Professional Documents
Culture Documents
Thesis Proposal by
Ge Zhiyang
CHAPTER ONE
INTRODUCTION
negative impact on the market; they also call into question the credibility of
accounting practices and the quality of corporate financial disclosure and oversight.
In his speech at the New York University Center for Law and Business, former
Securities and Exchange Committee (SEC) Chairman Arthur Levitt remarks:
I fear that we are witnessing an erosion in the quality of earnings,
and therefore, the quality of financial reporting If a company fails
to provide meaningful disclosure to investors about where it has
been, where it is and where it is going, a damaging pattern ensues.
The bond between shareholders and the company is shaken; investors
grow anxious; prices fluctuate for no discernible reasons; and the
trust that is the bedrock of our capital markets is severely tested
Much research on the financial analysts earnings forecasts (FAF) concludes that
compared with statistical and time-series forecast models, FAF reflect comprehensive
information, are relatively accurate and are associated with market returns and risk.
However, FAF are also documented to exhibit systematic upward bias, that is, they are
consistently higher than actual announced earnings. The incidence of earnings
restatement announcement provides a special setting to study financial analysts
earnings forecast (FAF), specifically, whether there is information in financial
analysts forecasts for the earnings restatement firms.
1.2 Objective of the study
Though earnings restatement can be initiated for various reasons (to be discussed in
details in the literature review part), this study limits its attention to the earnings
restatements due to accounting errors, aggressive accounting practices, accounting
irregularities and accounting fraud. Earnings restatements due to these reasons are
evident signals that the previous financial statements lack integrity and reliability, and
that the management lacks competence or credibility in their oversight. These kinds
of earnings restatement most often have negative effects on the firms. This study
aims to examine whether the financial analysts (as market intermediaries) have
predictive power of such kinds of earnings restatement, and whether the market has
aggregate wisdom about such events.
Specifically, this study aims to address four issues. The first issue is whether the
financial analysts can predict the subsequently corrected earnings of the restatement
firms during their misstated period. In other words, whether the financial analysts
have prior knowledge of the misstatement and therefore have better prediction of the
true earnings information of the restatement firms.
The second issue is whether there is any difference in the distribution of ex ante
financial analysts earnings forecasts (FAF) for restatement and non-restatement
firms. The ex ante FAF examined in this study is the forecast for the annual earnings
of the year immediately prior to the year of restatement announcement. Four aspects
of the ex ante FAF distribution are examined. Since the earnings restatement firms
3
may exhibit more uncertainty before their restatement announcements than the nonrestatement firms, we examine whether the uncertainty is reflected in a difference in
the distribution of the ex ante FAF for the restatement vs. non-restatement firms.
The third issue is whether the market has aggregate wisdom about the circumstances
leading to the restatement announcement. If the market in aggregate has prior
knowledge of the earnings restatement, there will be a pre-restatement announcement
drift in stock prices for the earnings restatement firms. Furthermore, if the markets
aggregate wisdom of the restatement announcement incorporates the information
conveyed through the ex ante FAF distribution, then the markets reaction to the
restatement announcement is to be mitigated by the information reflected in the ex
ante FAF distribution.
The fourth issue is whether the properties of ex ante FAF distribution for the
restatement firms capture risk aspects of these firms. To answer this question we
examine the relationship between the properties of ex ante FAF distribution and the
subsequent risk measures of the restatement firms.
indicate that the properties in FAF distribution for earnings restatement firms are
associated with their potential risk.
1.3 Potential contribution of the study
The phenomenon of earnings restatement has drawn researchers attention only in
recent years following the accounting scandals of large companies like Enron and
WorldCom in the late 1990s. The growing number of earnings restatement due to
accounting errors and irregularities in the recent years reflects the deterioration of the
problem and the scrutiny of the SEC. The extant research on earnings restatement
explores this issue from different aspects, for example, the capital market reaction to
the announcement of earnings restatement, the incentives for managers to apply
aggressive accounting methods in violation of GAAP that leads to earnings
restatement subsequently, the corporate governance characteristics of the restatement
firms, etc.
This study adds to the literature of earnings restatement by examining the role of
4
important intermediary in the capital market and are considered sophisticated and
efficient in information collection and procession.
evidence on whether the analysts earnings forecast can have predictive power of the
misstatement and the subsequent earnings restatement, or in other words, whether
there is information in financial analysts forecasts for earnings restatement firms.
This study contributes to the empirical studies on earnings restatement by its sample
coverage of the latest years. As the manual search for earnings restatement is tedious
and time-consuming work, most existing research covers restatements of annual
earnings merely and extends their sample collection to year 2000 only. As the number
of earnings restatement is shown to grow dramatically over time, the inclusion of
earnings restatements made in year 2001 and 2002 may provide more sample cases in
support of our analysis.
including the beginning and the burst of the economic bubbles in 1990s, making it
possible to illustrate, if any, the time-series properties of earnings restatements.
This study also contributes to the literature of financial analysts behavior by
examining the information content of financial analysts earnings forecasts for
earnings restatement firms.
models.
The data sources, sample selection procedure and research method are
outlined in Chapter Four. Chapter Five presents the empirical findings and analyze
research results. Finally, Chapter Six concludes the study with implications and
suggestions for future research.
earnings restatement due to accounting errors and irregularities has been growing in
number and in significance during the past decade (relative statistics and evidence
will be discussed in detail in the third section).
Restatements can involve SEC-filed annual reports, which are audited by independent
auditors, and quarterly reports (in most cases unaudited). They can also involve only
the interim quarters of the current fiscal year, including the unfiled quarterly reports
that were publicly announced before. The channels of correction of the misstatement
used by companies include amended filings (10K/A or 10Q/A), which supersede the
original financial statements, the 10K or 10Q in the subsequent period carrying the
corrected number, or the form 8-K.
