-
1
OFFICE-LEVEL AUDIT PARTNER ROTATION AND GOING CONCERN
OPINION ISSUANCE
John Goodwin
Associate Professor
School of Accounting and Finance
The Hong Kong Polytechnic University
8/F, Li Ka Shing Tower
Hunghom, Kowloon, HONG KONG
Tel: (852) 2766-7077 Fax: (852) 2330-9845
Email: [email protected]
Ferdinand A. Gul
Professor
School of Business
Monash University
Jalan Lagoon Selatan,
Bandar Sunway, 46150,
Selangor Darul Ehsan,
MALAYSIA
Tel: (603) 5514-6000 Fax: (603) 5514-6192
Email: [email protected]
Acknowledgements: The comments of Andy Chui, Robert Halperin,
Huang Xu,
Sydney Leung, Huiwen Liu, Chung-ki Min, Bin Srinidhi, Steven
Salterio, Srinivasan
Sankaraguruswamy, Nancy Lixin Su, Steve Wei, Franco Wong,
Donghui Wu, Wayne
Yu, workshop participants at the City University of Hong Kong,
The Hong Kong
Polytechnic University, participants at the 2011 European
Accounting Conference in
Rome, the editor, and two anonymous reviewers are much
appreciated. We are grate-
ful to The Hong Kong Polytechnic University for financial
support (project codes: G-
U514 and G-U774) and the research assistance of Susanna
Andreassen and Angel
Sung. We are responsible for any errors.
-
2
OFFICE-LEVEL AUDIT PARTNER ROTATION AND GOING CONCERN
OPINION ISSUANCE
We find that office-level audit partner rotation is negatively
related to going concern
opinion issuance for Australian, financially-distressed clients
audited by the Big4
audit firms. When office-level audit partner rotation is higher,
issued going concern
opinions are more accurate and the positive relation between
high total accruals and
going concern opinion issuance is weaker. But these relations
only exist for the years
just after introduction of mandatory partner rotation. We
interpret this evidence as
auditors behaving less conservatively with their clients as
office level audit partner
rotation increases, and believe this change in auditor behavior
is motivated by a desire
to reduce the risk of clients switching. We find no relation
between audit partner
rotation at the client level, partner tenure and going concern
opinion issuance and we
reconcile our results for partner tenure with prior studies
reporting a negative relation.
Keywords: office -level audit partner rotation; audit partner
tenure; auditor
conservatism; going concern opinion;
Data availability: data used in this study are available from
public sources as
identified in the paper.
-
3
1. INTRODUCTION
Whether auditor rotation can improve auditor independence has
attracted considerable
interest over the last two decades. Early proponents argue that
auditor rotation can
prevent client-auditor long term relationships that could impair
independence
(DeAngelo 1981; Johnson and Lys 1990; Deis and Giroux 1992;
Raghunathan et al.
1994). As a result of these early calls and corporate scandals
such as HIH in Australia
and Enron in the U.S., laws were passed requiring audit partners
to rotate off clients.
However these rules could have unintended consequences (Nelson
2006). In this
spirit we examine if there is a difference in audit quality
between audit offices that
rotate partners relatively more than others.
We use Australian data to examine this question as the names of
audit partners are
required to be publicly disclosed on the audit report in that
country, and to date
studies have only examined the client-specific link between
partner rotation and audit
quality with conflicting results. The evidence in Hamilton et
al. (2010), Carey and
Simnett (2006) and Ye et al. (2011) suggests that partner
rotation improves audit
quality, Lai and Cheuk (2005) and Chi and Huangs' (2005)
evidence suggests partner
rotation has no affect, and evidence in Chen et al. (2008) and
Chi et al. (2009)
suggests that partner rotation worsens audit quality.
Auditors do not work in isolation but in teams and across teams,
and auditor behavior
is pervasive across an office. For example, major events in the
workplace can change
staff behavior, and in their work auditors use decision aids to
provide consistency in
their decision making. The introduction of mandatory partner
rotation is an event that
can alter auditor behavior because it could have disruptive
effects for the office staff
-
4
or it could change auditor conservatism. If rotation has
reverberating effects then this
is an unintended consequence of audit partner rotation.
That changes in worker teams could have disruptive effects is
not new and has been
identified in the organizational literature (see for example,
Gully et al. 1995; Liang et
al. 1995 and Tesluk and Mathieu 1999). If these types of
disruptive effects
reverberate then rotation can have adverse consequences for
audit quality across an
audit office. Partners believe that rotation increases the
likelihood of client switching
to other audit firms (Daugherty et al. 2009). Faced with
impending client loss from
partner rotation, auditors could be less conservative with their
clients because auditor
conservatism increases the risk of clients switching (Krishnan
1994). That auditors
have differences in their levels of conservatism and change
those levels is also not
new. Cahan and Zhang (2006), Geiger et al. (2006), and Feldman
and Read (2010)
provide supporting evidence. We examine which of these two
explanations is more
likely in explaining the results.
For each audit office for each fiscal year we measure the office
rotation rate in four
different ways which we detail below in Section III, but
generally, the rate is the
number of scaled mandatory partner rotations. We then test
whether these rates (in
logarithmic form) are associated with the auditor’s propensity
to issue a going
concern opinion for a sample of financially-distressed
Australian clients. Our results
using 1,619 client years audited by Big 4 firms for 2004 to
2010, show that offices
with a higher rotation rate are associated with a lower
propensity to issue going
concern opinions. We find that going concern audit opinions
issued in high rotation
rate offices are more accurate than low rotation rate offices,
and that high rotation rate
-
5
offices are less conservative with their clients. We also find
that these relations are
not persistent. Collectively, this evidence suggests that high
rotation rate offices have
lower audit quality for the average client after the initial
shock of mandatory partner
rotation's introduction. We argue that auditors in high rotation
rate offices are
motivated to lower conservatism to reduce the risk of clients
switching.
This study makes two main contributions. In conducting the first
examination of the
office-level effects of partner rotation on audit quality, we
provide evidence of an
unintended consequence of partner rotation rules. That being
lower auditor
conservatism in the years immediately after introduction of the
rotation rule.1 Lower
conservatism is not observed after three years after
introduction. Second, we show
that partner tenure is unimportant in explaining going concern
opinion issuance,
mainly because tenure is correlated with board size and a proxy
for a partner's recent
workload, both of which are correlated with the going concern
dependent variable.
We believe that prior studies could have reached erroneous
conclusions about partner
tenure and going concern opinion issuance because we find that
long-tenured partners
are more accurate with their going concern opinions in the
presence of these two
variables.
The next section presents the background of the study and the
development of the
hypothesis, while Section III discusses the research
methodology. Section IV presents
and discusses the results and Sections V and VI present results
and provide discussion
of additional tests. Section VII is a reconciliation with
related studies and Section VIII
concludes.
1 This suggests that other audit quality proxies like earnings
management could be worse although we
do not examine them for brevity.
-
6
II. BACKGROUND AND HYPOTHESIS
Partner Rotation Requirements
In 2002 the Institute of Chartered Accountants in Australia and
CPA Australia issued
Professional Statement F1, requiring partner rotation (hereafter
rotation) after seven
consecutive years beginning with assurance reports dated after
December 30 2003.
For the vast majority of clients, this rule is equivalent to
"the audits of clients with
fiscal years ending on or after 31 December 2003."2 In 2004
rotation was enshrined
in legislation with the passing of the Corporate Law Economic
Reform Program 9
(CLERP 9 Commonwealth of Australia 2004), and this law requires
rotation after five
consecutive years for fiscal years beginning after June 30 2006.
For the vast majority
of clients, this rule is equivalent to "auditing clients with
fiscal years ending on or
after 30 June 2007". F1 was subsequently amended to a five-year
rotation period to
mirror the law. Similar requirements are included in the
Sarbanes-Oxley Act U.S.
House of Representatives 2002) where rotation is required every
five years, although
an important difference is that in Australia the partner may
begin auditing the client
after two years and the requirement is five years in the U.S.
The tenor of the
legislation is that rotation is beneficial for audit quality and
that more rotation is
preferred. There are no limits to the number of rotations in any
of these
pronouncements and non mandatory rotation continues at only
slightly higher rates
after 2003 than before 2003, as Table 2 shows.3
2 There are no listed clients with fiscal year ends of October
through November in 2003 and the annual report had
to be lodged within 75 days from fiscal year end in 2003,
meaning that most September year end clients would
have lodged their accounts (along with audit reports signed) by
31 December 2003. Perhaps the only exceptions
are clients who have not lodged within the required time and the
audit report is signed on or after 31 December
2003. These are probably only suspended clients and these are
more common with the non Big4 clients. 3 It is worth noting that
voluntary rotation could have occurred before F1's start date as a
result of the Australian
Government, in 1996, recommending rotation, and the
International Federation of Accountants (IFAC) updated
“Independence Requirements for Assurance Engagements” in
November 2001 specifying that lead partners should
rotate after seven consecutive years.
-
7
Prior Studies
While there are no studies that examine rotation at the office
level, we are aware of
seven that examine it at the client level. The most closely
related studies to the
present study are Carey and Simnett (2006), who report that
partner tenure (hereafter
tenure) more than seven years is associated with a lower
likelihood of issuing a going
concern opinion and tenure less than three years is associated
with a higher likelihood,
for a sample of distressed Australian clients. They conclude
that longer tenure is
associated with lower audit quality.4 And Ye et al. (2001), who
report that tenure
(measured continuously) is negatively associated with the
likelihood of going concern
opinion issuance, using a sample of distressed Australian
clients. As we find no
relation between tenure and going concern opinion issuance, we
reconcile these
studies' results with ours in Section VII.
