Earnings Management, Litigation Risk, and Asymmetric Audit Fee Responses. Lawrence J. Abbott, Susan Parker, and Gary F. Peters

Transcription

1 AUDITING: A JOURNAL OF PRACTICE & THEORY Vol. 25, No. 1 May 2006 pp Earnings Management, Litigation Risk, and Asymmetric Audit Fee Responses Lawrence J. Abbott, Susan Parker, and Gary F. Peters SUMMARY: This study examines the association between audit fees and earnings management, using publicly available fee data. We hypothesize that, due to asymmetric litigation effects, audit fees decrease (increase) with a client s risk of income-decreasing (increasing) earnings management risk. We also hypothesize that the positive relation between income-increasing earnings management risk and audit fees is heightened for clients that are high-growth firms. We test our hypotheses with a sample of 429 public, non-regulated, Big 5 audited companies, using fee data for the year We find that downward earnings management risk, as estimated by negative (i.e., incomedecreasing) discretionary accruals, is associated with lower audit fees. We also document that upward earnings management risk, as estimated by positive discretionary accruals, is associated with higher audit fees and that the interaction of this risk with an industry-adjusted price-earnings ratio has an incrementally significant, positive effect on fees. We interpret our findings as consistent with a conservative bias on the part of auditors. The conservative bias arises from asymmetric litigation risk in which income-increasing discretionary accruals exhibit greater expected litigation costs than income-decreasing discretionary accruals (Simunic and Stein 1996; Palmrose and Scholz 2004; Palmrose et al. 2004; Richardson et al. 2002; Heninger 2001). INTRODUCTION Litigation risk is a fundamental aspect of an auditor s fee-setting process (Simunic and Stein 1996). Prior archival audit fee research has generally focused on client financial health in proxying for the auditors litigation risk cost component (Pratt and Stice 1994; Simon and Francis 1988; Simunic 1980). However, recent audit-planning research suggests that auditors also consider earnings management risk in their riskassessment process. In particular, Bedard and Johnstone (2004) find that auditors respond to earnings management risk with ex ante increases in planned audit hours and billing rates. In this study, we expand this line of research and investigate whether auditors asymmetrically respond to income-increasing versus income-decreasing earnings management risk when considering litigation risk and ex post determination of audit fees. 1 1 We use the term asymmetric fee response to describe audit fee decreases (increases) in reaction to incomedecreasing (increasing) earnings management risk. Audit fee effects may be attributable to either audit scope revisions that address areas of assessed earnings management risk or simply a risk premium. Lawrence J. Abbott is an Associate Professor at the University of Memphis, Susan Parker is an Associate Professor at Santa Clara University, and Gary F. Peters is an Associate Professor at the University of Arkansas. 85 Submitted: October 2004 Accepted: December 2005

2 86 Abbott, Parker, and Peters We hypothesize that audit fees decrease (increase) with a client s risk of managing earnings downward (upward) due to asymmetric litigation risk outcomes. We draw on three research streams in forming our hypotheses. First, Heninger (2001) finds a significant positive relation between income-increasing abnormal accruals a common proxy for earnings management and subsequent auditor litigation. Second, Palmrose and Scholz (2004) examine restatements and demonstrate that: (1) annual restatements trigger auditor litigation at a higher rate than bankruptcies, and (2) 83 percent of restatements that elicit litigation involve reversals of previously income-increasing accounting. Finally, Barron et al. (2001) show that auditor assessments of litigation risk and planned audit investments are higher when potential errors overstate as opposed to understate financial performance. These studies suggest an asymmetric fee response on the part of the auditor to income-increasing versus income-decreasing earnings management. Following Barron et al. (2001), we also hypothesize that the audit fee effects resulting from a client s propensity to manage earnings upward will be magnified for clients within greater litigation risk environments. We categorize high price-earnings (P/E) clients as those facing greater litigation risk. High P/E clients likely have incentives to inflate their earnings to meet earnings growth expectations (Bedard and Johnstone 2004; Richardson et al. 2002; Dechow et al. 2000). In turn, the incentive to inflate earnings is greatest for high-growth clients since these firms are most severely punished by investors for missing earnings forecasts in the form of a stock market torpedo (Skinner and Sloan 2002). Moreover, such a torpedo effect could prompt future litigation, given the large, negative stock price reaction at the time of the earnings disappointment (Skinner and Sloan 2002; Stice 1991). To address our research question, we examine a sample of 429 audit fee disclosures from proxy statements filed in Consistent with prior research, we use negative (positive) discretionary accruals to proxy for income-decreasing (increasing) earnings management (Heninger 2001; DeFond and Park 1997; Subramanyam 1996). Following Bedard and Johnstone (2004), we also construct a client-specific, industry-adjusted P/E ratio. Using multivariate regressions controlling for client size, complexity, financial condition, and corporate governance structure, we document lower (greater) audit fees for client firms with income-decreasing (increasing) discretionary accruals. Moreover, the positive relation between income-increasing discretionary accruals and audit fees is incrementally greater for high P/E ratio firms. To our knowledge, our study is the first to provide archival evidence using audit fees to support arguments for an asymmetric response on the part of auditors to incomeincreasing versus income-decreasing earnings management risk. Barron et al. (2001) posit that asymmetrically planned audit investments are rational if the risk faced by the auditor is asymmetric with regard to the direction of misstatement. Although auditors may not be under-auditing income-decreasing discretionary accruals, they do appear to adjust audit effort or extract risk premiums depending on the direction of earnings management risk. Notably, this is somewhat at odds with the traditional audit risk model, which directs auditors to address the general risk of material misstatements or inherent risk (Auditing Standards Board [ASB] 1983). Generally Accepted Auditing Standards (GAAS) do not define inherent risk in terms of income-decreasing versus income-increasing components. Accordingly, the audit risk model suggests that any deviation from an unbiased report, regardless of direction, should result in greater auditor effort. Our results suggest that the conventional audit risk model may not be adequately capturing auditor pricing behavior with respect to earnings management risk.

