AN EMPIRICAL ANALYSIS OF AUDITOR INDEPENDENCE IN THE BANKING INDUSTRY Kiridaran Kanagaretnam DeGroote School of Business McMaster University 1280 Main Street West Hamilton, Ontario, Canada L8S 4M4 Phone: (905) 525-9140 ext 27857 Fax: (905) 521-8995 E-mail: giri@mcmaster.ca Gopal V. Krishnan Department of Accounting College of Business and Economics Lehigh University Bethlehem, PA 18015 Phone : (610)-758-3451 E-mail: gok208@lehigh.edu Gerald J. Lobo* C. T. Bauer College of Business University of Houston Houston, TX 77204-6021 Tel: (713) 743-4838 Fax: (713) 743-4828 E-mail: gjlobo@uh.edu October 2008 *Corresponding author We thank James Bierstaker, Jeff Chen, Chris Jones, Sok-Hyon Kang, Krishna Kumar, Ying Li, Lihong Liang, Jim Largay III, Erin Moore, Nandu Nayar, Dan Neeley, Mike Peters, H. Sami, Mary Sullivan, Bill Zhang and seminar participants at George Washington University, Hong Kong Polytechnic University, Lehigh University, McMaster University, University of Queensland, for their helpful suggestions. Kanagaretnam and Lobo thank the Social Sciences and Humanities Research Council of Canada (SSHRC) for its financial support. AN EMPIRICAL ANALYSIS OF AUDITOR INDEPENDENCE IN THE BANKING INDUSTRY Abstract We examine auditor independence in the banking industry by analyzing the relation between fees paid to the auditors of banks and the extent of earnings management through loan loss provisions. We also examine whether this relation differs across large banks whose managements are required under the Federal Deposit Insurance Corporation Improvement Act of 1991 to evaluate the internal control over financial reporting and whose auditors must attest to the report on the effectiveness of internal controls over financial reporting, and small banks that are not subject to such controls. Our results indicate a positive association between fees paid to the auditor and income-increasing earnings management through loan loss provisions. They suggest that, although banks face high levels of regulatory scrutiny, economic bonding between the auditor and the bank potentially impairs auditor independence. Our findings also indicate that this bonding is stronger for smaller banks that are subject to less regulatory oversight than are larger banks. Our results also suggest that the economic bond between the auditor and the bank is reflected in delayed recognition of loan write-offs and in a higher incidence of earnings benchmark beating behavior. Keywords: Auditor independence, earnings management, auditor fees, bank loan loss provision, FDICIA. 1 AN EMPIRICAL ANALYSIS OF AUDITOR INDEPENDENCE IN THE BANKING INDUSTRY I. INTRODUCTION Auditor independence is vital to maintaining public confidence in the capital markets and to the integrity of corporate financial statements. The objective of this study is to examine auditor independence in the banking industry. Banks represent more than 20% of the total public equity market and are vital to the functioning of the economy as a whole. Fields et al. (2004, p. 54) state, “Despite the economic importance of the banking industry, however, accounting researchers have done little to investigate the various relationships that exist between banks and their auditors.” Specifically, we provide empirical evidence on the relation between fees paid to auditors of banks and the extent of earnings management via loan loss provisions (LLP). The banking industry offers a unique context to study auditor independence for a number of reasons. First, banks are subject to the scrutiny of the FDIC, the Federal Reserve Board, and other governmental agencies. Whether this intense regulatory oversight enhances auditor independence is clearly an important and relevant question. Second, external audits are required for all national banks with $500 million or more in total assets.1 Additionally, the Federal Deposit Insurance Corporation Improvement Act of 1991 (FDICIA), which was passed in response to the savings and loan debacle of the 1980’s and became effective in 1992, imposed new auditing, corporate reporting, and governance reforms on each depository institution with assets exceeding $500 million and on its auditors (Murphy 2004). Section 112 of the Act requires the management of these institutions to evaluate the internal control over financial reporting and the auditor 1 The Office of the Comptroller of the Currency strongly encourages smaller national banks to have external audits performed by independent public accountants. Hawke (2000) reports that the vast majority of the smaller banks has established some type of external audit program. 2 must attest to the report on the effectiveness of internal controls over financial reporting. 2 Whether auditor independence is greater for larger banks that are subject to greater regulatory oversight relative to smaller banks is also an interesting and important question. Currently, there is very little empirical evidence on whether the reforms initiated by FDICIA contributed to enhanced auditor independence. Third, bank LLP is well-suited to studying earnings management for the following reasons. LLP is by far the largest and most important accrual for banks. The mean (median) ratio of LLP to earnings before LLP is 19.8% (15.1%) for our sample firms. Further, prior research indicates that banks use LLP to manage earnings (Wahlen 1994; Kanagaretnam et al. 2003; 2004). To the extent that banks can leverage “fee dependence” to influence their auditors to accept abnormal LLP, examining the relation between abnormal LLP and auditor fees is likely to reveal such a linkage. We believe that abnormal LLP is a better proxy for earnings management than the abnormal accrual measures used in prior research.3 This study mitigates error in measuring managerial discretion by focusing on a single accrual and a single industry. Focusing on a single accrual facilitates a sharper separation into its normal (nondiscretionary) and abnormal (discretionary) components. We use a number of industry-specific variables to better isolate the abnormal LLP from the normal LLP. Also, focusing on a single, relatively homogeneous industry provides control over other determinants of cross-sectional 2 However, FDICIA was a controversial legislation. While proponents of FDICIA argued that the reforms could improve the financial health of the banking industry through better monitoring, opponents argued that it is “regulatory overkill” (Kaufman 1994). 3 Prior research argues that measures of abnormal accruals commonly used to detect earnings management are subject to serious measurement error (Guay, Kothari, and Watts 1996; McNichols 2000; 2002; and Jones, Krishnan, and Melendrez 2008).3 For example, McNichols (2002, page 68) states, “the complexity associated with modeling the estimation errors in aggregate accruals is daunting, and the construct validity associated with a proxy based on aggregate accruals seems low.” 3 differences in accruals, thus increasing the reliability of the inferences from our empirical analysis. Our sample consists of 1,810 bank-year observations representing years 20002006. We estimate abnormal LLP as residuals from a regression of LLP on beginning loan loss allowance, total loans outstanding, changes in total loans outstanding, net loan charge-offs, beginning balance of non-performing loans, change in non-performing loans, loan mix, and control for years (see Kanagaretnam et al. 2004 and Wahlen 1994). Next, we estimate a regression of abnormal LLP on fee measures, an indicator variable for small banks (with total assets of $500 million or less) that are subject to less regulation, the interaction of the fee and the indicator variable for small banks, auditor type (Big 5 vs. non-Big 5 auditor), and several control variables. We use various measures of fees to capture an auditor’s economic dependence on the client, including audit fees, nonaudit fees, total fees, ratio of nonaudit fees to total fees (fee ratio), abnormal (unexpected) audit fees, abnormal nonaudit fees, and abnormal total fees. We report several key findings. First, we find that the level of total fee, nonaudit fee, and audit fee are negatively and significantly associated with negative (incomeincreasing) abnormal LLP for all banks. In other words, bank-year observations with higher fees exhibit higher income-increasing LLP. Similarly, total and audit fees are positively associated with positive (income-decreasing) abnormal LLP for all banks. However, when we use abnormal (unexpected) fee measures, we find a significant relation between abnormal total fee and abnormal nonaudit fee only with negative (income-increasing) abnormal LLP. Taken together, these results suggest that the economic bond (fee dependence) between the auditor and the bank appears to exacerbate earnings management by banks. 