Journal of Accounting and Economics 37 (2004) 139–165
An empirical analysis of auditor reporting and its
association with abnormal accruals$
MartyButler a, Andrew J. Leonea,*, Michael Willenborgb
a William E. Simon Graduate School of Business Administration, University of Rochester, Rochester,
NY 14627, USA
b School of Business Administration, University of Connecticut, Storrs, CT 06269, USA
Received 29 January2002; received in revised form 9 June 2003; accepted 25 June 2003
Abstract
In this paper, we use a web-based sampling methodologyto obtain and content analyze a
large sample of modified audit opinions. Based on this analysis, we re-examine whether certain
modified audit opinions are associated with abnormal accruals. We find that the documented
relation between modified opinions and abnormal accruals rests with companies that have goingconcern
opinions. These firms have large negative accruals that are likelyd ue to severe financial
distress. Overall, we find no evidence to support inferences in previous research that firms
receiving modified audit opinions manage earnings more than those receiving clean opinions.
r 2003 Elsevier B.V. All rights reserved.
JEL classification: M41; D21; C81
Keywords: Abnormal accruals; Audit firms; Audit opinions; Financial distress; Earnings management
ARTICLE IN PRESS
$We thank Mary Barth, Bill Beaver, Joe Carcello, Liz Demers, Bill Felix, Ed Kay, Bill Kinney, John
Hand, Paul Hribar (referee), Jagan Krishnan, Doug Skinner (editor), Richard Sloan, Christine Tan, Scott
Vandervelde, Charlie Wasley, Ross Watts, Joseph Weber, Tzachi Zach, Jerry Zimmerman, and workshop
participants at Dartmouth, Georgia State, Harvard, Iowa, Rochester, Stanford, Temple, the 2002 AAA
Audit Midyear Meeting, and the 2002 University of Illinois Symposium on Audit Research. We also thank
Paul Ames, VinayBassi, Natalya Bushneva, Rumen Efremov, Girts Freibergs, Junfeng Han, Treppy
Johnson, Winward Lewin, Ai Qin Liu, Trevor Lloyd, Chris Wang, and especially Kalina Berova for
excellent research assistance. Support was provided bythe John M. Olin Foundation and the Bradley
PolicyResearch Center at the Universityof Rochester. Portions of this paper were completed while the
second and third authors were visiting the Universityof Michigan and the Universiteit Maastricht,
respectively.
*Corresponding author. Tel.: +1-716-275-4714; fax: +1-716-442-6323.
E-mail address: leone@ssb.rochester.edu (A.J. Leone).
0165-4101/$ - see front matter r 2003 Elsevier B.V. All rights reserved.
doi:10.1016/j.jacceco.2003.06.004
1. Introduction
The audit opinion, as the observable output from an otherwise unobservable
process, represents a crucial piece of information for financial statement users. For
example, the presence of a going-concern (GC) opinion has been shown to be
informative with respect to stock returns (Dopuch et al., 1986; Choi and Jeter, 1992)
and bankruptcyeven ts (Hopwood et al., 1989; Kennedyand Shaw, 1991). More
recently, several studies examine the relation between accounting accruals and the
presence of certain modified audit opinions (Francis and Krishnan, 1999; Bartov
et al., 2000; Bradshaw et al., 2001). In essence, these studies test the hypothesis that
earnings management increases the likelihood of receiving a modified audit opinion.1
The direction of causalityis important because if auditor reporting conveys
information about earnings management, then a link can be made between audit
opinions and earnings quality. In this paper, we re-examine auditor reporting and its
association with abnormal accruals, assessing, in particular, the claim that modified
opinions are reliable signals of earnings management.
Although we expect auditing to limit earnings management, it is not obvious that
earnings management will typically lead to a modified audit opinion. Indeed, if an
auditor detects earnings management and firm managers refuse to adjust the
financial statements, the auditor’s reporting options under GenerallyAccept ed
Auditing Standards (GAAS) are to issue an adverse, disclaimed, or qualified
opinion, the consequences of which can be severe (e.g., per Rule 2-02 of Regulation
S-X, the SEC will not accept the statements). Moreover, firms with certain modified
opinions have unusual accruals for reasons other than earnings management. Due,
for example, to poor firm performance or liquidity-motivated survival tactics, the
presence of GC opinions is often contemporaneouslya ssociated with large negative
accruals. For these reasons, and consistent with the view that most earnings
management takes place within the boundaries of GAAP (Healy, 1985), we argue
that the opinion/accruals relation is not due to earnings management.
To examine the link between abnormal accruals and auditor opinion type, we
begin bypro viding descriptive information on Compustat’s auditor/auditor’s
opinion data item for the 20 years from 1980 to 1999, a period spanning a major
change in GAAS (i.e., SAS 58). Because auditors issue modified opinions for
numerous reasons that Compustat combines into just a few categories, we compare
Compustat opinion classifications for 1994–1999 to source documents on EDGAR.
We find that 1,385 (16%) of 8,478 opinions that Compustat classifies as modified are
actuallyclean opinions. For example, Compustat classifies 801 firm-year observations
as modified simplybecause auditors mention that they reviewed a
supplementarysched ule filed with the 10-K. Based on examining over 7,000
modified opinions from this period, we identifytwo factors that account for over
90% of modifications: consistencyissues requiring no restatements (e.g., adoption of
ARTICLE IN PRESS
1We use the term ‘‘modified opinions’’ to refer to both qualified opinions (scope limitations or
departures from GAAP) and unqualified opinions with explanatorylanguage (e.g., going concern and
consistency).
140 M. Butler et al. / Journal of Accounting and Economics 37 (2004) 139–165
new accounting standards) and GC opinions.2 In addition, we find that while
unqualified opinions with explanatorylangu age are relativelycommon , qualified
opinions are rare. Further, the reasons provided for these qualified opinions appear
unrelated to earnings management and is thus inconsistent with the view that
earnings management typically ‘‘causes’’ modified audit opinions.3
Based on our analysis of the reasons auditors depart from standard unqualified
opinions, we identifythe types of modified opinions that are associated with large
accruals. We find that companies with GC opinions drive the opinion/accruals
relation because such companies have extremely negative abnormal accruals.
However, we report no evidence of differences in abnormal accruals for GC firms
when compared to firms matched on a measure of financial distress. The evidence is
consistent with the view that firms in severe financial distress, independent of audit
opinion type, engage in liquidity-enhancing transactions (e.g., delaying payables or
factoring receivables) that result in large negative accruals. Because traditional
abnormal accrual models do not adequatelyco ntrol for distress-induced variation in
accruals, the large negative accruals characteristic of GC firms appear ‘‘abnormal’’
(i.e., indicative of earnings management), but theyare no different from similarly
distressed clean-opinion firms. Overall, we find no evidence that firms receiving
modified audit opinions manage earnings more than those receiving clean audit
opinions.
Our contribution is two-fold. First, we provide a large-sample descriptive analysis
of audit opinions byaudit firm type. This shows the impact of SAS 58 on auditor
reporting, points out problems with Compustat audit opinion data, and provides a
content analysis of the reasons why companies receive modified opinions. Second, we
show that the modified opinions/abnormal accruals relation stems from companies
with GC opinions, because theyhave negative accruals. As such, our findings cast
some doubt on the conclusions of previous research that has attributed the opinion/
accruals relation to more aggressive earnings management. In particular, we point
out how the Bartov et al. (2000) results appear to be due to severelydist ressed firms
(with GC opinions), rather than due to firms engaging in extreme earnings
management. Overall, we find no evidence to support the claim that firms receiving
modified audit opinions manage earnings more than those receiving clean audit
opinions.
Our findings also have implications for research on the relation between audit
opinion modifications and earnings management, as well as earnings management
in general. In particular, because the auditor’s role is not to assess earnings
quality, researchers should be cautious when drawing inferences about the
relation between auditors’ opinions and traditional measures of abnormal accruals
as proxies for earnings management. As we discuss above, modifications due to
ARTICLE IN PRESS
2We use the term ‘‘GC’’ opinion to refer both to cases in which auditors explicitlyuse the term ‘‘going
concern’’ in the text of the audit report (2,910) and to cases in which auditors mention either bankruptcy
proceedings or a material business uncertainty, but do not explicitly use the term ‘‘going concern’’ (2 5 1).
3Of course, the fact that a qualified audit opinion does not literallymention earnings management does
not mean that auditors fail to detect, or constrain, earnings management (see Section 2).
M. Butler et al. / Journal of Accounting and Economics 37 (2004) 139–165 141
earnings management should be exceedinglyrare. Further, because the negative
relation between certain modified audit opinions and abnormal accruals is
concentrated in firms in financial distress—that is, cases in which traditional
models of earnings management are especiallymis specified—using modified-opinion
firms to compare competing abnormal accrual models is more likelyto measure
the models’ relative misspecification than their power to detect earnings management.
More generally, because GC companies can substantially affect abnormal
accrual calculations, it is important to consider controlling for such opinions.
Finally, our findings highlight the importance of performance matching when the
companies of interest are experiencing extreme performance and demonstrate further
that the choice of performance control variable can be an important design
consideration.
The remainder of the paper is organized as follows. Section 2 discusses
institutional details and the opinion/accruals relation. Section 3 details our sampling
methodology. Section 4 provides a content analysis for those modified audit
opinions that we can find on EDGAR. Section 5 re-examines the opinion/accruals
relation. Section 6 summarizes our main findings and offers some implications and
suggestions for past and future earnings-management research.
