Sie sind auf Seite 1von 57

Journal of Accounting and Economics

Forthcoming, 2009

Earnings Quality: Some Evidence on the Role of


Auditor Tenure and Auditors Industry Expertise
Ferdinand A. Gul*
School of Accounting and Finance
The Hong Kong Polytechnic University
Hung Hom, Kowloon, Hong Kong
Phone: (852) 2766-7771
Fax: (852) 2365-9303
Email: afgul@inet.polyu.edu.hk

Simon Yu Kit Fung


School of Accounting and Finance
The Hong Kong Polytechnic University
Hung Hom, Kowloon, Hong Kong
Phone: (852) 2766-4246
Fax: (852) 2330-9845
Email: afsf@inet.polyu.edu.hk

Bikki Jaggi
School of Business
Rutgers University
Levin Building, Piscataway-NJ 08850
Tel : (732) 445-3539
Fax: (732) 445-3201
Email: Jaggi@rbsmail.rutgers.edu
&
The Hong Kong Polytechnic University
Hong Kong

______________________
* Corresponding author
Acknowledgement: We thank an anonymous reviewer, Jerry Zimmerman (the editor), Mark
Bliss, Shimin Chen, Peter Cheng, Richard Chung, Carol Dee, Jun Du, Kimberly Dunn, Jere
Francis, Stephen Gong, Ira Horowitz, Yuan Huang, Kam Wah Lai, Chung-kin Min, Al Nagy,
Bin Srinidhi, Nancy Su, Joanna Ho, Suresh Radhakrishnan, Cheong H. Yi, workshop
participants in The Hong Kong Polytechnic University, 2007 Journal of Contemporary
Accounting and Economics Symposium, 2007 AAA Auditing Midyear Conference and other
individuals for their helpful comments.

Earnings Quality: Some Evidence on the Role of


Auditor Tenure and Auditors Industry Expertise

ABSTRACT
Prior studies suggest that auditors with short tenure are associated with lower earnings quality
because of the lack of client-specific knowledge and/or low balling. In this study, we examine
whether industry specialization of auditors and low balling affect the association between
auditor tenure and earnings quality. We find that the association between shorter auditor
tenure and lower earnings quality is weaker for firms audited by industry specialists
compared to non-specialists. In addition, we do not find results consistent with the low
balling explanation.

Keywords: auditor tenure; auditor industry specialization; low balling; earnings quality
Data Availability: Data are available from sources identified in the paper.
JEL Classification Codes: M43, M49

Earnings Quality: Some Evidence on the Role of


Auditor Tenure and Auditors Industry Expertise

1. INTRODUCTION
Recent research shows that auditors with shorter tenure are associated with lower earnings
quality than auditors with longer tenure (e.g. Johnson et al. 2002; Myers et al. 2003; Ghosh
and Moon 2005). Auditor tenure is defined as the number of years an auditor is retained by
the firm (Myers et al. 2003). Three different explanations have been provided for this
relationship. The first is based on the argument that short-tenured auditors lack client-specific
knowledge that is necessary to conduct a high quality audit. In addition, the accounting
profession argues that short tenure may involve higher risk for audit failures, because
incoming auditors with insufficient client-specific knowledge will have to rely more heavily
on the estimates and representations made by client firms (e.g. PricewaterhouseCoopers 2002;
Gul et al. 2007). The second is based on low balling, whereby auditors charge lower audit
fees to obtain and retain new clients and then expect to recoup losses in later years of the
audit engagements (DeAngelo 1981). This argument suggests that auditors with short tenure
may be more lax in the early years of the auditor-client relationship so that they can retain the
job long enough to recoup the initial losses, resulting in lower quality audits and lower
quality earnings (Gul et al. 2007). The third is based on the argument that firms with higher
quality earnings are more likely to retain the incumbent (high quality) auditors, or high
quality auditors are more likely to drop risky clients that have lower quality earnings, who
will then move to lower quality auditors.

In this study, we first examine whether the positive relation between auditor tenure and
earnings quality, as documented in prior research, is weaker for firms audited by industry
specialists. Second, we evaluate whether the observed lower earnings quality in the early
3

years of the auditor-client relationship is consistent with the low balling argument. Our paper
is motivated by two primary considerations. First, both the financial press and policy makers
remain concerned about alleged earnings management in U.S. companies (e.g. Levitt 2007;
Karpoff et al. 2008). Related to these concerns is the fact that academics, practitioners, and
policy makers raise doubts about the quality of auditors with shorter tenure and its
implications for earnings quality (e.g. Arel et al. 2005; Dunham 2002; Geiger and
Raghunandan 2002; PricewaterhouseCoopers 2002). The importance of this research question
is further exemplified by the fact that high quality accounting information is central to the
efficient allocation of scarce capital resources in the market (e.g. Foster and Johnson 2001). It
would be useful for investors and other interested parties to be aware that situations may exist
in which the documented association between lower quality of earnings and shorter auditor
tenure is weaker. One such situation is the presence of an industry specialist auditor.1 Second,
while there is ample evidence to support the short tenure/low quality earnings relation, there
has been little attempt to examine whether this relation is related to low balling or lack of
client specific knowledge. This paper attempts to provide some understanding of this issue.

A maintained assumption in the linkage between auditors industry specialization and the
shorter auditor tenure/ lower earnings quality association is based on prior studies which
show that auditors industry specialization is associated with higher earnings quality (e.g.
Balsam et al. 2003; Krishnan 2003). Industry specialization is defined in terms of the
auditors market share (see Francis et al. 2005a). An auditor is considered a specialist in an
industry if the audit firm has the largest share of the industrys total assets (Mayhew and
Wilkins 2003). Based on this assumption, we expect that the association between lower

It should be noted that investigation of this type is a joint test of (1) the association between auditor tenure,
industry specialization and earnings quality, and (2) the adequacy of our empirical model in controlling for nonauditor sources of earnings quality.
4

earnings quality and shorter auditor tenure would be weaker if the auditors are also industry
specialists, ceteris paribus.

In our main tests, we use discretionary accruals as a proxy for earnings quality, which are
estimated based on a model suggested by Ball and Shivakumar (2006). In addition, as
sensitivity tests, we use other discretionary accruals models and the earnings benchmark tests.
Based on a sample of observations from 1993 through 2004, we show that earnings quality is
lower when auditor tenure is short. Additionally, we find that the positive association
between earnings quality and auditor tenure is significantly weaker for firms that are audited
by industry specialists. However, these results have to be viewed with some caution since the
data used in the study may be largely drawn from a presumably optimal matching between
auditors and clients i.e. the client retains the best auditor, and the auditor keeps the best
clients. We address this limitation by conducting tests of endogeneity.

In a separate set of tests, we examine whether the higher discretionary accruals in the earlier
years of auditor-client relationship (less than four years) can be explained by low balling.
Based on a sample of firms from 2000 through 2004, we identify low balling firms as the
firms with short auditor tenure (less than four years) and the auditor charges abnormally low
audit fees (when audit fees are one standard deviation below the predicted audit fees based on
an audit fees model). Results show that there is no significant association between low
balling and lower earnings quality.

To address the potential endogeneity problems in our main tests associated with hiring of
short-tenured versus long-tenured auditors, or specialists versus non-specialists (self-selection
bias), we use a two-stage least-squares (2SLS) estimation procedures. The results of these
tests are similar to our main findings. Our main test results are also robust to other
5

discretionary accruals models such as the accruals quality measure developed by Francis et al.
(2005b), the Jones discretionary accruals model (Jones 1991), or the performance-adjusted
model (e.g. Ashbaugh et al. 2003; Kothari et al. 2005). In addition, similar results are
obtained when we use meeting or beating the earnings benchmarks as an alternative measure
of earnings quality. Overall, our results suggest that industry specialization is likely to reduce
the association between shorter auditor tenure and lower earnings quality.

Our results contribute to the auditing literature in the following ways. First, our findings add
to the literature on the linkage between industry specialization and audit quality (Gramling
and Stone 2001; Balsam et al. 2003; Dunn and Mayhew 2004) by showing that industry
specialization has a role to play in the early years of auditor-client relationship. The results
also suggest that the evidence in prior studies linking short auditor tenure with poor earnings
quality (e.g. Myers et al. 2003; Johnson et al. 2002; Carcello and Nagy 2004) is likely to
apply more to firms audited by industry non-specialists. These results suggest that the
documented evidence linking short tenure to poor earnings quality may not be due to shorter
tenure per se, but rather it may be due to the auditors unfamiliarity with the clients business
that could affect the auditors ability to detect misrepresentations and/or misreporting. It
should, however, be noted that our results should not be interpreted to mean that all industry
specialists or the use of specialists services will always improve earnings quality. Second,
this study represents the first attempt to empirically distinguish the low balling argument
from the lack of client-specific knowledge argument.

The remainder of the paper is organized as follows. In Section 2, we discuss the background
and research questions for the study. Section 3 contains discussion on sample selection and
research design. Results are discussed in Section 4, and the conclusion is contained in
Section 5.
6

2. BACKGROUND AND RESEARCH QUESTIONS


2.1 Audit Quality and Earnings Quality
The assumption in this study that audit quality is positively linked to earnings quality is not
new and has been extensively documented in the accounting and auditing literature. Several
prior studies document an association between measures of higher quality auditors (such as
auditor size or industry expertise) and higher quality of financial reporting (e.g. Becker et al.
1998; Johnson et al. 2002; Krishnan 2003; Balsam et al. 2003; Myers et al. 2003; Ghosh and
Moon 2005). This linkage is based on the argument that high-quality auditors, as a result of
more effective monitoring, are more likely to detect questionable accounting practices and
misrepresentations by management than low-quality auditors. If managers are unwilling to
address the auditors concerns with regard to questionable accounting practices and
misrepresentations, high-quality auditors are more likely to issue qualified audit reports. In
this sense, the quality of financial reporting (earnings quality) may be viewed as a joint
product of managerial and auditor efforts.

2.2 Auditor Tenure and Earnings Quality


Recent studies in the auditing literature suggest that auditors with longer tenure are associated
with higher earnings quality (Geiger and Raghunandan, 2002; Gul et al. 2007). Johnson et al.
(2002), for example, document higher unexpected accruals when auditor tenure is short (two
to three years) than when it is medium (four to eight years). Moreover, they find no evidence
that a longer auditor-client relationship (i.e. nine years or more) is associated with lower
unexpected accruals compared to the medium auditor-client relationship. Myers et al. (2003)
find that a longer auditor-client relationship is associated with a lower dispersion in the
distributions of discretionary and current accruals, and there are greater constraints on both
income-increasing and income-decreasing discretionary accruals. Consistent with Myers et al.
7

(2003), Ghosh and Moon (2005) provide evidence that firms with longer auditor tenure are
associated with stronger earnings response coefficients, suggesting that investors perceive
earnings quality of firms with longer auditor tenure to be better than the earnings quality of
firms with shorter auditor tenure.

The above findings are consistent with learning theory in psychology (Glaser and Chi 1988;
Glaser and Bassok 1989; Lapre et al. 2000), which suggests that it takes time for auditors to
develop client-specific knowledge to perform an effective audit. For example, an auditor
conducting an audit for the same client over a number of years learns successively about
critical accounting issues that may require particular attention. However, there are other
alternative explanations. One interpretation of these results is that auditors in the early years
of the auditor-client relationship are less independent as a result of low balling. Another
interpretation is that clients may switch auditors or auditors may drop clients with poor
quality earnings. For example, it is possible that firms with higher earnings quality have a
tendency to retain the same auditor. This interpretation is consistent with the auditorswitching and opinion-shopping literature (e.g. DeFond and Subramanyam 1998; Krishnan
1994; Lennox 2000), which suggests that firms switch auditors after receiving a qualified
report (in some cases as a result of poor earnings quality). Alternatively, high quality
auditors may drop risky clients (in particular clients with large unexpected accruals or lower
earnings quality) in the first few years of their audit engagements.

2.3 Auditor Industry Specialization and Earnings Quality


In addition to auditor tenure, auditors industry specialization is considered to be another
proxy for audit quality. Several prior studies show that client firms with industry specialists
are associated with higher quality of financial reporting (e.g. Balsam et al. 2003; Krishnan
2003). These findings are consistent with the theory that auditors specialize in various
8

industries to achieve product differentiation and provide higher quality audits (Simunic and
Stein 1987; Dunn and Mayhew 2004). Higher quality of audits by industry specialists is also
attributed to the fact that they invest heavily in technologies, physical facilities, personnel,
and organizational control systems that enable them to detect irregularities and
misrepresentations more easily (Simunic and Stein 1987). Their ability to provide higher
quality audits comes from their experience in serving other clients in the same industry and
learning and sharing best practices across the industry (Maletta and Wright 1996, Dunn and
Mayhew 2004). Similarly, PricewaterhouseCoopers (2002) argue that audit quality depends
on numerous factors including an auditors knowledge and understanding of the company
being audited and the industry in which it operates. These arguments thus suggest that
auditors with industry expertise are more likely to detect misrepresentations and irregularities
than auditors without industry expertise, especially in the early years of the audit assignment.
The linkage is based on the assumption that industry specialist auditors have the industry
expertise that results in better understanding of the clients business (e.g. Kwon 1996). There
is, however, an alternative explanation for this association. It could be argued that firms with
higher earnings quality may hire industry specialists, and thus the observed positive
association between earnings quality and industry specialization may be due to self-selection
of specialist auditors. We address this issue in our sensitivity tests.

2.4. Auditor Tenure, Industry Specialization, and Earnings Quality


In this study, we integrate these two streams of research on auditor tenure and auditor
industry specialization to examine whether auditors industry specialization has an impact on
the association between auditor tenure and earnings quality. Though evidence suggests that
client-specific knowledge plays an important role in conducting an effective audit, it may be
argued that many audit-related issues are industry-specific and have unique industry features
e.g. forward sales contracts, off balance sheet financing arrangements, accounting systems,
9

tax rules or specialized reporting requirements in certain industries. Consequently, industry


expertise is also likely to play a role in improving audit quality in terms of auditor
competence, in addition to the general knowledge base required for all audits (e.g. Shockley
and Holt 1983; Craswell et al. 1995).

