TheInformationContentofDividendDecreases:EarningsorRiskNews?
DoronNissimColumbiaUniversityGraduateSchoolofBusiness3022Broadway,UrisHall604NewYork,NY10027(212)854-4249,dn75@columbia.edu February2005 IthankGeertBekaert,LawrenceGlosten,GurHuberman,ParthaMohanram,StephenPenman,DouglasSkinner,JacobThomas,andseminarparticipantsatColumbiaUniversityandtheInterdisciplinaryCenterforhelpfulsuggestionsments. TheInformationContentofDividendDecreases:EarningsorRiskNews?
Abstract Thispaperdemonstratesthatdividendcutannouncementsconveynewinformationregardingearningsinthecurrentandsubsequentyear.Whiledividenddecreasesarealsoassociatedwithanincreaseinfirmrisk,thechangeinriskurspriortothedividendannouncement.Consistentwiththesefindings,theabnormalstockreturnduringthethree-daydividendcutannouncementwindowisapproximatelyequaltothepresentvalueofunexpectedcurrentandnextyear’searnings.
I.IntroductionDividendcutannouncementstriggersubstantialnegativereturns(e.g.,Aharonyand Swary(1980),Michaelyetal.(1995)).Yet,priorresearchgenerallyfindsthatdividendreductionsareunrelatedtofutureearningsaftercontrollingforcurrentearnings(e.g.,HealyandPalepu(1988),Benartzietal.(1997),NissimandZiv(2001)).1Severalrecentstudiesinterpretthisevidenceasindicatingthatdividenddecreasesdonotconveynewinformationaboutfutureearningsbutratherimplyanincreaseinfirmrisk(e.g.,Benartzietal.(1997),AllenandMichaely(2002),Grullonetal.(2002),Grullonetal.(2003),Chenetal.(2004)).Thisstudyprovidesnewevidenceontheearningsandriskimplicationsofdividendcuts.Inparticular,Ishowthatessentiallyallofthemarketresponsetodividendcutannouncementsisduetonewinformationaboutcurrentandnextyear’searnings,andthatlittleifanyriskinformationisconveyedbytheseannouncements. Thedifferencesbetweenthefindingsofthecurrentandpriorstudiesaredueprimarilytomethodologicalchoices.Specifically,Iusealternativeapproachesformeasuringearningsexpectationsandexaminechangesinriskpremiumsusingshortratherthanlongintervals.Whilethesechoicespotentiallyinvolvetrading-offonebiasforanother,Iprovideevidencethatsupportsmychoicesanddemonstratetheirimpactontheresults.Inparticular,Idocumentthesourcesandextentofbiasassociatedwithusingdifferentmethodologiesformeasuringexpectedearnings.Ialsoshowthatusinglongintervalsforestimatingriskdisguisesthefactthatessentiallyalloftheincreaseintheriskinessofdividendcutfirmsursbeforethedividendannouncement.Finally,Idemonstratethatmyestimatesofthemagnitudesofearningsandrisknewsconveyedbydividendcutannouncementsareconsistentwithobservedmarketresponses. 1OneexceptionisDeAngeloetal.(1992).Focusingonfirmswithpositiveearningsanddividendsinthepriortenyearsandalossinthecurrentyear,DeAngeloetal.(1992)findthatthosefirmsthatcuttheirdividendsinthecurrentyearreportlowernextyearearningsevenaftercontrollingforcurrentyearearnings.
1 Thestudyproceedsasfollows.SectionIIdiscussesdifficultiesandtrade-offsinvolvedinmeasuringexpectedearningsfordividenddecreasefirms.SectionIIIdevelopsthemethodologyusedtoexaminethefutureearningsimplicationsofdividendcuts.SectionIVdescribesthesample,andSectionVpresentstheempiricalfindingsoftheearningsandriskanalyses.SectionVIconcludesthepaper. II.IssuesinMeasuringEarningsExpectationsforDividendDecreaseFirmsToexaminethefutureearningsimplicationsofdividendchangeannouncements, previousstudiesregressfutureearningsonthedividendchange,currentearnings(i.e.,earningsintheyearduringwhichthedividendchangeurs)andproxiesforexpectedfutureearnings.Theythenusetheestimatedcoefficientofthedividendchangeasameasureoftheinformationcontentofdividends.Currentearningsareincludedintheregressionbecausetheyhelpexplainfutureearningsandarepartiallypredictableatthetimeofthedividendchange(e.g.,usinginformationfromintermediatequarterlyreportsorfromthefinancialpress).Asmanyofthesestudiesrecognize,however,thischoiceinducesbiasagainstfindinginformationcontentindividendsbecausecurrentearnings,whichareonlypartiallyknownatthetimeofthedividendannouncement,arepositivelyrelatedtothedividendchange.Thus,thedecisiontocontrolforcurrentearningspotentiallyinvolvestrading-offbiasagainstversustowardseptingtheinformationcontentofdividendshypothesis. Howimportantisthischoice?
ordingtoMillerandRock(1985),thisissueisatthecoreoftheinformationcontentofdividendhypothesis.MillerandRockarguethatdividendchangesconveyinformationaboutfutureearningsindirectlybychangingthemarket’sestimateofcurrentearnings,whichinturncontributestotheestimateoffutureearnings.Thus,ifdividenddecreasespredictlowercurrentearnings,whichinturnimplylowerfutureearnings,thenegative
2 futureearningsimplicationsofthedividendchangewillbecapturedbythecoefficientofcurrentearningsratherthanbythedividendcoefficient.Inotherwords,controllingforcurrentearningsexposestheanalysistotheriskof“throwingthebabyoutwiththebathwater.” TheMillerandRockargumentappliesbothtodividendincreasesanddecreases.Butcontrollingforcurrentearningsmayintroduceanadditionalbiaswhichisuniquetodividenddecreases.Totheextentthatdividendcutfirmsaremorelikelytorecognizenegative“specialitems,”theirearningswillbelesspersistentthanthoseofotherfirms.2Ifthisisnotincorporatedintheanalysis,themodelwilloverstatethepersistenceofthenegativeshocktoearningsinthedividenddecreaseyear,leadingtoanunderstatementofexpectedfutureearnings.ordingly,unexpectedearnings(i.e.,realizedearningsminusexpectedearnings)ofdividenddecreasefirmswillbeoverstated,andthisbiasmayoffsetthenegativeearningsimplicationsofdividendcuts. Negativespecialitemsmaybemoreprevalentfordividenddecreasefirmsduetotwoountingphenomena:conservatismand“bigbath.”Conservatismisanountingconventionwhichrequirestherecognitionofanticipatedlosses,suchasthoseduetoimpairmentofassetsorrestructuringofoperations.Thus,ifdividendreductionssignallowerfuturecashflows,theyshouldbeassociatedwithimpairmentlosses,restructuringchargesandothernegativespecialitems.Thebigbatheffectisrelatedtomanagementbehavior.Theoretical,empiricalandanecdotalevidencesuggeststhatinthepresenceofbadnews(suchasthosecausingadividendcut),managersoftenchoosetotakea“bigbath;”thatis,theyoverstateestimatedliabilitiessuchasruedrestructuringcostsortheywritedownassets.3Thesechargesaregenerallynot 2Skinner(2003)showsthatthepersistenceofearningsispositivelyrelatedtothelevelofdividends,especiallyforlowlevelsofdividends.Thatis,lowdividendfirmshaveparticularlylowearningspersistence.Thisevidencesuggeststhatdividenddecreasefirmsmayalsohavelowearningspersistence.Thecurrentstudyexaminesthisconjecturebyfocusingonthemagnitudeoftransitoryearnings.
3 expectedtorecurinfutureperiodsandaretypicallyclassifiedas“specialitems.”Inextdiscussthemethodologyusedinthisstudytomitigatethesebiasesinmeasuringexpectedearnings. III.Methodology Withaperfectmeasureofexpectedearningsimmediatelypriortothedividend announcement,testingwhetherdividendcutsconveynewearningsinformationis straightforward;oneonlyneedstocalculatethet-statisticofthemeandifferencebetween realizedandexpectedearningsforasampleofdividenddecreasefirms.Equivalently,onecould runthefollowingregression: EPS(t)/P=E[EPS(t)]/P+λDIV_DEC+ε,
(1) whereEPS(t)isearningspershareinyeart(t=0,1,
2,…),E[.]istheexpectedvalueoperator conditionedonpubliclyavailableinformationpriortothedividendannouncement(inyear0),
P ispricepersharepriortothedividendannouncement,andDIV_DECisaqualitativevariable indicatingadividenddecrease.Realizedandexpectedearningspersharearedeflatedbyprice persharetomitigateheteroscedasticitysincethestandarddeviationofEPSisapproximately proportionaltoprice(e.g.,BakerandRuback(1999)). Unfortunately,E[EPS(t)]isunobservable.Ithereforeestimatethisquantityusing informationfromtwomarket-basedproxiesforexpectedearnings:analysts’forecastsandprice pershare.Specifically,ImodelexpectedEPSasfollows: E[EPS(t)]=βECR(t)×P+δAF(t),
(2) whereECR(t)(EarningsCapitalizationRate)isaproxyfortherateatwhichinvestorscapitalize expectedEPS(t)intoprice,andAF(t)(Analysts’Forecast)isameasureofavailableanalysts’ 3Forexamplesofanecdotal,empirical,andanalyticalevidence,seeJacksonthandPitman(2001),ElliottandHanna(1996)andKirschenheiterandMelumad(2002),respectively.
4 EPSforecastforyeartatthetimeofthedividendannouncement.Bothanalysts’forecastsand pricereflectexpectationsofmarketparticipantsregardingfutureearnings.However,prior researchestablishesthatbothmeasuresaresomewhatinefficientandbiased,andthateachcontainsincrementalinformationrelativetotheother.4Thus,byincorporatinginformationfrom bothmeasures,equation(2)maygenerateamoreefficientestimateofexpectedearningsthan estimatesbasedonpriceoranalysts’forecastsalone. Therateatwhichinvestorscapitalizeexpectedearnings,ECR(t),isafunctionof macroeconomicvariables(e.g.,interestrates,marketriskpremium)aswellasfirm-specificrisk,expectedgrowthandpayout.5IthereforespecifyECR(t)asabinationofthefollowing variables: ECR(t)=αyear+αindustry+αdistance+α1BETA+α2VOLAT +α3LEV+α4SIZE+α5BM+α6GROWTH+α7YIELD,
(3) whereαyearisayearfixedeffect,includedtocapturetheaverageeffectofmacroeconomic variables;αindustryisanindustry(2-digitSIC)fixedeffect,whichcontrolsforindustry-specific riskandgrowthprospects;andαdistanceisafixedeffectwhichcontrolsforthenumberofmonths betweenthedividendannouncementandtheendofthefiscalyear.Thislattereffectisimportant becausethepresentvalueofearningsdependsonthelengthoftimeuntiltheirrealization. Thecontinuousvariablesinequation(3)monlyusedmeasuresofrisk,expected growthandpayout.BETAisameasureofsystematicrisk,estimatedfrommonthlystockreturns 4Forareviewofthisliterature,seeKothari(2001).Notethattheprecisionoftheprice-basedmeasureofexpectedearningsisaffectedbythequalityoftheempiricalproxyforECR(t)inadditiontopriceefficiency.Similarly,analysts’forecastsmaycontainerrorduetotheirdiscretenature(i.e.,theymaybestaleatthetimeofthedividendannouncement)inadditiontoerrorduetoanalysts’inefficienciesinprocessinginformation. 5Forexample,undertheGordon(1962)model,ECR
(1)=(r–g)/d,whereristhecostofequitycapital,gisexpectedgrowth,anddisthedividendpayout.ForempiricalspecificationsofECRmodels,seeBeaverandMorse(1978),Zarowin(1990),Fairfield(1994),andGreenspan(1997).
