|
Citation |
- Permanent Link:
- http://ufdc.ufl.edu/AA00003865/00001
Material Information
- Title:
- Newness and imputed accuracy of management forecasts an empirical investigation
- Creator:
- Beshara, Robert L ( Robert Labib ), 1944-
- Publication Date:
- 1981
- Language:
- English
- Physical Description:
- vii, 142 leaves : ill. ; 28 cm.
Subjects
- Subjects / Keywords:
- Analytical forecasting ( jstor )
Earnings forecasting ( jstor ) Error rates ( jstor ) Forecasting standards ( jstor ) Information content ( jstor ) Investors ( jstor ) Net income ( jstor ) Null hypothesis ( jstor ) Statistical forecasts ( jstor ) Stock prices ( jstor ) Business forecasting ( lcsh ) Income forecasting ( lcsh )
- Genre:
- bibliography ( marcgt )
theses ( marcgt ) non-fiction ( marcgt )
Notes
- Thesis:
- Thesis (Ph. D.)--University of Florida, 1981.
- Bibliography:
- Includes bibliographical references (leaves 137-141).
- General Note:
- Typescript.
- General Note:
- Vita.
- Statement of Responsibility:
- by Robert L. Beshara.
Record Information
- Source Institution:
- University of Florida
- Holding Location:
- University of Florida
- Rights Management:
- Copyright [name of dissertation author]. Permission granted to the University of Florida to digitize, archive and distribute this item for non-profit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.
- Resource Identifier:
- 000297239 ( ALEPH )
ABS3612 ( NOTIS ) 08356777 ( OCLC )
|
Downloads |
This item has the following downloads:
|
Full Text |
NEWNESS AND IMPUTED ACCURACY OF MANAGEMENT
FORECASTS: AN EMPIRICAL
INVESTIGATION
by
ROBERT L. BESHARA
A DISSERTATION PRESENTED TO THE GRADUATE
COUNCIL OF THE UNIVERSITY OF FLORIDA IN PARTIAL
FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY
UNIVERSITY OF FLORIDA
1981
ACKNOWLEDGEMENTS
The author gratefully acknowledges the assistance and
support of the Supervisory Committee in the completion of this
dissertation.
Dr. Rashad Abdel-Khalik provided valuable guidance
and criticisms. Dr. Bipin Ajinky offered assistance and encouragement
far beyond the call of duty.
D. Robert Radcliffe offered helpful advice and
suggestions.
Gratitude is also expressed to Dr. Barnhill for his
editorial comments, and Ms. Susan Robinson for the cheerful typing
assistance.
Special acknowledgements are made to the Clarkson and
Gordon Foundation and Carleton University for providing financial
assistance.
TABLE OF CONTENTS
Page
ACKNOWLEDGMENTS............................................... ii
ABSTRACT........................ .................... ........... v
Chapter
I. STATEMENT OF THE PROBLEM AND PURPOSE OF THE
DISSERTATION.......................................... 1
Introduction........ ....... ...... ................... 1
Statement of the Problem............................. 4
Purpose of the Dissertation.................... ....... 8
II. PRIOR RESEARCH ON THE INFORMATION CONTENT OF
MANAGEMENT FORECASTS OF EARNINGS........................ 10
The Foster Study..................................... 10
The Gonedes, Dopuch and Penman Study.................. 13
The Patell Study.................................... 16
The Jaggi Study..................................... 20
The Penman Study.................................... 21
The Nichols and Tsay Study........................... 26
Summary and Limitations of Existing Research......... 30
Notes.. ...... ............................ .. ...... 35
III. DATA, METHODOLOGY, AND STATISTICAL TESTS................ 36
Data ................................................. 38
Definitions........................... ............... 42
Experimental Portfolios.............................. 42
Control Portfolios.................................. 44
Statistical Tests.................................... 46
Residual Analysis................................... 46
A Standardized Prediction Errors................ 47
B Unstandardized Prediction Errors .............. 52
Analysis of Variance................................ 52
Hypotheses....................... ............. ....... 56
A Testing the Effect of Forecasting Action...... 56
B Testing the Effect of Newness of Earnings.....
Forecasts..................................... 58
C Testing the Effect of Imputed Accuracy........ 59
D Testing the Effect of Management Forecasts
on Other Firms in the Same Industry............ 61
Notes................................................ 63
Chapter Page
Page
IV. RESULTS ............ .... ............................... 64
Introduction.......... ....... ...................... 64
Total Sample........................................ 65
The Effect of the Forecasting Action................. 68
The Effect of Newness of Management Earnings
Forecasts.... ...................................... 76
A The Positive Portfolio........................ 76
B The Negative Portfolio........................ 85
The Effect of Imputed Accuracy....................... 93
Positive Forecasts High Imputed Accuracy........ 93
Positive Forecasts Low Imputed Accuracy.......... 100
Comparison Between the High-level and the Low-
level Imputed Accuracy............................ 109
Negative Forecasts High Imputed Accuracy........ 113
Negative Forecasts Low Imputed Accuracy......... 120
Comparison Between the High-level and the Low-
level Imputed Accuracy........................... 125
V. SUMMARY AND CONCLUSIONS.................................. 131
BIBLIOGRAPHY.................................................. 137
BIOGRAPHICAL SKETCH... ....................................... 142
Abstract of Dissertation Presented to the Graduate Council of
the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy
NEWNESS AND IMPUTED ACCURACY OF MANAGEMENT FORECASTS:
AN EMPIRICAL INVESTIGATION
By
Robert L. Beshara
December, 1981
Chairman: Dr. Rashad Abdel-Khalik
Major Department: Accounting
The purpose of this dissertation was to empirically
examine the information content of management forecasts of earnings.
The following four research questions were examined:
1. Do management forecasts of earnings convey information in
excess of what is already available to market participants?
2. Does the forecasting action by itself convey favorable
information to investors?
3. Does the information content depend upon the level of
newness and imputed accuracy ascribed to management
earnings forecasts?
4. Do management earnings forecasts convey information
pertinent to establishing equilibrium prices of non-fore-
casting firms in the same, or similar, industry?
The experimental sample consisted of 401 management earnings forecasts
covering the period 1970-77. The total sample of management earnings
forecasts was classified into neutral, positive, and negative fore-
casts according to the magnitude and the sign of the new information
revealed by each forecast. Two control samples were utilized to
detect industry-wide and economy-wide effects. Results of the study
indicate that:
1. The positive (negative) news of forecasts induced an upward
(downward) stock price revisions during the forecasts
announcements week. This revision was found to be
statistically significant.
2. The forecasting action by itself induced an upward stock
price revision during the forecast announcement week. Such
revision, however, was neither sufficiently strong nor was
it the general pattern of the neutral portfolio to be
considered statistically significant.
3. Stock price revisions associated with the high-level
imputed accuracy portfolio were considerably higher than
those associated with the low-level imputed accuracy
portfolio for the positive news management earnings
forecasts but not for the negative news forecasts.
4. Negative news management earnings forecasts seem to convey
industry-wide information, while positive news forecasts
are perceived to be firm specific events. This conclusion,
however, should be tampered by the fact that the sample
size for this classification of forecasts is relatively
small.
CHAPTER I
STATEMENT OF THE PROBLEM AND
PURPOSE OF THE DISSERTATION
Introduction
The recent attempt by the Securities and Exchange Commission to
regulate the disclosure of management earnings forecasts has received
considerable attention at both the pragmatic and the academic levels.
The widely held belief that management forecasts of earnings are
relied upon in investment decisions and the assertion that management
has access to information not available to outsiders have motivated
several empirical research studies on the subject.
Disclosure regulations of management forecasts of earnings
involve two closely related issues. The first issue is the extent to
which management earnings forecasts convey information not already
available to investors. The second issue is whether disclosure
regulations of earnings forecasts by management are consistent with
the optimal allocation of resources in society. Gonedes et al. (1976)
point out the relationship between these two issues as follows
(p. 90):
If one is confident that forecasts of, say, income
convey no information pertinent to valuing firms,
- 1 -
-2 -
then the entire debate over forecasts disclosure
is of little interest, insofar as external
accounting is concerned. A necessary condition
for public disclosure of forecasts to be
consistent with Pareto optimality is that the
forecasts convey information pertinent to resource
allocation.
Empirical research directly concerned with the information
content of management earnings forecasts has been scanty. Most of the
existing research has focused on either the absolute accuracy of
management forecasts or the relative accuracy of management forecasts
as compared to historically-based models and/or analysts' forecasts.
Confirming this observation are Gonedes et al. when they state
(p. 93):
The available empirical work on information
content deals with earnings forecasts voluntarily
made available by firms. The bulk of this work
examines the prediction performance or 'accuracy'
of earnings forecasts. The implicit assumption
here is that greater accuracy, however measured,
corresponds to greater information content.
Empirical research dealing with management forecasts accuracy
suffers from a major deficiency. That deficiency is the lack of a
theoretical structure capable of explaining the relationship, if any,
between ex-post accuracy and the information content issue. Gonedes
et al. (p. 94) point this out when they state:
There is a more fundamental deficiency in these
prediction performance studies and the a priori
arguments on information content. Specifically,
they are not based upon any explicit theoretical
structure that connects their framework to
resource allocation decisions made under
uncertainty. In order to determine whether
forecasts convey information pertinent to valuing
a firm, one needs a theoretical model that
identifies the determinants for firms' values,
conditional on the way in which resource
allocation decisions are made.
-3-
The efficient-markets hypothesis discussed by Fama (1970) and
the capital asset pricing model developed by Sharpe (1964) and Lintner
(1965) provide the theoretical structure for studies dealing with the
effect of new information on equilibrium stock prices. The
semi-strong form of the efficient capital markets maintains that
prices of securities reflect all publicly available information and
that the market reaction to new information occurs instantaneously in
an unbiased fashion. The capital asset pricing model, under certain
restrictive assumptions asserts that the price (return) of a security
depends on the risk-return trade-off set by the market and the
probability distribution of the expected returns on the security
considered jointly with the market portfolio. Beaver (1968, p. 68)
considers an item to have information content "if it leads to a change
in investors' assessment of the probability distribution of future
returns (or prices) such that there is a change in equilibrium value
of the current market prices."
Studies by Foster (1973), Gonedes et al. (1976), Patell (1976),
Penman (1978), Jaggi (1978), and Nichols and Tsay (1979), comprise the
major empirical research on the information content of management
forecasts. All of these studies utilized the theoretical structure
provided by the efficient-markets hypothesis and the capital asset
pricing model. Results of these studies are somewhat conflicting and
incomplete. Several research questions on the information content of
voluntarily disclosed management forecasts remain unexplored.
-4-
Statement of the Problem
The information content of management earnings forecasts is a
function of several attributes. Patell (1976, p. 248) points out
that, "the extent to which an earnings forecast alters investors'
beliefs may be a function of (at least) three attributes." These are:
1 the imputed accuracy of the forecast,
2 newness of the forecast; that is, the extent to which the
forecast's content is not already available to market
participants from other sources, and
3 the motive underlying the voluntary disclosure of earnings
forecasts.
The above attributes are neither independent of each other, nor
are they totally exhaustive. For example, it is reasonable to expect
imputed accuracy to be correlated to the level of newness of the
forecast. Also, the imputed accuracy of forecasts may depend upon
industry volatility and the motive for forecast disclosure. Imputed
accuracy of management forecasts at the time of forecast disclosure is
difficult to assess and it has not been considered in the existing
literature.
The newness of management earnings forecasts has been typically
based upon naive forecasting models. The available research on the
information content of management forecasts defines newness as the
deviation of the forecast from the naive modelss. Foster (1973),
-5-
Patell (1976), and Penman (1978) utilized historically-based naive
models as a proxy for current market expectations of earnings. These
models, although practically convenient, do not incorporate existing
and expected factors affecting the firm, the industry in which the
firm operates, and the economy at large. These factors, however,
directly influence investors' expectations of earnings and should not
be ignored.
The motive attribute has been examined, albeit not
extensively. The signalling theory developed by Spence (1973, 1974),
Riley (1975, 1977), and extended to disclosure regulations in
financial markets by Ross (1977) provides a theoretical structure for
voluntary disclosure of inside information by corporate executives.
The basic contention of the signalling theory is that in competitive
markets incentives exist for management to publicly disclose inside
information relevant to the valuation of their firms by outsiders.
These incentives will exist in competitive markets irrespective of the
classification of the information as 'good'or'bad'news. The
incentive-signalling theory is best summarized by Ross (1977, p. 20),
as follows:
In general, there is a hierarchy of firms from
best to worst based on the relative changes in
values that would occur if their inside informa-
tion was made public. The incentive-signalling
mechanism provides a structure that managers use
to disclose their information in such a way that
outsiders in the market believe it. Those with
the best news distinguish their firms from those
with the next best and so on down the line. At
the bottom of the hierarchy, those with the worst
news would like to suppress it, but since it is
-6-
not in their interest to offer guarantees provided
by those with better news, the worst news will
also be signalled.
Several technologies are available to management in competitive
markets to sort their firms on value relevant attributes. Public
disclosure of quantitative forecasts of earnings per share is one
possible technology. Change in dividend policy, reporting quarterly
earnings, changes in capital structure, and providing a qualitative
statement regarding future earnings are examples of other signalling
technologies. On the other hand, earnings forecast disclosure can not
be totally ascribed to deliberate revaluation of the firm by manage-
ment. Other motives for forecast disclosure such as the desire to
raise new capital, justification for market expansion, merger
decisions, pressure from the financial community, and other factors,
can not be ruled out.
Penman (1978) adopted the incentive-signalling theory to
examine the voluntary disclosure of forecasted earnings. More
specifically, the author posed the following research question (p. 2):
"Do investors receive forecast information potentially available from
all firms?" The empirical results of the study do not provide an
affirmative answer to this question. These results, however, are not
inconsistent with the incentive-signalling theory mainly due to the
fact that several signalling technologies are available to management.
Furthermore, the iterative mechanism of the incentive-signalling
theory will result in the screening of companies which do not issue
-7-
quantitative forecasts of earnings based upon their own signals as
well as the forecasting action of other firms. Penman (p. 46) points
out that:
S the firms which do not issue forecasts are,
in fact, screened out as a group by the
forecasting action of others. A 'no forecast' is
in fact a forecast that screens firms into this
group and value them accordingly.
The above discussion suggests that the information content of
management earnings forecasts has implications for non-forecasting
firms. This is conceptually appealing for two reasons. First, the
mechanism of the incentive-signalling theory is an iterative process
through which firms will find it advantageous to reveal their value
relevant attributes. This process need not take place on an
individual firm basis. Instead, it can take place through grouping of
firms according to similarity of value attributes. Second, on an a
priori basis, it can be argued that an earnings forecast by management
is a quantification of the expected impact of general economic
conditions on their firm. To the extent that other firms in the same
industry are affected by common economic forces, the forecast can also
be viewed as an assessment of the industry situation. As such,
management forecasts convey information about the forecasting firms as
well as the industry in which they operate. That is, the information
content of management forecasts may extend to other firms in the same
(or similar) industry. This, however, is an empirical question which
has not been examined in the current literature.
-8 -
Purpose of the Dissertation
The purpose of this dissertation is to examine the extent to
which management forecasts convey information not already available to
market participants. via other sources. The methodology utilized in
this study explicitly considers market expectations of earnings per
share immediately prior to the release of management forecasts.
Market expectations are measured by available analysts' forecasts.
Thus, newness is defined as the deviation between the management
forecast and analysts' forecasts available at the time of the
management forecast release. Imputed accuracy is measured by the
revision of analysts' forecasts towards the management forecast. In
effect, a forecast is considered to possess a high level of imputed
accuracy if analysts revise their estimates of earnings per share
closer to that of management.
This study provides empirical evidence aimed at answering the
following four related research questions.
1. Do management earnings forecasts provide information in
excess of that already available in the market and
presumably incorporated into analysts' forecasts?
2. Do management earnings forecasts which merely confirm
investors' expectations result in an upward stock price
revision? In other words, does the forecasting action by
itself convey favorable information to investors.
3. Does the information content, if any, depend upon the
levels of newness and imputed accuracy ascribed to manage-
ment earnings forecasts?
-9-
4. Do management earnings forecasts convey information
pertinent to establishing equilibrium prices of
non-forecasting firms in the same, or similar, industry?
CHAPTER II
PRIOR RESEARCH ON THE INFORMATION CONTENT
OF MANAGEMENT FORECASTS OF EARNINGS
Empirical research directly concerned with the information
content of management forecasts has been relatively scarce. As stated
earlier, the majority of the empirical research to date deals with the
absolute and relative accuracy of management forecasts on an ex-post
basis.
Six published studies comprising the major empirical research
on the information content of earnings forecasts by managements are
identified and reviewed in this chapter. The purpose of this review
is to highlight the scope and limitations of the existing research on
the topic.
The Foster Study (1973)
This study examined the information content of estimates of
annual earnings per share by company officials. These estimates were
made after the end of the fiscal year but prior to the publication of
- 10 -
- 11 -
preliminary earnings per share (EPS). Thus, in effect Foster examined
the information content of unaudited EPS. This is evident from the
level of accuracy of EPS estimates. The average absolute percentage
error was 1.8 percent (p. 27). For all practical purposes, however,
such a minute error could be the result of the rounding of EPS
estimates by company officials.
The study examined 68 EPS estimates for the period 1968-1970.
The majority of estimates (87 percent) had a time-lapse between the
EPS estimates and the preliminary earnings announcement of 21 trading
days. The mean and the median for the time-lapse were 17 and 18
trading days respectively.
Foster tested for the individual and aggregate market behavior
using a methodology similar to that employed by Beaver (1968). Both
tests confirmed the hypothesis that the unaudited EPS convey
information to investors. Thus, abnormal returns can be achieved by
employing a trading strategy "based on knowledge of EPS estimates five
days before it is publicly released ." (p. 33).
The study utilized five annual and five quarterly naive models.
For each naive expectations model a first difference model was con-
structed. This resulted in a total of twenty models, ten level and
ten first difference. Forecasts were classified as positive (nega-
tive) news depending upon the sign of deviation from the naive expec-
tations model. Two purchase and short-sale strategies were adopted
for the positive and negative news forecasts respectively. The first
- 12 -
strategy was adopted five days prior to the EPS announcement to the
date the preliminary earnings report was announced. The second
strategy was adopted from the date the EPS estimate was announced to
the date of the preliminary earnings report.
The average abnormal return for the first trading strategy was
1.61 percent for the ten annual models and 1.48 percent for the
quarterly models. The average abnormal return for the second strategy
was much lower; 0.31 percent for the ten annual models and also 0.31
percent for the ten quarterly models. This result (p. 33) is
explained as follows:
Up to the time the company officials announce the
EPS estimate, there is still opportunity for an
abnormal return. ... .Once the official's EPS
estimate is publicly released, there is a stock
price adjustment such that trading strategy based
on that information yields lower abnormal returns.
The author points out, however, that abnormal returns "do not
indicate the extent to which the observed results were a general
pattern for all firms or was due to extreme observations" (p. 34). A
chi-square test was used to examine the frequency of abnormal
returns. One out of twenty models was significant at the .01 level;
two other models were significant at the .05 level and ten models were
significant at the .1 level. This result led to the conclusion that
estimates of earnings per share possess information content and that
the effect is a general pattern in the sample.
- 13 -
This study clearly highlights the effect of naive expectations
models on results obtained. Different models yield different results.
Although the author utilized a large number of models, the effect of
possible mis-classification of forecasts on the results of the study
remains unknown.
The Gonedes, Dopuch, and Penman Study (1976)
The purpose of this study was to evaluate the information
content of management earnings forecasts. The authors utilized a
methodology that (p. 107) "involves time-series analysis and tests for
information content." This required a series of forecasts for each
firm in the sample. Due to the fact that management earnings fore-
casts are not available from the same source and for several con-
secutive time periods, the authors used analysts' forecasts as
"proxies" for management forecasts.
The sample included 148 firms drawn from the set of firms
covered by the Standard and Poors Earnings Forecaster. One criterion
used in selecting the firms was that "each firm had to have con-
secutive forecasts from at least one source during the last half of
1967 and the last half of 1968" (p. 110). This approach led to
"having twenty-four forecasts for each of the 148 firms in the
sample" (p. 110).
- 14 -
Each firm's forecast was scaled by "the per share closing price
of the firm's common stock on the tenth trading day preceding the
forecast's publication date "(p. 113). Scaled forecasts were classi-
fied into four groups quartiless) according to the magnitude of the
scaled forecast. Portfolios were formed for each forecast date and
for each quartile (p. 113) "of the cross-sectional distribution of
scaled forecasts." One control portfolio, of equal size, was con-
structed to match each sampling portfolio. The result was 192 port-
folios (24 forecast dates x 4 quartiles x 2 portfolios), each composed
of 37 shares. All portfolios were minimum variance portfolios with
beta equal to unity.
The methodology used in this study is discussed in Gonedes and
Dopuch (1974), Gonedes (1975), and summarized in the current study
(pp. 111-114). Briefly, the approach involves a comparison of two
distribution functions; the first is the distribution of returns on
the sampling portfolios, and the second is the distribution of returns
on the control portfolios. Based on the assumption that the distri-
bution of returns is multivariate normal, the comparison involves
tests of equality of means and variances. However, an inequality
between the means of the two distributions is a sufficient condition
for the inequality of the two distributions. Accordingly, "only
results for tests on means are reported below" (p. 112).
Results of the study can be summarized as follows:
- 15 -
1 the null hypothesis of no information content was rejected
for the total sample. The value of the F statistic was
"substantially above the .995 fractile ." (p. 123).
2 disaggregating the data, the null hypothesis was rejected
for the first quartile (lowest scaled forecasts), but not
for the other three quartiles. This result led the authors
to conclude (p. 133) that:
our empirical results on income forecasts
are inconsistent with the statement that those
forecasts convey no such information. Moreover,
it appeared that much of the information content
of forecasted earnings can be ascribed to the
implication of extremely low (scaled) forecasts.
Forecasts not in this class seemed to reflect
little information beyond that already available,
such as the information reflected in available
forecasts of macro-economic conditions. This
observation seems to increase the importance of
assessing the extent to which substitutes are
available for forecasts of firms' earnings and the
properties of these substitutes.
3 testing for the effect of portfolio risk, the authors
reconstructed all sampling and control portfolios for
different levels of risk. They observed (p. 128) that:
In short it appears that the major effect of
letting risk depart from unity is to increase
dispersion, as measured by the generalized
variance. That implies that the value of risk
other than unity induces less powerful tests of
information content hypothesis than does risk
equal to unity. This is the major reason for our
heavy reliance on results for risk equal to unity.
- 16 -
4 disaggregating the total period into eight subperiods (30
observations each), the test results exhibited "a fair
amount of variability" (p. 130). This variability was
attributed, in part, to the increase in the estimated
standard error of the mean due to the decrease in the
number of observations. Although subperiod results "do not
seem to be consistent with the null hypothesis" (p. 130),
the value of the F statistic was below the .9 fractile for
four subperiods.
The above findings indicate that management forecasts of
earnings convey information not already available to investors. These
findings, however, should be tempered by the fact that the authors
utilized a proxy for management forecasts. Also, the data
were restricted to forecasts made during the second half of the fiscal
year. More importantly, however, the forecasts used were not
previously unavailable either from the same source or from some other
source (p. 109). That is, most of the forecasts were not first time
forecasts and were available to investors.
The Patell Study (1976)
This study examined 336 voluntary management forecasts of
annual EPS published in the Wall Street. Journal for the period 1963-
1968. Approximately 46 percent of the forecasts accompanied first and
second quarter results. The sample included 204 point or convertible
- 17 -
to point estimates and 132 of the form of greater (less) than X.XX
dollars (p. 250). the methodology utilized is essentially the same as
that of Beaver (1968) and May (1971). The study focused upon the
prediction error (abnormal performance), using the familiar market
model, for the forecast week as well as eight weeks prior to and eight
weeks after the forecast week.
The null hypothesis that the expected value of the prediction
error during the forecast week is equal to zero was rejected at a
statistically significant level, .0004. The null hypothesis could"not
be rejected at the .05 probability level for any other week in the
test period" (p. 260). Disaggregating the data the following results
were observed:
1 the mean prediction error for utility firms was much weaker
than that for non-utility firms.
2 the mean prediction error for forecasts that accompanied
quarterly earnings was weaker than it was for those that
were announced separately (p. 266).
3 the mean prediction error for open ended forecasts was
stronger than that for point estimate forecasts (p. 262).
4 the strongest mean prediction error was for forecasts
released during the third quarter. The prediction error of
the first and the second quarters exhibited a much weaker
effect and the null hypothesis of expected prediction error
- 18 -
equal to zero could not be rejected for the second quarter
at the .1949 level (see Table 4, p. 266).
Note also that first and second quarter forecasts account for
16 percent and 51 percent of the total sample respectively. The
explanation provided by the author for these results was that
(p. 266):
The firm-specific price changes are less signifi-
cant for first-quarter disclosure than for the
third-quarter disclosure, perhaps due in part to
greater overall market volatility in the first
quarter of December 31 fiscal years. However, the
second fiscal quarter results are even less
significant, and this cannot be easily attributed
to general market conditions.
The above results are somewhat perplexing and difficult to
explain given the methodology utilized in the study. Further
examination of the attributes of forecasts released at different time
periods (quarters) may provide an insight into the findings of this
study. The study intentionally avoided dealing with the imputed
accuracy attribute (p. 248), and the newness attribute was linked to
naive forecasting models. Accordingly, whether or not the varying
effects of each attribute provide a reasonable explanation of the
results obtained in this study remains an unanswered question.
Finally, the study examined the effect of deviations of
forecasts from investors' expectations. Three historically based
naive models were used as measures of investors' expectations.
Deviations from each model were ranked into high and low deviation
groups. The results were (p. 270):
- 19 -
The firms in the top 25 percent (firm forecasts
most exceeded naive expectations) experienced
significantly positive Uit values during the week
of forecast disclosure. However, the firms in the
bottom 25 percent (firm forecast was furthest
below naive expectations) also experienced (on
average) positive Uit; and for two of the three
naive forecasts, the average Uit had a positive
magnitude with a less than .02 probability of
occurrence.
Further examination of the forecast release week indicated that
(p. 270):
Thus, when attention is restricted to only the week
of forecast disclosure itself, there appears to be
very little to distinguish between forecasts on the
basis of sign or magnitude of the change in
investors expectations.
Although the sign of the prediction errors was positive for
both groups during the forecast week, examination of the eight week
period preceding the disclosure of the forecast revealed that
(p. 270):
Those firms whose management forecasts exceeded
market estimates appear to enjoy generally positive
price relative residual during the two months
proceeding the forecast announcement, while firms
whose forecasts fall short of market estimates
experience generally negative residuals during the
proceeding two months. However, the price-relative
prediction errors at the immediate announcement date
are positive for both sets of firms.
These findings were interpreted by Patell (p. 273) to be con-
sistent with the hypothesis that substitute sources of information are
publicly available to investors. However, the positive price revision
for the negative group during the forecast announcements week is
difficult to explain. One possible explanation offered by the author
- 20 -
(p. 273) is that the act of voluntary disclosure of earning forecasts
conveys information to investors. Whether or not the forecasting
action by itself conveys information is a research question that
requires further investigation.
The Jaggi Study (1978)
This study utilized daily residuals to examine the information
content of management forecasts of earnings. The sample consisted of
144 forecasts published during the first four calendar months of each
year for the period 1971-1974. The null hypothesis of no information
content was tested under the assumption of a symmetric stable distri-
bution. The daily residual for each stock was tested individually for
significance. This testing procedure covered the forecast date as
well as the ten days preceding and the ten days following the forecast
announcement date. The number of significant residuals for each day
was then tested under the assumption of a binomial distribution. The
author concluded that (p. 964):
The results show that statistically significant
price adjustments were evident on the day of
announcement of management forecasts and also
before and after this day.
Three annual naive forecasting models were used as a proxy for
market expectations. Forecasts were classified as positive or
negative depending upon the sign of deviation from the naive model.
Adopting a buy and short-sale strategy resulted in an abnormal return
for both groups.
- 21 -
Deviations from each naive model were ranked in defending order
and separated into two equal portfolios--top and bottom deciles. The
buy and short-sale strategy was repeated for stocks contained in both
the top and bottom portfolios. "This was motivated by the need to
determine whether stock market price behavior is directly related to
differences in expectations and forecasts" (p. 966). Results do not
indicate that the magnitude of deviation from naive models is related
to the size of abnormal returns. That is, the adoption of the buy and
short-sale strategy resulted in similar abnormal returns for both high
and low deviations from market expectations as measured by naive
models. Unfortunately, the study did not report the levels or the
signs of deviations from naive models for top and bottom portfolios.
Accordingly, interpretation of these results are somewhat difficult.
The author (p. 966) carefully provides the following interpretation of
the results:
In view of these results, it can be inferred that
voluntary disclosure of earnings forecasts may
provide some additional information to investors.
A caveat, however, is that even though price
adjustments were observed, this does not neces-
sarily imply that the contents of the forecasts
caused investors to revise their expectations.
These adjustments may be caused by other factors,
or even by the mere act of voluntary earnings
forecast announcements.
The Penman Study (1978)
This study examined two research questions. The first question
addressed the matter of full disclosure of earnings forecasts by all
- 22 -
firms in the economy. The second question examined the information
content of management forecasts of earnings.
The full disclosure is an adaptation of the incentive-sig-
nalling theory, mentioned earlier, to the capital market setting. The
theoretical argument developed by Penman is essentially the same as
that of Ross (1977). According to the incentive-signalling theory all
firms in the economy, except the one with the worst news in the
hierarchy, will be induced to disclose information relevant to the
valuation of their firms. The worst news firm will be revealed by
exception. Thus, full disclosure of inside information will result.
Penman examined the claim that "managements only publish fore-
casts when they have 'good news' and hence investors only receive
forecast information about firms which are doing relatively well"
(p. 5). In order to determine whether a firm is doing 'relatively
well' the author utilized a historically based martingale model with
drift as an earnings expectation model. The deviation between the
actual earnings for the year immediately prior to the forecast year
and the expectation model was calculated for 576 firms in the experi-
mental sample. The earnings deviations for each firm were then
standardized by the variance of prior years earnings. The standard-
ized deviations were ranked in defending order and divided into two
equal portfolios representing the top and bottom halves. Likewise,
newness of an earnings forecast was defined as the forecast deviation
from the expectation model. Following the same procedure, each of the
- 23 -
previous two portfolios were further divided into two portfolios based
on the ranked value of the standardized forecast deviations. This
classification scheme resulted in four portfolios of equal size.
The full disclosure issue was tested by comparing returns on
experimental and control portfolios. The total market index was used
as the control portfolio. Following Gonedes (1974, 1975) and Gonedes
et al. (1976), the author utilized the Hotelling T2 test as follows
(p. 40):
By using equal weights, the signals are 'averaged
out' so that it can be observed whether the infor-
mation sample differs from the market in terms of
attributes other than the signals.
Results were inconsistent with the full disclosure of earnings
forecasts by all firms in the economy. The four experimental port-
folios exhibited higher return than the market index used for control
even for low forecast portfolios. Examination of the data revealed
that (p. 39):
estimated return differences are positive
for all portfolios, reinforcing the belief that
the information sample as a whole is systemat-
ically different from the market. More specific-
ally, it appears to consist of firms whose return
performance is, on average, superior to that of
the market as a whole.
This finding should not be interpreted to mean that full dis-
closure of relevant economic attributes by all firms does not occur.
The appropriate interpretation, as pointed out by the author, is that
earnings forecasts by management is not the technology used by all
firms in the economy to signal their relevant economic attributes.
- 24 -
Other signalling technologies are available to firms which do not
issue earnings forecasts.
