Earnings Autocorrelation, Earnings Volatility, and Audit Fees

David B. Bryan Assistant Professor of Accounting - The University of North Florida

Terry W. Mason Assistant Professor of Accounting - Kansas State University

J. Kenneth Reynolds * Associate Professor of Accounting - Florida State University

ABSTRACT: This study posits that earnings autocorrelation and earnings volatility are associated with audit fees. Earnings autocorrelation and earnings volatility represent time-series of earnings characteristics that may affect an auditor’s perception of inherent risk. In response to greater inherent risk, auditors should conduct more extensive substantive testing in order to reduce the overall risk associated with the audit. We find a negative (positive) association between earnings autocorrelation (volatility) and audit fees. The results indicate that a change in the interquartile range in earnings autocorrelation and earnings volatility combined is associated with a change in audit fees of approximately $305,500 for the average firm in our sample. We also find that the relation between earnings autocorrelation and audit fees is amplified for specialist auditors, consistent with specialist auditors responding to uncertainty in earnings more thoroughly than non-specialists.

Draft Date: November 2015. Please do not cite without permission of the authors.

Keywords: audit fees; auditor specialization; earnings autocorrelation; earnings volatility

* Corresponding Author: [email protected]

We thank Bruce Billings, Allen Blay, Landon Mauler, Robbie Moon, Rick Morton, Josette Pelzer, and the workshop participants at Florida State University for their helpful comments on prior versions of this study.

Earnings Autocorrelation, Earnings Volatility, and Audit Fees

Keywords: audit fees; auditor specialization; earnings autocorrelation; earnings volatility

I. INTRODUCTION

We propose that earnings autocorrelation and earnings volatility are important earnings characteristics impacting audit risk. Although these characteristics may constitute elements of

inherent risk for the auditor, the associations and directions are ambiguous for reasons discussed

later. As inherent risk increases, auditors should perform additional substantive testing in order

to reduce the overall risk associated with the audit. We anticipate that this additional effort will

drive audit fees higher. Using audit fees as a measure of audit effort, our findings are consistent

with lower (higher) earnings autocorrelation (volatility) increasing the auditor’s assessment of

inherent risk, leading to a highly economically significant effect on audit fees.

Auditing standards specify audit risk as the product of three components: inherent risk,

control risk, and detection risk (AICPA 2006; PCAOB 2010). Inherent risk represents the

auditor’s assessment of the risk of a material misstatement without considering a firm’s internal

control over financial reporting, control risk is the auditor’s assessment of the risk that a material

misstatement will not be prevented or detected by the firm’s system of internal control, and

detection risk is the risk that the auditor will not detect a material misstatement during the course

of the audit (AICPA 2006). The audit risk model implies that if an auditor assesses a higher

level of inherent risk or control risk, then more substantive testing must be conducted during the

audit in order to lower detection risk so that overall audit risk is reduced to an acceptable level.

Therefore, factors that cause auditors to assess a higher degree of inherent risk or control risk

should increase the amount of effort the auditor has to put forth during the audit.

Earnings autocorrelation is a time-series characteristic of earnings that reflects the degree

to which prior-period earnings are informative about current-period earnings. Prior research

suggests that seasonal differences in quarterly earnings are correlated, and that the

autocorrelation is approximately 0.34 to 0.44, on average (Bernard and Thomas 1990; Ball and

Bartov 1996). Earnings volatility reflects the dispersion in earnings over time, with higher

volatility representing wider fluctuations in earnings. Prior research finds that earnings have

become more volatile over time (Givoly and Hayn 2000; Dichev and Tang 2008), making

volatility an increasingly important earnings characteristic. In the presence of low

autocorrelation and/or high volatility, earnings are less certain and more difficult for auditors to

assess against a benchmark expectation, thus making them more difficult to audit. On the other

hand, those same earnings characteristics may imply a lower risk of managerial manipulation of earnings, which will reduce the auditor's assessed risk of fraud in the financial statements. These competing potential drivers of audit risk create an interesting framework for examining whether and how auditors incorporate these characteristics into their audit risk assessments. Factors viewed as increasing risk exposure will lead the rational auditor to expend additional effort to

manage that risk.

Greater (or lower) audit effort should be reflected in higher (or lower) audit fees. Prior

audit fee research has identified associations between audit fees and several variables that

capture inherent risk. For example, research has found that a higher proportion of inventory and

accounts receivable to total assets is associated with higher audit fees (Simunic 1980), with the

explanation for this association being that these accounts are more difficult to audit and require

greater auditor effort. In this study, we argue that time-series of earnings characteristics can also

affect assessments of inherent risk and influence the extent of auditor effort. Further, prior

research also finds that analysts, who are among the most sophisticated users of financial

information, make systematic errors in assessing volatile earnings (Dichev and Tang 2009),

suggesting that volatility makes earnings more difficult to understand and interpret. Similarly,

auditors may also experience greater difficulty assessing highly volatile earnings. Hence, in

addition to impacting inherent risk itself, weak earnings autocorrelation and/or a wide dispersion

in the earnings distribution may make assessing inherent risk more difficult, in turn requiring

greater audit effort.

We also examine the impact of auditor specialization on the association between these

earnings characteristics and audit fees. Prior research suggests that in-depth industry knowledge allows specialist auditors to provide higher quality audits than non-specialists (e.g., Balsam et al.

2003; Reichelt and Wang 2010). In-depth industry knowledge may enable specialists to respond to low earnings autocorrelation or high earnings volatility more efficiently than non-specialists, suggesting the possibility of a weaker association between earnings autocorrelation (volatility) and audit fees for specialist auditors. Alternatively, specialist auditors may respond to low earnings autocorrelation or high earnings volatility by auditing more thoroughly than non- specialists. This could occur if, relative to non-specialists, specialists are either (1) better able to identify the risks associated with low earnings autocorrelation or high earnings volatility, or (2) more concerned about protecting a reputation for providing high quality audits. These reasons imply a strengthening in the association between earnings autocorrelation (volatility) and audit fees.

The results of our study provide strong evidence of a negative (positive) association between earnings autocorrelation (volatility) and audit fees, consistent with our hypotheses. We also find that specialist auditors charge higher fees than non-specialists in response to low earnings autocorrelation, consistent with specialists reacting more thoroughly to uncertainty in

3 earnings. However, we do not find evidence that earnings volatility affects audit fees differently for specialist auditors.

This paper provides two primary contributions to the literature. First, we identify two economically significant factors that affect audit fees. Our results suggest that a change in the interquartile range in earnings autocorrelation and earnings volatility is associated with a change in audit fees of $253,700 and $51,800, respectively, for the average firm in our primary sample, for a combined impact of $305,500. By comparison, a change in the interquartile range in total accruals, return on assets, and audit report lag is associated with a change in audit fees of only

$14,700, $90,400, and $128,400, respectively, for the average firm in our primary sample. Thus, the combined impact of earnings autocorrelation and earnings volatility on audit fees is greater in magnitude than several previously established audit fee model variables. Therefore, we add to the growing body of literature that examines the determinants of audit fees.

We also contribute to the auditor specialization literature. We provide evidence that earnings autocorrelation has a greater impact on audit fees when the auditor is an industry specialist. This result is consistent with specialist auditors reacting to uncertainty in earnings more thoroughly than non-specialists. Thus, we add to the literature that documents benefits associated with auditor specialization.

The remainder of this paper is organized as follows. Section 2 develops the hypotheses, section 3 presents our empirical methodology, section 4 discusses the results of the study, and section 5 concludes.