5 andtheCRSPvalue-weightedreturnsduringthefiveyearsthatendinthemonthpriortothe dividendannouncement.VOLATreflectsidiosyncraticriskandismeasuredastheroot-mean- squarederrorfromtheBETAregression.LEV,theratiooftotalliabilitiestothesumoftotal liabilitiesandthemarketvalueofequity,isaproxyforfinancialrisk.SIZE(logofmarketvalue ofequity)andBM(book-to-marketratio)captureriskandgrowthprospects.GROWTHisa measureofanalysts’long-termearningsgrowthforecastavailableatthetimeofthedividend announcement.YIELDistheratiooftheindicatedannualdividend(themostrecentquarterlydividendtimesfour)dividedbyclosingpricetwodayspriortothedividendannouncement.6 Inextsubstituteequation(3)intoequation(2)anddividebothsidesbyprice: E[EPS(t)]/P=γyear+γindustry+γdistance+γ1BETA+γ2VOLAT+γ3LEV +γ4SIZE+γ5BM+γ6GROWTH+γ7YIELD+δAF(t)/P
(4) Finally,Isubstituteequation(4)intoequation
(1): EPS(t)/P=γyear+γindustry+γdistance+γ1BETA+γ2VOLAT+γ3LEV+γ4SIZE +γ5BM+γ6GROWTH+γ7YIELD+δAF(t)/P+λDIV_DEC+ε
(5) Totestwhetherdividenddecreasesprovidenewinformationaboutcurrent(t=0)andfuture(t=
1,…,5)earnings,Iestimateequation(5)forasampleofdividendpayingfirmswhichincludes dividendincreases,dividenddecreases,andno-changeobservations.Thus,thecoefficientonthe DIV_DECindicatorvariablereflectstheincrementalearningsassociatedwithdividend decreases.Ifocusondividendpayingfirmsbecausethemagnitudesofbiasanderrorinthe 6Iusetheyieldratherthanthepayoutbecauseearningsarenegativeorclosetozeroformanydividenddecreasefirms.Moreover,totheextentthatpriceisaproxyfor“permanentearnings,”theyieldislikelytobeabetterproxyforexpectedfuturepayoutthantheactualpayoutduetothestabilityofdividendpayments.
6 proxiesforexpectedearningsmaybedifferentfortheseparedtofirmsthatdonotpaydividends.7 Whileitisimpossibletoremoveallerrorandbiasfromtheproxiesforexpectedearnings,itisimportanttonotethattheECR(t)variables(equation
(3))notonlyimprovetheprecisionoftheprice-basedproxyforexpectedearnings,butalsomitigatepotentialbiasesduetoinefficienciesinanalysts’forecasts.Forexample,priorresearchdocumentsthatthebiasinanalysts’forecastsdecreasesastheearningsannouncementdateapproaches.Thus,theinclusionofαdistanceintheregression(fixedeffectforthenumberofmonthsuntiltheendofthefiscalyear)mitigatespotentialbiasduetosystematicvariationintherelativefrequencyofdividenddecreasesduringthefiscalyear. Mostfirmspaydividendsonaquarterlybasis,sothesameannualearningsmaybeassociatedwithuptofourobservations.TheresultingoverlapinobservationshaslittleeffectontheDIV_DECcoefficientbecauseveryfewfirmshavemorethanonedividendreductioninanygivenyear(resultsarenotsensitivetotheexclusionoftheseobservations).Also,evenforoverlappingobservations,thevaluesofthedependentvariablearenotidenticalbecauseprice(thedeflator)ismeasuredatdifferentpointsintime(reflectingallinformationuptotheparticulardividendannouncement).Inanycase,tomitigatetheeffectofautocorrelationintheresidualduetooverlappinginformation,IreportNeweyandWest(1987)t-statisticswiththreelags. 7Firmsthatdonotpaydividendsaretypicallysmallwithlowprofitabilityandstronggrowthopportunities(seeFamaandFrench(2001)).
7 IV.DataTheinitialsampleincludesalldividendeventsintheCRSPfilesthatsatisfythefollowing criteria:(1)panypaidanordinarycashdividend(
U.S.dollars)inthecurrentquarterandinthepreviousquarter,(2)nootherdistributionswereannouncedbetweenthedeclarationofthepreviousdividendandthreedaysafterthedeclarationofthecurrentdividend,(3)therewerenoex-distributiondatesbetweentheex-distributiondatesofthepreviousandcurrentdividends,and(4)theindicatedannualdividendyieldpriortothedividendannouncementisatleast0.5%.Criteria(2)and(3)helpensurethatonly“clean”dividendchangesthatavoidconfoundingeffectsfromotherdistributionsareidentified.Criterion(4)issettorestrictthesampletofirmsthatpaysignificantdividends. Foreachofthedividendevents,IextractfromCRSPthemonthlystockreturnsinthepriorsixtymonthsandcalculatetheBETAandVOLATmeasuresdiscussedabove(atleast30returnobservationsarerequired).Next,ImatchthedividendeventsamplewithforecastandreporteddataextractedfromtheIBESDetail,SummaryandActualfiles.TheDetailfilecontainsanalyst-by-analystestimates,whiletheSummaryfilecontainstheconsensus(meanandmedian)analysts’forecastsineachmonth.ordingtoIBES,theDetailfileisareconstructionofarchiveddataandmaythereforecontainerrorsandomissions.Individualanalystforecastsmayalsobelessprecisethantheconsensusestimates.Ithereforeusetheconsensus(mean)forecastsintheprimaryanalysis,butcheckthesensitivityoftheresultstotheuseofindividualforecasts.Specifically,IrerunallanalysesusingthemostrecentEPSandgrowthforecastsavailableattimeofthedividendchangeandreporttheresultsofthesetestsinarobustnesssectionbelow. TheSummaryIBESfileconsistsofchronologicalsnapshotsofconsensusearningsforecaststakenontheThursdaybeforethethirdFridayofeverymonth.ordingtoIBES,theconsensusforecastsaremadeavailableinthefollowingweek.Ithereforematchdividend
8 announcementsurringafterthefourthSundayofeachmonthandbeforethefourthMondayofthefollowingmonthwiththeconsensusforecastsforthatmonth.IrequirethatthefollowingvariablesbeavailablefromIBES:meananalysts’EPSforecastforthecurrent(AF
(0))andnextyear(AF
(1)),aswellasmeanlong-termearningsgrowthforecast(GROWTH).8ordingtoIBES,thelong-termearningsgrowthforecastgenerallyreferstoaperiodofuptofiveyearsahead.Ithereforegenerateforecastsforearningsinfutureyears2through5byapplyingthemeanlong-termgrowthforecasttothemeanforecastfortheprioryearinthehorizon;i.e.,AF(t+s)=AF(t+s-1)×(1+GROWTH),fors=2,3,4,
5. TheresultingsampleisthenmatchedwiththeCOMPUSTATfilesbyassigningeachdividendeventtotheearliestfiscalyearforwhichearningshavenotyetbeendisclosed.Thatyearisdesignatedasyear0.Dividendsareoftenannouncedsimultaneouslywithquarterlyearnings,withapositivecorrelationbetweenthetwodisclosures(e.g.,firmsarelikelytoincreasedividendsfollowingearningsincreases;see,e.g.,Benartzietal.(1997)).Toassurethatthedividendeventsreflectonlydividendinformation,Iexcludeobservationsforwhichthedividendannouncementurswithinthreecalendardaysbeforeorafteraquarterlyearningsannouncement.9InextextracttherequiredfinancialstatementsinformationfromCOMPUSTATanddeleteobservationswithmissingvaluesforcurrentearningsorforanyoftheexplanatoryvariablesofequation
(5)(observationswithmissingvaluesforfutureearningsareretained,toavoidsurvivorshipbias).Finally,tomitigatetheeffectofinfluentialobservations,thevariablesarewinsorizedatthe1thand99thpercentilesoftheirempiricaldistributions.10 8Topreventduplication,Idonotincludeforecastswith“secondary”flag.IadjustallpersharenumbersforstocksplitsandstockdividendsusingtheIBESadjustmentfactor.Ifthefirmisfollowedonadilutedbasis,IusetheIBESdilutionfactortoconvertpersharevariablestoaprimarybasis.9EarningsannouncementdatesareextractedfromthequarterlyCOMPUSTATfiles.
9 Table1presentsdescriptivestatisticsforthefinalsample.Asreported,thereare282dividenddecreases,5,576dividendincreases,and35,769no-changeannouncementsthatsatisfyalltheabovecriteria.Thesedividenddeclarationsurredduringtheyears1982through2003(IBESdataareavailablesincetheearly1980s).Whiledividendreductionsaremuchmonthandividendincreases,theyareconsiderablylargerinmagnitude—themeanpercentagechangeindividendspershareis–45.6%fordividenddecreasesand12.9%fordividendincreases.Dividenddecreasefirmsarerelativelysmallandhavelargebook-to-marketratios,financialleverageandresidualvolatility,suggestingthattheyaremoreriskythanotherdividendpayingfirms.However,theaveragebetaofdividenddecreasefirmsissmallerthanthatofotherdividendpayingfirms.Inaddition,dividenddecreasefirmshavesubstantiallysmalleranalysts’long-termgrowthforecastsandhigherdividendyields.
V.EmpiricalFindingsA.Time-seriesBehaviorofEarningsandSpecialItems Pastlevelsofearningsprovidenaturalbenchmarksforcurrentandfutureearnings.IthusexaminesummarystatisticsfromcrosssectionaldistributionsofEPSintheyearssurroundingdividenddecreases,andinparticulartestthesignificanceofchangesinthesestatisticssubsequenttothedividendcut.11Figure1plotsthedistributionofEPSineachoftheelevenyearssurroundingthedividenddecrease,scaledbyclosingpricetwodayspriortothedividendannouncement(allpersharedataareadjustedforstocksplitsandstockdividends).Thestatisticsplottedarethe25th,50thand75thpercentiles,andthemean.PanelAofTable2presentsthemean 10Qualitativelysimilarresultsareobtainedwithoutwinsorizing,withtrimminginsteadofwinsorizing,andwithalternativepercentilecuts.11EPSismeasuredasCOMPUSTAT’sdataitem#58(basicearningspersharebeforeextraordinaryitemsanddiscontinuedoperationsandadjustedforpreferreddividends). 10 andmedianofearningsinthesevenyearssurroundingthedividenddecrease,aswellaschangesinthesequantitiesandthesignificanceofthechanges.12Asshown,currentearnings(i.e.,earningsinyear0)areconsiderablysmallerthaninthesurroundingyears,althoughthedecreaseinearningsstartsinyear–1(inyears–5through–2earningsarerelativelystable).Meanearningsdropsfrom0.109to0.068inyear–1(t-statisticforthechangeequals–7.5),andfrom0.068to–0.023inyear0(t-statisticforthechangeequals–10.7).Fromyear1on,earningsincreasemonotonicallybutataslowerpacethantherateofdeclineinyear0.Similarresultsareobtainedformedianearnings,indicatingthatthesetrendsarenotduetooutlierobservations. Consistentwiththeconjecturethatfirmsaremorelikelytoreportnegativespecialitemsindividenddecreaseyears,thelefttailoftheearningsdistributionflattensinyear0(e.g.,thedifferencebetweenthe50thand25thpercentilesinFigure1essubstantiallylarger).Theflatteningofthelefttailisalsoreflectedintherelationshipbetweenthemeanandthemedian:Meanearningsissimilartothemedianinyears–5through–1,butinyear0themeanisconsiderablysmallerthanthemedian.Toverifytheinferencethatfirmsreportabnormallevelsofnegativetransitoryitemsindividenddecreaseyears,Inextexaminethedistributionofspecialitems.13 Figure2plotsthecross-sectionaldistributionsofspecialitemspershareintheyearssurroundingthedividendcut,andPanelBofTable2presentssummaryandteststatistics(mean,frequencyofnegativevalues,changesinthesestatistics,andt-statisticsforthechanges).Toholdsizeconstant,specialitemsarescaledbyclosingpricetwodayspriortotheannouncementofthe 12Thechangestatisticsareslightlydifferentfromthecorrespondingdifferencesbetweentheconsecutivelevelsbecausetheyarecalculatedusingobservationswithnon-missingvaluesforbothyears.13SpecialitemsaremeasuredasCOMPUSTAT’sdataitem#17.Thisitemincludesdifferenttypesofrevenuesandexpensesfromcontinuingoperationswhichareclassifiedbyfirmsastransitory(e.g.,write-offs,impairmentsandrestructuringcharges).Specialitemsaremuchmonandlargerinmagnitudethanextraordinaryitems,whicharerarelyreportedpanies.Foradetaileddiscussionofspecialitems,seeBurgstahleretal.(2002). 11 dividendcut.Meanspecialitemsdecreasesfrom–0.013inyear–1to–0.043inyear0(t-statisticforthechangeequals–5.2)andincreasesbackto–0.017inyear1(t-statisticequals3.8).This“V”shapepatternisespeciallyapparentforthe10thpercentileofthedistribution(seeFigure2).Inaddition,thefrequencyofnegativespecialitemsincreasesfrom31.3percentinyear–1to50.2percentinyear0(t-statisticforthechangeequals4.2),anddecreasesbackto35.0percentinyear1(t-statisticequals–3.4).Afteryear1,thefrequencyofnegativespecialitemsremainsrelativelystable. Tosummarize,firmsthatcuttheirdividendsreportverylowearningsintheyearofthedividendchange,inlargepartduetotherecognitionofnegativespecialitems.Consistentwiththelowpersistenceofspecialitems(Burgstahleretal.(2002)),thesefirmsreportsubstantialEPSincreasesinsubsequentyears.Yet,earningsreachtheirpre-dividendcutlevelonlythreeyearsafterthedividenddecrease.Althoughtheseresultssuggestthatdividenddecreasespredictlowercurrentandfutureearnings,theydonotindicatewhetherthisinformationisrevealedbythedividendannouncementratherthanbypriordisclosuressuchasinterimearningsreports.Inextusemarket-basedinformationtoconstructmeasuresofexpectedearningsimmediatelypriortothedividendchange,whichallowmetotestwhetherdividendcutannouncementsconveynewearningsinformation.
