THE ACCOUNTING REVIEW Vol. 65, No. 1 January 1990 pp.

175-190

The Market Interpretation of Management Earnings Forecasts as a Predictor of Subsequent Financial Analyst Forecast Revision
Stephen P. Baginski
Florida State University

John M. Hcissell
University of Texas at Arlington
ABSTRACT: This study investigates the reiation between financial analyst earnings forecast revisions and two independent variables: ( D a measure of management earnings forecast news issued prior to analyst revisions, and (2) measures derived from the security market price reaction to that news. Results indicate that security price reactions to management forecasts are useful in predicting subsequent analyst forecast revisions. Furthermore, the explanatory power of price reaction is a function of the timing of the management forecast release.

ECENT studies (Beaver et al. 1980; Beaver et al. 1987; Collins et al. 1987) document that the accounting earnings and security price relation may be inverted so that security prices may be used to forecast earnings. Such a relation is possible if, for example, annual earnings releases convey infonnation that the securities market receives earlier from other sources. In addition. Brown et al. (1985) document that security prices contain information related to Einalysts' earnings forecast errors and subsequent revisions. Recent results also support the information content of security price changes occurring at the dates of
Helpful comments were received from Bob Jennings. Ken Lorek. workshop participants at the Florida State University, and an anonymous referee. We thank Zacks Investment Research, Inc., for providing a portion of the data for this study. Manuscrtpt received March 1988. Revisions received June 1988. December 1988. and May 1989. Accepted July 1989.

175

176

The Accounting Review, January 1990

voluntary disclosures. For example, Jennings' (1987) results suggest that security price changes and subsequent analyst forecast revisions around the dates of voluntary management earnings forecasts are associated. McNichols (1989) documents an association of security price responses to management forecasts and subsequent earnings forecast errors derived from management forecasts and actual earnings. This study examines the information in security price reactions to management forecast release conditioned by disclosure timing. Recent research finds timing to be a disclosure characteristic that affects the traditional unexpected earnings and unexpected price change relation, especially for bad news. Bathke and Lorek (1984) document positive abnormal returns for unexpected negative changes in earnings derived from time-series models and fourth quarter earnings releases. Mendenhall and Nichols (1988) also report a much weaker security price reaction to fourth quarter bad news, especially when measured relative to security analysts' forecasts. Two sets of tests are performed. The first set, closely related to Jennings (1987), provides a starting point for analysis of the timing issue.* These tests examine whether security price chemges reflect information about management earnings forecasts that explain financial analysts' subsequent earnings forecast revisions. This is possible if analysts observe and use price changes or use the same information underlying price changes to forecast earnings. This study does not distinguish between these alternative interpretations.* The second set of tests examines whether the pricing implications of management forecasts vary with the timing of forecast disclosure. Speciflcally, tests examine whether analysts are less likely to condition their forecast revisions on price changes or on the Information underlying price changes occurring later in the firm's reporting year. This result is suggested if components of the unexpected earnings in a management forecast are transitory in nature (and thus, not priced by the market) and if the likelihood that a management forecast conveys these components is greater in the fourth quarter. Alternatively, the tendency to delay bad news (Penman 1984, 1987; Chambers and Penman 1984; Kross and Schroeder 1984) and the early security price reflection of the news (McNichols 1989) may combine to produce this effect.

Basis for the Hypotheses

Analyst Revisions, Management Forecasts, and Security Prices Hassell et al. (1988) document an association in terms of sign and magnitude between the news conveyed by management forecasts and subsequent compos
Jennings (1987) does not construct tests of the Information In security prices. However, based on the coefficient relations implied by reverse regression (Maddala 1977; Beaver et al. 1987), the results of a test In this paper using a continuous measure of security price change (described later) are derivable from his results. Also, in general, the association of security price changes and subsequent analyst revision is suggested by Jennings' results. If either interpretation holds, analysts act "as if" they use security price in forecasting earnings. Beaver et al. (1980.26) identify the nature of this "as if" behavior as a link to "potenUai circularity" of analysts' forecasts.

