The changing relationship between audit firm size and going concern reporting

Description
As a result of gradual shifts in the market for audit services, we expect financially stressed
public companies to be increasingly audited by regional firms, who, in turn, will be increasingly
likely to issue going concern reports to their financially stressed public companies.
Our expectations challenge the view that larger audit firms, in order to avoid exposure
to litigation, report more conservatively. To address these issues, we examine the 22 years
between 1989 and 2010, which we classify into four ERAs (e.g., 1989–1994, 1995–2001,
2002–2005, and 2006–2010). We initially document that over time, financially stressed
public companies are shifting to regional audit firms, partly due to the actions of larger
audit firms shedding these clients, which represent ex-ante conservatism. In contrast, audit
firm reporting represents ex-post conservatism. We next show that over time, for their
financially stressed public clients, regional audit firms are increasingly more likely to issue
going concern reports, and BigN audit firms are increasingly less likely to issue going
concern reports. We also show that in more recent ERAs, regional audit firms have been
more likely than BigN and national audit firms to issue a going concern report to their
financially stressed pubic clients.

The changing relationship between audit ?rm size and going
concern reporting
Steven E. Kaplan
a
, David D. Williams
b,?
a
W.P. Carey School of Business, Arizona State University, Tempe, AZ 85287, United States
b
Fisher College of Business, Ohio State University, Columbus, OH 43210, United States
a b s t r a c t
As a result of gradual shifts in the market for audit services, we expect ?nancially stressed
public companies to be increasingly audited by regional ?rms, who, in turn, will be increas-
ingly likely to issue going concern reports to their ?nancially stressed public companies.
Our expectations challenge the view that larger audit ?rms, in order to avoid exposure
to litigation, report more conservatively. To address these issues, we examine the 22 years
between 1989 and 2010, which we classify into four ERAs (e.g., 1989–1994, 1995–2001,
2002–2005, and 2006–2010). We initially document that over time, ?nancially stressed
public companies are shifting to regional audit ?rms, partly due to the actions of larger
audit ?rms shedding these clients, which represent ex-ante conservatism. In contrast, audit
?rm reporting represents ex-post conservatism. We next show that over time, for their
?nancially stressed public clients, regional audit ?rms are increasingly more likely to issue
going concern reports, and BigN audit ?rms are increasingly less likely to issue going
concern reports. We also show that in more recent ERAs, regional audit ?rms have been
more likely than BigN and national audit ?rms to issue a going concern report to their
?nancially stressed pubic clients. Overall, our evidence suggests that more recently, larger
audit ?rms, relative to regional audit ?rms, acted more proactively to lessen their litigation
risks through increasing centralization of client selection and acceptance processes. How-
ever, our evidence suggests that more recently, to lessen their litigation risks, regional
audit ?rms, relative to BigN and national audit ?rms, acted more conservatively by issuing
more going concern reports to their ?nancially stressed public clients.
Ó 2012 Elsevier Ltd. All rights reserved.
Introduction
Under Generally Accepted Auditing Standards (GAAS),
audit ?rms have the responsibility to evaluate the going
concern status of each of their clients and to include
explanatory language in their report when they conclude
that there is ‘‘substantial doubt’’ about a client’s ability to
continue as a going concern (GC) over the next year. This
responsibility has been controversial, as well as conse-
quential. Generally, managers of public companies prefer
not to receive a GC report (Geiger & Rama, 2006; Mutchler,
1984), in part because equity markets react negatively
when a GC report is issued (Blay & Geiger, 2001; Menon
& Williams, 2010). However, issuing a GC report presum-
ably lessens the litigation risks audit ?rms face from inves-
tors seeking to recover their losses (Carcello & Palmrose,
1994). Audit researchers have a longstanding interest in
understanding the extent to which ?rm size moderates
the strength of the relation between litigation risks and
GC reporting (DeFond & Francis, 2005; Francis, 2004).
Audit researchers also have a longstanding interest in
understanding the market for audit services and how it
changes over time (Choi, Doogar, & Ganguly, 2004; Hogan
& Martin, 2009; Jones & Raghunandan, 1998; Landsman,
Nelson, & Rountree, 2009). Generally, the literature exam-
ines the extent to which the relative ?nancial risks for the
0361-3682/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved.http://dx.doi.org/10.1016/j.aos.2012.05.002
?
Corresponding author.
E-mail address: [email protected] (D.D. Williams).
Accounting, Organizations and Society 37 (2012) 322–341
Contents lists available at SciVerse ScienceDirect
Accounting, Organizations and Society
j our nal homepage: www. el sevi er. com/ l ocat e/ aos
portfolio of public clients differ by audit ?rm size and
changes to these ?nancial risks over time, but links to
going concern reporting are limited (Francis & Krishnan,
2002). Thus, further research is warranted on how changes
in the market for audit services impact GC reporting deci-
sions among differentially sized audit ?rms.
The purpose of the current study is to provide longitu-
dinal evidence on the changing relationship between audit
?rm size and auditor going concern reporting. Our longitu-
dinal analysis examines 22 years divided into four ERAs.
The years 1989–1994 represent the ?rst ERA preceding
passage of the Private Securities Litigation Reform Act
(PSLRA). The years 1995–2001 represent the second ERA,
following the PSLRA, but preceding the Sarbanes–Oxley
(SOX) legislation and the demise of Arthur Andersen. The
years 2002–2005 represent the third ERA, which includes
the immediate years after SOX while audit ?rms and public
companies were responding to these dramatic changes.
The years 2006–2010 represent the fourth ERA, which in-
cludes the time when audit ?rms and public companies
largely adjusted to temporary shocks in the audit environ-
ment occurring in 2002.
Our study focuses on three classes of audit ?rms (e.g.,
BigN, national, and regional
1
) across four different ERAs.
Over time, we expect that ?nancially stressed public compa-
nies will increasingly be audited by regional audit ?rms,
who in turn will increasingly report more conservatively
as demonstrated by a higher propensity to issue a GC audit
report. In addition, among ?nancially stressed public compa-
nies still audited by BigN audit ?rms, we expect a decreasing
likelihood of receiving a going concern report over time (e.g.,
less conservative GC reporting). Overall, at some point,
regional audit ?rms are expected to be more likely to issue
a going concern report compared to larger audit ?rms. Our
?ndings generally support these predictions.
We document that across the ERAs, ?nancially stressed
public companies are increasingly audited by regional
audit ?rms. Speci?cally, our evidence indicates that regio-
nal ?rms only audited approximately 16% of ?nancially
stressed public companies in the ?rst two ERAs, but over
30% in the last two ERAs. While a variety of factors are in-
volved, this change re?ects, in part, a decreasing willing-
ness by larger audit ?rms to audit ?nancially stressed
public companies. The BigN audit ?rms began using more
formal ?rm-wide screening practices in the early 1990s
(Arthur Andersen et al., 1992) and over time, placed more
emphasis on formal ?rm-wide screening as a means to
avoid associating with ‘‘risky’’ audit clients (Bell, Bedard,
Johnstone, & Smith, 2002; Winograd, Gerson, & Berlin,
2000). These screening practices focus, in part, on ?nancial
stress, which is generally observable to the audit ?rm and
also associated with audit litigation (Latham & Linville,
1998).
We also document that for their ?nancially stressed
public companies, regional audit ?rms were more likely
to issue a GC report in the latter ERAs, whereas BigN and
national audit ?rms were increasingly less likely to issue
a GC report. Krishnan, Raghunandan, and Joon (2007) refer
to the use of audit reporting to control their ?rm’s expo-
sure to litigation risks as ex-post conservatism. Results
are generally similar for an analysis of Type I accuracy
(e.g., issuing a GC report to a client that becomes bankrupt
in the subsequent year). In the case of regional ?rms, we
believe this change re?ects, in part, increases in regional
?rms’ exposure to catastrophic litigation costs (Francis &
Krishnan, 2002). We also believe the change among BigN
?rms re?ects, in part, increases in BigN ?rm reliance on
client screening as a mechanism to control their ?rm’s
exposure to litigation risks. Krishnan et al. (2007) refer to
the use of client screening for this purpose as ex-ante
conservatism.
Our study offers important contributions to the audit
markets and to the GC reporting literatures. Speci?cally,
the results of our study extend existing longitudinal audit-
ing research on portfolios (e.g., Choi et al., 2004; Francis &
Krishnan, 2002; Hogan & Martin, 2009; Landsman et al.,
2009). Our evidence considers a long time horizon, more
current data, and three categories of audit ?rms, which al-
lows us to assess the effects of gradual changes in the audit
environment on both the market for audit services and GC
reporting. Our evidence is particularly important because
it shows that in more recent time periods, BigN audit ?rms
did not issue GC reports more conservatively than smaller
audit ?rms for ?nancially stressed public companies.
Instead, in more recent time periods our evidence shows
that for ?nancially stressed public companies, regional
?rms issued GC reports more conservatively than did BigN
and national audit ?rms.
The remainder of this paper is organized as follows. The
next section discusses the auditing environment and
develops the hypotheses. The third section describes the
research methods and presents our results. The last section
provides a discussion of our results.
The auditing environment and hypothesis development
Substantial changes have occurred in the auditing envi-
ronment both before and during the years covered by our
investigation (Asthana, Balsam, & Kim, 2009; Cooper &
Robson, 2006; Kinney, 2005; Mills & Young, 1999; Preston,
Cooper, Scarbrough, & Chilton, 1995; Wyatt, 2004; Zeff,
2003). Kinney (2005, p. 91) identi?es three key events dur-
ing the 1970s that ‘‘set the stage for the next 25 years of
audit regulation.’’ First, the Department of Justice and the
Federal Trade Commission reached agreements with the
AICPA to increase competiveness in the market for audit
services. For example, restrictions on competitive bidding,
uninvited solicitation of new clients, and advertising were
eliminated and many additional changes occurred that
potentially impacted the market for audit services and/or
the conduct of audits. These changes were generally
1
We refer to the largest international audit ?rms as the BigN across all
time periods. During our four time periods, the largest auditing ?rms were
the Big6 in ERA1, the Big5 in ERA2, and the Big4 in ERAs 3 and 4. National
audit ?rms are de?ned as the next ?ve largest audit ?rms after the BigN,
based on client sales. For each of the four ERAs these ?ve ?rms were BDO
Seidman, Crowe Chizek, Grant Thornton, McGladrey & Pullen, and Moss
Adams. Regional audit ?rms are de?ned as all audit ?rms that were not
classi?ed as BigN or national. There were 1744 different regional audit
?rms over the 22 years.
S.E. Kaplan, D.D. Williams / Accounting, Organizations and Society 37 (2012) 322–341 323
successful in creating a more competitive environment. In
discussing the new environment, Leonard Spacek, a top
partner with Arthur Andersen & Co., wrote in 1984, ‘‘The
competition (today) is in fees only’’ (Zeff, 2003, p. 203).
Second, during the 1970s audit ?rms placed increased
emphasis on their nonaudit services and over the next sev-
eral decades nonaudit fees grew substantially, particularly
among larger audit ?rms. The growth and ?nancial success
of nonaudit services placed increasing pressures on audit
professionals within the ?rm to keep pace (Wyatt, 2004).
Third, in response to the potential adverse effects of in-
creased competition and concerns that nonaudit fees have
the potential to impair independence, in 1977 the AICPA
established the SEC Practice Section and the Public Over-
sight Board in an effort to enhance their ability to regulate
audits. These groups were established, in part, to ‘‘appease
the SEC’’ (Kinney, 2005, p. 91) and were responsible for
setting and enforcing quality control standards and a peer
review process. Fogarty (1996) applies institutional theory
to examine peer reviews.
Kinney (2005) also identi?es three key external changes
to the audit environment starting in the 1980s. First, public
companies were becoming larger. For example, mean cap-
italization of public companies rose from approximately
$1.9 billion in 1997 to over $3.8 billion in 2007 (Center
on Audit Quality, 2008). A second change involved rapid
innovations in information technology, fundamentally
changing the accounting procedures used to construct
?nancial statements. These changes impacted the audit
process by shifting emphasis away from substantive test-
ing and toward analytical procedures and tests of internal
controls. Third, the regulatory requirements and con-
straints for many industries, such as airlines, ?nancial ser-
vices, and energy, were substantially removed. Thus, ?rms
in these industries also faced more competition and oppor-
tunities to engage in an expanded set of business activities.
In addition, between the late 1980s and the early 2000s,
the largest (e.g., BigN) audit ?rms, partly in response to in-
creased competition, experienced consolidation from
eight, down to four ?rms.
2
There were further important changes to the auditing
environment triggered by several large accounting scan-
dals that occurred in a short period of time, such as Enron
(October 2001), Global Crossing (February 2002), and
WorldCom (March 2002). First, Arthur Andersen was
convicted in June 2002 of obstruction of justice and col-
lapsed, reducing the number of BigN ?rms from ?ve to
four. Second, SOX was passed in July 2002, substantially
adding to audit ?rms’ responsibilities and dramatically
changing the regulatory landscape. Under SOX, audit ?rms
were required to audit and report on public companies’
internal controls, as well as their ?nancial statements.
Landsman et al. (2009, p. 532) contend that these two
events resulted in ‘‘a temporary capacity constraint for
the BigN.’’ In addition, with the establishment of the Public
Company Accounting Oversight Board (PCAOB), SOX
largely shifted regulatory control away from the AICPA
(e.g., self-regulation) toward government regulation. De-
Fond and Lennox (2011, p. 23) claim that inspections by
the PCAOB increased costs to all audit ?rms ‘‘by increasing
regulatory scrutiny, requiring stricter compliance with
auditing standards, and by subjecting auditors to higher
penalties for misconduct.’’
Audit ?rm portfolios
An audit ?rm’s overall client portfolio is jointly in?u-
enced by the willingness of the audit ?rm and the audit
client to continue an existing relationship, as well as the
ability of the audit ?rm to attract new, desirable clients.
Portfolios of public companies will change over time as
audit ?rms and public companies decide to continue or
sever the relationship based on a variety of factors. These
factors include the quality of the relationship, cost of the
relationship, changes in the operating and/or ?nancing
environment of public companies, and changes in the legal
and economic environment. Audit ?rms presumably sever
an existing audit relationship, or decide not to initiate a po-
tential audit relationship with a public company, because
the risks from association (e.g., potential litigation) exceed
the bene?ts (e.g., audit fees) (Bockus & Gigler, 1998; Gen-
dron, 2002; Morgan & Stocken, 1998; Power, 2003). Large
audit ?rms commonly explain changes in their client port-
folios since the mid-1990s, in part as a proactive attempt
to manage risk (Hindo, 2003; MacDonald, 1997).
