Abnormal Audit Fee and Audit Quality

by Auditing: A Journal of Practice & Theory

Understanding the factors that lead auditors to compromise on audit quality is an important issue of concern to scholars, investors, and regulators. One possible factor that has received significant research attention is economic bonding between the auditor and client. The basic idea in these studies is that positive abnormal audit fees reflect the extent of economic bonding between the auditor and client, and greater economic bonding degrades audit quality by impairing auditor independence. Based on this premise, studies have examined the linkage between audit quality and positive abnormal audit fees, with such studies documenting a negative association.

In this study, we reexamine the association between abnormal audit fees and audit quality, and we do so in a way that allows our paper to offer two important contributions to the literature. First, we incorporate into our conceptual framework insights about client bargaining power (Casterella et al. 2004) in addition to the economic bonding story that forms the conceptual framework in prior studies. Expanding the framework to consider both bargaining power and economic bonding allows us to offer a novel prediction that would not be meaningful within a framework based exclusively on economic bonding. That prediction is: audit quality will decline as negative abnormal audit fees increase in magnitude, and the magnitude of the decline will increase as proxies for client bargaining power increase. The economic bonding literature that informs prior studies places the focus on positive abnormal audit fees, predicting no association between negative abnormal audit fees and audit quality, and a negative association between positive abnormal audit fees and audit quality.

Second, we include in our analysis data taken from the post- Sarbanes-Oxley (SOX) period (i.e., years 2004-2009), in addition to data from the pre-SOX period (i.e., years 2000-2003). The pre-SOX period has been the data source for virtually all of the prior studies. With data taken from both the pre- and post-SOX periods, we are able to probe for a dampened association between abnormal audit fees and audit quality that would be manifest if SOX reforms meaningfully increased auditor independence or strengthened the auditor's bargaining power, leading to higher audit quality. To date, there is only limited empirical evidence that speaks to the issue of whether SOX reforms increased audit quality.

Consistent with prior research (e.g., Choi et al. 2010; Hope et al. 2009; and Higgs and Skantz 2006), we decompose total audit fees into normal and abnormal components, and test for an association between audit quality and abnormal audit fees, conditioning our tests on the sign of the abnormal audit fee metric (i.e., above- normal audit fees and below-normal audit fees). Under the bargaining power story, we expect to find that audit quality declines as negative abnormal fees increase in magnitude, with the effect amplified as proxies for client bargaining power increase. Under the economic bonding story (and consistent with prior studies), we expect to find that audit quality declines as positive abnormal fees increase in magnitude. We also partition our analysis by regulatory regime (i.e., a pre-SOX reporting period or a post-SOX reporting period) in order to assess whether the sensitivity of audit quality to abnormal audit fees differs between these two regimes. Our two proxies for audit quality are absolute discretionary accruals and meeting or beating analysts' earnings forecasts, and our two proxies for client bargaining power reflect the importance of the client to the local practice office.

Our tests produce evidence consistent with both the economic bonding story and the bargaining power story. As predicted by the economic bonding story, we find that absolute discretionary accruals and the probability of meeting or beating earnings forecasts both increase as positive abnormal audit fees increase. As predicted by the client bargaining power story, we find that absolute discretionary accruals and the probability of meeting or beating earnings forecasts increase with the magnitude of negative abnormal audit fees, with the effect amplified as client bargaining power increases. With respect to the effects of SOX, we find that the effects of economic bonding and client bargaining power are both dampened in the post-SOX regime, suggesting that SOX was effective in enhancing auditor independence.

The evidence presented in our paper is important in at least two respects. First, our results suggest that investors, regulators, and others interested in assessing the effects of auditor remuneration on audit quality should be concerned with both above-normal and below-normal auditor remuneration, but for different reasons. The potential effect of above-normal audit fees in degrading audit quality by economically bonding the auditor with the client is well recognized and extensively investigated. Less recognized is the possibility that below-normal audit fees signal important nuances in the balance of power between the auditor and the client, and that such power may ultimately influence audit quality. Thus, our study highlights the importance of considering the bargaining power of the client when assessing audit quality. Second, our study presents the first evidence (of which we are aware) to suggest that reforms introduced by the Sarbanes-Oxley Act dampened the deleterious effects of economic bonding and client power on audit quality, and hence increased audit quality. Both of these important insights from our study increase understanding of the factors that may lead auditors to compromise on audit quality, and hence our paper should be of interest to accounting scholars, investors, and regulators.

The remainder of the study is organized as follows. The second section develops our theoretical framework and presents the hypotheses that we test. We present our research design in the third section, followed by a discussion of our sample in the fourth section and our results in the fifth section. We offer concluding comments in the sixth section.


