Apreviousdraftofthispaperwascirculatedwiththetitle,“StockSplits:InformationorLiquidity?”TheauthorsthankRayBall,PhilBerger,EugeneFama,AvnerKalay,ChristianLeuz,DougSkinner,theparticipantsattheUniversityofChicagoAccountingWorkshop,andattheAccountingandFinancebrownbags,forhelpfulcommentsandsuggestions.AlonKalay:phone:212‐854‐5315,email:ak3318@columbia.edu,MathiasKronlund:kronlund@illinois.edu

TheMarketReactiontoStockSplitAnnouncements:

EarningsInformationAfterAll

AlonKalayColumbiaSchoolofBusiness

ColumbiaUniversity

MathiasKronlundCollegeofBusiness

UniversityofIllinoisatUrbana‐Champaign

October10,2014

Classificationcode:G14Keywords:Stocksplits,eventstudy,analysts,earningsinformation

Abstract:Were‐examinewhethertheabnormalreturnsaroundstocksplitannouncementscanbeexplainedbyaninformationhypothesis.Ourevidenceestablishesalinkbetweentheabnormalreturnsandfutureearningsgrowth.Analystsreviseearningsforecastsby2.2‐2.5%aroundsplitannouncements,andthisrevisionissignificantlylargerthanthatformatchedfirms.Wefurthershowthattheearningsinformationinasplitlikelyarisesfromthefactthatsplittingfirmsexperiencelessmeanreversionintheirearningsgrowthrelativetomatchedfirms.Consistentwithanearningsinformationhypothesis,theanalystrevisionandtheabnormalreturnsarestrongerforfirmswithmoreopaqueinformationenvironments,andthecross‐sectionalvariationinanalystrevisionsisrelatedtothevariationinabnormalreturns.

1

1 Introduction

Many studies document abnormal returns around stock split announcements. However,

giventhatastocksplit issimplyasuperficialchangetoasecurity’spriceandsharesoutstanding,

the question why we observe these abnormal returns remains an unresolved puzzle. Various

theories have been proposed to explain the abnormal returns around split announcements. The

earliesttheorieshypothesizethatmarketparticipantslearninformationaboutfirms’fundamentals

e.g. dividends or earnings from stock splits. Later, alternative theories argue that it is not

information, but rather increased liquidity that stocks achieve via splits that is the cause of the

abnormalreturns.Amorerecentcateringtheoryarguesthatmanagerssplittheirstocktocaterto

investorswhoassignapremiumtolow‐pricedstocksduringcertaintimes.

Inthispaper,wepresentnewevidenceconsistentwiththeinformationhypothesis.Namely,

that the abnormal returns are caused by market participants inferring positive news about the

firm’sfundamentalsfromthesplitannouncement,specificallyaboutfutureearnings.Weshowthat

onaverage, stock splits result in an immediate increase in themarket’s expectationof the firm’s

future earnings, and that stock splits predict actual future earnings changes in amanner that is

consistentwiththeseexpectations.

Thispaperisnotthefirsttolinkstocksplitstoearnings see,forexample,Lakonishokand

Lev 1987 andAsquithet al. 1989 .Butmore recent literaturehasneverthelessdismissed the

linkbetweentheabnormalreturnsaroundstocksplitsand fundamentals,and instead focuseson

liquidity‐andcatering‐relatedtheoriestoexplainwhymanagerssplittheirstockandwhythestock

marketreactstostocksplitannouncements e.g.Bakeretal., 2009 ;Benartzietal., 2009 ;Linet

al., 2009 .

These alternative theories have received increased attention because the literature has

struggledtointerprettheexistingevidenceonwhetherstocksplitsinfactpredictimprovedfuture

2

fundamentalperformance.Forexample,LakonishokandLev 1987 andAsquithetal. 1989 finda

pattern of strong earnings growth prior to splits, followed by modest growth afterwards. But

Benartzietal. 2009 raisedoubtaboutwhether thisshouldbe interpretedasgoodnews,noting

that“Asquith,Healy,andPalepu 1989 findlargeearningincreasesandreturnspriortoasplit,but

nonethereafter.Dosplittingfirmstrytosignalthattheyhavealreadyreachedtheirpeakandtheir

growthrateshouldrevertbacktoalowerlevel?Thatinterpretationseemsunlikely.”

Prior research have also struggled to link stock splits to firms’ future abnormal earnings

growth,whethercomparedtoafirm’spastgrowth,orcomparedtoindustrypeers e.g.Huangetal.

2006 . However, one issue that plagues the analysis in previous studies is the lack of a clear

counterfactualforthepost‐splitearningsgrowth:Whatisthecorrectbenchmarkwhenweexamine

earningsofsplittingfirms?Furthermore,splitsareoftenannouncedtogetherwithotherfirmnews,

such as earnings or dividends, and many papers do not appropriately control for these events,

whichmakesithardtoattributetheirfindingstothestocksplitannouncement.Theseconceptual

andempiricalproblemshavecontributedtothesearchforalternativetheories.

Whether stock splits contain news about earnings is an important question to resolve

because many studies employ stock splits to study other questions, while making opposing

assumptionsaboutwhetherornotsplitsare informativeabout fundamentals.Forexample,Louis

and Robinson 2005 assume that stock splits represent a “credible signaling device” of

information,andstudywhetherthemarket interpretsaccrualsdifferentlyfor firmsthatalsosplit

their stock. In contrast, Baker et al. 2009 argue that stock splits provide a good laboratory for

studying managers’ behavior because splits “are not associated with any confounding, ‘real’

motivationinvolvingfirmfundamentals”.

In thispaper,weseek toaddress theseproblemsandreconcile the literatureonwhether

stocksplitscanbe interpretedasgoodnewsabout fundamentals.Wehypothesize thatmanagers

3

aremorelikelytosplittheirstockwhentheyareconfidentthatthefirm’spastearningsgrowthis

nottemporaryinnature,andthepost‐splitearningsarelikelytobemorepermanent.1Thus,market

participants should revise their future earnings expectations upwards following the split

announcement compared to their pre‐split expectations, even though the expected growth rate

after the splitmay still be lower compared to an industrybenchmarkor the same firm’s growth

priortothesplit particularlyifthepastgrowthwasveryhigh .

Theclearestwaytotestthishypothesisistoexaminehowinvestors’expectationsaboutthe

firm’s futureearningschangesaroundthesplitannouncement.Whilewecannotdirectlyobserve

every investor’s expectations about a firm’s earnings, we can observe forecasts by analysts at

regular intervals around a split. Therefore, to measure the change in earnings expectations we

calculate the difference between the pre‐split andpost‐split analyst estimates. Thismethod thus

alleviatestheneedtosearchfor“matches”toserveasabenchmarkorcounterfactual.