We briefly cover the other five studies using different audit
quality proxies to us to
highlight the mixed results among them. Hamilton et al. (2010)
find that rotation is
associated with less aggressive accounting for a sample of
Australian clients with high
prior-period accruals. They conclude that rotation results in
"some improvement in
accounting quality" (Hamilton et al. p 1). Lai and Cheuk (2005)
use a sample of
Australian clients to examine the relation between rotation and
financial reporting
timeliness, proxied by the audit reporting lag. They find no
relation between the two.
Using Taiwanese data, Chi and Huang (2005) initially find that
longer tenure is
associated with higher abnormal accruals but that audit firm
tenure subsumes much of
tenure's importance when added to their model. These two studies
suggest that
rotation has no effect on audit quality. Chen et al. (2008) find
that long tenure is
4 Carey and Simnett (2006) also find that long tenure is not
associated with discretionary accruals and
some evidence that long tenure is positively associated with
just missing and just beating earnings
benchmarks.
-
8
negatively associated with discretionary accruals after
controlling for audit firm
tenure using Taiwanese data, and Chi et al. (2009), find some
evidence that rotation is
positively associated with earnings management also using
Taiwanese data. These
two studies suggest that rotation is detrimental for audit
quality.
What the literature has not considered is whether rotation has
any effects for audit
quality across an audit office. A reverberating effect from
rotation could manifest
through a psychological effect on staff, including lower audit
team cohesiveness. As
noted, the management literature is supportive here. We call
this office-level
disruption and expand on it below. Another way is via risk
management, including
electronic decision aids and their application of them. Auditors
use electronic
decision aids in client continuance decisions because they can
provide consistency in
decision making (Bedard et al. 2008 p 200). These aids, their
outputs and auditors
application of the outputs, may not be static over time or
across offices. The evidence
in Kachelmeier and Messier (1990), Bedard and Graham (2002) and
O'Donnell and
Schultz (2005) can support these arguments. We also argue that
rotation can effect
auditor conservatism and we expand on this below. Either of the
above two
explanations could be observed in regressions because the
majority of clients (about
94 percent - see Table 2) are not subject to mandatory rotation
in a fiscal year. 5
Audit Quality at the Office Level
Office Level Disruption
Auditors work in teams and can work across teams. We are not
aware of any related
research but rotation could change the composition of teams in
non rotated clients. A
5 The total of within office and across office mandatory
rotations for the 2004 to 2010 years is 361 and
the total client years is 5,789 giving a percentage of about 6
percent (see Table 2).
-
9
partner may prefer to work with a particular team member,
thereby 'forcing' a switch
of a team member, and team members with particular types of
experience may move
to another client if the incoming partner is relatively less
experienced in the client's
industry or with a client of that size for example. Irrespective
of the reason, team
changes can alter the team's psychological traits such as
cohesiveness, norms, affect
and cognition, and the extent of within-group agreement; which,
in turn can affect the
team's performance, judgments and effectiveness. Turnover in
teams can also
adversely impact a team's ability to learn. The management
literature supports these
conjectures (see for example, Gully et al. 1995; Jehn 1995;
Forgas 1990; Liang et al.
1995; Tesluk and Mathieu 1999 and Carley 1992). On the other
hand, job rotation
can be regarded by employees as a career-enhancing experience.
In a study of 255
finance employees in a large pharmaceutical company, Campion et
al. (1994) found
job rotation to be useful for enhancing employees' business
skills, including an
improved knowledge of the organization's functional areas and
operations. This
suggests that rotation of other audit team members can have
beneficial effects for
audit quality. Another reason is that the introduction of
mandatory rotation could
have reduced the incidence of rotation of other audit team
members, perhaps because
the audit office believes that team continuity should be
preserved as far as possible in
a mandatory rotation regime. Importantly, Campion et al.'s
(1994) study also
identified costs of job rotation for non-rotated employees;
namely, a decrease in
productivity, job satisfaction and motivation in teams that both
gained and lost rotated
employees. Thus, even if the remaining audit team members do not
change when a
partner rotates, it does not necessarily follow that there will
be no deleterious effects
from rotation. To the extent that these costs of rotation are
pervasive then, on balance,
rotation is expected to have adverse consequences for audit
quality across the office.
-
10
While the above arguments could apply at the client level the
incentives are stronger
for auditors to behave otherwise at that level. Regulatory and
professional body
scrutiny focuses on rotation not on non rotation. For example,
the Australian
Securities and Investments Commission (ASIC) conducts periodic
reviews of audit
firms as part of its audit inspection program and issues policy
statements about
rotation, not specifically about non rotation. This scrutiny
could influence partners
with rotated clients, more than partners with non rotated
clients to perform within the
spirit of the legislation. Additionally, prior empirical
evidence supports a positive
relation between short tenure and going concern opinion issuance
(Carey and Simnett
2006 and Ye et al. 2011), although this relation is for the
average client and rotated
clients are a small percentage of them.6 With specific regard to
disruption, some prior
research is consistent with the link between rotation at the
client level and poor
quality audits. For example, using semi-structured interviews
with office manager
partners and surveys distributed to U.S. audit partners,
Daugherty et al. (2009) find
that rotation may have a negative overall impact on audit
quality through its dilution
of client specific and industry expertise. Consequently,
frequent rotation increases the
likelihood that partners will specialize in multiple industries
at the expense of industry
depth. This suggests that rotation increases the partner work
load and the workload of
other members of the audit team during the first few years of
the rotation, that could
result in lower audit quality. But partners believe that while
client-specific
information is lost at rotation this is not going to affect
audit quality (Daugherty et al.
2009).7
6 The total of mandatory and non mandatory rotations for the
2004 to 2010 years is 1,190 and the total
client years is 5,789 giving a percentage of about 21 percent.
Under a non mandatory rotation regime
the percentage is smaller (see Table 2). 7 Perhaps partners
believe that the knowledge lost can be compensated for by
increasing audit effort.
-
11
Auditor Conservatism
That auditors change their level of conservatism in response to
external shocks has
been shown. Relative to non Big4 auditors, Geiger et al. (2006)
reports that Big4
auditors were less conservative after introduction of the
Private Securities Litigation
Reform Act. Auditors were more conservative after introduction
of the Sarbanes
Oxley Act (2002) in the U.S. (Geiger et al. 2005 and Myers et
al. 2008) and Cahan
and Zhang (2006) report that ex-Arthur Andersen clients were
treated more
conservatively by their new auditors. Auditor conservatism is
also not static. While
auditor conservatism increased immediately post SOX, its level
returned to pre-Enron
levels in later years (Feldmann and Read 2010).
Auditor conservatism comes with the risk of the client switching
to another audit firm
(Krishnan 1994), and auditors may behave less conservatively
when faced with risk of
client loss (Vanstraelen 2002; Vanstraelen 2003).8 In the
present study we examine
financially-distressed clients as they are more likely to switch
(Haskins and Williams,
1990). Daugherty et al. (2009) report that the partners believe
that rotation increases
the risk of switching, but reasons are not provided.9
The introduction of mandatory rotation is an external shock to
audit firms and offices
that increases the risk of client switching because rotation
changes the dynamic
between the audit office and the client. It is a reasonable
assumption that offices
would permit their partners who have good client relations to
remain with the client
8 For samples of Belgian clients, Vanstraelen (2002) reports a
negative relation between going concern
opinion issuance and the proportion of clients lost by an audit
firm in the prior fiscal year; and
Vanstraelen (2003) reports that auditors are four times more
likely to switch after receiving a going
concern opinion in the last year of a mandated audit firm
period, than in other years of that period. 9 The survey also
examined the impact of the rotation rules on partners’ quality of
life.
-
12
more often than replace them, meaning that rotation is not
likely to improve this
dynamic. We are aware of no empirical studies on rotation and
client switching,
perhaps because disclosure of the partner name is not required
in most countries, but
results from surveys suggest that a client's relations with its
audit team could be
correlated with the rotation event. Addams and Davis (1994,
1996) and Addams et al.
(2002) report that personal relationships between the client and
the auditor featured
highly on reasons for audit firm selection, and that four
factors all related to the
incumbent auditor's poor service, were most important for
switching.10
Supporting
evidence is in Fried and Schiff (1981) and Eichenseher and
Shields (1983).
Mandatory rotation introduces uncertainty into the quality of
this relation.
A useful strategy for auditors to reduce the risk of client loss
is to lower their
conservatism. Lowering conservatism is more likely to succeed
than is lowering fees,
even for distressed clients, because a client's bankruptcy risk
can increase upon
receiving a going concern opinion for example (George et al.
1996; Pryor and Terza
2002; and Vanstraelen 2003). Thus if auditors behave less
conservatively with their
clients, perhaps due to the perceived threat of switching, then
the relation between the
office rotation rate and going concern opinion issuance should
be negative.
As noted above, regulatory pressures and prior research suggests
that the relation
could be different at the client level and these arguments seem
to apply equally
whether disruption or lower conservatism is the explanation.
With specific regard to
lower auditor conservatism, the threat of client loss is likely
higher for non rotated
clients because any 'damage has been done' with them. That is,
given that clients
10
The four factors in order of importance are: Not sufficiently
proactive, Lack of responsiveness, No
new ideas to help the company and Inadequate understanding of
the company (Addams et al. 2002 p
63).
-
13
place a high value of relations with the auditor, auditors can
do less to appease a
rotated client than a non rotated client. As a consequence there
is less incentive for
auditors to lower conservatism for rotated clients vis-à-vis non
rotated clients.