The second question examined in this study is the extent to
which management forecasts convey information beyond that which is
already public knowledge. Results of the Hotelling T2 test were
inconsistent with the hypothesis of no information content. Portfolio
by portfolio examination revealed that the information effect is
attributed to the two high forecast deviation portfolios. The
Hotelling T2 test, however, is not a direct test of information
content. Penman carefully points out that (p. 47):
The results of the T2 test suggest that earnings
forecasts passes information, in the sense that
foreknowledge of the forecast prior to the port-
folio holding period earns abnormal returns for a
given level of risk, on average. This does not
mean that the information is necessarily conveyed
to the market by means of the forecast. The
forecast is published, on average, about the
middle of the third fiscal quarter, so it is
possible that forecasts merely reflect information
which has become available through other signals
for example, interim earnings announcements.
Excess return tests were used to examine the information
content of management earnings forecasts. Daily abnormal returns for
a sample consisting of 1188 management forecasts were calculated for
121 days surrounding the forecast day. The sign of the abnormal
returns "indicate that the forecast announcements, on average, convey
information which is favorable to the firms that public them, relative
to the market's assessment prior to the announcement" (p. 51). The
cumulative abnormal returns indicate that (p. 53):
- 25 -
forecasting firms do, on average, enjoy
'good times' during the three months (approx-
imately) on either side of the forecast date, and
not only on the day of the forecast announcement.
A sample consisting of 725 management forecasts was disaggre-
gated into 'good' and 'bad' news forecasts depending upon the sign of
deviation from the expectations model. Results were summarized as
follows (p. 53):
'good' news firms apparently are continually
revalued upwards throughout the period and receive
a relatively sharp upward revaluation on or about
the date of the forecast. This is in contrast to
the 'bad' news group. They are subject to a
downward revaluation at the forecast date and, on
average, exhibit 'bad times' prior to the
announcement of the forecast.
The above findings indicate that 'good' and 'bad' news earnings
forecasts convey information to investors. It appears, however, that
the contents of these forecasts are properly anticipated by investors.
Stated differently, the anticipatory effect observed in this study
indicates that substitute information as to the content of the
management forecast of earnings is publicly available. The stock
price behavior of the bad news forecasts indicates that "incentive
exists for managements to publish forecasts which result in the
downward revaluation of their firms" (p. 54). This finding was
interpreted to be consistent with the iterative process of the
incentive-signalling theory which results in the disclosure of inside
information "irrespective of its implication for the value of the
firm" (p. 54). Atiase (1978, p. 22) provides a somewhat similar
interpretation of these results when he states:
- 26 -
even firms disclosing seemingly bad news may
be trying to prevent drastic downward price
revisions that the market might be making about
their prospects.
Finally, the author points out that the above results may be
influenced by other announcements accompanying earnings forecasts.
The tests were repeated after excluding forecasts which were
accompanied by other announcements. This reduced the total sample to
384 'good' news forecasts and 98 'bad' news forecasts. Results
differed little from the previous results for the unrestricted sample.
The Nichols and Tsay Study (1979)
This study examined the information content of management fore-
casts of earnings announced during the first half of the fiscal year.
The study covered a six year period beginning in 1968 and involved a
total sample of 83 forecasts. The selection criteria used for
inclusion in the sample differed from prior studies. The criteria
were set intentionally restrictive in an attempt to assure that "the
potential confounding influence of other types of information on the
results is less than in previous studies without such strict 'other
news' screening criteria" (p. 143). The selection criteria used in
the study were as follows (p. 143):
The forecast announcement must have been unaccom-
panied by other 'news' reports, and thus any firm
that had reported in the WALL STREET JOURNAL any
of the following was omitted from the study: (1)
annual or quarterly earnings reports or other
'news' within a week of the date the forecast was
published, (2) a stock split within eight weeks on
either side of the date of forecast, or (3) a cash
dividend within the week of forecast.
- 27 -
Two approaches were utilized to test the information content of
management forecasts. The first was the abnormal return method. For
each stock i, the absolute value of the residual for each time period
(eit) was calculated. The authors concluded that the average (et) for
the forecast week was larger than it was for the other sixteen weeks
in the test period.
The Kolmogorov-Smirnov one sample test indicated significance
at the 10 percent level. "Thus, while there is some evidence of
unusually large relative price changes during the announcement week,
the observed ranking was significant at the 10 percent level"
(p. 149).
The second approach utilized was an analysis of frequencies of
abnormal returns during the test period. The motive for the frequency
test is the belief that (p. 148):
very large abnormal returns of a few firms
may dominate averages of returns for a group of
firms. Although, again, this may be more of a
potential problem for small samples than for large
samples, the possibility exists in either. In any
event, analyses of averages may not reveal whether
observed abnormal return movements are pervasive
throughout the sample or whether they have been
produced primarily by a subset of the sample.
Analyses of frequencies will provide an indication
of the pervasiveness of any observed information
content.
The specific test used and the results are summarized as
follows (p. 148):
- 28 -
The weekly (eit)'s were ranked one through seven-
teen for each firm. If week-zero (eit) was the
largest for the seventeen-week period, a rank of
one was assigned; if it was the second largest, a
rank of two, and so on with the smallest receiving
a rank of seventeen. If very large relative
abnormal return movements occurred in period zero
as a result of reported forecasts, we might expect
an unusually large number of rank one's for period
zero, relating to the other week. The latter
possibility was easily checked and dismissed.
The apparent conflict between the results of the two tests
should be cautiously interpreted. The selection criteria allowed
other news to prevail during the sixteen weeks surrounding the fore-
cast week. Announcements of annual and quarterly earnings, cash
dividend, and other news were allowed to prevail except for the fore-
cast announcement week. Accordingly, the frequency test examined the
relative effect of management earnings forecasts on unexpected
earnings as compared to other news.
Six historically based naive models were used as a proxy for
current market expectations of earnings. Three of these models were
based on annual earnings, while the other three incorporated quarterly
earnings. Similar to Foster (1973), Patell (1976), and Penman (1978),
forecasts were classified as 'positive' or 'negative' news depending
on the sign of deviation from the naive model. Adopting a purchase
policy for the positive group and short-sale, for the negative group
eight weeks prior to the forecast announcements week resulted in
abnormal returns for both groups. "This would be consistent with the
statement that the executive forecasts possessed information content"
(p. 152).
- 29 -
A seventh group was constructed for comparison with results
reported by Patell (1976). An earnings forecast deviation was defined
as the difference between the forecast and the actual earning for the
previous year. Deviations were scaled by previous year's earning and
ranked in defending order. The top 50 percent were assumed to have
positive news, and forecasts in the bottom 50 percent were assumed to
have negative news. However, the sign of forecast deviation was
positive for both groups. The study reported a graphic presentation
of the cumulative average residuals without any statistical test. The
authors inferred that (p. 152):
The week-zero price movement is the largest weekly
movement for the purchase group, while for the
short-sale group week-zero has one of the smaller
movements. In contrast, Patell found relatively
large positive week-zero movements for both of the
groups.
Frequencies of gain and loss on both the purchase and
short-sale portfolios were examined jointly. The gain (loss) is
measured by the cumulative residual for the nine week period, up to
and including the forecast week.
The chi-square statistic was not significant at the 5 percent
level. The same analysis was repeated for the other six naive
forecasting models. The results were not significant for any of the
six models. These findings led to the conclusion that cumulative
average residuals are influenced by a small number of large residuals.
- 30 -
The above conclusion, however, should be tempered by the
methodological difficulties encountered in this study. Three main
deficiencies are worth noting. First, the gain (loss) on portfolios
was measured as a holding period return beginning eight weeks prior to
the forecast week. To the extent that other news were allowed to
prevail during that period, it would be inappropriate to attribute the
gain (loss) to the information content of management earnings fore-
casts. This is especially true given that the sample size of 83
stocks divided into two groups was not sufficiently large to randomize
the effect of other news. Second, the reliance on naive forecasting
models can result in misclassification of management forecasts, thus
adding another confounding effect to results of the frequency test.
Finally, there is no a priori reason to suspect that all management
forecasts convey information beyond that which is already available.
As demonstrated in the next chapter, the majority of management
earnings forecasts do not differ from publicly available analysts'
forecasts of earnings. Accordingly, the classification of forecasts
as either good or bad news is questionable, particularly when the test
involved is based on frequencies.
Summary and Limitations of Existing Research
Current research studies on the information content of manage-
ment forecasts support the contention that these forecasts convey
information in excess of what is already available to market
- 31 -
participants. One noted exception to this contention is the study by
Nichols and Tsay (1979). The behavior of stock prices prior to the
announcement date indicate an anticipation of the news released by
these forecasts. This is interpreted to mean that substitute
information to management earnings forecasts seems to exist and is
reflected in stock prices prior to the announcement of earnings
forecasts by management.
Conflicting results, however, have been reported with respect
to the information content of 'good' versus 'bad' news forecasts.
Findings of the Gonedes et al. study (1976) indicate that only 'bad'
news forecasts convey information, while results reported
independently by Patell (1976), Penman (1978), and Jaggi (1978)
indicate that both 'good' and 'bad' news forecasts convey information
not already available to investors.
A second area where conflicting results are observed is the bad
news earnings forecasts category. Patell reported an upward price
revision for the 'bad' news forecasts during the forecast week. This
is in contrast to Penman's study which reported a downward price
revision for this group on the day of the announcement as well as the
surrounding days.
The conflicts of research findings are not easily explainable
and may be attributed to several factor. One factor pointed out by
the empirical findings of prior research is the unreliability of
historically-based naive forecasting models as a proxy of market
- 32 -
expectations. Patell observed that "on average, the voluntarily
disclosed forecasts of earnings per share were greater than the naive
estimates of the market's expectations" (p. 268). Penman also
observed that "the earnings of forecasting firms are, on average,
higher than expectations and not representative of the market as a
whole" (p. 36). Thus, low earnings forecast portfolios (bad news
forecasts) are low relative to the sample studied but not low relative
to the measure of expectations. More importantly, however, the
empirical evidence provided by Gonedes et al., Patell, and Penman
suggest that substitute information for management forecasts is
publicly available. Naive forecasting models totally ignore informa-
tion available from sources other than past earnings and as such can
misclassify forecasts. This reasoning is supported by Atiase (1978,
p. 23) as follows:
Thus, it is easy to see the possible misclassifi-
cation that can result from relying on expecta-
tions models irrespective of predisclosure infor-
mation production and dissemination on or by a
firm.
Furthermore, all the relevant studies to date have classified
forecasts as either 'good' or 'bad' news forecasts and have ignored a
possible third classification for forecasts that do not deviate from,
or are approximately equal to, investors' expectations. The empirical
finding that substitute information for management forecasts is pub-
licly available provides support for the assertion that the majority
of management earnings forecasts do not deviate from investors'
- 33 -
expectations. This, of course, is an empirical question which has not
been examined. The validity of this assertion, however, adds further
evidence to the possibility of forecast misclassification by previous
studies.
Several research questions directly related to the information
content of management forecasts remain unexamined. Three of these
questions are of interest to this study. The first deals with the
information content of the voluntary forecasting action by itself.
That is, whether or not the voluntary forecasting action by itself
induces an upward stock price revision. On a theoretical basis it can
be argued that a management forecast which merely confirms investors
expectations of earnings reduces the variability associated with these
expectations causing a reduction of the risk attributed to the firm's
stock. This in turn causes a higher equilibrium value for the stock.
It should be noted, however, that this argument rests on the
assumption that inside information is perceived by investors to be
more reliable than information provided by outsiders to the firm, such
as financial analysts.1 Patell offered the possible effect of the
forecasting action as an explanation for the upward stock price
revision during the forecast week for the 'bad' news group.2 However,
the possible misclassification of forecasts and the conflicting
results of other studies highlight the need for a specific testing of
this proposition.
The second question is related to the effect of the perceived
accuracy of forecasts on their level of information content. Patell
- 34 -
points out that the imputed accuracy of an earnings forecast is an
attribute necessary to altering investors' expectations. Testing this
proposition, however, requires knowledge of investors' expectations
immediately prior to the earnings forecast announcement by management
and the revision of these expectations as a result of receiving the
earnings forecast. The reliance of prior research on naive fore-
casting models precluded empirical examination of the imputed accuracy
proposition.
The third question relates to the possible effect of earnings
forecasts on non-forecasting firms in the same industry. The
mechanism of the incentive-signalling theory alluded to earlier, and
the argument that a management forecast of earnings is a quantitive
assessment of several economic variables, some which are common to the
entire industry in which the forecasting firm operates, lead to the
assertion that the information content of an earnings forecast may
extend to other firms in the same industry. This assertion has not
been considered in the available empirical research on the information
content of management earnings forecasts.
To summarize, existing research findings support the contention
that management forecasts of earnings convey information to investors.
However, the anticipatory behavior of stock prices prior to the fore-
cast announcement date indicates that substitute information to
management forecasts do exist. The conflicting results reported in
the current literature may be attributed to several factors. The
- 35 -
reliance on naive forecasting models is one factor responsible for
misclassification of forecasts. The need for a more appropriate
measure of market expectations and hence proper classification of
forecasts is self evident. Moreover, the scope of available research
on the information content of management earnings forecasts has been
somewhat limited and several research questions remain unexamined.
The effect of the forecasting action by itself, the effect of the
imputed accuracy ascribed to forecasts, and the extent to which
management forecasts of earnings convey industry-wide information are
only three possible extensions to existing research. Examination of
these three extensions comprises the goal of this dissertation.
Notes
1. This argument follows directly from the properties of the capital
asset pricing model under the assumption that a reduction in the
variance associated with expected returns reduces the systematic
risk of the firm.
2. Jaggi (1978, p. 966) also points out the possible effect of the
forecasting action.
CHAPTER III
DATA, METHODOLOGY, AND STATISTICAL TESTS
This study explicitly considers the effects of newness and
imputed accuracy of management earnings forecasts on altering
investors' beliefs with regard to equilibrium stock prices. Opera-
tional measures of newness and imputed accuracy of management earnings
forecasts require a 'proxy' of investors' expectations before and
after the announcement of these forecasts. Publicly available fore-
casts of earnings by financial analysts seem to be an appropriate
measure of investors' expectations for four reasons. First, financial
analysts incorporate several factors in arriving at an estimate of
future earnings of a particular firm. The past earnings history of
the firm is only one such factor. Other external factors which have
impact on the potential profitability of the firm are also incorpo-
rated. Second, financial analysts communicate with managements
through interviews and private discussions and incorporate inside
information in their forecasts of earnings. Gonedes et al. point out
that (p. 108):
Indeed communication between management and ana-
lysts about forecasts seem to be one of the major
motivations for the SEC's current perspective on
- 36 -
- 37 -
forecast disclosure. .The use of such private
communication is also emphasized in a recent study
of earnings forecasts conducted by the Financial
Analysts Federation.
Third, the reliance of investors upon earnings forecasts published by
analysts as an input to investment decisions is widely supported.
Abdel-Khalik and Ajinkya (1979) point out that (p. 1):
the production of and demand for analysts fore-
casts clearly testifies to the importance of these
forecasts to the investment community.
Good (1975) explains the importance of the role of analysts to the
professional investors as follows (p. 42):
What ever the broad strategy used by the portfolio
manager, he is ordinarily dependent in varying
degree on analysts' recommendations concerning
individual stocks. He cannot accept analysts'
recommendations uncritically, but neither can he
investigate each security as thoroughly as the
analyst who specializes in the stock.
Fourth, the available empirical evidence suggests that earnings fore-
casts by analysts influence market prices of securities. Gonedes et
al. (1976) examined the information content of analysts' forecasts of
earnings and concluded that these forecasts convey information
relevant to establishing equilibrium prices of securities. A more
recent study by Abdel-Khalik and Ajinkya (1980) examined the informa-
tion content of analysts' revisions of earnings forecasts. The null
hypothesis of no information content was rejected at a statistically
significant level for the revision week. The hypothesis was rejected
for both positive and negative revisions.
- 38 -
To summarize, earnings forecasts published by financial ana-
lysts incorporate publicly available information as well as inside in-
formation privately communicated by managements. Moreover, empirical
research findings are consistent with the assertion that analysts'
forecasts are relied upon by market participants in making investment
decisions. Accordingly, analysts' forecasts of earnings represent a
reliable proxy of investors' expectations. This proxy is utilized in
this study to determine the newness and imputed accuracy of management
forecasts as defined in the following section of this chapter.
Data
The experimental sample of earnings forecasts was obtained by
searching the Wall Street Journal Index for eight years covering the
period 1970-1977. Only forecasts attributed to top management,
usually the president or the chairman of the board, were admitted into
the experimental sample. The following criteria were employed in the
sample selection process.
1 The forecast must be a point estimate or convertible to a
point estimate. The latter group are earnings forecasts
expressed either as a specific increase (decrease) over the
actual earnings of the prior years or as a range. Earnings
forecasts expressed as a range are bounded by a minimum and
a maximum value. These forecasts were admitted only if the
-39-
maximum value of the range did not exceed the minimum value
by more than ten percent. The mean value of the range was
considered to be the forecasted earnings. Open ended
forecasts in the form of greater (less) than a specific
amount were not included in the sample.
2 The forecast must not be accompanied by or fall within four
weeks prior to or four weeks after the date of any of the
following announcements:
a annual or quarterly earnings
b dividends
c stock splits
d capital issues.
3 Daily stock returns must be available on the daily CRSP tape
for at least one and one half years prior to the forecast
date and one year after the forecast date.
4 The forecast must be the first earnings forecast issued by
management during the fiscal year to which the forecast
refers. Second forecasts or forecast revisions by management
were excluded from the sample.
5 At least one analyst's forecast expressed in point form must
be available from the Standard and Poors Earnings Forecaster
within three weeks prior to the date of the management
forecast announcement. Another analyst's forecast must be
available no later than three weeks after the date of the
management forecast announcement. If two or more analysts'
forecasts were available, the one closest to the date of the
- 40 -
management forecast was selected. On very few occasions,
two or more analysts' forecasts appeared on the same date,
or very close together, in which case the mean value was
determined as the proxy for the market expectations.
The first criterion ensures that management forecasts of
earnings are quantifiable. The majority of forecasts predicted an
increase over the actual earnings of the prior year. Very few firms
predicted a decrease in earnings from the prior year. The second
criterion was imposed in an attempt to reduce the effect of other
announcements. It should be noted, however, that not all announce-
ments are published in the Wall Street Journal Index. The effect of
the undetected announcements should not influence the results of this
study. The large sample size utilized would be expected to randomize
the effect of these announcements over the test period. Penman (1978)
noted that other news announcements did not influence the results of
his study. The third criterion ensures that sufficient data are
available for statistical tests. The fourth criterion excluded
earnings forecast revisions by management in order to avoid the
possible confounding effects of mixing the two types of forecasts.
The fifth criterion is necessary for the operational measures of
imputed accuracy and newness of management earnings forecasts. Ana-
lysts forecasts prevailing within three weeks prior to the announce-
ment of an earnings forecast by management are considered a measure of
investors' expectations at that time. Similarly, analysts' forecasts
prevailing within three weeks after the announcement date are
- 41 -
considered a measure of investors' expectations conditional upon the
information revealed by management through the earnings forecast.
The selection criteria employed in this study are necessary to
reduce the confounding effects of other news and to focus on the
variables under examination. Nonetheless, as in the case for most
empirical studies, these selection criteria induce a bias towards
surviving firms. Criterion five induces another bias towards firms in
which analysts have exhibited interest.
The final experimental sample that satisfied the selection
criteria is composed of 401 earnings forecasts. Except for public
utility firms, no particular industry is over represented. This
observation is consistent with findings of prior studies. The sample
is not evenly distributed throughout the fiscal year and forecasts
tend to cluster in the fourth quarter. A monthly break-down of
forecasts is included in Table 3-1 below.
Table 3-1
Monthly Management Earnings Forecasts
Month in fiscal year 1 2 3 4 5 6 7 8 9 10 112
II I I I I I t I I
Number of Forecasts 5 12 30 40 143 27 1 18 16 41 1 58 51 (60
I P I T I 1 3I I I 1
Percent of Total I1.2 3.0 7.5 10.0110.71 6.714.5 14.0110.2114.5112.7115.0
I I I I I I I I I I I I
- 42 -
Definitions of Partitioning Variables
Newness and imputed accuracy of management forecasts of
earnings are operationally defined as follows:
Newness of a forecast = MF AP
Imputed accuracy
= MF AA
where
MF = management forecast of earnings,
AP = analyst's forecast of earnings prevailing within three
weeks prior to the announcement of the management
forecast, and
AA = analyst's forecast of earnings prevailing within three
weeks after the announcement of the management
forecast.
Experimental Portfolios
For the total sample, the mean and the standard
newness of forecasts are 0.0085 and 0.065 respectively.
the standard deviation for the absolute value of newness
0.05 respectively. This clearly indicates that the
deviation for
The mean and
are 0.042 and
majority of
- 43 -
management forecasts do not convey new information to investors.
Three portfolios were formed representing positive, negative, and
neutral news forecasts, depending upon the sign and the magnitude of
the level of newness. The positive news portfolio consists of 115
earnings forecasts with newness level > 3 percent, an arbitrary
choice. Similarly, the negative news portfolio consists of 85 manage-
ment forecasts with newness level< -3 percent. The neutral portfolio
consists of the remaining 201 management earnings forecasts.
The mean and the standard deviation for the absolute value of
imputed accuracy are 0.0198 and 0.045 respectively.
Observations in each of the three portfolios were ranked on the
level of imputed accuracy. If analysts revise their forecasts to
equal the forecasted earnings by management, the level of imputed
accuracy will be high and the results of Equation (2) will be zero.
The level of imputed accuracy of a positive news earnings forecast is
considered to be high if the result of Equation (2) was equal to or
less than 3 percent, and low if greater than 3 percent. The reverse
classification is true of the negative news forecasts. This classifi-
cation scheme yielded six portfolios. Table 3.2 presents the number
of forecasts in each portfolio.
- 44 -
Table 3.2
Imputed Accuracy and Newness of Management Earnings Forecasts
Imputed Accuracy POSITIVE NEGATIVE NEUTRAL
High 65 47 193
Low 50 38 8
Total 115 85 201
The neutral news low-level imputed accuracy portfolio contains
only eight forecasts. This result is not surprising, due to the fact
that this group of management forecasts represents a mere confirmation
of existing beliefs and, as such, no revision in analysts' forecasts
should be expected. The revision of the eight firms' forecasts may be
due to other factors that occurred within three weeks after the
announcement of the management earnings forecasts. This portfolio was
deleted from the study due to the small size of the sample.
Control Portfolios
For each experimental portfolio two control portfolios were
formed. The first control portfolio was selected by individually
matching each firm in the experimental portfolio with another firm,
- 45 -
for which no management forecast was known, from the same industry
classification. The matching was conducted using the first three
digits of the industry classification code. The selection of control
firms was random in the sense that any firm with the same industry
classification code had an equal chance of being chosen. Due to data
requirements, however, only firms with daily stock returns available
on the CRSP tape for the period 1970-1978 were admitted into the
control portfolio. If a firm did not satisfy the data availability
requirement another firm from the same industry classification code
was chosen at random. This process continued until all firms in the
experimental portfolios were individually matched. These control
portfolios are referred to hereafter as the matched control
portfolios.
The second set of control portfolios were randomly selected
firms without industry matching. To simplify the search process,
however, the matched control firms were randomly assigned to the
experimental firms. These control portfolios are referred to as
random control.
The matched control portfolios are designed to detect industry
effects and to assess the effect of management forecasts of earnings
on non-forecasting firms in the same, or similar, industry, the
random control portfolios are used to provide a bench-mark comparison
and to help detect industry effects.
- 46 -
Statistical Tests
Two statistical testing procedures are used in this study. The
first is residual analysis based on the familiar market model. The
second test involves a comparison of total returns of equal risk port-
folios. A two way analysis of variance is utilized for this test. The
use of two different testing procedures will indicate whether the
results obtained are sensitive to the statistical test used. Further-
more, two approaches to residual analysis are used to provide a
comparison with results reported by prior studies.
Residual Analysis
Information content can be attributed to management forecasts
of earnings if investors' beliefs with regard to equilibrium prices
are altered as a result of the information conveyed by these fore-
casts. Change in stock prices (and returns) during the forecast
announcements period is the empirical evidence of information content.
The returns observed on a security, however, cannot be totally
ascribed to a specific event, due to the fact that the return is also
influenced by economy-wide events affecting all securities. The
market model developed by Sharpe (1964) and Lintner (1965) is commonly
used to filter economy-wide factors. The market model is an ex-ante
relationship between the return on a specific security and the return
on the market portfolio. The model, however, is used on an ex-post
basis utilizing actual realizations of security returns to determine
- 47 -
abnormal performance. In this study, the market model is employed to
measure weekly abnormal returns using two related approaches. The
first approach measures standardized prediction errors and the second
examines the distribution of unstandardized errors during the forecast
period. While the latter approach is commonly used, the former is
theoretically more appropriate. The two tests are explained as
follows:
A Standardized Prediction Errors
This test was originally used by Beaver (1968) and expanded by
Patell (1976) using the market model which is given by:
Rit = ai+biRat+eit (1)
where
i = a firm index, i=1, ., N.
t = a week index for the non-forecast period, t=l, ., T.
Rit= the natural logarithm of the return on stock i for week t.
Rmt= the natural logarithm of the return on the market index
ai,bi the intercept and slope of the linear relationship
between Rit and Rmt.
eit = stochastic disturbance term for security i in week t.
Time series regression is performed on realizations of Rit and
Rmt to obtain estimates of ai and bi. These estimates are then used
- 48 -
to calculate the prediction error (abnormal return) for each stock i
and for 16 weeks surrounding the forecast announcements week
(hereafter called the test period). The prediction error is given by:
eit = Rit-ai-biRmt t = -8. .+8
The prediction error in the test period is assumed to be
normally distributed with the following properties:
E(eit) = 0 t = -8. .+8
Cov(eis,eit) = 0, s#t
citai s=t
ai2 is the variance of the regression during the estimation period.
The unbiased estimator of oi2 is si2 which equals:
si2 = ai2
T-2
Cit is the increase in the variance due to prediction outside the
estimation period. Cit = + 1+ (Rmt Rm)2
T t=1. .T
T
E (Rmt-Rm)2
t=l
- 49 -
The standardized prediction error during the test period is:
eit
Vit ~ t(T-2) (2)
si )cit
The cumulative standardized prediction error for the 17 weeks included
in the test period is given by:
L
C eit
WiL = t=1 ~ t(T-2) (3)
si /Lcit
where L < 17
Each eit is distributed independently of si2 and has a variance equal
to:
Ti-2
Ti-4
where Ti is the number of observations in the non-forecast period for
firm i. The prediction error is assumed to be an independent random
variable. According to the Central Limit Theorem, the cross-sectional
mean for any week t in the test period approaches a unit normal
distribution. The normalized sum can be expressed as:
N
E Vit
Zvt = i=l (4)
N Ti 2 )
i=l Ti 4
- 50 -
Similarly, the cross-sectional cumulative prediction error for week t
is
N
E WiL
ZwL = i=1 (5)
N Ti 2 \
i=l Ti 4
Equations (4) and (5) can be used to test the hypothesis that the
cross-sectional standardized prediction error is significantly
different from the expected mean of zero. One or two sided tests can
be performed, depending on whether an expected direction of eit can be
specified ex-ante. A statistical test that abstracts from direction
can be constructed directly from equation (2). Squaring Equation (2)
yields:
(eit)2
(Vit)2 = ~ F (1,Ti-2) (6)
si t
and
(eit)2
E = Ti 2
E si2cit 2 (7)
Ti 4
Equation (7) can be standardized to yield an expected value of 1. the
standardized statistic is:
- 51 -
Uit = (eit)2 Ti 4
sicit Ti 2
E(Uit) = 1
2(Ti 3)
VAR(Uit) =
Ti 6
Again, by applying the Central Limit Theorem, a unit normal cross-
sectional statistic can be constructed as follows:
N
E (Uit-1)
Zut = i=1 (8)
N ( 2(Ti-3)
i=1 (Ti-6)
Statistical tests developed in this section involve testing the
mean of the distribution of prediction errors under the normality
assumption. Nichols and Tsay (1979) questioned the validity of these
tests and pointed out that the mean may be influenced by a small
number of large prediction errors. To examine this claim, tests are
repeated after trimming the distribution by 10 percent on each side.
This test will indicate whether the observed effect is a general
pattern or merely due to a few extreme observations.
Another test similar to that used by Penman (1978) can be
developed as follows:
- 52 -
it ~ t (T-2)
V = 1 2 V.
N- Vit
N
E(Vt) = 0
Var (Vt) = at2
by appeal to the Central Limit Theorem (Vt //N)/I is approximately
standard normal distribution for large N. A Z test can be conducted
for each week in the test period. This test provides a comparison
with results reported by Penman (1978) and indicates the sensitivity
of results to the statistical test used.2
B Unstandardized Prediction Errors
As stated earlier, this test is commonly used in accounting
research. Foster (1973), Jaggi (1978), and Nichols and Tsay (1979)
utilized unstandardized prediction errors (eit). This test also
assumes cross-sectional independence and normal distribution of the
prediction error with a mean equal to zero. A normal distribution
test is used to examine the hypothesis that the mean value et is equal
to zero. Results of this test will be reported for comparison with
prior studies.
Analysis of Variance
An analysis of variance test is used in this study to examine
the information content of management forecasts of earnings using
- 53 -
actual returns observed on securities during the forecast announcement
week and the ten surrounding weeks. This test is designed to
supplement results of prediction errors tests and to provide a
comparison between results obtained using a different variable for
each test. In addition, the use of actual stock returns reduces the
reliance on the market model.
A two-factor analysis of variance is performed on the stock
returns of each experimental portfolio and its designated control
portfolios. The first factor is the type of portfolio and contains
three levels (i.e., experimental, matched control and random control).
Distinctions among the three portfolios are based on the information
set affecting each portfolio. Publicly available information
influencing stock returns is composed of three general components:
firm specific, industry specific, and economy-wide information. The
firm specific information distinguishes the experimental portfolio
from the two control portfolios. The industry specific information is
common to both the experimental and the matched control portfolios but
not to the random control portfolio. This provides a distinction
between the matched control and the random control portfolios. In
short, the three portfolios are influenced by different sets of
information and as such are viewed as three levels of the type of
portfolio factor.
The second factor is the new information effect and contains
eleven levels, t = -5 to t = +5. The forecast announcement week,
- 54 -
t = 0, is only one such level. Prior research indicates that the
information contained in earnings forecasts is partially anticipated
and influences stock returns prior to the forecast announcement date.
This implies a general flow of information produced and disseminated
by insiders and/or outsiders to the firm. Thus, the level of new
information made publicly available is distinguishable on a weekly
basis. It is important to note that stock returns observed every week
are influenced by the new information made public during that week and
not by the information revealed during prior weeks. This interpre-
tation follows directly from the empirical evidence of stock markets
efficiency in the semi-strong form, which asserts that stock prices
(returns) adjust instantaneously to all publicly available informa-
tion. Without the semi-strong form efficiency, the analysis of
variance tests would be questionable due to lack of independence among
weekly information effects.
Portfolio returns are not, strictly speaking, comparable
because of varying levels of systematic risk. To overcome this
difficulty, the systematic risk for all portfolios is equated to
unity. The procedure used is developed by Gonedes (1974) and involves
the following steps for each portfolio:
1 The systematic risk (beta) for each stock is calculated and
then ranked in defending order.
2 Two portfolios are constructed representing the top and
bottom halves of the ranked stocks. The mean beta for each
of the two portfolios is then calculated.
