Zr
:‘

 

 

 

 

 

 

I umm-

 

 

....

 

 

 

 

......‘.n....‘.‘

a?"

 

 

 

 

 

 

 

 

   

 

 

 

 

   

   

 

     

   

 

   

   

 

 

 

 

 

 

 

 

 

Ineas

MIIIIIUIIlllllllllllllllllllllllllIIIIIIIIIUIIIHIHHIIUI

293 01576 8512

This is to certify that the

dissertation entitled

The Effects of Corporate Disciosures on Firms'

Information Environments

presented by

Daqing D. 01

has been accepted towards fulfillment
of the requirements for

PhD degree in AggQuntj ng

 

9% 1,41%

C," y Major profess;

Date 7~ 2 2 “ 96

MSUis an Affirmative Action/Equal Opportunity Institution

 

0-12771

 

LIBRARY

Mlchlgan State
University

 

 

 

PLACE N RETURN BOX to romovo this checkout from your record.
To AVOID FINES mum on or before date duo.

DATE DUE DATE DUE DATE DUE

 

2 If" 13 “ f I
n n) ’

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

usu loAn Afflnnotivo Action/Equal Opponunlty lncthlon
mna-DJ

 

THE EFFECTS OF CORPORATE DISCLOSURES
ON FIRMS’ INFORMATION ENVIRONMENTS

By

Daqing D. Qi

A DISSERTATION

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

DOCTOR OF PHILOSOPHY

Department of Accounting

1 996

ABSTRACT

THE EFFECTS OF CORPORATE DISCLOSURES
ON FIRMS’ INFORMATION ENVIRONMENTS ‘

By

Daqing D. Qi

This dissertation seeks to provide input into the debate on the efi‘ectiveness of
corporate disclosures in financial reporting. It consists of two separate, but related papers
that investigate the effects of corporate disclosures on firms’ information environments.
The first paper is an association study on the efi‘ects of corporate disclosures on market
expectations of future earnings. It examines (1) whether stock prices anticipate earnings
information earlier for firms with more informative disclosures than for firms with less
informative disclosures, and (2) which alternative disclosure media contribute to such an
earlier anticipation. Empirical results indicate that market-adjusted returns of firms with
more informative disclosures start to reflect earnings changes 20 months prior to fiscal
year end, about three months ahead of firms with less informative disclosures. This lead is
statistically significant. Further analysis suggests that such an earlier anticipation of prices
over earnings mainly results fiom investor relations, instead of annual reports, quarterly
reports, analyst following, or other factors proxied by firm size.

The second paper is an event study that investigates the effects of preemption and
signal infonnativeness on the incremental information content of annual and lO-K reports.
It addresses the following two research questions: (1) whether stock returns exhibit

abnormal behavior in a three-day event period centered around the earlier of the dates on

which the Securities and Exchange Commission (SEC) receives and makes available to the
public annual reports to shareholders and IO-K reports, and (2) if abnormal returns
behavior is not observed around the receipt and release of these SEC filings, what
alternative explanations may account for its absence. Empirical results in general fail to
detect abnormal returns behavior in the three-day event period. However, evidence
consistent with the existence of incremental information content in the annual and lO-K
reports is found in both the annual earnings announcement period and the period
immediately before the event period, suggesting that firms have released either these

reports or the most relevant information in these reports before filing them with the SEC.

To Suqin

For her love, support, patience, and understanding

iv

ACKNOWLEDGMENTS

I would like to thank members of my dissertation committee, Professors Joe
Anthony (chair), Kathy Petroni, James Wiggins, and Jefi‘rey Wooldridge, for their
invaluable guidance, encouragement, and insights during the course of this dissertation. I
am grateful to Professor James Wheeler for providing me with the Association for
Investment Management and Research Corporate Information Committee Reports, 1983
to 1988, and to Professor Mark Zmijeslcwi for providing me with the SEC filing date data
base developed and maintained at the University of Chicago. I acknowledge the financial
support provided by the Department of Accounting for computing and data collection.

Special thanks go to the other members of my doctoral class, Brian Ballou,
Barbara Esteves, and Norman Godwin. I have enjoyed learning from, and studying with
them for the last four years.

I am forever indebted to my parents. They teach me the values of education,
persistence, and pursuit of excellence, are always proud of me, and never doubt that I will

succeed in life.

TABLE OF CONTENTS

LIST OF TABLE ......................................................................................................... viii
LIST OF FIGURES ......................................................................................................... x
CHAPTER 1
INTRODUCTION AND OVERVIEW ....... . ................. : ........................................... l
1.1 INTRODUCTION ......................................................................................... 1
1.2 OVERVIEW OF THE FIRST PAPER ........................................................... 2
1.3 OVERVIEW OF THE SECOND PAPER ...................................................... 6
CHAPTER 2
THE EFFECTS OF CORPORATE DISCLOSURES ON MARKET
EXPECTATIONS OF FUTURE EARNINGS ............................................................... 11
2.1 BACKGROUND AND MOTIVATION ...................................................... 11
2.2 THEORY AND HYPOTHESES ................................................................. 18
2.2.1 The Timing Hypothesis ................................................................. 21
2.2.2 Effects of Disclosures through Difl‘erent Media ............................. 21
2.3 STATISTICAL METHODS AND VARIABLE MEASUREMENTS .......... 22
2.3.1 Hypothesis 2.1 .............................................................................. 23
2.3.2 Hypothesis 2.2 ...................... ' ........................................................ 26
2.4 SAMPLE AND VARIABLE DESCRIPTIONS ........................................... 28
2.4.1 The AIMR Disclosure Data ........................................................... 28
2.4.2 Sample Selection Criteria and Procedure ....................................... 29
2.4.3 Summary Statistics of Variables .................................................... 32
2.5 EMPIRICAL RESULTS .............................................................................. 37
2.5.1 Testing of Hypothesis 2.1 .............................................................. 37
2.5.2 Sensitivity Analysis for the Testing of Hypothesis 2.1 .................... 43
2.5.3 Testing of Hypothesis 2.2 .............................................................. 47
2.5.4 Sensitivity Analysis for the Testing of Hypothesis 2.2 ................. 49
CHAPTER 3
PREEMPTION, SIGNAL INF ORMATIVENSSS AND THE INCREMENTAL
INFORMATION CONTENT OF THE ANNUAL AND lO-K REPORTS ..................... 51
3.1 BACKGROUND AND MOTIVATION ...................................................... 51
3.2 HYPOTHESIS DEVELOPMENT ............................................................... 55
3.2.1 Abnormal Volatility ....................................................................... 59
3.2.2 Abnormal Returns ......................................................................... 59

vi

3.2.3 The ERC ....................................................................................... 60
3.3 STATISTICAL METHODS AND VARIABLE MEASUREMENTS .......... 60

3.3.1 Proxies for the Informativeness of Disclosures .............................. 60
3.3.2 Dependent Variables ............................................... ' ...................... 62
3.3.3 Hypothesis 3.1 .............................................................................. 63
3.3.4 Hypothesis 3.2 .............................................................................. 64
3.3.5 Hypothesis 3.3 .............................................................................. 65
3.4 SAMPLE AND VARIABLE DESCRIPTIONS .......................................... 65
3.4.1 Sample Selection Criteria and Procedure ....................................... 65
3.4.2 Summary Statistics ........................................................................ 67
3.5 EMPIRICAL RESULTS .............................................................................. 72
3.5.1 Hypothesis 3.1 .............................................................................. 72
3.5.2 Hypothesis 3.2 .............................................................................. 78
3.5.3 Hypothesis 3.3 ......................................... 83
3.5.4 Prerelease before SEC Receipt Dates ............................................ 88
3.5.5 Delayed Responses ....................................................................... 92
CHAPTER 4 . .
SUMMARY AND CONCLUDING REMARKS ........................................................... 94
4.1 SUMMARY AND CONCLUDING REMARKS FOR
TI-EFIRST PAPER..........; ............................. i ............................................ 94
4.2 SUMMARY AND CONCLUDING REMARKS FOR
THE SECOND PAPER ............................................................................... 95
LIST OF REFERENCES ............................................................................................... 98

vii

2.1A

2.13

2.2

2.3A

2.33

2.4

2.5A

2.58

2.6

3.1A

3.18

3.2

3.3

3.4

3.5

3.6

3.7

LIST OF TABLES

Sample Selection Criteria and Procedure ............................................................ 30
Distribution of F inns and Industries. ................................................................... 31
Sample Surrunary Statistics, 1984-1992 .............................................................. 34
Correlation Coemcients between Dependent Variables ....................................... 36
Correlation Coeficients between Informativeness Variables ............................... 38
Results fiom Pooled Regressions for DL and DH Portfolios ............................... 41
Tests Based on the Beginnings of Positive Trends .............................................. 42
Tests Based on the Beginnings of Significantly Positive Correlation .................... 44
Tests for the Efi‘ects of Disclosures through Difi'erent Media .............................. 48
Sample Selection Criteria and Procedure ............................................................ 66
Distribution of Firms and Industries ................................................................... 68
Sample Summary Statistics, 1980-1984 .............................................................. ‘69
Pearson and Spearman Correlation Coeficients .................................................. 71
Values of U, for REC and AEA for Full Sample ................................................. 74
Values of U. for AEA with Full Sample Partitioned by mi. and SIZE“ ............ 76
Values of U. for REC and AEA for Portfolio PLAH ........................................... 77
Values of V, for REC and AEA with Full Sample Partitioned by

the Sign of UEM ............................................................................................... 79

viii

3.8A Values of V, for AEA with Full Sample Partitioned by the Sign of

UEiH] and RIRP" ................................................................................................ 82
3.83 Values of V, for AEA with Full Sample Partitioned by the Sign of

UEM and SIZE, ...................................................................... . ......................... 84
3 .9 Values of V, for REC and AEA with Portfolio PLAH Partitioned

by the Sign of UEM ................. . ......................................................................... 85
3.10 Regression Results for ABA and REC Periods .................................................... 86
3.11 Regression Results in Periods before ABA and REC ........................................... 90
3.12 Regression Results in Periods after ABA and REC ............................................. 93

2.1

2.2

3.1

3.2

3.3

3.4

LIST OF FIGURES

Average CMR for DL and DH Portfolios ........................................................... 39
CMR for DLSM and DHSM Hedge Portfolios ................................................... 46
Sequence of Events Modeled ............................................................................. 56
Values of U, for REC and AEA for the Full Sample ............................................ 73
Values of V, for AEA with Full Sample Partitioned by the Sign of UEM ............ 80
Values of V, for REC With Full Sample Partitioned by the Sign of UEM ............ 81

Chapter 1

INTRODUCTION AND OVERVIEW

1.1 INTRODUCTION

This dissertation seeks to provide input into the debate on the efi‘ectiveness of
corporate disclosures in financial reporting. ' It consists of tyre separate, but related papers
that investigate the efi‘ects of corporate disclosures on firms’ information environments.
The first paper is an association study on the efi'ects of corporate disclosures on market
expectations of fiiture earnings. It examines (1) whether stock prices anticipate earnings
information earlier for firms with more informative disclosures than for firms with less
informative disclosures, and (2) which alternative disclosure media contribute to such an
earlier anticipation. Empirical results indicate that market-adjusted returns of firms with
more informative disclosures start to reflect earnings changes 20 months prior to fiscal
year end, about three months ahead of firms with less informative disclosures. This lead is
' statistically significant. Further analysis suggests that such an earlier anticipation of prices
over earnings mainly results fi'om investor relations, instead of the annual reports to
shareholders (ARS), quarterly reports, analyst following, or other factors proxied by firm
size.

The second paper is an event study that investigates the efi‘ects of preemption and

signal inforrnativeness on the incremental information content of the ARS and lO-K

reports (IO-K). It addresses the following two research questions: (1) whether stock
returns exhibit abnormal behavior in a three-day event period centered around the earlier
of the dates on which the Securities and Exchange Commission (SEC) receives and makes
available to the public the ARS and lO-K, and (2) if abnormal returns behavior is not
observed around the receipt and release of these SEC filings, what alternative explanations
may account for its absence. Empirical results in general fail to detect abnormal returns
behavior in the three-day event period. However, evidence consistent with the existence .
of incremental information content in the ARS and lO-K is found in both the annual
earnings announcement period and the period immediately before the event period,
suggesting that firms have released either these reports to investors or the most relevant
information in these reports before filing them with the SEC.
1.2 OVERVIEW OF THE FIRST PAPER

The first research question in this paper examines the significance of disclosures as
a source of firm-specific information for the market to form expectations of fiiture
earnings. Finance and accounting research fiom an information economics perspective
generally assumes that managers have superior information on their firms’ current and
future performance relative to outside investors.‘ Healy and Palepu (1993) suggest that
disclosures constitute a unique, nonsubstitutable source of such information. In other
words, disclosures contain incremental information and their releases revise market

expectations of fixture earnings. Evidence from empirical research on management

 

‘ Examples include Jensen and Meckling (1976), .Fama and Jensen (l983a and 1983b), and
Holthausen and Leftwich (1983).

earnings forecasts is consistent with Healy and Palepu’s suggestion.2 However, it is not
clear to what degree insights gained fi'om such evidence can be extended to disclosures
other than management earnings forecasts. Most firms do not make management earnings
forecasts. Even for firms that make management earnings forecasts, such forecasts
constitute only a small portion of the overall disclosures released to the public. On the
other hand, empirical capital markets studies such as Ball and Brown (1968), Freeman
(1987), Collins and Kothari (1989), and Kothari and Sloan (1992) document that seCurity
returns anticipate accounting earnings long before their announcements. Moreover,
studies on the valuation implications of disclosures on pensions and fair value estimates
find that market values reflect information in these disclosures prior to their public
releases.3 Such price anticipation of earnings and other firm-specific information in
disclosures indicates the existence of other more timely sources of information, which may
or may not originate from the firms.

The presence of such pre-disclosure information makes it dificult to examine the
content of incremental information in disclosures. Imitigate this problem by focusing on
the effects of disclosures at the early stage of the empirical relation between returns and
accounting earnings. At this stage, prior information is too noisy to afl‘ect market

expectations of earnings in the fiscal year studied. Ceteris-paribus, if disclosures are a

 

2 Examples include Patell (1976), Penman (1980), Waymire (1934), and Pownall et al.
(1993)

3 For example, Barth (1994) examines the relation between fair value disclosures and
bank share prices. She find that, while fair values of investment securities possess
significant incremental explanatory power, such fair values are reflected in bank share
prices at fiscal-year ends before their public releases in the annual reports. Also see
Barth, Beaver, and Landsman (1994).

unique, nonsubstitutable source of firm-specific information, stock prices should reflect
earnings earlier for firms with more informative disclosures than for. firms with less
informative disclosures. Conversely, if the information in disclosures can be substituted by
information fiom other sources, the above empirical regularity should not be observed.

The second research question examines the relative importance of disclosures
released through different media as sources of firm-specific information that bear on
earnings expectations. Most prior studies on the efi'ects of disclosures on earnings
expectations focus on a single aspect of disclosures, such as management eamings
forecasts or segmental reporting. Managers, however, communicate with investors
through different media, such as annual reports, quarterly reports, and investor relations.
These media differ in management discretion, regulatory requirements, and timing
flexibility. Consequently, information disclosed through them may have difl'erent efi‘ects
on market expectations of future earnings. In this paper, I address this research question
by examining whether the earnings response coeficient (ERC), obtained from regressing
abnormal returns in the period of disclosures on unexpected earnings in the next period, is
an increasing filnction of the informativeness of annual reports, quarterly reports, and
investor relations.

Empirically testable hypotheses are derived from a model adapted fi'om Holthausen
and Verrecchia (1988). The Association for Investment Management and Research

Corporate Information Committee Reports‘ (AIMR Reports) are used to develop proxies

 

‘ Published by the Financial Analyst Federation (FAF) Corporate Information Committee
prior to 1989. The F AF has since merged with the Institute of Chartered Analysts to form
AIMR.

for the infonnativeness of finns’ disclosures. Stock exchange and industry memberships
are controlled for through sample selection. For the sample in this paper, market-adjusted
returns of firms with more informative disclosures start to reflect earnings changes 20
months prior to fiscal year end, about three months ahead of those of firms with less
informative disclosures. The lead is statistically significant and robust after controlling for
firm size and the degree of analyst following. Further analysis indicates that such an
earlier anticipation of prices over earnings mainly results from investor (analyst) relatiOns,
instead of annual reports, quarterly reports, analyst following, or other factors proxied by
firm size. Moreover, the size effect becomes statistically insignificant after the efi‘ects of
disclosures are controlled for.

This study contributes to the financial reporting and capital markets literature in
several ways. First, it provides additional evidence that disclosures constitute a unique,
nonsubstitutable source of firm-specific information for market participants to revise
expectations of future earnings. Second, it compares the efi‘ectiveness of disclosures
through difi‘erent media and finds that investor relations are more efi‘ective in
communicating firm-specific information to the market than annual and quarterly reports.
Such a finding suggests that policy makers such as the FASB and the SEC should
encourage firms to disclose more information voluntarily to investors by means such as
“safe harbor” regulations that reduce firms’ legal liabilities in case managers’ ex ante
forecasts do not materialize. Finally, it extends previous capital markets research on
returns-earnings relations by documenting that the degree to which prices lead earnings is

an increasing function of the informativeness of disclosures, and that the previously

documented efl‘ect of size on firms’ information environment may in part be attributable to
more informative voluntary disclosures by larger firms.

Taken together, this paper complements prior empirical research that considers the
relation between disclosures and capital market variables such as the cost of equity capital
(for example, Botosan 1995) and the bid-ask spreads of stocks (for example, Welker
1995). Most prior studies investigate the effects of disclosures on capital market variables
without testing whether the disclosures examined have assisted the stock market in
forming expectations of filture earnings’. This leads to uncertainty on whether, and to
what degree, the empirical results are causal relations as theorized. Most theoretical
models assume that informative signals afi‘ect security valuation through the efi‘ect on
market expectations of future earnings or liquidating dividends. By explicitly documenting
the efi‘ects of disclosures on market expectations of filture earnings, this paper provides
evidence that supports the theorized mechanisms through which disclosures afi‘ect capital
market variables.