II. DEVELOPMENT OF THE HYPOTHESES

Earnings Autocorrelation and Earnings Volatility, and Audit Fees

Autocorrelation and volatility both represent elements of uncertainty in earnings, yet are

distinct both conceptually and empirically. Earnings autocorrelation reflects the degree to which

prior-period earnings are predictive of current-period earnings. Firms with low earnings autocorrelation have prior-period earnings that provide little information about current-period

earnings, which does not necessarily imply that earnings are highly dispersed. That is, the same

firm can have both low earnings autocorrelation and low earnings volatility. For example,

38.0% of the firm-year observations in our primary sample have both below-average earnings autocorrelation and below-average earnings volatility. Likewise, earnings can be highly volatile, but have high autocorrelation if prior-period earnings have value in predicting current-period

earnings. To illustrate, we find that 12.2% of the firm-year observations in our primary sample

have both above-average earnings autocorrelation and above-average earnings volatility. We also find that earnings autocorrelation and earnings volatility have a correlation coefficient of only -0.05. Hence, either or both of these characteristics may constitute important elements of inherent risk for the auditor, but the associations and directions are ambiguous for reasons discussed later. If they are associated with the auditor's assessment of inherent risk, then we expect to see them reflected in audit fees.

A large body of literature, beginning with Simunic (1980), investigates the determinants of audit fees. An interesting question in this line of literature concerns how auditors respond to underlying risk drivers. Specifically regarding inherent risk assessments, research finds that auditors charge a higher fee in response to higher inherent risk, and that the higher fee is

attributable to a greater number of hours spent on the engagement (e.g., Simunic and Stein 1996;

Bell et al. 2001; Schelleman and Knechel 2010), or in other words, to greater audit effort.

Although all of the variables commonly encountered in the extant audit literature undoubtedly

indirectly capture some element of inherent risk, the levels of inventory and receivables represent

the primary direct measures of inherent risk (Simunic 1980; Hay et al. 2006). We suggest that

earnings autocorrelation and volatility are also directly related to inherent risk, and thus will

impact audit fees if the auditor perceives them as having an impact on risk.

The higher (lower) the absolute value of earnings autocorrelation, the more (less)

information prior-period earnings provides about current-period earnings. Early studies on the

time-series properties of earnings document that seasonal differences in quarterly earnings are

correlated (e.g., Bathke and Lorek 1984; Brown et al. 1987; Bernard and Thomas 1990). More

specifically, the literature shows that the average autocorrelation at the first lag is positive and

ranges between 0.34 and 0.44. Additionally, Bathke et al. (2014) find that approximately 5% of

the firms in their sample do not have statistically significant levels of first-order autocorrelation

in earnings. For these firms, prior-period earnings are uninformative about current-period

earnings. In summary, prior research suggests that (1) there is earnings autocorrelation on

average, and (2) the level of earnings autocorrelation can vary greatly, even to the point of being

statistically insignificant.

Low earnings autocorrelation implies more uncertainty in earnings. We hypothesize two

non-mutually exclusive explanations through which uncertainty in earnings could affect auditor

assessments of inherent risk. First, the uncertainty in earnings itself could be difficult for

auditors to interpret. When earnings are unpredictable from year-to-year, auditors are less able to benchmark current-year firm performance against prior-year performance. With a reduced

ability to use past earnings to judge the reasonableness of current earnings, an auditor has one

less analytical tool available to assess current-period earnings. Second, underlying business risks

faced by the client that tend to cause uncertainty in earnings may also cause an auditor to

perceive higher levels of inherent risk. For example, operating in a highly competitive

environment could lead to uncertainty in earnings, and could also cause an auditor to view

inherent risk as higher. Consequently, earnings autocorrelation could serve as a valuable proxy

to capture elements of the auditors risk exposure that may be difficult to directly measure

empirically, yet are reflected as an uncertain time-series of earnings.1 These factors all imply

higher risk exposure, and hence higher audit fees, when earnings autocorrelation is low.

Alternatively, a high degree of certainty in a time-series of earnings could reflect earnings

management. Prior research has provided evidence that earnings management is positively

associated with audit fees (e.g., Gul et al. 2003; Abbott et al. 2006; Knechel and Schelleman

2010; Krishnan et al. 2013).2 Intentional earnings manipulation is a direct causal factor in inherent risk. To the extent that greater certainty in earnings reflects managerial manipulation, we would expect to see auditors respond with heightened awareness and effort, and hence observe an increase in audit fees when autocorrelation is high rather than low.

Our rationale for predicting an association between audit fees and earnings volatility is similar to that for earnings autocorrelation. Volatile earnings may increase the inherent risk associated with an audit because more complex auditing issues are likely to arise during the

1 Note that this is similar to the intuition behind including audit report lag as a determinant in the audit fee model. Audit report lag captures problems or difficulties in an audit, which lead to a longer delay in filing the audit report and also lead to higher audit fees. Problems specific to each audit are impossible to directly empirically measure because that information is not available, however, audit report lag is a useful proxy for audit-specific problems. 2 Although we do not control for discretionary accruals in our main analysis, all of our results are robust to the inclusion of unsigned performance-adjusted discretionary accruals as a control variable.

audit. For example, as earnings volatility increases, auditors may have more difficulty with

issues associated with valuation, such as whether an adjustment for impairment expense is

necessary. In addition, prior research finds that analysts make systematic errors in evaluating

highly volatile earnings (Dichev and Tang 2009), and survey evidence suggests that one reason

executives view earnings volatility unfavorably is that they believe it makes assessing earnings

more difficult for analysts (Graham et al. 2005). Alternatively, low volatility in earnings could

reflect earnings management. Prior research has shown that smooth earnings are associated with

a lower cost of capital (Francis et al. 2004), providing managers with a strong incentive to

manipulate earnings volatility. Since earnings management has been shown to be associated

with higher audit fees, to the extent that higher volatility reflects less earnings manipulation, we would expect to observe a negative association between earnings volatility and audit fees.

Because of these competing effects for both earnings autocorrelation and earnings

volatility, we do not make directional predictions about their association with audit fees. In

addition, we do not view them as being mutually exclusive, and thus the competing effects could

potentially cancel out each other. Absent a perfect counterbalancing effect though, we should be

able to observe a broad effect with sufficient statistical power. Since autocorrelation and

volatility in earnings are conceptually and empirically distinct, the various theoretical factors

linking them with audit fees may weigh significantly differently in the assessment of inherent

risk. Hence, we express separate hypotheses for each, in alternative form as follows:

H1: Earnings autocorrelation is associated with audit fees.

H2: Earnings volatility is associated with audit fees.

Auditor Specialization

Prior research shows that auditor industry specialization is associated with higher quality audits (e.g., Balsam et al. 2003; Reichelt and Wang 2010) and audit fee premiums (e.g., Francis et al. 2005; Carson 2009; Fung et al. 2012), suggesting that specialist auditors are systematically

different from non-specialists. We investigate whether the association between earnings

autocorrelation (volatility) and audit fees is different depending on whether the auditor is an

industry specialist.

Prior research suggests that in-depth industry knowledge allows specialist auditors to

provide higher quality audits than non-specialists (e.g., Balsam et al. 2003; Reichelt and Wang

2010). Also, Willenborg (2002) suggests that auditors choose to specialize in an industry

because it allows them to audit both more effectively and more efficiently. Accordingly, greater

knowledge of industry characteristics may allow specialist auditors to respond to low earnings

autocorrelation or high earnings volatility more efficiently than non-specialist auditors. If less

substantive testing is required for specialist auditors to respond to low earnings autocorrelation

or high earnings volatility than non-specialists, then we expect a weakening in the association

between earnings autocorrelation (volatility) and audit fees for specialists.

Alternatively, greater industry knowledge may allow specialist auditors to better identify

the audit risks associated with low earnings autocorrelation or high earnings volatility than non-

specialists. If specialists are better able to recognize these risks, then they may audit to a more

appropriate extent than non-specialists, implying greater effort in response to earnings

autocorrelation and earnings volatility for specialist auditors. In addition, because specialization

is associated with an audit fee premium (e.g., Francis et al. 2005; Carson 2009; Fung et al. 2012),

specialist auditors have incentives to protect a reputation for providing high quality audits. If

specialist auditors are more concerned about reputation protection than non-specialists, they may

respond to the risk posed by earnings autocorrelation or earnings volatility by auditing more

vigilantly than non-specialists, leading to greater auditor effort. These reasons imply a

strengthening in the association between earnings autocorrelation (volatility) and audit fees for

specialist auditors.