Using analyst forecasts tomeasure changes in expectations around split announcements

nevertheless raises additional identification challenges which we also address in this paper.

Splittingfirmstendtoperformwellpriortothesplit.Therefore,becauseanalystsmayupdatetheir

forecastswith a lag, it is possible that increases in analyst forecasts following the split could be

attributedtothesplittingfirms’pastperformance.Weaddressthisconcernusingvariousmatched

sampletechniques.Furthermore,splitsareoftenaccompaniedbyother formsofnews.Thus, it is

important to focus on instances where analysts react to the split as opposed to earnings and

divided change announcements. We do so by focusing on analyst revisions that are not

accompanied by earnings announcements, dividend change announcements, or management

guidance. These concerns when examining analysts data around stock split have not been fully

addressedinpriorstudies e.g.McNicholsandDravid 1990 ,ConroyandHarris 1999 .

1ThishypothesisissimilartoAsquithetal. 1989

4

We thus seek to bridge the gap in views about the informativeness of stock split

announcementsbyproviding anew set of resultswhich attempts to address the conceptual and

empirical problems that contributed to the search for alternative theories to the information

hypothesis. Theaimofouranalysis is toshowthatmarketparticipantsreact tostocksplits ina

waythatisconsistentwithlearningaboutthefirm’searningsgrowth.

Theempiricalanalysisproceedsinseveralsteps.First,similartomanypreviouspapers,we

establishthatstocksplitannouncementsinoursampleareassociatedwithabnormalreturns,even

aftercontrollingforconfoundingnewseventsintheformofearningsannouncements,management

guidance,anddividendchangeannouncements.Themeanabnormalreturnforallthefirmsinour

sample over threedays around the split announcement is 1.7% t‐stat 19.50 .Whenwe exclude

observationsthatcoincidewithconfoundingannouncements,themeanabnormalreturnsremains

positiveat1.6% t‐stat12.36 .Second,consistentwithaninformationhypothesis,wefindthatthe

abnormalreturnsarehigherforfirmsforwhichthereislessotherpubliclyavailableinformation,as

measuredbyfeweranalystsfollowingthefirm.

Asathirdstep,toshowthatmarketparticipantsupdatetheirexpectationsaboutthefirm’s

futureearningswestudythechangeinanalystforecastsaroundthestocksplitannouncement.The

pre‐split analyst consensus provides an estimate of the market’s ex‐ante expectations, and the

change between the pre‐split and post‐split forecasts thusmeasures the change in expectations

conditionalonthesplit.Inourfullsample,thechangeintheconsensusforecastnormalizedbythe

pre‐splitstockprice ∆EPS/P isbetween0.13% t‐stat8.33 and0.14% t‐stat9.76 .Theincrease

remains positive and significant, between 0.13% t‐stat 4.22 and 0.15% t‐stat 5.71 , after we

excludeobservationsthatarepotentiallyconfoundedbycoincidingannouncements.Thus,analysts

arenotreactingsolelytoothernewsthattendstoaccompanystocksplitannouncements.Basedon

themedianP/Einoursample 17 ,theseresultstranslatetoapproximatelya2.2%‐2.5%increase

intheearningsforecasts.

5

Theserevisionswedocumentarealsosignificantlylargerthanthoseexperiencedbyfirms

withsimilarpastreturns,aswellasfirmswithasimilarpropensitytosplitthatdidnotsplitinthe

sameyear p‐matchedfirms .Forthesematchedfirms,analystsreviseearningsforecastsbetween

0.03% and 0.05%. These results alleviate the potential concern that the increase in the analyst

forecastsistheresultofdelayedanalystreactiontopastperformance.Furthermore,wefindcross‐

sectionaldifferences in themagnitudeof theanalyst forecast revisions.Firmswith loweranalyst

followingandlowermarketcapitalizationareassociatedwithhigherrevisions,implyingthatasplit

announcement is more informative to analysts for firms with more opaque information

environments. Finally, the cross‐sectional variation in analyst forecast revisions is positively

relatedtothecross‐sectionalvariationinabnormalreturns.

Asafourthstep,weexaminewhetherthefutureearningsperformanceofthesplittingfirms

is consistentwith the adjustment inmarket expectations following the split. Todo so, thepaper

examinestheactualpre‐andpost‐splitearningsgrowthofsplittingfirmscomparedtoamatched

sample matched on size, P/E, and past earnings growth . The results show splitting firms

experiencelowerlevelsofmean‐reversionintheirearningsgrowthafterthesplit,comparedtothe

matchedfirms.Eventhoughsplittingfirmsexperiencereducedearningsgrowthintheyearsafter

thesplitcomparedtotheirowngrowthbeforetothesplit,thepost‐splitearningsgrowthishigher

than the growth for firms with similar past performance. This implies that the prior earnings

growthexperiencedbythesplittingfirmsismorepermanentinnature.Therefore,theincreasein

analyst earnings expectations around splits appear to bewarranted, assuming that the pre‐split

expectedearningspathforthesplittingfirmsissimilartothatofthematchedfirms.

IkenberryandRamnath 2002 alsoanalyzeanalystforecastsaroundstocksplits.However,

theyexamineadifferentquestion:whyfirmsexperienceabnormalreturndriftsinthelong‐run i.e.,

overmanymonths afterstocksplits.Theyproposean“under‐reaction”explanation,andshowthat

analysts tend to provide downward‐biased earnings forecasts for splitting firms, and that this

6

downwardbiascontinueslongafterthesplit.Inotherwords,theirevidencesuggeststhatanalysts

“under‐react” to the split information, which is supportive of a broader stock market under‐

reaction. Incontrast,weshowthat irrespectiveofapotentialcontinuedbias in the forecast level

which we also confirm remains downward‐biased compared to matched firms after the split ,

analystsneverthelessdoreactpositivelytotheinformationcontainedinthestocksplitbyrevising

theirforecastsignificantlyupwardsfollowingtheannouncement.Theaverageanalystrevisionover

amontharoundthestocksplitannouncementismuchlargerthanthetypicalrevisionforthesame

firmsovereachofthe12monthsfollowingtheannouncement.Thus,thestocksplitannouncement

isameaningfulinformationeventassociatedwithsignificantpositiveforecastrevisions.

