.0001)(0.0001)(0.0001)

À0.4750.0850.0240.020(0.0001)(0.0001)(0.0001)(0.0001)

0.0430.0380.014(0.0001)(0.0001)(0.0003)

0.7920.613(0.0001)(0.0001)

0.696(0.0001)

0.180—(0.0001)

À0.354À0.470

(0.0001)(0.0001)—0.261(0.0001)0.140(0.0001)0.072(0.0001)

0.215(0.0001)0.106(0.0001)0.071(0.0001)

0.061—(0.0001)

0.0810.776—(0.0001)(0.0001)

0.0460.5660.668—(0.0001)(0.0001)(0.0001)

0.0520.0870.3010.272(0.0001)(0.0001)(0.0001)(0.0001)

À0.049À0.060À0.066(0.0001)(0.0001)(0.0001)

TACCistotalaccrualsfromthebalancesheetapproach.ItiscalculatedasDWorkingCapital

(DWC)+DNon-CurrentOperating(DNCO)+DFinancial(DFIN).Totalaccrualsandallofitscomponents(describedbelow)aredeﬂatedbyaveragetotalassets.

DWCisdeﬁnedasWCtÀWCtÀ1.WC¼CurrentOperatingAssets(COA)ÀCurrentOperatingLiabilities(COL)whereCOA¼CurrentAssets(CompustatItem#4)ÀCashandShortTermInvestments(STI)(CompustatItem#1).COL¼CurrentLiabilities(CompustatItem#5)ÀDebtinCurrentLiabilities(CompustatItem#34).

456

S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485

Table3(continued)

(CompustatItem#181)ÀCurrentLiabilities(CompustatItem#5)–Long-termdebt(CompustatItem#9).DFINisdeﬁnedasFINtÀFINtÀ1.FIN¼FinancialAssets(FINA)ÀFinancialLiabilities(FINL).FINA¼ShortTermInvestments(STI)(CompustatItem#193)+LongTermInvestments(LTI)(CompustatItem#32).FINL¼Longtermdebt(CompustatItem#9)+DebtinCurrentLiabilities(CompustatItem#34)+PreferredStock(CompustatItem#130).

ROAisoperatingincomeafterdepreciation(CompustatItem#178)deﬂatedbyaveragetotalassets.FROAistheaverageROAoverthenextfouryears.Forexample,FROAt+2istheaverageROAforyearst+2throught+5.Thereare72,475observationsforFROA.

RETistheannualbuy-holdsize-adjustedreturn.Thesize-adjustedreturniscalculatedbydeductingthevalue-weightedaveragereturnforallﬁrmsinthesamesize-matcheddecile,wheresizeismeasuredasmarketcapitalizationatthebeginningofthereturncumulationperiod.Thereturncumulationperiodbeginsfourmonthsaftertheendoftheﬁscalyear.

dominatedbyassetaccounts,whilevariationinﬁnancingaccrualstendstobedominatedbyliabilityaccounts.

ThecorrelationsinpanelBofTable4shedmorelightontherelationbet-weenoperatingassetsandﬁnancingliabilities.Ininterpretingtheseaccruals,itisimportanttorememberthattheliabilitycomponentofaccrualsissub-tractedfromtheassetcomponentofaccrualstoarriveatnetaccruals.Thus,apositivecorrelationbetweenanassetandaliabilitycategoryindicatesthattheytendtohaveoffsettingeffectsontotalaccruals.BothDCOAandDNCOAarestronglypositivelycorrelatedwithDINL,indicatingthatﬁnancingliabilitiesprovideanimportantsourceoffundingforgrowthinoperatingassets.DCOAandDNCOAarealsostronglypositivelycorrelatedwithDCOLandDNCOL,indicatingthatoperatingliabilitiesprovideanadditionalsourceoffundingforgrowthinoperatingassets.AnothernoteworthyresultisthatbothDCOLandDNCOLarepositivelycorrelatedwithTACC.ThedirecteffectofincreasesinDCOLandDNCOListoreduceTACC,soonemighthaveexpectedanegativecorrelationbetweenthesetwoliabilitycomponentsandTACC.Thereis,however,astraightforwardexplanationfortheobservedpositivecorrelations.Asmentionedabove,theassetandliabilitycomponentsofoperatingaccrualsarestronglypositivelycorrelated.Forexample,thePearson(Spearman)correlationbetweenDCOAandDCOLis0.583(0.560).Firmsthataregrowingtheiroperationstendtoincreaseboththeiroperatingassetsandtheiroperatingliabilities.However,ﬁrmstypicallyrequirepositiveworkingcapital,sotheincreaseinoperatingassetstendstobelargerthantheincreaseinoperatingliabilities.Firmgrowththere-foreresultsinanindirectpositiverelationbetweenDCOLandTACC.Wewillseeinthenextsectionthatthestrongcorrelationsbetweentheoperatingassetandoperatingliabilitycomponentsofaccrualscomplicatetheinterpretationofourregressionresults.

4.2.Earningspersistenceresults

Table5presentsouranalysisofthepersistenceofthecashﬂowandaccrualcomponentsofearnings.Section2.1generatestwokeypredictionsconcerningthese

S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485

457

persistencecoefﬁcients.First,theaccrualcomponentofearningsisexpectedtobelesspersistentthanthecashﬂowcomponentofearnings.Thecashcomponentofearningsperformanceissimplythedifferencebetweenearningsperformanceitselfandtheaccrualcomponentofearningsperformance.Accordingly,ourﬁrstpredictioncanbetestedthroughthefollowingregression:

ROAtþ1¼g0þg1ðROAtÀTACCtÞþg2TACCtþutþ1,(13)whereg1measuresthepersistenceofcashﬂowsandg2measuresthepersistenceofaccruals,andourﬁrstpredictionisðg22g1Þo0.Inordertofocusontherelativepersistenceofaccruals,weestimateamodiﬁedversionof(13)inwhichthecashﬂowcomponentofearningsperformanceisreplacedbyearningsperformanceitself:

ROAtþ1¼r0þr1ROAtþr2TACCtþutþ1.

(14)

NotethatthisregressionequationcanberewrittenintermsofthepersistenceparametersinEq.(13)as:

ROAtþ1¼g0þg1ðROAtÞþðg2Àg1ÞTACCtþutþ1.

(15)

Sor1¼g1andr2¼ðg22g1Þ.Theadvantageofestimatingthismodiﬁedversionisthatr2providesadirectestimateof(g22g1),soourﬁrstpredictionissimplythatr2o0.

Oursecondpredictionisthat(g22g1)willbemorenegativeforlessreliableaccruals.RecallthatTable2summarizesourassessmentsofthereliabilityofthevariousaccrualcategories.Inordertotestoursecondprediction,wemustexamineeachoftheaccrualcategoriesindividually.Wedothisthroughacombinationof‘‘univariate’’and‘‘multivariate’’regressionanalyses.Our‘‘univariate’’regressionanalysisfollowsfromEq.(14).Forexample,toexaminethepersistenceoftheworkingcapitalcomponentofaccruals,weestimatethefollowingregression:

ROAtþ1¼r0þr1ROAtþr2DWCtþutþ1.TheanalogtoEqs.(13)and(15)arenow:

ROAtþ1¼g0þg1ðROAtÀDWCtÞþg2DWCtþutþ1,(17)ROAtþ1¼g0þg1ðROAtÞþðg2Àg1ÞDWCtþutþ1.

(18)(16)

Sor1measuresthepersistenceofearningsexclusiveoftheworkingcapitalcomponentofaccruals,andr2measuresthedifferencebetweenthepersistenceoftheworkingcapitalcomponentofaccrualsandthepersistenceofearningsexclusiveoftheworkingcapitalcomponentofaccruals.NotethattheinterpretationofthecoefﬁcientsissubtlydifferentthaninEq.(13),becauser1nolongerrepresentsthepersistenceofcashﬂows,butinsteadrepresentsthepersistenceofearningsexclusiveoftheparticularaccrualcomponentthatisincludedintheregression(DWCinthiscase).Sincethisaccrualcomponentisasubsetoftotalaccruals,r1nowrepresentsthecombinedpersistenceofcashﬂowsandalloftheotheraccrualcomponents.Oursecondpredictionisthatr2willbemorenegativeforlessreliableaccruals.Itisalsopossiblethatr2maybepositiveformorereliableaccruals,thereasonbeingthatr1representsthecombinedpersistenceofcashﬂowsandtheother(potentiallyless

Table4

Descriptivestatisticsandcorrelationsforextendedaccrualdecomposition.Thesampleconsistsof108,617ﬁrm-yearobservationsfrom1962to2001PanelA:Descriptivestatistics

Mean

Std.Dev.25%Median75%TACCt0.0520.181À0.0070.0390.098DCOAt0.0470.133À0.0060.0310.094DCOLt0.0250.084À0.0080.0170.052DNCOAt0.0570.153À0.0060.0290.089DNCOLt0.0060.0470.0000.0020.011DFINAt0.0090.107À0.00300.007DFINLt0.0300.151À0.0210.0010.066DSTIt0.0060.096000DLTIt0.0030.046000ROAt0.0700.1750.0320.0930.151ROAt+10.0630.1770.0250.0890.146FROAt+20.0780.1300.0400.0910.140RETt+10.010

0.715

À0.330

À0.078

0.199

PanelB:Correlationmatrix—Pearson(abovediagonal)andSpearman(belowdiagonal)(p-valuesshowninitalicsbelowcorrelations)

TACCtDCOAtDCOLtDNCOAtDNCOLtDFINAtDFINLtROAtROAt+1FROAt+2RETt+1TACCt—0.3760.0900.401À0.0450.481À0.1750.2500.1250.072À0.058(0.0001)(0.0001)(0.0001)(0.0001)(0.0001)(0.0001)(0.0001)(0.0001)(0.0001)(0.0001)DCOAt0.441—

0.5830.2950.084À0.0090.3680.2220.1270.068À0.055(0.0001)

(0.0001)

