Earnings management and the underperformance of

《Earnings management and the underperformance of》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

JournalofFinancialEconomics50(1998)63Ð99Earningsmanagementandtheunderperformanceofseasonedequityo¤erings1SiewHongTeoh!,*,IvoWelch",T.J.Wong#!UniversityofMichiganBusinessSchool,AnnArbor,MI48109,USA"AndersonGraduateSchoolofManagement,UCLA,LosAngeles,CA90095,USA#HongKongUniversityofScienceandTechnology,ClearwaterBay,HongKongReceived28March1995;receivedinrevisedform26June1997AbstractSeasonedequityissuerscanraisereportedearningsbyalteringdiscretionaryac-countingaccruals.WeÞndthatissuerswhoadjustdiscretionarycurrentaccrualstoreporthighernetincomepriortotheo¤eringhavelowerpost-issuelong-runabnormalstockreturnsandnetincome.Interestingly,therelationbetweendiscretionarycurrentaccrualsandfuturereturns(adjustedforÞrmsizeandbook-to-marketratio)isstrongerandmorepersistentforseasonedequityissuersthanfornon-issuers.Theevidenceisconsistentwithinvestorsnaivelyextrapolatingpre-issueearningswithoutfullyadjustingforthepotentialmanipulationofreportedearnings.(1998ElsevierScienceS.A.Allrightsreserved.JELclassiÞcation:G14;G24;G32;M41Keywords:CorporateÞnance;Seasonedequityo¤erings;Earningsmanagement;Accountingaccruals;Anomalies;Markete¦ciency*Correspondingauthor.Tel.:313/763-1264;fax:313/764-2557;e-mail:siew.teoh@ccmail.bus.umich.edu.1WethankBradBarber(thereferee),RandyBeatty,VicBernard,K.C.Chan,KentDaniel,MarkDeFond,LauraField,DavidHeike,ChuanYangHwang,JonathanKarpo¤,S.P.Kothari,CharlesLee,WayneMikkelson(theeditor),TimOpler,KrishnaPalepu,K.Ramesh,JayRitter,TerryShevlin,DougSkinner,SheridanTitman,RossWatts,JerryZimmerman,andseminarparticipantsattheUniversityofCaliforniaFinanceandAccountingConference(Davis,March1995),theNBERCorporateFinanceConference(Boston,August1995),theCenterforResearchinSecurityPricesSeminar(Chicago,October1995),theAmericanFinanceAssociationConference(SanFrancisco,1996),theAmericanAccountingAssociationConference(Chicago,August1996),theUniversityofMichigan,andtheUniversityofRochesterforhelpfulcommentsanddiscussions.0304-405X/98/$19.00(1998ElsevierScienceS.A.AllrightsreservedPIIS0304-405X(98)00032-4 64S.H.Teoh,etal./JournalofFinancialEconomics50(1998)63Ð991.IntroductionLoughranandRitter(1995)andSpiessandA§eck-Graves(1995)documentthatÞrmsunderperformthestockmarketintheÞveyearsafteraseasonedequityissue.Forexample,LoughranandRitterreportaveragereturnsofonly7%peryear,whilecomparablenon-issuingÞrmsaverage15%peryear.Thereturndi¤erentialsaresolargethatonewonderswhyinvestorsbuytheseissues.Inthispaper,weexaminewhetherunusuallyaggressivemanagementofearningsthroughincome-increasingaccountingadjustmentsleadsinvestorstobeoverlyoptimisticabouttheissuerÕsprospects.Thatis,investorsmaymisinter-prethighearningsreportedatthetimeoftheo¤ering,andconsequentlyovervaluethenewissues.Whenhighpre-issueearningsarenotsustained,disappointedinvestorssubsequentlyrevaluetheÞrmdowntoaleveljustiÞedbyfundamentals.Thisearningsmanagementhypothesispredictsthatissuershaveunusuallyhighincome-increasingaccountingadjustmentspre-issueandunusuallypoorearningsandstockreturnperformancepost-issue.Further,thehypothesispredictsworseperformanceforissuerswithunusuallylargeincome-increasingaccountingadjustmentspriortotheo¤ering.Wereportevidenceconsistentwiththeearningsmanagementhypothesis.Forasampleofseasonedequityissuersfrom1976to1989,Table2documentshighernetincomegrowthintheissueyearforissuersthanforperformance-matchednon-issuingindustrypeers.Post-issue,however,issuerssigniÞcantlyunderperformtheirmatches.Forexample,theannualgrowthintheissuersÕasset-scalednetincomesigniÞcantlyexceedsthatofthematchednon-issuersbyamedianof1.69%intheissueyear,butissigniÞcantlylessthanthatofthematchednon-issuersbyamedianof1.60%and0.32%inthetwosub-sequentyears.Decomposingnetincomeintocashßowfromoperationsandaccountingadjustments(hereafterreferredtoasaccruals),weÞndthatitistheaccrualsthatcausetheat-issuepeakandpost-issueunderperformanceinnetincome.Incontrast,cashßowfromoperationsexhibitanoppositeproÞle.Table2reportsthatasset-scaledcashßowfromoperationsare3.8%belowtheindustrymedianintheissueyear,remainbelowtheindustrymedianforthenexttwoyears,andonlyriseto1%abovetheindustrymedianthreeyearsaftertheissue.Wedecomposeaccrualsintofourcategoriesjointlybytimeperiod(currentandlong-term)andmanagercontrol(discretionaryandnondiscretionary).Table3reportsthatofthefourcategories,discretionarycurrentaccruals(thecomponentmostsubjecttomanagerialmanipulation)drivethepost-issueearningsunderperformance.Intheo¤eringyear,theasset-scaleddiscretionarycurrentaccrualsofissuersexceedtheirpre-issueperformance-matchedindustrypeersby2.9%.Foreachofthethreesubsequentyears,theissuersÕdiscre-tionarycurrentaccrualsdeclinebymorethanthoseoftheirmatches.Ranking S.H.Teoh,etal./JournalofFinancialEconomics50(1998)63Ð9965issuersbydiscretionarycurrentaccruals,Table4reportsthatissuersinthemostaggressivequartile(i.e.withthelargestdiscretionarycurrentaccrualsinthepre-issueyear)underperformtheirmatchednon-issuersby7.50%inasset-scalednetincomeinthethreeyearsaftertheissueyear.Incontrast,issuersintheconservativequartileoutperformtheirmatchesby0.99%.Table5reportsthattheSpearmanrankcorrelationsbetweenpre-issuediscretionarycurrentac-crualsandpost-issuenetincomechanges(allasset-scaled)areapproximately!20%andstatisticallysigniÞcantinallthreepost-issueyears.Insum,theevidencesuggeststhatdiscretionarycurrentaccrualspredictpost-issueearningsunderperformance.Mostinterestingly,weÞndevidencethatdiscretionarycurrentaccrualsalsopredictunderperformanceinpost-issuestockreturns.For48monthsaftertheo¤ering,issuersintheaggressivequartileunderperformconservativeissuersbyarawreturndi¤erentialofabout40%,amarket-adjustedreturndi¤erentialofabout25%,andaFamaÐFrenchadjustedreturndi¤erentialofabout35%(seeTable6).Thesedi¤erencesareremarkableconsideringthattheearningsman-agementproxyisbasedonpublicinformationavailablefourto16monthsbeforetheperiodoverwhichreturnsaremeasured.Finally,inTable7,weuseaFamaÐMacbethtypeprocedureinasamplecontainingissuingandnon-issuingÞrms.Wetestwhetherpost-issueabnormalstockreturnsarenegativelyrelatedtolaggedaccruals,andwhetherthisrelationisstrongerforpre-issueaccruals.Theresultsindicatethatdiscretionarycurrentaccrualshaveastrongerandmorepersistentinßuenceonsubsequentreturnsforseasonedequityissuers.Therefore,consistentwithearningsmanagement,weÞndevidencethathighdiscretionarycurrentaccrualspredictpost-issuelong-runearningsandstockreturnunderperformance.2.Accruals-basedearningsmanagementproxiesToevaluatetheroleofearningsmanagement,weconstructaproxyfortheamountofaccountingadjustmentsundertakenbymanagement.Reportedearn-ingsintheÞnancialstatementconsistofcashßowfromoperationsplustotalaccruals:NetIncome"TotalAccruals#CashFlowfromOperations.(1)Theaccrualadjustmentsreßectbusinesstransactionsthata¤ectfuturecashßowseventhoughcashhasnotcurrentlychangedhands.Undergenerallyacceptedaccountingprinciples(GAAP),Þrmshavediscretiontorecognizethesetransactionsaseconomicevents,sothatreportedearningsreßectthetrueunderlyingbusinessconditionsoftheÞrmmoreaccurately.Withtheaccrualsystemofaccounting,reportedearningsaresupposedtobeinvarianttothetimingofcashreceiptsandpayments.However,managerialßexibilityinthe 66S.H.Teoh,etal./JournalofFinancialEconomics50(1998)63Ð99accrualssystemalsoopensopportunitiesforearningsmanagement.2Bytakingincome-increasingaccrualadjustmentsnow,managerscanraisecurrentre-portedearnings,butfuturereportedearningswillbelower.However,ac-countingregulations,suchastherequirementofanindependentaudit,limitthemanagerÕsdiscretionoverthetimingandmagnitudeofaccruals.AccrualscanbeclassiÞedintocategoriesbasedontimeperiodandmanageri-alcontrol.Currentaccrualsareadjustmentsinvolvingshort-termassetsandliabilitiesthatsupporttheday-to-dayoperationsoftheÞrm.Forexample,managerscanaltercurrentaccrualsbyadvancingrecognitionofrevenueswithcreditsales(beforecashisreceived),bydelayingrecognitionofexpensesaftercashisadvancedtosuppliers,andbyassumingalowprovisionforbaddebts.Incontrast,long-termaccrualsareadjustmentsinvolvinglong-termnetassets.Theseaccrualscanbealteredbydeceleratingdepreciation,decreasingdeferredtaxes(thedi¤erencebetweentaxexpenserecognizedforÞnancialreportingandactualtaxespaid),andrealizingunusualgains.Weconsidercurrentaccrualsandlong-termaccrualsseparatelybecauseaccountingresearchers(e.g.,Guenther,1994)havearguedthatmanagershavegreaterdiscretionovercurrentaccrualsthanoverlong-termaccruals.Althoughinvestorscanobserveaccruals,theycannotinferperfectlywhatportionofaccrualsisdiscretionary,i.e.,ÔmanagedÕ.Givenindustry-relatedandÞrm-speciÞcbusinessconditions,someaccrualadjustmentsarenecessary,andindeedexpectedbyinvestors;forexample,asset-intensiveÞrmshavehighdepreciation,andrapidlygrowingÞrmshaverevenuesthatexceedcashsales.ToextractthesenondiscretionaryaccrualsthataredictatedbyÞrmcondi-tionsandindependentofmanagerialmanipulation,weuseacross-sectionaladaptationTeohetal.(forthcomingb)ofthemodiÞedJones(1991)model.ThedetailsoftheprocedurearedescribedintheAppendix.Inessence,current2ForspeciÞcexamplesofhowearningscanbemanagedinanaccrualaccountingsystem,seeDavidsonetal.(1986),Teohetal.(forthcomingb),AppendixBinTeohetal.(forthcominga),KelloggandKelloggÕsFinancialStatementAlert,whichisaninvestornewsletterdevotedtoferretingoutsuspiciousaccountingadjustments,andmediaarticlessuchasÔTheSherlockHolmesofAccountingÕ,(BusinessWeek,pp.48Ð52;September5,1994),andtheseriesofarticlesinForbesinthesectionÔNumbersGameÕ,suchasÔLiesoftheBottomLineÕ(November12,1990),ÔSillyPussyfootingÕ(August21,1989),ÔNumbersPumpersÕ(November11,1991),andÔMysteryProÞtsÕ(April20,1987).Thesearticlesdetailearningsmanagementwithingenerallyacceptedaccountingprinciplesandnotnecessarilyfraudulentreporting.WedonotconsiderfraudulentreportingbehaviorspeciÞcally,becausethemajorityoftheseasonedequityissuersinoursampleappearstocomplywithgenerallyacceptedaccountingprinciples(GAAP).RelativelyfewÞrmsinthegeneralpopulationarecaughtnotcomplyingwithGAAP,andofthese,fewareseasonedequityissuers.MikeMahergenerouslyprovidedusthenamesof159SEC-reportingviolatorsfromJanuary1980toDecember1985,andonlysevenofourSEOÞrmsoverlappedwithhissampleofviolators. S.H.Teoh,etal./JournalofFinancialEconomics50(1998)63Ð9967accrualsareregressedonthechangeinsalesinacross-sectionalregressionusingallÞrmsinthesametwo-digitSICcodeastheissuer(butexcludingtheissuer).Thecross-sectionalregressionisperformedforeachÞscalyear,andallvariablesarescaledbybeginning-yearÞrmassets.Afteradjustingsalesgrowthfortheincreaseinaccountsreceivable,theissuerÕsÞttedcurrentaccrualslevel,termednondiscretionarycurrentaccruals(NDCA),isconsideredtypicalintheindus-~1tryforthelevelofsalesgrowth.SincetheremainingcurrentaccrualsarenotdictatedbyÞrmcondition,butaremanaged,theyaretermeddiscretion-arycurrentaccruals(DCA).