Journal of accounting, auditing finance tập 27, số 01, 2012 1

Earnings Revisions in SEC
Filings From Prior Preliminary
Announcements

Journal of Accounting,
Auditing & Finance
27(1) 3–31
Ó The Author(s) 2012
Reprints and permission:
sagepub.com/journalsPermissions.nav
DOI: 10.1177/0148558X11409006


Dana Hollie1, Joshua Livnat2, and Benjamin Segal3

Abstract
This article investigates earnings revisions that occur between preliminary earnings announcements and the immediate subsequent Securities and Exchange Commission (SEC) filings. On
average, the absolute value of the revision is 2.9% of the market value of equity where earnings were revised by more than US$100,000. The authors find that earnings revisions are
more likely to occur for firms that are more complex, are more financially leveraged, have
greater earnings volatility, have losses, and have switched auditors. They find that investors
react to the new information in the earnings revisions but find mixed evidence about whether
the act of revision itself indicates lower earnings quality to investors. The authors’ findings
suggest that financial analysts, investors, and regulators alike should pay close attention not
only to an earnings surprise at the preliminary earnings announcement date but also at the
SEC filing date to determine whether a subsequent earnings surprise occurs.
Keywords
market efficiency, asset pricing, earnings announcement, SEC filings, earnings revisions
Regulators and prior research have shown that investors have suffered significant losses as
market capitalizations have dropped by billions of dollars due to restatements of audited
financial statements (Levitt, 2000; Palmrose, Richardson, & Scholz, 2004). Less discernable restatements occur between the preliminary earnings announcement and the immediately subsequent Securities and Exchange Commission (SEC) filing (henceforth called

earnings revisions); however, very little is known in the academic literature about these
revisions. This study first examines the characteristics of firms most prone to earnings revisions and the type of income statement components that are most likely to be affected.
We then assess the market’s response to the additional earnings surprise (in the SEC filing)
and the total earnings surprise (i.e., the preliminary earnings surprise plus the additional
earnings surprise at the time of the SEC filing) to determine whether the market’s reactions
to the mere act of earnings revision is negative, irrespective of its content.
1

Louisiana State University, Baton Rouge, LA, USA
New York University, New York City, NY, USA
3
Boulevard de Constance, Fontainbleau, France
2

Corresponding Author:
Dana Hollie, E. J. Ourso College of Business, Louisiana State University, Baton Rouge, LA 70803
Email:

Downloaded from jaf.sagepub.com at Taylor's University on December 2, 2012


4

Journal of Accounting, Auditing & Finance

Typically, companies issue a press release to report preliminary earnings to the market
26 days after the fiscal quarter-end for the median company, with 1% of firms reporting in
less than 9 days.1 Subsequently, firms file a 10-Q/K Form with the SEC usually within the
last 2 days of the mandated filing period.2 A small but nontrivial portion of firms actually
revises its earnings from the preliminary earnings announcement to the subsequent SEC

filing for a variety of reasons. For example, a firm’s auditor may require the filing of different earnings figures with the SEC than those previously released to investors. Subsequent
information revealed after the preliminary announcement may also cause firms to modify
their earnings before the SEC filing. Although an infrequent phenomenon (1.98% of our
sample observations), some of these revisions are quite large. On average, the absolute
value of the revision is 2.9% of the market value of equity for our sample of 3,337 firm
revisions where earnings were revised by more than US$100,000.3 More than 5% of the
earnings revisions are greater in absolute value than 6.2% of the market value of equity at
quarter-end (mean market value for sample firms is US$4.864 billion). Although these earnings revisions are economically significant at the time of their revision, unlike typical earnings restatements, virtually none of the earnings revisions are preceded by a formal
announcement or a press release, nor are companies required to file an amendment such as
10-K/A or 10-Q/A. Furthermore, in many cases, firms do not even explain the reasons for
the earnings revisions in their SEC filings. Still, the phenomenon of earnings revisions is
important for academic research and practitioners because it indicates cases of potential
problems with earnings quality that have not previously been documented in the academic
literature. Furthermore, a study by Hollie, Livnat, and Segal (2005) shows that market participants seem to react adequately to the new earnings surprise contained in SEC filings of
firms that made earnings revisions. However, they do not study, as we do here, whether the
total market reaction to both sources of earnings surprise indicate that market participants
seem to treat firms with earnings revisions as if they have an inferior quality of earnings.
In our sample, approximately 64% of all significant earnings revisions are income
decreasing, and a large portion of the revisions (about 50%) occur in the fourth fiscal
quarter, consistent with both audit work at year-end and a longer period before SEC filings in which subsequent events may require an earnings revision. We find that the most
frequently revised components of net income are recurring items (e.g., sales; cost of
sales; Selling, General, and Administration [SG&A]; and depreciation) and that earnings
revisions are more likely to occur for firms that are more complex in nature, are more
financially leveraged, have greater earnings volatility, have losses, and have switched
auditors. We also find that the likelihood of an earnings revision is inversely related to
the persistence of earnings changes. These variables are also shown in the literature to be
associated with lower earnings quality and subsequent earnings restatements. However,
contrary to expectations, we do not find that significant earnings revisions occur in highlitigation risk industries. Most of the firms in our sample (about 68%) appear only once,
indicating that most sample firms are probably not attempting to manage earnings information strategically across their preliminary earnings announcements and SEC filings in
the long run.4

In regression results, we also find that market reactions to the total earnings surprise for
firms that revise their earnings (i.e., the associations of abnormal returns from a day before
the preliminary earnings announcement through a day after the SEC filing with the combined preliminary earnings surprise and SEC filing earnings surprise) do not differ significantly from those of non-Revisers. However, in matched-sample tests, we mostly find
evidence consistent with weaker market reaction to total earnings of Revisers than those of

Downloaded from jaf.sagepub.com at Taylor's University on December 2, 2012


Hollie et al.

5

non-Revisers. Thus, we cannot unambiguously conclude that Revisers are associated with
lower earnings quality.
Our empirical findings are particularly relevant to academics, financial analysts, regulators, and investors. The results of this study are important to academic researchers because it
focuses on cases of potential lower earnings quality, which have not been studied in the academic literature. We do find inconclusive results regarding whether firms with earnings revisions have lower earnings quality, as indicated by weaker market reactions to the total
earnings surprise. These cases may be of particular interest to the auditing profession, where
the identification of firms with a higher audit risk is extremely valuable. The evidence in this
study also suggests that financial analysts, investors, and regulators alike should pay close
attention not only to an earnings surprise at the preliminary earnings announcement date but
also at the SEC filing date to determine whether a subsequent earnings surprise occurs.
The remainder of this article is organized as follows. The section titled ‘‘Background,
Hypotheses, and Research Design’’ reviews the related literature and outlines our hypotheses and methodology. The section titled ‘‘Data and Results’’ describes the sample and presents the empirical findings. The section titled ‘‘Summary and Conclusions’’ summarizes
and concludes the article.

Background, Hypotheses, and Research Design
Most firms disclose their preliminary earnings for the quarter or year through a press
release, following it with an SEC filing several weeks later. Easton and Zmijewski (1993)
report a median lag between the balance sheet date and the preliminary earnings announcement of 28 days and a median SEC filing lag of 45 days for Forms 10-Q. Our sample
shows a similar pattern with a median preliminary earnings lag of about 25 to 26 days in

recent years. Some firms issue a press release to discuss earnings after their SEC filings
(Stice, 1991), and others do not issue any press release at all, relying on the information
available in the SEC filings alone (Amir & Livnat, 2006). When firms issue their preliminary earnings release prior to the SEC filing, investors implicitly assume that these earnings
will be identical to the SEC-filed earnings. However, as this study shows, a nontrivial portion of firms file significantly different earnings figures with the SEC than the one previously provided in their preliminary earnings announcement.5
Consider the following two examples, which are highlighted in Appendix A. R&B, Inc.
(NASDAQ: RBIN) reported net income of US$2.8 million in its second quarter preliminary
earnings release on July 31, 2002, but revised it upward to US$4.115 million on August 8,
2002, in its SEC filing, an increase of almost 150% from its preliminary earnings. An
examination of news items regarding R&B, Inc. (using LexisNexis) reveals that no press
releases were issued between the preliminary earnings announcement and the SEC filing
date. Hence, no public information seems to have been available between the earnings
announcement and the SEC filing date that would have indicated that an upward revision
in SEC-filed earnings was forthcoming. MEMC Electronic Materials, Inc. (NYSE: WFR)
issued its second quarter preliminary earnings release on August 10, 2001, reporting an
US$88.1 million loss, followed by an SEC filing (August 13, 2001) reporting a US$355.3
million loss, reducing the previous earnings figure by approximately 400% just days later.
There are no other news reports in LexisNexis between the preliminary earnings announcement and the SEC filing date to suggest that a downward revision was forthcoming. As
these two examples illustrate, some earnings revisions between the preliminary earnings

Downloaded from jaf.sagepub.com at Taylor's University on December 2, 2012


6

Journal of Accounting, Auditing & Finance

release and the SEC filing date are upward revisions, whereas others are downward revisions and can be quite large in magnitude.6
Plausible explanations (not necessarily specific to the two examples above) for earnings
revisions between the preliminary announcement and the SEC filing are the uncovering of
issues not known at the preliminary earnings release date due to audit work or new auditors

who are more likely to compel clients to revise previously issued earnings under the prior
auditor, information that becomes known subsequent to the preliminary earnings announcement, or discovery of accounting errors before the filing date. For example, several studies
document the existence of accounting errors that are discovered by auditors and corrected
before the public release of year-end financial statements.7 This is also consistent with
about 50% of our firm-quarter observations existing in the fourth quarter, when a full audit
is required (see Panel A of Table 2), and where the window between the preliminary earnings announcement and the SEC filing is typically longer, leading to greater opportunities
for discovering material subsequent to information that requires earnings revisions.