An audit firm is not a single person ''auditor''; rather, it is a decentralized organization in which individual audit partners act as agents for the audit firm (Liu and Simunic 2005). To the extent the partnership profit-sharing plan does not effectively align the interests of the partner with that of the audit firm partners as a whole, an uncontrolled moral hazard problem exists that might result in an individual audit partner succumbing to client pressure for earnings management. This is because the individual partner captures a significant portion of the expected benefit from acquiescing to client demands, while passing to the partnership as a whole the expected cost (Trompeter 1994).

What factors might lead an audit partner to compromise on audit quality? As described below, engagement profitability and client bargaining power are two possible explanations.

Engagement Profitability and Economic Bonding

Engagement profitability should influence audit quality for the following reason. Audit startup costs and client switching costs allow the auditor to price audit services at a price in excess of the avoidable cost of producing the audit, and thus create for the incumbent auditor clientspecific quasi-rents (DeAngelo 1981a, 1981b). The client-specific quasi-rents economically bond the auditor to the client, reduce auditor independence, and increase the likelihood that the auditor will acquiescence to a client's demand for earnings management. However, succumbing to client pressure risks audit firm forfeiture of some or all of the quasi-rents from the firm's entire client portfolio (if the earnings management is discovered), and additional economic loss through litigation and government penalties (DeAngelo 1981a, 1981b). The auditor will compromise audit integrity only if the expected gain (preserving the client-specific quasi-rent) exceeds the expected loss (forfeited quasi-rents from the overall client portfolio, litigation costs, and penalties), and thus the question of whether economic bonding undermines audit quality depends upon the relative magnitude of expected costs and benefits, which is an empirical question.1

Bargaining Power

Bargaining power should influence audit quality for the following reason. Audited financial statements, and hence audit quality, are the joint effort of the auditor and the client (Antle and Nalebuff1991) that arise from a process of negotiation between the two (Gibbins et al. 2001). The negotiation literature shows that when negotiators differ in bargaining power, the more powerful party expects greater concessions (e.g., Pruitt and Carnevale 1993; Hornstein 1965; Michener et al. 1975), and Barnes (2004) shows that audit quality may decrease as client bargaining power increases. In an experimental auditing setting, Hatfield et al. (2008) show that the effect of client bargaining power on the audited financial statements depends upon the negotiation strategy employed by the auditor, with a reciprocity negotiating strategy leading to more conservative financial statements. Thus, the question of whether client bargaining power undermines audit quality depends upon whether the auditor is able to employ a negotiating strategy that weakens the advantage held by a client with strong bargaining power. Abnormal Audit Fees

Simunic (1980) shows that the auditor's expected fee charged to the client is driven by the units of audit resources expended, the per-unit cost of those resources, and the auditor's expected future loss arising from the engagement (e.g., litigation losses, government penalties). Empirically, extant research models the expected audit fee as a function of observable factors that proxy for the auditor's cost in performing the audit, including auditor effort (i.e., resources expended and their cost), expected future litigation losses, and normal profit. If the audit fee model is well specified, the residual audit fee reflects abnormal profits from the audit engagement. To the extent that some factors are unobservable (and hence omitted from the audit fee model), the residual audit fee metric measures abnormal audit profitability with error.

Abnormal audit profitability should be associated with both client bargaining power and economic bonding. With respect to the former, Casterella et al. (2004) show a negative association between proxies for client bargaining power and audit fees earned by industry specialists. Their research suggests that, ceteris paribus, below-normal audit fees may reflect billing concessions granted by the auditor due to client bargaining power. With respect to the later, Kinney and Libby (2002) note that ''Unexpected fees may also better capture the profitability of the services provided . . . more insidious effects on economic bond may result from unexpected nonaudit and audit fees that may more accurately be likened to attempted bribes.''

Although there is scant evidence on the association between abnormally low audit fees and audit quality, a growing literature, described below, examines the association between abnormally high audit fees (as a proxy for economic bond) and audit quality. DeFond et al. (2002), Krishnan et al. (2005), Hoitash et al. (2007), and Hribar et al. (2010) test for a linear association between abnormal audit and/or engagement fees and audit quality (i.e., the curve relating audit quality to abnormal audit fees exhibits the same slope for both positive and negative abnormal audit fees).2 DeFond et al. (2002) find no association between abnormal engagement fees and auditors' going concern opinions during 2000-2001, while Krishnan et al. (2005) find that during year 2001, earnings response coefficients (a direct indicator of audit quality) decline as abnormal engagement fees increase. Hoitash et al. (2007) find during years 2000-2007 a positive association between abnormal engagement fees and two (inverse) audit quality metrics-the Dechow and Dichev (2002) accrual quality metric and the absolute value of performance- adjusted discretionary accruals. Hribar et al. (2010) find during years 2000 to 2007 a positive association between abnormal audit fees and accounting fraud, restatements, and SEC comment letters.3

