Accounting Research Center, Booth School of Business, University of Chicago
The Information Content of Annual Earnings Announcements Author(s): William H. Beaver Reviewed work(s): Source: Journal of Accounting Research, Vol. 6, Empirical Research in Accounting: Selected Studies 1968 (1968), pp. 67-92 Published by: Blackwell Publishing on behalf of Accounting Research Center, Booth School of Business, University of Chicago Stable URL: http://www.jstor.org/stable/2490070 . Accessed: 12/08/2012 12:50 Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp
. JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected]
Blackwell Publishing and Accounting Research Center, Booth School of Business, University of Chicago are collaborating with JSTOR to digitize, preserve and extend access to Journal of Accounting Research.
Contentof Annual The Information EarningsAnnouncements WILLIAM
contentofearningsis an issue ofobviousimportancean The information in accounting.Thi is a focal point formany measurementcontroversies paper empiricallyexaminesthe extentto which commonstock investo value. The study directsits at perceiveearningsto possessinformational as reflectedin th tentionto investorreactionto earningsannouncements, volume and pricemovementsof commonstocksin the weeks surroundi the announcementdate. Valuation theoryhas long posited a relationshipbetweenearningsan the value of commonstock. Miller and Modiglianipostulatethat one im thevalue ofcommonstockis theproducto portantelementin determining earningstimes the appropriateearningsmultiplierfor that risk class Graham,Dodd, and Cottle take a similarpositionwithrespectto the com putationof their"intrinsicvalue" of commonstock securities.2MM als provideempiricalevidencethat suggestsif reportedearningsare adjuste formeasurementerrorsthroughthe use of instrumentalvariables,the ad justed earningsare usefulin the predictionof the marketvalue of electr utilityfirms.In fact,the evidenceindicatedthat the earningstermwas th mostimportantexplanatoryvariablein the valuation equation.3The rel conten tionshipis a necessaryconditionforearningsto have information
AssistantProfessor,Universityof Chicago. 'Merton H. Millerand FrancoModigliani,"Some EstimatesoftheCost ofCapit to the ElectricUtilityIndustry,1954-57,"AmericanEconomicReview,LVI (Jun 1966),341. Analysis (New 2 BenjaminGraham,David L. Dodd, and Sidney Cottle,Security *
but the evidencedoes not precludethe possibilitythat the oppositemay be true. Althoughthereare many reasonsforadoptingthe positionthat earnin lack informationalvalue, two are frequentlyoffered.(1) Measuremen errorsin earningsare so largethat it would be betterto estimatethe valu of commonstock directlyfromthe instrumental variablesratherthan us earningsas an intermediatestep. (2) Even thoughearningsmightconve information, there are other sources available to investorsthat contai but are moretimely.By the time annua essentiallythe same information contenthas alreadybee earningsare released,any potentialinformation processedby investorsand is impoundedin the marketprice.The implic tion of both argumentsis earningsreportshave little or no informati content. The issue is of major concernto the accountingprofessionbecause it outcomedirectlyreflectsupon the utilityof the accountingactivity.On approachto examiningthisissue is to specifyan expectationsmodelofhow investorsrelate reportedearningsto marketprices. The paper present by Benston at last year's Conferencefollowedsuch an approach.4Bensto foundprice changes were largelyinsensitiveto earnings,which taken a face value is unfavorableto the utilityof earningsdata. But such resul to interpret are always difficult because the lack of an observedrelationsh may be due to eitherone or both of two factors.Either no relationsh existsor the expectationsmodel was improperlyspecified.It is impossib to determinethe extentto whichthe negativefindingsare due to the latte ratherthan the former. The approachtaken hereis to apply tests that requireno assumptio about the expectationsmodelsof investors.Note that the issue undercon siderationis of a positive ratherthan a normativenature-that is, th questionof concernis not whetherinvestorsshouldreact to earningsbu ratherwhetherinvestorsdo react to earnings.
Content Definitionsof Information
Information has been definedas a changein expectationsabout the out comeofan event.'Withinthe contextofthisstudy,a firm'searningsrepo is said to have informationcontentif it leads to a change in investor assessmentsof the probabilitydistributionof futurereturns(or prices suchthat thereis a changein equilibriumvalue ofthecurrentmarketprice
GeorgeJ. Benston,"Published CorporateAccountingData and Stock Prices,
Empirical Research in Accounting: Selected Studies, 1967, Supplement to Vol. 5 Journal of Accounting Research, pp. 1-54. I Henri Theil, Economics and Information Theory (Chicago and Amsterdam
Rand McNally and NorthHolland PublishingCompany,1967),Ch. 1. 6 A further stipulationis oftenmadethatinformation concernschangesin expecta tionsaboutan eventthatis a parameter ofa decisionmodel.Definingearningsinform tionin termsofits impacton futurereturns(or prices)is consistentwiththatfurth
Althoughneitherthe directionnor the magnitudeof the price changecan be specifiedwithoutknowingthe expectationsmodel(s) of investors,th variabilityof price changes is likelyto be greaterwhen earningsare an nouncedthan at othertimesduringthe year.7 states that not only must therebe a Anotherdefinition of information large to induc changein expectationsbut the changemust be sufficiently a changein the decision-maker's behavior.Accordingto this definitiona value only if it leads to an firm'searningsreportpossessesinformational alteringof the optimalholdingof that firm'sstock in the portfoliosof in dividualinvestors.The optimal adjustmentmightbe to buy more share or to sell some or all of the sharesalreadyheld. In eitherevent,the shiftin in thevolume.If earningsreportshav portfoliopositionwouldbe reflected information content,thenumberofsharestradedis likelyto be higherwhe the earningsreportis releasedthan at othertimesduringthe year.8
Relationshipsbetween Price and VolumeTests
The relationshipsposited above are consistentwith the economist notionthat volume reflectsa lack of consensusregardingthe price. Th the earning lack of consensusis induced by a new piece of information, in theway theyinterpret the report,som report.Since investorsmay differ timemay elapse beforea consensusis reached,duringwhichtimeincrease volumewould be observed.If consensuswere reachedon the firsttransac tion, therewould be a price reactionbut no volume reaction,assumin homogeneousrisk preferencesamong investors.If risk preferencesdiffe therestillcould be a volumereaction,even afterthe equilibriumpricehad been reached. An importantdistinctionbetweenthe priceand volumetestsis that th formerreflectschangesin the expectationsof the marketas a wholewhil the latter reflectschanges in the expectationsof individualinvestors.A
stipulation.For support,see the literatureon portfoliotheory,especiallyHarryM (New York Diversification of Investments Markowitz,PortfolioSelection:Efficient JohnWiley & Sons, 1959). 7 The changein equilibriumprice is in additionto any price changethat woul normallyoccur in the absence of any earningsannouncement.The assumptioni that the two price changesare positivelycorrelated,independent,or mildlycorr lated. If therewerestrongnegativecorrelation,the price changevariabilitymigh not be greaterat the announcementdate. In the light of previousresearchin th behaviorof securityprices,the assumptionof independenceis mostlikelyto be th correctone. See Eugene F. Fama, ''The Behaviorof StockMarketPrices," Journ of Business,XXXVIII (January,1965),34-105. 8 As a finalparenthetical of information, notethat redu commenton definitions was not one of the definitions chosen.It shouldbe apparenttha tionofuncertainty assessmentsare changin in a dynamicsituation(i.e., whereprobabilitydistribution overtime),a decisionmakermaybe moreuncertainabout a giveneventafterrecei ing a messageabout the event than he was beforehe receivedthe message.To us the entropymay increaseas a resultof a message,yet the me Theil's terminology, content.See Theil, op. cit.,Ch. 2. sage has information
may be neutralin the sense ofnot changingthe expec piece ofinformation tationsof the marketas a wholebut it may greatlyalter the expectation ofindividuals.In this situation,therewould be no pricereaction,but ther would be shiftsin portfoliopositionsreflectedin the volume. Because th pricereflectsthe expectationsof many investors,it may implya veryeff cientforecastof earningsforseveralweeks priorto the announcement.'If so, the pricetestmay be less sensitivethanvolumeto earningsreports. The foregoingdiscussionsuggeststhat a reactionmay be observedin only one of the tests or that the two tests may not respondequally. If neithertest responds,the utilityof earningsdata and the study's sampl designwill be suspect.