Characteristics of Firms With Earnings Revisions
Absent a theoretical model to guide the selection of potential variables that are associated
with the likelihood of earnings revisions in SEC filings, we use variables that may indicate
inferior earnings quality.
DeFond and Hung (2003) suggest earnings volatility and accruals magnitude as incentives to disseminate cash flow forecasts when earnings are of inferior quality. Chen,
DeFond, and Park (2002) claim that firms with losses and firms with greater earnings surprises are more likely to provide balance sheet information in their preliminary earnings to
strengthen their weaker earnings quality. We use these variables, as well as the persistence
of earnings surprises, to show that firms with a lower quality of earnings are more likely to
revise their earnings in the SEC filings.
Consistent with subsequent corrections during audit work (or review) between the preliminary earnings announcement and the SEC filing, we conjecture that the firm’s complexity
(measured by the number of its operating segments), size, number of analysts following, and
auditor changes are all positively associated with the likelihood of an earnings revision.
However, it may be argued that larger firms with a greater analyst following are managed
more carefully and are less likely to have earnings revisions.8 We also expect that less profitable firms (lower return on assets) are more likely to have earnings revisions because managers in such firms have greater incentives to manage preliminary earnings and because lower
profitability may indicate operational problems and potentially weaker accounting controls.
Studies of restatements in the literature (e.g., Richardson, Tuna, & Wu, 2002—annual
restatements—and Livnat & Tan, 2004—quarterly restatements) assert that financial leverage is positively associated with the likelihood of restatements due to management’s desire
to inflate earnings initially as a way to avoid debt restrictions. The literature also asserts that
growth firms are more likely to have earnings restatements because of their desire to show
continued earnings growth and ability to beat analyst expectations. Firms with high financial
leverage may also be subject to additional scrutiny by auditors and creditors, leading to
greater chances for earnings revisions. Thus, we expect financial leverage to be positively

associated with the likelihood of earnings revisions and the earnings-to-price ratio (an indicator of growth) to be negatively associated with the likelihood of earnings revisions.

Downloaded from jaf.sagepub.com at Taylor's University on December 2, 2012


Hollie et al.

7

Market Reactions to Preliminary Earnings Announcements and SEC Filing Dates
Prior research shows that the market responds to earnings surprises included in preliminary
earnings announcements and that stock prices incorporate this information instantaneously
(see Kothari, 2001, and Lev, 1989, for summaries of these studies). However, there is substantially less agreement about the incremental information content in 10-Q/Ks beyond preliminary earnings announcements. For example, Foster and Vickery (1978), as well as
Wilson (1987), document that 10-Ks have information content beyond earnings announcements, whereas other studies suggest that the market fails to react to earnings information
contained in SEC filings (e.g., Chung, Jacob, & Tang, 2003; Cready & Mynatt, 1991;
Easton & Zmijewski, 1993; Foster, Jenkins, & Vickery, 1983, 1986; Stice, 1991).
Easton and Zmijewski (1993) examine whether the 10-Q/K filing dates are associated
with abnormal returns, using squared market model prediction errors to avoid any predictions
about the direction of expected returns around the SEC filing dates. Their results show abnormal market returns that are significantly different from zero around preliminary earnings
announcements but show no significant difference for market reactions to SEC filings, except
in those cases where only the 10-Q dates are known but no preliminary earnings announcement dates are available on the Quarterly Compustat File. These results seem to imply that
SEC filings contain no incremental information beyond preliminary earnings announcements.
Stice (1991) examines whether the information content of an earnings announcement
can be affected by the method in which earnings are announced, concentrating on firms
that file their 10-Q/Ks several days before the earnings announcement. Stice finds that SEC
filings are not fully reflected in prices until subsequent earnings announcements are made.9
Chung et al. (2003) corroborate Stice’s findings and show that some firms in their sample
behave as if they manage earnings.
Balsam, Bartov, and Marquardt (2002) investigate whether investors in firms that are
suspected of engaging in earnings management are able to more rapidly incorporate the

information about accruals available in 10-Q filings and whether institutional investors
seem to respond even before the SEC filing dates. They find evidence consistent with no
investor reaction to the managed accruals around the preliminary earnings announcement
(with event windows up to 9 days later), with market reactions to discretionary accruals by
firms with at least 40% institutional investors during the window spanning 10 days after
the preliminary earnings announcement through 2 days before the SEC filing date, and
with market reactions to discretionary accruals in the window from a day prior to the SEC
filing date through 15 days afterward for firms with fewer institutional investors. Their
interpretation is that institutional investors seem to find the information necessary to
reverse accruals faster than other investors and prior to the SEC filing dates.
Qi, Wu, and Haw (2000) suggest that prior research’s inability to detect little, if any,
information content around the SEC filing date may be due to the SEC paper filing system
in place at the time of prior studies. Their study compares SEC paper filings with SEC
electronic filings to test whether the information content of 10-Ks has changed because of
electronically available SEC filings. In contrast to most of the prior research, Qi et al. provide evidence that 10-K filings through the Electronic Data Gathering, Analysis, and
Retrieval (EDGAR) system provide incremental information content that did not exist for
the paper filings. However, they study the years 1993 to 1995, in which the EDGAR
system was still voluntary (becoming mandatory in May 1996). In addition, their study is
limited to firms with available Association for Investment Management and Research analyst rankings.

Downloaded from jaf.sagepub.com at Taylor's University on December 2, 2012


8

Journal of Accounting, Auditing & Finance

In a recent study covering the period 1996 to 2001, Griffin (2003) finds that SEC filings
have abnormal market returns that are significantly different from zero, where abnormal
returns represent the absolute value of excess returns around the filing date compared with

the excess returns in a prior period. He finds greater market reactions to 10-Q/Ks, to
smaller firms, to firms with lower proportions of institutional holders, to firms that report
on days with many filings by other firms, and to firms with delayed filings. In addition, in
multivariate results, Griffin finds evidence of stronger market reactions to filings made in
recent periods and to delayed SEC filings.
Asthana, Balsam, and Sankaraguruswamy (2004) show that small trades increase in the
5-day period around the 10-K filing after EDGAR as compared with the pre-EDGAR
period, but not large trades, implying that small investors are better able to use the information in SEC filings in the post-EDGAR period. They also show that small investors seem to
incorporate better the information content of the 10-Ks in the post-EDGAR period than
before and provide evidence consistent with an erosion of the information advantage that
larger traders have as compared with small traders in the post-EDGAR period.
Callen, Livnat, and Segal (2006) show that earnings news is important in explaining the
volatility of unexpected returns around both the preliminary earnings announcements and
the SEC filing dates, with the cash flows and accruals components of earnings significantly
associated with the volatility of unexpected returns around the SEC filing dates.
Hollie et al. (2005) show that firms with significant earnings revisions in SEC filings
experience significant market reactions to the new earnings surprise in the SEC filings.
They also show that market reactions to a dollar of earnings surprise in the preliminary
earnings announcement are not statistically different from the market reactions to a dollar
of earnings surprise in the SEC filing for their sample with material earnings revisions.
Their study, although important in highlighting that market participants are able to identify
and react to the new earnings surprises in SEC filings, does not investigate whether the
total market reaction to the total earnings surprise (preliminary and revision) is weaker due
to the potential lower earnings quality.

Research Design
Figure 1 provides the time line of events in this study and highlights the range of periods
over which abnormal returns are calculated. This figure identifies the short windows
around the preliminary earnings announcement and SEC filing dates, the long window
between the preliminary earnings announcement and SEC filing, and the long window from

1 day before the preliminary earnings announcement through 1 day after the SEC filing date.
For some of our tests, we examine differences between two samples: a sample of firms with
earnings revisions (Revisers) and a sample of control firms from the same Fama and French
(1997) industry, closest in size (market value of equity) and in the same quarter (nonRevisers). For other tests, we use the entire population of non-Revisers.
We first assess the characteristics of firms that have earnings revisions between preliminary earnings announcements and SEC filings. We use a logistic regression model to determine the likelihood that a firm is a Reviser.
REVISERit 5b0 1b1 EARNVOLit 1b2 PERSEit 1b3 DEBTit 1b4 AUCit 1b5 SEGNUMit
1b6 LOSSit 1b7 ROAit 1b8 EPit 1b9 ACCPROPit 1eit

Downloaded from jaf.sagepub.com at Taylor's University on December 2, 2012

ð1Þ


Hollie et al.