Larcker and Richardson (2004), Higgs and Skantz (2006), Hope et al. (2009), Mitra et al. (2009), and Choi et al. (2010) test for a nonlinear association between abnormal audit fees and audit quality (i.e., the curve relating audit quality to abnormal audit fees exhibits different slope for positive as compared to negative abnormal audit fees). Larcker and Richardson (2004) use data from 2000 and 2001 to examine absolute discretionary accruals (an inverse indicator of audit quality), and find that audit quality increases as abnormal engagement fees increase in absolute magnitude. Higgs and Skantz (2006) find that during 2000-2002, earnings response coefficients (a direct indicator of audit quality) are greater in firms with positive abnormal audit fees. Hope et al. (2009) find that during 2000-2003, equity discount rates (an inverse indicator of audit quality) increase as positive abnormal engagement fees increase, but find no association with negative abnormal engagement fees. Mitra et al. (2009) find during years 2000-2005 a negative association between positive abnormal audit fees, and both absolute discretionary accruals and incomeincreasing accruals, but find no association between negative abnormal audit fees and discretionary accruals. Choi et al. (2010) find during 2000-2003 a positive association between positive abnormal audit fees and absolute discretionary accruals, but no association when abnormal audit fees are negative.

The weight of the preceding evidence suggests a negative association between audit quality and positive abnormal audit fees, and no association between audit quality and negative abnormal audit fees-findings consistent with the economic bonding hypothesis. However, as discussed above, negative abnormal audit fees may reflect client bargaining power that could degrade audit quality- with the degrading being larger in magnitude the greater the bargaining power of the client. Also as previously discussed, the question of whether client bargaining power undermines audit quality ultimately depends upon whether the auditor is able to employ a negotiating strategy that weakens the advantage held by a client with strong bargaining power. Thus, whether client bargaining power affects audit quality remains an open empirical question, leading us to specify and test the following client bargaining power hypotheses.

H1a: Audit quality will decline as below-normal audit fee increases in magnitude.

H1b: The association predicted in H1a will be amplified as proxies for client bargaining power increase.

For completeness, we also specify and test the following economic bonding hypothesis:

H2: Audit quality will decline as above-normal audit fees increase in magnitude.

Post-Sox Abnormal Audit Fees

In addition to the question of client bargaining power, another as yet unanswered question is whether the audit quality/abnormal audit fee association changed following SOX. Passed in 2002 following discovery of a series of high-profile financial reporting scandals, SOX seeks to improve corporate governance and enhance auditor independence by mandating federal government oversight of auditors, enhancing audit committee auditor oversight, and limiting the opportunity for auditors to sell nonaudit services to clients (U.S. House of Representatives 2002). If these reforms are sufficiently salient, they should manifest in a reduced association between abnormal audit fees and audit quality post-SOX relative to pre-SOX. This leads to our final hypothesis:

H3: The association between audit quality and abnormal audit fees will be attenuated in the post-SOX period as compared to the pre- SOX period.


To test our hypotheses, we need to measure abnormal audit fee and audit quality. We estimate abnormal audit fee (ABNAFEE) as the actual audit fee paid by the client to its auditor minus the predicted (normal) audit fee, with the difference deflated by the total audit fee revenue of the audit office conducting the client's audit.

We deflate the abnormal audit fee by total audit fee revenues of the practice office conducting the audit in order to capture the relative profitability of the engagement to the opining audit office. We do so since prior research (e.g., Reynolds and Francis 2001) suggests that economic incentives impacting audit quality are best measured at the local office level rather than at the national firm level. For example, the Enron audit failure largely stemmed from decisions made in the Houston office of Arthur Andersen (Chaney and Philipich 2002). We define two separate variables from ABNAFEE. If ABNAFEE . 0 then HIABNAFEE 1/4ABNAFEE, and 0 otherwise. If ABNAFEE , 0 then LOABNAFEE 1/4 jABNAFEEj, and 0 otherwise. This allows us to study the relationship of the dependent variables with the positive and negative abnormal audit fee, separately. The predicted audit fee is estimated from an audit fee model based on extant research.5 All the variables used in various tests are summarized in Table 1.

Audit quality is unobservable. Consistent with prior research, we define audit quality as the client's earnings quality (Higgs and Skantz 2006; Lim and Tan 2008; Davis et al. 2009; Francis and Yu 2009; Reichelt and Wang 2010; Choi et al. 2010). Following this research, we use two commonly used proxies for earnings quality: absolute discretionary accruals and propensity to meet or beat earnings expectations. We also conduct additional tests using an earnings response coefficient. The first two proxies are surrogates for actual earnings management, while the last one is related to perceived earnings quality. However, for the sake of brevity, we only report results for the first two, since the conclusions are similar. The tests for these proxies are discussed in more detail below.