Research examining audit ?rm resignations tends to
corroborate the ?rms’ public explanations, that they tend
to resign from high-risk clients, including clients with sub-
stantial ?nancial stress, and their successors tend to be
smaller (Krishnan & Krishnan, 1997; Landsman et al.,
2009; Rama & Read, 2006; Shu, 2000). For example, Rama
and Read (2006) report that for both 2001 and 2003, ?nan-
cial stress (Zscore) is positively and signi?cantly associated
with Big 4 audit resignations. These ?ndings were recently
corroborated by Landsman et al. (2009, p. 534), who report
that audit ?rm resignations ‘‘are more sensitive to client
risk than dismissal in both the pre- and post-Enron
periods.’’
Research ?ndings also indicate that an audit ?rm resig-
nation, as opposed to a client-initiated audit ?rm change,
substantially damages the client’s access to a similarly
sized audit ?rm (Rama & Read, 2006; Shu, 2000). For a
sample of clients switching audit ?rms between 1987
and 1996, Shu (2000) reports that among Big 6 resigna-
tions, less than 48% of the successors were Big 6 audit
?rms. In contrast, among client-initiated Big 6 switches,
more than 66% of the successors were Big 6 audit ?rms.
Rama and Read (2006), examining a more recent time per-
iod, report that following a resignation by a Big 4 audit
?rm, less than 16% subsequently hired a Big 4 audit ?rm.
These public companies which had selected (and presum-
ably wanted) a Big 4 audit ?rm were apparently unable
to ?nd another Big 4 audit ?rm that was willing to accept
2
Regarding the potential consequences of the consolidation of large
audit ?rms, The American Assembly (2005, 7) stated, ‘‘The Big4 clearly
constitute an oligopoly, but there is no indication that they have behaved
like one in terms of pricing, competition, and commoditization of services.’’
Similarly, Cox (2006) concludes that since 1972, when the American
Institute of Certi?ed Public Accountants eliminated anti-competitive
restrictions, competition among large audit ?rms has been ‘‘intense and
vicious.’’
324 S.E. Kaplan, D.D. Williams / Accounting, Organizations and Society 37 (2012) 322–341
the engagement after the incumbent Big 4 audit ?rm
resigned.
Gendron (2001, 2002) conducted ?eld studies to pro-
vide a rich set of evidence to better understand the client
acceptance process among three of the Big 6 audit ?rms.
Gendron (2001, 289) ?nds that even though each ?rm
had extensive written policies and decision aids to struc-
ture and guide client acceptance decisions, ‘‘decision pro-
cesses are largely organic’’ and involve signi?cant
consultation with colleagues. Gendron (2002) examines
the in?uence of audit ?rms’ organization (e.g., the extent
to which the ?rm’s policies and compensation schemes
re?ect professionalism and commercialism) on partners’
client acceptance decisions. While not intended to be
deterministic, an audit ?rm’s organization offers legiti-
macy when a partner’s concerns align with the underlying
logic of the ?rm’s organization.
Prior longitudinal studies on shifts in audit ?rm portfolios
Several studies (Choi et al., 2004; DeFond & Lennox,
2011; Francis & Krishnan, 2002; Landsman et al., 2009)
examine the aggregate effects of audit ?rm decisions (to
accept/retain client) and client decisions (to continue/dis-
continue with the audit ?rm) through an analysis of audit
?rms’ portfolio of public clients and changes in portfolio
over a relatively short period of time. Choi et al. (2004)
examine changes in the riskiness of Big 6 audit ?rms’ client
portfolios between 1975 and 1999, using several risk mea-
sures including the Zmijewski bankruptcy score to assess
?nancial stress. The authors identi?ed four separate time
periods based on changes in the relative litigation liability
pressures from one period to the next.
The ?rst period in Choi et al.’s (2004) study, 1975–1984,
establishes a benchmark. The researchers characterize the
second period, 1985–1989, as a time of ‘‘increasing con-
cern about professional liability exposure’’; the third peri-
od, 1990–1994, as a time of professional liability ‘‘relief’’;
and the fourth, 1995–1999, as a time of ‘‘relaxed concerns
about litigation pressure’’ (Choi et al., 2004, p. 754). Based
on this characterization, the authors predict that Big 6
audit portfolios in 1985–1989 will be less ?nancially risky
than the benchmark; and that audit portfolios in 1995–
1999 will be more ?nancially risky than the pre PSLRA
period of 1990–1994. While their ?ndings generally sup-
port their expectations, it is worth noting that in the ?rst
two periods, the number of new Big 6 clients substantially
exceeded the number of departing Big 6 clients, but that in
the third period (pre-PSLRA), the number of new and
departing clients was about the same; and in the ?nal
period of 1995–1999, the number of departing clients sub-
stantially exceeded the number of new clients. Further,
Choi et al. (2004) report that clients departing from Big 6
?rms are more ?nancially risky than newly accepted cli-
ents by Big 6 ?rms.
In addition, in an attempt to assess the impact of poten-
tial changes in the macro economy across the four periods,
Choi et al. (2004) examined the Big 6 share of most ?nan-
cially risky clients (e.g., a client located in the riskiest dec-
ile for at least one of the three ?nancial stress measures)
and the client’s overall market share. Their results show
that the overall Big 6 market share grew over time from
73% to almost 83%, but that their share of the ?nancially
riskiest clients relative to their overall market share de-
creased over time, suggesting that over time, the Big 6
were shedding their most ?nancially risky clients.
Francis and Krishnan (2002) provide descriptive evi-
dence about the number and ?nancial condition of ?nan-
cially stressed public companies in 1990 and 1997. The
authors indicate that the early 1990s represented a period
of increasing audit ?rm litigation risk until the passage of
the PSLRA in 1995, which decreased audit ?rm litigation
risk. Their descriptive evidence indicates that there were
more ?nancially stressed public companies in 1997 com-
pared to 1990, suggesting differences in the macro econ-
omy. However, among Big 6 audit ?rms, the ?nancial
condition of stressed public companies was stronger in
1997 than in 1990, suggesting that Big 6 audit ?rms had
disassociated themselves from their more ?nancially risky
public companies. In contrast, among non-Big 6 audit
?rms, the ?nancial condition of stressed public companies
was weaker in 1997 than in 1990, suggesting that non-Big
6 audit ?rms were servicing more ?nancially risky public
companies.
Landsman et al. (2009) examined audit ?rm switching
behavior before and after the Enron period. The post-Enron
period includes the years from 2002 to 2005. They contend
that the post-Enron audit market re?ected two key exoge-
nous shocks. The collapse of Arthur Andersen temporarily
reduced the supply of BigN audit ?rms and the passage
of SOX substantially increased the work of audit ?rms.
Their audit switching results are generally more consistent
with BigN audit ?rms rebalancing their portfolios, rather
than an increased sensitivity to client risk.
3
That is, BigN
audit ?rms maintained their sensitivity to the riskiness of
clients, but screened out their most risky clients post Enron
as they faced limited capacity.
Recently, DeFond and Lennox (2011) examined the ex-
tent to which small audit ?rms (e.g., a ?rm with fewer than
100 audits of public companies) quit auditing public com-
panies following the enactment of SOX. They document
that during the period of 2001–2008, over 600 small audit
?rms quit providing audit services to public companies,
and that the majority of small audit ?rms exited in the
years 2002 through 2004. The authors conclude that the
large number of exits by small audit ?rms represents ‘‘a
signi?cant shift in the composition of the market for small
auditors after the adoption of SOX’’ (DeFond & Lennox,
2011, p. 21–22).
Longitudinal shifts in audit ?rm portfolios: 1989–2010
The above discussion indicates that between 1989 and
2010, two signi?cant structural changes occurred (e.g.,
passage of the PSLRA and SOX). Following Choi et al.
(2004), we construct separate time periods, referred to as
‘‘ERAs,’’ to facilitate our examination of longitudinal shifts
in audit ?rm portfolios. Speci?cally, we include four
3
National audit ?rms have also experienced similar capacity constraints
during this time period (Karr, 2005).
S.E. Kaplan, D.D. Williams / Accounting, Organizations and Society 37 (2012) 322–341 325
separate ERAs, based, in part, on the occurrence of these
two signi?cant structural changes. ERA1 begins in 1989,
which coincides with the implementation of a new going
concern reporting standard (American Institute of Certi?ed
Public Accountants (AICPA), 1988), and continues through
1994. Thus, ERA1 precedes the passage of PSLRA, which
contained provisions of litigation relief to audit ?rms (Gei-
ger & Raghunandan, 2001). ERA2 includes 1995 through
2001, and precedes the passage of SOX, which substantially
increased audit ?rms’ responsibilities.
4
Landsman et al.
(2009) contend that the collapse of Arthur Andersen, as well
as the new reporting requirements of SOX, created tempo-
rary exogenous shocks to the supply of and demand for audit
services. Thus, we divide the post-SOX years into ERA3,
including 2002 through 2005, which are the years examined
in Landsman et al. (2009) and represent a period of both in-
creased audit ?rm reporting responsibilities and decreased
supply of BigN audit ?rms, and ERA4, including the years
2006 through 2010. Presumably, all audit ?rms had an
opportunity to respond to the temporary shocks related to
SOX by 2006.
We generally expect that over time, BigN ?rms will
increasingly shed their ?nancially risky clients. In particu-
lar, we expect that BigN ?rms will have continued disasso-
ciating from their ?nancially risky clients during ERA2,
even though audit ?rms received litigation relief from the
passage of PSLRA. Our expectation for ERA2 is based on
several considerations. During the mid- to late-1990s, BigN
?rms developed and introduced sophisticated technologies
to evaluate and manage the risks of client acceptance and
continuation (Bell et al., 2002; Winograd et al., 2000). As
described, these technologies were spurred by increased
competitiveness in the audit environment (e.g., there was
increasing downward pressure on audit fees), enhanced
professional standards placing greater emphasis on ?rm-
wide policies and procedures when making client accep-
tance and continuation decisions (American Institute of
Certi?ed Public Accountants (AICPA), 1997), and signi?cant
improvements in the functionality and cost-effectiveness
of information technologies. Use of ?rm-wide policies
and procedures to make client acceptance and continua-
tion decisions as a means to control the audit ?rm’s expo-
sure to litigation risks has been referred to as ex-ante
conservatism (Krishnan et al., 2007).
Changes also were occurring in the equity markets. As
discussed above, mean market capitalizations of public
companies were growing over time. Growth in market cap-
italizations is important because it may impact damages
suffered by audit ?rms. Potentially, public companies with
greater market capitalization suffer greater losses when
reporting negative news such as an accounting irregularity.
Greater losses, in turn, are likely to in?uence both the deci-
sion on whether to ?le a lawsuit against an audit ?rm and
the settlement amount, if the lawsuit goes against the
audit ?rm. In this regard, the Center on Audit Quality
(CAQ, 2008) reports that there were more public company
audit-related cases and Securities Class Actions ?led in
each of the 6 years after PSLRA (e.g., 1996–2001) compared
to any of the previous 5 years (e.g., 1991–1995).
Winograd et al. (2000, p. 176) discuss Pricewaterhous-
eCoopers use of the ?rm’s ?nancial risk assessment system
(FRISK), which is a tool ‘‘used to determine whether to ac-
cept or continue the client engagement.’’ Bell et al. (2002)
describe KRisk as a decision aid designed by KPMG to
structure and support the ?rm’s client acceptance and con-
tinuation decisions and to manage the overall riskiness of
the ?rm’s portfolio.
5
Johnstone and Bedard (2004) provide
further evidence on client acceptance/retention decisions
for a large Big 6 audit ?rm during ERA2. The study uses
the ?rm’s propriety data for 2000–2001 and shows that risk-
ier clients were culled from the portfolio, and that newly
accepted clients were less risky than either dropped or
continuing clients. Overall, we expect that from one ERA to
the next, BigN ?rms would have continued to cull ?nancially
stressed public companies from their portfolio.
Audit ?rm going concern reporting
Auditing standards require audit ?rms to evaluate the
going concern status on each audit engagement and when
appropriate, to modify their report to inform ?nancial
statement readers that there exists ‘‘substantial doubt’’
about their client’s ability to continue as a going concern
(American Institute of Certi?ed Public Accountants (AIC-
PA), 1988).
6
However in these standards, substantial doubt
is not de?ned, the going concern task is ill structured (Hoff-
man, Joe, & Moser, 2003), and the audit ?rm’s decision to is-
sue a going concern report impacts multiple stakeholders
(Geiger & Rama, 2006). Given the ambiguity of the
standards, we make two observations about audit ?rms’
decisions on whether to issue a going concern report. First,
audit partners, primarily responsible for making going con-
cern report decisions, will need to apply ‘‘seasoned judg-
ment’’ (Francis, 1994; Power, 2003). In this regard, we are
unaware of any decision aids or other ?rm-wide structuring
technologies that audit partners might use when making
going concern reporting decisions for their clients that exhi-
bit signs of ?nancial distress. Second, the seasoned judg-
ments of audit partners making going concern report
decisions are likely to include both strategic and non-strate-
gic considerations when making a going concern report
decision (Krishnan, Krishnan, & Stephens, 1996; Louwers,
1998; Matsumura, Subramanyam, & Tucker, 1997; Teoh,
1992).
Generally, managers of public companies prefer that the
audit report does not contain a going concern modi?cation
4
The passage of the SOX occurred on July 30, 2002, and our post-SOX
test period began for ERA2 was de?ned as FYE 2002-2005. A very small
timing difference occurred given that COMPUSTAT was the primary source
of data collection and COMPUSTAT uses the reporting convention of ?scal
year ends January through May are classi?ed with the prior calendar year
(e.g., a May 31, 2002 ending reporting period would be classi?ed by
COMPUSTAT as FYE 2001).
5
Costly screening technologies have been developed by other large audit
?rms, as well. Beneish, Hopkins, Jansen, and Martin (2005, 358) observe:
‘‘In an attempt to reduce potentially costly exposure to a subset of riskier
clients, all large audit ?rms implement these centralized, annual business-
risk-based reviews of their audit portfolios during the 1990s.’’
6
Holder-Webb and Wilkins (2000) discuss changes in GC reporting
decisions given SAS No. 59 (American Institute of Certi?ed Public
Accountants (AICPA), 1988).