- 55 -
3 To determine the weights to be assigned to stock returns of
the top and the bottom portfolios, the following equation
is solved:
B1X + B2 (1 X) = 1
where
B1 = the mean beta for the top portfolio
B2 = the mean beta for the bottom portfolio
X = the weight to be assigned to stock returns of the top
portfolio
(1-X) = the weight to be assigned to stock returns of the bottom
portfolio
For all portfolios, the weights were between 40 and 60
percent. This result is consistent with results reported by Gonedes
(1975). After restating all stock returns, the following two factor
linear model is used:
Rjik = u + Pi + tk + Yik + ejik
where
u = the grand mean
Pi = Portfolio effect, i=1, 2, 3
tk = the information effect, k=l,. ., 11
Yik = the interaction effect
ejik = random error term.
The analysis of variance test examines the overall equality of
means. If the overall hypothesis is rejected, further investigation
will be required to determine which of the thirty-three means, or
- 56 -
combination of means, are causing the rejection of the overall
hypothesis. The Duncan's multiple range test is used for comparison
of all means.
Hypotheses
A Testing the Effect of Forecasting Action
The neutral news forecasts convey a very low level of new
information to investors and as such should not be associated with
abnormal returns unless the forecasting action by itself does in fact
alter investors expectations. Thus, testing the effect of the fore-
casting action involves an examination of the behavior of prediction
errors associated with the "neutral" news earnings forecasts. It has
been argued that a neutral forecast may reduce the variance of
expected future returns and as such cause the price of the stock to
move to a higher equilibrium price. The validity of this argument
hypothesizes a positive mean prediction error for this group. The
following three tests are conducted on prediction errors:
Hol: The mean value of vt during the forecast disclosure week
is equal to zero
Hal: The mean value of vt during the forecast disclosure week
is greater than zero
Ho2: The trimmed mean value of vt during the forecast
disclosure week is equal to zero
- 57 -
Ha2: The trimmed mean value of vt during the forecast
disclosure week is greater than zero
Rejecting both Hol and Ho2 will indicate that the forecasting
action by itself conveys information favorable to stock prices and
that the effect is a general pattern of the sample, and vice versa.
If Hol is rejected but Ho2 is not rejected, this would mean that the
test is influenced by a few outliers. Failing to reject both Hol and
Ho2 would indicate that the forecasting action conveys no information
to the market. Results will be reported for the seventeen weeks in
the test period and for the experimental as well as the control
groups. Two other tests are conducted on the distribution of the
residuals and prediction errors respectively, (hereafter called
distribution tests), as follows:
Ho3: The mean value of et during the forecasts disclosure week
is equal to zero
Ha3: The mean value of et during the forecasts disclosure week
is greater than zero
Ho4: The mean value of vt during the forecasts disclosure week
is equal to zero
Ha4: The mean value of vt during the forecasts disclosure week
is greater than zero.
The analysis of variance test is conducted on portfolio returns
to test the following four hypotheses:
- 58 -
Ho5: The means of the three portfolios are equal
Ha5: The means of the three portfolios are not equal
Ho6: The means of the eleven information treatments are equal
Ha6: The means of the eleven information treatments are not
equal
Ho7: There is an interaction effect between the portfolio
effect and the information treatment effect
Ha7: There is no interaction effect
If the overall hypothesis of equality of means is rejected, the
following Duncan's Multiple Range Test will be examined.
Hog: All thirty-three means are equal
Hag: All thirty-three means are not equal.
B Testing the Effect of Newness of Earnings Forecasts
Two experimental portfolios are classified as high-level
newness portfolios: the positive and the negative news portfolios.
For each of the two portfolios the direction of abnormal returns can
be specified ex-ante and as such the alternative hypotheses should be
stated with the proper direction. That is, the alternative hypotheses
for abnormal returns tests should be greater (less) than zero for the
positive (negative) portfolio. The same eight hypotheses already
stated in conjunction with the "neutral" news portfolio are utilized
to test the effect of newness.
- 59 -
C Testing the Effect of Imputed Accuracy
As stated earlier, each of the two high-level newness
portfolios is subdivided into two unequal portfolios based on the
degree of imputed accuracy ascribed to the management forecasts of
earnings. Testing the effect of imputed accuracy is accomplished by
separate examination of results for the high and low imputed accuracy
portfolios. The same eight hypotheses are tested for each group.
Further, abnormal returns for the high-level and low-level imputed
accuracy portfolios for each week in the test period can be compared
under the assumption of cross-sectional independence. Three
statistical tests can be developed for comparison between abnormal
returns of the two types of imputed accuracy portfolios. The first
test follows directly from Equation (4). The difference between
standardized prediction errors for unequal samples is given by:
N1 N2
1 E 1 E Vj,
N i=1 Vit N2 j=l
ZdV, = (9)
1 N1 Ti-2 1 N2 T-2 1
N + + N 2 TZ 2
i=l Ti-4 j=l Tj-4
i1 j_4
where N1, and N2 are the number of firm forecasts designated as
hihg-level and low-level imputed accuracy respectively. Equation (9)
can be used to test the following two null hypotheses for the total
and the trimmed standardized prediction errors:
- 60 -
Hog: Zdvt during the forecast announcements week is equal to
zero
Hag: Zdvt is greater (less) than zero for the positive
(negative) forecasts.
Hol0: Zdvt for the trimmed standardized prediction errors during
the forecast announcements week is equal to zero
HalO: Zdvt for the trimmed standardized prediction errors during
the forecast announcements week is greater (less) than
zero for the positive (negative) forecasts.
The second test is designed to examine the observed distribu-
tion of weekly standardized prediction errors. The Z statistic for
the difference between the two means is given by:
1 N1 N2
N1 Z Vit 1 Z vj
i=1 N2 j=l
Zdvt = (10)
SN~it + S2jt i
N2
where Si2 and Sj2 are the weekly variances of the observed distribu-
tion of standardized prediction errors associated with the high-level
and the low-level imputed accuracy portfolios respectively. The third
test examines the distribution of unstandardized prediction errors
(et) and is given by:
- 61 -
1 N1 I N2
N1 eit N2 E ejt
i=l j=1
Zdet = (11)
K 2it K 2jt
N1 N2
where Ki2 and Kj2 are the weekly variances of the observed
distribution of unstandardized errors associated with the high-level
and low-level imputed accuracy portfolios respectively.
The following two null hypotheses are examined:
HOll: Zdvt during the forecast announcements week is equal to
zero
Hall: Zdvt during the forecast announcements week is greater
(less) than zero for the positive (negative) forecasts.
Ho12: Zdet during the forecast announcements week is equal to
zero
Hal2: Zdet during the forecast announcements week is greater
(less) than zero for the positive (negative) forecasts.
D Testing the Effect on Other Firms in the Same Industry
This hypothesis is examined by comparing the behavior of
abnormal returns and total stock returns of the matched control
portfolio to the other two portfolios--the experimental and the random
control.
- 62 -
As already stated, the random control portfolio is used to provide a
bench-mark comparison and to help detect economy-wide effects. The
comparison between the matched control portfolio and the experimental
portfolio will indicate the extent to which management forecasts
convey industry-wide information. If management forecasts of earnings
do in fact convey substantial industry-wide information, the abnormal
returns, and the total returns, of the matched portfolio should be
influenced in the same direction as the experimental portfolio during,
but not prior to, the forecast announcements week. On the other hand,
if significant abnormal returns are associated with the matched
control portfolio, but not the experimental portfolio, prior to the
forecast announcements week, and both portfolios exhibit significant
abnormal returns during the forecast announcements week, the inference
of industry-wide effect would not be justifiable. A more appropriate
inference would be that industry factors induce managements to issue
earnings forecasts in an attempt to revalue their firms relative to
other, non-forecasting, firms in the same industry. That is, the
observation of abnormal returns associated with the matched control
portfolio during the forecast announcements week is a necessary but
not a sufficient condition for the inference of industry-wide effects
of management forecasts of earnings. Examination of the behavior of
abnormal returns associated with each of the two portfolios prior to
the forecast announcements week will indicate whether management
forecasts of earnings convey industry-wide information or whether
industry factors induce managements to issue earnings forecasts.
- 63 -
The first eight hypotheses were tested for the three portfolios
to determine the effect of management forecasts on each portfolio and
to provide the comparison among the three portfolios necessary for
detecting the industry-wide effects.
Notes
1. The regression was performed on actual weekly realizations of
Ritand Rmtfor two hundred weeks prior to the test period.
2. Prediction errors tests assume stability of regression coeffi-
cients, a and b, during test period. Systematic violations of
this assumption biases the observed prediction errors during the
test period. In this study, such bias is not expected to be
systematic in nature for four reasons. First, the experimental
sample was drawn from observations which occurred at different
time periods, (8 years X 52 weeks), and as such any shift in the
regression parameters would not be expected to take a systematic
form. Second, the test period is relatively short (17 weeks) to
induce a serious shift in regression parameters. Third, the use
of two time matched control groups will highlight whether or not
any systematic bias in prediction errors did in fact take place.
Fourth, the large sample size utilized in this study would be
expected to randomize the effect of any shift in regression
parameters.
CHAPTER IV
RESULTS
Introduction
The first three research questions addressed in this study
examined the effect of newness and imputed accuracy of management
forecasts of earnings on equilibrium stock prices of forecasting
firms. The fourth research question examined the effect of management
earnings forecasts on equilibrium stock prices of non-forecasting
firms in the same, or similar, industry. On an a priori basis it can
be argued that the effect of management earnings forecasts on non-
forecasting firms is positively related to the level of newness as
well as to the level of imputed accuracy ascribed to these forecasts.
Accordingly, the effect on non-forecasting firms was examined in
conjunction with the first three research questions. This procedure
of joint examination of hypotheses provides a more comprehensive
analysis of results and avoids repetition in discussing the
implications of the empirical findings of this study.
Results are reported in this chapter for the total sample,
followed by results for the first three research questions; these are
- 64 -
- 65 -
examined in conjunction with the effect on non-forecasting firms. The
total sample represents an aggregation of different, and opposing,
classifications of management earnings forecasts and, therefore, the
results obtained are difficult to interpret. These results are
reported here, however, for comparison with prior studies and are
limited to standardized prediction errors tests.
Total Sample
The total sample consists of positive, negative, and neutral
management earnings forecasts and, as a result, the direction of
prediction errors cannot be specified ex-ante. A statistical test
that abstracts from direction (Zut) is given by equation (8) and
follows from squaring the standardized prediction errors.
Table 4-1 reports the Z values for standardized prediction
errors (Zvt), the mean value of the squared standardized prediction
errors Ut, and Zut for all three portfolios. The Zvt statistic
indicates rejection of the null hypothesis of no information content
during the forecast announcements week, t and week t+6 for the
experimental portfolio. The null hypothesis could not be rejected for
the matched control portfolio during the forecast announcements week
or any other week in the test period at a statistically significant
level. The null hypothesis is rejected, however, for the random
control portfolio during weeks t-1 and t+3, but not during the
forecast announcements week.
- 66 -
0 -l -i O .O N- 0 -A 0 0 0 0 -<
I I I I I
S 0 s 7 4 0 I. l 7 0 ..
'0 I !0'0 0 0 4N NU
0 0-ll -4 4'I 4 -4 0 0 '-0 O0 0 0 0 N '0
I I I
i O N 0 N S U, U, '0 m p N -
N .- o 0 .4 .-4 0 '0 .- .o U N O 0
I I I I I I I
Ue, o O 0 N o U, .- .
0 0 0 04o 0 0o o. 4 .
4M 'r 0 0 .7 0%
0% -4 .0 0 0< 0
S000 I I I
I I I I I
00% 0 0 4'.> r N
0 '0 0 N 0 .< '.0 0
I I I I
m 0 0 a '0 Ur 0 0 0 '0 0 C'4 .7 '.0 0T% 4t N
u f '00 O Q 4*1 .- .4 C% 44 I( Up
N 0 0O NO '-40 '0 Nt f- N N NO -4
I I I I I I I
El
4'(-. U, r-. '0 0% '0 0 r. ..t 0r C'. cM ct f. N' i, 4
0% 0% 0 0 .- 0t 4 0% O f 0 r. a 0o 0 % o4 4
'4
a LJ *
r- U 0 a 41-40N 0N N o 0 -4ON -
S|3 0 0 4 N 0 0 0 0 0 0 0 0 4
N4 -0 00 0 N 0 0 0 0 DF r
4 I I I I I
44 0 r. '. 1% -
iu a p O *n I I I ~ Ct c ( r m o rI I
I1___
JJ
N
'3
t0
'U -
- N
'U
u b
00
ca
'U
Dl
m
I I
- 67 -
The Zut Statistic should be cautiously interpreted for two
reasons. First, Zut is considered a test of the increased variance
during the test period only under the assumption that the expected
value of prediction errors is equal to zero. Accordingly, the
variance change interpretation for weeks which exhibit prediction
errors significantly different from zero is inappropriate (see Patell
1976, p. 261). Second, as would be expected when dealing with squared
values, Zut values are sensitive to a small proportion of large
prediction errors.
Results reported in Table 4-1 indicated that only the matched
control portfolio exhibits an increase in the variance during the
forecast announcements week unaccompanied by a mean prediction error
significantly different from zero. Furthermore, the observed increase
in the variance of prediction errors associated with the matched
control portfolio during the weeks immediately following the forecast
announcements week are considerably larger than those associated with
the other two portfolios. This indicates information effects during
these weeks for the matched control portfolio. Statistically
significant values of Zut unaccompanied by mean prediction errors
significantly different from zero indicate change in the variance and
signals rejection of the null hypothesis of no information content.
Disaggregation of the total sample is necessary to determine which of
the three classifications of management forecasts of earnings, or what
combination, is causing the observed information effects. Further-
more, removal of outliers is necessary to determine whether the
observed effect is also the general pattern of the sample.
- 68 -
Examination of the total sample was motivated by the need to
highlight the extent to which the sample of management earnings fore-
casts used in this study is similar to those used in prior studies.
Since prior studies did not utilize control groups, the comparison is
limited to the experimental portfolio. The following findings are
consistent with those of prior studies:
1 the observed mean prediction errors during the forecast
announcements week is statistically significant and
positive. This result indicates that, on average, manage-
ment forecasts of earnings are accompanied by an upward
stock price revision. This result is consistent with those
reported by Patell (1976) and Penman (1978).
2 the observed Ut values for weeks immediately following the
forecast announcements week are less than unity, indicating
a reduction in the variance for these weeks. This result
is consistent with findings reported by Patell (1976,
p. 261).
The Effect of the Forecasting Action
The neutral portfolio of earnings forecasts was examined to
determine whether the forecasting action by itself conveys favorable
information to market participants. Table 4-2 reports the Z values
for standardized production errors and cumulative prediction errors
for the three portfolios. The null hypothesis of no information
content, H01, was rejected at the 0.05 level for the experimental
portfolio during the forecast announcements week, whereas the two
- 69 -
Table 4-2
Total Neutral
Values for Zvt and ZWL
Week Experimental Matched Control Random Control
Zvt ZWL Zvt ZWL Zt ZWL
0.43
-1.07
0.50
0.75
0.28
0.60
0.88
0.56
1.93*
-0.51
-0.12
0.16
-0.78
1.22
0.91
0.81
0.94
0.43
-0.45
-0.08
0.31
0.40
0.61
0.90
1.04
1.62
1.38
1.27
1.27
1.00
1.29
1.48
1.64
1.82*
0.39
0.73
0..20
0.19
-0.84
-1.32
-0.77
0.18
-0.66
-0.20
-1.28
0.45
0.50
1.47
1.82*
0.17
0.99
0.39
0.79
0.76
0.76
0.30
-0.27
-0.54
-0.43
-0.63
-0.66
-1.02
-0.85
-0.67
-0.26
0.22
0.26
0.49
-1.84
0.68
-0.73
-0.93
0.31
1.05
0.69
0.23
-1.17
0.09
0.33
-1.64
1.04
-1.49
-0.58
0.38
2.60*
-1.84
-0.82
-1.09
-1.41
-1.12
-0.60
-0.29
0.19
-0.57
-0.51
-0.39
-0.84
-0.52
-0.90
-1.02
-0.90
-0.24
* Significant at the 0.05 level
L I I I I
- 70 -
control portfolios exhibited negative mean prediction errors during
the same week. Table 4-3 contains the Z values for the distribution
of weekly unstandardized prediction errors (et) and standardized
prediction errors (Vt). The results of the distribution tests
indicate rejection of the null hypotheses during the forecast
announcements week at the 0.07 and 0.05 level of significance for et
and vt respectively. These results suggest that the forecasting
action by itself has favorable implications for stock prices of
forecasting firms. Figure 4-1 portrays the behavior of cumulative
average standardized prediction errors during the test period. The
behavior of the prediction errors indicates that the experimental
portfolio enjoys a higher abnormal return relative to the two control
portfolios during the test period.
The trimmed mean test was conducted to determine whether the
observed effect was a general pattern of the experimental portfolio or
was due to the influence of a few large outliers. Table 4-4 reports
the Z statistic for the trimmed mean of standardized prediction
errors. The null hypothesis of no information content could not be
rejected for the forecast announcements week or any other week in the
test period at a statistically significant level. It should be noted,
however, that the largest positive weekly mean prediction errors were
observed during the forecast announcements week. These results lead
to the conclusion thatthe forecasting action by itself induces an
upward stock price adjustment, although such adjustment is not
statistically significant. The cumulative trimmed abnormal returns
associated with the experimental portfolio were larger than those
associated with the two control portfolios.
- 71 -
Table 4-3
Total Neutral
Values for Zet and Zvt
Week Experimental Matched Control Random Control
Zet Zvt Zet Zvt Zet Zvt
0.48
-0.79
0.38
0.60
0.81
1.04
0.77
1.02
1.52
-0.83
-0.21
0.61
-1.20
1.28
1.11
0.36
0.92
0.52
-1.17
0.17
0.93
0.67
0.67
0.82
0.33
1.72*
-0.53
-0.06
0.24
-0.94
0.99
1.15
0.16
0.81
-0.16
-0.18
0.44
-0.68
-0.61
-1.37
-0.25
0.49
-0.56
-0.49
-1.09
-0.08
-0.55
0.87
2.11 *
0.13
1.58
-0.12
0.65
0.20
-0.25
-0.87
-1.36
-0.85
0.19
-0.69
-0.20
-1.39
-0.44
-0.50
1.43
1.95*
0.19
0.95
-1.91
0.87
-1.12
0.34
0.87
0.15
-0.56
-0.03
-1.51
0.56
0.03
-2.12
0.85
-1.70
-0.75
-0.43
2.88*
-1.95
0.75
-0.66
0.11
0.32
0.63
-0.65
0.22
-1.20
0.09
0.36
-1.80
0.96
-1.49
-0.57
-0.35
3.35*
* Significant at the 0.05 level
- 72 -
1.00-
0.90
0.80
0.70
0.60
0.50
0.40
0.30
0.20-
0.10-
o. -8 N .0 -5 -4 0 1 2 3 4 5/ 6 7 8
-o0-".. ........ ,.. ,. **". / "
-0.20 **... *, ,-- *
**.....*
-0.30 -
-0.40
-0.50
-0.60-
Experimental
-0.70- Match Control -
Random Control ***
-0.80-
0.90-
1.00-
Figure 4-1
Neutral Group
Cumulative Standardized Prediction Errors (27 7)
1.1
- 73 -
Table 4-4
Total Neutral Trimmed Mean
Values for Zvt and ZWL
vt WL
eek Experimental Matched Control Random Control
Zvt ZWL Zvt ZL Zvt ZWL
0.62
-0.81
0.64
1.19
0.78
0.90
0.74
0.34
1.23
-0.45
-0.21
0.26
-1.11
1.15
0.69
0.77
0.83
0.62
-0.13
0.26
0.82
1.08
1.34
1.54
1.56
1.87*
1.64
1.50
1.51
1.14
1.41
1.54
1.68*
1.83*
-0.17
0.03
0.17
0.44
0.36
-1.02
-0.83
0.32
-0.54
0.16
0.99
-0.25
-0.12
0.69
1.50
0.62
0.79
-0.17
-0.09
0.02
0.24
0.37
-0.08
-0.38
-0.25
-0.41
-0.34
-0.03
-0.10
-0.04
0.06
0.45
0.59
0.76
-1.86
0.55
-0.83
-0.71
0.23
0.87
-0.54
0.26
-1.01
0.11
0.12
-1.21
0.87
-1.57
-0.55
-0.40
1.02
-1.86
-0.93
1.24
-1.43
-1.17
-0.72
-0.87
-0.72
-1.01
-0.93
-0.85
-1.16
-0.87
-1.26
-1.36
-1.42
-1.13
* Significant at the 0.05 level
- 74 -
Results of the two-way analysis of variance are reported in
Table 4-5. The null hypothesis of no difference among the means of
the three portfolios was rejected at the 0.036 level. The actual
risk-adjusted returns on the experimental portfolio over the entire
test period were considerably higher than those achieved on the two
control portfolios. These results support the earlier finding that
the abnormal returns on the experimental portfolio were larger than
the abnormal returns achieved on the two control portfolios during the
test period.
The null hypothesis of no difference among the eleven informa-
tion treatment effects and the null hypothesis of no interaction
effect could not be rejected at a statistically significant level.
The Duncan's Multiple Range Test was used to test the null hypothesis
that all thirty-three means are equal. This hypothesis could not be
rejected at the 0.05 level. These results suggest that the fore-
casting action by itself does not result in an upward stock price
revision significantly different from that occurring at random.
With respect to the industry-wide effect, the forecasting
action by itself does not convey favorable information to non-fore-
casting firms in the same, or similar, industry. The prediction
errors associated with the matched control portfolio were negative
during the forecast announcements week. The behavior of the
cumulative prediction errors prior to the forecasts disclosure week
does not indicate that the forecasting action was motivated by
industry-wide factors.
- 75 -
Table 4-5
Total Neutral Analysis
of Variance
Test F-Value PR>F
Type of portfolio effect 3.33 0.036
Information treatment effect 0.26 0.990
Interaction effect 0.87 0.624
Comparison of all means 0.83 0.73
- 76 -
To summarize, results of prediction errors tests indicate that
the forecasting action by itself results in statistically significant
abnormal returns during the forecast announcements week. The two-way
analysis of variance test, using total stock returns, provided a
contradictory result. Both of these tests, however, are sensitive to
the existence of outliers. Removal of outliers changed the conclusion
of the prediction errors test. The results of the trimmed prediction
errors test support the conclusion that the forecasting action by
itself does not result in a significant upward stock price revision.
All tests, however, suggest that abnormal returns, and actual
portfolio returns, during the test period are larger for the
experimental portfolio than for the two control portfolios. Finally,
the forecasting action by itself neither influences stock returns of
non-forecasting firms, nor is it motivated by industry-wide factors.
The Effect of Newness of Management Earnings Forecasts
The positive and the negative portfolios are the two high-level
newness portfolios. Results are reported below for each individual
portfolio.
A The Positive Portfolio
Table 4-6 reports the Z statistic associated with standardized
prediction errors for each of the three portfolios. Figure 4-2
portrays the behavior of the cumulative standardized prediction errors
- 77 -
Table 4-6
Total Positive
for Zvt and
Values
ZWL
Experimental Matched Control Random Control
Week
Zvt ZWL Zvt ZWL vt ZWL
1.08
1.70*
1.02
1.77*
-1.35
0.38
1.38
0.89
6.44*
0.57
-0.41
-0.11
-0.23
-0.43
0.51
0.38
1.16
1.08
1.96*
2.20*
2.79*
1.89*
1.88*
2.26*
2.43*
4.43*
4.39*
4.06*
3.85*
3.64*
3.39*
3.41*
3.20*
3.39*
-0.18
-1.05
1.31
0.43
0.14
-0.58
0.13
0.60
1.90*
0.91
0.79
0.73
0.60
-0.26
0.15
0.81
0.57
-0.18
-0.87
-0.05
0.26
0.30
0.03
0.08
0.29
0.69
1.15
0.86
1.03
1.15
1.04
1.05
1.21
1.04
0.31
0.34
0.80
1.82*
-0.11
-0.27
0.35
2.15*
0.36
1.54
-1.55
1.35
-1.11
-0.48
0.22
-0.61
1.57
0.31
0.46
0.84
1.63
1.42
1.18
1.22
1.91*
1.92*
2.30*
1.73*
2.05*
1.66*
1.47
1.48
1.28
1.22
* Significant at the 0.05 level
- 78 -
-5 -4 -3 -2 -1 0 1 2 3 4 5 6 7
Experimental
Match Control -
Random Control *
Figure 4-2
Positive Group
Cumulative Standardized Prediction Error (2: \7)
1.1
- 79 -
for the three portfolios. Results for the experimental portfolio can
be summarized as follows:
1 the null hypothesis of no information content was rejected
for the forecast announcements week and for two other weeks
(t-7 and t-5) at the 0.05 level.
2 the largest positive mean prediction error was associated
with the forecast announcements week.
3 with the exception of the first week, the cumulative
prediction errors were statistically significant at the
0.05 level for the entire test period. The size of the
cumulative prediction errors associated with the
experimental portfolio were considerably larger than those
associated with either of the two control portfolios.
Results for the control portfolios indicate that:
1 the null hypothesis of no information content during the
forecasts disclosure week was rejected at the 0.05 level
for the matched control portfolio but not for the random
control portfolio.
2 the largest positive mean prediction error was associated
with the matched control portfolio during the forecasts
disclosure week.
3 the random control portfolio exhibits statistically
significant positive prediction errors during weeks t-5 and
t-1.
- 80 -
Results of the distribution tests are shown in Table 4-7.
These results indicate that the null hypothesis of no information
content was rejected for both the experimental and the matched control
portfolios during the forecast announcements week but was not rejected
for the random control portfolio. Other weeks in the test period
exhibited slightly lower Z values than those reported for the
standardized predictio errors test.
The Z values for the trimmed standardized prediction errors
test are shown in Table 4-8. Results for each of the three portfolios
are:
1 the null hypothesis of no information content was rejected
for the experimental portfolio during the forecast an-
nouncements week but not for any other week in the test
period.
2 the null hypothesis of no information content could not be
rejected at the 0.05 level for the matched control port-
folio during the entire test period.
3 the null hypothesis of no information content could not be
rejected at the 0.05 level for the random control portfolio
during the entire test period.
The above results suggest that the positive prediction errors
observed for the experimental portfolio during the forecast announce-
ments week represent a general pattern and were not influenced by a
few large outliers. This conclusion is not equally valid for the
matched control portfolio where the observed effect seems to have been
influenced by a few large outliers.
- 81 -
Table 4-7
Total Positive
for Ze and
et
Values
Z
vt
Experimental Matched Control Random Control
Week
Zett t Zvt Zet Zvt
1.04
1.12
1.35
1.26
-1.44
0.14
1.70
0.97
5.73*
0.71
-0.02
0.03
-0.02
-0.24
0.56
-0.19
0.18
1.08
1.55
1.14
1.56
-1.49
0.26
1.38
0.83
5.96*
0.59
-0.44
0.05
-0.18
-0.58
0.50
-0.40
0.98
-0.03
-0.91
0.84
0.01
-0.59
-1.16
0.45
0.26
2.03*
1.07
-0.54
0.16
0.62
-0.48
0.33
0.92
-0.91
-0.09
-1.31
1.05
0.25
-0.21
-0.78
0.13
0.56
1.64*
0.89
-0.79
0.39
0.40
-0.23
0.18
0.84
-0.75
0.41
-0.21
1.20
1.57
-0.08
-0.36
0.92
2.13
0.11
1.39
-1.34
1.27
-0.70
-0.69
0.12
0.16
S1.22
0.45
0.28
0.93
1.60
-0.30
-0.25
0.53
1.88
0.30
1.58
-1.64
1.31
-1.05
-0.44
0.33
-0.42
1.54
* Significant at the 0.05 level
i
- 82 -
Table 4-8
Total Positive Trimmed
Mean Values for Zv
t
Week Experimental Matched Control Random Control
0.40
0.71
0.09
0.00
-1.14
-0.11
1.21
0.92
5.71*
0.41
-0.78
-0.46
0.34
-0.03
-0.17
-0.57
0.74
0.09
1.17
0.72
0.54
0.34
-0.57
0.22
0.75
1.18
0.20
-1.06
0.31
0.17
0.33
0.43
0.82
-0.38
-0.25
0.17
0.56
1.47
-0.77
-0.27
0.00
0.43
-0.25
1.38
1.20
0.93
1.36
0.49
0.65
0.62
1.11
* Significant at the 0.05 level
_ .1 _
- 83 -
Results of the two-way analysis of variance test are reported
in Table 4-9. These results confirm the finding of the prediction
errors tests. The null hypothesis of no difference among the means of
the three portfolios could not be rejected at the 0.05 level. The
null hypothesis of no difference among the eleven information treat-
ment effects was rejected at the 0.0005 level. The largest positive
overall weekly mean return was observed during the forecast announce-
ments week. The null hypothesis of no interaction effect could not be
rejected at a statistically significant level. The Duncan's Multiple
Range Test rejected the null hypothesis that the thirty-three weekly
mean returns are equal at the 0.011 level. The largest positive
weekly mean return was associated with the experimental portfolio
during the forecast announcements week.
In sum, findings of all tests suggest that:
1 management forecasts of earnings which exceed market expec-
tations cause a statistically significant upward stock
price revision for forecasting firms but not for non-
forecasting firms in the same, or similar, industry.
2 the observed effect for the experimental portfolio
represents the general pattern of the samples and is not
influenced by a small number of large outliers.
3 the size of the cumulative prediction errors for the three
portfolios prior to the forecast announcements week
indicates that the management earnings forecasts are not
motivated by industry-wide or economy-wide factors.
- 84 -
Table 4-9
Total Positive Portfolios
Analysis of Variance
Test F-value PR>F
Type of portfolio effect 0.71 0.49
Information treatment effect 3.15 0.0005
Interaction effect 1.01 0.445
Comparison of all means 1.66 0.011
- 85 -
B The Negative Portfolio
Table 4-10 reports the Z statistic for the standardized
prediction errors and the cumulative prediction errors. Figure 4-3
portrays the behavior of cumulative average standardized prediction
errors for the three portfolios. Results for the experimental
portfolio suggest that:
1 the null hypothesis of no information content was rejected
at the 0.05 level for the forecast announcements week and
for four other weeks in the test period. Three of these
four weeks preceded the forecast announcements week.
2 the largest negative weekly mean prediction errors occurred
during the forecast announcements week.
3 with the exception of t-8 and t-1, all weeks preceding the
forecast announcements week exhibited negative prediction
errors. As a consequence, the cumulative prediction error
was negative beginning at t-7 and continuing throughout the
test period.
4 the size of the negative cumulative prediction errors for
the experimental portfolio was considerably larger than
those reported for the two control portfolios.
Results for the two control portfolios suggest that:
1 the null hypothesis of no information content was rejected
for the matched control portfolio during the forecast
announcements week and during t-7. The same null
- 86 -
Table 4-10
Total Negative Values
for Zvt and ZWL
vt WL
Experimental Matched Control Random Control
Week
Zvt ZWL Zvt ZWL vt ZWL
0.56
-1.25
-2.14 *
-2.99 *
-0.33
-0.18
-1.79
0.26
-5.20*
-0.54
-1.04
-0.22
-0.06
-1.90*
1.80
0.41
0.05
0.56
-0.48
-1.63
-2.91*
-2.75*
-2.58*
-3.07*
-2.78*
-4.35*
-4.30*
-4.42*
-4.29*
-4.14*
-4.50*
-3.87*
-3.64*
-3.52*
-0.85
-2.09*
1.07
-0.89
1.12
-0.13
2.79
1.75
-2.82*
-1.59
1.50
-1.20
2.25
2.05
-0.15
-0.38
-0.61
-0.85
-2.16*
-1.15
-1.44
-0.80
-0.78
0.33
0.93
-0.07
-0.57
-0.09
-0.43
0.29
0.83
0.77
0.61
0.48
0.60
0.89
-0.12
0.04
0.47
-0.95
0.60
1.17
-0.12
-0.41
-0.50
-0.01
-0.32
0.43
2.15
0.82
1.24
0.60
1.48
0.79
0.70
0.84
0.42
0.12
0.53
0.43
0.27
0.11
0.11
0.01
0.13
0.68
0.86
0.52
* Significant at the 0.05 level
- 87 -
1.00
0.80
0.60
0.40
0.20
0
-0.20
-0.40
-0.60
-0.80
-1.00
-1.20
-1.40
-1.60
-1.80
-2.00
-2.20
-2.40
Figure 4-3
Negative Group
Cumulative Average Standardized Prediction Errors (2V7 \)
H1I
experimental -
iMach Control -
Random Control ** **
- 88 -
hypothesis could not be rejected during the test period for
the random control portfolio.