2.2.2 Reasons leading to earnings restatement
The restatement of financial statements can be initiated by a number of reasons. This
study limits its scope to the earnings restatements that correct the material
misstatement in previous financial results. These types of restatements result from
either unintentional accounting error, defined as mathematical mistakes, oversight,
7
or misuse of facts at the time the financial statements were originally prepared, 1 or
accounting irregularity, a term for intentional misstatements or omissions of amounts
or disclosures in financial statements done to deceive financial statement users, 2 or
the pursuit of aggressive accounting in violation of GAAP. Although some firms
admittedly acknowledge fraudulent financial reporting in their public announcement,
most firms will not do so. It is therefore hard to distinguish between intentional
manipulation and unintentional misinterpretation in some cases.
The reasons for material misstatements can be categorized into more elaborate
groups.
Accounting Office Report (GAO Report 2002 hereafter), restatements due to revenue
recognition problems constitute 38 per cent of the 919 restatements arising from
material misstatement including errors and fraud from 1997 to June 2002. Wu (2002)
find that of the 1,221 earnings restatements from 1977 to 2000 arising from
accounting errors and irregularities, 487 cases are caused by problems in revenue
recognition, representing the highest percentage (39 per cent) of the whole sample.
Besides accounting errors, irregularity and aggressive accounting methods, there are
other reasons arising from normal corporate activity or presentation issues that lead to
earnings restatement; for example, general changes in accounting principles under
GAAP, stock splits, dividend distributions, discontinuation of operations, change of
the accounting period, merger and acquisitions, and changes made for presentation
purposes. However, restatements caused by these reasons do not reveal previously
undisclosed and economically meaningful information to the investors, and do not
directly signal a lack of integrity or quality in previous financial statements.
Therefore they are excluded for the purpose of this study.
2.2.3 Growing number of restatements due to accounting misconduct
Early studies on earnings restatement find modest number of restatements in the
1970s and 1980s. Kinney and McDaniel (1989) examine firms that correct previously
reported quarterly earnings in a footnote to their annual reports because of accounting
errors covering the sample period from 1976 to 1985. They identify reports with
year-end restatement of previously issued quarterly financial statements sourced from
the National Automated Accounting Research System (NAARS) database of annual
reports. After excluding eight restatements related to prior fiscal year and two
extreme outliers, they observe 73 firms (178 quarters) that correct previous quarterly
earnings due to material errors.
DeFond and Jiambalvo (1991) examine firms making corrections of earnings
overstatement errors that existed in a prior years annual report from 1977 to 1988.
Their sample is obtained from a search of footnote disclosure of prior period
adjustments in NAARS and Accounting Trends and Techniques (ATT) database, and
41 firms are identified.
Recent studies on earnings restatement made since the late 1990s identify a
dramatically growing number of earnings restatements due to accounting errors and
irregularity.
financial statements in NAARS for restatements made from July 1987 to June 1995,
and supplement their search with the restatements in 1981 to 1987 that are examined
by DeFond and Jiambalvo (1991). After eliminating restatements that are other than
error corrections and not related to material misstatement, their final sample includes
116 error firms, with the highest frequency in 1988 when 21 firms are identified.
Palmrose and Scholz (2000) examine the companies that made first disclosure of
possible restatements between January 1, 1995 and June 30, 1999 and subsequently
filed amended 10-K or 10-Q with the SEC. They obtain their sample of restatements
from Lexis-Nexis News Library by using key-word searches for restatements and
include additional companies discussed for restatements in other unnamed sources.
Their final sample consists of 416 restatements for material misstatement, with 384
searched from Lexis-Nexis searches and 32 from other sources. The number of
restatements rises from 43 in 1995 to 136 in 1999. In the sample of restatements, 34
per cent are identified as having errors in revenue recognition, 28 per cent are for
operating expenses adjustments, and 23 per cent are for in-process R&D adjustments,
capturing the three categories with highest frequency of earnings restatements.
Turner, Dietrich, Anderson and Bailey (2001) examine the firms making earnings
restatement to prior annual earnings in their amended 10-K filings. They search 10-K
Wizard by key words for restatement within 10 words of financial statement. After
excluding those restatements not due to accounting errors or fraud, they identify a
final sample of 362 restatements, with 104 in year 1997, 116 in year 1998 and 142 in
year 1999. Of these restatements 15 per cent have problems in revenue recognition
and 7 per cent have errors in restructuring charges. Anderson and Yohn (2002) find
329 firms during the period 1997 to 1999 that restate financial statements and file a
10-K/A due to accounting errors, by searching 10-K Wizard for restated financial
statements. They include in their empirical tests only 161 restatements that have
available earnings and price data.
account for 17 per cent of the earnings restatements and 11 per cent are due to errors
in restructuring charges.
Similarly, Agrawal and Chadha (2002) examine a sample of restatements announced
from January 1, 2000 to December 31, 2001 by keyword searches from Lexis-Nexis,
Newspaper Source and Proquest Newspapers.
The GAO Report (2002) use Lexis-Nexis to search for restatement announcements
and identify 919 restatements from January1997 to June 2002 that involve accounting
errors or fraud resulting in material misstatement of financial results. The number
grows from 92 in 1997 to 225 in 2001, and the projected number for 2002 is 250.
When grouping these restatements by reasons, the error in revenue recognition is
found to be the most common reason, accounting for 38 per cent of the restatements.
Cost or expense-related issues are the next most frequently identified reason,
accounting for almost 16 per cent of all the restatements in the sample.
