+ All Categories
Home > Documents > Schroeder 1995

Schroeder 1995

Date post: 03-Jun-2018
Category:
Upload: tasleem-faraz-minhas
View: 238 times
Download: 0 times
Share this document with a friend

of 21

Transcript
  • 8/12/2019 Schroeder 1995

    1/21

    Journal of Business Finance Accoun ting, 22(7), October 1995, 0306-686 X

    EVIDENCE ON NEGATIVE EARNINGS RESPONSEGOEFEICIENTSDOUGLAS A. SCHR OED ER

    INTRODUCTIONThis paper provides empirical evidence on negative earnings responsecoefficients. There is a growing body of literature on cross-sectional andinterte m poral differences in earn ings response coefficients, Th is pa peraddresses an extreme case. Namely, the situation in which earnings responsecoefficients (ERCs) have the opposite sign of that normally predicted andempirically observed (i.e., negative). This extreme case is of interest since itillustrates the potential magnitude of cross-sectional differences in ERCs andmakes a strong case for the importance of informational interactions inunderstanding the relationship between accounting earnings and firm value(Antle, Demski and Ryan, 1995). That is, the paper focuses on situations inwhich accounting earnings reports are unconditionally good (bad) news but,conditionzi] on other information, earnings are bad (good) news.^

    Examples of informational interactions from the extant literature areabun da nt includ ing increases in banks loan loss reserves, information transferof intra-industry earnings announcements and the incremental informationconveyed by cash flows conditional on earnings reports. Unconditionally,increases in a ban k s loan loss reserves are bad news; how ever, there is anabu ndanc e of other information which reflects a ban k s troubled loans in a moretimely fashion than the allowance for loan loss. T hu s, cond itional on this other,more timely information, evidence such as in Beaver, Eger, Ryan and Wolfson(1989) and Liu and Ryan (1995) suggests that increases in loan loss reserves,conditional on nonperforming loans, are good news.Intra-industry information transfers of firms earnings reports may signalgood or bad news conditional on other information (Foster, 1981, and others).For instance, earnings increases for a firm may reflect good news for itscompetitors if the industry has benefited from increased market size. On theother hand, an earnings increase for a firm, conditional on a non-increasingm arke t size for the indu stry, is likely bad news for its com petitors as the firm sgain comes at the expense of its competitors. The author is from the Department of Accounting and MIS, the Ohio State University. Hewishesto thank Vic Bernard, Joel Dennski, Julia Gran t, Jae -O h K im, Bill Kross, Joh n Lyon, PaulNewman, Bill Scott, Katherine Schipper, Tom Stober, David Wallin, two anonymous referees,and especially Jerry Salamon for helpful comments and discussions which helped refine the ideasexpressed in this paper. (Paper received October 1993, revised and accepted MarchAddress for correspondence: Douglas A. Schroeder, Department of Accounting and MIS, TheOhio State University, 1775 College Road, Columbus, OH 43210-1399, USA. Blackwcl] Publishers Lid, 1995, 108 Cowley Road, Oxford OX4 l]F,UK 9 3 9and 238 Main Sirccl, Cambridge, MA 02142, USA.

  • 8/12/2019 Schroeder 1995

    2/21

    940 SCHROEDERWhether more cash flow for a given level of earnings is good or bad newsis also likely to be time period specific (Bernard and Stober, 1989). For instance,increases in cash flows when investment prospects are favorable may be bad

    news since the firm should be investing.These examples illustrate the importance of informational interactionsbetween accounting disclosures and other information in predicting the natureand magnitude ofthe information content in accounting disclosures. This studyemploys a simple, two signal model for identifying when earnings informationis unconditionally good (bad) news but conditionally is bad (good) news (i.e.,negative ERCs). x nte predictions of positive and negative earnings response coefficientsare developed by reference to a simple, stylized two-signal model following

    Holthausen and Verrecchia (1988). In particular, time series variance-covariance estimates of a firm s reported ea rnings per share and Va lue- Lin e searnings per share forecasts are employed to determine these predictions. Simplecorrelations, and multiple regressions (employing experimental control variablesfrom the extant literature) during a hold-out period support the predicteddifferences in association between price changes and unexpected earnings forthe ex nte identified positive and negative groups. Two return intervals areexamined: (i) an interval from the forecast date through the earnings reportdate and (ii) a two-day interval at the earnings announcement date. Aspredicted, tests employing the latter return interval (in conjunction withappro priate experim ental control variables) have greater discriminatory powerfor detecting a negative association between returns and earnings for thenegative group. The results are surprisingly supportive given the limitationsof the classification procedure (e . g., p aram eters are assume d to be firm-specificintertemporal constants, the reference model is a two-signal economy whichunder-specifies the information interactions for the economy in which the testfirms operate).Th e rem aind er ofth e pap er is organ ized as follows. The next section reviewsa stylized two-signal model and discusses its implications. The third sectiondevelops the implications of the two-signal model for the empirical analysisreported in the pape r. T he fourth section describes the data and the partitioningof firms based on time series parameter estimates. The fifth section presentshold-out period correlation, and multiple regression results for the ex anteidentified positive and negative ERC groups. The sixth section discussesimplications of these results for other sequential information issues includingmanagement forecasts, variation in ERC by firm size, and intra-industry

    information transfer. The final section offers a summary and conclusions ofthe paper.

    A TWO SIGNAL M ODEL OF CHANGES IN F IRM VALUEA number of papers (implicitly or explicitly) predict that the sign of the ERCcan be negative (e.g. Holthausen and Verrecchia, 1988; and Lundholm, 1988).