2 the largest negative weekly mean prediction errors occurred
for the matched control portfolio during the forecast
announcements week.
The above findings are fully supported by the results of the
distribution tests. Table 4-11 reports the Z statistic for et and vt
during the test period. The trimmed standardized prediction errors
test provides additional support for these findings and suggests that
the observed results are a general pattern of the data and not due to
the influence of a few large outliers. Table 4-12 reports essentially
the same results following the removal of outliers.
With respect to the industry-wide effect, all three tests
reported a significant abnormal return for the industry-matched
control portfolio during the forecast announcements week. These
results suggest that the management earnings forecasts contain infor-
mation relevant to establishing equilibrium prices of non-forecasting
firms in the same, or similar, industry. More importantly, however,
is the observed behaviour of abnormal returns associated with the
three portfolios prior to the forecast announcements week. The
cumulative abnormal returns for the eight weeks preceding the forecast
announcements week were positive for the industry-matched and the
random control portfolios. This suggests that neither industry-wide
factors nor economy-wide factors induced the negative management
- 89 -
Table 4-11
Total Negative Values
for Z eand Zv
et vt
Experimental Matched Control Random Control
Week -I.
Zet Zvt Zet Zvt Zet Zvt
-0.03
-1.19
-2.42*
-3.40*
- .76
0.35
-1.97*
0.08
-3.46*
-0.34
-2.43*
-0.80
-0.10
-1.38
1.55
0.84
-0.29
0.56
-1.38
-2.55*
-3.37*
-0.76
0.29
-2.26*
0.21
-3.71*
-0.19
-2.34*
-0.80
-0.02
- .74
1.34
0.64
-0.16
-1.25
-1.77*
1.68
-1.21
1.12
-0.8
2.21
1.14
-2.55*
-1.99*
0.56
-1.09
2.24
1.72
-0.78
-0.66
-1.40
-0.86
-1.72*
1.29
-1.00
1.19
-0.43
2.64
.94
-2.49*
-1.61
0.76
0.14
1.90
1.69
-0.47
-0.49
-1.46
0.48
0.55
0.44
-0.35
0.32
-1.06
-0.90
1.27
- .22
-0.66
-0.20
-0.27
0.09
0.48
2.77
1.06
0.22
0.42
0.78
0.27
-0.03
0.49
-1.03
-0.80
0.98
0.05
-0.75
-0.43
-0.40
0.01
0.29
2.37
1.03
-0.56
4 4
* Significant at the 0.05 level
- 90 -
Table 4-12
Total Negative
Trimmed Mean
Values for Zvt
Week Experimental Matched Control Random Control
0.66
-1.18
-2.06*
-2.62*
-0.65
0
1.26
0.51
-4.45*
-0.43
-0.78
-0.18
-0.35
-1.30
1.32
0.18
0.30
-0.16
-1.98*
1.00
-1.13
0.83
0.29
1.83
0.66
-2.99*
-1.02
-0.37
-1.19
1.42
0.49
-0.05
-0.05
-0.53
0.63
0.57
1.20
-0.46
0.28
0.97
-0.85
0.53
-0.22
0.43
0.47
-0.04
0.25
0.37
1.51
0.69
1.58
* Significant at the 0.05 level
- 91 -
earnings forecasts. Rather, the disclosure of these forecasts
resulted in a downward revaluation of other firms in the same, or
similar, industry.
The two-way analysis of variance test provides additional
evidence consistent with the results of prediction errors tests.
Results for the analysis of variance test are summarized in Table
4-13. The null hypothesis of no difference among the three portfolios
was rejected at the 0.01 level. The null hypothesis of no difference
among the eleven information treatment effects was rejected at the
0.0002 level. The largest negative weekly mean returns were observed
during the forecast announcements week. The null hypothesis of no
interaction effect could not rejected at a statistically significant
level. The Duncan's Multiple Range Test rejected the null hypothesis
that all thirty-three weakly mean returns are equal at the 0.0002
level. The two largest negative weekly mean returns occurred during
the forecast announcements week for the experimental and the industry
matched control portfolio respectively.
In sum, the positive (negative) news forecasts were accompanied
by significant upward (downward) stock price revisions during the
forecast announcements week. The behavior of the cumulative
prediction error indicated proper anticipation of the earnings fore-
casts content prior to the announcements week. Both classifications
of forecasts induced an industry-wide effect, although the observed
effect for the negative news forecasts was more profound than was the
observed effect for the positive news forecasts.
- 92 -
Table 4-13
Total Negative Portfolios
Analysis of Variance
Test F-Value PR > F
Type of portfolio effect 4.57 0.010
Information treatment effect 2.73 0.002
Interaction effect 1.19 0.254
Comparison of all means 1.88 0.002
- 93 -
The Effect of Imputed Accuracy of Management Earnings Forecasts
The two high-level newness forecasts are classified as either
positive or negative forecasts. Each of these two classifications was
subdivided into two unequal portfolios according to the degree of
imputed accuracy ascribed to these forecasts. The result is four
portfolios; with high and low imputed accuracy for both the positive
and the negative classification. Testing the effect of imputed
accuracy of management earnings forecasts is accomplished by an
individual examination of each of the four portfolios and by a
comparison of the high-level and low-level imputed accuracy portfolios
for each forecast classification. Separate examination of each port-
folio will indicatewhether the information content of management
earnings forecasts is a function of newness alone or the combination
of newness and imputed accuracy of these forecasts. The comparison
between the high and the low level imputed accuracy portfolios for
each classification of forecasts will indicate the extent to which the
information content is dependent upon the level of imputed accuracy
ascribed to forecasts.
Results are reported below for the two positive portfolios
followed by results for the two negative portfolios.
Positive Forecasts-High Imputed Accuracy
Table 4-14 reports the Z statistic for standardized prediction
errors and cumulative prediction errors. Figure 4-4 portrays the
|
Full Text |
PAGE 1
NEWNESS AND IMPUTED ACCURACY OF MANAGEMENT FORECASTS: AN EMPIRICAL INVESTIGATION by ROBERT L. BESHARA A DISSERT ATION PRESENTED TO THE GRADUATE COUNCIL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 1981
PAGE 2
ACKNOWLEDGEMENTS The author gratefully acknowledges the assistance and support of the Supervisory Committee in the completion of this dissertation. Dr. Rashad Abdel-Khalik provided valuable guidance and criticisms. Dr. Bipin Ajinky offered assistance and encouragement far beyond the call of duty. D. Robert Radcliffe offered helpful advice and suggestions. Gratitude is also expressed to Dr. Barnhill for his editorial comments, and Ms. Susan Robinson for the cheerful typing assistance. Special acknowledgements are made to the Clarkson and Gordon Foundation and Carleton University for providing financial assist a nce. ii
PAGE 3
TABLE OF CONTENTS ACKNOWLEDGMENTS .............................................. ABSTRACT ......................... ........................... .. Chapter I. STATEMENT OF THE PROBLEM AND PURPOSE OF THE DISSERTATION ... ......... ................ .......... ..... Introduction ....................................... Statement of the Problem ............................ Purpose of the Dissertation ......................... II. PRIOR RESEARCH ON THE INFORMATION CONTENT OF MANAGEMENT FORECASTS OF EARNINGS The Foster Study .................................... The Gonedes, Dopuch and Penman Study ................ The Pa tell Study ................................. The Jaggi Study .................................... The Penman Study .................................... The Nichols and Tsay Study .......................... Summary and Limitations of Existing Research ........ Not es .............................. III. DATA, METHODOLOGY, AND STATISTICAL TESTS ............... Data ......... ..................................... Definitions ....................................... . Experimental Portfol-ios ............................ Control Portfolios ................................. Statistical Tests~ ..................... ............ Residual Analysis ................. .. .......... ..... A Standardized Prediction Errors ........ ...... B Unstandardized Prediction Er:rors ......... ... Analysis of Variance ................ ............... Hypotheses .......................................... A Testing the Effect of Forecasting Action ... B Testing the Effect of Newness of Earnings .... Forecasts .................................... C Testing the Effect of Imputed Accuracy ....... D Testing the Effect of Management Forecasts on Other Firms in the Same Industry .......... Notes ............................. ................. iii Page ii V 1 1 4 8 10 10 13 16 20 21 26 30 35 36 38 42 42 44 46 46 47 52 52 56 56 58 59 61 63
PAGE 4
Chapter IV. RES~ TS Page 64 Introduction. . . . . . . . . . . . . . . . . . . . 64 Total Sample. . . . . . . . . . . . . . . . . . . 65 The Effect of the Forecasting Action................. 68 The Effect of Newness of Management Earnings Forecasts. . . . . . . . . . . . . . . . . . . . . . 76 A The Positive Portfolio........................ 76 B The Negative Portfolio........................ 85 The Effect of Imputed Accuracy....................... 93 Positive Forecasts High Imputed Accuracy........ 93 Positive Forecasts Low Imputed Accuracy......... 100 Comparison Between the High-level and the Lowlevel Imputed Accuracy. . . . . . . . . . . . . . 109 Negative Forecasts High Imputed Accuracy........ 113 Negative Forecasts Low Imputed Accuracy......... 120 Comparison Between the High-level and the Lowlevel Imputed Accuracy. . . . . . . . . . . . 125 V. SUMMARY AND CON CL US IONS . . . . . . . . . 131 BIBLIOGRAPHY................................................... 137 BIOGRAPH I CAL SKETCH . . . . . . . . . . . . 14 2 iv
PAGE 5
Abstract of Dissertation Presented to the Graduate Council of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy NEWNESS AND IMPUTED ACCURACY OF MANAGEMENT FORECASTS: AN EMPIRICAL INVESTIGATION By Robert L. Beshara December, 1981 Chairman: Dr. Rashad AbdelK hali k Major Department: A ccounting The purpose of th i s di s sert a ti o n w as to e mp i rically examine the information content of m anagement forecasts o f earni n gs. The following four research questions were exa mined: 1. Do management forecasts of earni ng s conve y in f or m ation in excess of what is alr ea dy availabl e to mar ke t pa rt icipan t s? 2. Does the forecasting action by i tsel f c on vey favora b le information to investors ? 3. Does the information content d e pend upon the level o f newness and i mputed accuracy a scribed to manageme nt earnings forecasts? V
PAGE 6
4. Do management earnings forecasts convey information pertinent to establishing equilibrium prices of non-fore casting firms in the same, or similar, industry? The experimental sample consisted of 401 management earnings forecasts covering the period 1970-77. The total sample 0f management earnings forecasts was classified into neutral, positive, and negative fore casts according to the magnitude and the sign of the new information revealed by each forecast. Two control samples were utilized to detect industry-wide and economy-wide effects. indicate that: Results of the study 1. The positive (negative) news of forecasts induced an upward (downward) stock price revisions during the forecasts announcements week. This revision was found to be statistically significant. 2. The forecasting action by itself induced an upward stock price revision during the forecast announcement week. Such revision, however, was neither sufficiently strong nor was it the general pattern of the neutral portfolio to be considered statistically significant. 3. Stock price revisions associated with the high-level imputed accuracy portfolio were considerably higher than those associated with the low-level imputed accuracy portfolio for the positive news management earnings forecasts but not for the negative news forecasts. vi
PAGE 7
4. Negative news management earnings forecasts seem to conve y industry-wide information, while positive news forecasts are perceived to be firm specific events. This conclusion, however, should be tampered by the fact that the sample size for this classification of forecasts is relatively small. vii
PAGE 8
CHAPTER I STATEMENT OF THE PROBLEM AND PURPOSE OF THE DISSERTATION Introduction The recent attempt by the Securities and Exchange Commission to regulate the disclosure of management earnings forecasts has received considerable attention at both the pragmatic and the academic levels. The widely held belief that management forecasts of earnings are relied upon in investment decisions and the assertion that management has access to information not available to outsiders have motivated several empirical research studies on the subject. Disclosure regulations of management forecasts of earnings involve two closely related issues. The first issue is the extent to which management earnings forecasts convey information not already available to investors. The second issue is whether disclosure regulations of earnings forecasts by management are consistent with the optimal allocation of resources in society. Gonedes et al. (1976) point out the relationship between these two issues as follows (p. 90): If one is confident that forecasts of, say, income convey no information pertinent to valuing firms, 1
PAGE 9
2 then the entire debate over forecasts disclosure is of little interest, insofar as e x ternal accounting is concerned. A necessar y condition for public disclosure of forecasts to be consistent with Pareto optimality is that the forecasts convey information pertinent to resource allocation. Empirical research directly concerned with the information content of management earnings forecasts has been scanty. M ost of the existing research has focused on either the absolute accuracy of management forecasts or the relative accuracy of management f orecasts as compared to historically-based models and/ or analysts' forecasts. Confirming this observation are Gonedes et al. when they state (p. 93): The available empirical work on information content deals with earnings forecasts voluntarily made available by firms. The bulk of this work examines the prediction performance or 'accuracy' of earnings forecasts. The implicit assumption here is that greater accuracy, however measured, corresponds to greater information content. Empirical research dealing with management forecasts accuracy suffers from a major deficiency. That deficiency is the lack of a theoretical structure capable of explaining the relationship, if any, between ex-post accuracy and the information c ontent issue. et al. (p. 94) point this out when they state: There is a more fundamental deficiency in th ese prediction performance studies and the a pri o r i arguments on information content. Speci f i ca l ly, they are not based upon any explicit theore tical structure that connects their fr a mew ork to resource allocation decisions ma d e under uncertainty. In order to determine whether forecasts convey information pertinent t o valuing a firm, one needs a theor e ti ca l m odel tha t identifies the determinants f or firms' val u e s, conditional on the way in whi ch res ou rce allocation decisions are made. G on e d e s
PAGE 10
3 The efficient-markets hypothesis discussed by Fama (1970) and the capital asset pricing model developed by Sharpe (1964) and Lintner (1965) provide the theoretical structure for studies dealing with the effect of new information on equilibrium stock prices. The semi-strong form of the efficient capital markets maintains that prices of securities reflect all publicly available information and that the market reaction to new information occurs instantaneously in an unbiased fashion. The capital asset pricing model, under certain restrictive assumptions asserts that the price (return) of a security depends on the risk-return trade-off set by the market and the probability distribution of the expected returns on the security considered jointly with the market portfolio. Beaver (1968, p. 68) considers an item to have information content "if it leads to a change in investors' assessment of the probability distribution of future returns (or prices) such that there is a change in equilibrium value of the current market prices." Studies by Foster (1973), Gonedes et al. (1976), Patell (1976), Penman (1978), Jaggi (1978), and Nichols and Tsay (1979), comprise the major empirical research on the information content of management forecasts. All of these studies utilized the theoretical structure provided by the efficient-markets hypothesis and the capital asset pricing model. Results of these studies are somewhat conflicting and incomplete. Several research questions on ~he information content of voluntarily disclosed management forecasts remain unexplored.
PAGE 11
4 Statement of the Problem The information content of management earnings forecasts is a function of several attributes. Patell (1976, p. 248) points out that, "the extent to which an earnings forecast alters investors' beliefs may be a function of (at least) three attributes," These are: 1 the imputed accuracy of the forecast, 2 newness of the forecast; that is, the extent to which the forecast's content is not already available to market participants from other sources, and 3 the motive underlying the voluntary disclosure of earnings forecasts. The above attributes are neither independent of each other, nor are they totally exhaustive. For example, it is reasonable to expect imputed accuracy to be correlated to the level of newness of the forecast. Also, the imputed accuracy of forecasts may depend upon industry volatility and the motive for forecast disclosure. Imputed accuracy of management forecasts at the time of forecast disclosure is difficult to assess and it has not been considered in the existing literature. The newness of management earnings forecasts has been typically based upon naive forecasting models. The available research on the information content of management forecasts defines newness as the deviation of the forecast from the naive model(s). Foster (1973),
PAGE 12
5 Patell (1976), and Penman (1978) utilized historically-based naive models as a proxy for current market expectations of earnings. These models, although practically convenient, do not incorporate existing and expected factors affecting the firm, the industry in which the firm operates, and the economy at large. These factors, however, directly influence investors' expectations of earnings and should not be ignored. The motive attribute has been examined, albeit not extensively. The signalling theory developed by Spence (1973, 1974), Riley (1975, 1977), and extended to disclosure regulations in financial markets by Ross (1977) provides a theoretical structure for voluntary disclosure of inside information by corporate executives. The basic contention of the signalling theory is that in competitive markets incentives exist for management to publicly disclose inside information relevant to the valuation of their firms by outsiders. These incentives will exist in competitive markets irrespective of the classification of the information as 'good'or'bad'news. The incentive-signalling theory is best summarized by Ross (1977, p. 20), as follows: In general, there is a hierarchy of firms from best to worst based on the relative changes in values that would occur if their inside informa tion was made public. The incentive-signalling mechanism provides a structure that managers use to disclose their information in such a way that outsiders in the market believe it. Those with the best news distinguish their firms from those with the next best and so on down the line. At the bottom of the hierarchy, those with the worst news would like to suppress it, but since it is
PAGE 13
6 not in their interest to offer guar a ntees provided by those with better news, the worst news will also be signalled. Several technologies are available to management in competitive markets to sort their firms on value relevant attributes. Public disclosure of quantitative forecasts of earnings per share is one possible technology. Change in dividend polic y reporting quarterly earnings, changes in capital structure, and prov i ding a qualitative statement regarding future earnings are examples of o ther signalling technologies. On the other hand, earnings f o r e cast d i sclosure c a n not be totally ascribed to deli berate revaluation of t h e f irm by management. Other motives for forecast disclosur e such a s the desire to raise new capital, justifi c ati o n f o r market ex pansion, merger decisions, pressure from the fi n ancial cornmuni t y an d other factors, can not be ruled out. Penman (1978) adopted the incenti ve sig n a ll i n g theory to examine the voluntary disclosur e of fo r eca st e d ea r n i n gs. M or e specifically, the author posed the following researc h q uest ion (p 2): Do investors receive forecast inform a tion po t en t ial l y available from all firms ? The empirical results of th e stu d y do not provi de a n affirmative answer to this question. The se results, however, are not inconsistent with the incentive-signalli ng theory mainly due t o the fact that several signalling technolo g i es a r e available to ma n age m e n t. Furthermore, the iterative mechani sm of t he incen t ive sig n a ll i n g theory will result in the screeni ng of companies w h ich d o not issue
PAGE 14
7 quantitative forecasts of earnings based upon their own signals as well as the forecasting action of other firms. Penman (p. 46) points out that: the firms which do not issue forecasts are, in fact, screened out as a group by the forecasting action of others. A 'no forecast' is in fact a forecast that screens firms into this group and value them accordingly. The above discussion suggests that the information content of management earnings forecasts has implications for non-forecasting firms. This is conceptually appealing for two reasons. First, the mechanism of the incentive-signalling theory is an iterative process through which firms will find it advantageous to reveal their v a lue relevant attributes. This process need not take place on an individual firm basis. Instead, it can take place through grouping of firms according to similarity of value attributes. Second, on an a priori basis, i t can be argued that an earnings forecast by management is a quantification o f the expected impact of general economic conditions on their firm. To the extent that other firms in the same industry are affected by common economic forces, the f orecast can also be vie wed as a n assessment of the industry s i tuat i on. As such, mana ge ment f or e casts conve y information about the forecasting firms as we l l as the i ndu st r y in which they operate. Th a t is, the informat i on cont ent of man ag ement forecasts may extend to other firms in the same (or similar) in du st r y This, however, is an empirical question whi c h has n ot b een exami ned in the current literature.
PAGE 15
8 Purpose of the Dissertation The purpose of this dissertation is to e x amine the extent to which management forecasts convey information not already available t o market participants. via other sources. The methodology utilized in this study explicitly considers market expectations of earnings per share immediately prior to the release of management forecasts. Market expectations are measured by available analysts' forecasts. Thus, newness is defined as the deviation between the management forecast and analysts' forecasts available at the time of the management forecast release. Imputed accuracy is measured by the revision of analysts' forecasts towards the management f orecast. In effect, a forecast is considered to possess a high level of imputed accuracy if analysts revise their estimates of earnings per share closer to that of management. This study provides empirical evidence aimed at answering the following four related research questions. 1. Do management earnings forecasts provide infor ma ti o n in excess of that already available in the m a r k et a n d presumably incorporated into anal y sts' forecasts ? 2. Do management earnings forecasts which mer ely co n firm investors' expectations result in an upward stock price revision? In other words, does the for e cast ing action by itself convey favorable information to i n vest ors. 3. Does the information content, if a ny, depend u pon the levels of newness and imputed accurac y a s cribed t o manage ment earnings forecasts?
PAGE 16
9 4. Do management earnings forecasts convey information pertinent to establishing equilibrium prices of non-forecasting firms in the same, or similar, industry?
PAGE 17
CHAPTER II PRIOR RESEARCH ON THE INFORMATION CONTENT OF MANAGEMENT FORECASTS OF EARNINGS Empirical research directly concerned with the information content of management forecasts has been relatively scarce. A s stated earlier, the majority of the empirical research to date deals with the absolute and relative accuracy of management forecasts on a n ex-post basis. Six published studies comprising the major empirical research on the information content of earnings forecasts by managements are identified and reviewed in this chapter. The purpose of this review is to highlight the scope and limitations of the e x isting research on the topic. The Foster Study (1973) This study examined the information content of estimates of annual earnings per share by company officials. These e stimat e s were made after the end of the fiscal year but prior to the publ i cation of 10
PAGE 18
11 preliminary earnings per share (EPS). Thus, in effect Foster examined the information content of unaudited EPS. This is evident from the level of accuracy of EPS estimates. The average absolute percentage error was 1.8 percent (p. 27). For all practical purposes, however, such a minute error could be the result of the rounding of EPS estimates by company officials. The study examined 68 EPS estimates for the period 1968-1970. The majority of estimates (87 percent) had a time-lapse between the EPS estimates and the preliminary earnings announcement of 21 trading days. The mean and the median for the time-lapse were 17 and 18 trading days respectively. Foster tested for the individual and aggregate market behavior using a methodology similar to that employed by Beaver (1968). Both tests confirmed the hypothesis that the unaudited EPS convey information to investors. Thus, abnormal returns can be achieved by employing a trading strategy "based on knowledge of EPS estimates five days before it is publicly released . ( p. 33). The study utilized five annual and five quarterly naive models. For each naive expect a tions model a first difference model was constructed. This resulted in a total of twenty models, ten level and ten first difference. Forecasts were classified as positive (negative) news depending upon the sign of deviation from the naive expectations model. Two purchase and short-sale strategies were adopted f or the positive and negative news forecasts respectively. The first
PAGE 19
12 strategy was adopted five days prior to the EPS announcement to the date the preliminary earnings report was announced. The second strategy was adopted from the date the EPS estimate was announced to the date of the preliminary earnings report. The average abnormal return for the first trading strategy was 1.61 percent for the ten annual models and 1.48 percent for the quarterly models. The average abnormal return for the second strategy was much lower; 0.31 percent for the ten annual models and also 0.31 percent for the ten quarterly models. explained as follows: This result (p 33) is Up to the time the company officials announce the EPS estimate, there is still opportunity fo r an abnormal return. . Once the official's EPS estimate is publicly released, there is a stock price adjustment such that trading strategy based on that information yields lower abnormal returns. The author points out, however, that abnormal returns "do not indicate the extent to which the observed results were a general pattern for all firms or was due to extreme observations" (p. 34). A chi-square test was used to examine the frequency of abnormal returns. One out of twenty models was significant at the .01 level; two other models were significant at the .OS level and ten models were significant at the .1 level. This result led to the conclusion that estimates of earnings per share possess information content and that the effect is a general pattern in the sample.
PAGE 20
13 This study clearly highlights the effect of naive expectations models on results obtained. Different models yield different results. Although the author utilized a large number of models, the effect of possible mis-classification of forecasts on the results of the study remains unknown. The Gonedes, Dopuch, and Penman Study (1976) The purpose of this study was to evaluate the information content of management earnings forecasts. The authors utilized a methodology that (p. 107) "involves time-series analysis and tests for information content." firm in the sample. This required a series of forecasts for each Due to the fact that management earnings forecasts are not available from the same source and for several con secutive time periods, the authors used analysts' forecasts as proxies" for management forecasts. The sample included 148 firms drawn from the set of firms covered by the Standard and Poors Earnings Forecaster. One criterion used in selecting the firms was that "each firm had to have con secutive forecasts from at least one source during the last half of 1967 and the last half of 1968" (p. llO). This approach led to "having twenty-four forecasts for each of the 148 firms . in the sample" ( p. llO).
PAGE 21
14 Each firm's forecast was scaled by "the per share closing price of the firm's common stock on the tenth trading day preceding the forecast's publication date "(p. 113). Scaled forecasts were classi fied into four groups (quartiles) according to the magnitude of the scaled forecast. Portfolios were formed for each forecast date and for each quartile (p. 113) "of the cross-sectional distribution of scaled forecasts." One control portfolio, of equal size, was constructed to match each sampling portfolio. The result was 192 portfolios (24 forecast dates x 4 quartiles x 2 portfolios), each composed of 37 shares. All portfolios were minimum variance portfolios with beta equal to unity. The methodology used in this study is discussed in Gonedes and Dopuch (1974), Gonedes (1975), and summarized in the current study (pp. 111-114). Briefly, the approach involves a comparison of two distribution functions; the first is the distribution of returns on the sampling portfolios, and the second is the distribution of returns on the control portfolios. Based on the assumption that the distribution of returns is multivariate normal, the comparison involves tests of equality of means and variances. However, an inequality between the means of the two distributions is a sufficient condition for the inequality of the two distributions. Acco rdingl y, "only results for tests on means are reported below" ( p. 112). Results of the study can be summarized as follow s:
PAGE 22
15 1 the null hypothesis of no information content was rejected for the total sample. The value of the F statistic was "substantially above the .995 fractile (p. 123). 2 disaggregating the data, the null hypothesis was rejected for the first quartile ( lowest scaled forecasts), but not for the other three quartiles. This result led the authors to conclude (p. 133) that: our empirical results on income forecasts are inconsistent with the statement that those forecasts convey no such information. M oreover, it appeared that much of the information content of forecasted earnings can be ascribed to the implication of extremely low ( scaled) forecasts. Forecasts not in this class seemed to reflect little information beyond that already available, such as the information reflected in available forecasts of macro-economic conditions. This observation seems to increase the importance of assessing the extent to which substitutes are available for forecasts of firms' earnings and the properties of these substitutes. 3 testing for the effect of portfolio risk, the authors reconstructed all sampling and control portfolios for different levels of risk. They observed (p. 1 2 8) that: I n short it appears that the major effect of letting risk depart from unity is to increase dispersion, a s measured by the generalized variance. T hat i mplies that the v a lue o f risk other than un i ty induces less powerful tests of information co ntent hypothesis than does risk equal to unity This is the major reason for our heavy reliance on results for risk equal to unity.
PAGE 23
16 4 disaggregating the total period into eight subperiods (30 observations each), the test results exhibited "a fair amount of variability" (p, 130). This variability was attributed, in part, to the increase in the estimated standard error of the mean due to the decrease in the number of observations. Although subperiod results "do not seem to be consistent with the null hypothesis" (p. 130), the value of the F statistic was below the .9 fractile for four subperiods. The above findings indicate that management forecasts of earnings convey information not already available to investors. These findings, however, should be tempered by the fact that the authors utilized a proxy for management forecasts. Als o, the data were restricted to forecasts made during the second half of the fiscal year. More importantly, however, the forecasts used were not previously unavailable either from the same source or from some other source (p. 109). That is, most of the forecasts were not first time forecasts and were available to investors, The Patell Study (1976) This study examined 336 voluntary management forecasts of annual EPS published in the Wall Street Journal for the period 19631968. Approximately 46 percent of the forecasts accompanied first and second quarter results. The sample included 204 point or convertible
PAGE 24
17 to point estimates and 132 of the form of greater (less) than X.X.X dollars (p. 250). the methodology utilized is essentially the same as that of Beaver (1968) and May (1971). The study focused upon the prediction error (abnormal performance), using the familiar market model, for the forecast week as well as eight weeks prior to and eight weeks after the forecast week. The null hypothesis that the expected value of the prediction error during the forecast week is equal to zero was rejected at a statistically significant level, .0004. The null hypothesis could "not be rejected at the .05 probability level for any other week in the test period" (p. 260). Disaggregating the data the following results were observed: 1 the mean prediction error for utility firms was much weaker than that for non-utility firms. 2 the mean prediction error for forecasts that accompanied quarterly earnings was weaker than it was for those that were announced separately (p. 266). 3 the mean prediction error for open ended forecasts was stronger than that for point estimate forecasts (p. 262). 4 the strongest mean prediction error was for forecasts released during the third quarter. The prediction error of the first and the second quarters exhibited a much weaker effect and the null hypothesis of expected prediction error
PAGE 25
18 equal to zero could not be rejected for the second quarter at the .1949 level (see Table 4, p. 266). Note also that first and second quarter forecasts account for 16 percent and 51 percent of the total sample respectively. The explanation provided by the author for these results was that (p. 266): The firm-specific price changes are less signifi cant for first-quarter disclosure than for the third-quarter disclosure, perhaps due in part to greater overall market volatility in the first quarter of December 31 fiscal ye ars. However, the second fiscal quarter results are even less significant, and this cannot be easily attributed to general market conditions. The above results are somewhat perplexing and difficult to explain given the methodology utilized in the study. Further examination of the attributes of forecasts released at different time periods (quarters) may provide an insight into the findings of this study. The study intentionally avoided dealing with the imputed accuracy attribute (p. 248), and the newness attribute was linked to naive forecasting models. Accordingly, whether or not the varying effects of each attribute provide a reasonable explanation of the results obtained in this study remains an unanswered question. Finally, the study examined the effect of deviations of forecasts from investors' expectations. Three historically based naive models were used as measures of investors' expectations. Deviations from each model were ranked into high and low deviation groups. The results were (p. 270):
PAGE 26
19 The firms in the top 25 percent (firm forecasts most exceeded naive expectations) experienced significantly positive Uit values during the week of forecast disclosure. However, the firms in the bottom 25 percent (firm forecast was furthest below naive expectations) also experienced (on average) positive Ui t; and for two of the three naive forecasts, the average Uit had a positive magnitude with a less than .02 probability of occurrence. Further examination of the forecast release week indicated that (p. 270): Thus, when attention is restricted to only the week of forecast disclosure itself, there appears to be very little to distinguish between forecasts on the basis of sign or magnitude of the change in investors expectations. Although the sign of the prediction errors was positive for both groups during the forecast week, examination of the eight week period preceding the disclosure of the forecast revealed that (p. 270) : Those firms whose management forecasts exceeded market estimates appear to enjoy generally positive price relative residual during the two months preceeding the forecast announcement, while firms whose forecasts fall short of market estimates e x perience generally negative residuals during the preceeding two months. However, the price-relative prediction errors at the immediate announcement date a re positive for both sets of firms. T hese findings were interpreted by Patell (p. 273) to be consist en t w i th the hypothesis that substitute sources of information ar e publicly available to investors. However, the positive price revision for the negative g roup during the forecast announc e ments week is difficult to explain. O ne possible explanation o ff ered by the author
PAGE 27
20 (p. 273) is that the act of voluntary disclosure of earning forecasts conveys information to investors. Whether or not the forecasting action by itself conveys information is a research question that requires further investigation. The Jaggi Study (1978) This study utilized daily residuals to examine the information content of management forecasts of earnings. The sample consisted of 144 forecasts published during the first fo ur calendar months of each year for the period 1971-1974. The null hypothesis of no information content was tested under the assumption of a symmetric stable distri bution. The daily residual for each stock was tested indivi dually for significance. This testing procedure covered the fo recast date as well as the ten days preceding and the ten days following the forecast announcement date. The number of signi ficant residuals for each day was then tested under the assumption of a binomial distribution. The author concluded that (p 964): The results show that statisticall y significant price adjustments were evident on the day of announcement of management forecasts and also before and after this day. Three annual naive forecasting models were used as a proxy for market expectations. Forecasts were classified as positive or negative depending upon the sign of deviation from the naive model. Adopting a buy and short-sale strategy resulted in an abnormal return for both groups.