1.3 OVERVIEW or THE SECOND PAPER

This paper differs in three ways from previous studies that examine the incremental
information content of ARS and IO-K. First, I explicitly control cross-sectional
difl‘erences in information disclosed prior to the release of these reports and. the
informativeness of these reports themselves. This increases the power of the empirical
tests since Holthausen and Verrecchia (1988) suggest that, under certain conditions, the

magnitude of price reaction to the release of new information is a decreasing function of

 

5 Lang and Lundholm (1994) is an exception. See discussions in the next section.

the inforrnativeness of prior information and an increasing firnction of the inforrnativeness
of the new information. Most previous studies do not explicitly control such factors (see,
for example, Cready and Mynatt 1991, Easton and Zmijewski 1993, Foster and.Vickrey
1978, Foster, Jenkins, and Vickrey 1986, and Stice 1991)‘. Second, I conduct tests based
on both squared market model prediction errors and the empirical relation between returns
and unexpected earnings. While the first test is the dominant methodology used in extant
research in this area, the second test employed in this paper allows the simultaneous
control of difi’erent information environment variables and provides more flexibility in
determining the length of the test period. Third, I explicitly examine alternatiVe
explanations that may have reduced the incremental information content of ARS and lO-K
on their receipt dates by the SEC. Easton and Zmijewski (1993) conjecture that this
information becomes available to the market in a multiple-day period surrounding these
dates. I empirically test two possibilities, that the market may already have had access to
the information contained in ARS and IO-K reports prior to their filings at the SEC, and
that the market may need time to access and evaluate the information and therefore delay
its responses.

The AIMR Reports are used to develop proxies for the infonnativeness of firms’
disclosures prior to and contained in ARS and lO-K. The SEC receipt dates of these
reports are obtained from the SEC filing date data base developed and maintained at the

University of Chicago. Empirical tests are based on a total of 933 firm-year observations

 

‘ Stice (1991) studies the incremental information content of 10-Q and lO-K reports that
are released before earnings announcements, but he does not explicitly control for prior
disclosures or the inforrnativeness of the 10-Q and lO-K.

from 1980 to 1984. F er comparison, most tests are also conducted for a three-day period
centered around the annual earnings announcement date.

Results for the annual earnings announcement period are consistent with
theoretical predictions in Holthausen and Verrecchia (1988) and results reported in
previous empirical studies on the efl‘ects of interim information on security returns
behavior surrounding earnings announcements (see, for example, Atiase 1985, Collins,
Kothari and Rayburn 1987, Foster, Olsen, and Shevlin 1984, Freeman 1987, and Shores
1990). The magnitude of the market response to earnings announcements, measured as
either squared market model errors, abnormal returns, or the earnings response coeficient
(ERC) from regressing abnormal returns on filture unexpected earnings, is smaller for
firms with more informative prior disclosures. Interestingly, the ERC is also significantly
higher for firms with more informative ARS and lO-K, indicating that some of the
information contained in these reports is released to the market in this period’.

I find evidence that is consistent with the existence of incremental information
content in the ARS. and lO-K in the period immediately prior to their SEC receipt and
release. The ERC, obtained from regressing abnormal returns accumulated in the three-
week (15 trading-day) period before the SEC receipt period on future unexpected
earnings, is significantly larger for firms with more informative ARS and IO-K reports, and
significantly smaller for firms with more informative prior disclosures. This finding

suggests that the market has already had access to the information contained in ARS and

 

7 For example, this result can obtain if firms with more informative ARS and IO-K
supplement their earnings announcements with information about revenues and segmental
disclosures. Wilson (1987) reports that some firms in his sample disclose information in
the ARS and IO-K in their earnings news releases.

lO—K reports before their filing with the SEC, indicating that firms have either released
these reports or the most relevant information in these reports to the market before filing
them with the SEC. Such an interpretation is firrther supported by the fact that no
abnormal returns behavior is detected in the three-week period immediately alter the
three-day SEC receipt period.

Empirical results in general do not support the existence of incremental
information content associated, with the SEC receipt and release of the ARS and lO-K.
The majority of tests find no significant incremental information content. Consistent with
results reported in prior studies, the standardized squared market model errors in the event
period are not significantly higher than their theoretical expectation or those during other
periods in the test interval for either the full or the partitioned samples. For the portfolio
of firms with more informative ARS and 10-K and less informative prior disclosures, the
standardized abnormal returns are not significantly positive (negative) for firms with
positive (negative) unexpected eamings. Moreover, the ERC is not significantly higher for
firms with more informative disclosures. On the other hand, one test based on the
standardized abnormal returns for the fill] sample suggests that the SEC receipt and
release of the ARS and lO-K provide incremental information for firms with positive
future unexpected earnings.

This paper makes several contributions to extant research on the incremental
information content of corporate disclosures. First and foremost, it documents systematic
returns behavior that can be attributed to the inforrnativeness of disclosures contained in

ARS and IO-K in both the earnings announcement period and the period prior to the SEC

10

receipt of these reports. The existence of such behavior is consistent with the conjecture
that a substantial number of firms release the ARS and lO-K or the most relevant
information in these reports to market participants before filing them with the SEC. This
explains the failure of previous studies in detecting abnormal returns behavior on the SEC
receipt dates and provides evidence that is consistent with the existence of incremental
information in these reports. Second, the absence of abnormal returns behavior under the
majority of tests on the SEC receipt dates under a refined, more powerful research design
lends filrther support to Easton and Zmijewski (1993)’s warning that “using the earlier of
the SEC ARS and IO-K receipt dates as the date of the first public disclosure of the
information in the annual report may introduce considerable error.” This is not a trivial
issue because it has implications on how to interpret the results fi'om empirical studies
which assume that non-earnings information is not available to the public until these or
similar dates'. Finally, this paper extends extant studies that use size as a proxy for the
availability of prior information to investigate market reaction to annual earnings
announcements and provides direct evidence that the magnitude of the response is
negatively associated with the inforrnativeness of prior disclosures.

For the rest of this dissertation, Chapter 2 examines the efl’ects of corporate
disclosures on market expectations of future earnings. Chapter 3 investigates the efi‘ects
of preemption and signal inforrnativeness on the incremental information content of the

ARS and IO-K. Concluding remarks are provided in Chapter 4.

 

' See footnote 1 in Easton and Zmijewski (1993) for a list of such studies.

Chapter 2

THE EFFECTS OF CORPORATE DISCLOSURES ON
MARKET EXPECTATIONS OF FUTURE EARNINGS

This chapter examines whether corporate disclosures assist investors in forming
expectations of firture earnings. It is organized as follows. Section 2.1 provides
background and motivation. Section 2.2 discusses a model that links disclosures, earnings
expectations, and stock returns and develops hypotheses. Statistical methods for
hypothesis testing and variable measurements are outlined in section 2.3. Section 2.4
describes the sample and variables. The last section reports empirical results and sensitivity
analysis.

2.1 BACKGROUND AND MOTIVATION

Ever since the passage of the Securities Act of 1933 and the Securities Exchange
Act of' 1934, financial reporting in the United States has been developed and designed to
protect the interests of stakeholders in publicly-traded corporations, particularly those of
stockholders, by providing them with decision-relevant information about these entities.
According to the Financial Accounting Standards Board (FASB), financial reporting
should “provide information to help present and potential investors and creditors and other

users in making rational investment, credit, and similar decisions,” and “the primary

ll

12.

focus of financial reporting is information about an enterprise’s performance provided by
measures of earnings and its components.”9

While financial statements are a central feature of financial reporting, a large
amount of information is communicated to the public through corporate disclosures.
Some disclosures, such as news releases and management’s earnings forecasts, are
voluntary and subject to management discretion. Others, however, are mandated by
either the Generally Accepted Accounting Principles (GAAP) or SEC regulations. Barth
and Murphy (1994) examine the purposes, subject, number, and trends of financial
statement disclosures required by the FASB and its predecessors. They report that 454
disclosure items are mandatory under GAAP through Statement of Financial Accounting
Standards (SFAS) 109. Moreover, there exists a clear trend of increasing disclosure
requirements over time, and few of the requirements have been eliminated once adopted.

The increasing number of disclosure requirements has, in recent years, led to
concerns and debates about disclosure overload, i.e., whether too many disclosure items
are required under GAAP given the costs of making such disclosures. In 1991, the
American Institute of Certified public Accountants (AICPA) formed the Special
Committee on Financial Reporting to address concerns about the relevance and usefulness
of business reporting In a report released in 1994, the committee states that

Because business reporting is not free, improving it requires considering the
relative costs and benefits of information, just as costs and benefits are key to

 

9 FASB. 1978. Objectives of Financial Reporting by Business Enterprises. Statement of
Financial Accounting Concepts No.1. Stamford, Conn: FASB.

l3

determining the features included in any product. Undisciplined expansion of
mandated reporting could result in large and needless costs. 1°

As a result, it calls for standard setters and regulators to better understand the costs and
benefits of business reporting and to “search for and eliminate less relevant disclosures.”ll

Currently, both the SEC and F ASB are examining the effectiveness of mandated
disclosures and searching for measures to improve the present system. The SEC has
formed an internal task force to review its existing corporate disclosure regulations and
seek detailed views from corporate leaders. The FASB, meanwhile, is considering
whether to add a formal project on disclosure efi‘ectiveness to its technical agenda and
calling for research inputs on this issue from all interested parties, especially academic
researchers. ‘2

To date, theoretical research has provided useful insights on the cost-benefit
tradeofi‘ of financial reporting. Audited mandatory disclosures can reduce transaction
costs and increase market liquidity by mitigating the incentives problems between
managers and investors, and between informed and uninformed investorsl3 . Moreover,
both mandated and voluntary disclosures can enable managers to better difi'erentiate their
firms fi'om the “lemons” to achieve costs of equity capital that are lower than otherwise, as
noted in Akerlof (1970) and Spence (1973). On the other hand, disclosures are costly,

incurring not only the costs of actually preparing and disclosing the information but also

 

1° The AICPA Special Committee on Financial Reporting. 1994. Improving Business
Reporting - A Customer Focus. Jersey City, NJ: AICPA.

ibid.
‘2 Beresford, D. and J. Hepp. 1995. Financial Statement Disclosures: Too Many or Too
Few? Financial Accounting Series (No. 149-B). Stamford, Conn: FASB.
is See Jensen and Meckling (1976), Hakansson (1977), Fama and Jensen (1983a and
1983b), Holthausen and Lefiwich(1983), and Beaver (1989).

14

the opportunity costs due to the loss of competitive advantage to competitors as a result
of publicly disclosing the information. Consequently, at the aggregate level, firms should
be required to disclose additional information only when the incremental social benefits are
greater than the incremental social costs. At the individual firm level, managers may
choose to disclose more information voluntarily until the marginal benefits accruing to the
firm equal the marginal costs.

While few dispute the theoretical importance of adequate financial reporting and
the cost-benefit tradeofi‘ involved in the process, empirical evidence on the benefits of
corporate disclosures has been limited. The majority of empirical accounting research on
capital markets and financial reporting focuses on accounting numbers recognized in the
financial statements, such as earnings, cash flows, and their components. Of the more
recent studies that deal directly with items disclosed but not recognized in financial
statements, most concentrate on the valuation implications, instead of the disclosure
efi'ectiveness, of a single disclosed item or set of items that relate to a single subject, such
as pensions, current cost accounting of oil and gas properties, and market value of
marketable securities. " Because these papers do not address the overall infonnativeness
of corporate disclosures and do not control for the efi’ects of information from other
sources, they provide only limited insights on the benefits of disclosures.

Several current manuscripts have taken a more global approach to investigate the
benefits of corporate disclosures. Lang and Lundholm (1994), for instance, indicate that

firms with more forthcoming disclosure policies have . a larger analyst following, more

 

“ See footnote 1 in Barth and Sweeney (1995) for a list of papers in this area of research.

15

accurate analyst forecasts for earnings in the same fiscal year, less dispersion among
analyst forecasts, and less variability in forecast revisions. Byrd, Johnson, and Johnson
(1994) present evidence that CEO presentations are positively correlated with analyst
following and, in the case of lightly followed firms, a reduction of the cost of equity capital
as measured by their equity beta. Welker (1995) documents that the stocks of firms with
more forthcoming disclosures have smaller bid-ask spreads. Botosan (1995) finds that
greater voluntary disclosure in annual reports is associated with a lower cost of equity
capital alter controlling for cross-sectional variation in systematic risk and size, provided
that a measure other than market value is used to proxy size. Healy, Palepu, and Sweeney
(1995) provide evidence that disclosure improvement is associated with a reduction of the
dispersion of analyst earnings forecasts, a decline in bid-ask spreads for the test firms, and
an increase in analyst following.

Taken together, these more recent studies present evidence consistent with
hypothesized relations between disclosures and analyst following, analyst forecast
dispersion, analyst forecast accuracy of earnings in the current fiscal year, and stock
liquidity. These tests also provide weak support for the hypothesis that the
inforrnativeness of disclosures is negatively correlated with the cost of equity capital.
However, they provide only limited evidence on whether the disclosures examined have

assisted the stock market in improving expectations of sample firms’ future events,
especially future earnings. Consequently, it remains rather uncertain whether, and to what
degree, such associations are causal relations as predicted by theoretical papers cited in

these empirical studies.

16.

Several theoretical papers have provided insights on the possible mechanisms
through which more informative disclosures lead to lower costs of equity capital. Barry
and Brown (1985) investigate the lack of information as a source of nondiversifiable risk.
Merton (1987) studies the relation between investor recognition and the cost of equity
capital. Lev (1988), Diamond and Verrecchia (1991), and Elliot and Jacobson (1994)
explore the efi‘ects of adverse selection on the cost of equity capital; While their focuses
difi’er, the existence of informative sigrxals is explicitly or implicitly assumed. These
informative signals enable investors to form better expectations of fixture events such as
future eamings or cash flows, leading to a lower cost of capital through reduced
nondiversifiable risk, additional investor recognition, mitigated adverse selection, or more
likely, a combination of these factors. Consequently, for disclosures to afi'ect stock
liquidity and the cost of equity capital, a prerequisite applies. That is, afier controlling for
the effects of other information, the items disclosed must provide additional information to
investors, and more informative disclosures should be more efi’ective in assisting investors
in the formation of expectations of fixture events such as earnings.

Prior research has provided some evidence on the efi‘ects of disclosures on the
formation of eamings expectations. Studies on management earnings forecasts show that
forecasts are price informative (i.e., Patell 1976; Penman 1980, Waymire 1984, Pownall et
al. 1993). Baldwin (1984) finds that analyst earnings forecasts become more accurate for
multisegment firms alter the adoption of segmental reporting requirements. Gill (1994)
finds that both analysts and the stock market react to firms’ qualitative comments on

earnings, but do not react to announcements on cost-cuttings, capital expenditures and

17.

price changes. Lang and Lundholm (1994) provide evidence that firms with more
forthcoming disclosure policies have more accurate analyst forecasts for earnings in the
same fiscal year.

Although these prior studies, taken together, indicate that disclosures lead to more
accurate earnings expectations, they are subject to two limitations. First, they only
address the effects of disclosures on the expectations of earnings in'the same fiscal year.
Consequently, it remains unresolved whether and when disclosures can assist investors in '
forming expectations of earnings beyond the current fiscal year. Evidence presented in
Freeman (1987), Collins and Kothari (1989), and Kothari and Sloan (1992) shows that
prices start to incorporate earnings information well before the beginning of the fiscal year.
The degree to which price anticipates earnings varies systematically with some variables,
such as firms’ market capitalization. ‘5 Because managers have superior information about
their firms’ fixture performance unobservable to outsiders, as noted in Healy and Palepu
(1993), disclosures can be an important determinant on the degree to which prices lead
earnings.

Second, most prior studies, with the exception of Lang and. Lundholm (1994),
focus on just one aspect of disclosures, such as management earnings forecasts or
segmental reporting. Managers, however, communicate with investors through several
media, including conversations with financial analysts, press releases, quarterly reports,

and annual and lO-K reports. As a result, when studying the overall efi’ects of disclosures

 

lsFreeman (1987) finds that the percentage of large-finn abnormal returns realized in
‘early’ months exceeds the percentage for small firms, but the difi‘erence in the lead times
of prices over earnings is not statistically significant.

18

on market expectations of fixture earnings, a comprehensive measure that can aggregate
disclosures through difi‘erent media seems more appropriate. On the other hand,
examining disclosures through different media separately provides no insights on their
relative contributions to more accurate earnings expectations. An empirical examination
of this issue is important and of interest to policy makers because while investor relations
are largely voluntary, almost all required disclosures are released through quarterly and
especially annual reports.

In this paper, I attempt to overcome the limitations by investigating two
interrelated research questions: (1) whether stock prices anticipate earnings information
earlier for firms with more informative disclosures than for firms with less informative
disclosures, and (2) which categories of disclosures contribute to such an earlier
anticipation.

2.2 THEORY AND HYPOTHESES

Even though changes in investors’ earnings expectations are not directly
observable, they can be inferred fiom abnormal security returns. Three assumptions are
made to link stock returns and earnings: (1) stock price equals the present value of
expected fixture dividends, (2) the discount rate is constant over time, and (3) the present
value of the revisions in expectations of fixture dividends is the same as the present value
of the revisions in expectations of future earnings. As Lipe (1990) notes, the first two
assumptions are commonly adopted in finance and accounting research, and the last
assumption can be interpreted as an extreme version of the statement that accounting

earnings provide information about the fixture dividend paying ability of the firm.