Baginski and HasseiiManagement Earnings Forecasts

177

ite analysts' revisions. This result is expected given the relative accuracy (Waymire 1986; Hassell and Jennings 1986) and informativeness (Patell 1976; Penman 1980) of management forecasts. However, this association between belief revisions due to management forecast releases and subsequent analyst forecast revisions may be attenuated or eliminated in certain cases. For example, managers may possess inferior information relative to other market participants or may not be truthful (Dye 1985; McNichols 1989). Timing Hypothesis Consistent with the Beaver et al. (1980) notions of permanent versus transitory components of earnings, Lipe (1986) argues that the persistence of earnings components affects security prices. It is possible that later forecasts by management may convey components of earnings expectations that are more transitory than earlier forecasts and, hence, that are priced to a lesser extent. Mendenhall and Nichols (1988) describe a reporting process whereby managers estimate accruals for interim reporting and have less discretion in disclosing annual results. Further, greater knowledge of transitory components is likely near yearend. For example, managers forecasting early in the year base their projections on expected (permanent) earnings adjusted for forecasted transitory components. Because transitory components are nonrecurring, they may be difficult to forecast. Thus, transitory components provide accurate adjustments to managers' earnings projections only as they are realized. Alternatively, discretioneuy manipulation of income may create transitory components. Healy (1985) provides a motivation for using fourth quarter discretionary adjustments to move annual earnings within a range of earnings subject to bonus. Finally, fourth quarter results may reflect corrections of errors during the flrst three quarters (Kinney and McDaniel 1989). The tendency of managers to delay bad news (Penman 1984, 1987; Chambers and Penman 1984; Kross and Schroeder 1984) suggests a second possible timing effect. However, McNichols (1989) reports a signiflcant association between news and price changes at the time of the forecast. Thus, while delaying bad news may weaken the power of security prices to explain analyst revisions, the effect probably is secondary in strength to the transitory component effect. Hypotheses Based upon the preceding arguments, three null hypotheses are tested: HOI: Changes in earnings expectations based on management forecasts have zero or negative association with subsequent revisions of EPS forecasts by flnanciai analysts. H02: Changes in earnings expectations implied by security price reactions to management forecasts have zero or negative association with subsequent revisions of EPS forecasts by flnanciai analysts. H03: The information in security price changes pursuant to management forecasts is not affected by the timing of the forecast. For Hoi and Hoi, the eiltemative hypotheses axe positive association of management forecast news and security prices with analysts' revisions. For H03. the

178

The Accounting Review. January 1990

Information In security prices Is expected to be reduced for late In the year forecasts.

Selection of Management and Analysts' Forecasts

Management forecasts are obtained from the Dow Jones News Retrieval Service database of articles appearing in The Wall Street Journal and Barrons for July 1,1979 to December 31,1982. Forecasts were included in the sample if they were attributed to a corporate official, were mnual earnings per share point estimates (or convertible into point estimates by averaging range endpoints or dividing earnings by shares outstanding), and were Issued prior to fiscal year-end. This sesa^ch resulted in over 500 forecasts. Anedysts' forecasts are gathered from an archival tape of Zacks Investment Company's Icarus Service.^ Seven weeks of analysts' forecast data were reqviired for each firm, beginning with the week before the management forecast and ending six weeks after the forecast. Management forecasts matched with the Zacks tape total 182." Of these, 23 do not have sufficient security return data on the CRSP file to estimate unexpected security returns, and 12 others have insufficient data on the COMPUSTAT Annual Tape to estimate earnings volatility and certain other variables. Table 1 provides the distribution of the flnal sample of 147 management forecasts by year, by flrm, by quarter of disclosure, and by exchange listing. The management forecasts are well distributed across flrms and years. However, 44 percent of the forecasts (64 to 147) are Issued In the fourth qusuter. Variable Measurement and Statistical Model Dependent VariableAnalysts' Forecast Revisions Deflning the management forecast week is week zero, revisions in consensus analyst EPS forecasts are computed as follows: AFR,,=[AFu-AF.oVSDEPS, for t=2, 4, 6 where: AFR(,=analysts' forecast revision for the firm issuing forecast t from week 0 through week t, AF,,=consensus analysts' forecast in week t for flrm t,* and SDEPS,=the standard deviation of EPS for the flrm Issuing forecast (. (1)

' This tape contains weekly consensus analyst forecasts for about 2,400 companies. * A detailed discussion of the search procedures Is provided In Hassell and Jennings (1986,61-65), This study's sample of 182 (up to this point) is greater than the Hassell ind Jennings sample because It Includes forecasts Issued in the last month of theflscalyear. Since Zacks provides emalysts' forecsists only up to the point at which actueil earnings are available, the sample decreases for certain tests after the management forecast. ' The Zacks tape reports a mean (consensus) forecast for all analysts Informing Zacks that they have made a forecast as of a given date.