B.DividendDecreasesandUnexpectedEarningsIstartbyexaminingtheassociationbetweenanalysts’forecasterrorsanddividend changes.Totheextentthatanalysts’EPSforecastsreflectthemarketexpectationofcurrentandfutureEPS,theaveragevalueoftheforecasterrors(i.e.,theaveragedifferencebetweenrealizedandforecastedEPS)fordividenddecreasefirmsshouldindicatewhetherdividendcutsconveynewearningsinformationtothemarket.Specifically,iftheaveragevalueofanalysts’forecast 12 errorsfordividenddecreasefirmsisnegativeandsignificant,theinferencewouldbethatdividendreductionsconveynegativeearningsnews. Thisapproachfortestingtheinformationcontentofdividendsisstraightforward(analysts’forecastsprovideadirectmeasureofexpectedearnings)andsimpletoimplement.However,ithasimportantings:Analysts’forecastsareonaveragebiasedupwardandtheydonotreflectallavailableinformation(forareviewoftheliteraturedocumentingtheseresults,seeKothari2001).Iparetheforecasterrorsofdividendcutfirmswiththoseofotherdividendpayingfirmsratherthanassumethattheunconditionalmeanoftheforecasterroriszero.Inaddition,Iconductmultivariateanalysesthatmitigatetheeffectofmeasurementerrorintheproxyforexpectedearningsbyincorporatinginformationfromstockpricesandothervariablesinadditiontoanalysts’forecasts. TheresultsoftheunivariateanalysisarereportedinTable3.Thistablepresentsthemeanandmediananalysts’EPSforecasterrorsfordividenddecreasefirmsandforallotherdividendpayingfirms.Analysts’forecasterrors(labeled“unexpectedIBESEPS”)aremeasuredasthedifferencebetweentheIBESmeasureofactualEPSandtheconsensusEPSforecastavailableatthetimeofthedividendannouncement.ActualEPSasmeasuredbyIBESisequaltoreportedEPSminusitemsclassifiedbyanalystsasnon-recurring(“IBESspecialitems”).ThetotaldifferencebetweenreportedandforecastedEPS(“unexpectedreportedEPS”)isequaltothesumof“unexpectedIBESEPS”and“IBESspecialitems.” Analysts’forecasterrorsarenegativeforbothdividenddecreasefirmsandotherdividendpayingfirms,andarelargerinmagnitudeinyearparedtoyear0.Bothresultsareconsistentwithpriorresearch,whichdemonstratesthatanalysts’forecastsarebiasedupwardwiththebiasincreasingintheforecasthorizon.However,unexpectedearningsaresubstantiallyandsignificantlysmallerfordividenddecreaseparedtootherdividendpayingfirms. 13 Specifically,unexpectedIBESearningsfordividenddecreasefirmsareonaverage–4.83%ofpriceforyear0and–5.89%foryear1,whilethecorrespondingnumbersforotherdividendpayingfirmsare–0.75%and–1.93%respectively.Theseresultsarenotduetooutliers,asthedifferencesinmedianarealsohighlysignificant. Themagnitudeofnegativespecialitems(middletwocolumnsinTable3)issubstantiallylargerfordividenddecreasefirmsthanforotherdividendpayingfirms,especiallyinyear0(–2.53%ofpricefordividenddecreaseparedto–0.30%forotherdividendpayingfirms).TheseresultsareconsistentwiththestatisticsinPanelBofTable2,whichdemonstratethatfirmsreportmorenegativespecialitemsindividenddecreaseyearsthaninotheryears.Becausespecialitemsarelesspersistentthanotherearningsitems(e.g.,Burgstahleretal.(2002)),thesefindingsimplythatdividenddecreasefirmsshouldexperienceearningsincreasesinfutureyears.Indeed,theresultsofpriorresearch(e.g.,Benartzietal.(1997))andthisstudy(PanelAofTable2)indicatethatdividendcutsarefollowedbyearningsincreasesinsubsequentyears.However,currentearnings,thestartingpointofthisearningsgrowth,aresubstantiallysmallerthanexpectedatthetimeofthedividendcut.Therefore,theleveloffutureearningsmaystillbelowerthanexpectedatthetimeofthedividendcut,inspiteofthefutureearningsincreases.ThestatisticsinTable3suggestthatunexpectedearningsinyeart=1ofdividenddecreasefirmsareindeednegative. ThusfarIhaveusedanalysts’EPSforecaststomeasurethemarketexpectationsofcurrentandfutureearnings.Asdiscussedabove,however,analysts’forecastscontainerrorandbias,whichmaybecorrelatedwiththecharacteristicsofdividenddecreasefirms(e.g.,smallsize,lowexpectedgrowth).Moreover,analysts’forecastsmaybestale,especiallyfordividenddecreasefirms(giventheirsmallsizeandotherlessattractivecharacteristics).Tomitigatetheseeffects,Inextestimateregressionmodel(5)whichincorporatesinformationfrommarketprices 14 andothervariablesinadditiontoanalysts’forecasts.Istartbyreplicatingtheresultsofpriorstudies(e.g.,Benartzietal.(1997),NissimandZiv(2001)),whichfindthatdividendreductionsarenotassociatedwithnegativefutureearningsaftercontrollingforcurrentearnings.Tothisend,Iincludeinequation(5)theratioofcurrentearningstopriceasanadditionalexplanatoryvariable. Table4presentstheregressionresults.Asshown,thecoefficientofthedividenddecreaseindicatorvariable(DIV_DEC)isnon-negativeineachofthefiveregressions(t=1,
2,…,5),confirmingthatdividenddecreasesdonotimplylowerfutureearningsaftercontrollingforcurrentearnings.Butshouldonecontrolforcurrentearnings?
Asdiscussedabove,ifrealizedcurrentearningsofdividenddecreasefirmsaresmallerthanexpectedatthetimeofthedividendchange,includingthemintheregressionwillbiastheresultsagainstfindinginformationcontentindividenddecreases.TheresultsinTables2and3suggestthatrealizedcurrentearningsareindeedsmallerthanexpectedatthetimeofthedividendchange.Moreover,thehighsignificanceofcurrentearningsinexplainingfutureearnings(Table4),evenaftercontrollingforanalysts’forecasts,priceandotherearningspredictors,suggeststhatcurrentearningsreflectinformationaboutfutureearningswhichisnotavailableatthetimeofthedividendchange. MillerandRock(1985)predicttheresultsofTable4.Theyarguethatdividendchangesconveyinformationaboutfutureearningsindirectlybychangingthemarket’sestimateofcurrentearnings,whichinturncontributestotheestimateoffutureearnings.Thus,whencurrentearningsareincludedintheregression,thecoefficientofthedividendchangeshouldbeinsignificant,asIindeedfind.Incontrast,whentheearningsexpectationmodelreflectsonlyinformationwhichisavailableatthetimeofthedividendchange,thecoefficientofthedividendchangeshouldbesignificant.Inextexaminethishypothesisbyestimatingequation
(5),whichexcludescurrentearnings.TheestimatesinTable5demonstratethatdividenddecreasesimply 15 lowercurrent(t=0)andnextyear(t=1)earnings,buthavenoimplicationsforearningsinlateryears. Assumingadiscountratebetween10to20percent,thedividendcoefficientsinTable5implythatdividendcutannouncementsshouldtriggernegativestockreturnsbetween–7.8%(=–0.0570/1.1.5+–0.0274/1.11.5)and–7.3%(=–0.0570/1.2.5+–0.0274/1.21.5).Theseestimates,however,overstatethepriceimpactofdividendcutannouncementsfortworeasons.First,theconsensusanalysts’EPSforecast,whichistheprimarycontrolvariable,reflectsanalysts’expectationsofrecurringearnings,whilethedependentvariablemeasuresrealizedtotalearnings.Thus,ifdividenddecreasefirmsareexpectedtoreportmorenegativespecialitemsthanotherfirms,thedividenddecreasecoefficientwillbebiaseddownward(itwillcaptureexpectedspecialitemsinadditiontounexpectedearningsandsooverstatethenegativeimplicationsofdividenddecreases).Second,negativespecialitemswhichreducereportedearningsoftenreflect“paperlosses”withlittlecashflowconsequences(e.g.,impairmentofgoodwill).Toaddresstheseconcerns,Inextrerunequation(5)usingCORE_EPS(EPSminusspecialitemspershare)insteadofreportedEPS.TheestimatesinTable6suggestasubstantiallysmallervalueeffect,between–3.6%(=–0.0176/1.1.5+–0.0219/1.11.5)and–3.3%(=–0.0176/1.2.5+–0.0219/1.21.5),assumingdiscountratebetween10to20percent.Yet,similartotheresultsinTable5,thedividendcoefficientishighlysignificantinboththet=0andt=1regressions.