PAGE 28
21 Deviations from each naive model were ranked in decending order and separated into two equal portfolios--top and bot tom deciles. The buy and short-sale strategy was repeated for stocks contained in both the top and bottom portfolios. "This was motivated by the need to determine whether stock market price behavior is directly related to differences in expectations and forecasts" (p. 966). Results do not indicate that the magnitude of deviation from naive models is related to the size of abnormal returns. That is, the adoption of the buy and short-sale strategy resulted in similar abnormal returns for both high and low deviations from market expectations as measured by naive models. Unfortunately, the study did not report the levels or the signs of deviations from naive models for top and bottom portfolios. Accordingly, interpretation of these results are somewhat difficult. The author (p. 966) carefully provides the following interpretation of the results: In view of these results, it can be inferred that voluntary disclosure of earnings forecasts may provide some additional information to investors. A caveat, however, is that even though price adjustments were observed, this does not neces sarily imply that the contents of the forecasts caused investors to revise their expectations. These adjustements may be caused by other factors, or even by the mere act of voluntary earnings forecast announcements. The Penman Study (1978) This study examined two research questions. The first question a ddressed the matter of full disclosure o f earnings forecasts by all
PAGE 29
22 firms in the economy. The second question examined the information content of management forecasts of earnings. The full disclosure is an adaptation of the incentive-sig nalling theory, mentioned earlier, to the capital market setting. The theoretical argument developed by Penman is essentially the same as that of Ross (1977). According to the incentive-signalling theory all firms in the economy, except the one with the worst news in the hierarchy, will be induced to disclose information relevant to the valuation of their firms. The worst news firm will be revealed by exception. Thus, full disclosure of inside information will result. Penman examined the claim that "managements only publish fore casts when they have 'good news' and hence investors only receive forecast information about firms which are doing relatively well" (p. 5). In order to determine whether a firm is doing 'relatively well' the author utilized a historically based martingale model with drift as an earnings expectation model. The deviation between the actual earnings for the year ilDIIlediately prior to the forecast year and the expectation model was calculated for 576 firms in the experimental sample. The earnings deviations for each firm were then standardized by the variance of prior years earnings. The standardized deviations were ranked in decending order and divided into two equal portfolios representing the top and bot tom halves. Likewise, newness of an earnings forecast was defined as the forecast de viation from the expectation model. Following the same procedure, each of the
PAGE 30
23 previous two portfolios were further divided into two portfolios based on the ranked value of the standardized forecast deviations. This classification scheme resulted in four portfolios of equal size. The full disclosure issue was tested by comparing returns on experimental and control portfolios. The total market index was used as the control portfolio. Following Gonedes (1974, 1975) and Gonedes et al. (1976), the author utilized the Hotelling T 2 test as follows (p. 40): By using equal weights, the signals are 'averaged out' so that it can be observed whether the infor mation sample differs from the market in terms of attributes other than the signals. Results were inconsistent with the full disclosure of earnings forecasts by all firms in the economy. The four experimental portfolios exhibited higher return than the market index used for c ontrol even for low forecast portfolios. Examination of the data revealed that ( p. 39): estimated return di f ferences are positive for all portfolios, reinforcing the belief that the information sample as a whole is systemat icall y different from the market. More speci f ic ally, it appears to consist of firms whose return performance is, on average superior to that of the m a rket as a whole. T his finding should not be interpreted to mean that full dis c losur e o f rele vant economic a ttribut e s by all firms does not occur. The appropriate int e rpretation, as pointed out by the author, is that earnings for e casts by man a g e ment is not th e te c hnology used by a ll f irms in the economy to si g nal their relevant economic attr i butes.
PAGE 31
24 Other signalling technologies are available to firms which do not issue earnings forecasts. The second question examined in this study is the extent to which management forecasts convey information beyond that which is already public knowledge. Results of the Hotelling T 2 test were inconsistent with the hypothesis of no information content. Portfolio by portfolio examination revealed that the information effect is attributed to the two high forecast deviation portfolios. The Hotelling T 2 test, however, is not a direct test of information content. Penman carefully points out that (p. 47): The results of the T 2 test suggest that earnings forecasts passes information, in the sense that foreknowledge of the forecast prior to the port folio holding period earns abnormal returns for a given level of risk, on average. This does not mean that the information is necessarily conveyed to the market by means of the forecast. The forecast is published, on average, about the middle of the third fiscal quarter, so it is possible that forecasts merely reflect information which has become available throu gh other signals for example, interim earnings announcements. Excess return tests were us ed to examine the information content of management earnings forec a sts. Daily abnormal returns for a sample consisting of 1188 management fo recasts were calculated for 121 days surrounding the forecast day. The sign o f the a bnorm al returns "indicate that the forecast announcements, on aver age convey information which is favorable to the firms that publ ic th em relative to the market's assessment prior to the announcement (p 51). The cumulative abnormal returns indicate that (p. 53):
PAGE 32
25 forecasting firms do, on average, enjoy 'good times' during the three months (approx imately) on either side of the forecast date, and not only on the day of the forecast announcement, A sample consisting of 725 management forecasts was disaggre gated into 'good' and 'bad' news forecasts depending upon the sign of deviation from the expectations model. Results were summarized as follows (p. 53): 'good' news firms apparently are continually revalued upwards throughout the period and receive a relatively sharp upward revaluation on or about the date of the forecast, This is in contrast to the 'bad' news group They are subject to a downward revaluation at the forecast date and, on average, exhibit 'bad times' prior to the announcement of the forecast. The above findings indicate that 'good' and 'bad' news earnings forecasts convey information to investors. It appears, however, that the contents of these forecasts are properly anticipated by investors. Stated differently, the anticipatory effect observed in this study indicates that substitute information as to the content of the management forecast of earnings is publicly available. The stock price behavior of the bad news forecasts indicates that "incentive exists for managements to publish forecasts which result in the downward revaluation of their firms" ( p. 54). This finding was interpreted to be consistent with the iterative process of the incentive-signalling theory which results in the disclosure of inside information "irrespective o f its implication for the value of the firm" (p. 54). Atiase (1978, p. 22) provides a somewhat similar interpretation of these results when he states:
PAGE 33
26 even firms disclosing seemingly bad news m a y be trying to revisions that their prospects. prevent drastic downward pri c e the market might be making about Finally, the author points out that the above results may b e influenced by other announcements accompanying earnings f o recas t s The tests were repeated after excluding forecasts which w ere accompanied by other announcements. This reduced the total sample to 384 'good' news forecasts and 98 'bad' news forecasts. Results differed little from the previous results for the unrestricted sample. The Nichols and Tsay Study (1979) This study examined the information content o f m ana ge ment fore casts of earnings announced during the first half of the f iscal y ear. The study covered a six year period beginnin g in 1968 and invol v ed a total sample of 83 forecasts. The selection criteria used fo r inclusion in the sample differed from prior studies. The criteria were set intentionally restrictive in an attempt to assure that the potential confounding influence of other types o f inf o rmat io n o n t h e results is less than in previous studies without such str ic t other news' screening criteria" (p. 143 ) the study were as follows (p. 1 4 3): The sel e ction crit e r ia used in The forecast a nnouncement must have been unaccom panied by other 'news' reports, and thu s any firm that had reported in the WALL STRE E T JOURNAL any of the following was omitted from the s t udy: (1) annual or quarterly earnings r e p o rt s or other 'news' within a week of the d a t e th e forecast was published, (2) a stock split within e i ght weeks on either side of the date of f orec a st or (3) a cash dividend within the week of forecast.
PAGE 34
27 Two approaches were utilized to test the information content of management forecasts. The first was the abnormal return method. For each stock i, the absolute value of the residual for each time period (eit) was calculated. The authors concluded that the average (et) for the forecast week was larger than it was for the other sixteen weeks in the test period. The Kolmogorov-Smirnov one sample test indicated significance at the 10 percent level. "Thus, while there is some evidence of unusually large relative price changes during the announcement week, the observed ranking was significant at the 10 percent level" (p. 149). The second approach utilized was an analysis of frequencies of abnormal returns during the test period. The motive for the frequency test is the belief that (p. 148): very large abnormal returns of a few firms may dominate averages of returns for a group of firms. Although, again, this may be more of a potential problem for small samples than for large samples, the possibility exists in either In any event, analyses of averages may not reveal whether observed abnormal return movements are pervasive throughout the sample or whether they have been produced primarily by a subset of the sample. Analyses of frequencies will provide a n indication of the pervasiveness of any observed information content. The specific test used and the results are summarized as follows ( p. 148):
PAGE 35
28 The weekly (eit)'s were ranked one through seven teen for each firm. If week-zero ( eit) was the largest for the seventeen-week period, a rank of one was assigned; if it was the second largest, a rank of two, and so on with the smallest receiving a rank of seventeen. If very large relative abnormal return movements occurred in period zero as a result of reported forecasts, we might expect an unusually large number of rank one's for period zero, relating to the other week. The latter possibility was easily checked and dismissed. The apparent conflict between the results of the two tests should be cautiously interpreted. The selection criteria allowed other news to prevail during the sixteen weeks surrounding the forecast week. Announcements of annual and quarterly earnings, cash dividend, and other news were allowed to prevail except for the fore cast announcement week. Accordingly, the frequency test examined the relative effect of management earnings forecasts on unexpected earnings as compared to other news. Six historically based naive models were used as a proxy for current market expectations of earnings. Three of these models were based on annual earnings, while the other three incorporated quarterly earnings. Similar to Foster (1973), Patell (1976), and Penman (1978), forecasts were classified as 'positive' or 'negative' news depending on the sign of deviation from the naive model. Adopting a purchas e policy for the positive group and short-sale, for the ne g ative grou p eight weeks prior to the forecast anno uncement s week resulted in abnormal returns for both groups. "This would be consistent with the statement that the executive forecasts possessed in f ormation co ntent" (p. 152).
PAGE 36
29 A seventh group was constructed for comparison with results reported by Patell (1976). An earnings forecast deviation was defined as the difference between the forecast and the actual earning for the previous year. Deviations were scaled by previous year's earning and ranked in decending order. The top SO percent were assumed to have positive news, and forecasts in the bottom SO percent were assumed to have negative news. However, the sign of forecast deviation was positive for both groups. The study reported a graphic presentation of the cumulative average residuals without any statistical test. The authors inferred that (p. 152): The week-zero price movement is the largest weekly movement for the purchase group, while for the short-sale group week-zero has one of the smaller movements. In contrast, Patell found relatively large positive week-zero movements for both of the groups. Frequencies of gain and loss on both the purchase and short-sale portfolios were examined jointly. The gain (loss) is measured by the cumulative residual for the nine week period, up to and including the forecast week. The chi-square statistic was not significant at the 5 percent level. The same analysis was repeated for the other six naive forecasting models. The results were not significant for any of the six models. These findings led to the conclusion that cumulative average residuals are influenced by a small number of large residuals.
PAGE 37
30 The above conclusion, however, should b e t e mper e d b y the methodological difficulties encountered in this st u d y T h ree mai n deficiencies are worth noting. First, the gain ( loss) o n portfolios was measured as a holding period return beginning eig h t w eeks prior to the forecast week, To the extent that other news were al lowed to prevail during that period, it would be inappropriate to a ttribute the gain (loss) to the information content of management earnings forecasts, This is especially true given that the sample size of 8 3 stocks divided into two groups was not sufficiently large to randomize the effect of other news. Second, the reliance on naive forecasting models can result in misclassification of m anagement forecasts, thus adding another confounding effect to results of the frequency test, Finally, there is no a priori reason to suspect that all management forecasts convey information beyond that which is already available. As demonstrated in the next chapter, the majority of management earnings forecasts do not differ from publicly a vailable anal y sts' forecasts of earnings. A ccord i ngly, the cla s s ifica tion o f for e cast s as either good or bad news is questionable, particu l arl y w hen t h e t e st involved is based on frequenci e s. Summary and Limitations of Existing Research Curre n t research studies on the informati o n cont en t of manage ment forecasts support the contention that the se fo r ecasts c onvey information in excess of what is already avai l able to market
PAGE 38
31 participants. One noted exception to this contention is the study by Nichols and Tsay (1979). The behavior of stock prices prior to the announcement date indicate an anticipation of the news released by these forecasts. This is interpreted to mean that substitute information to management earnings forecasts seems to exist and is reflected in stock prices prior to the announcement of earnings forecasts by management. Conflicting results, however, have been reported with respect to the information content of 'good' versus 'bad' news forecasts. Findings of the Gonedes et al. study (1976) indicate that only 'bad' news forecasts convey information, while results reported independently by Patell (1976), Penman (1978), and Jaggi (1978) indicate that both 'good' and 'bad' news forecasts convey information not already available to investors. A second area where conflicting results are observed i s the bad news earnings forecasts category. Patell reported an upward price revision for the 'bad' news forecasts during the forecast week. This is in contr a st to Penman's study which reported a downward price revision for this group on the day of the announcement as well as the surrounding days. The conflicts o f research findings are not easily e x plainable and may be a ttribut e d to several factor. One factor point e d out by the empirical find i ngs of prior research is the unreliability of histor i call y -b as ed n aiv e fo reca s ting models a s a proxy of market
PAGE 39
32 expectations. Patell observed that "on average, the voluntarily disclosed forecasts of earnings per share were greater than the naive estimates of the market's expectations" (p. 268). Penman also observed that "the earnings of forecasting firms are, on average, higher than expectations and not representative of the market as a whole" (p. 36). Thus, low earnings forecast portfolios ( bad news forecasts) are low relative to the sample studied but not low relative to the measure of expectations. More importantly, however, the empirical evidence provided by Gonedes et al., Pa tell, and Penman suggest that substitute information for management forecasts is publicly available. Naive forecasting models totally ignore information available from sources other than past earnings and as such can misclassify forecasts. This reasoning is supported by Atiase ( 1978, p. 23) as follows: Thus, it is easy to see the possible misclassifi cation that can result from relying on expec ta t ions models irrespective of predisclosure infor mation production and dissemination on or by a firm. Furthermore, all the relevant studies to date have classified forecasts as either 'good' or 'bad' news forecasts and have ignored a possible third classification for forecasts that do not deviate from, or are approximately equal to, investors' expectations. The empirical finding that substitute information for management forecasts is pub licly available provides support for the assertion that th e majority of management earnings forecasts do not deviate from investors
PAGE 40
33 expectations. This, of course, is an empirical question which has not been examined. The validity of this assertion, however, adds further evidence to the possibility of forecast misclassification by previous studies. Several research questions directly related to the information content of management forecasts remain unexamined. Three of these questions are of interest to this study. The first deals with the information content of the voluntary forecasting action by itself. That is, whether or not the voluntary forecasting action by itself induces an upward stock price revision. On a theoretical basis it can be argued that a management forecast which merely confirms investors expectations of earnings reduces the variability associated with these expectations causing a reduction of the risk attributed to the firm's stock. This in turn causes a higher equilibrium value for the stock. It should be noted, however, that this argument rests on the assumption that inside information is perceived by investors to be more reliable than information provided by outsiders to the firm, such as fi nancial analysts 1 Pa tell offered the possible ef feet of the forecasting action as an explanation for the upward stock price revision during the forecast week for the 'bad' news group 2 However, the possible misclassification of forecasts and the conflicting results of other studies highlight the need for a specific testing of this proposition. The second question is related to the effect of the perceived accuracy of forecasts on their level of information content. Patell
PAGE 41
34 points out that the imputed accuracy of an earnings f orecas t is an attribute necessary to altering investors' expectations. Testing t h i s proposition, however, requires knowledge of investors' e x pect a t io ns immediately prior to the earnings forecast announcement b y management and the revision of these expectations as a result of receiving the earnings forecast. The reliance of prior research on naive f o recasting models precluded empirical examination of the imput e d accurac y proposition. The third question relates to t h e possible effect o f earni n gs forecasts on non-forecasting firms in t h e same industr y The mechanism of the incentive-signalling the o r y alluded to e a r l ier and the argument that a management forecast of earnings is a q uantitive assessment of several economic variables, some which are c ommon to the entire industry in which the forecasting firm operates, lea d t o the assertion that the information content of an earnings for ec a st ma y extend to other firms in the same ind us try. T his a ssert io n ha s n ot been considered in the available empirical research o n t h e informa t ion content of management earnings forecasts, To summarize, existing research findings support t he contention that management forecasts of earnings convey informat i on to investors. However, the anticipatory behavior of stock prices pri o r to the fore cast announcement date indicates that substitut e information to management forecasts do exist. The conflicting results reported in the current literature may be attributed to se veral factors. The
PAGE 42
35 reliance on naive forecasting models is one factor responsible for misclassification of forecasts. The need for a more appropriate measure of market expectations and hence proper classification of forecasts is self evident. Moreover, the scope of available research on the information content of management earnings forecasts has been somewhat limited and several research questions remain unexamined. The effect of the forecasting action by itself, the effect of the imputed accuracy ascribed to forecasts, and the extent to which management forecasts of earnings convey industry-wide information are only three possible extensions to existing research. Examination of these three extensions comprises the goal of this dissertation. Notes 1. This argument follows directly from the properties of the capital asset pricing model under the assumption that a reduction in the variance associated with expected returns reduces the systematic risk of the firm. 2. Jaggi (1978, p. 966) also points out the possible effect of the forecasting action.
PAGE 43
CHAPTER III DATA, METHODOLOGY, AND STATISTICAL TESTS This study explicitly considers the effects of newness and imputed accuracy of management earnings forecasts on altering investors' beliefs with regard to equilibrium stock prices. O perational measures of newness and imputed accuracy of management earnings forecasts require a 'proxy' of investors' expectations before and after the announcement of these forecasts. Publicly available forecasts of earnings by financial analysts seem to be an appropriate measure of investors' expectations for four reasons. First, financial analysts incorporate several factors in arriving at an estimate of future earnings of a particular firm. The past earnings history of the firm is only one such factor. Other external factors which have impact on the potential profitability of the firm are also incorporated. Second, financial analysts communicate with managements through interviews and private discussions and incorporate i n s i de information in their forecasts of earnings. Gonedes et al. point out that (p. 108): Indeed communication between management and ana lysts about forecasts seem to be one of the major motivations for the SEC's current perspective on 36
PAGE 44
37 forecast disclosure The use of such private communication is also emphasized in a recent study of earnings forecasts conducted by the Financial Analysts Federation. Third, the reliance of investors upon earnings forecasts published by analysts as an input to investment decisions is widely supported. Abdel-Khalik and Ajinkya (1979) point out that (p. 1): the production of and demand for analysts fore casts clearly testifies to the importance of these forecasts to the investment community. Good (1975) explains the importance of the role of analysts to the professional investors as follows (p. 42): What ever the broad strategy used by the portfolio manager, he is ordinarily dependent in varying degree on analysts' recommendations concerning individual stocks. He cannot accept analysts' recommendations uncritically, but neither can he investigate each security as thoroughly as the analyst who specializes in the stock. Fourth, the available empirical evidence suggests that earnings fore casts by analysts influence market prices of securities. Gonedes et al. (1976) examined the information content of analysts' forecasts of earnings and concluded that these forecasts convey information relevant to establishing equilibrium prices of securities. A more recent study by Abdel-Khalik and Ajinkya (1980) examined the informa tion content of analysts' revisions of earnings forecasts. The null hypothesis of no information content was rejected at a statistically significant level for the revision week. The hypothesis was rejected for both positive and negative revisions.
PAGE 45
38 To summarize, earnings forecasts published by financial ana lysts incorporate publicly available information as well as inside in formation privately communicated by managements. Moreover, empirical research findings are consistent with the assertion that analysts' forecasts are relied upon by market participants in making investment decisions. Accordingly, analysts' forecasts of earnings represent a reliable proxy of investors' expectations. This proxy is utilized in this study to determine the newness and imputed accuracy of management forecasts as defined in the following section of this chapter. Data The experimental sample of earnings forecasts was obtained by searching the Wall Street Journal Index for eight years covering the period 1970-1977. Only forecasts attributed to top management, usually the president or the chairman of the board, were admitted into the experimental sample. The following criteria were employed in the sample selection process. 1 The forecast must be a point estimate or convertible to a point estimate. The latter group are earnings forecasts expressed either as a specific increase (decrease) over the actual earnings of the prior years or as a range. Earnings forecasts expressed as a range are bounded by a minimum and a maximum value. These forecasts were admitted only if the
PAGE 46
-39maximum value of the range did not exceed the minimum value by more than ten percent. The mean value of the range was considered to be the forecasted earnings. Open ended forecasts in the form of greater (less) than a specific amount were not included in the sample. 2 The forecast must not be accompanied by or fall within four weeks prior to or four weeks after the date of any of the following announcements: a annual or quarterly earnings b dividends c stock splits d capital issues. 3 Daily stock returns must be available on the daily CRSP tape for at least one and one half years prior to the forecast date and one year after the forecast date. 4 The forecast must be the first earnings forecast issued by management during the fiscal year to which the forecast refers. Second forecasts or forecast revisions b y management were excluded from the sample. 5 At least one analyst's forecast expressed in point form must be available from the Standard and Poors Earnings Forecaster within three weeks prior to the date of the management forecast announcement. Another anal y st's forecast must be available no later than three weeks after the date of the management forecast announcement. If two or more analysts' forecasts were available, the one closest to the date of the
PAGE 47
40 management forecast was selected. On very few occasions, two or more analysts' forecasts appeared on the same date, or very close together, in which case the mean value was determined as the proxy for the market expectations. The first criterion ensures that management forecasts of earnings are quantifiable. The majority of forecasts predicted an increase over the actual earnings of the prior year. Very few firms predicted a decrease in earnings from the prior year. The second criterion was imposed in an attempt to reduce the ef feet of other announcements. It should be noted, however, that not all announcements are published in the Wall Street Journal Index. The effect of the undetected announcements should not influence the results of this study. The large sample size utilized would be expected to randomize the effect of these announcements over the test period. Penman (1978) noted that other news announcements did not influence the results of his study. The third criterion ensures that sufficient data are available for statistical tests. The fourth criterion excluded earnings forecast revisions by management in order to avoid the possible confounding effects of mixing the two types of forecasts. The fifth criterion is necessary for the operational measures of imputed accuracy and newness of management earnings forecasts. Ana lysts forecasts prevailing within three weeks prior to the announce ment of an earnings forecast by management are considered a measure of investors' expectations at that time. Similarly, analysts' forecasts prevailing within three weeks after the announcement date are
PAGE 48
41 considered a measure of investors' expectations conditional upon the information revealed by management through the earnings forecast. The selection criteria employed in this study are necessary to reduce the confounding effects of other news and to focus on the variables under examination. Nonetheless, as in the case for most empirical studies, these selection criteria induce a bias towards surviving firms. Criterion five induces another bias towards firms in which analysts have exhibited interest The final experimental sample that satisfied the selection criteria is composed of 401 earnings forecasts. utility firms, no particular industry is over Except for public represented. This observation is consistent with findings of prior studies. The sample is not evenly distributed throughout the fiscal year and forecasts tend to cluster in the fourth quarter. A monthly break-down of forecasts is included in Table 3-1 below. Table 3-1 M onthly Management Earnings Forecasts M ont h in fi scal y ear j l I 2 I 3 I 4 I S 6 I 7 I 8 \ I 1 0 I 11 112 N umber of F oreca st s I S I 12 I 30 I 4 0 14 3 I 27 I 1 8 I 161 4 1 I S8 I Sl /6 0 Per c ent o f T ot al
PAGE 49
42 Definitions of Partitioning Variables Newness and imputed accuracy of management forecasts of earnings are operationally defined as follows: N e wness of a forecast= MF AP Imputed accuracy where MF = management AP = analyst's AP = MF-AA MF forecast of earnings, forecast of earnings prevailing (1) (2) within three weeks prior to the announcement of the management forecast, and AA = analyst's forecast of earnings prevailing within three weeks after the announcement of the manageme n t forecast. Experimental Portfolios For the total sample, the mean and the standard de v i ation fo r n e wness of forecasts are 0.0085 and 0 065 respectively. Th e mean a n d the standard deviation for the absolute value of newness ar e 0.042 an d 0.05 respectively. This clearly indicates that the majority of
PAGE 50
43 management forecasts do not convey new information to investors. Three portfolios were formed representing positive, negative, and neutral news forecasts, depending upon the sign and the magnitude of the level of newness. The positive news portfolio consists of ll5 earnings forecasts with newness level > 3 percent, an arbitrary choice. Similarly, the negative news portfolio consists of 85 manage ment forecasts with newness level :5_ 3 percent. The neutral portfolio consists of the remaining 201 management earnings forecasts. The mean and the standard deviation for the absolute value of imputed accuracy are 0.0198 and 0.045 respectively. Observations in each of the three portfolios were ranked on the level of imputed accuracy. If analysts reise their forecasts to equal the forecasted earnings by management, the level of imputed accuracy will be high and the results of Equation (2) will be zero. The level of imputed accuracy of a positive news earnings forecast is considered to be high if the result of Equation (2) was equal to or less than 3 percent, and low if greater than 3 percent. The reverse classification is true of the negative news forecasts. This classification scheme yielded six portfolios. of forecasts in each portfolio. Table 3.2 presents the number
PAGE 51
44 Table 3.2 Imputed Accuracy and Newness of Management Earnings Forecasts Imputed Accuracy POSITIVE NEGATIVE NEUTRAL High 65 47 1 93 Low 50 38 8 Total 115 85 201 The neutral news low-level imputed accuracy port fo lio contai n s only eight forecasts. This result is not surprising, due to the fact that this group of management forecasts represents a mere confirmation of existing beliefs and, as such, no revision in analysts' forecasts should be expected. The revision of the eight firms' forecasts may be due to other factors that occurred within three weeks after the announcement of the management earnings forecasts. Thi s portfolio was deleted from the study due to the small size of the sample. Control Portfolios For each experimental portfolio two contr ol portfolios were formed. The first control portfolio was selected by individually matching each firm in the experimental portfolio with another firm,
PAGE 52
45 for which no management forecast was known, from the same industry classification. The matching was conducted using the first three digits of the industry classification code. The selection of control firms was random in the sense that any firm with the same industry classification code had an equal chance of being chosen. Due to data requirements, however, only firms with daily stock returns available on the CRSP tape for the period 1970-1978 were admitted into the control portfolio. If a firm did not satisfy the data availability requirement another firm from the same industry classification code was chosen at random. This process continued until all firms in the experimental portfolios were individually matched. These control portfolios are referred portfolios. to hereafter as the matched control The second set of control portfolios were randomly selected firms without industry matching. To simplify the search process, however, the matched control firms were randomly assigned to the experimental firms. random control. These control portfolios are referred to as The matched control portfolios are designed to detect industry effects and to assess the effect of management forecasts of earnings on non-forecasting firms in the same, or similar, industry. the random control portfolios are used to provide a bench-mark comparison and to help detect industry effects.
PAGE 53
46 Statistical Tests Two statistical testing procedures are used in this study. The first is residual analysis based on the familiar market model. The second test involves a comparison of total returns o f equal risk port folios. A two way analysis of variance is utilized for this test. The use of two different testing procedures will indicate whether the results obtained are sensitive to the statistical test u sed. Further more, two approaches to residual analysis are used to provide a comparison with results reported by prior studies. Residual Analysis Information content can be attributed to man a ge m ent f o recasts of earnings if investors' beliefs with reg a rd to e quilibrium prices are altered as a result of the information co n veyed by these f o recasts. Change in stock prices (and returns) during the forecast announcements period is the empirical evide n ce of information content. The returns observed on a security, however, cannot be total ly ascribed to a specific event, due to the fact that the return is also influenced by economy-wide events affecting all securities. The market model developed by Sharpe (1964) and Lintner ( 1 965) is common ly used to filter economy-wide factors. The market model is an ex-ante relationship between the return on a specific security and the return on the market portfolio. The model, however, is used on an ex-post basis utilizing actual realizations of security returns to determine
PAGE 54
47 abnormal performance. In this study, the market model is employed to measure weekly abnormal returns using two related approaches. The first approach measures standardized prediction errors and the second examines the distribution of unstandardized errors during the forecast period. While the latter approach is commonly used, the former is theoretically more appropriate. follows: The two tests are explained as A Standardized Prediction Errors This test was originally used by Beaver (1968) and expanded by Patell (1976) using the market model which is given by: Rit = ai+biRmt+eit where ( 1) i = a firm index, i=l, N. t = a week index for the non-forecast period, t=l, .. T, Rit= the natural logarithm of the return on stock i for week ~t= the natural logarithm of the return on the market index ai,bi = the intercept and slope of the linear relationship between Rit and ~t eit = stochastic disturbance term for security i in week t. t. Time series regression is performed on realizations of Rit and ~t to obtain estimates of ai and bi. These estimates are then used
PAGE 55
48 to calculate the prediction error (abnormal return) for each stock i and for 16 weeks surrounding 1 the forecast announcements week (hereafter called the test period). The prediction error is given by: The prediction error in the test period is assumed to be normally distributed with the following properties: 0, s#t Ct a .2 l. l. t = -8. .+8 s=t a i 2 is the variance of the regression during the estimation period. The unbiased estimator of a i 2 is si 2 which equ a ls: = a. 2 l. Cit is the increase in the variance due to pred i ction outside the estimation period. Cit= 1 + 1 + (R.nt &..n) 2 T t=l. T
PAGE 56
49 The standardized prediction error during the test period is: t(T-2) (2) The cumulative standardized prediction error for the 17 weeks included in the test period is given by: L z: WiL = t=l where L < 17 t(T-2) (3) Each eit is distributed independently of si 2 and has a variance equal to: Ti-2 T -4 1. where Ti is the number of observations in the non-forecast period for firm i. The prediction error is assumed to be an independent random variable. According to the Central Limit Theorem, the cross-sectional mean for any week t in the test period approaches a unit normal distribution. The normalized sum can be expressed as: N I vit Zvt = i=l ( 4) i ~ l ( T 2 ) 1. T 4 1.