Baginskl and HassellManagement Earnings Forecasts

179

Table 1 Distribution of Management Forecast Sample

By Year 1979 1980 1981 1982 1979-1982 Number of Forecasts 32 45 37 33 147 Number of Firms 84 20 6 1 111 Number of Forecasts 84 40 18 5 147 Number of Forecasts 13 42 28 64 147 Number of Forecasts 144 3 147

By Firm Firms Issuing one forecast Firms Issuing two forecasts Firms Issuing three forecasts Firms Issuing flve forecasts Totals

By Quarter of Disclosure First quarter Second quarter Third quarter Fourth quarter Total

By Exchange Listing Forecast Issued by flrm on NYSE Forecast Issued by firm on AMSE Total

The calculation of analysts' forecast revisions uses windows of two, four, and six weeks for two reasons. First, little evidence exists on the rapidity of analyst adjustment to new information. Given that analysts manage the trade-off between timeliness and accuracy, afinancialanalyst may take some time to evaluate a management forecast, especially if it differs substantially from the analyst's forecast. Second, in some instances, an analyst's revision may precede a report to Zacks. The revisions are standardized by the standard deviation ofthe EPS stream for a number of years ending with the year preceding the forecast.*
' On average. 12.5 consecutive EPS numbers are used to estimate the standard deviation. Ninetyone percent ofthe standard deviations are estimated using between ten and 15 observations.

180

The Accounting Review. January 1990

This relates revision magnitude to variability of earnings over time and, hence, controls for cross-sectional differences." Independent Variables The two independent (information) variables are: (1) the change in earnings expectations conveyed by the management forecast for flrm i (C7EE,), which indicates the news in the management forecast; and (2) the unexpected security price change [CARi). CIEEi is measured by comparing the management forecast of EPS in week t = 0 (MF,.o) to the consensus analysts' forecast outstanding one week prior (AF,.,) and deflating by SDEPS: CIEE>=[ME,_o-AF,,-i)ISDEPS,. (2) To measure CARi, the market model is estimated via ordinary least squares regression over a 200-day period preceding the management forecast. CAR, is the unexpected return for three days centered on the management forecast date, and accumulated and standardized using the method in Patell (1976).* For one set of tests, categorical variable (AGREEi) compares the signs of signals CIEE, and CAR,. If the signs agree (disagree), then AGREE,= 1 {AGREE,=O). As discussed in the next section, use of the AGREE, variable allows simple direct tests of the effects of signal disagreement on the relation between analysts' revision and management forecast news. Table 2 provides descriptive statistics for the variables across all quarters and for late (fourth quarter) versus earlier quarters. Peuiel A reports negative means for analysts' forecast revisions {AFR,) through weeks two, four, and six after management forecast release. The mean management forecast news (CIEE,) is also negative. The security price reaction surrounding the forecast is nominally positive but insigniflcantly different from zero at a .05 level. Panel B reports that the signs of CAR, and CIEE, agreed for 95 of 147 observations.' Statistical ModelsOverall Analysis To test hypotheses Hoi and H02, two slightly different models relate security price changes and management forecast news to analysts' forecast revisions. The first examines the incremental information in security price changes holding management forecast news constant: AFR,=ao+ai(CIEE,)+a2[CAR,)+e,. (3)

' ImhofTand Pare (1982) provide evidence that this deflator ls a good surrogate for the standard deviation of forecasted amounts and performs as well as other potential earnings variability deflators. * CAR, IS measured over three days while CIEE, ls measured over approximately one week. However, a CIEE, measure for a shorter period could be oiitalned only by using a consensus analyst forecast from week zero. It is not possible to determine whether such a forecast was detennined before or after the management forecast. Therefore, a week minus one forecast is used as a proxy for analysts' beliefs at the date of the management forecast and, thus, is consistent in terms of timing with a three-day CAR,. All tests were repeated using a seven-day CAR, ending the day after the management foreceist. The results are generally consistent but weaker in some cases. This is expected since a seven-day CAR, would average out the management forecast price reaction over a longer period and would probably contain price reactions to a number ofother events. * The tendency for CIEE and CAR to agree In terms of sign is a general conclusion in the literature on management forecasts (Waymlre 1984).

Baginski and HassellManagement Earnings Forecasts

181

Table 2 Descriptive Statistics

Panel A. Continuous Variables: Number of Observations Variable AFR,:' Quarters 1-3 Fourth Quarter All Quarters Quarters 1-3 Fourth Quarter All Quarters AFRt: Quarters 1-3 Fourth Quarter All Quarters CIEE: Quarters 1-3 Fourth Queirter All Quarters CAR: Quarters 1-3 Fourth Quarter AU Quarters 82 60 142 82 42 124 82 32 114 82 63 145 83 64 147

Mean

Standard Deviation

Minimum

Maximum

-.032 -.020 -.027 -.047 -.070 -.055 -.068 -.099 -.077 -.109 -.000 -.062 .122 .134 .128