We test the alternate-form hypothesis (H1) as the predicted
relations under either
explanation are the same:
H1: There is a negative relation between the office-level
rotation rate and the
issuance of a going concern opinion
III. RESEARCH METHOD
An office-level rotation is coded as unity when a client has a
different audit sign-off
partner to that of the prior fiscal year and the audit firm and
office which audits the
client are the same for those two fiscal years. We separate
mandatory from non-
mandatory rotations and focus on mandatory rotations to provide
more meaningful
policy implications from the results.
We proxy for reverberating effects using office rotation rates.
We expect them to
capture the collective effects of rotation on an audit office's
behavior, similarly to
finance studies using the proportion of busy directors to
capture the collective effects
of directors' busyness on a client boards' behavior (see for
example, Ferris et al. 2003
and Fich and Shivdasani 2006).
There are no prior empirical studies or theory on office-level
partner rotation so we
treat the problem as an empirical exercise and use four
different measures.11
Our first
11 We are grateful to the reviewer for identifying different
measures of office-level rotation rates.
-
14
experimental variable is the number of mandatory rotations
divided by the number of
clients in the office (MCLI), the second is the number of unique
partners involved in
mandatory rotations divided by the number of clients in the
office (MUCLI). The
number of mandatory rotations divided by the number of partners
in the office is the
third (MPAR), and the number of unique partners involved in
mandatory rotations
divided by the number of partners in the office is the fourth
(MUPAR). The
corresponding non-mandatory rates are denoted NMCLI, NMUCLI,
NMPAR and
NMUPAR respectively. We include these variables in the models
because they are
generally correlated with the dependent variable GCREPORT, and
they are correlated
with their mandatory counterparts (see Table 4). In multivariate
tests we add unity to
these rates prior to taking the logarithm to improve symmetry
and to reduce the effect
of outliers.
Although the office rotation rates are strongly positively
correlated within each of the
mandatory and non mandatory groups (see Table 4), being the
proportion of clients
affected by rotations we favor MCLI and NMCLI as measures of the
impact of
rotation on an office. Audit offices generate revenue only from
clients, so auditor
behavior is probably more closely related to its client base
than with its partner base.
Using number of partners as the base can result in office
rotation rates that are
inconsistent with expectations when one is interested in the
effect of office rotation on
auditor behavior. Using unique partners in the numerator can
also result in
inconsistent numbers because an office that rotates a partner is
coded as unity
irrespective of the number of his rotations. For example, it is
difficult to argue that
any number of rotations of the same partner has the same effect
on auditor behavior in
-
15
the office ceteris paribus. To illustrate these points we
provide two examples for
mandatory rotations from a large and a smaller-sized audit
office in Appendix 1.
We examine only the Big4 audit firms as we do not have access to
the audit firm's
client lists, meaning that our measures of the office level
rotation rate is an estimate of
the 'true' rate. Measurement error in this rate is probably
lower using the Big4 as their
client base likely comprises a larger proportion of listed
clients. The coefficients for
the office rotation rates are insignificant in all models for
the non Big4 sample
(untabulated) which could be caused by higher measurement error
in the rates.
Data
The sample of Australian listed clients comprises electronic and
hand-collected data
for the years 1995 through 2010, from the Morningstar
Datanalysis and Finanalysis
databases. Our empirical tests begin with the first full fiscal
year of mandatory
rotation in Australia, namely 2004, as all rotations are subject
to endogeneity concerns
in the non-mandatory period and the 2003 mandatory rotations are
few (see Table 2).
We exclude banks and insurance companies since the Altman (1968)
Z-score control
variable is not appropriate for them. Since we also need usable
data for the control
variables, the final sample is 1,619 financially-distressed
client years comprising 581
unique clients. Table 1 shows the sample derivation.
Table 1 about here
Descriptive Statistics
Table 2 shows descriptive statistics for 1996 to 2010. The
percentages of listed
clients audited by the Big4 has declined from about 65 in 1999
to about 44 in 2010,
consistent with client shedding. The number of audit firms and
offices has declined
-
16
over the period due to the Price Waterhouse and Coopers and
Lybrand merger in 1998
and the end of Arthur Andersen in 2002. Tenure has also declined
slightly. It is
evident that rotations did occur before 2003 despite there being
no requirement to
rotate, and they occurred at about the same rate after 2003. The
means of the within-
office non mandatory rates before and after 2003 are about 0.13
for example
(untabulated).12
Our experimental variables use the mandatory within-office
rotations
which are shown in the fourth column from the right. There is a
'spike' in mandatory
rotations in the 2007 year, and in Section VII we find that the
2007 year does not only
provide the significance for the experimental variables. As
noted, the 2007 year is the
first that the rotation requirement was reduced from seven years
to five. Clients who
switched between offices or audit firms and who have new
partners as a consequence,
are excluded from the measurement of the experimental variables
in order to measure
within-office effects of partner rotation and because clients
who switch offices may be
treated differently by the new office. Rotations do occur across
offices, but the
second column from the right shows that mandatory rotations of
this type are rare,
with only 14 occurring in the sample period.13
We include both of these types of
across office rotations as a control variable denoted as
OTHERR.
Table 2 about here
The Model
The dependent variable, GCREPORT, is coded as unity if the
client received a going
concern opinion in its audit report for the fiscal year and zero
otherwise. As noted,
we have four variables for each type of office-level rotation
rate, namely mandatory
12
These raw data come from Table 2. For example, the mean for the
1996 to 2002 years is 681 non
mandatory rotations divided by 5,321 Big4 client years, which is
about 13 percent. 13
Rotations can also occur across audit firms when both the client
and the partner move to a new audit
firm but this type of rotation is very rare.
-
17
rotation rates (MCLI, MUCLI, MPAR and MUPAR) and non-mandatory
rotation
rates (NMCLI, NMUCLI, NMPAR and NMUPAR). All of these variables
are
positively skewed, often have zero values and are usually less
than unity. In the
regressions we measure each variable as the natural logarithm of
the rate plus unity to
improve their symmetry and to preserve their rankings.
With respect to the controls, we include three client-level
rotation variables
corresponding to the data in Table 2. MR is equal to unity if
the client had a within-
office mandatory rotation in a fiscal year and zero otherwise.
The corresponding non
mandatory rotation is denoted NMR. The data for these two
variables are shown in
the third and fourth columns from the right of Table 2. OTHERR
is the sum of the
first and second columns from the right of Table 2, and it
equals unity of the client
had a rotation that was not within the office in a fiscal year
and zero otherwise. As
noted in Section II we have no expectation for these
coefficients' signs. TENUREP is
the number of consecutive years that the incumbent audit partner
has signed the audit
report for the client and it is capped at 10 years as our
dataset begins in 1995. We
expect the TENUREP coefficient to be negative from Carey and
Simnett (2006) and
Ye et al. (2011). TENUREF is the number of consecutive years
that the incumbent
audit firm has audited the client and it is also capped at 10
years. Prior research is
inconclusive on this relation, (Levinthal and Fichman 1988;
Vanstraelen 2000; Geiger
and Raghunandan 2002; Knechel and Vanstraelen 2007), so we have
no expectation
for its coefficient. Francis and Krishnan (1999) report that
clients with high accruals
are more likely to receive a going concern opinion because
auditors act more
conservatively with those clients. We include an indicator equal
to unity if the client's
scaled accruals (net income less net operating cashflows divided
by lagged total
-
18
assets), is larger than the sample median and zero otherwise
(HIGHTA). Its
coefficient should be positive for consistency with Francis and
Krishnan (1999). The
office client-to-partner ratio is positively correlated with the
office level mandatory
rotation rates (see Table 4), so we include the natural
logarithm of the ratio of clients
to partners as a control for office busyness (CPRATIO).
Analogous to studies of
board busyness (Fich and Shivdasani, 2006), we expect its
coefficient to be negative.
Kallapur et al. (2010) report that earnings quality is lower
when audit market
competition is higher, suggesting that the relation between
competition and going
concern opinion issuance could be negative. We include the
Herfindahl index
(HERFIND) as a proxy for audit market competition and we expect
its coefficient to
be positive for consistency with Kallapur et al. (2010), as
higher numbers indicate
lower competition in the present study. The size of the audit
office is positively
related to going concern opinion issuance for Big4 firms
(Francis and Yu, 2009). So
we include OFFICESIZE, measured as the natural logarithm of the
sum of audit fees
paid to the office of the auditor which audits the client by all
clients of that office, and
expect the coefficient to be positive. Client influence at the
partner level, denoted
INFCLIPAR, is included because Chen et al. (2009) report this
variable is important
in explaining audit qualification likelihood in their sample of
Chinese listed
companies. It is measured as the ratio of a client's total fees
(audit fees plus non audit
fees) relative to aggregate annual fees paid to the practice
office for audits signed off
by the partner who audits the client. INCLIOFF is the ratio of a
client’s total fees
(audit fees plus non audit fees) relative to aggregate annual
fees paid to the practice
office which audits the client. We expect its coefficient to be
positive, relying on
similar reasoning from Reynolds and Francis (2000). NATLEADER is
an indicator
variable that equals unity if an auditor is the number one
auditor in an industry in
-
19
terms of aggregated audit fees in a specific fiscal year, and
zero otherwise, and
CITLEADER is an indicator variable that equals unity if an
office is the number one
auditor in terms of aggregated client audit fees in an industry
within that fiscal year.