C.MarketReactiontoDividendCutAnnouncementTheestimatesfromtheprevioussectionsuggestthatdividenddecreaseannouncements shouldtriggeranaveragestockdeclineofatleast3percent,duetotheirnegativeearningsimplications.Iparethisestimatewiththeactualmarketresponse.IusetheFamaandFrench(FF,1993)three-factormodeltoestimatetheaverageabnormalstockreturnineachof 16 the201tradingdayscenteredatthedividendcutannouncement.Specifically,foreachrelative tradingday(–100through+100),Iregressthefollowingmodelusingallfirmsthatcuttheir dividendsinday0: ER=γ1+γ2RMRF+γ3SMB+γ4HML+et,
(6) whereERisthedailyexcessstockreturn(rawreturnminustherisk-freereturn),RMRFisthe dailyexcessmarketreturn(marketreturnminustherisk-freereturn),SMBisthedailyreturnona portfoliolonginsmallstocksandshortinlargestocks,andHMListhedailyreturnonaportfoliolonginvaluestocksandshortingrowthstock.14Underthisapproach,theintercept(γ1)ofeach relativedayregressionmeasurestheaverageabnormalstockreturninthatdayforfirmsthat announceadividendcutinday0. Figure3plotsthecumulativeaverageabnormalreturnfordividenddecreasefirms(i.e., thecumulativesumoftheregressionintercepts).Firmsthatcuttheirdividendsintradingday0 experiencenegativestockreturnsintheprior100tradingdayswhichsumuptoaCARof–14% bytheendofday–
2.Thethreedaysannouncementreturn(–1,0,and1)isequalto–3.2%,andit isfollowedbyaveryslightdriftinthefollowingweeks.Theoverallmarketresponseis approximately–3.5%,consistentwiththeestimatesofthepresentvalueofunexpectedearningsreportedintheprevioussubsection.15 14FactorreturnsareobtainedfromWRDS.15Theabnormalreturnsareslightlysmallerthaninpriorstudies(e.g.,AharonyandSwary(1980))dueprimarilytotherequirementofavailabilityofanalysts’forecasts(morerecentsampleperiod,largerfirms).ThesmallmagnitudeofthepostannouncementdriftisconsistentwiththeevidenceinGrullonetal.(2002). 17
D.DividendDecreasesandEquityRisk Grullonetal.(2002)findthatdividendchangesareassociatedwithchangesinequity risk.Inparticular,paretheFFfactorloadingsofdividenddecreasefirmsinthethree yearssubsequenttothedividendchangewiththecorrespondingcoefficientsinthethreeprior yearsandreportsignificantincreasesineachofthethreecoefficients.Theyfurtherestimatethat theannualriskpremiumincreasesbyabout2%andarguethat“changesinriskpremiumofthis magnitudearesufficienttogeneratetheobservedannouncement-daypricereactions”(page388). Indeed,changesofthismagnitudeshouldtriggerverylargenegativereturnsiftheyareindicated bythedividendannouncement.InthissectionIexaminethetimelinessofthechangeinequity risk. Iusethefollowingproceduretoexaminechangesovertimeintheriskpremiumof dividenddecreasefirms.First,IcalculatetheaverageannualizedreturnassociatedwitheachofthethreeFFfactorsduringtheperiodJuly1,1963throughOctober31,2003.16Second,
I estimateequation(6)foreachoftherelativetradingdays–1,250through1,250forthesampleof dividenddecreasefirms.Third,Icalculatetheannualizedriskpremiumassociatedwitheach relativedayusingthefactorloadingsofequation(6)forthatdayandtheaverageannualized factorreturns: AnnualizedRiskPremium=γ2AAR(RMRF)+γ3AAR(SMB)+γ4AAR(HML)
(7) whereAAR(f)istheAverageAnnualizeddailyReturnonfactorfduring1963-2003,andγ
2,γ3and γ4aretheestimatedfactorloadingsfortheparticulartradingday.Finally,Icalculatethemean 16Theaverageannualizedreturnonfactorfiscalculatedasfollows:AAR(f)=exp{252×mean[ln(1+f)]}–1,where252istheaveragenumberoftradingdaysperyearandthemeaniscalculatedoveralldailyobservationsduringtheperiodJuly1,1963throughOctober31,2003. 18 valueoftheannualizedriskpremiumforgroupsof50consecutivetradingdays:(–1,250,–1,201),(–1,200,–1,151),…,(–50,–1),(0,49),…,(1,200,1,249). PanelAofFigure4plotstheestimates.Thex-axisdisplaystherelativetradingdaywhilethey-axispresentsthemeanvalueoftheriskpremiumforeachgroupof50consecutivetradingdays.ConsistentwithGrullonetal.(2002),theaverageriskpremiuminthethreeyearssubsequenttothedividendcut(tradingdays0through750)issubstantiallylargerthaninthepriorthreeyears.Specifically,theaverageriskpremiuminthethreeprioryearsis7.4%paredto9.0%afterthedividendcut(t-statisticforthedifferenceequals4.7).However,asisevidentfromthefigure,theincreaseinriskursgraduallyduringtheyearprecedingtheannouncementofthedividendcut,andlittleifanyadditionalincreasesurafterday–
1.Inparticular,duringtheintervals(–100,–51)and(–50,–1),theriskpremiumgraduallyincreasesfromlessthan8%tomorethan10%.Itremainsatthislevelforthenexttwoyears,andthendeclinesgraduallytoabout7%.Thisevidencesuggeststhatdividenddecreasesfollowratherthansignalincreasesinequityrisk. Toexaminetherobustnessofthisresult,Inextusealternativeapproachesforcalculatingtheriskpremium.First,Ireportthemediansoftherelativedayriskpremiumsinsteadofthemeans.Themedianestimates,plottedinPanelBofFigure4,areverysimilartothemeans(PanelA).Second,insteadofcalculatingmeansormediansforgroupsofconsecutivetradingdays,Ifitasplinefunctionwith49knots(correspondingto50intervals)forthewholesampleof2,500relativedayriskpremiums(days–1,250through1,249).PanelCofFigure4presentsthefittedvalueofthesplinefunction.Itisevidentthatalloftheincreaseintheriskpremiumursduringtheyearprecedingtheannouncementofthedividendcut. TheriskpremiumestimatesinPanelsAthroughCofFigure4arecalculatedbyrunningFFregressionsforeachrelativetradingday.Eachregressionincludesalldividenddecreasefirms 19 withavailablestockreturnsfortheparticularrelativeday,andtheestimatesarebasedontheassumptionthatallfirmshavethesamefactorloadings.Theprimaryadvantagesofthisapproacharethatitallowsthefactorloadingstovaryoverrelativetime,andthefactorloadingsareestimatedusingrelativelylargesamples(upto282dividenddecreasefirms).Yet,dividenddecreasefirmsarenotidenticalandmaythereforehavedifferentfactorloadings.Toaddressthisconcern,Ireruntheanalysisusingtime-seriesFFregressions.Specifically,foreachfirm,Iestimatetime-seriesFFregressionsforgroupsof50consecutivetradingdays((–1,250,–1,201),(–1,200,–1,151),…,(–50,–1),(0,49),…,(1,200,1,249)).Ithencalculatethefirm/intervalspecificriskpremiumusingtheestimatedfactorloadingsandtheaverageannualizeddailyfactorreturnsovertheperiod1963-2003.Finally,foreachinterval,Icalculatethemeanandmedianvaluesoftheriskpremiumacrossallfirms.TheresultingestimatesareplottedinPanelsD(means)andE(medians)ofFigure4.17 Consistentwiththesmallnumberofobservationsperregression,theestimatesinPanelsDandEofFigure4aremorevolatilethanthoseinPanelsAandB.Yet,inbothPanelsCandD,theriskpremiumclearlyincreasespriortothedividendannouncementanditsaveragevalueinthe150tradingdayspriortothedividendchange(thethreepointstotheleftofthey-axis)issimilartothepost-changevalues.Thus,thefindingthatdividendcutannouncementsfollowratherthansignalanincreaseinriskisrobusttotheuseofalternativeapproachesformeasuringriskpremium. 17Tomitigatetheimpactofnon-synchronoustrading,IincludeintheFFregressionstheone-daylagandleadvaluesofeachofthethreefactorsandusetheScholesandWilliams(1977)approachtoadjusttheestimatedfactorloadings. 20
E.RobustnessChecksInextconducttworobustnesschecksfortheearningsanalysis.First,Irerunalltests involvinganalysts’forecastsusingthemostrecentanalysts’EPSandgrowthforecastsinsteadoftherespectiveconsensusforecasts.Iobtainqualitativelysimilarresultstothosereportedabove.Inparticular,whenregressingequation(5)withCORE_EPSasthedependentvariable,Ifindthatthecoefficientofthedividenddecreasevariableis–0.021(t-statisticof–3.6)foryeart=
0,–0.029(t-statisticof–4.4)fort=1,andinsignificantfort=2,3,4,and5.Theseestimatesareslightlylargerinmagnitudethanthoseobtainedwiththeconsensusforecasts,suggestingthattheconsensusforecastsaremoreprecisethanthemostrecentforecasts.Theinference,however,remainsunchanged:Dividendcutannouncementssignallowercurrentandnextyearearnings. Second,toevaluatethemethodology,Iestimatetheregressionsforlargedividendincreases(greaterthanorequalto20percent)insteadofdividenddecreases.18IfindthatthedividendincreasecoefficientwhenexplainingCORE_EPSispositiveandsignificantforeachofthesixyears(t=0,
1,…,5),butitsmagnitudeisrelativelysmall(rangingbetween0.007and0.013witht-statisticsrangingbetween2.1and5.5).Thus,whilebothdividenddecreasesandincreasespredictsubsequentearnings,theinformationindividenddecreasesrelatesprimarilytothenearfuturewhilethatindividendincreasesismorepermanent.Thisdifferencecouldbeduetoountingconservatismwhichrequiresthatlossesberecognizedwhenanticipatedwhileprofitsshouldberecognizedonlywhenearned.Alternatively,theshortdurationofthenegativeearningsimplicationsofdividendcutscouldbeduetorealoptions,suchastheabandonmentoption,whichallowdividenddecreasefirmstoimproveordiscontinueessfuloperations. 18Ifocusonlargedividendincreasesbecause,asshowninTableI,dividenddecreasesaresubstantiallylargerinmagnitudethandividendincreases. 21 VI.ConclusionExtantresearchestablishesthatdividenddecreases:(1)triggernegativestockreturns,
(2) areunrelatedtofutureearningsaftercontrollingforcurrentearnings,and(3)areassociatedwithanincreaseinfirmrisk.Thisevidencehasbeeninterpretedassuggestingthatdividendcutannouncementssignalanincreaseinrisk,whichinturntriggersanegativemarketresponse(e.g.,Grullonetal.(2002)).Ireexaminetheinformationcontentofdividenddecreasesandfindthatrealizedcurrentearningsofdividenddecreasefirmscontainnegativespecialitemsandaresubstantiallysmallerthanexpectedatthetimeofthedividendchange.Theseresultsimplythatcontrollingforcurrentearningsinexplainingfutureearningsbiasesthedividendcoefficientagainstfindinginformationindividenddecreases. Ithusestimateanalternativemodelforexpectedearningswhichextractsinformationfromanalysts’forecasts,priceandothervariables,butexcludescurrentearnings.Usingthismodel,Ifindthatdividendcutannouncementsareassociatedwithnegativeunexpectedearningsinthesubsequentyear.Ifurthershowthatthepresentvalueofunexpectedearningsinthecurrentandnextyearisapproximatelyequaltothedividendcutannouncementreturn.Finally,Idemonstratethattheincreaseintheriskinessofdividendcutfirms,whichhasbeendocumentedbypreviousstudies,ursprimarilyintheyearpriortothedividendannouncement.Ithereforeconcludethatdividenddecreaseannouncementsconveyearningsratherthanriskinformation. 22 REFERENCES Aharony,Joseph,andItzhakSwary,1980,Quarterlydividendandearningsannouncementsandstockholders’returns:Anempiricalanalysis,JournalofFinance35,1-12. Allen,Franklin,andRoniMichaely,2002,Payoutpolicy,Workingpaper,UniversityofPennsylvania. Baker,Malcolm,andRichardS.Ruback,1999,Estimatingindustrymultiples,Workingpaper,HarvardUniversity. Beaver,William,andDaleMorse,1978,Whatdeterminesprice-earningsratios?
FinancialAnalystsJournal34,65-76. Benartzi,Shlomo,RoniMichaely,andRichardThaler,1997,Dochangesindividendssignalthefutureorthepast?
JournalofFinance52,1007-1034. Burgstahler,David,JamesJiambalvo,andTerryShevlin,2002,Dostockpricesfullyreflecttheimplicationsofspecialitemsforfutureearnings?
JournalofountingResearch40,585-612. Chen,Shuping,TerryShevlin,andYenHeeTong,2004,Whatistheinformationcontentofdividendchanges?
Anewinvestigationofanoldpuzzle,Workingpaper,UniversityofWashington. DeAngelo,Harry,LindaDeAngelo,andDouglasJ.Skinner,1992,Dividendsandlosses,JournalofFinance47,1837-1863. Elliott,JohnA.,andDouglasJ.Hanna,1996,Repeatedountingwrite-offsandtheinformationcontentofearnings,JournalofountingResearch34,135-155. Fairfield,PatriciaM.,1994,P/E,P/Bandthepresentvalueoffuturedividends,FinancialAnalystsJournal50,23-31. Fama,EugeneF.,andhR.French,1993,Commonriskfactorsinthereturnsonstocksandbonds,JournalofFinancialEconomics33,3-56. Fama,EugeneF.,andhR.French,2001,Disappearingdividends:Changingfirmcharacteristicsorlowerpropensitytopay,JournalofFinancialEconomics60,3-43. Gordon,MyronJ.,1962,TheInvestment,Financing,andValuationoftheCorporation(Irwin,Homewood,IL) Greenspan,Alan,1997,aryPolicyReporttoCongressPursuanttotheFullEmployment&BalancedGrowthActof1998,SuperintendentofDocuments. 23 Grullon,Gustavo,RoniMichaely,andBhaskaranSwaminathan,2002,Aredividendchangesasignoffirmmaturity?
JournalofBusiness75,387-424. Grullon,Gustavo,ShlomoBenartzi,RoniMichaely,andRichardH.Thaler,2003,Dividendchangesdonotsignalchangesinfutureprofitability,JournalofBusiness(ing). Healy,PaulM.,andKrishnaG.Palepu,1988,Earningsinformationconveyedbydividendinitiationsandomissions,JournalofFinancialEconomics21,149-175. Jacksonth,ScottB.,andMarshallK.Pitman,2001,Auditorsandearningsmanagement,TheCPAJournal71,38-44. Kirschenheiter,Michael,andNahumMelumad,2001,Can“bigbath”andearningssmoothingco-existasequilibriumfinancialreportingstrategies?
JournalofountingResearch40,761-96. Kothari,
S.P.,2001,Capitalmarketsresearchinounting,JournalofountingandEconomics31,105-231. Michaely,Roni,RichardH.Thaler,andKentL.Womack,1995,Pricereactionstodividendinitiationsandomissions:Overreactionordrift?
JournalofFinance50,573-608. Miller,MertonH.,andKevinRock,1985,Dividendpolicyunderasymmetricinformation,JournalofFinance40,1031-1051. Newey,WhitneyK.,andhD.West,1987,Asimplepositivesemi-definite,heteroskedasticityandautocorrelationconsistentcovariancematrix,Econometrica55,703-708. Nissim,Doron,andAmirZiv,2001,Dividendchangesandfutureprofitability,JournalofFinance56,2111-2134. Scholes,Myron,andJosephWilliams,1977,Estimatingbetafromnon-synchronousdata,JournalofFinancialEconomics5,309-327. Skinner,DouglasJ.,2003,Whatdodividendstellusaboutearningsquality?