4 Second, in a regression of negative (income-increasing) abnormal LLP, the coefficient on the interaction of fee and the indicator variable for small banks is negative and significant for all four fee measures (i.e., total fee, nonaudit fee, audit fee, and fee ratio). We do not observe such a relation for positive (income-decreasing) abnormal LLP. These results indicate that small banks engage in more income-increasing earnings management and there is stronger economic bonding between the small banks and their auditors. The results also indicate that, while the reforms of the FDICIA are effective at reducing auditor bonding for large banks, they do not fully eliminate it. Third, we examine whether current-period auditor fees are associated with nextperiod bank loan charge-offs. We find that after controlling for size, current-period LLP, and current-period charge-offs, the level of total fee, nonaudit fee, and audit fee are all positively and significantly associated with next-period charge-offs. We obtain similar results when we use unexpected fee measures. Given that bank managers have incentives to postpone writing off bad loans to future periods so that the current period’s financial position is presented in a favorable manner, our results suggest that the economic bond between the auditor and the bank is associated with delayed recognition of loan chargeoffs. This finding also holds when we replace actual fees with unexpected fees. Finally, for the (income-increasing) abnormal LLP sub-sample, we examine the association between auditor fees and beating earnings benchmarks as do Frankel et al. (2002) and Ashbaugh et al. (2003). The results indicate a significant, positive relation between our proxy for benchmark beating behavior and three fee measures – total fee, nonaudit fee, and audit fee. We obtain similar results when we use unexpected fee measures. Our findings indicate that economic bonding between the auditor and the bank is 5 associated with income-increasing earnings management through loan loss provisions. They also indicate that this bonding is stronger for small banks. Collectively, our results suggest that auditor fee dependence on the audit client is a threat to auditor independence, particularly among banks with less than $500 million in assets that are not subject to the same level of regulatory scrutiny as larger banks. The rest of this paper is organized as follows. The next section develops the empirical models used to estimate abnormal LLP and test the relation between abnormal LLP and fees paid to auditors. Section three describes the sample selection, section four discusses the results, and section five provides the conclusions of the study. II. RESEARCH DESIGN Our conceptual model, described in Figure 1, is based on the predictive validity model of Kinney and Libby (2002) that was developed by Runkel and McGrath (1972) and Libby (1981). As in prior research, the economic bond to the audit client and earnings management by the client (link 1) are the two theoretical constructs we examine. We identify proxies to measure the economic bond and earnings management because these two concepts are unobservable. We use multiple fee measures to capture the economic bond (link 2). Kinney and Libby (2002) state that “more insidious effects on the economic bond may result from unexpected nonaudit and audit fees that may more accurately be likened to attempted bribes.” We use unexpected audit, nonaudit, and total fees as measures of the economic bond and the abnormal component of a bank’s LLP as the measure of earnings management (link 3). We examine banking firms because they provide a setting where a stronger link 3 than used in prior research is possible. By using an industry-specific measure of earnings management, we are better able to separate the discretionary accruals from the nondiscretionary accruals than has prior research. We 6 control for several bank characteristics that may affect LLP (link 4). We then examine the relation between the various fee measures and abnormal LLP (link 5) to draw inferences about link 1. We also conduct two supplementary tests of link 5. First, we examine the relation between the various fee measures and future loan charge-offs. Second, we study the extent to which banks use abnormal LLP to meet earnings benchmarks. We describe these tests in a later section. [Insert Figure 1 About Here] We use a two-stage approach to examine link 5. First, we describe the model used to estimate abnormal LLP (link 4). We first estimate the normal or nondiscretionary component of LLP by regressing LLP on beginning loan loss allowance, beginning balance of non-performing loans, change in non-performing loans, net loan charge-offs, changes in total loans outstanding, total loans outstanding, loan mix, and controls for period effects using the following model:4 LLPit = γ0 + γ1 BEGLLA + γ2 BEGNPL + γ3 CHNPL + γ4 LCO + γ5 CHLOANS + γ6 LOANS + <LOAN CATEGORIES> + <YEAR CONTROLS> + eit (1) We define the variables as follows (all variables are deflated by beginning total assets):5 LLP = Provision for loan losses; BEGLLA = Beginning loan loss allowance; BEGNPL = Beginning nonperforming loans; CHNPL = Change in nonperforming loans; LCO = Net loan charge-offs; 4 These variables have also been used in several prior studies (Wahlen 1994; Beaver and Engel 1996; and Kim and Kross 1998) to estimate the normal component of LLP. 5 We also use beginning total loans as an alternate deflator. Our results are not sensitive to this choice of deflator. 7 CHLOANS = Change in total loans outstanding; LOANS = Total loans outstanding; and LOAN CATEGORIES = amount of commercial loans (COMM), consumer loans (CON), real estate loans (RESTATE), agriculture loans (AGRI), loans to foreign banks and governments (FBG), and loans to other depository institutions (DEPINS). The residuals from model (1) are the abnormal component of LLP, referred to as ALLP. We expect a negative coefficient on BEGLLA (i.e., the accumulated LLP less write-offs at the beginning of the year) as a higher initial loan loss allowance will require a lower LLP in the current period. Consistent with prior research, we expect γ2, γ3, γ4 and γ6 to be positive. Higher levels of nonperforming loans indicate problems with the loan portfolio will require higher loss provisions. Therefore, the beginning balance of nonperforming loans (BEGNPL) will be positively related to LLP. Change in nonperforming loans (CHNPL) in the current period will also have a positive effect on LLP because an increase in nonperforming loans will require a higher loss provision in the current period. The amount of net loan charge-offs (LCO) is positively related to LLP. As noted in Beaver and Engel (1996), “current loan charge-offs can provide information about future loan charge-offs which, in turn, may influence expectations of the collectability of current loans” and, hence, current LLP. A higher level of loans (LOANS) will also require higher provisions. We do not offer a prediction for γ5 because the effect of change in total loan portfolio on LLP is unpredictable due to the uncertainty in the quality of incremental loans. Although nonperforming loans and loan charge-offs serve as measures of risk, we include six additional variables to control for differences in loan composition which also 8 likely contribute to differences in risk. For example, banks with higher proportions of commercial and real estate loans are likely to have higher loan loss provisions than banks with higher proportions of consumer loans. Failure to account for these differences makes the residuals from the loan loss provision model a function of the bank type. This, in turn, may affect the inferences of our auditor independence tests because both audit fees and ALLP may systematically vary across banks based on their loan portfolio mix. The six loan portfolio composition variables included in the model are commercial loans (COMM), consumer loans (CON), real estate loans (RESTATE), agriculture loans (AGRI), loans to foreign banks and governments (FBG), and loans to other depository institutions (DEPINS). We also include six year-dummy variables representing years 2000 through 2005 in model (1) to control for period-specific effects. Next, we test the association between the signed values of abnormal LLP (ALLP) and auditor fees separately for negative (income-increasing) ALLP and positive (incomedecreasing) ALLP.6 Negative ALLP are of particular interest because of their positive impact on reported earnings. We control for the following factors that prior research has documented are associated with abnormal accruals (Ashbaugh et al. 2003): firm size, auditor type, market-to-book ratio, level of accruals, and performance. We use log of market value of equity to measure size. We represent performance by two variables, existence of loss and earnings before LLP, and growth by market-to-book ratio. We use past LLP to capture the reversal of accruals over time. To control for any capital management incentives, we include the beginning of year tier 1 capital ratio and total 6 Hribar and Nichols (2007) argue that the research design choices using unsigned measures of earnings management heighten the threat of correlated omitted variables and recommend using signed discretionary accruals to study earnings management. 