The evidence is mixed on these two variables with Francis and Yu
(2009) reporting
no relation and Reichelt and Wang (2009) reporting a negative
relation. We have no
expectation for either of these coefficients. The remaining
controls are more common
and we do not justify with prior research for brevity. Their
expected signs are
consistent with Francis and Yu (2009) and those expectations are
shown in Table 5.
SIZE is the natural logarithm of a client’s total assets. CASH
is the sum of a client’s
total cash and investments divided by total assets. PRIORGC is
an indicator variable
that equals unity if the client received a going concern opinion
in its audit report for
the previous fiscal year and zero otherwise. REPORTLAG is the
natural logarithm of
the number of days between a client’s fiscal year-end and its
earnings announcement
date. DEBT is total liabilities divided by total assets at the
end of the fiscal year.
LAGLOSS is an indicator variable that equals unity if earnings
before interest, tax
and depreciation is negative in the prior fiscal year and zero
otherwise. ALTMAN is
the Altman (1968) Z-score. RETURN is the annual raw stock return
measured to the
end of the fiscal year. VOLATILITY is the standard deviation of
monthly raw stock
returns measured to the end of the fiscal year. MB is the
natural logarithm of the
client’s market value of equity to its book value of equity at
fiscal year end. Some of
these variables are winsorized and we provide this detail in
Table 3 for brevity. The
HERFIND, OFFSIZE, INFCLIOFF, INFCLIPAR, NLEADER and CLEADER
variables are measured using all listed Australian clients from
2004 through 2010.
We use the following logit regression models for testing
hypothesis 1:
-
20
)1)(///()1( 10 εαα +++== CONTROLSMUCLIMCLIMUPARMPARfGCLogit
Model (1) is estimated using robust standard errors clustered by
client and fiscal year
as outlined in Gow et al. (2010), and year and industry fixed
effects using GICS
Industry Groups. Results for the industry and year indicator
variables are not reported
for brevity.
Table 3 shows the descriptive statistics for the regression
sample. It is evident that
even in a mandatory regime, some partners have not complied with
the rules as the
maximum tenure is 10 years.
Table 3 about here
Table 4 shows the Pearson and Spearman correlation coefficients
between the
regression variables. The experimental variables are generally
negatively correlated
with GC_REPORT and the MPAR and MUPAR correlations are stronger.
As
expected the correlations between the experimental variables are
generally high, but
they range from about 0.55 to about 0.85, for reasons like those
mentioned above and
illustrated in Appendix 1.
Table 4 about here
IV. RESULTS AND DISCUSSION
Table 5 shows results from estimating model (1) four times, once
for each of the
experimental variables. A financially-distressed client is
defined as a client that
reports negative net income and negative net operating cashflows
in the same fiscal
year. The experimental variables, shown in the top row of each
column in the
-
21
OFFICE LEVEL VARIABLES section, are all negative and significant
at .10 and we
cannot reject Ho1. These results indicate a lower propensity to
issue a going concern
opinion in offices with higher rates of mandatory rotation. The
significance of these
coefficients declines as one moves from rates that use clients
to those that user
partners as the base, probably because using the number of
clients as the base better
measures rotations' office-level effects on auditor behavior.
Significance also falls as
one moves within the two main rotation types from total
rotations to unique partner
rotations (for example from MCLI to MUCLI). This is expected
because, as noted
above, unique partner rotation rates are sometimes
unrepresentative of office-level
effects.
The client level rotation variables are not significant
indicating that the 'fresh eyes'
argument is not supported. One exception is OTHERR in equation
(2) but its
importance is subsumed by other variables in the other models.
Tenure is not
significant which is inconsistent with Carey and Simnett (2006)
and with Ye et al.
(2011). We revisit this relation in section VII as noted. We
find no evidence that
audit firm tenure is important consistent with our expectation.
HIGHTA is positive
and significant (at .05) consistent with Francis and Krishnan
(1999). We use this
variable in further tests below in Section V. Office busyness is
not important for
going concern opinion issuance as the results for CPRATIO show.
A caveat is that
there are other measures of office busyness (see Fich and
Shivdasani, 2006). The
HERFIND coefficient's results indicate that higher audit market
competition is
negatively related to going concern opinion issuance. Recall
that higher values
indicate weaker market competition here. As noted Kallapur et
al. (2010) find the
same directional relation with earnings quality. The OFFICESIZE
positive relation is
-
22
consistent with Francis and Yu (2009). Client influence at the
office or partner levels
are not significant. It is important to note here that the
result for INFCLIPAR is not
necessarily inconsistent with Chen et al. (2009) because they
examine the change in
sign and significance of that coefficient around an event. We
find no evidence that
auditor specialization at the City or at the National levels are
important, consistent
with Francis and Yu (2009). Of the remaining variables, SIZE,
CASH, PRIORGC,
DEBT, BANKRUPTCY, RETURN and to a lesser extent, MB are
significant and
consistent with expectations. REPORTLAG, LAGLOSS and VOLATILITY
are not
significant at .10 and only LAGLOSS has a sign inconsistent with
expectations. The
pseudo R-squared values are about 0.33 for the four models.
Table 4 about here
V. EXPLORING REASONS FOR THE RESULTS
In this section we rule out a competing explanation of knowledge
loss and then
attempt to differentiate between the office-level disruption and
lower auditor
conservatism explanations outlined above.
It can be argued that more rotations in an office results in
more client-specific
knowledge being lost across the office. Thus the observed
negative relations reported
in Table 5 could be due to knowledge loss and not due to office
disruption or lower
conservatism by audit partners. If knowledge loss is the reason,
we should observe a
more significant and negative coefficient for the rotation rates
for the sample of
mandatory rotations, compared to the rest of the sample. For
brevity, we do not report
results but for the mandatory rotation sample none of the
rotation rates' coefficients
are significant and all are negative and significant in the
other sample (at .10). This
-
23
suggests that audit quality could be lower for the non rotated
clients. Knowledge loss
does not explain the negative relations and we turn next to our
two possible
explanations.
We first examine the accuracy of going concern opinion issuance
because the test can
be used as a measure of auditor conservatism (Geiger and Rama
2006) and the
expected relation between office disruption and going concern
accuracy is obvious.
Specifically, the rates for Type I errors (going concern reports
issued to clients that do
not subsequently become bankrupt) will be higher if office-level
disruption is the
explanation for our results but these rates will be lower if
auditors behave less
conservatively. This is expected because clients 'at the margin'
of subsequent
bankruptcy are less likely to get a going concern report than
more financially-stressed
clients if the lower conservatism explanation holds. The rates
for Type II errors (not
issuing a going concern report to a client that subsequently
becomes bankrupt) should
be higher under either explanation. That is lower Type I coupled
with higher Type II
is evidence of lower auditor conservatism (Geiger and Rama
2006).
Univariate and multivariate logistic regression tests are used.
Bankrupt clients are
defined as those clients who entered voluntary administration,
liquidation or
receivership within twelve months of the audit report date
(bankrupt).14
From 2004 to
14
In another Australian study examining bankrupt clients, Jones
and Hensher (2004) include clients
that were suspended from quotation on the Australian Stock
Exchange because they did not pay their
annual fee, in their bankruptcy definition. We do not include
these clients because we believe our
measure more closely approximates that used in U.S. studies.
However, there are relatively few
instances of non-fee paying clients that were suspended, and
their exclusion did not alter Jones and
Hensher' (2004) inferences (see Jones and Hensher, 2004,
footnote 23). As the sample size is so small,
we include 2010 to increase the sample as much as possible, but
we only use clients with a fiscal year
end of June or earlier and measure the period to 9 months,
because at the time of writing, sufficient
time had not passed to measure other clients reliably. We do not
believe this poses any special
problems.
-
24
2010, there are 452 going concern opinions issued to distressed
clients of the Big4
and 30 distressed clients are bankrupt.
We split all mandatory office-level rotation rates at the median
for coding. Type I
error results, shown in Table 6, Panel A indicate that for 3 of
the 4 tests, offices with
high mandatory rotation rates could be more accurate with their
going concern
opinion issuance at the .05 level. For example, for MCLI, 6 (15)
of the 224 (228)
clients with going concern opinions in the audit report are
subsequently bankrupt in
the low (high) rotation rate offices, and the p-value is 0.049.
Unfortunately there are
only 30 distressed, bankrupt clients, precluding a meaningful
Chi-square test of Type
II errors so we report Fisher's exact test p-values. The
proportions of errors are higher
in high rotation rate offices but none of the differences are
significant at .10.
We can test Type I errors using logistic regression but not Type
II errors due to the
small sample size. We estimate a model where the dependent
variable is equal to
unity if the client is subsequently bankrupt after receiving a
going concern opinion
and include all controls from model (1). Initial estimation
proved unsuccessful due to
quasi complete separation, which is unsurprising given that we
have only 18 bankrupt
clients in this sample. The three year indicators causing the
problem were removed as
we are not aware of another solution. The results should be
interpreted with some
caution. Panel B of Table 6 shows the summary results.
These results also indicate that high rotation rate offices are
more accurate than low
rotation rate offices for MCLI and to a lesser extent MPAR but
MUCLI and MUPAR
are highly insignificant. This suggests that MCLI has more
accurate opinions at very
-
25
high levels of that rotation rate than do the other rates. To
reconcile with the chi-
square tests we split each rate at its sample median and the
results shown in Panel D
show that all are significant, consistent with the accurate
prediction observations for
MUCLI and MUPAR being more bunched around the median than the
other two
rates, despite all rates having more accurate observations above
than below the
median. These results are suggestive of lower auditor
conservatism, so we use
another test which does not have the concerns of small sample
size.
