You are on page 1of 50

A New Measure of Accounting Quality

Paul Hribar Tippie College of Business The University of Iowa, Iowa City, IA 52242 Todd Kravet University of Texas at Dallas Richardson, TX 75080 Ryan Wilson Tippie College of Business The University of Iowa, Iowa City, IA 52242

April 2010

This study examines an alternative approach to measuring accounting quality. Adopting a neoclassical view of the audit market, we argue that unexplained audit fees are a measure of accounting quality. We develop a comprehensive model of audit fee determinants to estimate unexplained audit fees. We use the framework developed by Cronbach and Meehl (1955) to examine the construct validity of our new measure. We find that our audit fee-based measure of quality correlates positively with other empirical measures of quality. We further show our measure is predictive of fraud, restatements, and SEC comment letters, even after controlling for other measures of accounting quality. Finally, we show that it is less associated with innate firm characteristics than measures based on realized earnings, making it a good candidate for studies examining the economic consequences of accounting quality.
We appreciate the helpful comments of, Andrew Acito, Daniel Beneish, Dan Collins, Scott Dyreng, Cristi Gleason, Irene Kim, Laureen Maines, Bill Mayew, Rick Mergenthaler, Katherine Schipper, Terry Shevlin, Ross Watts and workshop participants at Duke University, Georgetown University, Indiana University, The University of Iowa, and the University of Washington. We thank Andrew Leone for providing access to the restatement classification data.

I. Introduction This study presents a new approach to measuring accounting quality. Economic theory suggests that in a competitive equilibrium, audit fees incorporate the expected cost of poor quality earnings. Auditors face significant reputation and litigation risk if their clients financial reports are misstated. In response, auditors will increase audit hours and audit fees when they perceive accounting quality to be low. Isolating this part of the audit fee from the total audit fees can therefore provide a summary measure of the auditors assessment of the quality of the accounting system. We use a regression-based approach to remove the expected amount of audit fees based on the scope of the audit and other determinants, and use the unexplained fees as a measure of accounting quality. Our measure assumes that auditors charge higher fees to firms with lower quality accounting. We do not assume, however, that this increase in fees is necessarily in the form of a risk premium. For example, we expect auditors will increase the scope of their audit or the hours worked when faced with low quality accounting.1 Nor do we assume that increased audit effort is ineffectual at improving a firms accounting quality. Rather, we expect increased audit effort to improve accounting quality at the margin. However, our audit fee-based measure assumes that, on average, additional audit effort or audit procedures will not transform firms with ex-ante poor accounting quality into firms with ex-post high quality accounting. Thus, we expect a positive association between unexplained audit fees (UAF) and low quality accounting information will exist in the cross-section of firms, based on the endogeneity of audit fees with respect to accounting quality.

Using confidential survey data Bell, Landsman, and Shackelford (2001) find that when auditors perceive inherent risk to be high they increase the number of audit hours. This finding is consistent with the notion that auditors increase audit effort and therefore total audit fees in the cases of high inherent risk.

The hypothesized association between accounting quality and audit fees is analogous to the association between the number of police and crime rates examined in economics. Using panel data from 59 large U.S. cities Levitt (1997) documents a positive cross-city correlation between police and crime. Levitt goes on to demonstrate this finding is not a result of increased police leading to increased crime, but rather the simultaneous relationship between the number of police and crime rates. We believe a similar relationship exists between accounting quality and audit fees. In cases where auditors perceive a clients accounting quality to be low we expect them to increase both audit hours and audit fees resulting in improved accounting quality. However, because of inherent constraints in auditors ability to remediate effectively the low quality accounting and despite improvement from additional audit procedures, accounting quality will remain lower than firms with ex-ante high quality accounting. As a result, we expect to observe a positive association between UAF and low quality accounting in the cross-section of firms. Our primary goal is to introduce an alternative to the existing measures of quality in the literature (e.g., Dechow and Dichev 2002; Francis et al. 2004). A large body of research is interested in the role that accounting quality plays in the capital markets, and instances of poor quality accounting continue to headline in the popular press.2 Existing measures of accounting quality are typically based on realized earnings and earnings components. By developing a measure of accounting quality that originates from a different source (i.e., audit fees versus realized earnings), we hope to capture different aspects of accounting quality than existing measures and take advantage of the auditors private information. We also show that relative to existing measures, our measure has a different correlation structure with innate firm
For example, on March 12, 2010 headlines in the financial media discussed the U.S. bankruptcy-court examiners conclusion that Lehman Brothers had materially misled shareholders through an accounting gimmick referred to as Repo105.
2

characteristics, making it useful for triangulating research questions regarding accounting quality. There are several attractive features to an audit-fee based measure of accounting quality. First, auditors have inside knowledge of firms operations and how the financial reporting captures those operations. Although the audit opinion itself contains very little information, the fees charged by the auditor should incorporate the auditors assessment of the quality of the underlying accounting system, irrespective of whether the firm receives a clean opinion. Thus, an audit-fee-based measure of earnings quality summarizes the auditors unobservable private information about accounting quality, which potentially captures more information than measures of accounting quality that are based on realized financial statement data. Second, unexplained audit fees have the potential to capture a broader notion of accounting quality. The risk of misstatement that auditors are exposed to refers to a material misstatement in any part of the financial statements (e.g., income statement, balance sheet, and notes to the financial statements). Moreover, the auditors have expected legal and reputational costs that should be reflected in the audit fee. Thus, we expect that a measure based on unexplained audit fees provides a more global assessment of financial statement quality relative to measures that focus more on a specific dimension of accounting quality, such as how well accruals map into cash flows. Third, an important issue with measures of accounting quality based on realized earnings is that these measures capture other firm characteristics (e.g. operating risk) in addition to accounting quality (e.g., Liu and Wysocki 2007). This creates a significant hurdle to research linking accounting quality to economic constructs, such as asset pricing or investment efficiency, because the researcher has to convincingly rule out the known association between innate firm

