Sunteți pe pagina 1din 17

Accounting Research Center, Booth School of Business, University of Chicago

Predisclosure Information, Firm Capitalization, and Security Price Behavior Around


Earnings Announcements
Author(s): Rowland Kwame Atiase
Source: Journal of Accounting Research, Vol. 23, No. 1 (Spring, 1985), pp. 21-36
Published by: Wiley on behalf of Accounting Research Center, Booth School of Business,
University of Chicago
Stable URL: http://www.jstor.org/stable/2490905
Accessed: 08-07-2016 22:27 UTC
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
http://about.jstor.org/terms

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted
digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about
JSTOR, please contact support@jstor.org.

Accounting Research Center, Booth School of Business, University of Chicago, Wiley


are collaborating with JSTOR to digitize, preserve and extend access to Journal of Accounting Research

This content downloaded from 131.94.186.72 on Fri, 08 Jul 2016 22:27:21 UTC
All use subject to http://about.jstor.org/terms

Journal of Accounting Research


Vol. 23 No. 1 Spring 1985
Printed in U.S.A.

Predisclosure Information, Firm


Capitalization, and Security Price
Behavior Around Earnings
Announcements
ROWLAND KWAME ATIASE*

1. Introduction
Several studies on the relationship between earnings reports and
security price behavior provide evidence suggesting that a significant

portion of the information revealed through earnings reports is reflected


in security prices prior to the report month (e.g., Ball and Brown [1968]
and Brown and Kennelly [1972]). This has been attributed (at least
partly) to the existence of other more timely sources of information
which allow market agents to forecast earnings prior to their release.

The evidence above reflects an "average finding" across all firms. One
question of interest is whether there are significant systematic cross-

sectional differences in the security price reactions to earnings announcements of firms which are associated with specific firm characteristics

that lead to differential amounts of predisclosure information. This study

focuses on firm size (capitalization) as one such characteristic.' A "firm


* Assistant Professor, University of Florida. This paper is based on my doctoral dissertation completed at the University of California, Berkeley (committee members: Professors
James Ohlson [chairman], David Downes, and James Pierce, and workshop committee:
Professors Nils Hakansson [chairman], Robert Freeman, and G. Garcia). I gratefully
acknowledge their helpful comments, as well as those from Stephen Penman, Kwesi Odoom,
Rashad Abdel-khalik, Bipin Ajinkya, Senyo Tse, Prem Jain, and an anonymous referee.
Financial support was received from the University of Ghana and the Accounting Depart-

ment, University of California, Berkeley, for which I am grateful. [Accepted for publication
August 1984.]
1 Grant [1980] suggests that another firm-specific characteristic is the market in which

a firm's equity securities are traded (i.e., an "exchange effect"). He observed that the
average security price reaction to the earnings announcements of over-the-counter (OTC)
firms was significantly higher than those for New York Stock Exchange (NYSE) firms.
21
Copyright ?), Institute of Professional Accounting 1985

This content downloaded from 131.94.186.72 on Fri, 08 Jul 2016 22:27:21 UTC
All use subject to http://about.jstor.org/terms

22 JOURNAL OF ACCOUNTING RESEARCH, SPRING 1985

size-related differential information hypothesis" or "size effect" is pro-

vided by Atiase [1980], who argues that the amount of private predisclo-

sure information production and dissemination is an increasing function


of firm size (capitalization).2 Based on this notion, the amount of "un-

expected" information conveyed to the market by actual earnings reports


should be inversely related to firm capitalization, other things being
equal. The findings reported in this study strongly support the hypothesis
that the degree of unexpected security price changes in response to
earnings reports is inversely related to the capitalized value (size) of

firms. Several other researchers (Freeman [1983], Richardson [1984],

and Ro [1984]) have found results consistent with the findings reported
here.

In the next section, I describe the empirical testing procedures. Results

of the empirical study are presented in section 3, and conclusions appear


in section 4.

2. Empirical Procedures
The study covered the four-year period, 1969 through 1972, with weekly
security price revaluations examined in response to second-quarter earnings reports of 200 sampled firms in 1971 and 1972.3 Regression analysis
was used to test the hypothesis that the degree of a firm's security's price
revaluation in response to its earnings announcement was inversely
related to the firm's capitalization. Details of the sample design, variable

definitions, test statistics, and procedures are given below.


2.1 SAMPLE DESIGN

Because of differences in data availability and sources for NYSE and


American Stock Exchange (AMEX) firms vis-a-vis OTC firms, I used
two sets of overlapping criteria for sample inclusion.

2.1.1. Sample design: NYSE/AMEX firms. The following eight criteria


were used to select NYSE/AMEX firms used in the study. The firm had
to (1) be a member of the NYSE or the AMEX; (2) be listed on the daily
returns tape constructed by the Center for Research in Security Prices

(CRSP) at the University of Chicago for the period 1969-72; (3) be on


the Compustat tape; (4) have had a December 31 fiscal year-end; (5) have
had a capitalized value as of the beginning of 1971 or 1972 either less
2See Atiase [1980] for a review of the theoretical and empirical evidence in the
accounting-finance literature supporting this notion.
3 Second-quarter earnings were chosen to minimize confounding problems. For example,

first-quarter earnings tend to be released along with, or close to the time of, annual earnings
announcements. Similarly, May [1971] found that first announcements of dividend changes
are made more frequently during the fourth quarters of firms' operating years along with
or close to third-quarter earnings announcements. The number of years covered by the
study was largely constrained by the extensive data gathering involved in generating weekly
returns for the OTC firms. Nevertheless, there is no a priori reason to believe that the
research question of interest or the results of the study would be period specific.