326 S.E. Kaplan, D.D. Williams / Accounting, Organizations and Society 37 (2012) 322–341
(Geiger & Rama, 2006; Mutchler, 1984). Consequently,
receipt of a going concern audit report increases the likeli-
hood of management initiating a switch to a different audit
?rm (Carcello & Neal, 2003; Lennox, 2000). Alternatively,
going concern audit reports are believed to provide bene-
?ts to audit ?rms (Carcello & Palmrose, 1994; Frost,
1994; Palmrose, 1988; St. Pierre & Anderson, 1984). While
dated, evidence indicates that issuing a going concern re-
port to ?nancially stressed clients offers audit ?rms some
protection against litigation (Carcello & Palmrose, 1994;
Mutchler, 1984). In this regard, based on a sample from
1972 to 1992, Carcello and Palmrose (1994) report that
among bankrupt public companies that resulted (did not
result) in litigation against the audit ?rm, 36% (58%) re-
ceived going concern audit reports immediately prior to
the bankruptcy. Issuing a modi?ed report, however, was
not signi?cant in multivariate models of litigation against
the audit ?rm; overall, Carcello and Palmrose’s (1994, p.
3) state that their ‘‘evidence suggests a defensive role’’
for going concern reports.
Both Francis and Krishnan (2002) and Geiger, Raghun-
andan, and Rama (2006) examine the relationship between
audit ?rm size and going concern audit reports for ?nan-
cially stressed public companies before and after the pas-
sage of PSLRA in 1995. These studies contend that
litigation relief provided by the PSLRA would lead to less
conservative going concern reporting, and that the in?u-
ence of PSLRA on the going concern reporting decisions
would be stronger for Big 6 audit ?rms. Francis and Krish-
nan (2002) examine a sample of ?nancially stressed public
companies from1990 to 1997 and decompose the auditor’s
going concern reporting decision into two components.
They unexpectedly ?nd that starting in 1995, non-Big 6
audit ?rms would hypothetically issue more going concern
reports than reports actually issued by Big 6 audit ?rms,
suggesting that non-Big 6 audit ?rms report more conser-
vatively with respect to going concern reports. Geiger et al.
(2006) examine a sample of bankrupt ?rms and ?nd that
Big 6 audit ?rms were less likely to issue a going concern
audit report after the passage of PSLRA. In supplemental
testing, using only their post-reform sample, Geiger et al.
(2006) report that the Big 6 coef?cient is not signi?cant.
Hypotheses
As discussed above, over time, we expect larger audit
?rms to shed their ?nancially risky clients, and we expect
these clients to increasingly turn to regional audit ?rms.
While several prior studies have examined differences in
audit quality between BigN and non-BigN audit ?rms, we
further subdivide non-BigN audit ?rms between national
and regional ?rms. Boone, Khurana, and Raman (2010, p.
331) contend that because of their growth, national ?rms
‘‘appear to have emerged as an alternative to the Big 4,’’
suggesting that national and regional ?rms are different.
In this regard, Hogan and Martin (2009) show that new
clients of national ?rms
7
were signi?cantly more likely to
have been previously audited by BigN ?rms rather than by
regional ?rms. Speci?cally, between 2001 and 2004, they
?nd that for new clients of national ?rms, the predecessor
was a Big 4 (regional) audit ?rm for 84% (9%) and that for
departing national clients the successor was a Big 4 (regio-
nal) audit ?rm for 11% (41%).
8
However, because Hogan
and Martin’s (2009) research examines only the years
surrounding the passage of SOX, it is unclear whether their
?ndings can be generalized to earlier or later periods.
Francis and Krishnan (2002), based on their unexpected
?ndings that starting in 1995 non-Big 6 audit ?rms were
hypothetically more likely to issue going concern reports
than Big 6 audit ?rms, suggest that non-Big 6 audit ?rms
have fewer resources to withstand the ?nancial impact of
signi?cant litigation. They use Laventhol and Horwarth as
an example of a non-Big 6 audit ?rm that was unable to
survive the ?nancial impact of litigation damages. Presum-
ably, regional audit ?rms, on average, possess the fewest
?nancial resources and face the greatest bankruptcy risk
due to catastrophic litigation costs compared to either na-
tional or BigN audit ?rms. In this regard, recall that DeFond
and Lennox (2011) ?nd that during 2001–2008, over 600
small audit ?rms quit providing audit services to public
companies, suggesting that a large number of small audit
?rms assessed their bankruptcy risk to be unacceptably
high. Given their increasing exposure to catastrophic litiga-
tion costs, regional ?rms would be especially motivated to
act defensively, particularly with respect to their ?nan-
cially stressed public companies, by making going concern
reporting decisions in an increasingly conservative fashion.
We therefore formulate the following hypothesis.
H
1
. The likelihood of ?nancially stressed clients of smaller
(i.e., regional) audit ?rms receiving a going concern audit
report will increase over time.
Next, we consider changes in the likelihood of large
(BigN) audit ?rms issuing a going concern report to their
clients. Over time, large audit ?rms are increasingly shed-
ding some, but not all of their ?nancially risky clients.
Audit ?rms’ client acceptance/retention decisions, while
structured (Bell et al., 2002; Winograd et al., 2000), are
not deterministic (Gendron, 2001). As BigN audit ?rms
increasingly shed some, but not all of their ?nancially risky
clients, we expect their going concern reporting decisions
to be increasingly less conservative. As discussed above,
BigN ?rms have not developed decision aids or technolo-
gies to structure their going concern reporting decision.
Given the dif?culty and complexity of going concern
reporting decision and the absence of ?rm-wide structur-
ing technologies, BigN audit ?rm engagement partners
are likely to have reasonably broad discretion when mak-
ing going concern reporting decisions. Audit partners make
these going concern reporting decisions in a setting where
they receive a disproportionately large share of the imme-
diate pro?ts/bene?ts of their clients, but a smaller share of
any potential litigation costs that might be suffered in the
future (Ayers & Kaplan, 1998). Further, in explaining their
7
Hogan and Martin (2009) identify BDO Seidman, Crowe Chizek, Grant
Thornton, and McGladrey and Pullen as national ?rms.
8
Among new clients, 7% were from unknown sources. Among departing
clients, 9% were in bankruptcy, 9% were involved in a M&A, 21% were
deregistered, and 9% were unknown.
S.E. Kaplan, D.D. Williams / Accounting, Organizations and Society 37 (2012) 322–341 327
results, Francis and Krishnan (2002) speculate that since
only a few clients ?le for bankruptcy in any given year,
BigN audit ?rms and their engagement partners may
believe that once their ?rminstalls strong controls over cli-
ent acceptance and continuation decisions (e.g., ex-ante
conservatism), their exposure to litigation risk will be
effectively lessened. That is, by effectively screening out
overly risky clients, BigN ?rms may no longer see the need
to conservatively issue going concern reports. This discus-
sion leads to the following hypothesis:
H
2
. The likelihood of ?nancially stressed clients of larger
(i.e., BigN) audit ?rms receiving a going concern audit
report will decrease over time.
Hypotheses 1 and 2 focus on changes in the likelihood
of going concern reporting over time with respect to regio-
nal and BigN audit ?rms, respectively. We next consider
the implications of these changes over time on the relative
likelihood of larger vs. regional ?rms issuing going concern
reports. To the extent that regional audit ?rms are increas-
ingly more likely to issue a going concern report and BigN
audit ?rms are increasingly less likely to do so, we conjec-
ture that at some point this will be re?ected in a reversal of
the relative likelihood of regional and larger audit ?rms,
including national ?rms, issuing a going concern report.
That is, over time, compared to larger audit ?rms, we
expect regional audit ?rms to become more likely to issue
a going concern report to their ?nancially distressed cli-
ents. Consistent with this perspective, Geiger et al. (2006)
?nd that BigN ?rms were signi?cantly more likely to issue
a going concern report pre-PSLRA, but not in the post-
PSLRA period. This discussion leads to the following
hypothesis.
H
3
. Compared to larger (i.e., BigN and national) audit
?rms, smaller (i.e., regional) audit ?rms will be increas-
ingly more likely to issue a going concern audit report over
time.
Research method and results
Sample selection
We collected data for public companies over a 22-year
period between 1989 and 2010, primarily derived from
COMPUSTAT and supplemented by Compact Disclosure,
CRSP, AuditAnalytics, and SEC ?lings. Starting with all
COMPUSTAT ?rms, we excluded ADRs, trusts, funds,
limited partnerships, duplicate listings, ?rms with assets
under $1 million, and ?rms with incomplete data. Table
1, Panel A summarizes the sample selection process gener-
ating 199,921 ?rm-year observations.
We refer to the 199,921 ?rm-year observations as the
All Firms sample. Two other samples were also identi?ed.
The Stressed sample, which is a subset of the All Firms
Table 1
Sample characteristics.
Panel A: Sample selection criteria
All COMPUSTAT ?rms 1989–2010 259,525
– ADR ?rms, trusts, funds, limited partnerships, duplicates 42,649
– Firms with assets under $1million 8961
– Firms with missing data
7994
Final Sample 199,921
Test period All ?rms Stressed ?rms
a
GCAR ?rms
b
Panel B: size frequencies
1989–1994 (ERA1): ERA1 Total
48,919 9497 (19.4%) 2374 (25.0%)
1995–2001 (ERA2): ERA2 Total
73,691 17,503 (23.8%) 3377 (19.3%)
2002–2005 (ERA3): ERA3 Total
37,331 10,474 (28.1%) 3088 (29.5%)
2006–2010 (ERA4): ERA4 Total 39,980 10,603 (26.5%) 2826 (26.7%)
Total 199,921 48,077 (24.0%) 11,665 (24.3%)
Auditor
Panel C: Audit ?rm portfolios
BigN 157,379 (78.7%) 31,752 (20.2%) 4916 (15.5%)
National 15,544 (7.8%) 4521 (29.1%) 1157 (25.6%)
Regional
26,998 (13.5%) 11,804 (43.7%) 5592 (47.4%)
Total 199,921 48,077 (24.0%) 11,665 (24.3%)
Period BigN National Regional
All Stressed (%)
a
GCAR (%)
b
All Stressed (%)
a
GCAR (%)
b
All Stressed (%)
a
GCAR (%)
b
Panel D: Audit ?rm frequencies
ERA1 41,973 17.40 22.80 2583 24.60 27.90 4363 35.70 33.90
ERA2 61,553 21.30 4.30 4714 30.80 26.60 7424 39.40 38.90
ERA3 27,179 22.20 13.70 3530 31.30 28.70 6622 50.50 58.30
ERA4
26,674 19.80 10.20 4717 28.20 20.80 8589 46.30 50.50
Total 157,379 20.20 15.50 15,544 29.10 25.60 26,998 43.70 47.40
a
Percentage of stressed/ALL ?rms.
b
Percentage of GCAR/stressed ?rms.
328 S.E. Kaplan, D.D. Williams / Accounting, Organizations and Society 37 (2012) 322–341
sample, includes only ?rms exhibiting strong signs of
?nancial stress. Relying on similar indicators noted in
Louwers (1998) and Hopwood, McKeown, and Mutchler
(1994), a public ?rm is included in the Stressed sample
by having negative retained earnings and two consecutive
years of net losses.
9
This relatively strict de?nition gener-
ates a subsample of highly ?nancially stressed ?rms. The
Stressed sample is used to examine audit ?rm reporting
decisions. The GCAR sample includes only ?rms from the
Stressed sample that also received a GC report. Table 1, Panel
B shows the number of ?rm-year observations within each
ERA.
10,11
As shown, less than 25% of the All Firms sample
quali?ed for the Stressed sample, and less than 25% of the
Stressed sample received a GC report.
Longitudinal audit ?rm portfolios
We expect that as the market for audit services changes
over time, larger audit ?rms will become increasingly less
inclined to audit ?nancially risky clients. We identi?ed
four ERAs to examine changes occurring between 1989
and 2010. ERA1 includes the years 1989–1994, ERA2 in-
cludes the years 1995–2001, ERA3 includes the years
2002–2005, and ERA4 includes the years 2006–2010. We
employ three classes of audit ?rms, given that national
audit ?rms may possess characteristics that potentially dif-
fer from both the BigN and regional audit ?rms (Boone
et al., 2010; Geiger & Rama, 2006; Weber & Willenborg,
2003). Across the four ERAs, the three samples were parti-
tioned by audit ?rm class in Table 1, Panel C. As shown, the
number of Stressed clients as a percentage of the public
companies portfolio (e.g., All Firms sample) was substan-
tially smaller for BigN ?rms compared to either national
or regional ?rms. Also as shown, GC clients as a percentage
of ?nancially Stressed companies (e.g., Stressed ?rms)
were much smaller for BigN ?rms compared to either na-
tional or regional ?rms.
Table 1, Panel D presents initial descriptive evidence on
the three samples for each audit class. The data indicates
that BigN’s proportionate share of Stressed clients is
declining throughout the ERAs. For example in ERA1, for
the overall population, shown in Panel B, the percentage
of Stressed ?rms is 19.4%, whereas for the BigN sample,
shown in Panel D, the percentage of Stressed ?rms is
17.4% (i.e., a difference of 2%). Similarly, in ERA4, for the
overall population, shown in Panel B, the percentage of
Stressed ?rms is 26.5%, whereas for the BigN sample,
shown in Panel D, the percentage of Stressed ?rms is
19.8% (i.e., a difference of 6.7%). In contrast, regional audit
?rms showed the opposite pattern with sharp increases in
their proportionate share of Stressed ?rms across the ERAs.
For each ERA, the percentage of Stressed ?rms, shown in
Panel D, is higher for regional ?rms relative to the percent-
age of Stressed ?rms for the overall population, shown in
Panel B. A similar pattern is also noted with the propor-
tionate share of GC clients. Across ERAs, GC reports as a
percentage of the Stressed sample are generally decreasing
for BigN ?rms, but generally increasing for regional ?rms.
Table 1, Panel D also shows that regional ?rms’ proportion-
ate share of Stressed clients increased in ERA2, but then
remained relatively stable over the next two ERAs. For na-
tional ?rms, the proportionate share of GC clients was rel-
atively stable across the ?rst three ERAs before declining
somewhat in ERA4. This descriptive evidence provides
preliminary support for our contention that over time,
?nancially stressed ?rms were increasingly audited by re-
gional ?rms, and these ?nancially stressed ?rms increas-
ingly reported conservatively with respect to the GC
assumption.