.120 .147 .131 .173 .168 .171 .208 .198 .205 .430 .333 .393 1.659 1.319 1.515

-.519 -.800 -.800 -.519 -.800 -.800 -.969 -.854 -.969 -2.833 -.876 -2.833 -4.696 -4.622 -4.696

.300 .607 .607 .599 .090 .599 .674 .090 .674 .777 1.438 1.438 4.350 3.752 4.350

Panel B. Categorical Variable: Variable AGREE: Quarters 1-3 Fourth Quarter All Quarters ' Definitions: AFR, = standardized composite revision in earnings per share estimate made by flnanciai analysts from week zero to week t (week zero is the week of the forecast). CIEE change In earnings expectations cedculated by comparing a management foreceist to the preceding analyst forecast (standardized by the standard deviation of earnings per share). CAR= standardized unexpected price reaction to the management forecast measured from the day before through the day after the forecast. AGREE= indicator variable equal to one (zero) when the sign of changes in earnings expectations derived from a management earnings forecast agrees (disagrees) with the sign of the unexpected price reaction to the management forecast. Number of Observations 83 64 147 AGREE=1 59 36 95 AGREE=0 24 28 52

Under the Hm null. a,=0. The expectation for a, ls that It will be greater than zero. Under the H02 null, a2=0. Again, the alternative is an expected positive aj coefficient. " > Equation (3). however, restricts Oi across security price changes. To allow

182

The Accounting Review, January 1990

estimation of different associations of AFR, and CIEE, as a function of security price changes, the categorical variable AGREE, is used in a second model: AFR,=bo+bi[CIEE,) + bilAGREE,]+bi(CIEErAGREE,)+e,.
(4)

Testing for the difference from zero of slope coefficient bi and the sum of slope coefficients bi and b3 examines the hypothesis of association between management forecast news and subsequent analyst revision (Hoi), for the separate cases o{AGREE,=0 and AGREE,= l. Under HM. b2=0 and b3=0. If analysts' forecast revisions have a stronger association with management forecast news in the presence of signal agreement, then b3>0. Statistical ModelsTiming Effect To test the timing hypothesis, the coefficients in equations (3) and (4) are allowed to vary depending on whether the management forecast was issued in the fourth quarter. In equations (5) and (6), intercept and slope shift parameters are incorporated using an additional indicator variable, FQTR,. For fourth quarter forecasts, FQTR, equals one (and zero otherwise): AFR,=ao+a,[CIEE,)+a2{.CAR,)+ai[FQTR,)+a4[FQTR,*CIEE,) -i-a,{FQTR,*CAR,)+e,. AFR,=bo+b^(CIEE,) + bilAGREE,) + b^CIEE,* AGREE,) + b4(F9TR,) + bs(FQTR,*CIEE,) + bt,{FQTR,*CIEE,*AGREE,)+e,. (5)

(6)

In equation (5), the null hypothesis of no fourth quarter effect predicts that Oa and as equal zero. If transitory components of income included in fourth quarter management forecasts reduce the information in security price changes to predict subsequent analyst forecast revisions, then as will be less than 0. If the fourth quarter effect is sufficiently strong such that fourth quarter security price changes contain no infonnation, then a2-l-as=0. In equation (6), the null hypothesis of no fourth quarter effect predicts that b4 and be will equal zero. If transitory components in fourth quarter forecasts reduce the explanatory power of the AGREE, variable, then b6<0 and b3 + b6=0. While not a primary motivation of this study, an additional fourth quarter effect is captured by a* and bs in equations (5) eind (6), respectively. If analysts either perceive that fourth quarter management forecasts are more accurate than earlier forecasts or base their revisions on more accurate information that they received during the fourth quarter, then a4 and bj should be greater than zero.

Results
Table 3 presents results of the hypothesis tests before controlling for fourth quarter effects. The first column shows the three periods used to calculate euia' This model is related to a model in Jennings (1987). Jenningsflndsa signiflceint a i coefficient in a model with CAR, as the dependent variable and a measure similar to AFR, as the independent variable. Therefore, results of statistical tests using this equation provide a replication of certain of Jennings' results given the implications of reverse regression.