PAGE 57
50 Similarly, the cross-sectional cumulative prediction error for w ee k t is N t W-1 l. 2 wL = i=l (5) N ( T :t l. t i=l T i Equations (4) and (5) can be used to test the hypothesis that the cross-sectional standardized prediction error is significantly different from the expected mean of zero. One or two sided tests can be performed, depending on whether an expected direction of eit can be specified ex-ante. A statistical test that abstracts from direction can be constructed directly from equation (2). Squaring Equation (2) yields: and E T 4 l. (6) (7) Equ a tion ( 7 ) can be standardized to yield an expected value of 1 t he standardiz e d statistic is:
PAGE 58
2(Ti 3) T 6 l. 51 T 4 l. T 2 i Again, by applying the Central Limit Theorem, a unit normal crosssectional statistic can be constructed as follows: N E (Uit-1) 2 ut = i=l (8) i!l ~ 2(Ti-3) ) (Ti-6) Statistical tests developed in this section involve testing the mean of the distribution of prediction errors under the normality assumption. N ichols and Tsay (1979) questioned the validity of these tests and pointed out that the mean may be influenced by a small number of large prediction errors. To examine this claim, tests are repeated after trimming the distribution by 10 percent on each side. This test will indicate whether the observed effect is a general pattern or merely due to a few extreme observations. Another test similar to that used by Penman (1978) can be developed as follows:
PAGE 59
V it ~ t (T-2) Vt= 1 E Vit N 52 by appeal to the Central Limit Theorem ( V / I N) I a is t a p p ro xi ma t ely standard normal distribution for large N A Z test ca n be c o nduct e d for each week in the test period. This te st pro v id es a co m p a riso n with results reported by Penman ( 1 978) and indica t es t h e s e ns iti v ity of results to the statistical test used. 2 B Unstandardized Prediction Errors As stated earlier, this test is commo nly used in acco u n ting research. Foster (1973), Jaggi ( 1 97 8) and N ichols and T sa y (1 9 7 9 ) utilized unstandardized predi c tion errors (eit). Th i s test also assumes cross-sectional independence and nor m al distribu ti on of the prediction error with a mean equal to zero. A normal d i str ibution test is used to examine the hypothesis that the mean value e t is equal to zero. Results of this test will be reported for comp arison with prior studies. Analysis of Variance An analysis of variance test is used in thi s s t udy to ex am i ne the information content of management f orec a s t s of earni n gs using
PAGE 60
53 a c tual r eturns observed on securities during the forecast announcement w e ek a nd the ten surrounding weeks. This test is designed to s u pplement results of prediction errors tests and to provide a comparison between results obtained using a different variable for each test. In addition, the use of actual stock returns reduces the reliance on the market model. A two-factor analysis of variance is performed on the stock returns of each experimental portfolio and its designated control portfolios. The first factor is the type of portfolio and contains three levels (i.e., experimental, matched control and random control). Distinctions among the three portfolios are based on the information set affecting each portfolio. Publicly available information influencing stock returns is composed of three general components: firm specific, industry specific, and economy-wide information. The firm specific information distinguishes the experimental portfolio from the two control portfolios. The industry specific information is common to both the experimental and the matched control portfolios but not to the random control portfolio. This provides a distinction between the matched control and the random control portfolios. In short, the three portfolios are influenced by different sets of information and as such are viewed as three levels of the type of portfolio factor. The second factor is the new information effect and contains eleven levels, t = -5 to t = +5. The forecast announcement week,
PAGE 61
54 t = 0, is only one such level. Prior research indicates that the information contained in earnings forecasts is partially anticipated and influences stock returns prior to the forecast announcement date. This implies a general flow of information produced and disseminated by insiders and/ or outsiders to the firm. Thus the level of new information made publicly available is distinguishable on a weekly basis. It is important to note that stock returns observed every week are influenced by the new information made public during that week and not by the information revealed during prior weeks. This interpretation follows directly from the empirical evidence of stock markets efficiency in the semi-strong form, which asserts that stock prices (returns) adjust instantaneously to all publicly available information. Without the semi-strong form efficiency, the analysis of variance tests would be questionable due to lack of independence among weekly information effects. Portfolio returns are not, strictly speaking, comparable because of varying levels of systematic risk. To overcome this difficulty, the systematic risk for all portfolios is equated to unity. The procedure used is developed by Gonedes (1974) and involves the following steps for each portfolio: 1 The systematic risk (beta) for each stock is calculated and then ranked in decending order. 2 Two portfolios are constructed representing the top and bottom halves of the ranked stocks. The mean beta for each of the two portfolios is then calculated.
PAGE 62
where 55 3 To determine the weights to be assigned to stock returns of the top and the bottom portfolios, the following equation is solved: B 1 = the mean beta for the top portfolio B2 = the mean beta for the bottom portfolio X = the weight to be assigned to stock returns of the top portfolio (1-X) = the weight to be assigned to stock returns of the bottom portfolio For all portfolios, the weights were between 40 and 60 percent. This result is consistent with results reported by Gonedes (1975). After restating all stock returns, the following two factor linear model is used: Rjik = u +pi+ tk + Yik + ejik where u = the grand mean Pi= Portfolio effect, i=l, 2, 3 tk = the information effect, k=l, 11 Yik = the interaction effect ejik = random error term. The analysis of variance test examines the overall equality of means. If the overall hypothesis is rejected, further investigation will be required to determine which of the thirty-three means, or
PAGE 63
56 combination of means, are causing the rejection of the overall hypothesis. of all means. The Duncan's multiple range test is used for comparison Hypotheses A Testing the Effect of Forecasting Action The neutral news forecasts convey a very low level of new information to investors and as such should not be associated with abnormal returns unless the forecasting action by itself does in fact alter investors expectations. Thus, testing the effect of the forecasting action involves an examination of the behavior of prediction errors associated with the "neutral" news earnings forecasts. It has been argued that a neutral forecast may reduce the variance of expected future returns and as such cause the price of the stock to move to a higher equilibrium price. The validity of this argument hypothesizes a positive mean prediction error for this group. The following three tests are conducted on prediction errors: Ho1: The mean value of vt during the forecast disclosure week is equal to zero Ha 1 : The mean value of vt during the forecast disclosure week is greater than zero Hoz: The trimmed mean value of vt during the forecast disclosure week is equal to zero
PAGE 64
57 Ha 2 : The trimmed mean value of vt during the forecast disclosure week is greater than zero Rejecting both Ho1 and Ho2 Will indicate that the forecasting action by itself conveys information favorable to stock prices and that the effect is a general pattern of the sample, and vice versa. If H01 is rejected but H02 is not rejected, this would mean that the test is influenced by a few outliers. Failing to reject both Ho1 and Ho 2 would indicate that the forecasting action conveys no information to the market. Results will be reported for the seventeen weeks in the test period and for the experimental as well as the control groups. Two other tests are conducted on the distribution of the residuals and prediction errors respectively, distribution tests), as follows: (hereafter called H03: The mean value of et during the forecasts disclosure week is equal to zero Ha3: The mean value of et during the forecasts disclosure week is greater than zero H04: The mean value of vt during the forecasts disclosure week is equal to zero Ha4: The mean value of vt during the forecasts disclosure week is greater than zero. The analysis of variance test is conducted on portfolio returns to test the following four hypotheses:
PAGE 65
58 Ho5: The means of the three portfolios are equal Ha5: The means of the three portfolios are not equal Ho6: The means of the eleven information treatments are equal Ha6: The means of the eleven information treatments are not equal H07: There is an interaction effect between the portfolio effect and the information treatment effect Ha7: There is no interaction effect If the overall hypothesis of equality of means is rejected, the following Duncan's Multiple Range Test will be examined. Ho3: All thirty-three means are equal Ha3: All thirty-three means are not equal. B Testing the Effect of Newness of Earnings Forecasts Two experimental portfolios are classified as high-level newness portfolios: the positive and the negative news portfolios. For each of the two portfolios the direction of abnormal returns can be specified ex-ante and as such the alternative hypotheses should be stated with the proper direction. That is, the alternative hypotheses for abnormal returns tests should be greater (less) than zero for the positive (negative) portfolio. The same eight hypotheses already stated in conjunction with the "neutral" news portfol io are utilized to test the effect of newness.
PAGE 66
59 C Testing the Effect of Imputed Accuracy As stated earlier, each of the two high-level newness portfolios is subdivided into two unequal portfolios based on the degree of imputed accuracy ascribed to the management forecasts of earnings. Testing the effect of imputed accuracy is accomplished by separate examination of results for the high and low imputed accuracy portfolios. The same eight hypotheses are tested for each group. Further, abnormal returns for the high-level and low-level imputed accuracy portfolios for each week in the test period can be compared under the assumption of cross-sectional independence. Three statistical tests can be developed for comparison between abnormal returns of the two types of imputed accuracy portfolios. The first test follows directly from Equation (4). The difference between standardized prediction errors for unequal samples is given by: Nl Nz l E l E vjt N1 i=l vit Nz j=l ZdVt = (9) ( l Nl T -z l Nz T.-2)' N2 1 N22 J '2 l E + E i=l T,-4 j=l T.-4 1 J where N1, and Nz are the number of firm forecasts designated as hihg-level and low-level imputed accuracy respectively. Equation (9) can be used to test the following two null hypotheses for the total and the trimmed standardized prediction errors:
PAGE 67
60 Hog: Zdvt during the forecast announcements week is equal to zero Ha 9 : Zdvt is greater (less) than zero for the positive (negativ e ) forecasts. Ho10 : Zdvt for the trimmed standardi z ed prediction errors during the forecast announcements week is equal to zero Ha 1 o: Zdvt for the trimmed standardized prediction errors during the forecast announcements week is greater ( less) than z e ro for the positive (negative) forecasts. The second test is designed to examine the observed distribution of weekly standardized prediction errors. the difference between the two means is given by: l N1 N2 N1 [ vit 1 [ vjt i=l N2 j=l Zdvt = ( s it r 2 + s jt N1 N2 The Z statistic for (10) where Si 2 and s .2 J are the weekly variances of the observed distribution of standardized prediction errors associated wit h the high level an d the low level imputed accuracy portfolios respectively. The third test examine s the distribution of unstandardiz ed prediction errors (et) and i s given b y:
PAGE 68
where Zdet K.2 l. = ( and 1 N1 N1 E i=l 2 Kit K .2 J eit are 61 1 N2 Nz E ejt j=l (11) 2 K jt + the weekly variances of the observed distribution of unstandardized errors associated with the high-level and low-level imputed accuracy portfolios respectivel y The following two null hypotheses are examined: Holl: ZdV t during the forecast announcements week is equal t o zero Hall: ZdV t during the forecast announcements w ee k is greater (less) than zero for the positive (negative) f o recasts. Ho12: Zdet during the forecast announcements week is equal to zero Ha12: Zdet during the forecast announcements week is g reater (less) than zero for the positive (neg a tive) forecasts. D Testing the Effect on Other Firms in the Same Industry This hypothesis is examined by compar i ng the behavi o r of abnormal returns and total stock returns o f the matched control portfolio to the other two portfolios--the experimental a nd the random control.
PAGE 69
62 As already stated, the random control portfolio is used to provide a bench-mark comparison and to help detect economy-wide effects. The comparison between the matched control portfolio and the experimental portfolio will indicate the extent to which management forecasts convey industry-wide information. If management forecasts of earni n gs do in fact convey substantial industry-wide information, the abnormal returns, and the total returns, of the matched portfolio should be influenced in the same direction as the experimental portfolio during, but not prior to, the forecast announcements week. On the other h and, if significant abnormal returns are associated with the matched control portfolio, but not the experimental portfolio, prior to the forecast announcements week, and both portfolios exhibit significant abnormal returns during the forecast announcements week, the inference of industry-wide effect would not be justifiable. A more appropriate inference would be that industry factors induce managements to issue earnings forecasts in an attempt to revalue their firms relative to other, non-forecasting, firms in the same industry. That is, the observation of abnormal returns associated with the matched control portfolio during the forecast announcements week is a necessary but not a sufficient condition for the inference of industry-wide effects of management forecasts of earnings. Examination of the behavior of abnormal returns associated with each of the two portfolios prior to the forecast announcements week will indicate whether management forecasts of earnings convey industry-wide information or whether industry factors induce managements to issue earnings f orecasts.
PAGE 70
63 The first eight hypotheses were tested for the three portfolios to determine the effect of management forecasts on each portfolio and to provide the comparison among the three portfolios necessary for detecting the industry-wide effects. Notes 1. The regression was performed on actual weekly realizations of Ritand Rrotfor two hundred weeks prior to the test period. 2. Prediction errors tests assume stability of regression coeffi cients, a and b, during test period. Systematic violations of this assumption biases the observed prediction errors during the test period. In this study, such bias is not expected to be systematic in nature for four reasons. First, the experimental sample was drawn from observations which occurred at different time periods, (8 years X 52 weeks), and as such any shift in the regression parameters would not be expected to take a systematic form. Second, the test period is relatively short (17 weeks) to induce a serious shift in regression parameters. Thi rd, the use of two time matched control groups will highlight whether or not any systematic bias in prediction errors did in fact take place. Fourth, the large sample size utilized in this study would be expected to randomize the effect of any shift in regression parameters.
PAGE 71
CHAPTER IV RESULTS Introduction The first three research questions addressed in this study examined the effect of newness and imputed accuracy of management forecasts of earnings on equilibrium stock prices of forecasting firms. The fourth research question examined the effect of management earnings forecasts on equilibrium stock prices of non-forecasting firms in the same, or similar, in dustry. On an a priori basis it can be argued that the effect of management earnings forecasts on non forecasting firms is positivel y related to the l evel of newness as well as to the level of imputed accuracy ascribed to these forecasts. Accordingly, the effect on non-forecasting firms was examined in conjunction with the first three research questions. T his p rocedure of joint examination of hypotheses provides a more comprehensive analysis of results and avoids repet i ti on in discussing the implications of the empirical findi ngs of this study. Results are reported in this chapter for the t otal sample, followed by results for the first three re search questions; these are 64
PAGE 72
65 examined in conjunction with the effect on non-forecasting firms. The total sample represents an aggregation of different, and opposing, classifications of management earnings forecasts and, therefore, the results obtained are difficult to interpret. These results are reported here, however, for comparison with prior studies and are limited to standardized prediction errors tests. Total Sample The total sample consists of positive, negative, and neutral management earnings forecasts and, as a result, the direction of prediction errors cannot be specified ex-ante. A statistical test that abstracts from direction (Zut) is given by equation (8) and follows from squaring the standardized prediction errors. Table 4-1 reports the Z values for standardized prediction errors (Zvt), the mean value of the squared standardized prediction errors Ut, and Zut for all three portfolios. The Zvt statistic indicates rejection of the null hypothesis of no information content during the forecast announcements week, t and week t+6 for the experimental portfolio. The null hypothesis could not be rejected for the matched control portfolio during the forecast announcements week or any other week in the test period at a statistically significant level. The null hypothesis is rejected, however, for the random control portfolio during weeks t-1 and t+3, but not during the forecast announcements week.
PAGE 73
Experimental Week 2 vt Uc 2 ut -8 l. 27 0.97 -0 35 7 -0.45 .95 -0.68 -6 -0. 38 1.07 0.99 -5 0.24 1.06 0.83 -4 -0.43 1.19 2.66 -3 0.81 1.06 0.85 -2 o 29 1.10 1.38 -1 0.82 0.97 -0.39 9 2. 71 1. 51 6 .9 8 l o. 30 0.84 -2.16 2 0.72 o. 78 -3.10 3 0.02 0.92 -1.13 4 0.80 0.83 2.34 5 -0.17 1.17 2.36 6 1. 93 1.02 2.39 7 0.12 0.95 0. 73 8 l. 34 1. 3) 4.52 Table 4-1 Total Sam_le V alues for Z U and Z vt t ut Hatched Control 2 vt Uc 2 ut -0.93 0 85 2.15 -1.14 0.99 -0.21 1.16 1.08 1.07 -0.80 0.91 -1. 24 -0.04 1.05 o. 76 0.99 0.90 -1. 35 0.47 0 92 -1. 05 1.50 1.08 l. 21 -0.69 1.13 l. 92 -0.57 1.04 0.57 -0 .49 1.45 6 .34 -0.29 1.11 1.58 1. 30 1.33 4.66 1.60 1.41 5. 70 1.4 7 1.16 -2.27 0.48 .97 -0.36 -0. 02 1.01 0 14 Random Co ntrol 2 vt Uc 0.86 1.03 1.05 1.11 -0.15 1.08 0.37 1.16 0.38 0.94 0.16 0.91 0.03 1.04 1.85 1.17 0. 70 1.09 o. 70 1.07 0.89 o. 73 1.88 1.15 0.01 0.97 -1.13 1.04 0 82 1.05 0 .22 0 .9 7 1. 65 1.10 2 uc 0.47 l. 54 1. 1 2 1. 4~ -0.91 -1.32 0.62 2.63 l. 21 0.92 -1. 96 2.12 0.41 0.62 o. 73 -0.4 1 1.41 0\ 0\
PAGE 74
67 The Zut Statistic should be cautiously interpreted for two reasons. First, Zut is considered a test of the increased variance during the test period only under the assumption that the expected value of prediction errors is equal to zero. Accordingly, the variance change interpretation for weeks which exhibit prediction errors significantly different from zero is inappropriate (see Patell 1976, p. 261). Second, as would be expected when dealing with squared values, Zut values are sensitive to a small proportion o f large prediction errors. Results reported in Table 4-1 indicated that only the matched control portfolio exhibits an increase in the variance during the forecast announcements week unaccompanied b y a mean prediction error significantly different from zero. Furthermore, the observed increase in the variance of prediction errors associated with the matched control portfolio during the weeks immediatel y following the forecast announcements week are considerably larger than those associated with the other two portfolios. This indicates information effects during these weeks for the matched control portfolio. Statisticall y significant values of Zut unaccompanied by mean prediction errors significantly different from zero indicate change in the v ariance and signals rejection of the null hypothesis of no information content. Disaggregation of the total sample is necessar y to determine which of the three classifications of management forecasts of earnings, or w h a t c ombination, is causing the observed information effects. Furthermore, removal of outliers is necessar y to determine whether the observe d ef fect is also the general pattern of the sample.
PAGE 75
68 Examination of the total sample was motivated by the need to highlight the extent to which the sample of management earnings fore casts used in this study is similar to those used in prior studies. Since prior studies did not utilize control groups, the comparison is limited to the experimental portfolio. consistent with those of prior studies: The following findings are 1 the observed mean prediction errors during the forecast announcements week is statistically significant and positive. This result indicates that, on average, management forecasts of earnings are accompanied by an upward stock price revision. This result is consistent with those reported by Patell (1976) and Penman (1978). 2 the observed Ut Values for weeks immediately following the forecast announcements week are less than unity, indicating a reduction in the variance for these weeks, This result is consistent with findings reported by Patell (1976, p. 261). The Effect of the Forecasting Action The neutral portfolio of earnings forecasts was examined to determine whether the forecasting action by itself conveys favorable information to market participants. Table 4-2 reports the Z values for standardized preduction errors and cumulative prediction errors for the three portfolios. The null hypothesis of no information content, Ho1, was rejected at the O .05 level for the experimental portfolio during the forecast announcements week, whereas the two
PAGE 76
Week -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 6 9 Table 4-2 Total Neutral Values for Z and ZWL vt Experimental M atched Zvt ZWL Zvt 0.43 o : 43 0.39 -1. 07 -0.45 0.73 0.50 -0. 08 0. 2 0 o. 7 5 0.31 0.19 0.28 0.40 -0.84 0.60 0.61 -1.32 0.88 0.90 -0. 77 0.56 1.04 0.18 1. 93* 1. 62 -0.66 -0.51 1.38 -0.20 -0.1 2 1. 2 7 -1.28 0.16 1. 27 0.45 -0. 7 8 1.00 0.50 1. 2 2 1. 2 9 1. 4 7 0.91 1. 48 1. 82* 0.81 1. 64 0.17 0 94 1.8 2 0.99 Signific a nt a t the 0.05 level Control Random C o ntr o l ZWL Z v t Zwr.. 0.39 -1.8 4 -1.8 4 0. 7 9 0.68 -0.82 0.76 -0. 7 3 -1.09 0.76 -0.93 -1. 4 1 0.30 0.31 -1.12 -0.27 1.05 -0.60 -0.5 4 0.69 -0. 2 9 -0. 4 3 0. 2 3 0.19 -0.63 -1.1 7 -0.5 7 -0.66 0.09 -0.51 -1.0 2 0.33 0 .3 9 -0.85 1.64 0 84 -0.6 7 1. 0 4 -0.5 2 -0. 2 6 -1. 4 9 0 9 0 0. 22 -0.58 -1.0 2 0. 2 6 0. 3 8 0 90 0. 4 9 2 .60* 0 24
PAGE 77
70 control portfolios exhibited negative mean prediction errors durin g the same week. Table 4-3 contains the Z values for the distribution of weekly unstandardized prediction errors (et) and standardized prediction errors (Vt) The results of the distribution tests indicate rejection of the null hypotheses during the forecast announcements week at the 0.07 and 0.05 level of significance for et and vt respectively. These results suggest that the forecasting action by itself has favorable implications for stock prices of forecasting firms. Figure 4 1 portrays the behavior of cumulative averge standardized prediction errors during the test period. The behavior of the prediction errors indicates that the experimental portfolio enjoys a higher abnormal return relati ve to the two control portfolios during the test period. The trimmed mean test was conducted to determine whether the observed effect was a general pattern of the experimental portfolio or was due to the influence of a few large outliers. T able 4 4 reports the Z statistic for the trimmed mean of standardized prediction errors. The null hypothesis of no information content could not be rejected for the forecast announcements week or any other week in the test period at a statistically significant level. It should be noted, however, that the largest positive weekly mean prediction errors were observed during the forecast announcements week. These results lead to the conclusion thatthe forecasting action by itself induces an upward stock price adjustment, although such adjustment is not statistically significant. The cumulative trimmed abnormal returns associated with the experimental portfolio were larger th an those associated with the two control portfolios.
PAGE 78
Week -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 71 Table 4-3 Total Neutral Values for Z and Z et Vt Experimental Matched 2 et 2 vt 2 et 0.48 0.52 -0.16 -0.79 -1.17 -0.18 0.38 0.17 0. 4 4 0.60 0.93 -0.68 0.81 0.67 -0.61 1.04 0.67 -1.3 7 o. 77 0.82 -0.25 1.02 0.33 0.49 1.52 1. 72* -0.56 -0.83 -0.53 -0. 4 9 -0.21 -0.06 -1.09 0.61 0.24 -0.08 -1.20 -0.94 -0.55 1.28 0.99 0.87 1.11 1.15 2.11* 0.36 0.16 0.13 0.92 0.81 1. 58 Significant at the 0.05 level Control Random Control 2vt 2 et 2vt -0.12 -1.91 -1. 95 0.65 0.87 0. 75 0.20 -1.12 -0.66 -0.25 0.34 0.11 -0.87 0.87 0.32 -1.36 0.15 0.63 -0.85 -0.56 -0.65 0.19 -0.03 0.22 -0.69 -1.51 -1.20 -0.20 0.56 0.09 -1.39 0.03 0.36 -0. 44 -2.12 1.80 -0.50 0.85 0.96 1.43 -1. 70 -1. 49 1. 95* -0. 75 -0.5 7 0.19 -0. 43 -0.35 0.95 2.88* 3.35*
PAGE 79
1 00 0 90 0 80 0 70 0 60 0 50 0 40 -0 40 -0 50 -0 60 -0 70 -0 80 0 90 1 00 72 Experimental 1 1atc h C ontrol Random Contro l Figure 4-1 Neutral Group Cumulative Standardized Predict i on Errors ( l 7 V) l 1 t
PAGE 80
Week -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 73 Table 4-4 Total Neutral Trimmed Mean Values for Z and ZWL Vt Exp er imen tal Matched Zvt Zwt 2 vt 0.62 0.62 -0.17 -0.81 -0.13 0.03 0.64 0.26 0.17 1.19 0.82 0.44 0.78 1.08 0.36 0.90 1.34 -1.02 0.74 1.54 -0.83 0.34 1.56 0.32 1.23 1.87* -0.54 -0.45 1. 64 0.16 -0.21 1.50 0.99 0.26 1.51 -0.25 -1.11 1.14 -0.12 1.15 1.41 0.69 0.69 1. 54 1.50 0. 77 1.68* 0.62 0.83 1.83* 0.79 Significant at the 0.05 level Control Random Con;rol Zwt Zvt ZWL -0.17 -1. 86 -1. 86 -0. 09 0.55 -0.93 0.02 -0.83 1. 24 0.24 -0.71 -1. 4 3 0.37 0.23 -1.1 7 -0.08 0.87 -0. 7 2 -0.38 -0.54 -0.8 7 -0.25 0.26 -0. 72 -0.41 -1.01 -1.01 -0.34 0. 1 1 -0.93 -0.03 0.12 -0.85 -0.10 -1.21 -1.16 -0.04 0.87 -0.8 7 0.06 -1.57 -1. 26 0.45 -0.55 -1.36 0.59 -0. 4 0 -1. 4 2 0.76 1.02 -1. 1 3
PAGE 81
74 Results of the two-way analysis of variance are reported in Table 4-5. The null hypothesis of no differen ce among the means of the three portfolios was rejected at the 0.036 level. The actual risk-adjusted returns on the experimental portfolio over the entire test period were considerably higher than those achieved on the two control portfolios. These results support the earlier finding that the abnormal returns on the experimental portfolio were larger than the abnormal returns achieved on the two control portfolios during t he test period. The null hypothesis of no difference among the eleven informa tion treatment effects and the null hypothesis of no interaction effect could not be rejected at a statistically significant level. The Duncan's Multiple Range Test was used to test the null hypothesis that all thirty-three means are equal. This hypothesis could not be rejected at the O .OS level. These results suggest that the forecasting action by itself does not result in an upward stock price revision significantly different from that occurring at random. With respect to the industry-wide effect, the forecasting action by itself does not convey favorable information to non-forecasting firms in the same, or similar, industry. The prediction errors associated with the matched control portfolio were negative during the forecast announcements week. The behavior of the cumulative prediction errors prior to the forecasts disclosure week does not indicate that the forecasting action was motivat ed by industry-wide factors.
PAGE 82
Test Type of portfolio 75 Table 4-5 Total Neutral Anal y sis of Variance effect Information treatment effect Interaction effect Comparison of all means FV a l ue PR > F 3.33 o. 03 6 0. 2 6 0 .99 0 0 8 7 0. 624 0 .8 3 0 .7 3
PAGE 83
76 To summarize, results of prediction errors tests ind i cate th at the forecasting action by itself results in statistically signific an t abnormal returns during the forecast announcements week. The two-wa y analysis of variance test, using total stock returns, provided a contradictory result. Both of these tests, however, are sensitive to the existence of outliers. Removal of outliers changed the conclusion of the prediction errors test. The results of the trimmed prediction errors test support the conclusion that the forecasting action by itself does not result in a significant upward stock price revision. All tests, however, suggest that abnormal returns, and actual portfolio returns, during the test period are larger for the experimental portfolio than for the two control portfolios. Finally, the forecasting action by itself neither influences stock returns of non-forecasting firms, nor is it motivated by industr y -wide factors. The Effect of Newness of Management Earnings Forecasts The positive and the negative portfolios are the t w o high-level newness portfolios. portfolio. Results are reported below for each individual A The Positive Portfolio Table 4-6 reports the Z statistic associated with standardized prediction errors for each of the three p ort f olios. Figure 42 portrays the behavior of the cumulative standardized prediction errors
PAGE 84
Week -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 77 Table 4-6 Total Positive Values for Z and ZWL vt Experimental Matched Control Zvt ZWL Zvt ZWL 1.08 1.08 -0.18 -0.18 1.70* 1.96* -1.05 -0.87 1.02 2.20* 1. 31 -0.05 1. 77* 2.79* 0.43 0.26 -1.35 1. 89* 0.14 0.30 0.38 1. 88* -0.58 0.03 1. 38 2.26* 0.13 0.08 0.89 2.43* 0.60 0.29 6.44* 4.43* 1. 90* 0.69 0.57 4.39* 0.91 1.15 -0. 4 1 4.06* 0.79 0.86 -0.11 3.85* 0.73 1.03 -0.23 3.64* 0.60 1.15 -0.43 3.39* -0.26 1.04 0.51 3.41* 0.15 1.05 0.38 3.20* 0.81 1. 21 1.16 3.39* 0.57 1.04 Significant at the 0.05 level Random Control 2 vt ZWL 0.31 0.31 0.34 0.46 0.80 0.84 1.82* 1. 63 -0.11 1.42 -0.27 1.18 0.35 1. 22 2.15* 1. 91* 0.36 1. 92* 1. 5 4 2.30* -1.55 1. 73* 1.35 2.05* -1.11 1.66* -0. 48 1. 4 7 0.22 1. 48 -0.61 1. 28 1. 57 1. 22
PAGE 85
1 60 1 50 1.40 1 30 1 20 1 10 1 00 0 90 0.80 0 70 0 60 0 50 0 40 0 30 0 20 0 10 78 : .. --.. .. . ""' .. . .. .. .. ... .. . .. _/ _, ...... .;. ._.~,.... . _, ... : .,,.,, ....... "'i.,.,.,"'"' .. : ...... I : I : I .: I .. ,,, ... --~~ ,, I ,,,,., ... _,..,,,. I I 01111,A.:....1-___,1, _,_"'-....... ---'--.=.a~;.....-..i.... _.__ ________ __,;....___,, ___.. _._ _._ _. -0 10 -0 20 -0 30 --a. -7 r6 / / ... -5 -3 -2 1 0 2 3 4 5 6 7 8 Experimenta l l !atch Control Random Control Figure 4-2 Positive Group Cumulative Standardized Prediction Error ( 1;' 7 ) 1
PAGE 86
79 for the three portfolios. Results for the experimental portfolio can be summarized as follows: 1 the null hypothesis of no information content was rejected for the forecast announcements week and for two other weeks (t-7 and t-5) at the 0.05 level. 2 the largest positive mean prediction error was associated with the forecast announcements week. 3 with the exception of the first week, the cumulative prediction errors were statistically significant at the 0. 05 level for the entire test period. The size of the cumulative prediction errors associated with the experimental portfolio were considerably larger than those associated with either of the two control portfolios. Results for the control portfolios indicate that: 1 the null hypothesis of no information content during the forecasts disclosure week was rejected at the O. 05 level for the matched control portfolio but not for the random control portfolio. 2 the largest positive mean prediction error was associated with the matched control portfolio during the forecasts disclosure week. 3 the random control portfolio exhibits statistically significant positive prediction errors during weeks t-5 and t-1.
PAGE 87
80 Results of the distribution tests are shown in Table 4 -7. These results indicate that the null hypothesis of no information content was rejected for both the experimental a n d the matched control portfolios during the forecast announcements wee k but was not rejected for the random control portfolio. Ot h er weeks in the test period exhibited slightly lower Z values than those reported for t h e standardized predictio errors test. The Z values for the trimmed standardi z ed prediction err o rs test are shown in Table 4-8. Results for eac h o f the t h ree portfoli o s are: 1 the null hypothesis of no information content w as r ej ected for the e x perimental portfolio during t h e fo recast an nouncements week but not f or an y o t h er w eek i n t h e test period. 2 the null hypothesis of no i nf o rmati o n c o ntent c ould n o t be rejected at the O .OS level for the m atched contr ol port f olio during the entire test per i od. 3 the null hypothesis of no information content c oul d not b e rejected at the 0.05 level for the r an d o m c on tr ol p o r tfolio during the entire test per i od. The above results suggest that the p ositi v e p r e di c ti on errors observed for the experimental portfolio during the f orec a st annou n ce ments week represent a general pattern a nd w ere n o t influenced b y a few large outliers. This conclusion is not eq u a ll y valid fo r the matched control portfolio where the obser ve d e ffect seem s to have been influenced by a few large outli e rs.