458S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485DCOLt0.1730.560—0.3040.0580.0820.2070.0610.0650.033À0.028(0.0001)(0.0001)(0.0001)(0.0001)(0.0001)(0.0001)(0.0001)(0.0001)(0.0001)(0.0001)DNCOAt0.4090.3620.319—0.212À0.0200.5450.0900.0290.024À0.063(0.0001)(0.0001)(0.0001)(0.0001)(0.0001)(0.0001)(0.0001)(0.0001)(0.0001)(0.0001)DNCOLt0.1020.1280.0940.296—0.0450.0290.0200.0170.012À0.006(0.0001)(0.0001)(0.0001)(0.0001)(0.0001)(0.0001)(0.0001)(0.0001)(0.0016)(0.0544)DFINAt0.261À0.0330.068À0.0390.054—0.0090.0500.0310.003À0.022(0.0001)(0.0001)(0.0001)(0.0001)(0.0001)(0.0032)(0.0001)(0.0001)(0.3752)(0.0001)DFINLtÀ0.0560.3720.1740.5040.104À0.001—À0.008À0.018À0.012À0.040(0.0001)(0.0001)(0.0001)(0.0001)(0.0001)(0.8206)(0.0069)(0.0001)(0.0010)(0.0001)ROAt0.4370.3250.1830.2390.1860.1000.004—0.7920.613À0.003(0.0001)(0.0001)(0.0336)(0.0001)(0.0001)(0.0001)(0.1891)(0.0001)(0.0000)(0.3098)ROAt+10.2740.2050.1490.1230.1370.070À0.0380.776—0.6960.138(0.0001)(0.0001)(0.0001)(0.0001)(0.0001)(0.0001)(0.0001)(0.0001)(0.0001)(0.0001)FROAt+20.1640.1120.0820.0820.0900.044À0.0150.5660.668—0.126(0.0001)(0.0001)(0.0001)(0.0001)(0.0001)(0.0001)(0.0001)(0.0001)(0.0001)(0.0001)RETt+1À0.049À0.070À0.034À0.0590.0320.001À0.0570.0870.3010.272—

(0.0001)

(0.0001

(0.0001)

(0.7008)

(0.0001)

TACCistotalaccrualsfromthebalancesheetapproach.ItiscalculatedasDWorkingCapital(DWC)+DNon-CurrentOperating(DNCO)+DFinancial(DFIN).Thiscanbeequivalentlywrittenas(DCOA–DCOL)+(DNCOA–DNCOL)+(DFINA–DFINL).Totalaccrualsandallofitscomponents(describedbelow)aredeﬂatedbyaveragetotalassets.

DCOAischangeincurrentoperatingassetsdeﬁnedasCOAt–COAtÀ1.COA¼CurrentAssets(CompustatItem#4)ÀCashandShortTermInvestments(STI)(CompustatItem#1).

DCOLischangeincurrentoperatingliabilitiesdeﬁnedasCOLtÀCOLtÀ1.COL¼CurrentLiabilities(CompustatItem#5)ÀDebtinCurrentLiabilities(CompustatItem#34).

DNCOAischangeinnon-currentoperatingassetsdeﬁnedasNCOAt–NCOAtÀ1.NCOA¼TotalAssets(Compustatitem#6)ÀCurrentAssets(CompustatItem#4)–InvestmentsandAdvances(CompustatItem#32).

DNCOLischangeinnon-currentoperatingliabilitiesdeﬁnedasNCOLtÀNCOLtÀ1.NCOL¼TotalLiabilities(CompustatItem#181)ÀCurrentLiabilities(CompustatItem#5)ÀLong-termdebt(CompustatItem#9).

S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485459Table4(continued)

DFINAischangeinﬁnancialassetsdeﬁnedasFINAtÀFINAtÀ1.FINAt¼ShortTermInvestments(STI)(CompustatItem#193)+LongTermInvestments(LTI)(CompustatItem#32).

DFINLischangeinﬁnancialliabilitiesdeﬁnedasFINLtÀFINLtÀ1.FINL¼Longtermdebt(CompustatItem#9)+DebtinCurrentLiabilities(CompustatItem#34)+PreferredStock(CompustatItem#130).DSTIischangeinshortterminvestments.DLTIischangeinlong-terminvestments.

ROAisoperatingincomeafterdepreciation(CompustatItem#178)deﬂatedbyaveragetotalassets.

FROAistheaverageROAoverthenextfouryears.Forexample,FROAt+2istheaverageROAforyearst+2throught+5.Thereare72,475observationsforFROA.

460S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485Table5

Time-seriesmeansandt-statisticsforcoefﬁcientsfromannualcross-sectionalregressionsofnextyear’saccountingrateofreturnonthisyear’saccountingrateofreturnandthisyear’saccruals.Thesampleconsistsof108,617ﬁrm-yearobservationsfrom1962to2001PanelA:OLSregressionsfortotalaccrualsROAt+1¼r0+r1ROAt+r2TACCt+ut+1Intercept

ROATACC

Adj.R2Meancoefﬁcient0.0120.7660.582t-statistic

Meancoefﬁcient0.014

0.796

À0.0820.588

t-statistic

(À16.13)PanelB:OLSregressionsfortheinitialaccrualdecompositionROAt+1¼r0+r1ROAt+r2DWCt+r3DNCOt+r4DFINt+ut+1Intercept

ROA

DWCDNCODFINAdj.R2PredictedreliabilityMediumLow/Medium

High

Meancoef.0.0130.784À0.1160.589t-statistic(À15.05)

Meancoef.0.0140.772À0.0470.584t-statistic(À12.39)

Meancoef.0.0130.7640.0150.583t-statistic(2.86)Meancoef.0.015

0.802

À0.138À0.074À0.0520.593t-statistic

(À19.05)

(À11.18)

(À8.71)

S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485461Table5(continued)

PanelC:OLSregressionsfortheextendedaccrualdecomposition

ROAt+1¼r0+r1ROAt+r2DCOAt–r3DCOLt+r4DNCOAt–r5DNCOLt+r6DSTIt+r7DLTItÀr8DFINLt+ut+1Intercept

ROA

DCOAÀDCOLDNCOAÀDNCOLDSTIDLTIÀDFINLAdj.R2PredictedreliabilityLowHigh

Low

Medium

High

Medium

High

Meancoef.0.0140.780À0.0640.586t-statistic(À12.08)

Meancoef.0.0120.764À0.0350.583t-statistic(À4.74)

Meancoef.0.0140.772À0.0470.584t-statistic(À12.63)

Meancoef.0.0120.7660.0120.582t-statistic(0.96)

Meancoef.0.0120.7650.0360.582t-statistic(1.05)

Meancoef.0.0130.766À0.0450.583t-statistic(À4.11)

Meancoef.0.0130.7650.0230.584t-statistic(3.83)Meancoef.0.014

0.801

À0.137À0.178À0.084À0.0750.004À0.084À0.0560.595t-statistic

(À18.81)

(À18.13)

(À13.03)

(À6.48)

(0.10)

(À6.33)

(À8.17)

Thesampleconsistsof108,617ﬁrmyearsfrom1962to2001.Regressionresultsarebasedon40annualregressionusingtheFamaandMacbeth(1973)approach.

TACCistotalaccrualsfromthebalancesheetapproach.ItiscalculatedasDWorkingCapital(DWC)+DNon-CurrentOperating(DNCO)+DFinancial(DFIN).Thiscanbeequivalentlywrittenas(DCOAÀDCOL)+(DNCOAÀDNCOL)+(DFINAÀDFINL).Totalaccrualsandallofitscomponents(describedbelow)aredeﬂatedbyaveragetotalassets.

DWCisdeﬁnedasWCt–WCtÀ1.WC¼CurrentOperatingAssets(COA)ÀCurrentOperatingLiabilities(COL)whereCOA¼CurrentAssets(CompustatItem#4)ÀCashandShortTermInvestments(STI)(CompustatItem#1).COL¼CurrentLiabilities(CompustatItem#5)ÀDebtinCurrentLiabilities(CompustatItem#34).

DCOAischangeincurrentoperatingassetsdeﬁnedasCOAtÀCOAtÀ1.

462S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485Table5(continued)

DCOLischangeincurrentoperatingliabilitiesdeﬁnedasCOLtÀCOLtÀ1.

DNCOisdeﬁnedasNCOt–NCOtÀ1.NCO¼Non-CurrentOperatingAssets(NCOA)ÀNon-CurrentOperatingLiabilities(NCOL)whereNCOA¼TotalAssets(Compustatitem#6)ÀCurrentAssets(CompustatItem#4)ÀInvestmentsandAdvances(CompustatItem#32).NCOL¼TotalLiabilities(CompustatItem#181)ÀCurrentLiabilities(CompustatItem#5)–Long-termdebt(CompustatItem#9).DNCOAischangeinnon-currentoperatingassetsdeﬁnedasNCOAtÀNCOAtÀ1.DNCOLischangeinnon-currentoperatingliabilitiesdeﬁnedasNCOLtÀNCOLtÀ1.

DFINAischangeinﬁnancialassetsdeﬁnedasFINAt–FINAtÀ1.DFINLischangeinﬁnancialliabilitiesdeﬁnedasFINLt–FINLtÀ1.DSTIischangeinshortterminvestments.DLTIischangeinlong-terminvestments.

ROAisoperatingincomeafterdepreciation(CompustatItem#178)deﬂatedbyaveragetotalassets.

S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485463464

S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485

reliable)accruals.15Thepresenceofthelessreliableaccrualsmaydriveg1og2,inwhichcaser240.Nevertheless,r2isalwayspredictedtobelowerforlessreliableaccruals.

Theunivariateregressionshavetwoshortcomings.First,theydonotallowforformalstatisticaltestsofdifferencesinther2estimatesacrossaccrualcomponents.Second,asdescribedabove,theydonotallowustodirectlybenchmarkthepersistenceoftheaccrualcomponentsrelativetocashﬂows.Toaddresstheseshortcomings,wealsoestimatea‘‘multivariate’’versionofEq.(16)thatsimul-taneouslyincludesallcomponentsofaccruals:

ROAtþ1¼r0þr1ROAtþr2DWCtþr3DNCOtþr4DFINtþutþ1.TheanalogtoEqs.(17)and(18)arenow:

ROAtþ1¼g0þg1ðROAtÀDWCtÀDNCOtÀDFINtÞ

þg2DWCtþg3DNCOtþg4DFINtþutþ1,ROAtþ1¼g0þg1ðROAtÞþðg2Àg1ÞDWCt

þðg3Àg1ÞDNCOtþðg4Àg1ÞDFINtþutþ1.

