.5 it: D1:

 

. ‘ .17 :t...
, an.“ ..\. i
. ‘ y

a. .: r3. .
. .1! 3‘3 .15....
Vv!...§.§
t..5...£.
- 3.1.;
L

. \ u . a l V Ally ‘v OI.
Shufkh‘w ..n<2 , I 5.3. u . , iii: 2.9.3.? ..
.V in») H4 ‘ ‘ .H ., : .

 

 

 

 

I. ll:

 

 

cm H e l ‘66 5 ‘mlll?ll'AT’flli/lfillf/ll”Hill/7M I

3 1293 0062 08098

 

0 LIBRARY
Michigan Stat.

.L University

 

 

This is to certify that the

dissertation entitled

M‘Vk‘f KC$I¢M£¢ To Enfnoz); AKHOMnCCMLkflf‘:

TL»: ngcc-(s g; F’;M chfiffi‘fcv:¢f‘o‘3 0'

presentedby
[<1 Choc”) H6”

has been accepted towards fulfillment
of the requirements for

P’s? degree in F0.n4nce

7(a)" C—flufln

Majorcsprofs

Date Decoy—.1." 7" H89

 

MS U is an Affirmative Action/Equal Opportunity Institution 0-12771

 

 

 

 

PLACE IN RETURN BOX to remove this checkout from your record.
TO AVOID FINES return on or before duo due.

 

DATE DUE DATE DUE DATE DUE

J
|
l—HT

. r

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

[: ____
~ [fir—V7

MSU Is An Affirmdive Action/Equal Opportunity Institution

 

MARKET RESPONSE TO EARNINGS ANNOUNCEMENTS:

TEE EPEECTS OF FIRE CHARACTERISTICS

BY
Ki Choong Han

A DISSERTATION

Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY

Department of Finanoe and Insurance

1990

{in
.u(

I l
I l

i

@035

ABSTRACT

MARKET RESPONSE TO EARNINGS ANNOUNCEMENTS:
THE EFFECTS OF FIRM CHARACTERISTICS

BY
Ki Choong Ban

This study examines factors associated with stock price
response to earnings announcements. The study is developed
in two parts. First, a theoretical model of the firm is
developed to identify firm-specific determinants of the market
response to earnings announcements. The model allows for
differential market response depending upon the past and
expected future investment characteristics and performance of
the firm. The model contributes to an understanding of the
information which is important to the market and of the
transmission process by which earnings information affects
stock prices. Secondly, the factors identified in the
theoretical model are empirically examined. The empirical
analysis is extended to compare the relative importance of a
set of response factors identified in the theoretical and
empirical literature. Results show that dividend policy is
an important factor associated with stock price response to

earnings announcements.

dedicated to my mother, Lee Myong-Ja,
and my father, Han Seok-Woo

iii

ACKNOWLEDGEMENTS

First of all, I praise the Lord for His wonderful

guidance. He is my great shepherd.

My debts of gratitude to people who have helped me with
my dissertation are enormous. ‘First, I give my special thanks
to my dissertation chairman, Dr Butler. He has been so
patient and generous that he has read through each chapter a
number of times and offered extremely valuable suggestions and
criticisms. And I deeply appreciate Dr. Domian and Dr.
Moser's helpful comments. Also, I would like to thank Dr.
O'Donnell for his careful and sufficient support in many

respects.

I am also grateful to my sister for her endearing

encouragement, which she gave me whenever I needed it.

Particular thanks are due to my wife for the amount of
time that she gave up to let me study during the doctoral

program.

Finally, I would like to remember my parents, who
sacrificed and.devoted.a substantial portion of their life to
take care of me. 'Their love was of inestimable importance to

my life and gave birth to a lot of unforgettable memories.

iv

TABLE OF CONTENTS
List of Tables .............................. ............ vii
List of Figures .................... ... .................... x
Chapter 1. Introduction ..................................1

Chapter 2. Market Response to Unexpected Earnings ........ 6
2.1. The Market's Earnings Expectation ................7

l.a. Time-series Models ............. .......... .....7
l.b. Analysts' Forecasts ...........................9
1.c. Analyst Superiority

Relative to Time-series Models ........9

2.
2.
2

2.2. Information Content of Unexpected Earnings . ...... 12

2.2.a. Unexpected Annual Earnings ....................12
2.2.b. UnexpecteduQuarterly Earnings .................13

2.3. Cross-sectional Differences ............... ....... 15
2.4. Firm Characteristics as Explanatory Factors ......17
4a. FimSize......OOOOOOOOOOOOOOO...0.... ..... .0017
4.b. Earnings Predictability .......................23
4 c
4 d

. SystematicRisk 26
. Systematic Risk and Earnings Growth ...........27

Chapter 3. A Model of Market Response
to Unexpected Earnings . . . . . ...... . . 31
3.1. Information Content of Unexpected Earnings .......31
an Rational.Expectations .........................3l

1.
l.b. The Information Content
of an Earnings Announcement . . . . 33

3.
3.

3.2. Share Price Response in a Two-Period Model
underUncertainty .............35

3.2.a. Assumptions..................... .............. 35

3.2.b. Investment under Uncertainty ....... 37

3OZOCF1rmvaluatlon OOOOOOOOOOOOOOOOOOOOOOOOOOOOO00.39
3.2.d. Association between Earnings Surprise

and Stock Return .......... 40

3.3. Multi-period Model ................... 46

Chapter 4. Empirical Design ....... 49

4.1. Empirical Specification ........ ..49

4.2. Empirical Proxies and the Data Sample ............54
4.3. Model Specification and Parameterization .........S9

4 . 4 . Comparative Statics of the Empirical Model . . . . . . . 65

Chapter 5 . Empirical Results from the Theoretical Model . . 67
5.1. Entire Sample Results 67

5.2. Separate Analyses on Positive

and Negative Surprises ......... 74
Chapter 6. An Extension to Other Factors . ...... . ....... . .96
6. 1. Dividends versus Firm Size . . . . . . . . . . ..... . ...... .96

6 . 2 . Dividends versus Earnings Predictability . . . . . . . . . 113

6.3. DividendsversusGrowth..........................125

Chapter7. SummaryandConclusions 133

7. 1. Summary of the Empirical Results . . . . . . . . . . ..... . . 133
7.2. Conclusions .......... ............................137
Bibliography ......... ......... ................. ....... ....139

vi

Table
Table

Table

Table

Table

Table

Table

Table

Table

Table

Table

Table

Table

Table

Table

10

11

12

13

14

15

LIST OF TABLES

Summary statistics: 4972 firm-quarters ...........71
Effect of dividend, q ratio, and beta on the
market response to unexpected earnings:
4972 firm-quarters from 1983.4-1986.2 ............72
Effect of dividend, q ratio, and beta on the
market response to unexpected earnings:
OLS estimation with 4972 firm-quarters
froml983.4-l986.2 ...............................73
Descriptive statistics: Positive earnings
surprises (2126 firm-quarters) . . . . . . . . . . . . . . . . . . .84
Descriptive statistics: Negative earnings
surprises (2633 firm-quarters) . . . . . . . . . . . . . . . . . . .85
Market response to positive unexpected earnings:
2126 firm-quarters from 1983.4-1986.2
(Dependent variable: Raw'return) .................86
Market response to positive unexpected earnings:
2126 firm-quarters from 1983.4-1986.2
(Dependent variable:.Abnormal return) ............88
Market response to negative unexpected earnings:'
2633 firm-quarters from 1983.4—1986.2
(Dependent variable:IRaw'return) ............ ..... 90
Market response to negative unexpected earnings:
2633 firm-quarters from 1983.4-1986.2
(Dependent variable: Abnormal return) ............92
Pearson correlation coefficients:
2126 firm-quarters from 1983.4-1986.2
with positive earnings surprises . . . . . . . . . . . .
Pearson correlation coefficients:
2633 firm—quarters from 1983.4-1986.2
with negative earnings surprises . . . . . . . . . . . . . . . .95
Effect of dividend and firm size on the market
response to positive unexpected earnings:
2126 firm-quarters from l983.4-1986.2
(Dependent variable: Raw return) . . . . . . . . . . . . . . . . 103
Effect of dividend and firm size on the market
response to positive unexpected earnings:
2126 firm-quarters from 1983.4-1986.2
(Dependent variable: Abnormal return) . . . . . . . . . . . 104
Effect of dividend and firm size on the market
response to negative unexpected earnings:
2633 firm-quarters from 1983.4-1986.2
(Dependent variable: Raw return) . . . . . . . . . . . . . . . . 105
Effect of dividend and firm size on the market
response to negative unexpected earnings:
2633 firm-quarters from 1983.4-1986.2
(Dependent variable: Abnormal return) . . . . . . . . . . . 106

000.94

vii

Table

Table

Table

Table

Table

Table

Table

Table

Table

Table

Table

Table

Table

Table

17

18

19

20

21

22

23

24

25

26

27

28

29

Summary statistics: 830 firm-quarters

(All unexpected earnings are positive.) . . ....... 107
Summary statistics: 1001 firm-quarters

(All unexpected earnings are negative. ) . . . . . . . . . 108
Effect of dividend and firm size on the market
response to positive unexpected earnings:

830 firm-quarters from 1983.4-1986.2

(Dependent variable: Raw return) . . . . . . . . . . . . . . . . 109
Effect of dividend and firm size on the market
response to positive unexpected earnings:

830 firm-quarters from 1983.4-1986.2

(Dependent variable: Abnormal return) . . . . . . . . . . . 110
Effect of dividend and firm size on the market
response to negative unexpected earnings:

1001 firm-quarters from l983.4-1986.2

(Dependent variable: Raw return) . . . . . . . . . . . . . . . . 111
Effect of dividend and firm size on the market
response to negative unexpected earnings:
1001 firm-quarters from 1983.4-1986.2
(Dependent variable: Abnormal return) . . . . . . .
Summary statistics: 1879 firm-quarters
(All unexpected earnings are positive.) . . . . . . . . . 119
Summary statistics: 2323 firm-quarters

(All unexpected earnings are negative. ) . . . . . . . . . 120
Effect of dividend and earnings predictability on
the market response to positive unexpected
earnings: 1879 firm-quarters from 1983.4-1986.2
(Dependent variable: Raw return) . . . . . . . . . . . . . . . . 121
Effect of dividend and earnings predictability on
the market response to positive unexpected
earnings: 1879 firm-quarters from l983.4-1986.2
(Dependent variable: Abnormal return) . . . . . . . . . . . 122
Effect of dividend and earnings predictability on
the market response to negative unexpected
earnings: 2323 firm-quarters from 1983.4-1986.2.
(Dependent variable: Raw return) . . . . . . . . . . . . . . . . 123
Effect of dividend and earnings predictability on
the market response to negative unexpected
earnings: 2323 firm-quarters from 1983.4-1986.2
(Dependent variable: Abnormal return) . . . . . . . . . . . 124
Effect of dividend and growth on the market
response to positive unexpected earnings:

1879 firm-quarters from 1983.4-1986.2

(Dependent variable: Raw return) . . . . . . . . . . . . . . . . 129
Effect of dividend and growth on the market
response to positive unexpected earnings:

1879 firm-quarters from 1983.4-1986.2

(Dependent variable: Raw return) . . . . . . . . . . . . . . . . 130

....112

viii

Table 30

Table 31

Effect of dividend and growth on the market
response to negative unexpected earnings:

2323 firm-quarters from l983.4-1986.2

(Dependent variable: Raw return) ..........u....131
Effect of dividend and growth on the market
response to negative unexpected earnings:

2323 firm-quarters from 1983.4-1986.2

(Dependent variable: Raw return) . . . . . . . . . . . . . . . .132

ix

Figure
Figure
Figure

Figure
Figure

Figure
Figure
Figure

Figure

Figure

Figure
Figure
Figure
Figure

Figure

1.
2.
3.

4.
5.

10.

11.

12.

13.

14.

15.

LIST OF FIGURES

Positive unexpected earnings ...................38
Negative unexpected earnings ...................38
Predicted relationship between a factor and
shareprice....................................43
Comparison of positive and negative groups .....75
Contribution of each of dividend, q ratio, and
beta: Positivetearnings surprises ..............78
Contribution of each of dividend, g ratio, and
beta: Negative earnings surprises ..............81
Contribution of each of dividend and
firm size: Positive earnings surprises .. ....... 98
Contribution of each of dividend and
firm size: Negative earnings surprises .........99
Contribution of each of dividend and
firm size: Positive earnings surprises
(based on the reduced sample) .................101
Contribution of each of dividend and
firm size: Negative earnings surprises
(based on the reduced sample) ................102
Contribution of each of dividend and earnings
predictability: Positivetearnings surprises ..116
Contribution of each of dividend and earnings
predictability: Negativelearnings surprises ..117
Contribution of each of dividend and
growth: Positivetearnings surprises ..........126
Contribution of each of dividend and
growth: Negativetearnings surprises ..........127
Summary of results ..................... ..... .136

Chapter 1. Introduction

This study examines factors associated with stock price
response to earnings announcements. The study is developed
in two parts. First, a theoretical model of the firm is
developed to identify firm-specific determinants of the market
response to earnings announcement. The model allows for
differential market response depending upon the past and
expected future investment characteristics and performance of
the firm. The model contributes to an understanding of the
information which is important to the market and of the
transmission process by which earnings information affects
stock prices. Secondly, the factors identified in the
theoretical model are empirically examined. The empirical
analysis is extended to compare the relative importance of a
set of response factors identified in the theoretical and

empirical literature.

Studies of the information content of earnings
announcements have a long tradition in accounting and finance
research. Early studies (e.g. Ball and Brown [1968], Beaver
[1968]) determined that the sign and magnitude of stock price
responses are positively correlated with the sign and
magnitude of unexpected earnings. Subsequent articles
document cross-sectional differences in market response to
earnings announcements (Joy, Litzenberger, and McEnally
[1977], Beaver, Clarke, and Wright [1979], and Grant [1980]).

1

2
More recent studies have focused on the specific factors
associated with stock price response including firm size
(Atiase [1985]), earnings predictability (Pincus [1983]),
systematic risk (Collins and Kothari [1989]), and growth
(Collins and Kothari [1989]). However, these studies have
typically examined only one or two components of market
response and have not examined the relative importance of the
various determinants of a firm's stock price response to
earnings information. This study provides a comprehensive
analysis of the factors affecting the market reaction to

unexpected earnings.

Using a rational expectations framework, Chapter 3
develops a two-period model of the firm similar to that
proposed by Miller and Rock [1985]. This theoretical model
develops the economic determinants behind firm-specific
differences in market response to earnings announcements. In
the model, the firm bases its investment decisions on the
intersection of its investment opportunity schedule with its
cost of capital schedule. The firm continues to invest until
the expected marginal return on capital is equal to its
marginal cost of capital. Because of uncertainty, the firm
cannot guarantee that the chosen level of investment is
optimal. When actual investment outcomes deviate from the
expectations, the firm reflects this information in its

investment decision in the following period.

3
The unexpected investment outcome is realized as
unexpected earnings. If the unexpected earnings are positive,
it implies that the investment level was below the optimal
level. Consequently, the firm will adjust its next period
investment upward expecting identical marginal return and cost
and, therefore, higher expected.net return in the next period.
Since the market is aware of this, the share price of the firm
should increase with positive unexpected earnings. 0n the
other hand, if unexpected earnings are negative, the share

price of the firm should decrease.

This two-period model is then extended into a multi-
period setting within the rational expectations framework.
Both the two-period model and the multi-period model suggest
that market response to unexpected earnings is a function of

Tobin's q ratio, cost of capital, and dividend policy.

The theoretical model is translated into an empirical
regression equation in Chapter 4. The regression results of
Chapter 5 are then extended in Chapter 6 to include factors
identified by the empirical literature as being important but
which are not implied by the theoretical model.

The theoretical and empirical results of this study
indicate that dividend policy is a key determinant of market
response to an earnings announcement. Unexpected earnings and
dividend policy together provide useful information to the

market concerning management's inside information about

4

unrealized investment opportunities.

This study provides evidence which is contrary to the
empirical findings of some previous research. First, the
results show that market beta does not impact the
return/surprise relation. This is inconsistent with Collins
and Kothari [1989]. Second, the results indicate that
earnings predictability is related to market reaction but that
its relational direction is opposite that suggested by Pincus
[1983]. Finally, earnings growth turns out to be an
unimportant factor in the return/surprise relationship which

is inconsistent with Collins and Kothari [1989].

More importantly, the equity market reaction is
asymmetric between positive and negative earnings surprises.

The response is stronger to positive earnings surprise.

This study is organized as follows. In Chapter 2,
previous theoretical and empirical studies of market reactions
to earnings announcements are reviewed with an emphasis on
recent studies associating firm characteristics with market
response. Chapter 3 develops a two-period model within a
rational expectations framework. The two-period model is then
extended into a multi-period model. In Chapter 4, the
theoretical model is translated into an empirical
specification. Chapter 4 also identifies the sample used in
this study. Empirical results are reported in Chapters 5 and

6. Chapter 7 summarizes the empirical findings and discusses

5

some of the implications for past and future research.

Chapter 2. Market Response to Unexpected Earnings

A fundamental issue at the interface of finance and
accounting involves the relationship between a firm's reported
earnings and its stock returns. One question that has
received considerable attention is whether reported earnings
contain information used by the market in assessing the value
of a firm's common stock. Since the seminal work of Ball and
Brown [1968], numerous studies have addressed this question
by examining the contemporaneous relationship between stock
returns and unexpected earnings. The results of this large
body of research provide convincing empirical support for the
information content of accounting earnings. A number of
recent studies provide evidence that the market's response to
unexpected earnings varies across firms. As an attempt to
understand these cross-sectional differences, several recent
studies turn to firm characteristics such as firm size,

earnings predictability, systematic risk, and growth.

Share price reaction to earnings announcements may be
classified into contemporaneous, leading, and lagging
responses. Leading and lagging price responses are important
from a market efficiency perspective and in assessing the
value of accounting earnings. This study focuses on the
contemporaneous response of price to earnings announcements.
While the perspective of this study is consistent with market
efficiency, the empirical models of Chapters 4 through 6

6

7
require no assumption of efficiency. Furthermore, this study

focuses on share price response to earnings surprises.
2.1. The Market's Earnings Expectation

The theoretical and empirical literature focuses on the
relationship between unexpected earnings and share price.
Unexpected earnings is defined as actual earnings minus the
market's expectation of earnings. Examination of unexpected
earnings requires some proxy for the market's expectation of
earnings. Unfortunately, there exists no underlying theory
for the specification of market expectation of earnings. Many
early studies employ expectational models based on the past
time-series behavior of earnings. More recent studies employ
financial analysts' earnings forecasts which have been shown

to be superior to time-series expectational models.
2.1.a. Time-series Models

Since little established theory guides the selection of
an earnings expectations model, many researchers use a set of
time-series models. Time-series models which have been
applied to earnings expectations can be broadly categorized

into univariate time-series models and index models.

Within the class of univariate time-series models, Box-
Jenkins [1976] models are highly regarded for their ability
to make the most efficient use of time series data. A general

Box-Jenkins model has the following form:

8

EM (xt) . we," xv.” . . . .) ,
where
Ep,(.) - forecast of (.) made at time t-l,
Xt - realized earnings at time t.
The Box-Jenkins modelling technique enables one to select the
most appropriate time series model consistent with the process
generating each firm's time series of earnings. The Box-
Jenkins model, by not making 'a priori' assumptions about the
process generating earnings, subsumes autoregressive, moving
average and mixed models as special cases of the general
model. There are a number of other time-series models such
as random walk models, martingales, and seasonal martingales
which are special cases of the general Box-Jenkins model (see

Brown and Rozeff [1978]).1

An index model has the following general form:

3:409) ' xt-i + at + ”Jr-10‘s: ‘ x-M)!

where the indexitqIt is usually defined as the market's average
earnings. The use of this model is motivated by the
relationship that is found between first-differenced
individual company earnings and a first-differenced economy-
wide index of earnings such as average earnings across all

firms. By applying OLS, coefficients at and bt are obtained.

 

' Various forms of univariate time-series models are found
in Niger [1974], Elton and Gruber [1972], Joy, Litzenberger,
and McEnally [1977], Beaver, Clarke, and Wright [1979], and
Brown et al. [1987a].

9
Since Ball and Brown's [1968] seminal investigation of the
information content of earnings, a number of studies have used
this model (Beaver [1968], Beaver, Clarke, and Wright [1979],

and Fried and Givoly [1982]).
2.1.b. Analysts' Forecasts

Information content studies using analysts' forecasts as
a proxy for the market's earnings expectation rely on the
assumption that market participants use analysts' earnings
forecasts in stock valuation models. A considerable body of
circumstantial evidence suggests that this is the case (e.g.
Givoly and Lakonishok:[r984]). Annual and/or quarterly
earnings forecasts are provided to the market by major
brokerage houses and other financial service firms. Sources
of analysts' forecasts used by the investment community
include the Value Line Investment Survey, Standard and Poor's
Earnings Forecaster, and the Institutional Brokers Estimate

System (I/B/E/S) of Lynch, Jones and Ryan.2

2.1.c. Analyst Superiority Relative to Time-series Models

 

2 Studies employing these sources of analyst forecasts
include:

Value Line - Brown and Rozeff [1978], Brown et al.
[1987a], Brown, Richardson, and Schwager
[1987], and Brown et al. [1987b],
3 a P - Givoly and Lakonishok [1979], and Fried and
Givoly [1982],
I/B/E/S - Elton, Gruber, and Gultekin [1984], Brown,
Richardson, and Schwager [1987], and.0'Brien
[1988].

10

The selection of a time-series model as a surrogate for
market expectations is impaired by the underlying assumptions
that earnings generating processes are stationary and that the
model characteristics are applicable to all firms. Recent
empirical evidence (e.g. Brown, Richardson, and Schwager
[1987]) indicates that analysts' earnings forecasts are a
better surrogate for the market expectation of earnings than

time-series models.

Intuitively, it can be argued that analysts' forecasts
have an edge over time-series models because analysts utilize
all publicly available information while time-series models
rely exclusively on past earnings. Analysts use a broader
information set which includes non-accounting information on

the firm, its industry, and the general economy.