    Blackwdt Publishers Lid. 1995

  • 8/12/2019 Schroeder 1995

    3/21

    EVIDENCE ON NEGATIVE EARNINGS RESPONSE COEFFICIENTS 941Holthausen and Verrecchia (1988) consider sequential signalling of publicinformation, while Lundholm considers simultaneous public and private signals.Though there are some differences in their assumed economies,^ there are nosubstantive differences for purposes ofthe predictions adopted for this study.For simplicity of exposition, this paper adopts the model of Holthausen andVerrecchia (1988). There exists a single, risky asset which pays a liquidatingdividend u. It is known that u has a normal distribution with mean m andvarianc e u. Inform ation about the liquidating dividend is reporte d to allmarket participants at two dates from the time the market opens until liquidationofthe risky asset. This first signal^' communicates the liquidating dividendperturbed by noise I : y = u + e . has a normal distribution with meanzero, variance n', and uand I are uncorrelated . Th e second signal^ ' is alsoimperfect:^ = H + e . Residual uncertainty (noise) for signal two, e , alsohas a normal distribution with mean zero, variance n , and u and areuncorrelated. Thus, (u,y ,y ) has a trivariate normal distribution with mean(m, m, m) and covariance matrix:

    V V VV V + n v + cu u + c V + n

    wherecis the covariance between the noise in the two inform ation signals (i.e .c = Cov[ , ], where Cov refers to the covariance operator).The market prices prior to either signal {p ), after the first signal {p ), andfollowing the second signal {p ^) set by the risk-neutral, competitive marketmaker are as follows:pO = E[a] = m, (1)

    p = t j i u i y = y \ = m + v\u + n ) i v ' / ) t ^ J

    p= E[.|, y\f = / ] = . . p ^3

    The price chamges following the first and second information releases are8and 6^, respectively.6 ' = ^ ' - ^ = v{u +n y^(j>^-Tn), (4)

    2 ^ J2 _ ^1 = - u ( t > - t - O ( n ' - O C v ' - m ) + u(u + n ' ) ( n ' - O O ^ - m )(u-^7i')[(u-(-n')(u-t-n^) - {v + cf]

    Equation (5) can be rewritten in terms of the response to unexpected y*' -, K/^ 'r' , ,. \y - Etf^Ii' . / ) ] (6)

    { {where ^{y ^\y^ = y^) = m + (u-)-f)(i;-* n ' ) - ' ( > ' - m ) . (7) Blackwell Publishers Lid. 1995

  • 8/12/2019 Schroeder 1995

    4/21

    942 SCHROEDERAssume the second signal is earnings, the main thesis of this paper is basedon the sign of tscoefficient. If the noise for the two signals is sufficiently highlycorrelated and the variance of the noise for earnings is greater than that for

    signal one (i.e. c > n ) , then the E RC is negative.^ While this seems to beat odds with empirical evidence which documents a positive ERC, on average,that evidence does not preclude the possibility that some firm-earningsannouncements arepredictably negatively associated with contemporaneous pricechanges.Th e intuition for negative ER C s is that earn ings are not only informativeabout firm v alue but are also informative about the noise in the first signal smapping into firm value. Consequently, earnings are useful for (partially)resolving residual (following the first signal) uncertainty about firm value and

    sometimes the price response is opposite in direction to the sign of theunexpected earnings signal.For any imperfect signal, the sign alling effect is dam pen ed by the noisein the signal. More formally, since^^ = u -t- e^ and price changes are deter-mined by revision in expectationsofuonly, the stock price response to earn ing sfollows from the nature ofthe interaction between earnings and the first signal.Specifically,

    f\f = Cov(w, u\y^)

    Thus, consistent with intuition, the signalling effect is always positive and thenoise effect is negative; however, when the noise effect dominates the signallingeffect {c> n , the revision in expectations of u(and the change in stock price)has the opposite sign to the sign of the unexpected signal.The above discussion describes the interaction between two imperfect signalsgenerally. Now, we explore the intuition for the specific signals employed inthis study, namely analysts earnings forecasts and reported earnings. Inparticular, the discussion now considers the intuition reg arding how interactionbetween these two, largely redundant, signals leads to a negative associationbetween returns and unexpected earnings. Such negative association occurswhen covariance between the residual unce rtainty (noise) i n ^ an d^ ^ exceedsthe variance of noise in the first signal. This statement can be operationallydefined in terms ofthe observable signals. That is, Var(y ) Cov j ,y) =V + n^ v+c) = n^ cand c > rj when Cov y\ y^) > Var y^). Thus ,the question becomes, when is Co v(V L, EPS) > V ar( V L) , whe re V L refersto Value-Line searnings forecasts and EPS refers to reported earnings per share?Or, equivalently, when does the correlation between VL and EPS exceed theratio of the standard deviation of VL to the standard deviation of EPS?

    The important question is when is the above relation expected to be observed?Co nsider that analysts attemp t to identify tu rn in g points in their forecasts.If there is information available which helps analysts in identifying turningpoints, then one may expect to observe negative ERCs. For instance, suppose BlackweU Publishers Lid. 1995

  • 8/12/2019 Schroeder 1995

    5/21

    EVIDENCE ON NEGATIVE EARNINGS RESPONSE COEFFICIENTS 943a firm has recently experienced a slow (boom ) period and information becomesavailable which suggests that the trend is about to reverse (e.g. new geologicalsurvey evidence identifies that extractable reserves for a major oil or gas fieldwere previously significantly under- (over-) stated). This information may ormay not be reflected in transactions which flow into m easurem ent of accountingearnings for the period, but clearly has valuation implications. Further, ifanalysts are aware of such information, they are uncertain whether currentaccounting results will reflect such information. Thus, if analysts provideex nteunb iased forecasts, they will likely assign some non zero probability inanticipation of this new informa tion s impact on earn ings even though theremay be a small likelihood that the information is reflected in the curren t earningsreport. This implies that analysts earnings forecasts are subject to d mpenedshocks relative to repo rted ea rning s, while the time series of analy sts earn ingsforecasts and reported earnings remain highly correlated. These conditionsimply that Cov(VL, EPS) > Var(VL) and lead to predicted negative ERCsfor firms with such characteristics.

    An other example involves m an ag em ent s strategic reporting . Suppose thefirm has experienced tough times such that management will not achieve itsbonu s incentive, then m anag em ent m ay choose to take a big bath (i.e. write-off significant dollar a m ounts of assets) in order to embellish future pe rio d sfinancial reports. However, analysts might not know for certain whetherma nage me nt will take a ba th or the dollar am oun t of assets to be written off.Co nse qu entl y, an aly sts forecasts reflect some likelihood of this event bu t, onaverage, understate the magnitude of the write-down. This again results ina dampened variance of analysts earnings forecasts relative to the varianceof reported earnings while maintaining a high correlation between anlaystsforecasts and earnings.Putting this into perspective so that one does not conclude that a preponder-ance of negative ERCs are expected to occur, recall that turning points (large

    earn ings shocks) are relatively infrequent events. Th us , analysts forecasts andreported earnings may be highly correlated but for the majority of firms theircovariance will not exceed the varia nce in ana lysts earn ings forecasts.