9 capital ratio.7 Our model is as follows: ALLP = β0 + β1 FEE + β2 BIG5 + β3 MB + β4 LMVE+ β5 LOSS + β6 PASTLLP + β7 EBP + β8 TIER1t-1+ β9 TCAPt-1 + <YEAR CONTROLS> + ε (2) We define the variables as follows: ALLP = Abnormal loan loss provision; FEE = Natural log of audit fees (LAFEE) or total fees (LTOTFEE = ln (audit fees + nonaudit fees)) or nonaudit fees (LNAFEE), or fee ratio (FEERATIO = nonaudit fees/ total fees); BIG5 = Indicator variable set equal to 1 if audited by a Big 5 firm, and 0 otherwise; MB = Market-to-book ratio at the end of the year; LMVE = Natural log of market value of common equity; LOSS = Indicator variable set equal to 1 if net income < 0, and 0 otherwise; PASTLLP = prior year’s LLP divided by total assets at the beginning of the year; EBP = Net income before extraordinary items and loan loss provisions divided by total assets at the beginning of the year; TIER1 = Tier 1 risk-adjusted capital at the beginning of the year; and TCAP = Total risk adjusted capital at the beginning of the year. The variable of interest in model (2) is FEE. For negative (income-increasing) ALLPs, a negative coefficient on β1 is consistent with auditor fee dependence on the client, i.e., higher fees are associated with greater (more negative) income-increasing 7 Banks had incentives to influence regulatory capital through LLP prior to the change in bank capital adequacy requirements in 1990. This change altered banks’ incentives to manage capital through LLP as loan loss allowance is no longer considered part of tier I or core capital. Therefore, we do not expect β 8 to be significantly different from zero. Furthermore, loan loss allowance is included in tier II or supplementary capital only up to 1.25 percent of risk-adjusted assets. Thus, we predict a negative coefficient on β9 as higher capital ratios would require a lower LLP, ceteris paribus. 10 ALLP. Ashbaugh et al. (2003) argue that total fee, rather than the fee ratio is the more appropriate measure of economic bonding.8 In addition to total fee we also use audit fee, nonaudit fee, and the fee ratio (nonaudit fees/ total fees). Next, we examine the impact of the FDICIA on auditor independence, particularly on auditors of smaller banks, i.e., banks with total assets less than $500 million. Recall that FDICIA imposed new auditing, corporate reporting, and governance reforms on each depository institution with assets exceeding $500 million and on its auditors. We modify model (2) to include an indicator variable SMALL which equals one, when beginning of the year total assets is lower than $500 million and also interact the FEE variable with SMALL. We discuss sample selection in the next section. III. SAMPLE SELECTION We identify our sample banks from banks listed in the 2007 Bank Compustat annual data files and obtain fees paid to the auditors for the years 2000-2006 from Audit Analytics. The intersection of the Audit Analytics with Bank Compustat data results in an initial sample of 2,044 bank-year observations. We hand collect non-performing loans data for the period 1999-2006 from annual reports and obtain data on loan portfolio composition from the Federal Reserve Bank Holding Company Database (call reports). Our final sample with available data for all variables comprises 1,810 bank-year observations for 304 banks. Panel A of Table 1 reports descriptive statistics for the scaled variables used in the regressions. More than 52% of the sample observations are audited by Big 5 auditors. Note that the mean abnormal LLP (ALLP) is zero by construction. Turning to the bank 8 Kinney and Libby (2002) indicate that both the audit fee and the nonaudit fee are capable of increasing the economic bond. They reason that an auditor might be willing to decrease the audit fee or the nonaudit fee to maximize the potential total fee. Therefore, it is the total fee derived from a single client, rather than just the audit fee, that better reflects the economic bond. 11 loan variables, the ratios of average LLP, loan charge-offs and beginning nonperforming loans to beginning total assets are 0.003, 0.002 and 0.004, respectively. 9 [Insert Table 1 About Here] Panel B of Table 1 reports correlations for the scaled dependent and independent variables. As expected, LLP is positively correlated with non-performing loans (BEGNPL), change in non-performing loans (CHNPL) and loan charge-offs (LCO). Also, LLP is positively correlated with total loans (LOANS) and change in loans outstanding (CHLOANS). The correlations between LLP and nonaudit fees and fee ratio are positive and statistically significant at the 0.01 level while the correlations between LLP and total fee and audit fee are not significant. IV. RESULTS Estimation of Abnormal LLP We report the estimation results of model (1) in Table 2. The t-statistics reported in Table 2 and in other tables are based on standard errors adjusted for firm level clustering. The results in Table 2 show that the coefficients on the determinants of LLP have the expected signs. The coefficients on BEGLLA, CHNPL, LCO and LOANS are significant at the 0.01 level and the coefficient on BEGNPL is significant at the 0.10 level. Among the variables that reflect loan type, only commercial and consumer loans 9 These ratios are lower than the values reported in Wahlen (1994) and Kanagaretnam et al. (2004) due to an increase in overall quality of loans (Edwards and Mishkin 1995). Similarly, our sample data indicate that mean LLP equals 20 percent of earnings before provisions, considerably lower than the 148 percent reported for Wahlen’s (1994) sample and lower than the 31 percent reported for Kanagaretnam et al.’s (2004) sample. 12 are statistically significant. The explanatory power of the model is high (adjusted R2 = 63.15%) indicating that our model describes the variation in LLP quite well.10, 11 [Insert Table 2 About Here] Association Between Income-increasing (negative) Abnormal LLP and Fee Measures The results of model (2) relating abnormal LLP to the fee measures (total fee, nonaudit fee, audit fee, and fee ratio) are reported in Tables 3 and 4. Table 3 reports the results for negative (income-increasing) ALLP and Table 4 reports the results for positive (income-decreasing) ALLP. We first discuss the results in Table 3. For each of the four fee measure, we present the results for two specifications. The first specification does not distinguish between small and large banks, while the second specification distinguishes between small and large banks by addding the variable SMALL, which equals 1 for small banks (with total assets less than $500 million) and 0 for large banks, and its interaction with the fee variable to model (2). This specification examines whether the association between ALLP and fees is greater for small banks than for large banks. Recall that the FDICIA imposed new auditing, corporate reporting, and governance reforms on each depository institution with assets exceeding $500 million and on its auditors. Therefore, small banks are subject to less scrutiny than large banks, which may result in more pronounced fee dependence for small banks relative to large banks. 10 Our model fit compares favorably with the results of prior research that focus on total accruals rather than on a specific accrual as we do. For example, McNichols (2002) reports that when change in working capital accruals is regressed on changes in sales and property, plant, and equipment, the adjusted R 2 for the pooled sample is 7.3%. Ashbaugh et al. (2003) report mean R 2 values of 11% and 18% for their models of abnormal accruals. 