Baginski and HassellManagement Earnings Forecasts

00 CO
in es

183 ^
'H m CS

Q) .52

If

^ 1
CO

G O

C O

lai.

o> CD

co^

I ro CO -H
' *

e s 05

35 >-i

oJ^
' C)

CO CO

>

> CO ) CO tD N Q O

N ^ H ^^ CO

CO 00 ^^ 00

r '^
' I

1'^
' I

oo

o*

CDi.. rfU
ft)

-H^

-Ies ^ in ^

I p es

C ^ O Tji

C r^ O 'co

CO CO

'^es '^ro cs<

' isi

Tj" 0 0

>- <

in:

CO

in J~ in ^ * . ^ ^ o in -< (o C O O ^ O 00 O C O 'es ' i-i 'PH

+
I CO

5
)^
'es

i I

1^ ^ i s col, I M 4 lnes
I O O O O CS

I
.Q

Q ro t^ ro
-^ ^

I t p -' p C O

-H S 00
I es

ro^
I

CO - - I

- V

-H!L
in

p<N 'H ' ^ I CO I CO

-tII

.B fepp
u o
O II O II O
^

5! " S i ,
O II O
C Q

i^

<2 I

184

The Aecounting Review. January 1990

lysts' forecast revisions (AFR,). Panel A presents estimates for equation (3) using the continuous measure of security price reaction to the management forecast [CAR,). The measurement period length for AFR, does not affect the results. Signiflcant positive coefficients on ai (at the .01 level for each AFR, measurement period) indicate an association of management forecast news and subsequent analysts' revisions, thus rejecting Hw. Signiflcant positive coefficients on 02 (at the .01 level for the four-week revision period and at the .05 level for the other revision periods) indicate an incremental association of security price changes at the management forecast release and subsequent analyst revisions. These results allow rejection of H02. Panel B presents estimates of the alternative model (eq. (4]). Coefficient bs is always signiflcantly positive. This indicates a stronger association of AFR and CIEE when security price signals agree in sign with the management forecast news. Coefficient bi is signiflcantly different from zero for the two-week revision period only (.01 level). It is not clear why the earliest analysts' revisions would be associated with management forecast news when security prices disagree with the management forecast news (although the signiflcant bj coefflcient for the two-week AFR, indicates that they are more strongly associated when prices agree). One possible explanation is that some analysts experience a lag between making a revision and assigning a date to that revision. As a result, the data base may include some revisions made before the management forecast date and the security price reaction. Fourth Quarter Management Forecasts Table 4 examines the primary hypothesis (H03) that management forecast timing affects the relations documented in Table 3. In Panel A, coefficients aj, 04, and as allow fourth quarter shifts in the relations. The intercept shift coefficient ai is not signiflcantly different from zero for any AFR, measurement period. The slope shift coefficient on CIEE, (04) is weakly signiflcant for the two-week measurement period (04=.096, p=.082) and highly signiflcant for the four-week and six-week measurement periods (04=.337 and .328, respectively; p=.Ol). These results suggest that analysts follow management news more closely in the fourth quarter. An explanation consistent with these results is the tendency for analysts to revise more after later in the year releases (Stickel 1989)." Coefficient as provides a measure of the fourth quarter effect on the association of CAR, and emalyst forecast revision. A negative estimate is expected if the infonnation in security prices is less in the fourth quarter. For each AFR, measurement period, coefficient as is negative (p=.05 and .059 for the four-week and six-week measurement periods) and approximately equal in magnitude to the a2

" Stlckel's results dso may have Implications for the weaker evidence obtained for the two-week AFR, measurement period. As Indicated In Table 2. the four-week and six-week AFR revision periods contain 42 and 32 fourth quarter forecast observations, respectively. Reducing the AFK measurement period to two weeks increases the simple by over 40 percent (42 to 60 observations). These additional observations are very close to yeir-end and, thus, approach the time when fourth quarter earnings releases occur. Stickel documents only a small amount of analyst revision activity in the ten days preceding Interim earnings announcements.

Baginski and HassellManagement Earnings ForecEists

CO f-H CO

185
CO

(N

(N

CO

CO

rt C D

00 f

Is

P
lo ^

Ceo

qq

-H 0 0

00 V

CO GO

in Z.
^ rH

qq

qo>

CO.-, (N <N rH PH

M l , * IN t<. 03

r -

O ^ O * 00 C O

CON

ri.

(N<
CO

0 0

^ 00 -

C ^^ Ci OJ D - 00
CO

1^
" -H

CO 00 CO

q " eo

ae.