The linkage between auditor tenure, specialization and earnings quality is thus based on the
assumption that auditors ability to perform an audit effectively and efficiently will depend
on their expertise in the clients industry as well as on their client-specific knowledge.
Auditors expertise in the clients industry is therefore likely to be beneficial in a new
auditor-client relationship when the auditor lacks client-specific knowledge.

2.5 Low Balling


Since low balling has been identified as one of the factors that could be linked to the lower
quality of earnings for auditors in the first few years of an audit engagement (see, for
example, DeAngelo 1981), we also conduct tests to assess if our results are consistent with
predictions of the low balling argument. We construct a sample of low balling firms and
evaluate their association with earnings quality. The results of this evaluation will indicate
whether low balling is likely to contribute to the linkage between shorter auditor tenure and
lower earnings quality, ceteris paribus.

3. RESEARCH DESIGN
3.1 Sample
This study is based on a sample of firms from 1993 through 2004. All firms with sufficient
data on the Compustat annual industrial and research files for estimation of accruals are

10

included in the initial sample.2 We include only firms that are audited by Big 6/5/4 auditors3
(hereafter referred to as Big N auditors) because the focus of our research is on industry
specialization of auditors which is generally identified within the big accounting firms (e.g.
Francis et al. 2005a; Johnson et al. 2002).4 Firms with mergers and acquisitions (M&A) are
dropped from the sample because accruals in firms with mergers and acquisitions tend to be
larger for reasons unrelated to earnings management (Ashbaugh et al. 2003). 5 Our final
sample consists of 32,777 firm-year observations.

3.2 Calculation of Discretionary Accruals


Despite the extensive use of discretionary accruals as a measure of earnings quality (e.g.
Myers et al. 2003; Ashbaugh et al. 2003; Balsam et al. 2003; Johnson et al. 2002), there is
still little evidence documenting which discretionary accruals model is superior or more
appropriate. While we use different models of discretionary accruals in our sensitivity tests,
main tests are based on the discretionary accruals model suggested by Ball and Shivakumar
(2006). We modify this model slightly to include the components of growth and cash flows.
We use the following model for each of the two-digit SIC industry groups 6 to estimate
discretionary accruals, which are given by the residual term (t).

In an additional test, we follow Myers et al. (2003) in requiring at least six years of prior data to ensure that
any abnormal accruals behavior associated with start-up firms (Teoh et al. 1998a, 1998b) is not attributed to
short auditor tenure. In another sensitivity test, we also omit firms for which the auditor-client relationship
lasted for five years or less in our sample (e.g. Myers et al. 2003). These additional screening procedures reduce
the sample size, but our main results (untabulated) remain unchanged.
3
Big 6 auditors include Arthur Andersen, Coopers and Lybrand, Deloitte Touche Tohmatsu, Ernst and Young,
KPMG and Price Waterhouse.
4
Because auditor industry expertise is unobservable, prior studies rely on the auditors relative size in the
industry as a measure of industry specialization, based on the assumption that the auditors industry expertise
increases with their market shares and investment (e.g. Craswell et al. 1995; Hogan and Jeter 1999; Francis et al.
2005a). Since Big N auditors are substantially larger in size than other audit firms in almost all industries, using
industry size as a measure of auditor specialization will effectively render most non-Big N auditors nonspecialists, even if these non Big N auditors were indeed specialists in some industries. As a result, industry size
would be a better measure of an auditors industry expertise (e.g. Francis et al. 2005a) within big accounting
firms. We obtain similar results (untabulated) by including non-Big N firms in the sample in a sensitivity test.
5
In a sensitivity test we include M&A firms in our sample and add an indicator for M&A in our multivariate
analyses. Results (untabulated) show that the coefficient on M&A indicator is significantly positive, and our
main results still hold.
6
There should be at least 20 firms in each industry group for each year.
11

ACCt

= 1 + 2CFOt + 3CFOt-1 + 4CFOt+1 + 5 Revt + 6 PPEt + 7ROAt-1 +


8 CFOt + 9 DumCFOt + 10 CFOt *DumCFOt + t

(1)

where:
= Earnings before extraordinary items (Compustat annual data item 123)
cash flow from operations (Compustat annual data item 308);
= Cash flow from operations for periods t-1, t and t+1;
CFO
= Change in net sales revenue (Compustat annual data item 12); and
Rev
= Property, plant, and equipment - net (Compustat annual data item 8).
PPE
= One-period lag ROA
ROAt-1
DumCFO = Dummy variable, 1 when there is a negative change in the operating cash
flows, 0 otherwise.
ACC

All variables are scaled by average total assets, calculated as the average of beginning of year
and end of year total assets.

3.3 Auditor Tenure and Auditor Specialization


Auditor tenure is calculated based on the available data from Compustat, and is defined as the
number of years an auditor is retained by the firm. In the case of change in the audit firms
name as a result of audit firm mergers, the incumbent auditor-client relationship remains
unchanged.

A Big N audit firms industry specialization is measured based on its share of clients total
assets in the two-digit SIC industry group, and industry expertise is assumed when the audit
firms market share is the highest within the industry group (Hogan and Jeter 1999; Krishnan
2005). 7 Operationally, we rank Big N auditors based on their percentage of total assets
audited in the industry, and the audit firm capturing the largest market share (of total assets)
is identified as a specialist in that industry.

In an additional test, we use client sales as the basis to calculate market share captured by an auditor. Further,
we measure industry expertise based on city-level industry market shares and both national- and city-level
industry market shares as suggested in Francis et al. (1999; 2005a), The results based on these tests (untabulated)
are qualitatively similar to the results reported in the tables.
12

3.4 Construction of Low Balling Sub-sample


In order to assess if low balling is a viable explanation for the lower earnings quality of firms
with short auditor tenure, we construct a sub-sample of firms likely to be associated with low
balling. We define low balling as a condition when (1) auditors charge abnormally low audit
fees and (2) auditor tenure is short. We estimate the normal level of audit fees based on an
audit fee model (e.g. Gul and Tsui 1998; Francis et al. 2005a), 8 using observations with
auditor tenure of more than three years. Audit fees are considered to be abnormally low when
they are one standard deviation below the projected audit fees.9 Tenure is considered to be
short if it is less than or equal to three years (e.g. Johnson et al. 2002; Carcello and Nagy
2004). We include firms with auditor tenure less than or equal to three years during the
2000-2004 period for which audit fee data is publicly available. 10 The indicator variable
LowBall is coded as one if the actual audit fee paid by the firm is one standard deviation
lower than the projected audit fee, and zero otherwise.

3.5 Regression Model


Our main tests are conducted based on the absolute value of discretionary accruals (ABSTDA)
under the premise that upward as well as downward adjustments of reported earnings are
considered as earnings management that lowers the quality of reported earnings (Myers et al.
2003). The following regression model is used to estimate the association between absolute
discretionary accruals and auditor tenure:
ABSTDA = + 1Tenure + 2 Age + 3 Size + 4 Size2 + 5 Size3 + 6 IndGrow + 7 Grow
+ 8 CFO + j IndustryDum + k YearDum +
(2)
where:

ABSTDA

= absolute value of discretionary accruals scaled by average total


assets;

The audit fee model used is discussed in Appendix A of this paper.


Alternatively, we define abnormally low audit fees to be negative abnormal fees when they are in the lowest
25% of the distribution. We obtain qualitatively similar results (untabulated) using this alternative definition.
10
We also conduct sensitivity tests by splitting the sample to groups based on the pre- and post-SOX periods,
and obtain qualitatively similar results (discussed in section 4.5.5).
9

13

Tenure
Age
Size, Size2 & Size3
IndGrow

= log of auditor tenure, measured as the number of consecutive


years that the firm has retained the auditor;
= log of firm age, measured as the number of years that the firm
appeared in Compustat since 1950;11
= market value of equity (in 10 billions) to the power one, two and
three;
N
= N
Sales
/

i ,t Sales i ,t 1 by Fama and Frenchs (1997) 48 industry


i =1

Grow
CFO
IndustryDum
YearDum

=
=
=
=

i =1

groups;12
firm-specific sales growth, measured as Salest / Salest-1;
firms cash flow from operations divided by average total assets;
dummies for Fama and Frenchs (1997) 48 industry groups; and
dummies for fiscal year.

Since absolute values of discretionary accruals used as a dependent variable are truncated at
zero, a truncated regression approach is adopted to arrive at unbiased estimates of coefficients
in the model.13

In order to control for factors affecting discretionary accruals that are unrelated to auditor
tenure, we include several control variables (see also Myers et al. 2003). We use firm age
(Age) to control for the difference in discretionary accruals of firms with different life cycles
(Anthony and Ramesh 1992) and market value of equity14 (Size) to control for the differences
in the accrual behaviour of managers of large and small firms. While some researchers argue
that larger firms have more stable discretionary accruals (Dechow and Dichev 2002), others
document that the magnitude of discretionary accruals reported by larger firms is
systematically lower (e.g. Ashbaugh et al. 2003). Because Size is correlated strongly with

11

Alternatively, we use the number of years the firm has existed on the CRSP database to measure Age; the
results (untabulated) are qualitatively similar to our main findings.
12
In a sensitivity test, we compute IndGrow and include industry dummies based on two-digit SIC codes; the
results (untabulated) are qualitatively similar to our main findings.
13
See Myers et al. (2003), Greene (2000, pp. 682-690) and Maddala (1977, pp. 269-273) for further discussion
on the truncated regression approach.
14
Other measures of size, such as sales or total assets, are used in sensitivity analyses; the results (untabulated)
are qualitatively similar to our main results. In addition, we conduct tests on small and large client firms
separately based on median size. The test results (untabulated) are similar to the main results.
14

both ABSTDA and other firm characteristics, we use a third-degree polynomial (Size, Size2 &
Size3) of market value to capture the non-linear size effects.

Because growth firms in an industry may report systematically different levels of accruals
(Myers et al. 2003), the growth effect is controlled by using growth in industry sales
(IndGrow). Firm-specific growth in sales (Grow) is also included in order to capture the
possible difference in the accruals behaviour between firms with high and low growth that is
unrelated to earnings management.15 In addition, we include CFO in the model to take into
account the negative association between accruals and cash flows as documented in prior
studies (e.g. Dechow 1994). Industry and year fixed effects are also included in the model.
All continuous variables are winsorized at three standard deviations.

We expect the coefficient for Tenure to be negative, suggesting shorter tenure is associated
with higher discretionary accruals, i.e. lower earnings quality. Additionally, we expect this
association to be significantly weaker for firms audited by industry specialists. We conduct
tests based on both specialist and non-specialist sub-samples, as well as on the full sample
with an interaction term between tenure and specialists.

4. EMPIRICAL RESULTS
4.1 Descriptive Statistics

Table 1 presents descriptive statistics for discretionary accruals and other variables used in
the study.
(Insert Table 1 here)

15

Recent studies (e.g. Hribar and Nichols 2007) suggest that operating volatility is an additional determinant of
absolute discretionary accruals. As additional tests, we include various measures of operating volatility
(standard deviation of operating cash flows, standard deviation of sales and standard deviation of earnings over
the current and prior four years) and re-estimate all our models. The results (untabulated) are qualitatively
similar to our main findings.
15

Panel A shows that the means of absolute discretionary accruals (ABSTDA) are 3.6% and
4.0% for specialists and non-specialists respectively, indicating that the magnitude of
discretionary accruals is lower for the specialist sub-group. Though the means of absolute
discretionary accruals are consistent with prior literature (e.g. Balsam et al. 2003), their high
magnitudes (4% of total assets) may be due to problems associated with the specification of
discretionary accruals models (e.g. Dopuch et al. 2005; Ball and Shivakumar 2006).16

Univariate tests show that the mean of Tenure(log) is shorter for the specialist sub-sample
compared to the non-specialist sub-sample. In addition, firms audited by specialists are in
general older, larger in size, and have higher cash flows.

The correlations among variables used in the regression, as reported in Panel B of Table 1,
show that tenure is negatively associated with absolute discretionary accruals and similarly
firm age, firm size, industry growth, firm-specific growth and cash flows are negatively
correlated with discretionary accruals.

4.2 Auditor Tenure and ABSTDA

We first test the association between auditor tenure and absolute value of discretionary
accruals (ABSTDA) for the total sample. The results based on the total sample, reported in the
All column in Table 2, show that the coefficient for Tenure is significantly negative. This
finding is consistent with prior research (e.g. Myers et al. 2003), and indicates that longer
auditor tenure is associated with lower absolute discretionary accruals, suggesting higher
quality of reported earnings. Results on the control variables show that while the coefficient
for industry growth is insignificant, firm-specific growth is positively associated with

16

Caution is required in drawing inferences based on results using discretionary accruals with such high
magnitudes. To address this potential weakness, we conducted additional tests by using the likelihood of
meeting or beating earnings benchmarks and the results are qualitatively similar (discussed in Section 4.6.3).
16

ABSTDA, consistent with evidence in prior studies (e.g. Ashbaugh et al. 2003). Similar to
Myers et al. (2003), our results show that firm age is negatively associated with discretionary
accruals. The results on cash flows are also consistent with other prior studies (e.g. Dechow
1994). It is also shown that the relationship between size and discretionary accruals is nonlinear.17
(Insert Table 2 here)

4.3 Auditor Tenure, Auditor Industry Specialization and ABSTDA

In order to evaluate the role of auditor industry specialists on the association between
discretionary accruals and auditor tenure, we conduct two tests. First, we divide the total
sample into auditor industry specialists and auditor non-specialists sub-samples and conduct
tests separately for each of the sub-samples. Out of the total sample, 24,433 firm-year
observations are audited by non-specialists and 8,344 firm-year observations by industry
specialists. The results on the sub-samples, as reported in Table 2, show that the Tenure
coefficient continues to be negative and significant for the non-specialist sub-sample,
suggesting that discretionary accruals are higher when auditor tenure is short and firms are
audited by non-specialists (t-stat = 6.2). The Tenure coefficient for the specialist sub-sample
is, however, statistically insignificant based on the two-tailed test (t-stat = 1.6), suggesting
that shorter tenure is not associated with higher discretionary accruals when firms are audited
by industry specialists. Although the coefficient is statistically significant at 10% level with a
one-tailed test, the magnitude of coefficient for the non-specialist sub-sample (-0.382) is
significantly larger than that for the specialist sub-sample (-0.14) (F-value = 21.2). Consistent

17

We also conduct a number of tests to address some econometric concerns. For example, we compute clustered
standard errors, as suggested by Petersen (2009) to address the potential problems of non-independence of panel
observations. We also estimate average coefficients of annual regressions over the 12 years for all our tests
using the Fama-MacBeth procedure. These tests provide qualitatively similar results. We also obtain similar
results in other sensitivity tests, such as the inclusion of firm-level fixed effects, the use of ordinary least-squares
regression and the use of White-corrected statistics (1980).
17

with our expectation, this finding suggests that the association between auditor tenure and
discretionary accruals is significantly weaker for firms audited by industry specialists.