9

where REVISER is a dummy variable equal to one if a firm-quarter has an earnings revision and zero otherwise. The variables we use to discriminate Revisers and control firms
are as follows: (a) Earnings volatility (EARNVOL) is the absolute value of the ratio of the
standard deviation of Earnings per share (EPS)/Price over the most recent 12 months to the
average EPS/Price over the same period. It is winsorized to a maximum of 5; (b)
Persistence of earnings changes (PERSE) is estimated as the first autocorrelation between
scaled earnings surprises in the prior eight quarters. The earnings surprise is earnings in the
quarter minus earnings of the same quarter in the preceding year, scaled by market value at
the beginning of the quarter; (c) Debt/Assets (DEBT) is estimated as short-term plus longterm debt divided by total assets at the end of the quarter; (d) Auditor change (AUC) is a
dichotomous variable obtaining one if the firm’s auditor changed from the prior year
(mergers of audit firms such as Coopers and Lybrand and Price Waterhouse are not counted
as an auditor change); (e) The number of segments (SEGNUM) is from the Compustat
Segment file and is a surrogate for operating complexity; (f) ROA is the ratio of earnings
to total assets at quarter-end and proxies for firm profitability; (g) LOSS is a dummy variable obtaining one if earnings for the quarter are negative and zero otherwise; (h) EP is the

current earnings to market value at quarter-end; and (i) ACCPROP is the proportion of
accruals divided by sales over the prior four quarters.

Market Reactions to Earnings Revisions
As portrayed in Figure 1, this study focuses on four buy-and-hold abnormal returns
windows—the preliminary earnings announcement date (CARprelim), the SEC filing date
(CARfile), the window between the preliminary earnings announcement and the SEC filing
date (CARaf), and the window between a day before the preliminary earnings announcement and the day after the SEC filing date (CARtotal). The abnormal return is calculated as
the raw buy-and-hold return from the Center for Research in Security Prices (CRSP) minus
the buy-and-hold return on the portfolio of firms with the same size (the market value of
CARtotal
CARaf
CARfile

CARprelim
–1

–1

0 +1

Preliminary Earnings
Announcement Date

0

+1

SEC
Filing Date

Figure 1. Time line: Preliminary earnings announcements and SEC filings
Note: CARprelim = the buy-and-hold abnormal return from Day 21 through Day 11, where Day 0 is
the preliminary earnings announcement date. The abnormal return is the raw return minus the average return on a same size, book-to-market (B/M) portfolio (six portfolios), as provided by professor
French; CARaf = the buy-and-hold abnormal return between preliminary earnings announcements and
SEC filings. It controls for information released to the stock market between the two time periods;
CARfile = the buy-and-hold abnormal return from Day 21 through Day 11, where Day 0 is the SEC
filing date of 10-Q/10-K; CARtotal = the buy-and-hold abnormal return from 1 day before the preliminary earnings announcement through 1 day after the SEC filing date.

Downloaded from jaf.sagepub.com at Taylor's University on December 2, 2012


10

Journal of Accounting, Auditing & Finance

equity) and B/M ratio. The daily returns (and cutoff points) on the size and B/M portfolios
are obtained from Professor Kenneth French’s data library, based on classification of the
population into six (two size and three B/M) portfolios.10 We omit all observations with
abnormal returns in the top and bottom 0.5% of the population.
We then examine the stock returns around the preliminary earnings announcements to
provide initial evidence on the effect of earnings surprises in preliminary earnings
announcements. Consistent with prior research (e.g., Kothari, 2001; Lev, 1989; Penman,
1987), we expect a positive and significant relation between earnings announcements and
stock market returns. This analysis is intended only to show the consistency of our sample
with the results in prior studies. For this test, we examine the relation between the first
earnings surprise (FSURP) and CARprelim using the following regression:
CARprelim it 5b0 1b1 FSURPit 1eit

ð2Þ

where CARprelim is the 3-day (21 to 11) buy-and-hold abnormal returns centered on the
preliminary earnings announcement date (Date 0) and FSURP is the preliminary (or first)
earnings surprise, calculated as preliminary Compustat earnings at Quarter t minus
Compustat earnings at Quarter t 2 4, scaled by market value of equity at the end of
Quarter t. We expect a positive coefficient on b1, which represents the earnings response
coefficient (ERC) for the preliminary earnings surprise.
We then examine the stock market reaction to the additional earnings surprise in SEC
filings. We measure the additional earnings surprise (ASURP) as the SEC-filed earnings
(As-First-Reported [AFR] Compustat earnings at Quarter t) minus the preliminary
Compustat earnings, scaled by market value at quarter-end. In a similar manner to
CARprelim, we define CARfile as the 3-day (21 to 11) buy-and-hold abnormal returns centered on the SEC filing date (Date 0). To control for additional news possibly obtained by
market participants between the preliminary earnings announcement and the SEC filing
date, we include CARaf. CARaf is the buy-and-hold abnormal return from 2 days after the
preliminary earnings announcement through 2 days before the SEC filing date. It is
assumed that changes in stock prices capture all news during this event period.
Accordingly, we investigate the relationship between ASURP (additional surprise), FSURP,
CARaf, and CARfile using the following regression model:
CARfile it 5b0 1b1 FSURPit 1b2 ASURPit 1b3 CARaf it 1eit

ð3Þ

where ASURP is defined above. Our equation for total earnings surprise (TSURP) for nonRevisers and Revisers is provided below. Because non-Revisers do not have an additional
earnings surprise, ASURP, by definition, will always be equal to zero for the non-Reviser
control group of firms.
For Revisers and for non-Revisers:
Total earnings surprise5FSURP ðfirst surpriseÞ1ASURP ðadditional surpriseÞ
5½Preliminary Compustat earnings at Quarter t minus AFR
Compustat earnings at Quarter t À 4; scaled by market value of
equity at the end of Quarter tŠ1½AFR Compustat Earnings in the

SEC filing À Preliminary Compustat earningsŠ

Downloaded from jaf.sagepub.com at Taylor's University on December 2, 2012


Hollie et al.

11

The coefficient b1 captures the market reaction at the time of the SEC filing to the
already-known initial earnings surprise in preliminary earnings. We expect b1 to be insignificantly different from zero if stock prices have already fully impounded the earnings
information during the preliminary earnings announcement window. The coefficient on b2
represents the market reaction to the additional earnings surprise contained in the SEC
filing beyond earnings reported in the preliminary earnings announcement. It is expected to
be positive and statistically different from zero if the SEC filings are noticed by investors
who react to the additional earnings surprise, as is documented in Hollie et al. (2005). We
have no expectation about the sign or magnitude of the coefficient on b3 as CARaf is a control variable for information released to the market between the two event dates. This analysis is intended to show that our sample produces similar results to those of Hollie et al.
Next, we estimate the market reactions to the total earnings surprises of Revisers as
compared with those of control firms through a system of simultaneous equations.
Essentially, this method consists of the following steps: a first stage regression to get fitted
values for the probability that the firm is a Reviser and a second stage to run the regressions using these fitted values. The two endogenous variables are therefore REVISER and
CARtotal. The exogenous variables (i.e., ROA, DEBT, SEGNUM, EARNVOL, PERSE,
AUC, and LOSS), identified in the logistic analysis previously, are used as instrumental
variables to estimate whether Reviser equals ‘‘1’’ or ‘‘0.’’ In a similar manner to CARfile,
we define CARtotal as the buy-and-hold abnormal returns from 1 day before the earnings
announcement date through 1 day after the SEC filing date.
CARtotal it 5b0 1b1 REVISERit 1b2 TSURPit 1b3 REVISERit 3TSURPit 1eit

ð4Þ

CARtotal it 5b0 1b1 REVISERit 1b2 TSURPit 1b3 REVISERit 3TSURPit 1b4 ROAit
1b5 DEBTit 1b6 SEGNUMit 1b7 EARNVOLit 1b8 PERSEit 1b9 AUCit
1b10 LOSSit 1eit
ð5Þ
where TSURP is the sum of ASURP and FSURP as defined above. We assign each firmquarter into 10 deciles according to their TSURP. We then substitute the deciles rank for
TSURP, divide it by 9 and subtract 0.5, as is often done in the postearnings-announcement
drift literature. The coefficient on TSURP measures the return on a hedge portfolio that
holds long (short) positions in the top (bottom) decile of earnings surprises.11 To the extent
that the market views an earnings revision in a negative light, likely due to weaknesses in
accounting or control systems or perceived lower earnings quality, we expect the coefficient on REVISERit 3 TSURPit (b3) to be negative and statistically different from zero.
This would indicate that a dollar of earnings surprise in the case of Revisers contributes
less to the market reaction than a dollar of earnings surprise of non-Revisers.
We further test whether the total market reactions to the total earnings surprises for
Revisers are different from those of the matched control firms by comparing the returns of
Revisers and matched control firms in the same quarter. The matching is done by the total
surprise decile and size (market value of equity); total surprise decile and industry; total
surprise decile, size, and industry; and total surprise decile, size, and number of segments.
We also match the firm to itself in another quarter with no earnings revision but when the
total earnings surprise was in the same decile rank. If the market reacts negatively to the
act of earnings revisions, we should observe lower total market reactions for Revisers than
for the matched control firms.