This content downloaded from 131.94.186.72 on Fri, 08 Jul 2016 22:27:21 UTC
All use subject to http://about.jstor.org/terms

PREDISCLOSURE INFORMATION 23

than $20 million (relatively-small-capitalized-value [RSCV]) or greater


than $400 million (relatively-large-capitalized-value [RLCV]); (6) have
had 1971 and/or 1972 second-quarter earnings announcements published
in the Wall Street Journal (WSJ) on a day other than Monday;4 (7) have
had no dividend announcements in the same week as the earnings
announcement; and (8) have had no stock split announcements or
forecast of annual earnings during the seven-week period surrounding
the earnings announcements.
Criteria (1) through (4) provided assurance of easy access to weekly
returns data from the CRSP daily returns tape and appropriate financial
data from Compustat. Criterion (5) led to two final subsamples, one made
up of 100 RLCV firms and the other of 50 RSCV firms. Fifty additional
firms were selected from the OTC population, as described below. Criterion (6) was imposed because earnings announcements reported in the
WSJ on Mondays could have been released by firms and announced over
the Broad Tape on the previous Friday.5 Criteria (7) and (8) were imposed
to minimize any ambiguity associated with an observed security price
reaction in the week of the earnings announcement.
2.1.2. Sample design: OTC firms. Criteria (3) through (8) above were
also imposed in the selection of the OTC firms included in the study.
Regarding criterion (5), all the OTC firms selected fell in the subsample

of RSCV firms. Criteria (1) and (2) were replaced by (la) the firm's
equity securities had to have been traded OTC for the entire 1969-72
period; and (2a) the firm had to have been listed in the Investment
Statistics Laboratory (ISL) Daily Stock Price Index-OTC, with complete
data on stock prices, dividends, and capital changes available for the
entire 1969-72 period.
2.2 DATA COLLECTION

Weekly return relatives for each NYSE/AMEX firm, along with weekly
market return relatives, were computed from daily CRSP tapes for the
period 1969-72. Fisher's Value-Weighted Index was used for the market
returns. For the 50 OTC firms, data on weekly closing bid prices, cash
dividends, stock dividends, and stock splits were manually gathered from
the ISL Daily Stock Price Index-OTC for the four-year period, 1969
through 1972 inclusive, and used to compute weekly return relatives for
each of these OTC firms.
The final data consisted of 200 second-quarter earnings announcements, one each for 100 RSCV and 100 RLCV firms. The capitalized
values (CVs) of the firms included in the sample ranged from $1.80
million to $19.98 million (RSCV firms) and from $426.23 million to
$38.88 billion (RLCV firms).
4 The same restriction would have been applied to earnings announcements published
in a Tuesday WSJ if the previous Monday turned out to be a public holiday; but, this
problem was not encountered.

In fact, personal consultation with the San Francisco office of Dow Jones, Inc.,
publishers of the WSJ, confirmed that is generally the case.

This content downloaded from 131.94.186.72 on Fri, 08 Jul 2016 22:27:21 UTC
All use subject to http://about.jstor.org/terms

24 ROWLAND KWAME ATIASE


TABLE 1

Number of Weeks Between Second-Quarter End and Announcement Week


No. of Percentage of Announcements Cumulative Percentage
Weeks RSCV Firms RLCV Firms RSCV Firms RLCV Firms

2
3

7
8

11

31

19

19

34

38

21

17

59

15

10

8
9

9
3

10

4
5

2
1

74
84

93
96

100

6
37
71

88
92
97

99

100

100

The dates of the earnings announcements were obtained from the Wall

Street Journal Index (WSJI) and were cross-checked with the WSJ.
Table 1 reports comparative time lags between the financial statement

dates (June 30, 1971 [1972]) and the announcement dates for the RSCV
firms and the RLCV firms. For the RSCV and RLCV firms, the median
time lags were five and four weeks respectively, and all announcements
were made by the RSCV and RLCV firms by the end of ten and nine
weeks respectively.

The returns history was divided into an estimation period (EP) of 104
weeks and a test period (TP), which was further divided into a predisclo-

sure period (PP), and a report period (RP). The report period (RP) was
defined as the seven weeks surrounding the announcement date (ranging
from three weeks before the announcement week, week zero, to three
weeks after). Two basic definitions of the predisclosure period (PP) were
employed: (i) the period between the beginning of the fiscal period and
the beginning of the report period (P); and (ii) the period between the
beginning of the fiscal period and the end of the second quarter (P').
Thus, P takes into consideration the differences in the time lag between
the financial statement date and the announcement dates of the various
sampled firms, while P' does not. Figure 1 depicts the research design.