Our question,why the stockmarket reacts to split announcements, is different from the

relatedquestionwhymanagersdecide tosplit theirstock. It ispossible thatmanagerssplit their

stock for a variety of reasons, but that the abnormal returns are caused bymarket participants

reactingtoonlyasubsetofthesereasons,ortosomeotherinferredinformation.Infact,CEOsoften

citemultiplereasonsforsplitting.Forexample,whendiscussingCompaq’s5‐for‐2splitin1997,the

firm’sChairmanBenjaminM.Rosenstatedthatthesplit"reflectsourconfidenceinCompaq'slong‐

termgrowth… and thelowerpost‐splitsharepricewillmakeiteasierforindividualinvestorsto

purchasethestock,thushelpingbroadenthecompany'sownershipbase.”Butwhileamanagermay

havemultiplereasonsfordoingastocksplit,ourevidenceisconsistentwiththehypothesisthatthe

abnormalreturnsaroundsplitsarearesultofmarketparticipantsinferringpositivenewsaboutthe

firm’s fundamentals from the split announcement. Consistentwith analysts interpreting splits as

goodnewsaboutfundamentals,WilliamConroy,ananalystinHouston,commentedontheCompaq

split that “ a split is a good sign as companies don't split unless they are feeling good about

themselves.” 2Thus,as longasamanager’sdecision tosplit is correlatedwith,orconditionalon,

2HoustonChronicle,July2,1997

7

goodprivateinformation,marketparticipantscantheninferpositivenewsaboutfundamentsfrom

the split announcement—even though conveying such information may not have been the

manager’sintentwhendecidingonthesplit.3

This paper contributes to the literature by using improved identification techniques to

address the conceptual and empirical problems that contributed to the search for alternative

theoriestotheinformationhypothesis.Wepresentanewsetofresultsconsistentwiththeoriginal

informationhypothesisthatrelatestheabnormalreturnsaroundstocksplitannouncementstothe

firm’searningsperformance.Althoughrecentresearchonstocksplitshaslargelydismissedthelink

betweensplitsandearnings,thispaperprovidesempiricalevidencethatsupportstheassumption

madeinstudiesthatemploystocksplitsasinformationevents e.g.,LouisandRobinson 2005 .

Our paper proceeds as follows: Section 2 describes our sample. Section 3 presents our

findings related to the abnormal returns and analyst forecast revisions around stock split

announcements.Section4describesourearningsrelatedresults.Section5concludes.

2 Sampleselectionanddescriptivestatistics

OursampleconsistsofallcommonstocksplitsinCRSP eventcode5523 ontheNYSEwith

asplitfactor≥5:4andadeclarationdatebetween1January1988and31December2007.Westart

oursample in1988since IBEScoverage whichwerequire forouranalyst forecast tests isvery

limitedpriortothisyear.Theresultingsampleconsistsof2097splits.

Table 1 reports descriptive statistics for the splitting firms. The total number of unique

firms in our sample is 1203. The mean number of splits per firm is 1.74 median 1 , and the

maximumnumberofsplitsbyanyfirmoverthisperiodis9.Amajorityofsplits,1184 56% ,inour

3Variousmotivescouldleadtosuchacorrelationbetweenthesplittingdecisionandpositivefundamentals.Forexample,ifmanagersseektousesplitstomaintainadesiredsharepricerange,thentheywouldbemorelikelytosplittheirstockwhentheybelievethestockpricewillnototherwisedeclinetothetargetedpricerange Grinblattetal. 1984 ,Ikenberryetal. 1996 .

8

samplehaveasplitfactorof2:1.816splits 39% haveasplitfactoroflessthan2:1,97splits 5%

haveasplit factorofgreaterthan2:1,andthevastmajorityof theseare3:2and3:1respectively

untabulated . Themeanpre‐split price as of twodays prior to the declarationdate is $59.20,

withaminimumof$5.89andamaximumof$726.30.Themediansplittingfirminoursamplehas

10 analysts following the firm, measured as of the closest consensus estimate to the split

declarationdate.Onaverage, thesplitting firmsarealso largerandmoreheavily traded than the

averagefirmontheNYSEduringthisperiod.

InsertTable1aroundhere .

Inouranalysisofthemarketreactiontostocksplitannouncements,wemeasureabnormal

returnsaroundtheannouncementdatesasthecumulativereturnnetofthevalue‐weightedmarket

returnoverthreetradingdays ‐1to 1days aroundthesplitannouncementdate.Forthesplit

announcement date, we use the “declaration date” from CRSP. In some cases, it is possible that

news about the split leaks prior to the official declaration date, but such leaks should only bias

againstfindinganysignificantabnormalreturnsinthethree‐daywindow.

An important caveat in studies of stock splits is that splits are often announced in

conjunctionwithconfoundingannouncements,whichcanmakeitdifficulttodisentanglewhether

themarketisreactingtothesplititselfortotheothernewsreleasedatthesametime.Therefore,to

measure abnormal returns which are less likely to be contaminated by other information, we

separately analyze stock split announcements that do not coincide with quarterly earnings

announcements,theissuanceofguidance,orannouncementsofdividendchangeswithinthethree‐

day window around the split announcement. We obtain earnings announcement dates from

Compustat.Inoursampleof2097splits,weareabletolink2087splitstofirmsinCompustat.For

the10splits thatwecannot link toCompustat,weconservativelyassumethat therehasbeenan

earnings announcement in the window. We obtain dividend announcement dates and dividend

9

amountsfromCRSPanddefineadividendchangeannouncementasanannouncementofanycash

dividend CRSPdistributioncode12xxor13xx forwhichthepreviousannouncementofthesame

typewasnotof thesameamount.Weobtaindataonthe issuanceofguidancefromtheFirstCall

database.

Table 2 reports the number of splits that coincidewith confounding announcements.We

seethat22%ofthesplitscoincidewithanearningsannouncement,39.6%coincidewithadividend

change, and 4.8% coincide with the issuance of guidance in the three days around the split

announcement. Since some of the announcements overlap, 52.3% of the splits in our sample

coincidewithatleastoneoftheseannouncements.

InsertTable2aroundhere

Toanalyze the change in themarket’s expectationof futureperformance for the splitting

firms, we measure the revision in analyst earnings forecasts following the split declaration. To

calculate the analyst forecast revision, we use the IBES detailed file to compute an outstanding

analyst consensus EPS forecast before and after the split announcement details on the

computationoftheanalystconsensusisprovidedinAppendixA1 .WeanalyzechangesintheEPS

forecast for thenext full fiscalyearafter thesplitannouncement toensure that the forecastsare

madeatleastoneyearbeforetheannouncementofthatfiscalyear’searnings.Forexample,ifasplit

isannouncedinMarch2005forafirmwithaDecemberfiscalyear‐end,wecompareEPSestimates

adjustedforthesplitfactoriftheex‐datefallsbetweenanyoftheestimates madebeforeandafter

the split announcement for the fiscal year‐endDecember31, 2006.Weuse this forecast horizon

because longer horizons are more likely to be concerned with fundamental long‐term firm

performance, as opposed to temporary changes in earnings expectations e.g., resulting from

accruals, seasonality, and one‐off charges or revenues . We also focus on annual forecasts as

opposedtoquarterlyforecastsforthesamereason.Toconstructaconsensusestimate,werequire

10

at least three individualanalystEPSestimates in IBES in the30daysbefore,aswellas in the30

days after the split announcement. We can thus construct both pre‐ and post‐split consensus

estimatesfor727ofthesplitsinoursample.