Discretionary Accruals Model

The level of discretionary accruals has commonly been used as a surrogate for managers' exercise of discretion provide by GAAP. To the extent the discretionary component of accruals is used by managers to opportunistically manage earnings and auditors allow the manipulation to remain uncorrected, discretionary accruals adversely reflect on the audit and earnings quality (Schipper 1989; Jones 1991; Levitt 2000; DeFond and Park 2001). Discretionary accruals can be used for increasing or decreasing earnings depending on the incentives of managers. Since we are not looking at any specific managerial incentives, we have no directional predictions for accruals. We therefore use the absolute value of discretionary accruals (jDACCj) as the independent variable in our next test. Discretionary accruals (DACC) are calculated using the cross- sectional modified version of the Jones model (Jones 1991; Dechow et al. 1995), deflated by total assets, and estimated by year and for each industry. We adjust discretionary accruals for performance as suggested by Kothari et al. (2005). Following Hribar and Collins (2002), we use the difference between net income and cash from operations, deflated by lagged assets, as our measure of total accruals (TACC). Thus: TACC 1/4 [eth]IBC L OANCF[Thorn]=Lag[eth]AT[Thorn];

where IBC is the income before extraordinary items (Compustat cash flow item), OANCF is net cash flow from operating activities, and AT is total assets. The model to estimate discretionary accruals is:

... (1)

where Lag(AT) is total assets of prior year; DSALE is change in revenue; RECCH is the decrease in accounts receivables; PPEGT is property plant and equipment (gross total); and ROA is return on assets, calculated as IBC deflated by AT. Equation (1) is estimated by year for each industry (twodigit SIC code). Then, TACC minus the predicted value from the above regression is our measure of discretionary accruals.

Our test of jDACCj is based on the following model:

... (2)

The control variables are from extant research. Francis and Yu (2009) show that larger offices of Big 4 auditors have higher audit quality. The logarithm of total office-specific audit fee of all clients in a given year (LOFFICE) is included to capture this effect. Reynolds and Francis (2001) provide evidence that auditors report more conservatively for larger clients. Consistent with this research, the variable INFLUENCE, defined as the ratio of a client's total fee relative to the total annual fee of the practice office that audits the client, is included as an independent variable. TENURE (1/0 dummy variable for audit tenures of three years or less) controls for potential effect of short auditorclient association on audit quality (Johnson et al. 2002; Carey and Simnett 2006).

Balsam et al. (2003) argue that industry expertise increases audit quality. We include USLEADER and CITYLEADER, consistent with Francis and Yu (2009), to control for national-level and city-level auditor industry expertise. USLEADER is an indicator variable that is coded 1 if the/ auditor is the national audit fee leader in that industry. CITYLEADER is similarly defined at the city level. BUSSEG (GEOSEG) is the number of business (geographic) segments reported in the Compustat segment file. LOGMV is the natural logarithm of market value of equity at the end of the fiscal year. This variable controls for any size-related effects. Prior research (Ashbaugh et al. 2003; Butler et al. 2004; Menon and Williams 2004; Geiger and North 2006) finds LOGMV to be negatively associated to discretionary accruals. SGROWTH is the annual growth in sales and has been found to be positively related to discretionary accruals (Menon and Williams 2004).

CFFO is the cash flow from operations deflated by total assets. Previous researchers (Frankel et al. 2002; Chung and Kallapur 2003) find a negative association between discretionary accruals and CFFO. Hribar and Nichols (2007) find that accrual volatility may be related to firm-specific operating characteristics as measured by the volatility of the firm's cash flows and sales. Hence, we include SDSALES and SDCFFO (calculated as the standard deviations over the current and past four years of cash flow from operations and sales, respectively, deflated by the total assets) as control variables. Doyle et al. (2007) suggest that earnings quality may be a function of the quality of the firm's internal control. ICOPINION (number of material internal control weaknesses reported in Audit Analytics in the post-SOX period) is, therefore, included as a control variable.