2. Earnings-management and the audit opinions/accounting accruals relation
As a form of monitoring, auditing mitigates incentive problems between managers
and outsiders. Following such an agencycost argument, several recent papers
(Francis and Krishnan, 1999; Bartov et al., 2000; Bradshaw et al., 2001) argue
that modified audit opinions are either influenced by, or provide evidence of,
more pervasive earnings management. In general, these papers posit that
modified audit opinions should be a function of accounting accruals.4 However,
given the nature of the audit process and federal securities laws governing accounting
reports, it is unlikelythat earnings management will lead to a modified audit
opinion.
Consider a case where an auditor believes that certain accounts are misstated.
If the suspected earnings management involves no material misstatement or
conforms to GAAP, then there is no basis for the auditor to qualifyor otherwise
modifythe opinion. For example, the auditor may believe that the allowance for
doubtful accounts is understated but deem the understatement to be either
immaterial or due to a difference in judgment within the confines of GAAP. In
either of these cases, the auditor will not issue a modified opinion even if the
ARTICLE IN PRESS
4While the empirical design of these papers differs somewhat, theyall regress audit opinion type on
accounting accruals. Francis and Krishnan (1999) consider firms with audit opinions modified for asset
realization and GC uncertainties as a function of indicator variables that capture extreme levels or
components of total accruals. Bartov et al. (2000) consider firms with audit opinions modified for scope
limitations and GAAP departures as a function of the absolute value of different abnormal accrual
metrics. Bradshaw et al. (2001) consider firms with anyty pe of modified (or ‘‘unclean’’) audit opinion as a
function of signed working capital accruals.
142 M. Butler et al. / Journal of Accounting and Economics 37 (2004) 139–165
accounting seems aggressive. Auditors do not have latitude to comment or elaborate
on the overall qualityof earnings as long as the financial statements materially
conform to GAAP.
If, on the other hand, the auditor determines that the financial statements
include a material misstatement or departure from GAAP, then GAAS oblige the
auditor to communicate the deficiencies to management—or directlyto the audit
committee or SEC in cases of fraud—and request that theyadjust the financial
statements accordingly( O’Malleyet al., 2000, p. 82). If management makes
the necessaryadjustment s, then the monitoring role of the auditor is unobservable
and the audit opinion will contain no earnings-management-based modification.
However, should management refuse to adjust the financial statements, then
the auditor’s choice is either to resign or to issue an adverse, disclaimed, or qualified
audit opinion. The consequences of each option are severe. For example,
per Regulation S-X, the SEC will not accept a registrant’s financial statements
that have been qualified or disclaimed bytheir auditor. Moreover, because the
financial statements contain a material misstatement, ‘‘the registrant [is] knowingly
... filing a false and misleading document,’’ an act punishable under SEC Rule 10b-5
(Herz et al., 1997). The auditor and the companylikelyresol ve earningsmanagement-
related issues before the audit report is issued because the costs
of not doing so are so high. As a result, given the nature of the audit process,
the auditor’s reporting choices under GAAS, and institutional features of the
audit environment, an audit opinion modification due to earnings management
should be rare.5
Evidence in Nelson et al. (2002) suggests that auditors and management virtually
always resolve earnings-management issues before opinions are issued. They report
the results of a surveyof 253 audit partners and managers of a Big 5 firm, who
describe 515 specific incidences of potential earnings management detected during
the course of their audits. Of these instances of possible earnings management, 44%
of the time there was an adjustment, 21% of the time there was no adjustment
because the client was able to demonstrate compliance with GAAP, 17% of the time
there was no adjustment because the auditor was unable to convincinglyshow that
the client’s position was incorrect, and 18% of the time there was no adjustment due
to immateriality(or some other reason). In onlyseven of the 515 instances did the
putative earnings management attempt lead to an opinion modification.6 Moreover,
these seven modifications could be due to disagreements, between management and
the auditor, about the application of GAAP rather than earnings management. The
other 99% of cases were resolved prior to the issuance of the financial statements and
audit report.
ARTICLE IN PRESS
5We thank Ed Kay, Managing Partner at PricewaterhouseCoopers, Rochester, NY, for informative
discussions about the audit process and, in particular, the steps that auditors take when a material
misstatement is identified.
6There was one instance of an adverse audit opinion and six instances of audit opinions qualified for
departure from GAAP. We thank Mark Nelson for providing this information, which is not reported in
Nelson et al. (2002).
M. Butler et al. / Journal of Accounting and Economics 37 (2004) 139–165 143
3. Web-based sample of modified audit opinions
3.1. Compustat’s audit data item, before and after SAS58
We begin with the merged Compustat database (i.e., full coverage, industrial
annual, and research) for the period from 1980 to 1999 and identify147,926 firmyear
observations for which the auditor/auditor’s opinion data item (x149) is not
missing.7 Compustat uses this data item to designate a company’s audit firm—
companies with Big5 firms comprise 80.1% of the sample—and the type of audit
opinion the companyrecei ves: unaudited (0), unqualified (1), qualified (2),
disclaimer/no opinion (3), unqualified with explanatorylanguage (4), or adverse
(5). The majorityof the sample (77.3%) is unqualified opinions.
Several changes in auditing standards have influenced the format of audit reports
and, consequently, Compustat’s classification of audit opinions. Most importantly,
SAS 58, effective for reports issued after January1,1989 , changes the conditions
under which ‘‘qualified’’ opinions are issued. Prior to SAS 58, there were two broad
classes of qualified audit opinions: ‘‘except for’’ opinions were issued when there
were scope limitations, departures from GAAP, or a lack of consistency(e.g ., a
change from one generallyaccep ted accounting method to another); and ‘‘subject
to’’ opinions were issued in the presence of material uncertainties, such as asset
realization, contingent liabilities, or substantial doubt concerning an entity’s ability
to continue as a going concern. Under SAS 58, changes from one acceptable
accounting method to another and everyty pe of ‘‘subject to’’ opinion no longer give
rise to ‘‘qualified’’ opinions. Instead, under present-dayGAA S, there is now a
categoryfor ‘‘unqualified opinions with explanatorylanguag e’’ that discusses either
the accounting change or material uncertainty. Consequently, only the format and
not the information contained in the audit opinion has changed.
As a result, prior to SAS 58, a Compustat ‘‘qualified’’ opinion (Compustat
opinion code 2) related to both ‘‘except for’’ as well as ‘‘subject to’’ audit opinions.
Under the SAS 58 regime, however, Compustat code 2 opinions include only
qualifications for scope limitations and departures from GAAP. Other modifications,
such as changes from one generallyaccep ted accounting method to another
and material uncertainties, should now be classified as unqualified opinions with
explanatorylanguage (Compustat code 4).
Table 1 presents Compustat opinion codings for the 1980–1999 period,
categorized bySAS 58 time frame and audit firm type. Because our sample includes
both pre-SAS 58 years (1980–1987), as well as years in which SAS 58 is effective
(1988–1999), we present annualized counts to aid comparability. As expected, the
most significant shift across the two time periods is the frequencyof qualified
ARTICLE IN PRESS
7Compustat creates an observation for the entire 20-year period for each firm present in any of the years
covered. So, if Compustat began including a firm in 1990, Compustat adds records containing missing
values for the firm for each year from 1980 to 1989. To determine the number of valid firm-year
observations, we count onlyobservatio ns with non-missing values for both total assets and audit opinion
data items (x6 and x149, respectively).
144 M. Butler et al. / Journal of Accounting and Economics 37 (2004) 139–165
opinions (code 2) and unqualified opinions with explanatorylangu age (code 4).
Looking at Big5-audited companies, 11.1% had qualified opinions in the pre-SAS
regime versus only0 .2% in the SAS 58 regime. In contrast, there were virtually no
Big5-audited companies with Compustat code 4 (i.e., unqualified with explanatory
language) opinions in the pre-SAS 58 regime, but 25.6% of such companies received
these opinions in the SAS 58 regime. Notably, in the SAS 58 period, we detect no
meaningful distinction between Compustat’s code 2 and code 4 opinions. That is, we
identifycompanie s with qualified opinions that have code 4 (i.e., unqualified opinion
with explanatorylanguage ) classifications, as well as companies with code 2 (i.e.,
qualified opinion) classifications that have unqualified opinions with explanatory
language. Overall patterns are similar for non-Big5 firms. The differences we report
in the pre-SAS 58 and SAS 58 periods likelyhave implications for the comparability
of our findings to past research, which we examine in Section 5.4.