12
Multivariate analysis
We use the following multivariate model to provide fur-
ther evidence on the relationship between client ?nancial
stress and audit ?rm class size, controlling for other factors
[Model 1]:
CPA ¼ b
0a
þ b
0b
þ b
1
LASSETS þ b
2
MVBV þ b
3
GROWTH
þ b
4
EXCHANGE þ b
5
TENURE þ b
6
LIT INDUSTRY
þ b
7
FOREIGN þ b
8
PROBANK þ e
The dependent variable, CPA, indicates whether the client
engaged a BigN audit ?rm (=1), a national audit ?rm
(=0.5), or a regional audit ?rm (=0). Based on prior re-
search, the model also includes other control variables
potentially related to audit ?rm class.
The log of total assets (LASSETS) is used to control for
client size (Weber & Willenborg, 2003). The market-to-
book ratio (MVBV) was included as a measure of account-
ing conservatism (Beaver & Ryan, 2000) and larger audit
?rms have been associated with more conservative
accounting treatments (Francis & Krishnan, 1999). Gul
and Tsui (1998) show that high growth ?rms were more
frequently audited by BigN auditors. They computed
GROWTH as the change in sales from the preceding to the
current FYE. EXCHANGE is used to signify whether a client
is traded on a national exchange (coded as 1 for NYSE,
AMEX, or NASDAQ exchanges, otherwise = 0), given more
stringent audit requirements on the national exchanges
(Keinath & Walo, 2004). TENURE (=1 if the auditor–client
relationship was 4 years or longer, otherwise = 0) is used
as a measure of auditor longevity. Ghosh and Moon
(2005) indicate a relationship between auditor tenure
and audit quality. To capture potential industry-speci?c
9
Included in our de?nition of STRESSED ?rms were those ?rms that
received a going concern audit report. There were 312 ?rms that received
the going concern report and did not meet our selection criteria of negative
retained earnings and 2 year of consecutive losses. We reran our analysis
excluding these 312 ?rms from our STRESSED sample and the results did
not qualitatively change.
10
For our 199,921 ?rm-year observations, we have 21,733 unique ?rms
and 1466 ?rms remained in our sampled during the entire 22 years.
11
The number of ?rm-year observations peaked in ERA2 with the largest
number of ?rm year observations in 1996 for the ALL FIRMS sample, and
2001 for both the STRESSED and GCAR samples.
12
There are various factors that could in?uence the overall trends in the
number or percentage of stressed clients across ERAs. Factors such as
changes in economic conditions (recession or expansion) or market
changes (the Internet bubble) could signi?cantly impact the number or
percentage of stressed clients within an audit ?rm’s portfolio. However, we
would not expect the portfolios of audit ?rms of varying sizes to be
differentially impacted by these factors.
S.E. Kaplan, D.D. Williams / Accounting, Organizations and Society 37 (2012) 322–341 329
effects, we used a binary variable (LIT_INDUSTRY) (=1 if a
client belongs to one of the high litigation risk industries
identi?ed by Francis, Philbrick, and Schipper (1994)). Blou-
in, Grein, and Rountree (2007) ?nd that more complex
?rms struggle with more overall reporting transparency,
making it more dif?cult to audit them. To capture com-
plexity, we included FOREIGN as a binary variable (=1) for
?rms with foreign business activities.
The independent variable of interest is PROBANK, a mea-
sure of ?nancial stress based upon the Zmijewski (1984)
model.
13
Higher values of PROBANK indicate a higher likeli-
hood of bankruptcy. As discussed above, we expect the sign
of PROBANK to be increasingly negative. That is, over time,
among highly ?nancially Stressed public companies we
expect a shift away from BigN and national ?rms to regional
?rms.
Table 2 provides descriptive statistics for the indepen-
dent variables used in our multivariate testing for each of
the three samples.
14
Comparing across samples, it is no
surprise that Stressed clients are smaller and poorer per-
formers compared to the All Firms sample with the GCAR
sample having the smallest and most ?nancially stressed
?rms.
15
Table 2 also includes the percentage of ?nancial
institutions (SIC codes 6000–6999) in each of the three
samples. As discussed below, there is not a consensus within
the auditing literature on whether to include ?nancial
institutions for purposes of statistical analysis.
Table 2
Descriptive statistics.
Mean St. Dev. Q1 Median Q3
All ?rms (n = 199,921)
Assets (in $ millions) 2721 9768 26.3 157.9 921.2
Debt/assets 0.617 0.452 0.331 0.567 0.801
Market value/book value 2.629 3.163 0.565 1.537 3.262
Sales growth 0.076 0.659 À0.124 0.056 0.23
Distress À0.641 2.469 À2.354 À0.985 0.612
Firm age (years) 12.4 12.6 3 8 17
Stock return 0.151 0.685 À0.197 0.017 0.362
Exchange 58.80% – – – –
Auditor tenure 59.00% – – – –
Litigation industry 26.60% – – – –
Foreign 17.60% – – – –
Bankrupt 0.90% – – – –
Financial institutions 17.40% – – – –
Stressed ?rms (n = 48,077)
Assets (in $ millions) 437 2947 5.5 23.5 107.3
Debt/assets 0.759 0.707 0.252 0.582 0.954
Sales growth 0.015 1.005 À0.399 À0.023 0.339
Distress 1.097 3.159 À1.45 0.57 3.449
Firm age (years) 8.9 8.7 3 6 12
Stock return À0.053 0.961 À0.808 À0.343 0.288
Exchange 37.40% – – – –
Auditor tenure 52.70% – – – –
Litigation industry 40.20% – – – –
Foreign 16.90% – – – –
Prior GCAR 16.20% – – – –
Bankrupt 2.60% – – – –
Financial institutions 6.60% – – – –
GCAR ?rms (n = 11,665)
Assets (in $ millions) 239 1797 2.5 5.9 27.8
Debt/assets 1.243 0.896 0.573 0.954 1.776
Sales growth À0.037 0.97 À0.792 À0.158 0.156
Distress 3.267 3.048 0.858 3.835 6.21
Firm age (years) 10.1 9.4 4 7 14
Stock return À0.168 0.948 À0.872 À0.504 0.083
Exchange 16.20% – – – –
Auditor tenure 47.20% – – – –
Litigation industry 35.40% – – – –
Foreign 7.80% – – – –
Prior GCAR 54.20% – – – –
Bankrupt 7.10% – – – –
Financial institutions 7.40% – – – –
13
The Choi et al. (2004) and Carcello and Palmrose (1994) papers
represent some of the foundation for our research. These prior studies
employed the Zmijewski (1984) bankruptcy prediction score for all ?rms,
?nancial and non-?nancial, and while there are different versions of
Zmijewski’s model we’ve following this prior literature de?ning DISTRESS =
À4.336 + (À4.512
Ã
return on assets) + (5.679
Ã
debt/assets) + (0.004
Ã
cur-
rent ratio).
14
All continuous variables were winsorized at the 1% and 99% percentiles.
15
Mean and median tests for differences between the ALL FIRMS and
STRESSED samples for the size and ?nancial health variables were
signi?cant at the p < .01 level.
330 S.E. Kaplan, D.D. Williams / Accounting, Organizations and Society 37 (2012) 322–341
Consequently, we test our hypotheses using samples includ-
ing and excluding ?nancial institutions.
We conduct two multivariate analyses to explore the
relationship between audit ?rm class size and client ?nan-
cial stress. First, our analysis of Model 1 for each of the four
ERAs using the All Firms sample provides evidence for the
sign and signi?cance of PROBANK in each ERA, controlling
for other variables potentially related to the audit ?rm cat-
egory. Our second analysis, from pooling the data from two
consecutive ERAs (e.g., ERA1 and ERA2), yields three
distinct pooled data sets. In this pooled analysis (Model
1a), we modi?ed Model 1 to include a dummy variable
for ERA
0
(=1) for the latter of the two time periods and an
interaction term PROBANK
Ã
ERA
0
. This analysis provides
evidence for the sign and signi?cance of the interaction
term to determine whether the relationship between PRO-
BANK and audit ?rm category differ signi?cantly across the
two pooled ERAs. The interaction term is expected to be
negative. Results from ordered logistic regression, used to
statistically analyze Models 1 and 1a, are presented in
Table 3.
Table 3, Panel A shows a marked increase in the explan-
atory power of the model over time from a low pseudo-R
2
of 23.7% in ERA1 to a high of 48.4% in ERA4. This pattern is
consistent with audit ?rms within the same class acting
increasingly more homogenously over time with respect
to their portfolio management decisions. While client size
was the major determinant of audit ?rm choice, all of the
control measures were signi?cant and in the expected
direction.
The primary variable of interest, PROBANK, is negative
and signi?cant for all four ERAs. These results indicate that
explicitly (e.g., resignation) or implicitly (e.g., dismissal),
larger audit ?rms tend to shed ?nancially unhealthy
companies from their portfolios of public companies. The
results for the control variables in Model 1a are virtually
identical to those in Model 1. In Model 1a, the coef?cients
for PROBANK and ERA
0
each show a decline across the three
Table 3
An ordered logistic regression analysis on auditor selection decisions (all ?rms).
Model 1 : CPA
¼ b
0a
þ b
0b
þ b
1
LASSETS þ b
2
MVBV þ b
3
GROWTH þ b
4
EXCHANGE þ b
5
TENURE þ b
6
LIT INDUSTRY þ b
7
FOREIGN
þ b
8
PROBANK þ e
Model 1a : CPA
¼ b
0a
þ b
0b
þ b
1
LASSETS þ b
2
MVBV þ b
3
GROWTH þ b
4
EXCHANGE þ b
5
TENURE þ b
6
LIT INDUSTRY þ b
7
FOREIGN
þ b
8
PROBANK þ b
9
ERA
0
þ b
10
PROBANK Ã ERA
0
þ e
.
ERA1 (n = 48,919) ERA2 (n = 73,691) ERA3 (n = 37,331) ERA4 (n = 39,980)
Coef. t-Value Coef. t-Value Coef. t-Value Coef. t-Value
Model 1
LASSETS 0.291 63.8
**
0.285 79.0
**
0.329 77.0
**
0.331 85.0
**
MVBV 0.018 7.5
**
0.012 6.6
**
0.011 3.1
**
0.024 9.3
**
GROWTH 0.055 5.3
**
0.038 2.8
**
0.076 7.0
**
0.091 8.6
**
EXCHANGE 0.022 11.4
**
0.052 10.0
**
0.077 13.9
**
0.107 26.1
**
TENURE 0.272 17.5
**
0.264 21.3
**
0.324 20.4
**
0.452 30.2
**
LIT_INDUSTRY 0.379 20.3
**
0.541 37.8
**
0.600 32.5
**
0.495 28.6
**
FOREIGN 0.179 5.7
**
0.378 15.9
**
0.563 25.0
**
0.689 39.5
**
PROBANK À0.008 À2.6
**
À0.027 À12.4
**
À0.055 À18.8
**
À0.027 À9.6
**
Psuedo-R
2
0.237 0.292 0.454 0.484
ERA1 and ERA2 (n = 112,610) ERA2 and ERA3 (n = 111,022) ERA3 and ERA4 (n = 77,311)
Coef. t-Value Coef. t-Value Coef. t-Value
Model 1a
(control variables not listed)
PROBANK À0.006 À2.1
*
À0.024 À11.1
**
À0.079 À15.7
**
ERA
0
À0.239 À23.4
**
À0.637 À62.0
**
À0.957 À47.6
**
PROBANK
Ã
ERA
0
À0.024 À6.6
**
À0.036 À10.1
**
À0.027 À3.9
**
Psuedo-R
2
0.270 0.366 0.469
Note: CPA = 1 if the client engaged a BIGN audit ?rm, =0.5 if the client engaged a national audit ?rm, and =0 if the client engaged a regional audit ?rm;
LASSETS = the natural log of assets; MVBV = market value/book value; GROWTH = change in sales from the preceding year; EXCHANGE = 1 if the ?rm’s stock is
traded on a national exchange, otherwise = 0; TENURE = 1 if the auditor–client relationship was 4 years, or longer, otherwise = 0; LIT_INDUSTRY = 1 if the
?rm operates in SIC codes 2833–2836, 3570–3674, 5200–5961, 7370–7374, and 8731–8734 otherwise = 0; FOREIGN = 1 if foreign operations were reported,
otherwise = 0; PROBANK = a composite bankruptcy probability prediction metric based upon Zmijewski (1984); and ERA
0
= 1 to signify the latter period of
the pooled data sets (i.e., ERA2, ERA3, or ERA4, respectively), otherwise = 0.
*
p < 0.05 (two-tailed).
**
p < 0.01 (two-tailed).
S.E. Kaplan, D.D. Williams / Accounting, Organizations and Society 37 (2012) 322–341 331
pooled data sets. PROBANK and ERA
0
are signi?cant in all
three pooled samples. The primary variable of interest in
Model 1a, the interaction term, PROBANK
Ã
ERA
0
, is signi?-
cant and negative in all three pooled samples. These results
indicate that larger audit ?rms were consistently less likely
to audit ?nancially stressed public clients over the four
ERAs.
16
These results are consistent with our contention
that there was a gradual, ongoing shift over many years,
such that larger audit ?rms were becoming less willing to
audit ?nancially stressed public companies. In turn, over
time, regional ?rms were increasingly accepting ?nancially
stressed public companies as clients.
Client portfolio strategies
The above results do not directly speak to whether the
outcomes were a consequence of larger audit ?rms actively
shedding risky public clients. We provide two forms of evi-
dence documenting that large audit ?rms were shedding
Stressed public companies. The ?rst analysis is presented
in Table 4, Panel A showing the percentage of audit-?rm-
initiated audit ?rm changes for each ERA among ?nancially
Stressed ?rms and partitioned by audit ?rm class.
17
As
shown, for All Firms the percentage is increasing across
the ?rst three ERAs before declining in ERA4. The percentage
of audit-?rm-initiated changes may have peaked in ERA3, as
audit ?rms were responding to the temporary external
shocks occurring in 2002. Subsequently, audit-?rm-initiated
changes may have declined in ERA4, as there were no further
key changes in the audit environment. As shown, BigN ?rms
followed a similar pattern as the All Firms sample of
Stressed ?rms. Also, while the percentage was increasing
across ERAs for regional ?rms, regional ?rms initiated a low-
er percentage of changes than BigN ?rms for the ?rst three
ERAs. Thus, except for ERA4, BigN were more aggressively
dropping undesirable (e.g., ?nancially stressed) clients from
their portfolio of public companies compared to regional
?rms.