3 Earnings Management, Litigation Risk, and Asymmetric Audit Fee Responses 87 The remainder of our paper is as follows. The next section reviews prior research and develops our hypotheses. The third section presents sample selection and research design. The fourth section discusses results and limitations, and the final section concludes. PRIOR RESEARCH AND HYPOTHESES The auditor s cost of gathering evidence and producing an audit opinion is generally broken down into two elements: a resource cost component, which is increasing in the level of auditor effort, and an expected future loss component (Simunic and Stein 1996). Simunic (1980) models the expected loss component as the product of (1) the likelihood of litigation and (2) the expected present value of possible future losses that may arise from auditing this period s financial statements. Several research streams suggest that the auditor s expected loss component may be influenced by the direction of earnings management. Heninger (2001) documents a positive association between income-increasing abnormal accruals and ex post auditor litigation. Complementing these findings, Palmrose and Scholz (2004) show higher litigation rates against auditors for annual restatements than for client business failures. They also find that, on average, restatements of previously incomeincreasing accounting choices generated cumulative, negative abnormal returns of 13 percent, compared to negative 3 percent for restatements of income-decreasing accounting choices. Combined, these studies suggest that the likelihood of auditor litigation depends on the magnitude and direction of earnings management. Accordingly, it appears reasonable that the auditor s assessment of expected losses would vary depending on whether the result of the accounting discretion is income-increasing or income-decreasing. Houston et al. (1999) investigate the conditions in which accounting choices influence the auditor s assessment of audit and litigation risk. Using an experimental design, Houston et al. (1999) find that the presence of accounting choices reflecting higher risks of accounting irregularities leads to higher litigation risk assessments and fee premiums. More importantly, these authors find when potential accounting irregularities are present, auditor litigation risk assessments dominate the audit risk model s inherent risk assessments in explaining auditor planning behavior. Similarly, Barron et al. (2001) posit a conservative bias model of auditor behavior. They show auditors assessments of litigation risk and planned audit investments were higher (lower) when potential errors increased (decreased) financial performance. However, the results of Houston et al. (1999) and Barron et al. (2001) are constrained in that their earnings management construct does not have a publicly available analog. In an empirical setting that utilizes a publicly available proxy for earnings management, Lee and Mande (2003) examine how the passage of the Private Securities Litigation Act (PSLRA) affects auditors incentives to curtail earnings management. Assuming that the PSLRA reduced auditor litigation exposure, their tests find that after the PSLRA, incomeincreasing discretionary accruals rise for auditees of Big 6 firms (i.e., those firms with the most reputational capital). The results of Heninger (2001) and Palmrose and Scholz (2004) suggest that the likelihood of auditor litigation depends on the magnitude and direction of earnings management. The results of Houston et al. (1999) and Lee and Mande (2003) suggest that auditors incorporate the magnitude and direction of discretionary accruals into their litigation risk assessments. Finally, Barron et al. (2001) argue that an increase (decrease) in audit investments for misstatements that increase (decrease) net income is rational if litigation outcomes are correspondingly asymmetric. Thus, the asymmetric impact of income-increasing versus income-decreasing earnings management on the auditor s expected loss function suggests