The Huron Consulting Group (2003) performs a keyword search for all 10-K/A and
10-Q/A filings in the EDGAR database from 1998 through 2002. They refine their
search to restatements due to accounting errors and exclude restatements due to
accounting principles changes and non-financial related restatements. Their results
show 993 restatements filed from 1998 to 2002, and the number rises from 158 in
1998 to 330 in 2002. Revenue recognition is shown to be the leading cause for
earnings restatement, accounting for 20 per cent of the sample restatements.
Wu (2002) obtains the sample of restatement firms by manual search of online news
libraries including Lexis-Nexis, Dow-Jones Library and ABI/Info databases. She
identifies a total number of 1221 earnings restatements due to accounting
misrepresentation, irregularities, fraud or errors from January 1, 1977 to December
31, 2001. In her sample, the number of restatement firms is in single digit from 1977
to 1982, and remains stable at less than 50 from mid-1980s to mid-1990s. The
number soars to 96 in 1998 and reaches 153 in 2001, with a peak of 204 in 1999.
When classified by reasons, 487 (40 per cent) of the restatements are caused by errors
in revenue recognition, representing the largest category.
This is followed by
improper record of costs or expenses, accounting for 463 (38 per cent) of all the
restatements in the sample.
In summary, various studies uncover a growing number of earnings restatement due
to accounting errors or fraud from the late 1990s. They document consensually that
problems in revenue recognition is the leading reason for earnings restatement.
11
trading days following the restatement announcement. They find that the severity of
the reaction is associated with restatements that include negative information about
management integrity and competence.
negative raw returns of -11% over the three-day window around the earnings
restatement announcements for firms that restate annual or quarterly earnings in
amended filings from 1995 to 1999. They provide evidence that the negative
reactions are associated with shareholder litigation against the firms.
Anderson and Yohn (2002) analyze 161 firms that restated financial statements and
filed a 10-K/A due to accounting errors during the period 1997 to 1999. They find a
significant CAR of -3.5% on average during the seven-day window surrounding the
announcement of earnings restatement. In the official report of GAO (2002), the 689
publicly traded firms identified as having announced financial statement restatements
between January 1997 and March 2002 have suffered a 10% fall in stock prices over
the three-day window around restatement announcement.
Wu (2002) report that on average the market reacts significantly negative to the
announcements of earnings restatements with a CAR of -11% over a three-day
window. Moreover, her results show a significant downward pattern starting about
half a year prior to the restatement announcements, and a persistent negative postannouncement drift for up to four months. She suggests that the preannouncement
pattern hints at other value-reducing events the restating firms have experienced
before their announcements, for example, earnings warnings, missing analysts
forecasts, analysts downward revisions or SEC investigations or enquiries. She also
suggests that the post-announcement drift is due to the release of additional details
pertaining to the restatement and investors incessant revision of their beliefs about
the firms economic prospects.
However, Turner and Wheatley (2003) find that when firms correct the previous
misstatement in the subsequent annual statement without interim announcement
regarding the restatement, they can have positive benefits and help achieve the
objectives of the aggressive managers. They name this effect as stealth earnings
management and document that minimal disclosures of recounted earnings are
overlooked by securities markets.
recognition of revenue with minimal disclosure, not only can the same recounted
income be recognized twice; comparison of the two years is improved by moving the
income from one year to the next.
In summary, many empirical studies on earnings restatements report significantly
negative abnormal market returns for the restatement firms over the short-window
around the announcement of earnings restatement.
14
analysts expectations.3
Distinct from previous studies using earnings or revenue growth to measure the financial difficulties
of restatement firms, Richardson et al (2002) use E/P and B/M ratios to examine the markets
perceptions of future growth of these firms, and their results indicate that restatement firms have
significantly lower E/P and B/M ratios compared with non-restatement firms.
4
They measure four different variables as related to capital markets incentives: external funds raised;
ex ante need for financing; historical trend in EPS growth; pattern of quarterly earnings surprises.
15
predictions that are more accurate than those generated by time-series models (Brown
and Rozeff, 1979; Fried and Givoly 1982; OBrien 1988; and Alexander 1995).
Previous studies on the superiority of FAF to time-series models suggest that FAF
contains comprehensive information including macroeconomic events, industry
information and firm-specific non-accounting information, while time-series models
rely exclusively on accounting information. Compared with time-series models, FAF
appears to have both a contemporaneous advantage and a timing advantage (Brown,
Hagerman, Griffin and Zmijewski 1987a). The contemporaneous advantage means
that financial analysts can better use information existing on the date that time-series
models can be initiated, and the timing advantage means that the financial analysts
can use information that occurs after the cut-off date for the time-series data but
before the report of the analysts forecast.
Some research has been extended to examine the relationship between the superiority
of FAF and the firms information environment. Brown et al (1987) argue that
financial analysts superiority is positively related to the dimension of the FAF
information set and negatively related to both variance of the interim observations of
earnings and the correlation between the information variables. They use firm size,
divergence of analysts opinions and number of lines of business as proxies for the
above three variables. Their sample consists of 168 quarterly forecasts from Value
Line and 702 annual forecasts from I/B/E/S from 1977 to 1982. Their results are
consistent with the information interpretation of the FAF superiority. Kross et al
(1990) collected FAF from Value Line for 279 firms from 1973 to 1981. Their results
show that the advantage of FAF over time-series models is related to the variability of
historical earnings and the extent of coverage in The Wall Street Journal, which is
consistent with the proposition that analyst advantage increases with increases in
information gathering incentives and information dissemination activities. However,
they fail to find a positive relation between analyst advantage and firm size, as
documented in Brown et al (1987c).