18
Our second analysis focuses on stressed ?rms and pro-
vides descriptive evidence on the median probability of
bankruptcy among ongoing audit clients for the four
ERAs.
19
In this analysis, an ‘‘ongoing audit client’’ is de?ned
as ‘‘an audit ?rm–client relationship of four or more years.’’
We focus on ongoing clients to assess whether the ?nancial
condition of ?nancially stressed ?rms generally improved
across ERAs for BigN ?rms and generally declined for regio-
nal ?rms. Presumably, if BigN ?rms tended to increasingly
shed their more ?nancially risky clients, then the ?nancial
condition of the remaining stressed ?rms would be expected
Table 4
Audit portfolios of stressed ?rms.
Client audit ?rm ERA1 ERA2 ERA3 ERA4
Panel A: Frequency (percentage) of auditor-initiated auditor changes among stressed ?rms
All stressed clients: Audit ?rm resignations
145 408 287 193
(1.5%) (2.3%) (2.7%) (1.8%)
BigN stressed clients: Audit ?rm resignations
114 339 187 54
(1.6%) (2.6%) (3.1%) (1.0%)
National stressed clients: Audit ?rm resignations
11 27 43 46
(1.7%) (1.9%) (3.9%) (3.5%)
Regional stressed ?rms: Audit ?rm resignations
20 42 57 93
(1.3%) (1.4%) (1.7%) (2.3%)
Panel B: Median probability of bankruptcy of ongoing audit clients among stressed ?rms
Client audit ?rm ERA1 ERA2 ERA3 ERA4
All stressed clients: Ongoing audit clients
67.1% 61.2% 62.4% 60.2%
BigN stressed clients: Ongoing audit clients
67.2% 57.4% 52.4% 49.4%
National stressed clients: Ongoing audit clients
73.8% 81.2% 67.2% 61.7%
Regional stressed clients: Ongoing audit clients
61.1% 81.6% 83.6% 84.5%
ERA BigN stressed clients National stressed clients Regional stressed clients
Panel C: Mean (Median) market capitalization by audit ?rm portfolios (in millions) among stressed ?rms
ERA1 $316 ($21) $28 ($9) $16 ($6)
ERA2 $490 ($53) $48 ($10) $20 ($6)
ERA3 $610 ($85) $69 ($21) $30 ($8)
ERA4 $601 ($115) $119 ($34) $47 ($14)
Note: Ongoing audit clients are those ?rms where the audit relationship is four or more years.
16
We reran the Model 1 analyses comparing BigN to only national, then
BigN to only regional, and ?nally, national to only regional audit ?rms. The
DISTRESS and DISTRESS
Ã
ERA
0
results were unchanged from those reported,
including insigni?cant results for DISTRESS
Ã
ERA
0
for the ?nal pooled
sample.
17
We identi?ed audit ?rm resignations from Compact Disclosure from
1989 to 2005, and Auditor_Trak, and the popular press for resignations
between 1989 and 2010. Starting in 1996, we included searches of the SEC
EDGAR database examining Form 8-k ?lings. Finally, starting in 2000, our
search protocol was supplemented with AuditAnalytics. We excluded any
audit ?rm resignation relating to ?rm dissolution (e.g., Laventhol &
Horwath and Arthur Andersen), of?ce closures, independence issues, or
when an audit ?rm declined to service publicly traded clients.
18
We extended the analysis and noted that BigN stressed clients change
audit ?rms more frequently than non-stressed clients (11.1% compared to
7.1%, respectively) and similar results were noted for the national audit
?rms (13.0–9.0%, respectively). However regional audit ?rms differ from
the larger audit ?rms in that the stressed clients change less frequently
than the non-stressed clients (15.8–17.6%, respectively).
19
Following DeFond, Raghunandan, and Subramanyam (2002), we
converted the Zmijewski (1984) metric into a probability of bankruptcy
measure that ranges from 0% to 100%. The results were qualitatively
unchanged.
332 S.E. Kaplan, D.D. Williams / Accounting, Organizations and Society 37 (2012) 322–341
to improve across ERAs. As shown in Table 4, Panel B, among
all stressed ?rms, the median probability of bankruptcy
score for ongoing clients declined in ERA2 and has remained
about the same for the last two ERAs. However, while the
median probability of bankruptcy score for ongoing BigN
clients is about the same as the All Firms median in ERA1,
it has consistently declined across the subsequent three
ERAs. This pattern is consistent with BigN ?rms culling out
their more ?nancially risky clients. That is, to the extent that
over time BigN ?rms disassociate from their most ?nancially
stressed clients, the median probability of bankruptcy score
on the remaining ongoing clients will increase. Additionally,
Table 4, Panel B again shows that for regional ?rms the med-
ian bankruptcy score for ongoing clients was about the same
as the All Firms median in ERA1; it jumps substantially in
ERA2; and remains high in subsequent ERAs. This pattern
suggests that regional ?rms were not actively culling their
most ?nancially risky clients. Overall, the results presented
in Table 4, Panels A and B provide further evidence consis-
tent with our contention that larger audit ?rms were shed-
ding their most ?nancially risky clients.
Table 4, Panel C presents descriptive evidence (i.e.,
means and medians) about the market capitalization of
Stressed ?rms by audit ?rm class. We expect the market
capitalization of these companies to be increasing across
ERAs, particularly for BigNclients. As shown, for BigN?nan-
cially stressedclients, the mean(median) market capitaliza-
tion increases from $316 ($21) in ERA1 to $490 ($53), $610
($85), and $601 ($115) in the subsequent three ERAs. While
the magnitude of increases across ERAs is generally larger
for BigNclients, it is also important to consider the percent-
age increases across ERAs. In this regard, the percentage in-
crease across ERAs using means is largest for regional ?rms
(e.g. $47vs. $16), consistent withthe notionthat bankruptcy
risk for regional ?rms was increasing over time, and partic-
ularly during the last two ERAs. However, increases among
BigN stressed clients also are worth noting because it sug-
gests that the expected litigation-related costs for ?nan-
cially stressed BigN companies were not necessarily
decreasinginERA2, as their potential losses wereincreasing.
Hypothesis tests – audit ?rm going concern reporting
Our hypotheses focus on audit ?rms’ going concern
reporting, concentrating only on the Stressed sample of
public clients. As discussed above, there is not a consensus
among auditing researchers on whether to include ?nan-
cial institutions when performing statistical analysis. For
example, Choi et al. (2004) and Hogan and Martin (2009)
include ?nancial institutions in their sample of ?rms, while
Jones and Raghunandan (1998) and Landsman et al. (2009)
exclude ?nancial institutions from their sample of ?rms.
Consequently, we test H
1
, H
2
, and H
3
using samples includ-
ing and excluding ?nancial institutions. As shown in Table
1, Panel C, the percentage of Stressed clients receiving a GC
report differs within audit ?rm class and ERA. As shown in
Table 1, Panel D, the GC report percentage is generally
decreasing across ERAs for BigN ?rms, and generally
increasing for regional ?rms. This descriptive evidence
provides initial indications that the GC decision varies by
audit ?rm class, especially in the latter time periods.
To test H
1
and H
2
, we use a multivariate model includ-
ing ERA
0
[Model 2]:
GCAR ¼ b
0
þ b
1
LASSETS þ b
2
LðAgeÞ þ b
3
GROWTH
þ b
4
RETURN þ b
5
TENURE þ b
6
LIT INDUSTRY
þ b
7
FOREIGN þ b
8
PriorGCAR þ b
9
PROBANK
þ b
10
ERA
0
þ e
The dependent variable is the going concern audit report;
GCAR is 1 when a going concern report is issued, and 0
otherwise. Based on prior research (see Butler, Leone, &
Willenborg, 2004; Hopwood et al., 1994; Weber & Willen-
borg, 2003), the model also includes the previously de?ned
control variables of LASSETS, GROWTH, TENURE, LIT_INDUS-
TRY, FOREIGN, PROBANK, and ERA
0
. Following DeFond et al.
(2002) and Francis and Krishnan (1999), we also include
the log of the number of years the ?rm has been publicly
traded (L(Age)), the ?rm’s stock return over the ?scal year
(RETURN), and PriorGCAR (=1) if the ?rm had a GCAR in the
preceding year.
H
1
is restricted to clients of regional audit ?rms. As in
Model 1a, the data fromtwo consecutive ERAs were pooled.
For each pooled data set, a dummy variable for ERA
0
(=1)
represents the latter of the two ERAs. The primary variable
of interest is the ERA
0
term to determine whether GC
reporting differed signi?cantly across the two pooled ERAs.
H
1
predicts that the ERA
0
termwill be positive. Results from
logistic regression analysis are presented in Table 5, Panel A
for all stressed ?rms and Table 5, Panel B for all stressed
?rms excluding ?nancial institutions.
As shown toward the bottom of Table 5, Panel A, using
the sample of all stressed ?rms, ERA
0
is positive and signif-
icant for all three pooled data sets for regional audit clients.
Results shown toward the bottom of Table 5, Panel B, using
the sample of stressed ?rms excluding ?nancial institu-
tions also shows that ERA
0
is positive and signi?cant for
all three pooled data sets for regional audit clients. These
results indicate that, controlling for other factors, regional
?rms were signi?cantly more likely to issue GC reports in
subsequent ERAs. These results provide strong support
for H
1
. In addition, the control variables that were signi?-
cant had the expected sign.
H
2
is tested using only BigN audit clients. Again, the pri-
mary variable of interest is ERA
0
, and the hypothesis pre-
dicts a negative relationship. Results for BigN ?rms,
presented toward the top of Table 5, Panel A for the sample
of all stressed ?rms, show that ERA
0
is negative and signif-
icant for all three pooled data sets. Similarly, results for
BigN ?rms for the sample of stressed ?rms excluding
?nancial institutions, presented toward the top of Table
5, Panel B, again show that ERA
0
is negative and signi?cant
for all three pooled data sets. These results indicate that,
controlling for other factors, BigN ?rms were signi?cantly
less likely to issue GC reports in subsequent ERAs. The re-
sults from the pooled data sets provide strong support for
H
2
. In addition, the control variables that were signi?cant
had the expected sign.
Panels A and B in Table 5 also present results for na-
tional ?rms. As shown, ERA
0
is insigni?cant for the ?rst
two pooled data sets for national ?rms in both Panels A
and B. Also, as shown, ERA
0
is insigni?cant (signi?cant)
S.E. Kaplan, D.D. Williams / Accounting, Organizations and Society 37 (2012) 322–341 333
Table 5
Logistic regression on going concern audit reports by audit class (stressed ?rms).
Model 2 : GCAR
¼ b
0
þ b
1
LASSETS þ b
2
LðAgeÞ þ b
3
GROWTH þ b
4
RETURN þ b
5
TENURE þ b
6
LIT INDUSTRY þ b
7
FOREIGN
þ b
8
PriorGCAR þ b
9
PROBANK þ b
10
ERA
0
þ e
.
ERA1 and ERA2 ERA2 and ERA3 ERA3 and ERA4
Coef. t-Value Coef. t-Value Coef. t-Value
Panel A: All stressed ?rms
BigN
INTERCEPT À2.108 À26.9
**
À2.739 À30.0
**
À2.844 À19.8
**
LASSETS À0.063 À5.1
**
À0.045 À3.3
**
À0.073 À4.0
**
L(Age) À0.184 À6.0
**
À0.314 À8.9
**
À0.327 À6.4
**
GROWTH À0.255 À9.9
**
À0.231 À8.1
**
À0.219 À4.7
**
RETURN À0.323 À11.0
**
À0.382 À11.8
**
À0.495 À10.8
**
TENURE À0.113 À2.3
*
À0.049 À0.9 0.031 0.3
LIT_INDUSTRY À0.268 À5.8
**
À0.372 À7.1
**
À0.505 À6.5
**
FOREIGN À0.338 À4.4
**
À0.425 À5.5
**
À0.624 À7.3
**
PriorGCAR 2.820 44.7
**
3.054 41.5
**
3.304 33.5
**
PROBANK 0.198 26.0
**
0.234 27.0
**
0.267 21.1
**
ERA
0
À0.214 À4.7
**
À0.112 À2.1
*
À0.172 À2.3
**
Psuedo-R
2
0.352 0.360 0.425
N 20,431 19,155 11,321
GCAR ?rms 17.4% 14.1% 12.1%
National (control variables not listed)
PROBANK 0.186 9.1
**
0.184 9.7
**
0.163 7.9
**
ERA
0
À0.028 À0.2 0.004 0.1 À0.103 À0.8
Psuedo-R
2
0.232 0.398 0.441
N 2085 2555 2436
GCAR ?rms 27.0% 27.5% 24.4%
Regional (control variables not listed)
PROBANK 0.127 12.8
**
0.159 14.7
**
0.161 15.1
**
ERA
0
0.192 2.3
*
0.392 5.6
**
0.261 2.5
*
Psuedo-R
2
0.380 0.497 0.578
N 4484 6267 7320
GCAR ?rms 36.5% 48.7% 54.0%
Panel B: Stressed ?rms, excluding ?nancial institutions
BigN (control variables not listed)
PROBANK 1.422 11.8
**
1.786 13.1
**
1.945 10.5
**
ERA
0
À0.425 À6.9
**
À0.215 À2.3
*
À0.267 À2.9
**
Psuedo-R
2
0.390 0.359 0.316
N 19,232 18,330 10.712
GCAR Firms 17.1% 14.0% 11.9%
National (control variables not listed)
PROBANK 1.350 4.5
**
1.417 5.4
**
0.966 3.5
**
ERA
0
À0.215 À1.4 0.078 1.6 À0.316 À2.1
*
Psuedo-R
2
0.402 0.394 0.411
N 1926 2410 2228
GCAR ?rms 27.3% 28.0% 24.7%
Regional (control variables not listed)
PROBANK 0.990 5.9
**
0.969 7.5
**
0.744 6.6
**
ERA
0
0.421 2.4
*
1.363 18.9
**
0.138 2.2
*
Psuedo-R
2
0.393 0.373 0.347
N 4061 5743 6746
GCAR ?rms 36.8% 49.0% 54.2%
Note: GCAR = 1 if the client received a going concern audit report, otherwise = 0; LASSETS = the natural log of assets; L(Age) = the natural log of the number of
years the ?rm has been publicly traded; net change in accruals, scaled by sales; RETURN = the ?rm’s stock return over the ?scal year; PriorGCAR = 1 if the
?rm received a GCAR in the prior year, otherwise = 0; and ERA
0
= 1 to signify the latter period of the pooled data sets (i.e., ERA2, ERA3, or ERA4, respectively),
otherwise = 0. See Note to Table 3 for variable descriptions of GROWTH, TENURE, LIT_INDUSTRY, FOREIGN, and PROBANK.