Empirically, however, early comparisons of analysts'
forecasts to time-series models conclude that analysts'
forecasts are not more accurate than time-series forecasts.
Cragg and Malkiel [1968] compare five-year earnings growth
rates forecast by five investment houses with two sets of
naive models, one predicting no change and the other a change
equal to past change. The results indicate that forecasts
based on the naive models do not perform better or worse than
the analysts' predictions. Elton and Gruber [1972], comparing
time-series models to annual earnings forecasts made by

analysts in a large pension fund, an investment advisory

11
service and a big brokerage house, reach a similar conclusion.
But the period over which the performance of the competing
models is compared is very short for both of these studies.
Later studies employing more refined techniques and longer
test periods find.that.analysts' forecasts.are superior to the
prediction of time-series models. Brown and Rozeff [1978] and
Collins and Hopwood [1980] compare the performance of Value
Line forecasts for up to five quarters ahead with forecasts
made by fairly sophisticated time series models and find that
the Value Line forecasts are more accurate. Collins and
Hopwood [1980] find that Value Line predictions produce fewer
extreme errors. Using Standard and Poor's Earnings
Forecaster, Fried and Givoly [1982] compare the accuracy of
analysts' annual earnings per share estimates with a
univariate time-series model and an index model. Results show
that the analysts' forecasts are more accurate, on average,
than the timeseries models. The analysts' forecasts had a
mean relative absolute error over the test period of 16.4 per
cent. This was significantly lower than the mean error of
either the univariate time-series model (19.3 per cent) or the
index model (20.3 per cent). In more recent studies, Brown
et al. [1987a] (value Line), and Brown, Richardson, and
Schwager [1987] (Value Line and I/B/E/S) reach similar

conclusions.

12

2.2. Information Content of Unexpected Earnings
2.2.a. Unexpected Annual Earnings

The Ball and Brown [1968] study provides the first
comprehensive evidence of the adjustment of stock prices to
earnings announcements. For a sample of 261 COMPUSTAT firms
over the period 1946 through 1968, each annual earnings
announcement is classified either as favorable or as
unfavorable using an index model. Abnormal risk-adjusted
monthly rates of return for these two groups are examined.3
The major finding is that positive (negative) earnings
forecast errors are associated with positive (negative)
abnormal returns. The information content of unexpected
earnings is supported by Beaver [1968] in a different way.
Using abnormal weekly returns of 143 firms over the period
of 1961-1965, he shows that the magnitude of stock returns
during the weeks in which an earnings announcement is made is

much larger than those during the nonreport period. However,

 

3 Stock price movements are measured by the abnormal monthly
returns where the expected return is defined according to the
market model,

3(Rn) ' 6.4'ihfiu:

where Rn denotes the return of security i for period t, a, and
31 are parameters and R“ is the actual market rate of return
for period t. Menthly abnormal returns are measured by the
difference

“It ' Rt: ' EUR"),

where a, and B, are estimated by the OLS.

13
neither Ball and Brown [1968] nor Beaver [1968] provide
insight into whether abnormal returns vary with the magnitude

of the forecast error.

While Ball and Brown [1968] investigate the association
between the sign of the earnings forecast errors and abnormal
stock returns, Beaver, Clarke, and Wright [1979] also consider
the magnitude of the forecast error. Their sample of earnings
announcement-stock return pairs is categorized into 25
nonoverlapping groups based on the size of forecast errors
obtained by an index model and a martingale model with drift.
Significance tests are conducted based on an ordinal (rank-
order) relationship between monthly returns and forecast
errors. The magnitude of stock.price response to an earnings
announcement is significantly associated with the magnitude

of the earnings surprise.
2.2.b. Unexpected Quarterly Earnings

A study by May [1971] assesses whether quarterly earnings
announcements have a significant impact on market prices. His
results indicate that the magnitude of price changes in
announcement weeks is greater than the magnitude of price
changes in nonannouncement weeks, and support Beaver's [1968]
study. May [1971] does not attempt to distinguish between
favorable and unfavorable earnings reports. Niger [1974] also
presents evidence that there are substantial price reactions

to quarterly earnings announcements for a sample of 87 firms

14
(1966-69).

Joy, Litzenberger, and McEnally [1977] further develop
the association between the sign and magnitude of stock
returns and those of earnings forecast errors. In their
study, martingale models are used to categorize observed
earnings announcements as favorable, neutral, or unfavorable.
For each earnings announcement, weekly rates of return are
observed, ‘Unanticipated quarterly earnings announcements are
shown to have a statistically significant association with
abnormal price changes over the subsequent twenty weeks. More
importantly, the percentage deviation of reported earnings
from the expectation is shown to have a significant
association with the magnitude of the subsequent price

adjustment.

15

2.3 Cross-sectional Differences

Ball and Brown [1968] find that, while on average the
sign of earnings forecast errors is associated with the sign
of abnormal stock returns, not all firms that have positive
(negative) earnings forecast errors have positive (negative)
abnormal returns. Joy, Litzenberger, and McEnally [1977]
suggest that differences in magnitudes as well as signs of
earnings forecast errors are associated with differences in
abnormal stock returns. But stock returns behavior around
earnings reports is not fully explained by earnings forecast

errors e

Grant [1980] assesses the differences in the information
content of annual earnings announcements between a sample of
OTC firms and a sample of NYSE firms based on the notion that
the amount of interim information available on OTC firms may
be systematically less than that available on NYSE firms. He
argues that OTC investors rely more heavily on the earnings
announcement as a source of information for decision making.
Following Beaver's methodology, Grant compares the price
response to earnings announcements of OTC and NYSE firms.
Results reveal that the annual earnings announcements of OTC
firms possess more information content, as measured by share
price response, than those of NYSE firms. His findings
support the argument that differences in prior knowledge about

firms lead to cross-sectional differences in stock market

16

behavior associated with earnings announcements.

More recently, Kormendi and Lipe [1987] contend that
differences in time-series properties of accounting earnings
result in cross-sectionally different reactions to earnings
reports. To empirically estimate how stock.prices respond to
earnings changes, Kormendi and.Lipe [1987] adopt the following

model for each of 145 firms (1947-80):

Rn ' “M + an*(an/Pu4) T ‘nr
where
Rn - abnormal annual return for firm i and period t
Ux,t a unexpected annual earnings for firm i and period t
P",1 - price of firm i's stock at time t-l.
Kormendi and Lipe [1987] estimate the above equation for each

firm. Results show that a1i (market response to unexpected

earnings) varies across firms.

17

2.4. Firm Characteristics as Explanatory Factors

Several recent studies examine firm characteristics in
an attempt to understand the cross-sectional differences in
share price response to unexpected earnings. Firm
characteristics which have been identified as influencing
share price response include firm size (Atiase [1985] and
Freeman [1987]), earnings predictability (Pincus [1983]),
systematic risk (Collins and Kothari [1989]), and earnings

growth (Collins and Kothari [1989]).
2.4.a. Firm Size

The differential information hypothesis (e.g. Freeman
[1987]) suggests that differences in prior knowledge about
firms, due to differential availability of information
relevant for the assessment of share value, are one reason to
expect cross-sectional differences in stock market behavior
associated with earnings announcements. An earnings
announcement for a less widely-followed firm is likely to
provide proportionally more information than an earnings
announcement for a widely-followed firm. Under this
hypothesis, an earnings announcement results in more stock
price response for the less widely-followed firm. Thus,
information availability is inversely related to stock

variability at the time of an earnings announcement.

Atiase [1985] uses firm size as a proxy for differential

18

information,‘ ‘With weekly stock returns, Atiase [1985]
examines security price revaluations in response to second-
quarter earnings reports of 200 firms in 1971 and 1972.
Regression analysis is used to test the hypothesis that the
degree of a firm's price revaluation is inversely related to
its market capitalization. For this purpose he dichotomizes
his sample into large and small firms. The former group
consists of 100 large NYSE/AMEX firms. The latter group is

made up of 50 smaller NYSE/AMEX firms and 50 OTC firms.

To measure price revaluations in response to earnings
reports, Atiase uses three alternative indices: RI
(Revaluation Index), SRIl, and SRIZ (Standardized Revaluation
Indices One and Two). Atiase identifies a test period and an
estimation period consisting of the 104 weeks prior to the
test period. The test period is divided into a predisclosure
period and a report period consisting of seven weeks
surrounding the second—quarter earnings report date (ranging
from three weeks before the announcement week to three weeks
after). Two alternative definitions are used for the
predisclosure period. The predisclosure period is defined as
the period between the beginning of the fiscal year and either
1) the beginning of the report period or 2) the end of the
second quarter. If a firm has a December 31 fiscal year-end,

it has the following time horizon for the 1971 second-quarter

 

‘ Bamber [1986] suggests that private-search activities are
concentrated on relatively large firms.

19
earnings report:
Estimation Period (EP) Test Period (TP)
(104 weeks)
I ------------------------- I ---------------------------- I

December 31
1970

Predisclosure Report
Period (P) Period (RP)
(7 weeks)
alternative 1 I ----------------- I .......... I

Predisclosure
Period (P')

alternative 2 I ----------------- I
June 30
1971

Using the estimation period, the market model's coefficients

are estimated:

R =ai+BiRm+eit i=1...N(firms)

" t = 1 ... T (weeks),

where eit = stochastic individual component of Rn°

Estimated coefficients are used to compute'u", the unexpected

price change during the test period as follows:
“it = Ric " (at + biRm)‘

The three revaluation indices are given below:

RI. = -------------- , t* belonging to RP,

RI“. p.-2
SRIlu 8 ---------------------- , t* belonging to RP,
1 P, Pi t belonging to P,
---- z 111‘t
In t-l
RI". ry'- 2
SRIZR = ---------------------- , t* belonging to RP,
1 P,' P,' t belonging to P',
---- z RIit
Ifi' t=1

where

Sf = sample variance of eit for the estimation period,

1 (K's - i.)2
Cit. a 1 + --" + ----------------- .
T T

2( -‘)‘
t_1 Ru Ft

RI measures the ratio of new information conveyed to the
market by the earnings report to the average information over
the estimation period for a given firm. Thus, an RI greater
than 1.0 would imply that the earnings report conveyed more
than average information during the estimation period.
Results show that all three indices are significantly higher

for small firms than for large firms.

Atiase then specifies three regression models as follows:

Model 1 . RIi - ac + a1LCV, + a,,

Model 2 . SRIli - a0 + a1LCV, + (1.,
Model 3 : SRIZi - a0 + a1ch, + d,,

where LCV}=- natural logarithm of market value of firm i's
common stock.

21
In all the regression models, the estimated coefficients (afi
of the capitalized value variable Icwq are negative and
statistically significant at an alpha level of 0.0001. These
results are consistent with the hypothesis that the degree of
a security's price revaluation in response to its second-
quarter earnings announcement is inversely related to the

capitalized value of the firm, ceteris paribus.

Atiase's [1985] differential information hypothesis is
further examined by Freeman [1987]. Based on the notion.that
private information production increases with firm size,
Freeman [1987] examines whether the magnitude of abnormal
returns is inversely related to firm size. While Atiase
[1985] ignores the direction of stock price change, Freeman
[1987] investigates the relationship between the sign and

magnitude of stock price changes and firm size.

Freeman's sample (2263 firm-year observations covering
1966 - 1982) is divided into the following four portfolios:
large firms with good news (positive unexpected return on
equity), large firms with bad news (negative unexpected return
on equity), small firms with good news, and small firms with
bad news. In each year, large firms are the top market value
quartile of all firms. Large firms are then ranked by change
in return on equity (dROE). The top dROE quartile forms the
large-firm.good-news portfolio while the lowest dROE quartile

forms the large-firm bad-news portfolio. For each year of

22

dROE classifications, separate portfolios of large (L) and
small (8) firms are constructed of long'positions in.good news
firms and short positions in bad news firms. Average abnormal
returns (AAR) are computed over the interval beginning 24
months prior to the end of the fiscal year and ending twelve
months after the end of the fiscal year. The AAR in month
t - 1, .... 36 for a portfolio of large firms with i - 1, ...I
good news stocks and k - 1, .... K bad news stocks in year y

is written as

I K
AAR(I.~)Yt 8 (l/I)151AR““ - (1/K)RE1AR"“'

where ARWt is abnormal return (Rm - (aw + BWRMH .

Then cumulative average abnormal returns (CAAR) are computed
for each month t - 1, .... 36 for large firms.
t
CAAR(L)n - E AAR(L)W,
r-l
CAAR(S) is computed in an analogous manner for the lowest

market-value quartile of firms. Freeman [1987] tests whether

CAAR(S) exceeds CAAR(L).

Results show that CAAR(S) significantly exceeds CAAR(L)
at the 0.05 level for months 19 through 36. Also, a matching
technique is performed to see if the higher CAAR(S) is due to
the relatively high volatility of small-firm earnings.
Results for the matched subsample support the inverse

relationship between firm size and share price response.

23
2.4.b. Earnings Predictability

Based on the notion that return variability at the time
of therearnings announcement.should increase (decrease) as the
precision of the announcement increases (decreases) relative
to the precision of the prior information set, Pincus [1983]
hypothesizes that hard-to-predict earnings announcements
should exhibit greater return variability than easy-to-predict

announcements.s

In other words, easy-to-predict announcements
provide less news at the announcement than hard-to-predict
announcements so that the easy-to-predict announcements are

less valuable than those of hard-to-predict earnings.

Pincus [1983] employs the Earnings Predictability (E.P.)
Index, published as part of Value Line's Investment Survey,
as a proxy for earnings predictability. This index is based
on the standard deviation of percentage changes of earnings
over a five year period. ‘Value Line ranks firms from 5 to 100
in intervals of 5 with the highest number representing the
most predictable earnings. A high E.P. Index value implies
that individuals' priors regarding the dollar amount of firms'
forthcoming earnings releases reflect little remaining
uncertainty. Conversely, individuals' priors of firms
receiving low E.P. Index values reflect high degrees of

remaining uncertainty regarding the dollar amount of an

 

5 Precision is, in Pincus's [1983] words, referred to as
the extent that uncertainty is reduced by an announcement.

24
upcoming earnings announcement. Within his framework it is
expected that price change variability is inversely related
to the E.P. Index.

Pincus [1983] identifies an estimation period and a
forecast period. The forecast period is divided into a

nonannouncement period and an announcement period:

Estimation period Forecast period
(60 weeks)
I ----------------------- I -------------------------------- I
Nonannouncement Announcement
period period
I ----------------- I -------------- I

A forecast period begins with the first trading day of
the week following the week of the announcement of the
preceding quarter's earnings, and ends with the last trading
day of the week immediately prior to the week of the
announcement of the current quarter's earnings. Using the
estimation period, the market model's coefficients are
estimated and the estimated coefficients are used to compute
abnormal returns in the nonannouncement and announcement
periods. As a proxy for price change variability, Pincus
[1983] uses the ratio of variance of abnormal returns in the
announcement period to variance of abnormal returns in the
nonannouncement period. Based on the price change
variability, a sample of 136 firms (1978-1979) are grouped
into a large price change (in absolute value) group and a

small price change group. Also, the sample is divided into

25
an easy-to-predict group and a hard-to-predict group based on
the E.P. Index. Tests are conducted on the independence of
hard and easy predictability versus small and large price

change groups.

Results of a nonparametric rank correlation test provide
weak support (significance at the 0.09 level) for the
hypothesis for interim (lst, 2nd, and 3rd quarter)
announcements. Significance at the 0.05 level is indicated
for annual announcements, but the directional relation is
opposite that predicted. Another nonparametric test based on
relative frequency histograms provide the same results: hard-
to-predict firms are associated with large price changes for
interim announcements but the reverse is observed for annual

announcements.

As one explanation of the inconsistency between interim
and annual announcements, Pincus [1983] suggests that easy-
to-predict annuals are announced an average of three and one-
half trading days earlier than hard-to-predict annuals such
that early earnings announcements provide information useful
in revising priors regarding the earnings of later-announcing
firms. However, for quarterly announcements, he argues, hard-
to-predict announcements are announced less than one day
earlier than easy-to-predict announcements. Thus he maintains
that the results for annuals are not valid because the results

are confounded by the early/late-announcing phenomenon.

26
Pincus [1983] concludes that the interim announcement results

support his hypothesis.
2.4.c. Systematic Risk

If information about earnings is an important element in
valuing equity, then earnings announcements trigger price
adjustments to the extent that the announced earnings
realizations differ from those anticipated by the market. The
direction and size of the price adjustments depend on how the
market forms its anticipations. Assuming that the market's
anticipation of earnings is consistent with rational
expectations in the sense of Muth [1961], Miller and Rock
[1985] derive a two-period model describing the earnings
announcement effect. The model suggests that price change
following the disclosure of a firm's earnings is a function
of earnings surprise, the market required return, and a

persistence parameter:6

P1- E0(P1) " (x1- Eo(x1))*(1 + (k/(1 + 1)))I
where

‘5 - equity value after earnings announcement,

Eo(P1) - equity value before earnings announcement,

x, - actual earnings,

Eo(x1) - earnings forecast for x,,

k = persistence parameter representing the extent to which the
current earnings surprise continues in the future (Miller
and Rock [1985] assume that the persistence parameter is
known to the market),

 

6 The persistence parameter represents the present value of
revisions in expected future earnings given one dollar of
current earnings surprise.

27

i a market (risk-adjusted) discount rate.

Thus, in Miller and Rock [1985], the earnings announcement

effect is inversely related with the discount rate.

Focusing on the link between the time-series properties
of earnings, Kormendi and Lipe [1987] develop a multi-period
model for the earnings announcement effect. Kormendi and
Lipe's [1987] model is consistent with Miller and Rock's model
in that price changes associated with earnings announcements
are a function of earnings surprise, discount rate, and the
persistence parameter. The multi-period model states

Q

Rn = (1 +3218.Q')*(Ux"/P‘°1)'
where
RIt - stock return associated earnings announcement,

8 - l/(1 + i),

i - risk-adjusted discount rate,

Q, - persistent parameter representing the present value of
revisions in earnings forecasted for time t+s given a
dollar current earnings surprise.

In this model stock return is positively related to earnings

surprise since the earnings multiplier is positive, and the

earnings announcement effect is negatively correlated with the

discount rate, ceteris paribus.7

2.4.d. Systematic Risk and Earnings Growth

 

7 Kormendi and Lipe [1987] call the terms in the first

Q

brackets, (1 + E 82%), the earnings multiplier. It measures
881

a ratio of stock price change to earnings surprise.

28

The theoretical models of Section 2.4.c predict an
inverse relationship between the earnings announcement effect
and the discount rate. Collins and Kothari [1989] examine the
return/surprise relation and systematic risk using beta as a
proxy for the market discount rate. Assuming that annual
earnings realizations follow a random walk, change in annual
earnings is used as a proxy for unexpected earnings. As a
return metric, raw return (not abnormal return) is adopted
following Beaver et al. [1980] and Beaver, Lambert, and Ryan
[1987]. These studies report that the return/surprise
relation is essentially the same whether one uses raw return

or abnormal return.

Examining temporal as well as cross-sectional
determinants of the return/surprise relation, Collins and
Kothari [1989] propose that the return/surprise relation is
also positively associated with growth opportunities. Collins
and Kothari [1989] differentiate 'normal growth' from
'economic growth': 'normal growth' is commensurate with the
riskiness of investments in a competitive industry whereas
'economic growth' is growth resulting from projects yielding
above 'normal growth' (positive net present value projects).
Since the normal growth can be more easily predicted than the
economic growth, earnings surprise may be more informative of
the economic growth than of the normal growth. That is, a
positive (negative) earnings surprise may provide positive

(negative) information about the changing spread between

29
normal growth and economic growth. Whether an earnings
surprise provides useful information about economic growth
can be determined by the existence of economic growth
opportunities. If a firm has economic growth opportunities and
a positive earnings surprise, then the earnings surprise is
more likely to be informative of economic growth. Within this
framework, growth opportunities is positively associated with
the return/surprise relation. As a proxy for a firm's
economic growth opportunities, Collins and Kothari [1989] use
the ratio of market to book value of equity based on the
notion that the market to book value ratio depends on the
extent to which a firm's return on its existing assets and
expected future investments exceeds its required rate of
return on equity. It is hypothesized that the higher the
market to book value of equity ratio, the larger the share

price response to an earnings surprise.

To reduce measurement error problems, Collins and Kothari
[1989] employ a reverse regression (see Maddala [1977] and
Klepper and Leamer [1984]). Earnings changes are regressed
on returns and two terms representing equity return

interactions with risk and growth:

”Kn/Pita ' a0 + 3131: + “23133:: 1’ 3331333“!
where

uxn - change in annual earnings per share,
P“,1 - stock price as of t-l,

Ru 3 stock return,

Bit - beta risk,

3O

MB,t - market value to book value of equity ratio.

In a reverse regression, the functional relation between

independent and dependent variables is inverted.

Since beta is negatively related to the earnings
announcement effect, the coefficient a2 is expected to be
positive. Results show that the coefficient a2 is positive
and significant at the usual level (5%) of significance, which
indicates a negative relationship between beta and stock price
response. Risky firms have a smaller price response than
firms of lower risk. Results also reveal that the coefficient
a3 is negative and significant at the usual level of
significance, which indicates a positive relationship between
growth and the return/surprise relation. High growth firms
have a larger price response to an earnings surprise than low

growth firms.

Chapter 3. A Model of Market Response to Unexpected Earnings

The voluminous literature on unexpected earnings
generally concludes that the sign and magnitude of stock price
changes are a positive function of the sign and magnitude of
unexpected earnings. Section 1 of this chapter examines the
information content of unexpected earnings within a rational
expectations framework. A two-period rational expectations
model describing the nature of the information content is
developed in Section 2. This implements the rational
expectations framework of Section 1 in an explicit two-period
model of an unexpected earnings announcement. In Section 3,

the two-period model is extended to a multiperiod setting.
3.1. Information Content of Unexpected Earnings
3.1.a. Rational Expectations

Rational expectations are characterized by three
properties: ”unbiasedness", ”efficiency", and "consistency".1
Suppose that the market forecast at time t-1 of time t
earnings based on the information set Lp1 is related to the

actual earnings realization, x,, by

x: ' Et-io‘til‘t-l) + UXt' (3'1)

where

 

' For a discussion of the rational expectations conditions,
see Muth [1961], Sargent [1973], Shiller [1978], Lakonishok
[1980], or Nakamura and Nakamura [1985].