    HYPOTHESIS DEVELOPMENTThe above discussion suggests that current unexpected earnings should havean empirically detectable negative association with stock price changes duringthe earnings report period when the covariance ofthe noise between earningsand an alysts forecasts exceeds the varianc e of the noise in analy sts earn ingsforecasts. T his pred iction require s at least two qualifications. First, observableeconomies involve more than two information signals so that the foregoinganalysis is not strictly representative. Secondly, the parameters are assumedconstant through time in the empirical analysis and this could be a poor BlackweU Publishers Ltd. 1995

  • 8/12/2019 Schroeder 1995

    6/21

    944 SCHROEDERapproximation for some firms. These qualifiers suggest that there is some(perhaps substantial) likelihood of firm misclassification. Nonetheless, theearn ing s response coefficient is predicted tohepositivewhen Cov(VL, EPS) Var(VL).

    These two groups are operationally identified by time series estimation ofthe variances and covariance of observed quarterly earnings per share andanalysts' forecasts of quarterly earnings per share over the period 1973 through1979 {28 quarters). The years 1980 and 1981 (8 quarters) are held out for testpurposes.

    DATA DESCRIPTIONQuarterly earnings per share (EPS) and the latest available one-quarter aheadforecast of EPS (adjusted for stock splits and dividends) from he Value LineInvestment Survey from the first quarter of 1973 through the fourth quarter of1981 for 269 firms were obtained for the analysis. The 269 firm sample meetsthe following criteria:

    1. Th e firm is included in Va lue-L ine con tinuously from 1973 through 1981.2. The firm is a member of a Value-Line industry that contained at leasteight firms.3. The firm is included on Standard and Poors ' COMPUSTAT tapes.4. Daily returns data are available on the current CRSP fUe during the1980-1981 period.5. Monthly returns data are available on the current CRSP file for at least20 months for the period ending September 30, 1979.

    The first three criteria are the same as those employed in Kross, Ro andSchroeder (1990). The fourth criterion was necessary to complete data collectionon returns over the earnings report periods. The last criterion was necessaryfor computing firm's beta (systematic risk) prior to the test period. Incompletedata for 179 firm-quarters reduce the sample size to 1,973 observations duringthe two-year hold-out periods.Eirms are classified intoex antepositive (negative) earnings response coefficientgroups when the variance of the first signal, Value-Line earnings per shareforecasts (V L) , is large r (smaller) than its covariance w ith repo rted earn ingsper share (EPS). Thus, the negative ERC group is identified as follows

    V ar(V L) - Co v(V L, EPS) < 0. (8)These parameters are estimated from each firm's time series over the 1973through 1979 period. The series are assumed to be stationary in either theirfirst difference or seasonal difference. The choice of differencing for both seriesis determ ined based on the smaller estimated variance from the first or seasonal

    Btackw dl Pu blishers Ltd . 1 )95

  • 8/12/2019 Schroeder 1995

    7/21

    EVIDENCE ON NEGATIVE EARNINGS RESPONSE COEFFICIENTS 945difference of the VL earnings forecast series.^ For example, the covariancebetween reported EPS and VL earnings forecasts is estimated as the samplecovariance between (the regular- or seasonal-difference, whichever VL firmvariance is smaller) reported EPS and VL earnings forecasts during theestimation period. T he difference between the variance of Va lue-L ine forecastsand its covariance with EPS is positive for 223 firms (82 .9 ) and n onpositivefor 46 firms (1 7 .1 ). ' Th ese firm groups are employed for the tests ofassociation between earnings and stock returns discussed in the next section.

    An industry profile of the sample broken down by positive/negative groupsis reported in Table 1. Except for a proportionately greater representation infoods and textiles and a lesser represen tation in utilities by negative group firms,there is little difference in industry representation by negative and positive groupfirms.Ta ble 2 reports descriptive statistics on firm charac teristics for the classifiedsample. On average, the negative and positive groups are similar in market

    Table 1Industry Profile of the Sample

    IndustryFoods, TextilesPaper, ChemicaJsPetroleum

    ManufacturingMaterialsTransportat ion,Communicat ionsUtilitiesWholesale, RetailStoresFitiEincial InstitutionsServices

    Four Digit SI100,2000-23002600-28991311, 2911

    3000-38254200-48904911-49405199-5980

    6022-60257372-8911

    Total

    FullSample38(14.1 )33(12.3 )22(8.2 )43(16.0 )28(10.4 )51(19.0 )33(12.3 )18(6.7 )3(1.1 )

    269(100 )

    NumberofFirmsNegativeGroup

    11(23.9 )5(10.9 )4(8.7 )7(15.2 )8(17.4 )4(8.7 )3(6.5 )4(8.7 )0(0.0 )

    46( 1 7 . 1 )

    PositiveGroup27(12.1 )28(12.6 )18(8.1 )36(16.1 )20(9.0 )47(21.1 )30(13.5 )14(6.3 )

    3(1.3 )223(82.9 )

    Blackweli Publishers Lid. 1995

  • 8/12/2019 Schroeder 1995

    8/21

    946 SCHROEDERTable 2

    Sample Characteristics

    VariableSample size

    D / - f\\ 1f l [ 2 , V 1

    lEFEl/Price,

    UE|/Price,

    / - . ,0)1|EFE|/Price _j

    |UE|/Price_2

    Beta

    M arket value common equityWSJI

    Forecast lead time(in calendar days)Var

  • 8/12/2019 Schroeder 1995

    9/21

    EVIDENCE ON NEGATIVE EARNINGS RESPONSE COEFFIGIENTS 94 7

    Notei:|flj( 1 0)1 is the absolute value of the sum of firm ^ 's two daily returns arou nd the earn ingsdate (day 0). ' ',(2> 0)1 = U ^( 0~ 2) n( /y /P j)| = the absolute value of firm j ' s average daily continuouslycompounded return over the forecast horizon (i.e, from the forecast date ^ to the earningsannouncem ent d ale day 0).EPS is reported EPS before extraordinary items and discontinued operations (adjusted for sLocksplits and dividends, etc).VL is Value-Line's forecast of EPS,|EFE| is absolute value of EPS less VL.|UE| is absolute value of EPS less E[EPS|VL], where

    E [ E P S |V L ] = E P S , _ , + G o v ( E P S ^ - E P S ^ _ i , V L , - V L , _ , 1 *

    i f V a r [ V L , - V L , ^ , ] S V a r | V L ^ - V L , _ , ] . orE[EPS|VL1 = EPS^_, + C