Downloaded from jaf.sagepub.com at Taylor's University on December 2, 2012


12

Journal of Accounting, Auditing & Finance

Finally, we examine whether these earnings revisions are possibly a result of strategic

behavior by management. In this study, we examine two specific strategic behaviors, which
are (a) whether a firm initially meets or beats earnings forecasts but misses the forecast
after the earnings revisions and (b) whether it appears that the firm has initially taken a big
bath falling below its forecast but subsequently revised earnings upward above the forecast.
We use the previously described approaches to evaluate these noted behaviors.

Data and Results
Original Compustat Quarterly Data
Compustat’s database data entry procedure is as follows: When a firm releases its preliminary
earnings announcement, Compustat takes as many line items as possible from the preliminary
announcement and enters them into the quarterly database. The preliminary earnings data are
denoted by a code of 2, until the firm files its Form 10-Q/K with the SEC or releases it to
the public, at which point Compustat updates all available information with an update code
of 3. The Compustat quarterly database is further updated when a firm restates its previously
reported earnings. For example, if a firm engages in mergers, acquisitions, or divestitures and
restates previously reported earnings, Compustat inserts the restated data into the database
expunging the previously reported numbers. Similarly, if the annual audit requires a restatement of its previously reported earnings, Compustat updates the quarterly database to reflect
these restated numbers—again expunging the previously reported earnings numbers.
Charter Oak Investment Systems, Inc. (Charter Oak) has compiled the weekly original CDs
that Compustat distributed to its PC clients to create a database similar to Compustat that contains all original and updated information without expunging any previously recorded earnings
information. This database contains, for each firm in the Compustat quarterly database, three
numbers for each Compustat line item in each quarter: (a) preliminary earnings announcement
(Compustat update code of 2), (b) the AFR figure when Compustat first changed the update
code to 3 for that firm-quarter, and (c) the number that exists in the current version of
Compustat, which is what most investors and researchers utilize. Therefore, the Charter Oak
database permits us to establish whether an earnings revision has occurred in any particular
quarter by comparing the preliminary announced earnings to the first-reported 10-Q/K
earnings.

Sample Selection

The population consists of firm-quarter observations in the Compustat database between
the fourth quarter of 1993 (the 1st year of available SEC filing dates) and third quarter of
2009 and were traded on the NYSE, AMEX, or NASDAQ. Returns are calculated from
CRSP. We exclude firms with missing CUSIP; firms for which we were unable to find a
CRSP PERMNO for the GVKEY; firms with a subsequent event footnote, market value or
total assets below US$1 million at quarter-end, price per share less than US$1 at quarterend, or sales or total assets below US$1 million at the end of the prior quarter; firms that
are incorporated outside the United States, that have a missing preliminary earnings report
date, or that have a missing value for preliminary earnings before extraordinary items and
discontinued operations (Compustat quarterly Item No. 8) or the SEC filed earnings (the
AFR earnings in Charter Oak database), or for which we have no SEC filing date. We further delete observations with unavailable buy-and-hold returns for all our event windows,

Downloaded from jaf.sagepub.com at Taylor's University on December 2, 2012


Hollie et al.

13

or where the buy-and-hold returns for any of the windows is in the top or bottom 0.5% of
the observations, and obtain 168,436 observations. Out of these observations, we have
3,337 firm-quarters with differences between preliminary earnings and first-reported earnings in excess of US$100,000, which meets the additional criteria described below. This is
not a trivial proportion; about 1.98% of the relevant population files a different earnings
figure with the SEC than the one disclosed in the preliminary earnings announcement just a
few days/weeks earlier. We tested 100 of these earnings revisions to determine whether the
Compustat preliminary earnings announcement and SEC-filed earnings as captured by
Charter Oak is reliable. We were able to corroborate all cases we checked by hand to the
source documents.
To examine the characteristics of firms that materially revise their earnings on the SEC
filing after publicly disclosing a different earnings figure, we matched our sample of
Revisers with a control sample of non-Revisers from the same industry (Fama & French,

1997; based on 48 industries), where the initial earnings surprise decile is the same (earnings for the quarter minus earnings for Quarter t 2 4, scaled by market value), and where
the Revisers and non-Revisers are no more than one size (market value of equity at
quarter-end) decile apart. If we cannot find a matching firm from the same industry, we
eliminate the Reviser from our sample.
In addition, the following requirements must be met for Revisers and control firms
alike: (a) the absolute value of earnings to market value of equity (E/P) and to total assets
(ROA) at quarter-end is below one—this is intended to eliminate extreme cases of E/P or
ROA, (b) the absolute value of the scaled earnings surprise at the preliminary earnings
announcement is available and is below one, and (c) the preliminary earnings announcement is available on Compustat and is prior to the SEC filing date (eliminating observations subject to the Stice, 1991, effect). Thus, our sample selection criteria yield a final
sample of 6,674 observations (firm-quarters) for Revisers matched by control firm observations. In addition, we examine the full Compustat and matched sample as part of our sensitivity analysis throughout this study.

Descriptive Statistics
Panels A, B, and C of Table 1 contain descriptive statistics about Revisers and nonRevisers (i.e., control firms). It shows that Revisers have greater earnings volatility, proportion of accruals, debt, number of segments, frequency of auditor changes, analyst following,
and losses than control firms. Preliminary earnings surprises are more negative for Reviser
firms. The table also shows that Revisers have lower E/P ratios, persistence of earnings surprises, and ROA. CARprelim is lower for Revisers than control firms, consistent with a
lower earnings surprise on that date. The return for the window between the preliminary
earnings announcement and SEC filing date (CARaf) is positive (.00056) for the control
group but negative (2.003) for Revisers, which is consistent with more negative news
during that period for Revisers.
Panel C of Table 1 provides univariate tests of mean differences for Revisers versus
control firms for all variables. As can be seen from the table, earnings volatility, proportion
of accruals, financial leverage, frequency of auditor changes, number of segments, number
of analyst following, and losses are significantly larger for Revisers than control firms. The
earnings-to-price ratio, persistence of earnings surprises, profitability, and buy-and-hold
returns for the preliminary earnings as well as through the SEC filing periods are all significantly larger for control firms than Revisers. Consistent with our matched-pair design, we

Downloaded from jaf.sagepub.com at Taylor's University on December 2, 2012


14

Journal of Accounting, Auditing & Finance

Table 1. Summary Statistics on Variables for Sample and Control Firms
Panel A: Descriptive Statistics for Sample Firms
Variable
EARNVOL
E/P
ACCPROP
COREOCF
PERSE
DEBT
AUC
MARKET VAL
SEGNUM
ROA
IBESN
LOSS
BIG
FSURP
ASURP
CARprelim
CARfile
CARtotal
CARaf

n
3,231
3,337
2,552

2,527
3,157
3,165
2,928
3,337
2,768
3,337
3,337
3,337
3,337
3,337
3,337
3,337
3,337
3,337
3,337

M

SD

1.658
1.589
20.028
0.164
0.227
0.419
0.206
0.434
0.130

0.376
0.244
0.209
0.075
0.264
4,864.54 16,951.71
2.662
1.813
20.00671
0.072
6.664
6.504
0.277
0.448
0.797
0.402
20.020
0.228
20.017
0.111
20.005
0.075
20.003
0.047
20.011
0.128
20.003
0.094

10th

25th
50th
75th
90th
percentile percentile percentile percentile percentile
0.237
20.076
0.045
20.390
20.368
0
0
66.096
1.000
20.042
0
0
0
20.057
20.018
20.089
20.055
20.162
20.109

0.425
20.010
0.071
20.117
20.138

0.061
0
218.677
1.000
20.006
1.000
0
1.000
20.011
20.004
20.038
20.024
20.076
20.050

0.947
0.0080
0.123
0.236
0.115
0.215

2.527
5.000
0.018
0.030
0.223
0.423
0.563
0.764

0.412
0.644
0.378
0.515
0
0
826.041 3,442.30 11,164.13
2.000
4.000
5.000
0.004
0.014
0.029
5.000
10.000
16.000
1.000
1.000
1.000
1.000
1.000
0.001
0.007
0.026
20.00042 0.000394
0.004
20.003
0.029
0.076
20.002

0.018
0.047
20.011
0.051
0.134
20.003
0.039
0.099

Panel B: Descriptive Statistics for Control Firms
Variable
EARNVOL
E/P
ACCPROP
COREOCF
PERSE
DEBT
AUC
MARKET VAL
SEGNUM
ROA
IBESN
LOSS
BIG
FSURP
ASURP
CARprelim
CARfile
CARtotal
CARaf

n

M

SD

3,208
1.288
1.451
3,337
0.002
0.070
2,618
0.209
0.435
2,573
0.208
0.434
3,133
0.149
0.372
3,132
0.223
0.202
2,954
0.057
0.232
3,337 4,098.63 11,580.79
2,802

2.509
1.757
3,337
0.008
0.042
3,337
6.301
6.306
3,337
0.177
0.381
3,337
0.800
0.400
3,337 20.004
0.104
3,337
0
0
3,337
0.003
0.074
3,337 20.00023
0.047
3,337
0.003
0.118
3,337
0.00056
0.082

10th
25th
50th
75th
90th
percentile percentile percentile percentile percentile
0.186
20.017
0.040
20.420
20.338
0
0
66.180
1.000
20.013
0
0
0
20.021
0
20.076
20.049
20.127
20.090

0.308
0.005
0.064

20.103
20.125
0.045
0
216.334
1.000
0.002
1.000
0
1.000
20.004
0
20.030
20.022
20.057
20.037

0.639
0.012
0.107
0.243
0.133
0.192
807.559
2.000
0.008
5.000
1.000
0.001
0.002

20.001
20.002
20.002

Downloaded from jaf.sagepub.com at Taylor's University on December 2, 2012

1.663
0.020
0.200
0.547
0.439
0.350
0
3,168.98
4.000
0.019
9.000
0
1.000
0.006
0
0.035
0.018
0.057
0.034

4.130
0.029
0.371
0.763

0.654
0.473
0
10,648.02
5.000
0.034
15.000
1.000
1.000
0.016
0
0.085
0.048
0.142
0.091


Hollie et al.