Second, we conduct a test on the total sample by including the Specialist variable and an
interaction term between Specialist and Tenure.18 To allow the coefficients of the control
variables to vary between specialists and non-specialists, we also include interaction terms
between Specialist and all control variables, including the industry and year fixed effect
variables. The results reported in Table 3 show that the coefficient for Tenure continues to be
significantly negative. The coefficient for Specialist is also negative, but marginally
significant using a one-tailed test (t-stat = 1.4). 19 In addition, the coefficient for the
interaction term between Tenure and Specialist is significantly positive, suggesting that the
negative association between auditor tenure and discretionary accruals is weaker for client
firms audited by industry specialists.
(Insert Table 3 here)

4.4 Low Balling Tests

In order to evaluate the low balling explanation, we identify a sample of firms that are more
likely to be associated with low balling. We limit our analyses to firms with auditor tenure of
not more than three years in the period 2000-2004 for which audit fee data are publicly
available. This results in a sample of 1,087 firm-year observations. Of the 1,087 observations,
178 report audit fees below one standard deviation from the predicted level, and they fall into

18

Tests based on full sample with an interaction variable may be less precise under certain circumstances when
coefficients of control variables may differ between the two groups (e.g. Hardy 1993). In order to overcome this
weakness, it has been suggested that separate regression tests on the two groups be conducted, which will
provide better results when the association between the X variable (auditor tenure) and Y variable (discretionary
accruals measures) is hypothesized to be contingent on the moderator variable Z (auditor specialisation) which
assumes two values (specialists or non-specialists) (e.g. Staw and Oldham 1978; Wright et al. 1996, p. 452).
Consequently our inferences are mainly drawn based on the sub-sample analyses.
19
As an additional test (untabulated), we remove all interaction terms (except for Tenure*Specialist) and find
that the coefficient for Specialist becomes significantly negative, consistent with prior studies (e.g. Balsam et al.
2003).
18

the low balling sub-sample (LowBall = 1). The rest of the observations (N = 909) is
considered non-low balling (LowBall = 0).

To provide some construct validity for the LowBall variable, we compare the mean audit fees
for the low balling firms during the low balling period (i.e. first three years of their
engagement) with the later period, i.e. after three years of their engagement. If the LowBall
variable correctly identifies the firms that are associated with low balling, we expect their
audit fees to be lower in the early years of audit engagement (low balling period) compared to
later years (non-low balling period). We find that the mean audit fees for the low ball period
is 0.17% of total assets and that for the later period is 0.27%, and the difference is
statistically significant (t-stat = 5.3). 20 This result is consistent with the low balling
phenomenon which our LowBall variable is intended to capture.

We first compare absolute discretionary accruals between the firms that are likely to be
associated with low balling and those that are not. The means of ABSTDA for low balling
and non-low balling firms are 4.35% and 4.20% respectively, and the difference is not
statistically significant (t-stat = 0.85). The univariate test results thus suggest that our finding
on the association between short tenure and earnings quality is not consistent with the low
balling explanation.

Next, we run a regression test on the reduced sample of 1,087 observations with ABSTDA as
the dependent variable and include the LowBall dummy variable as an independent variable

20

We also compare audit fees of the non-low balling firms for the first three years of their audit engagements
with the later years of their audit engagements. The mean audit fee for the early years is 0.38% of total assets
and that for the later period is 0.34%. The difference in mean audit fees is not statistically significant (t-stat =
1.4), suggesting that auditors of the non-low balling firms, unlike their low balling counterparts, do not charge
significantly lower audit fees in the early years of their audit engagements.
19

(descriptive statistics of this smaller sample are reported in Panel A of Table 4). The results
(Model 1) reported in Panel B of Table 4 show that the coefficient for LowBall is statistically
insignificant, suggesting that there is no significant association between ABSTDA and firms
that are likely to be associated with low balling. We also include the LowBall dummy
variable in the full sample with both short tenure and long tenure observations during the
period 2000-2004 (N = 13,062) in the model. Results for Model 2 show that the coefficient
for Tenure is significantly negative, as expected, while the LowBall coefficient continues to
be insignificantly different from zero. In addition, we repeat our test in Table 2 using 20002004 data by including the LowBall variable as an additional control. Results in Panel C of
Table 4 are qualitatively similar to that in Table 2, in that the negative association between
Tenure and ABSTDA is stronger in the non-specialist group when compared to the specialist
group. The coefficients for LowBall are not statistically significant in both sub-samples. The
R2 of these regressions are similar to those reported in Table 2. 21 The results in Table 4,
which are also consistent with Coate and Loeb (1997) based on their two-period analytical
model, indicate that low balling is not associated with lower audit quality.
(Insert Table 4 here)

The low balling tests results, however, need to be viewed with caution for the following
reasons. First, our conclusion is only valid to the extent that our LowBall variable correctly
identifies the firms that are associated with low balling. Classification of firms as low balling
thus depends on correct estimation of the normal level of audit fees, which, in turn,
depends on the extent to which the audit fee model is well-specified. Second, although we

21

In a sensitivity test, we re-estimate Table 2 results after excluding observations with auditor tenure less than
four years, since low balling affects auditors in their early years of audit engagement. As a result of excluding
potential low balling firms, our sample drops to 27,804. The regression results (untabulated) are similar to the
results contained in Table 2, suggesting that the positive association between discretionary accruals and
relatively shorter tenure is weaker for auditor industry specialists. We obtained similar results when we
exclude firms with auditor tenure less than three years. This is consistent with our earlier evidence of not finding
a significant association between low balling and discretionary accruals.
20

obtain similar results by using different cut-offs (e.g. the bottom quartile of negative
abnormal fees, one standard deviation below the projected audit fees) for identification of low
balling firms, the fact remains that we do not find significant results for the LowBall variable,
which raises concerns with regard to the power of our tests. Lastly, even if the low balling
explanation is not supported by our results, we cannot unambiguously exclude other possible
alternative explanations for the short tenure/low earnings quality relationship using this test.

4.5 Additional Analyses


4.5.1 Self-selection and Endogeneity

It is possible that discretionary accruals and auditor tenure are endogenously determined, i.e.
firms with lower discretionary accruals may be motivated to retain auditors for a longer
period or auditors are less likely to drop clients with lower discretionary accruals. To control
for endogeneity, we employ a two-stage least-squares regression (2SLS) approach to obtain
consistent and efficient estimators. In the first stage, we obtain the predicted value of Tenure,
which is then used in the second stage regression. Our Tenure prediction model is augmented
from the auditor choice model used by Chaney et al. (2004). Specifically, we estimate the
following first stage equation:
Tenure = + 1 Aturn + 2 DA + 3Curr + 4 Quick + 5 ROA + 6 ROA * LOSS
+ 7 Export + 8 Litigation + 9 Lag ( ABSTDA) + 10 Age + 11 Size + 12 Size2 + 13 Size3

+ 14 IndGrow + 15 Grow + 16 CFO + j IndustryDum + k YearDum +


(3)
where:
Tenure
Aturn
DA
Curr
Quick
ROA
Loss
Export

= log of auditor tenure, measured as the number of consecutive years


that the firm has retained the auditor;
= asset turnover, measured as sales divided by total assets;
= debt-asset ratio, measured as long-term debt divided by total assets;
= current ratio, measured as current assets divided by total assets;
= quick ratio, measured as current assets minus inventory divided by
current liabilities;
= earnings before interest and taxes divided by total assets;
= dummy variable, 1 if the firm incurred a loss in the previous year, 0
otherwise;
= foreign sales divided by total sales;
21

= dummy variable, 1 of the firm operates in a high-litigation industry,


and 0 otherwise. High-litigation industries are industries with SIC
codes of 2833-2836, 3570-3577, 3600-3674, 5200-5961, and 73707370 (Ashbaugh et al. 2003);
= lag values of ABSTDA;
Lag(ABSTDA)
= log of firm age, measured as the number of years that the firm
Age
appeared in Compustat since 1950;
Size, Size2 & Size3 = market value of equity (in 10 billions) to the power one, two and
three;
N
= N
IndGrow
Salesi,t / Salesi,t 1 by Fama and Frenchs (1997) 48 industry
Litigation

i =1

Grow
CFO

i =1

groups;
= firm-specific sales growth, measured as Salest / Salest-1; and
= firms cash flow from operations divided by average total assets.

The model controls for firm size (Size, Size2 & Size3), firm complexity (Aturn, Curr, Quick,
Export), and firm risk (DA, ROA, ROA*LOSS), which are likely to be associated with auditor
retention, as the incumbent auditor would have a better understanding of the clients business.
Firm age (Age) is included because mature firms are more likely to retain an auditor that is
likely to provide high audit quality. Direction of the Litigation variable a priori is not clear.
It is possible that firms in more litigious industries are motivated to retain the incumbent
auditors to provide more credibility to their reports. On the other hand, it is also possible that
auditors may be motivated to drop the client firms from more litigious industries to reduce
their litigation risk. Firms with higher earnings quality (Lag(ABSTDA)) are likely to retain the
incumbent auditor to provide positive signals to the market on their earnings quality. The
model also controls for industry and year fixed effects. In addition, control variables in the
second stage equation are also included, as suggested in Larcker and Rusticus (2008). The
sample for this model drops to 26,756 observations because of additional data requirements.
The results for the first and second stage estimations are presented in Panel A of Table 5.
(Insert Table 5 here)

22

The results for the first stage estimation (Panel A) show that Tenure is positively associated
with Size, Age, CFO, Cur and Quick, 22 and negatively associated with Lag(ABSTDA). In
addition, the results show that partial R2 is reasonably high and partial F is statistically
significant, suggesting that the model is unlikely to be subject to weak instrument problems
(i.e. situations where the instrument and the regressor exhibit a small correlation) (see
Larcker and Rusticus 2008).

The second stage results are consistent with our earlier findings. Although the magnitude of
the coefficients for predTenure are much smaller in the second-stage results, the coefficient
for predTenure is still significantly negatively associated with ABSTDA for the nonspecialists sub-sample. While the predTenure coefficient for the sub-sample of industry
specialists is also statistically significant, the magnitude of this coefficient is significantly
smaller (F-stat = 23.46), consistent with our earlier evidence.

Hiring of specialist auditors may also be endogenously determined,23 i.e. firms with high
earnings quality hire audit specialists. We address this issue by conducting 2SLS tests with an
interaction term. The predicted values for Specialist and Tenure are used in the second stage
estimation to evaluate the joint effect of industry specialists and tenure on ABSTDA. The
results are reported in Panel B of Table 5. The results for the Specialist model show that
there is a significantly positive association between industry specialists and Size, Litigation
and DA, suggesting that larger firms, firms from more litigious industry groups, and firms
with higher debt-to-assets ratios are likely to hire industry specialists. The coefficients for
Aturn, Lag(ABSTDA) and Age are significantly negative, suggesting that mature firms, firms
with higher asset turnover and firms with higher levels of discretionary accruals are more

22

This is consistent with Chaney et al. (2004).


We thank the anonymous reviewer and the editor for drawing attention to the endogenous nature of hiring
industry specialists.

23

23

likely to hire non-specialists. While the partial F (6.03) is still statistically significant
(suggesting that the predicted variable does not suffer from a weak instrument problem), the
partial R2 is very low (0.025), implying that our model for specialists endogeneity is weak.

In the second stage estimation, we include predTenure, predSpecialist and the interaction
term between predTenure and predSpecialist in the model for the full sample, as in Table 3.24
The results for this second stage regression (Panel B) show that predTenure is negatively
associated with discretionary accruals, and the coefficient is significant. The coefficient for
the interaction term between predTenure and predSpecialist is significantly positive,
suggesting that auditor industry specialization moderates the negative association between
predTenure and ABSTDA. Consistent with our earlier results, the association between
discretionary accruals and auditor tenure is significantly weaker for firms audited by industry
specialists.

While 2SLS results are consistent with our main findings, some cautions in their
interpretation are warranted. Low R2 of the Specialist model suggests that validity of the
instrument for Specialist may be questionable. Unfortunately, there is no complete structural
model on the determinants of Specialist in the literature to provide us with guidance. In order
to test endogeneity for both first-stage models, we perform the over-identifying restriction
test (see Hausman 1978 and Godfrey and Hutton 1994) and also the Hausman (1978) test.
The results (untabulated) show that the Hausman (1978) test for endogeneity is significant for
both Tenure (stat = 51.18, p-value < 0.01) and Specialist (stat = 10.76, p-value < 0.01)

24

Discussions with some econometricians suggested that there should be no problem in using two predicted
variables in the same second stage equation, even though the two variables are estimated utilizing identical
variables. This might cause problems only when the coefficients of the two first-stage equations are the same,
and therefore the two predicted variables are essentially the same variable. Panel B of Table 5 shows that the
estimated results of the two equations are different, so are the R2 and F-values. In addition, we also find that the
correlation between the two predicted variables (PredTenure and PredSpecialist) is rather low (0.09).
24

models, rejecting the hypothesis of no endogeneity. The results of the over-identifying


restriction test for both Tenure (stat = 15.22, p-value < 0.01) and Specialist (stat = 10.59, pvalue < 0.01) models are also significant, which implies that the instruments constructed in
this study are not totally exogenous and uncorrelated with the error term.