~1Todecomposelong-termaccruals,weapplyanequivalentprocedure.WeÞrstdecomposetotalaccrualsintoadiscretionaryandanondiscretionarycompon-entbasedonsalesgrowthandproperty,plant,andequipment(toadjustfordepreciation).Thedi¤erencebetweendiscretionarytotalaccrualsanddis-cretionarycurrentaccrualsistermeddiscretionarylong-termaccruals(D¸A);~1thedi¤erencebetweennondiscretionarytotalaccrualsandnondiscretionarycurrentaccrualsistermednondiscretionarylong-termaccruals(ND¸A).~1Thecross-sectionalapproachautomaticallyadjustsforchangingindus-trywideeconomicconditionswhichinßuenceaccrualsindependentlyofearningsmanagement.Usingindustrybenchmarkstomeasurediscretionaryaccrualsissuggestedbythecommonpracticeofunderwriters,whopricenewequityissuesbycomparingmarketpricesandaccountingvariablesofsimilarÞrms.Tosummarize,accrualsaredecomposedintofourcomponents:discretionaryandnondiscretionarycurrentaccruals,anddiscretionaryandnondiscretionarylong-termaccruals.Thenondiscretionaryaccrualsareproxiesforaccrualrecog-nitionoutsidethecontrolofmanagementandthediscretionaryaccrualsareproxiesforearningsmanagement.33.SampleselectionandsamplecharacteristicsOurinitialsampleconsistsof6386seasonedequityissuesbetweenJanuary1970andSeptember1989fromtheSecuritiesDataCorporation.Ofthese,only3032issuesareavailableontheprimary,fullcoverage,andresearchCompustat1993tapesandontheCenterforResearchinSecurityPrices(CRSP)1993tapes.ForinclusionintheÞnalsample,werequireavailablestockreturnsdataand3TherobustnessofmodelsofearningsmanagementmeasureshasbeendiscussedbyDechowetal.(1995),Guayetal.(1996),andHealy(1996).Dechow,Sloan,andSweeneyconcludesuperiorityofthemodiÞedJonesmodeloverallothercurrentlyavailablemodels,thoughtheJonesmodelremainsimperfect. 68S.H.Teoh,etal./JournalofFinancialEconomics50(1998)63Ð99su¦cientdatatocomputediscretionaryaccountingaccrualsfortheyearpriortotheo¤ering.Toavoidsurvivorshipbias,wedonotrequirethatÞrmshaveaccrualsdatafortheentireperiodofthreeyearsbeforetothreeyearsaftertheissueyear.Becausebankingandutilitiesindustrieshaveuniquedisclosurerequirements,weeliminateÞrmsintheseindustriesfromoursample.IfaÞrmhasmultipleissues,weincludeonlytheearliestissuetoavoidusingoverlappingdatatoestimatethereturns-accrualsrelation.Ofthe3032issuersavailableonCRSPandCompustat,2645havesu¦cientdatatocomputeaccrualsinatleastoneyearbetweenthe1974Ð1993periodcoveredbyCompustat1993tapes.Ofthese,1285haveatleasttenotherÞrmsinthesametwo-digitSICcodeindustrygrouptoallowestimationoftheintra-industryregressiontocalculateexpectedaccruals.TheÞnalsampleconsistsof1265issuerswithavailableaccrualsdataintheÞscalyearpriortotheÞscalyearofthenewissue.Theactualsamplesizevariesdependingonthetestproceduresandaccrualsmeasuresused.Only1248Þrmshaveavailablereturnsduringtheissuemonth.Table1reportsthesamplestatisticsanddatacharacteristicsforourÞrms.TheearliestdataavailableonCompustat1993areforÞscalyear1973.BecauseweexaminetheaccrualsbehaviorofseasonedissuersfromÞscalyear!3to#3relativetotheÞscalyearoftheo¤ering,oursamplebeginsin1976.Seasonedequityissuesareclusteredbyindustriesandtimeperiods.Fourofthesampleyears(1980,1982,1983and1986)areveryactiveandcontainmorethan10%ofthesample,with1983carrying22%oftheissues.Furthermore,thecomputerandelectronicsindustriesaccountforalargefractionoftheissueswithapproximately31%ofthesample.Earningsmanagementmaybeprevalentintheserelativelynewindustriesbecausehighinformationasymmetryandlimitedpasthistorymakeitdi¦culttojudgetheappropriatenessoftheaccountingchoices.PanelCofTable1reportssizestatisticsforthesampleintheÞscalyearpriortotheissue.Themeanandmedianoftotalbookvalueofassetsare$625millionand$40million.Themeanandmedianofmarketcapitalizationofequityare$284millionand$52million.Assetsizevariesconsiderablyinthesampleasevidencedbythelargestandarddeviations.Themeanandmedianofsalesgrowthscaledbyassets,anexplanatoryvariableintheJones(1991)modelforaccruals,are54%and28%.LoughranandRitter(1995)alsoreporthighsalesgrowthfornewissuers.4.Post-issuepredictabilityofearningsInthissection,weÞrstexamineifthereisearningsunderperformanceafteraseasonedequityissue.WethenexaminethetimeproÞleoftheaccrualandcashßowcomponentsofnetincomearoundthetimeoftheissuetoevaluatethe S.H.Teoh,etal./JournalofFinancialEconomics50(1998)63Ð9969Table1Characteristicsof1265Þrmsconductingseasonedequityo¤eringsfrom1976to1989PanelA:SICdistributionIndustryCodesFreq%OilandGas13624.9FoodProducts20262.1PaperandPaperProducts24,25,26,27574.5ChemicalProducts28755.9Manufacturing30Ð34977.7ComputerEquipmentandServices35,7324519.4ElectronicEquipment3614111.1Transportation37,39,40Ð42,44,45987.7ScientiÞcInstruments381068.4Communications48302.4DurableGoods50332.6Retail53,54,56,57,59594.7EatingandDrinkingEstablishments58393.1Financialservices61,62,64,65352.8Entertainmentservices70,78,79332.6Health80342.7Allothers10,15,16,22,23,51,87,99957.5PanelB:¹imedistributionYearFreq%CumFreq%197620.220.21977352.8372.91978514.0887.01979483.813610.8198014411.428022.119811007.938030.0198213110.451140.4198327621.878762.21984534.284066.419851018.094174.4198614511.5108685.819871018.0118793.81988443.5123197.31989342.71265100.0PanelC:SizecharacteristicsTotalMarketBookSalesassetsvaluevaluegrowthMean625.2284.2207.20.537Median40.451.818.00.283Std.dev.2,653.9971.6884.81.107Sizecharacteristicsaremeasuredinmillionsofdollars.TotalassetsareobtainedattheendoftheÞscalyearpriortotheseasonedequityo¤ering,orÞscalyear!1.MarketvaluesarethenumberofsharesoutstandingtimesthestockpriceatendofÞscalyear!1.BookvalueofequityismeasuredatendofÞscalyear!1.SalesgrowthisthechangeinsalesinÞscalyear!1deßatedbytotalassetsinyear!2. 70S.H.Teoh,etal./JournalofFinancialEconomics50(1998)63Ð99relativecontributionofcashßowsandaccrualstothepost-issuenetincomeperformance.Forevidenceontherelativemagnitudeoftheearningsunderper-formance,wecomparethepost-issueearningsperformanceofissuersrankedbytheirdiscretionarycurrentaccrualsinthepre-issueyear.Finally,weexaminetheSpearmanrankordercorrelationbetweenpre-issueaccrualsandpost-issueearningsunderperformancetotestwhetherpre-issueaccrualsexplainthecross-sectionalvariationinpost-issueearningsunderperformance.4.1.Post-IssueearningsunderperformanceintimeseriesTable2reportsthreemeasuresofnetincomeperformanceinthesixyearssurroundingtheissueyear:netincomeasapercentageofprioryeartotalassets,asset-scalednetincomeminustheindustrymedianasset-scalednetincome,andtheannualchangeinasset-scalednetincomeoftheissuerminusthechangeforapre-issueperformance-matchednon-issuer.NetincomeisCompustatitem172,whichisthenumberreportedinanearningsannouncementintheWallStreetJournalandwhichcapturesmorefullythee¤ectsofdiscretionaryreport-ingchoices,suchasextraordinaryitems,ontheearningsperformance.Theresultsusingearningsbeforeinterestandtaxes(item18)arequalitativelysimilar,andarenotreportedhere.Thesecondmeasureadjustsforchangingbusinessconditionsintheindustry.ThisadjustmentcouldbeimportantgiventheevidenceinRitter(1991)thatsomeindustriesexperiencedsigniÞcantdeclinesinoperatingperformanceinthe1980s.ThethirdmeasureisrecommendedbyBarberandLyon(1997)forremovingnormalmeanreversioninnetincome.Thestatisticalmeansareobtainedafterwinsorizingthedataatthe1%and99%leveltoreducethee¤ectofafewlargevalues.Themeanswithoutwinsorizingaresimilar,butlessstatisticallysigniÞcant.Thepatternofunadjustedasset-scalednetincomeindicatesimprovingpre-issueperformancebutdeterioratingpost-issueperformance.Themediangrowsfrom6.50%inyear!3toapeakof9.00%inyear0,thendeclinesto3.80%byyear#3.Theequivalentmeansare6.33%inyear!3,6.63%inyear0,and0.71%inyear#3.Theindustry-adjustedperformancemeasuresindicateasimilarproÞleofpre-issueimprovementandpost-issuedecline.Themedianindustry-adjustedasset-scalednetincomegrowsfrom1.40%inyear!3to5.00%inyear0,anddeclinesto1.40%byyear#3.Theequivalentmeansare1.23%inyear!3,3.01%inyear0,and!1.42%inyear#3.Tomatcheachissuerwithanon-issuerofcomparablepre-issueperformanceforthethirdmeasure,weÞrstselectthenon-issuerfromthesameindustrywithasset-scalednetincomeclosesttothatoftheissuerinyear!1.Webeginwiththefour-digitSICcode;ifnomatchisavailable,wesearchamongthree-digitSICcodes,thentwo-digit,andÞnallyone-digit.Althoughwerequiretheasset-scalednetincomeofthenon-issuertobeatleast80%oftheasset-scalednetincomeoftheissuer,wedonotimposeanupperbound.Themajorityofthe S.H.Teoh,etal./JournalofFinancialEconomics50(1998)63Ð9971Table2TimeseriesproÞleofasset-scalednetincomeandcashßowfromoperations,inpercent,fromyear!3to#3relativetotheseasonedequityo¤ering(year0)Thistablepresentsthreemeasuresoftheperformanceoftwoaccountingvariables,netincomeandcashßowfromoperations,fromyears!3to#3relativetotheseasonedequityo¤ering.TheÞrstistheleveloftheaccountingvariablescaledbypriortotalassets,thesecondistheissuerÕsasset-scaledaccountingvariableminusthatofthemedianÞrminthesameindustry,andthethirdistheissuersÕyear-to-yearchangeintheasset-scaledaccountingvariableminusthatofamatchednon-issuer.Thematchednon-issuermustbeinthesameindustryandhavetheclosestasset-scalednetincometotheissuerinthepre-o¤eringÞscalyear.ThethirdnetincomeperformancemeasureiscomputedasNINININIi,t!i,t~1!m,t!m,t~1,A¹A¹ABA¹A¹ABi,t~1i,t~2m,t~1m,t~2whereiandmaretheissuerandmatchedÞrm,tistheÞscalyear,andNIisnetincome(Compustatitem172),and¹Aisbeginning-periodtotalassets.Theperformancemeasuresforthecashßowfromoperationsiscalculatedlikewise,usingcashßowfromoperations(usuallyCompustatitem308exceptasnotedintheAppendix)insteadofnetincome.Year!3!2!10123PanelA:NetincomeperformanceºnadjustednetincomeMedian6.50!7.40!8.00!9.00!6.10!4.80"3.80Mean6.33!5.49!6.71!6.63!3.33!0.99!0.71!N877103812651247120011341071Issuers+netincomeÐIndustrymediannetincomeMedian1.40!2.40!3.70!5.00!2.40!1.70!1.40"Mean1.23!0.802.82!3.01!!0.01!1.56!!1.42!N877103812651247120011341071Issuers+netincomechangeÐPerformance-matchednon-issuers+netincomechangeMedian.0.04!0.031.69!!1.60!!0.32"!0.01Mean.!0.70!0.693.15!!2.16!!2.00!1.79"N.78697111651028905798PanelB:CashßowfromoperationsºnadjustedcashßowfromoperationsMedian9.25!8.60!8.40!7.70!6.20!7.60!7.70!Mean8.12!5.40!3.94!2.69!3.86!4.82!6.17!N876103712651247119911281067Issuers+ßowfromoperationsÐIndustrymediancashßowfromOperationsMedian1.50!0.801.200.600.00#1.75!1.70!Mean0.46!1.97!!3.05!!3.84!!2.20!!0.780.98"N876103712651247119911281067 72S.H.Teoh,etal./JournalofFinancialEconomics50(1998)63Ð99Table2.Continued.Year!3!2!10123Issuers+cashßowchangeÐPerformance-matchednon-issuers+cashßowchangeMedian.!0.56!0.75!0.76!0.991.14"0.51Mean.!1.85#1.14!3.60!1.481.482.06"N.78396611601021895789!representstatisticalsigniÞcancelevelsatthe1%levels,usingt-testsforthemeanandWilcoxonp-valuesforthemedian."representstatisticalsigniÞcancelevelsatthe5%levels,usingt-testsforthemeanandWilcoxonp-valuesforthemedian.