2.3 VARIABLE DEFINITIONS AND MODEL SPECIFICATIONS

Sharpe's [1964] market model was used to eliminate marketwide

sources of price change:6


6The propriety of using the same market return relative, Rmt, and the same market
model (1) for both RLCV firms and RSCV firms, particularly for the OTC firms, may be
questionable since the market index, Rmt, does not even include the returns on securities
traded on the OTC market. Either a market index that embraces returns on all securities
(including those traded OTC) or a separate market index constructed for OTC firms might
have been more appropriate. Evidence presented later indicates that the results of this
study are robust with respect to the above.

This content downloaded from 131.94.186.72 on Fri, 08 Jul 2016 22:27:21 UTC
All use subject to http://about.jstor.org/terms

PREDISCLOSURE INFORMATION 25

Rit = ai + 3iRmt + e&t i = 1 ...N firm index (1)

t = 1 ... T, week index


where:

Rit = the natural logarithm of the return relative for firm i in week t;
Rmt = the natural logarithm of the return relative for the market
portfolio in week t;

ejt is the stochastic individual component of Rjt; and


ai and f3 are parameters unique to firm i, representing the
intercept and systematic risk coefficients, respectively.
Ordinary Least Squares (OLS) regression was performed on the ex post

realizations of Rd and Rmt observed during the 104-week estimation period


in order to obtain the estimates a, and bi of a, and f3, respectively.7 The
estimated coefficients were used to compute ui, the unexpected price
changes during the test period (TP) as follows:

uit = Rit - (ai + biRmt). (2)


Given that the unexpected price changes were estimated using observa-

tions which were not used in the estimation of ai and bi, they do not
Predisclosure period (P) Ea

P Loq
13cjipiiing 1969 (1970) End 1970 (19-71)

ki-

Copitolized Jdne 30 Reiport


values

1971

week

(1972)

Eshlinotion period (EP) - 104 weeks Test period (TP)

FIG. 1.-Research design.

'The average bi over the entire sample was 0.97, and 0.99 and 0.95 for the RSCV and
RLCV firms respectively. None was significantly different from 1.00 at the 10% level.
Hence, the sampled firms did not appear to be members of a particular segment of the
systematic risk spectrum. For the entire sample, the average R2 for the market model
regressions was 0.214, which is higher than that observed for NYSE firms by Beaver [1968],
Patell [1976], and Grant [1980] for weekly returns data in periods 1961-65, 1963-69, and

1960-64, respectively. However, the average R2 for the RSCV firms was only 0.113, compared
to the corresponding figure of 0.315 for the RLCV firms. Accordingly, it seemed necessary
to test the sensitivity of my results for these firms. (Results reported below show that the
average report week price revaluation of the RSCV firms is significantly [uniquely] higher
than the theoretical expectation and the averages in the other weeks in the report period.)

This content downloaded from 131.94.186.72 on Fri, 08 Jul 2016 22:27:21 UTC
All use subject to http://about.jstor.org/terms

26 ROWLAND KWAME ATIASE

represent residuals in the OLS sense. However, under the OLS regression
assumptions, and the hypothesis that these assumptions hold during the
TP, the ait are distributed as:

E(ait) = 0 t E TP, i = 1 ... N (3)

CoV(Uiss~~ Ii* - C*f -* ,t TP, i= 1 ... N (4)

Cov(&8, ~ = cist*ua, S =$t

COv(aitfmt = t EE TP, i-1 .............. N (5)


where:
T ~2

-.i2 S.2 =
E tEPS 6 (6)
E
t EEEP
tiT - 2

Cist*
= (Rmt
Rm)(Rms - Rm) t* E TP, (7)
Cs*T+
T
E (Rmt -Rm)2

t=1

Cit* = 1T+T1 + (Rmt* -Rm)2 t* E TP, (8)


E (Rmt - Rm)2

t=1

T= number of weeks in the estimation period (T = 104 weeks for all


sampled firms)

Rm =- Rmt t E EP. (9)


T t=1

Further, in order to apply a test of significance, the more restrictive

assumptions were made that the return distributions were Normal with:

Cov(&it*, Qit*) -{ 2 i u i- t* E TP, i, j = 1... N. (10)


In general, the over time covariance term, Cist* will be small (in the order
of 2/T) and will diminish as T increases. Thus for a large enough T (as

in the case of this study), cov(ais, ait) 1. 0, for s $ t*. The Cit* term
reflects the increase in variance due to prediction outside the estimation

period (EP). It is a function of the number of observations, T, used in

the EP, and how far market returns deviate from past averages.' Notice
however, that (10) implies cross-sectional independence of residuals in
the TP and ignores correlations such as the industry effects noted by
King [1966], and the extra-market covariance return arising from common factors noted by Rosenberg [1974] and by Rosenberg and Marathe
[1976]. These are controlled for later, although in any event, recent
evidence by Collins, Rozeff, and Salatka [1982] and Oppong [1980]
8 See Collins and Dent [1984].

This content downloaded from 131.94.186.72 on Fri, 08 Jul 2016 22:27:21 UTC
All use subject to http://about.jstor.org/terms

PREDISCLOSURE INFORMATION 27

suggests that cross-sectional correlation in weekly residuals may not be

significant, particularly at the individual security level.'