19

Taken together, these assumptions imply that the releases of signals that provide
usefixl information on fixture earnings lead to share price revisions reflected as abnormal
returns. If earnings announcements are the only source available for earnings information,
abnormal returns will only be observed when accounting earnings are announced. This
scenario, of course, is unrealistic. For publicly-traded firms, relevant information about
earnings is available from many other more timely sources. They include, but are not
limited to, articles in trade journals, earnings releases by competitors, analyst earnings
forecasts, and mandatory and voluntary corporate disclosures.

Under the current financial accounting and reporting system, corporate disclosures
are assumed to be an important source of incremental information about fixture earnings.
FASB believes that corporate disclosures serve four purposes, (1) to describe and provide
additional relevant measures of items that are recognized on the face of the financial
statements, (2) to describe and provide usefixl measures of items that are not recognized in
the financial statements, (3) to provide information to help investors and creditors assess
risks and potentials of both recognized and unrecognized items, and (4) to provide
important information in the interim while other accounting issues are being studied in
more depth. “5 Because the existing accrual accounting system under GAAP is based on
historical transaction data and emphasizes objectivity, verifiability, and conservatism,
instead of unbiased estimation of fixture earnings, disclosure items can reveal, either

directly or indirectly, relevant information about fixture earnings in addition to financial

 

‘6 FASB. 1990. Disclosure of Information about Financial Instruments with OE-Balance-
Sheet Risk and Financial Instruments with Concentrations of Credit Risk. Statement of
Financial Accounting Standards No. 105. Norwalk, Conn.: FASB.

20

statements. As a result, in year t-l investors can update and improve their expectations of
earnings in year t. Under the three valuation assumptions, the revision can be observed in
the form of abnormal stock returns.

The discussion above can be formalized using a model adapted fi'om Holthausen
and Verrecchia (1988). While Holthausen and Verrecchia (1988) consider the sequential
release of information, a single-signal model is suficient for an investigation of the return-
earnings relation in its early stage because signals previously released are too noisy to
provide information about unexpected earnings in the period studied. For simplicity,
assume that the earnings generation process is a random walk such that:

EPS, eEPS,_, + e,, (2.1)
where e, is normally distributed with a mean of 0 and a variance of v.. Let D,.. be a signal
contained in a disclosure at t-s, such as a management discussion of a new product with
analysts or the release of a quarterly report. It provides information about e, such that:

DM = e, +d,_,, (2.2)
where d ,. is a normally distributed random variable that is uncorrelated with e, and has a
mean of O and a variance of v4. .

‘ Before the release of D... the expectation of e, is zero. It can be shown that, after
the release of D..., the expectation of e, is:

E(etIDr-s)= LE1“; ‘ (23)

vs+vd

21.

For simplicity, assume that 0,. provides no additional information about EPSt-ts ignore the
time value of money between t-s and t- l , and normalize the share price before the release
of D... to one. The stock return associated with the release of D,.. is therefore:

.#
RET =%*:' D“ . (2.4)

t-s
+ Va

 

where r is the cost of equity capital.
2.2.1 The Timing Hypothesis _

In the above expression, the return associated with the release of D... is an
increasing fixnction of its precision, I/V¢ The more precise the signal, the greater is the
price revision. Iffirm A’s disclosure practice is more informative than that of firm B, i.e.,
for a given value of s, l/v..(s, A) is always greater than 1/va(s, B), then the magnitude of
RET...(A) is always greater than that of RET,..(B) for the same realization of D". If RET..
, must reach a minimum level of magnitude to be empirically observable, then it would be
first observed for firm A. This leads to the first hypothesis, in its alternative form:

Ceteris paribus, the abnormal returns associated with unexpected accounting

earnings of year t begins earlier for firms with more informative disclosures than

for firms with less informative disclosures.
2.2.2 Effects of Disclosures through Different Media

Information about e, can be disclosed to the public through either the annual
reports quarterly reports, direct communications twith investors, or a combination of the
above. The concern is which media of disclosures are efi‘ective in conveying information

about e, to the investors in year H. In equation (2.3), let D,.. represent the aggregate of

all disclosures released in year t-l. Assuming that information disclosed before year H is

22.

not informative about e,, which is consistent with findings in Freeman (1987) and Collins

and Kothari (1989)", equation (2.4) can be rewritten as:

 

*v,‘D,_, _ l ,[vfl'e +v.*d,_,]

Vi-Vd r v.-+-v,l v.+v, (2.5)

Notice that, by definition, e, and do, are uncorrelated. As a result, the earnings response
coeficient (ERC) in regression of RETH on e, is then:

ERCH =5 V' . (2.6)
r ve-i-vd

 

It is easy to verify that ERC“ is an increasing fixnction of UV... Because more informative
annual reports, quarterly reports and better investor relations provide more precise
information about e,, the relation in (2.6) leads to the second hypothesis:
Ceteris paribus, the ERC from regressing cumulative abnormal returns in year H
on unemected earnings in year t is positively correlated with the infonxrativeness
of annual reports, the informativeness of quaterly reports, and the
infannativeness of investor relations.
2.3 STATISTICAL METHODS AND VARIABLE MEASUREMENTS
Nine years of data, from 1984 to 1992, are used for hypothesis testing. A proxy
for the overall inforrnativeness of disclosures in the prior year is obtained for each
observation based on the relative industry rankings of analysts’ total evaluation scores in
the AIMR Reports (RIRTM). Proxies for the inforrnativeness of disclosures through

annual reports, quarterly reports and investor relations are based on the relative industry

 

‘7 Freeman (1987) reports that abnormal returns begin to reflect earnings changes 22 and
19 months before fiscal year end for large firms and small firms respectively, which are
consistent with results of figures 1 and 2 in Collins and Kothari (1989).

23

rankings of analyst evaluation scores for the three categories and are denoted as RIRAH,
RIRQH, and RIRIM respectively.
2.3.1 Hypothesis 2.1

Hypothesis one states that returns should anticipate earnings earlier for firms with
more informative disclosures than for firms with less informative disclosures. To test
hypothesis one, each annual subsample is divided into three portfolios, DH, DM and DL,
according to RIRTH, with DH being the most informative and DL the least informative.
RIRTa. 1 is used because it captures the overall effectiveness of the finn’s disclosure
practice. A matched-pair design is used to document the difi‘erence regarding when
returns start to reflect the change of earnings of year t for DH and DL portfolios.
Statistical tests are based on the intertemporal distributions of the difi’erence.

If earnings-relevant signals exist for multiple firms simultaneously, abnormal
returns can be realized based on foreknowledge of such signals. A zero-investment hedge
portfolio can be formed by taking an equally-weighted long position in firms with good
news and a similar short position in firms with bad news, with the nature of the news
derived fiom the sigrxals. In empirical research, however, the signals themselves are
dimcult, if not impossible, to observe. As a result, ex-post' earnings realizations are used
to separate firms into goods news and bad news groups, as in Freeman (1987). The first
test of hypothesis one will be based on the behavior of cumulative market-adjusted returns

of such earnings-based hedge portfolios. 1'

 

1' Since the hedge portfolio is formed by taking an equally-weighted long position in good
news firms and an equally-weighted short position in bad news firms, cumulative market-
adjusted return for the portfolios is the same as cumulative raw return in the first test of

24

Define UE;, as scaled unexpected earnings per share. Assuming that the earnings-
generating process follows a random walk, UK, can be obtained by subtracting primary
earnings per share before extraordinary items (EPS) of year t-l fiom that of year t and
then scaling the difference by share price at the beginning of year t-l. Both EPS and share
prices are adjusted for stock splits and stock dividends.

For each sample year t, hedge portfolios are constructed separately for DH and DL
firms by taking long positions in firms with (IE, >0, and short positions in firms with
UE;,<O. Equal weights are given to firms within the same news groups. Market-adjusted
returns (MR) are calculated for each firm in each of the 36 months fi'om the beginning of
year H to the end of year t +1. For firm i in sample year t, market-adjusted retum in
month t is:

MR... = r... - r...
where r,,, is actual return of firm i in month t, and r..., is the CRSP equally weighted
market index. The monthly market-adjusted returns for the two hedge portfolios can then
be calculated. For instance, if the DH hedge portfolio contains j = 1, ..., J good news
firms and j’ = l, J’ bad news firms, its monthly market-adjusted return in month t is:

MR(DH),,=-l-"iMR "grim.
l I in J I

in.
Once monthly market-adjusted returns are obtained, cumulative market-adjusted returns
(CMR), fi'om the beginning of year M to the end of month t, are computed for each

month, t = l, 36, as:

 

hypothesis one. The former is used to be consistent with the second test of hypothesis
one, where the cumulative market-adjusted return is used.

25

CMR(DH),, = 5:;MR(DH)&.

With CMR so defined, six-month cumulative market-adjusted return starting a months

afier the beginning of year M is: ‘
ACMR(DH), = CMR(DH)_,, —CMR(DH)_.

As an example, in sample year 1984, the six-month cumulative market-adjusted return for

the DH hedge portfolio starting four months after the beginning of year t-l (1983) is

ACMR(DH)4 = CMR(DH)""n -'-CMR(DH)""..

For the DL hedge portfolio, MR(DL),,, CMR(DL)., , and ACMR(DL). are similarly
defined. Under hypothesis one, the abnormal returns associated with accounting earnings
of year t begin earlier for DH firms than for DL firms. This means that, empirically,
CMR(DH) should exhibit a positive trend earlier than CMR(DL). The statistical test is
based on the intertemporal distribution of the difi'erence in the beginnings of positive
trends for DL and DH hedge portfolios. Define the beginning of a positive trend for a
hedge portfolio as the point in time from which cumulative market-adjusted returns remain
positive over time until reaching its maximum. Assume that, in sample year t, positive
trends start m(DL), and m(DH), months after the beginning of year t-l for DH and DL
portfolios respectively. Their difference is denoted as:
Am, = m(DL), - m(DH),.

For example, for sample year 1985, if positive trends start six and two months after the
beginning of 1984 for the DL and DH portfolios respectively, then m(DL) = 6, m(DI-I) =

2, and Am, = 4, indicating that the positive trend starts four months earlier for the DH

26_

portfolio. Under H1, Am, should be significantly positive. Since the distribution of Am, is
unknown and the number of years is limited to nine, both the Wilcoxon signed rank test
and the t test are conducted.

The first test of hypothesis one ignores the magnitude of UEt. and is limited by the
definition of the beginning of a positive trend. The cumulative market-adjusted return
could have increased monotonically for several months until being disrupted by a large
negative random shock. On the other hand, a positive trend as defined above can remain
tentative for a substantial period of time. To mitigate this limitation, hypothesis one is also
tested using a second method. The following regression, I

ACMR(DH)i,n = al.,I + BuUEi, +6m,

is cenducted in each sample year separately for DH and DL firms for n = 0, 1,..., 24. In
the regression, ACMRtu is cumulative market-adjusted returns over a six-month moving
window for firm i, similarly defined as ACMR(DH) and ACMR(DL). In sample year t,
the first window starts fiom the beginning of year t-l (n = 0), and the last window starts
24 months later (11 = 24). The first 11 value for which B. is statistically positive at the 0.05
level . (one-sided) is noted as -n(DH), for the DH firms and n(DL), for. DL firms
respectively. Their difl'erence is defined as An, = n(DL), - n(DH),. Under hypothesis one,
An should be significantly positive, which is tested based on the intertemporal distribution
of Ant using both the ercoxon signed rank test and the t test.
2.3.2 Hypothesis 2.2

Hypothesis two posits that the ERC fi'om regressing cumulative abnormal returns

in year H on unexpected earnings in year t is positively correlated with the

27,

inforrnativeness of annual reports, the inforrnativeness of quarterly reports, and the
inforrnativeness of investor relations. It is tested via the following regression in which
observations are pooled both cross-sectionally and intertemporally:

CMR,“ = a + Ex, * p, +p,UE, + p,araA,_, * UE, + p,1uao,_, * UE, + p,1mu,_, * us,
.5
+1, ‘LSIZEM *UE“ +1, "'ANA,,_I ‘UE, +8“.

where

CMR,“ = market-adjusted returns in year t-l for firm i cumulated from the fifth
month of year H to the fourth month of year t, obtained fiom CRSP
tapes",

UEg, = the difference between earnings per share before extraordinary items of-
year t and that of year t-l, scaled by per share price at the beginning of
year t-l, obtained fi'om the Compustat tapes, ‘

D, = a dummy variable that equals one for year t, and zero otherwise,

RIRA.“ = relative industry ranking of the inforrnativeness of annual report, for
firm i in year H, to be defined in the next section,

RIRQM = relative industry making of the inforrnativeness of quarterly reports, for
firm i in year t-l, to be defined in the next section,

Rm“. 1 = relative industry ranking of the inforrnativeness of investor relations, for
firm i in year H, to be defined in the next section,

LSIZEM = the logarithm of SIZE“, the market value at the beginning of year H
for firm i, obtained fi'om Compustat tapes,

ANA“ = the total number of analyst forecasts made in year t-l for BPS,” of firm
i, obtained fi'om IBES tapes,

8;, = a random, normally distributed error term.

 

‘9 Equally-weighted market return index is used. The window is selected to capture
disclosures made in year H and exclude those released in other time periods. Note that
annual and IO-K reports of year M are not released until early months of year t. See
Alford, Jones and Zmijewski (1994) for fixrther reference.

28

LSIZEM and ANA,“ are included to control for the potential efi‘ects of size and
analyst following on firms’ information environments. Under hypothesis two, 32. Ba, and
34 should be significantly positive.

2.4 SAMPLE AND VARIABLE DESCRIPTIONS .
2.4.1 The AIMR Disclosure Data

The sample years in this study range fi'om 1984 to 1992. The AIMR Reports,
fiom 1983 to 1991, are used to obtain proxies for the inforrnativeness of firms’ disclosure
practices in the year prior to each sample year. According to the AIMR Corporate
Information Committee, each year an industry-specific subcommittee20 evaluates the
inforrnativeness of selected firms’ disclosures along three dimensions: annual published
information, quarterly and other published information, and investor relations and other
aspects. Characteristics and issues unique to the industry are taken into consideration in
the evaluation process. Scores along these three dimensions are then weighted to obtain
an overall score about the inforrnativeness of the firrn’s disclosure practices. The weights
are in general 40-50 percent for the annual published information, 30-40 percent for the
quarterly and other published information, and 20-30 percent for investor relations and
other aspects. 'While the majority of the subcommittees report both overall and category
scores, some subcommittees publish only the overall scores. As a result, about one-third

of firm years in the AIMR Reports have no category scores reported.

 

2° The AIMR reports that these subcommittees are composed of leading analysts following
the industries being evaluated.

29

2.4.2 Sample Selection Criteria and Procedure

Table 2.1A describes the sample selection process. Filters are imposed to
eliminate or reduce potential confounding factors. The 1983-1991 AIMR Reports contain
4,390 firm-years, of which 168 firm-years are eliminated due to the unavailability of CRSP
data.21 The filtering process then eliminates 408 firm-years due to the lack of Compustat
data, 1,036 non-NYSE firm-years to control for the efi’ects of difl‘erent stock exchanges
(Grant 1980), 276 firm-years due to the lack of analyst following data per IBES tapes, 18
firm-years with (IE, larger than one, 564 firm-years with non-December 31 fiscal year
end, 532 firm-years in which the industry subcommittees do not report category scores,22
and finally, 39 firm-years with only two observations in their respective industry-year
groups.23

The selection procedure described above yields a sample of 1,349 firm-years from
287 firms. As shown in Table 2. 18, the number of firms in each year ranges fi'om 99 in
1986 to 214 in 1991, while the number of industries in each year ranges fiom 14 in 1986
to 26 in 1991. The number of observations in each industry-year group is three at the
minimum by research design. It is 16 at the maximum, indicating that, in any given year,

the sample is not dominated by a small number of industries.

 

2‘ Mon are in either international banking or international pharmaceutical industries.

22 This is due to three considerations. First, it is not clear why the subcommittees do not
disclose the category rankings, and one possible reason is that the industries involved have
information environments different from the remaining industries. Second, and more
importantly, the majority of firm-years without category ranking data belong to financial
services industries such as banking and insurance. Since financial services industries are
regulated, their information environments are expected to be difi‘erent from other
industries. Finally, category scores are needed for the tests of hypothesis two.

23 At least three observations are needed in each industry year to construct DL, DM, and
DH portfolios.

30

Table 2.1A
Sample Selection Criteria and Procedure

 

Numbcrof .

 

firm-year
Firm-years covered in AIMR Reports,
1983-1991
4,390
Less firm-years:
( 168)

without CU SIPs

without Compustat data ( 408)

with at least one monthly return from

the 1st month of year tel to the last month (1,036)

of year t+1 missing from CRSP

monthly return files

without analyst following data ( 276)

with sealed unexpected earnings ( 18)

larger than one

with non-December 31 fiscal year ( 564)

ends

without AIMR category rankings 1 ( 532)

withlessthanthreefirmsinannual ( 39)

industry groups

Firm-years included in the sample ' 1,349

 

31.

Table 2.1B
Distribution of Firms and Industries

 

 

Year Number of firms Number of industries
in the sample in the sample

1984 124 16

1985 128 17

1986 99 14

1987 109 . 14

1988 128 18

1989 160 22

1990 199 25

1991 214 26

1992 m 3;
Totals 1,349 175

 

32.

2.4.3 Summary Statistics of Variables

Because firms in difi‘erent industries are evaluated by difi‘erent subcommittees and
members of the same subcommittees are not necessarily the same in difl'erent years, the
raw scores reported by the subcommittees must be standardized to provide a meaningfiil
proxy for disclosure practices in the year prior to the sample year. This is achieved by
defining relative industry rankings of overall inforrnativeness of disclosures (RIRTM) as
the following:

RANKT},_
“Rn-I =‘fi—T
jr-I

where N,“ is the number of firms for industryj in year t-l, and RANKTM is the rank, in
ascending order, of firm i in year t-l within industry j based on its overall disclosure score.
RIRAM, RIRQM, and RIRIM are similarly obtained as proxies for the inforrnativeness of
annual reports, quarterly reports, and investor relations.