However, following Larcker and Rusticus (2008) and Francis and Lennox (2008) we conduct
tests to assess the extent of the endogeneity problem and sensitivity of our results to
endogeneity. These additional tests are conducted by including different combinations of the
explanatory variables in the first-stage models (exclude each of the explanatory variable from
the models [one at a time] except for the obvious determinants, such as size). Intuition for
these tests is that if the instruments are not valid, tests based on a different set of determinants
might produce different estimates of the true coefficients. The results of these tests show that
our main results are not sensitive to these alternative specifications.

4.5.2 Signed Discretionary Accruals

In our main analyses we use the absolute value of discretionary accruals as a measure of
earnings quality, which captures the combined effect of income-increasing and incomedecreasing earnings management decisions (Warfield et al. 1995; Myers et al. 2003). Some
researchers (e.g. Ashbaugh et al. 2003), however, argue that earnings overstatements are
more frequent and of greater concern to auditors, as they are likely to be associated with
opportunistic earnings management. On the other hand, while the use of negative
discretionary accruals (i.e. downward adjustment of reported earnings) could be opportunistic,
it may also be considered as a form of conservative accounting (Ashbaugh et al. 2003). Thus,
we conduct separate tests on observations with positive (income-increasing) and negative
(income-decreasing) discretionary accruals. The results are reported in Table 6.
(Insert Table 6 here)
25

The results for the sub-sample of positive discretionary accruals (POSTDA) are similar to the
results for the full sample reported in Table 2 i.e. the Tenure coefficient is negative and
significant for the non-specialists sub-sample, and insignificant for the specialists sub-sample.
The results for the sub-sample of negative discretionary accruals (NEGTDA*-1), however,
show that while shorter auditor tenure is associated with more negative discretionary accruals
(in the All column), the positive association is stronger for the non-specialists sub-sample
compared to the specialists sub-sample. If more negative discretionary accruals are
interpreted as a more conservative application of GAAP, this result by itself is not supportive
of our expectation.

In the current development of the literature, it is not clear ex-ante whether NEGTDA should
be interpreted as downward earnings management or conservative reporting. To further
investigate this issue, we identify a smaller sample in which managers have a stronger
incentive to opportunistically manage earnings using NEGTDA. For example, prior studies
(e.g. Healy 1985; Gaver et al. 1995; Holthausen et al. 1995) suggest that managers have an
incentive to smooth earnings or manage income downwards when it far exceeds certain
thresholds. In the other extreme, managers will have an incentive to take a big bath when
income is very low, so that they can report larger earnings in the following years (Healy 1985;
Walsh et al. 1991). We divide the NEGTDA sample into two groups based on the managerial
incentives to opportunistically manage earnings downwards, using the magnitude and sign of
the pre-managed earnings (income NEGTDA). In particular, we argue that when the premanaged earnings are either very high or very low, managers will have greater incentives to
smooth income or take a big bath using NEGTDA. Thus, firms with very high or very low
pre-managed earnings are more likely to be the firms that opportunistically manage earnings
using NEGTDA. When pre-managed earnings are not extreme, managerial incentives to
26

manage earnings downward will be lower, and the NEGTDA in these firms is less likely to be
opportunistic earnings management that lowers the quality of earnings.25

To operationalize this test, we select the top and bottom 20% of pre-managed earnings to
represent extreme earnings; the rest of the observations in the NEGTDA sample are
considered less extreme earnings.26 Regression results for each group are reported in Panel
B of Table 6. The coefficient for Tenure in the extreme pre-managed earnings group is
significantly negative, consistent with the results for the total negative discretionary accruals
group contained in Panel A, whereas it is insignificant for the group of non-extreme premanaged earnings. These results thus suggest that managers of the extreme pre-managed
earnings group have a stronger incentive to manage earnings downward using NEGTDA. To
the extent that extreme earnings is an appropriate measure of the stronger managerial
incentive to use NEGTDA opportunistically, our findings for NEGTDA provide evidence
consistent with our expectations and earlier results for POSTDA and ABSTDA.

4.5.3 Alternative Specifications of Auditor Tenure

As an additional test, we differentiate between medium and long audit tenure following the
procedure used by Johnson et al. (2002). Instead of a continuous variable, we use two
indicator variables for medium length of the auditor-client relationship (MEDIUM equals one
when the length of the auditor-client relationship is four to eight years, and zero otherwise)
and long tenure (LONG equals one when the length of the auditor-client relationship is nine
years or longer, and zero otherwise), with short tenure (three years or below) as the default

25

The NEGTDA in this group might reflect conservative application of GAAP, and might also reflect other
motives for downward earnings management. As a result, we do not have an expectation for this group of firms
in terms of their association between earnings quality, auditor tenure, and industry specialization.
26
As sensitivity analyses we use other cut-offs such as top and bottom quartiles, 33% and one standard
deviation, and results (untabulated) are qualitatively similar.
27

group.27 In addition, we also adopt a more restrictive definition of longer auditor tenure
(TenDum), as suggested in Carcello and Nagy (2004). Auditor tenure is considered long
(TenDum = 1) when an auditor is on the job for nine years or longer, and considered short
(TenDum = 0) when the auditor is on the job for three years or less. The observations with
auditor tenure between four and eight years are dropped from the analyses, reducing the
sample size to 13,723. The results are reported in Table 7.
(Insert Table 7 here)

The results of these tests are similar to our earlier findings. Specifically, we find that auditors
with both medium tenure (MEDIUM = 1) and long tenure (LONG = 1) are associated with
lower ABSTDA if the firms are audited by industry non-specialists, but not when firms are
audited by an industry specialists. Although the coefficient for LONG is also marginally
significant for the specialist group using a one-tailed test (t-stat = 1.5), we find that the
magnitude of the coefficients for both MEDIUM and LONG are significantly larger for the
non-specialist sub-sample than that for the specialist sub-sample. In addition, we find that the
magnitude of the coefficients for LONG is also significantly larger than that for MEDIUM for
both the specialists (F-stat = 4.6) and the non-specialists (F-stat = 10.3) samples, suggesting
that the level of ABSTDA is further lower when auditor tenure is nine years or more. The
results on TenDum are also qualitatively similar to our main results.28

In theory, non-specialists are expected to be able to accumulate sufficient client-specific


knowledge at some point of time to ensure high quality earnings, assuming that the

27

Johnson et al. (2002) use the medium length of auditor-client relationship as the default group. However,
since our focus is on the short tenure/lower earnings quality relationship, we use auditor tenure fewer than four
years as the default group.
28
Interaction analyses are again sensitive to alternative specifications due to possible multicollinearity problems.
For example, our results (untabulated) show that the interaction term between MEDIUM (LONG) and Specialist
is significantly negative (positive), which is consistent with our main findings, while the interaction term
between TenDum and Specialist is statistically insignificant.
28

operations as well as the business environment of the client remain constant over time. Our
results, however, suggest that four to eight years may not be a sufficient time for nonspecialist auditors to acquire enough client-specific knowledge. This is not surprising given
the dynamic nature of business operations and the changing business environments. It is
possible that auditors benefit from the accumulation of new client-specific knowledge as
time goes. As such, it may be difficult to predict at what point the auditor learning is
sufficient. We leave this issue for future research.

4.5.4 Pre- and Post-SOX Periods

Since our sample covers the periods before and after the Sarbanes-Oxley (SOX) Act of 2002,
we repeat our analyses separately for each period. We include in this analysis only firms that
exist in both the pre-SOX (1993-2001) and post-SOX (2002-2004) periods to partially control
for the difference in the two samples caused by the change in sample composition. Based on
this smaller sample, we investigate if the changes in the regulatory environment and auditing
profession after the passage of SOX have any effect on the relationship between auditor
tenure, auditors industry specialization and earnings quality. The results (untabulated) show
that the magnitudes of coefficients for Tenure are negative and significant for both periods
but much smaller for the post-SOX period (coeff = -0.261, t-stat = 2.19) compared to the preSOX period (coeff = -0.606, t-stat = 6.05). These results thus indicate a comparatively
weaker association in the post-SOX period. 29 In addition, consistent with our main findings,
the test results (untabulated) for the low balling group show that the association between
LowBall and ABSTDA is statistically insignificant for both pre-SOX (t-stat = 0.49) and postSOX (t-stat = 0.65) periods.

29

A possible reason for weaker results in the post-SOX period is that the more stringent financial reporting and
auditing requirements mandated by SOX forced all auditors to offer higher quality services irrespective of the
length of tenure and/or industry specialization (Ettredge et al. 2007).
29

4.6 Other Earnings Quality Proxies


4.6.1 Alternative Specifications of Discretionary Accruals

Robustness tests are also conducted using other measures of discretionary accruals. We reestimate discretionary accruals using the modified Jones (1991) model (ABSJONES), as well
as discretionary accruals based on the performance-adjusted modified Jones model
(Ashbaugh et al. 2003; Kothari et al. 2005), which controls for the mechanical relation
between current periods discretionary accrual estimate and firm performance. Additionally,
we conduct tests using the accruals quality measure (AQ) developed in Dechow and Dichev
(2002) and Francis et al (2005b).

In addition, following Francis et al. (2005b), we

decompose AQ into innate and discretionary parts of AQ, and replace AQ with discretionary
AQ (DISCAQ) since it is the discretionary component that is more likely to be associated with
opportunistic earnings management behaviour. These results (untabulated) using all these
alternative estimations are similar to those reported in Table 2, suggesting that our main
results are not sensitive to different ways of estimating discretionary accruals.30

Recent research suggests that discretionary accruals estimated cross-sectionally can be noisy
and bias the resulting tests if the firms in the industry are not homogeneous (Dopuch et al.
2005). As a robustness check, we repeat our tests based on discretionary accruals estimated
for each firm using the time-series model. The regression results (untabulated) based on these
discretionary accruals are qualitatively similar to our main findings, suggesting that the
measurement errors associated with industry heterogeneity is not likely to affect our
conclusions.

30

We find that the results (untabulated) of a regression test with an interaction between Tenure and Specialist
are similar to those reported in Table 3 for all models except for discretionary accruals estimated using the
modified Jones model (1991).
30

We conduct pair-wise correlation tests on ABSTDA with discretionary accruals based on other
models, and find that the absolute discretionary accruals based on all models are significant
(p < 0.001) and highly correlated with each other (the correlation between ABSTDA and all
the other five models ranged from 0.33 [ABSTDA and ABSJONES] to 0.78 [ABSTDA and
ABSTIME]). High correlations could indicate that the use of all discretionary models suffers
from some specification problems, as has pointed out in the literature (e.g. Dopuch et al. 2005;
Ball and Shivakumar 2008). The extent and the type of the problems are, however, likely to
be different for each measure, especially in cases when their correlations are relatively low.
To the extent that each measure reflects some different facet of the truly discretionary
component, the consistent results we obtained based on different models show that our
findings are not sensitive to these alternative measures and the use of these measures does not
work against our conclusions.

4.6.2 Tests based on Absolute Value of Non-discretionary Accruals (ABSNDA)

In order to evaluate the behaviour of non-discretionary accruals, we conduct tests with


absolute value of these accruals (ABSNDA) as the dependent variable. The results
(untabulated) show that, consistent with our main findings contained in Table 2, Tenure is
negatively associated with ABSNDA (coeff = -0.02, t-stat = 2.52). In addition, the coefficient
for Tenure is still significantly negative for the non-specialist group (coef = -0.03, t-stat =
2.84), whereas it is insignificant for the specialist group (coef = 0.00, t-stat = 0.29). It should,
however, be noted that the magnitude of the Tenure coefficients in this test are much lower
than for discretionary accruals reported in Table 2, which may suggest that auditors are less
concerned about non-discretionary accruals. We also conduct a test on ABSNDA with an
interaction term between Tenure and Specialist (similar to Table 3), and the results
(untabulated) show that the interaction term is insignificant (t-stat = 0.11).

Similar to the

results contained in Table 4, the results of a test on the low balling group show that the
31

association between LowBall and ABSNDA continues to be statistically insignificant (t-stat =


0.30).

4.6.3 Meeting or Beating Earnings Benchmarks

In view of the concerns expressed in the literature with regard to the problems associated
with the specification of discretionary accruals models (e.g. Dopuch et al. 2005; Ball and
Shivakumar 2008), we use the meeting or beating earnings benchmarks as alternative
measures of earnings quality. A stream of research (e.g. Burgstahler and Dichev 1997;
Ashbaugh et al. 2003) argues that firms with small increase in earnings or firms with earnings
just meeting or beating analyst earnings forecasts are more likely to engage in earnings
management. Therefore, we use the likelihood of firms reporting small earnings increases
(INCREASE) and the likelihood of firms meeting or beating analyst earnings forecasts
(SURPRISE) as additional earnings quality proxies. We expect that firms with longer tenured
auditors are less likely to be associated with INCREASE or SURPRISE, and this association is
expected to be significantly weaker for firms audited by industry specialists. We conduct our
tests based on the following logit model (see also Ashbaugh et al. 2003 who use a similar
model):
INCREASE / SURPRISE = + 1Tenure + 2 Litigation + 3 MB + 4 LnMVE + 5 IH
+ 6 LOSS + 7TDA + j IndustryDum + k YearDum +
(4)
where:

INCREASE
SURPRISE
Tenure
Litigation
MB

= dummy variable, 1 when the increase in net income, scaled by average


total assets, falls in the interval [0.00, 0.02], and 0 otherwise;
= dummy variable, 1 when a firm meets or beats by 1 cent the mean
consensus analysts forecast as reported in IBES database, and 0
otherwise;
= log of auditor tenure, measured as the number of consecutive years that
the firm has retained the auditor;
= dummy variable, 1 if the firm operates in a high-litigation industry
(industries with SIC codes of 2833-2836, 3570-3577, 3600-3674, 52005961 and 7370-7370), and 0 otherwise;
= market to book ratio, defined as market value of equity divided by
stockholders equity of common shareholders;
32

LnMVE
Loss
IH
TDA

= log of market value of equity;


= dummy variable, 1 if the firm incurred a loss in the previous year, 0
otherwise;
= percentage of shares held by institutional owners; and
= Discretionary accruals scaled by average total assets.