Consistent with DeFond and Jiambalvo (1994), Reynolds and Francis (2001), and Francis and Yu (2009), LEVERAGE, LOSS, and DISTRESS are included to control for the effects of debt and financial distress. LEVERAGE is total debt deflated by total assets. LOSS is a dichotomous variable with a value of 1 if client has a negative net income before extraordinary items, and 0 otherwise. DISTRESS is Zmijewski's (1984) measure of financial distress. B2M is the book- to-market value and captures the growth opportunities. Ashbaugh et al. (2003), Butler et al. (2004), Menon and Williams (2004), and Geiger and North (2006) suggest B2M and DACC to be negatively associated. Following Matsumoto (2002), Hribar and Nichols (2007), and Francis and Yu (2009), we include stock-return volatility (VOLATILITY) to proxy for capital market pressures that can result in increased earnings management. FINANCED and ACQUIRED are dichotomous variables that have values of 1 if the company was involved in significant financing activities or acquisitions, respectively, and 0 otherwise. Ashbaugh et al. (2003) and Chung and Kallapur (2003) find positive coefficients on these two variables. LAGROA is the previous year's return on assets and is included to control for prior performance. BIG-N (a 0/1 dummy variable) and QUALIFIED (a 0/1 dummy variable) capture the effect of auditor size and qualified opinions on earnings quality. Natural logarithm of 1 [thorn] number of calendar days from fiscal year-end to date of auditor's report (LDELAY) controls for the effect of extra effort by the auditor on earnings quality. Since restatements can influence earnings management, the dichotomous variable RESTATEMENT is added as a control variable. MBEX is a dichotomous variable with a value of 1 if the firm meets or beats the earnings expectation (proxied by the most recent median consensus analysts forecast available on I/B/ E/S file) by two cents or less, and 0 otherwise. Since jDACCj and MBEX can be jointly determined, we use Maddala's (1988) two-stage- least-squares estimation (2SLS) for Equations (2) and (3) to avoid any endogeneity problems. I_MBEX denotes the instrumented variable for MBEX from the 2SLS.6

Meet-or-Beat Earnings Expectation Model

There is evidence in extant literature that managers are rewarded (penalized) for meeting (missing) earnings forecasts (Bartov et al. 2002; Kasznik and McNichols 2002; Lopez and Rees 2002) and this leads to incentives for managers to manage earnings. If auditors' incentives to curtail earnings management vary with abnormal audit fee, the propensity for meeting or beating analysts' earnings forecasts will be a function of the abnormal audit fee. We estimate the following logit model to test this conjecture:

... (3)

In Model (3), F(|) denotes the logistic cumulative probability distribution function. I_jDACCj denotes the instrumented variable for jDACCj from the 2SLS. Following Reichelt and Wang (2010), we include the standard deviation of analysts' earnings forecasts (SDFOR) and natural logarithm of number of analysts' forecasts (LNUMFOR) to control for characteristics of the forecasts.

Additional Tests

To test H1b, we further interact LOABNAFEE and HIABNAFEE with two measures of clients' bargaining powers in Models (2) and (3). LARGEST is a dichotomous variable with value of 1 if the client pays the highest audit fee in the practice office that audits the client; TOP10%INFL is a dichotomous variable with a value of 1 if INFLUENCE is in the top 10 percent, and 0 otherwise. To make the interpretations of the interactions easier, we use the dichotomous versions of abnormal audit fees, DLOABNAFEE and DHIABNAFEE, where DLOABNAFEE (DHIABNAFEE) is a dichotomous variable with a value of one if LOABNAFEE (DHIABNAFEE) - median value, and 0 otherwise.

To test our conjecture that SOX has dampened the association between abnormal audit fees and earnings management (H3), we rerun Models (2) and (3) with an additional independent variable DSOX and interaction terms, DLOABNAFEE * DSOX and DHIABNAFEE * DSOX, where DSOX is a dichotomous variable with a value of 1 for the period 2004- 2009, and 0 otherwise.7

Adjustment for Clustering

Since our data are pooled over time, the same firm may appear more than once in the sample, resulting in clustering. Clustered samples can lead to underestimation of standard errors and overestimation of significance levels (Cameron et al. 2011). We therefore estimate the t-statistics adjusted for clustering using robust standard errors corrected for firm-level clustering and heteroscedasticity, consistent with Petersen (2009) and Gow et al. (2010).


The sample selection procedure is outlined in Table 2. We start with 61,953 firm-year observations for the period 2000-2009 available in the Audit Analytics database for non-Andersen clients. We exclude Andersen since the auditor-client relationship might be atypical, given the woes of Andersen. After merging with Compustat we are leftwith 35,081 observations. We then exclude the financial (SIC codes 6000-6900) and utility (SIC codes 4400-4900) industries, since the incentives of managers from these regulated industries may be different. This leaves us with 28,925 observations. Missing data for estimating the variables in a modified Jones (1991) model and Equation (2) further reduce our sample size to 18,873 observations (Sample 1). Finally, I/B/E/S and CRSP data are needed for Equation (3), which results in Sample 2 with 14,796 observations.