31

32

EP,( . ) - forecast of ( . ) made at time t-l,
U1!t - forecast error realized at time t.
The three properties may be stated formally as:
a) unbiasedness

EM(UXt:LM) =- 0 for any t:
b) efficiency

Cov(pr1,Ing) - 0, where Cov( . ) - covariance of ( . )3
c) consistency

E:(xmiI-'t) - EM(XM:LM) for any 5 > k if L: =- LN.

The first property, unbiasedness, implies that, on
average, the market forecasts correctly; it neither
systematically underestimates nor overestimates the actual
future value. By the property of unbiasedness, the expected
value of the forecast error is zero, on average. That is,
although the forecast error in each period, Uxt,‘will
generally be non-zero, it will be zero on average over a

number of forecasts.

The second property, efficiency, requires that the market
uses all relevant information available at time t-l (the
information set L24) to form a forecast of Xt. Therefore, the
forecast error, UX}, is independent of any information in
Lb,. The forecast error is due to unpredictable shocks and

will be unrelated to any information available at time t-l.

One way of stating the consistency property is that it

is impossible to know in advance how the market will change

33
its expectations. Only new information results in an
alteration of expectations. This requires that the nature of
expectations is stationary with respect to a given information
set. That is, the nature of the expectational process does

not change over time.
3.1.b. The Information Content of an Earnings Announcement

The classical valuation model (Gordon and Shapiro [1956])
says that stock prices are equal to the present values of
expected future cash flows to equity. Under some conditions,
this is equivalent to saying that the stock price equals the
present value of expected future dividend payments.2 However,
Campbell and Shiller [1988] show that earnings data help to
forecast future real dividends. Consequently, it can be
argued that earnings should be included in the information set

in forecasting stock prices.

Suppose that time t-1 is a time immediately prior to an
earnings announcement for a stock and time t is the time
of (or immediately after) an earnings announcement. If the
market value of stock at time t is denoted as Pu then at time
t-1 the stock price assessed by the market given the
information set, L“, must be EM(Pt}LM) . Following Sargent

[1987], the share price response to an earnings announcement

 

2 If the transversality condition holds, the present value
of future cash flows is equivalent to the present value of
future dividend.payments. See West (1987), Mankiw, Romer, and
Shapiro (1985), and Shiller (1981).

34

is given by

Et(Pt:Lt-i’ UXt) " Et-‘l‘PtiLt-l) '
EtEPt " Et-1(Pt:I‘t-1):Uxt " Et-1(Uxt:Lt-1)]' (3'2)
where
Eng - unexpected earnings at time t, added to the information
set, Lb1, to form the new information set, In: (If the
earnings announcement is the only information arriving
between times t-l and t, then the information set
consists of information available at time t-1, LN, plus
the unexpected earnings realization UX}.)
The left hand side of Equation (3.2) is the price change due
to the earnings announcement. Due to the property of
unbiasedness, Et_,(Uxt:LM) - 0. Hence, the right hand side
reduces to EJP: - EM(Pt:LM) :uxt]. As a result, the price will

change as long as Pt - EM(Pt:LM) and 0x, are not orthogonal.

35

3.2. Share Price Response in a Two-Period Model under
Uncertainty

3.2.a. Assumptions

A.1. Two-period Horizon:

Initially, assume that the firm has a two-period time horizon.

period 1 period 2

At the end of period 1, the assets of the firm are liquidated.
The firm then reinvests in period 2 depending on its success
in period 2. There is no risk of bankruptcy, so the firm
continues to operate as a going concern throughout the two
periods. The two period model is extended to a multiperiod

model in a later section.

A.2. NO Tax:
Following Miller and Rock [1985], there is neither corporate
nor personal tax. If included, taxes would affect the
magnitude but not the direction of price response to
unexpected earnings. This abstraction allows a more direct
analysis of the impact of investment decision revisions due

to unexpected earnings on share price.

A.3. Income Function:
The firm's periodic cash flow from operations (operating
income or operating earnings), x, is a stochastic, specified

function of investment, I. Let

36
x, - Eamon,” + ux,, (3-3)

x2 - E,(G,(I,)) + 0x2, (3-4)

where

IDA - unexpected income realized at time t,

X =- income or earnings realized at time t,

It 8 investment to be input at time t,

G (. ) - income generating function corresponding to 1,, which
is random such that G (. ) - v (G, + X) - K, where v
is a positive random variable with mean 1, and X
represents fixed operating costs that are assumed to
be constant throughout the two periods.

The stochastic operating income generating function, G, is

continuous and twice differentiable and provides diminishing

positive marginal returns to investment such that G' > 0 and

G" < Go A150 Gt(0) - -Ke

A.4. Riskless Debt:
Debt issued by the firm is assumed to be l-period, riskless
debt sold on a discount basis. Bankruptcy and agency costs

are explicitly omitted from the model.

A.5. Constant Discount Rate on Operating Cash Flow:
The firm has a known, constant risk-adjusted discount rate,
i, on operating income. The required return i is determined
(by the firm's investment opportunity set and is assumed to be
known and fixed for all time.

A.6. Constant Payout Ratio Dividend Policy:
A constant proportion of expected and unexpected earnings are

retained in the firm. The remainder of earnings are paid out

37
as aidividend to shareholdersw This does not require a firm's
earnings to be positive. If earnings are negative, the firm
can raise new equity from old shareholders through a rights

offering in order to continue operating as a going concern.
3.2.b. Investment under Uncertainty

At time 0 (the beginning of period 1), the firm makes an
investment decision, Io, expecting that the marginal return on
the investment will be equal to the marginal cost of capital.
This is a consequence of net present value maximization.
However, the firm does not know with certainty that the
marginal return on capital will be equal to the marginal cost

of capital. It only knows that
Eo(Go'(Io)) - 1 + i. (3.5)

In terms of capital budgeting, the firm expects that at I0 its
investment opportunity schedule will intersect with its
marginal cost of capital schedule. However, actual investment
15 may turn out to be different from the value-maximizing

level of investment.

If Go' (In) > 1 + i at time 1 (see Figure 1) , then the firm
underestimated its investment opportunity or its income
function. It expected at time 0 that its marginal net return
would be zero at 15, but it realized at time 1 that the
marginal net return is still positive at 10' As a result,

realized earnings exceed expected earnings. Conversely, if

38
Go' (Io) < 1 + i, then actual earnings are less than expected

earnings (see Figure 2).

Income Function
realized at t-1

  
  
 

(A <—- Income Function
s45/1+i expected at t-0

L\~—_ Positive Unexpected
Earnings

 

 

 

Figure 1. Positive unexpected earnings

1+1 \

’ "K— Income Function
expected at t=0

    
 
 

Income Function
realized at tel

Negative Unexpected
Earnings

 

 

 

Figure 2. Negative unexpected earnings

In addition, by the property of consistency, the
following should hold:

39
E°(Go'(I,,)) =- EO(G,'(I,)) a 1 + 1 and (3.6)

Eon.) .. E0(I,) - Io (3.7)

At time 0 the cost of capital and the level of investment are
expected to remain constant over the two periods. However,
actual I, may differ from Eo(I,) based on information received

in UX, .
3.2.c. Firm Valuation

At time 1, the firm realizes X,, pays dividend D,, borrows

capital F,, and invests I,, so that

The left hand side of (3.8) represents the firm's total cash
inflows and the right hand side the firm's total cash

outflows. The value of the original shares at time 1, V,, is

V1” D1" 1’1“" EHOW/(1 4' i). (3.9)

Making use of (3.4) and (3.8) , Equation (3.9) may be rewritten

as

V, - X, - I, + E.[E,(G,(I,)) + UXZJ/(1 + i)

= x, - I, + E,(G,(I,))/(l + i). (3.10)

Before time 1, when X, is not yet known, the value of the

original shares is the expected value of the shares, E0(V,):

4O
E0(V,) 3 Eo(X,) - EO(I,) + Eo(E,(G,(I,))/(1 + 1)

3.2 .d. Association between Earnings Surprise and Stock Return

Equation (3.10) represents the value of the original
shares immediately after the earnings announcement.

Therefore, making use of Equation (3.3),

v, a [EO(GO(I,,)) + UX,] - I, + E,(G,(I,))/(l + i)

IEO(G°(IO)) + UX, - I, + E,(G,(I,))/(1 + l). (3.12)

Equation (3.11) shows the value of the original shares before
the earnings announcement. Using Equations (3.3) and (3.7),

Equation (3.11) may be rewritten as

EO(V,) 3 E0[EO(G°(IO)) + UX,] - Eo(I,) + E0(G,(I,))/(1 + i)
a 30(Go(Io)) ' 30(1)), + Eo(G1(I1))/(1 T i)

=- Eo(c,‘,(Io)) - I(, + Eo(Go(Io))/(1 + 1). (3.13)

The change in value in response to the earnings announcement

is then V, - E0(V,) .

If UX, - 0, then Go'(I,,) - (1 + i) and the firm invested
exactly the right amount of capital. Since the income
generating function is a random walk, no adjustment of the
investment level is necessary or desirable (dI - 0). ‘The firm
invested the value-maximizing amount. Consequently, equity
value does not change at the time of the earnings

announcement.

41
If UX, differs from 0, then the firm either under- or
over-invested relative to its optimal level of investment.
Equity value should either rise to reflect the unexpected
investment opportunities from underinvestment (E,(G,' (10)) > 1
+ i) or fall because income was insufficient to meet costs due
to overinvestment (E,(G,' (10)) < 1 + i). The firm will revise

its next investment I, so that
I, = I,) + dI, (3.14)

where dI represents either net new investment (dI>0) or
disinvestment (dI<0) relative to the level of investment I,) in
the previous period. The value of the firm immediately after

the announcement will be

v, 'Eo(Go(Io)) + UX, - Io - dI

+E,(G,(Io+dI))/(l + 1). (3.15)

Since E,(c,(1,, + dI)) - E,(G,(Io)) + E,((dG,/dIo)*dI) for small

changes in the level of investment,

v, - Eo(G,,(Io)) + ux, - Io - d1
4» E,(G,(Io))/(l + 1) + E,((dG,/dIo)*dI)/(l + 1) , (3.16)

where th/dI,K - th/dI at the level of investment I,‘ (for
notational convenience let dG,/dI - dG,/dIo) . Because

E1(G1(Io)) ' 30(Go(10)) + UX,,

v, - 20(co(Io)) + ux, - I0 - dI + 30(co(1,,))/(1 + 1)
+ UX,/(1 + 1) + E,((dG,/dI)*dI)/(l + 1). (3.17)

42
The change in value at the time of an earnings announcement
is equal to Equation (3.17) minus Equation (3.13):
z, ( (dG,/dI) *dI) 0x,

v1 - Eo(v1) - UX‘ - dI + ............... + ...... e (3e18)
1 + i 1 + 1

Assuming d1 = a*UX,, where 0 < a < 1,

E,(dG,/dI) UX,
v, - EO(V,) - UX, - a*UX, + ------------ *a*UX, + -------
1 + i 1 + i
E,(dG,/dI) 1
- UX, * [l - a + a* ----------- + ------ ]
1 + i 1 + i
E,(dG,/dI) 1
=Ux,*[1+a( ----------- -1)+ ------ ].(3.19)
1 + i , 1 + 1

Equation (3.19) holds for both positive and negative
unexpected earnings. Regardless of the sign of earnings
surprise UX,,jprice responds to an earnings surprise through
the same mechanism. The change in value in response to an
earnings surprise is proportional to the magnitude of the
surprise in earnings as in the Miller and Rock model.
Equation (3.19) may be rewritten as the following:

mama/ax) - awn-(1+1) + 11
V, - Eo(V,) - UX, * [1 + ---------------------------- ].(3.20)
1 + i
The term, [a*E,(dG,/dI) - a*(1 + i) + 1], corresponds to Miller
and Rock's ”persistence parameter". Whereas Miller and Rock

provide no economic rationale behind the parameter, this model

shows that the parameter is a function of marginal

 

 

43
productivity E,(dG,/dI) , cost of capital 1, and the firm's
earnings retention rate a. The market responds to the
earnings surprise by relating that surprise to the firm's
marginal productivity, cost of capital, and earnings retention
rate a. The market's reaction depends on the firm's
unrealized investment opportunities whether the earnings

surprise is positive or negative.

While the functional form of share price response to
earnings surprises is the same for both positive and negative
unexpected earnings, it is important to note that the
comparative statics are different. The reason for this is
twofold: 1) the association between the sign of the change in
value and the sign of [E,(dG,/dI)/(1+i) - 1], and 2) the

multiplicative form of the return/surprise relationship. If

 

 

 

 

 

Relationship with (V, - E,,(V,))
FaCtor ------------------ 1r ----------------- q
Itux,>o IfUX,<0
UX, + +
a + +
E,(dG,/dI) + -
i - +

 

 

 

 

 

Figure 3. Predicted relationship
between a factor and share price

44
UX, > 0, then E, (dG,/dI)>(1+i) and the term in the brackets is
positive. If UX, < 0, then E,(dG,/dI)<(1+i) and the term in
the brackets is negative. This results in an asymmetry
between positive and negative earnings surprises. The

relationships are summarized in Figure 3:

If UX, > 0, then [E,(dG,/dI)/(1+i) - 1] is positive since
marginal productivity must have been greater than cost of
capital. Given the other factors, higher earnings retention
rates a should be associated with higher equity returns.
Marginal productivity should also be positively related to
equity return. Cost of capital should be negatively related

to return, ceteris paribus.

The comparative statics of negative earnings surprises
are not as straightforward. If UX, < 0, then [E,(dG,/dI)/(1+i)
-1] is negative since marginal productivity must have been
less than marginal cost. Since V,-E0(V,), UX,, and
[E,(dG,/dI)/(1+i) - 1] are all negative and 0 < a < 1, higher
earnings retention rates should be associated with higher

3 That is, share price

(i.e. less negative) equity returns.
should.not fall as much for firms with.high earnings retention
rates as for firms with low retention rates, ceteris paribus.

For negative earnings surprises, marginal productivity and

 

3 According to the model, equity value would increase in
response to a negative surprise if [1 + a*E (dG,/dI)/(1+i) +
1/(1+i)] < 0. This would require a*E,(dG,/di) < -2-i, which
will never occur if marginal productivity is a positive but
diminishing function of investment.

45
cost of capital are negatively and.positively related.to»stock

return, respectively.

46

3.3. Multi-period Model

The basic approach of the two-period model can be
extended into a multi-period setting. The multi-period model
differs from the two-period model only in the time horizon.

Identity (3.8) still holds:

Xt+Ft=It+DU

X + F = I + Dug for s = 1, 2, ...T (3.21)

t+s t+s tr!

subject to IN = 0 due to liquidation at time t+T.

At time t, the value of the shares outstanding at time t-l,

Vt, is

T D,“

Vt = D: - Ft + Et ( z ----------- )
s=l (1 + 1)s
T X + F - I

t t t
= Xt ' It + Et ( 2 ————— 1. ----- + ...----:i-.. ) (3.22)
s=1 (1 + i)s
for T > 1.

Hence, V, is the value after the earnings report for X,. On
the other hand, the value just before the report, EF,OQ), is

expressed as

Et-1(vt) = Et-1(Xt) ' Et-1(It)

Assume the firm does not change its financing policy at the

time of the earnings announcement. If Gb,'(Ib,) > 1 + i, then

47

T X,” T xt+s

+ E [ z --------- 1 - 13,, [ z ---------
s=1 (1 + 1)8 s=1 (1 + 1)8
T-l It, T-l 1m

- E [ z --------- 1 + EM [ z ---------
s=1 (1 + 1)’ s=1 (1 + 1)8

for' T > 1. (3.24)

Due to the property of consistency, the following should hold:

for s = 1, 2, ....T

Et-1(It) = Et-1(It+s) = It-l'
Et(It) = Et(It+s) g It'
E,(Im) - swam) = (11,, (3.25)

E,(Xm) - E,_,(Xm) = E,((dG,/dI,,,)dI,) + UX,_ (3.26)
Therefore, Equation (3.26) can be rewritten as

v - E,,,(v,) = uxt - dIt

t

r E,(dG,/dI,,,)dI, + UX,
+ z ------------------------
s=1 (1 + 1)8
T-l dI,
-' E ----------- for T > 1. (3.27)
s=1 (1 + 1)8

Assuming d1, = a*UX,, where 0 < a < 1,

T E,(dG,/dI,,,)*a + 1

V '- E _(V') = Uth [l - a + E ---------------------
t t1 t t s=1 (1 + 1)’
T-l a
- E ----------- ] for T > 1

48
'r E,(th/dIM)
=- uxtvnr [1 + a*( z -------------- - 1)
8-1 (1 + 1)‘
T-l 1 - a l
— z --------- + -------- 1 for T > l.(3.28)
s-1 (1 + 1)’ (1 + 1)’
Equation (3.28) is the multi-period model that associates
unexpected earnings with share price change. By the same
argument as in the two-period model, stock response to
negative unexpected earnings is given by Equation (3.28) . The
implications of this model are the same as those of the two-
period model. When.T - 1, the model reduces to the two-period
model. The term in the brackets is an expression
corresponding to Kormendi and Lipe's [1987] earnings

multiplier. It is nonnegative as in Kormendi and Lipe [1987]

since variable a cannot be greater than one.

Chapter 4. Empirical Design
4.1. Empirical Specification

Although Equation (3.19) demonstrates a mechanism by
which the market reacts to unexpected earnings, it requires
some modification for the purpose of empirical research.
First, the left-hand side should be standardized to reduce
problems arising from cross-sectional differences in the level
and variability of equity value. Dividing both sides by E0(V,)
and defining V, and UX, as per share values,

v, - E°(V,) UX, E,(dG,/dI) 1

----------- = ----—-*[1 - a + a*---------- + ----- ].(4.1)

E0(V,) E0(V,) 1 + 1 1 + 1

The left hand side of Equation (4.1) represents the percentage
change in share price (stock return). UX, is similarly
standardized as unexpected earnings as a percent of share
price. Equation (4.1) has a form consistent with Christie's
[1987] argument that share price is the appropriate deflator
to measure the association between security returns and
accounting variables. There are three other key variables in
Equation (3.19) . By assumption, a - dI/UX, represents the
proportion of unexpected earnings which are reinvested in the
firm. Retention rate (1 - payout ratio) is used as an
empirical proxy for the proportion of unexpected earnings
reinvested in the firm. The second empirical proxy represents
the term E,(dG,/dI)/(1+i) . This term is equivalent to Tobin's

49

50

q ratio. Tobin [1969] and Ciccolo [1975] show that the ratio

of the marginal return on capital to the marginal cost of
capital is another version of q ratio, originally defined as
the ratio of the valuation of a corporation in security
markets to the replacement cost of its physical assets. The
third empirical proxy is the market required rate of return
for the discount rate 1 in the model. Thus, stock price
response to unexpected earnings is a function of a firm's

earnings retention rate, Tobin's q ratio, and cost of capital.

For notational convenience, define

v, - E0(V,)
R a Stock Return = --------------- ,
UX,
S = Earnings Surprise 8 ---------- ,
E0(V,)
E,(dG,/dI)
Q = Tobin's q ratio a ------------ .
1 + 1

Equation (4.1) can be rewritten as

R - s t [1 + a*(Q - 1) + ---1-- 1. (4.2)
1 + 1
Equation (4.2) shows that R is a nonlinear function of S, a,
Q! and i. The concern of this study is to identify the
factors associated with market response to earnings surprise,
not the particular functional form, and to compare these
factors with factors suggested by other studies. Therefore,

a general linear model is adopted for the regression following

studies such as Leftwich [1981], Jain [1982], Easton (1985],

51
and Christie [1987]:

R,t - intercept + b,Sit + bza,t + 1’3th + b,i,, + “th (4.3)
for firm j - 1,.....N and time t - 1, ..... T,
where u - stochastic disturbance following a white noise
process.

There are three reasons why a linear model like that
proposed in this chapter may not capture the true relationship
between share price response and earnings announcement.
First, the model could be misspecified. This could occur, for
instance, if the true relationship is nonlinear or if

variables are omitted from the model. This study assumes

linearity in the return/surprise relationship in the hope that
the 1::rue relationship is not too badly misspecified. Chapter
6 examines several variables which may have been omitted from

the theoretical model in Chapter 3.

Second, the model could be correctly specified but the

emDirical proxies could contain measurement error. For
instance, the errors-in-variables could arise in the
“neXpected earnings measure because of an inadequate proxy for
the market's earnings expectations and/or because accounting
earn ings fail to reflect economic earnings. Measurement error
in unexpected earnings is probably not a major problem and
prObably does not materially affect the results. Easton

[1985] reports a strong link between security price and

ac-‘-c<>unting earnings which implies that accounting earnings are

52
a reliable proxy for economic earnings. Brown and Rozeff
[1978] find the financial analysts' earnings forecasts used
as market expectations in this study compare favorably to
other expectational models. O'Brien [1988] further develops

analysts' forecasts as a proxy for the market's expectation.

Also of concern is the measurement error in the three
firm-specific variables of the model in Equation (4.3) . While
the earnings retention ratio is perfectly negatively related
to the dividend payout ratio, the dividend payout ratio is a
function of a firm's level of earnings which varies over time.
Tobin's q ratio is estimated with Lindenberg and Ross' [1981]
algorithm and is subject to measurement error. Beta will be
used as a proxy for cost of capital. While beta is linearly
related to required return under the CAPM, the impact of the
discount rate in the theoretical model of Chapter 3 is
nonlinear. These measurement errors may have an important
impact on the empirical results. Yet the proxies represent

reasonable empirical alternatives to the true variables.