Given these offsetting effects, we do not make a directional prediction about the impact

of auditor specialization on the association between earnings autocorrelation (volatility) and audit fees. This discussion leads to our third and fourth hypotheses, stated in alternative form.

H3: Auditor industry specialization affects the association between earnings

autocorrelation and audit fees.

H4: Auditor industry specialization affects the association between earnings volatility

and audit fees.

III. METHODOLOGY

Sample

We form our sample from the intersection of the Compustat, CRSP, and Audit Analytics databases. We exclude financial firms and utilities (SIC codes 6000-6999 and 4900-4999) as well as foreign firms. Consistent with several recent audit fee model studies, we also exclude observations with non-Big 4 auditors in order to make our sample more homogenous (e.g.,

Krishnan et al. 2013; Fung et al. 2012; Francis et al. 2005).3 Lastly, we delete observations that

are missing required data. The sample period begins in 2004 and extends through 2012. Our

primary sample includes 10,510 firm-year observations from 1,994 unique firms.

In testing our auditor specialization hypotheses, we follow the sample selection procedure

from Fung et al. (2012). Consistent with recent auditor specialization research, we measure

specialization at the auditor city-level (e.g., Francis et al. 2005; Reichelt and Wang 2010; Fung et

al. 2012). We eliminate city-industry-year combinations if (1) there is only one observation in

the industry (defined using 2-digit SIC code) in the auditor’s city in that year, or (2) there is only one Big 4 auditor in the city in that year. This data requirement eliminates 3,356 firm-year

observations,4 leaving us with 7,154 firm-year observations from 1,559 unique firms in our

specialization sample. Panel A of Table 1 provides summary statistics and Panel B of Table 1 provides sample attrition information for our primary and specialization samples. Table 2 provides Pearson correlation matrices for our samples.

Calculation of Earnings Autocorrelation

In calculating earnings autocorrelation, we utilize all earnings (net income before extraordinary items) data that is available from the 48 quarters preceding the firm-year. To ensure reasonably precise measurement of autocorrelation, we exclude a firm-year if there are less than 36 prior quarters of earnings information available. We calculate earnings

3 Our inferences are unchanged when including non-Big 4 auditors and controlling for Big 4 auditors using an indicator variable. 4 Fung et al. (2012) report having 25,287 observations remaining after applying all of their data screens except for the specialization data screen. They then lose 8,080 observations from the specialization data screen, which is approximately 32% of their sample (8,080/25,287). Similarly, we lose 31.9% of our observations from the specialization data screen (3,356/10,510).

autocorrelation on a rolling basis for each firm-year. That is, for each firm, the earnings autocorrelation in any given year is based on the preceding 48 quarters of earnings information

for that firm. We estimate earnings autocorrelation as the correlation between the seasonally-

differenced quarterly earnings series at the first lag.

For instance, to find the autocorrelation for firm j in 2001, assuming firm j has complete

data, we first take earnings for each quarter and subtract earnings from four quarters ago for each

firm-quarter from 1989 through 2000 to arrive at seasonally adjusted earnings. We then divide

the seasonally adjusted earnings by the standard deviation of the earnings series to arrive at

standardized unexpected earnings (SUE). Finally, we estimate the correlation between the time-

series of SUEs in the current period and the time-series of SUEs in the preceding period. To

estimate the autocorrelation for firm j in year 2002, we would repeat this procedure using

earnings from 1990 through 2001. Additionally, since our predictions are based on the

magnitude of earnings autocorrelation, and not whether the autocorrelation is positive or

negative, we use the absolute value of earnings autocorrelation in our models.

Autocorrelation and Earnings Volatility

One question in our study is whether earnings autocorrelation and earnings volatility are

inversely capturing the same construct. In other words, does low earnings autocorrelation

always imply high earnings volatility (and vice versa)? The earnings autocorrelation coefficient

can range from -1 to +1. A coefficient of +1 would be consistent with a percentage change in prior-period earnings leading to the exact same percentage change in current-period earnings; whereas, a coefficient of -1 would be consistent with a percentage change in prior-

period earnings leading to the exact same percentage change in current-period earnings, just in the opposite direction. A coefficient near zero would suggest that the percentage change in prior-

period earnings does not carry over to the current-period. Earnings volatility, however, is simply

the dispersion of earnings for a given firm over time. Therefore, highly dispersed earnings result

in high earnings volatility while relatively stable earnings result in low earnings volatility. This

implies that firms with high earnings volatility do not necessarily have low autocorrelation, and

vice-versa. That is, a firm might have highly correlated earnings changes even though earnings

are highly dispersed. Additionally, a firm with uncorrelated earnings changes could have

relatively stable earnings.

The following evidence from our sample is consistent with the notion that earnings

autocorrelation and earnings volatility are different: 38.0% (12.2%) of the firm-year observations in our primary sample have both below-average (above-average) earnings autocorrelation and below-average (above-average) earnings volatility. Additionally, earnings autocorrelation and earnings volatility have a correlation coefficient of only -0.05. This evidence provides support that these variables are distinct.

Empirical Models

Our empirical model is based on audit fee model variables explored in the prior literature

(e.g., Simunic 1980; Whisenant et al. 2003; Francis et al. 2005; Antle et al. 2006; Hay et al.

2006; Krishnan et al. 2011; Ball et al. 2012; Krishnan et al. 2013). Since earnings

autocorrelation and earnings volatility are likely to be substantially affected by industry, we use

industry fixed effects and define industry by 2-digit SIC code.5 We also use year fixed effects to

control for differences in time periods. We cluster standard errors by firm. To minimize the

influence of extreme observations, we winsorize all continuous variables at the 1st and 99th

percentiles. We test the first two hypotheses by estimating the following model, using the

10,510 observations in our primary sample.

LNFEES = α + β1AUTO + β2IBVOL + β3LNAT + β4ROA + β5ACCR + β6CURR + β7FSALES

+ β8SEG + β9LEV + β10LOSS + β11DEC + β12LAG + β13TENURE + β14GC + β15XIDUM +

β16ACQ + β17NYSE + β18NEWDEBT + β19NEWEQ + β20LIT + β21MTB + β22LNNAF +

β23RETVOL + β24RETSUM + β25MATWEAK + βiINDUSTRY + βjYEAR + ε (1)

Where:

LNFEES = natural logarithm of total audit fees

AUTO = the absolute value of earnings autocorrelation, estimated as the correlation

between the seasonally-differenced quarterly earnings series at the first lag, using

a 48 quarter estimation window, as described previously

IBVOL = the standard deviation of annual earnings before extraordinary items scaled by

total assets at the beginning of the year, measured from years t-4 through t

LNAT = natural logarithm of total assets (in millions)

5 Additionally, in order to use the most refined definition of industry available, we also try using industry fixed effects with industry defined by 4-digit SIC code. We find that our results are robust to this alternative industry specification.