> rt O CO O C I C O O O O

O -^

Q ^

3 V

<2 o

88 82 88
Koe, oolU coi, ^ V . IN C "t C in O O qco qCO qq
"ei
CO
'CN

CD V 00 1 , C i n m CD O M rt C f.; O

(N _ i n -H ^ q

36

OO
E4C4 1 , O) . ^ in 00 * (N IM in 00 * (N q C O q tt ^ q c o qCO q ^ e4 ' -< ' -<

^ <y "w t- Cw O Cu V

c o ^ int.. <N^
CO CO

O 00 OOM (0 0) -1 C O M i-nO O

* S tN

qq

C r> O

qq

' d I r "^ ~

tfit
^ u 42 o 3] ^ J

C , N V. 10 1 . O

Sg SS S!^

'1 '1 ii
CD

q 00 q l'' ci i

ool, in (N

ml, t~ o ooi

1^1. 00 t* q C O

' ci i

' i

't

C D

I?
2
i-^

+ eT +5^ +
O II O It O
C Q

22 22

2-

O II O II O II

I I I

Illilll
> fc,
eg 5 1)

" IJ"^ B B B

.2 I

186

The Accounting Review, January 1990

coefficient on CAR,, For the fourth quarter {FQTR=1), the following estimates are obtained: AFR,=(ao+a,) + (a. -I- a4)CIEE,+(02 + a,)CAR,, AFR2= -.022 +. 199 CIEE,+.003 CAR,, AFi?4= -.045 + .422 CIEE,-.005 CAR., and AFJ?6= -.056+.490 C/EE,-.O11 CAR,. (7) The coefficients on CIEE, in equation (7) are each significantly different from zero at the .01 level. The coefficients on CAR, are not signiflcantiy different from zero at even the .10 level. Thus, security price changes pursuant to fourth quarter management forecasts do not provide incremental explanatory power for subsequent analyst forecast revisions. For earlier quarters (FQTR=O), coefficients a, and ai in panel A of Table 4 provide tests of the associations of CIEE, and CAR, with AFR,. As in the regression for all quarters in Table 3, the coefflcients are signiflcantiy positive at p < .05. The results in panel B of Table 4 lead to qualitative conclusions that are similar to those present in the panel A model, as well as some additional insights. First, recall that coefficient b, on CIEE, in Table 3 is significantly positive only for the two-week AFR, period. This indicates analyst revision association with management forecast news in the presence of security price reactions whose signs do not agree with the management forecast's news. In Table 4, b, is only nominally positive (.049), indicating that the relation documented in Table 3 does not hold for the flrst three quarters. That is, signal disagreement eliminates signiflcant association of analysts' forecast revisions and management forecast news. However, coefficient bs is signiflcantiy positive for each AFR, measurement period indicating a strong fourth quarter effect on the association of analyst revision and management forecast news. Equally as strong in the opposite direction is coefflcient be (significant at the .01 level for the four-week and six-week AFJR. periods). This indicates that signal disagreement reduces the fourth quarter association of CIEE, and AFR,. For fourth quarter forecasts (F9TR=1) when signal agreement exists, the equations for each measurement period are: AFR2= - .001 -I- .211 CIEE, AFi?4= -.025 + .338 CIEE, and AFRi = - .034 + .389 CIEE. (8) These results are similar to those in Table 3 for all quarters. However, for fourth quarter forecasts and signal disagreement, the equations are: AFR2 = - .052 +. 198 CIEE, AFR^=-.074 + .722 CIEE, and AFi?6 = - .082 + .837 CIEE. (9) These equations differ substantially from the signal disagreement case reported in Table 3 in that they imply signiflcant analysts' revisions. Thus, the signal disagreement case (AGREE=0) appears to drive the fourth quarter (timing) effect.

Baginski and HasseiiManagement Earnings Forecasts

187

Table S Binomial Test of Positive Association Between News Conveyed by Management Forecast and Sign of Unexpected Price Reaction (By Type of News, By Quarter) (1979-1982)
Null Hypothesis: There Is no association between sign of management foreeast news and sign of unexpected priee reaction. Percentage of One-Tailed Unexpected Price Binomial Test Number of Reactions That p-vaiue Forecasts Agree Quarters 1-3 Management Forecast Conveys Good News Management Forecast Conveys Bad News Quarter 4 Management Foreeast Conveys Good News Management Foreeast Conveys Bad News

33 50

.788 .660

.001 .012

30 34

.667 .470

.035 .242

Table 5 examines the nature ofthe AGREE variable by quarter and by type of management forecast news. If a positive sign relation exists between CAR, and CIEE,, then AGREE should equal one more than 50 percent of the time." For quarters one through three, the signs of the management forecast news and unexpected price reactions agree 78.8 and 66.0 percent ofthe time for good and bad news, respectively (p=.001 and p=.012 for rejection ofthe null hypothesis that p< .5). For the fourth quarter, good news forecasts yield similar results. The percentage of signs in agreement equals 66.7 (p=.035). However, the null hypothesis is not rejected for bad news forecasts, as only 47 percent ofthe signs agree." A possible explanation for the fourth quarter effect in panel B is that a high association of mEinagement forecast news and analysts' forecast revisions combines with a tendency for signal disagreement to occur in the fourth quarter (especially for bad news). Additional Tests Three additional tests were performed. The objective ofthe first is to provide direct evidence as to whether the transitory components of earnings hypothesis
" Research efforts by Waymlre (1984) and AJlnkya and Gift (1984) document symmetric belief adjustment by the securities market to management forecast news. This Is consistent with a positive sign relation between CARi and CIEE,. " Overall, the tendency for the disagreement case {AGREE,=0) to prevail across all forecasts Is positively associated with the fourth quarter.