Studies also show that the accuracy of FAF is related to firms financial risk and
business risk, and the error in earnings forecasts is analytically associated with the
uncertainty that a firm faces. Cukierman and Givoly (1982) develop a model of
17
earnings expectations and they show that the cross-sectional error in earnings
forecasts is the correct empirical counterpart of uncertainty; that is, of the dispersion
of the distribution of expected earnings. Their model also implies (and is confirmed
by empirical tests) that the cross-sectional error is positively associated with the
dispersion of forecasts across forecasters. Ciccone (2003) tests three measures of
forecast error, namely, the absolute value of the difference between the mean annual
earnings forecast at fiscal year end and the actual annual earnings, the absolute value
scaled by actual earnings, and the absolute value scaled by price as of fiscal year end.
He finds that for all the US firms listed on the NYSE, AMEX and NASDAQ from
1978 to 1996, the forecast error has a positive relationship with the standard deviation
of annual earnings in the three previous years prior to the year of forecast, no matter
which measure of forecast error is used. Moreover, the firms with large forecast
errors are more likely to have negative earnings and earnings declines. He concludes
that firms that are distressed have systematically higher forecast error. Additionally,
firms with high business risk, as measured by earnings standard deviation, tend
towards higher forecast error.
Information content of FAF
Studies on the information content of analyst forecast have examined the association
between FAF and the securities market, and they conclude that FAF is a better
surrogate for market expectation. Fried and Givoly (1982) provide evidence that the
forecast errors of FAF are more closely associated with security price movements
than are the forecast errors from the submartingale model and the index model. They
conclude that FAF provides a better surrogate for market expectations than forecasts
generated by time-series models. Brown, Hagerman, Griffin and Zmijewski (1987b)
investigate the relationship between abnormal returns and five alternative proxies for
the markets assessment of unexpected quarterly earnings. Their results show that the
unexpected earnings based on FAF explains abnormal returns better than other
proxies. Alexander (1995) finds that the forecast errors of the most recent analysts
forecasts are more closely correlated with the abnormal returns than the errors of the
less recent analysts forecasts and the forecasts derived from time-series models.
Dispersion of FAF
18
Prior research has examined the relationship between dispersion in analysts earnings
forecasts and the uncertainty about firms future economic performance and provided
empirical evidence on such relationship. Givoly and Lakonishok (1984) argue that
the level of forecast dispersion is perceived by investors as valuable information
about the level of uncertainty concerning firms future economic performance.
Forecast dispersion is suggested to reflect both uncertainty and lack of consensus
among market participants about future events (Barry and Jennings 1992; Barron,
Kim, Lim and Stevens 1998). Givoly and Lakonishok (1988) examine the
relationship between dispersion of FAF, used as a measure of uncertainty, and the
stock properties, particularly risk characteristics, such as the beta (computed over the
two years preceding the year for which the correlation is compared), marketability
(shares traded during the year as a percentage of shares outstanding), firm size
(natural logarithm of the market value of the firms equity at the end of the year), and
earnings growth variability (measured as the standard deviation of the rate of growth
in EPS over the years 1961-1980). They find a positive and significant association
between forecast dispersion and the traditional market-based risk measure (beta) and
the accounting-based risk measure (earnings growth variability), and a negative
although insignificant correlation between size and forecast dispersion. They also
find a positive association between forecast dispersion and marketability.
Malkiel (1981) uses a measure of the dispersion among Wall Street security analysts
concerning the future earnings and dividend growth of the company as a risk variable,
and he compares this risk variable with other risk variables such as beta, inflation
risk, interest rate risk, and economic activity risk with respect to expected returns.
His results show that dispersion of analysts forecast produces the highest correlations
with expected returns with the highest significance. He suggests that companies for
which there is a broad consensus with respect to future earnings and dividends seem
be less risky than companies for which there is little agreement among security
analysts. He concludes that dispersion of FAF is the best single measure of systematic
risk available.
Imhoff and Lobo (1992) use dispersion in analyst forecast as a measure of ex ante
earnings uncertainty, which may reflect either the fundamental uncertainty of a firms
future cash flows or noise in the financial reporting system. They measure dispersion
19
uncertainty reflected in the forecast dispersion originates largely from noise in the
earnings signals and that the greater ex ante earnings uncertainty is a signal of lower
quality of the earnings information.
Barron and Stuerke (1998) construct a forecast dispersion measure from forecasts that
are revised during the first 30 days following announcements of either prior year
annual earnings or current year interim earnings, and calculate it as the standard
deviation of revised forecasts following the earnings announcements divided by the
mean revised forecast. By doing so, they argue, the spurious dispersion caused by
stale forecasts can be controlled, and the forecasts reported are conditioned on a
public announcement. They compile their forecast observations from I/B/E/S Detail
data from 1990 to 1994, and find a positive association between ex ante dispersion
and the magnitude of price reactions around subsequent earnings releases, even after
controlling for other measures of uncertainty like beta and the variance of stock
returns. They postulate that dispersion in FAF serves as a useful indicator of
uncertainty about the price relevant component of firms future earnings.
In summary, dispersion among analyst forecasts is indicated to reflect uncertainty of
the firms future economic performance, though whether such uncertainty originates
from the uncertainty of the underlying future cash flows or the noise in the
accounting information is not resolved.
2.3.2 Analysts forecasts and irregular events
Some studies have related the research on analysts forecasts to certain events outside
the security market to test how the properties of analysts forecasts change with
20
respect to these events. They have drawn inference on the association between
analysts forecast for the firm and the specific event.