4 88 Abbott, Parker, and Peters a correspondingly asymmetric impact on audit fees. This leads to our first hypothesis (stated in alternative form): H1: Audit fees decrease (increase) with income-decreasing (increasing) discretionary accruals. While the above hypothesis is similar in spirit to the earnings management risk hypothesis of Bedard and Johnstone (2004), there are three important differences that merit discussion. First, Bedard and Johnstone (2004) do not consider auditor behavior in response to actual managerial financial reporting decisions encompassing discretionary accruals. Second, the earnings management risk construct of Bedard and Johnstone (2004) was essentially a dichotomous variable that did not address income-increasing versus incomedecreasing earnings management. 2 This distinction is critical since our income-increasing and income-decreasing partition allows us to directly test the competing predictions of the audit risk model and the conservative bias model of Barron et al. (2001). 3 Third, Bedard and Johnstone (2004) only investigate the relation between earnings management risk factors and ex ante planned audit effort and fee premiums. 4 Our setting allows us to explicitly investigate the association between the actual management accounting choices and ex post auditor actions, namely charged audit fees. Following Barron et al. (2001), we also hypothesize that the risk effect resulting from a client s propensity to manage earnings is magnified for clients in greater litigation risk environments. Much prior research recognizes accruals as a means of managing earnings (Dechow et al. 2000; Dechow and Skinner 2000; Jones 1991). Dechow et al. (2000) find that high-growth firms characterized by high earnings-to-price multiples are more likely to use large accruals to manage earnings. The incentive for managing earnings on the part of these high-growth firms is to meet earnings growth expectations and avoid negative earnings surprises (Bedard and Johnstone 2004; Richardson et al. 2002; Dechow et al. 2000). In turn, the incentive to inflate earnings is greatest for high-growth clients since these firms are most severely punished by investors for missing earnings forecasts in the form of a stock market torpedo, which could prompt future litigation (Skinner and Sloan 2002). Collectively, this research suggests high P/E clients face greater litigation risk environments, thus leading to our second hypothesis (stated in alternative form): H2: Audit fee effects resulting from a client s propensity to manage earnings upward are magnified for clients within greater litigation risk environments as identified by high-growth or high price-earnings (P/E) clients. 2 Bedard and Johnstone (2004) derive an earnings management risk-assessment score from nine dichotomous yes/ no audit-planning questions. Although their measure was additive in nature, approximately 81 percent of their observations had zero earnings management risk factors and less than 4 percent had more than 2. Also, the additive nature of their measure does not distinguish between income-increasing and -decreasing earnings management risk. 3 The audit risk model would suggest that any misstatement regardless of direction should result in greater audit work. Thus, a positive relation between income-decreasing discretionary accruals and audit fees is predicted by the audit risk model. 4 Discussions with Big 4 audit personnel suggest that audit procedures and scope often change with financial events and/ or reporting decisions occurring after the completion of audit planning.

5 Earnings Management, Litigation Risk, and Asymmetric Audit Fee Responses 89 METHOD Sample Selection SEC rules requiring fee disclosures became effective for proxies filed on or after February 5, We first examined all proxies (excluding mutual funds and other financial registrants) filed with the SEC between February 5, 2001 and March 16, 2001 for which there was a corresponding fiscal year-end 10-K available by March 16, To expand the original sample of 310 firms, we then selected a random sample of 250 proxy filings from March 19, 2001 to June 30, 2001, resulting in a preliminary sample of 560. After deleting observations with missing variables on Compustat (primarily those needed to estimate discretionary accruals), or those from 15 two-digit industries that contained fewer than 20 non-sample firms over which to estimate nondiscretionary accruals, 473 companies remained. We also eliminated 44 firms that engaged non-big 5 auditors, resulting in a final sample size of 429 firms. Regression Model and Variable Definitions Following prior research, our regression model is as follows: LNAFEE b 0 b 1 DISCACC*INCR b 2 DISCACC*DECR b 3 P/E*DISCACC*INCR b 4 SIZE b 5 SQSUBS b 6 FORSUBS b 7 INVREC b 8 RECURLOS b 9 LEVERAGE b 10 GOINGCON b 11 PERCOSBD b 12 BORDMEET b 13 ACINDACT e. Variable definitions are summarized in Table 1. Consistent with prior studies, our dependent variable (LNAFEE) is the natural log of audit fees (in 000s). Our primary test variables are the firm s estimated discretionary accruals (DISCACC) and the industry-adjusted price-earnings ratio (P/ E). We estimated discretionary accruals (DISCACC) for each sample firm using the cross-sectional variation of the Jones (1991) model, as modified by Dechow et al. (1995). The cross-sectional method is appropriate here because it allows comparisons across firms with different corporate governance characteristics and does not assume that the coefficients will be stationary over time. We calculated our price-earnings ratio (P/ E) as the fiscal year-end closing price (Compustat item 199) for 2000 divided by the year 2000 net income (not including extraordinary items, Compustat item 18). We then industry adjust the P/E ratio by taking the difference between the client s P/E ratio and the mean P/E ratio for all firms within the same industry (two-digit SIC code). 5 In subsequent regressions, the P/E ratio is retained only if the individual sample firm P/E exceeds the industry, and enters the regression as the amount by which the firm P/E exceeds the industry interacted with positive discretionary accruals. Control Variables The research on the determinants of audit fees is well developed, and explanatory models have adjusted R 2 s in the percent range (Craswell et al. 1995; Simon and Francis 1988; Francis and Simon 1987). Recent research by Carson et al. (2003) confirms that audit fee models are generally well specified. Following the above studies we include control variables for size, complexity, health, governance structures, and auditor type. SIZE 5 Industry P/ E ratios were calculated by using all firms on Compustat for each two-digit SIC code.