188

The Accounting Review, January 1990

explains the fourth quarter effect. The original 147 management forecasts are reviewed to determine if more statements about transitory items existed in the fourth quarter relative to earlier quarters. The incidence of clearly transitory items is low for several reasons. First, management forecasts that include extraordinary items are omitted from the original sample of 147 so that the sample would be comparable to Zacks' analysts' forecasts. Second, management usually does not cite any reason for a forecast. Third, reasons given for a management forecast may be highly contextual. For example, a manager citing increased industry demand as a basis for a more favorable forecast may not comment on whether the increased demand is likely to persist. Thus, expected persistence ofa component of income may be industry, nrm, or year specific. The implication of the second and third reasons for the hypothesis tests is that misclassification of a transitory item is likely. Only 11 of the 147 forecasts are classified as conveying clearly transitory effects on income. Most of these relate to land transactions, securities transactions, or lawsuit settlements. Transitory items are included in eight of 64 fourth quarter forecasts and three of the 83 earlier forecasts. Although the number of transitory components is small, a chi-square test rejects the hypothesis of equal incidence of transitory items in favor of more fourth quarter transitory item disclosure (p< .05). Thus, with the clear limitation of classification error in mind, direct evidence supports the transitory component hypothesis. However, this test does not support the idea that transitory components underlie bad news forecasts more often in the fourth quarter. Of the eight fourth quarter forecasts with transitory components, four accompanied good news forecasts and four accompanied bad news forecasts. A second issue relating to the results in Tables 3 and 4 concerns the presence of simultaneous information releases during the three-day period used to measure the sign of CAR,. Waymire (1984) documents that simultaneous releases are associated as to sign with the news in management forecasts. Of the 147 management forecasts in Table 1, only 18 are issued with simultaneous disclosures of quarterly forecasts, quarterly earnings, or dividend changes. Moreover, only six of the unexpected components of these three types of disclosures have signs that disagree with the news in management's forecast.'* Therefore, the likely impact of simultaneous disclosures on the measurement of signal agreement is small. Finally, it is possible that other characteristics of the information environment are associated with the analyst revision process. For example, if analysts* revisions subsequent to a management forecast are merely a function of firm size and AGREE, is cross-sectionally associated with size, then the tests for slope shift would incorrectly indicate information in security prices. To assess the possibility that the AGREE, variable proxies for other characteristics, variables identified in prior research as characterizing some aspect of the information environment were included as independent variables in a logistic regression with AGREE, as the binary dependent variable. The variables selected were: (1) the value of com" Expectations for these disclosures are obtained from simple models described in Waymlre (1984).

Baginski and HassellManagement Earnings Forecasts

189

mon equity for the year-end preceding the management forecast to proxy for size; (2) the standard deviation of EPS to proxy for the precision of prior information; and (3) a dichotomous variable indicating whether the management forecast disclosure was made in the fourth quarter (FQTR)." The hypothesis that all parameter estimates in the logistic regression are zero is not rejected at a conventional significance level. In addition, a negative estimate for FQTR indicates a tendency for signal disagreement in the fourth quarter and for signal agreement earlier in the year. Concluding Remarks The results of this study are consistent with a world in which analysts act as if they use security price reactions to management forecasts to revise beliefs about future earnings. This implies that analysts either use the information in security price chinges or obtain this information from other sources. These results do not conflict with the existence of information in analysts' forecast revisions. The news conveyed by the management forecast and the agreement ofthe management forecast news and the unexpected price reaction to that news explains only a portion of analysts' forecast revisions. In fact, financial analysts may provide the additional service of translating security price information into information on future earnings prospects. The results also indicate that disclosure timing affects the ability of minagement forecast news and security price changes to predict analysts' forecast revisions. While the association of forecast news and analysts' revisions is stronger in the fourth quarter, the association of security prices and aneilysts' revisions is eliminated. These results are consistent with the notion that fourth quarter management forecasts contain more earnings components with little or no pricing implications than forecasts issued earlier in the yeir.
" The size meeisure Is suggested by the work of Atiase (1985). Exchange listing Is not Included In the model (Grant 1980) due to lack of variation (see Table 1). The volatility measure is similar to the earnings predictabiUty measure In Pincus (1983).