As in theanalysisof abnormal returns,wecontrol for confoundingannouncementsmade

between the pre‐ and post‐split analyst estimates. To do this, we exclude individual analyst

estimatesmadebeforeandafterthedeclarationdateforwhichaconfoundingannouncementtook

place between the estimate and the split announcement.We then retain all consensus estimates

thatconsistofatleastthree“uncontaminated”forecastsbothbeforeandafterthedeclarationdate.

The final sample consists of 198 splitswith pre‐ andpost‐split consensusEPS forecasts that are

uncontaminatedbyearnings,guidance,ordividendchangeannouncements Table2 .

3 Theearningsinformationhypothesis,abnormalreturns,

andanalystforecastrevisions

The information hypothesis attributes the abnormal returns around stock split

announcements to information themarket learnsabout firm fundamentals. Fama,Fisher, Jensen,

andRoll 1969 introducedthisideaintheireventstudyonstocksplits,showingthatsplitspredict

increases in future dividends. One reason why market participants may rationally update their

expectations of future performance following a split announcement is thatwhilemanagersmay

split their stock formultiple reasons, these reasons are likely to be coupledwith themanager’s

beliefthatthefirmisperformingwell.Forexample,ifmanagersaimtokeeptheirstockpriceina

certainrange BakerandGallagher 1980 ,BakerandPowell 1992 ,andmanagerschoosetosplit

theirfirm’sstockwhentheythinkthatitisunlikelythatthestockpricewilldeclinetothedesired

range without a split. As a result, the market can infer from the split that the future earnings

performanceshouldimproverelativetothepre‐splitexpectations.Ikenberryetal. 1996 referto

11

the theory wheremanagers condition their decision to split the firm’s stock on their optimistic

viewsaboutthefirm’sfutureperformanceasthe“self‐selectionhypothesis”.

Areasonwhysomehaveconsideredstocksplitstobedubiousinformationeventsrelatesto

thedifficulty in identifyingasignificantcostassociatedwithastocksplit,which isrequired fora

splittobeacrediblesignal e.g.,Benartzietal. 2009 4.However,aninformationhypothesisdoes

not require thatmanagers split their stockwith a signalingmotive inmind, or in an attempt to

distinguishtheirfirm e.g.,asinthemodelbyBrennanandCopeland 1988 .Especiallysinceitis

unclearwhystocksplitswouldbeamanager’spreferredwaytosignalthefirm’sunderlyingvalue.

As long as the manager’s underlying motives for doing a stock split is correlated with, or

conditionalon,somepositiveprivateinformationaboutthefuture e.g.,ifmanagerswanttocater

toinvestorswhopreferlowprices,andemploystocksplitsiftheydonotexpectthestockpriceto

decreasesufficientlyonitsown ,asplitwillbeinterpretedaspositivenewsbythemarket.Thiscan

occureventhoughconveyingthisprivateinformationmaynotnecessarilybethemanager’sintent.

Therefore, it isultimatelyanempiricalquestionwhetherthemarket interpretsstocksplit

announcements as positive news about fundamentals. In this paper, we specifically examine if

market participants update their beliefs about a firm’s future earnings growth around split

announcements, and if theabnormal returnsobservedaround stock split announcements canbe

explainedbysuchchangesinexpectationsaboutthefirm’sfundamentals.

3.1 Abnormalreturnsandtheearningsinformationhypothesis

To test the earnings information hypothesis, we begin by re‐examining the abnormal

returnsaroundstocksplitannouncements.Acrossallobservations,themeanabnormalreturnover 4Somepossiblecoststhathavebeensuggested,inadditiontosomenon‐trivialadministrativecostsassociatedwithasplit,resultfromafirm’sstockpricefallingbelowcertainthresholds.Atoolowpricemayfailtomeettheexchange’sminimumpricerequirement,orcouldleadastocktobeexcludedfromtheholdingsofsomeinstitutionalinvestorsthathaveminimum‐pricerules.Furthermore,amanagermaysufferreputationalcostsifs/hesplitsthefirm’sstockandthenfailstodeliverthe‘positive’performancethatthemarketinfersfromthesplitannouncement.

12

three days around the split announcements is 1.7% t‐stat 19.50 Table 3, Panel A . Themean

abnormal return is slightly lower, 1.6%, but still positive and significant t‐stat 12.39 whenwe

exclude observations that coincide with earnings announcements, management guidance, or

dividendchangeannouncements PanelB .

InsertTable3aroundhere

Ifthemarket’sreactiontothesplitannouncementisduetonewinformation,thereaction

shouldbestrongerforfirmswherethereisotherwiselesspubliclyavailableinformation.Thisidea

issimilartothatofGrinblatt,Masulis,andTitman 1984 ,whofindthatnon‐dividendpayershave

higherabnormalreturnsaroundsplits.InTable3,wefindthattheabnormalreturnsaroundsplit

announcementsarehigher forsmaller firms Ikenberryetal. 1996 also findanegativerelation

between split announcement returns and the firm’s size decile . Similarly, abnormal returns are

alsohigherforfirmsthathavefeweranalystsfollowingthem.Whenweincludebothvariables,as

wellascontrolvariablesfortheoverallliquidityofthestock tradingvolumeandbid‐askspreads ,

weseethatthevariationinthenumberofanalystsfollowingthefirmexplainsmostofthecross‐

sectional variation in announcement returns. In particular, in Panel B, where we exclude

observationsthatcoincidewithotherannouncements,thenumberofanalystsfollowingthefirmis

theonlyvariablewhichremainssignificant.Furthermore,inthemultipleregressionspecifications

5 ‐ 7 theliquidityproxies dollarvolumeandbid‐askspread areeitherinsignificant,orsuggest

thatfirmswithhigherliquiditybeforethesplitannouncementareassociatedwithhigherabnormal

returns.5

This new result is important because it implies that the relation between the abnormal

returnsandthenumberofanalystsfollowingthefirmdoesnotresultsolelyfromaliquidityeffect, 5Oneexceptionisweaklypositivecoefficientonthebid‐askspread,suggestingthatfirmswithhigherpre‐splitbid‐askspreadsareassociatedwithhighersplitannouncementreturns.Becausebid‐askspreadsaremorelikelytobemissinginoursample,goingforwardweonlyusetradingvolumeasaproxyforliquidity.Nevertheless,wefindsimilarresultswhenwealsoincludethebid‐askspreadvariable.

13

andcanbeattributedtotheinformationenvironmentofthefirm.Whenmovingfromthe25thtothe

75thpercentileofanalyst following inoursample, thenumberofanalysts increases from5 to17

see the descriptive statistics in Table 1 . Thus, the coefficient on the log number of analysts in

Table3,between‐0.007and‐0.008,correspondstoapredicteddifferenceinabnormalreturnsof

around90basispointsacrosstheinnerquartilerangeofanalystfollowing.