3.2. Tracking Compustat’s audit data to EDGAR source documents via the Web
To independentlyverifyand further refine Compustat’s audit opinion classifications,
we construct a large database of modified opinions. To do this, we develop a
Web application that automates the process of collecting and classifying audit
opinions. The application searches EDGAR for the 10-K corresponding to a given
companyname and fiscal year-end. Obtaining opinions from EDGAR reduces data
collection costs; however, it also limits the time period we cover from 1980–1999 to
1994–1999. Using keywords to identify the firm’s audit opinion, the program
ARTICLE IN PRESS
Table 1
Average annual Compustat audit opinion code types by SAS 58 regime and audit firm type (1980–1999)a
Opinion
code
Compustat
opinion type
Pre-SAS 58 (1980–1987) SAS 58 (1988–1999)
Big5 Non-Big5 Big5 Non-Big5
0 Unaudited 2 (0.05%) 1 (0.08%) 1 (0.01%) 2 (0.11%)
1 Unqualified 4,277 (88.22%) 1,271 (79.91%) 4,911 (73.98%) 912 (65.27%)
2 Qualified 540 (11.14%) 293 (18.44%) 12 (0.18%) 17 (1.24%)
3 Disclaimer/no opinion 27 (0.56%) 23 (1.47%) 16 (0.25%) 7 (0.51%)
4 Unqualified with
explanatorylanguage
1 (0.03%) 2 (0.10%) 1,698 (25.58%) 459 (32.84%)
5 Adverse 0 (0.00%) 0 (0.00%) 0 (0.00%) 0 (0.03%)
Totals 4,847 (100.00%) 1,590 (100.00%) 6,638 (100.00%) 1,397 (100.00%)
a This table reports the frequencydistribution of audit opinions by Compustat classification, auditor
type, and pre- and post-SAS 58 implementation. Under SAS 58, ‘‘subject to’’ and ‘‘except for’’ consistency
opinions are no longer considered qualified opinions. Instead, these situations give rise to unqualified
opinions with explanatorylanguage. Audit firm/audit opinion data is from the merged Compustat
database (i.e., full coverage, industrial annual, and research) for the 20-year period from 1980 to 1999
(Compustat data item #149). Because the pre-SAS 58 regime contains 8 years of observations and the SAS
58 regime contains 12 years, we present annualized information to facilitate comparisons.
M. Butler et al. / Journal of Accounting and Economics 37 (2004) 139–165 145
extracts the opinion from the 10-K and copies it to a database. If the 10-K search
fails to find the opinion, the program searches other potentiallyrelev ant SEC filings
(e.g., S-l). In addition to facilitating collection of a large sample of audit opinions,
the application provides a systematic and efficient method (using keyword
algorithms) of categorizing opinions.8
Table 2 summarizes sample construction. Using the Web application, we attempt
to collect and classifythe 12,588 modified audit opinions of firms categorized by
Compustat from 1994 to 1999 as having either a qualified opinion (code 2) or an
unqualified opinion with explanatorylanguage (code 4). Of the 12,588 audit
opinions, 900 relate to utilities and financial firms that we eliminate because of
inherent differences associated with accounting accruals for these firms. Approximately7
3% (8,478) of the remaining 11,688 opinions are usable, as we discard 3,210
data records because we are unable to find either the filing or audit opinion on
EDGAR. Interestingly, given the criteria we use to select opinions of interest, we also
eliminate another 1,385 observations representing essentiallycle an opinions;9 that is,
approximately16% (1,385C8,478) of the opinions classified byComp ustat as
modified opinions (code-2 and code-4) from 1994 to 1999 are actuallycle an
opinions.10 It seems that Compustat simplyflags all audit opinions as ‘‘modified’’ if
theydev iate in any way from a standard ‘‘boiler plate’’ clean opinion. In summary,
using a Web-enabled, data collection routine, we collect and reliablycla ssify7,0 93
modified audit opinions for the 1994–1999 period (i.e., approximately1,2 00 per year).
4. Content analysis of modified audit opinions
Table 3 provides a content analysis, by audit firm type, of modified audit opinions.
Following GAAS, we broadlycategor ize opinions into two categories (qualified
opinions and unqualified opinions with explanatorylanguage ), which we then
partition into subcategories. The content analysis serves two purposes. First, it
provides a large-sample analysis of the reasons underlying auditors’ decisions to
depart from a standard, unqualified opinion. Second, it allows a detailed reexamination
of the audit opinion/accounting accruals relation.
4.1. Reasons for departures from a standard unqualified audit report
Table 3 provides several insights into auditor reporting. Audit opinion
qualifications occur much less frequentlythan unqualified opinions with explanatory
ARTICLE IN PRESS
8 Specific keywords used for classification are available from the authors.
9 ‘‘Clean’’ opinions either contain no qualifying or explanatory language or contain explanatory
language that is fundamentallydifferent from that contained in other opinions. Explanatory language in
the latter opinion types was often trivial (e.g., 801 observations noted, without further qualification, that
ancillaryschedules were audited in addition to financial statements; 33 observations noted the audited
companywas a development stage enterprise).
10We also review a large number of opinions classified as ‘‘clean’’ to verifythat these are, in fact, clean
opinions.
146 M. Butler et al. / Journal of Accounting and Economics 37 (2004) 139–165
language. In addition, almost half of all modifications during our sample period are
GC opinions. Note that we classifysome opinions that do not explicitly state ‘‘going
concern’’ as GC opinions if the opinion’s wording mentions severe business distress.
There were 251 (3,161 minus 2,910) cases where the auditors refer to either
bankruptcyor the presence of material business uncertainties without explicit
reference to the term ‘‘going concern.’’ None of the results we report later, however,
are sensitive to the exclusion of opinions in which the auditor does not explicitlyuse
the term ‘‘going concern.’’ Consistent with claims that the largest audit firms are
shedding riskier clients (Berton, 1995), we document a lower frequencyof GC
opinions for companies audited byBig 5 firms. In addition, ‘‘material uncertainty’’
opinions are infrequent, probablybecau se most of our sample is drawn from the
time period for which SAS 79 is effective.11 SAS 79 eliminates the requirement that,
ARTICLE IN PRESS
Table 2
Sample selection for content analysis and accruals analysis (1994–1999)
The sample is obtained by identifying all firm-year observations with Compustat audit opinion codes of
either 2 (i.e., qualified) or 4 (i.e., unqualified with explanatorylanguage) for the 6-year period from 1994 to
1999. After eliminating 900 observations relating to utilities and financial services firms, we attempt to
identifythe actual audit opinion on the SEC’s EDGAR database. Of the 11,688 firm-year observations
(i.e., 12,588 minus 900), we find 8,478 (i.e., 11,688 minus 3,210, or approximately73%) on EDGAR. Of
these 8,478, we classified 1,385 firm-year observations (i.e., 801 plus 33 plus 551) as ‘‘clean’’ opinions
(Note: The opinions we characterize as ‘‘clean’’ either (a) contain no qualifying or explanatory language or
(b) contain trivial explanatorylanguage fundamentally different from that contained in other collected
opinions. In other words, these opinions do not contain explanatorylanguage that is consistent with
GAAS for designation as an unqualified opinion with explanatorylanguage). We perform a content
analysis (see Table 3) using the resulting 7,093 firm-year observations, by audit firm type, to identify the
specific reason the audit opinion was either qualified or unqualified with explanatorylanguage. We then
perform an analysis of the association between modified audit opinions (per GAAS) and abnormal
accounting accruals using the 4,205 firm-year observations (of the 7,093) with adequate Compustat
information (see Tables 4–7).
11SAS 79 is effective for reports issued on or after February29, 1996, but earlier application was
encouraged.
M. Butler et al. / Journal of Accounting and Economics 37 (2004) 139–165 147
ARTICLE IN PRESS
Table 3
Content analysis of modified audit opinions by audit firm type (1994–1999)
Observations are the 7,093 firm years in Table 2 that, guided byGAAS, we categorize as either ‘‘qualified
opinions’’ or ‘‘unqualified opinions with explanatorylanguage. ’’ We then partition these two categories
into subcategories based on the underlying reason for the opinion modification.
148 M. Butler et al. / Journal of Accounting and Economics 37 (2004) 139–165
when certain criteria are met relating to uncertainties, the auditor’s report must
include an explanatoryparagra ph. Under SAS 79, auditors mayadd an emphasis of
matter paragraph to highlight an uncertaintythat is alreadyapp ropriately disclosed,
but evidentlyfew opinions contain such paragraphs. Lastly, and due largelyto
changes in GAAP (e.g., deferred taxes, OPEB), we document a high percentage of
opinions that refer to accounting changes not involving restatements. Approximately
48% of our sample firms with modified opinions report an accounting change;
however, most of these opinions refer to the adoption of a new standard rather than
to a change from one generallyaccepte d method to another.
4.2. A simple test of the relation between certain audit opinions and accruals
While auditors likelyplayan important role in monitoring and controlling
managers’ accrual choices, the content analysis provides little support for the
rationale that audit opinions commonlyconvey the presence of earnings management.
Bartov et al. (2000) assert that qualifications for scope limitations or GAAP
departures proxyfor extreme earnings management.12 However, our content
analysis identifies only 19 and 14 instances of opinion qualifications for scope
limitation and GAAP departures, respectively, from 1994 to 1999.13 We also
identify147 unqualified opinions with explanatory language referencing a
‘‘permissible’’ departure from GAAP, meaning that the SEC will accept the
financial statements. Of these opinions, 69 refer to the use of international
accounting standards, 75 pertain to trusts using a modified cash basis of accounting,
and three refer to the auditing of supplemental schedules that do not conform to
GAAP.14
Thus, the majorityof opinions qualified for either scope or departure from GAAP
bear no discernible connection to extreme earnings management bythe client.