Third, the relevant variables and the functional form of
the return/surprise relationship could be correctly specified,
but the slope coefficients may differ across firms,
conditions, or over time. Chapter 5 reviews the set of
conditions from the theoretical model including the size and
direction of earnings surprise, dividend yield (or the

earnings retention rate), q ratio, and the systematic risk of

53
the firm. Chapter 6 extends the analysis to other conditions
or factors identified in the empirical literature as
contributing to share price response including firm size
(Atiase [1985]), earnings predictability (Pincus [1983]), and

growth (Collins and Kothari [1989]).

This study focuses on the sign and significance of the
coefficient on each variable in Equation (4.3). As noted in
Chapter 3, Equation (4.3) has different implications for

positive and negative earnings surprises.

54

4.2. Empirical Proxies and the Data Sample

The earnings (income) in Equation (3.19) represents net
economic earnings (see Ohlson [1983]), not merely accounting
earnings. Accounting earnings are employed as a proxy for
unobserved economic earnings as in other studies. Thus, the
relation of earnings announcements with share prices partly
depends on the correspondence of accounting earnings to
economic earnings. Due to the superiority of financial
analysts' forecasts over time-series models in forecasting
quarterly earnings (see Chapter 2), financial analysts'
consensus quarterly earnings forecasts are used.1 'The source
of the consensus earnings forecasts is the I/B/E/S database

2 Unexpected earnings are measured

of Lynch, Jones and Ryan.
as actual earnings minus the consensus earnings forecast.
Unexpected earnings are scaled by share price (derived from
the CRSP daily returns tape) as of three trading days prior
to the earnings announcement date as a proxy for S,t in

Equation (4.3). Actual earnings information is derived from

accounting data on the COMPUSTAT quarterly tape.

 

‘ Consensus (average) estimates of earnings are believed to
play a key role in share price determination. See Elton,
Gruber and Gultekin [1981 s 1984], Crichfield, Dyckman and
Lakonishok [1978], and Cragg and Malkiel [1968]).

2 Lynch, Jones and Ryan collect, on a monthly basis,
earnings estimates from major brokerage firms on over 2,000
corporations. Forecast statistics include, among others, the
arithmetic mean, median, range, and standard deviation of the
individual analyst estimates of primary earnings per share
before extraordinary items for each corporation.

 

55

Empirical research has found that unexpected earnings are
associated with contemporaneous changes in price (Ball and
Brown [1968]) as well as price changes which both lead
(Beaver, Lambert and Morse [1980]) and lag (Collins and
Kothari [1989]) earnings surprises. According to the
efficient market hypothesis, however, a full response of stock
prices to news occurs essentially immediately.3 This study
focuses on contemporaneous price responses and their
determinants. In their study of stock price adjustment to
earnings and dividend announcements, Patell and Wolfson [1984]
report the effects of the announcements are captured within
two days. Christie [1987] reports that if the return interval
is sufficiently short, it is not necessary to consider the
confounding effects of'other'announcements. In light of these
reports, a two-day return is employed. Following Pincus
[1983], the geometric average return of the day of the
earnings announcement and the day preceding the announcement
of firm j at time t serves as a proxy for R,t in Equation
(4.3). Although raw returns (not risk adjusted) are better
than abnormal returns in the spirit of this study, both
returns are regressed on the same set of independent variables

4

for the sake of comparison with other studies. To measure

 

3 In their study of stock price adjustment to economic news,
Pearce and Roley [1985] use one-day returns to capture the
effects of economic data announcements.

‘ Equation (4.1) shows that raw return should be used as a
dependent variable. But if market-adjusted returns are
desired, using abnormal returns should make little difference.

56
abnormal returns (ARR), the conventional market model is

employed. That is, OLS estimates are obtained for the model:

Rjt = aj-rl%Rm:+ en, (4.4)

where

R,t 2 return of firm j in period t,

'11., - value-weighted market return obtained from the CRSP daily
index tape in period t,

0,, B =- estimated regression coefficients of firm j,

ejt - estimated error term of firm j in period t.

To obtain parameter estimates, the model is run over a 180-

day estimation period.5 .Abnormal returns are then
ARjt = 12,, - (a, + 3,12“). (4.5)
This approach is repeated for each quarter and for each firm.

0,, (Tobin's q ratio) is estimated using the algorithm6 by

 

5 The last day of the estimation period is three days prior
to the earnings announcement. A 180-day estimation period is
arbitrarily chosen. A 90-day estimation period is also used
resulting in basically the same measure of parameters.

6 Lindenberg and Ross [1981] estimate
q ratio - (market value)/ (replacement cost) with the following
proxies:

Market value - MVD, + MVP, + MVS,,

where MVD,, MVP" and MVSt are estimated market values of
debt and preferred stock and market value of common
stock, and

Replacement cost - TAt + RNP, - HNP, + RINV, - 1mm,

where TA - total assets in year t,
RNP, - net plant at its historical value in year t,
HNP =- net plant at its historical value in year t,
RI , =- inventories at replacement value in year t,
HINVt - inventories at historical value in year t.

 

57
Lindenberg and Ross [1981]. The numerator of the q ratio is
the firm's market value defined as the sum of the actual
market value of common stock and estimated market values of
preferred stock and debt. The denominator of the q-ratio is
the replacement cost of the firm's assets. All the
information about the numerator and the denominator are

readily available from the NBER's R&D Master File.7

The firm's earnings retention rate (1 - payout ratio) is
used as the empirical proxy for a". Quarterly dividend
payments are derived from the COMPUSTAT quarterly tape. These
dividend payments should be divided by earnings to measure
payout ratios. However, many firms report negative earnings
resulting in negative payout ratios. To overcome this
problem, dividends are divided by stock price instead of
earnings, which, in fact, measures dividend yields, not
dividend payout ratios. Dividend yield is a more appropriate
variable for regression in the spirit of Christie [1987].

As a proxy for the required rate of return in, market
model betas are used following Collins and Kothari [1989].
According to the Capital Asset Pricing Model, beta is linearly
related to required return. Equation (4.4) is adopted to

 

7 NBER's R&D Master File is created and updated at the
National Bureau of Economic Research. The file consists of
about 2600 large manufacturing firms. There are approximately
70 variables for each firm including market values of common
stock, preferred stock, and debt, and inventories adjusted
for inflation. For detail, see Hall et al. [1988].

58

compute betas.

The following criteria are used to select the firms in

the sample. The firm must

(1) be listed on the daily returns tape constructed by the
Center for Research in Security Prices (CRSP) at the
University of Chicago:

(2) be listed on the COMPUSTAT quarterly tape;

(3) be listed on the I/B/E/S/ consensus tape:

(4) be listed on the NBER's RED Master File:
(5) report quarterly earnings announcements in the Wall Street

Journal Index.
Criteria (1), (2), (3), and (4) provide assurance of easy
access to readily available data from the data bases. With
criteria (5), earnings announcement dates can be verified.
These criteria yield a sample of 4972 firm-quarters (452 firms
with consecutive 11 quarters) over the period of the fourth
quarter of 1983 through the second quarter of 1986. This
sample period is mainly determined by the data availability

on the tapes.

Using the proxies suggested in this section, Equation

(4.3) is rewritten in the following form:
th - intercept + b,S,t + b.,D,t + 1330,, + b,B,t + u,,, (4.6)

for firm j - 1,.....N and time t a 1,.....T,

where

D,t - quarterly (contemporaneous) dividend scaled by price as
of three days prior to the earnings announcement,

a market betas estimated on a 180-day period before the
earnings announcement.

B"

59
Equation (4.6) is the model to be used for empirical analysis
in this study.

4.3. Model Specification and Parameterization

The sample described in the preceding section is a panel
data sample of time series and cross-sectional observations.
The model typically used for testing the association between
stock price changes and unexpected earnings is a simple linear

model. The linear model in the context of this study states

where

R,, - geometric average stock return of the day of the earnings
announcement and the day preceding the announcement of
firm j at time t,

S,, - (actual earnings - earnings forecast)/share price of firm
j at time t,

D" - dividend yield of firm j at time t,

Tobin's q-ratio of firm j at time t,

market beta of firm j at time t.

one
To":
Ill

The parameters are assumed to be identical across firms and
over time in Equation (4.4). Temporal variation in firm-
specific response to earnings announcements is not of interest
in this study and is assumed to be insignificant. Instead,
it is supposed that there exist important cross-sectional
differences across firms. In the presence of such cross-
sectional firm differences, OLS estimates according to

Equation (4.4) are inefficient.

In general, there are two approaches for incorporating

60
cross-sectional firm.heterogeneity into the regression model.
The fixed effects model accounts for cross-sectional effects
which are not readily attributable to individual causal
variables by absorbing these effects into a firm-specific
intercept term. The fixed effects model states

Within estimation (see Mundlak [1961] or Hoch [1962]) is then
applied to the model. However, in the presence of between
firm variation in price response, within estimators are

inefficient.

The random effects model explicitly accounts for between
firm variation by treating the firm effects bj (captured by
the intercept terms in the fixed effects model) as random
variables with zero mean and homoskedastic variance Var[b].

Formally,

R,, =- b0 + b,S,, + szj, + 1330,, + b‘Bj, + w,,, (4.9)

where w,, - b, + u,,.

Since the firm effects b, are unrelated to the random error

u the covariance matrix C is block diagonal such that

it’

[ a
M 0
o o O O O
O M

03°
00°

C - (Var[b]+ Var[u]) * ,(4.10)

 

 

61

where

1 P
p 1
P P

(I!

ll
'U'U
'O'U'O

(4.11)

[999

He

 

 

and p a Var[b]/(Var[b] + Var[u]).

The generalized least squares (GLS) estimator is consistent
and efficient in this situation as long as the firm effects

are not correlated with the regressors.a

Rewriting (4.9) in vector form,

z,==tg c + w,, j a 1, 2, ..... N, (4.12)
where
z, - (R,), a Txl matrix,
W, - (e, F,) - (e, S , D,, Q , ), a Tx5 matrix,
e a (1, 1, .... , 1;, a 1xT un t vector,
F, - (8,, 0,, 0,, 8,), a Tx4 matrix,
G - (b,,, m') - (b,,, b,, b2, b,, b,,), a 1x5 matrix,

m' - (b,, b2, b,, b,), a 1x4 matrix,
w,' - (w,,, ,w,,), a le matrix.

The covariance matrix of w, (C,) is

C, - (Var[b] + Var[u])*M

- Var(u)*I,-+‘Var(b)*ee'. (4.13)

Its inverse is (see Wallace and Hussain [1969] and Nerlove

 

s For GLS estimation with the random effects model, see
Balestra and Nerlove (1964), Wallace and Hussain (1969), and
Maddala (1971).

62

[1971])
.1 1 [ Var(b) ]
c, =- ------ I, - ------------------ ee' (4.14)
Var(u) Var(u) + T Var(b)
1 1 1
= ------ l ( I, - --- ee' ) + r --- ee' ]
‘Var(u) T T
1 1
= ...... H + r --- ee'] , (4-15)
Var(u) T
where

H - I, - (1/T)ee' and
r - Var[b]/(Var[u] + T Var[b]).

Then the normal equations for the GLS estimators can be

written as

N 1 b,, N 1
E wxg‘w, .- 2 wgg'z, .. (4.16)
j-l m j-l

us

Solving (4.13),

 

W, 1
l N N _ _ _ _
mm5=I - Z W Ufll + r 2 (P.- )(F2 - F)’
T j-lj 1 j=1 1 J
J
1 N N _ _ _ _
* - 2 W,'HZ, + r E (F, - F) (F, - F)‘ , (4.17)
T j-l j-l

bOGLs . 2.- no“. Fr- (4.18)

where

63

_ 1 'r _ 1 N 'r
F, - -- z r,,, r - --- z 2 r,,,
r t-1 NT j-1 t-l
_ 1 'r __ 1 N 'r
z.---zz.,, z----z zz.
’ 'r t-1’ NT j-l t-l 1:

Using the between-group estimator (mb) and the "within

estimator" (mg), the GLS estimator (4.14) can be rewritten as9

m,Ls = dim, + (I, - d)*m,,, (4.19)
where
4
N N _ _ _ __
d=rT 2w,'nw,+r'r2(r,- )(F,-F)'
j-l j-l
N - —- —
* 2(r,-r)(r,-i-*T' .
i=1

The GLS estimator (4.19) is a weighted average of the between-
group and within estimators. In essence, r measures the
weight given to the between-group variation. In OLS
estimation of Equation (4.7), the between-group and within-
group variations are just added up, so that the proper weight
r is ignored. In the within estimation for Equation (4.8),

 

9 The within estimator is also called least-squares dummy-
variable estimator. Only the variation within each group
(firm in this study) is utilized in forming this estimator.
On the other hand, between-group estimation ignores variation
within the group. For further discussion, see Wallace and
Hussain [1969], Maddala [1971], or Hausman and Taylor [1981].

64
the between-group variation is completely ignored. Therefore,
GLS estimation of equation (4.9) gives efficient estimates.
If r ~ 1, however, "as reduces to the OLS estimator. If

r e 0, mm, reduces to the within estimator.

The problem with GLS estimation is that both Var[b] and
Var[u] are unknown. This study adopts a two-step GLS

procedure following Taylor [1980] to overcome this problem.

65
4.4. Comparative Statics of the Empirical Model

Rewriting Equation (4.6),

R,, a intercept + b,S,, + 6,0,, + 1339,, + b,B,, + u,,. (4.6)
The predictions on b,, b2, b3, and b, are based on Equation
(3.19), but the predictions differ for UX,:> 0 and UX, < 0.

Suppose actual earnings are greater than forecasted
earnings (positive unexpected earnings) in Equation (3.19).
Since the term in the brackets of Equation (3.19) is always
greater than zero, S“ is positively correlated with R“: that
is, an earnings surprise (good news in this case) always leads
to a positive stock return. Therefore, b, should be positive.
While the variable a is positively related to stock return for
positive surprises, dividend yield is used as a proxy for the
earnings retention rate a. Hence, the coefficient on D“, ha,
is predicted to have a negative sign. The coefficients b3 and

in should be positive and negative, respectively.

If actual earnings are less than forecasted earnings
(negative unexpected earnings), the comparative statics of
Equation (3.19) are slightly different. SR is still
positively related to R": negative earnings surprise (bad
news) should result in a negative return. A negative sign is
expected for b,. Since variable a is positively related to
stock return, dividend should be negatively related to stock

return: the higher the dividend, the lower the return, ceteris

66
paribus. Therefore, b2 should be negative. The coefficients

b3 and b, should be negative and positive respectively.

Chapter 5. Empirical Results from the Theoretical Model
5.1 Entire Sample Results

Many authors (Brown, Richardson, and Schwager [1987],
O'Brien [1988], Collins and Kothari [1989], etc.) estimate an
earnings response coefficient by pooling positive and negative
earnings surprises. While this is inconsistent with the
theoretical model of Chapter 3, it is worthwhile investigating
the factor coefficients of the theoretical model in this
pooled context. This approach is intended to provide
comparability to other research using pooled earnings
surprises. Descriptive statistics for the entire sample
including both positive and negative earnings surprises are

presented in Table 1.

Summary statistics for raw and abnormal returns reveal
that the average stock return associated with earnings
announcement is negative. This is expected because the sample
has more observations for negative earnings (2633) than
positive earnings (2126) surprises. But the average return
is not significantly different from zero given the standard
deviation. Summary statistics for unexpected earnings show
that the average earnings surprise (scaled by price) is
-0.01330, and the standard deviation is 0.15102. The average
earnings surprise is, therefore, not significantly different

from zero. This is consistent with analysts' forecasts being,

67

68
on average, unbiased estimates of actual earnings during the
sample period. Dividend yield (dividend scaled by price)
ranges from 0 to 0.05828. The sample mean beta of 0.93135
indicates that the sample is slightly less risky than the CRSP
value weighted average. This is expected since the sample
selection criteria are biased towards including larger firms
and previous evidence indicates that firm size and beta are
negatively related (Banz [1981]). Given the standard
deviation of 0.45341, however, the average beta is not
significantly lower than one. 0 ratio is not significantly

different from one, either.

First, stock returns (Rn) are regressed on earnings
surprises (S,,), dividends (D,,), q-ratios (Q,,), and betas (B,,)
using OLS on the combined sample of positive and negative
surprises.1 Results are reported in Table 2 (Panel A). All
the coefficients on the independent variables are

significantly different from zero at the usual level of

 

' Several studies employ reverse regression to reduce the
problem in measuring earnings surprise (see, for example,
Beaver, Lambert, and Ryan [1987] and Collins and Kothari
[1989]) . To use reverse regression the following model should

be estimated:
8,, =- c, + c212,, + c3D,,*R,, + c,Q,,*R,, + c,B,,*R,,.

However, R,,, D,*R,,, Q,,*R,,, and B,*R,, are seriously
correlated with one another for my sample. The highest
correlation is 0.87640 between R, and B,,*R,. Besides, c3,
cg, or’cg'might be significantly different rom zero only
because of the strong relationship between S,, and R,,.

69
significance.2 Note, however, that the model in Chapter 3
predicts that the sign of the price response relationship of
q ratio and cost of capital (or beta) depends on whether the
surprise is positive or negative. Coefficients on earnings

surprise and dividend are as predicted by the model.

Generalized least squares (GLS) estimators are consistent
and efficient and would control for heteroskedasticity
problems. To obtain the GLS estimate, a two-step procedure
is employed following Taylor [1980]. Within and between
estimations are adopted to determine the weight r (see Chapter
4). The results are provided in Panels B and C, respectively.
The mean square errors are 0.00039 and 0.00003 respectively
for within and between estimates. This implies that r
approaches one, so that the GLS estimate reduces to the OLS
estimate. This is consistent with Christie's [1987] argument
that heteroskedasticity does not appear to create serious

problems in returns studies.

Beaver, Lambert and Morse [1980], Beaver, Lambert and
Ryan [1987], and Collins and Kothari [1989] argue that the
return/surprise relation is essentially the same whether one
uses raw return or abnormal return (beta risk adjusted).

Their argument is that, relative to the variability in raw

 

2 A time-series model is employed to see whether the
results are sensitive to the use of earnings forecast. The
model suggested by Brown and Rozeff [1978] is used for
earnings forecasts. No difference is found (not reported
here).

'70

return around earnings announcements, the variability in

expected return based on beta risk is small such that the use
of abnormal return amounts to using raw return. To verify
their argument, raw return is replaced with abnormal return
and the regression is repeated utilizing the same independent
variables. Basically, the same results are obtained (Table
3). Nevertheless, results based on abnormal returns are also
presented throughout this study for comparison because most

other studies employ abnormal return as a return metric.

Overall, the results provide evidence that dividend
policy, q ratio, and beta significantly impact the market
response to earnings surprise. As demonstrated in Chapter 3,
however, predictions on the coefficients of the independent
variables should be different depending on the sign of the
earnings surprise. Hence, whether the factors identified by
the model are important is not clear until an analysis is made

on each group of positive and negative surprise observations.

71

 

 

Table 1
Summary statistics: 4972 firm-quarters
Variable N Mean Standard Minimum Maximum
Deviation
Return 4972 -0.00029 0.02094 -0.17135 0.30384
Abnormal 4972 -0.00071 0.00020 -0.l6724 0.30464
return
Unexpected 4972 -0.01330 0.15102 -6.02842 0.48380
earnings
scaled by
price
Dividend 4972 0.00672 0.00503 0.00000 0.05828
scaled by .
price
q ratio 4972 0.92127 0.60196 0.02784 7.65163
Beta 4972 0.93135 0.45341 -0.55150 2.87790

 

72
Table 2
Effect of dividend, q ratio, and beta on the
market response to unexpected earnings:
4972 firm-quarters from 1983.4-1986.2*
R,, =- b0 + b,S,, + sz,, + 1330,, + b,B,, + u,,**

Dependent variable: Raw return (Rn)

 

Independent Variables Regression Statistics

 

 

 

Earnings Dividend 0 ratio Beta F value R-square MSE
surprise scaled (on) (ER) (Prob>F) (Adjusted)
scaled by
by price
price (DR)
(8,.)
A. OLS estimation
0.0109# -0.l836# 0.00145 -0.0014# 13.916 0.0111 0.00043
0.0019 0.0613 0.0005 0.0006 (0.0001) (0.0103)
5.561 -2.993 2.881 -2.195
0.0001 0.0028 0.0040 0.0282
0.0000 0.0014 0.0020 0.0141
8. Within estimation
0.01074 -0.6473# 0.0040? -0.0039# 22.065 0.0175 0.00039
0.0020 0.1231 0.0011 0.0010 (0.0001) (0.0167)
0.0001 0.0001 0.0002 0.0002
0.0000 0.0000 0.0001 0.0001
C. Between estimation
0.0076 -0.0271 0.0009 0.0002 1.801 0.0159 0.00003
0.0050 0.0715 0.0005 0.0008 (0.1275) (0.0071)
1.521 -0.379 1.649 0.268
0.1290 0.7046 0.0999 0.7891
0.0645 0.3523 0.0499 0.3945

 

* Standard errors, t-statistics, p value based on 2-tailed
test, and p value based on 1-tailed test, respectively,
appear beneath the independent variable parameter
estimates.

** The coefficient estimate on intercept (b,,) is not reported.

5 Significant at 5% level based on 29tailed tests.

73

Table 3

Effect of dividend, q ratio, and beta on the
market response to unexpected earnings:
OLS estimation with 4972 firm-quarters from 1983.4-1986.2*
AR,, - b,, + b,S,, + sz,, + 1330,, + b‘B,, + u,,**

Dependent variable: Abnormal return (ARR)

 

 

Independent Variables Regression Statistics
Earnings Dividend Q ratio Beta F value R-square MSE
surprise scaled (QR) (BR) (Prob>F) (Adjusted)
scaled by ‘
by price
price (DR)

(s,,)

 

0.0082# '0.1561# 0.0010# -0.0008 8.413 0.0067 0.00040
0.0018 0.0590 0.0004 0.0006 (0.0001) (0.0059)

4.335 -2.646 2.033 -1.292

0.0001 0.0082 0.0421 0.1965

0.0000 0.0041 0.0210 0.0982

 

* Standard errors, t-statistics, p value based on 2-tailed
test, and p value based on 1-tailed test, respectively,
appear beneath the independent variable parameter
estimates.