3.2 Analystforecastrevisionsandtheearningsinformationhypothesis

Toexamineifthemarketupdatesitsexpectationaboutthefirm’searningsgrowthfollowing

the split announcement,we analyzewhether analysts revise theirEPS forecasts around the split

announcement.Wereportresultsusingboththemeanandthemediananalystforecastasthepre‐

andpost‐split consensusaroundeachsplitannouncement. InPanelAofTable4,wesee that the

analyst consensus estimates increase following the split announcement. Across all split

observations, the average change in themean analyst forecast normalized by the pre‐split stock

price ∆EPS/P is0.14%,andhighlysignificant t‐stat9.76 .Whatis importantforouranalysisis

that the change remains positive and significant, 0.15% t‐stat 5.71 , even after we exclude

observations thatcoincidewithearningsannouncements,managementguidanceannouncements,

ordividendchangeannouncements; indicatingthatanalystsarenotreactingsolelytotheseother

typesofcoincidingnews.Ourresultsusingconsensusestimatesbasedonthemedianforecast so

thataconsensusisnotdrivenbyoutlieranalysts areverysimilar.Weobserveasignificantaverage

increase of 0.13% t‐stats of 8.33 and 4.22 using the median forecasts for the full and non‐

confoundedsamples.

InsertTable4aroundhere .

Toprovide someeconomic intuition for themagnitudeof theearnings forecast revisions,

wecanmultiplytheestimateof∆EPS/PbythemedianP/Evalueinoursample 17 totranslate

these units into percent changes in forecasted earnings at least approximately, as

14

∆E

P* average

P

E≈∆E

E. By this logic, the estimate for ∆EPS/P of 0.14% corresponds to

approximately a 2.4% increase in the average consensus forecast in the full sample, while the

estimate after controlling for alternative announcements corresponds to a 2.2% increase in the

average earnings forecast. Themagnitude of the percent revision in forecasted earnings is very

similartotheabnormalreturnsaroundsplitannouncements 1.5‐1.6% ,sothechangeinearnings

expectationscouldeasily justify theabnormalreturns inasimplediscount‐discountmodelof the

stockpriceifmarketparticipantsassumethatthehigherlevelofexpectedearningsispersistent.

Onepotentialalternativeexplanation for the increase inanalystearnings forecasts is that

analysts are revising their earnings estimateswith some lag following thehighpast returns that

oftenprecedesplits,orbasedonotherobservablesthatcharacterizesplittingfirms,asopposedto

thesplitannouncementperse.Totestthisalternativeexplanation,wecomparetheanalystforecast

revisionofthesplittingfirmstothatoftwoseparatesetsofmatchedfirms.First,splittingfirmson

average experience very high returns prior to the split. Thus, to control for the possibility that

analystsareadjusting theirestimatesbasedonpastreturns,we identifymatched firmsbasedon

pastone‐yearreturnspriortothesplitannouncement.Wealsorequirepotentialmatchestobein

the same Fama French 49 industry and size tercile as the splitting firm6. As a secondmatching

method, we minimize the distance of an estimated propensity to split between the potential

matches that did not split in a given year and the splitting firm similar to the methodology

discussed in Armstrong et al. 2010 . The propensity to split is estimated through annual logit

regressionsofasplitdummyonstockpricesandpricessquared,returnsandlaggedreturns and

theirsquares ,thestandarddeviationofreturns,marketcapitalization,dollarvolumetraded,and

6WeusetheFama‐Frenchsizebreakpoints.

15

market‐to‐book.Werunthesepredictiveregressionsusingtheentiresample,butexceptBerkshire

Hathaway PERMNO 17778,83443 7.

Weanalyze thechange inanalyst consensus fora splitting firmand itsmatcharound the

samedate 30daysbeforeand30daysafterthesplitannouncementdate .Forboththematched

firmsandthesplittingfirms,wecomputetheanalystforecastrevisiononlyforfirmsforwhichat

least three analyst estimateswere updated or reiterated both before and after the split i.e. we

exclude stale estimates . In total, we find 597 and 583 matched consensus estimates when we

matchonreturnsandthepropensitytosplitrespectively.

Whiletheaveragechangeinthemean median forecastispositiveforthematchedsetsof

firms, it is much smaller than for the splitting firms Table 4, Panel A . The average change in

forecastedEPS/Priceusingthemean median forecastis0.04% 0.03% forthematchedsample

basedonpastreturns,and0.04% 0.05% whenwematchonthepropensitytosplit;therevisionis

onlysignificantforthep‐matchedfirms.Thedifferencebetweentheaverageconsensusrevisionfor

the splitting firms and that for both of thematched sets of firms is statistically significant. This

result holds regardless of the method used for matching, or whether we construct consensus

estimatesbasedonmeanormediananalystforecasts.

PanelBofTable4reportsdescriptivestatisticsforthesplittingfirmsandthematchedsets

of firms. Themean andmedian size of the splitting firms is also similar to that of both sets of

matchedfirms,asaretheprice‐to‐earningsandmarket‐to‐bookratios onesetofthematcheshasa

slightlyloweraveragemarket‐to‐bookwhereastheothermatchedsetoffirmshasaslightlyhigher

market‐to‐book . The prior one‐year returns are very high both for the splitting firms and the

matches.Thepropensity‐matchedfirmshavemoresimilarpastreturnscomparedtothefirmsthat

7TheBerkshireHathawayshareclassesareextremeoutliersinthesepredictiveregressions,astheirpriceswereveryhighbutthecompanyneverthelessdidnotsplittheirsharesduringoursampleperiod.ThefirmneverthelessannouncedasplitforitsB‐classsharesinJanuary2010.

16

areexplicitlymatchedonpastreturns,duetotherequirementthatthereturns‐basedmatchesalso

beinthesameFama‐Frenchindustryandsizetercile whichwedonotrequireofthepropensity‐

matched firms . Based on these results, although some of these differences are statistically

significant,thereappeartobenoimportantdifferencesacrossobservablecharacteristicsbetween

thesplittingandmatchedfirmsthatcoulddrivetheanalystrevisionresult.Thus,thesignificantly

smaller revision experienced by the matched firms alleviates the potential concern that the

observedincreaseintheanalystforecastsforthesplittingfirmsistheresultofadelayedreactionto

pastperformance,orduetosomeotherspecial observable characteristicofthesplittingfirms.