characteristics. In contrast, our analysis is designed to remove any association with innate characteristics, and consequently reveals no discernable pattern between audit fees and measures of cash flow volatility and firm size. Thus, if you obtain similar inferences when using an earnings-based measure of quality and UAF, it is more difficult to argue that these associations are due to omitted variables or innate firm characteristics. Using audit fees to measure accounting quality also presents challenges. For example, the total audit fee is a function of factors that may not be related to underlying accounting quality. To develop a useful measure, we need to remove from total audit fees, the variation that is unrelated to differences in accounting quality while leaving in the variation that predicts accounting quality. In several cases, it is debatable whether a variable that explains total audit fees is indicative of differences in accounting quality, and should therefore be removed from total fees. For example, Venkataraman et al. (2008) show that litigation risk is priced in the audit fee. Litigation risk could reflect risk that is related to concerns about accounting quality, which we would want to retain in an audit-fee based measure of quality. Alternatively, it could stem from the nature of the audit (e.g., IPO firms who fall under the Securities Act of 1933) and be unrelated to accounting quality, which we would want to remove from our measure. Other variables present similar challenges, and we generally rely on prior research to remove the audit fee determinants that are primarily driven by reasons other than accounting quality. We follow the suggestion of Defond (2009) and evaluate the construct validity of our audit-fee based measure using the approach prescribed by Cronbach and Meehl (1955). We begin by testing the measures convergent validity by examining whether it is associated with other measures of accounting quality. For other accounting quality measures, we use Dechow and Dichevs (2002) measure of accrual quality, the absolute value of discretionary accruals,

earnings smoothness, and total accruals. We show that UAF is significantly associated with each of these measures, both unconditionally, and conditional on the other measures, although the magnitude of the correlations are modest. Thus, although there is some commonality across the measures, there is additional information about accounting quality embedded in the UAF measure. We next examine whether our measure is associated with the expected antecedents and consequences of low quality accounting. As an antecedent, we examine whether poor internal controls are associated with higher unexplained audit fees. Prior research shows that poor internal controls result in poor quality accounting. For example, Ashbaugh-Skaife, Collins, Kinney, and LaFond (2008) find the quality of internal controls has an effect on accrual quality. As such, we expect firms reporting lower quality accounting information will be more likely to have previously reported an internal control deficiency. Consistent with our expectation, we observe a significant positive correlation between the reporting of internal control deficiencies and subsequent period UAF. As consequences of poor accounting quality, we investigate whether UAF is associated with ex-post realizations of fraud, accounting restatements, and SEC comment letters. We show that UAF is significantly associated with each of these measures, both unconditionally, and conditional on the other measures. Although there is some commonality between UAF and the other measures (i.e., convergent validity), the fact that it is significant in the presence of the other measures further indicates that there is unique information in UAF. This suggests that unexplained audit fees capture a dimension of accounting quality not captured by the other measures.

Because these aforementioned consequences tend to capture only poor quality accounting, we also examine a more symmetric measure of ex-post quality based on the markets ability to forecast future earnings, using the forward earnings-response coefficient (Collins et al. 1994; Lundholm and Myers 2002). We find the amount of future earnings information

embedded in price is positively associated with accounting quality measured using UAF, suggesting that when UAF is low (indicating high quality accounting), investors are better able to predict future earnings. We conclude our analysis by examining the extent to which UAF is confounded by innate characteristics, which can complicate statistical inferences and limit the attribution of causality. For example, the AQ measure over-identifies small firms as having low quality accounting, while the UAF measure does not appear to be associated with firm size. Thus, triangulating research results using both measures will help rule out alternative explanations and help validate existing research documenting a link between accounting quality and economic constructs, such as the cost of capital.

II. Prior Literature and Research Design 2.1 Measures of accounting quality The accounting literature includes numerous studies examining different aspects of accounting quality. Although there is no agreed upon definition of accounting quality, different dimensions of this construct have been operationalized. These studies typically develop measures of quality using reported earnings and earnings components, and define the construct as earnings quality or accruals quality. Examples of past definitions of earnings quality include high earnings persistence in a time-series; earnings that accurately represent the

economic implications of underlying transactions; the proportion of earnings relative to operating cash flows; or how well working capital accruals map into to past, current, and future cash flows (see McNichols 2002 for further discussion). A frequently used proxy for earnings and accrual quality is the Dechow and Dichev (2002) accrual quality (AQ) measure based on how well accruals map into cash flows. This measure defines accrual quality as the error variance from a regression of working capital accruals on past, current, and lagged cash flows. Prior research uses the accrual quality measure in a variety of settings, often as a measure of information or financial reporting quality. For example, Francis et al. (2005) use AQ as a proxy for information risk and examine whether the cost of capital is associated with AQ. Chen, Shevlin, and Tong (2007) examine the loading on an AQ risk factor around dividend changes, and find evidence that it behaves consistent with information risk. Kravet and Shevlin (2009) examine the loading on a discretionary AQ risk factor around restatements, and find evidence consistent with information risk increasing after restatements. Lui, Markov, and Tamayo (2007) examine sell-side analysts risk ratings and

show that firms with higher AQ are rated as riskier by the analysts. Biddle et al. (2008) use AQ as a proxy for financial reporting quality and show that higher quality reporting mitigates both over- and under-investment, and show firms with better AQ measures deviate less from predicted investment levels. In addition to AQ, various other empirical measures are used to capture aspects of accounting quality. These measures include total accruals; the absolute value of discretionary accruals (ABS(DA)) calculated using a cross-sectional version of the modified Jones model; and earnings smoothness measured as the ratio of income variability to cash flow variability (e.g.,