15

Table 1. (continued)

___________________________________________________________________________
Panel C: Tests of Mean Differences: Control Minus Sample Firms
Variables
EARNVOL
E/P
ACCPROP
COREOCF

PERSE
DEBT
AUC
MARKET VAL
SEGNUM
ROA
IBESN
LOSS
BIG
FSURP
ASURP
CARprelim
CARfile
CARtotal
CARaf

t statistic

p

29.76
9.79
21.53
0.18
1.94
23.96
22.82
20.75
23.20
10.00

22.31
29.89
0.31
3.79
8.73
4.44
2.21
4.64
1.85

\.0001
\.0001
.0126
.8584
.0524
\.0001
.0048
.4503
.0014
\.0001
.0209
\.0001
.7604
.0002
\.0001
\.0001
.0272
\.0001
.0637

Notes:
1. Sample observations are for firm-quarters during 1994 to 2009 with earnings revisions between preliminary
announcements dates and subsequent Securities and Exchange Commission (SEC) filings of at least US$100,000.
2. Control firms are from the same Fama and French (1997) industry (48 industries) that are closest in size to the
revising firm in the same quarter but without an earnings revision.
3. EARNVOL = earnings volatility, which is estimated as the absolute value of the ratio of the standard deviation of
earnings per share (EPS)/price over the most recent 12 months to the average EPS/price over the same period.
4. E/P = current quarterly earnings before extraordinary items and discontinued operations divided by market
value at the end of the quarter.
5. ACCPROP = proportion of accruals. This variable is the absolute value of total accruals divided by sales, averaged over the previous four quarters. Total accruals are income before extraordinary items and discontinued operations minus net operating cash flow (OCF).
6. COREOCF = correlation of quarterly earnings and OCF, estimated over the eight previous quarters.
7. PERSE = persistence of earnings surprises. Estimated as the first autocorrelation between earnings surprises in
the previous eight quarters. Earnings surprises are earnings at Quarter t minus earnings at Quarter t 2 4, scaled
by market value of equity at the beginning of the quarter.
8. DEBT = debt/assets. Estimated as short-term plus long-term debt divided by total assets at the end of the quarter.
9. AUC = auditor change. A dichotomous variable obtaining one if the firm’s auditor was changed from the prior year.
10. MARKET VAL = market value of equity in millions of dollars as of quarter-end.
11. SEGNUM = number of segments. Taken from the Compustat segment file and is a surrogate for operating complexity.
12. ROA = ratio of earnings to total assets at quarter-end.
13. IBESN = number of analysts on Institutional Brokers Estimate System (IBES). A measure of coverage and informational environment.
14. LOSS = dummy variable obtaining one if earnings for the quarter are negative.
15. BIG = an indicator variable equals to one if the firm is audited by a big auditor (currently the Big 4).
16. FSURP = The First (Preliminary) Surprise. Calculated as the Compustat preliminary quarterly earnings (mnemonic IBQ) minus the As-First-Reported (AFR) Compustat quarterly earnings for t 2 4, scaled by market value at
quarter-end.
17. ASURP = additional surprise. Calculated as earnings in the SEC filing (AFR Compustat earnings) minus earnings
reported in the preliminary earnings release (preliminary Compustat earnings), scaled by market value at quarterend. By definition, it is zero for control firms.

Downloaded from jaf.sagepub.com at Taylor's University on December 2, 2012

16

Journal of Accounting, Auditing & Finance

Table 1. (continued)

___________________________________________________________________________
18. CARprelim = the buy-and-hold abnormal return from Day 21 through Day 11, where Day 0 is the preliminary
earnings announcement date. The abnormal return is the raw buy-and-hold return minus the buy-and-hold return
on a same size book-to-market (B/M) portfolio (six portfolios), as provided by professor French.
19. CARfile = the buy-and-hold abnormal return from Day 21 through Day 11, where Day 0 is the SEC filing date
of 10-Q/10-K.
20. CARtotal = the buy-and-hold abnormal return from 1 day before the preliminary earnings announcement
through 1 day after the SEC filing date.
21. CARaf = the buy-and-hold abnormal return between preliminary announcements and SEC filings. This variable
controls for information released to the market between the two time periods.
22. The t-statistics test that the mean of the sample firm is smaller than the mean of control firms based on
unequal variances.

do not find significant differences for core operating cash flows, firm size, and whether a
firm is audited by a big audit firm.
Panel A of Table 2 reports the number of observations by year and fiscal quarters. There
is an indication of an increasing trend in the number of Revisers across years through 2008,
likely due to more extensive coverage of Compustat firms than a real increase in the proportion of Revisers to total Compustat firms. A noticeable trend is that there are about three
times as many Revisers in the fourth fiscal quarter as in other quarters. This may be attributed to revisions required by the auditor as part of the year-end audit work. It may also be
attributed to a longer period between the balance sheet date and the SEC filing for the fourth
quarter, when material errors may become known requiring a revision in earnings.
Panel B of Table 2 shows that about 64% (2,132 out of 3,337) of the earnings revisions
from the preliminary earnings announcements to the SEC filings are downward revisions,
resulting in lower filed earnings. This is not only consistent with strategic reporting by

management that is intended to inflate earnings initially but also with additional uncovered
errors that are more likely to cause downward revisions due to the conservative nature of
accounting. Panel B also indicates that about 54% of all Revisers (1,800 of 3,337) had a
negative initial earnings surprise, as compared with about 61% (2,043 of 3,337) for the
control firms. Earnings revisions (like restatements) may indicate problems in managerial
information systems and potentially other performance problems. These problems are likely
to be manifested in future periods as well. However, unlike restatements that may be disclosed many months after the problems occurred, the problems may surface very quickly
after earnings revisions are disclosed. Hence, untabulated tests show that Revisers have a
statistically significant higher proportion of firms (6% more than the matched nonRevisers) with negative earnings surprises in the subsequent quarter after the earnings revision. Revisers, on average, have 8% more negative surprises over the subsequent four quarters than the matched non-Revisers (a statistically significant difference between the two
groups). Thus, earnings revisions imply a higher likelihood for earnings disappointments in
the immediate subsequent quarters.
Panel C of Table 2 shows the distribution of earnings revisions across firms. The majority of firms, 68%, have only one revision during the 16-year (64 quarters) period. Another
18% have two earnings revisions during the sample period, and fewer than 14% have three
or more earnings revisions during the sample period. Although the overall frequency of
earnings revisions is low (about 2%), we find that 25% of all firms in the Compustat database (with the same data requirements as our sample firms) have at least one revision
during the study period, indicating the breadth of this phenomenon. In sensitivity tests, we

Downloaded from jaf.sagepub.com at Taylor's University on December 2, 2012


Hollie et al.

17

Table 2. Sample Composition
Panel A: Observations by Year and Fiscal Quarters
Year

Control firms

Sample firms

Total firms

Quarter 1

Quarter 2

Quarter 3

Quarter 4

1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
All

19
35
96
139
183
206
204
206
218
231
208
286
354
286
272
277
117
3,337

19
35
96
139
183
206
204
206
218
231

208
286
354
286
272
277
117
3,337

38
70
192
278
366
412
408
412
436
462
416
572
708
572
544
554
234
6,674

0
11

18
33
33
40
36
43
48
33
25
34
49
47
40
31
40
561

0
4
11
24
32
38
45
34
42
41
28
25
51

44
35
43
28
525

0
10
21
22
42
36
35
37
29
40
40
32
54
37
45
41
35
556

19
10
46
60
76

92
88
92
99
117
115
195
200
158
152
162
14
1,695

Panel B: Upward and Downward Earnings Revisions and Initial Earnings Surprise
Sample firms
Year
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005

2006
2007
2008
2009
All

Control firms

Sample firms

Upward
revisions

Downward
revisions

Positive
FSURP

Negative
FSURP

Positive
FSURP

Negative
FSURP

10
20

51
58
73
84
74
64
71
76
83
105
131
101
95
72
37
1,205

9
15
45
81
110
122
130
142
147
155
125
181
223

185
177
205
80
2,132

5
14
30
41
44
78
77
94
123
71
77
59
122
107
122
164
66
1,294

14
21
66
98
139

128
127
112
95
160
131
227
232
179
150
113
51
2,043

5
10
47
44
66
90
83
99
114
91
90
105
149
112
148
202

82
1,537

14
25
49
95
117
116
121
107
104
140
118
181
205
174
124
75
35
1,800

Downloaded from jaf.sagepub.com at Taylor's University on December 2, 2012


18

Journal of Accounting, Auditing & Finance

Table 2. (continued)

___________________________________________________________________________
Panel C: Frequency of Revisions per Reviser-Only Sample
Revisions

Firms

1
2
3
4
5
6
7
8
9
10
11
12
13
15
16
19
23
26
40

1,348
360
138

62
25
12
4
8
5
6
3
1
1
1
2
1
1
1
1

%
68.08
18.18
6.97
3.13
1.26
0.61
0.20
0.40
0.25
0.30
0.15
0.05

0.05
0.05
0.10
0.05
0.05
0.05
0.05

Cumulative n
1,348
1,708
1,846
1,908
1,933
1,945
1,949
1,957
1,962
1,968
1,971
1,972
1,973
1,974
1,976
1,977
1,978
1,979
1,980

Notes: FSURP = first earnings surprise.