SelectionofSample. The studyis based upon a sample of annual earning announcementsreleasedby 143 firmsduringthe years 1961 through1965 Six criteriawereused in the selectionofthe samplefirms. (1) The firmmust be on the Compustat tape; (2) the firmmust be a memberof the New York Stock Exchange; (3) the fiscalyear mustend on a date otherthan December 31; (4) no dividendswere announcedin th same week as the earnings announcement; (5) no stock splits wer announcedduringthe 17 week period surroundingthe announcemento earnings;and (6) therewere less than 20 news announcementsper yea appearingin the Wall StreetJournal.Table 1 indicatesthe extentto whic each criterionaffectedthe sample size. Criterion(1) was selectedbecause the Compustatpopulationrepresen over90 per centofthe total marketvalue ofthe commonstocksofpublicl held corporationsand henceis a relevantpopulationforstudy.A secondar reasonis the ease withwhichfinancialstatementitemscan be obtainedfo the Compustatfirmsrelativeto firmsnot on the tapes. Althoughno finan cial statementdata are needed forthe earlierphases of this study,even tuallythe scope will be extendedto relatingmarketpricesto the financi statementitems,namelythe earningsnumbers. Criterion(2) was used because weeklypriceand volume data on NYSE firmsare relativelyeasy to obtain. The Center for Research in Securit Prices (CRSP) provided tapes which contain daily price, volume, and transactioninformationon all firmson the NYSE for the years 196 through1965.10 of announceme Criterion(3) was selectedin orderto avoid a clustering
9Efficiency is definedin termsof E(x6- x*)2, wherex is the forecastedvalue o reportedearningsand x* is actual value. The closerthe expectationis to zero, th the forecastis. Note that a forecastmay be unbiasedbut veryinef moreefficient and unbiasednessis moreimportantto th cient.The distinctionbetweenefficiency presentedlater. of the findings interpretation 10Withoutthe cooperationof CRSP, the data collectionchorewould have bee overwhelming.
dates duringany time period. Without this criterion,the sample data would exhibita large clusteringof announcementsin the monthsof February,March, and April because two out of three Compustat firmsare 12/31firms.In subsequentanalysis,an attemptwill be made to remov the effectsof market-wideevents fromthe individualsecurity'svolum cluster,theybecomea form and pricedata. When earningsannouncements of market-wideprice indexes and the volume statistics.Hence, any attemptto removethe effectsof market-wideevents would eliminatethe ofthe earningsreportas well. effects The purposesof criteria(4) and (5) are similarin that they attemptto minimizeany ambiguityassociatedwithan observedreactionin the week of the earningsannouncement.If these criteriawere not applied, ther to separatethe would be a joint effect,and it would be extremelydifficult effectsofdividendsor stocksplitsfromthoseofthe earning announcement report."Criterion(6) was chosenso that therewould be weekswherefew if any, announcementswere released. To the extentthat news items are announcedin weeksotherthan the earningsannouncementweek,compar ing thoseweekswiththe earningsannouncementweek comparesthe infor mation contentof the earningsreportswith that of other types of new whichis not the issue understudy. announcements, Both the directionand magnitudeofany potentialbias introducedby the to assess. Criteria(1) and (2) led to the selec selectioncriteriaare difficult tionofthe largerfirmsin the economy.The averagetotal assets (perfinan cial statements)forthe 143 firmsin 1965 was 167 milliondollars,and thei averagemarketvalue ofcommonstockoutstandingin 1965 was 189 millio dollars. The effectof selectinglargerfirmswould tend to induce a bias against earningsreportsbecause the largerfirmsare generallyassociated than smallerfirms. witha greaterflowof additionalinformation The effectof criterion(3) was twofold:(a) Out of the subpopulationof Compustat,NYSE firms,the criteriontended to select the smallerfirm even thoughthey are probablystill largerthan average forthe econom as a whole. In 1965, the average total assets forComputsat,NYSE firm were 441 milliondollars,and the average marketvalue of commonstock outstandingwas 564 milliondollars. (b) A greaterproportionof retailer and foodprocessorsappears in the sample than would have been obtained if firmshad not been restrictednon-12/31firms.Retailers comprise14.6 per centof the sample,while 17.5 per cent of the firmsare foodprocessor The expectedpercentageswould have been 6.8 and 10.0, respectively,if
11A pilot studywithsimilarobjectivesdid not excludefirmswith dividendanin thesame weekas earnings.The investorreactionin termsofvolum nouncements was almosttwice as large as the reactionobservedin this study.Stock splits wer excludedbecause previousresearchhas foundthat stocksplits possess informati content.See Eugene F. Fama, et al., "The Adjustmentof Stock Prices to New Information,"Report 6705 (Centerfor MathematicalStudies in Business and Ecoin nomics,GraduateSchool of Business,Universityof Dhiiago,1967),forthcoming EconomicReview. the International
a randomsample were drawnfromthe Compustat,NYSE subpopulatio With respectto the implicationsforinformation content,thereare to b no obviousreasonswhythesefirmswould constitutea biased sample,wit the one exceptionthat retailerstend to reportfinancialstatementdat monthlywhichwould tend to induce a bias againstfindinginformatio value in annual earningsreports.In fact,in the analysis describedlater boththe priceand volumereactionswereless dramaticforthe retailersan foodprocessorsthan forthe otherfirmsin the sample. It is possiblethat the selectioncriteria,especiallycriterion(6), may in duce some bias in the oppositedirection.As long as the criteriaare visib ex ante, the populationforwhichthe study'sfindingsare relevantcan b easily identified.Also, the sample criteriacan be relaxedin futurestudie to discoverthe generalityof the findingspresentedhereforotherpopula tions. Data Collection. The firststep was the identification of firmsthat woul comprisethe sample. Meeting the criteriain any one of the five year (1961-1965) was a sufficient conditionfora firm'sinclusionforthat year The resultwas a sample of 143 firms.Because all firmsdid not meet th criteriain everyyear, the 143 firmsgave rise to 506 annual earningsan nouncements.The date of the earningsannouncementwas obtainedfro the Wall StreetJournalIndex. The distributions of financialstatementdates and announcementdate