11 We also use an alternate specification where the dependent variable is LLP net of LCO. The adjusted R2 for this specification is 0.23. As before, the following variables continue to be significant at the 0.10 level or better: BEGLLA, BEGNPL, CHNPL, LOANS, COMM, and CON. When we use ALLP obtained from this specification in model (2), our results for negative ALLP indicate that the coefficient on the fee measure for the main effect is not significant for any of the four fee measures. However, the coefficient on the interaction of SMALL and the fee measure is negative and significant at the 0.001 for total fee, nonaudit fee, and fee ratio (for a two-tailed test), and at the 0.05 level for audit fee. 13 The results indicate that three of the four fee measures (LTOTFEE, LNAFEE, and LAFEE) are negatively and significantly (p < 0.01) associated with ALLP, indicating that our results are not sensitive to the definition of the fee measure. Overall, the results indicate that abnormal LLP, our measure of earnings management, is more negative (i.e., more income-increasing) for banks that pay higher fees to their auditors. In terms of the control variables, LMVE, PASTLLP, and MB are significant. Interestingly there is no difference in abnormal LLP between clients of BIG5 and non-BIG5 auditors. As expected, the capital ratios are not significantly associated with ALLP, confirming the reduced capital management incentives through ALLP in the post-1990 period. Turning to the second specification, the results for the main effect (fee variable) continue to hold for total fee (p < 0.10, one-tailed test), nonaudit fee, and audit fee. More importantly, the coefficient on the interaction of SMALL and the fee measure is negative and significant at the 0.05 level for total fee, nonaudit fee, and fee ratio. Furthermore, the sum of the coefficients on FEE and FEE×SMALL for total fee and nonaudit fee is significant at the 0.01 level. Overall, these results support the notion that small banks engage in greater earnings management via income-increasing LLPs relative to large banks, and suggest that the potential impairment of auditor independence is more serious for small banks than for large banks. [Insert Table 3 About Here] Association Between Income-decreasing (positive) Abnormal LLP and Fee Measures The results reported in Table 4 indicate that two of the four fee measures (log of total fees (LTOTFEE) and log of audit fees (LAFEE) are positively and significantly associated with ALLP at the 0.01 level. Nonaudit fee is marginally significant at the 0.10 14 level for a one-tailed test. The interesting finding in Table 4 is that the coefficient on the interaction of SMALL and the fee measure is not significant for any of the four fee measures. Recall that the interaction of SMALL and the fee measure is significant for total fee, nonaudit fee, and fee ratio for income-increasing ALLP. These findings suggest that auditor fee dependence is more problematic for small banks relative to large banks. Further, note that after controlling for the effect of SMALL, LTOTFEE and LAFEE have a stronger positive association with ALLP, indicating that the associations between LTOTFEE and LAFEE and income-increasing ALLP are more pronounced for large banks. [Insert Table 4 About Here] We conduct several sensitivity checks of model (1). First, Ahmed et al. (1999) suggest that the earnings management through LLP reported in prior research is conditional on the inclusion of beginning non-performing loans (NPLt-1) in the discretionary LLP model. When we re-estimate model (2) for income-increasing ALLP after excluding NPLt-1 in model (1), LTOTFEE, LNAFEE and FEERATIO are significant at the 0.01 level and LAFEE is significant at the 0.10 level. Second, we assess the sensitivity of our results to extreme values. We re-estimate model (1) after deleting observations in the top 1% and bottom 1% for each variable (LLP, CHLOAN, LCO, BEGNPL and CHNPL). This reduces the number of observations from 1,810 to 1709 and the adjusted R2 is 68.94%. The test results for this reduced sample are consistent with those reported in Table 3. Third, we scale the variables in model (1) by beginning total loans instead of beginning total assets. Once again, our results for income-increasing ALLP are robust to the choice of scaling variable. LTOTFEE and LNAFEE are significant 15 at the 0.01 level, and LAFEE and FEERATIO are significant at the 0.05 level. Fourth, we estimate model (1) separately for small and large banks. The results based on this specification indicate that negative ALLP are significantly associated with all four fee measures for small banks. The results are weaker for large banks. Unexpected Fee Measures Auditor fees are also influenced by various economic determinants including size and complexity of audit task. Therefore, controlling for the economic determinants of auditor fees and using unexpected (abnormal) auditor fees may be a better measure of auditor fee dependence than using actual fees. For example, Kinney and Libby (2002) argue that unexpected fees are a better measure of the auditor-client economic bond because they reflect the “excess” profit derived from an audit client. Unexpected fees are estimated in two steps. First, we estimate the expected (normal) fees as the predicted values from a regression of audit fee or nonaudit fee or total fee on a set of firm characteristics. We then compute the residuals from this regression which represent the unexpected fees. Prior research models auditor fees as a function of a firm’s auditor choice, audit complexity, and audit risk, in addition to other variables (Firth 1997 and Ashbaugh et al. 2003). Fields et al. (2004) examine the determinants of normal audit fees in the banking industry. Using the variables identified in Fields et al. (2004) as determinants of audit fees, we estimate unexpected fees for three of our fee measures (LAFEE, LTOTFEE and LNAFEE). Audit fees and other fees are likely to be higher when the auditor is a Big 5 auditor. Auditor size also proxies for client size. We measure firm size as the natural log of total assets. The normal audit fee is directly related to a bank’s credit risk, operating risk, liquidity risk and capital risk. We include NPL, LCO, COMM, CON and RESTATE 16 as proxies for a bank’s credit risk and the efficiency ratio (EFFICIENCY) as a proxy for operating risk. We measure the efficiency ratio as the ratio of total operating expenses to total revenues. As in Fields et al. (2004), we use SECURITIES as a proxy for liquidity risk, and intangible assets (INTANG) and total capital ratio (TCAP) to account for capital risk. We estimate the following model:12 FEE = α0 + α1 BIG5 + α2 LASSETS + α3 SECURITIES + α4 NPL + α5 LOSS + α6 INTANG + α7 EFFICIENCY + α8 LCO+ α9 COMM + α10 CON + α11 RESTATE + α12 TCAP + <YEAR CONTROLS> + ε (3) We define the variables as follows: FEE = Natural log of audit fees (LAFEE) or total fees (LTOTFEE = ln (audit fees + nonaudit fees)) or nonaudit fees (LNAFEE); BIG5 = Indicator variable set equal to 1 if audited by a Big 5 firm, and 0 otherwise; LASSETS = Natural log of total assets; SECURITIES = [1-(total securities/total assets)]; NPL = Nonperforming loans over lagged total loans; LOSS = Indicator variable set equal to 1 if ROA < 0, and 0 otherwise; INTANG = Intangible assets over total assets; EFFICIENCY = Total operating expenses over total revenues; LCO = Net loan charge-offs over loan loss allowance; COMM = Total commercial and agriculture loans over total loans; CON = Total consumer loans over total loans; RESTATE = Total real estate loans over total loans; and 12 As in Fields et al. (2004), this model includes three loan categories. We also estimate unexpected fees after including all the six loan categories used in model (1). The relations between ALLP and UFEE are qualitatively the same when we use all six loan categories to estimate abnormal fees. 17 TCAP = Total risk-adjusted capital ratio. We also use alternate specifications of model (3) where we include audit fees (nonaudit fees) as a control variable when we estimate unexpected nonaudit fees (audit fees). Those results are discussed in a later section. Panel A of Table 5 reports the results of estimating model (3). As expected, we find a positive relation between our three fee measures and BIG5, LASSETS, SECURITIES, NPL, EFFICIENCY, TCAP and COMM. The signs of the coefficients are generally consistent with Fields et al. (2004). For the audit and total fee models, the adjusted R2 are, respectively, 82.52% and 74.99%, indicating a very good fit. For the nonaudit fee model the adjusted R2 is 51.45%. These R2 values are higher than the adjusted R2 values reported in Ashbaugh et al. (2003) for industrial firms and consistent with the R2 values reported in Fields et al. (2004). [Insert Table 5 About Here] We use the residuals from model (3) as the unexpected (abnormal) fees for our three fee measures. We then use the following model to examine whether banks’ abnormal accrual choices are associated with abnormal fees paid to the auditors: ALLP = χ0 + χ1UFEE + χ2 BIG5 + χ3 MB + χ4 LMVE+ χ5 LOSS + χ6 PASTLLP + χ7 EBP + χ8 TIER1t-1+ χ9 TCAPt-1 + <YEAR CONTROLS> + ε (4) We define the variables as follows: ALLP = Abnormal loan loss provision; UFEE = Unexpected audit fees or total fees or nonaudit