References Ajlnkya, B. B., and M. J. Gift. 1984. Corporate managers' earnings forecasts and symmetrical adjustments of market expectations. Journal of Accounting Research. (Autumn): 425-444. Atiase, R. K. 1985. Predlsclosure Information, firm capitalization and security price behavior around earnings announcements. Journal of Accounting Research. (Spring): 21-36. Bathke, A. W., Jr., and K. S. Lorek. 1984. The relationship between time-series models and the security market's expectation of quarterly earnings. The Accounting Review. (April): 163-176. Beaver, W. H., R. A. Lambert, and D. Morse. 1980. The information content of security prices. Joumal of Accounting and Ek:onomics. (March): 3-28. , , and S. G. Ryan. 1987. The information content of security prices: A second look. Joumal of Accounting and Economics. (July): 139-157. Brown, P., G. Foster, and E. Noreen. 1985. Security analyst multi-year earnings forecasts and the cc^ital market. Studies in accounting research no. 21. American Accounting Association.

190

The Accounting Review, January 1990

Chambers. A. E.. and S. H. Penman. 1984. Timeliness of reporting and the stock price reaction to earnings announcements. Joumai ofAjccounting Research. (Spring): 21-47. Collins, D. W., S. P. Kothari, and J. D. Rayburn. 1987. Firm size and the infonnation content of prices with respect to earnings. Journai ofAccounting and Economics. (July): 111138. Dye, R. A. 1985. Disclosure of nonproprietary information. Journai of Accounting Research. (Spring): 123-145. Grant E. B. 1980. Market implications of differential amounts of interim infonnation. Journai of Accounting Research. (Spring): 255-268. Hassell, J. M., and R. H. Jennings. 1986. Relative forecast accuracy and the timing of earnings forecast announcements. The Accounting Review. (January): 58-75. , , and D. Lasser. 1988. Management earnings forecasts: Their usefulness as a source of firm-specifLc information to security analysts. Joumai ofFinancialResearch. (Winter): 303-320. Healy, P. M. 1985. The effect of bonus schemes on accounting decisions. Joumai of Accounting andEconomtcs. (April): 85-107. Imhoff, E. A., Jr., and P. V. Pare. 1982. Analysis and comparison of earnings forecast agents. Joumai of Accounting Research. (Autumn): 429-439. Jennings, R. 1987. Unsystematic security price movements, management earnings forecasts, and revisions in consensus analyst earnings forecasts. Joumai of Accounting Research. (Spring): 90-110. Kinney, W. R., Jr., and L. S. McDaniels. 1989. Characteristics of firms correcting previously reported quarterly earnings. Joumai of Accounting and Ek:ononilcs. (February): 7193. Kross, W.. and D. A. Schroeder. 1984. An empirical investigation of the effect of quarterly earnings announcement timing on stock returns. Joumai of Accounting Research. (Spring): 153-176. Lipe, R. C. 1986. The information contained in the components of earnings. Joumai ofAccounting Research. (Supplement): 37-64. Maddala, G. S. 1977. Econometrics. McGraw-Hill. McNichols, M. 1989. Evidence of informational asymmetries from management earnings forecasts and stock returns. The Accounting Review. (January): 1-27. Mendenhcill, R. R., and W. O. Nichols. 1988. Bad news and differential market reactions to announcements of earlier-quarters versus fourth-quarter earnings. Joumai of Accounting Research. (Supplement): 63-86. Patell, J. M. 1976. Corporate forecasts of earnings per share and stock price behavior: Empirical tests. Joumai of Accounting Research. (Autumn): 246-276. Penman, S. H. 1980. An empirical investigation of the voluntary disclosure of corporate earnings forecasts. Joumai of Accounting Research. (Spring): 132-160. . 1984. Abnormal returns to investment strategies based on the timing of earnings reports. Journai of Accounting and Economics. (December) 165-183. 1987. The distribution of eeirnings news over time and seasonalities in aggregate stock returns. Joumai ofFinanciai Economics. (June): 199-228. Pincus, M. 1983. Information characteristics of earnings announcements and stock market behavior. Journal o/Accounting Research. (Spring): 155-183. Stickel, S. E. 1989. The timing of and incentives for emnual earnings forecasts near interim earnings announcements. Joumai of Accounting and Economics. (July): 275-292. Waymire, G. 1984. Additional evidence on the information content of management earnings forecasts. Joumai of Accounting Research. (Autumn): 703-718. . 1986. Additional evidence on the accuracy of analyst forecasts before and after voluntary management earnings forecasts. The Accounting Review. (January): 129-142.