Moses (1990) examines differences in FAF properties between failing and healthy
firms and investigates whether measures developed from FAF are useful indicators of
impending bankruptcy. He studies firms that declared bankruptcy from 1977 through
1985 and collects FAF data from I/B/E/S Summary data for these firms for four years
prior to bankruptcy. He then matches each bankrupt firm with a non-bankrupt firm
from the same industry and of approximately the same size resulting in a total sample
of 136 firms. His results show that compared with the healthy firms, the failing firms
have significantly larger error in forecast EPS up to as early as 4 years prior to failure
and significantly greater increase in forecast errors from year 2 to year 1 prior to
bankruptcy. They also have larger forecast dispersion from as early as three years
prior to failure than the healthy firms do. The bankrupt firms have consistently
increasing dispersion in forecasts both within and across years in the three years prior
to failure. These results are consistent with the notion that uncertainty increases as
failure approaches. He concludes that analysts forecasts do reflect conditions that
are associated with failure, and analysts forecasts are of poorer quality for firms
approaching failure.
Dechow, Sloan and Sweeney (1996) examine the forecast dispersion for firms subject
to alleged violations of GAAP according to AAER. They measure forecast dispersion
as the standard deviation of analysts forecast of current-year earnings reported in the
month of the firms fiscal year-ends.
analysts forecasts in the three years prior to the median dispersion of analysts
forecasts in the three years following the year allegation of earnings manipulation is
announced. They find a significant increase in the dispersion of analysts forecasts
for the alleged firms from pre-announcement period to post-announcement period,
but not for the control firms.
revising downward their beliefs about both the firms future economic prospects and
the credibility of the firms financial disclosures once the earnings manipulation is
disclosed.
21
Griffin (2002) examines the response of First Call financial analysts to corrective
restatements and disclosures that lead to securities fraud litigation and measures the
response in terms of forecast coverage and forecast accuracy around a corrective
disclosure. His sample is composed of 731 U.S. exchange-traded firms that are
alleged of fraud in federal class actions with end of class period dates between June
27, 1994 and March 31, 2001. He uses median EPS forecast reported in each month
for the current fiscal year as the forecasted earnings to derive the forecast error. His
results show that the number of analysts covering companies with corrective
disclosures declines significantly in the months after the disclosure, but not in
anticipation. Moreover, the analyst forecast error is essentially unchanged prior to a
corrective disclosure month, decreases significantly in the disclosure month and the
following month, and changes little thereafter. He suggests that financial analysts are
reluctant to follow companies with the bad news of corrective disclosure, and that
financial analysts demonstrate little ability to anticipate such bad news.
However, few studies have provided comprehensive evidence on the information
content of financial analysts forecasts for earnings restatement firms by examining
the FAF prior to the restatement announcements. The analysis on the properties of
analyst forecasts for the earnings restatement firms is even limited in the literature.
22
interested to examine whether the financial analysts (as market intermediaries) have
predictive knowledge of the misstatement and the subsequent earnings restatement,
and whether the market in aggregate has prior wisdom of the restatement.
Furthermore, we compare the distributions of analysts ex ante forecasts for the
restatement firms versus for the non-restatement firms.
address the following four questions:
Specifically, we aim to
subsequently corrected earnings of the restatement firms when the erroneous numbers
were reported? 2). Is there any difference in the distribution of ex ante financial
analysts earnings forecasts (FAF) for restatement and non-restatement firms? 3).
Does the market have aggregate wisdom about the circumstances leading to the
restatement announcement, and moreover, does the market incorporate the
information conveyed through the distribution of ex ante FAF for earnings
restatement firms?
23
Misstated period
To avoid the effect of small denominator, we do not deflate the forecast error by initial price or actual
earning, as some previous studies do (e.g., Easterwood and Nutt 1999). However, we include these
different measures of forecast errors in our robust tests.
24
and
processing
comprehensive
(accounting
and
non-accounting)
information about the company, the industry and the economy in forming earnings
expectations (Fired and Givoly 1982; Brown et al 1987c; Alexander 1995), and they
report that analysts forecasts are superior to nave time-series models in terms of
accuracy. If this is the case, we expect that the financial analysts have predictive
ability of the misstatement so as to detect the true earnings information of the
restatement firms during the misstated period. In other words, the FAF for the
misstated period is closer to the restated earnings than to the misstated earnings, i.e.
FEt FE 't .
In reality however, the relationship between forecast error and financial analysts
ability to collect and process information may be manipulated by management.
Specifically there are two scenarios. The managers may purposely misstate their
earnings to meet the financial analysts earnings forecasts (e.g., Degeorge et al 1999).
They may also intentionally provide earnings guidance to analysts, misleading the
analysts forecasts to the misstated earnings (e.g., Matsumoto 2002). If these two
scenarios dominate, it is possible that the FAF for the misstated period is closer to the
misstated earnings than to the corrected earnings, i.e., FEt FE 't .
Another overlaying reason that may cause the forecast errors measured using restated
earnings numbers to be larger than the forecast errors using the misstated earnings
25
earnings are measured as the misstated earnings than as the restated earnings.
The rejection of this hypothesis is an indication that the analyst forecasts are closer to
the restated earnings numbers and that the financial analysts have predictive ability of
the true earnings information despite the manipulation or misleading guidance from
the management. On the contrary, the failure to reject this hypothesis may result from
the management manipulation or the inefficiency of analyst forecasts as mentioned
earlier.
3.2.2
The properties (e.g., accuracy, efficiency and dispersion) of FAF are shown to be
associated with firms future performance and capture aspects of risk (Givoly and
Lakonishok 1984). As FAF are expected to reflect macroeconomic, industry and firmspecific information, the distribution of FAF may differ systematically across firms
depending on the conditions faced by the firms. In a sense FAF distribution may
reflect information relevant to predicting future events.