fees from model (3); 18 BIG5 = Indicator variable set equal to 1 if audited by a Big 5 firm, and 0 otherwise; MB = Market-to-book ratio at the end of the year; LMVE = Natural log of market value of common equity; LOSS = Indicator variable set equal to 1 if net income < 0, and 0 otherwise; PASTLLP = prior year’s LLP divided by total assets at the beginning of the year; EBP = Net income before extraordinary items and loan loss provisions divided by total assets at the beginning of the year; TIER1 = Tier 1 risk adjusted capital at the beginning of the year; and TCAP = Total risk adjusted capital at the beginning of the year. Model (4) is identical to model (2) except that we substitute UFEE for FEE where UFEE is unexpected fees, i.e., the residuals from model (3). Again, we use three measures of unexpected fees, total fees, audit fees, and nonaudit fees. Panel B of Table 5 reports the results of model (4) for income-increasing (negative) ALLP. The relation between income-increasing ALLP and unexpected total fee and unexpected nonaudit fee are both positive and significant at the 0.05 level. There is no significant relation between income-increasing ALLP and unexpected audit fee. Overall, the results in panel B of Table 5 along with the results in Table 3 provide strong evidence that income-increasing ALLP is higher for those audit clients that pay higher fees to their auditors.13 To examine whether the strength of the association between unexpected fees and income-increasing ALLP is greater for small banks, we add the variables SMALL and 13 When unexpected nonaudit fees are estimated after controlling for audit fees, the relation between negative ALLP and unexpected nonaudit fees is negative and significant at the 0.01 level. 19 SMALL× UFEE to model (4). We find that the coefficient on the interactive term SMALL×UFEE is more negative for both unexpected total fee and unexpected nonaudit fee. Also, the sum of the coefficients on UFEE and SMALL×UFEE is negative and significant at the 0.01 level for unexpected total fee and unexpected nonaudit fee, a result which confirms the finding in Table 3 that income-increasing earnings management is higher for those audit clients of small banks that pay higher fees to their auditors. We also examine the relation between ALLP and UFEE for income-decreasing (positive) ALLP. Untabulated results indicate that none of the three abnormal fee measures is related to income-decreasing ALLP. This is also true when we introduce controls for small banks. Analysis of Next Period Loan Charge-Offs As an additional test we examine the relation between fees paid to auditors in the current period and the next period’s loan charge-offs. Bank managers do have some discretion in recognizing loan charge-offs (Wahlen 1994) and might postpone charging off bad loans to future periods so that the current period’s financial position is presented in a favorable manner. We model next-period loan charge-offs as a function of fees paid to auditors, current period LLP, current-period loan charge-offs (LCO) and the natural log of total assets, a proxy for bank size. In addition, we include the indicator variable SMALL to control for differences between small and large banks. Following Altamuro and Beatty (2006), we use LLP, LCO, SMALL and the natural log of total assets as control variables in the model explaining future charge-offs which is formulated as follows LCOt+1 = χ0 + χ1 FEEt + χ2 SMALLt + χ3 LLPt + χ4 LCOt + χ5 SIZEt-1 + <YEAR CONTROLS> + ε 20 (5) We define the variables as follows (all variables are deflated by beginning total assets): LCO = Net loan charge-offs; FEE = Natural log of audit fees (LAFEE) or total fees (LTOTFEE = ln (audit fees + nonaudit fees)) or nonaudit fees (LNAFEE); SMALL= 1 if beginning total assets are less than $500 million; LLP Provision for loan losses; and = SIZE = Natural log of beginning total assets. Panel A of Table 6 reports the results of model (4) for the full sample. As expected, all three of our fee variables (LTOTFEE, LAFEE and LNAFEE) have a significant (p<0.01), positive relation with next-period loan charge-offs (FLCO). The two control variables, current-period loan loss provisions (LLP) and loan charge-offs (LCO), have a strong, positive association with FLCO. [Insert Table 6 About Here] In Panel B of Table 6, we repeat our analysis using unexpected fees estimated from model (3). Again, all three of the unexpected fee measures (UTOTFEE, UAFEE and UNAFEE) have strong, positive relations with next-period loan charge-offs (FLCO). Overall, the significant association between current-period fees and future loan chargeoffs suggests that bank managers delay recognition of loan charge-offs when the auditorclient economic bond is stronger. These results are consistent with our earlier results on the association between ALLP and fees in that auditors appear more tolerant of earnings management via income-increasing LLP or discretionary loan charge-offs when their fees are higher. 21 Earnings Benchmark Tests Frankel et al. (2002) and Ashbaugh et al. (2003) examine the association between several fee measures and earnings benchmark beating behavior of firms. Following prior research, we investigate the association between auditor fees and the likelihood of banks reporting a small earnings increase (INCREASE) for the income-increasing ALLP subsample.14 We estimate the following logit regression for the earnings benchmark tests: INCREASEt = χ0 + χ1 FEEt + χ2 SMALLt + χ3 ALLPt + χ4 MBt + χ5 LMVEt + χ6 BIG5t + χ7 TIER1t-1+ χ8TCAPt-1 + <YEAR CONTROLS> + ε (6) We define the variables as follows: INCREASE = 1 when the change in net income scaled by beginning of year assets falls in the interval [0.000, 0.002] and 0 otherwise. FEE= Natural log of audit fees (LAFEE) or total fees (LTOTFEE = ln (audit fees + nonaudit fees)) or nonaudit fees (LNAFEE); SMALL= 1 if beginning total assets are less than $500 million; ALLP = Abnormal loan loss provision; MB= Market-to-book ratio at the end of the year; LMVE= Natural log of market value of common equity; BIG5= Indicator variable set equal to 1 if audited by a Big 5 firm, and 0 otherwise; TIER1= Tier 1 risk adjusted capital at the beginning of the year; and TCAP= Total risk adjusted capital at the beginning of the year. The variable of interest in model (6) is FEE. A positive association between FEE 14 A parallel test for bench mark beating is examining the relation between fees and meeting or just beating the analyst forecast. However, for our income-increasing ALLP sub-sample, only 107 of the 724 bank-year observations fall in the meeting or just beating the analyst forecast group. Given this small number of observations, we focus our benchmark beating tests on meeting or beating the preceding year’s earnings. 22 and INCREASE will indicate that auditor fee dependence increases the bank’s earnings management behavior to meet or narrowly beat the previous year’s earnings. Since small banks have less regulatory monitoring, we expect a positive association between SMALL and INCREASE. Following Ashbaugh et al. (2003), we include current period abnormal accruals (ALLP) in the model. More negative ALLP for the income increasing sub-sample will result in a higher probability of beating the earnings benchmark; therefore, we expect a negative association between ALLP and INCREASE. Consistent with Frankel et al. (2002) and Asbaugh et al. (2003), we include controls for growth (MB), size (LMVE) and Big 5 auditors (BIG5). We also include beginning capital ratios as bank specific controls. Panel A of Table 7 reports the results of model (6) for the income increasing ALLP sub-sample. As expected, all three of the fee variables (LTOTFEE, LAFEE and LNAFEE) are significantly, positively related with bank’s earnings management behavior to meet or narrowly beat the previous year’s earnings. The indicator variable for small banks (SMALL) has a positive and significant (p<0.05) relation with INCREASE. As predicted ALLP is negatively associated with INCREASE and this relation is significant at the 0.01 level. [Insert Table 7 About Here] In Panel B of Table 7, we repeat our analysis using the unexpected fees from model (3). Again, all three of the unexpected fee measures (UTOTFEE, UAFEE and UNAFEE) have a strong, positive relation with earnings benchmark beating behavior. Overall, the significant association between both fees and abnormal fees and earnings benchmark beating behavior is consistent with the argument that auditors who are paid higher fees are more tolerant of earnings management via income-increasing LLP. 23 V. SUMMARY AND CONCLUSIONS We examine auditor independence in the banking industry. Banks are subject to the scrutiny of the FDIC, the Federal Reserve Board, and other governmental