As Lang and Lundholm (1994) note, relative industry ranking of disclosure scores
reflects only intra-industry variation in disclosure inforrnativeness. However, this
limitation does not pose a problem for this study, particularly for hypothesis one. Because
firms’ infomation environments are affected by industry membership, inter-industry
variation in disclosure inforrnativeness must be controlled, even if proxies for disclosure
inforrmtiveness are available across industries. The industry effect is mitigated by the
following procedure. Each year, firms in the same industries are placed into one of three

portfolios, DL, DM, and DH, based on the ascending order of their RIRTM values. The

33

number of firms in DL and DH are always kept the same.“ Consequently, tests based on

the difi‘erences in how the returns of DL and ‘DH firms anticipate firture earnings

information are not likely to be severely affected by the industry efi‘ect.
Table 2.2 presents summary statistics for the variables. Most have been previously
defined, except for the following:

Common Stock Beta (BETAM) = beta of common stock calculated by regressing
monthly returns on the NYSE equally-weighted
monthly index, at the beginning of year t-l and
obtained fi'om the CRSP tapes,

Book-to-Market Ratio (BMM) = the book value of the firm i divided by the market

value at the beginning of year t-l, obtained from the
Compustat tapes,

Earnings-to-Price Ratio (13PM) = EPS divided by per share price at the beginning of
year t-l, obtained fi'om the Compustat tapes.

These three variables are included because they have been used in previous studies as
proxies for the cost of equity capital, which under several theories are directly or indirectly
afi‘ected by disclosure inforrnativeness. Descriptive statistics for RIRTM, NRA“. 1,
RIRQM, and RIRIM are not included because statistics for relative rankings are not
meaningful. ,
As Table 2.2 shows, the average firm in the sample is rather actively followed by
the analysts. The mean (median) of ANA“ is 21.27 (21). This is of little surprise since

many subcommittees exclude firms that are not evaluated by a minimum number of

 

2‘ For instance, of the 11 airlines evaluated by the airline subcommittee in 1984, seven
pass the sample selection procedure. 0f the seven firms, the bottom two firms are
assigned to DL, the top two firms to DH, and the rest to DM. Due to this procedure, D1.
and DH have exactly the same industry composition.

34

Table 2.2

Sample Summary Statistics, 1984-1992

 

 

 

 

Percentiles‘

N Mean Sthev. 1% 25% 50% 75% 99%
ANA... 1,349 21.274 9.202 4 14 21 27 43
BETA... 1,310 1.015 0.303 0.403 0.816 1.019 1.195 1.853
BM... 1,349 0.661 I 0.375 0.027 0.423 0.627 0.865 1.632
EP... 1,349 0.062 0.135 -0.3 16 0.052 0.074 0.101 0.205
SIZE... 1,349 4,670 7,874 1 14 895 2,140 4,929 41,546
UE. 1,349 -0.005 0.106 -0.389 -0.029 0.005 0.021 0.273
Variables definitions:

ANA... =thetotalnumberofforecastsmadebyanalystsforEPS... offirmiinyeart-l,

BETA... = beta of common stock calculated by regressing monthly returns on the NYSE equally weighted

BM'a-l
EPit-l

monthly index, at the beginning of year t-l,

=EPSdividedbypersharepriceatthebeginningofyeart-1,

=thebookvalueofthefirmidividedbythemarketvalueatthebeginningofyeart-l,

SIZE... = the market value ofthefirm atthebeginning ofyear H, in millions ofdollars,

UE. =unexpectedearningsoffirmiinyeartscaledbystockpriceatthebeginningofyeart-l.

35

subcommittee members. Descriptive statistics also indicate that the distribution of SIZE...
. is skewed to the right, as indicated by a mean ($4,670 million) that is much larger than
the median ($2,140 million), and a standard deviation ($7,874 million) that is greater than
the mean. As a result, its logarithm is used in correlation analysis.

Table 2.3A reports both Pearson (below the diagonal) and Spearman rank (above
the diagonal) correlation coeficients between variables in Table 2.2 along with RIRTM
The two sets of correlation coemcients of RIRTg... with other variables are similar and the
discussions are based on Spearman rank correlation coeficients. Consistent with results
reported in previous studies, RIRTg... is positively correlated with both analyst following
(p=0.0001, two-sided) and size (p=0.0028, two-sided). man... is not correlated with
UEa... (p=0.1075, two-sided), indicating that results in this study are unlikely to be afi‘ected
by the empirical regularity that firms with good news tend to announce earnings earlier
than those with bad news. Because the correlation between RIRTi... and BETA... is not
statistically significant (p=0.2683, two-sided), the use of market-adjusted returns is
appropriate. RIRT... is marginally positively associated with EP... (p=0.0757, two-sided).
As discussed in Penman (1994), EP.... is substantially influenced by temporary deviations
from permanent earnings and is a poor proxy of the cost of capital. Consequently,
the positive ' correlation between RIRT.... and EP.... should not be interpreted
as evidence that firms with more informative disclosures on average have higher
costs of capital. Interestingly, while RIRTM is not significantly correlated with

BM... (p = 0.1673, two-sided), it is significantly negatively correlated with BM.

36.

Table 2.3A
Correlation Coefiicients between Dependent Variables'

 

ANAg... BET A... BM“ EP... RIRT... LSIZE“ UE-g

 

ANA-.4 -0.2566 0. 1789 -0.0169
0.0001 0.0001 0.5354

0.7500 0.0167
0.0001 0.5395

BETA... -0.2600 0.1718 0.0769 0.2918 -0.0808
. 0.0001 0.0001 0.0054 0.0001 0.0034
EMMI’ -0. 1748 0.1776 0.3000 -0.2702 -0.0197
0.0001 0.0001 0.0001 0.0001 0.4708

EP... 0.0766 -0.0094 0.0476
0.0049 0.7337 0.0819

0.0541 -0.1735
0.0472 0.0001

RIRT...
LSIZEM 0.7413 -0.3106 -0.2769 0.1633 -0.0067
0.0001 0.0001 0.0001 0.0001 0.8046
UEi 0.0193 -0.09 17 -0.0993 -0.0834 0.0269
0.4792 0.0009 0.0003 0.0022 0.3233

 

 

a. Both Pearson (below the diagonal) and Spearman rank (above the diagonal) correlation coefficients are
presented p values are reported below the coefl'rcients. Number of observations ranges from 1,299 to
1,349.

b. 11 observations with BM... < 0 are excluded.

Variable definitions:

LSIZE... .. the logarithm ofthe market value ofthe firm at the beginning ofyear t-l,

RIRT... = relative industry ranking of overall disclosure inforrnativeness in year t-l,

Refer to Table 2.2 for the definition of other variables.

37.

(p=0.0121, two-sided)”. This is consistent with Barth and Sweeney (1995) and suggests
that more forthcoming disclosures can lead to lower costs of equity capital.

Table 2.3B reports Pearson (below the diagonal) and Spearman rank (above the
diagonal) correlation coeficients among the four measures of disclosure inforrnativeness.
The two sets of correlation coemcients are essentially the same and discussions are based
on Spearman rank correlation coeficients. Because RIRT... is a weighted average of the
other three, it is not surprising that it is highly correlated with RIRAm, RIRQiei, and
RIRIR... Consistent with Lang and Lundholm (1993), the three category relative industry
rankings are all significantly correlated with one another at the 0.0001 level. The highest
correlation is 0.6043 between RM and RIRQ... and the lowest is 0.4462 between
mo, and arm... This indicates that firms tend to coordinate their disclosures through
difi‘erent channels.

2.5 EMPIRICAL RESULTS

All tests presented in this section are one-sided if the hypotheses are directional,
and two-sided otherwise.

2.5.1 Testing or Hypothesis 2.1

Hypothesis one states that the abnormal returns associated with unexpected
accounting earnings of year t begin earlier for firms with more informative disclosures than -
for firms with less informative disclosures. Figure 2.1 graphs the intertemporal averages
of CMR(DH) and CMR(DL) fi'om the beginning of year M to April of year t+1.

Hypothesis one is clearly supported. The positive trend for the average of CMR(DH)

 

2’ Not reported in Table 2.3A.

38

Table 2.33
Correlation Coefficients between Informativeness Variables‘

 

 

RIRTi. RIRA-. RIRQ; RIRL.
RIRT; 0.8143 0.7684 0.7096
0.0001 0.0001 0.0001
NRA. 0.8143 0.6043 0.4881
.0001 .0001 0.0001
RIRQi. 0.7678 0.6041 0.4462
.0001 0.0001 0.0001
RIRIg 0.7104 0.4885 0.4457
.0001 0.0001 0.0001 ‘

 

a BothPearson(belowthediagonal)andSpearmanrank(abovethediagonal)correlationcoeficients are
presented. p values are reported below the coeficients. '

RIRT... = relative industry ranking (RlR) of overall disclosure inforrnativeness for firm i in year t-l,
REA... =- RIR of the inforrnativeness of discloatres via annual reports for firm i in year t-l,
RIRQ... - RIR of the inforrnativeness of disclosures via quarterly reports for firm i in year t-l,

RIRI... = RIR of the inforrnativeness of disclosures via investor relations for firm i-in year t-l.

39

33:...— E— e... .5 so. :20 ounces.
3 2:2...

7. core gas a: see 2.8: .

 

 

 

 

 

 

 

 

_ was?
Encgoisrsemzulo;
88347223223222:22:2 e a e e an. ._
“anelsaheh+h+.iiha +, A‘ii‘c
:3...
E6
:3
..no

«6

40

starts four months afier the beginning of year t-l while the positive trend for the average
of CMR(DL) does not begin until seven months after the beginning of year t- l. The
cumulative market-adjusted returns over the positive-trend period for DL, portfolio is
0.233, compared with that for the DH portfolio at 0.211. However, 33.2 percent of the
overall returns in the positive-trend period are realized in year t-1 for the DH portfolio,
compared with only 20.3 percent for the DL portfolio.

Table 2.4 presents results from regressing cumulative market-adjusted returns in
six-month moving windows, denoted as ACME... on UEi. for both DL and DH portfolios.
The first window starts at the beginning of year t-l and observations in difl'erent years are
pooled together. Hypothesis one is again supported. The association between ACMR...
and UB9. is significantly positive at the 0.05 level five months afier the beginning of year t-
l for DH firms. On the other hand, for DL firms, the association between ACMR... and
UP... is not statistically positive at the 0.05 level until nine months after the beginning of
year t- 1.

Table 2.5A reports test results fi'om both the t test and the Wilcoxon signed rank
test based on the intertemporal distribution of the differences in the beginning of positive
trends for DH and DL hedge portfolios. Under hypothesis one, the positive trend should
start earlier for the DH than for the DL hedge portfolios. On average, the positive trend
starts 4.22 months after the beginning of year t-l for the DH hedge portfolio, compared to
V 6.56 months for the DL portfolio. The difference is 2.34 months and statistically

significant at the 0.05 level under both tests.

41

Table 2.4
Results from Pooled Regressions for DL and DH Portfolios

 

 

 

n DL DH
8,, Std Err. t 5,, Std Err. t
0 -0.2015 0.0698 -2.8850 «0.0926 0.0890 -1.0410
1 -0. 1925 0.0709 -2.7170 -0. 1213 0.0858 -1.4140
2 -0. 1261 0.0662 -1.9060 0. 1228 0.0805 -1.5250
3 -0. 1628 0.0706 -2.3040 0.0495 0.0827 0.5980
4 00292 0.0677 «0.43 10 0.1011 0.0850 1.1900
5 -0.0184 0.0657 -0.2800 0.1991 0.0820 2.4280'
6 -0.0756 0.0664 -1. 1380 0.1424 0.0788 1.8070'
7 -0.0017 0.0668 ' -0.0260 0.1347 - 0.0753 1.7900‘
8 0.0424 0.0715 0.5940 0.1931 0.0807 2.3950'
9 0.1453 0.0781 1.8600' 0.2539 0.0866 2.9310‘
10 0.1733 0.0750 2.3100' 0.4070 0.0849 4.7920‘
11 0.1210 0.0751 1.6110 0.3741 0.0829 4.5140'
12 0.3021 0.0694 4.3500‘ 0.4377 0.0797 5.4930“
13 0.3614 0.0741 4.8740‘ 0.5436 0.0828 6.5630‘
14 0.3548 0.0708 5.0130‘ 0.4605 0.0859 5.3590‘
15 . 0.2936 0.0750 3.9160‘ 0.3664 0.0862 . 4.2500'
16 0.3095 0.0741 4.1770“ 0.2923 0.0863 3.3870'
17 0.3938 0.0673 5.8500‘ 0.3534 0.0811 4.3590‘
18 0.2715 0.0684 3.9690‘ 0.2440 0.0774 3.1540'
19 0.2276 0.0676 3.3680‘I 0.1416 0.0786 1.8010‘
20 0.2298 0.0668 3.4390‘ 0.2585 0.0868 2.9800‘
21 0.1839 0.0741 ' 2.4820‘ 0.2791 0.0946 2.9500‘
22 0.2163 0.0712 3.0380‘ 0.1733 0.0931 1.8620‘
23 0.1980 0.0733 2.7010‘ 0.0798 0.0907 0.8800
24 0.2802 0.0752 3.7260‘ 0.1219 0.0899 1.3560

 

a significantly positive with a p-value smaller than 0.05, one-sided

42_

Table 2.5A
Tests Based on the Beginnings of Positive Trends

\

 

 

Year m(DL) mdDH) Am. lAmd rt‘ 2.” r32.
84 13 14 -1 l 2.5 o o
85 5 4 1 1 2.5 1 2.5
86 7 o 7 7 7.5 1 7.5
87 9 10 -1 l 2.5 o o
as 7 o 7 7 7.5 1 7.5
89 6 4 2 2 5 1 5
90 l l 0 0 - o o
91 l o l l 2.5 1 2.5
92 10 5 5 5 6 1 6

Average 6.56 4.22 2.34
t / Sum 2.19‘ 31"

 

a.r.istherankof|Am.|or|An.l.
b.z.equalsoneifAm.orAn.ispositive,andzerootherwise.

c. significantly positive with a p-value of 0.05 or smaller, one-sided '

43

Table 2.5B presents results of tests of hypothesis one based on the intertemporal

distribution of the differences in when [3... in the following regression: .
CMRtin=ata+BuUEti+8m

is first statistically positive at the 0.05 level. Under hypothesis one, B... should be
statistically positive earlier for the DH than for the DL portfolios. On average, cumulative
market-adjusted returns over a six-month window are positively correlated with
unexpected earnings of year t 4.34 months earlier for DH firms than for DL firms. The
lead is statistically significant at the 0.05 level using either the t test or the Wilcoxon
signed rank test.

Taken together, results reported in tables 5A and 5B strongly support hypothesis
one.
2.5.2 Sensitivity Analysis for the Testing of Hypothesis 2.1

Some potential confounding factors, such as exchange and industry efi‘ects, fiscal
year end differences, and fiscal year end changes, have already been controlled for through
the research design. In general, such control mechanisms tend to bias against the

maintained hypotheses by introducing noise to the relative industry ranking measures.26

 

2‘ For instance, Southwest Airline was ranked the fourth among eight airlines evaluated in
1990 for the lack of market and segment data and limited access to senior management.
Such a ranking would put it in DM and exclude it fi'om further analysis. However, two
airlines with higher rankings are eliminated in the sample selection process for not being
listed on the NYSE (British Airways PLC) and non-December fiscal year end (Delta). As
a result, Southwest’s ranking is elevated to the second highest, placing it as a DH firm in
the 1991 sample. Classified as a bad news firm in 1991 (UE<0 ), its 13-month cumulative
market-adjusted return starting from the beginning of 1990 is substantially positive, at 60
percent, distorting the cumulative market-adjusted returns for the DH hedge portfolio in
the 1991 sample year. While such observations can be excluded as anomalies, they are
kept in the sample to avoid making subjective judgments.

44

Table 2.51!
Tests Based on the Beginnings of Significantly Positive Correlation

 

 

Year m(DL) m(DH) An. IAnd r.‘ 24" the
84 9 3 6 6 4.5 1 4.5
85 6 o 6 6 4.5 1 4.5
86 14 9 5 5 3 1 3
87 12° 4 8 8 7.5 1 7.5
88 10 l 9 9 9 1 9
89 9 7 2 2 l 1 1
9o 14 7 7 7 6 1 6
91 2 10 -8 8 7.5 o o
92 10 6 4 4 2 1 2

Average 9.56 5.22 4.34
t / Sum 2.56‘ 37.5d

 

a r. is the rank of lAmJ or |An.|.
b. 2. equals one if Armor An. is positive, and zero otherwise.

c. month with the smallest p-value for n - 0, 1, 24 (p =- 0.066, one-sided) because no Bmis significantly
positive at the 0.05 level for the DL portfolio in sample year 1987.

d significantly positive with a p-value smaller than 0.05, one-sided

45

Two variables that are not controlled for by the research design are firm size and analyst
following As shown in Table 2.3, RIRTi. is positively correlated with both LSIZEM
(p=0.0028, two-sided) and ANA.“ (p=0.0001, two-sided). This leads to the concern that
the results reported above may in fact be attributable to size and analyst following efl‘ects,
even though the magnitudes of the correlation coeficients are not substantial, only 0.0813
between RIRTM and LSIZEg... and 0.1296 between RIRT.... and ANA“...

To examine the efi‘ect of size on sample firms’ information environments, firms in
the same industries are placed into one of three portfolios, SS (small), SM(medium), and
SL (large), based on the ascending order of their market values at the beginning of year t.
The numbers of firms in SS and SL are always kept the same. For firms in the middle-
sized portfolio SM, RIRT.... is not significantly correlated 'with either LSIZEM (p =
0.5237, two-sided) or ANAit-1(P = 0.1243, two-sided) at conventional levels. Because the
numbers of observations for DLSM and DHSM firms are small for individual sample
years, observations in difi‘erent years are pooled together, as in Freeman (1987).
Arbitrage portfolios are them formed for the pooled sample by taking a long position in
firms with good news and a short position for firms with bad news. The cumulative
market-adjusted returns for both portfolios are graphed in Figure 2.2. Consistent with
hypothesis one, the positive trend starts at least three months earlier for CMR(DHSM)
‘ than for CMR(DLSM). This suggests that the previously reported results for hypothesis

one is unlikely to be caused by the lack of control for size or analyst following.