Moreover, we identifyno modified opinions that refer to the future consequences of
ARTICLE IN PRESS
12 Bartov et al. (2000) include all firms receiving qualified opinions (code 2 on Compustat). However, the
time period theystudyspans the pre-SAS 58 period. At that time, a qualified opinion could be issued due
to scope limitations, departures from GAAP, lack of consistency, going concern, asset realization and
contingent liabilities. Therefore, while theycorrectlyused code 2 to identifyqualified opinions, the
definition of a qualified opinion during much of their study-period is different from the definition in the
SAS 58 period. Because the code 2 opinion code is used infrequentlyafter 1988 (see our Table 1), it is likely
that most of their sample came from the pre-SAS 58 time period. In addition, Bartov et al. describe in their
paper that Compustat Code 3 is assigned when a companyreceives a going concern opinion. However,
code 3 is assigned when the auditor chooses not to issue an opinion. This can happen if going concern
issues are extremelysevere, but in most cases the auditor modifies the opinion for going concern opinions
(code 2 in the pre-SAS 58 regime and code 4 in the SAS 58 regime).
13The 14 qualified opinions mentioning GAAP departures involve three companies. One companyelects
not to consolidate certain subsidiaries; the other two do not recognize deferred taxes in accordance with
US GAAP. Of these 33 qualified opinion observations, 17 have adequate Compustat information to
compute cross-sectional Jones (1991) model abnormal accruals. Of these 17, 10 (7) observations have
positive (negative) abnormal accruals as per Eq. (1) (see footnote 16). The mean for these 17 observations
is positive, though not significantlydifferent from zero.
14 Financial statements prepared in accordance with foreign GAAP are acceptable onlyif reconciled to
US GAAP. A modified cash basis of accounting is permissible for trusts.
M. Butler et al. / Journal of Accounting and Economics 37 (2004) 139–165 149
high positive accruals (Bradshaw et al., 2001). As such, our findings do not support
the underlying rationale for the opinion/accruals relation put forth by Bartov et al.
(2000). Theyare, however, consistent with one explanation given by Bradshaw et al.
(2001) for the unexpected negative relation theyfind between modified opinions
and working capital accruals, that reporting on the qualityof earnings is beyond the
scope of the audit report. Nevertheless, even though earnings management does not
explain the abnormallylarge negative accruals of modified-opinion firms, the
underlying cause of the opinion/accruals relation remains unidentified. In the
following section we investigate the source of this relation.
5. Accruals analysis of modified audit opinions
5.1. Descriptive statistics by audit opinion type
Table 4 reports mean and median statistics for firm-year observations from
1994 to 1999 with adequate Compustat information to compute the variables
of interest. We divide the observations into five audit opinion categories: unqualified,
going concern, material uncertainty, accounting change (without restatements),
and other.15 We provide information regarding total and abnormal accruals
per Jones (1991).16 Following Hribar and Collins (2002), we use the statement of
cash flows approach in our tables, although our findings are robust to using the
balance sheet approach to obtain total accruals (see also Section 5.4). Table 4 also
provides information for several of the independent variables in our accruals
regressions.
Firms in the GC categorydiffer on almost everydimension from those in other
categories. For example, firms with GC opinions tend to be smaller, less profitable,
less liquid, and to have more negative accruals. Highlighting the poor performance
of GC companies (e.g., mean and median lagged ROA is 0.60 and 0.32,
respectively) is the fact that over 85% of these observations report a loss in the year
prior to the GC. The peculiarities of companies with GC opinions reported in Table
4 are borne out in the multivariate analysis that follows.
ARTICLE IN PRESS
15The sum of observations in modified-opinion categories in Table 4 (5,122) exceeds the total number of
modified opinions in Table 2 (4,205) because some firms have modifications for multiple reasons (e.g., an
opinion referencing GC status and an accounting change would be classified as both GoingConcern and
OtherModified).
16 Following DeFond and Jiambalvo (1994), we estimate a cross-sectional version of the Jones (1991)
model for each (two-digit SIC) industryand year. After winsorizing at the 1st and 99th percentiles, we take
the residuals from the following regressions as our measures of abnormal accruals, which we denote as
AbnormalAccruals:
TotalAccrualsi;t ¼ a0 þ a1 DSalesi;t þ a2PPEi;t þ ei;t; ð1Þ
where TotalAccruals=earnings before extraordinaryitems and discontinued operations minus operating
cash flows from continuing operations (Compustat 123(308124)Ctotal assets (beginning of year), per
Hribar and Collins (2002)); D Sales=change in sales from year t1 to year tCtotal assets (beginning of
year); PPE=net property, plant and equipment as of year tCtotal assets (beginning of year).
150 M. Butler et al. / Journal of Accounting and Economics 37 (2004) 139–165
5.2. Regression analysis—full sample
We begin byregres sing absolute abnormal accruals on control variables and audit
opinion type. This is essentially the reverse of the Bartov et al. (2000) specification in
which audit opinion type (qualified or unqualified) is the dependent variable and
absolute abnormal accruals is an independent variable (along with control
variables). We estimate the model this waybecause institutional characteristics of
ARTICLE IN PRESS
Table 4
Descriptive statistics byaudit opinion category(1994–1999)
Variable Unqualified
opinions
Going
concern
opinions
Material
uncertainty
opinions
Accounting
change
opinions
Other
modified
opinions
TotalAccrualst Mean 0.06 0.27 0.10 0.06 0.15
Median 0.05 0.20 0.04 0.05 0.10
AbnormalAccrualst Mean 0.01 0.14 0.03 0.02 0.06
Median 0.02 0.10 0.03 0.03 0.01
MktCapt Mean 2.06 1.10 1.84 2.30 1.65
Median 2.01 1.12 1.78 2.25 1.54
Book/Mktt Mean 0.57 0.31 0.49 0.61 0.46
Median 0.44 0.20 0.40 0.52 0.38
ROAt1 Mean 0.03 0.60 0.08 0.01 0.27
Median 0.04 0.32 0.03 0.04 0.03
Debt/Assetst Mean 0.17 0.19 0.17 0.21 0.18
Median 0.11 0.07 0.12 0.18 0.11
CurrRatiot Mean 2.89 1.37 2.29 2.33 1.84
Median 2.03 0.79 1.76 1.74 1.29
Assetst Mean 920.75 91.02 530.16 1,579.02 561.54
Median 86.99 9.90 49.30 249.87 40.81
Big5t Mean 0.85 0.55 0.70 0.93 0.69
Observations 22,653 1,823 105 2,509 685
Observations are those firm-years from 1994 to 1999 for which adequate Compustat information is
available. For the modified audit opinion categories, observations are the 4,205 in Table 2; although,
instances of multiple classifications are now reflected in the above columns. These modified audit opinion
categories correspond to the categories in Table 3, with the following exceptions: Accounting Change
Opinion observations relate to the Consistency—non-restatement category in Table 3, and the other
opinion observations come from the qualified, consistency—restatements, reliance on another auditor, and
related partytransaction categories in Table 3.
Variables are defined as follows:
TotalAccruals=earnings before extraordinaryitems and discontinued operations minus operating cash
flows from continuing operations (Compustat 123–(308–124)Ctotal assets (beginning of year), per Hribar
and Collins (2002)).
AbnormalAccruals=cross-sectional Jones (1991) abnormal accruals (Eq. (1)).
MktCap =Log(Market value of equity).
Book/Mkt=book value of equityCmarket value of equity.
ROA=income before extraordinaryitems Ctotal assets (beginning of year).
Debt/Assets=long-term debtCtotal assets (beginning of year).
CurrRatio=current assetsCcurrent liabilities.
Assets=total assets (beginning of year).
Big5=one if a Big5 audit firm and zero otherwise.
M. Butler et al. / Journal of Accounting and Economics 37 (2004) 139–165 151
the audit process and our content analysis do not support a causal relation in which
earnings management leads to an audit opinion modification.17 Consistent with
Bartov et al., we control for market capitalization, book-to-market, profitability,
and leverage. Due to the characteristics of companies with GC opinions (see
Table 4), we also control for liquidity( CurrRatio). In addition, to consider the nonlinear
relation between performance and abnormal accruals, as described by Kothari
et al. (2003), we include ROA2
t1 in addition to ROAt1: 18 To lessen the effect of
outliers, we winsorize extreme observations for all variables bysetting the values in
the bottom and top 1% of observations of each measure to the values of the 1st and
99th percentiles of their respective distributions.19
jAbnormalAccrualsji;t ¼b0 þ b1 MktCapi;t þ b2 Book=Mkti;t þ b3 ROAi;t1
þ b4 ROA2
i;t1 þ b5 Debt=Assetsi;t þ b6 CurrRatioi;t
þ b7 Big5i;t þ b8 Modifiedi;t þ ei;t; ð2aÞ
where
AbnormalAccruals =cross-sectional Jones (1991) abnormal accruals (Eq. (1)).
MktCap =Log10 (Market value of equity).
Book/Mkt =book value of equityCmarket value of equity.
ROA =lagged income before extraordinaryite msCtotal assets (beginning
of year).
ROA2 =square of ROA.
Debt/Assets =long-term debtCtotal assets (beginning of year).
CurrRatio =current assetsCcurrent liabilities.
Big5 =one if a Big5 audit firm and zero otherwise.
Modified =one if opinion is qualified or unqualified with explanatory
language and zero otherwise.