IkenberryandRamnath 2002 alsoanalyzeanalystforecastsaroundstocksplits.However,

theyexamineadifferentquestion:whyfirmsexperienceabnormalreturndriftsinthelong‐run i.e.,

overmanymonths afterstocksplits.Theyproposean“under‐reaction”explanation,andshowthat

analysts tend to provide downward‐biased earnings forecasts for splitting firms, and that this

downwardbiascontinueslongafterthesplit.Inotherwords,theirevidencesuggeststhatanalysts

“under‐react” to the split information, which is supportive of a broader stock market under‐

reaction. Incontrast,weshowthat irrespectiveofapotentialcontinuedbias in the forecast level

which we also confirm remains downward‐biased compared to matched firms , analysts

nevertheless do react positively to the information contained in the stock split by revising their

forecastsignificantlyupwardsfollowingtheannouncement,andthisrevisionislargerthanthatfor

firmswithsimilarpastperformance.

Furthermore,whileanalystsalsocontinuetorevisetheirforecastsovertimeafterthesplit,

therevisionaroundthesplitisparticularlylargecomparedtofuturerevisions.InTable5,wefind

thattheanalystrevisionoverthemontharoundthesplitannouncementisalmosttwiceaslargeas

mostoftherevisionsthatoccurovereachofthe12monthsfollowingthesplitannouncement.The

resultsaresimilarbothwhenweexaminetheentiresampleandwhenweexcludesplits thatare

17

announcedwith confounding events.8 These results indicate thatwhile analystmaybe biased in

their level of expectations for splitting firms, the split announcement is a significant information

eventthatanalystsreactto.

InsertTable5aroundhere

Next, we analyze whether there is a relation between the analyst forecast revisions and

splittingfirmcharacteristics Table6 .Theincreaseinanalystearningsforecastsishigherforfirms

withmore opaque information environments as measured by fewer analysts and lowermarket

capitalization. We find a significant negative coefficient in simple regressions of the earnings

revision on the number of analysts andmarket capitalization specifications 1 , 2 .Whenwe

includeboththenumberofanalystsandmarketcapitalizationinamultipleregression,aswellasa

controlforliquidity dollarvolume andyearfixedeffects specifications 3 ,onlythecoefficient

onmarketcapitalizationremainssignificantlynegative.Thisresultissomewhatstrongerwhenwe

useallobservations,butremainssignificantatthe10%levelwhenweexcludeallobservationsthat

coincidewithconfoundingannouncements.TheseresultsareconsistentwiththeresultsinTable3,

wherewe find that theabnormal returnsaroundstock split announcementsarehigher for firms

withfeweranalystsandlowermarketcapitalization.

Finally, we examine the relation between the abnormal returns around stock split

announcementsand theanalyst forecasts revisions Table6,PanelB . Theprior results suggest

thatsplitannouncementsbymoreopaquefirmsleadtostrongerpositivereactionsbystockmarket

participantsandanalysts.Theseresultssuggestthatthereisarelationbetweenthecross‐sectional

variationinannouncementreturns,andthecross‐sectionalvariationinanalystforecastrevisions.

8Therevisionovermonth1 therevisionoverthemonththatincludesthesplitannouncement isslightlydifferentthantherevisionpresentedinTable4,becauseinconstructingthetimeseriesweneedtouseregularmonthlysnapshotsoftheconsensusforecasts.Therefore,weuseforecastsfromtheIBESsummaryfileinTable5.Incontrast,inTable4wedoabefore‐afteranalysisimmediatelyaroundthespecificsplitdateusingindividualanalystestimatesfromtheIBESdetailedfile.

18

We find that across our entire sample, on average, larger forecast revisions following the split

announcement are significantly associated with higher abnormal returns around the split

specifications 1 and 2 .Thecoefficientsimplythataninter‐quartilechangeinthemagnitudeof

the revision in the full sample is associated with approximately a 50 basis point increase in

abnormalreturns.

InsertTable6aroundhere

As the information content of the split announcement should be larger formore opaque

firms, we interact the forecast revision with firm size to examine whether the positive relation

between forecast revisions and abnormal returns is stronger amongmoreopaque firms.Weuse

size as our proxy for the information environment of the firm, because the measure of the

consensus forecast revision requires at least three analysts forecasts in a relatively short period

bothbeforeandafterthesplitannouncement,whichmeansthatwehavemuchlessvariationinthe

numberofanalystscomparedtoouranalysisinTable3.Acrosstheentiresample,theinteraction

termhasalimitedeffectontherelationbetweentheforecastrevisionsandtheabnormalreturns.

Thecoefficientfortheinteractiontermisclosetozeroandstatisticallyinsignificant.Thecoefficient

fortheforecastrevisionremainspositivebutthet‐statisticdropsto1.54inthisspecification.When

wefocusonthesamplethathasnoconfoundingannouncements,wefindapositiverelationthatis

more pronounced for more opaque firms. We find a positive and significant coefficient for the

forecastrevisionsandanegativeandsignificantcoefficientfortheinteractionterm.Takentogether,

theseresultssuggestthatthereiscross‐sectionalvariationintheinformationcontentofstocksplit

announcements. Firms that experience larger forecast revisions also experience higher

announcementreturns.Forthesamplewherethestocksplitannouncementisnotaccompaniedby

confounding announcements, this relation comes primarily from smaller more opaque firms for

whichthesplitcanbemoreinformative.Theseresultsarealsoconsistentwiththeearningsbased

informationhypothesis.

19

Insum,wefindthatanalystssignificantlyreviseearningsforecastsforthenextfiscalyear

aroundstocksplitannouncements,whilethereisamuchsmallerrevisionformatchedfirms.The

magnitude of the analyst revisions 2‐2.5% of earnings around split announcements is

economically plausible given the average abnormal returns associatedwith split announcements

1.6%‐1.7% .Furthermore,theseanalystrevisionsarenegativelycorrelatedwithourproxiesfor

theavailabilityofotherpublicinformationaboutthefirm e.g.,thenumberofanalystsfollowingthe

firm and the firm’s market capitalization . Finally, the cross‐sectional variation in the analyst

forecast revisions is positively correlated with the cross‐sectional variation in announcement

returns.

4 Earningsperformance

To relate the analyst revisions following stock split announcements to Asquith et al.

1989 ’shypothesisthatthestrongearningsgrowthexperiencedbysplittingfirmspriortothesplit

announcement ismorepermanent innature,weexamine the futureearningsperformanceof the

splitting firms. If theearningshypothesis iscorrect,weshouldobservethatwhile firmsthatsplit

their stock experience lower earnings growth in the years following the split, relative to the

acceleratedgrowth theyexperienceprior to the split, theearningsgrowth is still larger than the

unconditional expectation. This would cause analysts to infer that the prior earnings growth

experiencedbythesplittingfirmislesslikelytobetransitoryinnature i.e.asmallerportionofthe

prior earnings is likely to reverse , thus resulting in an upwards revision in the analysts’

expectationsoffutureearnings.