Healy 1985; Becker et al. 1998; Francis et al. 2005). Similar to AQ, these measures have been used in a variety of contexts, which are too numerous to list. These measures commonly use realized earnings and earnings components to measure quality. As such, they provide a measure based on ex-post earnings realizations. In addition, they are often intended to measure the quality of reported earnings or accruals, not necessarily the quality of the entire accounting system. To complement these measures, we develop a measure of accounting quality that instead uses the audit market to provide an indicator of accounting quality. 2.2 Audit Fees and Accounting Quality Our measure of accounting quality uses the fees charged by auditors to infer the auditors ex-ante assessment of the quality of the accounting system. Our approach builds from the premise that lower quality accounting systems increase the risk exposure for auditors, and therefore will be priced in equilibrium. Moreover, auditors are arguably in the best position to estimate the quality of the accounting system because of their proprietary information and access to management. Before evaluating the construct validity of our measure of accounting quality, we begin by defining accounting quality for the purpose of this study. We acknowledge that accounting quality is a difficult concept to define and to measure. Despite the ambiguous nature of accounting quality, evidence suggests auditors do think in terms of the quality and not just the acceptability of their clients financial information. For example, the 1999 Blue Ribbon Committee on improving audit effectiveness makes the following recommendation (No. 8): a companys outside auditor should discuss with the audit committee the auditors judgments about the quality, not just the acceptability, of the companys accounting principles as applied in its financial reporting; the discussion should include such issues as the clarity of the companys financial

disclosures and degree of aggressiveness or conservatism of the companys accounting policies and underlying estimates and other significant decisions made by management in preparing the financial disclosure (emphasis added) In response to this recommendation, the Auditing Standards Board amended Statement of Auditing Standards No. 61 to include a discussion of audit quality. It is notable that the committees recommendation discusses elements of accounting quality that are commonly referenced in academic research such as the clarity of disclosures and the degree of conservatism or aggressiveness of the companys accounting policies. Following this logic, we adopt a broad definition of accounting quality which is similar to the definition provided by Dechow and Schrand (2004): Accounting quality is the extent to which accounting information accurately reflects the companys current operating performance, is useful in predicting future performance, and helps assess firm value. Note that this definition captures more than the acceptability of financial statements. It captures aspects such as transparency, representational faithfulness, and decision usefulness. The discussion of accounting quality in SAS No. 61 suggests auditors are concerned with this broader notion of accounting quality. This point is illustrated by the recent criticism of Lehman Brothers auditor Ernst & Young in the U.S. bankruptcy-court examiners report for serious lapses related to Repo 105 transactions.3 This criticism comes despite Ernst & Youngs claims that Lehmans financial statements were presented fairly under GAAP (Merced and Sorkin 2010). It is reasonable to conclude Ernst & Young will incur significant reputational and legal costs associated with the Lehman Brothers case even if the financial statements are ultimately found to
Lehman employed offbalance sheet devices, known as Repo 105 transactions, to temporarily remove securities inventory from its balance sheet, usually for a period of seven to ten days. Repo 105 transactions were nearly identical to standard repurchase transactions that Lehman and other investment banks use to secure shorttermfinancing, with a critical difference: Lehman accounted for Repo 105 transactions as sales as opposed to financing transactions based upon their overcollateralization. By characterizing the Repo 105 transaction as a sale, Lehman removed the inventory from its balance sheet and improved its leverage ratios. These transactions were deemed within GAAP by Ernst and Young, but were not separately disclosed in the 10-K. (Valukas 2010).
3

have conformed with GAAP. Similarly, Bell et al. (2001) show that auditors concerns about their professional reputation and the risk of litigation are significant factors when assessing the inherent risk of a client. We believe that auditors reputational and legal costs, if they are associated with clients that produce financial information with low quality, will therefore be reflected in the fees charged to the client. Figure 1 illustrates the link between auditors assessment of accounting quality and audit fees. When auditors observe that a client exhibits poor quality accounting they have two choices if they are going to continue providing audit services for that client. First, the auditors can charge a fee premium in exchange for accepting higher engagement risk. Second, auditors can increase testing in an effort to reduce engagement risk to an acceptable level. As discussed previously, we believe additional audit effort can improve accounting quality at the margin, but will not transform firms with ex-ante poor accounting quality into firms with ex-post high quality accounting. These two potential responses by auditors to poor quality accounting are not mutually exclusive and it is likely auditors would pursue some combination of these actions. However, the important point from Figure 1 is that in either case when auditors identify poor quality accounting the result is higher audit fees. Therefore, after controlling for non-quality related determinants of audit fees, the unexplained portion will provide a summary measure of the auditors proprietary assessment of the firms accounting quality. We deliberately exclude measures of accounting quality from our set of determinants because we want the association between audit fees and accounting quality to be captured in the residual. There is a related branch of research examining the association between fees for audit and non-audit services and auditor independence (e.g. Frankel et al. 2002; Ashbaugh et al. 2003;

10

Larcker and Richardson 2004). These studies investigate the contention that increasing fees for audit and non-audit services increases the economic bond between the auditor and client, thereby reducing auditor independence. As a result, earnings management will be associated with higher fees. Evidence supporting this theory is mixed. Frankel et al. (2002) document an association between non-audit fees and earnings management. In contrast, Ashbaugh et al. (2003) and Larker and Richardson (2004) find no support for the economic bonding hypothesis, using different methodologies. The theoretical basis for our model of accounting quality contrasts with the notion that higher fees impair independence. We adopt a neoclassical perspective of the audit market, which assumes that auditors will increase fees to compensate for the additional risk and additional audit hours required for firms with poor quality accounting systems. In essence, we reverse the direction of causality linking audit fees and accounting quality. Prior studies conjecture that impaired auditor independence resulting from higher fees leads to lower quality accounting information, while we argue that lower quality accounting systems result in higher fees charged by auditors. Additional tests presented in section five attempt to rule out economic bonding as an alternative explanation for our findings. 2.3 Estimating Unexplained Audit Fees To construct our measure of accounting quality, we need to remove the variation in audit fees that is unrelated to accounting quality. Using prior literature we identify a set of determinants that are associated with audit fees (e.g., Simunic 1980; Larcker and Richardson 2004; Hanlon and Krishnan 2009; Venkataraman et al. 2008). The determinants are intended to measure the resources required to complete the audit, with various proxies for size and complexity. For example, Venkataraman et al. (2008) find that the audit fee is nearly twice as