1. Sample observations are for firm-quarters during 1994 to 2009 with earnings revisions between preliminary
announcements dates and subsequent Securities and Exchange Commission (SEC) filings of at least US$100,000.
2. Control firms are from the same Fama and French (1997) industry (48 industries) that are closest in size to the
revising firm in the same quarter but without an earnings revision.
3. Compustat firms are those observations that fit the initial sample selection criteria during the sample period.
The sample selection criteria require market value at quarter-end of at least US$1 million, price per share at
quarter-end in excess of US$1, sales and total assets at the beginning of the quarter in excess of US$1 million,
availability of a preliminary earnings announcement date on Compustat that is prior to the SEC filing date, availability of preliminary earnings and As-First-Reported earnings on Charter Oak, and availability of the number of shares
used to calculate earnings per share (EPS).

report on the robustness of our results to using only firms with just one earnings revision
during our sample period.
Untabulated results indicate that business services, telecommunications, trading, and
utilities have greater sample representation (proportionately more Revisers) than other
Compustat industries, where the first two industries are mentioned in prior studies as
having a greater litigation risk. In contrast, insurance, retail, pharmaceutical products, steel,
consumer goods, and medical equipment industries are represented less often in the
Revisers sample. Thus, one cannot strongly conclude that industry membership can explain
satisfactorily earnings revisions. However, it seems that more stable industries are slightly
underrepresented in the Revisers sample, whereas industries that are more volatile tend to
be slightly overrepresented.
Table 3 reports summary statistics about the components of earnings that are revised
between the preliminary earnings release and the SEC filing. This table summarizes revisions
only for those items that had been disclosed in the preliminary earnings release and subsequently revised; if Compustat could not find an earnings component in the preliminary earnings release, a component revision is not recorded for that firm, although the earnings

Downloaded from jaf.sagepub.com at Taylor's University on December 2, 2012


Hollie et al.

19

Table 3. Revisions to Components of Earnings
Component of earnings
Total revision
Sales
Cost of sales
Selling, general, and administration
Depreciation
Nonoperating gains and losses
Special items
Tax
Extraordinary items and discontinued operation
Recurring
Nonrecurring

n

M

Median

3,337
1,029
1,450
703
212
653
840
1,673

501
3,337
3,337

0.0293
0.0250
0.0217
0.0082
0.0028
0.0088
0.0858
0.0118
0.0172
65%
15%

0.0018
0.0020
0.0061
0.0016
0.0006
0.0022
0.0046
0.0013
0.0017

Notes:
1. Sample observations are for firm-quarters during 1994 to 2009 with earnings revisions between preliminary
announcements dates and subsequent Securities and Exchange Commission (SEC) filings of at least US$100,000.
2. The table reports statistics about the components of earnings that are revised between the preliminary earnings

announcement and the subsequent SEC filing. Table entries are for the absolute value of the revision, scaled by
market value of equity at quarter-end.
3. Recurring items are sales; cost of sales; selling, general, and administration (SG&A); and depreciation expenses.
Firms are classified as recurring when they have a revision in a recurring item and there is no revision to nonoperating gains and losses, special items or extraordinary items, and discontinued operations. Nonrecurring is the opposite.

component may in fact have been revised for that firm. We classify revisions to sales, cost of
sales, SG&A, and depreciation as recurring items if there are no revisions to nonrecurring
items for these firms. We classify revisions to nonoperating gains and losses, special items,
and extraordinary items and discontinued operations as nonrecurring items if there are no
revisions to recurring items for these firms.12 Table 3 shows that the tax component is subject
to the most revisions followed by cost of sales and sales components of net income.

Determinants for Revisers
Table 4 provides the results of the logistic regression analysis, where the dependent variable
is one for Revisers and zero for control firms. We provide two sets of tests. In the first, we
report all significant variables from the univariate tests. In the second, we exclude all insignificant variables from the first test. Our regression model has high explanatory power with
a significant likelihood ratio. Approximately 61% of the observations are predicted correctly. All variables in the model are statistically significant and in the predicted direction,
except proportion of accruals, E/P in the full matched sample, and earnings persistence in
the abbreviated matched-sample logistic regression. In particular, we find that the more
complex a firm (indicated by the number of segments), the greater the likelihood of a subsequent earnings revision. Similarly, the likelihood of a revision is higher for firms with losses
or more financial leverage, possibly because these firms are more likely to have operational
and accounting control problems and greater incentives to report strategically on the preliminary earnings announcement date. Firms with greater earnings volatility are also more likely
to have subsequent earnings revisions, possibly because subsequent information after the
preliminary earnings announcements is more likely to become available. Firms with greater
profitability (ROA) have a lower likelihood of a revision. Firms with auditor changes are
more likely to have a subsequent earnings revision, possibly because new auditors are more

Downloaded from jaf.sagepub.com at Taylor's University on December 2, 2012

20

Journal of Accounting, Auditing & Finance

Table 4. Logistic Regression to Predict Earnings Revisions Using a Matched-Pair Sample and a Full
Sample

Variables

Expected

Intercept

?

ROA

?

DEBT

1

SEGNUM

1

EARNVOL

1

PERSE

2

AUC

?

LOSS

1

E/P

?

ACCPROP

1

No. of observations
Likelihood ratio
p value
Percentage concordant

Matched pair
sample
Parameter
estimate

Matched pair
sample
Parameter
estimate

20.5405
\.0001
22.2021
.0035
0.3401
.0195
0.0695
.0001
0.1402
\.0001
20.1259
.1347
0.2613
.0402
0.2505
.0050
20.5763
.0972
20.0705
.4013
4,327
181.52
\.0001
61.0

20.4988
\.0001
22.8365
\.0001
0.4311
.0020
0.0536
.0012
0.1307
\.0001
20.0728
.3566
0.2499
.0343
0.2300
.0060

4,913
169.04
\.0001
60.1

Full sample
Parameter
estimate

Full sample
Parameter
estimate

24.5711
\.0001
21.8394
\.0001
0.2478
.0114
0.1631
\.0001
0.0700
\.0001
20.1411
.0167
0.1537
.0660
0.1304
.0269
21.6833
\.0001
0.0114
.8295
116,143
471.63
\.0001
57.2

24.5330
\.0001
23.2767
\.0001

0.5674
\.0001
0.1745
\.0001
0.0522
\.0001
20.1708
.0020
0.1794
.0210
0.1594
.0041

120,646
476.05
\.0001
58.7

Notes: Precise descriptions of sample and control firms as well as variable definitions are in notes to Table 1. The
table provides the results of Logistic Regression where the dependent variable equals one if a sample firm (earnings
revision) and zero if control. p values (in italics) are shown below each estimate. Bold entries are significant below
5%. The table provides results for a matched-pair sample, where each Reviser is matched with a non-Reviser from
the same industry and closest in size, and for the full sample of firms.

likely to uncover issues not known at the preliminary earnings release date. Firms with a
higher persistence of earnings surprises are less likely to have subsequent revisions, probably because of their greater earnings stability. Thus, firms with earnings revisions are those
that have operations that are more complex, greater earnings volatility, higher leverage, and
losses, and those that experienced an auditor change. Profitable firms and firms with persistent earnings changes are less likely to have an earnings revision.

Market Reactions to Earnings Revision

Panel A of Table 5 presents the results of regression equations for the full Reviser sample,
where the dependent variable is the market reaction and the independent variables include
various earnings surprises for Revisers. Column 2 provides information about market reactions to the preliminary earnings announcement. Consistent with prior studies, there is a significantly positive association between the scaled earnings surprise (FSURP) and the 3-day

Downloaded from jaf.sagepub.com at Taylor's University on December 2, 2012


Hollie et al.