Our second test relies on the arguments in Francis and Krishan
(1999). If auditors
behave less conservatively in high rotation rate offices, their
threshold for issuing
going concern opinions should be higher, that is, these auditors
should be willing to
accept more inherent risk (Francis and Krishan 1999). If high
office rotation causes
disruption the auditor’s threshold for issue going concern
opinions should not be
higher. We include the joint effect of the accruals indicator
variable (HIGHTA) and
the office rotation rates. A negative coefficient for the joint
effect is expected if
conservatism is the explanation but we do not expect a negative
coefficient if
disruption is the explanation.
Panel A of Table 7 shows summary results for the coefficients.
Except for MULCI,
the coefficients for the joint effect are negative and
significant at .10, suggesting that
the relation between going concern issuance and high accruals
could be weaker in
high rotation rate offices. To be more confident, we estimate
the predicted
probabilities of going concern opinion issuance at quartiles of
office rotation rates, for
the significant joint effect coefficients. All other explanatory
variables are set to their
-
26
medians. The industry and year indicator variables, MR, NMR,
OTHERR,
NLEADER and CLEADER and PRIORGC all have zero medians .
Results in Panel B show that probabilities are higher for high
accrual clients (except at
the maximum values), but the difference between the
probabilities declines as the
rotation rate increases. For example at the 25th percentile of
MCLI the difference is
about 10 percentage points and it is about 4 percentage points
at the 75th percentile.
The other rotation rates show similar patterns. This means that
the likelihood of
receiving a going concern opinion for clients with low versus
high accruals becomes
closer as the rotation rate increases. Alternatively, the change
in the probability as
one moves to higher percentiles is much larger (about three
times as large from the
25th to the 75th for MCLI), for high rotation rate offices. This
evidence is consistent
with lower auditor conservatism in higher rotation rate offices
and supports the
accuracy results above. We conclude that the lower likelihood of
going concern
issuance is caused by lower auditor conservatism and not by
office-level disruption.
VI. IS THE LOWER CONSERVATISM AN EQUILIBRIUM CONDITION?15
The year 2007 is an outlier with 127 mandatory rotations (see
Table 2). To examine
the influence of that year on the results and the persistence of
lower conservatism, we
estimate model (1) and model (1) including the accruals joint
effect with rotation rates
for the years 2004 to 2006, 2007 and 2008 to 2010. The results
for the rotation rates
from model 1, shown on the top row of each Panel in Table 8,
reveal that office
rotation rates are all negative and significant at (.05) for the
pre-2007 years sample,
insignificant for the 2007 year and for the 2008 to 2010 years
sample. Thus the 2007
15 We are grateful to the reviewer for pointing this out.
-
27
year is not causing the main results in Table 5 and the negative
relation does not
persist after about 3 years after the introduction of mandatory
rotation. In the bottom
row of each panel the interaction coefficients show that high
rotation rate offices are
less likely to issue a going concern opinion for the 2004 to
2006 years sample, and the
2007 and the 2008 to 2010 years samples, all the coefficients
for the joint effects are
insignificant, indicating that the lower conservatism in high
rotation rate offices does
not persist either. This is consistent with an initial shock of
mandatory rotation
temporarily reducing auditor independence. Small samples rule
out meaningful going
concern opinion accuracy tests.
It is important to note that other measures of audit quality may
not give the same
results for later years. Auditors in high rotation rate offices
could be more willing to
tolerate higher accruals management for example in later years
after rotation because
accruals management is within GAAP and a less risky choice for
auditors than is
lowering their independence. As DeFond (2010) suggests,
triangulation of audit
quality measures need not always be expected.
VII. RECONCILIATION WITH OTHER STUDIES
In this section we attempt to reconcile our findings for partner
tenure with findings
from two studies reporting results that conflict with ours. In
reporting a negative
relation between going concern issuance and tenure, Carey and
Simnett (2006) and
Ye et al.s' (2011) results could support a rotation policy due
to its client-specific
effect. Our results do not support a rotation policy due to its
client-specific effect.16
16
Our results do not support a partner rotation policy in the
initial years after introduction of the
rotation rules. Whether the reported lower conservatism behavior
persists is another matter. Auditors
in high rotation rate offices could become more conservative if
they perceive that their clients are less
likely to switch to other audit firms for example.
-
28
Carey and Simnett (2006) estimate a cross sectional logit
regression using data from
1995, and Ye et al. (2011) estimate the same type of model using
data from 2002.
Their estimations do not control for correlation in the
regression residuals across
clients. Standard errors will be inefficient in the presence of
across client residual
correlation (Petersen 2009 and Gow et al. 2010). Nor do their
models include some
important explanatory variables so their estimates could be
inefficient and biased.
Lastly, our results in Table 6, Panel C, indicating that long
tenured partners are more
accurate is not consistent with lower audit quality as partner
tenure increases.
We can only measure tenure from 1995, so we use the 2002 year,
recode the tenure
variable to cap it at eight years, and estimate Ye et al.'s
(2011) model as closely as
possible. We use the Ye et al. (2011) model since they also use
a continuous measure
for tenure (their two models are quite similar).17
One caveat is we do not have access
to their ALUMNI variable and we do not expect its omission to
pose any special
problems. Results of the replication, shown in the second column
from the left in
Table 9, reveal that tenure is negative and significant only at
.09 and Ye et al. (2011)
report a p-value of about .05 (far left column). Our sample size
is 15 fewer (611
versus 626), possibly because we also need legitimate values for
other controls for our
tests here, so this could explain this difference. Nevertheless,
it is the change in
tenure's significance that matters in the present analysis.
Estimating the model with
robust standard errors clustered by client and including all
significant omitted
variables results in a more significant and negative
coefficient, similar to Ye et al.
(2011) as the middle column shows. To arrive at this model
specification we initially
17
Carey and Simnett (2006) use short and long tenured indicator
variables (less than 3 and longer than
7 years respectively) and when we use their model in the
subsequent analysis in the present section the
results are qualitatively the same.
-
29
add all our extra variables (except the duplicated variables)
and then progressively
delete the most insignificant variable after each estimation,
unless the variable is well
supported by theory or prior research (SIZE remains despite its
insignificance for
example). After improving the model specification these results
could still be used to
support rotation due to a client specific effect, so we conduct
further tests.
We observe that it is the 56 clients with tenure longer than
seven years that result in
the tenure coefficient's importance, because when we remove
those clients its
coefficient is highly insignificant (p-values > 0.50),
whether we use Ye et al.'s (2011)
model or our extended version (untabulated). And we observe
client and partner
specific differences between these 56 clients and the rest of
the sample. These 56
clients are smaller and have lower market-to-book ratios
suggesting that their
governance mechanisms could be weaker. These partners have done
more audits in
recent times, suggesting that they could have more expertise and
experience, that they
could be overworked or that they could be a dominant partner in
the office for
example.18
We add a board size variable, because larger boards can be
associated
with better monitoring (Klein, 2002). We only have data to 2007
for board size. The
experimental variable (BOARDSIZE) is the natural logarithm of
the number of
directors on the board. A variable proxying for the
abovementioned partner attributes
has not been used before, so for each partner for each fiscal
year, we measure the total
of the number of audits he did in the five fiscal years prior to
the current year using a
rolling window, from 2002 to 2007. Five years is chosen only
because it equals the
maximum number of consecutive years for a partner to audit a
client under current
18 For the short and long (more than 7 years) tenured samples
respectively, the means of some variables
are, Number of directors on the board: 4.34 and 3.98, Market to
book ratio: 3.22 and 2.60, natural
logarithm of client size: 16.19 and 15.78 and the total of the
number of audits signed off on by the
partner in the five fiscal years before the current year: 29.66
and 39.71. All these differences are
significant at .05.
-
30
rules. The experimental variable (PWLOAD) is the natural
logarithm of the
aforementioned total plus unity. Consistent with our
expectations, BOARDSIZE is
negatively correlated with tenure (Pearson = -0.05) and PWLOAD
is positively
correlated with tenure (Pearson = 0.33), and they are both
negatively correlated with
the going concern opinion issuance dependent variable (Pearsons
= -0.12 and -0.08
respectively).19
When we add BOARDSIZE and PWLOAD to the model a different
picture emerges.
These extra two variables' coefficients are negative and
significant (at .05) as the
second column from the right shows, and tenure becomes
significant only at .24. We
repeated the analysis using the pooled sample from 2002 to 2007
with more dramatic
effect, and these results are shown in the far right column of
Table 9. Further,
untabulated results showed a negative, significant (at .10)
tenure coefficient using Ye
et al.'s (2011) model and using the model without the extra two
variables for this
pooled sample, and when we add these variables to the model
predicting Type I errors
(see Panel C, Table 6), tenure remains significant at the .05
level. Board size and the
total of a partner's recent number of audits subsumes much of
the information in
partner tenure when explaining going concern opinion issuance.
Examining the
reasons for the negative relations for BSIZE and PWLOAD is
beyond the scope of the
present paper.
Why do our results for tenure differ without the BOARDSIZE and
PWLOAD
variables? The answer lies in other omitted variables and the
Big4 sample that we use.
19 These reported Pearson correlation coefficients are for a
pooled sample from 2002 to 2007.
Spearman coefficients are similar. All coefficients are
significant at .01 except that the Spearman
coefficient between BOARDSIZE and TENUREP is significant at .10
in the pooled sample. The
coefficients for 2002 are almost the same except for the Pearson
and Spearman correlation coefficients
between PWLOAD and GC_OPINION, which are negative and
insignificant.