Untabulated statistics show that the sample industry composition is very close to that of the Compustat population. Across years, untabulated statistics generally show an even distribution of observations. Sample characteristics, including measures of central tendency, are presented in Table 3. Mean (median) jDACCj and MBEX are 0.0655 (0.0418) and 0.2039 (0.0000), respectively. Mean LOABNAFEE (HIABNAFEE) is 0.0259 (0.0267). Both variables have median values , 0.0001. A typical client accounts for under 2 percent of the audit office's revenue (median value of INFLUENCE). A mean value of over 10 percent for this variable suggests some large influential clients. Over 23 percent of the clients have been with their auditors for three years or less. On average, firms have between two and three business and geographic segments. Mean (median) log of firm market value (in $ million) is 6.15 (6.14). Clients experienced a mean of 13 percent sales growth during the sample period. The median firm had one internal control weakness reported in Audit Analytics. On average, total debt was 19 percent of firm assets and 30 percent of the firm-years reported losses. Book value was around 59 percent of market value of the firm. While 31 percent of the firms financed during the year, 19 percent were involved in acquisition activities. As expected, a large proportion (80 percent) of the clients are audited by Big 4 auditors. RESULTS

Results of estimation for Models (2) and (3) are presented in Panel A of Table 4. For the sake of comparison, regressions are reported with and without control variables. The adjusted R2 of models with controls range from 18.39 percent to 20.39 percent.8 For the jDACCj regression (Equation (2)), 14 of the 26 control variables are significant (at 10 percent level or better). SGROWTH, SDSALES, CFFO, SDCFFO, LOSS, DISTRESS, VOLATILITY, FINANCED, and MBEX are significantly positive; LOGMV, LEVERAGE, B2M, LAGROA, and BIG-N are significantly negative. Thus, firms with more sales growth, sales volatility, cash from operations, cash volatility, losses, distress, stock price volatility, and financing activities are more likely to have higher jDACCj. There is also a positive association between jDACCj and MBEX. On the other hand, firms with larger size, higher leverage, lower growth potential, profitability, and with BIG-N auditors are more likely to have lower jDACCj.

Both LOABNAFEE and HIABNAFEE are significant and positive. This suggests that as the magnitude of ABNAFEE increases on either side (positive or negative), the magnitude of jDACCj increases. This supports H1a and H2.

In Model (3), since the dependent variable MBEX is a dichotomous variable, we use logistic regression to estimate the propensity to meet or beat earnings expectations. The results for Model (3) are exhibited in the last two columns of Panel A of Table 4. Twenty-one of the 28 control variables are significant: CFFO, FINANCED, ACQUIRED, SDFOR, and jDACCj are significantly positive; LOFFICE, INFLUENCE, USLEADER, BUSSEG, LOGMV, ICOPINION, LEVERAGE, LOSS, B2M, VOLATILITY, LAGROA, BIG-N, QUALIFIED, LDELAY, RESTATEMENT, and LNUMFOR are significantly negative. LOABNAFEE and HIABNAFEE are significantly positive. Thus, this test also supports H1a and H2.

Tests for H1b are reported in Panels B and C of Table 4. DLOABNAFEE and DHIABNAFEE are positive and significant as predicted by H1a and H2. DLOABNAFEE * LARGEST and DLOABNAFEE * TOP10%INFL are both positive and significant at a 5 percent level or better for jDACCj and MBEX. This suggests that highly influential firms with more bargaining power get more freedom to manage their earnings, for a given level of LOABNAFEE. This supports H1b. Moreover, DHIABNAFEE * LARGEST and DHIABNAFEE * TOP10%INFL are insignificant in Panel C, as expected, since the bargaining effect is expected on the below- normal audit fee side (firms getting discounts) and not on the above- normal audit fee side (firms paying premiums).

To test H3, we rerun Equations (2) and (3), with LOABNAFEE * DSOX and HIABNAFEE * DSOX added as additional independent variables (see Panel D of Table 4). Both of these interaction terms are significantly negative for Model (2), suggesting that the management of jDACCj postulated in H1a and H2 has dampened post-SOX, supporting H3. However, the terms LOABNAFEE [thorn] LOABNAFEE * DSOX and HIABNAFEE[thorn]HIABNAFEE * DSOX are both significantly positive at a 1 percent level, suggesting that the effects have not yet been completely erased as a result of SOX. On the other hand, estimation of Model (3) with interactions yields insignificant interactions, suggesting that the association of abnormal fees with MBEX is unaffected by SOX.

Additional Analyses with Earnings Response Coefficient

Lim and Tan (2008) use the ERC (earnings response coefficient) as a proxy for investors' perception of earnings quality and, in turn, audit quality. We run the ERC model with interactions of unexpected earnings (UE) with LOABNAFEE and HIABNAFEE. The independent variables in this model are based on Francis andKe (2006), Limand Tan (2008), Wilson (2008), and Ghosh et al. (2009). We do not report detailed results for the sake of brevity. However, coefficients of UE * LOABNAFEE and UE * HIABNAFEE are both negative, suggesting that the market perceives that audit quality declines with LOABNAFEE and HIABNAFEE, consistent with results reported for jDACCj and MBEX.

Tests of Robustness

We conduct several tests of robustness discussed below to convince ourselves that our results are not driven by any bias or model misspecification.