ð21Þ

Sor1nowmeasuresthepersistenceofthecashﬂowcomponentofearnings,whiler2,r3andr4eachmeasurethepersistenceoftheirrespectiveaccrualcomponentrelativetothecashﬂowcomponent.F-testsallowforformalstatisticaltestsoftheequalityofthecoefﬁcientsacrosstheaccrualcomponents.Thekeyshortcomingofthismultivariateregressionisthattheerrors-in-variablesmodeldevelopedinSection2.1onlyappliestothecaseoftwovariables.Wehavenowdecomposedaccrualsintomultiplevariables,allofwhicharepotentiallymeasuredwitherror.Inthecaseofmultiplevariablesmeasuredwitherror,themagnitudeoftheresultingbiaseswillalsodependonthecorrelationsbetweenthesevariables(seeKlepperandLeamer,1984).RecallfrompanelBofTable4thatseveraloftheaccrualcomponentsarehighlycorrelated.Thus,caremustbetakenininterpretingthepersistencecoefﬁcientsinthemultivariatespeciﬁcation.

WeconductallofourregressionanalysesfollowingtheFamaandMacbeth(1973)procedureofestimatingannualcross-sectionalregressionsandreportingthetime-seriesaveragesoftheresultingregressioncoefﬁcients.PanelAofTable5reportsregressionresultsforTACC,thetotalaccrualcomponentofearnings.Theﬁrstregressionmodelisasimpleearningsautoregressionandprovidesabenchmarkforthesubsequentregressions.ConsistentwithSloan(1996),earningsisslowlymeanrevertingwithapersistencecoefﬁcientofabout0.77.Thesecondregressionesti-matesthemodelinEq.(14).Consistentwithourﬁrstprediction,r2isnegative(À0.082)andstatisticallysigniﬁcant(t¼À16:13),indicatingthattheaccrualcomponentofearningsislesspersistentthanthecashﬂowcomponentofearnings.

RecallfromSection2.2thatwhenbothcomponentsofearningsaremeasuredwitherror,thedownwardbiasintheestimatedcoefﬁcientstendstobegreatestforthevariablethatismeasuredwiththemosterror.

15(19)

ð20Þ

S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485

465

ThisresultcorroboratesSloan’soriginalresultusingourcomprehensivedeﬁnitionofaccruals.16PanelBofTable5reportsresultsfromtheestimationofEq.(16)forthecomponentsofourinitialaccrualdecomposition.RecallfromTable2thatDWCandDNCOareassessedtohavemediumtolowreliabilityandsoarepredictedtohaverelativelylowpersistencecoefﬁcients,whileDFINisassessedtohavehighreliabilityandhencearelativelyhighpersistencecoefﬁcient.Thecoefﬁcientsineachoftheunivariateregressionsareconsistentwiththesepredictions.ThecoefﬁcientonDWCisnegativeandhighlystatisticallysigniﬁcant(À0.116withat-statisticofÀ15.05),thecoefﬁcientonDNCOisalsonegativeandstatisticallysigniﬁcant(À0.047withat-statisticofÀ12.39)andthecoefﬁcientonDFINisclosetozero(0.015withat-statisticof2.86).

TheﬁnalrowofpanelBofTable5reportsresultsforthemultivariatespeciﬁcationinEq.(19).Consistentwithourpredictions,themultivariateregressionanalysisbasicallymaintainsthesamerelativerankingofaccrualpersistencecoefﬁcientsastheunivariateregressions.UntabulatedF-testsconﬁrmthatthecoefﬁcientsonDWCandDNCOaresigniﬁcantlydifferentfromthecoefﬁcientonDFIN.Themultivariateregressions,however,differfromtheunivariateregressionsinacoupleofsigniﬁcantrespects.First,thecoefﬁcientsonalloftheaccrualcomponentsaresomewhatlowerthanintheunivariatespeciﬁcations.Second,thecoefﬁcientonDFINisnegativeandstatisticallysigniﬁcant.Asdescribedearlier,thereisanaturalexplanationforthesedifferences.Intheunivariateregressions,theomittedaccrualcomponentsareimplicitlygroupedwithcashﬂowsandthelowerpersistenceoftheseaccrualcomponentscausesr1tobelower.Themultivariatespeciﬁcationallowsr1torepresentthecoefﬁcientonthepurecashﬂowcomponentofearnings,causingr1toincreaseandthecoefﬁcientsoneachofthelessreliableaccrualcomponentstocorrespondinglydecrease.

PanelCofTable5presentsresultsfortheextendedaccrualdecomposition.RecallfromTable2thattheassetcomponentsoftheoperatingaccrualsareassessedtobetheleastreliable,sowepredictthatthecoefﬁcientsontheoperatingassetcomponentsofaccrualswillbethemostnegative.Theresultsfromtheunivariateregressionsconﬁrmthesepredictions.ThepersistencecoefﬁcientsonDCOAandDNCOAarethetwomostnegativeandthemoststatisticallysigniﬁcantcoefﬁcients.However,inconsistentwithourassessmentthatDCOLismorereliablethanDNCOL,weﬁndthatDCOLhasthemorenegativecoefﬁcient.OnepossibleexplanationforthisresultisthattheaccrualsinDNCOLconsistofaccrualsrelating

Fairﬁeldetal.(2003b)showthatthelowerpersistenceofaccrualsrelativetocashﬂowsisprimarilyattributabletogrowthintheinvestmentbasethatisnotmatchedbygrowthinincome.Inotherwords,ﬁrmswithhighoperatingaccrualsareﬁrmsthataregrowingtheiroperatinginvestments,butearnaloweraveragereturnontheirincreasedoperatinginvestments,leadingtolowerpersistenceinearningsperformance.Toinvestigatetheimportanceofthisissueforourresults,wereplicatedalloftheresultsinTable5usingtheFairﬁeld,WhisenantandYohnapproachofmeasuringourdependentvariableasnextyear’sincomedeﬂatedbylastyear’saveragetotalassets.Ourresultsaresurprisinglyrobustwithrespecttothisprocedure.Forexample,ourtotalaccrualsvariablecontinuestohaveasigniﬁcantlynegativecoefﬁcientandtheoperatingassetaccrualsvariablescontinuetohavethemostnegativecoefﬁcients.

16466

S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485

todeferredtaxesandretirementobligationsthatcantakemanyperiodstoreverse.Thispossibilityisexaminedinmoredetailinthenextsubsection.

Ourextendeddecompositionofﬁnancingaccrualsisathree-waydecompositionintoshort-terminvestingaccruals(DSTI),long-terminvestingaccruals(DLTI)andﬁnancialliabilityaccruals(DFINL).RecallfromTable2thatweassessbothDSTIandDFINLtobeofhighreliabilityandhencehavesimilarpersistencetocashﬂows.Incontrast,weassessDLTItohaveonlymediumreliabilityandhencehaveasigniﬁcantlylowercoefﬁcientthancashﬂows.Theresultsaregenerallyconsistentwiththesepredictions.ThecoefﬁcientonDLTIissigniﬁcantlynegative(À0.045,t¼À4:11)andthelowestofthecoefﬁcientsonthethreeﬁnancingcomponents.Incontrast,thecoefﬁcientonDSTIismuchclosertozero(0.036,t¼1:05),whilethecoefﬁcientonDFINLissigniﬁcantlypositive(0.023,t¼3:83).ThesigniﬁcantlypositivecoefﬁcientonDFINLmayseemsomewhatsurprising,butrememberthatintheunivariatespeciﬁcation,r1representsthecombinedpersistenceofallcomponentsofearningsotherthantheincludedaccrualcomponent.Thesecomponentsincludetherelativelyunreliableoperatingassetaccruals,whichwillcauser1tobelowerandthecoefﬁcientonDFINLtobehigher.

TheﬁnalregressioninpanelCofTable5isthemultivariateregressionfortheextendedaccrualdecomposition.Sincethisregressionincludesallaccrualcomponents,r1representsthepersistenceofthepurecashﬂowcomponentofearnings.Consequently,thecoefﬁcientof0.801ishigherthaninanyoftheprecedingunivariateregressions.Thecoefﬁcientsinthemultivariateregressionconﬁrmseveralofourkeypredictions.ThecoefﬁcientsonDCOA,DNCOAandDLTIcontinuetobesigniﬁcantlynegative,consistentwiththeirlowtomediumreliabilityassessments.Also,thehighestcoefﬁcientisreportedonDSTI,anaccrualcomponentweassesstohavehighreliability.However,bothDCOL,DNCOLhaverelativelyhighpersistencecoefﬁcientsintheunivariateregressions,buthavemuchlowerpersistencecoefﬁcientsinthemultivariateregression.WeprovideanexplanationforthediscrepancyinthesepersistencecoefﬁcientsbetweentheunivariateandmultivariateregressionsinSection4.3.

4.3.Additionaltestsonearningspersistence

TheearningspersistenceresultsinSection4.2raisetwoissuesthatareworthyofadditionalexamination.First,itispossiblethatthepersistencecoefﬁcientsonnon-currentoperatingaccrualsaresensitivetothehorizonoverwhichearningspersistenceismeasured.Second,itappearsthatthereareimportantinteractionsbetweenoperatingassetandoperatingliabilityaccrualsthatcausedramaticshiftsinthepersistencecoefﬁcientsontheoperatingliabilityaccrualsbetweentheunivariateandmultivariateregressionspeciﬁcations.Thissubsectionaddresseseachoftheseissuesinturn.