11

large in the year of an IPO, reflecting the heightened litigation risk associated with an IPO. This results in large differences in audit fees from the IPO year to the post-IPO year, but would not be indicative of differences in accounting quality. One difficulty is that some variables correlate with audit fees for reasons both related and unrelated to accounting quality. For example, we include measures of the level of inventory and accounts receivable, using the logic that larger amounts of these assets will require more auditing hours. However, these accounts have also been linked to earnings management and restatements. Thus, there is the danger that some of our variables take away part of the differences in accounting quality that we are intending to capture in the residual. For the most part, we err to the side of including these variables in the model, because any association between these characteristics and accounting quality will limit the power of UAF as a measure of accounting quality and bias against finding results in our tests validating UAF as a measure of accounting quality.4 Picconi and Reynolds (2009) find that regressing the logarithm of audit fees on a set of predictors that includes the logarithm of assets produces biased estimates of actual audit fees. They demonstrate that regressing by year and size decile improves the estimation and explanatory power of the audit model. Consequently, we follow their suggestion and estimate the following model by year and size decile.5 Ln(AUDIT FEE)i,t = 0 + 1BIG4i,t + 2Ln(ASSETS)i,t + 3BUS SEGi,t + 4FGNi,t + 5INVi,t + 6RECi,t + 7DEBTi,t + 8INCOMEi,t + 9LOSSi,t + 10AUD OPINi,t + 11CLIENTi,t + 12IPOi,t + 13SEOi,t + 14ISSUANCEi,t + 15LITRISKi,t + INDi,t + i,t The variables are defined as follows: Ln(AUDIT FEE)i,t
4 5

(1)

Log of audit fee;

Two variables that we intentionally omit from the model are internal control quality and corporate governance. Estimating equation (1) by year allows the intercept and coefficients to vary by year and, thus, controls for any year-specific events, such as the implementation of the Sarbanes-Oxley Act.

12

BIG 4 i,t Ln(ASSETS)i,t BUS SEG i,t FGN i,t INV i,t REC i,t DEBT i,t INCOME i,t LOSS i,t AUD OPIN i,t

= = = = = = = = = =

CLIENT i,t IPO i,t SEO i,t ISSUANCE i,t LITRISK i,t IND i,t

= = = = = =

A dummy variable set to one when the firms auditor is a member of the Big 5 (or Big 4 after the exit of Arthur Andersen), and zero otherwise (data #149); Log of total assets (data #6); Square root of the number of business segments of the firm, reported on the Compustat Segment Data File; Ratio of foreign sales (Compustat Segment Data) to total sales (data #12); Inventory (data #3) scaled by lagged total assets; Receivables (data #2) scaled by lagged total assets; Sum of short-term debt and long-term debt scaled by lagged total assets (data #34 + data #9); Operating income after depreciation (data #178) scaled by lagged total assets; A dummy variable that is equal to one if income before extraordinary items and discontinued operations (data #18) is negative in the current or two previous years, and zero otherwise; A dummy variables that is equal to one if the firm receives a modified audit opinion (data #149), and zero otherwise, where a modified opinion is defined as anything other than a standard unqualified audit opinion coded as one by Compustat; Square root of the number of years that the firm has been a client of their current auditor; Indicator variable equal to one if in the year of the initial public offering, reported by SDC Platinum. Indicator variable equal to one if in the year of a seasoned equity offering, reported by SDC Platinum. Indicator variable equal to one in the year the firm issues debt, reported by SDC Platinum. Indicator variable equal to one for high litigation risk industries, as defined in Francis et al. (1994). Industry fixed effects based on two digit SIC codes.

Ln(ASSETS), BUS SEG, FGN, INV, REC, and DEBT are included to proxy for the complexity of the audit and the resources required to perform the audit. We include CLIENT to control for the importance of the client to the audit firm. INCOME and LOSS are included as variables that increase inherent risk and likely lead to greater audit effort, though not necessarily indicative of poor quality. IPO, SEO, and ISSUANCE capture specific corporate events that increase the litigation risk of the auditor for reasons other than differences in accounting quality. We proxy for litigation risk using the industry-based indicator variable (LITRISK) in Francis et

13

al. (1994), to control for the fact that auditors will charge a premium in industries that are more litigious, regardless of their assessment of the accounting quality. The residual from this model (UAF, or unexplained audit fees) is our measure of accounting quality, where larger values of the residual indicate lower accounting quality. We exclude measures of internal control quality and corporate governance from our set of audit fee determinants despite an expectation that these will be associated with audit fees. Ashbaugh-Skaife et al. (2008) find firms with internal control deficiencies report lower quality accruals, as measured by the absolute value of abnormal accruals, than firms not reporting internal control problems. They argue that weak internal controls make it difficult for managers to provide reliable accrual amounts and as a result lead to more noise and less reliable financial reporting.6 As a result, we expect audit fees will increase for firms with identified internal control deficiencies. Because we believe internal control quality will be associated with both audit fees and accounting quality we exclude measures of internal control quality to ensure that the association between audit fees and accounting quality is captured in the residual.

III. Data and Descriptive Statistics 3.1 Sample composition and descriptive statistics The data for estimating the audit fee model comes from various sources. We use Audit Analytics to obtain audit fee data for the years from 2000 to 2007.7 Financial data is obtained from the Compustat database and matched to the audit fee data. Our final sample of firm years

Under Section 404 of the Sarbanes-Oxley Act (SOX) management is required to provide an assessment of the firms internal controls in their annual report and auditors are required to provide an opinion on managements assessment. 7 The SEC mandated in 2000 that firms disclose audit and non-audit fees paid to their auditors.