The event of earnings restatement provides a special setting to examine the
information conveyed through analysts earnings forecast. The announcement of
earnings restatement itself can bring tremendously negative consequences to the
restatement firms like stock price decline and shareholder litigation. It is a signal of
potential cost to be incurred on the restatement firms, such as litigation cost, SEC
penalty, and monitoring cost.
influence the investors expectations on the firms future cash flows and therefore
26
properties of FAF from three aspects, namely, the size of the forecast error, the
forecast dispersion, and the skewness of FAF distribution.
The earnings forecast studied hereof is the current-year forecast for the annual
earnings of the fiscal year preceding the fiscal year in which the restatement is
announced, and is therefore ex ante to the restatement announcement, as depicted in
Figure 2.
we take the analyst forecasts for annual earnings of fiscal year t-1 as the ex ante
earnings forecast.
Ex ante earnings forecast
t-3
t-2
t-1
time
accounting misstatement in fiscal year t, which means the beginning of the misstated
period is within the same fiscal year of the restatement announcement. This happens
when a firm decides to restate its first three quarters financial results at the fiscal year
end. In this case the firms have not yet began their accounting misconduct at the time
the analysts make forecasts for annual earnings of year t-1 and therefore their
forecasts are not influenced by the release of the erroneous financial statements. We
include the first two categories of restatement firms in our analysis while excluding
the third category of restatement firms. The latter offers too short a time lag between
the misstatement and the restatement for any significant impact on the analysts ex
ante forecasts.
To construct a group of non-restatement firms for future analysis, we match each
restatement firm with a firm from the same industry and of approximately the same
size without a history of earnings restatement. Matching on industry is desirable to
control for industry-specific endogeneity. Forecast uncertainty may be related to
industry and forecast revisions may result from industry-wide information events
(Bhushan 1989). Matching on size is justified by the association size can have with
risk and analyst attention (Fama and French 1992; Bhushan 1989; Imhoff and Lobo
1992).
Forecast error
Forecast error has been shown analytically to be an appropriate indicator of
uncertainty (Cukierman and Givoly 1982; Lang and Lundholm 1993; Ciccone 2003).
The issue of interest here is whether there is difference in forecasts accuracy for
restatement firms versus non-restatement firms. If restatement firms are associated
with conditions characterized by more uncertainty about their future performance and
credibility in financial reporting prior to their restatement announcement than the
non-restatement firms, and this uncertainty is reflected in the forecast error for the
fiscal year prior to the earnings restatement, then we expect the ex ante forecast error
for restatement firms to be larger than that for the non-restatement firms.
28
H 2 : The forecast error of ex ante FAF is larger for the restatement firms than for the
non-restatement firms.
Recall that forecast error was previously measured as6
FE t 1 E t 1 Ft 1 ,
Ft 1 is the mean forecast for year t-1 across forecasters, with year t being the fiscal
year in which the earnings restatement is announced. The above hypothesis expects
that restatement firms have larger FE t 1 than non-restatement firms.
Forecast dispersion
Forecast dispersion measures the cross-sectional variation of analysts earnings
forecasts around the average forecasts. It reflects the divergence of the analysts
beliefs about the firms future economic performance and is often interpreted as an
earnings uncertainty measure. Studies on forecast dispersion find empirical evidence
that forecast dispersion is associated with the firms earnings uncertainty and other
commonly employed risk variables (Givoly and Lakonishok 1984, 1988; Daley 1988;
Swaminathan 1991; Imhoff and Lobo 1992; Barron and Stuerke 1998). Uncertainty
about future earnings stems from two sources: one is the difficulty of predicting
future cash flows, the other is the noise created by the accounting system itself
(Givoly and Laknoshishok 1988). When the financial analysts formulate earnings
forecasts for the restatement firms, they may be influenced by the misstated earnings
information which has been released publicly. Compared with the non-restatement
firms, there is more earnings uncertainty about the restatement firms. Uncertainty
arises because of the firms underlying economic performance as well as the
credibility of their financial reporting. Therefore we hypothesize a higher forecast
dispersion of ex ante FAF for restatement firms than for non-restatement firms.
H 3 : The forecast dispersion of ex ante FAF is larger for restatement firms than for
non-restatement firms.
6
Since the stock price at the time of the forecast may be inflated artificially by the misstatement of
earnings information, deflating the forecast difference by the stock price will be misleading. Yet we
will include the measures of forecast error as the absolute difference between actual and forecasted
earnings, deflated by initial price and absolute value of actual earnings respectively in the robustness
tests.
29
Hypothesis 3 suggests that the restatement firms have a higher DISPt 1 than that of
the non-restatement firms.
This study extends the study of ex ante FAF distribution to its skewness, though few
previous studies have made such effort.8 We try to make a thorough study of the ex
ante FAF distribution by further examining whether there is difference in the
skewness of the distribution of ex ante FAF for restatement firms and non-restatement
firms.
The skewness measures how the distribution of ex ante FAF deviates from symmetry.
A left-skewed distribution indicates that there are more extreme values in the lower
end of the distribution, while the opposite holds for the right-skewed distribution. If
the uncertainty over earnings of restatement firms were to result in more cases of
extreme low earnings forecasts, we would expect the distribution of ex ante FAF for
restatement firms to be more left-skewed than that for the non-restatement firms.
H 4 a : The FAF distribution is more left-skewed for the restatement firms than for the
non-restatement firms.
We develop a measure of skewness of analysts earnings forecast as the follows:
SKWt 1
n
(( Ft 1 Ft 1 ) / S t 1 ) 3 ,
( n 1)(n 2)
We include alternative measures of forecast dispersion such as the standard deviation deflated by the
absolute value of mean forecasts and the price at fiscal year end respectively in the robustness tests.
8
Abarnell and Lehavy (2003) examine the skewness of the cross-sectional distribution of analyst
forecast errors, and their purpose of using the skewness measure is different from ours. They try to use
the statistical inferences of the forecast error distributions to explain the inefficiency of FAF.