#representstatisticalsigniÞcancelevelsatthe10%levels,usingt-testsforthemeanandWilcoxonp-valuesforthemedian.Þrmsarematchedbythree-digitSICcodes,and94%arematchedbyatleasttwo-digitcodes.Themedianandmeanasset-scalednetincomelevelsforissuersforwhichmatchesareavailable,8.00%and5.34%,areslightlysmallerthanthematched-Þrmmedianandmeanof8.39%and5.96%.Byallowinghigherincomesfornon-issuers,weexaggeratethedropinnetincomeforthematchedÞrmifasset-scalednetincomenormallymean-reverts.Thus,theperformance-matchedmeasureisconservative,becauseitrequiresabiggerdropintheissuerÕspost-o¤eringnetincometoindicateunderperformance.4MatchedÞrmsandissuersshowsimilarincomeperformanceinthepre-issueyears.Thenetincomeperformanceofissuersdivergesfromtheirmatchesbeginninginyear0.IssuersgrowsigniÞcantlyfasterthantheirmatchednon-issuersbyamedianandmeanof1.69%and3.15%intheissueyear,respectively.Ineachofthetwoyearsaftertheo¤ering,issuerssigniÞcantlyunderperformtheirmatches.ThegrowthintheissuersÕasset-scalednetincomeislessthanthatofthematchednon-issuersbyasigniÞcantmedianandmeanof1.60%and2.16%intheÞrstyearaftertheo¤eringand0.32%and2.00%inthesecondyear(alsosigniÞcant).Thus,weconcludethatthepreviouslydocumentedpost-issuestockreturnunderperformanceofissuersisaccompaniedbyunusuallypoorearningsperformance.54Themeasureisalsoconservativeforasecondreason.Bysubtractingallofprioryearchange,themeasureunderestimatestheearningsunderperformanceinlaterpost-issueyearswhenaccrualsborrowedinpre-issueyearsarepaidbackoverthesubsequentyearsaftertheissue.5Poorperformancefollowingaseasonedo¤eringisalsoreportedbyHansenandCrutchley(1990)andLoughranandRitter(1997),butnotHealyandPalepu(1990).ThesamplesarerelativelysmallinHealyandPalepu(93issues)andinHansenandCrutchley(109issues).LoughranandRitter(1997)usedi¤erentmeasuresofperformance.Poorpost-issueperformancehasalsobeendocumentedforinitialpublico¤eringsbyJainandKini(1995),Mikkelsonetal.(1997),andTeohetal.(forthcomingb). S.H.Teoh,etal./JournalofFinancialEconomics50(1998)63Ð9973Next,weexaminewhichofthetwocomponents,cashßowfromoperationsoraccruals,inducestheobservednetincomepattern.TheresultsinPanelBofTable2indicatethatthenetincomeproÞleisnotmirroredbycashßowfromoperations.Themediansofallthreemeasuresofasset-scaledcashßows(unad-justed,industry-adjusted,andperformance-matched)showamonotonicdeclinefrompre-issueperiodstothelowestlevelsintheissueyearbeforeimprovinginyears#2and#3.Themeansalsoshowsimilarpatterns,withlowlevelsinyears0and#1.Therefore,newissuesoccurwhencashßowsfromoperationsaredeclining,notwhentheyareatapeak.Consequently,astheremainingcomponentofearnings,accrualsmustbedrivingtheobservednetincomeproÞlefornewissues.Wenowevaluatewhichofthefouraccrualmeasuresistheprimarycon-tributortothenetincomeproÞle.Table3presentstheproÞlesofthefouraccrualmeasures,withlevelsofasset-scaledaccrualsshowninPanelAandyear-to-yearchangesinPanelB.TheproÞleofdiscretionarycurrentaccrualsshowsthemostdramaticchange,suggestingmanipulationofcurrentaccrualsduringanewissue.DiscretionarycurrentaccrualsaresigniÞcantlypositive,monotonicallyrisingtoapeakintheo¤eringyearbeforedecreasingsigniÞ-cantlyinyears#2and#3.(Theyear-to-yearchangescannotbederivedfromthemeanlevelsinPanelAbecausethenumberofissuersvariesinoursampleeachyear.)Theyear0peakinasset-scaleddiscretionarycurrentaccrualsisstatisticallysigniÞcantatameanandmedianof5.59%and2.50%.6Inpost-issueyears#1through#3,thediscretionaryaccrualsdeclinemonotonicallyuntilyear#3,whentheyarenolongerstatisticallysigniÞcantlydi¤erentfromzero.ThenondiscretionarycurrentaccrualsshowasomewhatsimilarproÞle.ThenondiscretionarycurrentaccrualspeakintheissueyearanddeclinesigniÞcantlyinyear#1.Nondiscretionarycurrentaccrualsareapositivelinearfunctionofsalesgrowth(seetheAppendix),sotheevidenceisconsistentwithnewissuerstimingo¤eringsforwhensalesgrowthpeaks.Thepre-issuemeanandmedianchangesinPanelB,however,areusuallynegative,andsodonotsuggestamonotonicimprovementpriortotheo¤ering.AsreportedlaterinTable7,pre-issuenondiscretionarycurrentaccrualsdonotpredictpost-issueunderper-formance.6Thelevelofaccrualsdonotturnnegativeimmediatelyaftertheo¤ering,suggestingthatissuersavoidimmediatereversalsinaccruals.Asimilarpeakinyear0wasreportedforinitialpublico¤eringsbyTeohetal.(forthcomingb).Theysuggestinstitutionalexplanationssuchasthethreatoflawsuitsifreversalsoccurimmediately,commitmentbyunderwriterstostabilizepriceneartheo¤eringprice,andtheexistenceoflock-upperiodswheninsiderscommitnottosell.Interestingly,LoughranandRitter(1995)andSection5ofthispaperdocumentthatthedeclineinstockreturnperformancealsodoesnotoccurimmediatelyaftertheo¤ering. 74S.H.Teoh,etal./JournalofFinancialEconomics50(1998)63Ð99Table3Time-seriesproÞleofasset-scaledaccruals,inpercent,fromyear!3to#3relativetotheseasonalequityo¤ering(year0).Thistablepresentsthediscretionaryandnondiscretionarycurrentandlong-termaccrualsofÞrmso¤eringseasonedequityo¤eringsfromthethreeyearsbeforetothreeyearsaftertheo¤ering.Thenondiscretionaryaccrualsreßectaccrualschoiceslargelydictatedbyeconomicconditions,whereasthediscretionaryaccrualsaredesignedtopickupreportingchoicesthatarelargelycontrolled(ÔmanagedÕ)bytheÞrm.Theaccrualsmeasuresarescaledbybeginning-periodtotalassets,andreportedinpercent.InthelastthreerowsofPanelsAandB,analternativematched-pairmethodisused.ThediscretionarycurrentaccrualsarecalculatedasthemodiÞedJonesmodeldiscretionarycurrentaccrualsoftheissuerminusthemodiÞedJonesmodeldiscretionarycurrentaccrualsofthematchednon-issuingindustrypeer.Thematchednon-issuerhassimilarnetincomeperformanceastheissuerinyear!1andisselectedusingthematchingprocedureasdescribedforthethirdnetincomeperformancemeasureinTable2.SeetheAppendixfordetailsofthemodeltodecomposeaccrualsintodiscretionaryandnon-discretionarycomponents.FiscalYear!3!2!10#1#2#3PanelA:Accruals(levels)Discretionarycurrentaccruals(DCA)Median0.90!1.30!2.05!2.50!2.20!0.70!0.10Mean2.21"3.32!5.37!5.59!4.18!1.59#!0.24N863102012481234118311221064Discretionarylong-termaccruals(D¸A)Median!1.10!!1.00!!1.00"!1.20!!1.00!!1.30!!1.50!Mean!1.00!1.70!!0.83!1.33!!1.51!!3.33!!1.86!N857101212411218117511031054Nondiscretionarycurrentaccruals(NDCA)Median0.90!1.40!1.50!2.20!1.20!0.70!0.80!Mean2.59!3.80!4.95!5.98!2.24!1.76!2.06!N863102012481234118311221064Nondiscretionarylong-termaccruals(ND¸A)Median!3.70!!4.20!!4.70!!4.60!!4.30!!4.20!!4.10!Mean!4.49!!5.15!!6.80!!6.32!!5.54!!4.52!!5.24!N857101212411218117511031054Discretionarycurrentaccruals(DCA)ofIssuerÐDCAofmatchednon-issuerMedian0.410.86"1.64"2.85"1.420.180.26Mean0.601.83#3.90"4.90"2.01"!0.61!0.89N773954124811541017900797PanelB:Accruals(changes)FiscalYear!2!10#1#2#3Discretionarycurrentaccruals(DCA)Median0.250.400.70!0.45#!1.20!!1.10!Mean0.890.620.30!1.37!2.62!!1.98"N86210171228117611111057 S.H.Teoh,etal./JournalofFinancialEconomics50(1998)63Ð9975Table3.Continued.FiscalYear!2!10#1#2#3Discretionarylong-termaccruals(D¸A)Median0.000.200.200.00!0.30"0.10Mean!0.780.86!0.61!0.21!1.82!1.23N85310051206116310971040Nondiscretionarycurrentaccruals(NDCA)Median0.15!0.30"0.20"!0.60!!0.50!0.00Mean!0.08!0.301.00!3.85!!0.500.14N86210171228117611111057Nondiscretionarylong-termaccruals(ND¸A)Median!0.20!0.40!!0.40"0.30!0.00"!0.20Mean!0.48!0.960.370.691.07!!0.51N85310051206116310971040ChangeinIssuer+sDCA-ChangeinMatchedNon-Issuer+sDCAMedian!0.01!0.000.89#!0.58!0.98"!0.57Mean0.781.012.13#!2.36"!1.96#!1.23N77295111481011886785!representstatisticalsigniÞcancelevelsatthe1%levels,usingt-testsforthemeanandWilcoxonp-valuesforthemedian."representstatisticalsigniÞcancelevelsatthe5%levels,usingt-testsforthemeanandWilcoxonp-valuesforthemedian.#representstatisticalsigniÞcancelevelsatthe10%levels,usingt-testsforthemeanandWilcoxonp-valuesforthemedian.NeitherPanelAnorPanelBshowsatime-seriespatternfullyconsistentwithmanagementofdiscretionarylong-termaccruals.PriorresearchbyKreutzfeldtandWallace(1986)andGuenther(1994)hasalsofoundthatlong-termaccrualsarelesssubjecttomanipulationbymanagers.Perhapsmoreleadtimeisrequiredtochangelong-termaccrualsthanisavailablebeforeanequityo¤ering.Inaddition,issuersmaybemorereluctanttomanipulatelong-termaccrualsbecausetheyaremorevisiblethancurrentaccruals.Finally,themeanandmediannondiscretionarylong-termaccrualsarenegativeinallyears,aresultconsistentwithalargedepreciationcomponent.Theyear-to-yearchangesdonotshowanyremarkablepattern,andsodonotcontributetothehumppatternfornetincome.Becausediscretionarycurrentaccrualsappeartobemanaged,wereportanalternativemeasurefordiscretionarycurrentaccrualsasarobustnesscheck.InthebottompanelentriesinPanelsAandBofTable3,thediscretionarycurrentaccrualsofmatchedÞrmsaresubtractedfromtheissuerÕsdiscretionarycurrentaccrualstoremovepotentialbiasintheJonesmodelforhigh-performanceÞrms.