14

for which we can compute UAF is 38,385, which reduces to 25,415 when we require sufficient data to compute AQ.8 Table 1, Panel A presents descriptive statistics for the variables used to estimate UAF. Panel B presents the results of the regression used to estimate UAF. As expected, most variables are significant and in the predicted direction. Audit fees are positively associated with Big 4 auditors, larger firms, more business segments, larger foreign operations, larger inventories, larger receivables, the presence of losses, and any non-standard opinion. We also find that IPO years have significantly higher audit fees. The mean adjusted R2 over the 70 estimations by year and size deciles is 32.79%. Table 2 provides details on the characteristics of firms with high accounting quality (lowest UAF quartile) and low accounting quality (highest UAF quartile). Firms with estimated low quality accounting tend to be larger, both in terms of assets and sales, than the firms with high quality accounting. Notice that the directional association with size is the opposite of other measures, such as AQ, which classifies smaller firms as having lower quality earnings (Francis et al 2005, Liu and Wysocki 2008). Profitability and leverage appear comparable between the highest and lowest accounting quality quintiles, with median ROA = 1% and LEV = 19% in the lowest quality quintile versus median ROA = 1% and LEV = 15% in the highest quality quintile. Firms in the high accounting quality group also tend to be more value firms than firms in the low accounting quality group, with median book-to-market of 0.46 versus 0.36. 3.2 Tests of Convergent Validity Table 3 compares our audit-based measure of accounting quality to other measures of earnings quality used in past research. Although we expect our measure of accounting quality to
The estimation of UAF requires a maximum of two years of data which is a less stringent data requirement than other accounting quality proxies (e.g., AQ requires seven years of data). However, because audit fee data is only available beginning in 2000 it is not possible to calculate UAF for years before 2000.
8

15

be broader in scope than some other measures of quality, it should still capture aspects of quality reflected in the other measures, and we therefore expect a positive association between UAF and the other measures. Following the framework provided by Cronbach and Meehl (1955), we view this analysis as a test of the convergent validity of UAF. We compare UAF to four other measures of quality. The first measure is Dechow and Dichevs (2002) model of accrual quality (AQ), which captures the extent to which working capital accruals map into operating cash flows. McNichols (2002) augments this model by including property, plant and equipment (PPE) and the change in revenue (Rev), arguing that these variables are incrementally important in forming expectations about current accruals. Following McNichols (2002), we estimate the following model annually, for each two-digit SIC code with at least 20 observations: TCAi,t = 1 + 2CFOi,t-1 + 3CFOi,t +4CFOi,t+1 +5Revi,t +6PPEi,t + i,t Accrual quality (AQ) is computed by taking the standard deviation of firm is residuals, calculated over years t-4 through t. Consistent with UAF, larger values of AQ indicate lower earnings quality. The second measure is the absolute value of discretionary accruals (ABS(DA)), estimated using the modified Jones model adjusted for performance (Kothari et al. 2005). This measure is used in numerous papers as a measure of the general tendency of firms to manage earnings (see Hribar and Nichols 2007 for a list of citations). The third measure we examine is earnings smoothness (SMOOTH), measured as the ratio of the standard deviation of net income before extraordinary items divided by the standard deviation of cash flow from operations (Leuz et al. 2003). The fourth measure we examine is total accruals (TACC), measured as the difference between net income and cash from operations, scaled by total assets. We also include the (2)

16

standard deviation of operating cash flows as a measure of innate operating volatility, computed using the past five years of data. This allows us to examine the extent to which each accounting quality measure correlates with operating volatility. All proxies for accounting quality are adjusted, if need be, so that larger values represent lower accounting quality. Table 3 presents the results comparing UAF to the other measures of earnings quality. Because of the longer time series required to compute AQ due to its inclusion of lead and lag explanatory variables, the number of observations available drops to 25,415. Statistics summarizing the distributions of each earnings quality measure are presented in Panel A. Because UAF is a residual, it is centered on zero, while measures such as AQ, ABS(DA), and SMOOTH are strictly positive.9 Panel B estimates Pearson and Spearman correlations between our measures of accounting quality and the other measures of earnings quality. Examining Pearson correlations, UAF is significantly associated with all other measures, although the magnitudes of the correlations are modest, ranging from 0.05 with ABS(DA) to 0.08 with AQ. Similarly, Spearman correlations range from 0.03 with ABS(DA) to 0.08 with SMOOTH. Overall, the low correlations are likely due to both noise in the estimates of the traditional accounting quality measures and the measures capturing different aspects of accounting quality. Panel C regresses our measure of accounting quality on all of the other quality measures in a multiple regression to examine the extent to which UAF is associated with the unique information in each of the other measures. The standard deviation of cash flows (CFO) is included to remove innate operating volatility from each of the measures. UAF appears to be positively associated with the unique information in each of the other measures except total
We note that in Panel A of Table 3 the mean value of UAF is 0.01 rather than 0.00 because we estimate the audit fee model over the entire sample of 38,385 and then eliminate firms without the necessary data to calculate the other earnings quality measures.
9

17

accruals. The modest R2 of 1.05% is consistent with the low correlations reported above. In summary, the results in Table 3 show that UAF is related to the other measures of earnings quality, thereby demonstrating convergent validity among the measures.