Additional data required for these tests reduce the sample size to 25,868 (19,493)
observations for the INCREASE (SURPRISE) regression. The results are reported in Table 8.
(Insert Table 8 here)

After controlling for other factors, Tenure is negatively associated with both INCREASE and

SURPRISE in the total sample, suggesting that firms with longer tenured auditors are less
likely to meet or beat earnings benchmarks. We then split the sample into the specialists and
non-specialists groups. As shown in Table 8, the negative association between tenure and
both INCREASE and SURPRISE is significantly stronger in the non-specialists sub-sample
than in the specialists sub-sample. We also estimate regression on the full sample with
interaction terms (similar to Table 3), and find insignificant results (untabulated) for the
interaction terms. However, the variance inflation factors (VIFs) for regressions with the
interaction terms are very high, indicating significant multicollinearity.

To the extent that meeting or beating earnings benchmarks is an appropriate alternative


measure for earnings quality in the audit quality context, the results of these tests are
consistent with our main findings and provide additional support for our expectation that
auditors industry specialization moderates the association between auditor tenure and
earnings quality.

33

5. CONCLUSION

This paper provides evidence that auditors industry specialization affects the relationship
between auditor tenure and earnings quality, as documented in prior studies (e.g. Myers et al.
2003; Johnson et al. 2002). Our main results show that the association between shorter
auditor tenure and lower quality of reported earnings is weaker for firms audited by industry
specialists. One possible explanation of these results might be that auditors with industry
expertise in the clients business are more likely to detect irregularities and
misrepresentations and provide higher quality audits, even if auditors lack client-specific
knowledge as a result of short auditor-client relationships. Our results do not provide support
for the low balling explanation for the short tenure/low quality earnings relationship.

While our study is subject to the usual limitations of empirical studies of this type, some
important caveats are worth emphasizing. First, validity of our results depends on the
appropriateness of the proxies for earnings quality. There is no consensus on the superiority
or higher reliability of any particular model for estimating discretionary accruals (Bartov et al.
2000; Dechow et al. 1995; Nichols 2000). Some recent studies argue that discretionary
accrual models are mis-specified and suffer from errors-in-variables problems (e.g. Dopuch
et al. 2005; Ball and Shivakumar 2008). In order to address these concerns, we use earnings
benchmark tests as alternative proxies for earnings quality. Second, our results for the low
balling tests may be influenced by the lack of power of our tests. Third, it is also possible that
there may be alternative explanations for the association between shorter auditor tenure and
earnings quality, which have not been explored or tested in this paper. For example, the
lower quality earnings for firms audited by short-tenured non-specialists may be due to high
quality auditors dropping risky clients (especially those with high discretionary accruals) in
the first few years of audit engagements. We believe that this is not a significant concern in
the results documented in this study, since our tests are based on a sample that include only
34

firms audited by Big N auditors, and exclude non-Big N auditors where dropped (riskier)
clients are likely to go. Lastly, in view of the difficulties in specifying correct models for
determining the choice of Tenure and Specialist, our results using 2SLS procedures have not
fully overcome the underlying endogeneity problem. We leave these and other issues for
future research.

35

APPENDIX A: Estimation of Expected Audit Fees

Following prior studies (e.g. Gul and Tsui 1998, Francis et al. 2005a), we estimate the
following regression model to obtain the predicted audit fees for the period 2000-2004, where
audit fee data is available:

LAF = 0 + 1 LTA + 2 LSEG + 3 CATA + 4 QUICK + 5 DE + 6 ROI


+ 7 FOREIGN + 8 OPINION + 9YE + 10 LOSS + fixedeffects +
where:

LAF
LTA
LSEG
CATA
QUICK
DE
ROI
FOREIGN
OPINION
YE
LOSS
fixedeffects

=
=
=
=
=
=
=
=
=
=
=
=
=

log of audit fees (in thousands of dollars);


log of total assets (in millions of dollars);
log of the number of unique business segments;
ratio of current assets to total assets;
ratio of current assets (excluding inventories) to current liabilities;
ratio of long-term debt to total assets;
ratio of earnings before interest and tax to total assets;
proportion of total sales from foreign operations;
dummy variable coded 1 for modified audit report, and 0 otherwise;
dummy variable coded 1 for non-Dec. 31 year-end, and 0 otherwise;
dummy variable coded 1 if loss in current fiscal year, and 0 otherwise;
year and industry dummy variables.
random-error term with the usual normality properties.

Control variables represent firm complexity, audit risk, firm size (LTA), the audit opinion
(OPINION) and a December fiscal year end (YE). The firm-complexity variables are LSEG
and FOREIGN. The audit-risk variables include CATA, DE, LOSS, QUICK and ROI.
Audit fee data is extracted from Audit Analytics. Firms with auditor tenure below four years
are excluded from the estimation. The summary statistics of the estimation sample as well as
the estimation result are reported below in Tables A1 and A2 respectively. Our results are
consistent with the results reported in the prior studies (Francis et al. 2005a).

36

Table A1:

Descriptive Statistics of the Audit Fee Estimation Sample

Variable
Audit fees (in millions)
LAF
Total assets (in millions)
LTA
No. of Segments
LSEG
CATA
QUICK
DE
ROI
FOREIGN
OPINION
YE
LOSS

Table A2:

Mean
0.866
-1.070
2092.380
5.538
2.010
0.471
0.538
3.010
0.159
-0.050
0.125
0.039
0.282
0.450

Median
0.285
-1.256
214.030
5.366
1.000
0.000
0.548
1.602
0.079
0.045
0.000
0.000
0.000
0.000

Std Dev
2.256
1.196
8564.180
1.987
1.545
0.634
0.255
5.010
0.200
0.324
0.231
0.194
0.450
0.498

Regression Estimation of Expected Audit Fees

Dependent variable: LAF

Intercept
LTA
LSEG
CATA
QUICK
DE
ROI
FOREIGN
OPINION
YE
LOSS
Fixedeffects
Adj. R2
N

Estimate

t-value

p-value

-4.055
0.514
0.192
0.454
-0.028
0.164
-0.310
0.223
0.297
-0.214
0.115
Included

-31.8
106.0
15.0
10.2
-17.1
3.9
-10.8
6.8
7.5
-12.4
6.6

(0.00)
(0.00)
(0.00)
(0.00)
(0.00)
(0.00)
(0.00)
(0.00)
(0.00)
(0.00)
(0.00)

0.764
6739

Variables are as previously defined.

37

REFERENCES

Anthony, J. H., and K. Ramesh. (1992). Association Between Accounting Performance


Measures and Stock Prices: A Test of the Life Cycle Hypothesis. Journal of
Accounting and Economics, 15(June/September), 203227.
Arel, B., R.G. Brody and K. Pany. (2005). Audit Firm Rotation and Audit Quality. CPA
Journal 75(1), 36-39.
Ashbaugh, H., R. LaFond, and B. W. Mayhew. (2003), Do Nonaudit Services Compromise
Auditor Independence? Further Evidence. The Accounting Review, 78(3), 611-639.
Ball, R. and L. Shivakumar. (2006). The Role of Accruals in Asymmetrically Timely Gain
and Loss Recognition. Journal of Accounting Research 44(2), 207-242.
Ball, R. and L. Shivakumar. (2008). Earnings Quality at Initial Public Offerings. Journal of
Accounting and Economics 45(2-3), 324-349.
Balsam, S., J. Krishnan and J.S. Yang. (2003). Auditor Industry Specialization and Earnings
Quality. Auditing: A Journal of Practice and Theory 22(2), 71-97.
Bartov, E., F.A. Gul and J.S.L. Tsui. (2000). Discretionary Accruals Models and Audit
Qualifications. Journal of Accounting and Economics 30(3), 421-452.
Becker, C.L., M.L. DeFond, J. Jiambalvo and K.R. Subramanyam. (1998). The effect of audit
quality on earnings management. Contemporary Accounting Research 15(1), 4-24.
Burgstahler, D. and I. Dichev. (1997). Earnings management to avoid earnings decreases and
losses. Journal of Accounting and Economics 24, 99-126.
Carcello, J.V. and A.L. Nagy. (2004). Audit Firm Tenure and Fraudulent Financial
Reporting. Auditing. A Journal of Practice & Theory 23 (September), 55-69.
Chaney, P.K., D.C. Jeter and L. Shivakumar. (2004). Self-selection of Auditors and Audit
Pricing in Private Firms. The Accounting Review 79(1), 51-72.
Coate, C.J and M. P. Loeb. (1997) Audit pricing, auditor changes, and the winners curse.
The British Accounting Review 29(4), 315-335.
Congress, (2002). The Sarbanes-Oxley Act of 2002. United States of America.
Craswell, A., J. Francis and S. Taylor. (1995). Auditor Brand Name Reputations and Industry
Specialisations. Journal of Accounting and Economics 20, 297-322.
DeAngelo, L.E. (1981). Auditor Independence, Lowballing and Disclosure Regulation.
Journal of Accounting and Economics 3, 113-127.
Dechow, P. M. (1994). Accounting Earnings and Cash Flows as Measures of Firm
Performance: The Role of Accounting Accruals. Journal of Accounting and
Economics 18, 3-42.
Dechow, P.M. and I. Dichev. (2002). The Quality of Accruals and Earnings: The Role of
Accrual Estimation Errors. The Accounting Review 77, 35-60.
Dechow, P.M., R.G. Sloan, and A.P. Sweeney. (1995). Detecting Earnings Management. The
Accounting Review 70(2), 193-225.
DeFond, M. L., Subramanyam, K. R., (1998). Auditor changes and discretionary accruals.
Journal of Accounting and Economics 25, 35-67.
Dopuch, N., R. Mashruwala, C. Seethamraju, T. Zach. (2005). Accrual determinants, sales
changes and their impact on empirical accrual models. Working paper, Washington
University in St. Louis.
Dunham, K.J. (2002). Firms That Want To Switch Auditors Find It Takes Time, Money And
Faith. Wall Street Journal (March 15).
Dunn, K.A. and B.H. Mayhew, (2004). Audit firm Industry Specialization and Client
Disclosure Quality. Review of Accounting Studies, 9, 35-58.
Ettredge, M.L., L. Chan and S. Scholz. (2007). Audit Fees and Auditor Dismissals in the
Sarbanes-Oxley Era. Accounting Horizons 21(4), 371-386.
38

Fama, E. and K. French. (1997). Industry Costs of Equity. Journal of Financial Economics 43,
153193.
Foster J. M. and L.T. Johnson, (2001), Understanding the Issues: Why does the FASB have a
Conceptual Framework? Financial Accounting Standard Board.
Francis, J. and C. Lennox. (2008). Selection Models in Accounting Research. Working paper,
University
of
Missouri
at
Columnia.
Available
at
SSRN:
http://ssrn.com/abstract=1120796.
Francis, J., E.L. Maydew and H.C. Sparks. (1999). The Role of Big 6 Auditors in the
Credible Reporting of Accruals. Auditing: A Journal of Practice and Theory, 18(2),
17-34.
Francis, J.R., K. Reichelt and D. Wang. (2005a). The Pricing of National and City-Specific
Reputations for Industry Expertise in the US Audit Market. The Accounting Review
80(1), 113-136.
Francis, J., R. LaFond, P. Olsson and K. Schipper. (2005b). The Market Pricing of Accruals
Quality. Journal of Accounting and Economics 39, 295-327.
Gaver, J., K.M. Gaver and J.R. Austin. (1995). Additional evidence on bonus plans and
income management. Journal of Accounting and Economics 19(1), 3-28.
Geiger, M. and K. Raghunandan. (2002). Auditor Tenure and Auditor Reporting Failures.
Auditing: A Journal of Practice and Theory 21(1), 67-78.
Ghosh, A. and D. Moon. (2005). Auditor Tenure and Perceptions of Audit Quality. The
Accounting Review 80(2), 585-612.
Glaser, R. and M. Bassok. (1989). Learning Theory and the Study of Instruction. Annual
Review of Psychology, 636-666.
Glaser, R. and M.T.H. Chi. (1988). Overview. In The Nature of Expertise, ed. M.T.H. Chi, R.
Glaser and M.J. Farr, xv-xxviii, Hillsdale, NJ: Erlbaum.
Godfrey, L.G., and J.P. Hutton. (1994). Discriminating between errors-in-variables /
simultaneity and misspecification in linear regression models. Economic Letters 44,
359-364.
Gramling, A.A. and D.N. Stone. (2001). Audit Firm Industry Expertise: A Review and
Synthesis of the Archival Literature. Journal of Accounting Literature 20, 1-29.
Greene, W.H. (2000). Econometric Analysis. Englewood Cliffs, NJ: Prentice Hall.
Gul, A.F., B. Jaggi and G. Krishnan. (2007). Auditor Independence: Evidence on the Joint
Effects of Auditor Tenure and Nonaudit fees. Auditing: A Journal of Practice and
Theory 26(2), 117-142.
Gul, F.A. and J.S.L. Tsui. (1998). A test of the free cash flow and debt monitoring hypotheses:
Evidence from audit pricing. Journal of Accounting and Economics 24(2), 219-237.
Hardy, M.A. (1993). Regression with dummy variables. Newbury Park, California: Sage
Publications.
Hausman, J.A.. (1978). Specification tests in econometrics. Econometrica 46, 1251-1271.
Healy, P.M. (1985). The effect of bonus schemes on accounting decisions. Journal of
Accounting and Economics 7, 85-107.
Hogan, C.E. and D.C. Jeter. (1999). Industry Specialization by Auditors. Auditing: A Journal
of Practice and Theory, 18, 1-17.
Holthausen, R.W., D.F. Larcker and R.G. Sloan. (1995). Annual bonus schemes and the
manipulation of earnings. Journal of Accounting and Economics 19, 29-74.
Hribar, P. and C. Nichols. (2007). The Use of Unsigned Earnings Quality Measures in Tests
of Earnings Management. Journal of Accounting Research 45 (5), 10171053.
Johnson, V.E., I.K. Khurana and J.K. Reynolds. (2002). Audit-Firm Tenure and the Quality
of Financial Reports. Contemporary Accounting Research, 19(4), 637-660.
Jones, J. J. (1991). Earnings Management During Import Relief Investigations. Journal of
Accounting Research, 29, 193-228.
39