1. We first test for possible endogeneity bias in our audit fee model that potentially could introduce error into our measurement of ABNAFEE. The concern arises because Whisenant et al. (2003) suggest that audit and nonaudit fees are simultaneously determined, and thus not exogenous. To rule out this possibility, we run Davidson and MacKinnon's (1993) endogeneity test. The null hypothesis of no endogeneity is not rejected.

2. Unlike prior studies, we focus on audit fee revenue rather than nonaudit fee revenue as a source of impaired audit quality. We do so for two reasons. First, audit fee revenue is a recurring annuity whereas nonaudit fee revenue, with the exception of revenues from tax services, is not. Recurring engagements (as compared to nonrecurring engagements) create greater incentive for the auditor to compromise independence. Second, reforms enacted by the Sarbanes- Oxley Act of 2002 have significantly reduced the scope of nonaudit services that auditors can provide to their clients while also expanding the scope of work required in the audit. Together, these two factors suggest that in the current milieu, audit fees emerge as a potentially important determinant of audit quality.10 Recent research (Choi et al. 2010) has, therefore, focused on audit fees. However, as a robustness check, we repeat all of our tests with total engagement fee (i.e., the sum of audit and nonaudit fees) and arrive at similar conclusions.

3. We try an alternate specification for our models. Instead of the split-linear regression, we estimate the arctan regressions suggested by Freeman and Tse (1992) as below: ...

This specification yields interesting conclusions. The relationship of jDACCj with HIABNAFEE (LOABNAFEE) is S-shaped (inverse S-shaped). The flat curve around HIABNAFEE 1/4 LOABNAFEE 1/4 0 suggests that the auditors do not impair their independence until positive abnormal fee from a client exceeds what the auditor could earn by simply resigning from this client and contracting with a new one.11 Similarly, on the negative abnormal audit fee side, the auditor may have a pecking order. Initially they would grant their influential clients (with bargaining powers) fee discounts, and as the client gains more bargaining power, the auditor may also allow earnings management. The flattening of the curve for extreme departures from normal fee suggests that the auditors have a threshold or tolerance level for earnings management, and as the firm tries to report extremely high levels of discretionary, accruals the auditors start resisting.

4. Since audit fee data became available on Audit Analytics in 2000, one could question how the investors derive their ''normal'' audit fees to assess the audit quality in the year 2000. We rerun our analyses without the year 2000 data and our conclusions do not change qualitatively. For Model (2), the coefficients of LOABNAFEE and HIABNAFEE are 0.0251 (p 1/40.0049) and 0.0316 (p 1/40.0136), respectively; for Model (3), the coefficients of LOABNAFEE and HIABNAFEE are 1.9900 (p , 0.0001) and 1.2004 (p 1/4 0.0004), respectively.

5. Since quality of corporate governance may also affect earnings quality, as a sensitivity analysis we add the Gompers et al. (2003) index of corporate governance to our regressions and obtain similar results. We do not report these findings as our main results since the sample size after merging with the Gompers' database was considerably smaller.

6. To the extent that the first few years of the auditor-client relationship might be atypical, we run our tests excluding the first three years of new clients. The results are not altered qualitatively. For Model (2), the coefficients of LOABNAFEE and HIABNAFEE are 0.0153 (p 1/40.0016) and 0.0233 (p 1/4 0.0011), respectively; for Model (3), the coefficients of LOABNAFEE and HIABNAFEE are 1.3200 (p 1/40.0002) and 0.3991 (p 1/40.0813), respectively.

7. Firms that have consistent positive or negative abnormal fees across years might have different motivations from firms that switch between the two categories. We split the sample into firms that have five or more years with abnormal audit fees of the same sign and those that do not. Our hypotheses hold for both subsets, suggesting that our results are applicable to both categories of firms.

8. We try different cutoffs for the tenure variable to see if the results are sensitive to the definition of tenure. We try five years and also two dummy variables for three years or under and more than nine years. Our main conclusions are not affected.

9. The pricing of audit engagements for accelerated filers may be different from that of nonaccelerated filers, since accelerated filers could have a higher audit fee due to more internal control work. The pricing of material weaknesses may also be different. If the size variable in the audit fee model is unable to control for these effects, the conclusions might be biased. To ascertain that the results are not different for accelerated/non-accelerated filers, we run the tests separately for the two filing categories. We do not find any evidence of any category leading the results. For example, in Model (2) for accelerated filers, the coefficients of LOABNAFEE and HIABNAFEE are 0.0095 (significant at 1 percent) and 0.0054 (significant at 10 percent), respectively; for non- accelerated filers, the coefficients of LOABNAFEE and HIABNAFEE are 0.0147 (significant at 10 percent) and 0.0202 (significant at 5 percent), respectively. 10. As with any ratio, the scaling of abnormal audit fee by the total revenue of the audit office might introduce inflation (deflation) of the same numerator for smaller (larger) audit offices. To ascertain if this scaling affects our results, we rerun all the tests with unscaled abnormal audit fees. The coefficients of LOABNAFEE and HIABNAFEE continue to be positive and significant at a 5 percent level or better for both Models (2) and (3).