46

3:828 oueo: 2mm: 2... :3: as :20 .
a." cue»:—

3 core 9.538 us use case:

- _AY

 

A€&mizuxm§fivta-I.Awqmqncmazurlart

 

 

 

bu 6N nN VN an «N _N am a— w— 6— o— m— v— n— N— . a a

r b p . L P N p F _ h P n P F p
a d . a q u q q q q a q iiq hi

..
11-
qr
+-

noaY

Jr-
urb-
‘r
«l-
41-
«1-

.. wage

...wc mag“.

.. n—gv

tr N.°

e no

 

 

mA.

47

2.5.3 Testing of Hypothesis 2.2

Hypothesis two investigates which categories of disclosures _ are efi'ective in
communicating firm-specific information to investors. Test results are presented in Table
2.6. Year dummies and an intercept are included in all regressions but are not reported
because they do not relate to the hypothesis tested. In the table, Model 1 is a simple
regression of CMR... on UK. with an adjusted R2 of 0.0478. In Models 2 to 4, only one
disclosure proxy at a time is included. The results indicate that both annual reperts
(t=2.29) and investor relations (t=3.95) are efi‘ective in communicating future earnings
information to investors while quarterly reports (t=0.90) are not”. However, in Model 8,
while all the three disclosure proxies are included in the regression at the same time, only
B4 is significantly positive at the conventional levels (r- 3.40), indicating that in year t-l
only investor relations assist investors in the formation of future earnings expectation, and
that the positive efi‘ects of the annual reports indicated by results from Model 2 are caused
by the correlation between RIRAt... and MRI... Results based on Model 9, which
controls for the efl‘ects of size and analyst following, suggest that the effect of investor
relations is not a disguised size or analyst following etrect because Bais still significantly
positive at conventional levels (t= 3.24).

Results presented in Table 2.6 also suggests that the efi‘ect of size on firms’
information environments reported in previous studies (see, for example, Freeman 1987,

Collins and Kothari 1989, Kothari and Sloan 1992), which do not control for the effect of

 

27 It should be noted that same result could obtain if RIRQi. fails to capture cross-sectional
variations of the inforrnativeness of quarterly reports. The same caveat applies to other
tests of hypothesis two.

48

8:8... 5 8:82 2e 835% co 82.5 .e

 

 

:98 Ga: and 5.3 see :03

82... e8... :3... 22. 23. o: .o 3...? e coo:
84.9 an: 5.3 see

83... - - 8:. «no. 83 «8.: a .862
, 32: and $4.3

23¢ 83. a: _ .e - - - 3.3. n .382
. an: 898

:8... :2. . - - - Boo e .862
and 84.3

as... - 3. .e . - - 3...? e 3o:
33 :2:

:3... - - :3 - - e86. o .682
see so:

2.3... - . - a: .o - 23 n .822
and 3...:

83° - - . - :2. 23 N .68:
cos

”5:. - - - - - an... _ .682

a... 8.3.3 ...<z<. .m: ...msw... .m: :22. .m: 395.. .m: ..52. .m: .m:

 

.582 .885: case... .2323: be 385 o... as any
on 2.3.

49

disclosures, may in part be caused by the fact that larger firms tend to have better investor
relations. In Model 5, where only LSIZEitd‘I U5“ is included (in addition to UEa), y. is
significantly positive (t = 2.58), which is consistent with prior evidence that investors can
form more precise earnings expectation. for larger than for smaller firms in year t-l. In
Model 7, where ANA...‘UE.. is also included in the regression, the size efi‘ect is still
significant with a 1 value of 1.88. However, in model 9, where disclosure variables are
included in the regressions, the size effect is no longer significant at conventional levels (t
= 1.26).
2.5.4 Sensitivity Analysis for the Testing of Hypothesis 2.2

Bernard (1987) showed that, in cross-sectional regressions of return on
unexpected earnings, the cross-sectional dependence of returns may cause the levels ,of
statistical significance for the regression coeficients to be overstated. To alleviate this
potential problem as well as the concern that the results in Table 2.6 may have been
affected by a small number of outliers, annual regressions based on Model 9 are conducted
for each sample year and hypothesis two is then tested using the intertemporal distribution
of the coemcients in the nine sample years. The results are essentially the same as
reported in Table 2.6, with investor relations as the only statistically significant efl‘ect (t =
2.062). '

Collins and Kothari (1989) provide evidence that the ERC varies systematically
with the book-to-market ratio and equity beta. To make certain that the results reported
in Table 2.6 are not caused by the lack of control for these two variables, a regression that

includes the three disclosure proxies, the logarithm of size, analyst following, book-to-

50

market ratio, and equity beta is re-estimated. The results remain essentially the same as
reported in Table 2.6.

Finally, while the results reported. in Table 2.6 are based on regressions where the
CRSP equally weighted market index is used in the calculation of the CRSP cumulative
market adjusted returns, they remain essentially the same if the value weighted index is

used instead.

Chapter 3
PREEMPTION, SIGNAL INFORMATIVENSSS, AND THE INCREMENTAL
INFORMATION CONTENT OF THE ANNUAL AND 10-K REPORTS

This chapter examines the efi‘ects of preemption and signal inforrnativeness on the
incremental information content of annual and lO-K reports. It is organized as follows.
Section 3.1 provides background and motivation. Section 3.2 uses a model adapted from
Holthausen and Verrecchia (1988) to develop hypotheses. Section 3.3 outlines methods
for hypothesis testing and variable measurements. Section 3.4 describes the sample and
variables. Empirical results are presented in Section 3.5.
3.1 BACKGROUND AND MOTIVATION

Under the current financial reporting system in the United States, publicly-traded
firms must provide shareholders with the ARS and file the lO-K with the SEC. These
reports. represent the formal public release of a firrn’s detailed financial statements.
Moreover, they contain disclosures mandated by the Financial Accounting . Standards
Board (FASB) and/or the SEC, and other voluntary disclosures. Such disclosures include,
but are not limited to, management discussions of past and expected firm performances,
segmental data, accounting policy choices and changes, research and development,
and the auditor’s opinion. Therefore, even though ARS and lO-K are not released
to the public until several weeks after the announcements of summary earnings

information, they may still contain incremental information that is usefirl for market

51

52

participants to revise their expectations for future earnings. Such a conjecture is
consistent with results reported in previous survey studies. Lees (1981), for instance,
finds that analysts cite ARS and lO-K as important sources of firm information. Knutson
( 1992) reports that analysts perceive ARS as the most important reporting document.
Despite the perceptions of analysts, a group of primary users of financial reports,
prior studies on the incremental informationcontent of annual and lO-K reports have been
inconsistent. Foster and Vickrey (1978) report evidence that lO-Ks released at least 10
days alter the filing of ARS contain incremental information. However, most later studies
have been unable to reach the same conclusion. Foster et al. (1983) find no abnormal
price volatility in the week of lO-K release after the release of ARS, which represents a
reversal of the conclusion in Foster and Vickrey (1978). Foster, Jenkins, and Vickrey
(1986) do not find unusual price volatility in the week of ARS release, which may or may
not have been preceded by the release of lO-K. Cready and Mynatt (1991) reports that
“No evidence of a price response and little evidence of a volume of shares response at
annual report dates is found.” Stice (1991) finds no abnormal returns behavior on 10-K
and lO-Q dates that precede earnings announcement dates by at least four days, but
detects abnormal returns behavior on the later earnings announcement dates. Easton and
Zmijewski (1993) examine the incremental information content of ARS, 10-K and lO-Q
with a very large sample that covers SEC filings from 1966 to 1985. They report that,
overall, there exists little evidence of incremental information around ARS and lO-K
disclosure dates. More recently, Lang and Lundholm (1994) find that analyst forecasts are

not more accurate for firms with more forthcoming annual and lO-K reports afier

53

controlling for the inforrnativeness of quarterly reports and investor relations.” On the
other hand, Han and Huang (1995) find that security returns on the earlier of the release
dates of either the ARS or 10-K are positively associated with an indicator of future
earnings growth, but the significant association is attributed mainly to small firms. Taken
together, the collective evidence from prior research weighs against the existence of
incremental information associated with the release of the ARS and lO-K.”

Several research design issues may contribute to previous studies’ inability to
document abnormal returns behavior on the ARS and lO-K release dates. First, the lack
of control for the cross-sectional difi'erences in previously released information reduces the
power of statistical analysis. Managers communicate with investors through media other
than the ARS and lO-K, such as conversations with financial analysts, press releases, and
quarterly reports. The existence of more timely alternative sources of earnings-relevant
information means that returns can incorporate certain information about future earnings
well before the release of annual and lO-K reports. Han, Jennings, and Noel (1992), for
instance, report that there exists a significant association between security returns and
revisions in the probability of bankruptcy due to non-earnings data, but market
participants have largely revised their estimate of the probability of bankruptcy prior to the

earlier of the SEC receipt dates of the ARS and lO-K.

 

2' It should be noted that the event period in Lang and Lundholm (1994) precedes the
release of the ARS and lO-K.

2’ There exists an extensive literature providing evidence that the information in
disclosures contained in the ARS and lO-K are useful for the valuation of firms. See
footnote 1 in Barth and Sweeney (1995) for examples. However, since most papers in
this literature are association or level studies, it remains unclear whether the market
derives the information from the disclosures or fi'om other sources that may not even
originate fi'om the firms.

54

Second, most prior studies treat annual and lO-K reports fiom difi‘erent firms as if
they contain the same amount of information, which introduces noise into statistical tests.
The ARS and lO-K across firms differ in the amount and quality of information they
contain. It is conceivable that the ARS and lO-K fiom a subset of firms provide
incremental information to the market because the disclosures in these reports are more
informative than disclosures in similar reports fi'om other firms. Because firms tend to
coordinate disclosure policies, i.e., firms with more informative ARS and lO-K tend to
have more informative interim disclosures and better investor relations, the
inforrnativeness of the ARS and 10-1( and that of prior disclosures must be controlled
simultaneously to be efi‘ective.

The third issue relates to the specification of the event dates on which the
information in the ARS and lO-K first becomes known to the market. In general, they are
assumed to be the earlier of the SEC receipt dates of the ARS and lO-K. However, some
firms disclose information in the ARS and lO-K prior to the SEC receipt dates through
news releases. Wilson (1987) reports that in the 1981-82 period, 63 percent of firms in his
sample made such news releases. Of those releases, 79 percent contained earnings,
revenues and additional information such as segment data, and 13 percent contained
preliminary earnings statements and balance sheets. Moreover, some firms send their ARS
and lO-K to investors and analysts before sending them to the SEC. Han, Jennings, and
Noel (1992) report that only 63 percent of firms that responded to their survey indicated
that they do not send their ARS or lO-K to either shareholders or analysts/brokers more

than three days before mailing it to the SEC. On the other hand, investors’ response to

55

information in the ARS and lO-K may be delayed until days after their receipt by the SEC.
In general, the ARS and lO-K are not made available to the public by the SEC until
several days afier the receipt dates.3° ‘ Furthermore, because the information in the
financial statements and disclosures is contextual and firm-specific, it may take market
participants a certain period of time to first interpret and assimilate and then respond to
the information.

In this paper, I attempt to overcome these limitations by explicitly controlling .
cross-sectional differences in prior information and the inforrnativeness of the ARS and
lO-K. I also empirically investigate the conjecture that the ARS and lO-K, or the
information in them, may have been released to the market prior to the SEC receipt dates,
and the possibility of a delayed market response to the information in the ARS and lO-K
3.2 HYPOTHESIS DEVELOPMENT

The above discussion on the efi‘ects of cross-sectional difl‘erences in prior
information and the inforrnativeness of the ARS and lO-K on information content tests can
be formalized with a model adapted fiom Holthausen and Verrecchia (1988). Figure 3.1
illustrates the sequence of events modeled below, including fiscal year ends (FYE) in years
M, t and t+1 as well as earnings announcement dates (BAD) in years t and t+1. Assume
that the earnings generation process is a random walk such that:

raps. = raps, + em, (3.1)

 

3° Refer to Easton and Zmijewski (1993) for more information on this issue.

56_

 

 

 

mu ma. BAD. D... rm... BAD...
I i L I i
I l I I l
B...
Figure 3.1

Sequence of Events Modeled

57

where e... is normally distributed with a mean of 0 and a variance of v..31 Let Bt+s be the
aggregate of information released before t + s and takes the form:

Bm = e"I + b (3.2)

where b... is normally distributed, uncorrelated with e... , and has a mean of 0 and a
variance of v...32 In reality, B... includes, but is not limited to, earnings realizations,
quarterly reports, news releases, and analyst earnings forecasts prior to t + 8. Let D... be
new information in the ARS or lO-K released at t + s. It provides information about e...

such that:

D = e1+1 + dtes’ (3'3)

where d... is a normally distributed random variable that is uncorrelated with e... and has a
mean of 0 and a variance of v... In general, 11... and b... are correlated with a covariance of
c. Intuitively, this means some information in D... has been preceded by previously
released information. For instance, some of the information contained in the ARS and 10-
K may has been preceded by quarterly reports and/or management discussions with
analysts. .

It can be shown that, before the release of D..., the expectation of e... for investors

who can observe B... is:

E(etel lat") = fl?- ° (3'4)

v.+vb

 

3‘ Because EPS. is known before the release of the ARS and lO-K, this assumption is
equivalent to assuming that BPS... has a normal distribution with an unconditional mean of
EPS. and a variance of v..

32 Because t + s represents the release dates of the earlier of the ARS and lO-K of year t,
the value of s should normally be between zero and one.

58

After the release of D..., the expectation of e... for investors who can observe both B... and

D... is be:

 

v,(v. -C)B... +V.(Vr. '0”)... ' ‘ i (3.5)

E(e
(v. +vb)(v. +vd)-(v. +0)z

lB D...)=

NI to! ’

The revision of the expectation of e... due to the release Of D... is then:

v.(v, -c)Bm +v.(v.- c)D v, *8.“
(v. +vb)(v. +v‘)—(v. +c)’ v. +vb '

‘4'. _

F-(e... | B D... ) - 5(6... | B... ) = (3-6)

tn’

For simplicity, ignore the time value of money between t+s and t, and normalize the share
price before the release of D... to one. Stock return associated with the release of D... is

therefore:

REF

‘0‘

__1_¢ Ye(vd-C)Btn+ve(vb-C)Dt+s _ve‘Bt+s (37)
r (v.+v,)(v.+v‘)-(v.+c)’ v.+vb ’ °

where r is the cost of equity capital.

In the above expression, the return associated with the release of D... depends on
its precision, l/vd, the extent to which it has been preceded by B... due to the correlation
between b... and d..., the precision of B... measured by l/v.,, and the variance of the
earnings process due to the existence of noise. It can be simplified under the assumption
that the information contained in the ARS and lO-K subsumes all the information about
e... that has been previously released.33 Under the assumption, equation (3.7) can be

rewritten as the following:

 

3’ Such an assumption is reasonable since the ARS and lO-K are the most comprehensive
financial reporting documents about publicly listed firms. Mathematically, this means that
6: V4 and V5 > Va.

59

REF

'0'

 

I 1"D ’B
=_#|:ve tn _ V. be], (38)
r v. +vd v. +vb

3.2.1 Abnormal Volatility

It can be shown that the variance of RET..., conditional on having first observed

8..., is:

2 2

Var(RETm)=%'[ "' — V. ] (3.9)

v.+vd V.+Vb

 

It is a decreasing function of l/v., the inforrnativeness .of prior information, and an
increasing function of l/vd, the inforrnativeness of the ARS and lO-K. This indicates that
abnormal returns variance is most likely to be observed for firms with less informative
prior disclosures and more informative ARS and lO-K, leading to the first hypothesis in
its alternative form:
F or firms with more informative ARS and 10-K and less informative prior
disclosures, abnormal stock returns exhibit abnormal volatility on the earlier
of the dates when the ARS and IO-K are released
3.2.2 Abnormal Returns

Equation (8) can be rewritten as the following:

i *
RET =l.[ VI _ v. J‘et+' +1.[V. db+I _ v. bCOI], (3.10)
. v.+vd v.+vb r v.+v‘I v.+vh

 

 

where e... is not correlated with either d... or b.... Equation (10) shows that, on average,
if the ARS and lO-K for year t indeed' provide information about e..., then the abnormal
returns in the event period should on average be positive for firms with positive
unexpected earnings in year t+l, and negative for firms with negative unexpected earnings

in year t+1. Moreover, such effects should be most pronounced for firms with more

60_

informative ARS and 10-K and less informative prior disclosures. This leads to the second
hypothesis:
F or firms with more informative ARS and 10-K and less informative prior
disclosures, the abnormal stock returns on the earlier of the ARS and IO-K
release dates are, on average, positive for firms with positive future
unexpected earnings and negative for firms with negative unexpected fitture
earnings.
3.2.3 The ERC
Equation (3.10) also indicates that, in a regression of RET... on e..., the ERC is an
increasing function of UV. and a decreasing firnction of IN... This leads to the third
hypothesis: '
C eteris pm'ibus, the ERC from regressing abnormal returns on the earlier of
the ARS and IO-K release dates onfilture unexpected earnings is an
increasing fimction of the inforrnativeness of the annual report and a
decreasing junction of the inforrnativeness of prior disclosures.
The above hypothesis will also be tested in the period immediately before the event period
to investigate the conjecture that market participants may already have had access to the
ARS and lO-K or the information in these reports before their receipt and release by the
SEC. ° It will also be tested in the period immediately afier the events period to detect the
possibility of a delayed market response to the information in the ARS and lO-K.
3.3 STATISTICAL METHODS AND VARIABLE MEASUREMENTS
3.3.1 Proxies for the Informativeness of Disclosures
Proxies for the inforrnativeness of disclosures through the ARS, 10-K, and prior
disclosures are based on analysts’ evaluation scores for the sample firms’ annual and 10-

K reports, quarterly reports, and investor relations as published in the AIMR Reports.