** The coefficient estimate on intercept (b,,) is not reported.

# Significant at 5% level based on 2-tailed tests.

 

74

5.2. Separate Analyses on Positive and Negative Surprises

The theoretical model of Chapter 3 reveals an asymmetric
price response to positive and negative earnings
announcements. This study analyzes positive and negative
earnings surprises separately. The entire sample is divided
into three groups according to the signs of earnings
surprises: positive (2126 observations), negative (2633»
observations), and zero (213 observations) surprise groups.
The zero surprise group (Sn - 0) is of no interest in this
study and is ignored. Summary statistics for positive and
negative groups are reported in Tables 4 and 5, respectively.
A matched-sample t-statistic comparison of means is used to
examine if the firm characteristics are different for the two
groupsc3 Results (see Figure 4) provide evidence that
dividend and beta are not significantly different for the two

groups. Q ratio, however, shows a significant difference: q

ratio is significantly higher for the positive group.

 

3 The t-statistic is computed as follows:

t - ------------------------- ,

Sp * /(1/“1) 3’ (1/32)

s - /((n,-1)*S,z+ ( -1)*sf)/(n,+n,-2)

M: - mean of a variab e for the positive group,
'42

S

 

 

- mean of a variable for the negative group,
, - standard deviation of a variable for the
positive group,
Sz=- standard deviation of a variable for the
negative group,
In a observations for the positive group,
In - observations for the negative group,

75

 

 

 

 

Variable Matched-sample t-statistic Decision
Dividend -1.733 Not different
scaled by

price

q ratio 57.528 Different
Beta 0.617 Not different

 

 

 

 

 

Figure 4. Comparison of positive and negative groups

Predictions on the coefficients of the regressors with
positive unexpected earnings are summarized as follows:
+ - + -
Rjt g f( sit! Dig! ngr ng)° (5.1)

First, stock returns (R,t) are regressed on earnings surprises

alone:

Results are reported in Table 6 (Panel A). As predicted, the
estimate of coefficient b, is positive and statistically
significant (at 2% level): the higher the positive earnings
(good news), the higher the stock return. The same regression
is repeated with abnormal return (Panel A of Table 7).
Results are basically the same whether the dependent variable
is raw return or abnormal return. Equation (5.2) is the base

case, and as firm characteristics (DR, 0", and/or B“) are

76
added to this equation, partial F-statistics‘ are examined to
investigate the incremental importance of each

characteristic.s

In order to see if a variable is an important factor, the

following equation is estimated:

where X“ represents one of the factors (DR, 0", and B”).

Calculation of a partial F-statistic based on Equations (5.2)
and (5.3) reveals the incremental contribution of the variable
X" given S". Similarly, to see the incremental contribution

of a variable Y" given 8" and X", the following equation is

 

‘ The partial F-statistic is computed as follows:

(SSE, - sszu) / k
r- ----------------------- ,
sszu/ (n-l-h-k)

where SSE, - sum of squared errors for a restricted model,
-a? + a,:q + ...... + apq,
SSE - sum 0 squared errors for an unrestricted model,

y - a0 + a,x, + ...... + aux, + and“, + ......

Tami
k - number of variables being tested,
h - number of variables not being tested,
n - number of observations.

3 This procedure is similar to a stepwise regression, which
is used to determine which variable to include in a model.
It is, however, different from the stepwise regression in that
the partial F test in this study is used only to determine the
relative importance of each factor determined by the model
while the stepwise regression is adopted to choose independent
variables from a number of possible candidates without
theoretical relationship. See Berenson, Levine, and Goldstein
[1983] for stepwise regression.

 

77

employed:
R,, - b,, + b,S,, + bZX,, + b3Y,, + u,,, (5.4)

where Y,, represents one of the factors (0,, Q,,, or B,,)
differing from X". A partial F-statistic based on Equations
(5.3) and (5.4) shows if YB is an important factor given
the S" and X“. For the same purpose, the following equation

is adopted:

where z" differs from either of X“ and Y". The results from
estimating Equations (5.3), (5.4), and (5.5) are reported in
Table 6 (Table 7 with abnormal return). Using all these
results, partial F-statistics are computed to examine the
incremental contribution of each factor given positive

earnings surprise and/or other variables.

Figure 5 summarizes the incremental contribution of each
variable given a positive earnings surprise by presenting
partial F-statistics with corresponding p values and adjusted
R-squeres for unrestricted equations.6 :Each cell in Figure 5
contains the independent variables for an unrestricted model.

The factor being tested for its incremental contribution is

 

3 Figure 5 is developed based on the results with raw
returns: Table 6. Partial F-statistics and adjusted R-squares
are basically the same whether raw returns or abnormal returns
are used.

 

 

78

 

 

 

 

 

 

 

 

Equation Equation ’ Equation Equation
(5.2) (5.3) (5.4) (5.5)
8"! D"! Q" s"! D"! Q"! B].
Adj R2 - 0.0105 Adj Rz - 0.0106
s,,, 12,, r - 2.039 r - 1.102
P - 0.1837 R - 0.3130
Adj R2 .............. .. ................
r - 16.995 2 z
p =- 0.0001 Adj R - 0.0099 Adj R - 0.0106)
r - 0.605 r - 2.535
P - 0.4566 R - 0.1188
s"! Q"! D" s"! 0": Di" B"
5,, A43 R2 - 0.0105 Adj Rz - 0.010
2 s,,, 0,, r - 12.512 r - 1.102
Adj R P - 0.0004 F - 0.3130
:- 0.0026 Adj R2 --------------- .. ----------------
F -0.0061 8.0.8 s,0,a,n,
.. 6.510' r‘ - 6.496 1' " 1' n n it I
p b P - 0.0112 Adj R2 - 0.0048 Adj Rz - 0.0106
- 0.0108 F - 0.164 r - 13.449
P - 0.7060 R - 0.0003
----------------------------- q)--------—---——--
31:! 31:! 9:: SW 31:! Dlt' 9::
Adj R2 - 0.0099 Adj Rz - 0.0106)
F - 17.543 3 - 2.535
s,,, 3,, p - 0.0001 F - 0.1188
--------------- q)-------—---—----
2
Adj 3 0 0°21 Sig! B", Q]: s"! B"! Q“! D]:
r‘ - 0.054 Adj Rz - 0.0048 Adj.Rz - 0.0106
F - 0.8337 F - 6.603 r - 13.449
P - 0.0100 F - 0.0003
' F' - partial F-statistic
b P a p-value

Figure 5.

Contribution of each of dividend, q ratio, and
beta: Positive earnings surprises

79
underlined, so that a restricted model includes only the
independent variables which are not underlined. Below the
independent variables are an adjusted R-square for the
unrestricted model, a partial F-statistic for the underlined
variable, and the p-value for the F-statistic. Correlation

coefficients among the variables are reported in Table 24.

Panel D of Table 6 shows that the coefficient estimates
of all independent variables have the same sign as predicted.
The coefficients on q ratio 0" and beta 3" are not
significantly different from zero. IHowever, the coefficients
on earnings surprise S" and dividends D“ are statistically
significant at the 2% and 1% level, respectively. This is
vividly demonstrated in Figure 5, which suggests that only
dividend policy has a significant impact on share price
response to positive earnings surprises. Firms with higher
dividend yields are likely to have a smaller response to a

positive earnings surprise.

Figure 5 allows a comparison of the relative explanatory
power of q ratio and dividend policy. Both variables are
significant at the 2% level when included with earnings
surprise alone. The two variables are correlated (See Table
10) and q ratio does not add significant explanatory power to
a regression which already includes dividend. However, adding
dividend to earnings surprise and q ratio adds additional

explanatory power at the 1% significance level. Dividend

80
policy dominates q ratio as a determinant of share price

response to positive earnings surprises.

The result for systematic risk conflicts with the
findings of Collins and Kothari [1989]. Using a reverse
regression, these authors find that beta has a negative impact
on the return/surprise relation and that the effect is
statistically significant at the 5% level (based on a 2-tailed
test). Several interpretations are possible. First, Collins
and Kothari use a reverse regression which may produce
spurious results. Collins and Kothari [1989] adopt the

following reverse regression:
8,, - a0 + a,R,, + azR,,*B,, + a,R,,*MB,, +u,,. (5.6)

If the relationship between and S" and R" is strong enough to
dominate an effect of beta on return/surprise relation, the
coefficient a2 can be positive (negative association between
beta and return/surprise relation) and significant regardless
of the effect of beta. Second, Collins and Kothari's reverse
regression is run on a sample of both positive and negative
earnings surprises. Third, the empirical models of this

and/or Collins-Kothari and the theoretical model of Chapter

3 could be misspecified.

For negative earnings surprises, predictions on the
coefficients of the regressors are the following:

+ - - +
R]: I f( S"! big! ngo Bjdo (5.7)

81

 

 

 

Equation Equation Equation Equation
(5.2) (5.3) (5.4) (5.5)
s"! D"! Q," 81:! ngl ngl a]:
Adj Rz - 0.0038 Adj R2 =- 0.0046
s,,, 12,, r - 1.442 F - 3.099
2 P - 0.2368 R - 0.0827
Adj R --------------- I ---------------
F. : 888:6 Sin D,,, B" sjt' Djt' 3):! Q):
R - 0.95 Adj R2 - 0. 0042 Adj Rz - 0. 0046
r' - 4. 523 R' - 2. 144
R - 0. 0355 R - 0.1610
............................. q) --------------u
8]" Q"! D]: S"! Q"! Dig! a):
5,, Adj R2 - 0.0038 Adj R2 - 0.0046
2 s,,, 0,, F - 0.042 r - 3.099
Adj R 2 R - 0.0004 R - 0.0827
.- 0.0040 Adj R --------------- -» ---------------
F - 0.0042 8 p Q ' a S I Q I B I D
- 11.637' I" - 1.400 n 1' n It "2 n n
R R - 0.2413 Agj Rz - 0.0050 Adj R - 0. 0046
- 0.0007” r - 3.079 F - 0. 021
R - 0.0719 R - 0. 8912
----------------------------- 1+ ---------------
Sn! 3n: 9n Sn! 3n! D", 9n
Adj Rz - 0.0042 Adj Rz - o. 0046
I" - 0. 148 F - 2.144
s,,, 8,, R - 0. 7170 R - 0.1610
2 D --------------- d) ----------------
Adj R S"! B)" Q" s"! Blt’ Q"! Dig
. - 0.0045 2 2
F - 2.250 Adj R - o. 0050 Adj R - 0.0046
R - 0.1496 F - 2. 272 r - 0.021
R - 0.1472 R - 0.8912

 

 

 

;F' a partial F-statistic
bP - p-value

Figure 6.

 

 

Contribution of each of dividend, q ratio, and
beta: Negative earnings surprises

82

The coefficient estimates are reported in Tables 8 and 9.
Figure 6 summarizes the empirical findings in a format like
that of Figure 5. Figure 6 demonstrates that no variable has
a significant coefficient except earnings surprise. The
coefficient of beta is relatively significant (at 8% in Panel
D of Table 8).7 But the relation is opposite that predicted,
which may be consistent with Jain [1982].3 For g ratio, the
coefficient estimate is insignificant with a positive sign.
Only the dividend has the predicted sign. Correlation

coefficients among the variables are reported in Table 11.

Interestingly, the coefficient on positive earnings
surprise (0.0387 in Panel D of Table 6) is significantly
different from that on negative earnings surprise (0.0065 in
Panel D of Table 8) .9 A comparison of other panels in Tables

6 and 8 leads to the same conclusion. This implies that the

 

7 Figure 6 is built based on the results with raw returns:
Table 7. Partial F-statistics and adjusted R-squares are
basically the same whether raw returns or abnormal returns
are used.

3 Jain [1982] empirically finds that given a positive
(negative) abnormal return, the higher the variance of a stock
return, the higher (lower) the return. If beta is positively
related to variance of stock return, it is predicted that the
higher the beta, the lower the stock return with negative
unexpected earnings. However, it is not determined whether
the negative coefficient supports Jain [1982] since it is
significant merely at 8%.

9 A two-sample t-statistic is used to investigate whether
the coefficient on positive earnings surprise is significantly
different from that on negative earnings surprise (see
footnote 4). The t-statistic is equal to 107.812.

83
equity market response to earnings announcements is
asymmetric. The results indicate that the equity market, on
average, responds more strongly to positive earnings surprises

than to negative earnings surprises.

To summarize, the model developed in Chapter 3 is
supported by the positive earnings surprise group but is not
supported by the negative surprise group. The results provide
some evidence that dividend policy is an important factor to
be considered in return/surprise studies. However, whether
dividend impacts the relation between earnings surprise and
stock return should be determined after additional analyses.
Dividend may be a proxy for factors suggested by other studies
such as firm size, growth, or earnings predictability.
Importantly, the separate analyses provide evidence of
asymmetric market response to unexpected earnings: positive
earnings surprises yield.more response than negative earnings

surprises.

84

Table 4

Descriptive statistics: Positive earnings surprises

(2126 firm-quarters)

 

 

Variable N Mean Standard Minimum Maximum
Deviation

Return 2126 0.00454 0.01914 -0.10039 0.10351

Abnormal 2126 0.00368 0.01824 -0.10536 0.09527

return

Unexpected 2126 0.00964 0.02642 0.00001 0.48380

earnings

scaled by

price

Dividend 2126 0.00661 0.00507 0.00000 0.05828

scaled by

price

q ratio 2126 0.95692 0.65056 0.03598 7.65163

Beta 2126 0.92337 0.44839 -0.29091 2.83044

 

85

Table 5

Descriptive statistics: Negative earnings surprises

(2633 firm-quarters)

 

 

Variable N Mean Standard Minimum Maximum
Deviation

Return 2633 -0.00430 0.02175 -0.l7135 0.30384

Abnormal 2633 -0.00438 0.02080 -0.16724 0.30464

return

Unexpected 2633 -0.03290 0.20417 -6.02842 -0.00001

earnings

scaled by

price

Dividend 2633 0.00687 0.00505 0.00000 0.03232

scaled by

price

q ratio 2633 0.86660 0.54931 0.02784 6.58644

Beta 2633 0.92842 0.45681 -0.55150 2.87790

 

86
Table 6
Market response to positive unexpected earnings:

2126 firm-quarters from 1983.4-1986.2*
(Dependent variable: Raw return (Rn))

 

 

Independent Variables Regression Statistics
Earnings Dividend 0 ratio Beta F value R-square MSE
surprise scaled (QM) (Bk) (Prob>F) (Adjusted)
scaled by
by price
price (Du)

(Sit)

 

A. R,, =- b,, + 1:),8,t + uh“:

0.0387# 6.510 0.0031 0.00036
0.0160 (0.0108) (0.0026)

2.411

0.0160

0.0080

8. R,, - b,, + b,S,, + bZX,, + u,,**, X belonging to (D, Q, B}

0.0338# -0.3369# 11.768 0.0110 0.00036
0.0157 0.0817 (0.0001) (0.0100)

2.151 -4.120

0.0316 0.0001

0.0108 0.0000

0.0472# 0.00165 6.502 0.0061 0.00036
0.0159 0.0006 (0.0015) (0.0052)

2.964 2.545

0.0031 0.0110

0.0015 0.0055

0.04008 0.0001 3.276 0.0031 0.00036
0.0157 0.0009 (0.0380) (0.0021)

2.553 0.213

0.0108 0.8312

0.0054 0.4156

 

* Standard errors, t-statistics, p value based on 2-tailed
test, and p value based on 1-tailed test, respectively,
appear beneath the independent variable parameter
estimates.

** The coefficient estimate on intercept (b,,) is not reported.

# Significant at 5% level based on 2-tailed tests.

 

87
Table 6 (cont'd.)

 

 

Independent Variables Regression Statistics
Earnings Dividend 0 ratio Beta F value R-square MSE
surprise scaled (QR) (BR) (Prob>F) (Adjusted)
scaled by
by price
price (DR)

(5].)

 

C: R): " 1’o + 1’13): 1' bZX,, + bBYit 3' (1,“,

t
X and belonging to (D, Q, B}
0.0386# -0.3019# 0.0009 8.530 0.0119 0.00036
0.0160 0.0853 0.0006 (0.0001) (0.0105)

2.404 -3.538 1.429
0.0163 0.0004 0.1532
0.0081 0.0002 0.0766

0.0334# '0.3525# -0.0007 8.047 0.0112 0.00036
0.0157 0.0841 0.0009 (0.0001) (0.0099)

2.125 -4.188 -0.781

0.0337 0.0001 0.4351

0.0168 0.0000 0.2175

0.04744 0.00173 '0.0003 4.386 0.0062 0.00036
0.0159 0.0006 0.0009 (0.0046) (0.0048)

2.976 2.567 -0.398

0.0030 0.0103 0.6907

0.0015 0.0051 0.3453

D“ R]: ' l5’0 "' b1S): 1' b2”): 3' h3°): “' P43): "’ “1:“

0.03875 '0.31858 0.0010 -0.0010 6.677 0.0124 0.00036
0.0160 0.0867 0.0006 0.0009 (0.0001) (0.0106)

2.411 -3.670 1.596 -1.056

0.0160 0.0002 0.1106 0.2910

0.0080 0.0001 0.0553 0.1455

 

* Standard errors, t-statistics, p value based on 2-tailed
test, and p value based on 1-tailed test, respectively,
appear beneath the independent variable parameter
estimates.

** The coefficient estimate on intercept (b,,) is not reported.

# Significant at 5% level based on 2-tailed tests.

88
Table 7
Market response to positive unexpected earnings:

2126 firm-quarters from 1983.4-1986.2*
(Dependent variable: Abnormal return (ARn))

 

 

Independent Variables Regression Statistics
Earnings Dividend 0 ratio Beta F value R-square MSE
surprise scaled (on) (8") (Prob>F) (Adjusted)
scaled by
by price
price (DR)

(Sn)

 

0.0421: 7.960 0.0037 0.00033
0.0149 (0.0048) (0.0033)

2.821

0.0048

0.0024

8. AR,, - b,, + b,S,, + bZX,, + u,,**, X belonging to (D, Q, 8}

0.03688 -0.2866# 10.766 0.0100 0.00032
0.0149 0.0779 (0.0001) (0.0091)

2.461 '3.678

0.0139 0.0002

0.0069 0.0001

0.04758 0.0012! 5.996 0.0056 0.00033
0.0151 0.0006 (0.0025) (0.0047)

3.133 2.005

0.0018 0.0451

0.0009 0.0225

0.04228 0.0002 4.018 0.0038 0.00033
0.0149 0.0008 (0.0181) (0.0028)

2.823 0.281

0.0048 0.7785

0.0024 0.3892

 

* Standard errors, t-statistics, p value based on 2-tailed
test, and p value based on 1-tailed test, respectively,
appear beneath the independent variable parameter
estimates.

** The coefficient estimate on intercept (b,,) is not reported.

# Significant at 5% level based on 2-tailed tests.

 

89
Table 7 (cont'd.)

 

Independent Variables

Regression Statistics

 

 

Earnings Dividend Q ratio Beta F value R-square MSE
surprise scaled (QR) (Bu) (Prob>F) (Adjusted)
scaled by
by price
price (DR)
(8,.)
C. AR =ll’b -+1:s., 4-13X +- Y +
n o 1 1t 2 It b3Xi‘andu i belonging to (D, Q, 8)
0.0400# -0.2633# 0.0006 7.508 0.0105 0.00032
0.0153 0.0813 0.0006 (0.0001) (0.0091)
2.616 -3.237 0.996
0.0090 0.0012 0.3194
0.0045 0.0006 0.1597
0.0365# -0.2980# -0.0005 7.296 0.0102 0.00032
0.0149 0.0802 0.0009 (0.0001) (0.0088)
2.440 -3.715 -0.603
0.0148 0.0002 0.5467
0.0074 0.0001 0.2733
0.04768 0.00124 -0.0001 4.009 0.0056 0.00033
0.0151 0.0006 0.0009 (0.0076) (0.0042)
3.138 1.995 -0.l97
0.0017 0.0462 0.8439
0.0008 0.0231 0.4219
D. AR,, - b,, + b,S,, + sz,, + b,Q,, + b,B,, + u,,**
0.04014 '0.2752# 0.0007 -0.0007 5.788 0.0108 0.00032
0.0153 0.0827 0.0006 0.0009 (0.0001) (0.0089)
2.621 -3.327 1.123 -0.796
0.0088 0.0009 0.2615 0.4262
0.0044 0.0004 0.1307 0.2131

 

* Standard errors, t-statistics, p value based on 2-tailed
test, and p value based on 1-tailed test, respectively,
appear beneath the independent variable parameter
estimates.

** The coefficient estimate on intercept (b,, ) is not reported.

# Significant at 5% level based on 2-tailed tests.

90
Table 8
Market response to negative unexpected earnings:

2633 firm-quarters from l983.4-l986.2*
(Dependent variable: Raw return (Rn))

 

 

Independent Variables Regression Statistics
Earnings Dividend Q ratio Beta P value R-square MSE
surprise scaled (QR) (BR) (Prob>F) (Adjusted)
scaled by
by price
price (DR)

(8,.)

 

A. R,, - bo + b,s,t + “it”.

0.0070: 11.637 0.0044 0.00047
0.0020 (0.0007) (0.0040)

3.411

0.0007

0.0080

8. R,, - b,, + b,S,, + bzX,, + u,,**, X belonging to (D, Q, 8}

0.00708 -0.0058 5.819 0.0044 0.00047
0.0020 0.0839 (0.0030) (0.0036)

3.405 -0.069

0.0007 0.9447

0.0003 0.4723

0.0067# 0.0009 6.527 0.0049 0.00047
0.0020 0.0007 (0.0015) (0.0042)

3.235 1.190

0.0012 0.2343

0.0006 0.1271

0.00698 -0.0013 6.950 0.0053 0.00047
0.0020 0.0009 (0.0010) (0.0045)

3.334 -l.502

0.0009 0.1331

0.0004 0.0665

 

* Standard errors, t-statistics, p value based on 2-tailed
test, and p value based on 1-tailed test, respectively,
appear beneath the independent variable parameter
estimates.