Karpoff, J.M., D.S. Lee., and G.S. Martin. (2008). The cost to firms of cooking the books.
Journal of Financial and Quantitative Analysis 43(2), 581-612.
Kothari, S., A. Leone, and C. Wasley. (2005). Performance Matched Discretionary Accruals
Measures. Journal of Accounting and Economics 39(1), 163-197.
Krishnan, G.V. (2003). Does Big 6 Auditor Industry Expertise Constrain Earnings
Management? Accounting Horizons, supplement, 1-16.
Krishnan, J., (1994). Auditor switching and conservatism. The Accounting Review 69, 20015.
Krishnan, J., (2005). Client industry competition and auditor industry concentration. Journal
of Contemporary Accounting and Economics 1(2), 171-192.
Kwon, S. (1996). The Impact of Competition within the Clients Industry on the Auditor
Selection Decision. Auditing: A Journal of Practice and Theory (Spring), 53-70.
Lapre, M. A., A.S. Mukkerjee and L.N. Van Wassenhove (2000). Behind the learning curve:
linking learning activities to waste reduction. Management Science 46, 597-611.
Larcker, D.F. and T.O. Rusticus. (2008). On the use of instrumental variables in accounting
research. Working paper, Stanford University. Available at SSRN:
http://ssrn.com/abstract=694824.
Lennox, C., 2000. Do companies successfully engage in opinion shopping: Evidence from the
UK? Journal of Accounting and Economics 29, 321-37.
Levitt, A (2007). Standards deviation. The Asian Wall Street Journal, March 12.
Maddala, G.S. (1977). Introduction to Econometrics. New York, NY: MacMillan Publishing
Company.
Maletta, M. and A. Wright. (1996). Audit Evidence Planning: An Examination of Industry
Error Characteristics. Auditing: A Journal of Practice and Theory (Spring), 71-86.
Mayhew, B.W. and M.S. Wilkins. (2003). Audit Firm Industry Specialization as a
Differentiation Strategy: Evidence from Fees Charged to Firms Going Public.
Auditing: A Journal of Practice and Theory, 22(2), 33-52.
Myers, J., L. Myers and T. Omer. (2003). Exploring the Term of the Auditor-Client
Relationship and the Quality of Earnings: A Case for Mandatory Auditor Rotation?
The Accounting Review 78, 779-799.
Nichols, M.F. (2000). Research Design Issues in Earnings Management Studies. Journal of
Accounting and Public Policy 19(4,5), 313-345.
Petersen, M.A. (2009). Estimating Standard Errors in Finance Panel Data Sets: Comparing
Approaches. Review of Financial Studies 22(1), 435-481.
PricewaterhouseCoopers. (2002). Mandatory Rotation of Audit Firms: Will It Improve Audit
Quality? New York, NY: PricewaterhouseCoopers LLP.
Shockley, R. and R. Holt. (1983). A Behavioral Investigation of Supplier Differentiation in
the Market for Audit Services. Journal of Accounting Research 21, 545-564.
Simunic, D. and M. Stein. (1987). Production Differentiation in Auditing: A Study of Auditor
Choice in the Market for New Issues. Canadian Certified General Accountants
Research Foundation.
Staw, B. M., and G. R. Oldham. (1978). Reconsidering Our Dependent Variables A
Critique And Empirical Study. Academy of Management Journal 21, 539-559.
Teoh, S. H., I. Welch, and T. J. Wong. (1998a). Earnings Management and the Long-Run
Market Performance of Initial Public Offerings. The Journal of Finance 53
(December), 19351974.
Teoh, S.H., T. J. Wong, and G. Rao. (1998b). Are Accruals during Initial Public Offerings
Opportunistic? Review of Accounting Studies 3, 175208.
Walsh, P., R. Craig and F. Clarke. (1991). Big bath accounting using extraordinary items
adjustments: Australian empirical evidence. Journal of Business Finance &
Accounting 18(2), 173-189.
40

Warfield, T., J. Wild, and K. Wild. (1995). Managerial ownership, accounting choices, and
informativeness of earnings. Journal of Accounting and Economics 20, 6191.
White, H. (1980). A Heteroskedasticity-consistent Covariance Matrix Estimator and a Direct
Test for Heteroskedasticity. Econometrica 48, 817-838.
Wright, P., S. P. Ferris, A. Sarin, and V. Awasthi. (1996). Impact Of Corporate Insider,
Blockholder, And Institutional Equity Ownership On Firm Risk Taking. Academy of
Management Journal 39 (2), 441-463.

41

Table 1: Descriptive Statistics

Panel A: Variables in main regressions

Variable
ABSTDA
Tenure (raw)
Tenure (log)
Age (raw)
Age (log)
Size
Size2
Size3
IndGrow
Grow
CFO

Mean
4.0%
8.016
1.910
10.474
2.176
0.180
0.941
14.956
1.093
-0.479
0.021

Non-Specialists
(N = 24,433)
Median
SD
2.6%
4.477
7.000
5.020
1.946
0.588
9.000
7.481
2.197
0.569
0.015
0.953
0.000
19.349
0.000
660.414
1.080
0.125
1.074
14.055
0.067
0.202

Mean
3.6%
8.072
1.876
11.370
2.214
0.301
1.671
22.876
1.093
-0.343
0.035

Specialists
(N = 8,344)
Median
SD
2.3%
4.046
7.000
5.840
1.946
0.645
9.000
9.422
2.197
0.623
0.023
1.257
0.001
20.989
0.000
501.326
1.081
0.137
1.075
13.772
0.072
0.178

Mean
differences
t-stat
7.43 ***
0.79
4.35 ***
7.88 ***
4.96 ***
8.04 ***
2.80 ***
1.14
0.07
0.77
5.77 ***

*, ** and *** represent significance at p < 0.10, < 0.05 and < 0.01 respectively.

Panel B: Pair-wise correlations among variables used in regressions


Variable
ABSTDA

Tenure

Age

Tenure
-0.10
(0.00)

Age
-0.12
(0.00)

Size
-0.08
(0.00)

Size2
-0.02
(0.00)

Size3
-0.01
(0.24)

IndGrow
-0.02
(0.00)

Grow
-0.03
(0.00)

CFO
-0.26
(0.00)

Specialist
-0.04
(0.00)

0.64
(0.00)

0.11
(0.00)

0.05
(0.00)

0.03
(0.00)

-0.03
(0.00)

0.04
(0.00)

0.15
(0.00)

-0.03
(0.00)

0.13
(0.00)

0.05
(0.00)

0.02
(0.00)

-0.04
(0.00)

0.04
(0.00)

0.19
(0.00)

0.03
(0.00)

0.81
(0.00)

0.60
(0.00)

0.00
(0.58)

0.02
(0.00)

0.11
(0.00)

0.05
(0.00)

0.93
(0.00)

0.00
(0.93)

0.01
(0.25)

0.04
(0.00)

0.02
(0.00)

0.00
(0.64)

0.00
(0.58)

0.02
(0.00)

0.01
(0.32)

0.00
(0.69)

0.04
(0.00)

0.00
(0.94)

0.21
(0.00)

0.00
(0.44)

Size

Size2

Size3

IndGrow

Grow

CFO

0.03
(0.00)

p values (2-tailed) are reported in parentheses.

42

Table 1: Descriptive Statistics (continued)


Variable Definitions:
ABSTDA
Tenure
Age
Size, Size2 & Size3
IndGrow

=
=
=
=
=

absolute value of discretionary accruals scaled by average total assets (based on Ball
and Shivakumar 2006);
log of auditor tenure, measured as the number of consecutive years that the firm has
retained the auditor;
log of firm age, measured as the number of years that the firm appeared in
Compustat since 1950;
market value of equity (in 10 billions) to the power one, two and three;
N

i =1

Grow
CFO
Specialist

=
=
=

Sales / Sales
i ,t

i =1

i ,t 1

by Fama and Frenchs (1997) 48 industry groups;

firm-specific sales growth, measured as Salest / Salest-1;


firms cash flow from operations divided by average total assets; and
dummy variable, equals 1 for firms audited by auditor whose market shares is
highest in terms of clients total asset for each industry group, 0 otherwise.

43

Table 2: Association between ABSTDA and Auditor Tenure, Partitioned by Auditor


Industry Specialization
Dependent variable: ABSTDA

Intercept
Tenure
Age
Size
Size2
Size3
IndGrow
Grow
CFO
YearDum
IndustryDum

Coef.
2.497
-0.299
-0.209
-0.611
0.051
-0.001
-0.085
0.005
-4.730
Yes
Yes

Pseudo R2
N

0.135
32777

All
t-stat
4.2
-5.9
-3.7
-10.3
7.5
-5.7
-0.4
3.2
-36.1

p-value
(0.00)
(0.00)
(0.00)
(0.00)
(0.00)
(0.00)
(0.65)
(0.00)
(0.00)

Coef.
2.097
-0.382
-0.142
-0.726
0.064
-0.001
0.047
0.006
-4.597
Yes
Yes
0.131
24433

Diff in coef.
F-value
p value

Non-Specialists
t-stat
p-value
2.7
(0.01)
-6.2
(0.00)
-2.1
(0.04)
-9.3
(0.00)
7.2
(0.00)
-5.9
(0.00)
0.2
(0.84)
2.8
(0.00)
-30.4
(0.00)

Coef.
3.337
-0.140
-0.357
-0.486
0.045
-0.001
-0.423
0.004
-5.051
Yes
Yes

Specialists
t-stat
p-value
3.7
(0.00)
-1.6
(0.11)
-3.6
(0.00)
-4.6
(0.00)
2.9
(0.00)
-2.2
(0.03)
-1.3
(0.19)
1.4
(0.16)
-19.0
(0.00)

0.151
8344

21.2
(0.00)

Estimations are based on truncated regression approach. t values are reported in italics. p values (two-tailed) are reported in
parentheses.

Variable Definitions:
ABSTDA

Tenure

Age

Size, Size2 & Size3


IndGrow

=
=

Grow
CFO
Specialist

=
=
=

YearDum
IndustryDum

=
=

absolute value of discretionary accruals scaled by average total assets (based on Ball
and Shivakumar 2006);
log of auditor tenure, measured as the number of consecutive years that the firm has
retained the auditor;
log of firm age, measured as the number of years that the firm appeared in
Compustat since 1950;
market value of equity (in 10 billions) to the power one, two and three;
N

i =1

i =1

Salesi,t / Salesi,t 1 by Fama and Frenchs (1997) 48 industry groups;


firm-specific sales growth, measured as Salest / Salest-1;
firms cash flow from operations divided by average total assets;
dummy variable, equals 1 for firms audited by auditor whose market shares is
highest in terms of clients total asset for each industry group, 0 otherwise;
dummies for fiscal years;
dummies for Fama and Frenchs (1997) 48 industry groups.

44

Table 3: Association between ABSTDA, Auditor Tenure and Auditor Industry


Specialization: Analyses using Interaction Terms
Dependent variable: ABSTDA

Intercept
Tenure
Specialist
Tenure*Specialist
Age
Size
Size2
Size3
IndGrow
Grow
CFO
Age*Specialist
Size*Specialist
Size2*Specialist
Size3*Specialist
IndGrow*Specialist
Grow*Specialist
CFO*Specialist
YearDum
IndustryDum
YearDum*Specialist
IndustryDum*Specialist

Coef.
1.371
-0.382
-1.699
0.241
-0.142
-0.726
0.064
-0.001
0.047
0.006
-4.597
-0.215
0.240
-0.019
0.000
-0.470
-0.001
-0.454
Yes
Yes
Yes
Yes

Pseudo R2
N

0.137
32777

t-stat
1.8
-6.3
-1.4
2.1
-2.1
-9.5
7.4
-6.1
0.2
2.9
-31.2
-1.7
1.7
-1.0
0.2
-1.1
-0.3
-1.4

p-value
(0.07)
(0.00)
(0.17)
(0.03)
(0.03)
(0.00)
(0.00)
(0.00)
(0.83)
(0.00)
(0.00)
(0.09)
(0.08)
(0.31)
(0.87)
(0.26)
(0.77)
(0.16)

Estimation is based on truncated regression approach. t values are reported in italics. p values (two-tailed) are reported in
parentheses.

Variable Definitions:
ABSTDA

Tenure

Age

Size, Size2 & Size3


IndGrow

=
=

absolute value of discretionary accruals scaled by average total assets (based on Ball
and Shivakumar 2006);
log of auditor tenure, measured as the number of consecutive years that the firm has
retained the auditor;
log of firm age, measured as the number of years that the firm appeared in
Compustat since 1950;
market value of equity (in 10 billions) to the power one, two and three;
N

i =1

Grow
CFO
Specialist

=
=
=

YearDum
IndustryDum

=
=

Sales / Sales
i ,t

i =1

i ,t 1

by Fama and Frenchs (1997) 48 industry groups;

firm-specific sales growth, measured as Salest / Salest-1;


firms cash flow from operations divided by average total assets;
dummy variable, equals 1 for firms audited by auditor whose market shares is
highest in terms of clients total asset for each industry group, 0 otherwise;
dummies for fiscal years;
dummies for Fama and Frenchs (1997) 48 industry groups.