TheregressionsinTable5examineearningspersistenceoverconsecutiveannualintervals.Itispossiblethatsomeaccrualerrorstakeseveralyearstoreverse,particularlyinthecaseofaccrualsrelatingtonon-currentassetsandliabilities.Toinvestigatethesensitivityoftheresultswithrespecttolonger-termpersistence,

S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485

467

Table6replicatestheregressionanalysisinTable5measuringthedependentvariableasearningsperformanceaveragedfrom2through5yearsahead(denotedFROA).Asexpected,theuseoflong-termearningsleadstoareductioninthepersistencecoefﬁcientsonROA.Thereismodestevidencethatthepersistencecoefﬁcientsonthelong-termaccrualsfallrelativetothecurrentaccruals.Forexample,intheunivariateregressionsinpanelC,thecoefﬁcientonDNCOLswitchesfrombeinglargerthanthecoefﬁcientonDCOLinTable5tobeingsmallerthanthecoefﬁcientonDCOLinTable6.ThisisconsistentwithlowerreliabilityofDNCOLmanifestingitselfoverlongerhorizonsduetothelong-termnatureoftheunderlyingaccruals.

TheresultsinTables5and6provideconsistentevidenceconcerningthelowerearningspersistenceassociatedwithoperatingassetaccruals.Theresultsforoperatingliabilityaccruals,however,aremixed.Inparticular,recallfrompanelCofTable5thatbothDCOLandDNCOLhaverelativelyhighpersistencecoefﬁcientsintheunivariateregressions,butrelativelylowpersistencecoefﬁcientsinthemultivariateregressions.Fromastatisticalperspective,theseresultsarisebecauseofthestrongcorrelationsbetweentheoperatingliabilityaccrualsandtheotheraccrualcomponents.Inparticular,panelBofTable4indicatesthatbothDCOLandDNCOLarehighlypositivelycorrelatedwithbothDCOAandDNCOA.Table7conductsadditionalteststhatprovideinsightsintotheeconomicrationalebehindthesestatistics.

Theeconomicreasoningbehindthepositivecorrelationbetweenoperatingliabilityaccrualsandoperatingassetsaccrualsisstraightforward.Growthinthescaleofbusinessoperationsresultsingrowthinbothoperatingassetsandoperatingliabilities.Forexample,aretailingbusinessthatopensnewlocationswilltypicallyexperienceincreasesinbothinventoriesandaccountspayable.Thepositivecorrelationsbetweenoperatingassetsandoperatingliabilitiesleadtocorrelatedomittedvariablesbiasesintheunivariateregressions.Thesebiasesarenotpresentinthemultivariateregressions.Assuch,onlythepersistencecoefﬁcientsinthemultivariateregressionsrepresentthemarginalpersistenceofspeciﬁcaccrualcomponents(i.e.,afterholdingallotheraccrualcomponentsconstant).Inparticular,thenegativepersistencecoefﬁcientsonoperatingliabilityaccrualsindicatethatashifttowardtheuseofoperatingliabilitiestoﬁnanceoperatingassetsresultsinhigherearningspersistence.17TheresultsinpanelAofTable7provideadditionalinsightsintothisresult.

PanelAofTable7examinesthepersistenceofoperatingassetaccrualsaccordingtothemannerinwhichtheaccrualsareﬁnanced.Usingthebalancesheetidentityandourearlierterminology,wecandecomposeoperatingassetaccruals(DOA)intoitsﬁnancingsourcesasfollows:

DOA¼DOLþDFINLþDEQUITYÀDFA;

17(22)

Changesinliabilitiesentertheregressionswithanegativesign,becauseincreases(decreases)inliabilitiesrepresentnegative(positive)accruals.Thatiswhythenegativecoefﬁcientsonoperatingliabilityaccrualsindicatethatincreasesinoperatingliabilitiesleadtoincreasesinearningspersistence.

Table6

Time-seriesmeansandt-statisticsforcoefﬁcientsfromannualcross-sectionalregressionsoftheaveragefutureaccountingrateofreturn(FROA)onthisyear’saccountingrateofreturnandthisyear’saccruals.Thesampleconsistsof72,475ﬁrm-yearobservationsfrom1962to2001PanelA:OLSregressionsfortotalaccrualsFROAt+2¼r0+r1ROAt+r2TACC,t+ut+2Intercept

ROATACC

Adj.R2Meancoefﬁcient0.0380.4940.342t-statistic

Meancoefﬁcient0.039

0.539

À0.1130.358

t-statistic

(À14.63)PanelB:OLSregressionsfortheinitialaccrualdecompositionFROAt+2¼r0+r1ROAt+r2DWCt+r3DNCOt+r4DFINt+ut+2Intercept

ROA

DWCDNCODFINAdj.R2PredictedreliabilityMediumLow/Medium

High

Meancoef.0.0390.510À0.0930.350t-statistic(À11.02)

Meancoef.0.0400.500À0.0500.346t-statistic(À7.83)

Meancoef.0.0380.494À0.0060.343t-statistic(À1.07)Meancoef.0.040

0.541

À0.139À0.106À0.0930.361t-statistic

(À15.17)

(À11.55)

(À12.98)

468S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485PanelC:OLSregressionsfortheextendedaccrualdecomposition

FROAt+2¼r0+r1ROAt+r2DCOAt–r3DCOLt+r4DNCOAt–r5DNCOLt+r6DSTIt+r7DLTIt–r8DFINLt+u

t+2Intercept

ROA

DCOAÀDCOLDNCOAÀDNCOLDSTIDLTIÀDFINLAdj.R2PredictedreliabilityLowHigh

Low

Medium

High

Medium

High

Meancoef.0.0400.516À0.0850.352t-statistic(À13.48)

Meancoef.0.0390.5010.0570.345t-statistic(5.22)

Meancoef.0.0400.501À0.0490.346t-statistic(À7.72)

Meancoef.0.0380.5000.0290.343t-statistic(1.65)

Meancoef.0.0380.4950.0150.344t-statistic(0.26)

Meancoef.0.0380.494À0.0700.343t-statistic(À4.48)

Meancoef.0.0390.4940.0130.343t-statistic(2.43)Meancoef.0.040

0.547

À0.146À0.093À0.103À0.088À0.018À0.131À0.0980.366

t-statistic

(À16.44)

(À7.00)

(À11.78)

(À5.11)

(À0.28)

(À8.31)

(À13.89)

Thesampleconsistsof72,475ﬁrmyearsfrom1962to2001.Regressionresultsarebasedon40annualregressionusingtheFamaandMacbeth(1973)approach.

DCOAischangeincurrentoperatingassetsdeﬁnedasCOAtÀCOAtÀ1.DCOLischangeincurrentoperatingliabilitiesdeﬁnedasCOLtÀCOLtÀ1.

S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485469Table6(continued)

DNCOisdeﬁnedasNCOt–NCOtÀ1.NCO¼Non-CurrentOperatingAssets(NCOA)ÀNon-CurrentOperatingLiabilities(NCOL)whereNCOA¼TotalAssets(Compustatitem#6)ÀCurrentAssets(CompustatItem#4)ÀInvestmentsandAdvances(CompustatItem#32).NCOL¼TotalLiabilities(CompustatItem#181)ÀCurrentLiabilities(CompustatItem#5)–Long-termdebt(CompustatItem#9).DNCOAischangeinnon-currentoperatingassetsdeﬁnedasNCOAt–NCOAtÀ1.DNCOLischangeinnon-currentoperatingliabilitiesdeﬁnedasNCOLt–NCOLtÀ1.

DFINAischangeinﬁnancialassetsdeﬁnedasFINAtÀFINAtÀ1.DFINLischangeinﬁnancialliabilitiesdeﬁnedasFINLtÀFINLtÀ1.DSTIischangeinshortterminvestments.DLTIischangeinlong-terminvestments.

ROAisoperatingincomeafterdepreciation(CompustatItem#178)deﬂatedbyaveragetotalassets.

FROAistheaverageROAoverthenextfouryears.Forexample,FROAt+2istheaverageROAforyearst+2throught+5.

470S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485Table7

Additionaltestsinvestigatinghowthechangeinoperatingassets(DOA)andthechangeinoperatingliabilities(DOL)relatetonextyear’saccountingrateofreturn.Thesampleconsistsof108,617ﬁrm-yearobservationsfrom1962to2001

PanelA:Meancoefﬁcientsandt-statisticsfromannualcross-sectionalregressionsanalyzingaﬁnancingdecompositionofDOA.ROAt+1¼r0+r1ROAt+r2DOAt+ut+1ROAt+1¼r0+r1ROAt+r2DOLt+r3DFINLt–r4DFAt+r5DEquityt+ut+1Intercept

ROADOADOL

DFINL

ÀDFA

DEquity

Adj.R2Meancoef.0.0150.782À0.0440.586t-statistic(À14.93)

Meancoef.0.013

0.817

0.047À0.039À0.049À0.1060.595

t-statistic

(6.61)

(À7.84)

(À8.73)

(À12.22)

PanelB:Meanfuturechangeintheaccountingrateofreturn(DROAt+1)partitionedbyindependentquintilesortsonDOAandDOL(Numberofobservationsinparentheses).DOA

DOLLow

234HighTotalavg.Low0.0100.0150.0230.0320.0650.018(10,991)(4809)(2709)(1837)(1364)(21,710)2À0.011À0.0050.0000.0050.027À0.001(4763)(6997)(5326)(3179)(1465)(21,730)3À0.019À0.011À0.007À0.0030.008À0.007(2807)(5132)(6337)(5142)(2313)(21,731)4

À0.037À0.024À0.017À0.015À0.006À0.017(2009)

(3282)

(4927)

(6580)

(4931)

(21,729)

S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485471Table7(continued)

PanelB:Meanfuturechangeintheaccountingrateofreturn(DROAt+1)partitionedbyindependentquintilesortsonDOAandDOL(Numberofobservationsinparentheses).

DOADOLLow

234HighTotalavg.HighÀ0.040À0.035À0.030À0.031À0.026À0.029(1140)(1510)(2431)(4992)(11,644)(21,717)

TotalÀ0.006À0.007À0.006À0.009À0.009Average(21,710)

(21,730)

(21,717)

DOAisthechangeinoperatingassets,deﬁnedasDCOA+DNCOA.