6 90 Abbott, Parker, and Peters Variable Name Dependent Variable LNAFEE Test Variables DISCACC*INCR DISCACC*DECR TABLE 1 Definition of Variables Description The natural log of audit fees (000s). The absolute value of discretionary accruals times an indicator variable with the value 1 if the accruals are income-increasing, and 0 otherwise. The absolute value of discretionary accruals times an indicator variable with the value 1 if the accruals are income-decreasing, and 0 otherwise. P/ E*DISCACC*INCR Industry-adjusted PE ratio (only if in excess of industry-median P/ E ratio, and 0 otherwise) times the absolute value of discretionary accruals, if positive. Control Variables SIZE SQSUBS FORSUBS INVREC RECURLOS LEVERAGE GOINGCON PERCOSBD BORDMEET ACINDACT Natural log of total assets (in millions). Square root of the number of consolidated subsidiaries. Proportion of foreign subsidiaries to total subsidiaries. Proportion of total assets in inventory and accounts receivable. An indicator variable equal to 1 if the sample firm has experienced at least two annual net losses in the past three years. The sample firm s debt/ asset ratio in the sample year. An indicator variable equal to 1 if the firm received a going concern modification in the sample year, and 0 otherwise. The percentage of outside directors on the board. The number of full board meetings in the sample year. An indicator variable equal to 1 if the audit committee is fully independent, meets at least four times in the sample year, and includes at least one financial expert. is measured as the natural log of total assets (Compustat item 6). We control for client complexity by including the square root of the number of consolidated subsidiaries (SQSUBS) and the proportion of foreign subsidiaries to total subsidiaries (FORSUBS, both from firm 10-Ks). INVREC measures the proportion of total assets in inventory (Compustat item 3) and accounts receivable (Compustat item 2). We use an indicator variable (RE- CURLOS) to control for client health. RECURLOS is coded 1 if the client has experienced a loss (Compustat item 18) in two of the prior three years. LEVERAGE (the sample firm s debt/asset ratio, using Compustat items 9 and 6) is used to measure the client s business risk, related to their financial structure and debt level. Finally, we include an indicator variable (GOINGCON) coded 1 if the client received a going-concern modification in the sample year, and 0 otherwise (Compustat item 149 and the firm s 10-Ks). Recent U.S. audit market research has documented several significant associations between board/audit committee characteristics and audit fees. Boards and audit committees that are more independent of management are hypothesized to have a greater demand for audit scope and quality, resulting in higher audit fees (Carcello et al. 2002, Abbott et al. 2003; and others). We measure board composition (PERCOSBD) and activity (BORD- MEET) using the proportion of non-employees on the full board and the number of board meetings held in the sample year, respectively. Following Abbott et al. (2003), we code ACINDACT as 1 when the audit committee is entirely independent, includes at least one

7 Earnings Management, Litigation Risk, and Asymmetric Audit Fee Responses 91 expert, and meets at least four times per year. 6 All corporate governance information was collected from the sample firms proxy statements. RESULTS Descriptive Statistics Table 2 (Panel A) shows the distribution of sample firms by focus industry, and the comparison with the Compustat population. Thirty-three percent of the observations are manufacturing firms (SIC codes ), and 27 percent are in consumer products and TABLE 2 Sample Selection Results Panel A: Distribution of Observations by Focus Industry Focus Industry Corresponding Two-Digit SIC Codes Number of Observations (%) Compustat Population (%) a Construction (1.4) 1.1 Consumer products and food (26.8) 19.3 Energy 10 14, 46, (10.0) 9.3 Financial Services b 60 64, 67 2 (0.5) Information and Communication 48, 73, 78, 79, (15.4) 24.4 Manufacturing (32.6) 24.3 Personal services, healthcare 72, 80, 83 4 (0.9) 2.3 Commercial services, education 75, 76, 82, 87, 89 8 (1.9) 3.1 Real Estate 65, 70 3 (0.7) 1.7 Retail and Wholesale (6.3) 10.4 Transportation 40 42, 44, 45, (3.0) 2.4 All other 1, 2, 7, 8, 99 2 (0.5) 1.6 Total Firms (rounded) Panel B: Distribution of Observations by Auditor Number of Auditor Observations (%) Compustat Population (%) c Andersen 88 (20.5) 20.4 Deloitte & Touche 69 (16.1) 16.0 Ernst & Young 95 (22.1) 22.5 KPMG Peat Marwick 71 (16.6) 16.8 PricewaterhouseCoopers 106 (24.7) 24.3 Total a Calculated based upon all firms in the active Compustat database, excluding financial services firms. b These firms are financial information services firms such as Dow Jones and Moody s, which are not subject to the same regulatory environment as banks and other financial services companies. c Calculated based upon all firms in the active Compustat database, excluding those firms lacking auditor information, and those audited by non-big 5 auditors. 6 Our audit committee measures control for the possibility of an omitted, correlated variable. To wit, Xie et al. (2003) find that companies with effective audit committees exhibit lower discretionary accruals.