IV. Antecedents and Consequences of Low Quality Accounting 4.1 Antecedents We next examine whether our proxy for earnings quality is associated with expected antecedents and consequences of low quality accounting. Antecedents are the circumstances or firm characteristics that lead to differences in accounting quality, while consequences are the expected outcomes of differences in accounting quality. In the spirit of Cronbach and Meehl (1955), we attempt to examine both antecedents and consequences although we acknowledge that it is easier to objectively identify consequences of poor quality accounting than to identify antecedents. Therefore, our main analysis in this section is the tests examining whether UAF is associated with consequences such as accounting fraud, restatements, and SEC comment letters. Nonetheless, we believe the antecedent tests can help establish the construct validity of UAF and compliment our primary tests, and therefore begin with this analysis. Our antecedent test examines the association between prior internal control deficiencies and UAF. Ashbaugh-Skaife, et al. (2008) find firms reporting internal control deficiencies exhibit lower quality accruals relative to firms that do not report internal control problems and conclude the quality of internal controls has an effect on the quality of accruals. If weak internal controls result in less reliable financial information we would expect to observe an association between our measure of accounting quality and prior year incidence of internal control deficiencies.

18

Table 4 reports the results of our analysis of the association between incidence of internal control deficiencies and our proxy for accounting quality. We use data on internal control deficiencies from the Audit Analytics database beginning with the disclosure of internal control data following the passage of the Sarbanes-Oxley Act in 2002. We are limited to 4,883 firm-year observations with both data on internal controls in period t-1 and the necessary data to calculate UAFt. Panel A of Table 5 reports descriptive statistics and indicates that 13% of our sample firmyear observations report an internal control deficiency. Panel B reports the results of regressing UAFt on an indicator variable (ICDt-1) set equal to 1 for firm-years reporting an internal control deficiency in year t-1, and set to 0 for all other observations. Consistent with our prediction, we observe a significant positive association between internal control deficiencies and our proxy for accounting quality. Taken together, the results of our antecedent tests are consistent with expectations if UAFt is a reasonable proxy for accounting quality. 4.2 Consequences To validate whether unexplained audit fees provide a measure of accounting quality, we focus on three measures that capture ex-post differences in accounting quality. First, we examine the ability of UAF to predict accounting restatements and accounting fraud, both unconditionally and in the presence of other earnings quality measures. Second, we examine whether UAF is associated with future comment letters issued by the SEC regarding the financial statements. Because both settings provide ex-post indicators of earnings quality, they are suitable for validating ex-ante measures of quality, such as UAF, but not useful as measures of quality in and of themselves. Our first two settings for examining differences in accounting quality focus on incidence of poor accounting quality. We supplement this analysis by examining the association between UAF and the forward earnings response coefficient (FERC) a continuous

19

measure of accounting quality. FERC measures the extent to which information in current returns captures information in future earnings. 4.2.1 Restatements and Accounting Fraud Restatements and fraud represent instances where firms issue misstated financial statements. For restatements, we use Hennes, Leone, and Millers (2008) classification of the GAO restatement database into errors and irregularities. Hennes et al. (2008) classify restatements as irregularities if 1) the restatement is described as an irregularity or fraud through self-disclosure, 2) there is a related SEC investigation, or 3) there is a related independent investigation. For our restatement variable (RESTATEMENT), we only include the restatements that are due to irregularities. These represent the most severe restatements where we expect the usefulness of financial statements to be significantly impaired. The advantage of using this sample as opposed to a sample of all restatement types is that it does not include minor restatements that are due to error that are not necessarily indicative of low quality accounting information. Hennes et al. (2008) report the restatement announcement market reaction for the restatements they classify as irregularities (-14%) is significantly lower than that for restatements classified as errors (-2%). One complication is that the GAO database only provides the date of the restatement announcement, not the periods for which the financial statements were misstated. Our tests require knowing the years of the financial statement malfeasance. To determine the fiscal periods that were misstated, we merge the Audit Analytics data, which contains the restated fiscal periods, to the Hennes et al. (2008) list of irregularities. If firms in the Hennes et al. sample have multiple years restated in the Audit Analytics database, we only classify as irregularities those

20

restatements where the fiscal year end of the last misstated period is before the restatement announcement date. We also examine a separate sample of restatements that are defined by Audit Analytics as occurring from financial fraud (FRAUD). These restatement samples represent the most severe types of restatements where we expect the usefulness of financial statements to be significantly impaired, thus providing a powerful setting to validate our ex-ante measure of accounting quality. We test the association between our proxy for accounting quality and the probability of firms having one of the two types of restatements using the following model: Pr(RESTATEMENTi,t or FRAUDi,t) = 0 + 1UAFi,t + 2AQi,t + 3ABS(DA)i,t + 4SMOOTHi,t + 5TACCi,t + CONTROLSi,t + i,t (3)

RESTATEMENTi,t and FRAUDi,t are indicator variables that are equal to one when any period during firm is fiscal year t is subsequently restated or restated due to fraud, respectively, and zero otherwise. We use lagged values of UAF for firm-years where the restatement is announced prior to the date the audit opinion is signed for that firm-year, to avoid the fact that audit fees during the year of the restatement will be higher simply because of the audit procedures related to the audit itself. We expect 1 to be positive which is consistent with higher quality accounting firms being less likely to experience a restatement. We examine UAF unconditionally, and in the presence of the four other earnings quality measures (AQ, ABS(DA), SMOOTH, and TACC) and other controls that are predictive of restatements, to see whether it is incrementally informative. Audit Analytics identifies the fiscal periods for the transgression in addition to the date of the restatement, allowing us to examine unexplained audit fees in the year of the transgression, not the year of the restatement. In addition to the earnings quality measures, the other variables that are predictive of accounting restatements and fraud are taken from Beneish (1999) and