Workingpaper,UniversityofMichigan. Za
DoronNissimColumbiaUniversityGraduateSchoolofBusiness3022Broadway,UrisHall604NewYork,NY10027(212)854-4249,dn75@columbia.edu February2005 IthankGeertBekaert,LawrenceGlosten,GurHuberman,ParthaMohanram,StephenPenman,DouglasSkinner,JacobThomas,andseminarparticipantsatColumbiaUniversityandtheInterdisciplinaryCenterforhelpfulsuggestionsments. TheInformationContentofDividendDecreases:EarningsorRiskNews?
Abstract Thispaperdemonstratesthatdividendcutannouncementsconveynewinformationregardingearningsinthecurrentandsubsequentyear.Whiledividenddecreasesarealsoassociatedwithanincreaseinfirmrisk,thechangeinriskurspriortothedividendannouncement.Consistentwiththesefindings,theabnormalstockreturnduringthethree-daydividendcutannouncementwindowisapproximatelyequaltothepresentvalueofunexpectedcurrentandnextyear’searnings.
I.IntroductionDividendcutannouncementstriggersubstantialnegativereturns(e.g.,Aharonyand Swary(1980),Michaelyetal.(1995)).Yet,priorresearchgenerallyfindsthatdividendreductionsareunrelatedtofutureearningsaftercontrollingforcurrentearnings(e.g.,HealyandPalepu(1988),Benartzietal.(1997),NissimandZiv(2001)).1Severalrecentstudiesinterpretthisevidenceasindicatingthatdividenddecreasesdonotconveynewinformationaboutfutureearningsbutratherimplyanincreaseinfirmrisk(e.g.,Benartzietal.(1997),AllenandMichaely(2002),Grullonetal.(2002),Grullonetal.(2003),Chenetal.(2004)).Thisstudyprovidesnewevidenceontheearningsandriskimplicationsofdividendcuts.Inparticular,Ishowthatessentiallyallofthemarketresponsetodividendcutannouncementsisduetonewinformationaboutcurrentandnextyear’searnings,andthatlittleifanyriskinformationisconveyedbytheseannouncements. Thedifferencesbetweenthefindingsofthecurrentandpriorstudiesaredueprimarilytomethodologicalchoices.Specifically,Iusealternativeapproachesformeasuringearningsexpectationsandexaminechangesinriskpremiumsusingshortratherthanlongintervals.Whilethesechoicespotentiallyinvolvetrading-offonebiasforanother,Iprovideevidencethatsupportsmychoicesanddemonstratetheirimpactontheresults.Inparticular,Idocumentthesourcesandextentofbiasassociatedwithusingdifferentmethodologiesformeasuringexpectedearnings.Ialsoshowthatusinglongintervalsforestimatingriskdisguisesthefactthatessentiallyalloftheincreaseintheriskinessofdividendcutfirmsursbeforethedividendannouncement.Finally,Idemonstratethatmyestimatesofthemagnitudesofearningsandrisknewsconveyedbydividendcutannouncementsareconsistentwithobservedmarketresponses. 1OneexceptionisDeAngeloetal.(1992).Focusingonfirmswithpositiveearningsanddividendsinthepriortenyearsandalossinthecurrentyear,DeAngeloetal.(1992)findthatthosefirmsthatcuttheirdividendsinthecurrentyearreportlowernextyearearningsevenaftercontrollingforcurrentyearearnings.
1 Thestudyproceedsasfollows.SectionIIdiscussesdifficultiesandtrade-offsinvolvedinmeasuringexpectedearningsfordividenddecreasefirms.SectionIIIdevelopsthemethodologyusedtoexaminethefutureearningsimplicationsofdividendcuts.SectionIVdescribesthesample,andSectionVpresentstheempiricalfindingsoftheearningsandriskanalyses.SectionVIconcludesthepaper. II.IssuesinMeasuringEarningsExpectationsforDividendDecreaseFirmsToexaminethefutureearningsimplicationsofdividendchangeannouncements, previousstudiesregressfutureearningsonthedividendchange,currentearnings(i.e.,earningsintheyearduringwhichthedividendchangeurs)andproxiesforexpectedfutureearnings.Theythenusetheestimatedcoefficientofthedividendchangeasameasureoftheinformationcontentofdividends.Currentearningsareincludedintheregressionbecausetheyhelpexplainfutureearningsandarepartiallypredictableatthetimeofthedividendchange(e.g.,usinginformationfromintermediatequarterlyreportsorfromthefinancialpress).Asmanyofthesestudiesrecognize,however,thischoiceinducesbiasagainstfindinginformationcontentindividendsbecausecurrentearnings,whichareonlypartiallyknownatthetimeofthedividendannouncement,arepositivelyrelatedtothedividendchange.Thus,thedecisiontocontrolforcurrentearningspotentiallyinvolvestrading-offbiasagainstversustowardseptingtheinformationcontentofdividendshypothesis. Howimportantisthischoice?
ordingtoMillerandRock(1985),thisissueisatthecoreoftheinformationcontentofdividendhypothesis.MillerandRockarguethatdividendchangesconveyinformationaboutfutureearningsindirectlybychangingthemarket’sestimateofcurrentearnings,whichinturncontributestotheestimateoffutureearnings.Thus,ifdividenddecreasespredictlowercurrentearnings,whichinturnimplylowerfutureearnings,thenegative
2 futureearningsimplicationsofthedividendchangewillbecapturedbythecoefficientofcurrentearningsratherthanbythedividendcoefficient.Inotherwords,controllingforcurrentearningsexposestheanalysistotheriskof“throwingthebabyoutwiththebathwater.” TheMillerandRockargumentappliesbothtodividendincreasesanddecreases.Butcontrollingforcurrentearningsmayintroduceanadditionalbiaswhichisuniquetodividenddecreases.Totheextentthatdividendcutfirmsaremorelikelytorecognizenegative“specialitems,”theirearningswillbelesspersistentthanthoseofotherfirms.2Ifthisisnotincorporatedintheanalysis,themodelwilloverstatethepersistenceofthenegativeshocktoearningsinthedividenddecreaseyear,leadingtoanunderstatementofexpectedfutureearnings.ordingly,unexpectedearnings(i.e.,realizedearningsminusexpectedearnings)ofdividenddecreasefirmswillbeoverstated,andthisbiasmayoffsetthenegativeearningsimplicationsofdividendcuts. Negativespecialitemsmaybemoreprevalentfordividenddecreasefirmsduetotwoountingphenomena:conservatismand“bigbath.”Conservatismisanountingconventionwhichrequirestherecognitionofanticipatedlosses,suchasthoseduetoimpairmentofassetsorrestructuringofoperations.Thus,ifdividendreductionssignallowerfuturecashflows,theyshouldbeassociatedwithimpairmentlosses,restructuringchargesandothernegativespecialitems.Thebigbatheffectisrelatedtomanagementbehavior.Theoretical,empiricalandanecdotalevidencesuggeststhatinthepresenceofbadnews(suchasthosecausingadividendcut),managersoftenchoosetotakea“bigbath;”thatis,theyoverstateestimatedliabilitiessuchasruedrestructuringcostsortheywritedownassets.3Thesechargesaregenerallynot 2Skinner(2003)showsthatthepersistenceofearningsispositivelyrelatedtothelevelofdividends,especiallyforlowlevelsofdividends.Thatis,lowdividendfirmshaveparticularlylowearningspersistence.Thisevidencesuggeststhatdividenddecreasefirmsmayalsohavelowearningspersistence.Thecurrentstudyexaminesthisconjecturebyfocusingonthemagnitudeoftransitoryearnings.
3 expectedtorecurinfutureperiodsandaretypicallyclassifiedas“specialitems.”Inextdiscussthemethodologyusedinthisstudytomitigatethesebiasesinmeasuringexpectedearnings. III.Methodology Withaperfectmeasureofexpectedearningsimmediatelypriortothedividend announcement,testingwhetherdividendcutsconveynewearningsinformationis straightforward;oneonlyneedstocalculatethet-statisticofthemeandifferencebetween realizedandexpectedearningsforasampleofdividenddecreasefirms.Equivalently,onecould runthefollowingregression: EPS(t)/P=E[EPS(t)]/P+λDIV_DEC+ε,
(1) whereEPS(t)isearningspershareinyeart(t=0,1,
2,…),E[.]istheexpectedvalueoperator conditionedonpubliclyavailableinformationpriortothedividendannouncement(inyear0),
P ispricepersharepriortothedividendannouncement,andDIV_DECisaqualitativevariable indicatingadividenddecrease.Realizedandexpectedearningspersharearedeflatedbyprice persharetomitigateheteroscedasticitysincethestandarddeviationofEPSisapproximately proportionaltoprice(e.g.,BakerandRuback(1999)). Unfortunately,E[EPS(t)]isunobservable.Ithereforeestimatethisquantityusing informationfromtwomarket-basedproxiesforexpectedearnings:analysts’forecastsandprice pershare.Specifically,ImodelexpectedEPSasfollows: E[EPS(t)]=βECR(t)×P+δAF(t),
(2) whereECR(t)(EarningsCapitalizationRate)isaproxyfortherateatwhichinvestorscapitalize expectedEPS(t)intoprice,andAF(t)(Analysts’Forecast)isameasureofavailableanalysts’ 3Forexamplesofanecdotal,empirical,andanalyticalevidence,seeJacksonthandPitman(2001),ElliottandHanna(1996)andKirschenheiterandMelumad(2002),respectively.
4 EPSforecastforyeartatthetimeofthedividendannouncement.Bothanalysts’forecastsand pricereflectexpectationsofmarketparticipantsregardingfutureearnings.However,prior researchestablishesthatbothmeasuresaresomewhatinefficientandbiased,andthateachcontainsincrementalinformationrelativetotheother.4Thus,byincorporatinginformationfrom bothmeasures,equation(2)maygenerateamoreefficientestimateofexpectedearningsthan estimatesbasedonpriceoranalysts’forecastsalone. Therateatwhichinvestorscapitalizeexpectedearnings,ECR(t),isafunctionof macroeconomicvariables(e.g.,interestrates,marketriskpremium)aswellasfirm-specificrisk,expectedgrowthandpayout.5IthereforespecifyECR(t)asabinationofthefollowing variables: ECR(t)=αyear+αindustry+αdistance+α1BETA+α2VOLAT +α3LEV+α4SIZE+α5BM+α6GROWTH+α7YIELD,
(3) whereαyearisayearfixedeffect,includedtocapturetheaverageeffectofmacroeconomic variables;αindustryisanindustry(2-digitSIC)fixedeffect,whichcontrolsforindustry-specific riskandgrowthprospects;andαdistanceisafixedeffectwhichcontrolsforthenumberofmonths betweenthedividendannouncementandtheendofthefiscalyear.Thislattereffectisimportant becausethepresentvalueofearningsdependsonthelengthoftimeuntiltheirrealization. Thecontinuousvariablesinequation(3)monlyusedmeasuresofrisk,expected growthandpayout.BETAisameasureofsystematicrisk,estimatedfrommonthlystockreturns 4Forareviewofthisliterature,seeKothari(2001).Notethattheprecisionoftheprice-basedmeasureofexpectedearningsisaffectedbythequalityoftheempiricalproxyforECR(t)inadditiontopriceefficiency.Similarly,analysts’forecastsmaycontainerrorduetotheirdiscretenature(i.e.,theymaybestaleatthetimeofthedividendannouncement)inadditiontoerrorduetoanalysts’inefficienciesinprocessinginformation. 5Forexample,undertheGordon(1962)model,ECR
(1)=(r–g)/d,whereristhecostofequitycapital,gisexpectedgrowth,anddisthedividendpayout.ForempiricalspecificationsofECRmodels,seeBeaverandMorse(1978),Zarowin(1990),Fairfield(1994),andGreenspan(1997).