To consider the effects of the specific type of audit opinion modification, we
augment Eq. (2a) byreplac ing the Modified variable with variables corresponding to
the audit opinion categories in Table 4:
jAbnormalAccrualsji;t ¼b0 þ b1 MktCapi;t þ b2 Book=Mkti;t þ b3 ROAi;t1
þ b4 ROA2
i;t1 þ b5 Debt=Assetsi;t þ b6 CurrRatioi;t
þ b7 Big5i;t þ b8 GoingConcerni;t
þ b9 MatUncerti;t þ b10 AcctChangei;t
þ b11 OtherModifiedi;t þ ei;t; ð2bÞ
ARTICLE IN PRESS
17 Following Bartov et al. (2000) we also estimate logit models. Inferences are unchanged (see
Section 5.4).
18The inclusion of this variable, the squared of lagged ROA, has no effect on the paper’s inferences.
19As an additional sensitivitytest, we exclude outliers per Belsleyet al. (1980). This data trimming
increases the regression’s explanatorypower but, again, has no effect on inferences.
152 M. Butler et al. / Journal of Accounting and Economics 37 (2004) 139–165
where
GoingConcern =one if opinion is unqualified with explanatorylanguage pertaining
to going concern status, bankruptcy, or material business
uncertainty, and zero otherwise.
MatUncert =one if opinion is unqualified with explanatorylanguage pertaining
to a material uncertainty(e.g., asset realization or contingent
liability), and zero otherwise.
AcctChange =one if opinion is unqualified with explanatorylanguage pertaining
to lack of consistencynot requiring restatement, and zero
otherwise.
OtherModified =one if opinion is qualified (scope limitations or GAAP departure)
or unqualified with explanatorylanguage pertaining to lack of
consistencyrequiring restatement or reliance on another auditor or
related partytransacti ons and zero otherwise.
The first two columns of Table 5 present the results of these regressions. Results in
the first column, where the coefficient onModified is positive and significant, support
Bartov et al.’s (2000) contention that modified audit opinions are positively
associated with the magnitude of abnormal accounting accruals. However, results in
column two demonstrate that the positive coefficient on Modified stems primarily
from companies with GC opinions. Companies with material uncertaintyopinions
also have larger absolute abnormal accruals. However, as Table 3 shows, most
material-uncertaintyopinions arise from litigation risk, suggesting that incomeincreasing
earnings management is unlikely.
Next, given the findings of Bradshaw et al. (2001) that firms with modified
(unclean) opinions have more negative working capital accruals, we change the
dependent variable in Eqs. (2a) and (2b) from absolute abnormal accruals to signed
abnormal accruals:
AbnormalAccuralsi;t ¼b0 þ b1 MktCapi;t þ b2 Book=Mkti;t þ b3 ROAi;t1
þ b4 ROA2
i;t1 þ b5 Debt=Assetsi;t þ b6 CurrRatioi;t
þ b7 Big5i;t þ b8 Modifiedi;t þ ei;t; ð3aÞ
AbnormalAccrualsi;t ¼b0 þ b1 MktCapi;t þ b2 Book=Mkti;t þ b3 ROAi;t1
þ b4 ROA2
i;t1 þ b5 Debt=Assetsi;t þ b6 CurrRatioi;t
þ b7 Big5i;t þ b8 GoingConcerni;t þ b9 MatUncerti;t
þ b10 AcctChangei;t þ b11 OtherModifiedi;t þ ei;t: ð3bÞ
The last two columns of Table 5 present the results of these regressions. As column
three shows, companies with modified opinions have significantlymore negative
abnormal accruals than companies with unqualified opinions. Column four shows
that the negative coefficient on Modified stems from firms with GoingConcern
opinions. Therefore, even after controlling for other factors, the association between
ARTICLE IN PRESS
M. Butler et al. / Journal of Accounting and Economics 37 (2004) 139–165 153
modified opinions and abnormal accruals is due to the large negative (i.e., incomedecreasing)
accruals of companies with GC opinions. We also find that MatUncert
(AcctChange) firms have more negative (positive) abnormal accruals than firms with
ARTICLE IN PRESS
Table 5
OLS on absolute value and signed abnormal accruals (full sample)
jAbnormalAccrualsi;tj ¼b0 þ b1 MktCapi;t þ b2 Book=Mkti;t þ b3 ROAi;t1 þ b4 ROA2
i;t1 þ b5 Debt=Assetsi;t
þb6 CurrRatioi;t þ b7 Big5i;t þ b8 Modifiedi;t þ ei;t ; ð2aÞ
jAbnormalAccrualsi;tj ¼b0 þ b1 MktCapi;t þ b2 Book=Mkti;t þ b3 ROAi;t1 þ b4 ROA2
i;t1 þ b5 Debt=Assetsi;t
þ b6 CurrRatioi;t þ b7 Big5i;t þ b8 GoingConcerni;t þ b9 MatUncerti;t
þ b10 AcctChangei;t þ b11 OtherModifiedi;t þ ei;t ; ð2bÞ
AbnormalAccrualsi;t ¼b0 þ b1 MktCapi;t þ b2 Book=Mkti;t þ b3 ROAi;t1 þ b4 ROA2
i;t1 þ b5 Debt=Assetsi;t
þ b6 CurrRatioi;t þ b7 Big5i;t þ b8 Modifiedi;t þ ei;t ; ð3aÞ
AbnormalAccrualsi;t ¼b0 þ b1 MktCapi;t þ b2 Book=Mkti;t þ b3 ROAi;t1 þ b4 ROA2
i;t1 þ b5 Debt=Assetsi;t
þ b6 CurrRatioi;t þ b7 Big5i;t þ b8 GoingConcerni;t þ b9 MatUncerti;t
þ b10 AcctChangei;t þ b11 OtherModifiedi;t þ ei;t : ð3bÞ
Variables Eq. (2a) Eq. (2b) Eq. (3a) Eq. (3b)
Constant 0.204 0.191 0.024 0.046
(73.98)a (67.12)a (6.72)a (12.29)a
MktCap 0.025 0.022 0.003 0.008
(26.58)a (23.09)a (2.25)b (6.22)a
Book/Mkt 0.033 0.030 0.007 0.013
(25.17)a (22.50)a (4.10)a (7.29)a
ROA 0.074 0.061 0.129 0.108
(21.12)a (17.26)a (28.01)a (23.32)a
ROA2 0.011 0.009 0.025 0.022
(9.12)a (7.24)a (15.87)a (13.59)a
Debt/Assets 0.054 0.051 0.012 0.008
(13.89)a (13.24)a (2.39)b (1.53)
CurrRatio 0.033 0.030 0.006 0.005
(25.17)a (22.50)a (16.66)a (14.53)a
Big5 0.003 0.002 0.006 0.012
(11.46)a (9.66)a (1.91)c (4.19)a
Modified 0.014 0.039
(7.03)a (14.48)a
GoingConcern 0.060 0.113
(18.07)a (25.79)a
MatUncert 0.024 0.041
(2.05)a (2.64)a
AcctChange 0.014 0.008
(5.49)a (2.36)a
OtherModified 0.005 0.007
(1.13) (1.16)
Observations 27,219 27,219 27,219 27,219
Adjusted R2 12.2% 13.2% 6.5% 8.2%
154 M. Butler et al. / Journal of Accounting and Economics 37 (2004) 139–165
clean opinions, although the association between abnormal accruals and these
opinion types is not nearly as economically nor statistically significant as the GCaccrual
association.
Overall, these findings suggest that Bartov et al.’s (2000) results stem from
firms with GC opinions. That is, the positive relation between modified opinions
and the magnitude of accruals relates primarilyto companies receiving GC
opinions, which have large negative accruals. These findings are also inconsistent
with the explanation for the opinion/accruals relation in Francis and Krishnan
(1999). Francis and Krishnan argue that larger, higher-qualityaud itors assess highaccrual
firms as inherentlymore risky—because managers of such firms are more
likelyto have exercised discretion—and, as a result, adjust opinion materiality
thresholds downward. However, this implies that the signed abnormal accruals
regressions for firms with modified opinions should be positive, not negative as our
results show.
These findings highlight the importance of testing both signed and absolute
abnormal accruals. If, as in this setting, absolute abnormal accruals are dominated
byone side of the distribution (i.e., negative or positive abnormal accruals),
inferences likelydiffer from the case in which absolute abnormal accruals are larger
but symmetric.
ARTICLE IN PRESS
t-Statistics are in parentheses. Observations are all firm-year observations from 1994 to 1999 with
adequate Compustat information.
Variables are defined as follows:
AbnormalAccruals=cross-sectional Jones (1991) abnormal accruals (Eq. (1)).
MktCap=Log10(Market value of equity).
Book/Mkt=book value of equityCmarket value of equity.
ROA=income before extraordinaryitems Ctotal assets (beginning of year).
ROA2=square of ROA.
Debt/Assets=long-term debtCtotal assets (beginning of year).
CurrRatio=current assetsCcurrent liabilities.
Big5=one if a Big5 audit firm and zero otherwise.
Modified=one if opinion is qualified or unqualified with explanatorylanguage and zero otherwise.
GoingConcern=one if opinion is unqualified with explanatorylanguage pertaining to going-concern
status, bankruptcy, or material business uncertainty and zero otherwise.
MatUncert=one if opinion is unqualified with explanatorylanguage pertaining to a material uncertainty
(e.g., asset realization or contingent liability) and zero otherwise.
AcctChange=one if opinion is unqualified with explanatorylanguage pertaining to lack of consistency not
requiring restatement and zero otherwise.