2.3.1. The revaluation indices (alternative dependent variables). Of
particular importance to testing the main hypothesis in this study is the
construction of an adequate index of a security's earnings report week

price revaluation which endogenously controls for predisclosure infor-

mation. A motivation for controlling for predisclosure information en-

dogenously is to avoid various conceptual problems associated with using


exogenous measures of predisclosure information. For example, an ex-

ogenous measure of predisclosure information employed in several prior


studies is the number of news items that appeared in the WSJ about a
firm during the predisclosure period.10 However, there are other channels

besides the WSJ for communicating information to investors. Similarly,


privately acquired information that is reflected in prices during the
predisclosure period may not appear in the WSJ. Finally, news items
appearing in the WSJ on various firms are probably not comparable,
given that not all news items will be equally pertinent to valuing firms.

To avoid these problems I controlled for predisclosure information by


constructing an index that measures security price revaluations in response to earnings announcements relative to average price revaluations

during predisclosure periods. This index provides a completely endogenous control for predisclosure information.

Since no assumptions could be made about investors' predisclosure

expectations, the sign of the report week unexpected returns, uit, could
not be specified ex ante. This motivated the use of u U , which is essentially
an estimate of the variance of the unexpected return in the report week.
u. for the report week for each firm was standardized by the EP residual
variance. Dividing the resulting F statistic by its expected value provides

the Revaluation Index (RI), distributed as given below:"


U?* T -4

RIit* = C iS 2- T t-2' E report period (11)


E(RIit*) = 1

Var(RIit*) --(T
2(T-6)3)
'Indeed, Collins, Rozeff, and Salatka [1982] found that even for abnormal returns
measured at a common point in calendar time for firms from the same basic industry (oil
and gas production), the average level of cross-sectional correlation in weekly residuals was
only 0.067. Besides assumption (10), Oppong [1980] also provides evidence in support of
assumption (5).

10 See, e.g., Grant [1980].

" RI is similar to measures used by Beaver [1968] and Patell [1976]. For a detailed
derivation of RI and its distributional properties, see Atiase [1980] or Patell [1976].

This content downloaded from 131.94.186.72 on Fri, 08 Jul 2016 22:27:21 UTC
All use subject to http://about.jstor.org/terms

28 ROWLAND KWAME ATIASE

Note that the variance of RI is a function of T only. Since the value of


T is the same for all sampled firms, RIi has constant expectation and
constant theoretical variance for all sampled firms, i (throughout the
report and predisclosure periods).12

For an intuitive interpretation, the RIi for the report week, RIjo, may

be viewed as a surrogate for the ratio of the "new" information conveyed


to the market by the earnings report (numerator), to the average information over the estimation period (denominator) for firm i. Thus, an

RIjo greater than 1.0 would imply that the earnings report conveyed more
than average information during the estimation period, and vice versa.

Similarly, the average RIi, over the predisclosure period may be intuitively

viewed as the average predisclosure (period) information relative to the


average information over the estimation period. Thus, the report period

RIi,* for each sampled firm can be standardized by the average RIj, over

the predisclosure periods (P and P'), thereby yielding the following F

statistics:

_ _ _ F(_ Pi), t* E report period (RP) (12)

1 1 ' t E predisclosure period (Pi).


Pi t=1

R F(1, Pt'), t* E report period (RP) (13)

P1 ' t E predisclosure period (Pi').

PiIX
RIt
E t= t
For comparability across firms whose Pi values differ, the statistics in
(12) and (13) are scaled by their expected values to give the Standardized
Revaluation Indices-SRI1 and SRI2 respectively, distributed as given
below:

SRI1~* = RIit* Pi - 2 t~ *E RP,


1

Pi

PtEP

tEXRIit
Pi t=1

E(SRI1st*)

1,

(14)

Var(SRIlit*) = (Pi - 1)

SRI2it* = RIt* P'- 2 t *E RP,


i Pi' Pi t EPi,
X RIat

Pi. t= 1

E(SRI2it*)

1,

(15)

Var(SRI2it*) = 2(P' - 1)
12 This is a convenient property for the final regression models. In particular, if one
assumes that the model specifications are translation invariant, the constant variance
property ensures homoscedasticity.

This content downloaded from 131.94.186.72 on Fri, 08 Jul 2016 22:27:21 UTC
All use subject to http://about.jstor.org/terms

PREDISCLOSURE INFORMATION 29

Thus RIjo, the measure of new information conveyed to the market by


the earnings announcements, is assessed relative to the average information during the predisclosure periods P and P' for each firm i, through

SRI1io and SRI2jo. Hence, each firm i acts as its own endogenous control.
RIj, SRI1j, and SRI2j were computed for each of the seven weeks of the
report period.

A priori, the superiority of SRI1 over SRI2 lies in the fact that the
former allows control for differences in the time lag between the financial

statement date and the earnings announcement weeks of the various

sampled firms, while the latter does not.


2.3.2. Capitalized value (the explanatory variable). The measure of

capitalized value employed here was the "market value of a firm's


common stock" (CV). In the final regression models (specified in section
3.2) and test of hypothesis, the CVs were transformed into their natural
logs (LCV).