ROA = return on assets, measured as net income scaled by the average of total assets at

the beginning and end of the fiscal year

ACCR = the absolute value of total accruals scaled by total assets

CURR = current assets divided by total assets

FSALES = foreign sales divided by total sales

SEG = number of business segments

LEV = total liabilities scaled by total assets

LOSS = an indicator variable that takes the value of 1 if earnings before extraordinary

items is negative and 0 otherwise

DEC = an indicator variable that takes the value of 1 if a firm has a December fiscal year-

end and 0 otherwise

LAG = number of days between the firm’s fiscal year-end and the issuance of the audit

report

TENURE = the length of the auditor-client relationship in years

GC = an indicator variable that takes the value of 1 if a firm received a going-concern

audit opinion during the year and 0 otherwise

XIDUM = an indicator variable that takes the value of 1 if extraordinary items is non-zero

and 0 otherwise

ACQ = an indicator variable that takes the value of 1 if a firm engages in merger or

acquisition activity during the year and 0 otherwise

NYSE = an indicator variable that takes the value of 1 if a firm is listed on the New York

Stock Exchange and 0 otherwise

NEWDEBT = an indicator variable that takes the value of 1 if a firm issues long-term debt

during the year and 0 otherwise

NEWEQ = an indicator variable that takes the value of 1 if a firm issues common stock in

excess of five percent of its market value of equity during the year and 0

otherwise

LIT = an indicator variable that takes the value of 1 if a firm is in a high litigation risk

industry (SIC codes 2833-2836, 3570-3577, 3600-3674, 5200-5961, and 7370-

7374) and 0 otherwise

MTB = market value of equity divided by book value of equity

LNNAF = the natural logarithm of non-audit fees

RETVOL = the standard deviation of monthly stock returns during the fiscal year

RETSUM = the sum of monthly stock returns during the fiscal year

MATWEAK= an indicator variable that takes the value of 1 if a firm reports a material weakness

in internal control over financial reporting and 0 otherwise

INDUSTRY = indicator variables for each 2-digit SIC code in the sample

YEAR = indicator variables for each year in the sample

In testing the first and second hypotheses, we are interested in the coefficients on AUTO

and IBVOL. In accordance with our hypotheses, we do not make directional predictions about the coefficients.

To test the third and fourth hypotheses, we follow Fung et al. (2012) in calculating auditor industry specialization. We measure specialization at the auditor city-level, with industry

defined using 2-digit SIC code. If an auditor has the highest market share (in terms of total audit

fees) in an industry in its city, then we consider the auditor to be an industry specialist for that particular year. In our regression, we utilize an indicator variable, SPEC, which takes the value

of 1 if a firm in a given industry is audited by an auditor that specializes in that industry and 0 otherwise. Panel A of Table 1 shows that 56% of our sample observations are audited by a

specialist auditor.

Since the association between audit fees and each of the control variables can vary based

on the firm being audited by a specialist auditor, we utilize a fully interacted model whereby all

of our test variables and control variables are interacted with SPEC, except for the year and

industry indicator variables.6 A fully interacted model allows the slope of each control variable

to vary based on there being a specialist auditor, which reduces the risk that our interactions of interest are picking up effects that are actually attributable to systematic differences in the association between the control variables and audit fees that occur when a company is audited by

6 Our results are robust to using a model that is not fully interacted (i.e., a model where auditor specialization is only interacted with earnings autocorrelation and earnings volatility.

17 a specialist auditor. This approach provides a stronger test of our hypotheses. We estimate the following model to test our third and fourth hypotheses, using the 7,154 observations in our specialization sample.

LNFEES = α + β1AUTO + β2IBVOL + β3SPEC*AUTO + β4SPEC*IBVOL + β5SPEC +

β6LNAT + β7ROA + β8ACCR + β9CURR + β10FSALES + β11SEG + β12LEV + β13LOSS +

β14DEC + β15LAG + β16TENURE + β17GC + β18XIDUM + β19ACQ + β20NYSE +

β21NEWDEBT + β22NEWEQ + β23LIT + β24MTB + β25LNNAF + β26RETVOL + β27RETSUM +

β28MATWEAK + ∑β29-51SPEC*CONTROL + βiINDUSTRY + βjYEAR + ε (2)

Where:

SPEC = an indicator variable that takes the value of 1 if a firm in a given industry is

audited by an auditor that specializes in that industry (as discussed previously)

and 0 otherwise

All other variables are as previously defined.

In testing the third and fourth hypotheses, we are interested in the coefficients on the interactions SPEC*AUTO and SPEC*IBVOL. In accordance with our hypotheses, we do not make directional predictions about these coefficients.

IV. RESULTS

Earnings Autocorrelation and Earnings Volatility

Table 3 presents the results of estimating the audit fee model presented in equation (1)

using our primary sample of 10,510 firm-year observations. The first column shows the results from estimating the model while excluding our test variables. Consistent with prior research, the model explains a very high proportion of the variation in audit fees, with an R2 of 81.35%.

Additionally, all of our statistically significant control variables load in the expected direction.

Next, we add earnings autocorrelation and earnings volatility into the model. In order to

further assess whether AUTO and IBVOL are capturing different constructs, we add each

variable into the regression separately in columns 2 and 3, and then combined in column 4 of

Table 3. The results for AUTO (IBVOL) when in the model without IBVOL (AUTO) are

similar to when both variables are in the model together, which provides even further evidence

that these variables are distinct. The results reveal a negative and highly significant coefficient

on AUTO (t = -6.77), and a positive and highly significant coefficient on IBVOL (t = 3.08).

These results indicate that auditors respond to low earnings autocorrelation or high earnings

volatility by increasing audit fees, which provides strong support for our first two hypotheses.

We also estimate standardized regression coefficients in order to assess the economic

significance of each variable. The results show that a one standard deviation (interquartile

range) change in earnings autocorrelation and earnings volatility is associated with an change in

audit fees of $158,700 and $75,500 ($253,700 and $51,800), respectively, for the average firm in

our primary sample, for a combined effect on audit fees of $234,200 ($305,500). Notably, the

impact of earnings autocorrelation and earnings volatility on audit fees is larger in magnitude

than several well-established audit fee model variables. For example, a one standard deviation

(interquartile range) change in total accruals, return on assets, and audit report lag is associated

with a change in audit fees of only $18,800, $152,100, and $114,100 ($14,700, $90,400, and

$128,400), respectively, for the average firm in our sample. Therefore, earnings autocorrelation

and earnings volatility combined appear to be of greater economic significance than several other

audit fee model variables.

Auditor Specialization

We test our third and fourth hypotheses by incorporating auditor specialization into the

model. Before testing our third and fourth hypotheses, we first examine whether our results for

auditor specialization are consistent with the prior research that finds a fee premium for

specialists. The first column in Table 4 presents the results of estimating the audit fee model using our specialization sample of 7,154 firm-year observations while only adding the main effect of auditor specialization into the model. Consistent with prior research, we find a positive and highly significant coefficient on auditor specialization (t = 3.96), suggesting a fee premium for specialists.

Columns 2-4 in Table 4 incorporate the interactions with auditor specialization.

Although we interact each control variable with SPEC, except the year and industry indicator variables, for brevity we present only the interactions with AUTO and IBVOL. The results reveal a negative and marginally significant coefficient on SPEC*AUTO (t = -1.78), which suggests that the influence of earnings autocorrelation on audit fees is strengthened for specialist

auditors. This implies that specialist auditors respond more thoroughly to uncertainty in earnings

than non-specialists, supporting our third hypothesis. However, we do not find a statistically

significant coefficient on SPEC*IBVOL (t = -0.48). Thus, we fail to find support for our fourth

hypothesis.

Further Analyses

Autocorrelation measured using annual earnings

In our empirical tests, we calculate earnings autocorrelation using quarterly earnings; however, autocorrelation could also be calculated by using annual earnings. As an additional test, we measure earnings autocorrelation using annual earnings with a minimum required estimation period of 9 years and a maximum estimation period of 12 years, which corresponds to the length of time chosen for the quarterly data. A significant drawback in using annual earnings is that there are far fewer observations available for the calculation (a maximum of 12 observations instead of 48), which leads to a less precise measurement of autocorrelation. When using annual earnings to calculate autocorrelation, our results (untabulated) continue to indicate a negative and significant association (t = -3.04) between earnings autocorrelation and audit fees.

Thus, our results are robust to measuring autocorrelation using annual earnings.

The influence of auditor tenure

Our results suggest that low earnings autocorrelation or high earnings volatility make auditor assessments of inherent risk more difficult, and that auditors respond to these time-series of earnings characteristics by increasing auditor effort and charging a higher audit fee. However, over time as an auditor gains experience with a client, the auditor may learn how to better

21 interpret earnings autocorrelation and earnings volatility. If auditor learning occurs, then we expect that earnings autocorrelation and earnings volatility will have less of an effect of audit fees when there is longer auditor tenure.