21

Table 5. Market Reactions to Earnings Surprises in Preliminary Earnings Announcements and SEC
Filings
Panel A: Market Reactions to Earnings Revision
Reviser sample
Revisers only

Revisers only

CARprelim

CARfile

Dependent variable
Intercept
FSURP

2.0037
.0044
.0367

\.0001

2.0024
.0039
.0028
.2556
.0225
.0023
2.0299
.0006
\.0001
.0057
3,337

ASURP
CARaf
F value
Adjusted R2
Observation

\.0001
.0267
3,337

Panel B: Market Reactions to Upward/Downward Earnings Revision
Full reviser sample
Earnings revision:
Dependent Variables
Intercept
FSURP

Downward
CARprelim

Downward
CARfile

Upward
CARprelim

Upward
CARfile

2.0047
.0035
.0393
\.0001

2.0033
.0026
.0002
.9464
.0241
.0024
2.0271
.0133
.0016
.0058
2,132

2.0020
.3554
.0325
\.0001

2.0002
.8859
.0073
.0632
2.0339
.3099
2.0388
.0076
.0123
.0065
1,205

ASURP
CARaf
F value
Adjusted R2
Observation

\.0001
.0310
2,132

\.0001
.0197
1,205

Note: Sample observations (Reviser = 1) are for firm-quarters during 1994 to 2009 with earnings revisions
between preliminary earnings and subsequent Securities and Exchange Commission (SEC) filings. p values (in italics)
are shown below each estimate. Bolded entries are significant below 5%.
FSURP = The First (Preliminary) Surprise. Calculated as the Compustat preliminary quarterly earnings (mnemonic IBQ)
minus the As-First-Reported (AFR) Compustat quarterly earnings for t 2 4, scaled by market value at quarter-end.
ASURP = additional surprise. Calculated as earnings in the SEC filing (AFR Compustat earnings) minus earnings
reported in the preliminary earnings release (preliminary Compustat earnings), scaled by market value at quarterend. By definition, it is zero for control firms.
CARprelim = the buy-and-hold abnormal return from Day 21 through Day 11, where Day 0 is the preliminary
earnings announcement date. The abnormal return is the raw buy-and-hold return minus the buy-and-hold return
on a same size, book-to-market (B/M) portfolio (six portfolios), as provided by professor French.
CARfile = the buy-and-hold abnormal return from Day 21 through Day 11, where Day 0 is the SEC filing date of
10-Q/10-K.
CARaf = the buy-and-hold abnormal return between preliminary earnings announcements and SEC filings. It controls for information released to the market between the two time periods.

Downloaded from jaf.sagepub.com at Taylor's University on December 2, 2012


22

Journal of Accounting, Auditing & Finance

buy-and-hold return centered on the preliminary announcement date (0.0367, p \ .0001).
Column 3 presents the results of regressing 3-day buy-and-hold excess returns around the
SEC filing date where the earnings surprise includes also the additional earnings surprise
(ASURP). We also control for the information flow to the market between the preliminary
earnings release and the SEC filing date by adding CARaf, the buy-and-hold abnormal return
between the two dates, as an independent variable in the regression. We find that the additional
surprise on the revised earnings on the SEC filing date, ASURP, has a positive and significantly different from zero association with the 3-day abnormal return around the filing date
(0.0225, p = .0023). Thus, consistent with the findings of Hollie et al. (2005), investors react to

this new earnings information on the SEC filing date. As expected, we find that there is no statistically significant market reaction to the already-known initial earnings surprise (FSURP).
Also, note that the information flow to the market between the preliminary earnings release
date and the SEC filing date (CARaf) is significantly negative (2.0299, p = .00006).13
In untabulated results, we also examine the lag between preliminary earnings and
quarter-end, as compared with the average lag in the same quarter of the prior 4 years.14
We find that Revisers have an average lag increase of about 1 day more than the matched
sample in each of the next four quarters after the revision (all differences between Revisers
and controls are statistically significant), indicating that managers of Revisers take more
time in future quarters to disclose preliminary earnings. This indicates both an economic
consequence of the earnings revision (less timely information to market participants) and a
unique feature of earnings revision as compared with restatements. Similarly, we examine
the lag between the preliminary earnings announcement and the subsequent SEC filing for
Revisers and non-Revisers in Quarter t 1 4 (because many revisions occur in the fourth
fiscal quarter). The lag had decreased by more than 2 days on average for Revisers than
non-Revisers (statistically significant), indicating that managers of Revisers delay the voluntary disclosure of preliminary earnings closer to the SEC filing.
We also investigate whether the market reactions differ for Revisers with upward versus
downward earnings revisions. In Panel B of Table 5, we find that the significant market
reactions to the additional earnings surprises in SEC filings, ASURP, are only significant
when earnings are revised downward. When earnings are revised upward, there is an insignificant association between the SEC filing returns and the additional earnings surprise,
ASURP. This may be related to market participants’ suspicions about the strength of the
accounting and reporting system of firms with an earnings revision, even if the revision is
supposed to convey ‘‘good’’ news to the market. Hence, there is a positive and significant
association between market returns around the SEC filings and the additional earnings surprise when firms revise earnings, although this association is mainly concentrated in firms
that revise earnings downward.
Table 6 presents the two-stage least squares (2SLS) parameter estimates in column 2
and the ordinary least squares (OLS) parameter estimates in columns 3 and 4. As a firm
can appear several times during the sample period, we use robust standard errors clustered
by firm to control for the dependence among the regression residuals for the OLS estimates. The dependent variable in this table is the total return from a day before the preliminary earnings announcement through one day after the SEC filing date. This window
captures the effects of the preliminary earnings surprise and the new information that
appears in SEC filings. For non-Revisers, the independent variables include the earnings

surprise, TSURP, which will be equal to the preliminary earnings surprise. For Revisers, it
is equal to the sum of the preliminary earnings surprise, FSURP, and the additional earnings surprise in the SEC filing, ASURP. If the market does not penalize Revisers, the

Downloaded from jaf.sagepub.com at Taylor's University on December 2, 2012


Hollie et al.

23

Table 6. Market Reactions to Total Earnings Surprises From the Preliminary Earnings Announcement
Through the SEC Filing
Variables
Intercept
REVISER
TSURP
TSURP 3 REVISER
ROA
DEBT
SEGNUM
EARNVOL
PERSE
AUC
LOSS
F value
Stage 1 adjusted R2
Stage 2 adjusted R2
Number of clusters
No. of observations

Full sample
(2SLS)

Full sample
(clustered by firm)

Full sample
(clustered by firm)

.0052
\.0001
2.2038
\.0001
.0476
\.0001
2.0039
.5879
2.0804
\.0001
.0124
\.0001
.0039
\.0001
.0010
\.0001
2.0031
.0044
.0039
.0180
.0032

.0061

.0007
.0146
2.008
.0001
.0478
\.0001
.0013
.8452

.0016
.0608
2.006
.0231
.0422
\.0001
2.0082
.3139
.0289
.0133
2.0014
.4589
.0010
\.0001
2.0000
.9220
.0016
.0993
2.0030

.0412
2.0127
\.0001

\.0001
.00422
.01532

\.0001
.0158

\.0001
.0186

9,703

7,275

120,646

Note: The dependent variable in this table is the buy-and-hold abnormal return from 1 day prior to the preliminary
earnings announcement through 1 day after the Securities and Exchange Commission (SEC) filing date. The abnormal return is raw buy-and-hold return minus the buy-and-hold return on the matched size and book-to-market
(B/M) portfolio (six portfolios) as obtained from professor French’s data library. REVISER is an indicator variable
which equals to one if an earnings revision in excess of US$100,000 took place in any quarter between 1991 and
2004. TSURP is the total earnings surprise, consisting of FSURP plus ASURP. FSURP, the First (Preliminary)
Surprise, is calculated as the Compustat preliminary quarterly earnings (mnemonic IBQ) minus the As-FirstReported (AFR) Compustat quarterly earnings for t 2 4, scaled by market value at quarter-end. ASURP is the
Additional surprise, calculated as earnings in the SEC filing (AFR Compustat earnings) minus earnings reported in
the preliminary earnings release (preliminary Compustat earnings), scaled by market value at quarter-end. By definition, it is zero for control firms. p values (in italics) are shown below each estimate. Bolded entries are significant
below 5%.
All firms are ranked according to TSURP each quarter and are assigned into 10 deciles. TSURP in the table represents the decile rank divided by 9 minus 0.5. Thus, the coefficient on TSURP measures the abnormal buy-and-hold

return on a ‘‘hedge’’ portfolio holding long (short) positions in the most positive (negative) decile of total surprises.
TSURP 3 REVISER is an interaction variable of TSURP and REVISER, and is used to assess the additional market
reaction to the total surprise of Revisers as compared with that of other firms. For the definition of other control
variables, see the notes to Table 1. The table reports the results of a two-stage least squares (2SLS) where
REVISER is an endogenous variable and is estimated by the control variables and is then used as an independent
variable in the market reactions regression, as well as two separate ordinary least squares (OLS) regressions, with
and without controls.