45

Table 4: Tests for Low Balling

Panel A: Descriptive statistics of low balling samples

Sample
Variable
ABSTDA
Tenure (raw)
Tenure (log)
Age (raw)
Age (log)
Size
Size2
Size3
IndGrow
Grow
CFO
LowBall

Short Tenure (< 4 years) Sample, 2000-2004


(N = 1,087)
Mean
4.57%
2.848
1.037
10.054
1.999
0.187
0.924
9.861
1.076
0.469
-0.013
0.492

Median
3.10%
3.000
1.099
8.000
2.079
0.023
0.001
0.000
1.078
1.141
0.045
0.000

SD
4.637
0.359
0.146
9.220
0.767
0.943
11.092
150.981
0.138
12.389
0.220
0.500

Full Sample, 2000-2004


(N = 13,062)
Mean
4.21%
9.709
2.042
14.148
2.405
0.257
1.615
28.893
1.075
-0.873
0.005
0.056

Median
2.65%
8.000
2.079
1.000
2.398
0.024
0.001
0.000
1.076
1.060
0.060
0.000

SD
5.194
6.954
0.680
11.235
0.684
1.245
27.804
52.140
0.141
15.199
0.229
0.230

Panel B: Association between ABSTDA and low balling


Dependent variable: ABSTDA

Sample

Model 1:
Short Tenure (< 4 years) Sample, 2000-2004

Model 2:
Full Sample, 2000-2004

Coef.

t-stat

p-value

Coef.

t-stat

p-value

2.458
-0.260
-0.254
-0.170
-0.541
0.041
-0.001
-0.150
0.003
-5.633
Yes
Yes

3.2
-2.9
-1.3
-1.9
-5.4
3.7
-2.8
-0.4
1.2
-26.9

(0.00)
(0.00)
(0.20)
(0.06)
(0.00)
(0.00)
(0.00)
(0.66)
(0.24)
(0.00)

Intercept
Tenure
LowBall
Age
Size
Size2
Size3
IndGrow
Grow
CFO
YearDum
IndustryDum

2.912

1.0

(0.32)

-0.208
-0.415
-1.808
0.337
-0.015
-0.852
0.043
-2.199
Yes
Yes

-0.8
-2.0
-2.4
1.9
-1.6
-0.7
3.9
-3.2

(0.45)
(0.05)
(0.02)
(0.06)
(0.12)
(0.48)
(0.00)
(0.00)

Pseudo R2
N

0.106
1087

0.144
13062

46

Table 4: Tests for Low Balling (continued)

Panel C: Association between ABSTDA and auditor tenure, partitioned by auditor industry
specialization (controlling for low balling)
Dependent variable: ABSTDA

Coef.
Intercept
Tenure
Lowball
Age
Size
Size2
Size3
IndGrow
Grow
CFO
YearDum
IndustryDum

1.925
-0.308
-0.223
-0.133
-0.706
0.064
-0.001
0.092
0.003
-5.495
Yes
Yes

Pseudo R2
N

0.139
9335

Non-Specialists
t-stat

p-value

Coef.

1.8
-2.7
-0.9
-1.2
-4.8
3.8
-3.2
0.2
0.8
-22.3

(0.06)
(0.01)
(0.38)
(0.24)
(0.00)
(0.00)
(0.00)
(0.84)
(0.40)
(0.00)

3.523
-0.140
-0.312
-0.308
-0.430
0.037
-0.001
-0.610
0.002
-5.958
Yes
Yes

Specialists
t-stat

p-value

3.3
-1.0
-1.0
-2.1
-2.6
1.6
-1.3
-1.3
0.4
-14.6

(0.00)
(0.33)
(0.31)
(0.03)
(0.01)
(0.10)
(0.21)
(0.20)
(0.67)
(0.00)

0.162
3727

Diff in coef.
F-value
p value

6.75
(0.00)

Estimations are based on truncated regression approach. t values are reported in italics. p values (two-tailed) are reported in
parentheses.

The sample includes observations with audit fee data in the period 2000-2004.
Variable Definitions:
ABSTDA

Tenure

LowBall

Age

Size, Size2 & Size3


IndGrow

=
=

Grow
CFO
Specialist

=
=
=

YearDum
IndustryDum

=
=

absolute value of discretionary accruals scaled by average total assets (based on Ball
and Shivakumar 2006);
log of auditor tenure, measured as the number of consecutive years that the firm has
retained the auditor;
dummy variable, equals 1 for firms with auditor tenure below or equal to three
years and the audit fee is below one standard deviation from the predicted audit fee
(predicted using model discussed in Appendix A), 0 otherwise;
log of firm age, measured as the number of years that the firm appeared in
Compustat since 1950;
Market value of equity (in 10 billions) to the power one, two and three;
N

i =1

i =1

Salesi,t / Salesi,t 1 by Fama and Frenchs (1997) 48 industry groups;


firm-specific sales growth, measured as Salest / Salest-1;
firms cash flow from operations divided by average total assets;
dummy variable, equals 1 for firms audited by auditor whose market shares is
highest in terms of clients total asset for each industry group, 0 otherwise;
dummies for fiscal years;
dummies for Fama and Frenchs (1997) 48 industry groups.
47

Table 5: Tests of Endogeneity

Panel A: Partition analyses using two stage least squares


First-stage
(Dep. Var. = Tenure)

Second-stage
(Dep. Var. = ABSTDA)
Non-Specialists

Variable
Intercept
predTenure
Aturn
DA
Curr
Quick
ROA
ROA*LOSS
Export
Litigation
Lag(ABSTDA)
Age
Size
Size2
Size3
IndGrow
Grow
CFO
YearDum
IndustryDum

Coef.
0.207

t-stat
2.8

p-value
(0.01)

-0.001
0.011
0.040
0.002
0.027
-0.021
0.080
0.016
-0.004
0.684
0.051
-0.004
0.000
0.036
-0.001
0.056
Yes
Yes

-0.2
0.7
2.3
3.4
0.8
-0.6
1.5
1.3
-3.4
122.6
6.9
-4.6
3.9
1.6
-1.5
2.6

(0.82)
(0.51)
(0.02)
(0.00)
(0.44)
(0.54)
(0.14)
(0.20)
(0.00)
(0.00)
(0.00)
(0.00)
(0.00)
(0.12)
(0.13)
(0.01)

Adj. R2
(Partial R2)
N
F-stat
(Partial F-stat)

0.461
(0.460)
26756
305.82
(298.5)

Specialists

Coef.
2.796
-0.065

t-stat
2.980
-6.320

p-value
(0.00)
(0.00)

Coef.
2.814
-0.031

t-stat
2.740
-2.080

p-value
(0.01)
(0.04)

-0.489
-0.700
0.063
-0.001
-0.081
-0.002
-4.479
Yes
Yes

-7.980
-7.800
6.290
-5.230
-0.310
-0.180
-26.560

(0.00)
(0.00)
(0.00)
(0.00)
(0.76)
(0.86)
(0.00)

-0.390
-0.511
0.045
-0.001
-0.596
0.004
-5.274
Yes
Yes

-4.140
-4.210
2.590
-1.920
-1.660
0.270
-17.330

(0.00)
(0.00)
(0.01)
(0.05)
(0.10)
(0.79)
(0.00)

0.128

0.143

20062

6694

45.660

18.240

Diff in coef.
F-value
p value

23.46
(0.00)

Estimation(s) of the first-stage model is (second-stage models are) based on OLS approach (truncated regression approach). t
values are reported in italics. p values (two-tailed) are reported in parentheses.

48

Table 5: Tests of Endogeneity (continued)

Panel B: Test of interaction term using two stage least squares


First-stage
(Dep. Var. = Tenure)
Variable
Intercept
predTenure
predSpecialist
predTenure*
predSpecialist
Aturn
DA
Curr
Quick
ROA
ROA*LOSS
Export
Litigation
Lag(ABSTDA)
Age
Size
Size2
Size3
IndGrow
Grow
CFO
Age*
predSpecialist
Size*
predSpecialist
Size2*
predSpecialist
Size3*
predSpecialist
IndGrow*
predSpecialist
Grow*
predSpecialist
CFO*
predSpecialist
YearDum
IndustryDum
Adj. R2
(Partial R2)
N
F-stat
(Partial F-stat)

Coef.
0.207

-0.001
0.011
0.040
0.002
0.027
-0.021
0.080
0.016
-0.004
0.684
0.051
-0.004
0.000
0.036
-0.001
0.056

t-stat
2.8

-0.2
0.7
2.3
3.4
0.8
-0.6
1.5
1.3
-3.4
122.6
6.9
-4.6
3.9
1.6
-1.5
2.6

p-value
(0.01)

(0.82)
(0.51)
(0.02)
(0.00)
(0.44)
(0.54)
(0.14)
(0.20)
(0.00)
(0.00)
(0.00)
(0.00)
(0.00)
(0.12)
(0.13)
(0.01)

First-stage
(Dep. Var. = Specialist)
Coef.
0.509

-0.022
0.046
-0.004
0.000
0.039
-0.022
-0.033
0.020
-0.004
-0.016
0.066
-0.005
0.000
-0.010
0.001
0.024

Yes
Yes

Yes
Yes

0.461
(0.460)
26756
305.82
(298.5)

0.026
(0.025)
26756
10.420
(6.03)

t-stat
7.0

-5.3
2.7
-0.3
0.7
1.2
-0.7
-0.6
1.7
-3.8
-3.0
9.2
-6.3
4.8
-0.4
1.2
1.2

p-value
(0.00)

(0.00)
(0.01)
(0.79)
(0.50)
(0.25)
(0.51)
(0.53)
(0.09)
(0.00)
(0.00)
(0.00)
(0.00)
(0.00)
(0.67)
(0.22)
(0.25)

Second-stage
(Dep. Var. = ABSTDA)
Coef.
2.796
-0.065
0.018

t-stat
3.1
-6.5
0.0

p-value
(0.00)
(0.00)
(0.99)

0.033

1.7

(0.08)

-0.489
-0.700
0.063
-0.001
-0.081
-0.002
-4.479

-8.2
-8.0
6.5
-5.4
-0.3
-0.2
-27.2

(0.00)
(0.00)
(0.00)
(0.00)
(0.75)
(0.86)
(0.00)

0.099

0.8

(0.40)

0.188

1.2

(0.23)

-0.018

-0.8

(0.40)

0.000

0.2

(0.86)

-0.515

-1.1

(0.27)

0.006

0.3

(0.76)

-0.795

-2.2

(0.03)

Yes
Yes

0.133
26756
31.980

Estimations of the first-stage models are (second-stage model is) based on OLS approach (truncated regression approach). t
values are reported in italics. p values (two-tailed) are reported in parentheses.

49

Table 5: Tests of Endogeneity (continued)


Variable Definitions:
ABSTDA

= absolute value of discretionary accruals scaled by average total assets (based on Ball
and Shivakumar 2006);
Tenure
= log of auditor tenure, measured as the number of consecutive years that the firm has
retained the auditor;
Specialist
= dummy variable, equals 1 for firms audited by auditor whose market shares is highest
in terms of clients total asset for each industry group, 0 otherwise;
Aturn
= asset turnover, measured as sales divided by total assets;
DA
= debt-asset ratio, measured as long-term debt divided by total assets;
Curr
= current ratio, measured as current assets divided by total assets;
Quick
= quick ratio, measured as current assets minus inventory divided by current liabilities;
ROA
= earnings before interest and taxes divided by total assets;
Loss
= dummy variable, 1 if the firm incurred a loss in the previous year, 0 otherwise;
Export
= foreign sales divided by total sales;
Age
= log of firm age, measured as the number of years that the firm appeared in Compustat
since 1950;
Litigation
= dummy variable, 1 of the firm operates in a high-litigation industry, and 0 otherwise.
High-litigation industries are industries with SIC codes of 2833-2836, 3570-3577,
3600-3674, 5200-5961, and 7370-7370 (Ashbaugh et al. 2003);
Lag(ABSTDA)
= lag values of ABSTDA;
Age
= log of firm age, measured as the number of years that the firm appeared in Compustat
since 1950;
Size, Size2 & Size3 = Market value of equity (in 10 billions) to the power one, two and three;
N
IndGrow
= N
Salesi ,t / Salesi ,t 1 by Fama and Frenchs (1997) 48 industry groups;

i =1

Grow
CFO
YearDum
IndustryDum

=
=
=
=

i =1

firm-specific sales growth, measured as Salest / Salest-1;


firms cash flow from operations divided by average total assets;
dummies for fiscal years;
dummies for Fama and Frenchs (1997) 48 industry groups.