-
31
If we estimate Ye et al.'s (2011) model on the Big4 sample from
2002 to 2010, the
coefficient is -0.06 (p = .04), adding the significant omitted
variables (except for
BOARDSIZE and PWLOAD), industry indicators and using
two-dimension
clustering by client and year gives a negative coefficient with
a p-value of about 0.16
(untabulated).
We do not use the same samples as these two studies, and we
exclude ALUMNI from
our model due to data limitations, but the analysis in this
section suggests that model
misspecification in those studies could be the reason for the
differences in results. A
negative relation for tenure after controlling for board size
and partner history may be
found, but long tenured partners are more accurate with their
opinions as the results in
Table 7 show. A negative coefficient in explaining going concern
issuance does not,
of itself, indicate lower audit quality. It could be that fewer,
less-accurate opinions
are issued.
VIII. CONCLUSION
This study examines for the first time, the association between
mandatory rotation at
the office level and going concern opinion issuance. Using a
sample of financially-
distressed Australian clients from 2004 through 2010, audited by
the Big4 audit firms,
we make several contributions to the auditing literature.
We find that audit offices with high rotation rates are less
likely to issue a going
concern opinion. We find that the going concern audit opinions
are generally more
accurate in high rotation rate offices. Tests using total
accruals provide evidence
consistent with lower auditor conservatism in high rotation rate
offices. We also find
-
32
that these relations are only observed in the years immediately
after the introduction
of mandatory rotation. We argue that relatively high office
rotation creates an
incentive for auditors to behave less conservatively to reduce
the risk that clients will
switch to other audit firms. It seems that mandatory rotation's
introduction
temporarily exacerbated the risk of client loss and the
consequential change in auditor
conservatism is observable. We cannot test the switching
explanation due to paper
constraints but this is an interesting area for future research.
We find that the
propensity to issue a going concern opinion is not associated
with rotations at the
client level, nor is it associated with tenure. A reconciliation
with prior studies shows
that board size and a proxy for the size of the partner's recent
workload render the
tenure variable unimportant in explaining going concern opinion
issuance.
Reinvestigation of prior studies might be warranted. Further
exploration of these two
extra variables' relations, including the underlying reasons for
them, is left to future
research. We suggest monitoring as a reason for the former and
partner expertise,
partner dominance in the office and overwork as reasons for the
latter. Although not
our focus, we provide some insight into audit market competition
and going concern
opinion issuance. As noted, Kallapur et al. (2010) report that
higher audit market
competition is associated with higher earnings management, and
we report that higher
audit market competition is associated with a lower propensity
to issue a going
concern opinion. These results appear consistent but we did not
examine the accuracy
of those opinions for paper scope reasons. We leave this to
future research.
There are limitations to note. Although endogeneity concerns are
probably weak for
mandatory rotations we cannot be as confident about
non-mandatory rotations. If
endogeneity is present then inferences could change. Results for
the four office
-
33
rotation rate variables are not always consistent. Nevertheless,
we do not observe any
case where they predict opposite relations. We believe that MCLI
is the best of the
four to proxy for the office-level effects of rotation.
-
34
TABLE 1
DERIVATION OF THE SAMPLE - 2004 - 2010
Client
years
Unique
clients
ASX-listed clients audited by Australian auditors 12,157
2,294
Sample after excluding non Big4 clients 5,789 1,297
Sample after excluding non-distressed clients 2,028 706
Sample after excluding clients in banks and insurance industries
2,018 701
Final sample after excluding clients with unusable control
variables 1,619 581
A client with negative net income and negative net operating
cash flow in the same fiscal year is defined as a distressed
client.
-
35
TABLE 2 - DESCRIPTIVE STATISTICS FOR ALL BIG4 CLIENTS 1996 -
2010
Year Clients Audit
Firms
Audit
Offices Partner Tenure
Within Office
Rotations
Within Audit Firm,
Across Office Rotations
N % of
total Mean Median
Mandatory
Non-
mandatory
Mandatory
Non-
mandatory
1996 687 65 6 42 - - 0 92 0 19
1997 701 65 6 39 - - 0 76 0 12
1998 734 65 5 33 - - 0 61 0 9
1999 758 65 5 33 - - 0 95 0 15
2000 800 63 5 35 - - 0 96 0 18
2001 828 63 5 35 - - 0 143 0 13
2002 816 62 4 30 - - 0 118 0 15
2003 796 59 4 30 - - 5 156 0 14
2004 808 55 4 32 2.81 2.00 29 85 3 16
2005 821 52 4 31 2.89 2.00 16 98 1 14
2006 863 51 4 30 2.77 2.00 30 114 2 9
2007 860 47 4 30 2.15 2.00 127 111 3 11
2008 836 45 4 29 2.26 2.00 54 114 1 15
2009 788 43 4 30 2.29 2.00 52 147 1 12
2010 813 44 4 30 2.58 2.00 39 78 3 5
Totals
2004-2010 5,789 48 - - 2.54 2.00 347 747 14 82
All 11,909 55 - - 2.54 2.00 352 1,584 14 197 Partner tenure is
probably biased below the 'true' partner tenure more in earlier
years, as we begin tenure's measurement in 1995 and it is capped at
10 years. We
say probably because we use only clients that are listed, as do
all other studies that we have read, meaning the 'true' partner
tenure is unknown when only listed
client data are used. A within-office rotation is recorded when
the audit firm and audit office which audits the client are the
same and the audit partner who audits
the client is different over two consecutive fiscal years. A
within audit firm, across office rotation is recorded when the
audit firm which audits the client is the
same and the audit office which audits the client and the audit
partner who audits the client are different over those two
consecutive fiscal years. The year 1995 is
'lost' because we need to measure partner changes for rotations.
With regard to the Price Waterhouse and Coopers and Lybrand merger
in 1998 and the Arthur
Andersen and Ernst and Young merger in 2002, the new merged
audit firms are treated as different audit firms to the pre-merged
firms.
-
36
TABLE 3 - DESCRIPTIVE STATISTICS FOR THE SAMPLE 2004 - 2010
Variables N Min Median Mean Max StdDev
GCREPORT 1,619 0 0 0.248 1.000 0.043
MCLI 1,619 0 0.047 0.057 0.470 0.060
NMCLI 1,619 0 0.104 0.119 0.442 0.082
MUCLI 1,619 0 0.037 0.044 0.470 0.041
NMUCLI 1,619 0 0.076 0.078 0.405 0.046
MPAR 1,619 0 0.134 0.214 1.124 0.234
NMPAR 1,619 0 0.318 0.413 1.386 0.316
MUPAR 1,619 0 0.115 0.165 0.693 0.139
NMUPAR 1,619 0 0.248 0.276 0.693 0.158
MR 1,619 0 0 0.051 1.000 0.219
NMR 1,619 0 0 0.150 1.000 0.357
OTHERR 1,619 0 0 0.024 1.000 0.153
TENUREP 1,619 1.000 2.000 2.566 10.000 1.545
TENUREF 1,619 1.000 5.000 5.503 10.000 3.073
HIGHTA 1,619 0 0 0.500 1.000 0.500
CPRATIO 1,619 0 1.153 1.390 2.386 0.592
HERFIND 1,619 0.159 0.233 0.251 1.000 0.104
OFFSIZE 1,619 9.259 16.564 16.470 18.659 1.267
INFCLIOFF 1,619 0.000 0.004 0.018 1.000 0.084