Downloaded from jaf.sagepub.com at Taylor's University on December 2, 2012


24

Journal of Accounting, Auditing & Finance

coefficient on the interaction term TSURP 3 REVISER should be insignificantly different
from zero. If we find that the coefficient on TSURP 3 REVISER is negative and significant, it indicates that the market reacts less strongly to a dollar of earnings surprise for
Revisers, suggesting a penalty for their lower earnings quality. The regression also controls
for variables that were previously shown to affect earnings revision.
We find that the coefficient on REVISER is negative and significant. The coefficient on
TSURP 3 REVISER is statistically insignificant, which suggests that the total market reaction to the total earnings surprise does not differ significantly when an earnings revision
occurs. This is consistent with no penalty for earnings revisions. Our OLS regression results
are consistent with our findings using a 2SLS model.
Table 7 presents the average differences in excess total returns (preliminary earnings
announcement through SEC filing) for firms with earnings revisions and matched control
firms. We find a negative and significant (t statistic = 23.35, 24.06, and 22.79) difference
between Revisers and control firms, when matched on (a) total earnings surprise and industry, (b) total earnings surprise and size, and (c) earnings surprise, size, and number of segments, which suggests that firms that had earnings revisions have weaker market reactions
than firms without earnings revisions. This weaker market reaction indicates that market
participants likely interpret the earnings revision as a weakness in the quality of earnings or
the firms’ reporting and control systems. We find no significant difference between the two

when matched on earning surprise, size, and industry, although the returns are lower for
Revisers than non-Revisers. When we compare a Reviser with itself in another quarter with
no revision and with the same earnings surprise decile rank, we find that market reactions
during the earnings revision quarter are significantly weaker (t statistic = 24.16), likely
due to the penalty imposed on Revisers for their lower earnings quality. These results are
inconsistent with those in Table 6, which compare the Revisers with all non-Reviser firms.

Sensitivity Analyses
We delete all Revisers with more than one revision. This is likely to eliminate cases where
Compustat did not find sufficient data in the preliminary earnings release but was able to
report a different income figure when the SEC filings became available. The main results
are virtually the same; there is a positive and significant market reaction to ASURP, with
no market reaction to FSURP after controlling for information between the preliminary
earnings release and SEC filing (CARaf). There is a significantly weaker market reaction to
the total earnings surprise of Revisers than to those of non-Revisers.
We examine the results in the pre- and post-Regulation Fair Disclosure periods. Firms
may behave differently after the year 2000, where private channels of communication with
special groups of investors are restricted. Nevertheless, our main results hold in the period
through 2000 and during the years 2001 to 2009.
We repeat our main tests separately for earnings revisions in the fourth quarter, when an
audit is required, and all other quarterly earnings revisions as one group. For fourth quarter
observations, the logistic regression coefficients are all in the same direction as our main
results, but the debt, size, and persistence of earnings are insignificant. The additional earnings surprise is positively associated with abnormal returns around the filing date, but the
market reaction to the total earnings surprise, although weaker than for non-Revisers, is not
statistically different. For earnings revisions in other quarters, the logistic regression coefficients are all in the same direction and significant as our main results, except for earnings
volatility, which is insignificant. The additional earnings surprise for other quarterly

Downloaded from jaf.sagepub.com at Taylor's University on December 2, 2012

Hollie et al.

25

Table 7. Average Differences in Excess Total Returns on Revisers (CARtotal) and Matched Control
Firms (CARtotal2matched)
Matching

n

Total earnings surprise and
industry
Total earnings surprise, size,
and industry
Total earnings surprise and size
Total earnings surprise, size, and
number of segments
Same firm in a different quarter,
matched by total earnings
surprise

CARtotal CARtotal2matched Differences t statistics

p

3,249 20.0111

20.0032

20.0079

23.35

.0008

1,171 20.0103

20.0039

20.0065

21.30

.1934

3,180 20.0107
2,172 20.0118

20.0007
20.0024

20.0100
20.0094

24.06
22.79

\.0001
.0053

2,820 20.0111

0.0001

20.0111

24.16

\.0001

Notes:
1. The table presents excess total stock returns for a sample of companies with earnings revisions and a matched
control sample of companies. The table also presents the average differences in excess total returns and their significance levels. p values (in italics) are shown below each estimate. Bolded entries are significant below 5%.
2. CARtotal, CARtotal2matched: the buy-and-hold abnormal return from 1 day before the preliminary earnings
announcement through 1 day after the Securities and Exchange Commission (SEC) filing date.
3. Matching techniques (a) matching by earnings surprise decile and industry; (b) matching by size (size difference is
capped at 10%) and earnings surprise decile; (c) matching by size, earnings surprise decile, and the number of segments; (d) matching by size, earnings surprise decile, and industry; and (e) matching the Reviser with itself in
another quarter that has the same total earnings surprise decile rank.
4. Excess returns are measured in the 3-day period (21, 11) around the release of preliminary earnings. Excess
buy-and-hold returns are calculated as the buy-and-hold return from Center for Research in Security Prices
(CRSP) minus the buy-and-hold return on the portfolio of firms with the same size (market value of equity) and
book-to-market (B/M) ratio. Daily returns and cutoff points on the size and B/M portfolios are obtained from professor Kenneth French’s data library, based on classification of the population into six (two size and three B/M)
portfolios. Observations in the top and bottom 0.5% of excess return are deleted from the sample to ensure that
our results are not driven by outlying returns. Portfolio returns are computed each quarter. The table presents
average quarterly returns and t statistics based on a Fama and MacBeth (1973) approach.
TSURP is the total earnings surprise, consisting of FSURP plus ASURP. FSURP, the First (Preliminary) Surprise, is
calculated as the Compustat preliminary quarterly earnings (mnemonic IBQ) minus the As-First-Reported (AFR)
Compustat quarterly earnings for t 2 4, scaled by market value at quarter-end.
ASURP is the Additional surprise, calculated as earnings in the SEC filing (AFR Compustat earnings) minus earnings
reported in the preliminary earnings release (preliminary Compustat earnings), scaled by market value at quarterend. By definition, it is zero for control firms.

All firms are ranked according to TSURP each quarter and are assigned into 10 deciles.

revisions is positively and significantly associated with abnormal returns around the filing
date, and the market reactions to the total earnings surprises of Revisers are significantly
weaker than for non-Revisers.
Next, all Revisers’ public announcements between the preliminary earnings announcement and the subsequent SEC filing dates were examined to determine whether the firm
issued a press release that announced the upcoming earnings revision in the SEC filing.
Only 102 Revisers with sufficient return information issued such revision announcements,
with 68 (66.7%) of them having a negative additional SEC earnings surprise (as compared
with 62.2% for all nonannouncing Revisers). Forty-two revision announcements occurred
within the SEC filing window, the 3-day window centered on the SEC filing date.15 The
results for the nonannouncers are very similar to those reported for the main sample. Even

Downloaded from jaf.sagepub.com at Taylor's University on December 2, 2012


26

Journal of Accounting, Auditing & Finance

for those firms that announced an upcoming earnings revision in the SEC filing, we
observe a positive and significant market reaction to the additional SEC earnings surprise.
We examine the sensitivity of our results to the calculation of the earnings surprise as
being either based on Institutional Brokers Estimate System (IBES) forecasts or on timeseries forecasts when the firm is not followed by analysts. Our main results are stronger for
the subsample of firms followed by at least one analyst; the market reaction to the additional earnings surprise is positive and significant and there is a significantly weaker
market reaction to the total earnings surprise of Revisers as compared with non-Revisers.
We examine the robustness of our results to whether revisions are to recurring or nonrecurring items. As expected, when the revisions are to recurring items the reduction in
market reactions to the total earnings surprise of Revisers as compared with non-Revisers
is more pronounced than for nonrecurring items.
We examine the sensitivity of the results to the various reporting lags between Revisers

and control firms, with no significant role for any reporting lag. Specifically, we find the
that Revisers have a mean (median) lag of 32 (28) days between the quarter-end and the
preliminary earnings announcement, as compared with 29 (27) for control firms. Similarly,
the lag from the preliminary earnings announcement to the SEC filing has a mean (median)
of 32 (28) for Revisers and 24 (20) for control firms. However, in spite of the greater lags
for Revisers when compared with control firms, these lags are not statistically significant in
any of the return regressions we use in the study.

Summary and Conclusions
In this article, we provide evidence on the occasions where earnings revisions are more
likely to occur, whether the market reacts to these additional earnings surprises and whether
market reactions to the total earnings surprises are weaker, likely due to indications of
lower earnings quality or imperfections in the firm’s reporting and control systems.
Contrary to what might be expected, the sample of earnings revisions we use does not
contain just minor, insignificant corrections to the previously issued earnings figure. The
median company reports an earnings revision of more than 2.9% of market value of equity at
quarter-end. More than 25% of our revisions are above 0.7% of market value, and more than
10% are above 2.6% of market value. These are not trivial numbers and potentially can (and
in fact do) cause significant market reactions. We find conflicting results about the market
reactions to the mere act of an earnings revision. When we compare the total market reactions with the preliminary and subsequent additional surprise of Revisers and the full population of non-Revisers, we find no significantly weaker market reactions to Revisers. However,
in matched-pair designs (including using the revising firm as its own control in other quarters), we find significantly weaker market reactions, consistent with potentially lower quality
of earnings. We find that earnings revisions are more likely to occur for firms that are more
complex in nature, are more financially leveraged, have greater earnings volatility, and have
losses, and where an auditor change has occurred. We also find that most of the revisions
occur in the sales, tax, and cost of sales component of earnings.
Our results contribute to the literature on earnings quality, showing another manifestation of a potentially deficient accounting/control system within a firm. Researchers that
study market reactions around SEC filings should specifically exclude cases of earnings
revisions, which likely contaminate the sample as a whole (e.g., Callen et al., 2006). This
study goes further than Hollie et al. (2005) by showing that mangers cannot ‘‘game the
system’’ by initially reporting higher earnings that are subsequently revised in SEC filings;

Downloaded from jaf.sagepub.com at Taylor's University on December 2, 2012