appear in Table 2. Restrictingthe sample to non-12/31firmswas succes fulin reducingthe clusteringof dates. The mostfrequentmonthin whic the fiscalyear ended (June) representsonly 23.8 per cent of the sampl whilean unrestricted sample would have resultedin 67 per centin a sing month (December). With respect to announcementdates, the highe three-month period(September,October,and November)contains37.6 pe centofthe announcements, whileunderan unrestricted samplingprocedu the highestpercentagewouldhave been approximately 67 per cent (durin February,March, and April). The mostfrequentmonthof announceme (October) represents13.4 per cent of the announcements,which is onl slightlyhigherthan the percentagethat would be obtained under a com pletelyuniformdistribution throughoutthe year (9.1 per cent). One by-productof the data gatheringwas some insightinto the tim lag betweenthe financialstatementdate and the announcementdate (se Table 3). The median lag was 9 weeks,only 3 per cent of the announc mentsweremade by the end of 4 weeks,and 93 per cent of the earnin had been reportedby the end of 13 weeks. A possible avenue forfutu researchwould be to study the information contentof the time lag itse (e.g., is "bad" news reportedless rapidlythan "good" news?). Definitionof Variables. The next step was to compute the followi variablesforeach firmon a weeklybasis forthe 261 weeks (fromJanuar 1, 1961 to December31, 1965):
no. of sharesof firmi tradedin week t X it= I no. of shares outstanding no. of trading forfirmi in week t days in week t
no. of sharestraded forall NYSE firms in week t I no. of shares outstanding no. of trading days in week t forall NYSE firms in week t In [Dit + P]
Rmt = In
L(SP) t-1J = cash dividend"paid" on share of firmi in week t, Dit Pit = closingpriceforshare offirmi at end of week t, = closingprice at end of week t - 1, adjusted forcapital chang (e.g., stocksplitsand stockdividends), (SP) = closingvalue ofStandardand Poor's PriceIndex at end ofweek (SP) t-1 = closingvalue at end of week t - 1.
Vit is a weeklyaverageof the daily percentageofsharestraded.Weekl volumewas dividedby the numberofsharesoutstandingso that the resul would not be dominatedby those firmswiththe largestnumberof share outstanding.The percentageof shares traded per week were then divide by the numberof tradingdays in orderto adjust forthe fact that not a weekshave the same numberof tradingdays. thelevelofvolumeforall NYSE firms.The weightingschem VMtreflects implicitin thisvolumeindexassignsgreaterweightto percentageof shar traded of firmswith the largernumberof shares outstanding.While th featureis not entirelysatisfying,its use is defendedon the groundstha thisindexis mucheasierto obtainthan an indexthat assignsequal weigh to all firmsand because thereis no reason to believe the use of this inde leads to eitheran upwardor a downwardbias in the findingsregardingth contentof earningsreports. information Jit is the naturallogarithmof the pricerelativeand can be viewedas measureof price change or as the rate of returnof the securityassumin RMt is a similar measure for 425 industr continuouscompounding.12
12 The properties ofRit are furtherdescribedin Fama, op. cit.; BenjaminF. King "Market and IndustryFactors in Stock Price Behavior," Journal of Busines XXXIX (January,1966), 139-90;JamesH. Lorie and LawrenceFisher, "Rates o Returnon Investmentin CommonStocks,"JournalofBusiness,XXXVII (Januar 1964),1-21.
NYSE firms.This statistichas some limitationsas a market-wideinde of pricechangeand in many respectsis less preferablethan some recent developed indexes,notably Fisher's Link Relative.'3 However, again it index o use is defendedon the same groundsas thoseforthe market-wide volume. The Fisher Link Relative has been computedformonthlydat only. To constructa similarindex on a weeklybasis is a researchprojec in itself.Not onlyis the S & P index easier to obtain but it was foundi otherstudiesthat resultswereinsensitiveto whichindex is used.'4Withi the contextof this study,thereis no reason to believethat the use of th of th or understatement S & P based indexwill lead to an overstatement information contentof earningsreports.
Vjt was computedforeach week t in the reportperiodforeach of the 50 j. The reportperiodis definedas the 17 weekperio earningsannouncement surroundingthe announcementdate (8 weeks beforethe announceme week,and 8 weeks after).Then the '[t (averagingacrossj) was compute foreach ofthe 17 weeks,and the resultsappear in Figure1. The dottedlin denotesthe value of Vt in the nonreportperiod (i.e., that portionof th 261 weeksnot includedin the 17 week reportperiods). The evidenceindicatesa ratherdramaticincreasein volume in the an nouncementweek (week 0). In fact,the mean volumein week 0 is 33 pe cent largerthan the mean volume duringthe nonreportperiod,and it i by far the largestvalue observedduringthe 17 weeks. Investorsdo shi and thisshi portfoliopositionsat the timeof the earningsannouncement, is consistentwith the contentionthat earningsreportshave informat content. The contentionis furthersupportedby the behaviorof investorsin th other weeks. Eight weeks prior to the announcement,volume is belo normal,whichsuggeststhat investorsmay postponetheirpurchasesan sales of the securityuntil the earningsreportis released.The fourweek after the announcement,when the annual reportsare received,exhib slightlyabove normalvolumeand hencepermita morethoroughevaluatio of the earningsdata. The investorresponseappears to be very rapid, for almost all of th above-normalactivityoccursduringweek0. This findingsupportsprevio studies that also show investorsrespond quickly (as reflectedin pric changes) to new pieces of information(see Fama, et al.). Perhaps some commentis in orderregardingthe overalllevel of volum throughoutthe year. The volumestatisticsreportedin Figure 1 are mult
13 For a discussionof the S & P Index vis-A-vis Fisher's Index, see Lawren Fisher,"Some New StockMarketIndices," JournalofBusiness,XXXIX (Januar 1966),191-225. I4 Fama, et al., op. cit.
plied by the factorof 103.The average volume in the nonreportperiod .00112-that is, the average daily percentageof shares traded is slight greaterthan one-tenthof one per cent of the sharesoutstanding.This im plies an annual turnoverof approximately25 per cent and a weeklytur X has 10 millionsharesoutstandin overof.5 ofonepercent.If corporation duringa normalweek50,000shareswillbe tradedwithan expectedvolum of 66,667sharesduringthe earningsreportweek.