agencies. In addition, the Federal Deposit Insurance Corporation Improvement Act (FDICIA) which became effective in 1992, requires the management of depository institutions with assets exceeding $500 million to evaluate the internal control over financial reporting and the auditor to attest to the report on the effectiveness of internal controls over financial reporting. We provide empirical evidence on the relation between fees paid to the auditor and the extent of earnings management via loan loss provisions in the banking industry. We also study whether this relation differs across small and large banks. Our findings indicate a positive association between fees paid to the auditor and income-increasing earnings management through loan loss provisions. They suggest that economic bonding between the auditor and the bank potentially impairs auditor independence. This result is especially interesting given the high level of regulatory scrutiny faced by banks. Our findings also indicate that this bonding is stronger for small banks that are subject to less regulatory oversight than large banks. Collectively, our results suggest that auditor fee dependence on the audit client is a threat to auditor independence, particularly among banks with less than $500 million in assets that are not subject to the same level of regulatory scrutiny as larger banks. They suggest that although the high level of regulatory oversight of banks does not eliminate economic bonding, it does so to a greater extent for large banks subject to more stringent controls. Our results have several implications. First, they suggest that the FDIC and other banking regulators and inspectors should closely review the loan loss provisions of banks where the fee dependence is high. Second, that banks can engage in earnings 24 management despite the high level of scrutiny by the FDIC and other regulatory agencies, suggests that the extent of earnings management in industries that are less closely regulated could be even greater. Third, the FDICIA can be viewed as a precursor to the Sarbanes-Oxley Act of 2002 for large banks. There has been considerable discussion about relaxing the requirements of the Sarbanes-Oxley Act for small firms. If generalizable to other industries, our results of increased earnings management by small banks that are less closely regulated suggest that reducing the requirements of the Sarbanes-Oxley Act for small firms should be approached with caution. The results of two additional tests corroborate our findings of an economic bond between the auditor and the bank. First, we find a significant, positive association between the level of total fee and the next-period charge-offs indicating that the fee level is positively related to delaying the recognition of loan charge-offs. Second, we observe a significant, positive relation between various fee measures and benchmark beating behavior. We note that one limitation of our study is that the results could be driven by an alternative hypothesis that the audit fees reflect audit risk that is captured by the discretionary loan loss provision. Auditors are likely to charge higher fees to firms that are more difficult to audit and, if firms that are more difficult to audit have higher incentives to engage in earnings management, this will be manifested in a positive relationship between audit fees and discretionary loan loss provisions. We attempt to address this concern by including variables that reflect differences in audit risk across banks as well examining the relationship using abnormal fees after explicitly taking account of possible factors driving the normal fees. Nevertheless, we cannot completely rule out this alternative explanation and recognize it as a limitation of our study. 25 References Ahmed, A.S., C. Takeda, S. Thomas. 1999. Bank loan loss provisions: a reexamination of capital management, earnings management and signaling effects. Journal of Accounting & Economics (November): 1-25. Altamuro, J., A.L. Beatty. 2006. Do internal control reforms improve earnings quality? Working paper, Ohio State University, available on the SSRN. Ashbaugh, H., R. LaFond, B. Mayhew. 2003. Do nonaudit services compromise auditor independence? Further evidence. The Accounting Review 78 (July): 611-639. Beaver, W.H., and E. Engel. 1996. Discretionary behavior with respect to allowances for loan losses and the behavior of security prices. Journal of Accounting & Economics 22 (August-December): 177-206. Edwards, F.R., and F.S. Mishkin. 1995. The decline of traditional banking: Implications for financial stability and regulatory policy. Federal Reserve Bank of New York Economic Policy Review 1 (2): 27-45. Fields, L.P., D.R. Fraser, M.S. Wilkins. 2004. An investigation of the pricing of audit services for financial institutions. Journal of Accounting and Public Policy 23 (January/February): 53-77. Firth, M. 1997. Provision of nonaudit services by accounting firms to their clients. Contemporary Accounting Research (Summer): 1-21. Frankel, R., M. Johnson, K. Nelson. 2002. The relation between auditors’ fees for nonaudit services and earnings management. The Accounting Review 77 (Supplement): 71-105. Guay, W.R., S.P. Kothari, R.L. Watts. 1996. A market-based evaluation of discretionary accrual models. Journal of Accounting Research (34): 83-105. Hawke, J. 2000. Statement before the SEC. http://www.occ.treas.gov/ftp/release/200057b.txt Hribar, P., and D. C. Nichols. 2007. The use of unsigned earnings quality measures in tests of earnings management. Journal of Accounting Research (45): 1017-1053. Jones, K.L., G.V. Krishnan, K. Melendrez. 2008. Do discretionary accruals models detect actual cases of fraudulent and restated earnings? An empirical evaluation. Contemporary Accounting Research (Summer): 499-531. Kanagaretnam, K., G. Lobo, R. Mathieu. 2003. Managerial incentives for income smoothing through loan loss provisions. Review of Quantitative Finance and Accounting (20): 63-80. 26 _____, _____, D. Yang. 2004. Joint tests of signaling and income smoothing through bank loan loss provisions. Contemporary Accounting Research (Winter): 843-84. Kauffman, G. 1994. “FDICIA: the early evidence.” Challenge 37: 53-58. Kim, M., and W. Kross. 1998. The impact of the 1989 change in bank capital standards on loan loss provisions and loan write-offs. Journal of Accounting and Economics 25 (1): 69-99. Kinney, W.R. Jr., R. Libby. 2002. Discussion of the relation between auditors’ fees for nonaudit services and earnings management. The Accounting Review (Supplement): 107114. Libby, R. 1981. Accounting and Human Information Processing: Theory and Applications. Englewood Cliffs, NJ: Prentice Hall, Inc. McNichols, M. F. 2000. Research design issues in earnings management studies. Journal of Accounting and Public Policy 19 (4-5): 313-345. _____. 2002. Discussion of the quality of accruals and earnings: the role of accrual estimation errors. The Accounting Review 77 (Supplement): 61-69. Murphy, C.W. 2004. “Goodbye FDICIA, Hello Sarbox – Sort of.” PricewaterhouseCoopers Publications. Runkel, P., and J. McGrath. 1972. Research on Human Behavior – A Systematic Guide to Method. New York, NY: Holt, Rinehart, and Winston, Inc. Wahlen, J.M. 1994. The nature of information in commercial bank loan loss disclosures. The Accounting Review (July): 455-478. 27 FIGURE 1 Conceptual Model Independent Variable Concepts Dependent Variable Economic Bond to Client 1 2 Operational Measures Earnings Management 3 Multiple Fee Measures 5 Abnormal Loan Loss Provision 4 Bank characteristics affecting independent and dependent variables 28 TABLE 1 Panel A: Variables Used in Regressions (N = 1,810) Variable Mean Std. Dev Minimum Median Maximum LTOTFEE 12.6539 1.1814 10.50 12.50 16.42 LAFEE 12.2121 1.1629 9.81 12.07 15.75 LNAFEE 11.2175 1.5063 7.85 11.07 15.96 FEERATIO 0.3153 0.2044 0.0105 0.2691 0.8413 LLP 0.0028 0.0032 -0.0009 0.0021 0.0131 ALLP 0.0000 0.0017 -0.0033 -0.0001 0.0052 LOANS BEGLLA 0.7542 0.1960 0.3324 0.7318 1.3489 0.0092 0.0033 0.0031 0.0088 0.0194 CHLOANS 0.1055 0.1350 -0.0968 0.0733 0.6433 BEGNPL 0.0037 0.0038 0.0000 0.0028 0.0190 CHNPL 0.0006 0.0030 -0.0063 0.0000 0.0113 LCO 0.0019 0.0025 -0.0003 0.0013 0.0107 SMALL 0.1923 0.3942 0 0 1 BIG5 0.5232 0.4996 0 1 1 MB 3.1062 2.6696 0.9258 2.3950 13.7300 LMVE 19.7910 1.5136 17 20 25 LOSS 0.0121 0.1096 0 0 0 PASTLLP 0.0026 0.0030 -0.0005 0.0020 0.0124 EBP 0.0148 0.0056 -0.0565 0.0145 0.0292 TIER1 11.8939 3.4777 7 10 20 TCAP 13.5537 3.3327 10 10 20 Variable Definitions: LAFEE LTOTFEE LNAFEE FEERATIO LLP ALLP LOANS BEGLLA CHLOANS BEGNPL CHNPL LCO SMALL BIG5 MB LMVE LOSS PASTLLP EBP TIER1 TCAP = Natural log of audit fees = Natural log of total fees = Natural log of non-audit fees = Non-audit fees divided by total fees = Provision