3. Results
Analyses were performed at a portfolio level, to test the uniqueness of

the report week price revaluations, and at the individual-firm level, to


test the main hypothesis.
3.1 UNIQUENESS OF THE AVERAGE RIO, SRI1o, SRI2O

Tables 2, 3, and 4 report the estimated averages of RI, SRI1, and SRI2
(i.e., RI,SRI1 and SRI2 respectively) in each week in the report period
for the following five portfolios: (1) total sample, (2) total subsample of

RSCV firms, (3) RSCV OTC firms only, (4) RSCV NYSE/AMEX firms
only, and (5) the subsample of RLCV firms.
The results obtained for SRI1 (table 3) are consistently stronger than
those for SRI2 (table 4), as expected. Nevertheless both sets of results
show essentially the same patterns. For ease of exposition, then, the
discussion is limited to RI and SRI1.

Compareqd to their theoretical expectations of 1.000, the average report


week RI (RIO) and SRI1 (SRI10) for the portfolio of the entire sample
TABLE 2

Estimated Average RI
Sample Description

Week in Report Period Relative to Report Week


-3

-2

-1

+1

+2

+3

Total sample ............ 0.767 0.549 1.133 2.026a 0.805 0.866 0.837
RSCV firms:

Total subsample ......... 0.763 0.564 1.256 3.2268 0.547 0.674 0.556
OTC firms only ......... 0.532 0.282 0.837 3.843a 0.581 0.759 0.457

NYSE/AMEX firms only. 1.019 0.877 1.720 2.543a 0.510 0.579 0.666
RLCV firms .............. 0.771 0.533 1.009 0.814b 1.064 1.061 1.120
aSignificantly higher than both the E(RI) and the RIs in the other weeks in the report period at an
a-level of less than or equal to 0.005.

b Not significantly different from the E(RL) or the RIs in the other weeks in the report period at any
meaningful level of a.

This content downloaded from 131.94.186.72 on Fri, 08 Jul 2016 22:27:21 UTC
All use subject to http://about.jstor.org/terms

30 ROWLAND KWAME ATIASE


TABLE 3

Estimated Average SRI1


Week in Report Period Relative to Report Week

Sample Description

-3

-2

-1

+1

+2

+3

Total sample ............ 0.764 0.540 0.952 1.611a 0.794 0.795 0.800
RSCV firms:

Total subsample ......... 0.613 0.438 0.814 2.377a 0.433 0.457 0.451
OTC firms only ......... 0.550 0.344 0.809 2.9398 0.437 0.603 0.413
NYSE/AMEX firms only. 0.684 0.543 0.819 1.7548 0.427 0.296 0.493
RLCV firms .............. 0.917 0.642 1.091 0.838b 1.159 1.137 1.151
a Significantly higher than both the E(SRI1) and the SRIls in the other weeks in the report period
at an a-level of less than or equal to 0.005.

b Not significantly different from the E(SRI2) or the SRIls in the other weeks in the report period
at any meaningful level of a.

TABLE 4

Estimated Average SRI2


Week in Report Period Relative to Report Week
Sample Description

-3

-2

-1

+1

+2

+3

Total sample ............ 0.745 0.532 0.950 1.5808 0.786 0.785 0.788
RSCV firms:

Total subsample ......... 0.594 0.426 0.817 2.3238 0.424 0.445 0.434
OTC firms only ......... 0.524 0.329 0.815 2.8918 0.438 0.590 0.397

NYSE/AMEX firms only. 0.671 0.533 0.820 1.6598 0.408 0.284 0.475
RLCV firms .............. 0.897 0.638 1.083 0.829b 1.152 1.131 1.146

aSignificantly higher than both the E(SRI2) and the SRI2s in the other weeks in the report period
at an a-level of less than or equal to 0.005.

b Not significantly different from the E(SRI2) or the SRI2s in the other weeks in the report period
at any meaningful level of a.

were 2.026 and 1.611, 103% and 61% higher than their theoretical
expectations respectively. The theoretical variances of the RIo

and SRIl0 for this portfolio were 0.0103 and 0.0114 respectively.
Hence both the RIo and SRIlo were significantly higher than 1.000 at an
al-level of less than or equal to 0.005. As a result, we can conclude that
irrespective of the extent of information production and dissemination
about individual sampled firms, the average earnings report week price
revaluations (for the entire sample) were 103% higher than the average
during the estimation period (104 weeks) and 61% higher than the
average during the predisclosure period (26 weeks on average). These
results are consistent with those reported by Beaver [1968] and May
[1971].

The RIo of the portfolios of the "total subsample of RSCV firms,"


"RSCV OTC firms only," and "RSCV NYSE/AMEX firms only" were:
3.226, 3.843, and 2.543 respectively (223%, 284%, and 154% higher than

the E(RI)). The respective SRITo of the above portfolios were 2.377,
2.939, and 1.754 (138%, 194%, and 75% higher than the E(SRI1)). Again,
the averages were significantly higher (a c 0.005) than their theoretical

expectation of 1.000 and their RI and SRI1 in the other weeks in the

This content downloaded from 131.94.186.72 on Fri, 08 Jul 2016 22:27:21 UTC
All use subject to http://about.jstor.org/terms

PREDISCLOSURE INFORMATION 31

report period. These results suggest that tests of the main hypothesis
below should be robust enough even with the use of a single market
model and market return index for all the sampled firms (including the
OTC firms).'3