5 andtheCRSPvalue-weightedreturnsduringthefiveyearsthatendinthemonthpriortothe dividendannouncement.VOLATreflectsidiosyncraticriskandismeasuredastheroot-mean- squarederrorfromtheBETAregression.LEV,theratiooftotalliabilitiestothesumoftotal liabilitiesandthemarketvalueofequity,isaproxyforfinancialrisk.SIZE(logofmarketvalue ofequity)andBM(book-to-marketratio)captureriskandgrowthprospects.GROWTHisa measureofanalysts’long-termearningsgrowthforecastavailableatthetimeofthedividend announcement.YIELDistheratiooftheindicatedannualdividend(themostrecentquarterlydividendtimesfour)dividedbyclosingpricetwodayspriortothedividendannouncement.6 Inextsubstituteequation(3)intoequation(2)anddividebothsidesbyprice: E[EPS(t)]/P=γyear+γindustry+γdistance+γ1BETA+γ2VOLAT+γ3LEV +γ4SIZE+γ5BM+γ6GROWTH+γ7YIELD+δAF(t)/P
(4) Finally,Isubstituteequation(4)intoequation
(1): EPS(t)/P=γyear+γindustry+γdistance+γ1BETA+γ2VOLAT+γ3LEV+γ4SIZE +γ5BM+γ6GROWTH+γ7YIELD+δAF(t)/P+λDIV_DEC+ε
(5) Totestwhetherdividenddecreasesprovidenewinformationaboutcurrent(t=0)andfuture(t=
1,…,5)earnings,Iestimateequation(5)forasampleofdividendpayingfirmswhichincludes dividendincreases,dividenddecreases,andno-changeobservations.Thus,thecoefficientonthe DIV_DECindicatorvariablereflectstheincrementalearningsassociatedwithdividend decreases.Ifocusondividendpayingfirmsbecausethemagnitudesofbiasanderrorinthe 6Iusetheyieldratherthanthepayoutbecauseearningsarenegativeorclosetozeroformanydividenddecreasefirms.Moreover,totheextentthatpriceisaproxyfor“permanentearnings,”theyieldislikelytobeabetterproxyforexpectedfuturepayoutthantheactualpayoutduetothestabilityofdividendpayments.
6 proxiesforexpectedearningsmaybedifferentfortheseparedtofirmsthatdonotpaydividends.7 Whileitisimpossibletoremoveallerrorandbiasfromtheproxiesforexpectedearnings,itisimportanttonotethattheECR(t)variables(equation
(3))notonlyimprovetheprecisionoftheprice-basedproxyforexpectedearnings,butalsomitigatepotentialbiasesduetoinefficienciesinanalysts’forecasts.Forexample,priorresearchdocumentsthatthebiasinanalysts’forecastsdecreasesastheearningsannouncementdateapproaches.Thus,theinclusionofαdistanceintheregression(fixedeffectforthenumberofmonthsuntiltheendofthefiscalyear)mitigatespotentialbiasduetosystematicvariationintherelativefrequencyofdividenddecreasesduringthefiscalyear. Mostfirmspaydividendsonaquarterlybasis,sothesameannualearningsmaybeassociatedwithuptofourobservations.TheresultingoverlapinobservationshaslittleeffectontheDIV_DECcoefficientbecauseveryfewfirmshavemorethanonedividendreductioninanygivenyear(resultsarenotsensitivetotheexclusionoftheseobservations).Also,evenforoverlappingobservations,thevaluesofthedependentvariablearenotidenticalbecauseprice(thedeflator)ismeasuredatdifferentpointsintime(reflectingallinformationuptotheparticulardividendannouncement).Inanycase,tomitigatetheeffectofautocorrelationintheresidualduetooverlappinginformation,IreportNeweyandWest(1987)t-statisticswiththreelags. 7Firmsthatdonotpaydividendsaretypicallysmallwithlowprofitabilityandstronggrowthopportunities(seeFamaandFrench(2001)).
7 IV.DataTheinitialsampleincludesalldividendeventsintheCRSPfilesthatsatisfythefollowing criteria:(1)panypaidanordinarycashdividend(
U.S.dollars)inthecurrentquarterandinthepreviousquarter,(2)nootherdistributionswereannouncedbetweenthedeclarationofthepreviousdividendandthreedaysafterthedeclarationofthecurrentdividend,(3)therewerenoex-distributiondatesbetweentheex-distributiondatesofthepreviousandcurrentdividends,and(4)theindicatedannualdividendyieldpriortothedividendannouncementisatleast0.5%.Criteria(2)and(3)helpensurethatonly“clean”dividendchangesthatavoidconfoundingeffectsfromotherdistributionsareidentified.Criterion(4)issettorestrictthesampletofirmsthatpaysignificantdividends. Foreachofthedividendevents,IextractfromCRSPthemonthlystockreturnsinthepriorsixtymonthsandcalculatetheBETAandVOLATmeasuresdiscussedabove(atleast30returnobservationsarerequired).Next,ImatchthedividendeventsamplewithforecastandreporteddataextractedfromtheIBESDetail,SummaryandActualfiles.TheDetailfilecontainsanalyst-by-analystestimates,whiletheSummaryfilecontainstheconsensus(meanandmedian)analysts’forecastsineachmonth.ordingtoIBES,theDetailfileisareconstructionofarchiveddataandmaythereforecontainerrorsandomissions.Individualanalystforecastsmayalsobelessprecisethantheconsensusestimates.Ithereforeusetheconsensus(mean)forecastsintheprimaryanalysis,butcheckthesensitivityoftheresultstotheuseofindividualforecasts.Specifically,IrerunallanalysesusingthemostrecentEPSandgrowthforecastsavailableattimeofthedividendchangeandreporttheresultsofthesetestsinarobustnesssectionbelow. TheSummaryIBESfileconsistsofchronologicalsnapshotsofconsensusearningsforecaststakenontheThursdaybeforethethirdFridayofeverymonth.ordingtoIBES,theconsensusforecastsaremadeavailableinthefollowingweek.Ithereforematchdividend
8 announcementsurringafterthefourthSundayofeachmonthandbeforethefourthMondayofthefollowingmonthwiththeconsensusforecastsforthatmonth.IrequirethatthefollowingvariablesbeavailablefromIBES:meananalysts’EPSforecastforthecurrent(AF
(0))andnextyear(AF
(1)),aswellasmeanlong-termearningsgrowthforecast(GROWTH).8ordingtoIBES,thelong-termearningsgrowthforecastgenerallyreferstoaperiodofuptofiveyearsahead.Ithereforegenerateforecastsforearningsinfutureyears2through5byapplyingthemeanlong-termgrowthforecasttothemeanforecastfortheprioryearinthehorizon;i.e.,AF(t+s)=AF(t+s-1)×(1+GROWTH),fors=2,3,4,
5. TheresultingsampleisthenmatchedwiththeCOMPUSTATfilesbyassigningeachdividendeventtotheearliestfiscalyearforwhichearningshavenotyetbeendisclosed.Thatyearisdesignatedasyear0.Dividendsareoftenannouncedsimultaneouslywithquarterlyearnings,withapositivecorrelationbetweenthetwodisclosures(e.g.,firmsarelikelytoincreasedividendsfollowingearningsincreases;see,e.g.,Benartzietal.(1997)).Toassurethatthedividendeventsreflectonlydividendinformation,Iexcludeobservationsforwhichthedividendannouncementurswithinthreecalendardaysbeforeorafteraquarterlyearningsannouncement.9InextextracttherequiredfinancialstatementsinformationfromCOMPUSTATanddeleteobservationswithmissingvaluesforcurrentearningsorforanyoftheexplanatoryvariablesofequation
(5)(observationswithmissingvaluesforfutureearningsareretained,toavoidsurvivorshipbias).Finally,tomitigatetheeffectofinfluentialobservations,thevariablesarewinsorizedatthe1thand99thpercentilesoftheirempiricaldistributions.10 8Topreventduplication,Idonotincludeforecastswith“secondary”flag.IadjustallpersharenumbersforstocksplitsandstockdividendsusingtheIBESadjustmentfactor.Ifthefirmisfollowedonadilutedbasis,IusetheIBESdilutionfactortoconvertpersharevariablestoaprimarybasis.9EarningsannouncementdatesareextractedfromthequarterlyCOMPUSTATfiles.
9 Table1presentsdescriptivestatisticsforthefinalsample.Asreported,thereare282dividenddecreases,5,576dividendincreases,and35,769no-changeannouncementsthatsatisfyalltheabovecriteria.Thesedividenddeclarationsurredduringtheyears1982through2003(IBESdataareavailablesincetheearly1980s).Whiledividendreductionsaremuchmonthandividendincreases,theyareconsiderablylargerinmagnitude—themeanpercentagechangeindividendspershareis–45.6%fordividenddecreasesand12.9%fordividendincreases.Dividenddecreasefirmsarerelativelysmallandhavelargebook-to-marketratios,financialleverageandresidualvolatility,suggestingthattheyaremoreriskythanotherdividendpayingfirms.However,theaveragebetaofdividenddecreasefirmsissmallerthanthatofotherdividendpayingfirms.Inaddition,dividenddecreasefirmshavesubstantiallysmalleranalysts’long-termgrowthforecastsandhigherdividendyields.
V.EmpiricalFindingsA.Time-seriesBehaviorofEarningsandSpecialItems Pastlevelsofearningsprovidenaturalbenchmarksforcurrentandfutureearnings.IthusexaminesummarystatisticsfromcrosssectionaldistributionsofEPSintheyearssurroundingdividenddecreases,andinparticulartestthesignificanceofchangesinthesestatisticssubsequenttothedividendcut.11Figure1plotsthedistributionofEPSineachoftheelevenyearssurroundingthedividenddecrease,scaledbyclosingpricetwodayspriortothedividendannouncement(allpersharedataareadjustedforstocksplitsandstockdividends).Thestatisticsplottedarethe25th,50thand75thpercentiles,andthemean.PanelAofTable2presentsthemean 10Qualitativelysimilarresultsareobtainedwithoutwinsorizing,withtrimminginsteadofwinsorizing,andwithalternativepercentilecuts.11EPSismeasuredasCOMPUSTAT’sdataitem#58(basicearningspersharebeforeextraordinaryitemsanddiscontinuedoperationsandadjustedforpreferreddividends). 10 andmedianofearningsinthesevenyearssurroundingthedividenddecrease,aswellaschangesinthesequantitiesandthesignificanceofthechanges.12Asshown,currentearnings(i.e.,earningsinyear0)areconsiderablysmallerthaninthesurroundingyears,althoughthedecreaseinearningsstartsinyear–1(inyears–5through–2earningsarerelativelystable).Meanearningsdropsfrom0.109to0.068inyear–1(t-statisticforthechangeequals–7.5),andfrom0.068to–0.023inyear0(t-statisticforthechangeequals–10.7).Fromyear1on,earningsincreasemonotonicallybutataslowerpacethantherateofdeclineinyear0.Similarresultsareobtainedformedianearnings,indicatingthatthesetrendsarenotduetooutlierobservations. Consistentwiththeconjecturethatfirmsaremorelikelytoreportnegativespecialitemsindividenddecreaseyears,thelefttailoftheearningsdistributionflattensinyear0(e.g.,thedifferencebetweenthe50thand25thpercentilesinFigure1essubstantiallylarger).Theflatteningofthelefttailisalsoreflectedintherelationshipbetweenthemeanandthemedian:Meanearningsissimilartothemedianinyears–5through–1,butinyear0themeanisconsiderablysmallerthanthemedian.Toverifytheinferencethatfirmsreportabnormallevelsofnegativetransitoryitemsindividenddecreaseyears,Inextexaminethedistributionofspecialitems.13 Figure2plotsthecross-sectionaldistributionsofspecialitemspershareintheyearssurroundingthedividendcut,andPanelBofTable2presentssummaryandteststatistics(mean,frequencyofnegativevalues,changesinthesestatistics,andt-statisticsforthechanges).Toholdsizeconstant,specialitemsarescaledbyclosingpricetwodayspriortotheannouncementofthe 12Thechangestatisticsareslightlydifferentfromthecorrespondingdifferencesbetweentheconsecutivelevelsbecausetheyarecalculatedusingobservationswithnon-missingvaluesforbothyears.13SpecialitemsaremeasuredasCOMPUSTAT’sdataitem#17.Thisitemincludesdifferenttypesofrevenuesandexpensesfromcontinuingoperationswhichareclassifiedbyfirmsastransitory(e.g.,write-offs,impairmentsandrestructuringcharges).Specialitemsaremuchmonandlargerinmagnitudethanextraordinaryitems,whicharerarelyreportedpanies.Foradetaileddiscussionofspecialitems,seeBurgstahleretal.(2002). 11 dividendcut.Meanspecialitemsdecreasesfrom–0.013inyear–1to–0.043inyear0(t-statisticforthechangeequals–5.2)andincreasesbackto–0.017inyear1(t-statisticequals3.8).This“V”shapepatternisespeciallyapparentforthe10thpercentileofthedistribution(seeFigure2).Inaddition,thefrequencyofnegativespecialitemsincreasesfrom31.3percentinyear–1to50.2percentinyear0(t-statisticforthechangeequals4.2),anddecreasesbackto35.0percentinyear1(t-statisticequals–3.4).Afteryear1,thefrequencyofnegativespecialitemsremainsrelativelystable. Tosummarize,firmsthatcuttheirdividendsreportverylowearningsintheyearofthedividendchange,inlargepartduetotherecognitionofnegativespecialitems.Consistentwiththelowpersistenceofspecialitems(Burgstahleretal.(2002)),thesefirmsreportsubstantialEPSincreasesinsubsequentyears.Yet,earningsreachtheirpre-dividendcutlevelonlythreeyearsafterthedividenddecrease.Althoughtheseresultssuggestthatdividenddecreasespredictlowercurrentandfutureearnings,theydonotindicatewhetherthisinformationisrevealedbythedividendannouncementratherthanbypriordisclosuressuchasinterimearningsreports.Inextusemarket-basedinformationtoconstructmeasuresofexpectedearningsimmediatelypriortothedividendchange,whichallowmetotestwhetherdividendcutannouncementsconveynewearningsinformation.