    Price^ is stock price on day dBeta is the firm's systematic risk (estimated from approximately 60 monthly returns in advanceof the test period).Mark et v alue comm on equity is the firm's end of qu arte r stock price times the nu mb er ofoutstanding common shares (in millions) during the test period,WSJI is the number of column inches of coverage on the firm in the annual WallStreet Journalndex during ihc test period.Forecast lead time is the number of calendar days between the release of the Vaiue-Line earningsforecasts and the earnings report date during the test period.Var(VL) is the variance of (the regular- or seasonal-difTcrence, whichever VL firm variance issmaller) Value-Line's EPS forecast during the estimation period,Var ( reported EPS) is the variance of (the regular- or seasonal-difference, whichever V L firm varianceis smaller) reported EPS during the estimation period,Co v(V L, E PS) is the covariance between (the regular- or seasonal-difference, whichever VL firmvariance is smaller) Value-Line's EPS forecast and reported EPS during the estimation period.

    value of common equity, systematic risk and reported and forecasted EPS scaledby stock price. The negative group, on average, has a greater return variabilitydu ring the earnings announcem ent period in conjunction with greater variabilityin unexpected earnings, somewhat greater financial press coverage and moretimely earnings forecasts than the positive group. (Timely forecasts mayfacilitate the prediction of turning points by analysts.) By construction, thecovariance between reported and forecasted EPS is greater (less) than thevariance of forecasted EPS for the negative (positive) group. The variance ofreported E PS and the covariance between reported and forecasted EPS ap pearto be much larger for the negative than the positive group, while the varianceof forecasted EPS appears to be similar for the two groups (however, thesevariables are not scaled for cross-sectional comparability). Blackwdl Publishers Ltd- 1995

  • 8/12/2019 Schroeder 1995

    10/21

    948 SGHROEDERANALYSIS AND RESULTS

    This section examines whether a difference in the association between unex-pected ea rnings with stock price changes is empirically de tectable for the positiveand negative groups. Simple correlations and regressions of returns with twomeasures of earnings surprise are examined. Th e first earnings surprise m easureis the earnings forecast error (EFE = EPS VL) and provides comparabilitywith the extant litera ture . The second earnings surprise measu re is draw n fromequation (7). Unexpected earnings (UE) is defined to be reported EPS lessE [E P S | VL ] , i.e. expected earnings conditional on the Value Line earningsforecast based on the time series variance-covariance estimates employed toclassify positive and negative ER C firms. Since price chan ges are deflatedby initial price to determ ine ret ur ns , ^ the earn ings forecast erro r is alsodeflated by price at the beginning of the return interval.

    Two return intervals are examined: (i) the price change from the time ofthe V L forecast throug h the earning s report da te (the day reported in theWallStreetJournal ,and (ii) the sum of the two daily retu rns d urin g the two-dayanno unc em ent period (day before and the day reported in the WallStreetJournal .The former corresponds with the model in the section headed HypothesisDevelopment in that the price change is aligned with the timing of disclosurefor the two signals. However, the model assumes that the second signalinstantaneously follows the first signal so that issues of discounting, changesin equilibrium returns and, importantly, other information do not alterinvestors perceptions. Since the median num ber of calendar days between theVL forecast date and the earnings report date is 40 days, it is very likely thatprice changes du ring this inter\ al are contam inated by other information eventsso that any differential E R C tests are likely to have low pow er. T he choiceof the second (shorter) return interval is expected to mitigate this omittedvariables problem.

    Simple Co rrelationsTable 3 reports simple (Pearson-product moment) correlations between returnsand earnings surprise as well as the information primitives (reported EP S, V Lforecasts and E[EPS|VL]) for the two groups. Two return measures areexamined for each return interval: (i) raw returns, and (ii) market-adjustedret urn s, that is, the firm s raw re turn less the contem poran eous return on ama rket index. Th e two retu rn m etrics yield similar correlation s for a givenreturn interval. Differences in correlations across return intervals are asexpected. The forecast-to-earnings report date return interval demonstratesdampened (toward zero) correlations relative to the two-day earnings dateinte rva l. As pred icted , there are very substan tial differences between thepositive and negative groups correlations for all return measures and intervalsand both measures of earnings surprise. For example, the correlations for the

    Blackwetl Publishers Ltd. 1995

  • 8/12/2019 Schroeder 1995

    11/21

    Table 3 Association Between Returns and Unexpected Earnings by Positive and

    Negative GroupsPearson Product-Moment Correlations (p-values)

    Negative Group (344Return Metric

    Rj{z, 0) - R^\

    ; ( - . 0)^ , ( -1 .0) - JPositive GroupReturn AlelrifR,{z, 0)

    RM 0 - R^^/ * / - . 0)^ , ( - 1 . 0 ) - ^

    (z , 0)

    ?. - ,

    (1,629

    :^.0)

    observations)EEE/P^

    - 0 . 0 0 0 4 5(0.9933)- 0 . 0 1 6 9 0(0.7548)- 0 . 0 9 6 2 3(0.0747)

    0) -0 .13068(0.0153)observations)

    EFE/P^0.09032(0.0003)0.09919(0.0001)0.14051(O.OOOI)

    0) 0.14820(0.0001)

    UE/P,- 0 . 0 0 6 6 8(0.9018)- 0 . 0 1 4 0 4(0.7953)- 0 . 0 3 1 9 6(0.5547)- 0 . 0 7 1 3 3(0.1869)

    UE/P^0.04617(0.0624)0.05998(0.0155)0.18796(0.0001)0.19496(0.0001)

    EPS/P,0

    (00,(0,

    .04098.4487)

    .03942.4661)- 0 . 0 1 3 9 5(0.7966)- 0(0, 02439,6521)

    EPS/P,0,

    (0.0.(0 .

    .14885.0001)

    .16087.000 )0.15278(0.0001)0.16553(0.0001)

    VL/P,0.06221(0.2499)0.07981(0.1396)0.09907(0.0665)0.12623(0.0192)

    VL/P,0.12099(0.0001)0.12897(0.0001)0.07643(0.0020)0.08739(0.0004)

    EE/P^0.03462(0.5223)0.04119(0.4463)0.02442(0.6518)0.05904(0.2749)

    EE/P,0.13759(0.0001)0.13606(0.0001)

    - 0 . 0 1 7 8 6(0.4714)- 0 . 0 0 9 4 4(0.7034)

    Rj{z.O) = 1/(0i)ln(F(/^,) = firmj's average daily continuously compotjndcd return over theforecast horizon (i.e. from theforecast dat e zto theearnings announcement date day 0) .Rj{z, 0) - R^{z,0) = l / (O-r) [ln(P^.o/P^,,) - ^n{hp, Jhp.z)] here Isp_,= the S&P 500 Indexlevel on dayz (relative toearnings annou ncem ent date equal to dayzero)./?.( , 0) is thesum of firm; ' s two daily returns around che earnings date (day0) .R^{ \. 0) isthe sum ofthe two daily retur ns onCR SP 's value-weighted market index aroundthe earnings date (day d .EFE = EPS VL - unexpected earnings, reported earnings per share (EPS) less Value-Line'searnings pershare forecast (V L).UE isEPS less E[E PS|V L], where