According to the AIMR Corporate Information Committee, an industry-specific

61

subcommittee composed of leading analysts following the industry evaluates, on an annual
basis, the inforrnativeness of selected firms’ disclosures along three dimensions: annual
published information, quarterly and other published information, and investor relations
and other aspects. Characteristics and issues unique to the industry are taken into
consideration in the evaluation process. Scores along these three dimensions are then
weighted to obtain an overall score about the inforrnativeness of the finn’s disclosure
practices. The weights are in general 40-50 percent for the annual published information,
30-40 percent for the quarterly and other published information, and 20-30 percent for
investor relations and other aspects. While the majority of the subcommittees report both
overall and category scores, some subcommittees publish only i the overall scores. As a
result, about one-third of firm years “in the AIMR Reports have no category scores
reported and are therefore not included in the sample.

Because firms in difi‘erent industries are evaluated by difi‘erent subcommittees and
members of the same subcommittees are not necessarily the same in difi‘erent years, the
raw scores reported by the subcommittees must be. standardized to provide meaningful
proxies for disclosure inforrnativeness. This is achieved by defining relative industry
rankings for the ARS and lO-K (RIRAa) as the following:

mfiRANKA,
N,-1

-l

9

 

where N,. is the number of firms for industry j in year t, and RANKA. is the rank, in
ascending order, of firm i in year t within industry j based on analysts’ evaluation score for
the ARS and lO-K. Analysts’ evaluation scores for quarterly reports and investor

relations are combined to obtain a weighted score for disclosures released prior to the

62

release of the earlier of the ARS and lO-K, and relative industry rankings for prior
disclosures (RIRPR) are then obtained similarly as RIRA... RIRQg. and RIRI. are also
similarly defined as RIRA. as proxies for the inforrnativeness of quarterly reports and
investor relations.
3.3.2 Dependent Variables

Dependent variables for statistical analysis are defined following Patell (1976). Let
r = 0 represent the three-day event period centered around the date on which either the
ARS or lO-K, whichever is earlier, is received by the SEC. The test interval, composed of
three-day periods and indexed by r, ranges from r = -5 to t = +5. The following market
model is first estimated over an estimation interval consisting of 60 three-day periods prior
and 60 three-day periods following the test interval:

Rm = a, +fluRm + e“, , where

R... = the return for firm i in sample year t for period I, obtained by summing up daily
returns from the CRSP daily returns tapes,

R... = the equally-weighted market return index in sample year t during period 1,
obtained by summing up equally weighted daily market return index from the
CRSP daily returns tapes
e... = a random, normally distributed error term.
Let T. be the number of observations in the regression, and 3..2 be the variance of
a... estimated from the above regression. The abnormal returns for period 1: = -5, ..., +5 in

the test interval are calculated from:

“in = Ritr _(ait +flitRnr)’

63

As in Patell (1976), the following variable is distributed as a Student t statistic with T..- 2
degrees of freedom:
u

where C... is an adjusting factor for making predictions outside the estimation interval.

Vite =

 

The average of V... for a given value of 1 over all firm-years is denoted as V.. The

distribution of V... means the following variable should have a normal distribution:

 

XXV.
Zv'_ I ‘Tntz.
izg‘rit -4

Also, the following variable:

[1.2 T. -4

Um. = rtr 2 . It ,
C its 8 i' Tit _ 2
has an expectation of one and an variance of 2(T..-3 )/(T..-6). This means that the

following variable has an approximately unit Normal distribution:

222:1”... -1>

2‘“ 2(1~ 3) ’
F? 1:-..

where U. is the average of U... for a given value of I over all firm-years.

 

3.3.3 Hypothesis 3.1
To test hypotheses one, I first divided the full sample into two portfolios according
to the inforrnativeness of the ARS and lO-K, with AH being the more informative and AL

the less informative. This is achieved by putting observations with RIRA.. above or equal

64

to their respective industry-year medians in AH, and the rest in AL. A similar procedure is
performed to divide the full sample into another two portfolios based on the
inforrnativeness of prior disclosures, with PH being the more informative and PL, the less
informative. Consequently, portfolio PLAH contains firms with less informative prior
disclosures and more informative ARS and lO-K.

Under H1, the abnormal returns of firms in the portfolio PLAH should exhibit
abnormal volatility in the event period. This is tested by whether U0 is significantly larger
than one. .

3.3.4 Hypothesis 3.2

Define us... as scaled unexpected earnings offinn i in year t + 1. Assuming that
the earnings-generating process follows a random walla UE.... can be obtained by
subtracting primary earnings per share before extraordinary items (EPS) of year t fiom
that of year t + l and then scaling the difi‘erence by share price at the beginning of year t +
1. Both EPS and share prices are adjusted for stock splits and stock dividends, obtained
from the Compustat tapes.

Under H2, for observations in the portfolio PLAH, the abnormal stock returns in
the event period are, on average, positive for firms with UE.... larger than zero and
negative for firms with UE.... smaller than zero. This is tested by whether V0 is
significantly positive for the subset of firms in portfolio PLAH with positive UE.... (UEP)

and signifiCantly negative for the subset of firms in portfolio PLAH with negative UE....

(UEN).

65

3.3.5 Hypothesis 3.3

Hypothesis three is tested using the following regression in which observations are
pooled both cross-sectionally and intertemporally at the same time during the event
period, 1: = 0:

u“ =a+B.UEn.. +BleRA. "U13.M +B,RIRP. ‘UEN +1 ‘LSIZE. ‘UEM +a‘.
In this regression, LSIZE.. is the logarithm of SIZE... the market value for firm i at the end
of year t, and 6.. is a random, normally distributed error term. Other variables are as
previously defined. Under hypothesis three, [3. is. positive and [33 is negative. LSIZE.. is
included in the regression to control the efi'ects of other variables, that may afi‘ect firms’ 6
information environment, but no prediction is made for the regression coeficient.

To investigate the conjecture that market participants may already have had access
to the ARS and lO-K or the information in these reports before their receipt and release by
the SEC, the above regression will be estimated with abnormal returns accumulated over 1:
= -S, -l. The possibility of a delayed market response to the information in the ARS
and 10-K will be examined by repeating the above regression with abnormal returns
accumulated over 1 = +1, ...,+5. . i
3.4 SAMPLE AND VARIABLE DESCRIPTIONS
3.4.1 Sample Selection Criteria and Procedure

Table 3.1A describes the sample selection process. The 1980-1984 AIMR
Reports contain 2,190 firm-years with CUSIP numbers identifiable fi'om the CRSP tapes.

The following eight selection criteria reduce this initial sample to its final total of 933. (1)

The industry subcommittees must report category scores. (2) The SEC receipt dates

66.

Table 3.1A
Sample Selection Criteria and Procedure

 

 

Number of
firm-year
Firm-years covered in AIMR Reports,
1983-1991, with CUSIPs from CRSP tapes 2,190
Less firm-years:
without category scores ( 772)
without either the SEC ARS or lO-K receipt
date ( 323)
withatleastonedailyreturninthe393 .
trading days centered around the event date ( 62)
. . 8 ‘
without either annual earning
announcementdateorthenextfirstquarter ( 9)
esmings announcement date
withanearnings announcementthatisless ( 22)

than eight days away from the event date

withatleastonedailyreturninthe393
trading days centered around the annual ( 3)
earnings announcement date missing

withoutCompustatdatatoobtainUE... -( 65)_
withlessthantwofirmsinannual ( I)
indusu'ygroups

Firm-years included in the sample 933

 

67

for the ARS and lO-K are both available. (3) All daily returns in a 393-day period
centered around the event date are available from the CRSP daily returns tapes. (4)
Annual earnings announcement dates are available from either the Compustat tapes or the
Wall Street Journal Index. (5) No earnings announcement occurs less than eight days
away from the event date. (6) All daily returns in a 393-day period centered around the
fourth-quarter earnings announcement date are available from the CRSP daily returns
tapes. (7) Data are available from the Compustat tapes for the calculation of UE..... (8)
There are at least two observations in the industry-year. Criterion (l) is imposed because
proxies are needed for both the inforrnativeness of prior disclosures and the ARS and 10-
K. Criteria (3) and (6) are imposed to exclude firms that are either not listed until shortly
before or delisted right after the sample year. Criterion (i5) is to eliminate potential
confounding efi‘ects attributable to earnings announcements. Criterion (8) makes the
calculation of relative industry rankings possible. All the other criteria are selected for
data availability.

As shown in Table 3. 18, the number of firms in each year ranges fiom 157 in 1980
to 204 in 1983, while the number of industries in each year ranges from 19 in 1980 and
1981 to 24 in 1984. The number of observations in each industry-year group is two at the
minimum and 19 at the maximum, indicating that, in any given year, the sample is not
dominated by a small. number of industries.

3.4.2 Summary Statistics
Table 3.2 presents summary statistics for the variables. Descriptive statistics for

RIRA.., RIRQ.., RIRI. and RIRP.. are not included because statistics for relative rankings

68

Table 3.1B
Distribution of Firms and Industries

 

Numberoffirms Numberofindustries .

 

Year
in the sample in the sample
1980 157 19
1981 187 19
1982 195 21
1983 204 21
1984 i9_o_ 24
Totals 933 104

 

69

Table 3.2
Sample Summary Statistics, 1980-1984

 

 

 

Pereentiles

Variable Mean Std.Dev. 1% 25% 50% 75% 99%
Reporting lag from
FYE to EAD' 37.1 11.9 17 28 36 45 68
Reporting lag from
FYEtoSECARS 81.5 13.1 49 74 81 89 117
receipt dateb
Reporting lag from
FYE to SEC lO-K 85.5 8.3 57 84 88 90 93
receipt date°
DEVE. 0.632 0.482 0 0 1 1 1
DEXE. 0.975 0.155 0 1 1 l 1
DFYE. 0.741 0.439 0 0 1 1 1
SIZE... 2,160 3,709 58 433 979 2,151 17,999
UE... -0.005 0.270 -0.535 -0.027 0.005 0.018 0.354

 

a. =thelagfromafirm’sfisealyearendaiYE)toitsannualeamingsannouneementdatemAD),
obtainedfromtheSECfilingdatedatabaseattheUniversityofChieago,samedatasoureefor
bandc.

DEVE. =adummyvariablethatequalsoneiftheSECreeeivestheARSoffirmiforfiseslyeartearlier
thanthe lO-K,andOotherWise,obtainedfi'omtheSECfi1ingdatedatabaseattheUniversity
ofChieago.

DEXE. -adummyvariablethatequalsoneiffirrniislistedintheNYSEandAMEXattheendoffiseal
yeartandOOthenvise,obtainedfromtheCRSPtapes.

DFYB. -adummyvariab1ethatequalsoneiffirmihasaDeeemberFYEinyeartandOotherwise,
obtainedfromtheCompustattapes..

srzra. =themarketvalueoffirmiattheendofyeart, inmfllionsofdoflaraobtainedfi'omthe
Compustat tapes.

UE... =unexpectedearningsforfirmiinyeart+1,0btainedbysubtractingprimaryearningspershare
beforeextraordinaryitemsofywtfromtharofyesrfilandthensalingthedifierenceby
shareprieeatthebeginningofyeart+ 1. BothEPSandsharepricesareadjustedforstock
splits and stock dividends and obtained from the Compustat tapes.

70

are not very meaningiiil. Following Easton and Zmijewski (1993), reporting lags are
defined as the number of calendar days from the firms’ fiscal year end. The mean (median)
annual earnings announcement lag is 37.1 (3 6) days. The mean (median) reporting lag is
81.5 (81) days for the ARS, and 85.5 (88) days for the lO-K.“ About 63.2 percent of
ARS filings occur before lO-K filings. Only 2.5 percent of firm-years are listed on the
NASDAQ stock exchange, making it not feasible to study this sub-sample separately.
Slightly less than three quarters of firm-years have a December fiscal year end. The -
distribution of SIZE... is skewed to the right, as indicated by a mean ($2,160 million) that is
much larger than the median ($979 million), and a standard deviation ($3,709 million) that
is greater than the mean. As a result, it logarithm, LSIZE.., is used in correlation analysis.
Table 3.3 reports both Pearson (below the diagonal). and Spearman rank (above
the diagonal) correlation coeficients among the inforrnativeness proxies as well as
LSIZE... The two sets of correlation coefiicients are essentially the same and the
discussions are based on Spearman rank correlation coeficients. Consistent with results
reported in previous studies, all the inforrnativeness variables are significantly correlated
with LSIZE... This indicates that larger firms tend to have more informative disclosures
and justifies the inclusion of LSIZE.. in regression analysis as a control variable.
Consistent with Lang and Lundholm (1993), firms tend to coordinate their disclosures

through difi‘erent channels. The three category relative industry rankings are all

 

3‘ These statistics are comparable to, but slightly smaller than, those reported in Easton
and Zmijewski (1993).

Pearson and Spearman Correlation Coefficients

71.

Table 3.3

 

 

REA. REQ. REI; REP. LSIZE.
REA; 0.5977 0.5160 0.6250 0.1648
0.0001 0.0001 0.0001 0.0001
REQ; 0.5985 0.4628 0.7821 0.1210
0.0001 0.0001 0.0001 0.0002 _
REL. 0.5154 0.4632 0.8326 0.1242
0.0001 0.0001 0.0001 0.0001
REP. 0.6248 0.7825 0.8330 0.1471
0.0001 0.0001 0.0001 0.0001
LSIZE. 0.1781 0.1257 - 0.1335 0.1537
0.0001 0.0001 0.0001 0.0001

 

a. Both Pearson (below the diagonal) and Spearman rank (above the diagonal) correlation coeficients are
presented p values are reported below the eoefiicients.

REA. =therelativeindustryrankingfortheinformativenessforfirmi’sARSand lOoKinyesrtbased
on analysts’ evaluation scores as in the AIMRReports,

REQ. = the relative industry making for the informativeness for firm i’s quarterly reports in year t
based on analysts’ evaluation sooresasintheAIMRReports,

REL. = the relative industry making for the inforrnativeness for firm i’s investor relations in year t
basedonanalysts’ evaluation seoresas intheAIMRReports,

REP. - the relative industry ranking for the inforrnativeness for firm i’s disclosures prior to the release
of the earlier of the ARS and lO—K in year t, based on a weighted average of analysts’ evaluation
scores for quarterly reports and investor relations.

LSIZE. - the logarithm ofSIZE. as defined in Table 3.2.

72

significantly correlated with one another at the 0.0001 level. The highest correlation is
0.5977 between REA“ and REQ.., and the lowest is 0.4628 between REQ.. and RIM...
3.5 EMPIRICAL RESULTS

This section reports empirical results for tests of hypotheses one, two, and three.
When appropriate, tests are performed for both the annual earnings announcement (ABA)
and the SEC receipts of the earlier of ARS and lO-K (REC) for comparison and additional
insights. All p values and levels of significance are one-sided if directional predictions
are made, and two-sided otherwise. For tests involving 2... and 2..., t-test results based on
the empirical cross-sectional distributions of U... and V... are reported in brackets, in
addition to results based on their theoretical distributions. To reduce the efl‘ects of
nonlinearity in the returns-eamings relation, UE.... values above the 99 percentile are set at
the 99 percentile value. Similarly, UE.... values below the one percentile are set at the one
percentile value.
3.5.1 Hypothesis 3.1

Figure 3.2 graphs U. in the test interval for both ABA and REC based on the full
sample, as reported in Table 3.4. For AEA, U0 is larger than U. in any other period in the
test interval. Its value of 1.48 is statistically significantly at conventional levels, with a Z...
(t...) value of 10.40 (5.80). For REC, however, no clear pattern emerges for U. in the test
interval. Its value of 0.96 is also smaller than its theoretical expectation of one, though the

difference is not statistically different”.

 

3’ It is also smaller than U. in seven of the other 10 periods in the test interval.

73

can—am ==h 05 .3.— <H< .3... DH: 3.. p: be 8...!»

 

 

an 23E
P
n v n N Z c T N. n. Y n-
. . u . .. n a n n 8...
0mm+ (Elli
.. and
-. 3.6
. 86

 

 

 

 

 

74

Table 3.4
Values of U. for REC and AEA for the Full Sample

 

 

 

REC (sac Receipt) ABA (Annual Earnings Announcement)
Period U. z... to. U. 20. tn.
-5 1.14 3.03 2.02 1.03 0.68 0.45
.4 1.18 3.89 0.94 0.99 0.25 -0. 18
-3 0.99 . -0.28 -0.21 1.08 . 1.02 0.99
-2 1.01 0.18 0.14 0.91 -194 -177
.1 1.01 0.24 0.12 0.88 ~2.58 -2.43

.-W<«c¢:~:gr

.
4-2»:v:-:<-'-.¢:-"- '. 5 .

 

risivix‘

+1 0.93 -l.4l -1.09 1.05 1.15 0.94
+2 0.91 -1.85 -1.48 1.01 0.17 0.14
+3 1.10 2.12 1.50 1.07 1.55 0.95
+4 0.86 -2.93 -2.77 1.01 0.32 0.21

+5 0.97 -0.62 -0.47 1.00 0.00 0.00

 

75

Results in Table 3.5, which are for AEA, provide some assurance that REP.. is indeed a
valid proxy for the inforrnativeness of prior disclosures not attributable, to its correlation
with firm size. First presented are the values of U. for PL and PH portfolios. The U0
value for the PL portfolio is 1.70, significantly larger than that for the PH portfolio at 1.29
as indicated by a t test ( t = 2.47, not reported in Table 3.5). The last five columns are the
values of U. for the portfolios SS (with SIZE. below and equal to the median) and SL
(with SIZE. above the median) respectively. A t test indicates that the U0 value of 1.52
for the SS portfolio is not significantly larger than that for the SL portfolio at 1.45 (t =
0.409, not reported in Table 3.5). I

Table 3.6 reports test results for hypothesis one based on firm-years in the
portfolio PLAH, i.e., observations with less informative prior disclosures and more
informative annual reports. The U0 value for the portfolio is 0.75, significantly smaller,
instead of larger, than its theoretical value of one, as indicated by a Z... (t...) value of -1 .95
(-2.97). Moreover, there is no evidence that abnormal returns are more volatile than other
periods in the test interval. The value of U0 is smaller than nine of the 10 other U. values
in the test interval. This compares to a U0 value of 2.10 for AEA, which is .significantly
larger than one at conventional levels?"5 In sum, there is no evidence supporting the
existence of unusual returns volatility in the three-day event period in the subsample that

such an efi‘ect is most likely to exist.