*
p < 0.05 (two-tailed).
**
p < 0.01 (two-tailed).
334 S.E. Kaplan, D.D. Williams / Accounting, Organizations and Society 37 (2012) 322–341
for national ?rms in the last pooled data set using all
stressed ?rms (stressed ?rms excluding ?nancial institu-
tions). These results generally indicate that across ERAs
there were limited changes over time in national audit
?rms’ propensity to issue GC reports.
When testing H
3,
we use a multivariate model to
control for other factors. We use two models in testing
the hypothesis, the ?rst of which, Model 3, is as follows:
GCAR ¼ b
0
þ b
1
LASSETS þ b
2
DEBTAS þ b
3
LðAgeÞ
þ b
4
GROWTH þ b
5
RETURN þ b
6
TENURE
þ b
7
LIT INDUSTRY þ b
8
FOREIGN þ b
9
PriorGCAR
þ b
10
PROBANK þ b
11
BIGN þ b
12
NATIONAL þ e
Model 3 includes dummy variables for BIGN (=1) and NA-
TIONAL (=1) audit ?rms, but otherwise is identical to Model
2. For each ERA, this model is analyzed using both the
sample of all stressed ?rms and the sample of stressed
?rms excluding ?nancial institutions. This analysis pro-
vides evidence for each ERA on the sign and signi?cance
of the BIGN and NATIONAL terms to determine whether
GC reporting differed signi?cantly across audit ?rm clas-
ses. The primary variables of interest in Model 3 are the
BIGN and NATIONAL terms. H
3
predicts that in the latter
time periods, BIGN and NATIONAL will each be signi?cant
and negative. Table 6, Panel A presents logistic regression
results for Model 3 using the sample of all stressed ?rms.
Table 7, Panel A presents logistic regression results, also
Table 6
Logistic regression on going concern audit reports (all stressed ?rms).
Model 3 : GCAR
¼ b
0
þ b
1
LASSETS þ b
2
LðAgeÞ þ b
3
GROWTH þ b
4
RETURN þ b
5
TENURE þ b
6
LIT INDUSTRY þ b
7
FOREIGN
þ b
8
PriorGCAR þ b
9
PROBANK þ b
10
BIGN þ b
11
NATIONAL þ e
Model 3a : GCAR
¼ b
0
þ b
1
LASSETS þ b
2
LðAgeÞ þ b
3
GROWTH þ b
4
RETURN þ b
5
TENURE þ b
6
LIT INDUSTRY þ b
7
FOREIGN
þ b
8
PriorGCAR þ b
9
PROBANK þ b
10
BIGN þ b
11
NATIONAL þ b
12
ERA
0
þ b
13
BIGN Ã ERA
0
þ b
14
NATIONAL Ã ERA
0
þ e
.
ERA1 ERA2 ERA3 ERA4
Coef. t-Value Coef. t-Value Coef. t-Value Coef. t-Value
Panel A: Results for Model 3
INTERCEPT À1.584 À15.5
**
À1.519 À18.7
**
À1.380 À12.8
**
À1.139 À9.8
**
LASSETS À0.084 À4.9
**
À0.054 À2.9
**
À0.069 À3.4
**
À0.034 À2.7
**
L(Age) À0.067 À1.7 À0.002 À0.1 À0.086 À2.1
*
0.107 2.5
*
GROWTH À0.202 À5.8
**
À0.263 À10.8
**
À0.214 À6.3
**
À0.239 À6.7
**
RETURN À0.261 À6.9
**
À0.266 À9.5
**
À0.417 À12.3
**
À0.419 À11.2
**
TENURE À0.314 À5.0
**
À0.004 À0.1 0.033 0.5 0.015 0.2
LIT_INDUSTRY À0.124 À1.9 À0.281 À5.8
**
À0.219 À3.4
**
À0.342 À5.0
**
FOREIGN À0.189 À1.7 À0.409 À4.7
**
À0.556 À6.1
**
À0.843 À9.9
**
PriorGCAR 2.725 36.5
**
2.735 40.6
**
3.152 40.8
**
3.227 39.8
**
PROBANK 0.164 16.4
**
0.211 26.9
**
0.209 20.3
**
0.208 19.8
**
BIGN À0.062 À0.8 À0.678 À10.7
**
À0.953 À10.9
**
À0.985 À11.3
**
NATIONAL À0.126 À1.0 À0.337 À3.9
**
À0.636 À6.0
**
À0.517 À5.1
**
Psuedo-R
2
0.375 0.373 0.427 0.517
n 9497 17,503 10,474 10,603
GCAR ?rms 25.0% 19.3% 29.5% 26.7%
ERA1 and ERA2 ERA2 and ERA3 ERA3 and ERA4
Coef. t-Value Coef. t-Value Coef. t-Value
Panel B: Results for Model 3a
Model 3a (control variables not listed)
ERA
0
0.329 5.8
**
0.719 12.8
**
0.331 5.9
**
BIGN
Ã
ERA
0
À0.613 À10.4
**
À0.877 À12.2
**
À0.747 À9.6
**
NATIONAL
Ã
ERA
0
À0.325 À3.8
**
À0.604 À6.0
**
À0.397 À4.0
**
Psuedo-R
2
0.376 0.394 0.418
N 27,000 27,977 21,077
GCAR ?rms 21.3% 23.1% 28.6%
Note: See Notes to Tables 3 and 5 for variable descriptions.
*
p < 0.05 (two-tailed).
**
p < 0.01 (two-tailed).
S.E. Kaplan, D.D. Williams / Accounting, Organizations and Society 37 (2012) 322–341 335
for Model 3, using the sample of stressed ?rms excluding
?nancial institutions.
Model 3a is similar to Model 3, except that it includes a
time variable, ERA
0
and interaction terms between ERA
0
and
two of the audit class variables (i.e., BIGN and NATIONAL).
Using pooled data sets, the primary variables of interest
in Model 3a are the BIGN
Ã
ERA
0
and NATIONAL
Ã
ERA
0
inter-
action terms. These interaction terms provide evidence for
each pooled sample on whether the relationship between
audit ?rm type and GC reporting differed across the two
time periods. H
3
predicts that in the latter pooled samples,
the two interaction terms will each be signi?cant and
negative. This would indicate that the BIGN(NATIONAL), rel-
ative to regional audit ?rms, are signi?cantly more likely to
issue a GC report in the latter time period compared to the
earlier time period. Again using the sample of all stressed
?rms and stressed ?rms excluding ?nancial institutions,
Tables 6 and 7 Panel B, respectively, presents logistic
regression results for Model 3a.
Tables 6 and 7, Panel A show that for the last three ERAs
BIGN and NATIONAL are each negative and signi?cant.
These results indicate that relative to regional ?rms, BIGN
and NATIONAL ?rms are signi?cantly less likely to issue a
GC report. As shown in Tables 6 and 7, Panel B, the two
interaction terms are negative and signi?cant in all three
pooled samples. These results indicate that compared to
regional ?rms, BIGN and NATIONAL ?rms are signi?cantly
less likely to issue a GC report when ERA
0
= 1.
In addition, we also test H
3
using the GCAR sample.
Using the GCAR sample allows us to assess factors associ-
Table 7
Logistic regression on going concern audit reports (stressed ?rms excluding ?nancial institutions).
Model 3 : GCAR
¼ b
0
þ b
1
LASSETS þ b
2
LðAgeÞ þ b
3
GROWTH þ b
4
RETURN þ b
5
TENURE þ b
6
LIT INDUSTRY þ b
7
FOREIGN
þ b
8
PriorGCAR þ b
9
PROBANK þ b
10
BIGN þ b
11
NATIONAL þ e
Model 3a : GCAR
¼ b
0
þ b
1
LASSETS þ b
2
LðAgeÞ þ b
3
GROWTH þ b
4
RETURN þ b
5
TENURE þ b
6
LIT INDUSTRY þ b
7
FOREIGN
þ b
8
PriorGCAR þ b
9
PROBANK þ b
10
BIGN þ b
11
NATIONAL þ b
12
ERA
0
þ b
13
BIGN Ã ERA
0
þ b
14
NATIONAL Ã ERA0 þ e
.
ERA1 ERA2 ERA3 ERA4
Coef. t-Value Coef. t-Value Coef. t-Value Coef. t-Value
Panel A: Results for Model 3
Model 3
INTERCEPT 7.095 19.2
**
8.049 21.2
**
3.593 10.1
**
8.722 22.7
**
LASSETS À0.212 À16.3
**
À0.146 À12.6
**
À0.136 À11.3
**
À0.226 À16.8
**
L(Age) À0.067 À7.2
**
À0.068 À3.2
**
À0.106 À4.4
**
0.044 1.7
GROWTH À0.107 À4.7
**
À0.170 À10.2
**
À0.059 À3.2
**
À0.074 À3.8
**
RETURN À0.077 À3.3
**
À0.008 À0.5 À0.018 À1.1 À0.012 À0.6
TENURE À0.169 À4.1
**
À0.059 À1.6 0.025 0.7 0.153 À3.8
**
LIT_INDUSTRY À0.080 À1.9 À0.181 À5.4
**
À0.083 À2.3
*
À0.153 À3.9
**
FOREIGN À0.098 À1.3 À0.238 À3.9
**
À0.451 À8.5
**
À0.593 À11.7
**
PriorGCAR 8.483 29.3
**
9.034 42.9
**
4.403 12.6
**
9.112 32.7
**
PROBANK 1.062 18.2
**
1.086 21.7
**
1.345 25.1
**
1.247 22.2
**
BIGN À0.084 À1.6 À0.433 À10.2
**
À0.982 À19.7
**
À0.753 À14.4
**
NATIONAL À0.144 À1.7 À0.242 À4.3
**
À0.628 À10.9
**
À0.656 À10.5
**
Psuedo-R
2
0.398 0.403 0.426 0.453
n 8655 16,564 9919 9767
GCAR ?rms 24.8% 19.1% 29.2% 26.5%
ERA1 and ERA2 ERA2 and ERA3 ERA3 and ERA4
Coef. t-Value Coef. t-Value Coef. t-Value
Panel B: Results for Model 3a
Model 3a (control variables not listed)
ERA
0
0.298 4.3
**
1.722 30.3
**
0.487 8.9
**
BIGN
Ã
ERA
0
À0.608 À8.2
**
À1.414 À17.9
??
À1.219 À13.5
**
NATIONAL
Ã
ERA
0
À0.345 À3.3
**
À0.847 À8.5
**
À1.003 À8.8
**
Psuedo-R
2
0.405 0.422 0.435
N 25,219 26,483 19,686
GCAR ?rms 21.0% 22.9% 28.0%
Note: See Notes to Tables 3 and 5 for variable descriptions.
*
p < 0.05 (two-tailed).
**
p < 0.01 (two-tailed).
336 S.E. Kaplan, D.D. Williams / Accounting, Organizations and Society 37 (2012) 322–341
ated with Type I errors, which occurs when the audit ?rm
issues a GC report to a ?nancially stressed client that is still
viable the next ?scal year. Type I errors represent a form of
ex-post conservative reporting. Following Geiger and Rama
(2006) and Myers, Schmidt, and Wilkins (2010), we identi-
?ed ?rms reporting on COMPUSTAT, CRSP, and Compact
Disclosure that declared bankruptcy, or involuntarily liqui-
dated, within 1 year of the audit ?rm’s report to assess
Type I errors. Following a research design primarily
derived from Geiger and Rama (2006),
20
using the GCAR
sample, we examined the relation between Type I errors
and audit ?rm size using the following model [Model 4]:
BANKRUPT ¼ b
0
þ b
1
LSALES þ b
2
EXCHANGE
þ b
3
LIT INDUSTRY þ b
4
RETURN
þ b
5
PROBANK þ b
6
BIGN þ b
7
NATIONAL þ e
The dependent variable is BANKRUPT = 1 if the ?rm entered
bankruptcy or involuntary delisting within 1 year, and = 0
otherwise (e.g., remained viable). For the previously unde-
?ned independent variables, LSALES = the natural log of
sales and EXCHANGE = 1 if the ?rm was listed on the New
York or American Stock Exchange, and = 0 otherwise.
Model 4 is evaluated using the sample including all GCAR
?rms and the sample of GCAR ?rms excluding ?nancial
institutions. Table 8 reports the results for the Type I errors
for the sample including all GCAR ?rms. As shown, there
was no signi?cant difference in audit ?rm reporting accu-
racy among audit ?rm classes in ERA1. However, for
ERA2 through ERA4, Type I error was signi?cantly higher
for regional ?rms compared to BigN and national ?rms.
Our results show that public companies receiving a GC
report by regional ?rms, relative to public companies
receiving a GC report by BigN and national ?rms, were
more likely to remain viable for 12 months. This is consis-
tent with our prediction that regional ?rms report in a
more conservative fashion compared to larger ?rms. The
second panel of Table 8 reports the trend over ERAs. Both
interaction terms are signi?cant and positive in the three
Table 8
Logistic regression on bankruptcy (Type I Accuracy) (GCAR Firms).
Model 4 : BANKRUPT
¼ b
0
þ b
1
LSALES þ b
2
EXCHANGE þ b
3
LIT INDUSTRY þ b
4
RETURN þ b
5
PROBANK þ b
6
BIGN þ b
7
NATIONAL þ e
Model 4a : BANKRUPT
¼ b
0
þ b
1
LSALES þ b
2
EXCHANGE þ b
3
LIT INDUSTRY þ b
4
RETURN þ b
5
PROBANK þ b
6
BIGN þ b
7
NATIONAL þ b
8
ERA
0
þ b
9
BIGN Ã ERA
0
þ b
10
NATIONAL Ã ERA
0
þ e
.