** The coefficient estimate on intercept (b,,) is not reported.

# Significant at 5% level based on 2-tailed tests.

91
Table 8 (cont'd)

 

Independent Variables

Regression Statistics

 

Earnings
surprise

scaled
by
price
(8,.)

Dividend Q ratio Beta MSE
scaled (Q,,) (B,,)
by

price

(0,.)

F value R-square
(Prob>F) (Adjusted)

 

it
tbelonging to (D, Q, B)

X and

0.0067# 0.0138 0.0009 4.359 0.0049 0.00047
0.0021 0.0855 0.0007 (0.0047) (0.0038)

3.200 0.162 1.198

0.0014 0.8710 0.2308

0.0007 0.4355 0.1154

0.00693 -0.0323 -0.0014 4.679 0.0053 0.00047
0.0020 0.0856 0.0009 (0.0031) (0.0042)

3.352 -0.378 -1.547

0.0008 0.7053 0.1219

0.0004 0.3526 0.0609

0.00644 0.0011 -0.0016 5.392 0.0061 0.00047
0.0020 0.0007 0.0009 (0.0012) (0.0050)

3.096 1.507 -1.764

0.0020 0.1320 0.0779

0.0010 0.0660 0.0389
D. R,, - b,, + b,S,, + sz,, + 1339,, + b,B,, + u,,**

0.00654 -0.0120 0.0011 -0.0016 4.047 0.0061 0.00047
0.0021 0.0867 0.0008 0.0009 (0.0028) (0.0046)

3.095 -0.139 1.465 -1.762

0.0020 0.8894 0.1431 0.0782

0.0010 0.4447 0.0715 0.0396

 

Standard errors, t-statistics, p value based on 2-tailed
test, and p value based on 1-tailed test, respectively,
appear beneath the independent variable parameter
estimates.
** The coefficient estimate on intercept (b,,) is not reported.
# Significant at 5% level based on 2-tailed tests.

 

92
Table 9

Market response to negative unexpected earnings:
2633 firm-quarters from 1983.4-1986.2*

(Dependent variable: Abnormal return (ARn))

 

Independent Variables

Regression Statistics

 

Earnings
surprise

scaled
by
price
(8,.)

Dividend Q ratio Beta
scaled (Q,,) (B,,)
bY

price

(Dig)

F value
(Prob>F) (Adjusted)

R-square MSE

 

A. AR“ -

B. AR" =

0.0043#
0.0020
2.173
0.0298
0.0149

0.0043#
0.0020
2.169
0.0302
0.0151

0.00415
0.0020
2.063
0.0392
0.0196

0.00435
0.0020
2.138
0.0326
0.0168

b,, + b,S,, + 132x,t + ujt**'

-0.0030

0.0815

-0.037

0.9702
0.4856

0.0005
0.0007
0.739

0.4597
0.2298

-0.0005 2.577

4.723
(0.0298) (0.0014)

2.361
(0.0945) (0.0010)

2.635
(0.0719) (0.0012)

0.0018 0.00044

X belonging to (D, Q, 8)

0.0018 0.00044

0.00044

0.0020

0.0020 0.00044

0.0009 (0.0762) (0.0012)

-O.657
0.5113
0.2556

 

*

Standard errors, t-statistics, p value based on 2-tailed
test, and p value based on 1-tailed test, respectively,
appear beneath the independent variable parameter
estimates.
** The coefficient estimate on intercept (b,,) is not reported.
# Significant at 5% level based on 2-tailed tests.

 

93

Table 9 (cont'd)

 

Independent Variables

Regression Statistics

 

 

Earnings Dividend Q ratio Beta F value R-square MSE
surprise scaled (QR) (BR) (Prob>F) (Adjusted)
scaled by

by price

price (DR)

(8,.)

C. AR 8 b -+iDS. + bi! -+ Y + u **

It 0 1 It 2 1' b3thand it belonging to (D, Q, 8}
0.00414 0.0088 0.0005 1.760 0.0020 0.00044
0.0020 0.0830 0.0007 (0.1509) (0.0009)

2.040 0.107 0.746
0.0414 0.9150 0.4557
0.0207 0.4575 0.2278
0.00434 -0.0143 -0.0006 1.727 0.0020 0.00044
0.0020 0.0832 0.0009 (0.1574) (0.0008)
2.144 -0.l73 -0.678
0.0321 0.8630 0.4978
0.0160 0.4315 0.2489
0.0040# 0.0006 -0.0007 1.977 0.0023 0.00044
0.0020 0.0007 0.0009 (0.1136) (0.0011)
1.996 0.882 -0.814
0.0460 0.3779 0.4158
0.0230 0.1889 0.2079
D. AR,, - b,, + b,S,, + 8,0,, + 830,, + 8,8,, 4» 8,,“
0.00408 -0.0026 0.0006 -0.0007 1.482 0.0023 0.00044
0.0020 0.0843 0.0007 0.0009 (0.2048) (0.0007)
1.990 -0.032 0.865 -0.807
0.0466 0.9745 0.3870 0.4195
0.0233 0.4872 0.1935 0.2097

 

* Standard errors, t-statistics, p value based on 2-tailed
test, and p value based on 1-tailed test, respectively,
appear beneath the independent variable parameter
estimates.

** The coefficient estimate on intercept (b,,) is not reported.

# Significant at 5% level based on 2-tailed tests.

 

94

Table 10

Pearson correlation coefficients:
2126 firm-quarters from 1983.4-1986.2
with positive earnings surprises

 

 

Raw Earnings Dividend q ratio Beta

return surprise scaled by (QR) (Bk)

(Rn) scaled by price

price (DR)
(8,.)

Raw 1.00000 0.05528 -0.09386 0.04443 0.00414
return (0.0000) (0.0108) (0.0001) (0.0405) (0.8486)
(8,.)
Earnings 0.05528 1.00000 -0.09656 -0.l7682 -0.00864
surprise (0.0108) (0.0000) (0.0001) (0.0001) (0.6905)
scaled by
price
(8,.)
Dividend -0.09386 -0.09656 1.00000 -0.26419 -0.23412
scaled by (0.0001) (0.0001) (0.0000) (0.0001) (0.0001)
price
(Di!)
q ratio 0.04443 -0.l7682 -0.26419 1.00000 0.23367
(QR) (0.0405) (0.0001) (0.0001) (0.0000) (0.0001)
Beta 0.00414 -0.00864 -0.23412 0.23367 1.0000
(8") (0.8486) (0.6905) (0.0001) (0.0001) (0.0000)

 

95

Table 11

Pearson correlation coefficients:
2633 firm-quarters from 1983.4-1986.2
with negative earnings surprises

 

 

Raw Earnings Dividend q ratio Beta

return surprise scaled by (QR) (8")

(RR) scaled by price

price (Du)
(sit)

Raw 1.00000 0.06636 0.00390 0.03130 -0.03246
return (0.0000) (0.0007) (0.8416) (0.1083) (0.0959)
(8,.)
Earnings 0.06636 1.00000 0.07899 0.12576 -0.04939
surprise (0.0007) (0.0000) (0.0001) (0.0001) (0.0113)
scaled by
price
(8,.)
Dividend 0.00390 0.07899 1.00000 -0.18025 -0.20354
scaled by (0.8416) (0.0001) (0.0000) (0.0001) (0.0001)
price
(0,.)
q ratio 0.03130 0.12576 -0.18025 1.00000 0.18401
(9n) (0.1083) (0.0001) (0.0001) (0.0000) (0.0001)
Beta -0.03246 -0.04939 -0.20354 0.18401 1.0000
(BR) (0.0959) (0.0113) (0.0001) (0.0001) (0.0000)

 

Chapter 6. An Extension to Other Factors
6.1. Dividends versus Firm Size

The differential information hypothesis (Freeman [1987])
suggests that differential information affects stock returns
at the time of an earnings announcement. Atiase [1985] and
Freeman [1987] argue that firm size is a good proxy for
differential information such that the absolute value of stock
price change is negatively related to firm size. In the
context of this study, stock return should be negatively
related to firm size given the positive unexpected earnings
as well as positively related given the negative unexpected

earnings.

To investigate whether the effect of firm size is
supported on the basis of the sample, stock return is
regressed on positive earnings surprise and firm size. The
coefficient estimate on firm size is predicted negative. As
a proxy for firm size, market value of common stock is used.
Following Atiase [1985] and Collins and Rothari [1989],
natural logarithm is adopted so that larger market values are
stated as percentage increase. The following equation is

estimated:
where F" is the natural logarithm of the market value of

96

97
common stock. Regression results are found in Panel A of
Table 12 (Table 13 with abnormal return). As expected, the
coefficient estimate on firm size is negative. On the basis
of the sample, the information hypothesis is confirmed for

positive earnings surprises.

On average, large firms maintain higher payout ratios
than small firms. A dividend effect could be found merely
because dividend is positively related to firm size. In other
words, dividend may be a proxy for firm size. Therefore, the
analysis should be controlled for firm.size to see‘whether'the
dividend effect found in Chapter 5 essentially reflects
differential information or a firm size effect. One effective
way to control for firm size is to include dividend and firm
size simultaneously in a regression (see Jain [1982]). The

following equation is adopted for this purpose:

The results are reported in Panel B of Table 12 (Table 13 with
abnormal return). The coefficient of dividend is still
negative and significant. On the contrary, the coefficient
on firm.size is no longer significant although it has the same
sign as predicted. Comparing Panels A and B of Table 12
results in the conclusion for positive surprises that, given
dividend, firm size is not an important explanatory factor.
That is, the effect of firm size on the return/surprise
relation is dominated by the dividend effect. Figure 7

 

98

 

 

Equation Equation Equation
(5.2) (5.3) (5.4)
Sn! Du 5n: On! in
Adj R2 - 0. 0100 Adj Rz - 0.0101
F-16.97l r-1.074
8,, R - 0. 0001 R - 0. 3210
Adj R2 =- 0.0026 --------------------------------------- 1
r =- 6. 510' 5
R - 0. 0108 s,,, 2,, s,,, F,,, 12,,
, Adj R2 =- 0.0046 Adj R2 = 0.0101
F* a 5.307 r' - 12.726
R - 0.0223 R - 0. 004

 

 

 

 

 

: F. - partial F-statistic
P - p-value

Figure 7. Contribution of each of dividend and
firm size: Positive earnings surprises
summarizes the incremental explanatory power of dividend and

firm size.

The same procedure is repeated with the negative earnings
surprise group. iResults in.Tables 14 and 15 (Panel A) support
the differential information hypothesis. The coefficient on
firm size is positive and significant as expected. The
results of estimating Equation (6.2) reported in Tables 14 and
15 (Panel B) also show that the coefficient on firm size is
highly significant even after including dividend in the

regression. This is summarized in Figure 8.

As an additional analysis, this study investigates

 

99

 

 

 

 

Adj R2 - 0.0046
F* - 13.366
R - 0.0003

 

 

Equation Equation Equation
(5-2) (5.3) (5.4)
8“, DR snv ”8' in
Adj Rz - 0.0036 Adj R2 - 0.0091
F - 5.819 r - 15.355
5" R - 0.0030 R - 0.0001
Adj R2 =- 0.0040 ----------------------------------------
I
F - 11.637,

Adj R2 - 0.0091
F - 1.983
R - 0.1583

 

 

 

' F. a partial F-statistic
b P - p-value

Figure 8. Contribution of each of dividend and
firm size: Negative earnings surprises
whether the effect of firm size found in this section is in
fact a differential information effect. One way of
controlling for differential information is to examine those
firms with a minimal set of public information available.
Assuming that the Wall Street Journal is the source of public

information,1 firms for which the Wall Street Journal

publishes other information than earnings and dividend

 

' Several studies such as Grant [1980] use the Wall Street
Journal to investigate differential information about firms
based on the implicit assumption that the Wall Street.Journal
is the public information source.

100
announcements are eliminated.2 In so doing, firms in the
reduced sample should have an identical information
environment. This procedure causes a reduction in
sample size from 2126 to 830 observations for the positive
group and from 2633 to 1001 observations for the negative
group. This procedure does not affect the variation in firm
size which ranges from about $9 million to $5 billion. Only
the information environment is controlled. Descriptive
statistics for each reduced sample are presented in Table 16

(positive group) and Table 17 (negative group).

First, stock return is regressed on earnings surprise and
firm size using Equation (6.1). Secondly, Equation (6.2) is
estimated based on the reduced sample. Panel A of Table 18
(Table 19 with abnormal return) reveals that firm size is no
longer important given the positive unexpected earnings when
the information environment is controlled. Panel A of Table
20 (Table 21 with abnormal return) also shows that firm size
is not an important factor given negative unexpected earnings
when the information environment is controlled. This supports
the differential information hypothesis. However, the
coefficient on dividend is not affected by the information
control. The dividend effect with positive earnings surprise
is still significant after the information control. These

 

2 The reason is that earnings and dividend announcements
are the minimum amount of information published in the Wall
Street Journal Index.

 

101

results are presented in Tables 18 through 21 (Panel 8).

Figures 9 and 10 are produced based on these results.

 

—q)

 

 

 

 

Equation Equation Equation
(5.2) (5.3) (5.4)
31:! 9): 3):! ”1:! 5):
Adj R2 - 0.0391 Agj R2 - 0.0387
R - 6.549 R - 0.657
5,, R - 0.0106 R - 0.4181
Adj Rz - 0.0338 -------------------------------------- a
F - 28.969'b
R =- 0.0001 s,,, E,. s,,, R,,, 12,,
Adj R2 - 0.0317 Adj R2 - 0.0387
F* - 0.217 R - 6.862
R - 0.6362 R - 0.0083

 

' F. - partial F-statistic

b P a p-value

Figure 9.

Contribution of each of dividend and

firm size: Positive earnings surprises
(based on the reduced sample)

 

102

 

 

 

l
I Equation Equation Equation
(5-2) (5.3) (5.4)
Sm 9): 3w Dlt' 2):
Adj R2 - 0. 0076 Adj R2 - 0.0078
R' - 0. 655 R - 1.196
S,. P - 0.4190 R - 0. 2712
Adj R2 :- 0. 0080 ....................................... ,
R - 9. 014'
R - 0. 0027b s,,, 2,, S,.. P‘,.) 12,.
Adj Rz - 0.0077 Adj R2 - 0. 0078
F* - 0.713 R - 1.138
R - 0.3997 R - 0. 2826

 

 

 

 

 

' bF - partial F-statistic
bP -p-value

Figure 10. Contribution of each of dividend and
firm size: Negative earnings surprises
(based on the reduced sample)

103

Table 12

Effect of dividend and firm size on the
market response to positive unexpected earnings:
2126 firm-quarters from 1983.4-1986.2*
(Dependent variable: Raw return (Rn))

 

 

Independent Variables Regression Statistics
Earnings Dividend Natural F value R-square MSE
surprise scaled logarithm of (Prob>F) (Adjusted)
scaled by firm size
by price (FR)
price (DR)

(8,.)

 

0.0316# -0.0006# 5.918 0.0055 0.00036
0.0161 0.0002 (0.0027) (0.0046)

1.964 -2.305

0.0496 0.0213

0.0248 0.0106

3' RI: ' be + btsit + b2'3): "' 33:5th + “1:“

0.0303 -0.3080# -0.0002 8.209 0.0115 0.00036
0.0160 0.0863 0.0002 (0.0001) (0.0101)

1.888 -3.567 -1.044

0.0592 0.0004 0.2967

0.0296 0.0002 0.1488

 

* Standard errors, t-statistics, p value based on 2-tailed
test, and p value based on 1-tailed test, respectively,
appear beneath the independent variable parameter
estimates.

** The coefficient estimate on intercept (b,,) is not reported.

4 Significant at 5% level based on 2-tailed tests.

 

104

Table 13

Effect of dividend and firm size on the
market response to positive unexpected earnings:
2126 firm-quarters from 1983.4-1986.2*
(Dependent variable: Abnormal return (ARR))

 

 

Independent Variables Regression Statistics
Earnings Dividend Natural F value R-square MSE
surprise scaled logarithm of (Prob>F) (Adjusted)
scaled by firm size
by price (FR)
price (DR)

(8,.)

 

0.0322# -0.0007# 8.085 0.0076 0.00033
0.0153 0.0002 (0.0003) (0.0066)

2.103 -2.861

0.0355 0.0043

0.0177 0.0021

B. AR,, - b,, + b,S,, + 1321),, + b,,?“ + unis

0.03128 '0.2396# -0.0004 8.239 0.0115 0.00032
0.0153 0.0822 0.0002 (0.0001) (0.0101)

2.040 -2.914 -1.778

0.0415 0.0036 0.0755

0.0207 0.0018 0.0377

 

* Standard errors, t-statistics, p value based on 2-tailed
test, and p value based on l-tailed test, respectively,
appear beneath the independent variable parameter
estimates.

** The coefficient estimate on intercept (b,,) is not reported.

# Significant at 5% level based on 2-tailed tests.

 

105

Table 14

Effect of dividend and firm size on the
market response to negative unexpected earnings:
2633 firm-quarters from 1983.4-1986.2*
(Dependent variable: Raw return (Rn))

 

 

 

Independent Variables Regression Statistics
Earnings Dividend Natural F value R-square MSE
surprise scaled logarithm of (Prob>F) (Adjusted)
scaled by firm size
by price (FR)
price (DR)

(8,.)

 

0.00604 0.00094 12.527 0.0094 0.00046
0.0028 0.0002 (0.0001) (0.0087)

2.893 3.656

0.0038 0.0003

0.0019 0.0001

0.0061 -0.1257 0.0011: 9.019 0.0102 0.00046
0.0020 0.0891 0.0002 (0.0001) (0.0091)

2.940 -1.411 3.919

0.0033 0.1583 0.0001

0.0016 0.0791 0.0000

 

* Standard errors, t-statistics, p value based on 2-tailed
test, and p value based on l-tailed test, respectively,
appear beneath the independent variable parameter
estimates.

** The coefficient estimate on intercept (b,,) is not reported.

# Significant at 5% level based on 2-tailed tests.

106

Table 15

Effect of dividend and firm size on the
market response to negative unexpected earnings:
2633 firm-quarters from 1983.4-1986.2*
(Dependent variable: Abnormal return (AR"))

 

 

Independent Variables Regression Statistics
Earnings Dividend Natural F value R-square MSE
surprise scaled logarithm of (Prob>F) (Adjusted)
scaled by firm size
by price (FR)
price (DR)

(8,.)

 

A. AR,, =- b(, + b,S,, + 1523,, 4- “It!"

0.0035 0.00088 7.414 0.0056 0.00044
0.0020 0.0002 (0.0006) (0.0049)

1.727 3.176

0.0843 0.0015

0.0421 0.0007

0.00358 -0.1041 0.00098 5.425 0.0062 0.00044
0.0020 0.0866 0.0002 (0.0012) (0.0050)

1.767 -1.202 3.396

0.0774 0.2295 0.0755

0.0387 0.1147 0.0007

 

* Standard errors, t-statistics, p value based on 2-tailed
test, and p value based on 1-tailed test, respectively,
appear beneath the independent variable parameter
estimates.

** The coefficient estimate on intercept (b,,) is not reported.

8 Significant at 5% level based on 2-tailed tests.

 

107

Table 16

Summary statistics: 830 firm-quarters

(All unexpected earnings are positive.)

 

 

Variable N Mean Standard Minimum Maximum
Deviation

Return 830 0.00639 0.02067 -0.10039 0.08526

Abnormal 830 0.00541 0.02004 -0.10536 0.08588

return

Unexpected 830 0.00822 0.01945 0.00001 0.32320

earnings

scaled by

price

Dividend 830 0.00540 0.00423 0.00000 0.02691

scaled by

price

Firm size* 830 181.31394 508.79328 10.54762 4535.99617

 

* Market value of common stock (in million)

 

108

Table 17

Summary statistics: 1001 firm-quarters
(All unexpected earnings are negative.)

 

 

Variable N Mean Standard Minimum Maximum
Deviation

Return 1001 -0.00489 0.02285 -0.12790 0.30384

Abnormal 1001 -0.00503 0.02258 -0.12436 0.30464

return

Unexpected 1001 -0.02073 0.13429 -3.67145 -0.00004

earnings

scaled by

price

Dividend 1001 0.00607 0.00478 0.00000 0.03232

scaled by

price

Firm size* 1001 165.62563 463.81707 9.93218 3958.45866

 

* Market value of common stock (in million)

 

 

109

Table 18

Effect of dividend and firm size on the
market response to positive unexpected earnings:
830 firm-quarters from 1983.4-1986.2*
(Dependent variable: Raw return (Rn))

 

 

Independent Variables Regression Statistics
Earnings Dividend Natural F value R-square MSE
surprise scaled logarithm of (Prob>F) (Adjusted)
scaled by firm size
by price (FR)
price (DR)

(8,.)

 

A. R,, - b,, + b,S,, + b,R,, + u,,**

0.2003# 0.0003 14.583 0.0341 0.00041
0.0377 0.0006 (0.0001) (0.0317)

5.301 0.473

0.0001 0.6362

0.0000 0.3686

0.19178 -0.44678 0.0005 12.123 0.0422 0.00041
0.0377 0.1689 0.0006 (0.0001) (0.0387)

5.073 -2.644 0.810

0.0001 0.0083 0.4181

0.0000 0.0041 0.2090

 

* Standard errors, t-statistics, p value based on 2-tailed
test, and p value based on 1-tailed test, respectively,
appear beneath the independent variable parameter
estimates.