PAGE 88
Week -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 81 Table 4-7 Total Positive Values for Z and Z et Vt Experimental Matched 2 et 2 vt 2 et 1.04 1.08 -0.03 1.12 1. 55 -0.91 1.35 1.14 0.84 1. 26 1. 56 0.01 -1.44 -1. 49 -0.59 0.14 0.26 -1.16 1. 70 1.38 0.45 0.97 0.83 0.26 5. 73* 5.96* 2.03* 0. 71 0.59 1.07 -0.02 -0.44 -0.54 0.03 0.05 0.16 -0.02 -0.18 0.62 -0.24 -0.58 -0.48 0.56 0.50 0.33 -0.19 -0.40 0.92 0.18 0.98 -0.91 Significant at the 0.05 level Control Random Control 2 vt 2 et 2 vt -0.09 0.41 0.45 -1.31 -0.21 0 .28 1.05 1.20 0.93 0.25 1.57 1.60 -0.21 -0.08 -0. 30 -0.78 -0.36 -0.25 0.13 0.92 0.53 0.56 2.13 1. 88 1. 64* 0.11 0.30 0.89 1.39 1. 58 -0. 7 9 -1.34 -1. 64 0. 39 1.27 1. 31 0. 4 0 -0. 70 -1.05 -0.23 -0.69 -0. 44 0.18 0.12 0.33 0.84 0.16 -0. 42 -0.75 1.22 1. 5 4
PAGE 89
Week -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 82 Table 4-8 Total Positive Trimmed Mean Values for Zvt Experimental M atched Control 0.40 0.09 o. 71 1.17 0.09 0.72 0.00 0.54 -1.14 0.34 -0.11 -0 .57 1. 21 0.22 0.92 0.75 5. 71* 1.18 0.41 0.20 -0. 7 8 -1. 06 -0. 4 6 0.31 0.34 0.17 -0.03 0.33 -0.17 0.43 -0.57 0.82 0. 74 -0.38 Significant at the 0.05 level Random Control 0.25 0.17 0.56 1.47 -0.77 -0.27 0.00 0.43 0 25 1.38 1.20 0.93 1.36 0.49 0.65 0.62 1.11
PAGE 90
83 Results of the two-way analysis of variance test are reported in Table 4-9. These results confirm the finding of the prediction errors tests. The null hypothesis of no difference among the means of the three portfolios could not be rejected at the 0.05 level. The null hypothesis of no difference among the eleven information treatment effects was rejected at the 0.0005 level. The largest positive overall weekly mean return was observed during the forecast announce ments week. The null hypothesis of no interaction effect could not be rejected at a statistically significant level. The Duncan's Multiple Range Test rejected the null hypothesis that the thirty-three weekly mean returns are equal at the O.011 level. The largest positive weekly mean return was associated with the experimental portfolio during the forecast announcements week. In sum, findings of all tests suggest that: 1 management forecasts of earnings which exceed market expec tations cause a statistically significant upward stock price revision for forecasting firms but not for non forecasting firms in the same, or similar, industry. 2 the observed effect for the experimental portfolio represents the general pat tern of the samples and is not influenced by a small number of large outliers. 3 the size of the cumulative prediction errors for the three portfolios prior to the forecast announcements week indicates that the management earnings forecasts are not motivated by industry-wide or economy-wide factors.
PAGE 91
8 4 Table 4-9 Test Total Positive Portfolios Analysis of Variance Type of portfolio effect Information treatment effect Interaction effect Comparison of all means F v a lu e P R >F o. 71 0. 4 9 3. 1 5 0.0005 1. 0 1 0 4 45 1. 66 0 011
PAGE 92
85 B The Negative Portfolio Table 4-10 reports the Z statistic for the standardized prediction errors and the cumulative prediction errors. Figure 4-3 portrays the behavior of cumulative average standardized prediction errors for the three portfolios. portfolio suggest that: Results for the experimental 1 the null hypothesis of no information content was rejected at the 0.05 level for the forecast announcements week and for four other weeks in the test period. Three of these four weeks preceded the forecast announcements week. 2 the largest negative weekly mean prediction errors occurred during the forecast announcements week. 3 with the exception of t-8 and t-1, all weeks preceding the forecast announcements week exhibited negative prediction errors. As a consequence, the cumulative prediction error was negative beginning at t-7 and continuing throughout the test period. 4 the size of the negative cumulative prediction errors f or the experimental portfolio was considerabl y larger th a n those reported f or the two control portfolios. Resu l ts for the two control portfolios suggest that: 1 the null hypothesis of no i n f ormation content was rejected for the matched control portfolio during the f orecast announcements week and during t7 T he same null
PAGE 93
Week -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 86 Table 4-10 Total Negative Values for Z and Z vt WL Experimental Matched Control z vt Zwr. 2 vt Zwr. 0.56 0.56 -0.85 -0.85 -1.25 -0.48 -2. 09* 2.16* -2.14 -1. 63 1.07 -1.15 -2.99 -2.91* -0.89 1.44 -0.33 -2.75* 1.12 -0.80 -0.18 -2.58* -0.13 -0.78 -1. 7 9 -3.07* 2.79 0.33 0.26 -2.78* 1. 75 0.93 -5.20* -4.35* -2.82* -0.0 7 -0.54 -4. 30* -1.59 -0.5 7 -1.04 4. 42* 1. 50 -0. 09 -0.22 -4. 29* -1.20 -0. 43 -0.06 -4 .14* 2.25 0. 2 9 -1. 90* -4. 50* 2 .05 0.83 1. 80 -3.87* -0.15 0. 77 0.41 -3. 64* -0.38 0.61 0.05 -3.52* -0.61 0. 48 Significant at the 0.05 level Random Control 2 vt Zwr. 0.60 0.60 0.89 1.48 -0.12 0.79 0.04 o. 70 0.47 0.84 -0.95 0.42 0.60 0.12 1.17 0 .53 -0.12 0 4 3 -0. 41 0 27 0.50 0.11 -0. 01 0 .11 -0.32 0.01 0.43 0.13 2.15 0.68 0.82 0.86 1.24 0.52
PAGE 94
1 00 0.80 0.60 0.40 0 20 87 . . . . . ,,, /--------~ ,#/ .lr / -----.. .. ,. ... \ ....... ... .... ,.. \ 0 ,.::;;.....~,...i.-...i._ ....... _.__ ....... ~--------~-----...... -i.....-'-1---" 1 ..... _....1.,_~ -6 -5 -4 -3 I -2 -1 O'.._ 1 ~..,, 3., 4 5 -0 20 -0.40 -0 60 -0 80 1 00 -1 20 -1 40 -1 60 -1 80 -2.00 -2 20 -2 40 I ;,--J ,,_.;, rx perimental !! atch Control R andom Contro l Figure 4-3 Negative Group Cumulative Average Standar-dized Prediction 6 7 8 Errors ( 1; 1 v ) , t
PAGE 95
88 hypothesis could not be rejected during t h e test peri o d f o r the random control portfolio. 2 the largest negative weekly mean pred i ction errors occurred for the matched control portfolio during t h e f orecast announcements week. The above findings are fully supported b y t h e results of the distribution tests. Table 4 -11 reports the Z stat i stic for et and v t during the test period. The trimmed standardized pre di ction errors test provides additional support for these findin g s and s ugg ests that the observed results are a general pat tern of the data a nd not du e t o the influence of a few large outliers. Tabl e 4 12 r ep orts essen tiall y the same results following the removal o f outliers. With respect to the industr y -wide e f f e ct all th r e e t e sts reported a significant abnormal return f or t he indu st ry m at ch e d control portfolio during the forecast announceme n ts week The s e results suggest that the management earnings f o rec a sts c on t a in in fo mation relevant to establishing e qui l ibrium prices of n on fo r e cast ing firms in the same, or similar, industry. M ore i mpo r t a n t ly, however, is the observed behaviour of abnormal returns a ss ocia t e d with the three portfolios prior to the forecast ann o unceme nts week. The cumulative abnormal returns f or t h e eig h t weeks p rec eding the forecast announcements week were positive for the i n dustr y matched and the random control portfolios. This su g gests th a t n ei ther industry-wide factors nor economy-wide factors induced t h e negative management
PAGE 96
W e ek -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 89 Table 4-11 Total Negative Values for Z and Z et Vt Experimental M atched 2 et 2 vt 2 et -0.03 0.56 -1. 2 5 -1.19 -1. 38 -1. 77* -2. 4 2* -2.55* 1.68 -3. 4 0* -3.3 7 -1. 2 1 7 6 -0.76 1.12 0.35 0.29 -0.8 -1. 97* 2 .26* 2. 2 1 0 08 0. 2 1 1.1 4 -3.46* -3. 71* -2.55* -0.3 4 -0.19 -1. 99* 2 4 3* 2. 3 4 0.56 -0.80 -0.80 -1.09 -0 10 -0.0 2 2 24 -1. 38 7 4 1. 72 1. 55 1. 3 4 -0. 7 8 0 8 4 0.64 -0. 66 -0. 2 9 -0.16 1.4 0 Significant at the 0 05 level Control Random Control 2 v t 2 et 2v t -0.86 0. 4 8 0. 42 -1. 7 2* 0 55 0 .7 8 1.29 0. 44 0 .27 -1.00 -0.35 -0.03 1.19 0.32 0. 4 9 -0. 4 3 -1.06 -1.03 2.6 4 -0.90 -0.80 9 4 1. 27 0.98 -2. 4 9* 22 0 0 5 -1. 61 -0.66 0 7 5 0. 7 6 -0. 2 0 0. 4 3 0. 14 0 .27 -0. 4 0 1. 90 0 0 9 0 01 1. 69 0 .4 8 0. 2 9 -0. 47 2 7 7 2 .3 7 -0 .4 9 1. 06 1.03 -1. 4 6 0. 22 -0.56
PAGE 97
90 Table 4-12 Total Negative Trimmed Mean Values for Zvt Week Experimental M atched Control -8 0.66 -0.16 -7 -1.18 -1.98* -6 -2. 06* 1.00 -5 -2. 62* -1.13 -4 -0.65 0.83 -3 0 0.29 -2 1. 26 1.83 -1 0.51 0.66 0 -4.45* -:2.99* 1 -0.43 -1.02 2 -0.78 -0.37 3 -0.18 -1.19 4 -0.35 1.42 5 -1. 30 0.49 6 1.32 -0.05 7 0.18 -0.05 8 0.30 -0.53 Significant at the 0.05 level Random Control 0. 63 0.57 1.20 -0. 4 6 0.28 0.97 -0.85 0.53 -0. 22 0. 4 3 0.47 -0.0 4 0.25 0.37 1.51 0.69 1. 58
PAGE 98
91 earnings forecasts. Rather, the disclosure of these forecasts resulted in a downward revaluation of other firms in the same, or similar, industry. The two-way analysis of variance test provides additional evidence consistent with the results of prediction errors tests. Results for the analysis of variance test are summarized in Table 4-13. The null hypothesis of no difference among the three portfolios was rejected at the 0.01 level. The null hypothesis of no difference among the eleven information treatment effects was rejected at the 0.0002 level, The largest negative weekly mean returns were observed during the forecast announcements week. The null hypothesis of no interaction effect could not rejected at a statistically significant level. The Duncan's Multiple Range Test rejected the null hypothesis that all thirty-three weakly mean returns are equal at the 0 .0002 level. The two largest negative weekly mean returns occurred during the forecast announcements week for the experimental and the industry matched control portfolio respectively. In sum, the positive (negative) news forecasts were accompanied by significant upward (downward) stock price revisions during the forecast announcements week. The behavior of the cumulative prediction error indicated proper anticipation of the earnings fore casts content prior to the announcements week. Both classi f ications of forecasts induced an industry-wide effect, although the observed effect for the negative news forecasts wa~ more pro f ound than was the observed effect f or the positive news forecasts.
PAGE 99
92 Table 4-13 Total Negative Portfolios Analysis of Variance Test F V alue Type of portfolio ef f ect 4 .5 7 I Information treatment e ff ec t 2.7 3 Interaction effect 1. 19 Comparison of all means 1.88 PR> F 0 010 0 002 0.254 0 0 0 2
PAGE 100
93 The Effect of Imputed Accuracy of Management Earnings Forecasts The two high-level newness forecasts are classified as either positive or negative forecasts. Each of these two classifications was subdivided into two unequal portfolios according to the degree of imputed accuracy ascribed to these forecasts. The result is four portfolios; with high and low imputed accuracy for both the positive and the negative classification. Testing the effect of imputed accuracy of management earnings forecasts is accomplished by an individual examination of each of the four portfolios and by a comparison of the high-level and low-level imputed accuracy portfolios for each forecast classification. Separate examination of each port folio will indicatewhether the information content of management earnings forecasts is a function of newness alone or the combination of newness and imputed accuracy of these forecasts. The comparison between the high and the low level imputed accuracy portfolios for each classification of forecasts will indicate the extent to which the information content is dependent upon the level of imputed accuracy ascribed to forecasts. Results are reported below for the two positive portfolios followed by results for the two negative portfolios. Positive Forecasts-High Imputed Accuracy Table 4-14 reports the Z statistic for standardized prediction errors and cumulative prediction errors. Figure 4-4 portrays the
PAGE 101
W eek -8 7 -6 -5 4 -3 -2 -1 0 1 2 3 4 5 6 7 8 94 Table 4 -1 4 Positive High Imputed Accuracy Values for Z and ZWL vt Experimental Matched Con tr ol 2 vt ZWL 2 vt ZWL 1. 35 1. 35 -0.93 0 93 1. 20 1.80* -1. 65 1. 82 1.01 2.62* 1.01 0 .91 0.65 2.10* 0.13 0. 71 0.39 1. 7 0* -0.37 0.81 0. 4 6 1. 74 -0 .17 0 .8 1 1. 66 2 .24* 0.05 -0. 73 1. 39 2.59* 0.68 -0. 44 6.55* 4 .62* 1. 8 7* 0 2 1 0.65 4 .59* 1.58 o. 7 0 0.08 4.40* 1.31 0.27 -0.99 3.93* -0.65 0 07 0.15 3.82* -0. 39 0 .0 4 -0.14 3.64* 0.12 -0.01 -0.21 3.46* -0. 43 -0.12 -0.03 3 34* 0 .5 1 0 01 1.54 3. 62 0. 75 0 18 Significant at the 0.05 level Random Con tr ol 2 vt ZWL 0 72 0. 7 2 0 06 0 .5 6 0.63 0.82 2 .90* 2 .16 0.48 2 .15* -1. 18 1. 47 0 .3 1 1.48 2.03* 2.10* 0.03 1. 98* 1. 82* 2. 4 5* -1. 94 1.75* 1.98* 2. 25: : -1.39 1. 7 7* 2.08 1.15 1. 5 1 1. so 0. 41. 1.36 1.8 7 l. 77
PAGE 102
2.00 1.90 1 80 1.70 1 60 1 50 1.40 1 30 1.20 1 10 1.00 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0 1 ... , . . . . . : 95 . .. ... . . . :~ . . .. : : .. : . . . : .. . .. .. . .. ...... ... . . . . . . . . . . . . . 0 N . I\ I \ I \ I \ I \ I ', I o..,_..__ ..... __ ..... _.._ _. _____ ..__ ..... __ i...4,_.._ _. ____ .... """~~::: =-----~--, '""'-, -8 -7 -6 -5 -4 -3 -2 -1 Of 2 3 4---s-:--,s,--7 -0.1 -0.2 -0.3 -0.4 '\ ,,,,,,/ \ ,--.......... ;-" \ ,' ..._ ____ ,,, \,"' Figure 4-4 Positive H H F xperimental !l atch Control Random Control
PAGE 103
96 behavior of the cumulative average standardized prediction errors for the three portfolios. Results for the experimental portfolio indicate that: l the null hypothesis of no information content was rejected at the 0. 05 level for the forecast announcements week and week t-2 but not for any other week in the test period. 2 the largest positive weekly mean prediction errors occured during the forecast announce ments week 3 all weeks preceding the forecast announcemen ts week exhibited positive mean prediction errors. 4 with the exception of week t-1, the cumulative prediction errors were statistically significant at the 0 05 level. The size of the cumulative prediction errors was consider ably larger than that for the matched control portfolio but only slightly larger than the cumulative prediction errors for the random control portfolio. Results for the control portfolios indicate that: l the null hypothesis of no information content was rejected at the 0.05 level for the matched control portfolio during the forecast announcements week but not for any other week in the test period. 2 the null hypothesis of no information content could not be rejected for the random control portfolio during the fore cast announcements week. However, the null h y pothesis was rejected at the 0 .05 level for fiv e weeks in the test period.
PAGE 104
97 Results of the distribution tests for unstandardized and standardized prediction errors are shown in Table 4-15. These results indicate rejection of the null hypothesis of no information content during the forecast announcements week for both the experimental port folio and the matched control portfolio, at the 0.001 and 0.07 levels respectively. Other weeks in the test period exhibited lower Z values than those reported for the standardized prediction errors test. Results of the trimmed standardized prediction errors test are reported in Table 4-16. Results for the experimental portfolio indicate rejection of the null hypothesis of no information content at the 0.05 level during the forecast announcement week. The null hypothesis could not be rejected for any other week in the test period. Results for the matched control portfolio indicate that the largest positive mean prediction error occurred during the forecast announcements week. Nonetheless, the null hypothesis of no information content could not be rejected at a statistically significant level for the forecast announcements week or for any other week in the test period. Results for the random control portfolio indicate that the null hypothesis could not be rejected during the forecast announcements week. The null hypothesis, however, was rejected at the O .05 level for weeks t-5, t+l and t+2. The above results strongly indicate that an upward stock price re v i sion occurred during the forecast announcements week for the
PAGE 105
Week -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 98 Table 4-15 Positive High Imputed Accuracy Values for Z and Z et Vt Experimental Matched Control 2 et 2vt 2 et 2vt 1. 29 1.56 -1.8 -1.3 4 1.06 1.22 -2.48 -1.34 1.36 o. 92 0.48 0.62 0.20 o. 63 -0.56 -0. 23 -0. 45 -0.38 -0.36 -0.28 0.66 0.50 -0. 43 -0.41 1. 77* 1.69* 0.60 0.08 1. 38 1.27 0.58 0.66 5.44* 5.60* 1. 53 1.45 0.89 0.70 1.64* 1. 40 0.73 0.11 -0. 70 -1.12 -1.10 -0.94 -0.69 -0.64 0.43 0.15 -0. 15 -0.2 4 -0.02 -0.16 -0.12 0.20 -0.30 -0.2 4 -0. 46 -0. 44 0.25 -0.03 0.35 0.28 -.38 1.20 0.2 7 0.48 Significant at the 0.05 level Random Control 2 et 2vt 0.35 0.53 0.33 0.34 0.73 0. 4 9 1. 79-k 2. 08* 0.40 0.39 -0.99 -0. 96 0.56 0.31 1.90* 1.64* -0.58 0.02 1.30 1. 7 2* -1.80 2.07 1. 65* 1. 74 0. 62 -1.03 -2.33 -2.23 1.13 1. 4 3 0. 4 5 -0.22 1.16 1. 66
PAGE 106
99 Table 4-16 Positive High Imputed Accuracy Portfolios Trimmed Mean Values for Zv t Week Experimental Matched Control -8 1.11 -0.81 -7 1.03 -1.54 -6 0.26 0.51 -5 0.56 0.38 -4 -0.39 -0.02 -3 0.30 -0.02 -2 1.43 0.12 -1 1.25 1.00 0 5.51* 1.10 1 0.43 o. 64 2 0.10 -1.34 3 -0.94 -0.69 4 0.15 0.62 5 -0. 43 0.36 6 0.30 0.07 7 0.05 0.69 8 1.26 0.55 Significant at the 0.05 level Random Control 0.57 -0.19 0.25 2.06* -0.17 0.86 -0.11 0.99 -0.33 1.66* 1. 68* 1.39 -1.50 -1. 99 1.39 0.85 1. 4 5
PAGE 107
100 experimental portfolio. The matched control portfolio also exhibited an upward stock price revision during the forecast announcements week. Such revision, however, was neither sufficiently strong nor was it representative enough of the general pattern of the sample to be considered statistically significant. Accordingly, an industry-wide effect cannot be attributed to this classification of management earnings forecasts. Moreover, the behavior of the cumulative abnormal returns prior to the forecast disclosure week suggest that the fore casting action was not motivated by industry-wide factors. The results of the two-way analysis of variance test supported the above findings; results for this test are reported in Table 4-17. These results indicate that the null hypothesis of no difference among the three groups could not be rejected at the 0 .05 level. The null hypothesis of no difference among the eleven information effects was rejected at the 0.0001 level. The null hypothesis of no interaction effect could not be rejected at the 0.05 level. The results of the Duncan's Multiple Range Test indicate that all thirty three means are not equal and this hypothesis was rejected at the 0.0002 level. The largest weekly mean returns occurred during the forecast announcements week and were associated with the experimental portfolio. Positive Forecasts-Low Imputed Accuracy Table 4-18 reports the Z values for the standardized predict ion errors and the cumulative prediction errors for the three portfolios.
PAGE 108
101 Table 4-17 Positive High Imputed Accuracy Portfolios Analysis of Variance Test F-Value Type of Portfolio effect 1. 4 4 Information treatment effect 3.63 Interaction effect 1.49 Comparison of all means 2.16 PR > F 0.2 4 0.0001 0.0 7 0.0002
PAGE 109
Week -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 102 Table 4-18 Positive Low Imputed Accurac y Values for Z and ZWL vt Experimental Matched Control 2vt ZWL 2vt Zwt. -0.23 -0.23 0.78 0.78 1.14 0.64 0.30 0. 7 5 0.18 o. 63 0.84 1.10 1. 79* 1.44 0.51 1.21 -1.58 0.58 0.64 1.36 -0.32 0.40 -0.69 0.97 0.03 0.38 0.14 0.95 -0.05 0.34 0.14 0.94 2.61* 1.19 0.49 1.05 0.31 1.22 -0.16 0.94 -0.96 0.88 0.30 0.99 0.97 1.12 1.84* 1. 4 8 -0.52 0.93 1.36 1.80* -0.39 0.79 -0.53 1.59 0.55 0.91 0.70 1. 72* -0.69 0. 7 1 0.65 1.82* -0.18 o. 64 -1.67 1.36 Significant at the 0.05 level Random Control 2vt ZWL -0.58 -0.58 0.33 -0.18 0. 4 0 0.09 -0. 47 -0.16 -0.79 -0.50 0.92 -0.08 0.37 0.07 0.8 4 0.36 0.30 0. 44 0.23 0. 4 9 -0.25 0.39 -0. 4 0 0.30 -0. 4 3 0 .13 1. 65* 0 .5 7 -1.25 0.8 7 -0. 24 0 47 0.1 6 0. 4 9
PAGE 110
1 40 1 30 1 20 1 10 1.00 0.90 0.80 0.70 0.60 0.50 0.40 0.30 0 20 0.10 -0 .1 0 -0.20 -0 30 -0 40 -0.50 103 .,, ; ,.; ,-.. .. . . : ... ... : .. ..., --.. .. . . .. : ..... 0 Figure 4-5 Positive H-L 2 3 4 5 6 7 Exper i menta l l fatch Control Random Control 8
PAGE 111
104 Figure 4-5 portrays the behavior of the cumulative average standardized prediction errors for the three portfolios. Results for the experimental portfolio indicate that: 1 the null hypothesis of no information content was rejected at the O. 05 level for the forecast announcements week and week t-5 but not for any other week in the test period. 2 the cumulative prediction errors were not significantly different from zero at the 0 .05 level for the entire test period. Moreover, the size of the prediction errors associated with the experimental portfolio was not materially different from prediction errors associated with the two control portfolios. Results for the two control portfolios indicate that: 1 the null hypothesis of no information content could not be rejected during the forecast announcements week for either of the control portfolios. 2 the null hypothesis was rejected at the 0.05 level fo r the matched control portfolio during week t+3, and for the random control portfolio during week t+S. Table 4-19 reports the results of the distribution tests. These results are consistent with the above findings al though the Z values for the distribution tests were slightly lower t han those reported for the prediction errors test. Table 4-20 re ports results of the trimmed mean test. The null hypothesis o f no info rmation
PAGE 112
Week -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 105 Table 4-19 Positive Low Imputed Accuracy Values for Z and Z et vt Experimental Matched Control Zee zvt Zee zvt 0.07 -0.25 1.01 0.91 0.42 0.92 0.67 0.27 -0.05 0.23 0.64 0.79 1.53 1.37 0.63 0.49 -1.68 -1. 95 0.00 0.46 -0.57 -0.30 -1.00 -0.40 0.27 0.02 -0.15 -0.13 0.00 -0.6 -0.35 0.11 2.49* 2. 80* 0.89 0.32 0.38 0.40 -0.32 -0.20 -1. 25 -1.21 0.06 0.21 0.84 0.95 1. 78* 1.89* -0.59 -0.55 1.10 1.00 -0.07 -0.46 -0.31 -0.35 0.82 0.60 1.05 0.74 -0.81 -0.69 0.60 0.66 -0.21 -0.24 -1.86 -1. 73 Significant at the 0.05 level Random Control Zee zvt -0.46 -0.61 0.13 0.40 0. 77 0.71 -0.24 -0.66 -0.74 -0. 79 1.12 1.06 0.38 0.20 0.96 0.93 0. 63 0.41 -0.10 0.35 -0.14 -0.25 -0.52 -0.30 -0.34 -0.39 1. 69* 1. 58 -1. 64 -1. 54 -0.58 -0.36 0.33 0.19
PAGE 113
Week -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 106 Table 4-20 Positive Low Imputed Accuracy Trimmed Mean Values for Zv t Experimental Matched Control -0.56 0. 7 8 -0.11 -0.00 -0.16 0.49 0.63 0.37 -1.27 0.53 0.50 -0.82 0.19 0.19 -0.04 -0.02 2.28* 0.34 0. 1 2 -0. 4 3 -1.27 -0.06 0.38 1.26 0.69 0.96 -0.55 -0.09 0.08 0.55 -0.90 0. 4 4 -0.33 -1.20 Significant at the 0.05 level Random Control 1 .0 2 -0.03 0.55 0.2 7 -0.9 7 -0.5 7 0. 12 0. 4 8 0.0 1 0. J.. 8 -0. 1 3 -0.19 -0.33 1. 5 4 0. 6 0 0.03 0.01
PAGE 114
107 content was rejected for the experimental portfolio during the forecast announcements week but not for any other week in the test period. As for the two control portfolios the null hypothesis of no information content could not be rejected during any week in the test period. These results indicate that the information content of management forecasts of earnings which exceed market expectations induces an upward stock price revision despite investors assessment of low-level accuracy for these forecasts. Results of prediction errors tests clearly indicate that t h is classification of managements earnings forecasts does not conve y industry-wide information. The behavior of abnormal returns indicate that the matched control portfolio enjoyed a marginally hig h er abnormal returns than the experimental portfolio prior to the forecast announcement week. The trimmed standardized prediction errors test (see Table 4-20) indicates that such marginal difference is mainl y due to outliers. As such, the inference that industry-wide f actors induced the management earnings forecasts is not justifiable. Results of the two-way analysis of variance test are shown in T able 4-21. The null hypothes i s o f no di ff erence among the three port fo lios could not not be rejected. The null hypothesis of no d iffe rence among the eleven information treatment effects and the null h ypothesi s of no interaction ef f ect could not be rejected at a s tatistic a l l y significant level. These results are in conflict with res u l t s of pre d i ction errors tests. Th i s con f lict in results should,
PAGE 115
108 Table 4-21 Positive Low Imputed Accuracy Portfolios Analysis of Variance Test F-Value Type of portfolio effect 0 055 Information treatment effect 1.33 Interaction effect 0.71 Comparison of all means .86 PR> F 0.955 0.20 .82 0.69
PAGE 116
109 however, be tempered by the fact that the largest weekly mean stock return is observed for the experimental portfolio during the forecast announcements week. This suggests that an upward stock price revision occurred during the forecast announcements week. Such a revision, however, was not sufficiently strong to influence the overall mean for the three portfolios during the forecast announcements week and thus cause a rejection of the null hypothesis of no information treatment effect. Comparison Between the High-level and the Low-level Imputed Accuracy Portfolios Four tests were used to test the null hypothesis of no difference between abnormal returns associated with each of the highlevel and the low-level imputed accuracy portfol i os. The first two tests were based on Equation (9) and involved a comparison of weekl y mean standardized prediction errors and trimmed prediction errors. Results for these two tests are reported in Table 4-2 2 These resu l ts indicate that prediction errors associated with the highl evel imputed accurac y portfolio are significantly different from those associated with the low-level imputed accuracy portfolio at t h e 0.05 level during the forecast announcements week. Both tests also indicate that the null hypothesis of no difference between the t w o m eans could not be rejected at the 0.05 level for any other wee k in the test period. The third and fourth tests examined the weekl y distribu t i o n of observed prediction errors (et) and standardi z ed pr e diction errors ( vt ) respec tively. Table 4-23 reports the Z statistic f or the distribution tests
PAGE 117
110 Table 4-22 Positive Portfolios Standardized Prediction Errors Zdv value for the Difference Between High and Low Imputed Accuracy Portfolios Total Prediction Errors Trimmed Prediction Week Zdvt Zdvt -8 1.03 1.15 -7 -0.05 0.76 -6 0.52 0.29 -5 -0.87 -0.10 -4 0.89 0.70 -3 0.53 -0.18 -2 1.06 0.79 -1 0.94 0.84 0 2.35* 1.88* l 0.20 0.19 2 0.75 1.02 3 -1.35 -0.89 4 0.47 -0. 42 5 0.19 -0.51 6 -0.53 0.13 7 0.48 0. 71 8 1.13 1.0 7 Significant at the 0.05 level Errors
PAGE 118
Week -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 1 8 111 Table 4-23 Positive Portfolios Zdet and Zdvt for the Distributions Tests Zdet 0.70 0.30 1.09 -1.01 0.81 0.86 0.93 1.03 2.30* 0.34 1.45 -1.36 0.73 0.06 -0.85 o. 77 0.45 Significant at the 0.05 level Zdvt 1.20 -0.06 0.57 -0.78 1.03 0.55 1.02 0.97 2.29* 0.24 1.02 -1.36 0.51 0.23 -0.62 0.51 1.15
PAGE 119
112 based on Equations (10) and (11). Results for both tests indicate rejection of the null hypothesis at the 0.05 level during the forecast announcements week but not for any other week in the test period. To summarize, management forecasts of earnings which exceeded market expectations induced an upward stock price revision during the forecast announcements week, irrespective of the level of imputed accuracy ascribed to these forecasts. Prediction errors tests indicate that observed revision is statisticall y significant fo r both the high-level and the low-level imputed accuracy portfolios. The results of the two-way analysis of variance test indicate that only the revision associated with the high-level imputed accuracy portfolio is statistically significant. Comparison between abnormal returns associated with the two portfolios suggest that the upward stock price revision for the high level imputed accuracy portfolio was significantly greater than the revision associated with the low-level imputed accura c y portfolio. With respect to the industry-wide effe ct, the high-level imputed accuracy portfolio seems to induce positive abnormal returns for non-forecasting firms in the same, or similar, industry during the forecast announcements week. This industry-wide effect, however, was neither statistically significant nor was it a general pattern of the sample of non-forecasting firms. The low-level imputed accura c y portfolio did not appear to induce an industrywide effect.
PAGE 120
113 Negative Forecasts-High Imputed Accuracy Table 4-24 reports the Z statistic for standardized prediction errors and cumulative prediction errors for the three portfolios. Figure 4-6 portrays the behavior of the cumulative average standardi zed prediction errors for the three portfolios. experimental portfolio indicate that: Results for the 1 the null hypothesis of no information content was rejected at the 0.05 level for the forecast announcements week and for five other weeks in the test period. Four of the f ive weeks preceded the forecast announcements week. 2 the largest negative weekly mean prediction errors occurred during the forecast announcements week. 3 the size of the cumulative prediction error f or the experimental portfolio was considerably larger than those reported for the control portfolio. Results for the two control portfolios suggest that: 1 the null hypothesis of no information content was rejecte d at the 0.05 level for the matched control portfolio during the forecast announcements week and w eek t+3. 2 The null hypothesis could not be rejected during the fore cast announcements week for the random control portfolio. The null hypothesis is rejected, however, during week t+8 for the random control portfolio.