30
where Ft 1 is the mean forecast for the annual earnings of year t-1 across n
forecasters, S t 1 is the standard deviation of the analyst forecasts, and n is the
number of forecasters following a specific firm. A negative SKW means that the
distribution of FAF has a longer tail in the negative direction and therefore more
extreme values in the lower end of the distribution. The above hypothesis suggests
that the distribution of ex ante FAF for restatement firms has lower SKW than that for
the non-restatement firms.
3.2.3
31
empirical results support this hypothesis. Kim and Verrecchia (1991) also construct a
theoretical model demonstrating that public announcements that are anticipated
manifest relatively smaller price responses at the time of the disclosure than
announcements that are not anticipated. In the case of earnings restatement
announcement in our study, we suggest that if the market is able to aggregate
information relevant to the subsequent restatement from ex ante FAF properties prior
to the restatement announcement, we would expect the market to react less upon the
announcement of earnings restatement. In other words, the market response is
mollified by the information of subsequent restatement reflected in ex ante
uncertainty of FAF.
H 5 : The higher the uncertainty measures derived from the ex ante FAF properties
are, the less the market responds to the announcement of earnings restatement.
Define CAR (-1, 1) as the cumulative abnormal return over a short three-day window
around the event of restatement. According to our above examination of the three
aspects of FAF for restatement firms, we expect the following directions of
relationship between market response and the specific aspect of FAF in the following
basic regressions:
CAR ( 1,1) 0 1 DISP
expected sign:
1 (-)
1 (+)
Ex ante properties of FAF for restatement firms and their long-term risk
measures
32
would expect a correlation between the ex ante FAF properties and the restatement
firms subsequent risk measured by traditional risk measures.
H 6 : The greater the uncertainty is reflected in the ex ante FAF properties, the higher
consistent with the notion that forecast dispersion measures uncertainty. We extend
our study to examine whether the uncertainty reflected in the ex ante earnings
forecast can indicate subsequent risk using the traditional risk measures, i.e., the
return variance and beta.
Wu (2002) reports that the negative cumulative abnormal returns for the restatement
firms continue for as long as one year after the announcement of restatement.
Accordingly we extend the horizon of our study to one year after the announcement
of restatement. If the ex ante uncertainty reflected in the FAF distribution measures
the ex post risk of the restatement firms, we expect the following signs of correlation
between the following variables of interest:
DISP & BETA (+), DISP & RETVAR (+), SKW & BETA (-), SKW & RETVAR (-)
The beta and return variance are calculated using the daily stock returns from the first
day to the 250th day after the restatement announcement. Results contrary to the
predicted signs may indicate that the uncertainty reflected in ex ante FAF result more
from divergence of analysts beliefs about the restatement firms financial reporting
quality.
33
misstated period earnings as the forecasted earnings for the misstated period, as it has
been documented that the most current forecast available from an analyst dominates
34
both the mean and median forecast in absolute error terms. We also collect the latest
forecast from each analyst reported before the earnings announcement of year t-1 to
form a set of the ex ante forecasts. To calculate the standard deviation and skewness
of the ex ante forecasts we require that each firm be followed by at least four analysts
for the year t-1.
Daily stock price data are collected from CRSP. Set the restatement announcement
date as event date 0. We collect daily price for both restatement and non-restatement
firms from day -120 to 250 for further analysis.
4.3 Method
4.3.1 Analysts forecasts for the misstated period
Measure of forecast error for the misstated period using misstated earnings
For a restatement firm that restates its quarterly earnings, the forecast error of each
misstated quarter j is:
FE j E j F j
quarter j, and F j is the most current forecast available from analyst before interim
earnings announcement of quarter j. The forecast error of the misstated quarters in
FE j .
aggregate is calculated as: FE
j
For a restatement firm that restates its earnings for one or more years, the forecast
error of the misstated year
earnings for year
is:
FE E F
FE .
period for a restatement firm is thus: FE
Measures of forecast error for the misstated period based on restated earnings
35
We use the similar way to construct the forecast error measures for restatement firms
using the restated earnings. This forecast error is expressed as FE. As the misstated
and restated earnings are collected from the news reports and rechecked with the SEC
filings, potential problem of data inconsistency still exists when we calculate forecast
error using forecast earnings and actual earnings from different data sources. We are
unable to reconcile this problem by obtaining all the information from one source,
because the earnings provided by I/B/E/S are the misstated earnings but not the
restated ones. Yet we have made best efforts to adjust the forecast earnings and actual
earnings to the same basis. As the forecast EPS from I/B/E/S is the net income from
continuing operations divided by the weighted average number of shares outstanding
for the year, we accordingly evaluate firms with the misstated and restated earnings
from net income from continuing operations. This means that we eliminate the
restatement cases that restate earnings arising outside of continuing operations.
Furthermore, we impose a censoring rule developed by OBrien (1988): forecast
errors larger than $10.00 per share are deleted from the sample. In this way we try to
minimize the problems of data inconsistency in testing our first hypothesis.
There are different forms of forecast error measures in the analyst forecast literature,
including the absolute difference between the forecast and the actual earnings with no
deflation, deflated by the actual earnings or deflated by the initial price. We keep the
measure unscaled to avoid small denominator problem, but will include the other two
measures in the robustness tests.