The results for the portfolio of "RLCV NYSE/AMEX firms," however,

were 0.814 and 0.838, for RIo and SRIlo respectively, or only 19% and
16% lower than the theoretical expectation of 1.000, neither of which is

significant at any meaningful level of a. The RIO and SRI10 were also
not significantly (uniquely) different from the averages in the other
weeks in the report period. I should note that my results for the RLCV
firms and the RSCV OTC firms are consistent with the results reported
by Grant [1980], as are the relatively higher average report week price

revaluations of the RSCV OTC firms compared to the RLCV NYSE/


AMEX firms.14

To ensure that the portfolio-level results above were not driven by


extreme price revaluations to earnings announcements of a few firms,

each of the above parametric tests was replicated using a nonparametric


test based on a hypergeometric distribution.'5 The null hypothesis is that

each of the seven weeks in the report period of each firm has an equal

chance (P = 1/7) of containing the largest RIi,* or SRIlit* *16


For the portfolios of (1) the total sample of 200 firms, (2) the subsample
of 100 RSCV firms, (3) the 50 RSCV OTC firms, and (4) the 50 RSCV

NYSE/AMEX firms, the report week SRIli was largest in the sevenweek report period for 81, 66, 41, and 25 firms, respectively. Assuming
the null hypothesis that P = 1/7, the probability of observing equal or
greater frequencies than these observed frequencies for each of the

portfolios is less than 0.0001. Thus the abnormally high SRI10 (and RIo)
observed were not just due to extreme price revaluations around the
earnings announcements of a few firms. For the portfolio of the RLCV
NYSE/AMEX firms, however, the report week SRIli was the largest in
the report periods for only 15 out of 100 firms, which results in a rejection
of the null hypothesis at an a-level of 0.4129. Overall, then, the results
of the parametric and nonparametric tests are consistent with each other.
13 The uniqueness of the report week RI (SRI1) relative to the other weeks in the report
period (particularly for the RSCV firms) also suggests that the results of this study should
not be affected significantly by the "size effect" noted by Banz [1981] and Reinganum
[1981], or the "January/turn-of-the-year effect" noted in Keim [1983] and Roll [1983]. In
fact, the latter would tend to induce a bias against supporting the main hypothesis of this
study, since the residual variance of the RSCV firms, which are smaller than the RLCV
firms, would tend to be more biased upward during the estimation and predisclosure periods
(see the results of Banz [1981] and Reinganum [1981]).
14 Additional comparisons between Grant's "exchange effect" and the "size effect" are
made in a separate study (Atiase [1984]).

15 A similar test was employed by May [1971].


16 The transformation of RIj,* to SRI1jt or SRI2i,. is an order-preserving transformation
for each firm i. Thus, the results of this test are the same for RIgt* SRI1jit or SRI2it,.

This content downloaded from 131.94.186.72 on Fri, 08 Jul 2016 22:27:21 UTC
All use subject to http://about.jstor.org/terms

32 ROWLAND KWAME ATIASE


3.2 FINAL REGRESSION MODELS AND HYPOTHESIS

For comparison, three regression models were specified as follows:

Model 1: RIo = /3o + f1LCVi + bi. (16)


Model 2: SRIlio = i0b + f13LCVi + Ai. (17)

Model 3: SRI2io = /30 + f31LCVi + bi. (18)


A one-sided test of significance was performed on the primary hypothesis for each regression model stated as:

Ho: A3 = 0
versus:

H1: A1 < 0.
3.3 FINAL REGRESSION RESULTS (INDIVIDUAL-LEVEL ANALYSIS)

The deciles of the estimated values of the variables for the final

regression models for LCVj, RIjo, SRIlio, and SRI2jo across the total
sample, the subsample of RSCV firms, and the subsample of RLCV firms
TABLE 5

Data of Final Regression Variables (Total Sample)

Decile LCV, RIho SRI1io SRI2jo


.10

1.384

0.039

0.049

0.048

.20

1.711

0.175

0.211

0.202

.30

1.978

0.439

0.420

0.420

.40

2.400

0.681

0.677

0.639

.50

2.995

1.015

1.007

0.983

.60

6.549

1.374

1.215

1.207

.70

6.738

1.913

1.718

1.705

.80

7.009

2.745

2.588

2.531

.90

7.749

4.485

4.245

4.011

Maximum 10.568 28.788 10.887 12.540

TABLE 6

Data of Final Regression Variables (RSCV Firms)

Decile LCV, RI0o SRI1io SRI2jo


.10

1.067

0.755

0.490

0.490

.20

1.384

1.014

0.789

0.728

.30

1.556

1.229

1.027

0.986

.40

1.700

1.505

1.193

1.144
1.411

.50

1.851

1.838

1.533

.60

1.978

2.132

1.951

2.008

.70

2.189

3.115

2.636

2.613

.80

2.400

4.384

3.889

3.708

.90

2.573

6.046

4.840

4.945

Maximum

2.995

28.788

10.887

12.540

This content downloaded from 131.94.186.72 on Fri, 08 Jul 2016 22:27:21 UTC
All use subject to http://about.jstor.org/terms

PREDISCLOSURE INFORMATION 33
TABLE 7

Data of Final Regression Variables (RLCV Firms)

Decile

LCV,

RIo

SRI1,o

SRI2jo

.10

6.424

0.005

0.009

.20

6.549

0.039

0.049

0.006

0.048

.30

6.621

0.108

0.116

0.116

.40

6.738

0.174

0.211

0.202

.50

6.869

0.345

0.407

0.382

.60

7.009

0.505

0.599

0.595

.70

7.309

0.712

0.889

0.889

.80

7.749

1.185

1.327

1.363

.90

8.823

2.078

2.212

2.206

Maximum

10.568

4.829

5.936

5.936

TABLE 8
Final Regression Results (t-statistics in parentheses)

Regression

Model No.go0R
1

.................