21

Dechow, Ge, Larson, and Sloan (2008). This includes change in receivables (REC); change in inventory (INV); change in cash sales (CASH SALES); change in net income (NIBE); unexpected change in employees (EMPLOYEES); and the book to market ratio (BTM). The volatility of cash flows (CFO) is included to control for innate operating volatility. Specific definitions are provided in the notes to Table 6. For our restatement and fraud tests we use a matched pair control group. Specifically, we match a control firm-year observation to each incidence of restatement or fraud with the necessary data available to calculate the variables in equation (3). We select the control sample using the following three criteria: (1) fiscal year, (2) two-digit SIC code, and (3) nearest total assets in the current year.10 The results of estimating equation (3) are reported in Table 5. The first column shows only UAF and the second column includes the other measures of accounting quality, while the third column also includes additional control variables. When predicting restatements (column 1), UAF is significantly positive, suggesting unexplained audit fees are positively associated with the occurrence of restatements (1=0.169, p=0.053). After adding other earnings quality measures to the equation (column 2), UAF remains positive and significant (1=0.261, p=0.023). UAF also remains positive and significant (column 3) after additional control variables are included in the model (1=0.284, p=0.016). The only other accounting quality variable that is predictive of restatements is AQ (columns 2 and 3). When predicting fraud, the results are similar. By itself (column 4), UAF is positively and significantly related to FRAUD (1=0.527, p=0.003). When other earnings quality measures

10

In supplemental tests (not tabulated) we include all firm year observations over the sample period not identified as fraud or restatement firms with the necessary data to calculate the variables in equation (3) as an alternative control group. We continue to find UAF is significantly associated with incidence of fraud and restatements using this expanded control group.

22

are added to the equation (column 5), UAF remains positive and significant (1=1.023, p<.001). After additional control variables are included in the model UAF continues to be significant (1=1.084, p<.001). At the bottom of Table 5, we present the change in the probability of a restatement when the independent variables change from the first quartile to the third quartile. We focus on the results in column 3 and 6 for the sake of brevity. The probability of a restatement increases by 4.85% when UAF changes from the first quartile to the third quartile. Similarly, the probability of a fraudulent restatement increases by 17.14% when UAF changes from the first quartile to the third quartile. Therefore, the predictive ability of UAF with respect to restatements appears to be economically significant. 4.2.2 Comment Letters Our second setting for validating our measure of accounting quality uses the receipt of a comment letter from the Securities and Exchange Administration (SEC) as an indication of accounting quality. The SEC reviews financial reports submitted to them by public companies. When SEC staff identifies potential deficiencies during a review, they send a comment letter to the firm seeking additional information, clarification, or revision of the filing. Under the Sarbanes Oxley Act (SOX 2002) the SEC is required to review a firms filings at least once every three years. According to Chen and Johnston (2008, p. 6) the SECs stated objectives for the review program are to identify potential or actual material accounting, auditing, financial reporting or disclosure deficiencies; influence accounting standards and practices; propose new and amended disclosure rules; offer guidance and counseling, either informally or through no action letters. These objectives suggest firms receiving a comment letter likely exhibit low

23

quality accounting and therefore provides another setting for testing the usefulness of our measure. A limitation of our fraud prediction tests is the ability to generalize the results of those tests to less extreme cases of poor accounting quality. Using comment letter prediction tests helps to mitigate this limitation by validating the usefulness of our measure in a less extreme setting. While comment letters can lead to SEC enforcement actions, this is not typically the case, and more often comment letters are used to evaluate the quality of implementation of new standards and correct poor practices. Comment letter data is obtained from the Audit Analytic database and we limit our analysis to comment letters applying to either 10-K or 10-Q filings. We test the association between our proxy for accounting quality and the probability of firms receiving a comment letter from the SEC using the following model: Pr (COMMENT_LETTERi,t) = 0 + 1UAFi,t + 2 EPi,t + 3log(MARKET_VALUEi,t) + 4SALESVOLi,t + 5SYNCi,t + 6RATINGi,t + 7NO_RATINGi,t + 8RESTATEMENTi,t-1 (4) + 9PROPORTIONi,t + The dependent variable, COMMENT LETTERi,t, is an indicator variable that is set equal to one when firm is fiscal year t is subsequently subject to a comment letter by the SEC, and zero otherwise. We expect 1 to be positive which is consistent with lower quality accounting firms being more likely to be subject to a comment letter. Similar to the fraud prediction tests, we examine UAF unconditionally and in the presence of alternative measures of accrual quality (total accruals, AQ, absolute discretionary accruals, and a measure of smoothness). Section 408(b) of SOX indicates the review criteria for SEC staff. Following Chen and Johnston (2008), we use these review criteria to construct control variables for our comment letter prediction tests.11 Consistent with our fraud tests, we use a matched pair control group to evaluate the
11

According to Section 408(b) of SOX the commission shall consider: (1) issuers that issued a material restatement of financial results; (2) issuers that experience significant volatility in their share price as compared to other issuers; (3) issuers with the

24

association between incidence of comment letters and UAF. Specifically, we match a control firm-year observation to each firm receiving a comment letter with the necessary data available to calculate the variables in equation (4). We select the control sample using the following three criteria: (1) fiscal year, (2) two-digit SIC code, and (3) nearest total assets in the current year. In equation (4) above EPi,t is earnings per share to price ratio. log(MARKET_VALUEi,t) is the natural log of the firms market capitalization at the fiscal year-end prior to receiving the comment letter. SALESVOLi,t is the standard deviation of sales scaled by total assets over the five years prior to receiving the comment letter. SYNCi,t is the firms logistic transformed relative idiosyncratic volatility estimated from the market model one year prior to receiving the comment letter. 12 For firms with S&P credit ratings, RATINGi,t ranges from one for AAA to 22 for D. NO_RATINGi,t is an indicator variable set equal to one if the firm does not have an S&P credit rating and zero otherwise. RESTATEMENTi is an indicator variable equal to one if the firm had a restatement in any prior year. PROPORTIONi,t is firm is proportion of its two-digit SIC industry revenue at the fiscal year end prior to receiving the comment letter. Results of the comment letter tests are provided in Table 6. When included on its own, UAF is significantly positive (1=0.259, p < 0.001), indicating that firms with higher unexplained audit fees are more frequently the subject of SEC comment letters. When the other earnings quality measures and the control variables are added to the regression, UAF remains significantly positive with a coefficient of 0.224 (p < 0.001). None of the other measures of earnings quality are significantly associated with incidence of comment letters. Overall, the results from the tests intended to validate UAF as a measure of accounting quality appear to

largest market capitalization; (4) emerging companies with disparities in price to earnings ratios; (5)issuers whose operations significantly affect any material sector of the economy; and (6) any other factors that the Commission may consider relevant. 12 Following Durnev, Morck, and Yeung (2004) we measure idiosyncratic volatility as SYNC = ln ((1 R2 ) / R2), where R2 is the R-square from the market model.