 

3‘ The second highest U. in the ABA test interval is 1.20 for t = -3.

76

 

2.7 2.7 36 «ad 3”.— ecA and. :6. «ad ad 36 8A n+

3A SA a— A 2A.. 5.7 36 had 36 3A nod. wad. med v+
. 3A vnA 2A end 36 3A 2A and «A A and. 2.6.. med m...
emd and 8A cod. 8.? 8A 2.6. and. wed med 3 A «9A N+
36 3.6 8A 3.: ”AA 8A 2.9 3 6 3A «A A 91— 2A 3.

 

 

 

 

8..- .3- a; 2:- 3a- 1 Be 8.... Sn. 2.5 2 e. a ..v a... _-
Se. 23. 8.: a..- 8n. 8... 8.". «E. a; 8.9 8.9 «a... N-
a... .2 8.. Re a... 8.. as. :5. a... 3... c2 8.. m-
as. 8.9 a... a3 :5 a: a... .2- a; 8... a... 2.. r
:3. 5.. use .2 NE a: 63 a3 8.. a... 8... a: n-
s. 5N .2 5. 5~ .2 .3 5w 5 a. 5N .3 85..
A872 .muaa .a $72 5:55 mm 3.12 3.25.... 295 E 5.12 9:25.... an: .5
gm 3 8:35.. .52 3 83:8

 

..HN—m A...- sg .3 $23.32!— oEEam ==h 5m). «6:. .3.— .D ..e 83—3,
. md 033—.

77

Table 3.6
Valuu of U. for REC and AEA for Portfolio PLAH

 

 

 

Period REC (SEC Receipt) ABA (Annual Earnings Announcement)
U! 20‘ tUt Us zlk tU‘
-5 1.07 0.53 _ 0.46 1.12 0.94 0.78
-4 0.96 -0.32 -0. 17 1.14 1.12 0.93
-3 1.10 0.77 0.44 1.20 1.56 1.25
-2 0.77 -1.78 4.27 1.07 0.55 0.37
-1 0.86 -1.11 -0.84 1.09 0.70 0.50

 

+1 0.68 -2.46 -3.47 1.07 0.54 0.40
+2 1.01 0.07 0.04 0.90 -0.79 -0.83
+3 1.08 0.65 0.51 0.91 -0.67 -0.54
+4 1.00 -0.02 -0.01 1.00 0.02 0.02

+5 ' 1.16 1.27 0.63 0.75 -1.97 -2.66

 

78

3.5.2 Hypothesis 3.2

Figure 3.3 plots V. in the test interval for both ABA and REC With the full sample
divided into portfolios UEN and UEP, as reported in Table 3.7. The last six. columns are
results for ABA The value of V0 is -0.29 for UEN, significantly negative at conventional
levels with a Zvo (tvo) value of -5.65 (4.83). Also for AEA, the value of V0 is 0.11 with
regard to UEP, significantly positive at conventional levels with a Zvo (tvo)value of 2.56
(2.08). These results indicate that the annual earnings announcements can assist investors
in updating their expectations for EPSM.

Figure 3.4, which parallels Figure 3.3 but is for REC, is based on the first six
columns of Table 3.7 . For portfolio UEN, there is no. evidence supporting the existence of
incremental information content due to the release of the earlier of the ARS and lO-K.
The value of V0 is positive, instead of negative as expected. However, for portfolio UEP,
its V0 value of 0.09 is significantly positive at the 0.05 level, with a Zvo (tvo) value of 2.21
(2.27). This suggests that, for the full sample, the ARS and lO-K contain incremental
information for the future earnings of firms in the “good news” portfolio.

Table 3.8A reports V. values for ABA in the test interval. In the table, UEN is
fiirther separated into two portfolios, UENPL and UENPH, based on RIRP;.. The same
procedure applies for UEP. As expected, the V0 value for UENPL is -0.40, smaller (more
. negative) than that for UENPH at -0. 18. A t test indicates that the difl‘erence is significant
at the 0.05 level (one-sided, not reported in Table 3.8A). The V0 value for UEPPL is 0.20
and, as expected, larger (more positive) than that for UEPPH at 0.03. The difi'erence is

significant at the 0.05 level via a t test (one-sided, not reported in Table

79

 

::.:. o:.:.. 8.: NY: 3.: ~:.: 3.: ::A 3.: ...:A _:A m:.: n+

8A. ::A 8.: am: on: 8.: 3N Run «— .: :nA 2A 8.: A...
3A 3 .N 8.: 3.? mod. n:.:. nmA AAA 8.: 3.: n:A m:.: m+
3.: 3.: ~:.: 3 A S A 8.: 3.: Na: 3.: SA SA . n:.: ~+
«NA nmA 8.: «9:. 8.:- n:.¢. 3.: :v: «:6 man: «0.: m:.: . A+

 

 

 

 

K... 3... 8... a." .3 N... ...... ...... S... 2.... ...... ...... .-
8.... 8.... 8.... .....- .....- 8.... m3 2." ...... 2... ...... ...... N- .
8.... 8.... 8... ...: 2." a. ... ...... a... ...... 8.... 8.... . ...... m-
... .... ...... 8.... ...N... n... ...... 3.... an... 8... ...... ...... ...... ...
an..- a..- 8.... a... a... 3... 8... 8... 8... _ S..- Q..- «...... n-

.3 ..N .> .3 ..N .> .3 ..N .> .3 ..N .> 3......
33",: ...... ......uz. 2m: .....nuz. ..m: 3.72. 2m:
..=2=§ao§<\mu=1§m .252. <9. .2383. um... um...

 

Em: ... ..ua 2.. ... 3.3.3:... 9.9.3 ...... ...... <5. ...... ...... .... .> ... 8......»
5. 2......

80

2.5 ... =5 2.1.. 38.35.. ......Sm ...... ...... <5. .3. p> ... 8...;

...». as»...

 

 

mm:<m< III

Zm5<m< IOI

 

 

 

1h

. :m.:.

.. 8....

.. ......

 

 

. :A.:

. :N.:

- :n.:

 

 

8.: »>

81

2a.. ... am 2.. ... 38.5.... .356 ...... ....3 6B. .3. .S ... 8.3.5

...n 9...»...

:n .:.

 

.5305— III

ZmDUm—M IOI

 

 

 

+-

fil-

db

L.

~l

Ml

1r

L

J.

a 3....

.. 8....

.. 3....

 

:_ .:..
86:2

. 2...

a ..N...

r :m:

. :9:

 

 

82

 

8.7 8:. 8:.- na: 3.: 8.: 3.: :N: 8.: 3.: cm: 8.: n..-

::A 8A NA .: mm: «M: N:.: AN: AN: 5.: :N: AN: ~:.: v+
«FA 3A A A .: 8A 8A 8.: AN:- -.:. 8:. g.:. 8:. m::. m.-
n::. . 8:. 8:. 8A 3A :A.: :nA SA :A.: 6A.: 2.: A:.: ~+
8.: 8.: 8.: 8.: 8A 6:: :A .: :A .: 3.: 8. 7 8. 7 a:.:. :-

 

 

8... ...... 8... a... a... 8... no. .... ...... ....N ....N ... ... .-
2...- n.-.... ......- ....... ...... . 8... ......- 8... 8.... 8.... ......- 8...- N-
a... ...... N...... S...- »n... ......- ..n.. 3.. ...... ...... ...... ...... .-
..... ...... ...... z..- n...- 8.... 2.9 ...... ......- 3... S... 8... ...
Nu...- mn... 8.... .....- a..- 8.... 3... N2. ...... 2... ...... 8... n-
.3 3N .> .3 ..N .> .3 ...-N .> 3 am .> 3......

 

AnmNuz «02.5585 895 In 330-2 62.5585 «mo—v An— GEnZ .339585 895 I.— AaaAuZ .gugo-EA 9.8—: Am

 

mm: 2m:

 

.55. ...... in... ... ....m 2.. ... 3.3.8.... ....aqm ...... ...... 3.2 ......» ... 8...;
«...... 2......

83

3.8A). Table 3.88 presents results for “bad news” and “good news” observations further
partitioned based on size. The values of V0 are -0.291 and -0.287 for UENSS and
UENSL portfolios, essentially the same. For UEPSS and UEPSL portfolios, the values
are 0.14 and 0.08 respectively, the difl'erence between which is insignificant at
conventional levels (not reported in Table 3.88). In sum, results in Tables 3.8A and 3.83
are consistent with those in Table 3.5 and provide further assurance that RIRPu is a valid
proxy for prior information.

Table 3 .9 presents test results for hypothesis two based on firm-years in portfolio
PLAH. Consistent with test results for hypothesis one, there is no evidence supporting the
existence of incremental information content associated with the SEC receipts of the
earlier of the ARS and 10-K for observations in this portfolio. The V0 value for the “bad
news” portfolio is 0.11, instead of negative as predicted. The V0 value for the “good
news” portfolio is 0.04, positive but not statistically significant, which is inconsistent with
results in Table 3.7 . For comparison, the V0 values during the annual earnings
announcement period for the “bad news” and “good news” portfolios are -0.55 and 0.33
respectively, both of which are significant at the 0.05 level.
3.5.3 Hypothesis 3.3

Table 3.10 reports. test results for hypothesis three, which predicts that in a
regression of abnormal returns in the event period on UEN, the ERC should be an
increasing function of RIRA;. if the ARS or lO-K released in the event period provides
incremental information to the market. It also predicts that the ERC should be a

decreasing function of RM“, which proxies for the inforrnativeness of prior information.

84

 

3:.

an.—

ea:

3.:

5:.

an.

N:.:.

3.:

an:

8.:

ma.—

3.:

am..—

an:

3.:

::.N

no:

3.:

2.:

3.7

:5:

mm:

No.7

2.:

.n:

2:.

8.:

3.:

3.:

8:.

mm.—

3:.

5.7

m:.~

:N:.

2.7

2.—

2:.

.... ...
S...
8...
......

n:.:.

n+

v+

m...

N+

.+

 

 

3.. ...... 8.. a... ...... 8.. 3.. ...... ...... 8.. 2.. .... .-
.... .... ...... 3.... ...... 8... ...... S... ...... .....- .....- ...... N.
n... a... 8... 2.. .... ...... ...... 8.. 3.. on. a... 2.. ..-
...- .N..- .....w 8.. 8.. 8.. a... a... 8... 2.. ...... ...... ...
8... a... 3... ...... 8... 8... 2.... 2.... 8... ...,. 2.. 2.. n-
.2 ...N .> .>. SN .> s. ...N .> .>. .>~ .> .55..

 

.....Nuz duh... ..m

...Nuz .23.... mm

.....nz quasi

Gauz $2.55.: mm

 

mm:

2m:

 

gm .2... ...”... .. ....m 2.. ... 3.2.2.... 236 ...... ...: <3. 2....» ... .2...>

and 03:.

85.

 

an; sh.— n_.: 3:. 2:. 8:. mm: :v: m:.: 2.. ~:._ 2: T.

«0:. :v.:. n:.:. S .:. 3:. 8:. «ad SUN 2.: 3:. «9:. ::.:. v+
«N: :N: 8.: he. 7 0:. 7 3:. :m: :n: 2...: 3 ._ cm; 3 .: n+
Na: 3.: 3 .: 2.7 3.7 :_ .:. ::.: ::.: :_ .: ::.: ::.: ::.: a...
3:. 3:. 3:. uh:. , 3:. 2:. 3:. 3:. ::.:. am: we: 5:: I. .

 

 

 

 

3... an... 8.. 3.. n... ... ... ...... ...... ...... .....- an..- a... .-
...... .n... 8.. 8... .....- 2... a... ...... ...... a... S... ... ... N-
a... n... ...... 8.. ...... ... ... 2.... .:.. S... ...... S... ... ... m-
...... 8... ... ... «.... ...... 2.. 2... ...... ...... ...... an... 8... ...
.. ... .. ... 8.. x S... ...... ... ... a... ...... 8... 2...- «.....- 2... n-

... ..N .> s. .>N %> .3 .>~ .> .>. ..N .> 8......

8th: ..m: 51: 2m: .....uz. ...... 2W2. 2m:
9:38:32... 9.6.... 352.4”... 3.83. 0.5. om...

 

in... .. gum 2.. ... 3.3.2:... ...... 2.8.3.. ...: ...... .2... H... 2.. .> .. .2...>
a... 2...."

86

Table 3.10
Regression Results in ABA and REC Periods

 

 

 

REC (SEC Receipt) ABA (Annual Earnings Announcement)
Estimate Standard t Estimate Standard t
Error Error

Intercept .0024 .0014 2.42 0.0013 0.0013 -1.00

131 -.0214 .0504 -0.43 -0.0731 . 0.0605 -1.21

02 .0134 .0434 0.32 0.1025 0.0522 1.96

03 -.0354 .0404 -0.8‘7 -0. 1114 0.0491 -2.27

y .0084 .0094 0.90 0.0213 0.0109 1.96
adjusted-R2 -0.001 - - 0.016 - -
F statistic 0.767 - - 4.826 - -

 

87

This means that in the regression, B; is expected to be positive and [33 is expected to be
negative.

Results for the annual earnings announcement period are reported in the last three
columns. They are intended to serve as a benchmark for the specification of the regression
model and to provide additional insights into returns behavior in the period. The value of
B. is -0.0731, which is not significantly difi‘erent fi'om zero (t = -1.21, two-sided)”. As
expected, [3; is significantly negative at the 0.05 level (t = -2.27, one-sided), suggesting
that more informative prior disclosures can assist investors in forming expectation about
[IEiH-l before the release of BPS“. Interestingly, B2 is significantly positive at the 0.05 level
(t = 1.96, one-sided), suggesting that some firms release either non-audited ARS and lO—K
or some earnings relevant information contained in these reports”. The estimate for y is
significantly positive with a p value of 0.051 (two-sided). This is consistent with either
that more information is made available about BPS“ for larger firms in the period, or that
the earnings process is more predictable for larger firms, or both.

Unreported sensitivity analysis is conducted to provide assurance that these results
are robust. A test in the spirit of White (1980) fails to reject the null hypothesis that the

first and second moments of the model are well specified ( p = 0.4801). Results are

 

37 Note that because of the inclusion of LSIZE. in the regression, no observation in the
sample has an ERC that equals 31- In a simple regression of 11;... on UEN, the [3. value of
0.03 84 is significantly positive (t = 3.02, two-sided), which are consistent with results
reported in Table 3.10.

3' Another possibility is that the earnings process is more predictable for firms with more
informative ARS and 10-1( than for firms with less informative ARS and lO-K. However,
this is rather unlikely because if m is a proxy for earnings predictability and/or
quality, RIRP;. is also likely to be a proxy for the same construct, and [33 would be positive
instead of negative as predicted.

88

essentially the same if the effects of equity beta and book-to-market ratio on the ERC are
controlled for in the regression.” Similar results obtain if um is replaced by V... as the
independent variable in the regression, or if UEM values above the 99 percentile and
below the one percentile are deleted.

The first three columns of Table 3.10 report test results in the period when the
SEC receives the earlier of the ARS and lO-K. The regression has an adjusted R2 of -
0.001. Its F statistic is 0.767, which is not significant at conventional levels. B; is negative
as predicted but is not significantly different from zero (t = -0.87, one-sided). More
importantly, B; is not significantly positive as predicted (t = 0.32, one-sided). In sum,
consistent with test results for hypotheSes one and two, results in Table 3.10 show no
evidence supporting the existence of incremental information content associated with the
SEC receipts of the earlier of the ARS and lO-K.
3.5.4 Prerelease before SEC Receipt Dates

One possible explanation for the absence of abnormal returns behavior during the
SEC ARS and 10-K receipt periods is that firms may have released these reports and/or
the most relevant information in these reports to shareholders and analysts before
submitting them to the SEC. This is tested by regressing abnormal returns accumulated
over 1 -1, ..., -5 on UBM. The sample size is reduced to 913 because 20 observations
with annual earnings announcements in t = -3, -5 are removed.

Table 3.11 reports the test results. The regression has an adjusted R2 of 0.008

and an F statistic of 2.849. The value of B; is 0.0882 and not significantly difl‘erent fi'om

 

39 Collins and Kothari (1989) indicate that the ERC varies systematically with equity beta
and the book-to-market ratio.

89

zero (t = 0.81, two-sided). The value of 02 is 0.3089, significantly larger than zero at the
0.01 level (t = 3.23, one-sided). The 13; value is -0.1487, consistent with prediction and
significant at the 0.05 level (t = - 1.65, one-sided), suggesting that, on average, more
information about EPSM has been preempted for firms with more informative prior
disclosures than for firms with less informative prior disclosures in fiscal year t. The value
of y is -0.0223, not statistically difl'erent from zero.