4.019
(9.71)a

.................

2.763

-0.440

f1i
0.144

(5.74)a

-0.254

0.136

(11.29)a (5.53)a
3

.................

2.687

-0.244

0.127

(11.04)a (5.33)a
One-sided test:

a Statistically significant at an a-level of less than or equal to 0.0001.

Model 1: RI0o = gdo + O3LCVi + bi.


Model 2: SRI1,o = gdo + f1 LCVX + bi.
Model 3: SRI2,o = gdo + A1 LCVX + bi.

are reported in tables 5, 6, gnd 7 respectively. The reported values show


that although the SRIl0 (RIo) for the total sample were 1.611 (2.026),

only a little over 50% of the total sampled firms had SRI1io (RIjo) greater
than the theoretical expectation of 1.000 (table 5). Further, while the
SRIlio (RIio) of over 70% (80%) of the RSCV firms (table 6) were above

the expected value, over 70% (almost 80%) of the SRIlio (RIjo) of the
RLCV firms (table 7) were below the expected value.
The final regression results reflect the general relationship between
the report week price revaluations, earnings announcements, and firm
capitalization. The results of the three regression models17 are reported
in table 8. The results of the regression models were supplemented with
correlation tests based on the nonparametric Spearman rank-order correlation coefficient'8 and on the parametric Pearson product-moment
17 Diagnostic tests suggest that the regression models are fairly well specified, with little
evidence of heteroscedasticity and only moderate departures from normality (based on the
Kolmogorov-Smirnov goodness-of-fit test).

18 Nonparametric tests based on the equally powerful Kendall rank-order correlation


coefficient were also conducted. The results obtained (in terms of the statistical significance

levels of the negative coefficients) were the same as those for the estimated Spearman
rank-order correlation coefficients. Thus, only the latter results are reported here.

This content downloaded from 131.94.186.72 on Fri, 08 Jul 2016 22:27:21 UTC
All use subject to http://about.jstor.org/terms

34 ROWLAND KWAME ATIASE

TABLE 9

Results of Correlation Coefficients Between LCVi (the Size Variable) and Report Week
Revaluation Indices (Significance levels in parentheses)
Spearman Pearson
Dependent Rank-Order Product-Moment
Variables Correlation Correlation
Coefficient Coefficient

RI0o

.........

-0.548

(0.0001)*

SRIlio

........

-0.405

(0.0001)*

SRI2io

........

-0.401

(0.0001)*

-0.380
(0.0001)*

-0.368
(0.0001)*

-0.357
(0.0001)*

* Significance level.

correlation coefficient between LCVi and each of the dependent variables


(RIo, SRIlio, and SRI2io). Table 9 reports these results.

In all of the final regression models, the estimated coefficients (i13) of


the capitalized value variable were negative and statistically significant
at an a-level of less than or equal to 0.0001. These results are consistent

with those of both the nonparametric Spearman rank-order correlations

and the parametric Pearson product-moment correlations between LCVi


and each of the report week revaluation indices. That is, all of the
estimated correlation coefficients are negative and statistically significant at an a-level of 0.0001. Thus, these results strongly support the

hypothesis that the degree of a security's price revaluation in response


to its second-quarter earnings announcement is inversely related to the
capitalized value of the firm, other things equal.

As expected,19 the estimated coefficient of the intercept term, f03, in


each of the three regression models is positive and greater than zero at
an a-level of 0.0001.

The final regression results are quite robust across the various model
specifications, although model 2 (dependent variable is SRIlio) slightly

outperforms model 3 (dependent variable is SRI2io) in terms of both the


t-statistics associated with the estimated coefficients and R2. Recall that
SRI1io was standardized for the differences in the time lag between the
financial statement date and the report weeks of the various sampled
firms, whereas SRI2io was not. We would expect firm-specific predisclosure information production and dissemination closer to the report weeks
to be weighed more heavily by the market.

4. Conclusions
Subject to limitations that may be due to sample sizes, selection
criteria, and assumptions underlying the model specifications, the em19 In a simple regression model where, for example, Y = do + f1X + E, o - Y - flX.
Thus, since d, is negative, do = F + I d, 1. X, which should be greater than Y.