ERA1 ERA2 ERA3 ERA4
Coef. t-Value Coef. t-Value Coef. t-Value Coef. t-Value
Panel A: Results for Model 4
INTERCEPT À3.168 À9.9
**
À6.749 À14.3
**
À7.231 À13.1
**
À3.278 À8.4
**
LSALES 0.196 4.7
**
0.400 10.6
**
0.394 8.2
**
0.156 3.8
**
EXCHANGE 0.045 3.3
**
0.101 4.8
**
0.116 4.6
**
0.051 2.3
*
LIT_INDUSTRY 0.270 1.7 0.340 2.2
*
0.183 1.8 0.218 1.5
RETURN À0.312 À2.4
*
À0.293 À2.6
**
À0.337 À2.9
**
À0.108 À2.1
*
PROBANK 0.126 4.3
**
0.086 3.0
**
0.100 2.8
**
0.057 2.6
**
BIGN 0.304 1.3 0.162 4.5
**
1.261 4.9
**
0.977 2.3
*
NATIONAL 0.592 1.8 0.825 2.5
*
1.077 3.6
**
0.232 2.1
*
Psuedo-R
2
0.079 0.187 0.221 0.417
n 2734 3377 3088 2826
Bankrupt ?rms 8.4% 7.0% 5.1% 8.4%
ERA1 and ERA2 ERA2 and ERA3 ERA3 and ERA4
Coef. t-Value Coef. t-Value Coef. t-Value
Panel B: Results for Model 4a (control variables not listed)
ERA
0
À1.331 À5.4
**
À0.832 À4.2
**
À0.818 À6.5
**
BIGN
Ã
ERA
0
1.259 5.1
**
1.018 4.2
**
0.828 4.1
**
NATIONAL
Ã
ERA
0
0.882 2.7
**
1.001 3.4
**
0.622 2.2
*
Psuedo-R
2
0.120 0.192 0.194
N 5751 6465 5914
Bankrupt ?rms 7.6% 6.1% 6.7%
Note: BANKRUPT = 1 if the client declared bankruptcy or had an involuntary delisting, otherwise = 0; LSALES = the natural log of sales. See Notes to Tables 3
and 5 for other variable descriptions.
*
p < 0.05 (two-tailed).
**
p < 0.01 (two-tailed).
20
Geiger and Rama (2006) did not include LIT_INDUSTRY, but was a
control variable used by Myers et al. (2010). In addition, we included
RETURN that was not previously identi?ed and did not include a measure
for debt covenant defaults.
S.E. Kaplan, D.D. Williams / Accounting, Organizations and Society 37 (2012) 322–341 337
pooled data sets, indicating that over time Type I errors
were steadily increasing for regional audit ?rms compared
to BigN and national audit ?rms.
21
Table 9 reports the results for the Type I errors for the
GCAR sample excluding ?nancial institutions. Again, as
shown in Table 9 Panel A, there was no signi?cant differ-
ence in audit ?rm reporting accuracy among audit ?rm
classes in ERA1. Type I error was signi?cantly higher for
regional ?rms compared to BigN for ERA2 through ERA4,
and Type I error was signi?cantly higher for regional ?rms
compared to national ?rms for ERA2 and ERA3. The second
panel of Table 9 reports the trend over ERAs. The BigN
interaction term is signi?cant and positive in all three
pooled data sets and the national interaction term is signif-
icant and positive in the ?rst two pooled data sets. These
results indicate that Type I errors were steadily increasing
for regional audit ?rms compared to BigN and steadily
increasing for the ?rst three ERAs compared to national
audit ?rms. Overall, the results from both models generally
provide additional support for H
3
.
22
Research extensions
In this section, we report the results for three additional
extensions. The ?rst extension considers the potential ef-
fects of ?nancially stressed clients self-selecting an audit
?rm, which may introduce a bias in our analysis. Variables
Table 9
Logistic Regression on Bankruptcy (Type I Accuracy). (GCAR Firms excluding Financial Institutions).
Model 4 : BANKRUPT
¼ b
0
þ b
1
LSALES þ b
2
EXCHANGE þ b
3
LIT
I
NDUSTRY þ b
4
RETURN þ b
5
PROBANK þ b
6
BIGN þ b
7
NATIONAL þ e
Model 4a : BANKRUPT
¼ b
0
þ b
1
LSALES þ b
2
EXCHANGE þ b
3
LIT INDUSTRY þ b
4
RETURN þ b
5
PROBANK þ b
6
BIGN þ b
7
NATIONAL þ b
8
ERA
0
þ b
9
BIGN Ã ERA
0
þ b
10
NATIONAL Ã ERA
0
þ e
.
ERA1 ERA2 ERA3 ERA4
Coef. t-Value Coef. t-Value Coef. t-Value Coef. t-Value
Panel A: Results for Model 4
Model 4
INTERCEPT À3.676 À10.0
**
À7.155 À13.9
**
À7.572 À12.2
**
À3.082 À7.5
**
LSALES 0.233 5.2
**
0.421 10.5
**
0.431 7.8
**
0.122 3.0
**
EXCHANGE À0.020 À1.2 0.121 5.2
**
0.091 3.4
**
0.040 1.7
LIT_INDUSTRY 0.268 1.5 0.377 2.3
*
0.295 1.2 0.183 1.2
RETURN À0.230 À1.6 À0.308 À2.5
*
À0.258 À2.0
*
À0.197 À2.2
*
PROBANK 0.162 4.8
**
0.094 3.1
**
0.122 2.8
**
0.050 2.2
*
BIGN 0.106 0.4 1.090 3.8
**
1.492 4.5
**
1.224 2.0
*
NATIONAL 0.508 1.5 0.869 2.5
*
0.930 2.2
*
0.191 0.8
Psuedo-R
2
0.073 0.203 0.260 0.237
n 2145 3171 2896 2594
Bankrupt ?rms 7.3% 6.8% 3.8% 7.9%
ERA1 and ERA2 ERA2 and ERA3 ERA3 and ERA4
Coef. t-Value Coef. t-Value Coef. t-Value
Panel B: Results for Model 4a
Model 4a (control variables not listed)
ERA
0
À1.279 À4.8
**
À1.364 À5.0
**
À1.137 À7.9
**
BIGN
Ã
ERA
0
1.217 4.5
**
1.309 4.4
**
0.895 4.0
**
NATIONAL
Ã
ERA
0
0.929 2.6
**
0.878 2.1
*
À0.031 À0.1
Psuedo-R
2
0.133 0.223 0.136
N 5316 6067 5490
Bankrupt ?rms 7.0% 5.3% 5.7%
Note: See Notes to Tables 3, 5 and 8 for variable descriptions.
*
p < 0.05 (two-tailed).
**
p < 0.01 (two-tailed).
21
We reran Tables 8 and 9 replacing REGIONAL (=1) for a regional audit
?rm with the BIGN being re?ected in the intercept. The results were that
there were no signi?cant differences between BIGN and national ?rms.
22
We conducted several modi?cations to assess the robustness of our
results based on sample criteria, model speci?cation, variable modi?cation,
and sample partitioning. Our results remained robust following these
changes. In addition, to assess Type II misclassi?cations (e.g., the bankrupt
?rms that were not issued a GC report), we also reanalyzed Models 4 and 4a
(using the subsample of STRESSED ?rms that did not remain viable during
the next ?scal year. The untabulated results show no differences in
reporting accuracy in each ERA, and no trend in Type II misclassi?cations
over ERAs.
338 S.E. Kaplan, D.D. Williams / Accounting, Organizations and Society 37 (2012) 322–341
that may be associated with selecting an audit ?rm from a
particular audit ?rm class may also be associated with the
audit ?rm’s going concern reporting decision. To address
endogeneity concerns, we employed the Heckman (1979)
selection model. Following DeFond et al. (2002), we used
a two-stage regression with Model 1 used to estimate the
likelihood of engaging a high-quality audit ?rm, estimated
separately for each ERA, and then computed an inverse
Mills ratio using the parameters of this model.
23
The in-
verse Mills ratio was included in Models 2, 3, and 3a as a
control measure to test H
1
, H
2
, and H
3
. The results of the
two-stage regression are untabulated and are generally con-
sistent with the logistic regressions reported in Tables 5–7.
The second extension follows the research design of De-
Fond et al. (2002) and Carcello and Neal (2003), by exam-
ining only initial GC reports. We identi?ed 4691 ?rst-time
GC reports and reran Models 3 and 3a using ?rst-time GC
reports as the dependent variable.
24
The overall results
basically hold, with the exception of insigni?cant positive
?ndings for ERA
0
in Tables 6 and 7 for the national and regio-
nal audit ?rms in the last pooled regression (i.e., ERA3 and
ERA4). These insigni?cant ?ndings in this more stringent
subsample may be due to either a lack of power in the tests
or that national and regional ?rms reached their desired
level of client risk after ERA3.
The ?nal extension is based on a restricted sample of
Stressed ?rms. Given their size and scope, larger public
companies are expected to engage a BigN audit ?rm and
smaller public companies may be unable to attract a BigN
audit ?rm or be unwilling to pay the fee premiums asso-
ciated with a BigN audit ?rm. Stated differently, few, if
any, regional audit ?rms have the resources to audit lar-
ger public companies, and, until recently, neither did na-
tional ?rms (Boone et al., 2010). Given this relationship
between company size and audit ?rm class, we reduced
our Stressed sample by restricting it to only the bottom
quartile of BigN clients, and the top quartile of non-BigN
clients, by ERA. This subsample of clients presumably
would be potentially attractive to the largest number of
audit ?rms, and an analysis of this subsample is used to
provide further evidence on the relationship between
audit ?rm size and GC reporting across ERAs. Speci?cally,
we reran Models 2, 3, and 3a using this restricted sub-
sample of Stressed ?rms, and the untabulated results
were consistent with those reported in Tables 5–7.
Robustness tests
We tested the sensitivity of our results to different var-
iable measures. We substituted the log of sales (LSALES) for
our size measure, we computed long-term debt/assets
(LTDEBTAS) for our leverage measure, and we computed
net income/assets (ROA) for our pro?tability measure. We
also altered our de?nition of Stressed ?rms to include only
those ?rms with negative retained earnings and a net loss
in the current year. To control for the possibility of omitted
variables, we included DEBTAS, EBIT, and RETURN in Model
1, and market to book (MVBV) in Models 2, 3 and 3a. We
also included a measure for auditor expertise representing
the audit ?rm’s market share based on two-digit SIC codes,
scaled by sales (MSHARE). We deleted any Arthur Andersen
clients in ERA3. Finally, we reran the Type I misclassi?-
cation analysis based on COMPUSTAT/CRSP delistings,
excluding mergers and acquisitions, instead of bankruptcy
?lings. Consistent results were noted with these changes.
Discussion
Our research extends research on GC reporting by con-
sidering the strategic implications that result from gradual
changes in differentially sized audit ?rms’ portfolios over
time. We contend that over time BigN ?rms will be
increasingly unwilling to audit ?nancially stressed public
clients, and that regional ?rms will increasingly audit
?nancially stressed public clients. For ?nancially stressed
public clients, our research initially documents a gradual
shift away from BigN audit ?rms to regional audit ?rms.
Across the ERAs, increases in ?nancial stress were signi?-
cantly associated with a higher likelihood of being audited
by a regional ?rm. Our evidence also indicates that the
shift toward regional ?rms by ?nancially stressed public
clients is due, in part, to the actions of BigN ?rms. Thus,
our evidence indicates that BigN ?rms were increasingly
engaging in ex-ante conservatism (Krishnan et al., 2007)
through their unwillingness to audit ?nancially stressed
public companies.
Based on these changes, we tested three hypotheses
about audit ?rm reporting: regional audit ?rms will be
increasingly more likely to issue a GC report; BigN audit
?rms will be increasingly less likely to issue a GC report;
and in more recent time periods, regional audit ?rms will
be more likely to issue a GC report compared to larger
audit ?rms. Our results generally provide support for our
three hypotheses. In testing H
1
and H
2
, the results for the
last three pooled samples indicate an increasing tendency
for regional ?rms to issue GC reports, a decreasing ten-
dency for BigN ?rms to issue GC reports, and limited
changes in national ?rms propensity to issue a GC report.
Our tests for H
3
show that starting around ERA2, ?nan-
cially stressed public companies were signi?cantly more
likely to receive a GC report from a regional audit ?rm
compared to either BigN or national audit ?rms. Additional
analysis indicate that over time, Type I errors were gener-
ally increasing for regional ?rms, consistent with the no-
tion that regional ?rms exhibited increasingly
conservative GC reporting. Similar results were found
when controlling for the potential effects of clients self-
selecting an audit ?rm and a subset of ?nancially stressed
clients with the largest number of potential successor
audit ?rms.
Our ?ndings for the three hypotheses provide impor-
tant contributions. Our research shows that for a key
subset of the audit market, ?nancially stressed public cli-
23
Fargher and Jiang (2008) also attempt to control for endogeniety by
using a two-stage approach concerning the going concern decision of
Australian ?rms with the ?rst stage determining if the client should be a
potential receipt of the GC and the second stage if the audit actually issued
the going concern report.
24
Initial GC reports were identi?ed from Compact Disclosure and
AuditAnalytics.
S.E. Kaplan, D.D. Williams / Accounting, Organizations and Society 37 (2012) 322–341 339
ents, regional audit ?rms appear to engage in audit
behaviors that re?ect greater conservatism with respect
to going concern reporting compared to larger audit
?rms. Krishnan et al. (2007) refer to this activity as ex-
post conservatism. In this regard, our ?ndings indicate
that in more recent ERAs, regional ?rms are more likely
to issue a GC report to their ?nancially stressed public cli-
ents compared to larger ?rms. Because the economic con-
sequences for inferior audit quality in terms of a damaged
reputation and litigation costs are tied to audit ?rm size,
larger audit ?rms are believed to have the strongest
incentives to report in a conservative fashion (DeFond &
Francis, 2005). In recent ERAs, however, our results chal-
lenge this view. Speci?cally, our ?ndings document signif-
icant changes occurring to the riskiness of differentially
sized audit ?rms’ portfolios and relate these changes to
differentially sized audit ?rms’ tendencies to issue GC re-
ports more or less conservatively.
Several factors may motivate differentially sized audit
?rms to change the relative conservatism in their GC
reporting in response to gradual changes in the riskiness
of their portfolios. First, Francis and Krishnan (2002)
speculate that regional ?rms are likely to report more
conservatively (e.g., ex-post conservatism) as they audit
more risky clients and as the riskiness of the ?rm’s port-
folio increases. Regional audit ?rms, because they are
smaller, are less able to withstand and survive large lit-
igation-related damages. Second, Francis and Krishnan
(2002) also speculate that BigN ?rms, increasingly rely-
ing on ?rm-wide decision aids and technologies to make
acceptance and continuation decisions (e.g., ex-ante
conservatism), may believe that client screening has
effectively dealt with risks associated with ?nancially
stressed clients.