(ThematchedÞrmsareselectedasdiscussedinSection3.1forTable2.)The 76S.H.Teoh,etal./JournalofFinancialEconomics50(1998)63Ð99previouslyreportedhumpshapeisrobustwithrespecttothisalternativemeasure.Asbefore,themediansandmeansshowapeakinyear0andsigniÞcantpost-issuedeclinesbyyear#2.Thisevidencesuggeststhatthehumpshape(risingdiscretionarycurrentaccrualsuptoapeakinyear0followedbyasubsequentdecline)ismoreexaggeratedforissuersthanforcomparablenon-issuers.Overall,theevidencesuggeststhatnetincomeperformanceduringanewissueisdrivenlargelybydiscretionarycurrentaccruals,andnotcashßowfromoperations.4.2.Predictingpost-issuenetincomeunderperformancewithdiscretionaryaccrualsincross-sectionToevaluatewhetherpre-issuediscretionaryaccrualspredictpost-issuenetincomeunderperformance,werankissuersbytheirpre-issuediscretionarycurrentaccrualstoexaminedi¤erencesintheirpost-issuenetincomeperfor-manceincross-section.Table4examinestheperformanceoftheÔaggressiveÕquartilewiththehighestdiscretionarycurrentaccrualsandtheÔconservativeÕquartilewiththelowestdiscretionarycurrentaccruals.Bothextremequartilesofissuersperformwellimmediatelypriortotheissueyearandpoorlyimmediatelyafter.However,di¤erencesbetweenquartilesemergebyyear#3whenconser-vativequartileissuersbutnotaggressivequartileissuersreportsigniÞcantimprovement.Intheissueyear,themedianandmeangrowthinnetincomeis1.53%and4.72%fortheaggressiveissuersand1.56%and3.26%fortheconservativeissuers.Byyear#3,themedianandmeangrowthis!1.52%and!1.39%fortheaggressivequartilebutisasigniÞcant0.50%and13.24%fortheconservativequartile.Forasummarymeasureoflong-termperformance,wecomputeanabnormalcumulativenetincomeoverthethreepost-issueyearsrelativetobaseyear!1,asfollows:3NINININI+i,t!i,~1!m,t!m,~1,A¹A¹ABA¹A¹ABt/1i,t~1i,~2m,t~1m,~2whereNIisnetincomeinÞscalyeart,¹AistotalassetsinÞscalyeart,andi,ti,tiandmdenotetheissuerandmatchedÞrm.ThebottompanelofTable4reportsthataggressivequartileissuersunderperformmatchednon-issuersbyamedianandmeanof!7.50%and!6.29%overthethree-yearpost-issueperiod,whereasconservativeissuersoutperformtheirmatchesbyatotalmedianandmeanof0.99%and5.27%inthesameperiod.Thus,discretionarycurrentaccrualspredictpost-issueperformance;aggressivequartileissuersunderper-formconservativeissuersbyasmuchasamedianandmeanof8.49%and11.56%overthethreeyearsaftertheissueyear. S.H.Teoh,etal./JournalofFinancialEconomics50(1998)63Ð9977Table4Asset-scaledchangesinnetincomeandcashßowfromoperations,inpercent,ofaggressiveandconservativepre-issueaccruerquartilesfromyear!3to#3relativetotheseasonedequityo¤ering(year0)Thistablepresentstheaccountingperformanceoftheextremequartilesofissuersfromyear!3toyear#3relativetotheo¤eringyear.AggressiveandconservativequartilescontainÞrmswiththelargestandsmallestyear!1discretionarycurrentaccruals,respectively.ThetoppanelreportstheissuersÕyear-to-yearchangeinasset-scalednetincomeorcashßowfromoperationsrelativetoasampleofmatchednon-issuingindustrypeerswiththeclosestasset-scalednetincometotheissuerinthepre-o¤eringÞscalyear.SeedescriptionforthethirdperformancemeasureinTable2.Thebottompanelreportsthecumulativeabnormalperformance-matchednetincomerelativetobaseyear!1,calculatedas3NINININI+Ai,t!i,~1B!Am,t!m,~1B,¹A¹A¹A¹At/1i,t~1i,~2m,t~1m,~2whereNIisnetincomeinÞscalyeart,¹AistotalassetsinÞscalyeartandiandmdenotethei,ti,tissuerandmatchedÞrmrespectively.EventperiodDNIA!DNIADCFA!DCFAt~1,tit~1,tmt~1,tit~1,tmMedian/mean/observationsMedian/mean/observationsAggressiveConservativeAggressiveConservative(!3,!2)1.74#0.46!1.19!1.201.72!2.03!0.61!2.37159173159173(!2,!1)!1.40#0.53!11.73!12.02!!2.34!0.91!11.61!14.73!218222218222(!1,0)1.53!1.56!9.56!!8.31!4.72!3.26#8.16#!16.18!284284283284(0,1)!2.36!!3.39"1.99!2.54!4.53!!3.42!9.91!!1.07253253249253(1,2)!1.25"!0.62!0.101.77#!0.82!1.301.353.33230214227213(2,3)!1.520.50!1.140.22!1.3913.24!3.252.47203189201188Performance-matchedabnormalcumulative(overyears#1to#3)netincomelevels,inpercentAggressiveConservativeMedian!7.50"0.99Mean!6.29#5.27!representstatisticalsigniÞcancelevelsatthe1%levels."representstatisticalsigniÞcancelevelsatthe5%levels.#representstatisticalsigniÞcancelevelsatthe10%levels. 78S.H.Teoh,etal./JournalofFinancialEconomics50(1998)63Ð99Asbefore,weevaluatethecontributionofcashßowchangestothenetincomeunderperformance.Theevidencesuggeststhatcashßowsarenotresponsibleforthepoorerpost-issuenetincomeperformanceofaggressivequartiles.Infact,aggressivequartileissuersexperiencegreatercashßowimprovementbeginninginyear!1throughyear#3.Perhapsaggressivepre-issueaccrualmanipula-torsalsomanipulatecashßowfromoperationsthroughrealchangesinopera-tions.InTable5,wereportSpearmanrankcorrelationsbetweendiscretionaryaccrualsinyear!1withchangesinnetincomerelativetobaseyear!1foreachsubsequentyearfrom0to#3.Inthesecorrelations,weconsiderbothindustry-adjustednetincomeandperformance-adjustednetincome,tochecktherobustnessofourresults.Table5showsthatthecorrelationsofdiscretion-arycurrentaccrualswiththechangeinindustry-adjustednetincomerangebetween!20%and!15%,andareallnegativeandstatisticallysigniÞcantatthe1%level.Theperformance-adjustednetincomecorrelationsyieldsimilarstatisticallysigniÞcantÞndings,withtheresultsrangingbetween!16%and!6%.Thecorrelationsusingdiscretionarylong-termaccrualsarealsoallnegative,thoughsmallerinabsolutevalueandlesssigniÞcantthancorrelationsusingdiscretionarycurrentaccruals.Thus,onlydiscretionarycurrentaccruals,andnotdiscretionarylong-termaccruals,predictpost-issuenetincomeunder-performance.Table5Spearmanrankordercorrelationsofdiscretionarycurrentandlong-termpre-issueaccrualswithindustry-adjustedandperformance-matchedpost-issuenetincomeperformancePanelAliststheSpearmanrankcorrelationsbetweendiscretionarycurrentandlong-termaccrualsinyear!1relativetotheo¤eringyearwithchangeinindustry-adjustedasset-scalednetincomeinyears0through3relativetobaseyear!1.PanelBliststheSpearmanrankcorrelationswiththechangeinperformance-matchednetincomemeasure(thirdearningsperformancemeasureinTable2)inyears0through3relativetobaseyear!1.DCAarediscretionarycurrentaccruals,~1andD¸Aarediscretionarylong-termaccruals.~1DNIDNIDNIDNI~1,00~1,11~1,22~1,33PanelA:Spearmancorrelationswithindustry-adjustedasset-scalednetincomeDCA!0.150!!0.201!!0.196!!0.200!~1D¸A!0.055!0.051!0.036!0.062"~1PanelB:Spearmancorrelationswithperformance-matchedasset-scalednetincomeDCA!0.076!!0.065!!0.166!!0.133!~1D¸A!0.011!0.081!!0.058!0.065~1!representstatisticalsigniÞcancelevelsatthe1%levels."representstatisticalsigniÞcancelevelsatthe5%levels. S.H.Teoh,etal./JournalofFinancialEconomics50(1998)63Ð9979Overall,ourresultsareconsistentwiththefollowingscenario.Althoughcashßowsfromoperationsaredecliningpriortotheo¤ering,managersofissuingÞrmsreporthighandimprovingearningsbymanagingdiscretionarycurrentaccruals.Intheyearoftheo¤ering,reportedearningspeakdespiterelativelyweakcashßowfromoperationsbecausemanagerscontinuetotakelargepositivediscretionarycurrentaccruals.Post-o¤ering,highnetincomecannotbesustainedbecausecashßowsfromoperationsdonotimprovesu¦cientlyandissuerscannolongercontinuetotakelargediscretionaryaccruals.Issuerswiththehighestdiscretionarycurrentaccrualsinyear!1experiencethelargestdropinnetincomeaftertheissue.5.Predictingpost-issuestockreturnswithpre-issueaccrualsWenextexaminewhetherpre-issuediscretionaryaccrualspredictpost-issuestockreturnunderperformance.Addressingthistopicrequiresanappropriatemeasureforexpectedlong-runreturns,anissuethatisdebatedintheassetpricingliterature.Weusethreelong-runreturnmeasures:rawreturns,returnsnetofthereturnstothevalue-weightedmarketportfolio,andreturnsnetoftheFamaandFrench(1997)three-factormodelforexpectedreturns(describedintheAppendix).5.1.Post-issuereturnsbypre-issueaccrualquartilesWestudytherelationbetweenpre-issueaccrualsandpost-issuereturnsbyÞrstexaminingdi¤erencesinstockreturnperformanceamongfourquartileportfoliosgroupedbylevelsofpre-issuediscretionarycurrentaccruals.Eachquartileportfoliocontainsabout200Þrms.WethentrackeachportfolioÕsreturnperformancerelativetomonth0,whichiseitherthemonthoftheissueorfourmonthsafterthepreviousÞscalyearend,whicheverislater.Thefour-monthlagrepresentsatradeo¤:usingaccountinginformationwithshorterlagsmightmeanthatÞnancialstatementsarenotyetavailabletoinvestors,whilelongerlagsmightnotcapturetheperiodwheninvestorsreacttothereportcontainingmanipulatedearnings.Tochecktherobustnessofourresults,thepanelregres-sionsinSection6extendthewaitingperiodtosixmonths.Thelong-runreturnperformancereportedinTable6ismeasuredinthefollowingway.ForeachÞrmandyear,weÞrstcompoundmonthlyreturnsintoanannualreturn.WethenaveragethesereturnsacrossallsampleÞrmsintheportfoliotocomputetheoverallannualrawreturns.Fortheadjustedreturns,wesubtractthecompoundannualreturnonabenchmarkportfolio(eitherthemarketportfoliooraFamaandFrenchequivalentportfolio,describedintheAppendix)fromthecompoundannualrawreturnforeachÞrm.WhenasampleÞrmdisappearsduringtheyear,theremainingmonthlyraworadjustedreturns 80Table6Post-issuelong-runstockreturns,inpercent,bypre-issuediscretionarycurrentaccruals(DCA)quartilesS.H.Teoh,etal.~1AnnualReturnsarecomputedas1Nmfmf+<(1#r)!<(1#a),NCi,ti,tDi/1t/m1t/m1/whereNisthenumberofÞrmsineachquartilewithvalidreturndataintheÞrstmonthofeachyear,risthemonthlyreturnonsecurityiinmontht,aisJournalofFinancialEconomics50(1998)63i,ti,tanequivalentadjustmentreturn(identically0intherawreturnspart),andmistheÞrstmonthandm(nottoexceedm)isthelastmonthwithvalid1&%returndata.Forinclusion,returndatamustbeavailableintheÞrstmonthm.WhenasampleÞrmdropsfromCRSP,theremainingmonthsÕexcess1returnsarepresumedtobezero.Compoundreturnsarecomputedfromthereportedannualreturnsinthecolumntotheimmediateleft.Thus,ine¤ect,eachportfolioisassumedtoberebalancedonceayear.Allreturnsarereportedinpercent.mmRawreturnsMarket-adjustedreturnsFamaÐFrenchadjustedreturns1%ConservativeAggressiveConservativeAggressiveConservativeAggressiveAnnualCompoundAnnualCompoundAnnualCompoundAnnualCompoundAnnualCompoundAnnualCompound!12!179.5076.0757.2553.46!6.6648.2501113.2513.256.566.560.900.90!6.08!6.085.565.56!6.95!6.9512233.4617.170.957.57!10.99!10.19!13.47!18.73!8.20!3.10!8.75!15.09243517.0337.126.9315.03!0.94!11.03!10.90!27.593.580.37!10.04!23.62Ð9936479.7250.45!4.609.74!4.66!15.18!17.93!40.572.903.28!11.80!32.63485910.1565.727.4117.87!2.12!16.98!5.65!43.933.737.14!4.16!35.43 S.H.Teoh,etal./JournalofFinancialEconomics50(1998)63Ð9981areassumedtobezerountiltheendoftheyear.(SuchÞrmsdropfromourportfoliointhefollowingyear.)Finally,wecompoundtheannualrawandadjustedreturnaveragesintoÞve-yearcumulativereturns.Thisproceduremimicksatradingstrategythatrebalancestheportfolioannually,assigningequalweighttothosestocksstillinexistence.Table6showsthatourstrategynetsabouta66%rawreturnoverÞveyearsforÞrmsintheconservativeearningsmanagementquartile,and18%forÞrmsintheaggressivequartile.Whentheequivalentmarket-returnsaresubtracted,theconservativequartileportfolioearns!17%excessreturnsandtheaggres-sivequartileportfolioearns!44%.AmongÞrmswithsu¦cientpre-issuereturndatatocomputeFamaÐFrenchexposures,theconservativequartileportfolioearnsa#7%FamaÐFrenchadjustedreturnandtheaggressivequartileportfolioearns!35%.Long-runreturnsaresensitivetothecomputationmethod(BarberandLyon,1997;KothariandWarner,1997).WhenweÞrstcomputeanacross-ÞrmaveragereturnineachmonthandthencompoundoverÞveyears,theconservativeandaggressivequartilesnet83%and26%,respectively.