Totestthehypothesisthatsplittingfirmshavelessmeanreversioninearningsgrowth,we

compare theearningsgrowthexperiencedby firms that split their stock inoursample to thatof

matched firms.We compute earnings growth as the change in annual earnings scaled by lagged

totalassets.FollowingBarberandLyon 1996 ,wematchonsizeandpastperformance.Wefirst

20

requirepotentialmatchestobeinthesamesizetercileandP/Equintileasthesplittingfirmduring

theyearof the split sizeandP/Ebreakpointsare fromKennethFrench’swebsite .Outof these

potentialmatches,wepickthefirmthathastheclosestpercentageearningsgrowthtothesplitting

firmduringtheyearofthesplitannouncement betweenfiscalyearends‐1and0 .Asaseparate

setofmatches,wepickthefirmthathasthe“mostsimilar”growthoveruptofouryears,endingin

the fiscal year of the split announcement between fiscal year ends ‐4 and 0 . We do this by

minimizingthesumofsquareddifferencesinthepercentageearningsgrowthbetweenthesplitting

andmatchedfirmsovertheseyears.Theadvantageofminimizingthesumofsquareddifferences

overseveralyearsisthatthematchedfirmsaremorelikelytobeonasimilarearningspathwhich

extendsoverseveralyears into thepast,whereas the firmsmatchedonlyontheyearof thesplit

will naturally have more similar earnings growth for that specific year. Further details on our

matchingprocedurearedescribedinAppendixA2.9

Our resultsonearningsgrowtharepresented inTable7.Becauseour sampleof splitting

andmatchedfirmsislikelytobeaffectedbyextremeperformers outliers ,wefocusourdiscussion

onthemedianearningsgrowthandrelatedWilcoxontest‐statistics.

InsertTable7aroundhere

InPanelA,wereportresultsusingoperatingincomebeforedepreciationasthemeasureof

earnings. When we match based on four years of prior earnings growth, the matched firms

experience earnings growth that is similar and statistically indistinguishable from that of the

splitting firms in year three before to the split announcement. The splitting firms begin to

outperformthematchedfirmstwofiscalyearspriortothesplitandcontinuetodosofortwofiscal

years following the split. While both the splitting firms and the matched firms experience 9Inuntabulatedresultswealsomatchonindustry FamaFrench49industries ,inadditiontoPE,size,andpastearningsgrowth.Ourresultsrelatedtothesplittingfirm’sfutureearningsgrowthissimilar,however,inthiscasethesplittingfirmsoutperformthematchedfirmsduringpastyearsaswellasinthefuture.Hencewepresentresultsforthematchedsampledescribedabove.

21

significantreversionintheirmedianearningsgrowth,theearningsgrowthofthesplittingfirmsin

oursamplerevertsrelativelylessandreturnstonormallevels asdefinedbythematchedfirms in

the thirdyear following thesplit.Whenwematchononeyearearningsgrowth, thesplittingand

matched firms experience similar earnings growth during the year of the split, but the splitting

firmsstilloutperformthematchedfirmsinthetwoyearsfollowingthesplitannouncement.Inthe

thirdyearfollowingthesplitannouncement,thesplittingfirmsmaintainthesameearningsgrowth

rateasthatofthematchedfirms.

Forbothsetsofmatches,thesplittingfirmscontinuetooutperformthematchedfirmsfor

twofullyearsfollowingthesplitandmaintaintheirlevelofincreasedperformancegoingforward

inyearthreeafterthesplit,whentheygrowataratesimilartothatexperiencedbythematched

firms.We findsimilarresultswhenweuse incomebeforeextraordinary itemsas themeasureof

earnings PanelB ,exceptthat inthiscasethesplitting firmsrevert tonormal levels inyeartwo

followingthesplit.Finally,PanelCofTable7showsthatthesplittingandmatchedfirmsarevery

similar across several dimensions that could be associated with future earnings growth: size,

market‐to‐book,andprice‐to‐earningsratios.Themostsignificantdifferencesarethatbothsetsof

matched firms have slightly highermedian P/E,which if anything should predict slightly higher

futuregrowthforthematchedfirmsratherthanthelowergrowththatweobserve.

Theseresultsshowthattheearningsgrowthofthesplittingfirmsintheyearspriortothe

split is less transitory in nature than that of firms with similar past earnings growth and

performance.Thisfindingisconsistentwithmarketparticipantscorrectlyexpectingsplittingfirms

tohavehigherfutureearningsgrowthrelativetothemarket’sunconditionalestimatepriortothe

split announcement. These earnings‐based results support our conclusions that it is information

aboutthefirm’searningsgrowththatmarketparticipantsandanalystsreacttofollowingthestock

splitannouncement.

22

5 Conclusion

EversinceFama,Fisher,Jensen,andRoll’sseminalpaper 1969 ,financialeconomistshave

soughttounderstandwhymarketsreacttostockssplitannouncements,sinceastocksplitappears

tobemerelyacosmetictransactionthatincreasesthenumberofsharesoutstandingandreduces

thesharepricebythesplit factor.Takentogether,ourevidenceshowsthattheabnormalreturns

aroundsplitannouncementsareconsistentwithanearningsinformation‐basedexplanation.

Wefindthatanalystsincreasetheirearningsestimatesaroundstocksplitannouncements,

and that the revision is greater for firms with more opaque information environments.

Furthermore,theearningsforecastrevisionsforsplittingfirmsissignificantlyhigherthanthatfor

matched firms, indicating that the observed increase in earnings estimates does not result from

analystssluggishlyrevisingtheirforecastsinresponsetothesplittingfirms’pastperformance.The

results also show that the cross‐sectional variation in the analyst forecast revisions is positively

correlatedwiththecross‐sectionalvariationinannouncementreturns.

Finally,wefindthatthefutureearningsgrowthofthesplittingfirmsishigherthanthatof

matchedfirmswithsimilarpastearningsgrowth,foruptotwoyearsfollowingthesplit.Whileboth

thesplittingfirmsandthematchedfirmsexperiencelowerearningsgrowthinfutureperiodsafter

thesplitcomparedtotheirownpastearningsgrowth, the futureearningsgrowthof thesplitting

firms isneverthelesshigher than thatof thematched firms.This result implies that theearnings

growthexperiencedbythesplittingfirmsbeforethesplitislesstransitoryinnaturethanthepre‐

splitexpectations asproxiedbytheperformanceofex‐antecomparable firms .Thisresulthelps

explainwhyanalysts revise theirexpectationsof futureearnings followinga split announcement

andincreasetheirearningsestimates.Thispositivechangeinexpectationsislikelytobeaprimary

reasonwhythemarketviewsastocksplitannouncementasfavorablenews.

23

Ourevidencesupportsthehypothesisthatwhilemanagersoftenstatevariousmotivations

for splitting their stock, the market’s reaction to stock split announcements is likely driven by

information related to the firm’s earnings,which themarket infers from the split announcement

and views as favorable news. An earnings information hypothesis therefore warrants renewed

attentionasanexplanationforthemarket’sreactiontostocksplitannouncements.