11. Prior research (e.g., Bedard and Johnstone 2004; Johnstone and Bedard 2001, 2003) suggests that proxies for audit risk used in extant research may be inadequate. There may be above-normal effort in cases of earnings manipulation risk and corporate governance risk (Bedard and Johnstone 2004) that are not captured by the audit fee model. This extra effort may result in an abnormally high audit fee and lead to higher audit quality. Extending this argument, below- normal audit fees may represent abbreviated audit efforts, and not clients' bargaining power, leading to poorer quality audits. In the absence of data on engagement hours and staffmix, we utilize the number of days between fiscal year-end and release of the audit report as a proxy for audit effort. We include this variable, square of the variable, and natural logarithm of the variable in our audit fee model, in an effort to control for auditor effort that is in response to missing audit risk factors. All of our conclusions are supported as before. Additionally, unless the missing measures for audit risk are correlated with bargaining power, the differential variation of audit quality with clients' bargaining power (documented in Panel C of Table 4) gives additional credibility to our results. However, to the extent that we are unable to control for audit risks in our model, the factors discussed above continue as limitations of our research.

12. In Panels B, C, and D of Table 4, we use logistic regressions for the dichotomous dependent variable MBEX, along with interaction terms. Norton et al. (2004) caution that since logistic regressions are nonlinear models, the interaction terms must be interpreted in a different way. They show that the marginal effect of a change in both interacted variables is not equal to the marginal effect of the change in just the interaction term. They suggest applying the INTEFF function in Stata for estimating the correct marginal effect of the interaction term. As a robustness check, we run the INTEFF function as detailed in Norton et al. (2004). Our conclusions are unchanged, suggesting that the results reported in Panels B, C, and D are robust. One constraint with the INTEFF procedure is that it works for only one interaction term. In Panel D, for the SOX tests, we have two interaction terms. We therefore run two separate tests with INTEFF for each interaction term.

In addition, to convince ourselves that our interpretation of the interaction effects is correct, we run the above regressions with MBEX replaced with a continuous variable, excess earnings defined as the reported earnings per share minus the analyst forecast. Since this dependent variable is not a dichotomous variable, we no longer have to use the logistic regression. We are able to replicate all the results reported before. This further provides assurance that our interpretation of the interaction terms is reliable.


Auditors are hired and compensated by their clients, and this creates an economic bond between the two. The question of whether this economic bond ultimately leads to reduced audit quality remains an important public policy issue and has been the topic of prior research. We extend this research by examining the role of client bargaining power in addition to economic bonding as a determinant of audit quality. In our study, we examine the association between abnormal audit fees and two audit quality proxies: (1) abnormal discretionary accruals and (2) the probability of meeting or beating analysts' consensus earnings forecasts.

Above-normal audit fees suggest quasi-rents arising from a highly profitable audit engagement, while below-normal audit fees suggest the client has strong bargaining power (and hence is able to negotiate billing concessions). Both factors-quasi-rents and client bargaining power-may lead the auditor to succumb to client pressure for earnings management. Thus, absolute abnormal audit fees reflect the presence of these independence-impairing underlying factors and is our variable of interest.

As hypothesized, we find that audit quality declines as actual audit fees depart from ''normal'' fee levels. Our finding that audit quality declines as positive abnormal audit fees increase in magnitude is consistent with prior research. New to the literature is our finding that audit quality declines as negative abnormal audit fees increase in magnitude, with the decline increasing in magnitude as proxies for client bargaining power increase.

We also examine whether the audit quality/abnormal audit fee association changed following implementation of SOX, and use this examination as a means of testing whether SOX was effective in increasing auditor independence. We find evidence that this association in the post-SOX era is milder than in the pre-SOX era. Thus, SOX reforms increase the risk of and reduce the gain from succumbing to client pressure and compromising audit quality.

It is important to note that our results regarding abnormally high or low audit fees could be attributable not to economic bonding or client bargaining power, but rather to factors that are not captured by our audit fee model. Alternative explanations that we cannot investigate due to data limitations include such items as audit team composition (i.e., the relative allocation of audit hours between partners, managers, and staff), the audit work allocation between interim and year-end, the influence of internal audit assistance, the relative quality of client financial reporting systems, and individual differential audit firm production functions. Accordingly, the reader should remain mindful of this limitation when interpreting our evidence.

Overall, our study increases understanding of the factors that may lead auditors to compromise on audit quality, and hence our paper should be of interest to accounting scholars, investors, and regulators.

Copyright American Accounting Association Aug 2012

(c) 2012 Auditing: A Journal of Practice & Theory. Provided by ProQuest LLC. All rights Reserved.

Related Articles