for loan losses deflated by beginning total assets = Residuals estimated from model (1) = Total Loans deflated by beginning total assets = Beginning value of Loan Loss Allowance over beginning total assets = Change in total loans outstanding deflated by beginning total assets = Beginning Nonperforming loans deflated by beginning total assets = Change in Nonperforming loans deflated by beginning total assets = Loan charge-offs deflated by beginning total assets = An indicator variable = 1 if beginning assets < 500 million and 0 otherwise = An indicator variable = 1 if audited by a big 5 firm and 0 otherwise = Market-to-book ratio at the year end = Natural log of market value of common equity = An indicator variable set equal to 1 if net income < 0 and 0 otherwise = LLP at the beginning of the year deflated by beginning total assets = Net income before extraordinary items and LLP over beginning total assets = Beginning Tier 1 Capital = Beginning Total Capital 29 Panel B: Correlations between Dependent and Independent Variables (N=1,810) LLP ALLP LOANS BEGLLA CHLOANS BEGNPL CHNPL LCO LTOTFEE LAFEE ALLP LOANS BEGLLA CHLOANS BEGNPL CHNPL LCO 0.604 <.001 1.000 0.191 <.001 0.000 1.000 1.000 0.184 <.001 0.000 1.000 0.209 <.001 1.000 0.096 <.001 0.000 1.000 0.805 <.001 -0.026 0.265 1.000 0.255 <.001 0.000 1.000 0.021 0.363 0.331 <.001 -0.095 <.001 1.000 0.209 <.001 0.000 1.000 0.093 <.001 -0.051 0.030 0.067 0.005 -0.167 <.001 0.726 <.001 0.000 1.000 -0.040 0.086 0.428 <.001 -0.146 <.001 0.382 <.001 0.107 <.001 LTOTFEE LAFEE LNAFEE FEERATIO SMALL BIG5 MB LMVE LOSS PASTLLP EBP TCAP 0.031 0.192 -0.028 0.234 -0.164 <.001 0.023 0.322 -0.075 0.002 -0.001 0.979 -0.046 0.052 0.145 <.001 1.000 -0.015 0.510 -0.025 0.288 -0.174 <.001 -0.002 0.947 -0.073 0.002 0.003 0.912 -0.061 0.010 0.101 <.001 0.944 <.001 1.000 0.075 0.002 -0.041 0.081 -0.114 <.001 0.057 0.016 -0.065 0.006 -0.017 0.469 -0.027 0.254 0.167 <.001 0.806 <.001 0.611 <.001 1.000 0.131 <.001 -0.021 0.369 0.023 0.328 0.080 0.001 -0.013 0.579 -0.017 0.475 0.042 0.075 0.134 <.001 0.170 <.001 -0.155 <.001 0.640 <.001 1.000 0.015 0.536 -0.004 0.853 0.161 <.001 0.001 0.971 0.167 <.001 -0.015 0.515 0.024 0.313 -0.043 0.066 -0.444 <.001 -0.443 <.001 -0.320 <.001 -0.006 0.784 1.000 0.020 0.405 -0.029 0.211 -0.196 <.001 -0.017 0.482 -0.101 <.001 -0.019 0.425 -0.029 0.224 0.094 <.001 0.525 <.001 0.493 <.001 0.414 <.001 0.098 <.001 -0.340 <.001 1.000 0.076 0.001 0.003 0.891 0.207 <.001 0.018 0.434 0.253 <.001 -0.021 0.363 0.003 0.907 -0.013 0.583 -0.041 0.080 -0.096 <.001 0.056 0.017 0.172 <.001 0.124 <.001 0.133 <.001 1.000 -0.001 0.962 -0.039 0.101 -0.139 <.001 -0.024 0.309 -0.027 0.252 -0.065 0.006 -0.060 0.010 0.071 0.002 0.821 <.001 0.769 <.001 0.686 <.001 0.173 <.001 -0.450 <.001 0.530 <.001 0.213 <.001 1.000 0.058 0.013 0.012 0.597 -0.038 0.110 0.221 <.001 -0.055 0.020 0.039 0.101 -0.038 0.106 0.142 <.001 0.013 0.576 0.009 0.694 0.008 0.737 0.013 0.567 0.047 0.048 -0.004 0.872 0.015 0.514 -0.078 0.001 1.000 0.422 <.001 0.096 <.001 -0.006 0.792 0.480 <.001 -0.120 <.001 0.338 <.001 0.012 0.623 0.564 <.001 0.032 0.173 0.004 0.856 0.073 0.002 0.091 0.000 -0.004 0.873 0.044 0.061 0.027 0.249 -0.009 0.705 0.096 <.001 1.000 0.282 <.001 0.124 <.001 0.224 <.001 0.092 <.001 0.208 <.001 0.032 0.174 0.067 0.004 0.185 <.001 0.090 0.000 0.054 0.021 0.120 <.001 0.109 <.001 -0.082 0.001 0.072 0.002 0.151 <.001 0.283 <.001 -0.396 <.001 0.090 0.000 1.000 -0.117 <.001 0.034 0.154 -0.245 <.001 -0.134 <.001 -0.017 0.464 -0.068 0.004 -0.030 0.196 -0.117 <.001 -0.200 <.001 -0.169 <.001 -0.187 <.001 -0.090 0.000 0.179 <.001 -0.145 <.001 -0.159 <.001 -0.203 <.001 -0.016 0.507 -0.114 <.001 0.071 0.002 LNAFEE FEERATIO SMALL BIG5 MB LMVE LOSS PASTLLP EBP See panel A for variable definitions. Total number of observations equals 1,810 and the data are for the years 2000 through 2006. 30 TABLE 2 Estimation of Abnormal Loan Loss Provision (ALLP) Variable Expected Sign Intercept ? BEGLLA _ BEGNPL + CHNPL + LCO + CHLOANS ? LOANS + COMM ? CON ? RESTATE ? AGRI ? FBG ? DEPINS ? Coefficient (t-statistic) -0.0011 (-3.93)*** -0.2260 (-13.25)*** 0.0241 (1.76)* 0.0853 (5.26)*** 1.0081 (45.34)*** -0.0006 (-0.98) 0.0046 (10.26)*** 0.0005 (2.26)** -0.0017 (-2.76)*** 0.0000 (0.15) -0.0005 (-0.15) -0.0105 (-0.45) 0.0085 (0.55) Yes 1810 63.15% Year Controls N Adjusted R2 Variable Definitions: LLP BEGLLA BEGNPL CHNPL LCO CHLOANS LOANS COMM CON RESTATE AGRI FBG DEPINS = Provision for loan losses deflated by beginning total assets = Beginning value of Loan Loss Allowance deflated by beginning total assets = Beginning Nonperforming loans deflated by beginning total assets = Change in Nonperforming loans deflated by beginning total assets = Loan charge-offs deflated by beginning total assets = Change in total loans outstanding deflated by beginning total assets = Total loans outstanding deflated by beginning total assets = Commercial loans deflated by beginning total assets = Consumer loans deflated by beginning total assets = Real estate loans deflated by beginning total assets = Agriculture loans deflated by beginning total assets = Loans to foreign banks and governments at deflated by beginning total assets = Loans to other depository institutions deflated by beginning total assets ***, **, and * indicate respectively, 0.01, 0.05, and 0.10 significance levels for a two-tailed test. 31 TABLE 3 Relation between Income-Increasing (Negative) ALLP and Fee Measures Variable Intercept SMALL LTOTFEE SMALL×LTOTFEE Total Fee -0.0028*** -0.0025*** (3.15) (-3.3) 0.0106* (2.10) -0.0002*** -0.0001** (-3.61) (-2.35) -0.0009** (-2.11) LNAFEE Nonaudit Fee -0.0031*** -0.0027*** (-3.33) (-3.54) 0.0051** (2.29) -0.0001*** (-2.66) SMALL×LNAFEE Audit Fee -0.0029*** -0.0025*** (-3.08) (-3.12) 0.0031 (1.07) Fee Ratio -0.0030*** -0.0027*** (-3.3) (-3.46) 0.0007** (2.23) -0.00004 (-1.53) -0.0005** (-2.28) LAFEE -0.0001*** (-2.60) SMALL×LAFEE -0.0001** (-2.10) -0.0003 (-1.11) FEERATIO -0.0003 (-1.20) 0.0001 (1.47) -0.0001** (-2.47) 0.0003*** (3.86) -0.0040 (-1.02) -0.0695*** (-2.84) -0.0297* (-1.79) 0.00002 (0.52) -0.00001 (-0.40) Yes 0.0001 (1.61) -0.00003 (-1.32) 0.0002*** (3.93) -0.0036 (-0.99) -0.0717*** (-2.86) -0.0270* (-1.72) 0.00002 (0.55) -0.00001 (-0.29) Yes 0.0001 (0.92) -0.00004* (-1.75) 0.0002*** (3.08) -0.0040 (-1.03) -0.0678*** (-2.80) -0.0249 (-1.51) 0.00002 (0.56) -0.00001 (-0.39) Yes 0.0001 (1.12) -0.00002 (-0.77) 0.0001*** (3.21) -0.0036 (-0.99) -0.0666*** (-2.77) -0.0221 (-1.44) 0.00002 (0.58) -0.00001 (-0.24) Yes 0.0001 (1.18) -0.00005** (-2.12) 0.0002*** (4.27) -0.0040 (-1.02) -0.0714*** (-2.93) -0.0253 (-1.62) 0.00003 (0.62) -0.00001 (-0.48) Yes 0.0001 (1.05) -0.00004* (-1.80) 0.0002*** (4.16) -0.0039 (-1.10) -0.0729*** (-2.99) -0.0246 (-1.60) 0.00002 (0.51) -0.00001 (-0.22) Yes 0.0001 (0.79) -0.00003 (-1.39) 0.0001*** (2.84) -0.0040 (-1.03) -0.0694*** (-2.89) -0.0211 (-1.34) 0.00003 (0.74) -0.00002 (-0.59) Yes 0.00004 (0.30) -0.0025** (-2.30) 0.0001 (0.71) -0.00002 (-0.78) 0.0001*** (2.95) -0.0037 (-1.00) -0.0648*** (-2.69) -0.0192 (-1.29) 0.00003 (0.64) -0.00001 (-0.24) Yes 1004 13.19% 1004 15.53% 1004 12.76% 1004 15.16% 1004 12.49% 1004 12.90% 1004 12.22% 1004 14.51% SMALL×FEERATIO BIG5 MB LMVE LOSS PASTLLP EBP TIER1 TCAP Year Controls N R2 ***, **, and * indicate respectively, 0.01, 0.05, and 0.10 significance levels for a two-tailed test. 32 TABLE 4 Relation between Income-Decreasing (Positive) ALLP and Fee Measures Variable Intercept SMALL LTOTFEE SMALL×LTOTFEE Total Fee 0.0040*** 0.0040*** (4.51) (3.37) 0.0048 (1.46) 0.0004*** 0.0005*** (2.97) (3.56) -0.0004 (-1.48) LNAFEE Nonaudit Fee 0.0042*** 0.0044*** (4.54) (3.24) 0.0027 (1.37) 0.0001 (0.94) Audit Fee 0.0038*** (3.23) 0.0026 (0.91) Fee Ratio 0.0040*** 0.0042*** (4.50) (3.32) 0.0003 (0.99) 0.0001 (1.37) -0.0003 (-1.4) SMALL×LNAFEE LAFEE 0.0003*** (2.69) SMALL×LAFEE 0.0003*** (2.93) -0.0002 (-0.95) FEERATIO 0.0004 (0.65) -0.0002 (-1.53) 0.00005 (1.33) -0.0005*** (-4.02) 0.0048 (1.19) 0.0829 (1.59) 0.0686 (2.87) 0.0001 (1.50) -0.0001 (-1.44) Yes -0.0002 (-1.26) 0.0001 (1.63) -0.0005*** (-3.75) 0.0049 (1.21) 0.0807 (1.54) 0.0705*** (2.92) 0.0001* (1.70) -0.0001 (-1.39) Yes -0.0001 (-0.69) 0.000004 (0.09) -0.0003*** (-3.00) 0.0050 (1.20) 0.0900* (1.73) 0.0627*** (2.59) 0.0001 (1.58) -0.0001* (-1.68) Yes -0.0001 (-0.47) 0.00001 (0.31) -0.0003*** (-2.65) 0.0051 (1.23) 0.0876 (1.67) 0.0640*** (2.64) 0.0001* (1.80) -0.0001 (-1.63) Yes -0.0001 (-1.23) 0.0000 (0.95) -0.0004*** (-4.06) 0.0051 (1.27) 0.0880 (1.67) 0.0701*** (2.80) 0.0001 (1.45) -0.0001 (-1.54) Yes -0.0001 (-1.06) 0.00004 (1.16) -0.0004*** (-3.68) 0.0051 (1.27) 0.0872 (1.66) 0.0711*** (2.78) 0.0001 (1.62) -0.0001 (-1.46) Yes -0.0001 (-0.5) -0.00001 (-0.22) -0.0002*** (-4.37) 0.0050 (1.21) 0.0925* (1.75) 0.0612** (2.36) 0.0001 (1.55) -0.0001* (-1.69) Yes 