The sectionwillpresentan analysiswhichattemptsto removethe effe of market-wide eventsupon the individualsecurity'svolume.The motiv tionforthe analysisis two-fold.(1) It is possiblethat the abnormallyhig th volumemay be caused in part by market-widepieces of information are released at the same time as the earningsannouncements.Since th throughout theyea are releasedalmostuniformly earningsannouncements this is not a very plausible explanationof the findings.Nevertheless,r movingthe marketwide effectsshould allay any fearsthat this unlike the analys situationdoes accountforthe results.(2) More importantly, will serve to reduce "noise" in the volume data. Noise is any movemen in volume due to unspecifiedfactors,one of whichis market-wideeven that would cause increasesin the volume. Period.The followingmodelwas used to abstra AnalysisforNonreport factors:15 frommarket-wide Vit
ai + biVMt + eit.
Estimates of as and bi were obtained fromlinear regressionsbased upo observationsfromthe nonreportperiod.The observationsfromthe repo periodweredeletedfromthe regressionbecause if earningsannounceme mod content,the assumptionsof the classicalregression have information are violated duringthe reportperiod (e.g., E(eit) - 0). Some summarystatisticsrelatingto the regressionsappear in Table 4 The mean volume of the sample firmsis much higherthan that of th weightingschemeimplicitin eac marketindex.One reasonis the different measure.The marketindexassignsgreaterweightto firmswiththe great numberof sharesoutstanding.If thesefirmshave lowervolume (express as a percentageof shares outstanding),then the marketindex would b expected to have a lower mean. Anotherexplanationis that the samp
15 The rationale for using this particularmodel is two-fold:(1) It is a simp relationship,and thereis no obvious reasonwhy a morecomplexmodel would b moreappropriate.(2) It is analogousto themodelthatwillbe used to removeeffe ofmarket-wide eventsuponthepricechangesofindividualsecurities.The paperw laterindicatethatsuch a modelseemsto be a reasonableway to characterizepri changes.Hence, it wouldseem reasonableto assume a similarprocessis generat volumeover timeas well.
implicitly favoredhigherturnoversecurities.But it is no selectioncriterion obviouswhythat shouldbe truenor what implicationit has forinferenc content. regardinginformation was low, implyingthat removingth The averagecorrelationcoefficient influenceof VMtshouldhave littleeffectupon the analysis.In spite of th was positivefor 13 low association,the sign of the correlationcoefficient firmsand negativeforonly 4. These two findingstaken togethersugge that the marketinfluenceon an individualfirm'svolume is significan different fromzero but that its magnitudeis small.16 The residual,eit, is that portionof an individualsecurity'svolumetha in VMt. The mea eventsas reflected cannotbe explainedby market-wide of es (averagingacross time forgiven firmi) is forcedto be zero by th mechanicsof the regressioncomputations.However,the mean of et (aver age acrossfirmsfora givenweek t) may be nonzero.An inspectionof it distributionfor the 261 weeks provides some interestinginsights (se Figure2). is skewedto the right,as indicatedby the fact that 5 The distribution per cent of it are negativeand 42 per cent are positive.The median of e is -.02 and its mean is zero (again this must be true because of the me chanics of the regressioncomputations).The ei's are even more asym metrical,with64.6 per centnegativeand 35.4 per centpositive.One inte is providedto investorsin pretationof the asymmetryis that information discontinuous"lumps" ratherthan smoothlyor continuouslyovertime. ResidualAnalysisfortheReportPeriod.The residual,ejt, was compute foreach week t of the reportperiodforeach of the 506 earningsannounc mentsj in the followingmanner: ejt = Vjt -
in the nonreport period whereai and bi wereobtainedfromtheregressions Then the it was computedforeach of the 17 weeks,and the resultsappea in Figure 3. A positive residual implies above normalvolume; negativ belownormal;and zero,normalvolume. The behaviorof the volumeresidualis the same as that of the previou analysis. There is a large peak in week 0, wherethe mean volume is ap proximately30 per cent higherthan duringthe nonreportperiod (i.e .33/1.12,mean residualin week0/meanvolumein the nonreportperiod and is about 40 per cent higherthan the mean volumein the weeksprio
is less tha coefficient I6 The probabilitythat theexpectedvalue ofthe correlation or equal to zero is less than 1 chancein 100,000. 17Note thatthesubscripti refers to firmi or securityi, butj refersto an earning Hence as and bi may be used a maximumof fivetimes;its frequen announcement. of firmi or securit of use will dependupon the numberof earningsannouncements i includedin the sample of 506 announcements.
Again the volume duringtheseweeksis abnorma to the announcement. low, whileslightlyabove normalvolume persistsforfourweeksfollow is the same ofthesefindings week.The interpretation the announcement that of the previousanalysis.In short,the resultsare veryconsistentw content possessinformation the contentionthat earningsannouncements Because a comparisonofmean values can oftenbe misleading,two ad tional comparisonswere made to see how unusual is an it of .33, wh was the value observed in week 0. The firstcomparisonexamined t values of it in the reportperiod and those in the nonreportperiod (s Figure2). Out of the 261 nonreportperiodvalues of it, only4 had valu exceeding.33. Althoughsuch a comparisonis admittedlya crude appro mation,it does suggestthat the value in week0 is unusuallyhigh. Moreover this comparisontends to understatethe unusual nature the week 0 residual.The it duringthe nonreportperiodis based upon maximumof 143 observationsper mean, while the it in week 0 (as w as the rest of the reportperiod) was based upon 506 observations.Sin correlated,the dispersionof the distribu the eit's are less thanperfectly of et would decrease as the numberof observationsper mean increas Hence, if the distributionin the nonreportperiod were also based up 506 observationsper mean,its dispersionwould be smallerand the num of values above .33 would be fewer.Anotherfactorleading to an und statementis that et in the nonreportperiodwas based upon residualstak fromthe same week, while the mean residualin week 0 was based up weeks. If contemporaneousresidu observationstaken from different residuals (and t are more highlycorrelatedthan noncontemporaneous evidence suggeststhey are), then a distributionof it in the nonrep observationswouldhave a smal periodbased upon noncontemporaneous dispersionand fewervalues above .33.18The major pointis the compari indicatesthat the mean residualin week0 is unusuallyhigh,in spite oft fact that the comparisontendsto understatehow unusual it reallyis. A second comparisoninvolvedthe analysis of the frequencyof posit residualsin each reportperiodweek as comparedwiththe numberdur the nonreportperiod (see Figure4). The behaviorof the positiveresidu is consistentwith the previousrelationshipsobservedin Figures 1 and Prior to the announcement,the frequencyof positive residualsis bel that of the nonreportperiod,whilethe frequencyis slightlyabove norm By farthe largestfrequencyoccursin week 0, a afterthe announcement. smallprobabilitythat such a highnumberof posit thereis an extremely residualscould have occurredby chance.'9 This second comparisonsu geststhe same inferencedrawnfromthe first-namely,the volumein we 0 is an unusuallyhigh value. In sum, the behaviorof volume unifor
18 The serial correlation is reflectedin the positiveautocorrelationcoefficien theresiduals(see Table 4). Anotherindicationis thatthefourvalues ofe exceed .33 occurredin a five-week period. 19The probabilityis less than 1 chancein 100,000.