    E[EPS|VL| E P S , . , Cov[EPS^-EPS,_, , VL , -V L,_ , * VarJVL,-VL,_4].or

    Cov[EPS,-EPS,^ , , ^_^) * VarfVL,-

    EE = EIEPSIVL] (defined above).F,isthe firm's stock price ondayz{z 2orV L forecast d ate forthe two-day or forecast-to-eamings date return interval, respectively). Blackwell Publishers Ltd. 1995

  • 8/12/2019 Schroeder 1995

    12/21

    950 SCHROEDERpositive group are 0.141 and 0.148 (both with /^-values less than 0.0001) fortwo-day observed returns and market-adjusted returns with earnings forecasterror (deflated by price), respectively. On the other hand, the correlations forthe negative group are 0.096 and 0.131 (with two -tailedp-v alues of 0.0747and 0.0153) for two-day observed returns a nd market-adjusted retu rns , respec-tively. These differences between the ex nte identified positive and negativegroups are quite striking; however, they do not control for other factors withdocu me nted links to cross-sectional variation in ER C s. T his is considered next.Regression AnalysesThe extant literature documents cross-sectional variation in ERCs as a functionof various firm characteristics. For instance, Easton and Zmijewski (1989) arguethat E R C s are inversely related to a firm s systematic risk. Freem an (1987)and Ro (1989) provide evidence that ERCs are inversely related to firm size.Kross and Schroeder (1989) provide evidence that ERCs are inversely relatedto a firm s financial press coverage. Th ese charac teristics m ight offer analternative explanation to the posited dampening of ERCs for the negativegroup. That is, if the negative group firms are partitioned such that it isdisproportionately represented by risky, large firms which receive substantialfinancial press coverage, then this might explain the dampened correlationsobserved in Table 3 for the negative group relative to those for the positivegro up. Accordingly, experimental controls for beta , firm size and presscoverage are incorporated into the next set of tests via regression analysis.In particular, since the documented effects associated with these firm charac-teristics are most prominent for low beta firms, small firms and low financialpress coverage firms, an interaction between earnings surprise (EARNSUP,)and a dummy variable denoting the smallest third of the sample for each ofthese characteristics is incorporated in the regression.

    Rjt = 7o + 7iEARNSUP, + 72EARNSUP, * NEG +73EARNSUP, * LOWBETA + 74EARNSUP, SM LL +75EARNSUP, * LOWWSJI + 76^ . , + ^t (9 )

    Rjt returns for firm^ either measured over the interval from theforecast-to-earnings report date or over the two-day earningsreport period,EA RN SU P, = earnin gs surprise measured as either the V L forecast error orEPS less E[EPS|VL] both deflated by opening price (theforecast date price or price on day depending on the returninterval),N EG = one if the firm is classified as a negative gro up firm, and zerootherwise,LO W BE TA = one if the firm is in the lowest third of sample fir m s be ta, andzero otherwise,SM A LL = one if the firm is in the lowest third of firm size for the sam ple,and zero otherwise,

    BlackweU Publishers Ltd. 1995

  • 8/12/2019 Schroeder 1995

    13/21

    EVIDENCEONNEG TIVE E RNINGS RESPONSE COEFFICIENTS 951LOWWSJI = one ifthe firmis in thelowest thirdof financial press coverage

    (measured as column inchesin theannual Wall Street JournalIndex in the sample, and zero otherwise,

    Rmi = return on the market index contemporaneous with the firm'sreturn interval (either the return on the S P 500 Index overthe forecast-to-earnings report date or the return on CRSP'svalue-weighted index over the two-day earnings report period),

    ^ = a disturbance term.Table 4 reports results of regression analyses for four combinations of return

    intervals (forecast-to-earnings report date, or two-day earnings report dateinterval) and earnings surprise (VL earnings forecast error, or reported EPSless E[EPS|VL]). These results are very similar to the correlation resultsreported in Table 3. There is a substantial difference in ERCs between thepositive and negative groups after controlling for other cross-sectional influenceson ERCs. The difference is statistically significant in all four panels. ^ PanelsA and C report the forecast-to-earnings date interval regressed on the VLearnings forecast error and unexpected earnings conditional on the VL forecast,respectively. As expected, the differences between the negative and positivegroups' ERCs are dampened (but still significant) relative to the two-day returnintervals reported in Panels B and D. Further, the ERC is significantly negativefor the two-day return interval for both the VL earnings forecast error andunexpected earnings conditional on VL earnings surprise measures reportedin Panels B and D, respectively. For instance, the ERC on the negative groupreported in Panel B is -0.1 68(0 .0 35 - 0.203) which has ap-value of 0.0001(reported at the bottom of the panel as Negative ERC test).

    In summary, the regression analyses provide strong evidence that theex antepredicted negative ERC firms have smaller ERCs thanex antepredicted positiveERC firms after controlling for firm characteristics associated with cross-sectional variation in ERC and contemporaneous returns on a market index.It should be noted that low beta and low financial press coverage firms havesignificantly higher ERCs than their counterparts, but no difference is detectedfor small firms relative to medium and large firms. This latter result may bedue to Value-Linens propensity to cover larger firms. As expected, the powerof these tests appears to be strongest for the two-day return interval since theforecast-to-earnings date return interval is more likely contaminated by otherinformation events. Indeed, the two-day return interval results indicate asignificantly negative ERC for firms in the negative group.