50

Table 6: Tests on Signed Measures of Discretionary Accruals

Panel A: POSTDA and NEGTDA


Dependent variable:

POSTDA

[NEGTDA*-1]

All

NonSpecialists

Specialists

All

NonSpecialists

Specialists

YearDum
IndustryDum

2.462
2.9
(0.00)
-0.338
-4.8
(0.00)
-0.201
-2.5
(0.01)
-0.513
-6.2
(0.00)
0.044
4.8
(0.00)
-0.001
-3.8
(0.00)
-0.057
-0.2
(0.83)
0.011
4.8
(0.00)
-4.808
-27.0
(0.00)
Yes
Yes

2.195
2.1
(0.04)
-0.451
-5.2
(0.00)
-0.103
-1.1
(0.29)
-0.635
-6.0
(0.00)
0.058
4.9
(0.00)
-0.001
-4.2
(0.00)
-0.055
-0.2
(0.86)
0.015
5.6
(0.00)
-4.609
-22.4
(0.00)
Yes
Yes

2.983
2.2
(0.03)
-0.094
-0.8
(0.45)
-0.456
-3.3
(0.00)
-0.297
-1.9
(0.06)
0.026
1.1
(0.25)
-0.001
-0.8
(0.44)
-0.031
-0.1
(0.94)
-0.001
-0.1
(0.90)
-5.287
-14.7
(0.00)
Yes
Yes

2.484
2.9
(0.00)
-0.270
-3.7
(0.00)
-0.193
-2.4
(0.02)
-0.897
-8.9
(0.00)
0.095
6.2
(0.00)
-0.002
-4.9
(0.00)
-0.060
-0.2
(0.83)
0.000
-0.1
(0.92)
-4.635
-23.9
(0.00)
Yes
Yes

1.922
1.7
(0.09)
-0.340
-3.8
(0.00)
-0.135
-1.4
(0.17)
-1.137
-8.0
(0.00)
0.145
5.8
(0.00)
-0.004
-4.5
(0.00)
0.204
0.6
(0.54)
-0.004
-1.3
(0.19)
-4.555
-20.3
(0.00)
Yes
Yes

3.586
3.0
(0.00)
-0.214
-1.7
(0.09)
-0.221
-1.6
(0.12)
-0.648
-4.3
(0.00)
0.061
2.6
(0.01)
-0.001
-2.0
(0.05)
-0.791
-1.7
(0.09)
0.009
2.0
(0.04)
-4.736
-11.9
(0.00)
Yes
Yes

Pseudo R2
N

0.143
16865

0.144
12495

0.154
4370

0.128
15912

0.123
11938

0.152
3974

Intercept

Tenure

Age

Size

Size2

Size3

IndGrow

Grow

CFO

Diff in coef.
F-value
p value

14.4
(0.00)

9.1
(0.00)

Estimations are based on truncated regression approach. t values are reported in italics. p values (two-tailed) are reported in
parentheses.

51

Table 6: Tests on Signed Measures of Discretionary Accruals

Panel B: NEGTDA: Extreme pre-managed earnings versus non-extreme pre-managed


earnings
Dependent variable: [NEGTDA*-1]
Sample

Observations with
Extreme Pre-managed Earnings

Observations with
Non-Extreme Pre-managed Earnings

All

NonSpecialists

Specialists

All

NonSpecialists

Specialists

YearDum
IndustryDum

1.307
0.8
(0.45)
-0.429
-2.7
(0.01)
-0.196
-1.1
(0.26)
-2.602
-8.5
(0.00)
0.414
6.2
(0.00)
-0.016
-5.2
(0.00)
0.083
0.1
(0.90)
-0.002
-0.3
(0.74)
-5.157
-14.0
(0.00)
Yes
Yes

0.826
0.4
(0.70)
-0.547
-2.8
(0.01)
-0.096
-0.5
(0.65)
-2.883
-7.3
(0.00)
0.455
5.3
(0.00)
-0.017
-4.5
(0.00)
0.471
0.6
(0.54)
-0.007
-1.2
(0.21)
-5.107
-12.2
(0.00)
Yes
Yes

2.444
0.8
(0.41)
-0.487
-1.7
(0.09)
-0.185
-0.6
(0.56)
-2.823
-4.7
(0.00)
0.581
3.4
(0.00)
-0.032
-2.8
(0.01)
-1.342
-1.1
(0.27)
0.016
1.8
(0.08)
-5.097
-6.4
(0.00)
Yes
Yes

2.451
9.6
(0.00)
-0.019
-0.9
(0.39)
-0.072
-2.8
(0.00)
-0.074
-2.4
(0.01)
0.009
2.1
(0.03)
0.000
-2.0
(0.05)
0.037
0.4
(0.67)
0.001
1.9
(0.06)
-1.207
-16.8
(0.00)
Yes
Yes

2.595
8.0
(0.00)
-0.031
-1.1
(0.25)
-0.047
-1.5
(0.12)
-0.130
-2.6
(0.01)
0.022
1.8
(0.08)
-0.001
-1.4
(0.17)
0.063
0.6
(0.53)
0.002
1.9
(0.05)
-1.206
-14.9
(0.00)
Yes
Yes

2.490
5.8
(0.00)
0.001
0.0
(0.99)
-0.147
-2.9
(0.00)
-0.023
-0.5
(0.64)
0.004
0.7
(0.51)
0.000
-0.8
(0.41)
-0.141
-0.8
(0.44)
0.000
-0.1
(0.90)
-1.205
-7.5
(0.00)
Yes
Yes

Pseudo R2
N

0.192
6364

0.177
4837

0.249
1527

0.082
9548

0.082
7101

0.093
2447

Intercept

Tenure

Age

Size

Size2

Size3

IndGrow

Grow

CFO

Diff in coef.
F-value
p value

5.35
(0.01)

0.65
(0.52)

Estimations are based on truncated regression approach. t values are reported in italics. p values (two-tailed) are reported in
parentheses.

52

Table 6: Tests on Signed Measures of Discretionary Accruals (continued)

Pre-managed earnings is computed as the difference between earnings and NEDGTDA. It is considered
extreme when pre-managed earnings is either in the top or bottom quintiles.
Variable Definitions:
POSTDA

NEGTDA

Tenure

Age

Size, Size2 & Size3


IndGrow

=
=

Grow
CFO
Specialist

=
=
=

YearDum
IndustryDum

=
=

positive value of discretionary accruals scaled by average total assets (based on Ball
and Shivakumar 2006);
negative value of discretionary accruals scaled by average total assets (based on Ball
and Shivakumar 2006);
log of auditor tenure, measured as the number of consecutive years that the firm has
retained the auditor;
log of firm age, measured as the number of years that the firm appeared in
Compustat since 1950;
market value of equity (in 10 billions) to the power one, two and three;
N

i =1

i =1

Salesi,t / Salesi,t 1 by Fama and Frenchs (1997) 48 industry groups;


firm-specific sales growth, measured as Salest / Salest-1;
firms cash flow from operations divided by average total assets;
dummy variable, equals 1 for firms audited by auditor whose market shares is
highest in terms of clients total asset for each industry group, 0 otherwise;
dummies for fiscal years;
dummies for Fama and Frenchs (1997) 48 industry groups.

53

Table 7: Alternative Definitions for Auditor Tenure


Dependent variable: ABSTDA
Tenure variable

Intercept

MEDIUM

LONG

MEDIUM and LONG

TenDum (below 4 vs. above 9 years)

All

NonSpecialists

Specialists

All

NonSpecialists

Specialists

2.534
4.1
(0.00)
-0.306
-2.7
(0.01)
-0.530
-4.5
(0.00)

2.241
2.8
(0.00)
-0.459
-3.3
(0.00)
-0.688
-4.7
(0.00)

3.208
3.5
(0.00)
-0.073
-0.4
(0.70)
-0.309
-1.5
(0.12)

3.787
4.1
(0.00)

3.663
3.1
(0.00)

4.456
3.1
(0.00)

-0.621
-4.1
(0.00)
-0.409
-3.3
(0.00)
-0.660
-7.4
(0.00)
0.055
5.7
(0.00)
-0.001
-4.6
(0.00)
-0.011
0.0
(0.97)
0.011
3.1
(0.00)
-4.720
-17.4
(0.00)
Yes
Yes

-0.351
-1.7
(0.10)
-0.523
-3.3
(0.00)
-0.395
-3.3
(0.00)
0.036
2.2
(0.03)
-0.001
-1.7
(0.09)
-0.478
-0.9
(0.38)
0.009
1.9
(0.06)
-5.672
-12.2
(0.00)
Yes
Yes

0.127
10229

0.135
3494

YearDum
IndustryDum

-0.304
-5.7
(0.00)
-0.616
-10.3
(0.00)
0.051
7.5
(0.00)
-0.001
-5.8
(0.00)
-0.100
-0.5
(0.60)
0.005
3.2
(0.00)
-4.726
-36.0
(0.00)
Yes
Yes

-0.289
-4.5
(0.00)
-0.735
-9.4
(0.00)
0.065
7.3
(0.00)
-0.001
-6.0
(0.00)
0.044
0.2
(0.85)
0.005
2.8
(0.01)
-4.596
-30.3
(0.00)
Yes
Yes

-0.348
-3.7
(0.00)
-0.481
-4.6
(0.00)
0.044
2.9
(0.00)
-0.001
-2.2
(0.03)
-0.435
-1.4
(0.17)
0.004
1.4
(0.16)
-5.037
-19.0
(0.00)
Yes
Yes

-0.451
-3.7
(0.00)
-0.437
-4.5
(0.00)
-0.532
-7.9
(0.00)
0.041
5.7
(0.00)
-0.001
-4.3
(0.00)
-0.103
-0.4
(0.72)
0.010
3.6
(0.00)
-4.970
-21.2
(0.00)
Yes
Yes

Pseudo R2
N

0.135
32777

0.131
24433

0.151
8344

0.126
13723

TenDum

Age

Size

Size2

Size3

IndGrow

Grow

CFO

Diff in coef.
F-value
p value
F-value
p value

MEDIUM
LONG

5.97
(0.00)
12.84
(0.00)

TenDum

54

10.1
(0.00)

Table 7: Alternative Definitions for Auditor Tenure (continued)


Estimations are based on truncated regression approach. t values are reported in italics. p values (two-tailed) are reported in
parentheses.

Variable Definitions:
ABSTDA

MEDIUM

LONG

TenDum

Age

Size, Size2 & Size3


IndGrow

=
=

absolute value of discretionary accruals scaled by average total assets (based on Ball
and Shivakumar 2006);
dummy variable, equals 1 when the length of the auditor-client relationship is
medium (four to eight years); zero otherwise;
dummy variable, equals 1 when the length of the auditor-client relationship is long
(nine years or longer); zero otherwise;
dummy variable, equals 1 for auditor-client relationship is nine years or longer, 0 for
auditor-client relationship is three years or below;
log of firm age, measured as the number of years that the firm appeared in
Compustat since 1950;
market value of equity (in 10 billions) to the power one, two and three;
N

i =1

Grow
CFO
Specialist

=
=
=

YearDum
IndustryDum

=
=

Sales / Sales
i ,t

i =1

i ,t 1

by Fama and Frenchs (1997) 48 industry groups;

firm-specific sales growth, measured as Salest / Salest-1;


firms cash flow from operations divided by average total assets;
dummy variable, equals 1 for firms audited by auditor whose market shares is
highest in terms of clients total asset for each industry group, 0 otherwise;
dummies for fiscal years;
dummies for Fama and Frenchs (1997) 48 industry groups.

55

Table 8: Earnings Benchmark Tests


Dependent variable:

INCREASE

SURPRISE

All

NonSpecialists

Specialists

All

NonSpecialists

Specialists

YearDum
IndustryDum

-2.231
1023.0
(0.00)
-0.056
4.4
(0.04)
0.172
23.6
(0.00)
0.012
40.6
(0.00)
0.219
588.0
(0.00)
0.215
10.2
(0.00)
-1.538
1012.5
(0.00)
0.012
11.0
(0.00)
Yes
Yes

-2.354
800.0
(0.00)
-0.057
3.1
(0.08)
0.169
16.7
(0.00)
0.012
31.5
(0.00)
0.244
498.6
(0.00)
0.174
4.8
(0.03)
-1.481
708.7
(0.00)
0.010
6.3
(0.01)
Yes
Yes

-1.924
219.9
(0.00)
-0.055
1.3
(0.25)
0.173
6.2
(0.01)
0.010
8.3
(0.00)
0.165
99.8
(0.00)
0.279
4.9
(0.03)
-1.709
302.1
(0.00)
0.017
5.2
(0.02)
Yes
Yes

-1.820
541.5
(0.00)
-0.059
4.1
(0.04)
0.251
47.8
(0.00)
0.003
2.4
(0.13)
0.116
126.5
(0.00)
0.529
52.8
(0.00)
-0.639
219.4
(0.00)
0.014
14.9
(0.00)
Yes
Yes

-1.830
384.2
(0.00)
-0.068
3.6
(0.06)
0.256
36.4
(0.00)
0.004
2.2
(0.14)
0.122
94.6
(0.00)
0.510
34.6
(0.00)
-0.628
153.9
(0.00)
0.015
13.0
(0.00)
Yes
Yes

-1.793
153.3
(0.00)
-0.045
0.7
(0.39)
0.233
11.0
(0.00)
0.002
0.3
(0.61)
0.105
32.2
(0.00)
0.566
17.8
(0.00)
-0.668
65.2
(0.00)
0.011
2.2
(0.14)
Yes
Yes

Pseudo R2
N

0.177
25868

0.179
19196

0.175
6672

0.060
19493

0.060
14205

0.060
5288

Intercept

Tenure

Litigation

MB

LnMVE

IH

Loss

TDA

Diff in coef.
F-value
p value

3.66
(0.03)

12.3
(0.02)

Logistic estimations are conducted. Chi-squares are reported in italics. p values are reported in parentheses.

56

Table 8: Earnings Benchmark Tests (continued)


Variable Definitions:
INCREASE

SURPRISE

Tenure

Litigation

MB

LnMVE
IH
Loss
TDA
Specialist

=
=
=
=
=

YearDum
IndustryDum

=
=

dummy variable, 1 when the increase in net income, scaled by average total assets,
falls in the interval [0.00, 0.02], and 0 otherwise;
dummy variable, 1 when a firm meets or beats by 1 cent the mean consensus analysts
forecast as reported in IBES database, and 0 otherwise;
log of auditor tenure, measured as the number of consecutive years that the firm has
retained the auditor;
dummy variable, 1 if the firm operates in a high-litigation industry (industries with
SIC codes of 2833-2836, 3570-3577, 3600-3674, 5200-5961 and 7370-7370), and 0
otherwise;
market to book ratio, defined as market value of equity divided by stockholders
equity of common shareholders;
log of market value of equity;
percentage of shares held by institutional owners;
dummy variable, 1 if the firm incurred a loss in the previous year, 0 otherwise;
discretionary accruals scaled by average total assets;
dummy variable, equals 1 for firms audited by auditor whose market shares is highest
in terms of clients total asset for each industry group, 0 otherwise;
dummies for fiscal years;
dummies for Fama and Frenchs (1997) 48 industry groups.

57

Das könnte Ihnen auch gefallen