DCOAischangeincurrentoperatingassetsdeﬁnedasCOAt–COAtÀ1.COA¼CurrentAssets(CompustatItem#4)ÀCashandShortTermInvestments(STI)(CompustatItem#1).COAisdeﬂatedbyaveragetotalassets.

DNCOAischangeinnon-currentoperatingassetsdeﬁnedasNCOAt–NCOAtÀ1.NCOA¼TotalAssets(CompustatItem#6)ÀCurrentAssets(CompustatItem#4)ÀInvestmentsandAdvances(CompustatItem#32).NCOAisdeﬂatedbyaveragetotalassets.DOListhechangeinoperatingliabilities,deﬁnedasDCOL+DNCOL.

DCOLischangeincurrentoperatingliabilitiesdeﬁnedasCOLtÀCOLtÀ1.COL¼CurrentLiabilities(CompustatItem#5)ÀDebtinCurrentLiabilities(CompustatItem#34).COLisdeﬂatedbyaveragetotalassets.

DNCOLischangeinnon-currentoperatingliabilitiesdeﬁnedasNCOLtÀNCOLtÀ1.NCOL¼TotalLiabilities(CompustatItem#181)ÀCurrentLiabilities(CompustatItem#5)–Long-termdebt(CompustatItem#9).NCOLisdeﬂatedbyaveragetotalassets.

DFINListhechangeinﬁnancialliabilities(inclusiveofpreferredstock)deﬁnedasFINLtÀFINLtÀ1.FINL¼Long-termdebt(CompustatItem#9)+DebtinCurrentLiabilities(CompustatItem#34)+PreferredStock(CompustatItem#130).FINLisdeﬂatedbyaveragetotalassets.

DFAisthechangeinﬁnancialassetsdeﬁnedasFAtÀFAtÀ1.FA¼Cashandcashequivalents(CompustatItem#1)+InvestmentsandAdvances(CompustatItem#32).FAisdeﬂatedbyaveragetotalassets.

DEquityisthechangeinexternalequityﬁnancingdeﬁnedasEQtÀEQtÀ1.EQ¼Bookvalueofcommonequity(CompustatItem#60).EQisdeﬂatedbyaveragetotalassets.

DROAt+1isthechangeintheaccountingrateofreturndeﬁnedasROAt+1–ROAt,whereROAisoperatingincomeafterdepreciation(CompustatItem#178)deﬂatedbyaveragetotalassetsinyeart+1.

472S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485

473

whereDOAisthechangeinoperatingassets(DCOA+DNCOA),DOLthechangeinoperatingliabilities(DCOL+DNCOL),DFINLthechangeindebt(includingpreferredstock),DEQUITYthechangeincommonequityandDFAthechangeinﬁnancialassets(includingcash).TheﬁrstregressioninTable7conﬁrmsourearlierresultsthatoperatingassetaccrualsareassociatedwithlowerearningspersistence.Thenextregressionindicatesthatthelowerearningspersistenceisentirelyattributabletooperatingassetaccrualsthatareﬁnancedbydebt,ﬁnancialassetsorequity.Whenoperatingliabilitiesareusedtoﬁnanceoperatingassetaccruals,earningspersistenceisactuallyhigher,asindicatedbythesigniﬁcantlypositivecoefﬁcientonDOL.Incontrast,whendebt,equityorﬁnancialassetsareusedtoﬁnanceoperatingassetaccruals,persistenceislower,asindicatedbythesigniﬁcantlynegativecoefﬁcientsonthesesourcesofﬁnancing.NissimandPenman(2003)providerelatedevidence.Theyﬁndthatleveragearisingfromoperatingliabilitiesisassociatedwithhigherfutureproﬁtabilitythanleveragearisingfromdebt.

Theeconomicintuitionbehindtheseresultscanbeillustratedthoughanexample.Considertwocompaniesthateachreporthighoperatingassetaccrualsintheformoflargeincreasesininventory.Inonecompany,theinventoryincreaseresultsfromslowingsales,andthecompanyisforcedtoissuenewdebtinordertoﬁnancetheinventoryglut(becauseinventorytradepayablesbecomeduebeforetheinventoryissold).Intheothercompany,theincreaserepresentsnewinventorythatisbeingacquiredinresponsetoasurgeindemand,andthecompanyisabletoﬁnancetheincreaseusingtradecredit(becausetheinventoryisbeingsoldbeforetheassociatedtradepayablebecomesdue).Theformercompanyisexperiencingproblemsthatarelikelytoresultinreductionsinfutureproﬁtability,whilethelattercompanyisexperiencinghealthygrowththatislikelytoleadtoincreasesinfutureproﬁtability.Asecondandcomplementaryeconomicforcethatisalsolikelytobeatworkisthattradecreditorsareinauniquepositiontocontinuouslyverifythequalityoftheoperatingassetsthatareavailabletosatisfytheirclaims.Unlikeothersourcesofﬁnancing,tradecredittendstobeshortterminnatureandgrantedbysupplierswhohaveuniqueinsightsintotheﬁnancialhealthofthecompany.Asaresult,operatingassetaccrualsthatareﬁnancedbyoperatingliabilitiesareexpectedtobemorepersistentthanoperatingassetaccrualsthatareﬁnancedbydebtorequity.

PanelBofTable7providesadditionalevidencesupportingtheaboveinterpretationoftheinteractionbetweenoperatingassetsandoperatingliabilities.Toconstructthistable,weconductindependentquintilesortsofoursampleonDOA(thesumofDCOAandDNCOA)andDOL(thesumofDCOLandDNCOL).Thenumbersinparenthesesrepresentthenumberofobservationsineachcell.ThepositivecorrelationbetweenDOAandDOLleadstheobservationstobeconcentrateddownthemaindiagonal.Thereare,however,manyobservationsintheothercells,providinguswiththeopportunitytoexaminehowDOAandDOLinteracttoinformaboutearningspersistence.

ThestatisticreportedineachcellofpanelBofTable7isthemeanchangeinnextperiod’searningsperformance(DROAt+1)forobservationsinthatcell.ScanningdowneachcolumnrevealsanegativeassociationbetweenfutureearningschangesandDOA,consistentwiththelowerpersistenceandreliabilityofoperatingasset

474

S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485

accruals.Butwealsoseethatthemaximumspreadinfutureproﬁtabilityisobservedbetweenthe‘lowDOA,highDOL’cellandthe‘highDOA,lowDOL’cell.The‘lowDOA,highDOL’cellhasameanfutureearningschangeof0.065,comparedtoanaveragechangeof0.018foralllowDOAcells.Similarly,the‘highDOA,lowDOL’cellhasafutureearningschangeofÀ0.040comparedtoanaveragechangeofÀ0.029forallhighDOAcells.FocusingoncellswhereDOAandDOLmoveinoppositedirectionsmaximizesthespreadinfutureearningschanges.18FirmswhereDOAisrisingandDOLisfallingsubsequentlydomostpoorly,whileﬁrmswhereDOAisfallingandDOLisrisingsubsequentlydothebest.Tosummarize,DOLprovidesimportantinformationaboutthepersistenceofDOA,becausehigheroperatingliabilityaccrualssignalmorereliableoperatingassetaccruals.

4.4.Stockreturnresults

Inthissection,weinvestigatewhetherstockpricesactasifinvestorsanticipatetheimplicationsofaccrualreliabilityforearningspersistence.Ifinvestorsunderstandtheimplicationsofaccrualreliabilityforearningspersistence,thenthereshouldbenorelationbetweenaccrualsandfutureabnormalstockreturns.However,ifinvestorsnaıvelyfailtoanticipatethelowerpersistenceofthelessreliableaccrualcomponentsofearnings,therewillbeanegativerelationbetweentheseaccrualcomponentsandfutureabnormalstockreturns.Forexample,aﬁrmwithrelativelyhighaccrualsthisperiodisexpectedtohaverelativelylowearningsperformancenextperiod,butifinvestorsnaıvelyignorethehighaccruals,theywillbesurprisedbynextperiod’slowearningsperformance,resultinginnegativeabnormalreturnsinthenextperiod.Thus,thenaıveinvestorhypothesispredictsanegativerelationbetweenaccrualsandfuturestockreturns,withtherelationbeingstrongerforlessreliableaccrualsthatresultinlowerearningspersistence.

OurstockreturntestsarepresentedinTable8.ThistablereplicatestheanalysisinTable5afterreplacingthedependentvariablewiththenextyear’sannualsize-adjustedbuy-holdstockreturn(measuredstartingfourmonthsaftertheﬁscalyearend).PanelAofTable8reportstheresultsofthebasicregressionofstockreturnsonearningsandtheaccrualcomponentofearnings.Asinpreviousresearch,weobtainanegativeandsigniﬁcantcoefﬁcientonaccruals,whichisconsistentwiththenaıveinvestorhypothesis.

PanelBreportsregressionresultsfortheinitialbalancesheetdecomposition.Consistentwiththenaıveinvestorhypothesis,therelativerankingsofthecoefﬁcientsontheaccrualcomponentsofearningslineupcloselywiththeircounterpartsinTable5.ThecoefﬁcientsonDWCandDNCOareconsistentlythemostnegativeandarehighlystatisticallysigniﬁcant.

Weconductedt-testsofdifferencesinmeansbetweeneachpossiblecombinationofthe4cornercellsinthe5Â5DOA/DOLmatrix,conﬁrmingthattheyareallsigniﬁcantlydifferentfromeachotheratconventionallevels.Forexample,inthehighDOArowthedifferencebetweenlowDOL(À0.040)andhighDOL(À0.026)isstatisticallysigniﬁcantatthe5%level(two-tailedtest).