contentforindiv supportsthecontentionthat earningshave information ual investors. In somerespects,thesefindingsdo not reflectthe entireextentto wh activityis above normalin week0. Not all of the earningsannounceme of the 143 firmswereused-in fact,only 506 out of a possible715. The the remaining209 are includedin the nonrep week periodssurrounding period.This willtendto inducea bias againstearningsreportssincevolu activityis increasedin the nonreportperiod by the inclusionof the 2 "reportperiods." The extentof this bias could be seriousbecause one the reasonsforplacing a reportin the 209 groupwas the announcem of earningsand dividendsin the same week which would produce ev studied.How morepriceand volumeactivitythan the 506 announcements ever,thereare also compensatingfactors.Althoughthe activityin week was above normal,the activityin the weeks priorwere below normalf the 506 observations.If thistendsto be trueof the deletedannounceme as well,the bias may not be so great.If the 209 observationsweredelet fromthe nonreportperiod,to be completelyconsistent,othertypesofne announcementswould also have to be deletedfor the same reasons. Th resultwould be virtuallyno observationsin the nonreportperiod.Sin the nonreportperioddoes includethese events,it is importantto stresst fact that comparingthe earningsreportperiodswiththe nonreportperi contentof earningsreportswi involvesa comparisonof the information the average amount of informationbeing released duringthe nonrep period.By necessity,thisis a bias againstearningsreportssincethe appr at all. priatecomparisonwouldbe a nonreportperiodwithno information
in the sense ofleadingto chang If earningsreportsconveyinformation in the equilibriumvalue of the currentmarketprice,the magnitudeof t price change (withoutrespectto sign) should be largerin week 0 th duringthe nonreportperiod.The firststep in makingthispredictionoper tional is to removethe effectof market-wideevents upon the individ security'sprice change. The reasons for wishingto abstract fromth events are similarto those cited in the volume analysis.20The model us herewas firstsuggestedby Sharpe,and it providedthemotivationforusi an analogousmodelforvolume.2'The Sharpe modelstates: Rit = at + biRMt+ uit.
Rit is a measureof the pricechangeof securityi duringtimeperiodt, an RMt is a measureof averageprice change duringtime periodt for425 i dustrialNYSE firms.Both variables were definedearlier.The residu
See p. 75. WilliamF. Sharpe, "A SimplifiedModel forPortfolioAnalysis,"Managem Science,IX (January,1963),277-93. 20
that portionof the individualsecurity'sprice change tha uit, represents
cannotbe accountedforby the effectsof market-wideevents as reflect
The Sharpe model has been investigatedby Fama et al. and by Schole and was helpfulin abstractingfromthe influenceof market-widefactor King's studyof monthlypricechangesfoundthat, on the average,31 pe cent of the variationin an individualsecurity'sprice change can be ex indexofpric in a market-wide factorsas reflected plainedby market-wide change.22For these reasons,a price change analysis, unadjusted for th influenceof market-widefactors,was not conducted. The evidence wi laterindicatethatifsuchan analysishad beenconducted,the resultswoul be essentiallythe same as thosereportedhere. Since the directionof the pricechange cannotbe specified,a knowled of uit tha of the investors'expectationmodel(s), some transformation is the squar abstractsfromits sign,is needed. One such transformation conten of the residual (i.e., ui t). If earningsreportspossess information b 2 period.Th U2 t shouldbe greaterduringweek 0 than duringthe nonreport mean of U2t duringthe nonreportperiod is simplythe variance of tha variable (s,2).23 The relationshipbetweenthe squared residualin week0 and the averag squared residualduringthe nonreportperiodcan be expressedin the for of the ratio,Uit, wherethe numeratoris uit and the denominatoris si2. I the ratiois greaterthanone,the residualpricechangeis largerthannorma and converselyfora ratioofless than one. The predictionis the mean of U will be greaterthan one duringweek 0 (averagingacross announcements) content. if earningsreportspossessinformation Period.Estimatesof as, bi, and s,2wereobtaine AnalysisofNonreport fromregressionsbased upon the nonreportperiod.The observationsfro each announcement)wer the reportperiod(i.e., the 17 weekssurrounding conten deletedfromthe regressionbecause if earningshave information the assumptionsof the classical regressionmodel are violated duringth reportperiod (e.g., the variance of the residualsduringthe reportperio is not equal to the varianceduringthe nonreportperiod). Some summarystatisticsrelatingto the regressionsappear in Table 5 The mean pricechangestend to be lowerforthe samplefirmsthan forth as a rate of return,th marketindex.Since the Rit call also be interpreted lowerreturnsforthe sample firmswould suggestthat they are less risk
22 Fama, et al., op. cit.; MyronScholes, "The Effectsof SecondaryDistributio upon the Market Price" (paper presentedat the November,1967 session of th ConferencefortheStudyofSecurityPricesheldat the GraduateSchoolofBusines UniversityofChicago); and King,op. cit.The percentagerefersto theperiodAugus 1952throughDecember,1960. is the estimateof yj2, computedfro = 23 The variance (j2 [ sj2 ]/T, where T = number of weekly observations fo sample data. Sz2 = [Ze=i(uit)E(uj)]2. the nonreportperiodforsecurityi.