To examine this possibility, we split our sample based on the median value of 12 years of auditor tenure and estimate equation (1) for each of the subsamples. For the subsample of firms with less than 12 years of auditor tenure, our results (untabulated) continue to show a negative

(positive) and significant association between earnings autocorrelation (volatility) and audit fees.

However, for the subsample of firms with at least 12 years of auditor tenure, we find no association between earnings volatility and audit fees, but we do continue to find a negative and significant association between earnings autocorrelation and audit fees. These results suggest that auditors are able to learn to better interpret earnings volatility over time, but not earnings autocorrelation. A possible explanation for this result is that low earnings autocorrelation, by its nature, indicates earnings that are unable to be predicted by past results, so auditors are unable to learn from experience for firms with low earnings autocorrelation. However, although high earnings volatility may make earnings more difficult to interpret, high volatility itself does not necessary indicate unpredictability, so auditors may be able to learn from experience for firms with high earnings volatility.

Other additional tests

Although we have shown that earnings autocorrelation and earnings volatility are distinct variables, an unanswered question is whether each variable only affects audit fees at a certain level of the other variable. That is, perhaps earnings autocorrelation only matters when earnings

volatility is high, or perhaps earnings volatility only matters when earnings autocorrelation is

low. In order to examine this possibility, we split our primary sample of 10,510 observations

into two subsamples based on the median value of AUTO. We then estimate equation (1) for

each subsample and find (untabulated) that IBVOL is positive and significant in both subsamples. Similarly, when we estimate equation (1) on two subsamples created by splitting our primary sample on the median value of IBVOL, we find that AUTO is negative and significant in both subsamples. Therefore, the presence of a significant association between earnings autocorrelation (volatility) and audit fees does not depend on the level of earnings volatility (autocorrelation).

Lastly, in our main analysis we focus on the absolute value of earnings autocorrelation because we are interested in measuring the extent to which prior-period earnings are informative about current-period earnings, and not whether the autocorrelation is positive or negative.

However, we also conduct additional tests to examine whether our results are affected by negative autocorrelation.7 We continue to find a negative and significant association between

earnings autocorrelation and audit fees when we estimate equation (1) while either using signed

autocorrelation or excluding firm-year observations with negative autocorrelation. Thus, our results continue to hold when taking negative autocorrelation into account.

V. CONCLUSION

This study examines the effect of earnings autocorrelation and earnings volatility on audit fees. The results reveal a negative (positive) and highly significant association between earnings

7 Only 12.9% of the firm-year observations in our primary sample have negative earnings autocorrelation.

23 autocorrelation (volatility) and audit fees, which provides support for our hypotheses. We also investigate whether the association between earnings autocorrelation (volatility) and audit fees is affected by auditor specialization. We find that the negative association between earnings autocorrelation and audit fees is stronger for specialist auditors compared to non-specialists.

This study makes two primary contributions to the literature. First, we identify two economically significant determinants of audit fees. The results show that a change in the interquartile range in earnings autocorrelation and earnings volatility combined is associated a change in audit fees of $305,500 for the average firm in our primary sample. This impact is greater in magnitude than several previously established audit fee model variables. We also contribute to the auditor specialization literature by showing that specialization is associated with a heightened influence of earnings autocorrelation on audit fees, consistent with specialist auditors responding more thoroughly to uncertainty in earnings. Thus, we add to the literature that demonstrates the benefits of auditor specialization.

REFERENCES

Abbott, L. J., S. Parker, and G. F. Peters. 2006. Earnings management, litigation risk, and

asymmetric audit fee responses. Auditing: A Journal of Practice & Theory. 25: 88-98.

American Institute of Certified Public Accountants (AICPA). 2006. Audit Risk and Materiality in

Conducting an Audit. AU Section 312.

Antle, R., E. Gordon, G. Narayanamoorthy, and L. Zhou. 2006. The joint determination of audit

fees, non-audit fees, and abnormal accruals. Review of Quantitative Finance and

Accounting. 27: 235-266.

Ball, R., and E. Bartov. 1996. How naïve is the stock market’s use of earnings information?

Journal of Accounting and Economics. 21: 319-337.

Ball, R., S. Jayaraman, and L. Shivakumar. 2012. Audited financial reporting and voluntary

disclosure as complements: A test of the confirmation hypothesis. Journal of Accounting

and Economics. 53: 136-166.

Balsam, S., J. Krishnan, and J. Yang. 2003. Auditor industry specialization and earnings

quality. Auditing: A Journal of Practice and Theory. 22: 71-97.

Bathke, A. and K. Lorek. 1984. The relationship between time-series models and the security-

market’s expectation of quarterly earnings. The Accounting Review. 59: 163-176.

Bathke, A., T. Mason, and R. Morton. 2014. Post-earnings announcement reversal: The puzzling

case of seasonal random walk firms. Working paper. Florida State University.

Bernard, V., and J. Thomas, 1990, Evidence that stock prices do not fully reflect the implications

of current earnings for future earnings. Journal of Accounting and Economics. 13: 305-

340.

Bell, T., W. Landsman, and D. Shackelford. 2001. Auditor’s perceived business risk and audit

fees: Analysis and evidence. Journal of Accounting Research. 39(1): 35-43.

Brown, L., P. Griffin, R. Hagerman, and M. Zmijewski. 1987. Security analyst superiority

relative to univariate time-series models in forecasting quarterly earnings. Journal of

Accounting and Economics. 9: 61-87.

Carson, E. 2009. Industry specialization by global audit firm networks. The Accounting

Review. 84 (2): 355-382.

Dichev, I. and V. Tang. 2008. Matching and the changing properties of accounting earnings over

the last 40 years. The Accounting Review. 86(6): 1425-1460.

Dichev, I. and V. Tang. 2009. Earnings volatility and earnings predictability. Journal of

Accounting and Economics. 47: 160-181.

Francis, J., R. LaFond, P. Olsson, and K. Schipper. 2004. Costs of equity and earnings attributes.

The Accounting Review. 79: 967-1010.

Francis, J., K. Reichelt, and D. Wang. 2005. The pricing of national and city-specific

reputations for industry expertise in the U.S. audit market. The Accounting Review. 80

(1): 113-136.

Fung, S., F. Gul, and J. Krishnan. 2012. City-level auditor industry specialization, economies of

scale, and audit pricing. The Accounting Review. 87 (4): 1281-1307.

Givoly, D., and C. Hayn. 2000. The changing time-series properties of earnings, cash flows and

accruals: Has financial reporting become more conservative? Journal of Accounting and

Economics. 29: 287-320.

Graham, J., H. Campbell, and S. Rajgopal. 2005. The economic implications of corporate

financial reporting. Journal of Accounting and Economics. 40: 3-73.

Gul, F. A., C. Chen, and J. Tsui. 2003. Discretionary accounting accruals, managers’ incentives,

and audit fees. Contemporary Accounting Research. 20(3): 441-464.

Hay, D., W. R. Knechel, and N. Wong. 2006. Audit fees: A meta-analysis of the effect of supply

and demand attributes. Contemporary Accounting Research. 23 (1): 141-191.

Krishnan, J., J. Krishnan, and H. Song. 2011. The effect of Auditing Standard No. 5 on audit

fees. Auditing: A Journal of Practice & Theory. 30 (4): 1-27.

Krishnan, G., L. Sun, Q. Wang, and R. Yang. 2013. Client risk management: A pecking order

analysis of auditor response to upward earnings management risk. Auditing: A Journal of

Practice and Theory. 32(2): 147-169.

Public Company Accounting Oversight Board (PCAOB). 2010. Audit Risk. Auditing Standard

No. 8.

Reichelt, K., and D. Wang. 2010. National and office-specific measures of auditor industry

expertise and effects on audit quality. Journal of Accounting Research. 48 (3): 647-686.

Schelleman, C., and R. Knechel. 2010. Short-term accruals and the pricing and production of

audit services. Auditing: A Journal of Practice and Theory. 29(1): 221-250.