** The coefficient estimate on intercept (b,,) is not reported.

8 Significant at 5% level based on 2-tailed tests.

 

110

Table 19

Effect of dividend and firm size on the
market response to positive unexpected earnings:
830 firm-quarters from 1983.4-1986.2*
(Dependent variable: Abnormal return (AR"))

 

 

Independent Variables Regression Statistics
Earnings Dividend Natural F value R-square MSE
surprise scaled logarithm of (Prob>F) (Adjusted)
scaled by firm size
by price (Pu)
price (Du)

(sjg)

 

A. AR,, =- b0 + b,S,, + 13217,, + unse

0.19398 0.0000 15.101 0.0352 0.00038
0.0366 0.0006 (0.0001) (0.0329)

5.297 0.057

0.0001 0.9543

0.0000 0.4771

0.18798 -0.3121 0.0001 11.306 0.0394 0.00038
0.0366 0.1640 0.0006 (0.0001) (0.0360)

5.122 -1.903 0.301

0.0001 0.0574 0.7635

0.0000 0.0287 0.3817

 

* Standard errors, t-statistics, p value based on 2-tailed
test, and p value based on 1-tailed test, respectively,
appear beneath the independent variable parameter
estimates.

** The coefficient estimate on intercept (b,, ) is not reported.

8 Significant at 5% level based on 2-tailed tests.

 

111

Table 20

Effect of dividend and firm size on the
market response to negative unexpected earnings:
1001 firm-quarters from 1983.4-1986.2*
(Dependent variable: Raw return (R"))

 

 

Independent Variables Regression Statistics
Earnings Dividend Natural F value R-square MSE
surprise scaled logarithm of (Prob>F) (Adjusted)
scaled by firm size
by price (FR)
price (DR)

(8,.)

 

0.0154# 0.0005 4.861 0.0096 0.00051
0.0054 0.0006 (0.0079) (0.0077)

2.859 0.843

0.0043 0.3997

0.0021 0.1998

8. R,, - b,, + b,S,, + 1021),, + b314,t + nuts

0.01588 -0.1688 0.0007 3.626 0.0108 0.00051
0.0054 0.1570 0.0006 (0.0128) (0.0078)

2.923 -1.075 1.101

0.0035 0.2826 0.2712

0.0017 0.1413 0.1356

 

* Standard errors, t-statistics, p value based on 2-tailed
test, and p value based on 1-tailed test, respectively,
appear beneath the independent variable parameter
estimates.

** The coefficient estimate on intercept (b,,) is not reported.

8 Significant at 5% level based on 2-tailed tests.

112

Table 21

Effect of dividend and firm size on the
market response to negative unexpected earnings:
1001 firm-quarters from 1983.4-1986.2*
(Dependent variable: Abnormal return (AR,,))

 

 

Independent Variables Regression Statistics
Earnings Dividend Natural F value R-square MSE
surprise scaled logarithm of (Prob>F) (Adjusted)
scaled by firm size
by price (FR)
price (DR)

(S,,)

 

A. AR,, I b,, + b,S,, + b217,, + u,,**
0.01358 0.0002 3.487 0.0069 0.00050
0.0053 0.0006 (0.0310) (0.0049)
2.521 0.434
0.0118 0.6645
0.0059 0.3322

0.01378 -O.1041 0.0004 2.473 0.0074 0.00050
0.0053 0.1554 0.0006 (0.0593) (0.0044)

2.559 -0.670 0.598

0.0107 0.5032 0.5499

0.0053 0.2516 0.2749

 

* Standard errors, t-statistics, p value based on 2-tailed
test, and p value based on 1-tailed test, respectively,
appear beneath the independent variable parameter
estimates.

** The coefficient estimate on intercept (b,,) is not reported.

# Significant at 5% level based on 2-tailed tests.

 

113
6.2. Dividends versus Earnings Predictability

Pincus [1983] argues that variability of unexpected
returns is negatively related to predictability of earnings
at the time of an earnings announcement. In the context of
this study, earnings predictability should be negatively
related to stock returns given the positive earnings surprise
and positively related to returns given the negative earnings
surprise. In the spirit of Pincus [1983] one may argue that
the dividend effect demonstrated in this study is simply due
to the effect of earnings predictability. That is, high
dividend (large, stable) firms are those for which it is
relatively easy to predict earnings whereas low dividend

(small, growth) firms are relatively hard to predict.

To examine this argument, both ex post and ex ante
proxies for earnings predictability are employed. Following
Pincus [1983], stability of past earnings is adopted as a
proxy for earnings predictability. Stability of past earnings
(standard deviation of changes in earnings over a five year
period: SDC) is available from Lynch, Jones and Ryan's
I/B/E/S/ tape. This is an ex post measure of earnings
stability. Standard deviation of analysts' forecasts of
future earnings (SDF) is used as an ex ante measure of
earnings predictability. This proxy is also available from
the I/B/E/S/ tape. Both proxies are expected to be positively

related to stock return given a positive earnings surprise and

114
negatively related to stock return given a negative earnings
surprise. The results based on these two proxies indicate
that the ex ante measure (SDFR) is a better proxy than the ex

post measure (SDCR).3

Because the earnings predictability proxies are not
available for some firms, the sample size is reduced from 2126
to 1879 observations for the positive surprise group and from
2633 to 2323 observations for the negative surprise group.

Summary statistics are presented in Tables 22 and 23.

First, stock return is regressed on positive earnings
surprise on the basis of the reduced sample (1879
observations). As expected, the estimated coefficient is
positive and significant at the 5% level. The results are
presented in Panel A of Table 24 (Table 25 with abnormal

 

3 To determine which is a better proxy for earnings
predictability, the following two equations are estimated:

R,, - b,, + b,S,, + szDC,, 4» u,, and

The results reveal that no proxy variable has a significant
coefficient given positive earnings surprises. Given negative
earnings surprises, however, the coefficient estimate on SDF,
is statistically significant while the coefficient estimate
on SDC is insignificant again. This is more clearly
demonstrated when the coefficients of the two proxies are
estimated at the same time:

R,, - b,, + b,S,, + bZSDC,, + b,SDF,, + u,,.
The coefficient of SDF,, is significant at the 5% level while

that of SDC,, is not s gnificant (t-statistic - -0.445). In
this section, only the results with SDFR are reported.

115
return). To assess the effect of earnings predictability on
stock price response to positive unexpected earnings, the

following equation is estimated:

The results of'estimating'Equation (6.3) are reported in Panel
B of Table 24 (Table 25 with abnormal return). The
coefficient b2 is insignificant. The sign is opposite that

predicted by Pincus [1983].

To see if the dividend effect has anything to do with the
earnings predictability, dividend (Dn) is added to Equation

(6.3):
R,, - b,, + b,S,, + b,SDF,, + b3D,, + u,,. (6.4)

Panel C of Table 24 (Table 25 with abnormal return) reveals
that the coefficient on DR is still negative and significant
while the coefficient on SDFR is insignificant. This
indicates that the dividend effect found in Chapter 5 is
independent of the effect of earnings predictability for the
positive earning surprise group. This is accentuated in

Figure 11.

stock return is also regressed on negative earnings
surprise. Results are presented in Panel A.of Table 26 (Table
27 with abnormal return). Estimation of Equation (6.3) is

repeated with the negative surprise group resulting in Panel

116

 

 

 

 

 

Equation Equation Equation
(5.2) (5.3) (5.4)
s"! D]: s"! Di" SDI]:
Adj R2 - 0.0087 Adj R2 - 0.0082
F - 13.364 R - 13.357
8,, R - 0.0003 R - 0.0003
Agj R2 - 0.0022 --------------------------------------- J
O
R - 5.171 b
R =- 0.0231 s,,, 5122,. s,,, SDF,,, 0,,
Adj Rz - 0.0017 Adj R2 - 0.0082
F* . 0.005 R - 0.008
R - 0.9391 R - 0.9247

 

' F. - partial F-statistic

b P - p-value

Figure 11.

Contribution of each of dividend and

 

earnings predictability: Positive earnings surprises

B of Table 26 (Table 27 with abnormal return). The
coefficient estimate on SDFR is statistically significant at
the 5% level. But the directional relation is opposite that

predicted in the context of Pincus [1983]. This indicates
that, given the negative unexpected earnings, the easier to
predict the earnings, the lower the stock return: that is,
price change variability is positively related to

predictability of earnings.

Equation (6.4) is estimated to investigate the
relationship between the effects of earnings predictability
and dividend given negative unexpected earnings. Results are

presented in Panel C of Table 26 (Table 27 with abnormal

117

 

 

,1
: Equation Equation Equation
, (5.2) (5.3) (5.4)
S": D" 3") Dh' EDI“
Adj R2 - 0. 0046 Adj R2 - 0. 0063
, R - 0. 061 R' - 4. 977
5,, P - 0. 7861 R - 0. 0258
Adj R2 - 0. 0050 --------------------------------------- -
R - 12. 708'
R - 0. 0004b s,,, 555,, s,,, SDF,,, 5,,
Adj R2 - 0.0067 Adj R2 - 0.0063
F* - 4.979 R" - 0.061
R - 0.0256 R - 0. 8000

 

 

 

 

 

:r' - partial F-statistic
bP - p-value

Figure 12. Contribution of each of dividend and
earnings predictability: Negative earnings surprises
return) . The coefficient on SDF,, is statistically significant
(5% level) whereas the coefficient on dividend on is

insignificant as in Chapter 5 (see Figure 12).

It is worthy to note in Pincus [1983] that "Results of
the nonparametric rank correlation and chi-square independence
procedures provide weak support (significance at about the .09
level) for the hypothesis for interim announcements only. On
the other'hand, significance:at the .05 level is indicated for
annuals, but the directional relation is opposite that
predicted." Pincus [1983] argues that for interim earnings,
hard-to-predict earnings and easy-to-predict earnings are

announced almost at the same time (less than one day time

118
span), but for annuals, easy-to-predict earnings are announced
an average of three and one-half trading days earlier than
hard-to-predict earnings. The effect would be to reduce the
remaining uncertainty surrounding the earnings of late-
announcing firms such that the effect of earnings
predictability is confounded by the early/late-announcing
phenomenon. IPincus [1983] concludes that.only the results for
quarterly earnings are valid and the results support the
hypothesis that earnings predictability is negatively related
to price change variability. The conclusion is not
unambiguous. First, it is not easy to believe that the same
firm has significantly different behavior for interim and
annual earnings announcements. Secondly, even if they are so
different, Pincus's weekly return interval might capture the
timing effect of earnings announcements since easy-to-predict
annuals are announced an average of less than one week earlier

than hard-to-predict earnings.

One interpretation of the findings in this section is
that, given the earnings surprise, a given size surprise for
easy-to-predict earnings must be a real surprise and should
result in a large price change. On the other hand, a surprise
for hard-to-predict earnings cannot be new information because
that surprise is always expected. In other words, if one
lives in uncertainty, a surprise is not news to note. But
this is not the case when one lives in certainty. Other

interpretations may certainly be possible.

119

Table 22

Summary statistics: 1879 firm-quarters
(All unexpected earnings are positive.)

 

 

Variable N Mean Standard Minimum Maximum
Deviation

Return 1879 0.00443 0.01929 -0.10039 0.10351

Abnormal 1879 0.00354 0.01836 -0.10536 0.09527

return

Unexpected 1879 0.00950 0.02566 0.00001 0.48380

earnings

scaled by

price

Dividend 1879 0.00653 0.00487 0.00000 0.02691

scaled by

price

Standard 1879 0.22929 0.19219 0.00400 0.82600

deviation of

change in

earnings

(soc,.)

Standard 1879 0.03478 0.04413 0.00000 1.29900

deviation of
forecast of
future
earnings
(SDF,,)

 

120

Table 23

Summary statistics: 2323 firm-quarters
(All unexpected earnings are negative.)

 

 

Variable N Mean Standard Minimum Maximum
Deviation

Return 2323 -0.00440 0.02221 -0.17135 0.30384

Abnormal 2323 -0.00446 0.02156 -0.16724 0.30464

return

Unexpected 2323 -0.03304 0.20911 -6.02842 0.00001

earnings

scaled by

price

Dividend 2323 0.00675 0.00504 0.00000 0.03232

scaled by

price

Standard 2323 0.24253 0.19520 0.00400 0.83300

deviation of

change in

earnings

(soc,.)

Standard 2323 0.03428 0.03144 0.00000 0.36770

deviation of
forecast of

future
earnings
(SDF,,)

 

121

Table 24

Effect of dividend and earnings predictability on the

market response to positive unexpected earnings:

1879 firm-quarters from 1983.4-1986.2*
(Dependent variable: Raw return (R“))

 

Independent Variables

Regression Statistics

 

 

Earnings Dividend Std. Dev. of F value R-square MSE
surprise scaled forecasts of (Prob>F) (Adjusted)
scaled by future earnings

by price (SDFR)

price (DR)

(8,.)

 

A. R,, - b,, + b,s,t + unit

0.03948

0.0173
2.274

0.0231
0.0105

B. R,, - b,, + b,S,, + bZSDF,, + u,,..

0.03938

0.0173
2.273

0.0231
0.0105

c. R"

0.0320
0.0173
1.844

0.0654
0.0327

5.171 0.0027 0.00037
(0.0231) (0.0022)

0.0000 2.587 0.0028 0.00037
0.0001 (0.0755) (0.0017)

0.076

0.9391

0.4695

-0.33488 -0.0000 6.187 0.0098 0.00036
0.0916 0.0001 (0.0004) (0.0082)
-3.654 -0.095

0.0003 0.9247

0.0001 0.4623

 

Standard errors, t-statistics, p value based on 2-tailed
test, and p value based on 1-tailed test, respectively,
appear beneath the independent variable parameter
estimates.
** The coefficient estimate on intercept (b,,) is not reported.
8 Significant at 5% level based on 2-tailed tests.

122

Table 25

Effect of dividend and earnings predictability on the
market response to positive unexpected earnings:
1879 firm-quarters from 1983.4-1986.2*

(Dependent variable: Abnormal return (AR"))

 

Independent Variables Regression Statistics

 

Earnings Dividend Std. Dev. of F Value R-square MSE
surprise scaled forecasts of (Prob>F) (Adjusted)

scaled by future earnings
by price (SDFn)
price (DR)

(8,.)

 

A. AR,, =- b,, + b,S,, + u,,aa

0.0410# 6.184 0.0033 0.00033
0.0164 (0.0130) (0.0028)

2.487

0.0130

0.0065

Be AR]: ' b0 + b1sjt + bstFIt + U,,**

0.04108 0.0000 3.179 0.0034 0.00033
0.0164 0.0000 (0.0418) (0.0023)

2.489 0.421

0.0129 0.6740

0.0064 0.3370

0.03513 -0.2725: 0.0000 5.377 0.0085 0.00033
0.0165 0.0873 0.0000 (0.0012) (0.0069)
2.118 -3.121 0.406

0.0343 0.0018 0.6845
0.0171 0.0009 0.3422

 

* Standard errors, t-statistics, p value based on 2-tailed
test, and p value based on l-tailed test, respectively,
appear beneath the independent variable parameter
estimates.

** The coefficient estimate on intercept (b,,) is not reported.

8 Significant at 5% level based on 2-tailed tests.

123

Table 26

Effect of dividend and earnings predictability on the
market response to negative unexpected earnings:
2323 firm-quarters from 1983.4-1986.2*

(Dependent variable: Raw return (Rn))

 

Independent Variables Regression Statistics

 

Earnings Dividend Std. Dev. of F value R-square MSE

surprise scaled
scaled by

by price
price (DR)
(8,.)

forecasts of (Prob>F) (Adjusted)
future earnings
(sorn)

 

A. R,, I b,, + 13,8,t + u,,ss

0.00788
0.0021
3.565
0.0004
0.0002

12.708 0.0054 0.00049
(0.0004) (0.0050) '

0.00778
0.0021
3.534
0.0004
0.0002

0.0003: 8.859 0.0076 0.00049
0.0001 (0.0001) (0.0067)

2.233

0.0256

0.0128

0.00788 -0.0231
0.0022 0.0913
3.542 -0.253

0.0004 0.8000
0.0002 0.4000

0.00038 5.925 0.0076 0.00049
0.0001 (0.0006) (0.0063)

2.231

0.0258

0.0129

 

* Standard errors, t-statistics, p value based on 2-tailed
test, and p value based on l-tailed test, respectively,
appear beneath the independent variable parameter

estimates.

** The coefficient estimate on intercept (b,,) is not reported.
8 Significant at 5% level based on 2-tailed tests.

124

Table 27

Effect of dividend and earnings predictability on the
market response to negative unexpected earnings:
2323 firm-quarters from 1983.4-1986.2*

(Dependent variable: Abnormal return (AR"))

 

Independent Variables Regression Statistics

 

Earnings Dividend Std. Dev. of F value R-square MSE
surprise scaled forecasts of (Prob>F) (Adjusted)

scaled by future earnings
by price (SDFR)
price (DR)

(8,.)

 

A. AR,, - b,, + 13,5,t + uni-a

0.0050# 5.691 0.0024 0.00046
0.0021 (0.0171) (0.0020)

2.386

0.0171

0.0085

B. AR,, a b,, + b,S,, + bZSDF,, + u,,**

0.00508 0.00038 5.645 0.0048 0.00046
0.0021 0.0001 (0.0036) (0.0040)

2.352 2.364

0.0188 0.0182

0.0094 0.0091

c. AR,, a 13° + b,S,, + 1320,, + b,SDFn + u,,** ,
0.00508 ‘0.00408 0.00038 3.762 0.0048 0.00046

0.0021 0.0888 0.0001 (0.0105) (0.0036)
2.348 -0.045 2.363

0.0190 0.9641 0.0182
0.0095 0.4820 0.0091

 

* Standard errors, t-statistics, p value based on 2-tailed
test, and p value based on 1-tailed test, respectively,
appear beneath the independent variable parameter
estimates. 7

** The coefficient estimate on intercept (b,,) is not reported.

8 Significant at 5% level based on 2-tailed tests.

125

6.3. Dividends versus Growth

Collins and Kothari [1989] provide evidence that, given
a level of earnings surprise, growth is positively associated
with the magnitude of the price response. Given positive
unexpected earnings, therefore, the higher the growth, the
higher the stock return. Given negative unexpected earnings,
the higher the growth, the lower the stock return. In this
sense, one may argue that the dividend effect is found only
because dividends are a proxy for growth. Low dividends may
signal management's information about future growth
opportunities. To see whether this argument is correct, the

following two equations are estimated:

R,, - 8,, + b,S,, + 820,, + u,,, (6.5)

R,, - 8,, + b,S,, + 8,0,, + b3G,, + u,,, (6.6)

where G,, - a proxy for growth of firm j at time t. As a proxy
for growth, analysts' forecasts of long term growth obtained
from the I/B/E/S tape are used.‘

The results of estimating Equation (6.5) are presented
in Panel A of Table 28 (Table 29 with abnormal return) for
positive unexpected earnings. Interestingly, the coefficient

on growth G" is not significant (t-statistic - -1.016).

 

‘ The sample used in this section is basically the same as
that in Section 6.2. The sample size is 1879 observations for
the positive surprise group and 2323 observations for the
negative surprise group.

126

 

 

 

 

Equation Equation Equation
(5.2) (5.3) (5.4)
Sn! Du Sn! Dnv 9n
Adj R2 - 0. 0087 Adj Rz - 0. 0089
F-l3.364 R'-1.382
, 5,, R - 0. 0003 R - 0. 2379
Adj R2 - 0. 0022 ----------------------------------------
R - 5.171' b
R - 0. 0231 s,,, 5,, s,,, s,,, 5,,
Adj Rz - 0.0022 Adj Rz - 0.0089
F* - 1.023 R - 13.719
R - 0.3039 R --0.0002

 

 

 

' b'F - partial F-statistic

bP - p-value

Figure 13.

Contribution of each of dividend and
growth: Positive earnings surprises

Estimating Equation (6.6) reveals that the coefficient on

dividend remains significant while that on growth is not

significant (see Tables 28 and 29).

dividend effect is not affected by a growth effect.

13 conveys the same message.

This implies that the

Figure

The estimation of Equations (6.5) and (6.6) is repeated

with the negative earnings surprises resulting in Tables 30

and 31 Based on these results, Figure 14 is constructed.

Figure 14 reveals that growth does not contribute to

explaining the return/surprise relation given negative

earnings surprises.

Consequently, Figure 14 combined with

127

 

 

 

 

Equation Equation Equation
(5-2) (5.3) (5.4)
51v 9): 3w ”w 9):
Adj R2 - 0. 0046 A93 R2 - 0.0042
R' - 0. 056 R - 0.001
5,, R - 0.7861 R - 0.9881
Adj R2 - 0. 0050 --------------------------------------- -
R‘ =- 12. 708'
R . 0. 0004 s,,, 5,, S,.. G,.. 12,.
Adj R2 - 0.0046 Adj RZ - 0. 0042
F* a 0.001 R - 0.061
R - 0.9815 R - 0. 7865

 

 

 

 

;F' a partial F-statistic

bP a p-value

Figure 14.

Contribution of each of dividend and
growth: Negative earnings surprises

Figure 13 indicate that the dividend effect is independent of
a growth effect.

The results provide evidence that growth does not impact

the stock return given unexpected earnings. This is

inconsistent with Collins and Kothari [1989]. The
methodologies of this and the Collins and Kothari study differ
in several respects, and this may account for the inconsistent
conclusions. For instance, Collins and Kathari use market-
to-book value equity ratios as a proxy for growth whereas

analysts' forecasts of growth are used in this study. Collins

and Kothari's (market value/book value) ratio is related to

128
the q ratio (market value/replacement cost of assets) used in
this study. Either analysts' forecast or q ratio may be a
better proxy for growth than a market-to-book value ratio
since the former are market-determined figures. Also, Collins
and Kothari do not analyze positive and negative earnings

surprises separately.

129

Table 28

Effect of dividend and growth on the
market response to positive unexpected earnings:
1879 firm-quarters from 1983.4-1986.2*
(Dependent variable: Raw return (Rn))

 

Independent Variables Regression Statistics

 

Earnings Dividend Analysts' P value R-square MSE
surprise scaled forecasts of (Prob>F) (Adjusted)
scaled by growth

by price (GR)
price (on)

(3,.)