8 92 Abbott, Parker, and Peters food (SIC codes ), while the population includes 24 percent and 19 percent, respectively. Panel B shows the distribution of observations by auditor. The representation of these firms in the sample mirrors that of the overall population. Table 3 provides descriptive statistics for the sample. Mean (median) audit fees were $831,200 ($308,500). The mean (median) estimated discretionary accruals are incomedecreasing for sample firms at ( 0.005) and the raw price-earnings ratio has a mean (median) value of (11.75). The industry-adjusted P/E has a mean (median) of (2.21). The mean (median) firm size measured by assets is $4.6 billion ($573 million). Firms exhibit a mean of square root of subsidiaries of 3.9. The mean proportion of foreign subsidiaries to total subsidiaries is.18. The mean INVREC score was Thirty-one percent of firms experienced a loss in two of the three years preceding the sample year. The mean level of leverage is 21 percent. Ten percent of firms received a going-concern modification in the prior year. Sixty-nine percent of the directors of the full boards of sample firms are independent, and the mean (median) board meetings is 6.9 (6.0). Forty percent of the firms meet our joint threshold for expert presence, independence, and activity of the audit committee. 7 Regression Results Multivariate regression results are presented in Table 4. The high R 2 of the regression (79% percent) suggests a good model fit. 8 The coefficient estimates on all of our test variables are statistically significant, providing support for our hypotheses. We find that audit fees are significantly lower (higher) when the client firm s discretionary accruals are more income-decreasing (increasing) (p-values.001 and 0.024). 9 In economic terms, the dollar impact of moving from the 20th to 80th percentile of the discretionary accrual distribution is approximately 5.5 percent of median audit fees. 10 To put our results into further perspective, we note the results of Francis and Wang (2005). These authors find that the mean, inflation-adjusted audit fee deviation from their prediction model is approximately 3.3 percent of audit fees. Within this framework, Francis and Wang (2005) find that audit clients used mandated audit fee disclosures to demand changes in subsequent audit fees i.e., those audit clients who were systematically overcharged (undercharged) in 2000 paid relatively lower (higher) audit fees in In other words, it appears that seemingly modest audit fee magnitudes (and similar in size to those in the current study) are considered economically significant by market participants. We also find that the effect of positive discretionary accruals has an incremental explanatory effect when interacted with the industry-adjusted P/E ratio (p 0.053, twotailed), confirming our hypothesis that perceived earnings management is of greater concern (or perhaps more visible to auditors) for clients facing greater litigation risk resulting from performance expectations that exceeds industry norms. Although the economic significance 7 Eighty-one percent of firms included at least one financial expert, 77 percent were entirely comprised of independent directors, and 58 percent held at least four meetings in the sample year. 8 In terms of regression diagnostics, our highest documented pairwise correlation was VIF scores revealed no problems (all scores 2). The calculated index was According to Belsley et al. (1980), a condition index of 5 10 indicates weak dependencies. Therefore, it appears that multicollinearity is not a problem. Our Breuch-Pagan statistic was 10.15, suggesting that heteroscedasticity is not an issue. 9 Our empirical results differ from the Australian results of Gul et al. (2003), who find a significant relation only in income-increasing subsamples. 10 Discretionary accrual values at the 20th and 80th percentiles were and , respectively. The increase in audit fees resulting from moving between the 20th and 80th percentiles represents approximately $17,013 or 110 additional audit staff hours, if one assumes a billing rate of $150/audit staff hour.