OtherModified=one if opinion is qualified (scope limitation or GAAP departure) or unqualified with
explanatorylanguage pertaining to lack of consistency requiring restatement or reliance on another
auditor or related partytransaction s and zero otherwise.
a Significant beyond the 1% level (two-sided tests).
b Significant beyond the 5% level (two-sided tests).
c Significant beyond the 10% level (two-sided tests).
Table 5 (Continued)
M. Butler et al. / Journal of Accounting and Economics 37 (2004) 139–165 155
5.3. Why do GC firms have extremely negative abnormal accruals?
Several studies show that abnormal accruals are non-zero for firms with extreme
performance (e.g., Dechow et al., 1995; Kothari et al., 2003). In particular, abnormal
accruals are more negative (positive) for firms with extreme poor (good)
performance. These studies show that accrual models tend to overreject the null of
zero abnormal accruals. Given that financial distress is essentiallya necessary
condition for observing a GC opinion (Hopwood et al., 1994) and that certain
financiallytroubl ed firms have large negative accruals (DeAngelo et al., 1994),
companies with GC opinions are likelyto have more extreme negative accruals. That
is, GC companies likelyha ve negative accruals due to poor performance.
For example, financiallydistress ed firms often face liquidityco nstraints that force
them to reduce non-cash net working capital to survive. It is also likelythat these
firms have non-performing assets that theyare required to write off under GAAP.
To the extent that conventional accrual models do not account for the accrual
consequences of distress due to liquidity-related transactions, these accruals will be
misclassified as discretionary. This misspecification is likely a significant source of
the association between modified opinions and conventional estimates of abnormal
accruals.
As Table 4 shows, there are marked differences between companies with GC
opinions and other firms, whether theyrecei ve other types of modified opinions or
standard, unqualified opinions. In particular, GC firms are experiencing extremely
poor performance. To determine whether the abnormal accruals of GC firms are
reliablynegative we follow Kothari et al.’s (2003) performance-matching
approach.20 In a studyof the statistical properties of discretionaryaccrua l models,
Kothari et al. advocate matching to control for the impact of firm performance on
accruals. Theydemonst rate that performance matching leads to accrual-based tests
of earnings management that are well specified and powerful.
Given our unique sample, we consider two alternative measures of performance
for identifying a matched sample. First, as suggested by Kothari et al. (2003),
we use ROAt1 (i.e., lagged return on assets, or the ratio of prior-year income
before extraordinaryitem s to total assets). Second, we use the probabilityof
bankruptcyco nstructed from Shumway(2001) , which we label Distress. As
mentioned, GC firms tend to have more severe liquidityproblem s than other
firms. To survive, the GC firms are likelyto engage in liquidity-enhancing
transactions which lead to large negative accruals but which do not necessarily
affect ROA (e.g., delaying payments to suppliers or factoring receivables).21 It is
therefore important to control for differences in liquidityas well as profitability.
ARTICLE IN PRESS
20 Conditioning empirical analysis on financial distress is common in research on GC opinions (e.g.,
McKeown et al., 1991; Hopwood et al., 1994; Mutchler et al., 1997; Louwers, 1998).
21 Factoring receivables will likelyhave some impact on income but most of the related accrual will not.
For example, suppose a firm factors receivables with a book value of $100 at a price of $90. The firm will
record a negative accrual of $100 but only$10 of it will be reflected on the income statement.
156 M. Butler et al. / Journal of Accounting and Economics 37 (2004) 139–165
Consistent with this reasoning, Distress is more effective than ROAt1 for matching
in our sample.22
Our matching procedure is as follows. We construct a dataset from the merged
Compustat database that contains firms with unqualified opinions (without
explanatorylanguage ) and sufficient data to compute our performance measures,
abnormal accruals, and the regression covariates. Then, for each newly initiated GC
opinion observation (e.g., a GC opinion in year t, but not in t1), we identifythat
observation in the matched group with the closest performance (i.e., either ROAt1
or Distress). In other words, control firms are matched on year, industry, auditor
type, and a measure of performance. This procedure yields a sample size of 1,052:
526 firms with first-time GC opinions and 526 control firms with standard,
unqualified opinions. Using this matched sample, we estimate the following
regression:
AbnormalAccrualsi;t ¼b0 þ b1 MktCapi;t þ b2 Book=Mkti;t
þ b3 ROAi;t1 þ b4 ROA2
i;t1
þ b5 Debt=Assetsi;t þ b6 CurrRatioi;t
þ b7 Distressi;t þ b8 Big5i;t þ b9 GoingConcerni;t
þ b10 Big5GoingConcerni;t þ ei;t; ð4Þ
where
Distress =bankruptcyprob abilityper Shumway’s (2001) Table 4b hazard
model (1(1/exp(7.8116.307Net Income/Total Assets+
4.068Total Liabilities/Total Assets0.158Current Assets/
Current Liabilities+0.307Ln (CompanyAge))) .
Big5
Going Concern
=one if a Big5 audit firm issued the GoingConcern opinion and
zero otherwise.
We add the term interacting Big5 and GoingConcern opinions to consider an
alternative explanation to performance. An alternative explanation for companies
with GC opinions having large negative accruals is that auditors, in response to
heightened litigation risk, impose a higher level of conservatism on companies to
which theyrender GC opinions (DeFond and Subramanyam, 1999). Given that
larger audit firms are perceived to provide increased coverage (i.e., deeper pockets) in
the event of securities litigation (Arthur Andersen et al., 1992), one test of this
explanation is to examine whether income-decreasing accruals varybyaudit firm
type. Evidence that abnormal accruals are more negative for GC companies audited
bythe Big5 than for those audited by the non-Big5 would support the auditor
conservatism explanation. If accruals are not different for Big5 versus non-Big5
audited GC companies or if accruals are more negative for GC companies audited by
ARTICLE IN PRESS
22The mean and median values of ROA, current ratio, book-to-market and size of the sample matched
on distress are much closer to the GC sample than the sample matched on lagged ROA.
M. Butler et al. / Journal of Accounting and Economics 37 (2004) 139–165 157
the non-Big5, then the relation between accruals and troubled firms is less likelydue
to auditor conservatism.
Column one of Table 6 provides the results of estimating Eq. (4) for this sample of
companies with first-time GC opinions and a control sample matched on year,
auditor, industry, and ROAt1: As in column four of Table 5, the coefficient for
GoingConcern is negative and statisticallysigni ficant (beyond 0.001).23 That is, firms
with GC opinions have negative abnormal accruals compared to a matched sample
of firms with similar prior-year ROA, even after controlling for other factors
associated with abnormal accruals and GC opinions. The coefficient for
Big5
GoingConcern is positive and thus does not support a differential auditor
conservatism explanation. Of course, it is possible that auditor conservatism is
important, but equallyso across audit firm types. As with the Table 5 regression,
inferences are robust to considering the effects of influential observations per Belsley
et al. (1980).24
Column two of Table 6 reports the results of estimating Eq. (4) using the
alternative control sample matched on Distress (Shumway, 2001). For this
regression, the coefficient on GoingConcern is not significantlydiff erent from zero,
suggesting that the GC opinion itself is neither the cause nor the result of abnormal
accruals. Instead, firms in financial distress likelyha ve more negative accruals due to
poor performance and transactions arising to meet liquidityneed s (e.g., delaying
payables and factoring receivables). Again, the coefficient on Big5
GoingConcern is
positive. This is further evidence that auditor conservatism does not explain the
relation between abnormal accruals and GC opinions.
5.4. Adopting Bartov et al.’s (2000) empirical design
Based on our analysis of a large sample of firms receiving modified opinions
between 1994 and 1999, we conclude that the relation between accruals and modified
opinions is not due to earnings management but is instead driven byfirms receiving
ARTICLE IN PRESS
23We adopt an alternative form of Kothari et al.’s (2001) method. Following their specific technique, the
sample size is halved and the variable of interest is the intercept. In our context, the intercept captures the
change in abnormal accruals for a firm with a first-time GC opinion compared to a firm in the same
industry, audited by the same type of auditor with the closest lagged ROA (or distress). This method yields
results similar to those reported.
24We conduct additional robustness checks. First, following Hribar and Collins (2002), we exclude 359
firms with assets below $10 million to assess whether the inverse relation between GC opinions and
abnormal accruals stems from extremelysmall firms. However, when we rerun the Eq. (4) regression, the
GoingConcern coefficient remains significantlynegative. Second, we estimate total accruals using a balance
sheet approach in which total accruals are the change in non-cash current assets minus change in non-debt
current liabilities minus depreciation expense. We also estimate abnormal accruals using the modified
Jones model (Dechow et al., 1995). In both cases, the GoingConcern coefficient remains significantly
negative. Finally, we transform AbnormalAccruals into unscaled levels, add assets (i.e., the former deflator)
as an independent variable, and adjust standard errors per White (1980) (see Barth and Kallapur, 1995).
The coefficient on GoingConcern is again significantlynegative.
158 M. Butler et al. / Journal of Accounting and Economics 37 (2004) 139–165
GC opinions. As DeAngelo et al. (1994) find in their studyof loss firms that cut
dividends, we find that troubled firms with GC opinions have large negative,
performance-related accruals. Thus, the results of past research, in particular Bartov
et al. (2000), seem to be explained bypoor performance rather than, as theyimpl y,
aggressive accounting. However, there are several important differences between our
empirical design and that of Bartov et al.: (1) matching procedures used to obtain
control samples differ; (2) right-hand-side control variables differ; and (3) regression
specifications of the association between abnormal accruals and audit opinions differ
(Bartov et al. use a logit specification with the presence of a qualified opinion as the
dependent variable). In this section, we assess the sensitivityof our results to these
differences.