PAGE 121
Week -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 s 6 7 8 114 Table 4-24 Negative High Imputed Accuracy Values for Z and ZWL vt Experimental Matched Control Zvt Zwr.. Zvt Zwt 1.04 1.04 -1.03 -1. 03 -1. 98* -0.67 -0.98 -1. 4 2 -1. 74 -1.55 0.70 -0 .7 6 -2. 98* -2.83* -0.7 4 -1.03 -1.16 -3.05* 0.24 -0.81 -0.90 -3.15* -0.32 -0.8 7 -2. 49* -3.86* 2.11 0 0.49 -3.44* 2.27 0. 7 9 -3.76* -4.50* -1. 75* 0 1 7 -1.00 -4.58* 0.02 0.1 7 -0. 74 -4.59* 1. 65 0.65 -1. 21 -4.74* -1. 86* 0.09 0.21 -4.50* 0.88 0.33 -2.17* -4.92* 2.24 0.92 2.26 -4.17* 0.82 0.68 0.21 -4.09* 0.20 0. 7 0 0.38 -4.06* -0.62 0.53 Significant at the 0.05 level Random Contro l Zvt Zwr.. 0.34 0.34 -0.61 -0.19 0.94 0.39 0.01 0.34 0.69 0.62 -1.22 0.06 0. 4 3 0.22 1.13 0.61 -0.16 0.52 0.04 0.3 4 -0. 22 0. 4 1 0. 2 5 0 47 -0.6 7 0 2 6 0.0 4 0. 2 6 1. 77 0 71 0.19 0 74 -1. 71 0.30
PAGE 122
115 0.80 0.60 --........ / 0 40 0 20 I I . I' ,..._..._ I ,!, ~-I 1. ---... .... ~ / ,9a' -1 1 ~ -I I o~~~~~~-~------~~--,~------....... -~~~-----~----~-~ -6 -5 -4 -3 ; _, 0 2 3 4 5 6 7 8 -0.20 -0.40 -0 60 -0.80 ~,.oo -1 20 -1 40 -1 60 -1 80 -2.00 -2.20 -2 40 -2.60 -2.80 I ..-.., I Experimental Match Control Random Control Figure 4-6 Negative H-H
PAGE 123
116 Table 4-25 reports the Z statistic for the distribution tests. Results of these tests are consistent with the above findings. Table 4-26 reports the Z statistic for trimmed standardized prediction errors. These results confirmed the above findings and indicate that the observed effect was a general pattern of the samples. The null hypothesis of no information content was rejected at the 0. 05 level during the forecast announcements week for both the experimental and the matched control portfolios but not for the random control port folio. Results for the experimental portfolio indicate a strong down ward stock price revision prior to the forecast announcements week. These results suggest an industry-wide effect for negative news fore casts which are perceived to be accurate by market participants. Furthermore, the observed behavior of the cumulative abnormal returns associated with the three portfolios prior to the forecast announce ments week suggests that neither industry-wide factors nor economy wide factors induced a downward revaluation of firms in the same, or similar, industry. The two-way analysis of variance test provides additional support to the above findings. Table 4-27 reports the results for this test. The null hypothesis of no difference among the three portfolios was rejected at the 0.004 level. The null hypothesis of no difference among the eleven information treatment effects was rejected at the 0. 009 1 eve 1. The null hypothesis of no interaction effect could not be rejected. The Duncan's Multiple Range Test rejected the null hypothesis that all thirty three means are equal at the 0 05
PAGE 124
We e k -8 -7 -6 -5 4 -3 -2 -1 0 1 2 3 4 5 6 7 8 117 Table 4-25 Negative High Imputed Accuracy Values for Z and Z et Vt Experimental Matched Control 2 et 2vt iet 2vt 0.48 0.94 -0.71 -0.67 -2.38" -2.40* -1.07 -0.93 -2.91" -2.94* 0.71 0.23 -2.29" -2.25* -0.99 -1.19 -0.58 -0.80 0.61 0.48 -0.52 -0.40 -0.82 o.33 -1. 98" -1. 99 2.88 2.72 0.86 0.81 1.41 1.04 -2.50,. -2.85* -2.10* -1.94* -1.13 -0.94 -0.21 -0.10 :-1.18 -1.19 0.54 0.56 -1.19 -1.16 -1. 87* -1.68* 0.65 0.72 1.13 0.45 -1.31 -1.72* 1. 42 1.35 1.58 1.30 -1.22 1.22 -0.75 -0.62 -0.07 0.04 -0.14 0.01 -0.77 -0.64 Significant at the 0.05 level Random Control Zee Zvt 0.36 0.17 -1.12 -1.01 0.84 0.52 -0.79 -0.33 1. 31 0.88 -1.03 0.90 0.21 0.41 0. 84 0.86 -0.37 0.14 0.17 0.01 -0.41 -0.10 0.29 0.18 -0.43 -0. 4 3 0.06 0.16 2.18 1. 75 -0.21 -0.22 -0.88 -1.30
PAGE 125
Week -8 -7 -6 -5 -4 -3 -2 -1 0 l 2 3 4 5 6 7 8 118 Table 4-26 Negative High Imputed Accuracy Trimmed Mean Values for Zv t Experimental Matched Control 1.10 -0.65 -1. 89* -0.62 -1.1 4 -0.94 -2.49* -0.93 -1.2 7 0.01 -0.77 -0. 4 5 -2.35* 1.95 0.15 l.83 -3.36* -1.83* -0.41 0. 4 0 -0.80 0.80 -1.15 -1. 96* 0.29 0.81 -2.19* 0 36 2 03 0. 7 0 -0. 2 9 -0.53 -0.81 -0.50 Significant at the 0.05 level Random Control 0.3 4 -0.55 0.93 -0.55 1.11 1.24 -0.16 0.2 4 0.51 0.85 -0. 4 0 -0. 2 0 0. 2 5 -0.1 6 1.39 0. 63 -1. 4 5
PAGE 126
119 Table 4-27 Negative High Imputed Accuracy Portfolios Analysis of Variance Test F-Value Type of portfolio effect 5.59 Information treatment effect 2.36 Interaction effect 0.60 Comparison of all means 1.46 PR> F 0.004 0.009 0.914 0.047
PAGE 127
120 level. The largest negative weekly mean return occurred during the forecast announcements week and was associated with the experimental portfolio. Negative Low Imputed Accuracy Portfolio Results for the standardized prediction errors test are reported in Table 4-28 Figure 4-7 portrays the behavior of the cumulative average standardized prediction errors for the three portfolios. Results for the distribution tests are reported in Table 4-29. All three tests essentially indicate the same results. The null hypothesis of no information content is rejected at the 0 0 5 level for the experimental portfolio during the forecast announcements week and week t-6. Results for the matched control portfolio indicate rejection of the null hypothesis at the 0.05 level during t he forecas t announcements week, week t+l and week t7. The null hypothesis could not be rejected at the 0 .05 level during the fo recast anno uncements week or any other week in the test period for the random contro l portfolio. Results of the trimmed standardized prediction err or s test are reported in Table 4-30. These results indicate rejection of the null hypothesis of no information content at the 0 .05 level for both the experimental and the matched control port f olios dur ing the forecast announcements week. This finding suggests that the observed effect during the forecast announcements week is a general pattern of the two samples. The null hypothesis was also rejected at the 0.06 level for
PAGE 128
Week -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 121 Table 4-28 Nega tive Low Imputed Accuracy Values for Z and ZWL vt Experimental Matched Control 2 vt ZWL 2 vt ZWL 0.09 0.09 -0.21 -0.21 0.53 0.49 -2.13 -1. 65* -1. 84* -0. 76 0. 98 -0.62 -1. 26 -1.33 -0.84 -1.00 0.17 -1.04 1.05 0.31 0.57 -0. 70 0.93 0.09 0.42 -0.47 1.40 0.60 -0.02 -0. 45 -0.21 0 50 -2.88* -1. 46 -2. 94* -0.51 -0.06 -1. 4 0 -2.15* -1.16 -0.93 -1.64* 0.48 -0.97 0.82 -1. 32 -0.22 -0.98 -0.61 -1.45 2.86 -0.15 0.47 -1. 26 1.00 0.1 2 0.06 -1.20 o. 72 0.30 0.63 -0.84 -0.66 -0.13 1.09 -0. 71 -0.18 0.08 I Significant at the 0.05 level Random Control 2vt ZWL 0.51 0.51 2.00 1. 78 1.21 0.75 0.05 0.68 -0.08 0.57 0.07 0.49 -1.38 -0.06 0.49 0.11 -0.13 0.06 0.66 -0.15 -0.51 0.29 -0.30 -0.3 7 0.26 0.28 0.60 -0. 1 0 1. 24 0.22 1.01 0. 47 -0.04 0.4 4
PAGE 129
0 80 0 70 0 60 0 50 0 40 0 30 0.20 0 10 -0 10 -0.20 -0.30 -0.40 -0 50 -0 60 -0 70 -0 80 -0 90 122 .. .... : . . .. . .... : ... ... :... .. ... . . . . . .... -8 \ \ -5 -4 \ \ \ \ I \ I V Experimental Match Control Random Control .. .. . . ..... ,. ..... I \ I \ I \ I \ 0 2 3 4 // \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ ,, I \ I I \ I \ I \ I \/ \/ Figure 4-7 Negative H-L 6 7 8
PAGE 130
Week -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 123 Table 4-29 Negative Low Imputed Accuracy Values for Z and Z et vt Experimental Matched Control 2 et 2 vt 2 et 2 vt 0.01 0.12 -1. 21 -0.78 0.76 0.69 -1. 87* -2. 04* 2. 06:lr -2 .12* 1. 88 1.95 -1.26 -1.49 -0. 44 -0.12 0.24 0.30 0.84 1.01 0.98 0.55 0.89 0.98 0.70 0.52 1.14 1.27 0.00 -0.02 -0.24 -0.11 -1. 87* -2.23* -2.79* -2.96* 0.08 -0.08 -2.39* -2.26* -1.34 -0.93 -0.38 0.14 0.81 0.86 0.08 0.00 -0.87 -0.70 2.00 2.18 0.40 0.54 0.98 1.07 -0.17 0.07 0.35 0.89 1.85 1.45 -0.85 -0.93 0.57 0.36 -0.82 -0.83 I Significant at the 0.05 level Random Control 2 et 2 vt 0.60 0.60 1. 7 9 1.85 -0.81 -1. 20 0.00 0.06 -0.38 -0. 07 0.11 -0.07 -1.34 -1.64* 0.95 0.65 -0.12 -0.1 4 -0.98 -0. 7 5 -0.05 -0.53 -0.20 -0.29 -0.29 0.28 o. 71 0.54 1.40 1. 32 1. 56 1. 4 2 0.62 -0.0 4
PAGE 131
Week -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 124 Table 4-30 Negative Low Imputed Accurac y Trimmed Mean Values for Zv t Experimental Matched Control 0.01 -0.03 -0.02 -1.43 -1. 73* 1.00 -1.04 -0.60 -0.46 0. 7 3 0.51 0.89 0.13 1.01 -0.08 -0.36 -2.34* 2 .1 2 0.05 -1.56 -1.13 0. 2 6 o. 7 8 0. 4 3 -0.68 2 0 1 -0.13 0. 4 1 -0.60 0. 7 5 0.38 -1.0 2 -0.31 -0.89 Significant at the 0.05 level Random Control 0.52 1. 4 8 1. 03 -0.05 -0.85 0.03 1.12 0.50 0.31 -0.31 1. 08 -0. 1 5 0.08 0 7 3 0.63 0.31 -0.01
PAGE 132
125 the matched control portfolio during week t+l. These results indicate a strong industry-wide effect which continued over a two week period (tO and t+l). The results of the two-way analysis of variance test confirmed the above findings; Table 4-31 reports these results. The null hypothesis of no difference among the three portfolios could not be rejected at the O .05 level. The null hypothesis of no difference among the eleven information treatment effects was rejected at the 0.01 level. The null hypothesis of no interaction effect could not be rejected. The Duncan's Multiple Range Test indicates rejection of the null hypothesis that all thirty-three mean are equal but only at the 0.1 level. The two largest negative weekly mean returns were observed for the matched control portfolio and the experimental portfolio respectively during the forecast announcements week. Comparison Between the High-level and the Low-level Imputed Accuracy Portfolios. The same four tests used in conjunction with the two positive portfolios are repeated here for the two negative portfolios. Results for standardized prediction errors are reported in Table 4-32. Results for the distribution tests are reported in Table 4-33. The null hypothesis of no difference between the two levels of imputed accuracy portfolios could not be rejected during the forecast announcements week. The sign and magnitude of the difference between the prediction errors associated with the high-level and low-level
PAGE 133
1 2 6 Table 4-31 Negative Low Imputed Accurac y Portfolios Analysis of V ariance Test F V a l ue T y pe of portfolio effect 0. 09 Information treatment effect 2 .33 Interaction effect 0.95 Comparison of all means 1.33 PR > F 912 0 01 0 0 .5 24 0 10 6
PAGE 134
127 Table 4-32 Negative Portfolios Standardized Prediction Errors Zdvt values for the Difference Between High and Low Imputed Accuracy Portfolios Total Prediction Errors Trimmed Prediction Week Zdvt Zdvt -8 0.67 o. 77 -7 -1. 77* 1. 35 -6 0.07 0.42 -5 -1. 21 -1.03 -4 -0. 70 -0.57 -3 -1.04 -0.91 -2 2. 05* -1. 75* -1 -0.36 0.16 0 -0.62 -0.72 1 -0.66 -0.33 2 0.13 -1.36 3 -1.43 -1. 36 4 0.58 0.69 5 -1.87* -1.46 6 1.56 1. 86 7 -0. 30 0.06 8 -0.50 -0. 35 Significant at the 0.05 level Errors
PAGE 135
128 Table 4-33 Negative Portfolios Zdet and Zdvt for the Distribution Test Week Zdet Zdvt -8 0.32 0.58 -7 -2.18* 2.33* -6 0.15 -0.01 -5 -0.82 -0.66 -4 -0.57 -0. 7 8 -3 -1.11 -0.69 -2 -1.93* -1.82* -1 0.46 0.47 0 -0.01 -0.04 1 -0.91 -0. 76 2 0.53 0.06 3 -1.35 -1. 41 4 1.08 1.02 5 -1. 29 -1. 71 6 1. 28 0.92 7 -1. 92* -1. 48 8 -0.52 -0. 21 Significant at the 0.05 level
PAGE 136
129 imputed accuracy portfolios during the eight weeks preceding the forecast announcements week suggest that the former portfolio was subjected to a stronger downward stock price revision. This pattern continued during and after the forecast announcements week. To summarize, management earnings forecasts which fell short of market expectations induced a downward stock price revision during the forecast announcements week, irrespective of the level of imputed accuracy attributed to these forecasts. Prediction errors tests and the two-way analysis of variance test suggest that the observed down ward stock price revision during the forecast announcements week was statistically significant for both portfolios. The trimmed standardized prediction errors test indicates that the observed effect is a general pattern of both portfolios. The comparison between abnormal returns associated with the high-level and the low-level imputed accuracy portfolios indicate no significant difference during the forecast announcements week. However, the cumulative prediction errors prior to the forecast announce ments week were stronger for the high-level imputed accuracy port folio. Negative management earnings forecasts induce an industry-wide effect. This conclusion is equally valid for both the high-level and the the low-level imputed accuracy portfolios. The downward stock price revision of non-forecasting firms in the same, or similar, industry during the forecast announcements week is statisticall y significant and a general pattern of matched control portfolios.
PAGE 137
130 The above conclusion, with respect to the industry-wide effect of management earnings forecasts, should be tempered by the fact that the sample size utilized in this study is relatively small, A more elaborate examination of the industry-wide effect of management earnings forecasts is necessary for reaching a sound inference,
PAGE 138
CHAPTER V SUMMARY AND CONCLUSIONS The purpose of this study was to empirically examine the extent to which management forecasts of earnings convey information not already publicly available to investors. Four related research questions were examined. The methodology utilized in this study differs from those utilized in prior studies concerned with the infor mation content of management earnings forecasts in two respects. First, newness was measured by the deviation between earnings fore casted by management and earnings forecasted by analysts immediately prior to the date of the management forecast disclosure. Second, two control portfolios were utilized in this study; an industry matched control portfolio, and a random control portfolio. The industry matched control portfolio was designed to detect industry-wide effects and to assess the effect of management earnings forecasts on nonforecasting firms in the same, or similar, industry. The random control portfolio was employed to provide a bench-mark comparison and to filter economy-wide factors. The experimental sample consisted of 401 management earnings forecasts covering the period 1970 1977. The total samp l e of management earnings forecasts was classified into neutral, positive and negative forecasts according to the magnitude and the sign of the new information revealed by each forecast. Results indicate that 131
PAGE 139
132 approximately fifty percent of management earnings forecasts do not differ from investors' expectations prevailing immediately prior to the date of public disclosure of the forecast. Twenty-nine percent of management earnings forecasts exceeded investors' expectations and twenty-one percent fell short of investors' expectations. These results suggest that the majority of management earnings forecasts are a mere confirmation of investors' expectations and as such do not convey any new information. The neutral, or low-level, newness portfolio was u s ed to test the hypothesis that the forecasting action by itself induces an upward stock price revision. This hypothesis was based on the argument that confirmation of investors' expectations by management reduces the variance of expected future earnings. This in turn reduces the expected return on the firm's stock which is achieved through a higher present equilibrium value for the stock. Results indicate that an upward stock price revision did in fact occur during the forecast announcements week. Such revision, however, was neither sufficientl y strong nor was it representative enough of the general pattern of the neutral portfolio to be considered statisticall y significant. The positive news forecasts induced an upward st ock price revision during the forecast announcements week. This rev i sion was found to be a general pattern of the portfolio and statistically significant. Moreover, firms which issued posi tive news forecasts enjoyed an upward stock price revision prior to the forecast announcements week.
PAGE 140
133 The negative news forecasts induced a statistically significant downward stock price revision during the forecast announcements week. Removal of outliers indicates that the downward stock price revision was the general pattern of the portfolio. Firms which issued negative forecasts suffered a strong downward stock price revision during the weeks preceding the forecast announcements week. The above findings suggest that the sign and the magnitude of the stock price revision resulting from public disclosure of an earnings forecast by management are positively related to the level of newness of the forecast. Management earnings forecasts which mere ly confirmed investors' expectations did not result in statistically significant abnormal returns during the forecast announcements week. On the other hand, management earnings forecasts which exceeded ( fell short) of investors' expectations resulted in positive ( negative ) abnormal returns. Thus, earnings forecasts is an ef f ective technology for the deliberate revaluation of the firm by management. The fact that neutral, positive, and negative news management earnings forecasts were publicly disclosed is consistent with the iterative process of the incentive-signalling theor y which results in the disclosure of inside information irrespective of its implication to the value of the firm. Each of the positive and negative portf o lios was f urth e r subdivided int o high-level and low-level imputed accurac y classifications. Ex amination of the effect o f imp u ted accuracy w as
PAGE 141
134 accomplished by testing each of the four portfolios separately and by comparing the abnormal returns achieved on the high -le vel and the low level imputed accuracy portfolio for each classification of forecasts. Results for the positive portfolio indicate that bot h the high-level and the low-level imputed accuracy forecast s induce an upward stock price revision during the forecas t announcements week. All tests indicate that significant stock price revision is associated with the high-level imputed accuracy portfolio. Mixed results were reported for the low-level imputed accuracy portfolio. The results of prediction errors tests indicate rejection of the null hypothesis of no information content at the 0.05 level during the forecast announce ments week. The analysis of variance test could not reject the null hypothesis of no difference among the eleven information treatment effects. Furthermore, the Duncan's Multiple Range Test could not reject the hypothesis that all thirty-three means are equal at the 0. 05 level. Comparison between the magnitude of abnormal returns associated with each of the two imputed accuracy portfolios indicates that significantly larger abnormal returns are associated with the high-level imputed accuracy portfolio as compared to the low-level imputed accuracy portfolio. All four tests indicate rejection of the null hypothesis of no difference between abnormal returns asso ciat ed with each of the two portfolios at the 0 .0 5 level during the fo recast announcements week. The null hypothesis could not be rejected for any other week in the test period. These results suggest that the
PAGE 142
135 magnitude of abnormal returns induced by the public disclosure of positive news management earnings forecasts is an increasing function of the level of imputed accuracy ascribed to these forecasts. Results for the two negative portfolios indicate that both the high-level and the low-level imputed accuracy portfolios induced a downward stock price revision during the forecast announcements week. Results of prediction errors tests and the two-way analysis of variance test indicate statistically significant downward stock price revisions for both portfolios during the forecast announcements week. More importantly, however, all four comparison tests could not reject the null hypothesis of no difference between the magnitudes of abnormal returns associated with the high-level and the low-level imputed accuracy portf alias during the forecast announcements week. These results suggest that negative news management earnings forecasts are accompanied by a downward stock price revision during the forecast announcements week, irrespective of the level of imputed accurac y ascribed to these forecasts. Thus, it seems that investors react differently to positive and negative news released by management. The final hypothesis that was examined in this dissertation is the extent to which management forecasts of earnings convey industr y wide information. This hypothesis was motivated by the iterative mechanism of the incentive-signalling theory which can take p l ace through grouping of firms with similar economic attributes. It has
PAGE 143
136 been argued that a management earnings forecast is a quantitative assessment of firm-specific as well as industr y -wide fac tors. Accord ingly, management forecasts of earnings conve y information re l e van t t o establishing equilibrium prices of non-forecast ing firms in the same, or similar, industry. Results of this study indicate that the magnitude of the industry-wide effect of management earnings fo r ecasts appears to depend upon the classification of forecasts as either good or bad news forecasts. The positive forecasts induced an upward revaluation of non-forecasting firms during the forecast announcements week. This upward revaluation, however, was found to be due to the influence of a few outliers and not the general pattern of the portfolio Removal of these outliers led to the conclusion that positive news management earnings forecasts do not induce a statisticall y significant industry wide effect. The negative news forecasts seem to induce an industry wide effect. These results, however, should be caut io usl y interpre ted due to the sample size utilized in this stud y Further investigation of this research question is necessar y In closing, conclusions of this stud y should be tempered b y the fact that the sample selection criteria induce a bias towards surviving firms and firms in which analysts have exibited int e rest
PAGE 144
BIBLIOGRAPHY Abdel-Khalik, A. Rashad, "Forecast Revision, Informational Advan tages, and Implications for SEC Policy" (Accounting Research Centre, University of Florida, W.P., #1, 1980). Abdel-Khalik, A. Rashad, and Bipin Ajinkya, "Revision of Analysts' Earnings Forecasts and Stock Prices" (Accounting Research Centre, University of Florida, W.P., #7, 1979). Abdel-Khalik, A. Rashad, Robert B. Thompson and Jose Espejo, "Research on Earnings Forecasts: The State of the Art" (Accounting Research Centre, University of Florida, W.P., 118, 1977). Akerlof, George, "The Market for 'Lemons': Qualitative Uncertainty and Market Mechanism" Quarterly Journal of Economics (August 1970), pp. 488-500. Albrecht, William Steve, Crace Johnson and Larry Lookabill, "A Comparison of the Accuracy of Corporate and Security Analysts' Forecasts of Earnings: A Comment" The Accounting Review (July 1977), pp. 736-740. Ataise, Rowland Kwame, "Predisclosure Information Asymmetries, Firm Capitalization, Financial Reports, and Security Price Behavior" (Unpublished paper, University of California, Berkeley, December 1978). Ball, Ray, and Philip Brown, "An Empirical Evaluation of Accounting Income Numbers" Journal of Accounting Research (Autumn 1968), pp 159-178. Basi, Bart A., "A Comparison of the Accuracy of Corporate and Security Analysts' Forecasts of Earnings: A Reply" The Accounting Review (July 1977), pp. 741-745. 137
PAGE 145
138 Basi, Bart A., Kenneth A. Carey and Richard D. Twark, "A Comparison of the Accuracy of Corporate and Security Analysts' Forecasts of Earnings" The Accounting Review (April 1976), pp. 224-254. Beaver, William, "The Information Content of Annual Earnings Announcements" Empirical Research in Accounting: Selected Studies, supplement to Volume 6, Journal of Accounting Research (1968), pp. 67-92. Beaver, William, "The Time Series Behavior of Earnings" Empirical Research in Accounting: Selected Studies, supplement to Volume 8, Journal of Accounting Research (1970), pp. 62-94. Brown, Lawrence D., and M.S. Rozeff, "The Superiority of Analyst Forecasts as Measures of Expectations Evidence From Earnings" Journal of Finance (March 1978), pp. 1-15. Copeland, Ronald, and Robert Marioni, "Executives' Forecasts of Earnings per Share Versus Forecasts of Naive Models" Journal of Business (October 1972), pp. 497-512. Cragg, J.G., and Burton Malkiel, "The Consensus and Accuracy of Some Predictions of the Growth of Corporate Earnings" Journal of Finance (March 1968), pp. 67-84. Daily, R. Austin, "The Feasibility of Reporting Forecasted Information" The Accounting Review (O ctober 1971), pp. 686-692. Fama, Eugene, "Efficient Capital Markets: A Review of Theory and Empirical Work" Journal of Finance (May 1970), pp. 383-417. Foster, George, "Stock Market Reaction to Estimates of Earnings per Share by Company Officials" Journal of Accounting Research (Spring, 1973), pp. 25-37.
PAGE 146
139 Gonedes, N.J., "Capital Market Equilibrium and Annual Accounting Numbers: Empirical Evidence" Journal of Accounting Research (Spring 1974), pp. 26-62. Gonedes, N.J., "Risk, Information and the Effect of Special Accounting Items on Capital Market Equilibrium" Jounal of Accounting Research (Autumn 1975), pp. 220-256. Gonedes, N.J., and N. Dopuch, "Capital Market Equilibrium, Information Production and Selected Accounting Techniques: Theoretical Framework and Review of Empirical Work" Studies on Financial Accounting Objectives 1974. Supplement to Volume 12, Journal of Accounting Research, pp. 48-169. Gonedes, N. J. N. Dopuch and S. H. Penman, "Disclosure Rules, Infor mation-Production, and Capital Market Equilibrium: The Case of Forecast Disclosure Rules" Journal of Accounting Research (Spring 1976), pp. 89-137. Good, Walter, "Interpreting Analysts' Recommendations" Financial Analysts Journal (May/June 1975), pp. 42-48. Green, David, and Joel Segal, "The Predictive Power of FirstQ uart~r Earnings Reports" Journal of Business (January 1967), pp. 44 -55. Griffin, Paul A., "Common Stock Price Adjustments to Rating Changes: A Methodological Emphasis" (Graduate School of Business, Stanford University, Research Paper #439, April 197 7 ). Hakansson, Nils H., "Interim Disclosure and Public Forecasts: A n Economic Analysis and a Framework for Choice" Th e Accounting Review (April 1977), pp. 396-415. Hays, William, Statistics For the Social Sciences New York: Holt Rinehart and Winston, 1973
PAGE 147
140 Hopkins, Wayne A., "Publicly Disclosed Corporate Profit Forecasting and the Effect of Auditor Review: An Empirical Study" ( Faculty of Administrative Studies, University of Manitoba, Canada, W.P. December 5, 1977). Jaggi, Bikki, "A Note on the Information Content of Corporate Annual Earnings Forecasts" The Accounting Review ( October 1978 ) pp. 961-96 7. Kirk, Roger, Experimental Design: Procedures For the Behavioral Sciences. Belmont, California: Brooks / Cole, 1972. Lintner, John, "The Valuation of Risk Assets and the Selection of Risky Investments in Stock Portfolio and Capital Budgets" Review of Economics and Statistics (February 1965), pp. 13-37. Lorek, Kenneth, Charles McDonald and Dennis Patz, "A Comparative Examination of Management Forecasts and Box-Jenkins Forecasts of Earnings" The Accounting Review (April 1976), pp. 321-330. May, Robert, "The Influence of Quarterly Earnings Announcements on Investor Decisions as Reflected in Connnon Stock Price Changes" Empirical Research in Accounting: Selected Studies, Supplement to Volume 9, Journal of Accounting Research (1971), pp. 119-163. McDonald, Charles "An Empirical Examination of the Reliability of Published Predictions of Future Earnings" The Accounting Review (July 1973), pp. 502-510. Nichols, Donald, and Jeffrey Tsay, "Security Price Reaction to Long Range Executive Earnings Forecasts" Journal of Accounting Research (Spring 1979), pp. 140-155. Patell, James, "Corporate Forecasts of Earnings per Share and Stock Price Behavior: Empirical Tests" Journal of Accounting Research (Autumn 1976), pp. 246-276.
PAGE 148
141 Penman, Stephen, "Voluntary Disclosure of Corporate Earnings Fore casts" (Unpublished paper, University of California, Berkeley, September 1978). Reilly, F., D. Morgenson and M. West, "The Predictive Ability of Alternative Parts of Interim Financial Statements" Empirical Research in Accounting: Selected Studies, Supplement to Volume 10, (1972) Journal of Accounting Research, pp. 105-124. Riley, John, "Competitive Signalling" Journal of Economic Theory (June 1975), pp. 175-186. Riley, John, "Informational Equilibriunt 1 ~ Rank Corporation, No. R-2059-NSF, Rand Corporation, Santa Monica, California, 1977. Ross, Stephen, "Disclosure Regulations in Financial Markets: Implications of Modern Finance Theory and Signalling Theory" (University of Pennsylvania, Discussion paper #3, April 1977). Sharpe, William, "Capital Assets Prices: A Theory of Market Equil ibrium under Conditions of Risk" Journal of Finance (September 1964), pp. 425-442. Spence, Michael, "Job Market Signalling" Quarterly Journal of Economics (August 1973), pp. 355-374. Spence, Michael, "Competitive and Optimal Responses to Signals: An Analysis of Efficiency and Distribution" Journal of Economic Theory (March 1974), pp. 296-332. Sunder, Shyam, "Relationship Between Accounting Changes and Stock Prices: Problems of Measurement and Some Empirical Evidence" Empirical Research in Accounting: Selected Studies, Supplement to Volume 11, Journal of Accounting Research (1973), pp. 1-45.
PAGE 149
BIOGRAPHICAL SKETCH Robert L. Beshara was born March 30, 1944, in Cairo, Egypt. He graduated from Hilipolis High School, Cairo, Egypt, in 1960. Mr. Beshara received a Bachelor of Commerce with a major in accounting from Cairo University in 1964. He received a Master of Science degree from the University of Saskatchewan in 1969. He entered the Ph .D. program in business administration at the University of Florida in 1976. Mr. Beshara received his R.I.A. designation (Saskatchewan, Canada) in 1975. He taught accounting at the University of Regina and Carleton University since 1970. -142
PAGE 150
I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate in scope and quality as a dissertation for the degree of Doctor of Philosophy A. Rashad Abdel-khalik Chairman of Dissertation Co mm ittee Graduate Research Professor of Accounting I certify tha t I have r ea d this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate in scope and quality, as a dissertation for the degree of Doctor of Philosophy. Bipin B Ajinkya Assistant Professo r of Accounting I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality as a dissertation for the degree of Doctor of Ph ilosophy. Rober t C. Radcl iff e Associate Professor This di s sertat ion was submitted to the Graduate Faculty of the School of Accounting in the Coll ege of Business Administration a nd to the Graduate Council, and was accepte d as partial fulfillment of the requir e ments for the degree of Doctor of Philosophy. Decer:ibe r 1981 Dean for Graduate Studies a~d rtese arch
PAGE 151
111 11 1 11 11 ~ rn ~ 1 i 1 ~ 1 11 1' 11 11 1 1 !1 i1 111 1 11~ 11 '1 ~ 1~ 1 111 1 111 1 3 1262 08554 0622
xml version 1.0 encoding UTF-8
REPORT xmlns http:www.fcla.edudlsmddaitss xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.fcla.edudlsmddaitssdaitssReport.xsd
INGEST IEID EWUXV6DMU_D9M70B INGEST_TIME 2011-10-24T16:22:59Z PACKAGE AA00003865_00001
AGREEMENT_INFO ACCOUNT UF PROJECT UFDC
FILES
|
|