9 Earnings Management, Litigation Risk, and Asymmetric Audit Fee Responses 93 TABLE 3 Descriptive Data Variable Name Mean Median Standard Deviation 25th percentile 75th percentile AUDFEE (in $000s) DISCACC P/E (raw) P/E (industry-adjusted) ASSETS (in millions) 4, SQSUBS FORSUBS INVREC RECURLOS LEVERAGE GOINGCON PERCOSBD BORDMEET ACINDACT Variables are as defined in Table 1, with the exception of ASSETS (in millions). The natural log of total assets is utilized in regression analysis. of the coefficient is a matter of judgment, the results provide limited empirical support for H2. The majority of control variables are also significant. Consistent with prior research, the coefficient estimates on SIZE, INVREC, SQSUBS, and RECURLOS are positive and significant, as are corporate governance variables. Interestingly, we do not find support for two of our risk measures, GOINGCON and LEVERAGE. The shift away from the joint and several liability regime toward one of proportionate liability, combined with the defensive role played by the audit qualification (Carcello and Palmrose 1994), may explain the lack of significance for our GOINGCON variable. Menon and Williams (2001) also fail to find a relation between leverage and audit fees and posit that leverage may not be an adequate proxy for an auditor s litigation risk. Sensitivity Analysis We conduct a number of sensitivity tests (not reported) to confirm the robustness of our results. Prior research indicates that audit fees increase as a nonlinear (concave) function of size. Thus, we partitioned our samples into two subsamples based on company size. Regression results for both subsamples were qualitatively similar to those reported in Table 4. We also include additional variables from previous research in the primary and sizepartitioned regression tests. Our results are qualitatively unchanged when we include ROA, growth in total assets, an indicator variable signifying a litigious industry (Carcello and Palmrose 1994), and an indicator variable for whether the firm has a December 31 yearend. When we include inside ownership (Gul et al. 2003) and interact inside ownership with our test variables, our test variables continue to be significant at conventional levels. We also include measures of restructuring, acquisitions, and debt placement and find our primary results unchanged.

10 94 Abbott, Parker, and Peters Variable Name TABLE 4 Regression Results (n 429) LNAFEE b 0 b 1 DISCACC*INCR b 2 DISCACC*DECR b 3 P/E*DISCACC*INCR b 4 SIZE b 5 SQSUBS b 6 FORSUBS b 7 INVREC b 8 RECURLOS b 9 LEVERAGE b 10 GOINGCON b 11 PERCOSBD b 12 BORDMEET b 13 ACINDACT e Expected Sign Parameter Estimate t-statistic p-value Intercept DISCACC*INCR DISCACC*DECR P/E*DISCACC*INCR SIZE SQSUBS FORSUBS INVREC RECURLOS LEVERAGE GOINGCON PERCOSBD BORDMEET ACINDACT Adjusted R Model F-stat p Two-sided p-values are reported. Variables defined in Table 1. We also investigate the effect of additional auditor characteristics. First, we control for auditor tenures of 1, 2, or 3 years, since new engagements may result in lower fees. Second, we drop each of the Big 5 auditors from the sample in turn, to test whether any one auditor has an excessive influence on results. Third, some previous research has found a relation between nonaudit fees and audit fees, and thus we include nonaudit fees as an independent variable. Finally, we control for a potential industry-specialist fee premium (Craswell et al. 1995). None of these tests changes our results qualitatively. In terms of the homogeneity of our audit fee model across industries, we performed three sensitivity tests. First, consistent with Carcello et al. (2002), we included an industry dummy variable for each of the one-digit SIC codes per Table 2. Second, we segregated our sample into only those observations in the manufacturing sectors (SIC codes ). Third, we ran separate regressions for the Consumer Products and Food and Manufacturing focus industries. We acknowledge that our industry categories may introduce classification error. In all three sets of tests, our results remain qualitatively similar to those reported in Table 4. We test the robustness of the results to alternate definitions of discretionary accruals. We used the Kothari et al. (2005) estimation method to calculate performance-matched discretionary accruals and found results similar to those found in Table 4. Similar results were found when using the Hribar and Collins (2002) accruals estimation technique. Our