0.0006 (0.91) -0.0011 (-1.41) -0.00005 (-0.4) -0.00001 (-0.16) -0.0002*** (-3.07) 0.0051 (1.22) 0.0914* (1.72) 0.0611** (2.35) 0.0001* (1.74) -0.0001 (-1.61) Yes N 806 806 806 806 806 806 806 806 R2 11.44% 11.76% 10.30% 17.65% 10.74% 10.83% 10.16% 17.53% SMALL×FEERATIO BIG5 MB LMVE LOSS PASTLLP EBP TIER1 TCAP Year Controls ***, **, and * indicate respectively, 0.01, 0.05, and 0.10 significance levels for a two-tailed test. 33 TABLE 5 Panel A: Estimation of Unexpected (Abnormal) Fee Measures Variable Intercept BIG5 LASSETS SECURITIES NPL LOSS INTANG EFFICIENCY LCO COMM CON RESTATE TCAP Year controls N Adj. R2 Total Fee 5.2168*** (15.46) 0.3018*** (7.71) 0.6631*** (42.99) 0.9454*** (5.37) 20.6448*** (3.82) -0.3070 (-1.62) 2.9197*** (2.88) 1.5547*** (6.47) -0.0459 (-0.45) 0.1919** (2.58) -0.2163 (-0.90) -0.0807* (-1.73) 0.0114* (1.91) Yes 936 82.52% Nonaudit Fee 2.1344*** (3.15) 0.1892** (2.32) 0.7361*** (22.71) 1.2029*** (3.27) 38.6089*** (3.44) -0.7922** (-2.00) -4.9025** (-2.38) 2.8298*** (5.87) -0.2105 (-0.97) 0.2246 (1.42) -0.0760 (-0.15) -0.0938 (-0.95) -0.0060 (-0.47) Yes 936 51.45% Audit Fee 5.4565*** (14.22) 0.2282*** (4.95) 0.6299*** (34.33) 1.1125*** (5.35) 3.4956 (0.55) 0.1063 (0.47) 8.6393*** (7.42) 0.8193*** (3.00) 0.0803 (0.66) 0.2505*** (2.80) -0.4722 (-1.65) -0.0541 (-0.97) 0.0198*** (2.76) Yes 936 74.99% Variable Definitions: BIG5 LASSETS SECURITIES NPL LOSS INTANG EFFICIENCY LCO COMM CON RESTATE TCAP = Indicator variable set equal to 1 if audited by a Big 5 firm, and 0 otherwise; = Natural log of total assets; = [1-(total securities/total assets)]; = Nonperforming loans over lagged total loans; = Indicator variable set equal to 1 if ROA < 0, and 0 otherwise; = Intangible assets over total assets; = Total operating expenses over total revenues; = Net loan charge-offs over loan loss allowance; = Total commercial and agriculture loans over total loans; = Total consumer loans over total loans; = Total real estate loans over total loans; and = Total risk-adjusted capital ratio. ***, **, and * indicate respectively, 0.01, 0.05, and 0.10 significance levels for a two-tailed test. 34 Panel B: Relation between Income-Increasing (Negative) ALLP and Unexpected Fee Measures Total Fee Intercept -0.0031*** (-2.89) SMALL UTOTFEE -0.0001** (-1.99) SMALL×UTOTFEE -0.0027*** (-3.02) -0.0002 (-1.18) -0.00004 (-0.71) -0.0007 (-1.41) UAFEE Audit Fee -0.00309*** (-2.95) -0.0027*** (-3.16) -0.0001 (-1.12) 0.00001 (-0.20) 0.00003 (0.63) -0.0002 (-0.53) SMALL×UAFEE Nonaudit Fee -0.00317*** (-2.97) -0.0029*** (-3.11) -0.0001 (-0.98) -0.00009*** (-2.88) 0.00007 (0.79) -0.00004 (-1.48) 0.0001*** (2.87) -0.0033 (-1.16) -0.0625 (-1.46) -0.0283 (-1.85) 0.00004 (0.76) -0.00002 (-0.59) 0.0001 (0.66) -0.00002 (-0.98) 0.0001*** (3.05) -0.0033 (-1.16) -0.0673 (-1.67) -0.0266 (-1.80) 0.00002 (0.53) -0.000001 (-0.03) 0.00007 (0.75) -0.00004 (-1.49) 0.00013*** (2.93) -0.00327 (-1.15) -0.06136 (-1.41) -0.02806* (-1.83) 0.00004 (0.82) -0.00002 (-0.69) 0.0001 (0.61) -0.00003 (-1.29) 0.0001*** (3.11) -0.0033 (-1.14) -0.0636 (-1.58) -0.0275* (-1.82) 0.00003 (0.71) -0.00001 (-0.42) 0.00005 (0.62) -0.00004 (-1.50) 0.00013*** (2.94) -0.00326 (-1.15) -0.063 (-1.47) -0.02809* (-1.84) 0.00004 (0.72) -0.00002 (-0.56) -0.0001** (-1.77) -0.0003 (-1.61) 0.0001 (-.61) -0.00002 (-1.05) 0.0001*** (3.16) -0.0032 (-1.15) -0.0645 (-1.54) -0.0258* (-1.80) 0.00003 (0.57) -0.000002 (-0.06) Yes 936 13.53% Yes 936 14.37% Yes 936 13.29% Yes 936 13.52% Yes 936 13.80% Yes 936 14.68% UNAFEE SMALL×UNAFEE BIG5 MB LMVE LOSS PASTLLP EBP TIER1 TCAP Year Controls N R2 ***, **, and * indicate respectively, 0.01, 0.05, and 0.10 significance levels for a two-tailed test. 35 TABLE 6 Relation between Future Loan Charge-Offs on Fee Measures Panel A: Fee Regressions Variable Total Fee Nonaudit Fee Audit Fee Intercept -0.0031*** (-4.39) -0.0028*** (-3.57) -0.0016*** (-3.24) LTOTFEE 0.0004*** (4.53) 0.0003*** (3.47) LAFEE 0.0002*** (4.21) LNAFEE -0.0003* (-1.68) -0.0003* (-1.86) -0.0003* (-1.70) LLP 0.2884*** (11.03) 0.2893*** (11.13) 0.2933*** (10.93) LCO 0.2589*** (8.30) 0.2601*** (8.40) 0.2585*** (8.13) LASSETS -0.0002** (-2.10) -0.0001 (-1.16) -0.0001 (-1.11) Yes Yes Yes 101.92 100.83 96.20 1666 1651 1599 37.74% 37.70% 37.33% SMALL Year controls F value N Adj. R2 Variable definitions (all variables are deflated by beginning total assets): FLCO= FEE= SMALL= LLP = LCO= SIZE = Next period’s net loan charge-offs, Natural log of audit fees (LAFEE) or total fees (LTOTFEE = ln (audit fees + nonaudit fees)) or nonaudit fees (LNAFEE); 1 if beginning total assets are less than $500 million; Provision for loan losses; Net loan charge-offs; and Natural log of beginning total assets. ***, **, and * indicate respectively, 0.01, 0.05, and 0.10 significance levels for a two-tailed test. Panel B: Unexpected (Abnormal) Fee Regressions Variable Intercept UTOTFEE Total Fee Nonaudit Fee Audit Fee -0.0006 (-1.43) -0.0007 (-1.61) -0.0004 (-1.04) 0.0004*** 9 (3.55) 9 9 9 0.0003*** (2.68) -0.0003* (-1.75) -0.0003* (-1.69) -0.0003* (-1.77) LLP 0.2666*** (9.38) 0.2631*** (9.26) 0.2664*** (9.34) LCO 0.3042*** (8.87) 0.3052*** (8.88) 0.3102*** (9.05) LASSETS 0.0001*** (2.88) 0.0001*** (3.02) 0.0001*** (2.66) Yes Yes Yes F value 82.28 81.42 81.33 N 1404 1404 1404 36.68% 36.43% 36.41% UAFEE 0.0002*** (2.58) UNAFEE SMALL Year controls Adj. R2 Variable Definitions (all variables are deflated by beginning total assets): FLCO= UFEE= SMALL= LLP = LCO= SIZE = Next period’s net loan charge-offs, Unexpected (abnormal) fee (UTOTFEE, UAFEE and UNAFEE); 1 if beginning total assets are less than $500 million; Provision for loan losses; Net loan charge-offs; and Natural log of beginning total assets. ***, **, and * indicate respectively, 0.01, 0.05, and 0.10 significance levels for a two-tailed test. 37 TABLE 7 Relation between Benchmark-beating and Fee Measures for Income Increasing ALLP Panel A: Earnings Benchmark Tests and Fees Variable Total Fee Nonaudit Fee Audit Fee Intercept -5.9172*** (19.66) -4.014*** (11.78) -5.937*** (19.85) LTOTFEE 0.6647*** (25.63) 0.2703*** (16.31) LNAFEE 0.5607*** (18.04) LAFEE 0.4353** (4.35) 0.4075** (3.84) 0.4255** (4.20) ALLP -387.10*** (16.81) -391.80*** (17.40) -393.70*** (17.45) MB 0.1611*** (11.54) 0.1139*** (6.91) 0.1344*** (8.73) LMVE -0.1869** (3.68) 0.0225 (0.09) -0.1012 (1.18) BIG5 -0.1246 (0.54) 0.0706 (0.19) -0.1135 (0.44) TIER1 0.1820*** (7.87) 0.1598** (6.19) 0.1701*** (7.15) TCAP -0.1048 (2.62) -0.0853 (1.76) -0.1078* (2.84) Year controls Yes Yes Yes N 1004 1004 1004 SMALL Wald Chi-Square values in parentheses. Variable Definitions: INCREASE = FEE= SMALL= ALLP = MB= LMVE= BIG5= TIER1= TCAP= 1 when the change in net income scaled by beginning of year assets falls in the interval [0.000, 0.002] and 0 otherwise. Natural log of audit fees (LAFEE) or total fees (LTOTFEE = ln (audit fees + nonaudit fees)) or nonaudit fees (LNAFEE); 1 if beginning total assets are less than $500 million; Abnormal loan loss provision; Market-to-book ratio at the end of the year; Natural log of market value of common equity; Indicator variable set equal to 1 if audited by a Big 5 firm, and 0 otherwise; Tier 1 risk adjusted capital at the beginning of the year; and Total risk adjusted capital at the beginning of the year. ***, **, and * indicate respectively, p < 0.01, 0.01 < p < 0.05, and 0.05 < p < 0.10 significance levels. Panel B: Earnings Benchmark Tests and Unexpected (Abnormal) Fees Variable Total Fee Nonaudit Fee Audit Fee Intercept -5.2663*** (15.02) -5.1698*** (14.62) -5.0865*** (14.31) UTOTFEE 0.6111*** (17.01) 0.2498*** (12.38) UNAFEE 0.4496*** (10.00) UAFEE 0.4181** (3.79) 0.4001* (3.48) 0.4018* (3.53) -412.00*** (18.40) -402.10*** (17.62) -412.20*** (18.49) 0.0640 (2.38) 0.0648*** (2.50) 0.0608** (2.21) LMVE 0.2312*** (12.17) 0.2246 (11.62) 0.2239*** (11.64) BIG5 0.0692 (0.17) 0.0630 (0.14) 0.0652 (0.15) TIER1 0.1551** (5.87) 0.1540** (5.68) 0.1501** (5.61) TCAP -0.0991 (2.32) -0.0939 (2.05) -0.0980 (2.31) Year controls Yes Yes Yes N 936 936 936 SMALL ALLP MB Wald Chi-Square values in parentheses. Variable definitions: INCREASE = FEE= SMALL= ALLP = MB= LMVE= BIG5= TIER1= TCAP= 1 when the change in net income scaled by beginning of year assets falls in the interval [0.000, 0.002] and 0 otherwise. Unexpected (abnormal) fee (UTOTFEE, UAFEE and UNAFEE); 1 if beginning total assets are less than $500 million; Abnormal loan loss provision; Market-to-book ratio at the end of the year; Natural log of market value of common equity; Indicator variable set equal to 1 if audited by a Big 5 firm, and 0 otherwise; Tier 1 risk adjusted capital at the beginning of the year; and Total risk adjusted capital at the beginning of the year. ***, **, and * indicate respectively, p < 0.01, 0.01 < p < 0.05, and 0.05 < p < 0.10 significance levels. 39