    DISCUSSION

    Implications For Other Information EnvironmentsIn addition to providing empirical evidence on negative earnings responsecoefficients, these data also provide indirect evidence on other issues involving Blackwell Publishers Ltd. 1995

  • 8/12/2019 Schroeder 1995

    14/21

    952 SCHROEDER

    aVoa

    -9

    ^ 77 -^

    ra"D6VC

    t ccrau

    co

    uu(HV

    uuco?c

    E c c Eo I*3^ s

    o cII

    Vto ^CcomybnVPi

    5u>otun

    u

    isevd

    JJooare

    aOL^in

    re

    60C1VSiXI- QVse

    :?

    amin

    foe

    Co

    -J

    crau

    -;o

    ic

    13 .Ucc3

    0 r-.. O CN CO en o01 : O CO OO O4 o O i~- O OM c-i O O in O Oo d o d d o o

    O * O (N (N O OO IM O O eo O OOJ O O O c^ O Od d d d d d drt I-H O tN O lOin 1,0o * o Cl OO n r - i n - ^ c o in-*-0 0 4 ^ i n eoO c o o C - j r - c o oO O I J O )

    o- ; [T , fT ,fS W W

    cfl hJ _:

    w

    TI.0*1.0TIO Oi O

  • 8/12/2019 Schroeder 1995

    15/21

    EVIDENCE ON NEGATIVE EARNINGS RESPONSE COEFFICIENTS 953 j - r^.. CO CO OO eo Th in OO o CM OO O 1.0 O O

    O CJ o c ) O Oo o - o od d d d d

    O r~- tx Oin o CO o oo o o o od d d d d

    Th Tf eo i~- COO O O O lO O o o Oid d d < > to

    CO

    Q

    a = bo

    ts1

    / s .

    6 S

    E >O ;S W

    1 ^ CL,

    ]>-> 1

    1

    1

    >ou+

    1OHun

    -

    o > 1

    1

    1>

    1a.1

    a.W_0O+

    1tCLI

    '

    > ^

    0bC

    IDERleJJu

    -cu3uio3

    132Q .

    nu

    1

    >1

    ECl .V )

    'EnUtm

    Q

    b

    Miv

    VIVc

    O

    10

    zo

    2E

    11

    I

    E ';;IS C

    >S U UJ

    >1

    1

    'SeaII

    ObjZ

    i2) -i- Cov ji\ f) - Var(>;')] - CoYiy\ ff > 0. (13)The time series estimates indicate that this difference in (13) is positive for 231firms(85.9 of the sample) and it is nonpositive for 38 firms (14 .1 ). Thus ,this is consistent with Pownall and Waymire's (1989) finding that the associationbetween management's earnings forecasts and abnormal returns during theforecast announcement period is higher than the association between subsequentearnings reports and abnormal returns during the earnings report period.

    Similarly, the analysis predicts that the response coefficient for earnings (atthe time of the earnings report) not preceded by other valuation-relevantinformation disclosures (informationally-sparse economies) is greater than theresponse coefficient for earnings preceded by other valuation-relevant informa-tion (informationally-rich economies). Expressed in terms of observables,this is

    > CQ^f{y\f > 0 (14)wherej refers to earnings andj) is a prior signal in the informationally-richeconomy. Based on the same time series estimates as above, this relation issupported for 263 firms (97.8 ) and it is not supported for 6 firms (2 .2 ).Thus, this evidence is consistent with the relation between ERCs and firm sizeor fmancial press coverage documented in the literature (for example. Freeman,1987; Ro, 1989; and Kross and Schroeder, 1989).

    mplicationsfor nformation TransferThe empirical support for the predictions of the stylized, two-signal model,suggests an ex anteapproach for the study of information transfer. Frost

    Blackwell Publishers Ltd. 1995

  • 8/12/2019 Schroeder 1995

    17/21

    EVIDENCE ON NEGATIVE EARNINGS RESPONSE COEFFICIENTS 95 5(1989) argues that some studies overstate the significance of informationtransfers (e.g. Foster, 1981; Baginski, 1987; and Clinch and Sinclair, 1987)and other studies understate its significance (e.g. Han, Wild and Ramesh, 1989)by the nature of their controls for cross-sectional correlation and returnsimu ltaneity . F rost, also, finds sparse evidence of significant inform ationtransfers. How ever, as An tle, Demski and Rya n suggest, this may bebecause earnings reports sometimes signal industry-wide changes (circumstancesrepres entative of positive corre lations between firms future cash flows) andother times earnings reports signal strategic gains or losses by one firm (or groupof firms) at the expense of other firms (circum stances represen tative of negativecorrelation between firms future cash flows).^^ T he sign of the inform ationtransfer coefficient depends on this as well as the interaction of ecurnings withother signals as suggested in the foregoing analysis.A similar approach to that taken in this paper may help to identify ex ntecircumstances in which a positive or negative information transfer occurs. Forinstance, consider a two-asset, two-signal economy in which two signals areobserved in sequence for one firm (as discussed in the second section above).The change in price for firm two (the non-announcer) is very similar to thatfor firm one (the ann oun cer) given in equation s (4) and (5) wh ere is someother information andy is the ann oun cer s earnings report

    (y-i-n )[(i; n )(u + n^) - (u + O^lwhere y is the covariance of firm one s future cash flows with firm tw o sfuture cash flows.

    Therefore, the implications for information transfer are very similar to thosefor earnings response coefficients as indicated in this paper, except that it alsodepends on the sign (and m agnitude) of the covariance between the firms futurecash flows u . T ha t is, when the covariance i; is positive, the inform ationtransfer coefficient (the second term in the numerator of (15)) is positive(negative) for n > c(n < c), O n the other ha nd , when the covariance u is negative, the In formation transfe r coefficient is nega tive (positive) for n > {n S c). Such an approach allows for cross-sectional pooling of observationswhich may increase the power of information transfer tests.

    CONCLUSIONSThe empirical analysis in this paper is developed from the implications ofsequential signalling such as in the stylized models of Holthausen andVerrecchia (1988) and others. The empirical evidence provides strong evidenceof cross-sectional variation in earnings response coefficients. That is, while somegrou ps of firms have earnings me asure s which are significantly, positivelyassociated with chang es in stock price, other firms earn ings measure s havepredictably lower and even negative associations with stock price changes during

    Blackweli Publishers Ltd. 1995

  • 8/12/2019 Schroeder 1995

    18/21

    956 SCHROEDERthe earnings report period. The evidence is remarkably robust across simplecorrelation analyses, and regression analyses.

    These results provide empirical support for the importance of complexinformational interactions (Antle, Demski and Ryan, 1992; and Garman andOhlson, 1980) when evaluating the information content of accounting disclo-sures. This is a very difficult empirical issue since these interactions are likelyto vary across firms and intertemporally. This study treats informationalinteractions as an intertemporal constant for a given firm, an assumption whichis almost surely violated. Fu ture work toward relaxing such restrictions arelikely to help further untangle these complex informational interactions andimprove ou r understan ding of the information content of accounting disclosures.One promising avenue for such future work would be to more fully exploitthe impact of accounting structure for determ ining informational interactions.