24

Appendix

A.1 ComputingtheoutstandingEPSforecastconsensusfromtheI/B/E/S

detailfile

To compute the analyst consensus before after the split declaration date, we take the

mean of all the outstanding estimates in the IBES detail file adjusted for stopped and excluded

estimates thatweremadewithinthe30daysbefore after thesplitannouncement.Theforecast

period for earnings that we use for our analysis is the next full fiscal year after the split

announcement.Asanalternativemeasureof theconsensus,we take themediananalystestimate

beforeandafterthestocksplitannouncement.WeusetheunadjustedIBESdetaildatatoavoidthe

rounding error problem in I/B/E/S seeDiether et al. 2002;Barber andKang, 2002 and adjust

manually for splits using the CRSP share adjustments factor CFACSHR to make all estimates

comparable. We also exclude outliers that we suspect are mistakes, by excluding all analyst

estimates from the consensus that are either more than 1.5 times the mean estimate of all

estimates ,or less thanhalfof themeanestimate of all estimates .Thisapproach results in the

exclusionof0.5% 35observations of theanalystestimatesmadebeforetheannouncementand

0.7% 50observations of theanalystestimatesmadeafter theannouncement.Finally,weretain

thepre‐andpost‐split consensusestimatesonly if thereare at least threeanalyst forecastsboth

beforeandafterthesplitannouncementthatformthisconsensus.Theaverageageoftheindividual

forecastsincludedintheconsensusis12.2 11.6 daysbefore after thesplitannouncement.

A.2 Computingearningsgrowthforfirmsthatsplittheirstockandfora

matchedsampleoffirms

To compute the earnings growth for firms that split their stock we collect all available

annualearningsdataonCompustat for theperiodthatextends fromthree fiscalyearsbefore the

splitdeclarationdatetofourfiscalyearsfollowingit includingtheyearofthesplitdefinedasyear

25

zero . We collect data on operating income before depreciation data item 13 , income before

extraordinaryitems dataitem18 andtotalassets dataitem6 .Ourmeasureofearningsgrowth

is the changes in annual earnings scaled by lagged assets data item6 .We exclude all splitting

firmsfromtheanalysisthateitherhavefiscalyearchangesormissingconsecutiveannualearnings

observationsovertheperiod.Wealsoexcludeoneextremeoutlier PECorp,Permno86806,that

announcedasplitonJan20,2000 whichexperiencedanearningschangeof ‐760% ofassets or‐

7.6 ,intheyearbeforethesplit.

Tocreateamatchedsetof firmswe firstmatchonmarketcapitalization tercilesandP/E

quintiles, in the fiscal year of the split announcement date based on breakpoints for market

capitalizationandP/EfromKennethFrench’swebsite10 .Potentialmatchesarefromthepoolofall

annualearningsdataavailableonCompustat forregularshares sharecode10,11 listedon the

NYSEat the timeof theobservationbetween1982and2007,excludingsubsidiarieswithastock

ownershipcodeof1or2andother firms thatsplitduring thesameyear.Weuse thesame logic

describedaboveandeliminatefirmsfromconsiderationwhenthereareeitherfiscalyearchanges

or missing consecutive annual earnings observations over the potential period. As the matched

firm,wechoosethefirmthathastheclosestearningsgrowthduringtheyearofthesplit between

fiscalyearends‐1andyear0 orthefirmthathasthe“mostsimilar”growthoveruptofouryears

endinginthefiscalyearofthedeclarationdate betweenfiscalyearends‐4and0 ,byminimizing

thesumofsquareddifferencesinearningsgrowthbetweenthesplittingandmatchedfirmsoverall

availableobservations uptofour .Finally,wecomputethefutureearningsgrowthforthematched

firms.

10http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html

26

ReferencesAsquith,P.,P.Healy,andK.Palepu,1989,Earningsandstocksplits,TheAccountingReview44,387‐

403

Baker,H.,andP.L.Gallagher,1980,Management’sviewofstocksplits,FinancialManagement,9,73‐77.

Baker,H.,andG.E.Powell,1992,Whycompaniesissuestocksplits,FinancialManagement,21,11.

Baker,M.,R.Greenwood,and J.Wurgler,2009,Catering throughnominal shareprices, JournalofFinance,64,2559‐2590

Barber, B.M., J. D. Lyon, 1996,Detecting abnormal operating performance: The empirical powerandspecificationofteststatistics,JournalofFinancialEconomics41,359‐399

Barber,W.R.,andS.H.Kang,2002,Theimpactofsplitadjustingandroundingonanalysts’forecasterrorcalculations,AccountingHorizons16,277‐289.

Benartzi, S., R. Michaely, R. Thaler, W. Weld, 2009, The nominal share price puzzle. Journal ofEconomicPerspectives23 2 ,121‐42.

Brennan, M. J., T. E. Copeland, 1988, Stock splits, stock prices, and transaction costs. Journal ofFinancialEconomics22 1 ,83‐101.

Conroy,R.M.,andR.S.Harris,1999,TheRoleofSharePrice,FinancialManagement28,28‐40.

Fama,E.,L.Fisher,M.JensenandR.Roll,1969,Theadjustmentofstockpricestonewinformation,InternationalEconomicReview,10,1‐21

Fama,E.,K.R.French,1997,Industrycostsofequity,JournalofFinancialEconomics43,153‐193

Grinblatt, M., R. Masulis, and S. Titman, 1984, The valuation effects of stock splits and stockdividends,JournalofFinancialEconomics,13,461‐490

Huang, G., K. Liano, and M. Pan, 2006, Do stock splits signal future profitability? Review ofQuantitativeFinanceandAccounting,26,347‐367

Ikenberry,D.,G.Rankine,andE.Stice,1996,Whatdostocksplitreallysignal?JournalofFinancialandQuantitativeanalysis,31,357‐375

Ikenberry,D.,andS.Ramnath,2002,Underreactiontoself‐selectednewsevents:Thecaseofstocksplits.ReviewofFinancialStudies,15,489‐526

Lakonishok, J.,andB.Lev,1987,StockSplitsandstockdividends:Why,whoandwhen,JournalofFinance,42,913‐932

Lin, J.,A.K.Singh,andW.Yu,2009,Stocksplits, tradingcontinuityand thecostofequitycapital,JournalofFinancialEconomics,93474‐489

27

Louis,H.,andD.Robinson,2005,Domanagerscrediblyuseaccrualstosignalprivateinformation?Evidence from the pricing of discretionary accruals around stock splits, Journal ofAccountingandEconomics,39,361‐380

McNichols,M.,andA.Dravid,1990,StockDividends,StockSplits,andSignaling,JournalofFinancialEconomics,45 3 ,857‐879