Following Bartov et al., we identifytho se firms in our sample with a newly
initiated modified opinion (i.e., a modified opinion in year t, but not t1) that we
ARTICLE IN PRESS
Table 6
OLS on signed abnormal accruals (first-time GC opinion and performance-matched samples)
AbnormalAccrualsi;t ¼b0 þ b1 MktCapi;t þ b2 Book=Mkti;t þ b3 ROAi;t1 þ b4 ROA2
i;t1 þ b5 Debt=Assetsi;t þ b6 CurrRatioi;t
þb7 Distressi;t þ b8 Big5i;t þ b9 GoingConcerni;t þ b10 Big5GoingConcerni;t þ ei;t ; ð4Þ
.
Variable Eq. (3c) with match
on lagged ROA
Eq. (3c) with match
on distress
Constant 0.058 0.070
(2.43)b (2.32)b
MktCap 0.032 0.019
(2.82)a (1.33)
Book/Mkt 0.020 0.003
(2.25)b (0.24)
ROA 0.007 0.010
(0.32) (0.51)
ROA2 0.005 0.002
(1.18) (0.62)
Debt/Assets 0.054 0.028
(1.74)c (0.71)
CurrRatio 0.009 0.014
(3.50)a (4.54)a
Distress 0.003 0.003
(12.41)a (12.29)a
Big5 0.002 0.031
(0.07) (1.14)
GoingConcern 0.115 0.021
(4.70)a (0.73)
Big5GoingConcern 0.040 0.061
(1.35) (1.71)c
Observations 1,052 1,052
Adjusted R2 22.6% 15.8%
M. Butler et al. / Journal of Accounting and Economics 37 (2004) 139–165 159
match on year, industry, auditor type, and assets (i.e., size-matching on assets, not
performance-matching on ROAt1 or Distress) to construct our control sample. This
approach yields a sample size of 2,236: 1,118 firms with newly initiated modified
opinions (of which 523 are GCs) and 1,118 control firms with standard, unqualified
opinions. We then add a new regressor, Perform, that Bartov et al. specifyto
capture extreme earnings performance and estimate the following logistic
regression:
1 if Modifiedi;t
0 Otherwise
)
¼b0 þ b1 MktCapi;t þ b2 Book=Mkti;t þ b3 Debt=Assetsi;t
þ b4 Performi;t þ b5 jAbnormalAccrualsji;t þ ei;t; ð5Þ
where
Perform =absolute value of the change in income from continuing operations
Ctotal assets (beginning of year).
Table 7 reports the results of estimating Eq. (5) with various sub-samples. Column
one, which includes the entire sample of 2,236 observations, is consistent with
column one of Table 5 and Bartov et al. (2000). Abnormal accruals are greater for
firms that receive a modified audit opinion. However, columns two and three of
Table 7 show that this association stems from companies with GC opinions. Column
two replicates the logistic regression onlyfor companies with GC opinions (and their
ARTICLE IN PRESS
t-Statistics are in parentheses. Observations are all companies with newly-initiated GoingConcern opinions
(e.g., a GC in year t, but not t1) from 1994 to 1999 matched with a clean-opinion companyon year,
industry, auditor type and either closest lagged ROA (Kothari et al., 2001) or closest contemporaneous
Distress (Shumway, 2001). All observations must have adequate Compustat information.
Variables are defined as follows:
AbnormalAccruals=cross-sectional Jones (1991) abnormal accruals (Eq. (1)).
MktCap=Log10(Market value of equity).
Book/Mkt=book value of equityCmarket value of equity.
ROA=income before extraordinaryitems Ctotal assets (beginning of year).
ROA2=square of ROA.
Debt/Assets=long-term debtCtotal assets (beginning of year).
CurrRatio=current assetsCcurrent liabilities.
Distress=bankruptcyprobabi lityper Shumway’s (2001) Table 4b hazard model (1
(1/exp(7.8116.307Net Income/Total Assets+4.068Total Liabilities/Total Assets0.158Current
Assets/Current Liabilities+0.307Ln(CompanyAge))).
Big5=one if a Big5 audit firm and zero otherwise.
GoingConcern=one if opinion is unqualified with explanatorylanguage pertaining to going-concern
status, bankruptcyor material business uncertainty and zero otherwise.
Big5GoingConcern=one if a Big5 audit firm issued the GoingConcern opinion and zero otherwise.
a Significant beyond the 1% level (two-sided tests).
b Significant beyond the 5% level (two-sided tests).
c Significant beyond the 10% level (two-sided tests).
Table 6 (Continued)
160 M. Butler et al. / Journal of Accounting and Economics 37 (2004) 139–165
matched firms), and column three includes all other modifications (and their
matched firms). The coefficient on the absolute value of abnormal accruals is
significant for the GC opinion regression but not for the other opinion modifications
regression. From these results, we conclude that our main results are robust to
considering a different matching procedure, alternative control variables, and an
alternative econometric specification. As an additional check, we replicate the tests
ARTICLE IN PRESS
Table 7
Logit regressions on audit opinion type (first-time modified opinion and size-matched sample)
1 if Modifiedi;t
0 Otherwise
)
¼b0 þ b1 MktCapi;t þ b2 Book=Mkti;t þ b3 Debt=Assetsi;t þ b4 Performi;t
þ b5 jAbnormalAccrualsi;tj þ ei;t; ð5Þ
Variable All observations GC firms Non-GC firms
Constant 0.310 0.325 0.157
(6.86)a (3.80)c (0.72)
MktCap 0.043 0.178 0.032
(0.73) (3,57)c (0.22)
Book/Mkt 0.020 0.017 0.042
(0.91) (0.41) (0.77)
Debt/Assets 1.133 1.435 0.774
(26.91)a (22.71)a (5.76)b
Perform 0.098 0.140 0.024
(5.17)b (6.62)b (0.09)
1.172 1.524 0.611
|AbnormalAccruals| (14.61)a (16.52)a (1.12)
Observations 2,236 1,046 1,190
Log likelihood 50.47a 53.89a 7.33
Chi-square statistics are in parentheses. Observations are all companies with newlyinitiated Modified audit
opinions (i.e., a modified audit opinion in year t, but not t1) from 1994 to 1999 matched with a cleanopinion
company on year, industry, auditor type and total assets (Bartov et al. 2000). All observations
must have adequate Compustat information. In the case of the GC sample regression (column 2), there are
3 pairs of observations for which we are unable to compute Perform; and, as such, this regression has a
sample size of 1,046 as opposed to a sample size of 1,052 per Table 6. In the case of the Non-GC sample
regression (column 3), the 595 newlyinitiated Modified audit opinions are a subset of the 3,299 (i.e., 105
Material Uncertainty, 2,509 Accounting Change and 685 Other Modified) opinions per Table 4.
Variables are defined as follows:
Modified=one if opinion is qualified or unqualified with explanatorylanguage and zero otherwise.
MktCap=Log10(Market value of equity).
Book/Mkt=book value of equityCmarket value of equity.
Debt/Assets=long-term debtCtotal assets (beginning of year).
Perform=absolute value of the change in operating income (Compustat \#18) divided bybeginning of the
year assets.
AbnormalAccruals=cross-sectional Jones (1991) abnormal accruals (Eq. (1)).
a Significant beyond the 1% level (two-sided tests).
b Significant beyond the 5% level (two-sided tests).
c Significant beyond the 10% level (two-sided tests).
M. Butler et al. / Journal of Accounting and Economics 37 (2004) 139–165 161
reported in Table 7 using a balance sheet measure of total accruals (as in Bartov
et al.) and results are qualitativelysim ilar.25
6. Conclusion and implications
6.1. Conclusion
We conduct a large-sample studyof the relation between audit opinions and
abnormal accruals, after taking into account the variation in the types of
modified-opinions auditors render. Our particular interest is assessing whether
such opinions are a function of earnings management (i.e., whether observable
auditing judgments proxyfor the exercise of managerial discretion). In other
words, does it seem that managers of companies receiving modified opinions are
more likelyto have managed earnings than managers of companies receiving clean
opinions?
Consistent with our understanding of audit procedures and standards, we find no
evidence that auditors use their opinions to alert financial statement users of either
excessive earnings management or the consequences of high positive accruals. In
particular, based on our content analysis of modified audit opinions, we demonstrate
that few firms receive opinion qualifications for scope limitations or departures from
GAAP and that the opinions of those who do bear no literal or discernible relation
to income-increasing earnings management.
We find that the modified opinion/abnormal accruals relation stems from
companies with going-concern opinions, because such companies have negative
abnormal accruals. Our findings are inconsistent with earnings-management and
auditor-conservatism explanations for the audit opinion/abnormal accruals relation.
Importantly, we demonstrate that the negative accruals of companies with GC
opinions are no different from a matched sample of companies experiencing similar
severe financial distress.
ARTICLE IN PRESS
25 In addition to these differences, our EDGAR-based sample period is from 1994 to 1999, whereas
Bartov et al. studythe period from 1980 to 1997. To assess whether our main results are robust to
considering these different time periods, we randomlysample 1,000 firm-year observations from a pre-SAS
58 period (1980–1987) where Compustat flags the firm as having a ‘‘qualified’’ audit opinion (i.e., code 2).