Simunic, D. 1980. The pricing of audit services: Theory and evidence. Journal of Accounting

Research. 18(1): 161-190.

Simunic, D., and M. Stein. 1996. The impact of litigation risk on audit pricing: A review of the

economics and the evidence. Auditing: A Journal of Practice and Theory. 15

(supplement): 119-134.

Whisenant, S., S. Sankaraguruswamy, and K. Raghunandan. 2003. Evidence of the joint

determination of audit and non-audit fees. Journal of Accounting Research. 41: 721-744.

Willenborg, M. 2002. Discussion of: Brand name audit pricing, industry specialization, and

industry leadership premiums post-big 8 and big 6 mergers. Contemporary Accounting

Research. 19(1): 111-116.

Table 1 Descriptive Statistics Panel A: Summary Statistics Primary Sample Specialization Sample (N = 10,510) (N = 7,154) Mean Median S.D. Mean Median S.D.

AUDIT FEES 2,641,526 1,419,500 3,462,720 2,768,656 1,465,000 3,663,465 TOTAL ASSETS 4,939.59 1,041.64 14,755.32 5,356.98 1,052.69 16,725.17 LNFEES 14.26 14.17 0.98 14.30 14.20 0.99 AUTO 0.30 0.26 0.23 0.30 0.27 0.22 IBVOL 0.08 0.04 0.10 0.08 0.04 0.10 LNAT 7.02 6.95 1.66 7.03 6.96 1.69 ROA 0.02 0.05 0.14 0.02 0.05 0.14 ACCR 0.08 0.06 0.09 0.08 0.06 0.09 CURR 0.48 0.48 0.22 0.48 0.48 0.22 FSALES 0.27 0.17 0.29 0.29 0.22 0.29 SEG 2.84 3.00 1.98 2.86 3.00 2.03 LEV 0.51 0.50 0.24 0.50 0.49 0.24 LOSS 0.23 0.00 0.42 0.23 0.00 0.42 DEC 0.67 1.00 0.47 0.69 1.00 0.46 LAG 96.77 95.00 24.89 96.76 95.00 24.34 TENURE 14.51 12.00 9.86 14.54 12.00 9.80 GC 0.01 0.00 0.11 0.01 0.00 0.11 XIDUM 0.02 0.00 0.13 0.02 0.00 0.13 ACQ 0.49 0.00 0.50 0.49 0.00 0.50 NYSE 0.50 0.00 0.50 0.49 0.00 0.50 NEWDEBT 0.51 1.00 0.50 0.50 0.00 0.50 NEWEQ 0.07 0.00 0.25 0.07 0.00 0.26 LIT 0.36 0.00 0.48 0.38 0.00 0.48 MTB 2.72 2.08 3.39 2.79 2.13 3.38 LNNAF 12.25 12.35 1.64 12.30 12.40 1.65 RETVOL 0.12 0.10 0.07 0.12 0.10 0.07 RETSUM 0.13 0.12 0.46 0.13 0.13 0.45 MATWEAK 0.05 0.00 0.23 0.05 0.00 0.23 SPEC - - - 0.56 1.00 0.50

Panel B: Sample Attrition

Firm-years at the intersection of Compustat, CRSP, and Audit Analytics 67,973 for the period 2004 – 2012

Less: Financial firms, utilities, foreign firms, and firm-years (44,149) with non-Big 4 auditors

Less: Firm-years missing required data (13,314)

Primary sample 10,510

Less: Firm-years lost because (1) there is only one observation in the industry (3,356) in the auditor’s city in that year, or (2) there is only one Big 4 auditor in the city in that year.

Specialization sample 7,154

Table 1 provides summary statistics and sample attrition information for our primary (specialization) sample of 10,510 (7,154) firm-year observations for the period 2004 – 2012. All continuous variables are winsorized at the 1st and 99th percentiles. AUDIT FEES is total audit fees in dollars. TOTAL ASSETS is total assets in millions. LNFEES is the natural logarithm of total audit fees. AUTO is the absolute value of earnings autocorrelation, estimated as the correlation between the seasonally-differenced quarterly earnings series at the first lag, using a 48 quarter estimation window. IBVOL is the standard deviation of earnings before extraordinary items scaled by total assets at the beginning of the year, measured over the period t-4 through t. LNAT is the natural logarithm of total assets (in millions). ROA is net income scaled by the average of total assets at the beginning and end of the year. ACCR is the absolute value of total accruals scaled by total assets. CURR is current assets divided by total assets. FSALES is foreign sales divided by total sales. SEG is the number of business segments. LEV is total liabilities scaled by total assets. LOSS is in indicator variable that takes the value of 1 if income before extraordinary items is negative and 0 otherwise. DEC is an indicator variable that takes the value of 1 if a firm has a December fiscal year- end and 0 otherwise. LAG is the number of days in between a firm’s fiscal year-end and the issuance of the audit report. TENURE is the number of years the current auditor has audited a firm. GC is an indicator variable that takes the value of 1 if a firm received a going-concern audit opinion during the year and 0 otherwise. XIDUM is an indicator variable that takes the value of 1 if extraordinary items is non-zero and 0 otherwise. ACQ is an indicator variable that takes the value of 1 if a firm engages in merger or acquisition activity during the year and 0 otherwise. NYSE is an indicator variable that takes the value of 1 if a firm is listed on the New York Stock Exchange and 0 otherwise. NEWDEBT is an indicator variable that takes the value of 1 if a firm issues long-term debt during the year and 0 otherwise. NEWEQ is an indicator variable that takes the value of 1 if a firm issues common stock in excess of five percent of its market value of equity during the year and 0 otherwise. LIT is an indicator variable that takes the value of 1 if a firm is in a high litigation risk industry (SIC codes 2833-2836, 3570-3577, 3600-3674, 5200-5961, and 7370-7374) and 0 otherwise. MTB is the market value of equity divided by the book value of equity. LNNAF is the natural logarithm of total non-audit fees. RETVOL (RETSUM) is the standard deviation (sum) of monthly stock returns during the fiscal year. MATWEAK is an indicator variable that takes the value of 1 if a firm reports a material weakness in internal control over financial reporting and 0 otherwise. SPEC is an indicator variable that takes the value of 1 if a firm in a given industry is audited by an auditor that specializes in that industry and 0 otherwise. Auditors are classified as specialists if they have the highest market share (in terms of audit fees) in an industry in their city, during a particular year.