(Theequivalentmarketreturnsareabout115%,indicatingexcessreturnsof!30%and!90%,respectively;theequivalentFamaÐFrenchreturnsare103%and158%,respec-tively,indicatingexcessreturnsof!20%and!130%,respectively.)Whenwecomputeanacross-ÞrmexcessreturneachmonthandeithercompoundoraveragethisoverÞveyears,weÞndÞve-yearexcessreturnsof!5%to!15%fortheconservativequartileand!45%to!65%fortheaggressivequartile.ThereturnsinFig.1arecomputedusingthislastmethod,whichisequivalenttomonthlyrebalancingtheportfolio.Annualorevenlonger-horizonrebalancingsu¤ersfromthepresenceofmanystaleÞrms(withoutreturns).Inoursample,althoughportfoliosthatareneverrebalancedperformconsiderablyworsethanportfoliosthatarerebalanced,thedi¤erencebetweenaggressiveandconservativequartileÞrmsisnotsensitivetorebalancing.TheresultsinFig.1showthattheaggressivequartileÞrmshavemarginallyhigherpre-o¤eringreturnperformancethanconservativeissuers.Thesmalldi¤erenceindicatesthatitisunlikelythatourresultsaredrivenbythewinner/loserreversalphenomenon.Aftertheo¤ering,conservativeissuersunderperformonlymarginally(!7%),whileaggressiveissuersunderperformdramatically(!48%).BecauseoverlappingmultiyearÞrmreturnsarejointlyexposedtocontempor-aneousindustryshocks,wecannotcomputecross-sectionalstandarderrorstoassessthestatisticalsigniÞcanceofthesereturns.Instead,wecomputethestandarddeviationfromthetime-seriesrealizationsofeachquartileÕsportfolio.Thisassumesthatthereisnocross-sectionalpredictabilityacrosstime,i.e.,thereturnrealizationofÞrmXinmonthtdoesnothelppredictthestockreturnrealizationofÞrmYinmonthst#1,t#2,etc.Onaverage,theconservativequartileportfolioshowsanaveragemonthlyreturnof!0.165%,witha 82S.H.Teoh,etal./JournalofFinancialEconomics50(1998)63Ð99Fig.1.Time-seriesgraphofFamaÐFrenchadjustedreturnsclassiÞedbypre-issuediscretionarycurrentaccruals(DCA)quartiles.AnaveragemonthlyexcessreturnisconstructedbyÞrst~1subtractinganequivalentFamaÐFrenchbenchmarkreturnfromeachissuerÕsmonthlyreturnandthenaveragingtheseindividualÞrmexcessreturnsacrossallÞrmsintheportfolio.Thegraphedreturnsaretheloggedcumulativesumofthesemonthlyportfolioexcessreturns.Thereturnsarenormalizedsothattheevent-monthreturniszero.FirmsareclassiÞedintothefourquartileportfoliosusingDCA,thediscretionarycurrentaccrualsintheÞscalyearpriortotheseasoned~1equityo¤ering.Timeismeasuredfromthedateoftheseasonedequityo¤eringorfourmonthsafterthepriorÞscalyearend(whereDCAwasreported),whichevercomeslater.~1t-statisticof!0.96.Theaggressivequartileportfolioshowsanaveragemonthlyreturnof!1.346%,withat-statisticof!7.00.Theaveragereturndi¤erenceof1.18%permonthhasat-statisticof4.60,indicatingthatmoreaggressiveearningsmanagementpredictspoorerpost-issuereturnperformance.Insum,thepartitionedunivariateevidencesuggestsalargelong-runreturndi¤erencebetweenconservativeandaggressiveÞrms.Itthusappearsthatpoorpost-issueperformancecanbeexplainedpartiallybythepre-issueearningsmanagementofseasonednewissuers.BychoosingÞrmswithnegativeorlowdiscretionarycurrentaccruals,investorscanavoidinvestinginissuersthatdramaticallyunderperformtheirnon-issuingpeers. S.H.Teoh,etal./JournalofFinancialEconomics50(1998)63Ð99835.2.Regressionsofpost-issuereturnsonpre-issueaccrualsTable7displaystheresultsfromordinaryleastsquaresregressionsofpost-issueÞrmstockpriceperformanceonpre-issueaccountingaccruals.Thedependentvariableisthelogofthefour-yearcompoundedstockreturn(orcompoundedexcessstockreturn),beginningeitherfromtheissuedateorfourmonthsafterthepreviousÞscalyear,whichevercomeslater.Toavoidinßuentialeccentricobservations,wewinsorizeaccrualsatthe1%and99%percentiles.Ingeneral,ourresultsarerobusttowinsorization.Thediscretionaryaccrualscomponentsarethekeyexplanatoryvariablesofinterest.Weincludethenondiscretionaryaccrualscomponentsintheregressiontoevaluatetherelativeinformationcontentforreturnsbetweenthediscretion-aryandnondiscretionarycomponents.Wealsoincludeasetofindustryandyearcontroldummies(coe¦cientsarenotreported).Theindustrydummies,asoutlinedinTable1,accountforpost-issueperformancevarianceacrossindus-tries.Interceptdummiesforindividualyears1978through1989accountforbusinesscyclee¤ectsandcapturecontemporaneouscross-sectionalcorrelationbetweenfour-yearreturns.Logequity-sizeandlogbook-to-marketvariablescontrolforÞrmcharacteristics.Multi-yearreturnsofdi¤erentstocksoverlapacrossÞrms.(Wedonothaveduplicatereturns.)Contemporaneouscross-sectionalreturnscancontainspuriousresidualcorrelation,whichdoesnotbiascoe¦cientestimatesbutcouldbiascoe¦cientstandarderrorsiftheinducednonzeroo¤-diagonalcovariancescorrelatewithourmeasureofdiscretionarycurrentaccruals.Section6implementsamorecomplexpaneldatatestproced-urethattakesthecross-sectionalcorrelationsintheresidualsintoaccount.Finally,wedonotreportresultsforFamaÐFrenchadjustedreturns,becausetheyaresimilartothetworeportedregressions.WealsoincludeavariantofCheng(1995)measureofuseofproceeds.ChengÞndsthatequityissuersthatdonotinvestunderperformaftertheissue,whereasissuersthatinvestdonotunderperform.WeuseChengÕsmeasuretoexaminewhethertheearningsmanagementproxyinthispaperhasanincrementale¤ectonreturnsovertheChenge¤ect.Thecapitalexpendituregrowthbetweenpre-andpost-issueperiodsiscalculatedasDCAPEXPt`1(CAPEXP#CAPEXP)!(CAPEXP#CAPEXP)"tt`1t~1t~2,(2)2TAt~1whereCAPEXPistheissuerÕscapitalexpenditure(Compustatitem128)andtTAistheÞrmÕstotalassetsinyeart.Year0dataarefromtheÞnancialtstatementsfollowingtheissue.Consequently,DCAPEXPincorporatestwot`1numbersnotavailableatthetimeoftheseasonedo¤ering,andsohasatwo-yeartimingadvantageoverouraccrualsmeasures. 84S.H.Teoh,etal./JournalofFinancialEconomics50(1998)63Ð99Table7Ordinaryleast-squaresregressionspredictingfour-yearpost-issuereturnswithpre-issueaccrualsandcontrolsThedependentvariableisafour-yearaftermarketlogreturn,beginningattheo¤eringdateorfourmonthsafterthepreviousÞscalyear-end,whichevercomeslater.Formarket-adjustedreturns,monthlyrawreturnsareÞrstadjustedbysubtractingthemarketreturn,andthencontinuouslycompoundedforfouryears.Theindependentaccrualvariables(DCAthroughND¸A)are~1~1computedfromregressions(describedintheAppendix)andmeasuredintheÞscalyearprecedingtheissue(subscript!1).DCAPEXPisthesumofcapitalexpendituresinthe(post-issue)eventandt`1followingÞscalyearminusthesumofcapitalexpendituresinthetwoyearspriortotheevent,dividedbytwicetotalassetsintheyearpriortotheseasonedo¤ering.Thus,DCAPEXPusesinformationt`1fromtwoÞnancialstatementsnotavailabletotheseasonedissuepurchasers,unliketheaccrualsmeasureswhichrelyonlyoninformationknowntoinvestorsatthetimeoftheissue.Toadjustforsomecross-sectionalcontemporaneouscorrelationsbetweencomponentsofthecompoundedreturns,weincludebutdonotreportacompletesetofindustryandyeardummies(twoyeardummiesfor1983),aswellasÞrmsÕlogbook-marketvalueandlogequitysize.SmallandlargeÞrmsarethesmallestandlargestmarketcapitalizationtertials.White-tstatisticsarereportedinparentheses.IndependentvariableRawMarket-adj.Market-adj.returnsreturnsreturnsSmallÞrmsLargeÞrmsDiscretionaryCoef!0.3818!!0.3954!!0.5413!!0.1962currentaccruals(DCA)(t)(!2.61)(!2.68)(!2.51)(!0.92)~1DiscretionaryCoef!0.3620!0.3665!1.485!!0.1962long-termaccruals(D¸A)(t)(!1.28)(!1.26)(!3.20)(!0.33)~1NondiscretionaryCoef!0.1141!0.1200!0.1390!0.1377!currentaccruals(NDCA)(t)(!0.88)(!0.89)(!0.39)(!3.49)~1NondiscretionaryCoef!0.0714!0.0652!0.0516!0.7914long-termaccruals(ND¸A)(t)(!0.20)(!0.18)(!0.09)(!1.30)~1ChangeincapitalCoef0.08290.08010.3034!!0.1206#expenditures(DCAPEXP)(t)(1.59)(1.53)(3.30)(!1.65)t`1IndustrydummiesFullsetÐnotreportedYeardummiesFullsetÐnotreportedBook/Market,MarketvalueNotreportedN1,0351,035332355R215.89%13.73%27.45%20.47%R212.59%10.35%17.76%10.62%!representstatisticalsigniÞcancelevelsatthe1%level."representstatisticalsigniÞcancelevelsatthe5%level.#representstatisticalsigniÞcancelevelsatthe10%level.Theregressionresultsindicatethatamongthefouraccrualsmeasures,onlythediscretionarycurrentaccrualisstatisticallysigniÞcant.Itscoe¦cientisabout!0.4,itst-statisticisabout!2.6.Thisimpliesthatissuerswithaggres-sivelevelsofpre-issuediscretionarycurrentaccrualsperformsigniÞcantlyworse S.H.Teoh,etal./JournalofFinancialEconomics50(1998)63Ð9985aftertheo¤ering.Thecoe¦cientondiscretionarylong-termaccrualsisnegative,butnotstatisticallysigniÞcant.Thecoe¦cientsonthetwonondiscretionaryaccrualsarealsoinsigniÞcant,suggestingthatonlytheactivelymanageablecomponentofearningspredictsfuturereturns.AsforeconomicsigniÞcance,itisnotsurprisingthatindividualÞrmlong-runstockreturnsarehighlyvariableanddi¦culttoexplainincross-section.SpeciÞcally,thecross-sectionalstandard-deviationoffour-yearlogreturnsis71%forthe1035Þrmsinourregression.Becausethestandarddeviationofdiscretionarycurrentaccrualsinourregressionsis0.200,acoe¦cientof!0.4indicatesthataone-standard-deviationdi¤erenceinDCAexplainsafour-~1yearlog-returndi¤erentialofabout8%.Thus,onestandarddeviationinoursingleDCAvariableaccountsformorethan11%ofthefour-yearcross-~1sectionalstandarddeviationinlog-returns(71%).Adiagnosticcheckonthelinearityofourregressionisthefractionoftheactualamountofextremequartilereturnsthattheregressionpredicted.Divid-ingthe1035ÞrmsintheregressionsampleintofourDCAportfolios,weÞnd~1thatthemostaggressiveaccrualÞrmshaveapost-issuemeanlog-returnof!19.6%(market-adjusted!44.0%).Theexplanatoryvariablesfromourregressionpredicta!16.5%(!41.6%)post-issuelogreturnforthisgroup.Thus,theresidualunexplainedmeanisonly!3.1%forrawreturnsand!2.4%formarket-adjustedreturns.EquivalentcomputationsforthemostconservativeaccrualÞrmsindicateapost-issuereturnof!0.1%forrawreturnsand!25.6%formarket-adjustedreturns,mostofwhichiscapturedbytheregression.ThetworightmostcolumnsinTable7reproducethemarket-adjustedregres-sionsrunonthetwoextremeportfoliosgroupedbymarketcapitalization.Weignorethemiddlethirdtomaximizedi¤erencesinÞrmsizebetweenthetwoextremegroups.Iftheearningsmanagementhypothesisholds,weexpectpre-dictableperformancedi¤erencestobemorepronouncedintheregressionforsmallÞrmsthanforlargeÞrms.LowertransactioncostsforlargeÞrmsallowinvestorstotakeadvantageofthereturnanomalyfortheseÞrms.Thedis-cretionarycurrentanddiscretionarylong-termaccrualcoe¦cientsareindeedreliablynegativeonlyinthesmall-Þrmregressions.However,withonly332observationsineachportfolio,thecoe¦cientdi¤erencesareonlysuggestiveandshouldbeinterpretedwithcaution.Nondiscretionaryaccruals(bothcurrentandlong-term),hypothesizedtohavelittlepredictivepower,areinsigniÞcantforsmallÞrms.AsurpriseÞndingisthesigniÞcanceofnondiscretionarycurrentaccrualsforlargeÞrms.Thecorrelationoffuturereturnswithpre-issuesalesgrowthmightbepresentonlyinlargeÞrms.ChengÕscapitalexpendituremeasureispositiveandsigniÞcantonlyamongsmallÞrmsandissensitivetothemethodusedtocalculateabnormalreturns.ThepredictedpositiverelationisnotobservedinlargeÞrms,againconÞrmingthatpricinganomaliespersistwhenabnormalproÞtopportunitiescannotbeeasilyarbitragedaway. 