    N O T E S1 This statement on earnings response coefficients refers to earnings-returns association basedon return cumulaiion periods which (i) coincide with both ihe earnings measurement periodand report period (usually a quaner or year), and (ii) coincide with only the earnings reportor event period (usually two or three days). For clarity of exposition in this paper, the formercase is described as earn ings association coefficients (e.g. K orm endi and Lipe, 1987; Fre em an,1987; Collins and Kothari, 1989; Lipe, 1990; and Ohlson, 1991 and 1995) and the latter isdescribed as earn ing s response coefficients (e.g. Easton and Zmijewski, 1989; Kross andSchroeder, 1989; Ro, 1989; and K ross and Schroeder, 1990). Both cases examine the m appingbetween (unexpected) earnings and changes in stock prices. However, the empirical analysisand results reported in (his paper are restricted to the latter event period mapping.2 Accounting earnings reports convey good (bad) news when investors upwardly (downwardly)revise their beliefs reg ardin g a firm s future cash flows as reflected in the chang e in a firm sstock price.3 Lundholm (1988) assumes risk averse investors in a noisy rational expectations equilibrium,while Holthausen and Verrecchia (1988) assume a risk neutral market maker; both economiesare competitive.4 Ho lthausen and V errecchia also discuss a multi-asset, mu lti-signal economy in the latter partof their paper. While this is largely beyond the scope of this paper, preliminary analysisincorporating signals on other assets did not substantively alter the results reported in this paper.5 Note that the sign ofth coefficient in (6) is determ ined by the num erato r, since the de nom inatorin (6) is always nonnegative and only equal to zero when the correlation between the noiseterms equals one and their variances are equal.6 An extension to this work might consider relaxing the intertemporal stability assumptionregarding the variance-covariance parameters via Bayesian inference, or perhaps by using ARCH(autoregressive conditionally-heteroskedastic) models. This is left to future work.7 Missing observations on Value-Line forecasts resulted in the elimination of the quaner forthe firm, 244 firms had 36 usable quarters, 15 firms had 35 usable quarters, and 10 had 34usable quarters; a total of 9,649 quarterly observations were employed,8 Earn ings report dates for 35 firm-q uarters are unava ilable in theWallStreet ournalIndex priceson the VL forecast date are missing for two firm-quarters, and the forecast date is concurrentwith or later than the earnings report date for 142 firm-quarters. The latter are eliminatedto accommodate return measurement from the forecast date to the earnings report date.9 The motivation for this approach is threefold. First, seasonaiity in EPS is not homogeneousacross firms (Lorek and Bathke, 1984) and identifying the nature of any nonstationarity isexpected to increase the efficiency of classification. Second, the focus is on VL forecasts ratherthan reported EPS to determine the nature of the nonstationarity, since VL forecasts have sm aller

    Blatkwrll Publishers Ud. 1995

  • 8/12/2019 Schroeder 1995

    19/21

    EVIDENCE ON NEGATIVE EARNING S RESPONSE COEF FICIENT S 95 7variance on average than reported earnings and are less likely to be influenced by extremeobservations. Third, comparing variance-covariance estimates from assumed processes for whichsecond moments may not exist (say, the levels of the two series) is likely to lead to more frequentmisclassification. Indeed, assuming that the model predictions under the research hypothesishold, estim ates from th e non-differenced series result in frequent misdassifications of firms(relative to the difFerenced series estimates). Other approaches such as strictly applying regularor seasonal differencing on VL forecasts and reported earnings yield qualitatively similar resultsbut, as expected, suffer some loss in efficiency.

    10 Th e 1,973 firm-qua rter observations are divided into 344 (17 .4 ) firm -quar ters in the negativegroup and 1.629 (82 .6 ) flrm-quarters in the positive grou p.11 Recall that the choice of reg ular or seasonal difTerencing is dete rm ined by the differenced V Lseries with smaller variance, so that (7) becomesE[EPS1VL1 = EPS,_, + Cov[EPS^-EPS,^i, VL^-VL^_,]* Var[VL^-VL,_,] - ' * (VL,-VL^_,] )

    i fVar IV L-V L,_ , l Var (VL,-VL,_ , l . or

    12 The analysis was also conducted on returns cum dividends. The results are very similar tothose reported.13 The forecast date is the library receipt date; if the stamp is not available, the issue date is usedas the forecast date (there is typically no more than a day difference between the two dates).14 The market index employed for the two-day return interval is CRSP's value-weighted indexand the market index employed for the forecast-to-earnings date return interval is the S&P500 Index. The negative ERC results are not sensitive to the choice of market index or to itsinclusion.15 The forecast-to-earnings report date returns are measured as the geometric (daily) mean toequalize weights across firms and facilitate this (predominantly) cross-sectional analysis.16 Since some examples of informational interactions seem to be driven by timeliness of accountingdisclosures (e.g. increases in banks' loan loss reserves), the impact of the timeliness of reportedearnings relative to VL forecasts was also investigated. An interaction between forecast leadtime and earnings surprise was added to equation (9); the coefficient on the additional variableis significantly negative and the negative ERC results are very similar to those reported.

    17 Inclusion of additional dummy variable interaction terms with earnings surprise for the topone-third for each of the variables was also investigated. The results on negative ERCs is verysimilar. Since the extant literature suggests that the resuhs are most sensitive for the bottomone-third (and this is corroborated by these data), only the results employing the bottomone-third are reported (in the spirit of parsimony).

    18 Standard (two-tailed) /^-values are reported along with /j-values accommodating conditionalheteroskedasticity from Wh ite (1980). Wh ite's model specification test is repo rted at the bottomof each panel and suggests that conditional heteroskedasticity is not a serious problem for theseanalyses.19 Of course, it is presumptuous to assume that the time series behaviour of observed variableswould be the same if ihe information environment were so altered.20 Inform ation transfer refers to the signalling effects ofon firm's earnings report on other (usuallywithin the same industry or having a customer/supplier relationship) firms' future dividends(e.g. Firth, 1976; Foster, 1 981; and Oisen and Dietrich, 1985),21 This discussion appeared in the 1989 version of the paper.22 Frost allows for this possibility but her ex perim ental design may not be sufficiently powerfulto detect information transfers.23 One may be disillusioned by the small improvement in explanatory power afforded by thepositive/negative ERC classification. However, one should bear in mind that intertemporalstability assumptions are likely to severely limit gains in explanatory power. Nonetheless, thecrude methods employed here re able to (partially) separate firms with positive/negative ERC.Thus, there is potential that more refined methods which accommodate differences in ERCsacross firms and over time m ay increase, p erhaps gread y, the exp lanatory power of empirical tests.