Based on Bartov et al.’s description of their sample selection procedure, the majorityof their observations
are likelyfrom this time period. As Table 2 shows, ‘‘qualified’’ audit opinions were far more common
during the pre-SAS 58 regime; however, during this time period, a ‘‘qualified’’ opinion was appropriate for
scope limitations, GAAP departures, lack of consistency, and material uncertainties (e.g., asset realization,
going concern, and contingent liabilities). We obtain 743 of these 1,000 firm-year observations from either
the NAARS database or the microfiche collection at the Universityof Rochester library. We eliminate 38
(5.1%) observations because theyare unqualified opinions, resulting in a sample size of 705. Of these 705,
104 are newlyinitiated qualifications of which 54 are GC qualifications and 50 are non-GC qualifications.
We then rerun the Eq. (5) regressions for these observations (and their matched counterparts) and obtain
results consistent with those in Table 7.
162 M. Butler et al. / Journal of Accounting and Economics 37 (2004) 139–165
6.2. Implications for earnings-management research relating to audit opinions
Our findings have implications for research that attributes the audit opinions/
accounting accruals relation to more-pervasive earnings management. Because
auditing is thought to reduce agencycosts, several papers assert that certain
types of modified audit opinions are (or should be) a function of accounting
accruals. Francis and Krishnan (1999) provide evidence suggesting that
larger, higher-qualityauditor s assess high-accrual firms as inherentlymore risky—
because managers of such firms are more likelyto have exercised discretion—and,
as a result, adjust their asset realization and GC opinion materiality
thresholds downward. We show that the audit opinions/accounting accruals
association is reliably negative, yet that it dissipates upon matching on distress
and controlling for other factors associated with accruals. This suggests that the
conditions that lead to a GC opinion are contemporaneouslyassociated with
the conditions that lead to income-decreasing accruals (i.e., both the opinion and the
accruals are consequences of the firm being financiallydistress ed) and that no
causalityshould be inferred.
Bartov et al. (2000) compare the power of various accrual models in detecting
earnings management under the assumption that firms receive certain modified
opinions because theyengage in extreme earnings management. Our findings
suggest that the relation between opinion type and abnormal accruals is due
not to earnings management, but instead to model misspecification. This implies
that the models that Bartov et al. (2000) recommend as having the most
power—cross-sectional Jones and cross-sectional Modified Jones—are the
models that suffer most from misspecification because theyare most prone to
picking up distress-related accruals and hence are more confounded by
financial distress than other accruals models. As such, differences in the power
of models theyrep ort are likelyto be measures of relative misspecification,
not power.
With respect to Bradshaw et al. (2001), we help to distinguish between certain
explanations theyoffer for the ‘‘surprisingly’’ (p. 68) negative relation
between working capital accruals and unclean audit opinions. Theyare unable
to distinguish their explanations because theydo not have details on the
underlying cause of the modified (unclean) opinion. One explanation they offer is
that auditors ‘‘... lack the necessarysophist ication to understand the future
implications of high levels of accruals’’ (p. 46) or that ‘‘... auditors interpret
the higher earnings associated with higher accruals as a positive sign, and are
less likelyto issue modified opinions in such cases’’ (p. 68). Our findings are
consistent with another of their explanations: that the role of auditors is
not to modifytheir opinions based on an assessment about the qualityof earnings
as measured byac cruals (i.e., they mayunde rstand the implications of inflated
accruals, ‘‘... but are not to communicate this information to investors through their
audit opinions’’ (p. 46)). As our description of the audit process and our content
analysis imply, auditors are unlikely to issue modified opinions for earningsmanagement
reasons.
ARTICLE IN PRESS
M. Butler et al. / Journal of Accounting and Economics 37 (2004) 139–165 163
6.3. Implications for earnings-management research in general
Our findings also have broader implications for earnings-management research.
For one, as noted, GC companies can substantiallyimpac t abnormal accrual
calculations. For another, studies often use absolute abnormal accruals as a measure
of earnings management. Our results suggest that using absolute value maymask the
source of abnormal accruals. If absolute abnormal accruals are large because
accruals are especiallypos itive or negative, inferences likelydiffer from the case in
which absolute abnormal accruals are large but accruals are more symmetric.
Finally, our findings lend some support to the importance of performance matching
when the companies of interest are experiencing extreme performance. In our case of
extreme poor performance, matching on contemporaneous distress more effectively
captures accrual-sensitive differences in performance than matching on lagged ROA.
This is likelybecause matching on distress captures the negative accruals associated
with poor operating performance as well as the negative accruals arising from
liquidity-motivated transactions (e.g., delaying payables and factoring receivables).
References
Arthur Andersen & Co., Coopers & Lybrand, Deloitte & Touche, Ernst & Young, KPMG Peat Marwick,
Price Waterhouse, 1992. The liabilitycrisis in the United States: impact on the accounting profession.
A statement of position.
Barth, M., Kallapur, S., 1995. The effects of cross-sectional scale differences on regression results in
empirical accounting research. ContemporaryAccounting Research 13, 527–567.
Bartov, E., Gul, F., Tsui, J., 2000. Discretionary-accruals models and audit qualifications. Journal of
Accounting and Economics 30, 421–452.
Belsley, D., Kuh, E., Welsch, R., 1980. Regression Diagnostics. Wiley, New York, NY.
Berton, L., 1995. Big accounting firms weed out riskyclients. The Wall Street Journal (June 26), B1, B6.
Bradshaw, M., Richardson, S., Sloan, R., 2001. Do analysts and auditors use information in accruals?
Journal of Accounting Research 39, 45–74.
Choi, S., Jeter, D., 1992. The effects of qualified audit opinions on earnings response coefficients. Journal
of Accounting and Economics 15, 229–247.
DeAngelo, H., DeAngelo, L., Skinner, D., 1994. Accounting choice in troubled companies. Journal of
Accounting and Economics 17, 113–143.
Dechow, P., Sloan, R., Sweeney, A., 1995. Detecting earnings management. The Accounting Review 70,
193–225.
DeFond, M., Jiambalvo, J., 1994. Debt covenant violations and manipulation of accruals. Journal of
Accounting and Economics 17, 145–176.
DeFond, M., Subramanyam, K.R., 1999. Modified audit opinions, litigation risk and discretionary
accruals. Unpublished working paper, Universityof Southern California.
Dopuch, N., Holthausen, R., Leftwich, R., 1986. Abnormal stock returns associated with media
disclosures of ‘‘subject to’’ qualified audit opinions. Journal of Accounting and Economics 8, 93–117.
Francis, J., Krishnan, J., 1999. Accounting accruals and auditor reporting conservatism. Contemporary
Accounting Research 16, 135–165.
Healy, P., 1985. The effect of bonus schemes on accounting decisions. Journal of Accounting and
Economics 7, 85–107.
Herz, R., Dittmar, N., Lis, S., Decker, W., Murray, R., 1997. The Coopers & Lybrand SEC Manual 7th
Edition. Wiley, New York, NY.
ARTICLE IN PRESS
164 M. Butler et al. / Journal of Accounting and Economics 37 (2004) 139–165
Hopwood, W., McKeown, J., Mutchler, J., 1989. A test of the incremental explanatorypower of opinions
qualified for consistencyand uncertainty. The Accounting Review 64, 28–48.
Hopwood, W., McKeown, J., Mutchler, J., 1994. A reexamination of auditor versus model accuracy
within the context of the going-concern opinion decision. ContemporaryAccount ing Research 10,
409–431.
Hribar, P., Collins, D., 2002. Errors in estimating accruals: implications for empirical research. Journal of
Accounting Research 40, 105–134.
Jones, J., 1991. Earnings management during import relief investigations. Journal of Accounting Research
29, 193–228.
Kennedy, D., Shaw, W., 1991. Evaluating financial distress resolution using prior audit opinions.
ContemporaryAccount ing Research 8, 97–114.
Kothari, S.P., Leone, A., Wasley, C., 2003. Performance matched discretionary accrual measures. Journal
of Accounting and Economics, forthcoming.
Louwers, T., 1998. The relation between going-concern opinions and the auditor’s loss function. Journal
of Accounting Research 36, 143–156.
McKeown, J., Mutchler, J., Hopwood, W., 1991. Towards an explanation of auditor failure to modify
the audit opinions of bankrupt companies. Auditing: A Journal of Practice & Theory10 (Supplement),
1–13.
Mutchler, J., Hopwood, W., McKeown, J., 1997. The influence of contraryinformat ion and
mitigating factors on audit opinion decisions on bankrupt companies. Journal of Accounting
Research 35, 295–310.
Nelson, M., Elliott, J., Tarpley, R., 2002. Evidence from auditors about managers’ and auditors’ earnings
management decisions. The Accounting Review 77, 175–202.
O’Malley, S., Chookaszian, D., Kolton, P., Longstreth, B., Lowenstein, L., Palmrose, Z., Peters, A., Saul,
R., 2000. The public oversight board panel on audit effectiveness: report and recommendations.
Shumway, T., 2001. Forecasting bankruptcy more accurately: a simple hazard model. Journal of Business
74, 101–124.
White, H., 1980. A heteroskedasticity-consistent covariance matrix estimator and a direct test for
heteroskedasticity. Econometrica 48, 817–838.
ARTICLE IN PRESS
M. Butler et al. / Journal of Accounting and Economics 37 (2004) 139–165 165