86S.H.Teoh,etal./JournalofFinancialEconomics50(1998)63Ð99Insum,ourevidenceofpost-issuereturndi¤erencesincross-sectionisconsistentwithanearningsmanagementscenario.Rangan(1997)conÞrmsthisÞndingusingquarterlyaccruals.Amongourfouraccrualmeasures,discretion-ary(i.e.,managed)currentaccrualspredictsubsequentpoorstockpriceperfor-mancethebest.Thefactthatdiscretionarycurrentaccrualsareagoodpredictorisespeciallysurprisinginlightofthefactthattheyareanimperfectmeasureofearningsmanagementandcalculatedfrominformationavailableasearlyasfourtosixteenmonthsbeforetheissue.6.AFama±MacBethpanelprocedure6.1.MethodologyInthissection,weoutlineaprocedurethataddressestwopreviouslyneglectedissues.First,sincethepreviousregressionsuseoverlappingmultiyearreturns,theregressionerrorscouldbecorrelatedifallriskfactorshavenotbeenproperlyaccountedfor.(Theappropriaterisk-factoradjustmentsforex-pectedreturnsarecurrentlystilldebatedintheasset-pricingliterature,andaconsensushasyettoemerge.)Second,thesigniÞcanceoftheaccrualvariablesinthepreviousregressionsmayhavebeenduetotheabilityofdiscretionaryaccrualstoexplainsubsequentreturnsinallÞrms,notjustinperiodswhentherearenewissues.Thus,wewanttomeasuretheincrementalpredictivepowerofpost-issuereturnsbypre-issueaccrualspertainingtoperiodsofseasonedequityo¤erings.Toaddresstheaboveissues,weruncross-sectionalregressionsexplainingmonthlyreturnsfromJuly1975throughDecember1994withthefollowinglaggedvariables:(a)thelogoftheÞrmÕsbook-to-marketvalue,(b)thelogoftheÞrmÕsmarketvalueofequity,(c)thefouraccrualmeasuresdescribedearlier,and(d)fourinteractionvariables,accrual*SEOdummy,whicharethepre-issueaccrualmeasureduringseasonedequityo¤ering-relatedperiods,andzerootherwise.AllindependentvariablesarefromthesameÞscalyearandlagtheLHSreturnstheyseektoexplain.Followingthetraditionofusinglogar-ithmsfortheFamaÐFrenchvariables,thebook-to-marketandsizevariablesaretruncatedat0.0001.Accrualmeasuresarewinsorizedatthe5%and95%percentilestoavoidundueinßuenceofoutliers.Theresultsarerobusttowinsorizingatmoreextremepercentiles.Fig.2illustratesthetimeline,thelagstructurefortheindependentvariables,andthealgorithmforwhenthenewissuedummyissettoone.Wenowassumeanevenmoreconservativereportinglagofsixmonths(insteadofthefour-monthlagintheprevioussection).Thus,theregressionsrelatereturnstoaccrualsandcontrolsfromÞscalyearsendingatleastsixmonthspriortothe S.H.Teoh,etal./JournalofFinancialEconomics50(1998)63Ð9987Fig.2.Illustrationoftimealignmentinpanelregressions.ThisÞgureillustratestheregressionlead-lagstructurebetweenthedependentvariable,monthlyreturns,andthekeyindependentvariables,thefouraccrualmeasuresandtheirinteractivedummyvariables,accrual*SEO.Regres-sionsareperformedinmonthlycross-section,andoverallstatisticsareaggregatedoverallmonths.Wespecifyandaggregatecross-sectionalcoe¦cientsforfourdi¤erentsetsofregressions.IntheÔyear0Õregressions,thekeyindependentvariables,thefouraccrualmeasures,areobtainedfromtheÞscalyearending7to16monthspriortothemonthlyreturns.ForÔyear1Õregressions,theaccrualsarefromÞscalyearending17to28monthsprior,forÔyear2Õregressionsfrom29to42monthsprior,andforyear3regressionsfrom43to64monthsprior.Theinteractiveaccrual*SEOdummyturnsononlywhentheaccrualsarefromtheÞscalyearbeforetheseasonedissue(and,forÔyear0Õregressions,providedthedependentreturnobservationoccursaftertheseasonedequityissue).returns.7Fouralternativelagstructuresfortheindependentvariablesrelativetothereturnsareconsidered,fromonetofourÞscalyearspriortothereturns,afteraccountingforthereportinglag.Eachlagisconsideredseparatelyintheregressions.ThealternativeofusingonlyasinglesetofregressionsofmonthlyreturnsontheaccrualsfromallfourlaggedÞscalyearshassurvivorshipproblems.Thenamingconventionforthefoursetsofregressionsisyearn,wheren"0,1,2,and3,withn"0asthelowestlag.Thus,theyear0regressionsrelatereturnsinagivenmonthtoaccrualsandcontrolsfromÞscalyearsendingatleastsixmonthsandupto17monthsprior.Forexample,theregressionexplainingreturnsinJanuary1990woulduseaccrualsandcontrolsfromÞscal7FirmsarerequiredtoÞleform10-KreportingontheirannualÞnancialstatementswiththeSECwithin90daysoftheÞscalyearend.Alford,Jones,andZmijewski1994reportthat80%ofÞrmsÞletimelyreportsandonly2%ofÞrmsfailtoÞleafter150daysaftertheÞscalyearend. 88S.H.Teoh,etal./JournalofFinancialEconomics50(1998)63Ð99yearsendingnolaterthanJune1989.Equivalently,accrualsandcontrolsdatedJune1989wouldbeusedtoexplainreturnsineachmonthfromJanuary1990toDecember1990.Year1regressionsrelatereturnstoaccrualsandcontrolsfromÞscalyearsending18to29monthsearlier,year2regressionsrelatereturnstoaccrualsandcontrols30to41monthsearlier,andyear3regressionsrelatereturnstoaccrualsandcontrols42to53monthsearlier.Thus,wehaveatotaloffourregressionsetsofabout180monthlyregressionseach,witheachsetexplainingthepredictivepowerofaccrualsandcontrolsforvariouslead-periodreturns.Theyear0setofregressionspredictsreturnsfromJuly1975throughMay1991;theyear1setpredictsreturnsfromJune1976throughMay1992,andsoon.Ineachofthe180regressions,thetypicalnumberofÞrmsrangesbetween2700to3200observations.Theinteractiondummyforthepresenceofaseasonedequityo¤eringequalsoneonlyfortheÞscalyeardataimmediatelyprecedingtheo¤eringinallregressionsets.Consequently,thecoe¦cientsontheinteractionvariable,accrual*SEOdummy,measurethemarginalpredictiveabilityofpre-issueaccrualsforreturnsbetweenzeroandthreeyearsaftertheissue.Theinteractiondummyissettooneonlyaftertheo¤eringfortheyear0regressionset.Suppose,forexample,thenewissueoccursinFebruary1990andtheÞscalyearendsinJune.TheissuedummyissettooneinmonthlyreturnregressionsfromMarch1990toDecember1990onJune1989accrualsandothercontrolvariablesfortheyear0setofregressions.TheissuedummyisonealsointheotherthreesetsofregressionswheneverJune1989accrualsareused.Thus,theissuedummyisoneforthefollowingregressions:Jan-uary1991toDecember1991monthlyreturnregressionsonJune1989accrualsandcontrolsfortheyear1set,January1992toDecember1992monthlyreturnregressionsonJune1989accrualsandcontrolsfortheyear2set,andsoon.WeexcludemonthsintheperiodfromfourmonthsaftertheÞscalyearendtotheo¤eringdatefromtheregressions.ItwouldbemisleadingtoattributeexplanatorypowerinthesemonthsasaÔnon-issueÕe¤ect,justasitwouldbemisleadingtocallitÔpost-issueÕreturnpredictabilitywhentheequityo¤eringhasnotyetoccurred.Theresultsarerobusttoomittingorincludingthesemonths.Thetimeseriesoftheestimatedcoe¦cientsoneachindependentvariableareaveragedacrossthe180monthlyregressionsineachset,andanoverallt-statisticiscalculatedassumingserialindependence.Finally,agrandmeanandagrandt-statisticarecomputedacrossallfoursetsofregressionstoestimatetherelationbetweenaccrualsandfour-yearreturns.AttributingGaussianunit-normalitytotheaggregatedfour-yeart-statisticsine¤ectassumeszerocorrela-tioninthetimeseriesoftheaccrualsfortheissuers.Theobservedcorrelationsareoftheorderof5Ð10%,andthustheassumptionofuncorrelatedaccountingvariablesisunlikelytobeproblematic.But,towarnthereader,wehavebracketedthesigniÞcancelevelindicatorsinthetable. S.H.Teoh,etal./JournalofFinancialEconomics50(1998)63Ð9989Unliketheapproachintheprevioussection,inwhichmultimonthreturnsarecompoundedintolong-termholdingperiods,thecross-sectionalregressionsareperformedmonthly,sonooverlappingperiodsexisttoinducecross-sectionalcorrelations.InsteadoftheFamaÐFrenchfactorcoe¦cients,wenowusetheÞrmÕsownbook-to-marketandÞrmsizemeasurestocontrolforcross-sectionalvariationinexpectedreturns.8Becausetheexplanatoryvariablesareknownwhenreturnsaremeasured,thereisnoneedtorunpre-test-periodregressionstoestimateexante(beta)coe¦cientsorgroupÞrmsintoportfoliostoreducetheerrors-in-variablesproblem(asinFamaandMacBeth,1973).ByusingthisvariationoftheFamaÐFrenchmodel,wecanevaluatetherobustnessoftherelationbetweenaccrualsandfuturereturnswithrespecttoalternativemodelsofexpectedreturns.Finally,byperformingthecross-sectionalregres-siononallÞrms,withaccruals*SEOdummyvariablesasregressors,wecanestimatetheincrementalpredictabilityofpost-issuereturnsbypre-issueac-crualsbeyondanyaveragepredictabilityofaccrualsforfuturereturns.ThisallowsustoextendSloanÕs(1996)researchrelatingtotalaccrualstofuturereturnstoevaluatetherelativeimportanceofdiscretionaryversusnondis-cretionaryaccrualsandpre-issueperiodversusnon-issueperiodaccrualsinpredictingfuturereturns.6.2.ResultsTable8reportsaggregatemonthlyregressionresultsforthefoursetsofyear0throughyear3regressions,andoverallgrandmeanresultsoverthefoursetsofregressions.PanelAregressionsincludeonlybook-to-market,Þrmsize,andaccrual*SEOdummyasexplanatoryvariables.PanelBregressionsincludeadditionallythefouraccrualmeasurestocomparethepre-issueaccruals-returnrelationwiththeaccruals-returnrelationinthegeneralpopulation.Consistentwithotherresearch,Þrmsizeisonlymarginallyimportantinthesampleperiod,whilethebook-to-marketratioishighlysigniÞcant.PanelAshowsthatpre-issuediscretionarycurrentaccrualshaveasigniÞcantlynegativerelationwithstockreturnstwoandthreeyearsaftertheissue(t-statisticsof!2.94and!2.36).Inthetwoyearsimmediatelyfollowingtheissue,thecoe¦cientisalsonegativeandmarginallysigniÞcant(t-statisticsof!1.82and!1.61).Aggregatingoverthefouryears,theaveragecoe¦cientis!3.521with8FamaandFrench(1993,1997)arguethatthebook-to-market(HMLportfolioreturns)andmarket(SMBportfolioreturns)factorsdescribethecross-sectionalvariationinreturnsquitewell.However,DanielandTitman(1997)recommendusingtheÞrmÕsownbook-to-marketratiosandsizemeasuresinsteadoftheFamaÐFrenchfactorreturns.TheyreportthattheÞrmÕsownvariablesexplainthecross-sectionalvariationinreturnsbetterthantheFamaÐFrenchfactorreturns. 90S.H.Teoh,etal./JournalofFinancialEconomics50(1998)63Ð99!)!))"!))"2.33(4.37(1.491.55lcolumn!!!!0.1100.0998.34(3.5213.5180.1131.9912.11(CoefTÕs!!!!!""0.742.360.140.04-statisticsmaynotbeindependentt!!!!t-statisticisthesumofeachmonthÕs0.0590.0833.044.7570.934!!!!1.462.940.331.498.403!!!!