18Table8

Time-seriesmeansandt-statisticsforcoefﬁcientsfromannualcross-sectionalregressionsofnextyear’ssizeadjustedstockreturnonthisyear’saccruals.Thesampleconsistsof108,617ﬁrm-yearobservationsfortheperiod1962–2001PanelA:OLSregressionsfortotalaccrualsRETt+1¼r0+r1ROAt+r2TACCt+ut+1Intercept

ROATACC

Adj.R2Meancoefﬁcient0.0070.0200.006

t-statistic

(0.86)(0.36)Meancoefﬁcient0.0130.083À0.1970.011

t-statistic

(1.52)

(1.47)

(À6.38)

PanelB:OLSregressionsfortheinitialaccrualdecompositionRETt+1¼r0+r1ROAt+r2DWCt+r3DNCOt+r4DFINt+ut+1Intercept

ROA

DWCDNCODFINAdj.R2PredictedreliabilityMediumLow/Medium

High

Meancoef.0.0100.069À0.3090.011t-statistic(1.19)(1.23)(À8.72)

Meancoef.0.0190.048À0.2680.012t-statistic(2.14)(0.86)(À8.76)

Meancoef.0.0120.0070.1230.010t-statistic(1.40)(0.12)(5.85)Meancoef.0.0220.094À0.300À0.271À0.0540.016t-statistic

(2.45)

(1.69)(À7.54)(À6.77)

(À1.94)

S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485475Table8(continued)

PanelC:OLSregressionsfortheextendedaccrualdecomposition

RETt+1¼r0+r1ROAt+r2DCOAt–r3DCOLt+r4DNCOAt–r5DNCOLt+r6DSTIt+r7DLTIt–r8DFINLt+ut+1Intercept

ROA

DCOAÀDCOLDNCOAÀDNCOLDSTIDLTIÀDFINLAdj.R2PredictedreliabilityLowHigh

Low

Medium

High

Medium

High

Meancoef.0.0160.085À0.3000.012t-statistic(1.74)(1.55)(À8.71)

Meancoef.0.0120.0310.2130.008t-statistic(1.29)(0.57)(4.49)

Meancoef.0.0210.049À0.2630.012t-statistic(2.30)(0.88)(À8.44)

Meancoef.0.0080.0210.0530.007t-statistic(0.88)(0.37)(0.65)

Meancoef.0.0080.0160.0540.007t-statistic(0.94)(0.29)(0.43)

Meancoef.0.0080.021À0.2040.007t-statistic(0.92)(0.38)(À3.38)

Meancoef.0.0140.0150.2060.010t-statistic(1.64)(0.26)(8.01)Meancoef.0.0240.106À0.309À0.200À0.250À0.2330.026À0.253À0.0270.018t-statistic

(2.51)

(1.93)(À7.70)(À4.10)(À6.42)(À2.98)(0.17)(À4.18)

(À0.90)

Thesampleconsistsof108,617ﬁrmyearsfrom1962to2001.Regressionresultsarebasedon40annualregressionusingtheFamaandMacbeth(1973)approach.

TACCistotalaccrualsfromthebalancesheetapproach.ItiscalculatedasDWorkingCapital(DWC)+DNon-CurrentOperating(DNCO)+DFinancial(DFIN).Thiscanbeequivalentlywrittenas(DCOAÀDCOL)+(DNCOAÀDNCOL)+(DFINAÀDFINL).Totalaccrualsandallofitscomponents(describedbelow)aredeﬂatedbyaveragetotalassets.

DCOAischangeincurrentoperatingassetsdeﬁnedasCOAtÀCOAtÀ1.

476S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485Table8(continued)

DCOLischangeincurrentoperatingliabilitiesdeﬁnedasCOLtÀCOLtÀ1.

DNCOisdeﬁnedasNCOtÀNCOtÀ1.NCO¼Non-CurrentOperatingAssets(NCOA)ÀNon-CurrentOperatingLiabilities(NCOL)whereNCOA¼TotalAssets(Compustatitem#6)ÀCurrentAssets(CompustatItem#4)ÀInvestmentsandAdvances(CompustatItem#32).NCOL¼TotalLiabilities(CompustatItem#181)ÀCurrentLiabilities(CompustatItem#5)–Long-termdebt(CompustatItem#9).DNCOAischangeinnon-currentoperatingassetsdeﬁnedasNCOAt–NCOAtÀ1.DNCOLischangeinnon-currentoperatingliabilitiesdeﬁnedasNCOLt–NCOLtÀ1.

DFINAischangeinﬁnancialassetsdeﬁnedasFINAtÀFINAtÀ1.DFINLischangeinﬁnancialliabilitiesdeﬁnedasFINLtÀFINLtÀ1.DSTIischangeinshortterminvestments.DLTIischangeinlong-terminvestments.

ROAisoperatingincomeafterdepreciation(CompustatItem#178)deﬂatedbyaveragetotalassets.

S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485477478

S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485

PanelCofTable8reportsregressionresultsfortheextendedaccrualdecomposition.WeagainseeaclosecorrespondencebetweentherelativerankingsontheaccrualcoefﬁcientswiththosereportedinpanelCofTable5.Mostimportantly,thecoefﬁcientsonDCOA,DNCOAandDLTIareconsistentlythemostnegativeandsigniﬁcant.ThesearetheaccrualsthatwejudgetobetheleastreliableandwhichalsohavethelowestpersistencecoefﬁcientsinTable5.Overall,theresultsareconsistentwiththenaıveinvestorhypothesis.Futurestockreturnsarenegativelyrelatedtoaccruals,andthenegativerelationisstrongerforlessreliableaccruals.TheresultsforoperatingliabilityaccrualsinpanelCofTable8displaysimilarpropertiestothosefrompanelCofTable5.Inparticular,thepersistencecoefﬁcientsonDCOLandDNCOLfallsharplyfromtheunivariateregressionstothemultivariateregression.Table9replicatestheadditionalanalysisofoperatingassetandliabilityaccrualsinTable7,butfocusesonnextyear’sstockreturnsinsteadofnextyear’searnings.ThestockreturnresultsinTable9closelymirrortheearningsresultsfromTable7.PanelAshowsthatthenegativerelationbetweenoperatingassetaccrualsandfuturestockreturnsisnotsigniﬁcantwhentheseaccrualsareﬁnancedbyoperatingliabilities.Thenegativerelationisdrivenbytheothersourcesofﬁnancing.PanelBshowsthatthepredictablefuturestockreturnsaregreatestwhenoperatingassetaccrualsandoperatingliabilityaccrualsmoveinoppositedirections.Firmsinthe‘lowDOA,highDOL’cellhavemeansizeadjustedreturnsof10.2%.Conversely,ﬁrmsinthe‘highDOA,lowDOL’cellhavemeansizeadjustedreturnsofÀ9.4%.19Theseresultsindicatethatinvestorsdonotuseinformationinoperatingliabilityaccrualsconcerningthepersistenceofoperatingassetaccruals.Table10providesevidenceontheeconomicsigniﬁcanceofourresults,usingthecommonconventionofreportingabnormalreturnsforrankedportfolios.Werankobservationsannuallyoneachaccrualcomponentandusetherankstogroupobservationsintotenequal-numberedportfolios(deciles).Wecalculatesize-adjustedannualreturnsforeachdecile,withthereturncumulationperiodstarting4monthsafterﬁscalyear-end.Wealsocomputereturnsonahedgeportfolioconsistingofalongpositioninthelowestaccrualdecileandanequal-sizedshortpositioninthehighestaccrualdecile,toprovideasummarymeasureoftheeconomicmagnitudeofthemispricingforeachaccrualcomponent.

PanelAofTable10reportstheresultsfortheinitialaccrualdecomposition.Thehedgeportfolioreturnfortotalaccrualsis13.3%.Theinitialaccrualdecompositionprovidesevidenceonthesourceofthesereturns.TheDWCcomponenthasahedgeportfolioreturnof12.8%.ThiscomponentofaccrualsdiffersslightlyfromtheaccrualdeﬁnitionusedbySloan(1996)inthatitexcludesdepreciation.TheDNCOcomponentofaccrualshasahedgeportfolioreturnof16.5%.Thiscomponentofaccrualsislargelyunexploredbypreviousresearch.DNOA(thesumofDWCandDNCO)hasthelargesthedgeportfolioreturnof18.0%.Bycombiningcurrentandnon-currentoperatingaccruals,weobtainhedgeportfolioreturnsthatarelarger

Wealsoconductedt-testsofdifferencesinmeansforthe4cellsatthecornersofthis5Â5DOA/DOLmatrix.Alldifferencesaresigniﬁcantatconventionallevels,withtheexceptionofthedifferencebetweenthe‘highDOA,lowDOL’cell(À0.094)andthe‘highDOA,highDOL’cell(À0.075).

19Table9

Additionaltestsinvestigatinghowthechangeinoperatingassets(DOA)andthechangeinoperatingliabilities(DOL)relatetonextyear’ssizeadjustedannualstockreturns(RETt+1).Thesampleconsistsof108,617ﬁrm-yearobservationsfortheperiod1962–2001

PanelA:Meancoefﬁcientsandt-statisticsfromannualcross-sectionalregressionsanalyzingaﬁnancingdecompositionofDOA.RETt+1¼r0+r1ROAt+r2DOAt+ut+1RETt+1¼r0+r1ROAt+r2DOLt+r3DFINLt–r4DFAt+r5DEquityt+ut+1Intercept

ROADOADOL

DFINL

ÀDFA

DEquity

Adj.R2Meancoef.0.0240.096À0.2280.014t-statistic(À11.12)

Meancoef.0.024

0.105

À0.034À0.255À0.179À0.2690.017

t-statistic

(À1.04)

(À9.76)

(À7.29)

(À6.25)

PanelB:Nextyear’smeansize-adjustedannualstockreturn(RETt+1)partitionedbyindependentquintilesortsonDOAandDOL(numberofobservationsinparentheses)DOA

DOLLow

234HighTotalAvg.Low0.0620.0660.0990.0600.1020.069(10,991)(4809)(2709)(1837)(1364)(21,710)20.0260.0220.0410.0810.0920.041(4763)(6997)(5326)(3179)(1465)(21,730)3À0.0220.0120.0230.0420.0320.020(2807)(5132)(6337)(5142)(2313)(21,731)4

À0.062À0.021À0.016À0.0160.045À0.008(2009)

(3282)

(4927)

(6580)

(4931)

(21,729)

S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485479Table9(continued)

PanelB:Nextyear’smeansize-adjustedannualstockreturn(RETt+1)partitionedbyindependentquintilesortsonDOAandDOL(numberofobservationsinparentheses)

DOADOLLow

234HighTotalAvg.HighÀ0.094À0.093À0.057À0.066À0.075À0.073(1140)(1510)(2431)(4992)(11,644)(21,717)

Total0.0230.0150.0190.007À0.014Average(21,710)

(21,730)

(21,717)

DOAisthechangeinoperatingassets,deﬁnedasDCOA+DNCOA.