than the firmscomprisingthe index.An inspectionof the distribution of b also lends supportto that contention.Sharpe states that bi can be viewe as an operationalmeasureof a security'sriskiness,withlargervalues of b implyinggreaterriskiness.24 A bi of one denotes a securityof "average riskiness.The average bi forthe sample firmsis less than one (.89), whic suggeststhat the samplefirmsare less risky.However,the discussionin th sectionon definition ofvariablesindicatedthat the definition of RMt base upon the S & P indexmay be subject to measurementerror.An errorsvariablesmodelsuggeststhat measurementerrorin the independentvaria ble, even if it has a zero expectation,will induce a downwardbias in th estimates of the regressioncoefficientassociated with the independe variable (i.e., bi).25 Effortswere undertakento assess the extent of th downwardbias by computingbi forthe sample firms,usingmonthlydat and Fisher'sLink Relative as a definition ofRMt . The medianbi was .993 suggestingthe sample firmsare of average riskinessrelative to NYSE firms(i.e., the firmsthat comprisethe FisherIndex). On the average,the associationbetweenRit and RMt was low. Only 6 per cent of the variationin Rit can be explainedby the variationin RM The im as measuredby the square of the average correlationcoefficient. plicationis two-fold:(1) Removingthe influenceof RMtshould have litt effectupon the results,relativeto what would have been obtainedif M were analyzed ratherthan uit. (2) The explanatorypoweris much lowe than that obtainedby King, suggestingthat eitherweeklydata have mor noise than monthlydata or that RMt was not properlydefined,or both to detectany pric The presenceof eitherfactorwill make it moredifficult effectsof the earningsreports. The distributionof UC (averaging across 143 firms,t = 2, ,261 duringthe nonreportperiodis shownin Figure5. It will be used as a basi ofthe Ut's observedduringthe reportperiod forassessingthesignificance Price Residual Analysisfor ReportPeriod. The residual,ujt, was com puted foreach week t of the reportperiodand foreach of the 506 earnin announcements j in the followingmanner:
1,***, 143 1, ***,506 t =-8, **,+8. i=
The residualwas thensquared and dividedby the varianceof the residua forits firmduringthe nonreportperiod,as follows:
24WilliamF. Sharpe, "Capital Asset Prices: A Theory of Market Equilibriu underConditionsof Risk," JournalofFinance,XIX (September,1964),425-42. 25 J. Johnston, Methods(New York: McGraw-Hill,1963),148ff. Econometric 26 The distribution is skewedto the right.One explanationforthis phenomen is theleptokurticnatureoftheunderlying uit's (see Fama, op. cit.). The distributi of ustis also skewedin thesame direction.Althoughthe mean ofu~tis one for eac period,only26 percentoftheobservationsexceedon securityduringthenonreport
CONTENT OF ANNUAL EARNINGS ANNOUNCEMENTS
Ujt = u 1,t
1,.. , 143 i-= , 506 j t --8 ... ,+8.
U, (averagingacrossj) was computedforeach of the 17 weeksof therepo period,and the resultsappear in Figure6. The magnitudeofthe pricechangesin week0 is muchlarger(67 per cen higher)than the average duringthe nonreportperiod.The above norm price activityis what would be expectedif changesin equilibriumprice are more likelyto occur when earningsreportswere released,and henc the evidenceis very consistentwith earningsreportspossessinginform tionalvalue. Althoughthe priceactivityis highestin week 0, the next largestvalue occur in the weeks immediatelycontingentto week 0. Price changes ar whic above averagein the week immediatelypriorto the announcement, may reflectinformation leakage or the fact that the Wall StreetJourn was not the firstsourceto reportthe earningsin some cases. Above norm activityis also presentfortwoweeksafterthe announcement, duringwhic timethe annualreportsare releasedand are evaluatedby investors. The below price activityin weeks -8 through-2 is open to at leas two interpretations:(1) There is a below normal amount of informati comingonto the marketat this time. (2) The below normalprice activit is a result of the below normal volume also observed duringthe sam period.More will be said about both (1) and (2) later. The behaviorofthe mean residual,at , also indicatesgreaterpriceactiv ityinweek0 (see Table 6). The meanin week0 is .00500,whichis thelarge value observedduringthe 17 weeksand is fourtimeslargerthan the aver age value of Rit duringthe nonreportperiod (.00125, see Table 5). Th means give the impressionthat serial correlationmay be presentin th of the priceresidualswas quit data. However,the averageautocorrelation low (-.08) duringthe reportperiod. The low degree of autocorrelati supportsthe similarfindingsof Fama and his conclusionthat the mark Furtherevidenceof this i moves to new equilibriumpositionsquickly.27 in thefactthat the bulk ofthe pricereactiondoes occurin week reflected also suggeststhatthe pricechang (see Figure6). The low autocorrelation were permanentin natureand were not reversedin subsequentweeks. In fact, the autocorrelationof the residualsin the weeks immediatelyaft the announcementweek was slightlypositive.28 Two additional comparisons(analogous to those made in the volum analysis)wereconductedto see how unusualan Ut of 1.67 is. The firstcom parison examined UO in the nonreportperiod (see Figure 5). Out of 26 values, only 11 exceeded 1.67. The comparisonsuggeststhat the price ac
op. cit. cross-sectionalbasi autocorrelationwas examinedon a week-by-week, i.e., ot = -=i (ejtejetA)1/[Z'0 (eit)2], t = -8, *., +8. 27 Fama, 28 The
tivityin week0 is unusuallyhigh,in spiteof the fact that such a comp son tendsto understatehow unusual it reallyis.29The secondcompar examinedthe frequencyof Up's largerthan one relativeto the frequ that occurredduringthe nonreportperiod (see Figure 7). The frequ of values above one is greatestin week 0, withthe nexthighestvalues curringin the weeks adjacent to the announcementweek. There is an tremelysmall probabilitythat such a high number(181 in week 0) co have occurredby chance.30The interpretation is the same as that of mean analysis-namely,thereis above normalpriceactivitywhenearn reportsare released.What thisanalysisrevealsthat the mean analysisd not is the fact that the abnormallyhighmean is not caused by a few servationsdominatingthe resultsbut ratherby a substantialpropor of the sample data. In summary,the behaviorof the price changesuniformly supports contentionthat earningsreportspossess informationcontent.Observ a price reactionas well as a volume reactionindicatesthat not only expectationsof individualinvestorsalteredby the earningsreportbut a the expectationsof the marketas a whole,as reflectedin the change equilibrium prices.