-statistics.Thedependentvariablesarelogstockreturns,measuredint5dividedbytwo.Becausethefour0.0990.1253.982.8945.8175.222Ð!!!!statisticsinthelastcolumnsareinparentheses.t!"1.451.612.51.Theboldfacedlinescontainthestatisticsofmostinteresttous:theymeasurethe!!!years(plusanassumedreportinglagof6months)beforethemonthlyreturntheyseekto:AggregatedYear0to3½regressiondescribesthepredictivepoweroftheaccountingvariableintheleft-mostcolumnreported½½3.5030.1195.050.1044.270.6933.6228.1768.1414.6934.960.710.122.9207.5003.97!!!!t-statisticsfromcolumns2!!!FrenchvariablesandaccrualsforÞrmsissuingseasonedequityo¤erings-1.011.823.22!!!0.0825.7384.5091.7590.940.2115.495!!!Year0AverageCoefOverallAverageTYear1OverallAverageCoefOverallAverageT!OverallYear2CoefTYear35,andthesumofthefourÐ)0.1625.39yearsaftertheaccountingvariablewasY½~YY~Ystatisticdividedbythesquarerootofthenumberofmonthlyregressions.Theaggregatedcolumnsreporttheaverageofthefouraverage~Y~t)~Y~SEOSEOSEOExplainingreturnswithlaggedFama/MarketY*SEO**:~Y*Y~Y~~Y~Table8Time-seriesaveragesofmonthlycross-sectionalregressionsofreturnsonlaggedaccrualsandcontrolsforallÞrmsonCRSPwithsu¦cientdataTheindependentvariablesaremeasuredsimultaneouslyandexplain.SEOisaninteractiondummythatisonewhenimmediatepre-issueaccrualsareused(unlesstheissuehasnotyethappened)andzerootherwise.TimealignmentisillustratedinFig.2.Essentially,theyearonstockreturnsinßuenceofpre-issuediscretionarycurrentaccrualsonpost-issuereturns.Theothervariablesareasdescribedinprevioustables.Eachnumericarepresentstime-seriesstatisticsofabout180cross-sectionalregressioncoe¦cients/percent.Theprintedcoe¦cientisthetime-seriesaveragesofeachmonthÕscross-sectionalcoe¦cient.Thecross-sectionalcoe¦cientsfromcolumns2(althoughtheyarenothighlycorrelated),thesigniÞcancelevelstarsoftheVariablenameMonthlyreturnsareinyearPanelAConstantLog(BkLog(MarketDCADLANDCANDLA3.4314.58 S.H.Teoh,etal./JournalofFinancialEconomics50(1998)63Ð9991!)!)!))")"!)!)2.26(2.52(1.360.026.72(0.684.99(0.61!!!!!!!!3.3600.1069.23(0.0978.44(2.6982.5332.2080.8422.43(0.6991.593!!!!!!!!!0.681.920.220.990.061.690.200.098andaccrualsforallÞrms!!!!!!!,2.8920.0573.870.0803.014.5810.2396.6360.1332.260.3420.900!!!!!!!!"!!1.412.280.541.949.1191.613.210.551.261.3870.073!!!!!!!3.5460.0964.850.1234.042.4603.5002.9200.060.3472.0812.1881.100!!!!!!!!!!!1.430.301.573.780.122.880.10!!!!!!!accrualsforÞrmsissuingseasonedequityo¤erings,3.5390.1155.110.1034.348.7556.2994.3560.590.9380.280.4421.9878.3162.12!!!!!!!!!!!1.000.553.150.0266.732.19-Frenchvariables!!!!!0.0803.7785.6461.7651.531.9480.200.0681.2970.7530.430.340.66615.148!!!!!!)0.1575.49Y~YYY~Y~~)~Y/MarketY~SEOSEOSEOExplainingreturnswithlaggedFama~*SEO**:Y*Y~Y~YY~Y~~Y~Y~~PanelBConstantLog(BookLog(MarketDCADLANDCANDLADCA3.461DLANDCA4.63NDLA!representstatisticalsigniÞcancelevelsatthe1%levels.representstatisticalsigniÞcancelevelsatthe5%levels." 92S.H.Teoh,etal./JournalofFinancialEconomics50(1998)63Ð99at-statisticof!4.37,indicatingthathighpre-issuediscretionarycurrentaccrualspredictpoorstockreturnsduringthefour-yearperiodaftertheissue.ThesigniÞcanceofdiscretionarycurrentaccrualsissimilartothatofthebook-marketratio,andconsiderablyhigherthanthesigniÞcanceofÞrmsize.Incontrast,pre-issuenondiscretionarycurrentaccrualsaregenerallynotasigniÞ-cantpredictorofstockreturns,exceptinyear1.AsPanelBbelowshows,thisexplanatorypowerofpre-issuenondiscretionarycurrentaccrualsisgenerictoallÞrms,andnotspeciÞcallyrelatedtoseasonednewissues.Forthepre-issuelong-termaccruals,thecoe¦cientfordiscretionarylong-termaccrualsisnegativeandstatisticallysigniÞcantatthe1%levelonlyinthemonthsimmediatelyafterissue.Thiscoe¦cientisneithernegativenorsigniÞ-cantintheremainingthreeyears.Ontheotherhand,nondiscretionarylong-termaccrualshaveamoredelayedimpactonreturns;thecoe¦cientsfortheyear2andyear3regressionsareactuallypositiveandsigniÞcantataboutthe5%level.TheseresultssuggestthatÞrmswithpositivepre-issuenondiscretion-arylong-termaccrualsactuallyoutperformthemarketaftertheissue.PanelBcomparesthedi¤erenceinexplanatorypoweroftheaccrualvariablesbetweenissueperiodsandnon-issueperiods.Thecoe¦cientsonthefouraccrualvariableswithouttheinteractivedummiesarefornon-issuersinallmonthsandissuersinnon-issuemonths.Thecoe¦cientsonthefourinteractivedummyvariablesnowmeasuretheincrementalexplanatorypowerofpre-issueaccrualsaboveandbeyondtheexplanatorypowerofaccrualsinthegeneralpopulation.DiscretionarycurrentaccrualsappeartopredictreturnsforthegeneralpopulationofÞrmsintheshortterm.Thet-statisticsfortheyear0andyear1regressionsarehighlysigniÞcantat!6.73and!3.78,respectively.OfsigniÞcancetotheearningsmanagementhypothesis,theinteractionpre-issuediscretionarycurrentaccrualvariablehasasigniÞcantadditionale¤ectinyears2and3.Thecoe¦cientsfortheDCASEOvariablesareoneorderof~1*magnitudelargerthanthecoe¦cientsoftheDCAvariables(!2.46and~1!4.58,ascomparedwith!0.35and!0.13,respectively),andthet-statistic(!2.28)fortheyear2laggeddiscretionarycurrentaccrualsforissuersisstatisticallysigniÞcantatthe5%level.Ineconomicterms,the!2.5to!4.6magnitudeofthecoe¦cientsondiscretionaryaccrualssuggeststhataone-standard-deviationdi¤erenceindis-cretionarycurrentaccrualsbetweentwoissuers(about0.2)canexplaina0.5Ð0.9%monthlyreturndi¤erential.Aggregatingoverthefour-yearperiod,theDCASEOvariableisstatisticallysigniÞcantataboutthe1%level.Thus,~1*theevidenceindicatesthatthenegativerelationbetweendiscretionarycurrentaccrualsandsubsequentlong-termstockreturnsisstrongerduringissue-relatedperiods.Asfortheotherthreeaccrualvariables,discretionarylong-termaccrualsdonotpredictfuturestockreturnsineithernewissuersorinthegeneralpopulationofÞrms.Nondiscretionarycurrentaccrualspredictnegativesubsequentreturns S.H.Teoh,etal./JournalofFinancialEconomics50(1998)63Ð9993inthegeneralpopulation,perhapsfromcontaminationofthevariablebythediscretionarycomponentthatisleftinthevariable.Nondiscretionarylong-termaccruals,ontheotherhand,predictincrementallypositivefuturereturnsforissuersbutareunrelatedtostockreturnsforthegeneralpopulationofÞrms.Insum,thesimplecross-sectionalregressionsintheprevioussectionandthepanelregressionsheredocumentastrongabilityofpre-issuediscretionarycurrentaccrualstopredictmultiyearpost-issueabnormalreturns.Wealsodocumentthatdiscretionarycurrentaccrualshavea(time-diminishing)abilitytopredictsubsequentreturnsforallÞrms,althoughthispredictiveabilityissigniÞcantlygreaterforissuers.Incontrast,discretionarylong-termaccrualspredictreturnsonlyintheyearimmediatelyfollowingtheissue(whichresultsinalowersigniÞcanceinthesimplecross-sectionalregressions).Finally,nondis-cretionarypre-issueaccrualshavenoreliablepredictiveabilityonpost-issuestockmarketperformance.7.SummaryandconclusionThispaperexamineswhetherpre-issueearningsmanagement,asreßectedindiscretionaryaccruals,explainsthelong-termunderperformanceofsea-sonedequityissuers.Wedocumentthatdiscretionarycurrentaccrualsgrowbeforetheo¤ering,peakintheo¤eringyear,anddeclinethereafter.Thisaccrualspatterncausesnetincometogrowbefore,peakin,anddeclineaftertheo¤eringyear,despitelowpre-issueandimprovedpost-issuecashßowfromoperations.Thepost-issuenetincomedeclineisespeciallypronouncedforissuersthataggressivelymanagediscretionarycurrentaccrualsbeforetheissue.Mostimportantly,wedocumentanegativerelationbetweenpre-issuedis-cretionarycurrentaccrualsandpost-issueearningsandstockreturns.ThenegativerelationwithstockreturnsremainsaftercontrollingforÞrmsize,book-to-marketratio,andpost-issuecapitalexpenditures.ThenegativerelationbetweendiscretionarycurrentaccrualsandsubsequentreturnsiscommontoallÞrmsduringthe1976to1990testperiod.Theseresultsextendpriorresearchintwoways.First,weshowthatthediscretionarycomponentofcurrentaccrualsexplainsfuturereturnsaswellasthebook-to-marketratiodoesandconsider-ablybetterthanÞrmsize.Second,weshowthatthenegativerelationbetweendiscretionarycurrentaccrualsandfuturereturnsisstrongerandmorepersistentamongÞrmsengaginginnewequityo¤erings.Therelationbetweendiscretionarycurrentaccrualsandpost-issueunderper-formanceinstockreturnshasalsobeendocumentedbyTeohetal.(forthcomingb)andTeohetal.(forthcominga)forinitialpublico¤erings.Becauseseasonedequityissuersarealreadyfollowedbyanalysts,o¤ermorepublicandaudited 94S.H.Teoh,etal./JournalofFinancialEconomics50(1998)63Ð99information,havegreatermarketcapitalization,andareeasiertosellshort,itwouldseemplausiblethatincentivesandopportunitiesfordeceivinginvestorsbymanagingearningsaremorelimitedforseasonedequityissuesthanforinitialpublico¤erings.Thus,theevidencehereofpredictablepost-issuereturnsforseasonedissuersbasedonavailablediscretionaryaccrualinformationposesanevenstrongerchallengetothee¦cientmarketstheorythantheevidenceoninitialpublico¤erings.Insum,theevidenceisconsistentwiththehypothesisthatinvestorsnaivelyextrapolatepre-issueearnings,andignorerelevantinformationcon-tainedinpre-issuediscretionarycurrentaccruals.Inthisinterpretation,aninformationallyimperfectmarketistoooptimisticwhenaseasonedequityissueiso¤eredandlateronbecomesdisappointedwhenthehighearningscannotbesustained.TheseÞndingshaveimplicationsforinvestors,Þrms,andaccountingstandardsetters.Investorscanuseinformationcontainedinthepre-o¤eringaccountingaccrualstodiscriminateamongissuers.Man-agerscanconsiderpermissibleaccountingchoicestoreducetheÞrmÕscostofcapitalorincreasetheirownwelfare.Andaccountingstandardsettersmaywanttoconsiderthecostsofdiscretioninaccrualchoices,especiallywheninvestorshaveconsiderablylessinformationthandothemanagersofissuingÞrms.9AppendixA.A.1.ComputationoftheaccrualsmeasuresThisappendixexplainstheestimationofthefouraccrualmeasures.NumbersinparenthesesareCompustatitemnumbers.Totalaccruals(¹AC)consistofcurrentaccruals(CA)orworkingcapitalaccruals,andlong-termaccruals(¸A).¹ACiscalculatedas¹AC"CA#¸A"NetIncome(172)!CashFlowfromOperations(308).(A.1)Priorto1987,cashßowfromoperationsisnotavailableasitem(308)soitiscalculatedasthefundsßowfromoperations(item110)minuscurrentaccruals9AcurrentproposalattheFinancialAccountingStandardsBoard(FASB)suggestsrequiringonlyashort-formreport.Ourevidencesuggeststhatthismayhaveadverseconsequencesforinvestors.Theeliminationofprice-relevantaccountingitems(suchasitemsnecessarytocomputeaccruals)mayresultininvestorstradinginanevenlessinformationallye¦cientmarket. S.H.Teoh,etal./JournalofFinancialEconomics50(1998)63Ð9995asdeÞnedbelow(seepage111oftheCompustat1994manualforfurtherdetails).10Currentaccrualsarethechangeinnoncashcurrentassetsminusthechangeinoperatingcurrentliabilities:CA"D[currentassets(4)!cash(1)]!D[currentliabilities(5)!currentmaturityoflong!termdebt(44)].