Table 2 Correlation Matrix Panel A: Primary Sample (N = 10,510) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (1) LNFEES (2) AUTO -0.12 (3) IBVOL -0.22 -0.05 (4) LNAT 0.81 -0.05 -0.34 (5) ROA 0.18 0.08 -0.36 0.32 (6) ACCR -0.15 -0.01 0.31 -0.20 -0.52 (7) CURR -0.24 0.10 0.26 -0.45 -0.15 0.04 (8) FSALES 0.35 0.01 -0.01 0.16 0.09 -0.07 0.19 (9) SEG 0.42 -0.06 -0.18 0.34 0.13 -0.12 -0.18 0.09 (10) LEV 0.28 -0.17 -0.03 0.29 -0.16 0.13 -0.27 -0.12 0.09 (11) LOSS -0.18 -0.06 0.32 -0.29 -0.70 0.45 0.14 -0.04 -0.14 0.12 (12) DEC 0.03 -0.09 0.09 0.04 -0.06 0.04 -0.13 -0.01 0.03 0.12 0.05 (13) LAG -0.16 -0.02 0.20 -0.27 -0.19 0.11 0.09 -0.06 -0.12 -0.04 0.17 0.02 (14) TENURE 0.23 0.01 -0.13 0.24 0.10 -0.10 -0.03 0.11 0.18 0.04 -0.11 -0.03 -0.18 (15) GC -0.05 0.00 0.17 -0.10 -0.28 0.15 0.04 -0.02 -0.04 0.15 0.17 0.03 0.09 0.00 (16) XIDUM 0.08 -0.03 -0.02 0.07 0.00 -0.02 -0.04 -0.01 0.05 0.06 -0.01 0.01 -0.01 0.01 (17) ACQ 0.30 -0.02 -0.17 0.25 0.14 -0.12 -0.19 0.12 0.23 0.01 -0.16 -0.02 -0.03 0.07 (18) NYSE 0.44 -0.05 -0.27 0.51 0.19 -0.15 -0.31 0.02 0.27 0.26 -0.21 0.06 -0.27 0.20 (19) NEWDEBT 0.30 -0.08 -0.17 0.35 0.04 -0.06 -0.39 -0.02 0.19 0.38 -0.08 0.10 -0.08 0.04 (20) NEWEQ -0.12 -0.01 0.22 -0.15 -0.35 0.09 0.07 -0.08 -0.08 0.02 0.22 0.07 0.05 -0.08 (21) LIT -0.11 0.01 0.19 -0.13 -0.14 0.10 0.26 0.04 -0.22 -0.15 0.13 -0.15 0.11 -0.05 (22) MTB 0.00 0.03 0.09 -0.01 0.06 -0.03 0.12 0.01 -0.06 -0.03 -0.06 0.03 -0.03 0.01 (23) LNNAF 0.64 -0.06 -0.17 0.59 0.18 -0.14 -0.18 0.24 0.32 0.19 -0.18 0.03 -0.15 0.20 (24) RETVOL -0.23 -0.02 0.33 -0.32 -0.36 0.33 0.16 -0.02 -0.14 0.12 0.39 0.01 0.15 -0.14 (25) RETSUM -0.03 -0.01 -0.01 0.00 0.18 -0.11 0.03 -0.01 0.01 0.00 -0.15 0.00 -0.04 0.00 (26) MATWEAK 0.03 -0.03 0.09 -0.09 -0.06 0.04 0.02 -0.01 0.00 0.01 0.08 -0.01 0.19 -0.07

Panel A: Primary Sample (Cont.) (15) (16) (17) (18) (19) (20) (21) (22) (23) (24) (25) (15) GC (16) XIDUM 0.01 (17) ACQ -0.06 0.01 (18) NYSE -0.04 0.06 0.12 (19) NEWDEBT 0.00 0.05 0.16 0.29 (20) NEWEQ 0.12 0.00 -0.07 -0.08 0.00 (21) LIT 0.01 -0.03 -0.08 -0.32 -0.23 0.06 (22) MTB -0.03 -0.01 -0.02 -0.03 -0.06 0.04 0.08 (23) LNNAF -0.05 0.05 0.23 0.32 0.21 -0.09 -0.08 0.02 (24) RETVOL 0.19 -0.04 -0.19 -0.19 -0.08 0.21 0.06 -0.08 -0.20 (25) RETSUM -0.10 0.01 -0.04 0.03 -0.02 0.06 -0.02 0.18 0.00 0.18 (26) MATWEAK 0.05 0.03 -0.02 -0.07 -0.01 0.02 0.03 -0.01 -0.03 0.02 -0.04

Panel B: Specialization Sample (N = 7,154) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (1) LNFEES (2) AUTO -0.10 (3) IBVOL -0.25 -0.04 (4) LNAT 0.82 -0.04 -0.35 (5) ROA 0.21 0.07 -0.40 0.33 (6) ACCR -0.15 -0.01 0.31 -0.18 -0.51 (7) CURR -0.27 0.09 0.27 -0.48 -0.17 0.01 (8) FSALES 0.37 0.06 -0.05 0.20 0.13 -0.10 0.18 (9) SEG 0.45 -0.05 -0.20 0.37 0.14 -0.14 -0.18 0.10 (10) LEV 0.27 -0.16 -0.04 0.29 -0.17 0.13 -0.28 -0.11 0.11 (11) LOSS -0.19 -0.05 0.33 -0.29 -0.70 0.45 0.16 -0.07 -0.15 0.12 (12) DEC 0.04 -0.08 0.08 0.06 -0.08 0.05 -0.16 -0.08 0.00 0.15 0.05 (13) LAG -0.17 -0.02 0.21 -0.27 -0.19 0.12 0.11 -0.09 -0.14 -0.07 0.17 0.02 (14) TENURE 0.22 0.02 -0.13 0.21 0.10 -0.11 0.01 0.16 0.20 0.03 -0.12 -0.05 -0.18 (15) GC -0.05 0.02 0.20 -0.10 -0.29 0.16 0.02 -0.02 -0.03 0.18 0.17 0.03 0.10 0.00 (16) XIDUM 0.08 -0.02 -0.02 0.06 0.01 -0.03 -0.03 0.00 0.07 0.05 -0.02 0.01 -0.03 0.00 (17) ACQ 0.31 0.00 -0.19 0.26 0.15 -0.12 -0.20 0.10 0.23 0.01 -0.17 -0.04 -0.03 0.09 (18) NYSE 0.43 -0.04 -0.29 0.51 0.20 -0.12 -0.36 0.03 0.28 0.29 -0.21 0.10 -0.28 0.17 (19) NEWDEBT 0.31 -0.07 -0.17 0.38 0.05 -0.05 -0.43 -0.01 0.20 0.40 -0.08 0.13 -0.11 0.02 (20) NEWEQ -0.14 -0.02 0.24 -0.15 -0.36 0.10 0.07 -0.11 -0.09 0.02 0.24 0.07 0.06 -0.08 (21) LIT -0.12 -0.01 0.21 -0.17 -0.15 0.08 0.27 0.03 -0.18 -0.18 0.14 -0.15 0.13 -0.05 (22) MTB -0.02 0.02 0.08 -0.03 0.04 -0.04 0.13 0.01 -0.05 -0.02 -0.05 0.01 -0.02 0.01 (23) LNNAF 0.64 -0.05 -0.19 0.59 0.19 -0.14 -0.20 0.25 0.35 0.19 -0.18 0.01 -0.16 0.19 (24) RETVOL -0.24 -0.03 0.34 -0.31 -0.38 0.33 0.16 -0.05 -0.16 0.13 0.39 0.02 0.14 -0.13 (25) RETSUM -0.03 -0.01 -0.02 0.00 0.16 -0.12 0.02 0.00 0.00 0.00 -0.14 0.00 -0.06 0.00 (26) MATWEAK 0.03 -0.02 0.10 -0.09 -0.06 0.04 0.02 0.00 0.00 0.00 0.07 -0.02 0.20 -0.07 (27) SPEC 0.24 0.00 -0.03 0.21 0.06 -0.05 -0.06 0.07 0.12 0.05 -0.06 0.02 -0.03 0.07

Panel B: Specialization Sample (Cont.) (15) (16) (17) (18) (19) (20) (21) (22) (23) (24) (25) (26) (15) GC (16) XIDUM 0.01 (17) ACQ -0.06 0.01 (18) NYSE -0.03 0.06 0.12 (19) NEWDEBT 0.00 0.04 0.15 0.34 (20) NEWEQ 0.12 0.00 -0.09 -0.08 -0.01 (21) LIT 0.01 -0.04 -0.05 -0.38 -0.26 0.06 (22) MTB -0.02 -0.01 -0.03 -0.03 -0.06 0.06 0.09 (23) LNNAF -0.04 0.06 0.24 0.30 0.21 -0.10 -0.07 0.00 (24) RETVOL 0.21 -0.04 -0.20 -0.18 -0.08 0.23 0.07 -0.08 -0.21 (25) RETSUM -0.08 0.01 -0.04 0.04 -0.01 0.08 -0.04 0.17 0.00 0.17 (26) MATWEAK 0.06 0.02 -0.01 -0.09 -0.03 0.00 0.03 -0.01 -0.02 0.02 -0.04 (27) SPEC -0.02 0.02 0.09 0.08 0.06 -0.03 -0.01 0.01 0.17 -0.06 0.00 0.02

Panel A (Panel B) of Table 2 presents a Pearson correlation matrix for our primary (specialization) sample of 10,510 (7,154) firm-year observations for the period 2004 – 2012. All continuous variables are winsorized at the 1st and 99th percentiles. Bolded values indicate statistical significance at the p < 0.05 level using a 2-tailed test. Italicized values indicate statistically insignificant associations (p > 0.05). See Table 1 for a detailed description of the variables.