    Blackwell Publishers Lid. 1995

  • 8/12/2019 Schroeder 1995

    20/21

    958 SCHROEDERR E F E R E N C E S

    Antle, R., J.S . Demski and S.G. R yan (1995), 'Multiple Sources oflnfo rm ation, Valuation andAccounting Earnings',Journal ofAccounting Auditing Finance (Fall), pp. 675696.Baginski, S.P. (1987), 'Intra-Industry Information Transfers Associated with Management Forecastsof Earnings',Journal ofAccounting Research, Vol. 25, No. 2 (Autumn), pp. 196216.Beaver, W., C. Eger, S. Ryan and M. Wolfson (1989), 'Financial Reporting, SupplementalDisclosures and Bank Share Prices',Journalo fAccounting Research, Vol. 27, No. 2 (Autum n),pp. 157-178 .Bernard, V. and T. Stobcr (1989), 'The Nature and Amount oflnformation in Cash Flows andAccruals ' , TTie Accounting Revietv, Vol. 64, No. 4 (October), pp. 624652.Clinch, G.J. and N.A. Sinclair (1987), 'Intra-Industry Information Releases: A Recursive SystemsApproach' ,Journal ofAccounting Economics. Vol. 9, N o. 1 (April), p p. 89106.Collins, D.W. and S.P. Kothari (1989), 'An Analysis of Intertemporal and Cross-sectionalDeterm inants of Earnings ResponseCoeffiCKTits JoumtUofAccounting Economics,V ol. 11,Nos 2/3 (July), pp. 143-182.Easton, P.D. and M.E. Zmijewski (1989), 'Cross-sectional Variation in the Stock Market Responseto Accou nting Earnings An nou ncem ents', younta/q/'/lourtii'nf Economics, Vol. 11, Nos 2/3(July), pp. 117-142.Firth, M. (1976), 'The Impact of Earnings Announcements on the Share Price Behavior of SimilarType Firms ' , EconomicJournal Vol. 62, No. 2 (June), pp. 29 6-3 06.Foster, G. (1981), 'Intra-Industry Information Transfer Associated With Earnings Releases',_/ourWof Accounting Economics, V ol. 3, No . 3 (December), pp. 20 2-2 32 .Freeman, R.N. (1987), 'The Association Between Accounting Earnings and Security Returns forLarge and Small Firms',JournalofAccounling Economics,Vo l. 9, N o. 2 (July), p p. 195228.Frost, C.A. (1989), 'Intra-In dustry Information Transfer: An Analysis of Research Methods andAdditional Evidence', Working Paper (University of Michigan).G arm an , M . and J. O hlson (1980), 'Information and Sequential Valuation ofAssetsin Arbitrage-Free Economies',Journal ofAccounting Research, Vol. 18, No. 2 (Au tum n), pp. 420440.Ha n, J.C .Y ., J.J . W ild and K. Ramesh (1989), 'Man age rs' Earnings Forecasts and Intra-IndustryInformation Transfers',youmu/o/'i^ccounft'nf G?, onomtcj, Vol. 11, No. (February), pp. 334.Holthausen, R.W. and R.E. Verrecchia (1988), 'The Effect of Sequential Information Releaseson the Variance of Price Changes in an Intertemporal Multi-Asset^AajkeC,Journalof AccountingResearch. Vol. 26, No. 1 (Spring), pp. 82 -1 06 .Korm endi, R .C . and R . Lipe (1987), 'Earnings Innovations, Earnings Persistence and Stock Prices',Journal of Business. Vol. 60, No. 3 Quly), pp. 323-346.Kross, W., B. Ro and D A . S chroeder (1990), 'Earnin gs Expectations: Th e Analysts' InformationAdvantage' , The Accounting Review,Vol. 65, N o. 2 (April), pp. 461476.and D A . Schroeder (1989), 'Firm Prom inence and the Differential Information Con tentof Quarterly Earnings Anno unce men ts' , youm a/o/BiuinfiJ, Finance Accounting Vol. 16,No. 1 (Spr ing), pp . 5574.(1990), 'An Investigation of Seasonality in Stock Price Responses to QuarterlyEarnings Announcements',yoKma/o/BujjKMj Finance Accounting Vol. 17, No. 5 (Winter),pp. 649-676 .Lipe, R. (1990), 'The Relation Between Stock Retu rns and A ccounting Earnings Given A lternativeInformation' , Th eAccounting Revietv,Vol. 65, No. 1 (Janu ary), pp. 4971.LJu, C. and S. Ryan (1995), 'Th e Effect of Bank Loan Portfolio Com position on the Mark et Reactionto and Anticipation of Provisions for Loan Losses',Journal ofAccounling Research (Spring),pp. 7 7 - 9 4 .Lorek, K .S. and A.W . Bathke (1984), 'A Time-Series Analysis of Nonseasonal Qua rterly EarningsData ' ,youma/ ofAccounting Research, Vol. 22. No. 1 (Spring), pp. 369 -37 9.Lundholm, R.J. (1988), 'Price-Signal Relations in the Presence of Correlated Public and PrivateInformation' , Contemporary Accounting Research. V ol. 26, No . 1 (Sp ring), pp. 107118.Ohlson, J. (1991), 'The Theory of Value and Earnings and an Introduction to the BallBrownAnalysis ' , Contemporary Accounting Research, V ol. 8, No . 1 (Fall), pp. 19.(1995), 'Earnings, Book Values and Dividends in Security Valuation',Contemporary AccotmtingResearch (Spring).

    Blackwell Publishers Ltd. 1995

  • 8/12/2019 Schroeder 1995

    21/21

    EVIDENCE ON NEGATIVE EARNINGS RESPONSE COEF FICIENT S 9 5 9Ols en, C. and R. Dietrich (1985), Vertica l Informa tion Tran sfers: Th e Association BetweenRetailers Sales Announcements and Suppliers Security K etums , JotiriuUof ccountingResearch,Vol. 23 (Supplement), pp. 144-166.Pow nall, G. and G. W aym ire (1989), Vo lun tary D isclosure Credibility and Security Prices:

    Evidence from Managem ent Earnings Forecasts ,youma/ of ccountingResearch, Vol. 27, No. 2(Autumn), pp. 227-245.R o, B. (1989), Ea rnin gs News and the Firm Size Effect , Contemporary ccountingResearch, Vol. 6,No. 1 (Fall), pp. 177 -19 5.W hite, H. (1980), A Heteroskedasticity-Consistent C ovarian ce Matrix Estimato r and a DirectTest for H eterosk edasticity , Econometrica, Vol. 48, No. 4 (May), pp. 817-838.


Recommended