Relationshipbetweenthe Volumeand thePrice Findings. The prev sentenceraises the issue, "how muchof the increasedpriceactivitycan attributedmerelyto the fact that thereis more 'action' in the secur ratherthan to changesin equilibriumprices?" One way to approach this questionis to view the price change dur a giventimeperiodas a sum of pricechangeson each transactionthat the pricechangef curredduringthat period.In a worldof uncertainty, each transactioncan be treatedas an observationfroma probability tributionof the investor'sassessmentof what the price changeshould The pricechangeper period,then,is a sum of randomvariables.If tr actions occur as if they are independentover time (evidence on da weekly,and monthlyprice changessuggestthey do), the variance of weeldy price change will increase in directproportionto the numbe transactionsthat occurduringthe timeperiod.3"
29 The reasonsforunderstatement are similarto thosestated in thevolumean sis. See p. 77. 30 The probability is less than 1 chance in 100,000. 31The evidenceregarding serial correlationof daily and monthlyprice chan can be foundin Fama, op. cit. and Fama, et al., op. cit., respectively.The aver coefficient forweeklychangesin thissamplewas --.08, whichwo autocorrelation withthe numberof tran cause thevarianceto increaseless thanproportionately may be higher(e.g., beca tions.Withina giventradingday, the autocorrelation of certaininstitutionalfactors,such as clusteringoflimitordersor stoploss ord fromb However,theexistenceof arbitragers shouldpreventthe autocorrelation verylarge.In orderforthepriceactivityto be explainedentirelyby increasedtr action activity,the autocorrelationwould have to be one. This would be hi forarbit unlikelybecause of theempiricalevidencecitedand the opportunities
The issue now is what is the appropriatemeasure of the numberof transactionsoccurringduringa given period. If the volume is used as a measure,then Ut in week 0 would be expectedto be 1.30 merelybecaus of moreactionin the security.The remainingportionwould be attribute to changesin the equilibriumprices of the securities.However,it is no at all clearthat volume(or evennumberoftransactions)is the appropria measure,because it reflectsonly the explicittransactionsthat occur. It could be argued,with considerablesupportfromeconomictheory that the expectationsof all investorsinfluencethe marketprice,whethe or nottheyengagein a purchaseor a sale. If themarketacts in thismanne the total numberof transactions,explicitand implicit,are the same pe timeperiod.Hence all ofthe above averagepriceactivitycan be attribute to changesin equilibriumprices. Additionalempiricalresearchis needed beforethisissue will be resolved The researchwould consistof studyingincreasedvolume activitydue to comingonto the market.An initialanalysi reasonsotherthaninformation of the seasonal variationin volume (VMt) from1946 through1966 reveale that the volumeis greatestduringthe monthsDecemberand January.The explanationseems to stem fromtax considerationsratherthan an abov Researchalso indicatedthat the pricevariabil normalflowofinformation. ity ofRMt duringthesemonths(i.e., Ut) was only.996,indicatingno abov average pricevariabilityduringthese months.This findinglends suppor to the positionthat none of the price activityin week 0 is due merelyto moremotion. Beforeleavingthis topic,note that isolatingthe volume effectson pric changesis of concernonlyto the extentone wishesto distinguishbetwee that altersthe expectationsof the marketas a wholefromin information formationthat alters only the expectationsof individualinvestors.All o in the lattersense. the price activitycan be attributedto information
duringReportPeriod Frequencyof OtherNews Announcements
The purposeof this analysiswas to discoverif therewas any clusterin of othernews announcementsaroundweek 0 that mightpossiblyaccoun forthe volumeand pricereactions.As indicatedearlier,the sample desig excludedany firmsthat announceddividendsin the same week as earning or any firmsthat splittheirstockduringthe reportperiod.However,it is mightclusterin weeksimmed conceivablethat dividendsannouncements atelypriorto and afterweek 0 or that othertypesof announcements(e.g. managementearningsforecasts)mightclusterin week 0. To examinethi in the Wall Stre possibility,the occurrenceof othernews announcements Journalduringthe 506 reportperiodswas examined(see Table 7). By far the most frequenttype of announcementwas dividends,whic exceededthefrequencyof all othertypes of announcementsby a factoro
9 to 1. Withrespectto thepurposeof thisanalysis,thereis no clusteringo in weeks -1 or +1; in factthe oppositeseemst dividendannouncements a be true.Also thereis no clusteringof any othertype of announcements any timeduringthe period,includingweek 0. The volumeand pricereac tionin week0 does not appear to be attributableto the clusteringof othe news announcements.32
The dramaticpriceand volumereactionindicatesthat investorsdo look directlyat reportedearningsand do not use othervariablesto the exclusio of reportedearnings.The evidencealso indicatesthat news announcemen occurringpriorto the earningsreportdo not entirelypreemptthe inform tion contentof reportedearnings.Given these findings,one of the fir extensionsof the study will be to explorethe possibilityof constructi expectationsmodels that will permita predictionof the directionand magnitudeof the priceresidual. The resultsof a recentstudy by Ball and Brown in this area are ver encouraging.33 They used an earningsmodel similarin formto the pric and volumemodelsdescribedin thisstudy (e.g., changesin the earningso inde an individualsecuritywereviewedas a linearfunctionofmarket-wide of earningschanges). The sample was dividedinto two groups:instance wherethe earningsresidualwas positive(actual earningswerehigherthan "expected") and instanceswherethe earningsresidualwas negative(actua earningslower than "expected"). The behaviorof the price residualsfo thesetwo groupswas examined,and the findingswere: (1) The signof th cumulativeprice residual (summedover a 12 monthperiodincludingth announcementmonth)was highlyassociatedwiththe sign of the earning residual. (2) There was a persistentupward driftin the cumulativemean priceresidualsforthe positiveearningsresidualgroup.This driftstarted1 monthspriorto the earningsannouncement,and over 90 per cent of th drifthad taken place by the beginningof the announcementmonth.The negativeearningsgroupexhibitedan analogousbehaviorpattern. The findings indicatethat reportedearningsare associatedwithunderly ing events that are perceivedby investorsto affectthe marketprice.Be as cause earliernews announcements conveysome of the same information to revis the earningsreports,investorsare able to use this information theirforecastsof earningsand to adjust the price accordingly.In fact,by month,investorsformlargelyunbias the beginningof the announcement forecastsofreportedearnings,even thoughthe reportedearningsare abov
32 As measuredin termsof numberof news announcements per week,the flowo does not appearto be below information duringtheweekspriorto theannouncement normaland hencewouldnotaccountforthebelownormalpriceactivityduringweek -2 through-8. 33Ray Ball and Philip Brown,"An EmpiricalEvaluation of AccountingIncom
Numbers," Journal of Accounting Research, 6 (Autumn, 1968), pp. 159-78.