11 Earnings Management, Litigation Risk, and Asymmetric Audit Fee Responses 95 results remain qualitatively unchanged when we winsorize discretionary accrual values at the 95 percent and 99 percent levels. We also substitute the estimated discretionary accruals in 1999 for the year 2000 estimation, since it is unclear whether any audit fee effect might be due to auditors assessing risk based on the earnings management history. We fail to document any relation between prior-year accruals and current-year audit fees, suggesting that auditors respond to client accounting discretion during the course of the audit. The lack of a relation between prior-year accruals and current-year audit fees suggests a scenario consistent with our observed results. In particular, discussions with several Big 5(4) audit partners and managers indicate that final audit fees are often influenced by ongoing negotiations that relate to audit scope revisions that occur over the course of the audit. 11 Post-planning hours spent on an audit area may often change depending on the emerging analysis of the financial statement account, most likely during the fieldwork stage of the audit. 12 We believe that audit-scope revisions arising from fieldwork-stage analysis will follow the conservative bias model proposed by Barron et al. (2001). More specifically, those accounts that appear to be potentially overstating income during the fieldwork stage will compel senior audit personnel to recommend increases in audit scope. Those accounts that appear to be potentially understating income during the fieldwork stage are less likely to be the subject of increases in proposed audit scope. We believe this is due to the auditor s recognition that overstated income poses a greater threat to the auditor s expected loss function. As discussed in our Limitations section, our research design cannot distinguish between this explanation and other possible scenarios. Nonetheless, this description is generally consistent with our results and representations made to us by Big 5(4) audit partners and managers. Limitations Our dependence on a purely ex post research design creates several issues that merit discussion. First, we cannot definitively ascertain how much of our documented incremental audit fee response appears in the planning stage (with a concomitant increase in audit fieldwork) or fieldwork (discovery) stage of the audit. Given our discussion with auditors and the failure to document a relation between prior-year accruals and current-year audit fees, we believe that meaningful fee changes occur during the fieldwork stage. Second, we cannot determine whether our audit fee response to income-increasing accruals is due to a risk premium or an increase in audit effort. We believe it is most likely the latter as several recent studies have utilized audit fees as a proxy for audit effort (Abbott et al. 2003; Carcello et al. 2002). Third, our analyses cannot disentangle whether auditors are, indeed, doing sufficient work for income-decreasing accrual clients (as opposed to underauditing ), while simply doing more work or charging a higher risk-related fee premium for income-increasing accrual clients. Fourth, we cannot establish what, if any, adjustments were made to pre-audit discretionary accruals as a result of the audit. This weakness is common to any research design relying on publicly available audited financial 11 These revisions do not necessarily entail formally altering original risk assessments, but often involve supplementary audit procedures and/ or increased sample sizes to obtain additional audit coverage. Conversely, in cases of income-decreasing discretionary accruals, auditors may reduce sample sizes or rely more heavily on analytical procedures. Given that audit purchasers are often former Big 5 (4) personnel (Carcello et al. 2002), such purchasers are likely cognizant of the reduced audit coverage and would seek to negotiate a reduction in audit fees. 12 Das and Shroff (1996) find that firms performing poorly (well) in interim quarters increase (decrease) fourth quarter earnings to achieve a desired annual earnings target. The authors conclude the majority of earnings management transpires during the fourth quarter, which occurs after the audit-planning stage.

12 96 Abbott, Parker, and Peters statements. We believe prospective research design refinements addressing these limitations are avenues for future research. There are at least two other sample-related limitations to our paper, as well. First, our industry classification scheme and sample period restrictions may introduce sampling risk, thus hindering the generalizability of our results. Second, the exclusion of non-big 5 auditors from our tests may exacerbate such generalizability concerns. CONCLUSION This purpose of this study was to examine the impact of income-increasing versus income-decreasing earnings management on auditor behavior. Using discretionary accruals as our proxy for earnings management risk and concomitant litigation risk, we find that audit fees decrease (increase) with income-decreasing (increasing) discretionary accruals. We also find that the increase in audit fees for positive discretionary accruals is magnified for high P/E firms. We attribute these findings to a conservative bias on the part of auditors. This bias arises from asymmetric litigation outcomes concerning incomedecreasing versus income-increasing earnings management. Although this evidence is at odds with the GAAS-based audit risk model, it is consistent with an emerging stream of conservative bias literature, which provides evidence that the audit risk model does not adequately capture auditor behavior concerning audit planning and investment (Lee and Mande 2003; Hodge et al. 2002; Barron et al. 2001). For example, Houston et al. (1999) find that, in the presence of accounting irregularities, the auditor s assessment of litigation risk or the risk of loss or injury to an auditor s professional practice due to client relationships dominates the traditional elements of the audit risk model in the explanation of audit investment. From an academic research perspective, DeFond and Francis (2004) note the need for research to address the role and importance of litigation in maintaining high audit quality. To the extent income-increasing versus income-decreasing discretionary accruals proxy for auditor s assessment of higher versus lower litigation risk, our asymmetric findings suggest that the litigation risk component of the auditor s production function helps preserve high audit quality. However, litigation could also have detrimental effects on overall financial reporting quality to the extent that auditors attention over-emphasizes income-increasing earnings management vis-à-vis income-decreasing earnings management. As such, audits may not reflect the growing concerns exhibited by the SEC concerning income-decreasing earnings management such as Cookie-Jar reserves. Since our tests cannot completely disentangle the possible detrimental effects of litigation, we believe further research in these areas would be warranted. REFERENCES Abbott, L. J., S. Parker, G. F. Peters, and K. Raghunandan The association between audit committee characteristics and audit fees. Auditing: A Journal of Practice & Theory 22 (2): Auditing Standards Board (ASB) Audit Risk and Materiality in Conducting and Audit. Statement on Auditing Standards No. 47. New York, NY: ASB. Barron, O., J. Pratt, and J. D. Stice Misstatement direction, litigation risk, and planned audit investment. Journal of Accounting Research 39 (December): Bedard, J. C., and K. M. Johnstone Earnings management risk, corporate governance risk, and auditors planning and pricing decisions. The Accounting Review 79 (April):