4.3.2 Ex ante analyst forecasts
Measures of forecasted earnings for the year prior to the year of restatement
To examine the properties as well as the distribution of ex ante FAF for restatement
firms, we collect the most recent estimate from each analyst before the earnings
announcement of fiscal year t-1, the year prior to the restatement announcement year
t. As analyzed in Chapter 3, we exclude the cases of current-year interim earnings
restatement in which the misstated period falls in the same fiscal year of restatement
announcement. In other words, if the misstated period begins after the earnings
36
announcement of fiscal year t-1, we exclude such restatement. In these cases the
properties of ex ante FAF may not yet reflect the influence of earnings manipulation
on the uncertainty faced by the restatement firms.
sample, we take the mean of the most recent forecasts from each analyst as the
forecast earnings to calculate the forecast error for each firm. Our robustness tests
will include the median as a measure of forecast earnings. Our measures of forecast
error and forecast dispersion in the main study are not scaled by initial price and
actual earnings or mean earnings forecast, but we will include them in the robustness
tests as well.
E t 1 Ft 1
(F
t 1,i
Ft 1,i ) 2 / n 1 , i=1, 2, n
Ft 1,i
across n individual analysts for year t-1, n is the number of analysts following the
firm.
3. Skewness of ex ante FAF: SKWt 1
Ft 1,i
and
Ft 1,i
n
(( Ft 1,i Ft 1,i ) / S t 1 ) 3 ,
(n 1)(n 2) i
37
where
Rit = return for firm i on day t from the CRSP daily data,
using the daily returns over the year beginning one year before the beginning of the
misstated period and ending on the day before the misstated period. The magnitude of
abnormal returns is then computed as
CAR [ Rit ( i i Rmt )] , where t is the event window that we examine.
t
Multiple regressions
Besides using univariate tests to examine the relationship between market response
and the uncertainty measures derived from ex ante FAF properties, we also develop
multiple regressions by adding independent variables of information about the
restatement released in the restatement announcement, to control for the potential
influence of the restatement-related information.
38
variable is the market response to the restatement announcement CAR (-1, 1). The
independent variables include the specific measure of uncertainty derived from ex
ante FAF properties (DISP and SKW), the magnitude of the restatement, length of
restatement period, the firm size, the reason for restatement, case of concurrent
earnings announcements and the markets prior knowledge of restatement ((CAR (120, -1)). The regression is expressed as the follows:
CAR (1,1) 0 1 FAFPROP 2 AMOUNT 3 QRTS 4 SIZE
5 FRAUD 6 CONEA
where
CAR(-1,1)= the cumulative abnormal returns over three days around the restatement
announcement;
FAFPROP= DISP and SKW respectively;
AMOUNT=the magnitude of the restatement scaled by the firms market
capitalization six months earlier, the magnitude of the restatement is
calculated by subtracting originally reported net income from restated net
income;
QRTS= the number of quarters being restated; if a whole fiscal year has been restated,
the number of restated quarters is four;
SIZE= firm size, the log form of the firms market capitalization at the end of fiscal
year t-1
FRAUD= 1 when the restatement firm explicitly admitted the existence of fraudulent
practices or irregularities in the restatement announcement;
CONEA=1 if there is a concurrent earnings announcement within the event window
(-1, 1);
Supplementary multiple regressions
In addition to treating the uncertainty measures derived from ex ante FAF properties
as continuous variables, we employ supplementary multiple regressions treating the
uncertainty measures from ex ante FAF properties as a classificatory variable. First
39
we rank the restatement firms each year based on the dispersion and skewness
measure of the ex ante FAF distribution. Then the restatement firms are grouped into
three equal-size categories based on the ranked values, the LOW, MEDIUM and
HIGH category. The supplementary test employs the following multiple regressions
using cross-sectional firm-specific data:
CAR ( 1,1) 1 2 D M 3 D H 1 AMOUNT 2 QRTS 3 MKTCAPTi ,t
where
DM = a dummy variable equals to 1 when a restatement firm is in the MEDIUM
portfolio in year t-1 based on the uncertainty measure derived from ex ante
FAF properties (DISP and SKW respectively), and 0 otherwise;
D H = a dummy variable equals to 1 when a restatement firm is in the HIGH portfolio
in year t-1 based on the uncertainty measure derived from ex ante FAF
properties (DISP and SKW respectively), and 0 otherwise;
and all other variables are the same as defined earlier.
In support of Hypothesis 5, we test the following:
H 01 : 2 3 0,
H a1 : 2 0; 3 0, and
H 0 2 : 2 3 0,
H a2 : 3 2 .
The rejection of the above two hypotheses will be consistent with Hypothesis 5. In
other words, we expect the market to respond less to the firms with high and medium
uncertainty indicated by ex ante FAF properties than to those with low uncertainty,
ceteris paribus. Moreover, we expect the market to respond less to the firms with
high uncertainty indicated by ex ante FAF properties than to those with medium
uncertainty, ceteris paribus.
4.3.4 Research question 4
Univariate test
40
We employ simple univariate test to examine the association between the uncertainty
level reflected in ex ante properties of FAF for restatement firms and the firms
subsequent levels of risk.
The correlation of interest is that between DISP and BETA, DISP and RETVAR, SKW
and BETA, and SKW and RETVAR, where
BETA= the firms beta estimated over the period beginning from the earnings
restatement date and ending with day 250 after the earnings restatement
RETVAR= the variance of daily raw returns over the period from the earnings
restatement date and ending with day 250 after the earnings restatement
4.3.5 Robustness tests:
In robustness tests, the various tests are repeated using different measures of forecast
earnings, forecast error and forecast dispersion.
forecast for the mean forecast as a measure of forecast earnings in calculating the
forecast error. Secondly, we deflate the forecast error with both the absolute value of
the actual earnings (either misstated or restated in the two tests) and the price at the
end of fiscal year t-1. Thirdly, we deflate the forecast dispersion with both the
absolute value of mean earnings forecast and the price at the end of fiscal year t-1.
We evaluate whether our results are sensitive to the measure of forecast earnings,
forecast error and forecast dispersion.
41