DCOAischangeincurrentoperatingassetsdeﬁnedasCOAtÀCOAtÀ1.COA¼CurrentAssets(CompustatItem#4)ÀCashandShortTermInvestments(STI)(CompustatItem#1).COAisdeﬂatedbyaveragetotalassets.

DNCOAischangeinnon-currentoperatingassetsdeﬁnedasNCOAtÀNCOAtÀ1.NCOA¼TotalAssets(CompustatItem#6)ÀCurrentAssets(CompustatItem#4)ÀInvestmentsandAdvances(CompustatItem#32).NCOAisdeﬂatedbyaveragetotalassets.DOListhechangeinoperatingliabilities,deﬁnedasDCOL+DNCOL.

DCOLischangeincurrentoperatingliabilitiesdeﬁnedasCOLt–COLtÀ1.COL¼CurrentLiabilities(CompustatItem#5)ÀDebtinCurrentLiabilities(CompustatItem#34).COLisdeﬂatedbyaveragetotalassets.

DNCOLischangeinnon-currentoperatingliabilitiesdeﬁnedasNCOLt–NCOLtÀ1.NCOL¼TotalLiabilities(CompustatItem#181)ÀCurrentLiabilities(CompustatItem#5)–Long-termdebt(CompustatItem#9).NCOLisdeﬂatedbyaveragetotalassets.

DFINListhechangeinﬁnancialliabilities(inclusiveofpreferredstock)deﬁnedasFINLt–FINLtÀ1.FINL¼Long-termdebt(CompustatItem#9)+DebtinCurrentLiabilities(CompustatItem#34)+PreferredStock(CompustatItem#130).FINLisdeﬂatedbyaveragetotalassets.

DFAisthechangeinﬁnancialassetsdeﬁnedasFAtÀFAtÀ1.FA¼Cashandcashequivalents(CompustatItem#1)+InvestmentsandAdvances(CompustatItem#32).FAisdeﬂatedbyaveragetotalassets.

DEquityisthechangeinexternalequityﬁnancingdeﬁnedasEQtÀEQtÀ1.EQ¼Bookvalueofcommonequity(CompustatItem#60).EQisdeﬂatedbyaveragetotalassets.

480S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485

481

Table10

Nextyear’sannualmeansize-adjustedreturnsfordecileportfoliosformedontotalaccrualsanditscomponents.Thesamplecovers108,617ﬁrm-yearobservationsfortheperiod1962–2001PanelA:PortfoliosconstructedoncomponentsofinitialaccrualdecompositionPortfoliorankLow23456789HighHedget-statistic

TACC0.0660.0380.0290.0280.0160.0150.012À0.011À0.026À0.0670.13310.25

DWC0.0600.0560.0330.0200.0140.0240.004À0.010À0.032À0.0680.12810.56

DNCO0.0730.0530.0430.0420.0170.009À0.003À0.013À0.029À0.0920.16514.26

DNOA0.0800.0810.0440.0300.0300.019À0.008À0.032À0.045À0.1000.18014.91

DFIN-0.061-0.0130.0030.0160.0140.0250.0220.0340.0380.021À0.082À7.49

PanelB:PortfoliosconstructedoncomponentsofextendedaccrualdecompositionPortfoliorankLow23456789HighHedget-statistic

DCOA0.0690.0610.0290.0170.0260.0070.010À0.017À0.025À0.0760.14511.79

ÀDCOLÀ0.0270.0010.0010.0160.0160.0180.0100.0170.0200.028À0.055À4.64

DNCOA0.0700.0540.0490.0330.0220.004À0.001À0.009À0.029À0.0910.16114.02

ÀDNCOLÀ0.008À0.0030.0190.0070.0070.0160.0330.0070.0190.003À0.011À1.08

DSTI0.0220.0130.0050.0050.0050.0050.0050.0050.0160.0150.0070.71

DLTI0.0260.0080.0150.0150.0150.0150.0150.0150.002À0.0260.0525.40

ÀDFINLÀ0.076À0.024À0.0100.0070.0280.0350.0320.0380.0440.029À0.105À10.09

Portfoliosareformedbasedontotalaccrualsanditscomponents.Firm-yearobservationsarerankedannuallyandassignedinequalnumberstodecileportfoliosbasedontherespectiveaccrualcomponent.Hedgerepresentsthenetreturngeneratedbytakingalongpositioninthe‘Low’portfolioandanequalsizedshortpositioninthe‘High’portfolio.T-statistictestswhetherthehedgereturnisstatisticallydifferentfromzero.

TACCistotalaccrualsfromthebalancesheetapproach.ItiscalculatedasDWorkingCapital(DWC)+DNon-CurrentOperating(DNCO)+DFinancial(DFIN).Thiscanbeequivalentlywrittenas(DCOAÀDCOL)+(DNCOAÀDNCOL)+(DFINAÀDFINL).Totalaccrualsandallofitscomponents(describedbelow)aredeﬂatedbyaveragetotalassets.

DCOAischangeincurrentoperatingassetsdeﬁnedasCOAt–COAtÀ1.DCOLischangeincurrentoperatingliabilitiesdeﬁnedasCOLt–COLtÀ1.

482

S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485

Table10(continued)

DNCOisdeﬁnedasNCOtÀNCOtÀ1.NCO¼Non-CurrentOperatingAssets(NCOA)ÀNon-CurrentOperatingLiabilities(NCOL)whereNCOA¼TotalAssets(Compustatitem#6)ÀCurrentAssets(CompustatItem#4)ÀInvestmentsandAdvances(CompustatItem#32).NCOL¼TotalLiabilities(CompustatItem#181)ÀCurrentLiabilities(CompustatItem#5)–Long-termdebt(CompustatItem#9).DNCOAischangeinnon-currentoperatingassetsdeﬁnedasNCOAtÀNCOAtÀ1.DNCOLischangeinnon-currentoperatingliabilitiesdeﬁnedasNCOLtÀNCOLtÀ1.DNOAisdeﬁnedasthesumofDWCandDNCO.

DFINisdeﬁnedasFINt–FINtÀ1.FIN¼FinancialAssets(FINA)ÀFinancialLiabilities(FINL).FINA¼ShortTermInvestments(STI)(CompustatItem#193)+LongTermInvestments(LTI)(CompustatItem#32).FINL¼Longtermdebt(CompustatItem#9)+DebtinCurrentLiabilities(CompustatItem#34)+PreferredStock(CompustatItem#130).DFINAischangeinﬁnancialassetsdeﬁnedasFINAt–FINAtÀ1.DFINLischangeinﬁnancialliabilitiesdeﬁnedasFINLt–FINLtÀ1.DSTIischangeinshortterminvestments.DLTIischangeinlong-terminvestments.

Theportfolioreturnsareequal-weightedmeanannualbuy-holdsize-adjustedreturn.Thesize-adjustedreturniscalculatedbydeductingthevalue-weightedaveragereturnforallﬁrmsinthesamesize-matcheddecile,wheresizeismeasuredasmarketcapitalizationatthebeginningofthereturncumulationperiod.Thereturncumulationperiodbeginsfourmonthsaftertheendoftheﬁscalyear.

thanthoseobtainedusingeachcomponentindependently.inally,thehedgeportfolioreturnfortheDFINcomponentofaccrualsisÀ8.2%.Whileoperatingaccrualsarenegativelyassociatedwithfuturestockreturns,ﬁnancingaccrualsarepositivelyassociatedwithfuturestockreturns.ThisresultisexplainedbythenegativecorrelationbetweenDFINandtheoperatingaccruals,asreportedinTable3anddiscussedinSection4.1.AsortonDFINisessentiallyanoisyreversesortonDNOA,becauseDFINisaprimarysourceofﬁnancingforDNOA.

PanelBofTable10reportsthehedgeportfolioreturnsfortheextendeddecompositionofaccruals.Thestrongestmispricingisintheoperatingassetcomponentsofaccruals,DCOAandDNCOA,withhedgeportfolioreturnsof14.5%and16.1%,respectively.DecompositionofDFINintoDSTI,DLTIandDFINLshowsthatthesourceofnegativehedgeportfolioreturnsisDFINL,whichistheﬁnancingcomponenthavingthestrongestcorrelationwithoperatingaccruals.DLTIisassociatedwithapositivehedgeportfolioreturnof5.2%,consistentwithitslowerpersistenceandreliability.

Overall,theresultsinTable10conﬁrmthatourmorecomprehensivemeasureofaccrualsisassociatedwithevengreatermispricingthanhasbeendocumentedinpreviousaccrualresearchsuchasSloan(1996).Moreover,themispricingisdrivenbytheaccrualcomponentswiththelowestreliability.

5.Conclusionandimplications

Ourcomprehensiveexaminationofaccrualreliability,earningspersistenceandfuturestockreturnsgeneratesseveralnewinsights.First,wedocumentsigniﬁcantcostsfromrecognizinglessreliableinformationinﬁnancialstatements.Lessreliable

S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485

483

accrualsleadtolowerearningspersistenceandinvestorsdonotappeartofullyanticipatethislowerpersistence,leadingtosigniﬁcantsecuritymispricing.Second,weprovideacomprehensivedeﬁnitionandcategorizationofaccrualsthatincorporatesmanyaccrualsthathavebeenignoredbypreviousresearch.Ourresultsindicatethatsomeoftheaccrualcategoriesthathavebeenignoredbypreviousresearchhaveparticularlylowreliability.Third,weshowthatthemagnitudeofthesecuritymispricingrelatedtoaccrualsissigniﬁcantlygreaterthanoriginallydocumentedbySloan(1996).Forexample,asimpledecilesortonacombinationoftheleastreliableaccrualcategoriesproducesannualhedgeportfolioreturnsof18%.