Table 3 Audit Fee Model Regression DV = LNFEES (1) (2) (3) (4)

AUTO -0.266 -0.259 (-6.93)*** (-6.77)*** IBVOL 0.324 0.282 (3.49)*** (3.08)*** LNAT 0.433 0.434 0.435 0.436 (45.13)*** (45.74)*** (45.16)*** (45.71)*** ROA -0.424 -0.406 -0.413 -0.397 (-5.44)*** (-5.24)*** (-5.28)*** (-5.11)*** ACCR -0.056 -0.031 -0.113 -0.081 (-0.73) (-0.40) (-1.44) (-1.03) CURR 0.380 0.396 0.363 0.381 (6.21)*** (6.57)*** (5.95)*** (6.34)*** FSALES 0.507 0.503 0.509 0.504 (12.33)*** (12.39)*** (12.32)*** (12.38)*** SEG 0.051 0.049 0.052 0.050 (10.26)*** (10.07)*** (10.34)*** (10.15)*** LEV 0.246 0.214 0.246 0.214 (5.38)*** (4.68)*** (5.37)*** (4.69)*** LOSS 0.027 0.023 0.026 0.022 (1.56) (1.34) (1.47) (1.26) DEC 0.009 0.001 0.006 -0.002 (0.40) (0.04) (0.25) (-0.08) LAG 0.002 0.002 0.002 0.002 (4.89)*** (4.88)*** (4.71)*** (4.72)*** TENURE 0.001 0.001 0.001 0.001 (0.67) (0.90) (0.75) (0.96) GC 0.037 0.063 0.024 0.052 (0.69) (1.22) (0.45) (1.00) XIDUM 0.115 0.106 0.115 0.107 (3.75)*** (3.37)*** (3.78)*** (3.40)*** ACQ 0.070 0.071 0.071 0.072 (4.88)*** (5.06)*** (4.97)*** (5.13)*** NYSE 0.120 0.116 0.123 0.119 (4.89)*** (4.87)*** (5.03)*** (4.99)*** NEWDEBT 0.039 0.040 0.041 0.041 (2.78)*** (2.87)*** (2.90)*** (2.98)*** NEWEQ -0.011 -0.006 -0.020 -0.014 (-0.47) (-0.28) (-0.86) (-0.63) LIT -0.001 -0.013 -0.011 -0.022 (-0.02) (-0.35) (-0.30) (-0.59)

MTB -0.001 -0.001 -0.002 -0.001 (-0.47) (-0.25) (-0.77) (-0.52) LNNAF 0.075 0.074 0.075 0.074 (11.93)*** (11.94)*** (11.85)*** (11.86)*** RETVOL 0.439 0.440 0.348 0.360 (3.85)*** (3.88)*** (3.03)*** (3.15)*** RETSUM -0.025 -0.028 -0.022 -0.026 (-2.28)** (-2.58)** (-2.06)** (-2.37)** MATWEAK 0.319 0.316 0.314 0.311 (12.03)*** (11.90)*** (11.88)*** (11.76)*** INTERCEPT 9.326 9.385 9.327 9.384 (45.01)*** (46.73)*** (45.73)*** (47.35)***

R2 81.35% 81.68% 81.42% 81.74% N 10,510 10,510 10,510 10,510 ***, **, and * indicate statistical significance at the 0.01, 0.05, and 0.10 levels, respectively, using a 2-tailed test. T- statistics are in parentheses. Standard errors are clustered by firm. All continuous variables are winsorized at the 1st and 99th percentiles. Table 3 presents the results of estimating equation (1) for our sample of 10,510 firm-year observations for the period 2004 – 2012. Year and industry indicator variables are omitted for brevity. See Table 1 for a detailed description of the variables.

Table 4 Audit Fee Model Regression with Specialization DV = LNFEES (1) (2) (3) (4)

AUTO -0.183 -0.176 (-2.98)*** (-2.86)*** IBVOL 0.284 0.253 (2.31)** (2.08)** SPEC*AUTO -0.128 -0.132 (-1.72)* (-1.78)* SPEC*IBVOL -0.064 -0.083 (-0.37) (-0.48) SPEC 0.075 0.021 -0.004 0.022 (3.96)*** (0.11) (-0.02) (0.11) LNAT 0.428 0.413 0.415 0.415 (38.62)*** (23.60)*** (23.88)*** (23.50)*** ROA -0.473 -0.472 -0.452 -0.458 (-5.05)*** (-3.23)*** (-3.11)*** (-3.14)*** ACCR -0.078 -0.074 -0.126 -0.117 (-0.82) (-0.58) (-1.00) (-0.93) CURRENT 0.364 0.430 0.404 0.410 (5.22)*** (4.52)*** (4.24)*** (4.31)*** FSALES 0.535 0.531 0.531 0.534 (11.52)*** (8.78)*** (8.75)*** (8.79)*** SEGMENTS 0.057 0.051 0.053 0.052 (10.39)*** (6.61)*** (6.79)*** (6.66)*** LEV 0.208 0.181 0.208 0.185 (3.79)*** (2.37)** (2.74)*** (2.43)** LOSS 0.017 -0.012 -0.007 -0.012 (0.82) (-0.40) (-0.24) (-0.40) DEC 0.033 0.029 0.026 0.025 (1.27) (0.79) (0.71) (0.68) LAG 0.002 0.002 0.002 0.002 (4.86)*** (3.30)*** (3.30)*** (3.23)*** TENURE -0.000 -0.001 -0.001 -0.001 (-0.07) (-0.42) (-0.43) (-0.37) GC 0.032 0.067 -0.004 0.034 (0.52) (0.83) (-0.05) (0.42) XIDUM 0.102 0.049 0.052 0.051 (2.76)*** (0.76) (0.83) (0.80) ACQ 0.078 0.094 0.092 0.095 (4.58)*** (3.82)*** (3.74)*** (3.86)*** NYSE 0.136 0.161 0.165 0.161 (5.02)*** (4.39)*** (4.47)*** (4.40)***

NEWDEBT 0.043 0.060 0.060 0.059 (2.61)*** (2.60)*** (2.56)** (2.54)** NEWEQ -0.020 -0.046 -0.055 -0.051 (-0.75) (-1.21) (-1.44) (-1.35) LIT -0.009 -0.023 -0.024 -0.031 (-0.22) (-0.45) (-0.48) (-0.61) MTB -0.001 -0.005 -0.005 -0.005 (-0.38) (-1.10) (-1.24) (-1.23) LNNAF 0.068 0.067 0.067 0.066 (8.96)*** (6.13)*** (6.13)*** (6.08)*** RETVOL 0.348 0.544 0.479 0.472 (2.61)*** (3.11)*** (2.67)*** (2.65)*** RETSUM -0.018 -0.033 -0.031 -0.031 (-1.39) (-1.70)* (-1.61) (-1.61) MATWEAK 0.335 0.282 0.274 0.275 (10.91)*** (5.67)*** (5.49)*** (5.54)*** INTERCEPT 9.239 9.358 9.319 9.366 (35.12)*** (35.29)*** (33.93)*** (35.64)***

R2 82.24% 82.66% 82.39% 82.69% N 7,154 7,154 7,154 7,154 ***, **, and * indicate statistical significance at the 0.01, 0.05, and 0.10 levels, respectively, using a 2-tailed test. T- statistics are in parentheses. Standard errors are clustered by firm. All continuous variables are winsorized at the 1st and 99th percentiles. Table 4 presents the results of estimating equation (2) for our specialization sample of 7,154 firm-year observations for the period 2004 – 2012. Year and industry indicator variables are omitted for brevity. Also omitted for brevity are the control variable interactions with SPEC. See Table 1 for a detailed description of the variables.