 

A. R" 3 be + b,Sn + ban + unit

0.0356# -0.0097 3.101 0.0033 0.00037
0.0178 0.0083 (0.0452) (0.0022)

2.002 -1.169

0.0454 0.2425

0.0227 0.1212

0.0311 -0.3395# -0.0000 6.653 0.0105 0.00036
0.0174 0.0916 0.0000 (0.0002) (0.0089)
1.790 -3.703 -1.181

0.0737 0.0002 0.2379
0.0368 0.0001 0.1189

 

* Standard errors, t-statistics, p value based on 2-tailed
test, and p value based on l-tailed test, respectively,
appear beneath the independent variable parameter
estimates.

** The coefficient estimate on intercept (be) is not reported.

# Significant at 5% level based on z-tailed tests.

130

Table 29

Effect of dividend and growth on the
market response to positive unexpected earnings:
1879 firm-quarters from 1983.4-1986.2*

(Dependent variable: Abnormal return (ARR))

 

Independent Variables

Regression Statistics

 

Earnings
surprise
scaled
by
price

(8,.)

Dividend Analysts'

scaled forecasts of
by growth
price (GR)

(0,.)

P value R-square MSE
(Prob>F) (Adjusted)

 

A. A12It - b0 4» 5,3“ + ”23,-. + “1:"

0.0403#
0.0165
2.443
0.0147
0.0073

-0.0000
0.0000

-1.032
0.3021
0.1510

3.625 0.0038 0.00033
(0.0268) (0.0028)

8. AR" =- bo + b,Sjt + ”291: + 1336" + “It“:

0.0341#
0.0165
2.060
0.0395
0.0197

-0.2772# -0.0000
0.0873 0.0000

'3.173 -1.173
0.0015 0.2410
0.0007 0.1205

5.784 0.0092 0.00033

(0.0007) (0.0076)

 

* Standard errors, t-statistics, p value based on 2-tailed
test, and p value based on l-tailed test, respectively,
appear beneath the independent variable parameter
estimates.

** The coefficient estimate on intercept (b,,) is not reported.

# Significant at 5% level based on z-tailed tests.

131

Table 30

Effect of dividend and growth on the
market response to negative unexpected earnings:
2323 firm-quarters from 1983.4-1986.2*
(Dependent variable: Raw return (Rn))

 

 

Independent Variables Regression Statistics
Earnings Dividend Analysts' P value R-square MSE
surprise scaled forecasts of (Prob>F) (Adjusted)
scaled by growth
by price (Gk)
price (Du)

<s,.>

 

A. Rjt =- bo + b1Sjt + 1326it + unit

0.0078# -0.0000 6.352 0.0054 0.00049
0.0021 0.0000 (0.0018) (0.0046)

3.564 -0.023

0.0004 0.9815

0.0002 0.4907

B. R]: I bo + b,Sjt + b2”): + bBGJt + unit

0.0078# -0.0247 -0.0000 4.257 0.0055 0.00049
0.0022 0.0915 0.0000 (0.0054) (0.0042)
3.574 -0.271 -0.015

0.0004 0.7865 0.9881
0.0002 0.3932 0.4940

 

* Standard errors, t-statistics, p value based on 2-tailed
test, and p value based on l-tailed test, respectively,
appear beneath the independent variable parameter
estimates.

** The coefficient estimate on intercept (be) is not reported.

# Significant at 5% level based on 2-tailed tests.

132

Table 31

Effect of dividend and growth on the
market response to negative unexpected earnings:
2323 firm-quarters from l983.4-1986.2*
(Dependent variable: Abnormal return (ARR))

 

Independent Variables

Regression Statistics

 

 

Earnings Dividend Analysts' P value R-square MSE
surprise scaled forecasts of (Prob>F) (Adjusted)
scaled by growth
by price (Gk)
price (on)
(8,.)
A. ARit =- b0 + b,Sjt + bzc;jt + ujtfl
0.0051# 0.0000 2.849 0.0024 0.00046
0.0021 0.0000 (0.0581) (0.0016)
2.385 0.091
0.0171 0.9278
0.0085 0.4639

8. ARIt .. be + 16,3,t + ”291: + ban + an"

0.
0.
2.
0.
0.

0051* '0.0059 -0.0000
0021 0.0889 0.0000
383 -0.067 -0.093

0173 0.9465 0.9262
0086 0.4732 0.4631

1.900 0.0025 0.00046
(0.1257) (0.0012)

 

*

Standard errors, t-statistics, p value based on 2-tailed
test, and p value based on l-tailed test, respectively,
appear beneath the independent variable parameter

estimates.

** The coefficient estimate on intercept (be) is not reported.
# Significant at 5% level based on 2-tailed tests.

Chapter 7. Summary and Conclusions
7.1. Summary of the Empirical Results

The evidence of Chapter 5 indicates that, in a study of
market response to earnings announcements, data should be
analyzed separately according to the sign of earnings
surprise. When a regression is run on the entire sample, all
the factors (dividend, q ratio, beta) appear to impact the
return/surprise relation in section 5.1. However, a separate
analysis on positive and negative surprises provides different
results. The results of the matched-pair t-test in section
5.2 show that the equity market response is asymmetric between
positive and negative earnings surprises. The market reacts
more strongly to positive earnings surprise than to negative
earnings surprise. Given a positive earnings surprise, the
factors have a statistically significant impact on the
return/surprise relation. But all the factors of Chapter 5
turn out to be unimportant with a negative earnings surprise.
This suggests that the equity market reacts to earnings
surprise through different functional mechanisms depending on
whether the earnings surprise is positive (good news) or

negative (bad news).

The results of the separate analyses in section 5.2
show that dividend is the most important of the factors
examined. The results indicate that dividend policy is an

133

134
important factor to be considered in studies of the

relationship between equity return and earnings surprise.

In Chapter 6, analyses are made to examine whether the
dividend effect differs from the firm size, earnings
predictability, and growth effects found by other studies.
The results in section 6.1 reveal that the dividend effect is
distinguished from the effect of firm size found in Atiase
[1985] and Freeman [1987]. The addition of a firm size
variable to the regression (Equation (4.6)) does not weaken
the dividend effect. When the information environment is
controlled, the effect of dividend is still apparent for
positive surprises while the effect of firm size disappears.
The results suggest that the dividend effect is not a proxy

for differential information.

The evidence of section 6.2 indicates that dividend is
not a proxy for earnings predictability. The addition of an
earnings predictability proxy to the regression does not
change the results. The results of regressing stock return
on earnings surprise and earnings predictability provide
evidence inconsistent with Pincus [1983]. Given the positive
earnings surprise, earnings predictability has no association
with market response. Given the negative earnings surprise,
earnings predictability plays an important role.
Interestingly, however, the directional relation is opposite

that predicted by Pincus [1983]. Since this study uses

135
quarterly data, the timing effect suggested by Pincus should
not matter. Based on these results the following
interpretation is suggested. Firms with less predictable
earnings streams react less to a given earnings surprise than
firms with more predictable earnings streams. The surprise
means more to the predictable firm than to the unpredictable
firm. Another way of stating this is that the 'unexpected'
surprise delivers more information than the 'expected'

surprise.

In section 6.3, it is revealed that the dividend effect
is not the same as the growth effect identified.by Collins and
Kothari [1989]. A growth effect is not supported by this
study. The results are not consistent with Collins and

Kothari [1989].

This study finds that heteroskedasticity does not create
problems in return/surprise studies. This is consistent with
Christie [1987]. The GLS estimator controlling for

heteroskedasticity reduces to the OLS estimator in this study.

The results of Chapters 5 and 6 evidence that the
return/surprise relation is essentially the same whether one
uses raw return or abnormal return as a dependent variable.
This is consistent with Beaver, Lambert and Horse [1980],
Beaver, Lambert and Ryan [1987], and Collins and Kothari
[1989]. This implies that, relative to the variability in raw

return around earnings announcements, the variability in

136

expected return based on systematic risk is small.

Finally, the positive and negative earnings surprise
groups are different in terms of q ratio. The matched-pair
t-test reveals that q ratio is higher for the positive group
than for the negative group. But systematic risk and dividend

yield are not different for the two groups.

The findings for each variable examined in this study are

summarized below.

 

 

 

Factor Positive surprise Negative surpris
(good news) (bad news)

Dividend strong support no support
Q ratio weak support' no support
Systematic no support weak supportb
risk
Firm size weak support strong support
Earnings no support strong supportc
predictability ‘
Growth no support no support

 

 

 

 

a weak.support.is defined as significance at the 15% level.

b The directional relation is opposite that predicted by
Equation (3.19).

c The directional relation is opposite that predicted by
Pincus [1983] . Strong support is defined as significance
at the 5% level.

Figure 15. Summary of results

137

7.2. Conclusions

This study theoretically and empirically identifies
factors contributing to stock price response to earnings
announcements. Chapter 5 provides evidence that dividend
policy is an important factor to be considered in
return/surprise studies. One may argue that dividend is found
important simply because it is a proxy for firm size, earnings
predictability, or economic growth. But the empirical results
of Chapter’6 show’that.dividend.has its own.explanatory power.
The dividend effect is only apparent for positive earnings
surprises. Positive earnings surprise and dividend policy
combined provide very useful information to the investment
community. That information may signal management's inside
information about the firm's unrealized investment

opportunities.

Evidence in this study has several implications for
interpreting the findings in previous research and for future
research. First, the results show that market beta,
empirically found important by Collins and Kothari [1989],
does not impact the return/surprise relation. In particular,
with negative earnings surprise, the directional relation is
opposite that found by Collins and Kothari [1989]. The
results are consistent with Jain [1981] . This conflict should
be addressed by future research. As always, there is a need

to replicate the study on different sets of firms, over

138

different time periods, and using alternative methodologies.

Second, given the results of this study, the relationship
between earnings predictability and market reaction is
opposite that predicted by Pincus [1983]. The results
indicate that variability of price change is positively
related to earnings predictability. As an extension,
different data can be analyzed. More importantly, one may
examine returns variability associated with other types of

announcements.

Also, the evidence of the results on growth is
inconsistent with Collins and Kothari [1989]. Collins and
Kothari use the ratio of market to book value as a proxy for
growth while security analysts' forecasts of growth are used
in this study. This is an issue which will benefit from

further research.

One of the most important findings is that the equity
market reaction is asymmetric between positive and negative
earnings surprises. The response is stronger to positive
earnings surprise. It may reflect the psychology of the
market or general economic conditions. Again, it would be
fruitful to replicate the study on different sets of data and
using different methodologies. Differential market responses
to good and bad news may depend on whether the market is a)
bull market or a bear market. This can be found by

investigating different or expanded time series.

8181410033?“

BIBLIOGRAPHY

Aitken, A. C. ”On Least Squares and Linear Combinations of
Observations." Proceedings of the Royal Society, Edinburgh,
55 (1935), 42-48. '

Atiase, R. K. "Predisclosure Information, Firm Capitalization
and Security Price Behavior around Earnings Announcements."
Journal of Accounting Research, 23 (Spring 1985), 21-36.

Balestra, P. and M. Nerlove. "Pooling Cross Section and Time
Series Data in the Estimation of a Dynamic Model: The Demand
for Natural Gas." Econometrica, 34 (July 1966), 585-612.

Ball, R. and P. Brown. "An Empirical Evaluation of Accounting
Income Numbers." Journal of Accounting Research, 6 (Autumn
1968), 159-178.

Bamber, L. S. "The Information Content of Annual Earnings
Releases: A Trading Volume Approach.” Journal of Accounting
Research, 24 (Spring 1986), 40-56.

Banz, R. "The Relationship between Return and Market Values
of Common Stock." Journal Financial Economics, 9 (1981), 3-
18.

Basu, S. "The Effect of Earnings Yield on Assessments of the
Association between Annual Accounting Income Numbers and
Security Prices." Accounting Review, 53 (July 1978), 599-625.

Beaver, W. "The Information Content of Annual Earnings
Announcements.” Journal of Accounting Research, 6 (1968
Suppl.), 67-92.

Beaver, W., R. Clarke, and W. Wright. "The Association between
Unsystematic Security Returns and the Magnitude of Earnings
Forecast Errors." Journal of Accounting Research, 17 (Autumn
1979), 316-340.

Beaver, W., R. Lambert, and D. Morse. "The Information Content
of Security Prices." Journal of Accounting and Economics, 2
(1980), 3-28.

Beaver, W., R. Lambert, and 8. Ryan. "The Information Content

of Security Prices: A Second Look.” Journal of Accounting and
Economics, 9 (1987), 139-157.

139

140

Berenson, M. L., D. M. Levine, and.M. Goldstein. Intermediate
Statistical Methods. Prentice-Hall, Inc. Englewood Cliffs, New
Jersey, 1983.

Box G. E. P. and G. M. Jenkins. Time Series Analysis:
Forecasting and Control. 2nd Edition, Holden-Day, San
Francisco, 1976.

Brown, L. D., P. A. Griffin, R. L. Magerman, and M. E.
Zmijewski. "Security Analyst Superiority Relative to
Univariate Time-series Models in Forecasting Quarterly
Earnings." Journal of.Accounting'and.Economics, 9 (1987a), 61-
87.

. "An Evaluation of Alternative
Proxies for the Markets' Assessment of Unexpected Earnings."
Journal of Accounting and Economics, 9 (1987b), 159-193.

 

Brown, L. D., G. Richardson, and S. Schwager. ”An Information
Interpretation of Financial Analyst Superiority in Forecasting
Earnings." Journal of Accounting Research, 25 (Spring 1987),
49-67.

Brown, L. D. and M. Rozeff. "The Superiority of Analyst
Forecasts as Measures of Expectations: Evidence from
Earnings." Journal of Finance, 33 (March 1978), 1-16.

Campbell, J. Y. and R. Shiller. ”Stock Prices, Earnings, and
Expected Dividends.” Journal of Finance, 43 (1988), 661-676.

Christie, A. A. "On Cross-sectional Analysis in Accounting
Research." Journal of Accounting and Economics, 9 (1987),
231-258.

Ciccolo, J. ”A Linkage between Product and Financial Markets:
Investment and q." Chapter III of Ph.D. Dissertation, Yale
University, 1975.

Collins, W. A. and W. S. Hopwood. "A.Mu1tivariate Analysis of
Annual Earnings Forecasts Generated from Quarterly Forecasts
of Financial Analysts and Univariate Time-series Models.”
Journal of Accounting Research, 18 (1980), 390-405.

Collins, D. W. and S. P. Kothari. ”An Analysis of the
Intertemporal and Cross-sectional Determinants of Earnings
Response Coefficients." Working Paper, University of Iowa,
1989.

Cragg, J. and B. Malkiel. "The Consensus and Accuracy of Some
Predictions of the Growth of Corporate Earnings.” Journal of
Finance, 23 (March 1968), 67-84.

141

Crichfield, T., T. Dyckman, and.J. Lakonishok. "An Evaluation
of Security Analysts' Forecasts." Accounting Review, 53
(July 1978), 651-668.

Easton, P. D. "Accounting Earnings and Security Valuation:
Empirical Evidence of the Fundamental Links.” Journal of
Accounting Research, 23 (1985 suppl.), 54-77.

Elton, E. J., M. J. Gruber, and M. N. Gultekin. "Professional
Expectations: Accuracy and Diagnosis of Errors.” Journal of
Financial and Quantitative Analysis, 19 (December 1984), 351-
363.

. ”Earnings Expectations and Share
Prices.” Management Science, 27 (September 1981), 975-987.

 

Freeman, R. N. "The Association between Accounting Earnings
and Security Returns for Large and Small Firms." Journal of
Accounting and Economics, 9 (1987), 195-228.

Fried, D. and D. Givoly. "Financial Analysts' Forecast of
Earnings: A Better Surrogate for Earnings Expectations."
Journal of Accounting'andchonomicsq 4 (October 1982), 85-107.

Givoly, D. and J. Lakonishok. ”Properties of Analysts'
Forecasts of Earnings: A Review and Analysis of the Research. "
Journal of Accounting Literature, 3 (1984), 117-152.

Gordon, M.J. and E. Shapiro. ”Capital Equipment Analysis: The
Required Rate of Profit", Management Science, 3 (October
1956), 102-110.

Grant, E. B. "Market Implications of Differential Amounts of
Interim Information." Journal of Accounting Research, 18
(Spring 1980), 255-268.

Griliches, z. and.J. A. Hausman. "Errors in Variables in Panel
Data." Journal of Econometrics, 31 (February 1986), 93-118.

Hall, B. H., C. Cummins, E. S. Laderman, and.J. Mundy. The R&D
Master File Documentation. National Bureau of Economic
Research, 1988.

Hausman, J. A. ”Specification Test in Econometrics.”
Econometrica, 46 (November 1978), 1251-1271.

Hausman, J. A. and W. Taylor. ”Panel Data and Unobservable
Individual Effects." Econometrica, 49 (1981), 1378-1398.

Hoch, I. "Estimation of Production Function Parameters
Combining Time-Series and Cross-Section Data." Econometrica,
30 (January 1962), 34-53.

142

Jain, P. C. ”Cross-sectional Association between Abnormal
Returns and Firm Specific Variables." Journal of Accounting
and Economics, 4 (1982), 205-228.

Joy, 0., R. Litzenberger, and R, McEnally. "The Adjustment of
Stock Prices to Announcements of Unanticipated Changes in
Quarterly Earnings." Journal of Accounting Research, 15
(Autumn 1977), 107-225.

Kiger, J. E. "Volatility in Quarterly Accounting Data."
Accounting Review, 49 (January 1974), 1-7.

Klepper, S. and E. Leamer. "Consistent Sets of Estimates for
Regressions with Errors in All Variables.” Econometrica 52,
163-183.

Kormendi, R. and R. Lipe. ”Earnings Innovations, Earnings
Persistence, and Stock Returns." Journal of Business, 60
(1987), 323-345.

Lakonishok, J. "Stock Market Return Expectations: Some General
Properties." Journal of Finance, 35 (September 1980) , 921-931.

Lindenberg, E. B. and S. A. Ross. "Tobin's q Ratio and
Industrial Organization." Journal of Business, 54 (January
1981), 1-32.

Leftwich, R. ”Evidence of the Impact of Mandatory Changes in
Accounting Principles on Corporate Loan Agreements." 3 (1981) ,
3-36.

Maddala, G. S. "The Use of Variance Components Models in
Pooling Cross Section and Time Series Data." Econometrica, 39
(March 1971), 341-358.

. Econometrics. McGraw-Hill, New York. 1977.

 

Mankiw, N. G., D. Romer, and M. D. Shapiro. "An Unbiased
Reexamination of Stock Market Volatility. " Journal of Finance,
40 (July 1985), 677-689.

May, R. "The Influence of Quarterly Earnings Announcements on
Investor Decisions as Reflected in Common Stock Price
Changes." Journal of.Accounting Research, 9 (1971 suppl.) 119-
163.

Miller, M. H. and K. Rock. ”Dividend Policy under Asymmetric
Information." Journal of Finance, 40 (September 1985), 1031-
1051.

Mundlak, Y. "Empirical Production Function Free of Management
Bias." Journal of Farm Economics, 43 (February 1961), 44-56.

143

Muth, J. F. "Rational Expectations and the Theory of Price
Movements." Econometrica, 29 (July 1961), 313-335.

Nakamura, A. and M. Nakamura. "Rational Expectations and the
Firm's Dividend Behavior. " Review of Economics and Statistics,
67 (November 1985), 606-615.

Nerlove, M. "Further Evidence on the Estimation of Dynamic
Economic Relations from a Time Series of Cross Sections."
Econometrica, 39 (March 1971), 359-382.

O'Brien, P. " Analysts' Forecasts as Earnings Expectations."
Journal of Accounting and Economics, 10 (1988), 53-83.

Ohlson, J. A. "Price-Earnings Ratios and Earnings
Capitalization under Uncertainty.” Journal of Accounting
Research, 21 (Spring 1983), 141-154.

Patell, J. M. and M. A. Wolfson. ”The Intraday Speed of
Adjustment of Stock Prices to Earnings and Dividend
Announcements.” Journal of Financial Economics, 13 (1984),
223-252.

Pearce, D. K. and V. V. Roley. "Stock Prices and Economic
News." Journal of Business, 58 (1985), 49-67.

Pincus, M. "Information Characteristics of Earnings
Announcements and Stock Market Behavior." Journal of
Accounting Research, 21 (Spring 1983), 155-183.

Pyndyck, R. S. and D. L. Rubinfeld. Econometric Models and
Econometric Forecasts. McGraw-Mill, 1981.

Reinganum, M. "Misspecification of Capital Asset Pricing:
Empirical Anomalies Based on Earnings' Yields and Market
Values." Journal of Financial Economics, 9 (March 1981), 19-
46.

. "A Direct Test of Roll's Conjecture on the Firm
Size Effect.” Journal of Finance, 37 (March 1982), 27-35

 

Sargent, T. J. Macroeconomic Theory. 2nd Edition, Academic
Press, 1987.

. "Rational Expectations, the Real Rate of
Interest, and the Natural Rate of Unemployment.” Brookings
Papers on Economic Activity, 2 (1973), 429-472.

 

Shiller, R. "Do Stock Prices Move Too Much to be Justified by
Subsequent Changes in Dividends?" American Economic Review,
71 (June 1981), 421-436.

144

. ”Rational Expectations and the Dynamic Structure
of Macroeconomic Models: A Critical Review." Journal of
Monetary Economics, 4 (January 1978), 1-44.

 

Taylor, W. E. “Small Sample Consideration in Estimation from
Panel Data.” Journal of Econometrics, 13 (June 1980) , 203-223.

Tobin, J. ”A General Equilibrium Approach to Monetary Theory. "
Journal of Money, Credit, and Banking, 1 (February 1969), 15-
29.

Wallace, T.D. and A. Hussain. ”The Use of Error Components
Models in Combining Cross Section with Time Series Data."
Econometrica, 37 (1969), 55-72.

West, K. D. "A Specification Test for Speculative Bubbles.”
Quarterly Journal of Economics, 102 (August 1987), 554-580.

            

    

HIGRN STATE UN V.

HIC I LIBRRRIES
W W I} I’HIIIHWIIHHIWHI
9 8098

312 300620