B.DividendDecreasesandUnexpectedEarningsIstartbyexaminingtheassociationbetweenanalysts’forecasterrorsanddividend changes.Totheextentthatanalysts’EPSforecastsreflectthemarketexpectationofcurrentandfutureEPS,theaveragevalueoftheforecasterrors(i.e.,theaveragedifferencebetweenrealizedandforecastedEPS)fordividenddecreasefirmsshouldindicatewhetherdividendcutsconveynewearningsinformationtothemarket.Specifically,iftheaveragevalueofanalysts’forecast 12 errorsfordividenddecreasefirmsisnegativeandsignificant,theinferencewouldbethatdividendreductionsconveynegativeearningsnews. Thisapproachfortestingtheinformationcontentofdividendsisstraightforward(analysts’forecastsprovideadirectmeasureofexpectedearnings)andsimpletoimplement.However,ithasimportantings:Analysts’forecastsareonaveragebiasedupwardandtheydonotreflectallavailableinformation(forareviewoftheliteraturedocumentingtheseresults,seeKothari2001).Iparetheforecasterrorsofdividendcutfirmswiththoseofotherdividendpayingfirmsratherthanassumethattheunconditionalmeanoftheforecasterroriszero.Inaddition,Iconductmultivariateanalysesthatmitigatetheeffectofmeasurementerrorintheproxyforexpectedearningsbyincorporatinginformationfromstockpricesandothervariablesinadditiontoanalysts’forecasts. TheresultsoftheunivariateanalysisarereportedinTable3.Thistablepresentsthemeanandmediananalysts’EPSforecasterrorsfordividenddecreasefirmsandforallotherdividendpayingfirms.Analysts’forecasterrors(labeled“unexpectedIBESEPS”)aremeasuredasthedifferencebetweentheIBESmeasureofactualEPSandtheconsensusEPSforecastavailableatthetimeofthedividendannouncement.ActualEPSasmeasuredbyIBESisequaltoreportedEPSminusitemsclassifiedbyanalystsasnon-recurring(“IBESspecialitems”).ThetotaldifferencebetweenreportedandforecastedEPS(“unexpectedreportedEPS”)isequaltothesumof“unexpectedIBESEPS”and“IBESspecialitems.” Analysts’forecasterrorsarenegativeforbothdividenddecreasefirmsandotherdividendpayingfirms,andarelargerinmagnitudeinyearparedtoyear0.Bothresultsareconsistentwithpriorresearch,whichdemonstratesthatanalysts’forecastsarebiasedupwardwiththebiasincreasingintheforecasthorizon.However,unexpectedearningsaresubstantiallyandsignificantlysmallerfordividenddecreaseparedtootherdividendpayingfirms. 13 Specifically,unexpectedIBESearningsfordividenddecreasefirmsareonaverage–4.83%ofpriceforyear0and–5.89%foryear1,whilethecorrespondingnumbersforotherdividendpayingfirmsare–0.75%and–1.93%respectively.Theseresultsarenotduetooutliers,asthedifferencesinmedianarealsohighlysignificant. Themagnitudeofnegativespecialitems(middletwocolumnsinTable3)issubstantiallylargerfordividenddecreasefirmsthanforotherdividendpayingfirms,especiallyinyear0(–2.53%ofpricefordividenddecreaseparedto–0.30%forotherdividendpayingfirms).TheseresultsareconsistentwiththestatisticsinPanelBofTable2,whichdemonstratethatfirmsreportmorenegativespecialitemsindividenddecreaseyearsthaninotheryears.Becausespecialitemsarelesspersistentthanotherearningsitems(e.g.,Burgstahleretal.(2002)),thesefindingsimplythatdividenddecreasefirmsshouldexperienceearningsincreasesinfutureyears.Indeed,theresultsofpriorresearch(e.g.,Benartzietal.(1997))andthisstudy(PanelAofTable2)indicatethatdividendcutsarefollowedbyearningsincreasesinsubsequentyears.However,currentearnings,thestartingpointofthisearningsgrowth,aresubstantiallysmallerthanexpectedatthetimeofthedividendcut.Therefore,theleveloffutureearningsmaystillbelowerthanexpectedatthetimeofthedividendcut,inspiteofthefutureearningsincreases.ThestatisticsinTable3suggestthatunexpectedearningsinyeart=1ofdividenddecreasefirmsareindeednegative. ThusfarIhaveusedanalysts’EPSforecaststomeasurethemarketexpectationsofcurrentandfutureearnings.Asdiscussedabove,however,analysts’forecastscontainerrorandbias,whichmaybecorrelatedwiththecharacteristicsofdividenddecreasefirms(e.g.,smallsize,lowexpectedgrowth).Moreover,analysts’forecastsmaybestale,especiallyfordividenddecreasefirms(giventheirsmallsizeandotherlessattractivecharacteristics).Tomitigatetheseeffects,Inextestimateregressionmodel(5)whichincorporatesinformationfrommarketprices 14 andothervariablesinadditiontoanalysts’forecasts.Istartbyreplicatingtheresultsofpriorstudies(e.g.,Benartzietal.(1997),NissimandZiv(2001)),whichfindthatdividendreductionsarenotassociatedwithnegativefutureearningsaftercontrollingforcurrentearnings.Tothisend,Iincludeinequation(5)theratioofcurrentearningstopriceasanadditionalexplanatoryvariable. Table4presentstheregressionresults.Asshown,thecoefficientofthedividenddecreaseindicatorvariable(DIV_DEC)isnon-negativeineachofthefiveregressions(t=1,
2,…,5),confirmingthatdividenddecreasesdonotimplylowerfutureearningsaftercontrollingforcurrentearnings.Butshouldonecontrolforcurrentearnings?
Asdiscussedabove,ifrealizedcurrentearningsofdividenddecreasefirmsaresmallerthanexpectedatthetimeofthedividendchange,includingthemintheregressionwillbiastheresultsagainstfindinginformationcontentindividenddecreases.TheresultsinTables2and3suggestthatrealizedcurrentearningsareindeedsmallerthanexpectedatthetimeofthedividendchange.Moreover,thehighsignificanceofcurrentearningsinexplainingfutureearnings(Table4),evenaftercontrollingforanalysts’forecasts,priceandotherearningspredictors,suggeststhatcurrentearningsreflectinformationaboutfutureearningswhichisnotavailableatthetimeofthedividendchange. MillerandRock(1985)predicttheresultsofTable4.Theyarguethatdividendchangesconveyinformationaboutfutureearningsindirectlybychangingthemarket’sestimateofcurrentearnings,whichinturncontributestotheestimateoffutureearnings.Thus,whencurrentearningsareincludedintheregression,thecoefficientofthedividendchangeshouldbeinsignificant,asIindeedfind.Incontrast,whentheearningsexpectationmodelreflectsonlyinformationwhichisavailableatthetimeofthedividendchange,thecoefficientofthedividendchangeshouldbesignificant.Inextexaminethishypothesisbyestimatingequation
(5),whichexcludescurrentearnings.TheestimatesinTable5demonstratethatdividenddecreasesimply 15 lowercurrent(t=0)andnextyear(t=1)earnings,buthavenoimplicationsforearningsinlateryears. Assumingadiscountratebetween10to20percent,thedividendcoefficientsinTable5implythatdividendcutannouncementsshouldtriggernegativestockreturnsbetween–7.8%(=–0.0570/1.1.5+–0.0274/1.11.5)and–7.3%(=–0.0570/1.2.5+–0.0274/1.21.5).Theseestimates,however,overstatethepriceimpactofdividendcutannouncementsfortworeasons.First,theconsensusanalysts’EPSforecast,whichistheprimarycontrolvariable,reflectsanalysts’expectationsofrecurringearnings,whilethedependentvariablemeasuresrealizedtotalearnings.Thus,ifdividenddecreasefirmsareexpectedtoreportmorenegativespecialitemsthanotherfirms,thedividenddecreasecoefficientwillbebiaseddownward(itwillcaptureexpectedspecialitemsinadditiontounexpectedearningsandsooverstatethenegativeimplicationsofdividenddecreases).Second,negativespecialitemswhichreducereportedearningsoftenreflect“paperlosses”withlittlecashflowconsequences(e.g.,impairmentofgoodwill).Toaddresstheseconcerns,Inextrerunequation(5)usingCORE_EPS(EPSminusspecialitemspershare)insteadofreportedEPS.TheestimatesinTable6suggestasubstantiallysmallervalueeffect,between–3.6%(=–0.0176/1.1.5+–0.0219/1.11.5)and–3.3%(=–0.0176/1.2.5+–0.0219/1.21.5),assumingdiscountratebetween10to20percent.Yet,similartotheresultsinTable5,thedividendcoefficientishighlysignificantinboththet=0andt=1regressions.