In comparing our results to previous GC research, our
study is noteworthy, in part, as it provides longitudinal
evidence across four discreet ERAs covering 22 years,
which includes recent time periods. Also, our research con-
siders three classes of audit ?rms. While considering three
categories of audit ?rms is not novel, we believe that the
growth of national audit ?rms over time makes it an
increasingly important design choice. In this regard, our
results indicate that for several ERAs, GC reporting differs
between national and regional ?rms. For example, results
shown in Tables 6 and 7 indicate that national ?rms were
signi?cantly more likely to issue a GC report compared to
regional ?rms in the last three ERAs. Also, the results from
Table 5 generally indicate that there were no differences in
GC reporting across ERAs for national ?rms, but regional
?rms were increasingly more likely to issue a GC report
over ERAs. Thus, our results suggest that going forward,
it may be increasingly important to use three classes of
audit ?rms in GC research.
Lastly, our results raise an interesting policy issue re-
lated to the ability of ?nancially stressed clients to hire
an audit ?rm. While the results of our study indicate that
?nancially stressed clients are still able to hire an audit
?rm, their options appear to be decreasing over time. To
the extent that their audit ?rm options continue to shrink
over time, some ?nancially stressed public companies
may be unable to hire an audit ?rm in the future. Evalu-
ating the implications and potential consequences to
these ?rms represents an important area for further
research.
Acknowledgements
We would like to acknowledge helpful comments from
Christopher Chapman (Editor), two anonymous reviewers,
Anil Arya, Randy Beatty, Matt Hart, Krishnagopal Menon,
Kurt Pany, Janet Samuels, Eric Spires, Gary Taylor, Dale
Williams, and Teri Ziegler.
References
American Institute of Certi?ed Public Accountants (AICPA) (1988). The
auditor’s consideration of an entity’s ability to continue as a going
concern. SAS No. 59. New York, NY: AICPA.
American Institute of Certi?ed Public Accountants (AICPA) (1997).
Consideration of fraud in a ?nancial statement audit. SAS No. 82.
New York, NY: AICPA.
Arthur Andersen & Co., Coopers & Lybrand, Deloitte & Touché, Ernst &
Young, KPMG Peat Marwick, and Price Waterhouse (1992). The
liability crisis in the United States: Impact on the accounting
profession. A Statement of Position (August 6): pp. 1–8.
Asthana, S., Balsam, S., & Kim, S. (2009). The effect of Enron, Andersen, and
Sarbanes-Oxley on the US market for audit services. Accounting
Research Journal, 22, 4–26.
Ayers, S., & Kaplan, S. (1998). Potential differences between engagement
and risk review partners and their effect on client acceptance
judgments. Accounting Horizons, 12, 139–153.
Beaver, W., & Ryan, S. (2000). Biases and lags in book value and their
effects on the ability of the book-to-market ratio to predict book
return on equity. Journal of Accounting Research, 38, 137–148.
Bell, T., Bedard, J., Johnstone, K., & Smith, E. (2002). Krisk
SM
: A
computerized decision aid for client acceptance and continuance
risk assessments. Auditing: A Journal of Practice & Theory, 21, 97–113.
Beneish, M., Hopkins, P., Jansen, I., & Martin, R. (2005). Do auditor
resignations reduce uncertainty about the quality of ?rms’ ?nancial
reporting? Journal of Accounting and Public Policy, 24, 357–390.
Blay, A., & Geiger, M. (2001). Market expectations for ?rst-time going-
concern recipients. Journal of Accounting Auditing and Finance, 16,
209–226.
Blouin, J., Grein, B., & Rountree, B. (2007). An analysis of forced auditor
change: The case of former Arthur Andersen clients. The Accounting
Review, 82, 621–650.
Bockus, K., & Gigler, F. (1998). A theory of auditor resignation. Journal of
Accounting Research, 36, 191–208.
Boone, J., Khurana, I., & Raman, K. (2010). Do the Big 4 and the second-tier
?rmsprovide audits of similar quality? Journal of Accounting and Public
Policy, 29, 330–352.
Butler, M., Leone, A., & Willenborg, M. (2004). An empirical analysis of
auditor reporting and its association with abnormal accruals. Journal
of Accounting and Economics, 37, 139–165.
Carcello, J., & Neal, T. (2003). Audit committee characteristics and auditor
dismissals following ‘‘new’’ going-concern reports. The Accounting
Review, 78, 95–117.
Carcello, J., & Palmrose, Z. (1994). Auditor litigation and modi?ed
reporting on bankrupt clients. Journal of Accounting Research, 32, 1–30.
Center for Audit Quality (CAQ) (2008). Report of the major public
company audit ?rms to the department of the Treasury Advisory
Committee on the auditing profession (January 23). New York: CAQ.
Choi, J., Doogar, R., & Ganguly, A. (2004). The riskiness of large audit ?rm
client portfolios and changes in audit liability regimes: Evidence from
the US audit markets. Contemporary Accounting Research, 21, 747–785.
Cooper, D., & Robson, K. (2006). Accounting, professions and regulation:
Locating the sites of professionalization. Accounting, Organizations and
Society, 31, 415–444.
Cox, J. (2006). The oligopolistic gatekeeper: The U.S. accounting
profession. Duke Law School Legal Studies Paper No. 117.
DeFond, M., & Francis, J. (2005). Audit research after Sarbanes-Oxley.
Auditing: A Journal of Practice & Theory, 24, 5–30.
DeFond, M., & Lennox, C. (2011). The effect of SOX on small auditor exits
and audit quality. Journal of Accounting and Economics, 52, 21–40.
DeFond, M., Raghunandan, K., & Subramanyam, R. (2002). Do non-audit
service fees impair auditor independence? Evidence from going
340 S.E. Kaplan, D.D. Williams / Accounting, Organizations and Society 37 (2012) 322–341
concern audit opinions. Journal of Accounting Research, 40,
1247–1274.
Fargher, N., & Jiang, L. (2008). Changes in the audit environment and
auditors’ propensity to issue going-concern opinions. Auditing: A
Journal of Practice & Theory, 27, 55–78.
Fogarty, T. (1996). The imagery and reality of peer review in the U.S.:
Insights from institutional theory. Accounting, Organizations and
Society, 21, 243–267.
Francis, J. (1994). Auditing, hermeneutics and subjectivity. Accounting,
Organizations and Society, 19, 235–269.
Francis, J. (2004). What do we know about audit quality? The British
Accounting Review, 36, 345–368.
Francis, J., & Krishnan, J. (1999). Accounting accruals and auditor
reporting conservatism. Contemporary Accounting Research, 16,
135–165.
Francis, J., & Krishnan, J. (2002). Evidence on auditor risk-management
strategies before and after The Private Securities Litigation Reform
Act of 1995. Asia-Paci?c Journal of Accounting & Economics, 9,
135–157.
Francis, J., Philbrick, D., & Schipper, K. (1994). Shareholder litigation and
corporate disclosures. Journal of Accounting Research, 32, 137–164.
Frost, C. (1994). Uncertainty-modi?ed audit reports and future earnings.
Auditing: A Journal of Practice & Theory, 13, 22–35.
Geiger, M., & Raghunandan, K. (2001). Bankruptcies, audit reports, and the
Reform Act. Auditing: A Journal of Practice & Theory, 20, 187–196.
Geiger, M., Raghunandan, K., & Rama, D. (2006). Auditor decision-making
in different litigation environments: The Private Securities Litigation
Reform Act, audit reports and audit ?rm size. Journal of Accounting &
Public Policy, 25, 332–353.
Geiger, M., & Rama, D. (2006). Audit ?rm size and going-concern
reporting accuracy. Accounting Horizons, 20, 1–18.
Gendron, Y. (2001). The dif?cult client-acceptance decision in Canadian
audit ?rms: A ?eld investigation. Contemporary Accounting Research,
18, 283–310.
Gendron, Y. (2002). On the role of the organization in auditors’ client
acceptance decisions. Accounting, Organizations and Society, 27,
659–684.
Ghosh, A., & Moon, D. (2005). Auditor tenure and perceptions of audit
quality. The Accounting Review, 80, 585–612.
Gul, F., & Tsui, J. (1998). A test of the free cash ?ow and debt monitoring
hypotheses: Evidence from audit pricing. Journal of Accounting &
Economics, 24, 219–237.
Heckman, J. (1979). Sample selection bias as a speci?cation error.
Econometrica, 47, 153–161.
Hindo, B. (2003). Audit clients get the heave-ho. Business Week, 7.
Hoffman, V., Joe, J., & Moser, D. (2003). The effect of constrained
processing on auditors’ judgments. Accounting, Organizations and
Society, 28, 699–714.
Hogan, C., & Martin, R. (2009). Risk shifts in the market for audits: An
examination of changes in risk for ‘‘second tier’’ audit ?rms. Auditing:
A Journal of Practice & Theory, 28, 93–118.
Holder-Webb, L., & Wilkins, M. (2000). The incremental information
content of SAS No. 59 going-concern opinions. Journal of Accounting
Research, 38, 209–219.
Hopwood, W., McKeown, J., & Mutchler, J. (1994). A reexamination of
auditor versus model accuracy within the context of the going-
concern opinion decision. Contemporary Accounting Research, 10,
409–431.
Johnstone, K., & Bedard, J. (2004). Audit ?rm portfolio management
decisions. Journal of Accounting Research, 42, 659–690.
Jones, F., & Raghunandan, K. (1998). Client risk and recent changes in the
market for audit services. Journal of Accounting and Public Policy, 17,
169–181.
Karr, S. (2005). Auditor bandwidth affecting clients. Big 4, second six.
Compliance Week (February 1).
Keinath, A., & Walo, J. (2004). Audit committee responsibilities. The CPA
Journal, 74, 22–28.
Kinney, W. Jr., (2005). Twenty-?ve years of audit deregulation and re-
regulation: What does it mean for 2005 and beyond? Auditing: A
Journal of Practice & Theory, 24, 89–109.
Krishnan, J., & Krishnan, J. (1997). Litigation risk and auditor resignations.
The Accounting Review, 72, 539–560.
Krishnan, J., Krishnan, J., & Stephens, R. (1996). The simultaneous relation
between auditor switching and audit opinion: An empirical analysis.
Accounting and Business Research, 26, 224–236.
Krishnan, J., Raghunandan, K., & Joon, S. (2007). Were former Andersen
clients treated more leniently than other clients? Evidence from
going-concern modi?ed audit opinion. Accounting Horizons, 21,
423–435.
Landsman, W., Nelson, K., & Rountree, B. (2009). Auditor switches in the
pre- and post-Enron eras: Risk or realignment? The Accounting Review,
84, 531–558.
Latham, C., & Linville, M. (1998). A review of the literature in auditor
litigation. Journal of Accounting Literature, 17, 175–213.
Lennox, C. (2000). Do companies successfully engage in opinion-
shopping? Evidence from the UK. Journal of Accounting & Economics,
29, 321–337.
Louwers, T. (1998). The relation between going-concern opinions and the
auditor’s loss function. Journal of Accounting Research, 36, 143–156.
MacDonald, E. (1997). More accounting ?rms are dumping risky clients.
Wall Street Journal (April 25): Section 3, 2.
Matsumura, E., Subramanyam, K., & Tucker, R. (1997). Strategic auditor
behavior and going-concern decisions. Journal of Business Finance &
Accounting, 24, 727–758.
Menon, K., & Williams, D. (2010). Investor reaction to going concern audit
reports. The Accounting Review, 85, 2075–2106.
Mills, P., & Young, J. (1999). From contract to speech: The courts and CPA
licensing laws 1921–1996. Accounting, Organizations and Society, 24,
243–262.
Morgan, J., & Stocken, P. (1998). The effects of business risk on audit
pricing. Review of Accounting Studies, 3, 365–385.
Mutchler, J. (1984). Auditor perceptions of the going concern opinion.
Auditing: A Journal of Practice & Theory, 3, 17–30.
Myers, L., Schmidt, J., & Wilkins, M. (2010). An investigation of recent
changes in going concern reporting decision among BigN and non-
BigN auditors. Working paper: University of Arkansas.
Palmrose, Z. (1988). An analysis of auditor litigation and audit service
quality. The Accounting Review, 63, 55–73.
Power, M. (2003). Auditing and the production of legitimacy. Accounting,
Organizations and Society, 28, 379–394.
Preston, A., Cooper, D., Scarbrough, D., & Chilton, R. (1995). Changes in the
code of ethics of the U.S. accounting profession, 1917 and 1988: The
continual quest for legitimation. Accounting, Organizations and Society,
20, 507–546.
Rama, D., & Read, W. (2006). Resignations by the Big 4 and the market for
audit services. Accounting Horizons, 20, 97–109.
Shu, S. (2000). Auditor resignations: Clientele effects and legal liability.
Journal of Accounting and Economics, 29, 173–205.
St. Pierre, K., & Anderson, J. (1984). An analysis of the factors associated
with lawsuits against accountants. The Accounting Review, 59,
242–263.
Teoh, S. (1992). Auditor independence, dismissal threats, and the market
reaction to auditor switches. Journal of Accounting Research, 30, 1–23.
The American Assembly (2005). The future of the accounting profession:
Auditor concentration. The American Assembly, New York, NY:
Columbia University Press.
Weber, J., & Willenborg, M. (2003). Do expert informational
intermediaries add value? Evidence from auditors in microcap IPOs.
Journal of Accounting Research, 41, 681–720.
Winograd, B., Gerson, J., & Berlin, B. (2000). Audit practices of
PricewaterhouseCoopers. Auditing: A Journal of Practice & Theory, 19,
175–182.
Wyatt, A. (2004). Accounting professionalism – They just don’t get it!
Accounting Horizons, 18, 45–54.
Zeff, S. (2003). How the US accounting profession got to where it is today,
Part I. Accounting Horizons, 17, 189–205.
Zmijewski, M. (1984). Methodological issues related to the estimation of
?nancial distress prediction models. Journal of Accounting Research,
22, 59–82.
S.E. Kaplan, D.D. Williams / Accounting, Organizations and Society 37 (2012) 322–341 341

doc_881989043.pdf