Sensitivity analysis not reported in tables indicates that these results are not
attributable to model misspecifications. (le test as suggested by White (1980) fails to
reject the null hypothesis that the first and second moments of the model are well specified
(p = 0.5830). (2)The results remain qualitatively similar if the efl'ects of equity beta and
book-to-market ratio on the ERC are controlled for in the regression, if um is replaced by
V.» as the independent variables in the regression, or if UEM values above the 99
percentile and below the one percentile are deleted fi'om the sample. (3)The null
hypothesis that the distribution of um is normal is rejected at the 0.0001 level for the 913
observations used in the regression, but the same hypothesis can not be rejected at
conventional levels if seven observations with um values three standard deviations away
from the mean are deleted. The results based on the remaining observations are stronger.
(4) Results from univariate regressions of um on U13“... are consistent with multivariate
regression results reported in Table 11. The adjusted-R2 values for both the firll sample

and subsamples PLAL, PHAL and PHAH are negative. On the other hand, for subsample

90

Table 3.11
Regression Results in Periods before ABA and REC

 

 

 

REC (SEC Receipt) ABA (Annual Earnings Announcement)
Estimate Standard 1 Estimate Standard t
Error Error

Intercept 0.0050 0.0024 2.12 0.0008 0.0023 0.34

pi 0.0882 0.1096 0.81 . 0.0518 0.1084 0.48

02 0.3089 0.0957 3.23 0.1606 0.0946 -1.70

133 0.1487 0.0900 -1.65 0.0331 0.0890 0.37

7 0.0223 0.0197 .1. 13 0.0141 0.0195 0.72
adjusted-R2 0.008 - - 0.007 - -
F statistic 2.849 - - 2.674 - -

 

91

PLAH, i.e., observations with less informative prior disclosures and more informative
ARS and 10x, the adjusted-R2 value is 0.1210. The value of 13. is 0.3213, which is
significantly positive at the 0.0001 level (t = 4.22, one-sided). Further analysis indicates
that these univariate results for PLAH are not caused. by a small number of influential
observations.

The significantly positive value of B: indicates the ERC is, on average, larger for
firms with more informative ARS and 10—I( than for firms with less informative ARS and
lO-K. For firms that release their ARS and lO-K to investors (but not the SEC) in this
period, this provides direct evidence supporting the existence of incremental information
in these reports. For firms that release the most relevant information in these reports but
not the reports themselves, this result Can be interpreted as indirect evidence supporting
the existence of incremental information in these reports, i.e., the market would react to
the release of these reports later had the information not been released in this period.
Taken together, the result provides evidence that the inforrnativeness of the ARS and 10-
K systematically afl‘ect returns behavior in this period. .

.The same regression is also performed for the period right before the annual
earnings announcement periods, and results are also presented in Table 3.11. 03 is
negative but not significantly different form zero (t = -0.37, one-sided), and y is positive
but not significantly different form zero (t = 0.72, two-sided). The value for B; is -0.1606,
which is significantly different from zero at the 0.10 level (t = -1.70, two-sided). Further

analysis (not reported in Table 3.11) indicates that this is likely due to the correlation

92

between the inforrnativeness of the ARS and 10-K and that of the quarterly report“. The
fact that B; is not significantly positive in the pre-AEA period is not surprising. Firms
rarely release information in the ARS and lO-K before annual earnings announcements.
Moreover, this also provides some assurance about the validity of RIRAa as a proxy for
the inforrnativeness of the ARS and lO-K.‘l
3.5.5 Delayed Responses

Because it may take several days for the SEC to make the ARS and 10-K available
to the public, and that investors may need time to access and analyze the information in
the reports, it is conceivable that the market may respond to the information in the reports
days after their receipt by the SEC. This is tested by regressing abnormal returns
accumulated over 1: = +1, ..., +5 on UEH. The sample size is reduced to 919 because 14
observations with earnings ann0uncements for the first quarter of year t+1 in t = +3,
+5 are removed. A similar regression, with abnormal returns accumulated after the annual
earnings announcement period, is also performed for comparison. Results for both
regressions are reported in Table 3.12. No supporting evidence for delayed market
responses is found in either case. In both regressions, the adjusted R2 values are negative,

and the values of B2, B3, and B4 are not statistically difi‘erent fi'om zero.

 

‘° The regression is re-estimated with my '+1 replaced by RIRQi.‘ '+1 and
RIRIR‘UEM. The value of B2 is -0.0860 and not significantly difi'erent from zero (t =
0.42). The coefficient for RIRQn‘I '+1 is -0.2794 and significantly negative with a t
statistic of -2.87.

“ If B; for the pre-REC period is significantly positive due to the correlation of We
with some omitted variables, then it is also expected to be significantly positive in the pre-
AEA period.

93

Table 3.12
Regression Results in Periods after ABA and REC

 

 

 

REC (SEC Receipt) AEA (Annual Earnings Announcement)
Estimate Standard t Estimate Standard t
Error Error
Intercept 0.0094 0.0023 4.16 0.0048 0.0022 2.19
pr 0.1340 . 0.1050 1.28 0.0050 0.1021 0.05
132 0.0461 0.0937 0.49 0.1287 0.0910 -141
(33 0.0460 0.0876 0.53 0.0174 0.0852 0.20
7 0.0216 0.0188 .1. 15 0.0125 0.0183 0.68
adjusted-R2 0.0023 - - 0.0003 - -

F statistic 0.467 - - 0.0934 - -

 

Chapter 4

SUNIMARY AND CONCLUDING REMARKS

This chapter provides summary and concluding remarks for this dissertation.
4.1 SUMMARY AND CONCLUDING REMARKS FOR THE FIRST PAPER

The first paper examines the efl‘ects of corporate disclosures on market
expectations of future earnings by addressing two interrelated research questions, which
are (1) whether stock prices anticipate earnings information earlier for firms with more
informative disclosures than for firms with less informative disclosures, and (2) which
alternative disclosure media contribute to such an earlier anticipation. Results for the first
research question support Healy and Palepu’s (1993) suggestion that disclosures
constitute a unique, nonsubstitutable source of such information. The market-adjusted
returns of firms. with more informative disclosures start to reflect earnings changes about
three months ahead of those of firms with less informative disclosures. The lead is
statistically significant and still present after controlling for firm size and the degree of
analyst following. Tests for the second research question find that such an earlier
anticipation of prices over earnings mainly results fiom more informative investor (analyst)
relations, instead of annual reports, quarterly reports, analyst following, or other factors
proxied by firm size. The size effect becomes Statistically insignificant after the effects of

disclosures are controlled. These findings indicate that investor relations are more

94

9S

efl‘ective than quarterly and annual reports in communicating firm-specific information to
investors. They also suggest that policy makers such as the FASB and SEC should
encourage firms to make more voluntary disclosures so that investors can form better
expectations about firms’ future performances.

A direct extension of this paper is to study the efi‘ects of corporate disclosures on
finns’ cost of equity capital. To date, evidence for this issue is still very limited. The most
direct evidence is provided by Botosan (1995), who finds that greater voluntary disclosure
is associated with a lower cost of equity capital alter controlling for cross—sectional
variation in systematic risk and size, but a measure other than market value must be used
to proxy size. She attributes part of the weakness of statistical results to the small sample
size of 122 observations, but voices concern that pooling firm years from difi'erent
industries may introduce potential confounding effects. The larger sample size and the use
of relative industry ranking of disclosures in this paper can potentially overcome both
problems.

4.2 SUMMARY AND CONCLUDING REMARKS FOR THE SECOND PAPER

'In the second paper, I investigate the incremental information content of annual
and lO-K reports with a research design that explicitly controls the cross-sectional
difl‘erences in information disclosed prior to the release -of these reports and the
inforrnativeness of these reports themselves. I also empirically examine alternative
explanations that may account for the absence of abnormal returns behavior associated
with the SEC receipt and release of these reports, such as prerelease by firms of these

reports or the most relevant information they contain, and delayed market responses.

96

For annual earnings announcements, I find results that are consistent with
theoretical predictions in Holthausen and Verrecchia (198 8) and prior empirical studies.
The magnitude of market response, measured as either squared market model errors,
abnormal returns, or the ERC fiom regressing abnormal returns on unexpected earnings, is
smaller for firms with more informative interim disclosures. Moreover, the ERC is also
significantly higher for firms with more informative ARS and lO-K, indicating that the
inforrnativeness of the ARS and, 10-K systematically afl‘ects returns behavior in this period.
I also find evidence that is consistence with the existence of incremental information
content of these reports in the period immediately prior to the SEC receipt dates. The
ERC fi'om regressing abnormal returns accumulated in this period on future unexpected
earnings is significantly higher for firms with more informative ARS and lO-K reports, and
significantly smaller for firms with more informative prior disclosures. This supports the
conjecture that firms have either released these reports themselves or the information in
these reports to the market before filing them with the SEC. Such an interpretation is
further supported by the fact that no abnormal returns behavior has been detected in the
three-week period immediately after the three-day SEC receipt period. On the other hand,
results in the SEC receipt period provide little evidence for the incremental information
content of the ARS and lO-K. While one test supports the existence of incremental

information for observations in the “good news” portfolio, all other tests fail to document
any corroborating evidence.

This paper provides evidence that the infonnativeness of disclosures contained in

ARS and 10-K systematically affects returns behavior in both the earnings announcement

97

period and the period prior to the SEC receipt of these reports. These results explain the
failure of previous studies in detecting abnormal returns behavior on the SEC receipt dates
and are consistent with the existence of incremental information in these reports. This
paper has also extended extant studies that use firm size as a proxy for the availability of
prior information to investigate market reaction to annual earnings announcements by
providing direct evidence that the magnitude of the response is negatively associated with

the inforrnativeness of prior disclosures.

LIST OF REFERENCES

98

LIST OF REFERENCES

Akerlof; G. 1970. The Market for “Lemons”: Quality Uncertainty and the Market
Mechanism. Quarterly Journal of Economics 84, 488-500.

Alford, A., J. Jones, and M. Zmijewski. 1994. Extensions and Violations of the
Statutory SEC Form 10-K Filing Requirements. Journal of Accounting and
Economics 17: 229-254.

Association for Investment Management and Research (AIMR). 1980/1981 - 1991/ 1992.
Corporate Information Committee Reports. Charlottesville, Virginia.

Atiase, R. 1985. Predisclosure Information, Firm Capitalization, and Security Price
Behavior Around Earnings Announcement. Journal of Accounting Research
(Spring): 21-36.

Ball, R. and P. Brown, 1968. An Empirical Evaluation of Accounting Numbers. Journal
of Accounting Research 6, 157-178.

Baldwin, B. 1984. Segment Earnings Disclosure and the Ability of Security Analysts to
Forecast Earning Per Share. The Accounting Review (July): 3763 89.

Barry C. and S. Brown. 1985. Difi‘erential Information and Security Market
Equilibrium. Journal of Financial and Quantitative Analysis (December), 407-
422.

Barth, M. 1994. Fair Value Accounting: Evidence from Investment Securities and the
Market Valuation of Banks. The Accounting Review (January): 1-25.

and C. Murphy. 1994. Required Financial Statement Disclosure: Purposes,
Subject, Number, and Trends. Accounting Horizons 8: 1-22.

, W. Beaver, and W. Landsman. 1994. Value Relevance of Banks’ Fair Value
Disclosures under SFAS 107. Working paper, Stanford University.

and A. Sweeney. 1995. Differential Information Environments and the Relation
between Accounting Amounts and Shares Prices: Cost of Capital and Recognition
versus Disclosure. Working paper, Harvard Business School.

Beaver, W. 1989. Financial Reporting: An Accounting Revolution. 2nd ed. Englewood
Cliffs, NJ: Prentice-Hall.

Bernard, V. L. 1987. Cross-Sectional Dependence and Problems in Inference in Market-
Based Accounting Research. Journal of Accounting Research (Spring): 1-48.

99

Botosan, C. 1995. The Efi‘ect of Disclosure Level on the Cost of Equity Capital and
Stock Market Liquidity. Doctoral dissertation, the University of Michigan.

Byrd, J ., M. Johnson, and M. Johnson. 1993. Investor Relations and the Cost of Capital.
Working paper, the University of Michigan.

Collins, D., S. P. Kothari, and J. Rayburn. 1987. Firm Size and the Information Content
of Prices with Respect to Earnings. Journal of Accounting and Economics:
111-38.

Collins, D. and S. P. Kothari. 1989. An Analysis of Intertemporal and Cross-Sectional
Determinants of Earnings Response Coeficients. Journal of Accounting and
Economics 11, 143- 181.

Cready, W. and P. Mynatt. 1991. The Information Content of Annual Reports: A Price
and Trading Response Analysis. The Accounting Review 66 (April): 291-312. '

Diamond D. and R. Verrecchia. Disclosure, Liquidity and the Coat of Equity Capital.
The Journal of Finance (September), 1325-1359.

Easton, P. and M. Zmijewski. 1993. SEC Form 10K/10Q Reports and Annual Reports to
Shareholders: Reporting Lags and Squared Market Model Prediction Errors.
Journal of Accounting Research 31 (Spring): 113-129.

Elliot, R and P. Jacobson. 1994. Costs and Benefits of Business Information Disclosure.
Accounting Horizon 8, 80-96.

Fama, E. and M. Jensen. 1983 8. Separation of Ownership and Control. Journal of Law
.and Economics 26, 301-325.

and . 1983 b. Agency Problems and Residual Claims. Journal of Law
and Economics 26, 327-349.

Foster, 6., C. Olsen, and T. Shevlin. 1984. Earnings Releases, Anomalies, and the
Behavior of Security Returns. The Accounting Review: 574-603.

Foster, T. and D. Vickrey. 1978. The Incremental Information Content of the lO-K.
The Accounting Review 53 (January): 921-34.

D. Jenkins, and D. Vickrey. 1983. Additional Evidence on the Incremental
Information Content of the lO-K. Accounting and Business Research (Spring):
57-66.

 

100

, and _. 1986. The Incremental Information Content of the
Annual Report— Accounting and Business Research (Spring): 91-98.

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

Gill, S. 1994. Motivations for and Reactions to Voluntary Management Disclosures.
Doctoral dissertation, Michigan State University.

Grant, E. 1980. Market Implications of Difl‘erent Amounts of Interim Information.
Journal of Accounting Research (Spring), 255-269.

Hakansson, N. 1977. Interim Disclosure and Public Forecasts: An Economic Analysis
and Framework of Choice. The Accounting Review 52: 396-416.

Han, B. ,.R Jennings, and J. Noel. 1992. Communication of Nonearnings Information at
the Financial Statements Release Date. Journal of Accounting and Economics 15:
63- 86.

Han, J. and C. Huang. 1995. The Information Content of Corporate Annual Filings with
the SEC. Working paper, Hong Kong University of Science and Technology.

Healy, P. and K. Palepu. 1993. The Effect of Firms’ Financial Disclosure Strategies on
Stock Prices. Accounting Horizons 7, 1-11.

, and A. Sweeney. 1995. Causes and Consequences of Changes in
Disclosure Strategy. Working paper, Harvard Business School.

 

Holthausen, R. and R Leftwich. 1983. The Economic Consequences of Accounting '
Choices: Implications of Costly Contracting and Monitoring. Journal of
Accounting and Economics 5, 77-117.

and R Verrecchia 1988. The Efi‘ects of Sequential Information Releases on the
Variance of Price Changes in An Intertemporal Multi-Asset Market. Journal of
Accounting Research (Spring), 82-106.

 

Jensen, M. and W. Meckling. 1976. Theory of the Firm: Managerial Behavior, Agency
Costs and Ownership Structures. Journal of Financial Economics 3: 305-360.

Knutson, P. 1992. Financial Reporting in the 1990’s and Beyond. AIMR:
Charlottesville, Virginia.

Kothari, S. P. and R. Sloan. 1992. Information in Prices about Future Earnings. Journal
of Accounting and Economics 15, 143-171.

101

Lang, M and R Lundholm. 1993. Cross-Sectional Determinants of Analyst Ratings of
Corporate Disclosures. Journal of Accounting Research (Autumn), 246-271.

, and . 1994. Corporate Disclosure Policy and Analysts. Working paper,
Stanford University.

Lees, F. 1981. Public Disclosure of Corporate Earnings Forecasts. The Conference
Board: New York.

Lev, B. 1988. Toward A Theory of Equitable and Eficient Accounting Policy. The
Accounting Review 63, 1-22.

Lipe, R 1990. The Relation between Stock Returns and Accounting Earnings Given
Alternative Information. The Accounting Review 65, 49-71.

Merton, R. 1987. A Simple Model of Capital Market Equilibrium with Incomplete
Information. The Journal of Finance (July), 483-510.

Patell, J. 1976. Corporate Forecasts of Earnings Per Share and Stock Price Behavior:
Empirical Tests. Journal of Accounting Research (Autumn): 246-276.

Penman, S. 1980. An Empirical Investigation of the Voluntary Disclosure of Corporate
Earnings Forecasts. Journal of Accounting Research (Spring): 397-417.

. 1994. The Articulation of Price-Earnings Ratios and Market-to-Book Ratios
and the Evaluation of Growth. Working paper, University of California at
Berkeley.

Pownall, G., C. Wasley, and G. Waymire. 1993. The Stock Price Efl‘eCts of Alternative
Types of Management Earnings Forecasts. The Accounting Review (October):
896-912.

Shores, D. 1990. The Association between Interim Information and Security Returns
Surrounding Earnings Announcements. Journal of Accounting Research 28
(Spring): 164-81.

Spence M. 1973. Job Market Signaling. Quarterly Journal of Economics 87: 355-374.

Stice, K. 1991. The Market Reactions to 10-K and lO-Q Filings and to Subsequent Wall
Street Journal Eamings Announcements. The Accounting Review 66 (January):
42-55.

Waymire, A. 1984. Additional Evidence on the Information Content of Management
Earnings Forecasts. Journal of Accounting Research (Autumn): 703-718.

102

Welker, M. 1995. Disclosure Policy, Information Asymmetry and Liquidity in Equity
Market. Working paper, the University of Saskatchewan.

White, H 1980. A Heteroskedasticity-Consistent Covariance Matrix Estimator and
A Direct Test for Heteroskedasticity. Econometrics 48: 817-838.

Wilson, P. 1987. The Incremental Information Content of the Accrual and Funds
Components of Earnings Afier Controlling for Earnings. The Accounting Review
62 (April): 293-322.