C.MarketReactiontoDividendCutAnnouncementTheestimatesfromtheprevioussectionsuggestthatdividenddecreaseannouncements shouldtriggeranaveragestockdeclineofatleast3percent,duetotheirnegativeearningsimplications.Iparethisestimatewiththeactualmarketresponse.IusetheFamaandFrench(FF,1993)three-factormodeltoestimatetheaverageabnormalstockreturnineachof 16 the201tradingdayscenteredatthedividendcutannouncement.Specifically,foreachrelative tradingday(–100through+100),Iregressthefollowingmodelusingallfirmsthatcuttheir dividendsinday0: ER=γ1+γ2RMRF+γ3SMB+γ4HML+et,
(6) whereERisthedailyexcessstockreturn(rawreturnminustherisk-freereturn),RMRFisthe dailyexcessmarketreturn(marketreturnminustherisk-freereturn),SMBisthedailyreturnona portfoliolonginsmallstocksandshortinlargestocks,andHMListhedailyreturnonaportfoliolonginvaluestocksandshortingrowthstock.14Underthisapproach,theintercept(γ1)ofeach relativedayregressionmeasurestheaverageabnormalstockreturninthatdayforfirmsthat announceadividendcutinday0. Figure3plotsthecumulativeaverageabnormalreturnfordividenddecreasefirms(i.e., thecumulativesumoftheregressionintercepts).Firmsthatcuttheirdividendsintradingday0 experiencenegativestockreturnsintheprior100tradingdayswhichsumuptoaCARof–14% bytheendofday–
2.Thethreedaysannouncementreturn(–1,0,and1)isequalto–3.2%,andit isfollowedbyaveryslightdriftinthefollowingweeks.Theoverallmarketresponseis approximately–3.5%,consistentwiththeestimatesofthepresentvalueofunexpectedearningsreportedintheprevioussubsection.15 14FactorreturnsareobtainedfromWRDS.15Theabnormalreturnsareslightlysmallerthaninpriorstudies(e.g.,AharonyandSwary(1980))dueprimarilytotherequirementofavailabilityofanalysts’forecasts(morerecentsampleperiod,largerfirms).ThesmallmagnitudeofthepostannouncementdriftisconsistentwiththeevidenceinGrullonetal.(2002). 17
D.DividendDecreasesandEquityRisk Grullonetal.(2002)findthatdividendchangesareassociatedwithchangesinequity risk.Inparticular,paretheFFfactorloadingsofdividenddecreasefirmsinthethree yearssubsequenttothedividendchangewiththecorrespondingcoefficientsinthethreeprior yearsandreportsignificantincreasesineachofthethreecoefficients.Theyfurtherestimatethat theannualriskpremiumincreasesbyabout2%andarguethat“changesinriskpremiumofthis magnitudearesufficienttogeneratetheobservedannouncement-daypricereactions”(page388). Indeed,changesofthismagnitudeshouldtriggerverylargenegativereturnsiftheyareindicated bythedividendannouncement.InthissectionIexaminethetimelinessofthechangeinequity risk. Iusethefollowingproceduretoexaminechangesovertimeintheriskpremiumof dividenddecreasefirms.First,IcalculatetheaverageannualizedreturnassociatedwitheachofthethreeFFfactorsduringtheperiodJuly1,1963throughOctober31,2003.16Second,
I estimateequation(6)foreachoftherelativetradingdays–1,250through1,250forthesampleof dividenddecreasefirms.Third,Icalculatetheannualizedriskpremiumassociatedwitheach relativedayusingthefactorloadingsofequation(6)forthatdayandtheaverageannualized factorreturns: AnnualizedRiskPremium=γ2AAR(RMRF)+γ3AAR(SMB)+γ4AAR(HML)
(7) whereAAR(f)istheAverageAnnualizeddailyReturnonfactorfduring1963-2003,andγ
2,γ3and γ4aretheestimatedfactorloadingsfortheparticulartradingday.Finally,Icalculatethemean 16Theaverageannualizedreturnonfactorfiscalculatedasfollows:AAR(f)=exp{252×mean[ln(1+f)]}–1,where252istheaveragenumberoftradingdaysperyearandthemeaniscalculatedoveralldailyobservationsduringtheperiodJuly1,1963throughOctober31,2003. 18 valueoftheannualizedriskpremiumforgroupsof50consecutivetradingdays:(–1,250,–1,201),(–1,200,–1,151),…,(–50,–1),(0,49),…,(1,200,1,249). PanelAofFigure4plotstheestimates.Thex-axisdisplaystherelativetradingdaywhilethey-axispresentsthemeanvalueoftheriskpremiumforeachgroupof50consecutivetradingdays.ConsistentwithGrullonetal.(2002),theaverageriskpremiuminthethreeyearssubsequenttothedividendcut(tradingdays0through750)issubstantiallylargerthaninthepriorthreeyears.Specifically,theaverageriskpremiuminthethreeprioryearsis7.4%paredto9.0%afterthedividendcut(t-statisticforthedifferenceequals4.7).However,asisevidentfromthefigure,theincreaseinriskursgraduallyduringtheyearprecedingtheannouncementofthedividendcut,andlittleifanyadditionalincreasesurafterday–
1.Inparticular,duringtheintervals(–100,–51)and(–50,–1),theriskpremiumgraduallyincreasesfromlessthan8%tomorethan10%.Itremainsatthislevelforthenexttwoyears,andthendeclinesgraduallytoabout7%.Thisevidencesuggeststhatdividenddecreasesfollowratherthansignalincreasesinequityrisk. Toexaminetherobustnessofthisresult,Inextusealternativeapproachesforcalculatingtheriskpremium.First,Ireportthemediansoftherelativedayriskpremiumsinsteadofthemeans.Themedianestimates,plottedinPanelBofFigure4,areverysimilartothemeans(PanelA).Second,insteadofcalculatingmeansormediansforgroupsofconsecutivetradingdays,Ifitasplinefunctionwith49knots(correspondingto50intervals)forthewholesampleof2,500relativedayriskpremiums(days–1,250through1,249).PanelCofFigure4presentsthefittedvalueofthesplinefunction.Itisevidentthatalloftheincreaseintheriskpremiumursduringtheyearprecedingtheannouncementofthedividendcut. TheriskpremiumestimatesinPanelsAthroughCofFigure4arecalculatedbyrunningFFregressionsforeachrelativetradingday.Eachregressionincludesalldividenddecreasefirms 19 withavailablestockreturnsfortheparticularrelativeday,andtheestimatesarebasedontheassumptionthatallfirmshavethesamefactorloadings.Theprimaryadvantagesofthisapproacharethatitallowsthefactorloadingstovaryoverrelativetime,andthefactorloadingsareestimatedusingrelativelylargesamples(upto282dividenddecreasefirms).Yet,dividenddecreasefirmsarenotidenticalandmaythereforehavedifferentfactorloadings.Toaddressthisconcern,Ireruntheanalysisusingtime-seriesFFregressions.Specifically,foreachfirm,Iestimatetime-seriesFFregressionsforgroupsof50consecutivetradingdays((–1,250,–1,201),(–1,200,–1,151),…,(–50,–1),(0,49),…,(1,200,1,249)).Ithencalculatethefirm/intervalspecificriskpremiumusingtheestimatedfactorloadingsandtheaverageannualizeddailyfactorreturnsovertheperiod1963-2003.Finally,foreachinterval,Icalculatethemeanandmedianvaluesoftheriskpremiumacrossallfirms.TheresultingestimatesareplottedinPanelsD(means)andE(medians)ofFigure4.17 Consistentwiththesmallnumberofobservationsperregression,theestimatesinPanelsDandEofFigure4aremorevolatilethanthoseinPanelsAandB.Yet,inbothPanelsCandD,theriskpremiumclearlyincreasespriortothedividendannouncementanditsaveragevalueinthe150tradingdayspriortothedividendchange(thethreepointstotheleftofthey-axis)issimilartothepost-changevalues.Thus,thefindingthatdividendcutannouncementsfollowratherthansignalanincreaseinriskisrobusttotheuseofalternativeapproachesformeasuringriskpremium. 17Tomitigatetheimpactofnon-synchronoustrading,IincludeintheFFregressionstheone-daylagandleadvaluesofeachofthethreefactorsandusetheScholesandWilliams(1977)approachtoadjusttheestimatedfactorloadings. 20
E.RobustnessChecksInextconducttworobustnesschecksfortheearningsanalysis.First,Irerunalltests involvinganalysts’forecastsusingthemostrecentanalysts’EPSandgrowthforecastsinsteadoftherespectiveconsensusforecasts.Iobtainqualitativelysimilarresultstothosereportedabove.Inparticular,whenregressingequation(5)withCORE_EPSasthedependentvariable,Ifindthatthecoefficientofthedividenddecreasevariableis–0.021(t-statisticof–3.6)foryeart=
0,–0.029(t-statisticof–4.4)fort=1,andinsignificantfort=2,3,4,and5.Theseestimatesareslightlylargerinmagnitudethanthoseobtainedwiththeconsensusforecasts,suggestingthattheconsensusforecastsaremoreprecisethanthemostrecentforecasts.Theinference,however,remainsunchanged:Dividendcutannouncementssignallowercurrentandnextyearearnings. Second,toevaluatethemethodology,Iestimatetheregressionsforlargedividendincreases(greaterthanorequalto20percent)insteadofdividenddecreases.18IfindthatthedividendincreasecoefficientwhenexplainingCORE_EPSispositiveandsignificantforeachofthesixyears(t=0,
1,…,5),butitsmagnitudeisrelativelysmall(rangingbetween0.007and0.013witht-statisticsrangingbetween2.1and5.5).Thus,whilebothdividenddecreasesandincreasespredictsubsequentearnings,theinformationindividenddecreasesrelatesprimarilytothenearfuturewhilethatindividendincreasesismorepermanent.Thisdifferencecouldbeduetoountingconservatismwhichrequiresthatlossesberecognizedwhenanticipatedwhileprofitsshouldberecognizedonlywhenearned.Alternatively,theshortdurationofthenegativeearningsimplicationsofdividendcutscouldbeduetorealoptions,suchastheabandonmentoption,whichallowdividenddecreasefirmstoimproveordiscontinueessfuloperations. 18Ifocusonlargedividendincreasesbecause,asshowninTableI,dividenddecreasesaresubstantiallylargerinmagnitudethandividendincreases. 21 VI.ConclusionExtantresearchestablishesthatdividenddecreases:(1)triggernegativestockreturns,
(2) areunrelatedtofutureearningsaftercontrollingforcurrentearnings,and(3)areassociatedwithanincreaseinfirmrisk.Thisevidencehasbeeninterpretedassuggestingthatdividendcutannouncementssignalanincreaseinrisk,whichinturntriggersanegativemarketresponse(e.g.,Grullonetal.(2002)).Ireexaminetheinformationcontentofdividenddecreasesandfindthatrealizedcurrentearningsofdividenddecreasefirmscontainnegativespecialitemsandaresubstantiallysmallerthanexpectedatthetimeofthedividendchange.Theseresultsimplythatcontrollingforcurrentearningsinexplainingfutureearningsbiasesthedividendcoefficientagainstfindinginformationindividenddecreases. Ithusestimateanalternativemodelforexpectedearningswhichextractsinformationfromanalysts’forecasts,priceandothervariables,butexcludescurrentearnings.Usingthismodel,Ifindthatdividendcutannouncementsareassociatedwithnegativeunexpectedearningsinthesubsequentyear.Ifurthershowthatthepresentvalueofunexpectedearningsinthecurrentandnextyearisapproximatelyequaltothedividendcutannouncementreturn.Finally,Idemonstratethattheincreaseintheriskinessofdividendcutfirms,whichhasbeendocumentedbypreviousstudies,ursprimarilyintheyearpriortothedividendannouncement.Ithereforeconcludethatdividenddecreaseannouncementsconveyearningsratherthanriskinformation. 22 REFERENCES Aharony,Joseph,andItzhakSwary,1980,Quarterlydividendandearningsannouncementsandstockholders’returns:Anempiricalanalysis,JournalofFinance35,1-12. Allen,Franklin,andRoniMichaely,2002,Payoutpolicy,Workingpaper,UniversityofPennsylvania. Baker,Malcolm,andRichardS.Ruback,1999,Estimatingindustrymultiples,Workingpaper,HarvardUniversity. Beaver,William,andDaleMorse,1978,Whatdeterminesprice-earningsratios?
FinancialAnalystsJournal34,65-76. Benartzi,Shlomo,RoniMichaely,andRichardThaler,1997,Dochangesindividendssignalthefutureorthepast?
JournalofFinance52,1007-1034. Burgstahler,David,JamesJiambalvo,andTerryShevlin,2002,Dostockpricesfullyreflecttheimplicationsofspecialitemsforfutureearnings?
JournalofountingResearch40,585-612. Chen,Shuping,TerryShevlin,andYenHeeTong,2004,Whatistheinformationcontentofdividendchanges?
Anewinvestigationofanoldpuzzle,Workingpaper,UniversityofWashington. DeAngelo,Harry,LindaDeAngelo,andDouglasJ.Skinner,1992,Dividendsandlosses,JournalofFinance47,1837-1863. Elliott,JohnA.,andDouglasJ.Hanna,1996,Repeatedountingwrite-offsandtheinformationcontentofearnings,JournalofountingResearch34,135-155. Fairfield,PatriciaM.,1994,P/E,P/Bandthepresentvalueoffuturedividends,FinancialAnalystsJournal50,23-31. Fama,EugeneF.,andhR.French,1993,Commonriskfactorsinthereturnsonstocksandbonds,JournalofFinancialEconomics33,3-56. Fama,EugeneF.,andhR.French,2001,Disappearingdividends:Changingfirmcharacteristicsorlowerpropensitytopay,JournalofFinancialEconomics60,3-43. Gordon,MyronJ.,1962,TheInvestment,Financing,andValuationoftheCorporation(Irwin,Homewood,IL) Greenspan,Alan,1997,aryPolicyReporttoCongressPursuanttotheFullEmployment&BalancedGrowthActof1998,SuperintendentofDocuments. 23 Grullon,Gustavo,RoniMichaely,andBhaskaranSwaminathan,2002,Aredividendchangesasignoffirmmaturity?
JournalofBusiness75,387-424. Grullon,Gustavo,ShlomoBenartzi,RoniMichaely,andRichardH.Thaler,2003,Dividendchangesdonotsignalchangesinfutureprofitability,JournalofBusiness(ing). Healy,PaulM.,andKrishnaG.Palepu,1988,Earningsinformationconveyedbydividendinitiationsandomissions,JournalofFinancialEconomics21,149-175. Jacksonth,ScottB.,andMarshallK.Pitman,2001,Auditorsandearningsmanagement,TheCPAJournal71,38-44. Kirschenheiter,Michael,andNahumMelumad,2001,Can“bigbath”andearningssmoothingco-existasequilibriumfinancialreportingstrategies?
JournalofountingResearch40,761-96. Kothari,
S.P.,2001,Capitalmarketsresearchinounting,JournalofountingandEconomics31,105-231. Michaely,Roni,RichardH.Thaler,andKentL.Womack,1995,Pricereactionstodividendinitiationsandomissions:Overreactionordrift?
JournalofFinance50,573-608. Miller,MertonH.,andKevinRock,1985,Dividendpolicyunderasymmetricinformation,JournalofFinance40,1031-1051. Newey,WhitneyK.,andhD.West,1987,Asimplepositivesemi-definite,heteroskedasticityandautocorrelationconsistentcovariancematrix,Econometrica55,703-708. Nissim,Doron,andAmirZiv,2001,Dividendchangesandfutureprofitability,JournalofFinance56,2111-2134. Scholes,Myron,andJosephWilliams,1977,Estimatingbetafromnon-synchronousdata,JournalofFinancialEconomics5,309-327. Skinner,DouglasJ.,2003,Whatdodividendstellusaboutearningsquality?
Workingpaper,UniversityofMichigan. Za
声明：
该资讯来自于互联网网友发布，如有侵犯您的权益请联系我们。