This content downloaded from 131.94.186.72 on Fri, 08 Jul 2016 22:27:21 UTC
All use subject to http://about.jstor.org/terms

PREDISCLOSURE INFORMATION 35

pirical results presented in section 3 suggest: (1) that for the entire
sample, the average security price revaluation in the second-quarter

earnings report weeks, irrespective of the extent of predisclosure infor-

mation production and dissemination about the various sampled firms,


was significantly greater than the average security price revaluations

over the estimation period, the predisclosure period, and in the other
weeks in the report period. In this regard, the results presented in this
study are consistent with prior empirical findings of Beaver [1968] and

May [1971]; and (2) that the degree of a security's price revaluation in
response to its second-quarter earnings report is inversely related to the
capitalized value of the firm, other things being equal. This phenomenon
is attributed to differential levels of (private) predisclosure information
production and dissemination on firms of different capitalized values.
A basic implication of the results is that future empirical research
designs requiring control for (private) predisclosure information should
control for the capitalized value of sampled firms. This is particularly
important for studies aimed at testing the "economic consequence" and

/or "effectiveness" of new public information disclosure rules. For example, a finding of no effect at the time a new public disclosure is made
is consistent with several hypotheses, only one of which is that the new

public disclosure conveys no information pertinent to the assessment of

capital asset prices. It may be that the new disclosures have information
content for some firms (lower capitalized values) but not others.
As a final comment, theoretical and empirical research about finer
partitions of firm-specific characteristics that are systematically related
to predisclosure information as well as the economic circumstance(s)
that give rise to such cross-sectional correlation(s) would appear to be
fruitful avenues to pursue in the future.
REFERENCES
ATIASE, R. K. "Predisclosure Informational Asymmetries, Firm Capitalization, Financial

Reports, and Security Price Behavior." Ph.D. dissertation, University of California,


Berkeley, 1980.

. "Market for Shares, Firm Size, Earnings Reports, and Security Price Behavior:

Some Additional Evidence." Working paper, University of Florida, June 1984.


BALL, R., AND P. BROWN. "An Empirical Evaluation of Accounting Income Numbers."
Journal of Accounting Research (Autumn 1968): 159-77.

BANZ, R. W. "The Relationship Between Return and Market Value of Common Stocks."
Journal of Financial Economics (March 1981): 3-18.

BEAVER, W. "The Information Content of Annual Earnings Announcements." Journal of


Accounting Research (Supplement 1968): 67-92.

BROWN, P., AND J. W. KENNELLY. "The Information Content of Quarterly Earnings: An


Extension and Some Further Evidence." Journal of Business (July 1972): 403-15.

COLLINS, D. W., AND W. T. DENT. "A Comparison of Alternative Testing Methodologies


Used in Capital Market Research." Journal of Accounting Research (Spring 1984): 4884.

COLLINS, D. W., M. ROZEFF, AND W. SALATKA. "The SEC's Rejection of SFAS No. 19:
Tests of Market Price Reversal." The Accounting Review (January 1982): 1-17.
FREEMAN, R. N. "The Association Between Accounting Earnings and Security Returns for

This content downloaded from 131.94.186.72 on Fri, 08 Jul 2016 22:27:21 UTC
All use subject to http://about.jstor.org/terms

36 ROWLAND KWAME ATIASE

Large and Small Firms." Working paper, University of California, Berkeley, December
1983.

GRANT, E. B. "Market Implications of Differential Amounts of Interim Information."


Journal of Accounting Research (Spring 1980): 255-68.

KEIM, D. B. "Size Related Anomalies and Stock Return Seasonality: Further Empirical
Evidence." Journal of Financial Economics (June 1983): 13-32.

KING, B. F. "Market and Industry Factors in Stock Price Behavior." Journal of Business
(January 1966): 139-90.

MAY, R. "The Influence of Quarterly Earnings Announcements on Investor Decisions as

Reflected in Common Stock Price Changes." Journal of Accounting Research (Supplement


1971): 119-63.

OPPONG, A. "Information Content of Annual Earnings Announcements Revisited." Journal


of Accounting Research (Autumn 1980): 574-84.
PATELL, J. M. "Corporate Forecasts of Earnings per Share and Stock Price Behavior:
Empirical Tests." Journal of Accounting Research (Autumn 1976): 246-76.
REINGANUM, M. "Misspecification of Capital Asset Pricing: Empirical Anomalies Based

on Earnings' Yields and Market Values." Journal of Financial Economics (March 1981):
19-46.

RICHARDSON, G. "The Information Content of Annual Earnings for Large and Small Firms:
Further Empirical Evidence." Working paper, University of British Columbia, 1984.

Ro, B. T. "Firm Size and the Information Content of Annual Earnings Announcements."
Working paper, Purdue University, March 1984.

ROLL, R. "Vas Ist Das? The Turn-of-the-Year Effect and the Return Premia of Small
Firms." Journal of Portfolio Management (Winter 1983): 18-28.

ROSENBERG, B. "Extra-Market Components of Covariance in Security Markets." Journal


of Financial and Quantitative Analysis (March 1974): 263-74.
, AND V. MARATHE. "Common Factors in Security Returns: Microeconomic Determinants and Macroeconomic Correlates." Working paper, Institute of Business and
Economic Research, University of California, Berkeley, 1976.
SHARPE, W. F. "Capital Asset Prices: A Theory of Market Equilibrium Under Conditions
of Risk." Journal of Finance (September 1964): 425-44.

This content downloaded from 131.94.186.72 on Fri, 08 Jul 2016 22:27:21 UTC
All use subject to http://about.jstor.org/terms

S-ar putea să vă placă și