13 Earnings Management, Litigation Risk, and Asymmetric Audit Fee Responses 97 Belsley, D. A., E. Kuh, and R. E. Welsch Regression Diagnostics Identifying Influential Data and Sources of Collinearity. New York, NY: John Wiley & Sons. Carcello, J. V., and Palmrose, Z-V Auditor litigation and modified reporting on bankrupt clients. Journal of Accounting Research 32 (Supplement): 1 30., D. R. Hermanson, T. L. Neal, and R. R. Riley Board characteristics and audit fees. Contemporary Accounting Research 19 (Fall): Carson, E., R. Simnett, B. Soo, and A. M. Wright A longitudinal investigation of the audit and non-audit service fee markets (1984 to 1999). Working paper, The University of New South Wales. Craswell A. T., J. R. Francis, and S. L. Taylor Auditor brand name reputations and industry specializations. Journal of Accounting and Economics 20 (December): Das, S., and P. Shroff Fourth quarter reversals in earnings changes and earnings management. Working paper, University of Illinois at Chicago. Dechow, P. M., R. G. Sloan, and A. P. Sweeney Detecting earnings management. The Accounting Review 70: , S. A. Richardson, and I. A. Tuna Are benchmark beaters doing anything wrong? Working paper, University of Michigan., and D. J. Skinner Earnings management: Reconciling the views of accounting academics, practitioners, and regulators. Accounting Horizons 14 (June): DeFond, M., and C. W. Park Smoothing income in anticipation of future earnings. Journal of Accounting and Economics 23 (July): , and J. Francis Does auditing matter? Working paper, University of Southern California. Francis, J. R., and D. T. Simon A test of audit firm pricing in the small client segment of the U.S. audit market. The Accounting Review 62 (January): , and D. Wang Impact of the SEC s public fee disclosure requirement on subsequent period fees and implications for market efficiency. Auditing: A Journal of Practice & Theory (Supplement): Gul, F. A., C. J. P. Chen, and J. S. L. Tsui Discretionary accounting accruals, managers incentives and audit fees. Contemporary Accounting Research 20 (3): Heninger, W. G The association between auditor litigation and abnormal accruals. The Accounting Review 76 (January): Hodge, F., R. D. Martin, and J. H. Pratt Qualified accounting changes and investor assessments of financial performance and representational faithfulness. Working paper, Indiana University. Houston, R. W., M. F. Peters, and J. H. Pratt The audit risk model, business risk, and auditplanning decisions. The Accounting Review 74 (July): Hribar, P., and D. W. Collins Errors in estimating accruals: Implications for empirical research. Journal of Accounting Research 40 (March): Jones, J. J Earnings management during import relief investigations. Journal of Accounting Research 29 (Autumn): Kothari, S. P., A. J. Leone, and C. E. Wasley Performance-matched discretionary accrual measures. Journal of Accounting and Economics 39 (1): Lee, H. Y., and V. Mande The effect of the Private Securities Litigation Reform Act of 1995 on accounting discretion of client managers of Big 6 and non-big 6 auditors. Auditing: A Journal of Practice & Theory 22 (March): Menon, K., and D. D. Williams Long-term trends in audit fees. Auditing: A Journal of Practice & Theory 20 (March): Palmrose, Z-V., V. J. Richardson, and S. W. Scholz Determinants of market reaction to restatement announcements. Journal of Accounting and Economics 73 (1): , and S. W. Scholz The circumstances and legal consequences of non-gaap reporting: Evidence from restatements. Contemporary Accounting Research 21 (Spring): Pratt, J., and J. D. Stice The effects of client characteristics on auditor litigation risk judgments, required audit evidence, and recommended audit fees. The Accounting Review 69 (October):

14 98 Abbott, Parker, and Peters Richardson, S. A., I. A. Tuna, and M. Wu Predicting earnings management: The case of earnings restatements. Working paper, The University of Pennsylvania. Simon, D., and J. R. Francis The effects of auditor change on audit fees: Test of price cutting and price recovery. The Accounting Review 63 (October): Simunic, D The pricing of audit services: Theory and evidence. Journal of Accounting Research 18 (Spring): , and M. T. Stein The impact of litigation risk on audit pricing: A review of the economics and the evidence. Auditing: A Journal of Practice & Theory 15 (Supplement): Skinner, D. J., and R. G. Sloan Earnings surprises, growth expectations, and stock returns: Don t let an earnings torpedo sink your portfolio. Review of Accounting Studies 7 (2): Stice, J. D Using financial and market information to identify pre-engagement factors associated with lawsuits against auditors. The Accounting Review 66 (July): Subramanyam, K. R The pricing of discretionary accruals. Journal of Accounting and Economics 22 (1 3): Xie, B., W. N. Davidson III, and P. J. DaDalt Earnings management and corporate governance: The role of the board and the audit committee. Journal of Corporate Finance 9:

15