INFCLIPAR 1,619 0.001 0.088 0.207 1.000 0.280
NLEADER 1,619 0 0 0.276 1.000 0.447
CLEADER 1,619 0 0 0.366 1.000 0.482
SIZE 1,619 9.105 16.589 16.674 23.952 1.631
CASH 1,619 -0.002 0.248 0.346 3.977 0.310
PRIORGC 1,619 0 0 0.192 1.000 0.394
REPORTLAG 1,619 2.773 4.317 4.303 6.482 0.268
DEBT 1,619 0.001 0.137 0.305 2.908 0.432
LAGLOSS 1,619 0 1.000 0.899 1.000 0.301
ALTMAN 1,619 -67.602 0.879 1.976 82.606 15.302
RETURN 1,619 -0.899 -0.172 0.120 4.500 1.006
VOLATILITY 1,619 0.029 0.199 0.224 0.801 0.124
MB 1,619 -6.700 2.010 3.297 23.920 4.672
Variable Definitions:
Dependent Variable:
GCREPORT = indicator variable that equals unity if a client
receives a going-concern opinion
in its audit report in a fiscal year, and zero otherwise;
Experimental Variables:
MCLI = natural logarithm of the number of mandatory partner
rotations that occurred in an
office in a fiscal year divided by the number of clients in the
office plus 1;
MUCLI = natural logarithm of the number of unique partners
involved in mandatory rotations
that occurred in an office in a fiscal year divided by the
number of clients in the office plus 1;
MPAR = natural logarithm of the number of mandatory partner
rotations that occurred in an
office in a fiscal year divided by the number of partners in the
office plus 1;
-
37
MUPAR = natural logarithm of the number of unique partners
involved in mandatory
rotations that occurred in an office in a fiscal year divided by
the number of partners in the
office plus 1;
Control Variables (Office-level Partner Rotation Variables):
NMPAR = natural logarithm of the number of non mandatory partner
rotations that occurred
in an office in a fiscal year divided by the number of partners
in the office plus 1;
NMUPAR = natural logarithm of the number of unique partners
involved in non mandatory
rotations that occurred in an office in a fiscal year divided by
the number of partners in the
office plus 1;
NMCLI = natural logarithm of the number of non mandatory partner
rotations that occurred in
an office in a fiscal year divided by the number of clients in
the office plus 1;
NMUCLI = natural logarithm of the number of unique partners
involved in non mandatory
rotations that occurred in an office in a fiscal year divided by
the number of clients in the
office plus 1;
Control Variables (Client-level Partner Rotation Variables):
MR = indicator variable that equals unity if the client has a
within-office mandatory rotation
of an audit partner and zero otherwise;
NMR = indicator variable that equals unity if the client has a
within-office non mandatory
rotation of an audit partner and zero otherwise;
OTHERR = indicator variable that equals unity if the client has
an across-office or an across-
audit firm partner rotation and zero otherwise;
Other Control Variables:
TENUREP = number of consecutive years that the partner's name
appears at the bottom of
the audit report for the client capped at 10 years;
TENURF = number of consecutive years that the audit firm audits
the client capped at 10
years;
HIGHTA = indicator equal to unity if net income less net
operating cashflows divided by
lagged total assets is greater than the sample median and zero
otherwise;
CPRATIO = natural logarithm of the ratio of the number of
clients in the office to the
number of partners in the office;
HERFINDAHL = sum of the squared market shares (where market
share is measured using
audit fees paid to the incumbent audit office) of all audit
firms in the city divided by the
number of audit firms in that city;
OFFSIZE = natural logarithm of the sum of audit fees paid to the
office of the auditor which
audits the client, by all clients of that office;
INFCLIOFF = ratio of a specific client’s total fees (audit fees
plus non audit fees) relative to
aggregate annual fees generated by the practice office which
audits the client;
INFCLIPAR = ratio of a specific client’s total fees (audit fees
plus non audit fees) relative to
aggregate annual fees generated by the partner in that
office;
NLEADER = indicator variable that equals unity if an auditor is
the number one auditor in an
industry in terms of aggregated audit fees in a fiscal year, and
zero otherwise;
CLEADER = indicator variable that equals unity if an office is
the number one auditor in
terms of aggregated client audit fees in an industry within that
city in a fiscal year, and zero
otherwise;
SIZE = natural logarithm of a client’s total assets;
CASH = sum of a client’s total cash and investments divided by
total assets;
PRIORGC = indicator variable that equals unity if a client
received a going-concern opinion
in its prior fiscal year, and zero otherwise;
REPORTLAG = natural logarithm of the number of days between a
client’s fiscal year-end
and its earnings announcement date;
DEBT = client’s total liabilities deflated by total assets,
winsorized at the 1st and 99
th
percentiles;
LAGLOSS = indicator variable that equals unity if operating
income after depreciation in
previous fiscal year is negative, and zero otherwise;
ALTMAN = Altman (1968) Z-score winsorized at the 1st and 99
th percentiles;
-
38
RETURN = client's 12-month raw stock return for the fiscal year,
winsorized at the 1st and
99th percentiles;
VOLATILITY = standard deviation of the client's 12 monthly stock
returns for the fiscal year,
winsorized at the 1st and 99th percentiles;
MB = client’s market value of equity to its book value of
equity, winsorized at the 1st and
99th percentiles;
-
39
TABLE 4 -
PEARSON AND SPEARMAN CORRELATION COEFFICIENTS, (PEARSON ABOVE
THE DIAGONAL)
VARIABLES
GC
RE
PO
RT
MC
LI
NM
CL
I
MU
CL
I
NM
UC
LI
MP
AR
NM
PA
R
MU
PA
R
NM
UP
AR
MR
NM
R
OT
HE
RR
GCREPORT
-0.040* 0.025 -0.004 0.056* -0.078* -0.049* -0.062* -0.063*
-0.029 0.010 0.040*
MCLI -0.001
-0.085* 0.862* -0.001 0.830* -0.137* 0.771* -0.091* 0.243*
-0.019 -0.005
NMCLI 0.035 0.045*
0.019* 0.734* -0.113* 0.778* -0.022* 0.653* -0.010 0.358*
0.003
MUCLI 0.009 0.965* 0.122*
0.118* 0.548* -0.106* 0.711* -0.074* 0.196* 0.008 -0.008
NMUCLI 0.054* 0.135* 0.816* 0.199*
-0.197* 0.298* -0.146* 0.510* 0.011 0.234* 0.036
MPAR -0.048* 0.890* -0.055* 0.831* -0.092*
0.072* 0.884* 0.083* 0.202* -0.010 -0.010
NMPAR -0.058* -0.093* 0.739* -0.066* 0.415* 0.095*
0.186* 0.835* -0.013 0.294* -0.027
MUPAR -0.050* 0.853* -0.033 0.828* -0.102* 0.984* 0.137*
0.174* 0.184* 0.025 -0.017
NMUPAR -0.054* 0.015 0.704* 0.028 0.582* 0.155* 0.902*
0.180*
0.001 0.230* -0.015
MR -0.029 0.221* -0.003 0.195* 0.020 0.195* -0.033 0.178*
0.001
-0.097* -0.036
NMR 0.010 0.012 0.315* 0.028 0.213* 0.007* 0.255* 0.013 0.235*
-0.097*
-0.066*
OTHERR 0.040 -0.012 0.001 -0.005 0.017 -0.021 -0.034 -0.020
-0.021 -0.036 -0.066*
TENUREP -0.059* -0.114* -0.216* -0.125* -0.190* -0.087* -0.134*
-0.088* -0.159* -0.285* -0.518* -0.194*
TENUREF -0.047* 0.026 0.023 0.013 0.011 0.010 -0.008 -0.004
-0.028 0.161* 0.071 0.039
HIGHACC 0.179* -0.001 -0.032 -0.008 -0.041* 0.027 0.015 0.032
0.007 -0.045* -0.019 0.085*
CPRATIO -0.138* -0.072* -0.182* -0.150* -0.471* 0.322* 0.462*
0.350* 0.355* -0.018 0.001 -0.031
-
40
TABLE 4 - CONTINUED
SPEARMAN CORRELATION COEFFICIENTS
VARIABLES
GC
RE
PO
RT
MC
LI
NM
CL
I
MU
CL
I
NM
UC
LI
MP
AR
NM
PA
R
MU
PA
R
NM
UP
AR
MR
NM
R
OT
HE
RR
HERFIND 0.074* -0.176* -0.017 -0.098* 0.197* -0.376* -0.330*