or below normalrelativeto theirhistoricalrelationshipwithmarket-w earnings. for Althoughthe forecastsare unbiased,they are not very efficient, theywere,therewouldbe no volumeorpricereactionwhenearningsrepo were released.34 The Ball and Brown findingsand the findingspresen content here are mutuallysupportivewith respectto the information earningsreportsand also are uniformlyconsistentwith the finding previousstudiesin the behaviorof securityprices.One extensionof the searchpresentedherewill be to replicatethe Ball and Brownstudyon t sampleofnon-12/31firms(the Ball and Brownstudydealt exclusivelyw 12/31firms)and thento attemptto predictthe magnitude,as well as t sign,of the priceresidual. A second area of furtherresearchis the applicationof this methodol At an earliermeetingof the Co to othertypes of news announcements. contentof interim ference,Green and Segall exploredthe information ports.An analysisofvolumeand pricechangesduringthe announcemen approachto thissame issue. T interimearningswouldprovidea different contentof dividendannouncementsis anothertopic that h information receivedmuchattentionand stillis in needofadditionalempiricalinvest tion. Such researchwill indicate the importanceof annual earningsa nouncementsrelativeto otherkindsof information. Perhaps the most importantextensionof this study would be deal withthenormativeissue,"Should decisionmakersperceiveearningsrepo to possessinformational value?" The normativequestioncan be approach by selectingan event of interestto decisionmakers (preferablyas free possiblefromthe influenceof theirperceptions)and by investigatingt abilityofearningsdata to predictthatevent.A fewstudiesofthistypeha but muchmorewo been presentedat earliermeetingsof the Conference, is neededin this area.85Hopefully,the findingspresentedherewithresp to the positive question will provide greaterinsightinto the normat issue as well. 34 The
was discussedin footno distinctionbetweenunbiasednessand efficiency
35James0. Horrigan,"The Determinationof Long-TermCredit Standingw
SelectedStudies,1966,Supp Financial Ratios," EmpiricalResearchin Accounting: mentto Vol. 4, Journalof AccountingResearch,pp. 44-62, and William Beav "Financial Ratios as Predictorsof Failure," ibid,pp. 71-102.
ofSelectionCriteriauponSampleSize Effect No. offirms
Compustatfirms(step 1)a.896 .599 Less: 12/31firms Non-12/31firms(step 2).297 Less: Non-NYSE firms
.242 NYSE and non-12/31(step 3) Less: More than 20 announcements per year... week................39 Dividends in earningsannouncement Stock split duringreportperiod. Otherb........................................ Sample size (step 4)
48 7 99 ...................5
a Sample criteriawere applied sequentiallyin fourstages. The sample size afte each stage is denotedby parentheticalcomment(e.g., steps 1, etc.). b Miscellaneousreasonssuch as firm'searningswere not reportedin Wall Stree Journal. 2 TABLE Dates and Announcement Distributionof FinancialStatement Month
Pecntageof timeseari Percentageoffirmswhosefiscal Percents wre arnings yearendedin each month in each month
.... January February......................... March..... .
7.0 6.3 7.8
May. June..... July.......................... August. September. .... October. November........................ .. December.
1.4 23.8 9.6 7.8 15.3 9.1 5.6 0.0
7.5 .................... 2.3 2.8 5.0
8.7 6.5 .................... 6.8 11.3 11.9 13.4 12.3 11.5 100.0
a Total numberoffirmsequals 143,and total numberofannouncements equals 506
3 TABLE Fiscal Year-Endand Date of Announcement Numberof Weeksbetween No. ofweeks
Less than 4 4
18.9 32.9 46.7 57.9 68.9 77.5 86.1 93.0 96.0 98.2 100.0
11.6 14.0 13.8 11.2 11.0 8.6 8.6 6.9 3.0 2.2 1.8
6 7 8 9 10 11 12 13 14 15 More than 15
is 506. Total numberof announcements 4 TABLE SummaryofRegressionStatisticsVolumeAnalysis Item Item
Fractile .10 .25 .50 .75 .90
No. ofobser- Mean ofdepend- Mean of indeAutocorrelat per ent variable pendentvariable Correlation coefficient of fvations -coefficient in non~~firm coefficientl (Vi) X 10lrsdul (Vi) x lo0 reportperiod
165 176 193 210 227
.33 .53 .88 1.56
.577 .583 .588 .595 .608
.06 .16 .28 .39
.21 .29 .39 .50
5 TABLE StatisticsPrice Analysis SummaryofRegression Item
No. ofobser- Mean ofdepend- Mean of inde- Regressioncoef- Correlatio ent - variable pendetvral coefficien etaibeficient ofRmt (Rm) X 10' (Ri) X 108 reportperiod
per ~~vations in non~~firm
Fractile .10 .25 .50 .75 .90
165 176 193 210 227
1.25 -2.13 -.26
1.51 2.88 3.98
.96 1.25 1.51 2.04 2.96
.42 .62 .87 1.13 1.44
.13 .22 .27 .32 .37
TABLE 6 Analysisof Mean Price Residual Mean residual
~~~~~~~~(25iz6 juitIS06) .00183 .00105 .00029 .00064 .00096 .00019 - .00047 .00229 .00500 .00204 .00163 .00120 .00109 .00354 - .00040 .00257 .00343
-8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8
TABLE 7 Occurrence ofOtherNews Announcements Week
No ofdividend announcements a
No. of all other typesof
-4 -3 -2 -1 0 1 2 3 4
43 39 42 16 0 16 33 32 41
3 2 4 5 4 4 4 3 2
CONTENT OF ANNUAL EARNINGS ANNOUNCEMENTS
VtX 1O L5 , =2;6 ( /506,wheret =-8,**., +8.
i~~~~l X ~~~~~report a
Ft X 103 during nonperiod= 1.12.
+4 0 +2 -2 Weeksafterannouncement
FIG. 1. Volume Analysis Relative frequency et1 oi8
C /143, t =1,, 261
ssTil= -aa- biVt
.10 .06 1 observation
FIG. 2. Distribution of et in the Nonreport Period
/06 j62j e26)/506,wheret
-Mean j, X 10' duringnon reportperiod= 0.
-2 0 +2 Week afterannouncement
FIG. 3. Residual VolumeAnalysis No. of Vejts
260 no.of positive epys -Expected based on relative frequencyin nonreportperiod.
180 s _
FIG. 4. Frequencyof Positive eit's-Residual VolumeAnalysis
CONTENT OF ANNUAL EARNINGS ANNOUNCEMENTS Relative frequency
1.6 1.2 _ 1.4 Valueof U, FIG. 5. Ut in Nonreport Period 1.0
t= -8, '
-Mean U, during nonreport period= L00.
uji = Rj, - ai -
varianceofresidualin the nonreport period.
1.2 1.1 1.0----?
.90 .80 -8
+2 +4 -2 0 Weekafterannouncement FIG. 6. Price Residual Analysis
no. of Uj,'s > 1.00bas on relative frequencyin nonre period
FIG. 7. Frequencyof Uit's > 1.00-Residual Price Analysis