(A.2)Toobtainthediscretionaryandnondiscretionaryaccrualsforagivenyear,weusethecross-sectionaladaptationofthemodiÞedJones(1991)modelasinTeohetal.(forthcomingb).AnordinaryleastsquaresregressionofcurrentaccrualsforagivenyearisregressedonthechangeinsalesforthatyearusingallÞrmsinthesametwo-digitSICcodeastheseasonednewissuer,butexcludingtheissuer.Thisintra-industrycross-sectionalregressionisreestimatedforeachyearinthetestperiod(fromyears!3to#3relativetotheÞscalyearoftheissue).Consistentwiththeuseofthemodelintheaccountingliterature,allvariablesincludingtheinterceptterminthecross-sectionalregressionaredeßatedbybeginningtotalassetstoreduceheteroskedasticity:CA1DSA¸ESjt"a#ajt#e,(A.3)¹A0A¹AB1A¹ABjtj,t~1j,t~1j,t~1wherejÞrmsbelonginthesametwo-digitSICcodeastheissuingÞrmbutexcludingtheissuer,¹Aistotalassetsinyeart!1,andDSA¸ESisthej,t~1j,tchangeinsalesinyeartforÞrmj.Thenondiscretionarycurrentaccruals(scaledbybeginningassets),NDCA,~1representtheportionofcurrentaccrualsdictatedbyÞrmsalesgrowth,andisviewedasindependentofmanagerialcontrol.Itiscomputedas1DSA¸ES!DA/RNDCA"aL#aLitit,(A.4)it0A¹AB1A¹ABi,t~1i,t~1whereDA/Risthechangeintradereceivablesinyeartforissueri.Wesubtractittheincreaseinaccountsreceivablefromsalesgrowthtoallowforthepossibilityofcreditsalesmanipulationbytheissuer,whomightallowgenerouscreditpoliciestoobtainhighsalespriortotheo¤ering(seeDechowetal.,1995).Themainresultsofourpaperarerobusttoomittingthisadjustment.10Item(110)isusedasthecashßowfromoperationsincaseswhenCompustatalreadyexcludescurrentaccrualsfromitem(110).Thisoccurswhenfundsfromoperationsarereportedeitherasonetotalorgroupedtogetherinthefundsßowstatement(formatcodeforitem318is2)orinthecashbyactivitystatement(formatcodeforitem318is3). 96S.H.Teoh,etal./JournalofFinancialEconomics50(1998)63Ð99Theremainingcurrentaccrualsarethescaleddiscretionarycurrentaccruals(DCA)andaretheportionofcurrentaccrualssubjecttomanipulationby~1management:CADCA"it!NDCA.(A.5)it¹Aiti,t~1Forlong-termaccruals,weÞrstestimatetotalaccrualsbyusingaregressionsimilartoEq.(A.3).Thedependentvariableistotalaccruals.Weincludeproperty,plant,andequipmentasanadditionalregressorbecauselong-termaccruals(e.g.depreciationlevels)area¤ectedbytheamountoflong-termassetsasinJones(1991):¹AC1DSA¸ESPPEjt"b#bjt#bjt#e,(A.6)¹A0A¹AB1A¹AB2A¹ABjtj,t~1j,t~1j,t~1j,t~1wherejÞrmsbelonginthesametwo-digitSICcodeastheissuer(excludingtheissuer),¹ACistotalaccruals,andPPEisgrossproperty,plant,andequip-jtjtmentforÞrmjinyeart.Thenondiscretionarytotalaccrualsscaledbyassets(ND¹AC)andthediscretionarytotalaccrualsscaledbyassets(D¹AC)arecomputedas1DSA¸ES!DA/RPPEND¹AC"bK#bKitit#bKit,it0A¹AB1A¹AB2A¹ABi,t~1i,t~1i,t~1(A.7)and¹ACD¹AC"it!ND¹AC.(A.8)itA¹ABititAsinEq.(A.4),theincreaseinaccountsreceivableissubtractedfromsalesgrowthtoallowforthemanipulationofcreditsales.Long-termaccrualsaretotalaccrualsnetofcurrentaccruals.Thus,non-discretionarylong-termaccrualsscaledbyassets(ND¸A)willbethedi¤er-~1encebetweenND¹ACandNDCA.Discretionarylong-termaccrualsscaled~1byassets(D¸A)willbethedi¤erencebetweenlong-termaccrualsand~1ND¸A.~1Tosummarize,thefouraccrualsmeasuresarediscretionarycurrentaccruals(DCA),discretionarylong-termaccruals(D¸A),nondiscretionarycurrent~1~1accruals(NDCA)andnondiscretionarylong-termaccruals(ND¸A).The~1~1twodiscretionaryaccrualmeasuresareproxiesforearningsmanagementwhere-asthetwonondiscretionaryaccrualmeasuresareproxiesforaccrualrecogni-tionoutsidethecontrolofmanagement. S.H.Teoh,etal./JournalofFinancialEconomics50(1998)63Ð9997Table9providessomedescriptivestatisticsonthepropertiesoftheestimatedregressions(coe¦cients,t-statistics,regressionÞts,andnumbersofobservations)fromEqs.(A.3)and(A.6).Ingeneral,theregressionstatisticsarecomparablewiththosereportedinpreviousstudies(seePerryandWilliams,1994).ThemedianadjustedR2fortheregressionforcurrentaccrualsis20%andfortotalaccrualsis32%,whichisencouragingasanindicatoroftheexplanatorypowerofthecross-sectionalJonesmodel.A.2.FamaÐFrenchreturns11ToextractindividualÞrmfactorloadings(c)onthethreeFamaÐFrench(1993)factors,werunatimeseriesregressionforeachÞrmofthemonthlyexcessreturnovertherisk-freerateonthethreeFamaÐFrenchfactorsfrommonth!36tomonth!12relativetotheÞlingdateoftheo¤ering:R!r"c(R!r)#cR#cR#e,(A.9)i,tf,t1m,tf,t24.",t3).-,ti,twheretisamonthindex,Risthereturnonasmall-capitalizationportfolio4.",tminusalarge-capitalizationportfolio,Risthereturnonahighbook-to-).-,tmarketportfoliominusalowbook-to-marketportfolio,Risthereturnonm,tavalue-weightedmarketindex,ristheone-monthTreasurybill(risk-free)ratef,tofreturn,andRisthereturnforeachissuer.Aminimumof12availablei,tmonthsisrequiredtoestimateeachregression.Theexpectedreturnforeachmonthiscomputedfrommonth!11to#60,usingtheestimatedcoe¦cientsfromthefactorregression,therelevantmonthfactorreturns,andreplacingtheinterceptwiththerisk-freerateofreturn:ER"r#cL(R!r)#cLR#cLR.(A.10)i,tf,t1m,tf,t24.",t3).-,tTheabnormalreturnistherealizedreturnminustheFamaÐFrenchexpectedreturn.Inoursample,thecoe¦cientsindicatethattheFamaÐFrenchvariablesareunlikelytoexplaindi¤erencesinobservedreturnsbetweenthetwoextremeaccrualquartiles.ThebetasforthemarketpremiumandÞrmequitysizedi¤erby0.06and0.03,respectively.Thebetasforthebook-to-marketvariabledi¤erbyamodest0.35.Furthermore,Table6indicatesthatthetwoquartileshavesimilarpre-issuereturnperformance,makingitunlikelythatourresultsaredrivenbythewinner/loserreversalphenomenon.11ThereissomeargumentastowhethertheFamaÐFrenchmodelexplainssystematicrisk(covariance).However,itisknowntoexplainthecross-sectionalvariationinaveragereturnsintheCRSPdatasetquitewell.ByincludingabnormalreturnsrelativetotheFamaÐFrenchfactorsforexpectedreturns,weshowthattheaccrualsÕinßuenceonreturnsisdistinctfromthatexertedbythemarket-return,theÞrmÕsbook-to-marketratioandtheÞrmÕssize.LoughranandRitter(1995)followthesamestrategy. 98S.H.Teoh,etal./JournalofFinancialEconomics50(1998)63Ð99Table9DescriptivestatisticsonestimatedparametersfortheexpectedaccrualsmodelThistableprovidesdescriptivestatisticsontheparametersfromtheregressionsthatestimateexpected(currentandtotal)accruals.ForeachissuingÞrm,theparametersareestimatedfromacross-sectionalregressionusingÞrmsinthesametwo-digitSICcodeastheissuingÞrm(over60two-digitSICcodeindustriesarerepresented).NisthenumberofÞrmsintheissuerÕstwo-digitSICindustryusedinthecross-sectionalregressions.ThestatisticsontheparametersareroughlycomparablewiththosereportedinPerryandWilliams(1994)whoestimateatimeseriesregressionÞrmbyÞrm.(Nocomparablestatisticsforthecross-sectionalJones(1991)modelhavebeenreportedincurrentpublishedpapers.)ParameterMeanQuartilecuto¤s1st2nd3rdPanelA:Currentaccrualsregression,Eq.(5)a0.05!0.08!0.000.100t-stat0.52!1.37!0.021.42a0.050.010.080.161t-stat3.100.132.105.00Adj.R227%4%20%41%N882757111PanelB:¹otalaccrualsregression,Eq.(8)b!0.01!0.11!0.010.070t-stat!0.84!1.75!0.240.92b0.03!0.020.050.131t-stat2.00!0.331.403.66b!0.07!0.10!0.07!0.042t-stat!3.23!5.12!4.66!1.28Adj.R237%15%32%56%N872757105ReferencesAlford,A,Jones,J,Zmijewski,M,1994.ExtensionsandviolationsofthestatutorySECform10-KÞlingrequirements.JournalofAccountingandEconomics17,228Ð254.Barber,B,Lyon,J,1997.Detectinglong-runabnormalstockreturns:theempiricalpowerandspeciÞcationofteststatistics.JournalofFinancialEconomics43(3),341Ð372.Cheng,L.,1995.Equityissueunderperformanceandthetimingofsecurityissues.Unpublishedworkingpaper.MassachusettsInstituteofTechnology,Cambridge,MA.Daniel,K,Titman,S,1997.Evidenceonthecharacteristicsofcrosssectionalvariationinstockreturns.JournalofFinance52(1),1Ð33.Davidson,S,Stickney,C,Weil,R,1986.Accounting:TheLanguageofBusiness.ThomasHortonandDaughters,SunLakes,AZ. S.H.Teoh,etal./JournalofFinancialEconomics50(1998)63Ð9999Dechow,P,Sloan,R,Sweeney,A,1995.Detectingearningsmanagement.AccountingReview70(2),193Ð225.DeFond,M,Jiambalvo,J,1994.Debtcovenantviolationandmanipulationofaccruals.JournalofAccountingandEconomics18(2),145Ð176.Fama,E,French,K,1993.Commonriskfactorsinthereturnsonstocksandbonds.JournalofFinancialEconomics33,3Ð56.Fama,E,French,K,1997.Industrycostsofequity.JournalofFinancialEconomics43(2),153Ð194.Fama,E,MacBeth,J,1973.Risk,returnandequilibrium:empiricaltests.JournalofPoliticalEconomy81,607Ð636.Guay,W,Kothari,S,Watts,R,1996.Amarket-basedevaluationofdiscretionaryaccrualmodels.JournalofAccountingResearchSupplement34,83Ð105.Guenther,D,1994.Measuringearningsmanagementinresponsetocorporatetaxratechanges:evidencefromthe1986taxreformact.AccountingReview69,230Ð243.Hansen,R,Crutchley,C,1990.CorporateearningsandÞnancing:anempiricalanalysis.JournalofBusiness63,349Ð371.Healy,P,1996.Discussionofamarket-basedevaluationofdiscretionaryaccrualmodels.JournalofAccountingResearchSupplement34,107Ð115.Healy,P,Palepu,K,1990.Earningsandriskchangessurroundingprimarystocko¤ers.JournalofAccountingResearch28,25Ð48.Jain,B,Kini,O,1995.Thepost-issueoperatingperformanceofipoÞrms.TheJournalofFinance49(2),1699Ð1726.Jones,J,1991.Earningsmanagementduringimportreliefinvestigation.JournalofAccountingResearch29(2),193Ð228.Kothari,S,Warner,J,1997.Measuringlong-horizonsecuritypriceperformance.JournalofFinancialEconomics43(4),301Ð340.Kreutzfeldt,R,Wallace,W,1986.Errorcharacteristicsinauditpopulations:theirproÞleandrelationshiptoenvironmentalfactors.Auditing:AJournalofPracticeandTheory6,20Ð43.Loughran,T,Ritter,J,1995.Thenewissuespuzzle.JournalofFinance50(1),23Ð51.LoughranT.,Ritter,J.,1997.TheoperatingperformanceofÞrmsconductingseasonedequityo¤erings.JournalofFinance52,forthcoming.Mikkelson,W,Partch,M,Shah,K,1997.Ownershipandoperatingperformanceofcompaniesthatgopublic.JournalofFinancialEconomics44,281Ð307.Perry,S,Williams,T,1994.Earningsmanagementprecedingmanagementbuyouto¤ers.JournalofAccountingandEconomics18,157Ð179.Rangan,S.,1997.EarningsAroundSeasonedEquityO¤erings:Aretheyoverstated?JournalofFinancialEconomics,forthcoming.Ritter,J,1991.Thelong-runperformanceofinitialpublico¤erings.JournalofFinance66,3Ð27.Spiess,K,A§eck-Graves,J,1995.Underperformanceinlong-runstockreturnsfollowingseasonedequityo¤erings.JournalofFinancialEconomics38,243Ð267.Sloan,R,1996.Dostockpricesfullyimpoundinformationinaccrualsaboutfutureearnings?AccountingReview71,289Ð315.Teoh,S.H,Welch,I,Wong,T.J,forthcominga.Earningsmanagementandthelong-termmarketperformanceofinitialpublico¤erings.JournalofFinance.TeohS.H.,Wong,T.J.,Rao,G.,forthcomingb.Areaccrualsduringinitialpublico¤eringsopportun-istic?ReviewofAccountingStudies.