-0.370* -0.289* -0.028* -0.073* 0.053*
OFFSIZE 0.123* 0.284* 0.087* 0.306* 0.257* 0.070* -0.234* 0.037
-0.182* 0.047* -0.037 0.007
INFCLIOFF -0.008 -0.144* 0.015 -0.154* -0.067* -0.065* 0.124*
-0.053* 0.102* -0.006 0.055* -0.008
INFCLIPAR 0.137* 0.063* 0.118* 0.108* 0.253* -0.122* -0.196*
-0.131* -0.136* 0.077* 0.122* 0.033
NLEADER -0.014 0.030 -0.078* 0.044* -0.056* 0.031 -0.075* 0.025
-0.093* 0.002 -0.020 -0.034
CLEADER 0.010 0.107* -0.073* 0.098* -0.103* 0.152* -0.037 0.145*
-0.022 0.006 0.011 0.014
SIZE -0.165* 0.067* 0.097* 0.076* 0.133* 0.001 -0.016 -0.009
0.001 0.039 -0.018 -0.043*
CASH -0.245* 0.029 -0.083* 0.016 -0.083* 0.063* -0.026 0.059*
-0.017 0.015 0.016 -0.006
PRIORGC 0.488* -0.009 -0.025 -0.007 -0.004 -0.032 -0.081* -0.032
-0.068* 0.002 -0.029 0.077*
REPORTLAG 0.014 -0.049* -0.072* -0.072* -0.192* 0.115* 0.171*
0.130* 0.139* 0.008 0.012 0.038
DEBT 0.331* -0.001 0.050* 0.023 0.098* -0.080* -0.074* -0.078*
-0.077* 0.007 -0.015 0.014
LAGLOSS 0.003 -0.033 -0.025 -0.046* -0.070* 0.012 0.046* 0.014
0.028 -0.026 0.003 0.026
ALTMAN -0.386* -0.005 0.014 -0.022 -0.012 0.035 0.080* 0.030
0.086* -0.018 0.007 -0.032
RETURN -0.252* -0.043* -0.118* -0.060* -0.106* 0.013 -0.015
0.014 -0.017 0.038 -0.084* -0.038
VOLATILITY 0.093* 0.084* 0.035 0.082* -0.026 0.141* 0.092*
0.149* 0.075* -0.003 0.004 0.003
MB -0.065* 0.016 -0.088* 0.001 -0.106* 0.060* -0.005 0.062*
-0.021 0.024 -0.003 0.027
-
41
TABLE 4 - CONTINUED
PEARSON AND SPEARMAN CORRELATION COEFFICIENTS, (PEARSON ABOVE
THE DIAGONAL)
VARIABLES T
EN
UR
EP
TE
NU
RE
F
HIG
HT
A
CP
RA
TIO
HE
RF
IND
OF
FS
IZE
INF
CL
IOF
F
INF
CL
IPA
R
NL
EA
DE
R
CL
EA
DE
R
SIZ
E
CA
SH
GCREPORT -0.076* -0.046* 0.179* -0.129* 0.039 0.110* -0.009
0.085* -0.014 0.010 -0.156* -0.177*
MCLI -0.096* 0.032 0.001 -0.042* -0.047* 0.141* -0.056* 0.030
0.012 0.122* 0.054* 0.031
NMCLI -0.202* 0.029 -0.037 -0.059* -0.125* -0.001 -0.100* 0.044*
-0.101* -0.050* 0.063* -0.058*
MUCLI -0.108* 0.022 -0.022 -0.205* 0.049* 0.188* -0.053* 0.091*
0.049* 0.076* 0.069* 0.021
NMUCLI -0.177* 0.008 -0.044* -0.417* 0.025 0.163* -0.114* 0.211*
-0.061* -0.088* 0.128* -0.047*
MPAR -0.067* 0.014 0.037* 0.379* -0.246* 0.015* -0.082* -0.135*
-0.042* 0.181* -0.010 0.048*
NMPAR -0.116* 0.016 0.009* 0.516* -0.262* -0.190* -0.112*
-0.193* -0.106* -0.024 -0.050* -0.036
MUPAR -0.088* 0.004 0.025* 0.384* -0.240* 0.039 -0.102* -0.137*
0.002 0.172* -0.014 0.044*
NMUPAR -0.097* 0.000 0.011 0.464* -0.231* -0.148* -0.149*
-0.158* -0.126* -0.050* -0.030 -0.027
MR -0.234* 0.161* -0.045* -0.014 0.006 0.041* -0.004 0.031 0.002
0.006 0.036 0.006
NMR -0.426* 0.066* -0.019 0.008 -0.069* -0.025 -0.034 0.102*
-0.020 0.011 -0.009 0.020
OTHERR -0.159* 0.037 0.085* -0.047* 0.035 -0.010 0.033 0.066*
-0.034 0.014 -0.051* -0.015
TENUREP
0.227* -0.021 0.059* 0.036 0.001 0.020 -0.107* 0.046* -0.006
0.000 -0.011
TENUREF 0.227*
-0.041* -0.020 0.003 0.001 0.026 0.034 0.062* -0.037 -0.009
-0.021
HIGHACC -0.027 -0.045*
0.074* -0.055* 0.025 -0.028 0.003 -0.040 0.002 -0.158* 0.014
CPRATIO 0.059* -0.013 0.066*
-0.313* -0.293* -0.190* -0.430* -0.036 0.068* -0.161* 0.024
-
42
TABLE 4 - CONTINUED
PEARSON AND SPEARMAN CORRELATION COEFFICIENTS, (PEARSON ABOVE
THE DIAGONAL)
VARIABLES
TE
NU
RE
P
TE
NU
RE
F
HIG
HT
A
CP
RA
TIO
HE
RF
IND
OF
FS
IZE
INF
CL
IOF
F
INF
CL
IPA
R
NL
EA
DE
R
CL
EA
DE
R
SIZ
E
CA
SH
HERFIND 0.065 0.017 -0.074* -0.478* -0.282* 0.324* 0.148* 0.010
0.133* -0.011 -0.003
OFFSIZE -0.008 0.003 0.001 -0.458* 0.133* -0.465* 0.070* 0.083*
0.069* 0.143* 0.018
INFCLIOFF -0.028 0.008 0.001 0.156* -0.024 -0.684* 0.289* -0.019
0.126* 0.071* -0.028
INFCLIPAR -0.145* 0.020 -0.027 -0.477* 0.250* 0.148* 0.289*
-0.024 0.051* 0.250* -0.087*
NLEADER 0.042* 0.065* -0.040 -0.011 0.044* 0.081* -0.035 -0.005
0.216* 0.002 0.049*
CLEADER -0.005 -0.032 0.002 0.087* -0.021 0.110* -0.014 0.018
0.216* 0.055* 0.040*
SIZE -0.002 -0.009 -0.151* -0.161* 0.031 0.122* 0.321* 0.280*
-0.007 0.047* -0.327*
CASH -0.020 -0.013 0.024 0.079* -0.008 -0.007 -0.190* -0.117*
0.027 0.035 -0.331*
PRIORGC -0.044* 0.017 0.159* -0.087* 0.026 0.062* -0.021 0.064*
-0.003 0.004 -0.214* -0.136*
REPORTLAG -0.077* -0.072* 0.058* 0.343* -0.280* -0.242* 0.019
-0.269* -0.023 -0.048* -0.124* -0.071*
DEBT -0.004 0.022 0.160* -0.180* 0.147* 0.116* 0.229* 0.276*
0.001 0.076* 0.127* -0.308*
LAGLOSS -0.018 -0.048* 0.031 0.115* -0.055* -0.066* -0.112*
-0.139* 0.018 -0.023 -0.210* 0.139*
ALTMAN 0.015 -0.067* -0.245* 0.090* -0.080* -0.076* -0.073*
-0.148* 0.001 -0.054* 0.247* 0.026
RETURN 0.058* 0.024 -0.089* 0.123* 0.001 -0.081* -0.063* -0.135*
0.023 -0.001 0.041* 0.136*
VOLATILITY -0.094* -0.104* 0.152* 0.087* -0.125* -0.060* -0.065*
-0.090* 0.006 -0.038 -0.230* 0.082*
MB -0.011 -0.004 0.096* 0.130* -0.047* -0.065* -0.055* -0.109*
-0.025 -0.012 -0.189* 0.323*
-
43
TABLE 4 - CONTINUED PEARSON CORRELATION COEFFICIENTS
PR
IOR
GC
RE
PO
RT
LA
G
DE
BT
LA
GL
OS
S
AL
TM
AN
RE
TU
RN
VO
LA
TIL
ITY
GCREPORT 0.488* 0.032 0.338* 0.003 -0.284* -0.194* 0.080*
MCLI -0.016 -0.031 -0.005 -0.016 -0.010 0.089* 0.047*
NMCLI -0.032 -0.014 0.023 0.004 0.028 -0.100* 0.074*
MUCLI -0.011 -0.052* 0.007 -0.035 -0.009 0.035 0.064*
NMUCLI -0.010 -0.126* 0.073* -0.051* -0.018 -0.065* -0.005
MPAR -0.030 0.094* -0.070* 0.029 0.034 0.110* 0.091*
NMPAR -0.062* 0.156* -0.058* 0.063* 0.078* -0.028 0.117*
MUPAR -0.030* 0.121* -0.081* 0.028 0.042* 0.088* 0.131*
NMUPAR -0.064* 0.132* -0.052* 0.043* 0.073* 0.023 0.072*
MR 0.002 0.012 -0.010 -0.026 -0.012 0.021 0.005
NMR -0.029 0.016 -0.029 0.003 0.037 -0.061* 0.007
OTHERR 0.077* 0.050* 0.010 0.026 -0.041* -0.031 -0.015
TENUREP -0.061* -0.062* -0.016 -0.012 -0.004 0.034 -0.110*
TENUREF 0.018 -0.082* 0.001 -0.050* -0.072* -0.001 -0.086*
HIGHTA 0.159* 0.059* 0.167* 0.031 -0.195* -0.018* 0.149*
CPRATIO -0.076* 0.293* -0.132* 0.115* 0.090* 0.106* 0.082*
-
44
TABLE 4 - CONTINUED
PEARSON AND SPEARMAN CORRELATION COEFFICIENTS,
(PEARSON ABOVE THE DIAGONAL)
PR
IOR
GC
RE
PO
RT
LA
G
DE
BT
LA
GL
OS
S
AL
TM
AN
RE
TU
RN
VO
LA
TIL
ITY
HERFIND 0.008 -0.085* 0.101* -0.060* -0.064* -0.031* -0.087*
OFFSIZE 0.050* -0.198* 0.048* -0.036 -0.023 -0.073* -0.033
INFCLIOFF -0.004 -0.023 0.063* -0.129* -0.048* -0.041*
-0.048*
INFCLIPAR 0.036 -0.179* 0.191* -0.146* -0.073* -0.086*
-0.079*
NLEADER -0.003 -0.015 0.005 0.018 0.018 0.011 0.007
CLEADER 0.004 -0.018 0.075* -0.023 -0.031 -0.012 -0.009
SIZE -0.217 -0.157* -0.012 -0.240* 0.307* 0.017 -0.202*
CASH -0.077* -0.100* -0.185* 0.121* -0.082* 0.131* 0.056*
PRIORGC
0.059* 0.292* 0.095* -0.277* -0.063* 0.104*
REPORTLAG 0.027
0.032 0.107* 0.002 0.082* 0.133*
DEBT 0.240* -0.183*
-0.078* -0.506* -0.162* -0.037
LAGLOSS 0.095* 0.144* -0.121
-0.037 0.088* 0.058*
ALTMAN -0.318* 0.142* -0.788* -0.047*
0.122* 0.018
RETURN -0.106* 0.091* -0.173* 0.093* 0.172*
0.277*
VOLATILITY 0.102 0.204* -0.122* 0.064* 0.008 0.106*
MB -0.018 0.034 0.035 0.150* -0.202* 0.427* 0.065*
* = significant at .10
-
45
TABLE 5 - GOING CONCERN AUDIT OPINION TESTS
FOR DISTRESSED CLIENTS, 2004 - 2010 Variable (1) (2) (3) (4)
INTERCEPT
-5.501 -5.455 -5.437 -5.655
(0.100) (0.106) (0.104) (0.108)
OFFICE-LEVEL ROTATION VARIABLES
MCLI
-
-5.577
(0.040)
NMCLI
? 0.161
(0.858)
MUCLI
-
-6.547
(0.056)
NMUCLI
? -0.084
(0.965)
MPAR
-
-1.293
(0.066)
NMPAR
?
0.062
(0.751)
MUPAR
-
-2.006
(0.099)
NMUPAR
? -0.193
(0.775)
PARTNER-LEVEL ROTATI