Ouranalysishasseveralimplicationsforexistingresearch.First,ithighlightsthecrucialtrade-offbetweenrelevanceandreliabilityinaccrualaccounting.Weshowthatlessreliableaccrualsintroducecostsintheformoflowerearningspersistenceandtheassociatedmispricing.OurﬁndingsprovideanaturalcounterpointtoresearchcallingfornewcategoriesofrelevantbutrelativelyunreliableinformationtobeallowedunderGAAP(e.g.,LevandSougiannis,1996).Inparticular,ourresultssupportWatts’(2003)contentionthatrecentmovesbytheFASBtointroduceevenlessveriﬁableinformationintoGAAPmayproveextremelycostly.20WereiterateWatts’conclusionthatthecaseforallowinglessreliablecategoriesofaccrualsmustrecognizetheproblemsthatreliabilityevolvedtoaddress.Wenoteinthisrespectthatouranalysistreatsmeasurementerrorinaccrualsasanindependentrandomvariable.Totheextentthatstrategicmanagerialreactionand/oraccountingconventionssuchasconservatismcausethemeasurementerrortobehaveinanon-randomfashion,ouranalysismayunderestimatetheeffectsandcosts.

Thesecondkeyimplicationofouranalysisisthataccruals-basedresearchshouldconsiderbroaderdeﬁnitionsofaccrualsthanthedeﬁnitionoriginallyproposedbyHealy(1985).Healy’sdeﬁnitionhasbecomesynonymouswithaccrualsintheaccountingliterature.Yetweshowthathisdeﬁnitionexcludesseveraleconomicallysigniﬁcantcategoriesofaccrualswithlowreliability.Non-currentoperatingassetaccruals(e.g.,capitalizationofoperatingcostsaslong-termassets)representagoodexample.Suchaccrualswereattheheartofthewell-knownaccountingdebacleatWorldCom.Werecommendthatfutureresearchemploybroadermeasuresofaccrualsinordertoprovidemorepowerfultests.Ouranalysisalsohighlightsthedifferingdegreesofreliabilityassociatedwithdifferentcategoriesofaccruals.Inthisrespectouranalysisisstillquitecrude,andfurtherreﬁnementsshouldleadtoadditionalinsightsintoaccrualreliabilityandearningspersistence.

Third,ourﬁndingscorroborateclaimsconcerningtheimportanceofdistinguish-ingbetweenearningsand‘freecashﬂow’inevaluatingﬁrmperformance.21Thestandardtextbookdeﬁnitionoffreecashﬂowadjustsearningsbyaddingbackdepreciationandamortizationandsubtractingchangesinworkingcapitaland

ThespeciﬁcexamplesusedbyWattsaretheFASBsrecentstandardsSFASNo.141andSFASNo.142.Thesestandardsrequiremanagerstoestimatethefuturecashﬂowsassociatedwiththeﬁrm’soperatingassetsand(undercertaincircumstances)recognizetheseestimatesintheﬁnancialstatements.21Foranexample,see‘CashisKing’,Chapter5ofCopelandetal.(2000).

20484

S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485

capitalexpenditures.Assuch,theseadjustmentsaresubstantiallysimilartotheoperatingaccrualcomponentsthatwedevelopinthispaper.Financingaccrualsaretypicallynotbackedoutinarrivingatfreecashﬂow.Thus,theaccrualsthatweﬁndtobetheleastreliablecorrespondcloselytotheadjustmentsthataremadetoearningsinarrivingatfreecashﬂow.Freecashﬂowrepresentsacombinationofactualcashﬂowsplustherelativelyreliableﬁnancingaccruals.

Finally,ouranalysisissubjecttonumerouslimitations.First,wedonotdevelopandtestpotentialalternativeeconomicexplanationsforthedifferentialpersistencecoefﬁcientsonthecashﬂowandaccrualcomponentsofearnings.Inparticular,theaccrualcomponentofearningsisassociatedwithsalesgrowth,anditispossiblethateconomicfactorscauseearningstobelesspersistentingrowthﬁrms.ResearchbyXie(2001)andbyRichardsonetal.(2004)ﬁndsthatthelowerpersistenceofaccrualsremainsevenaftercontrollingforsalesgrowth.Nevertheless,additionalresearchtoevaluatetherelativeimportanceofcompetingexplanationswouldbeuseful.Second,weassumethatmeasurementerrorinaccrualsisanindependentrandomvariable.Inreality,measurementerrorislikelytobeinﬂuencedbystrategicmanagerialmanipulationofearningsandbyaccountingconventionssuchasconservatism.Modelingandendogenizingthemeasurementerrorshouldleadtoadditionalinsights.Finally,ouranalysisassumesthatanymeasurementerrorincashﬂowsisofsecond-orderimportance.Recentresearchsuggeststhatthecashﬂowsarealsosubjecttomanagerialmanipulation(seeRoychowdhury,2003;Grahametal.,2005).Thesimultaneousmodelingofmeasurementerrorincashﬂowsandaccrualsshouldprovideanimprovedunderstandingofearningspersistence.

References

Alford,A.W.,Jones,J.J.,Zmijewski,M.E.,1994.ExtensionsandviolationsofthestatutorySECForm

10-Kﬁlingrequirements.JournalofAccountingandEconomics17,229–256.

Barth,M.E.,Kallapur,S.,1996.Theeffectsofcross-sectionalscaledifferencesonregressionresultsin

empiricalaccountingresearch.ContemporaryAccountingResearch13,527–567.

Copeland,T.,Koller,T.,Murrin,J.,2000.Valuation:MeasuringandManagingtheValueofCompanies.

Wiley,NewYork.

Dechow,P.M.,1994.Accountingearningsandcashﬂowsasmeasuresofﬁrmperformance:theroleof

accountingaccruals.JournalofAccountingandEconomics18,3–42.

Dechow,P.M.,Sloan,R.G.,Sweeney,A.,1996.Causesandconsequencesofearningsmanipulation:an

analysisofﬁrmssubjecttoenforcementactionsbytheSEC.ContemporaryAccountingResearch13,1–36.

Dechow,P.M.,Kothari,S.P.,Watts,R.L.,1998.Therelationbetweenearningsandcashﬂows.Journalof

AccountingandEconomics25,133–168.

Fairﬁeld,P.M.,Whisenant,J.S.,Yohn,T.L.,2003a.Accruedearningsandgrowth:implicationsforfuture

proﬁtabilityandmarketmispricing.AccountingReview78,353–371.

Fairﬁeld,P.M.,Whisenant,J.S.,Yohn,T.L.,2003b.Thedifferentialpersistenceofaccrualsandcashﬂows

forfutureoperatingincomeversusfutureproﬁtability.ReviewofAccountingStudies8,221–243.Fama,E.F.,Macbeth,J.D.,1973.Risk,returnandequilibrium—empiricaltests.TheJournalofPolitical

Economy81,607–636.

Graham,J.R.,Harvey,C.R.,Rajgopal,S.,2005.Theeconomicimplicationsofcorporateﬁnancial

reporting.TheJournalofAccountingandEconomicsforthcoming.

S.A.Richardsonetal./JournalofAccountingandEconomics39(2005)437–485

485

Griliches,Z.,Ringstad,V.,1971.EconomiesofScalesandtheFormoftheProductionFunction.North

Holland,Amsterdam.

Healy,P.,1985.Theeffectofbonusschemesonaccountingdecisions.JournalofAccountingand

Economics7,85–107.

Holthausen,R.W.,Watts,R.L.,2001.Therelevanceofthevaluerelevanceliteratureforﬁnancial

accountingstandardsetting.JournalofAccountingandEconomics31,3–75.

Hribar,P.,Collins,D.,2002.Errorsinestimatingaccruals:implicationsforempiricalresearch.Journalof

AccountingResearch40,105–134.

Jones,J.,1991.Earningsmanagementduringimportreliefinvestigations.JournalofAccountingResearch

29,193–228.

Klepper,S.,Leamer,E.E.,1984.Consistentsetsofestimatesforregressionswitherrorsinallvariables.

Econometrica52,163–183.

Kothari,2001.Capitalmarketsresearchinaccounting.JournalofAccountingandEconomics31,

105–231.

Lev,B.,Sougiannis,T.,1996.Thecapitalization,amortizationandvaluerelevanceofR&D.Journalof

AccountingandEconomics21,107–138.

Nissim,D.,Penman,S.H.,2001.Ratioanalysisandequityvaluation:fromresearchtopractice.Reviewof

AccountingStudies6,109–154.

Nissim,D.,Penman,S.H.,2003.Financialstatementanalysisofleverageandhowitinformsabout

proﬁtabilityandprice-to-bookratios.ReviewofAccountingStudies8,531–560.

Palepu,K.G.,Healy,P.M.,Bernard,V.L.,2000.BusinessAnalysisandValuation,Seconded.

SouthwesternCollegePublishing.

Penman,S.H.,Zhang,X.J.,2002.Accountingconservatism,thequalityofearningsandstockreturns.

AccountingReview77,237–264.

˙.,2004.TheimplicationsofﬁrmgrowthandRichardson,S.A.,Sloan,R.G.,Soliman,M.T.,Tuna,I

accountingdistortionsforaccrualsandproﬁtability.Workingpaper,UniversityofMichigan.

Roychowdhury,S.,2003.ManagementofEarningsThroughtheManipulationofRealActivitiesthat

AffectCashFlowsfromOperations.MITPress,Cambridge,MAWorkingpaper.Shumway,T.,1997.ThedelistingbiasinCRSPdata.JournalofFinance52,327–340.

Sloan,R.G.,1996.Dostockpricesfullyreﬂectinformationinaccrualsandcashﬂowsaboutfuture

earnings?TheAccountingReview71,289–315.

Thomas,J.K.,Zhang,H.,2002.Inventorychangesandfuturereturns.ReviewofAccountingStudies7,

163–187.

Watts,R.L.,2003.ConservatisminaccountingpartI:explanationsandimplications.Accounting

Horizons17,207–221.

Xie,H.,2001.Arediscretionaryaccrualsmispriced?Areexaamination.TheAccountingReview76,

357–373.

因篇幅问题不能全部显示，请点此查看更多更全内容

查看全文