25

support the notion that a measure of accounting quality that is based on audit fees is incrementally informative to existing measures of accounting quality. 4.2.3 Forward Earnings Response Coefficient Our final test of the consequences of accounting quality examines the association between UAF and the forward earnings response coefficient (FERC). The FERC is a measure of the extent to which information in current returns captures information in future earnings, introduced by Collins, Kothari, Shanken, and Sloan (1994). The FERC captures the decisionusefulness aspect of accounting information by measuring the extent to which the market predicts future earnings. Prior research show that firms with higher quality or more frequent disclosures exhibit higher FERC values (Gelb and Zarowin 2002; Lundholm and Myers 2002; Tucker and Zarowin 2006; Choi et al. 2008; and Orpurt and Zang 2009). Better quality accounting should allow investors to improve their future earnings forecasts, leading to a stronger association between current stock returns and future earnings, conditional on the information included in the current financial statements. Following Lundholm and Myers (2002) and Ettredge et al. (2005) we estimate the FERC using equation (5) below. Rt = a + b1Et-1 + b2Et + b3Et+1 + b4Rt+1 + b5UAFt + b6Et-1*UAFt + b7Et*UAFt + b8Et+1*UAFt + b9Rt+1*UAFt + t (5)

Where Rt is the total annual stock return measured over the 12 month period ending three months after fiscal year t. Et-1, Et, and Et+1 represent the net income for fiscal years t-1, t, and t+1, respectively, and all are scaled by the stock price at three months after the beginning of fiscal year t. Et-1 is included to function as an expectation for earnings in the current year and is included as a separate variable so that a specific form of the time-series process of earnings is not imposed (e.g., to impose a random walk Et-1 is subtracted from Et). b2 is the contemporaneous earnings response coefficient and is expected to be positive while b1 is expected to be negative.

26

b3 is the forward earnings-response coefficient (FERC), and as such, is the primary variable of interest. It is expected to be positive. Rt+1 is future period returns and is included because of the error-in-variables problem with Et+1 (Collins et al. 1994).13 If higher quality accounting translates into investors ability to incorporate future earnings information into price, then we expect b8 to be negative. This would indicate that for firms with higher accounting quality investors are able to better predict future earnings because higher values of UAF indicate lower accounting quality. We do not have predictions for the other interactions in this equation. Table 7 presents the results of estimating the forward-earnings response regressions. Panel A presents descriptive statistics for the variables used in the regression. Because of requirements for seven consecutive years of data and forward earnings and returns, the sample size is reduced to 19,713 firm-year observations. Panel B provides the results of estimating equation (5). Our primary variable of interest, the forward-earnings response coefficient, is identified with a border in Panel B. As predicted, there is a significant negative interaction between UAF and Et+1, indicating that firms characterized as having low (high) quality accounting have a significantly smaller (larger) forward earnings-response coefficients (b8 = -0.15, t-stat=-2.70). V. Additional properties of UAF and robustness checks 5.1 Association with innate firm characteristics Past research has examined the role of accounting quality in various settings such as asset pricing, capital allocation, CEO reputation (e.g., Francis et al. 2005; Chen, Shevlin, Tong 2007; Biddle et al. 2008). One of the challenges researchers face when documenting these associations
13

Theoretically, the variable we want to include is expected future earnings but Et+1 measures this with error. Rt+1 is correlated with the portion of Et+1 that is related to unexpected earnings news in period t+1 and, therefore, is included to control for this component of Et+1.

27

is that the proxy for earnings quality is capturing systematic firm characteristics that are unrelated to financial reporting, and it is these innate characteristics that drive the association (i.e., correlated omitted variables). Although variables can be included in a regression to try to rule out correlated omitted variables, it is often unclear which variables to include, how many to include, or the correct specification. Therefore, an important aspect when developing a measure of accounting quality is not just the association with the unobservable construct of interest (i.e., accounting quality), but also the lack of association with other potentially confounding constructs (i.e., innate firm characteristics). Recognizing that omitted variables bias is a source of endogeneity (i.e., correlation between the independent variable and the error term), an econometric concern when testing the effect of earnings quality is finding a good instrument, where the ideal instrument is correlated with the underlying construct of interest, but uncorrelated with the endogenous variables. Because the list of potentially omitted variables is limitless, we rely on documenting the association with known innate characteristics to provide the reader some assurance that UAF does not proxy for other obvious constructs that have been associated with other measures of quality. Figure 2 graphically displays the values of UAF and AQ, relative to percentiles of CFO, Ln(ASSETS), and ROA. AQ is provided as a benchmark for the association with the innate factors. Because the scaling of the quality measures differs, the left vertical axis displays the values of AQ, while the right vertical axis displays the values of UAF. This analysis reveals some interesting patterns. With respect to cash flow volatility in Panel A, AQ appears to be an increasing function of cash flow volatility, with high quality (low AQ) firms having very low

28

cash flow volatility, and low quality firms (high AQ) having high cash flow volatility. In contrast, there does not appear to be a discernable pattern between UAF and cash flow volatility. Panel B shows the association with size. Here again, AQ is a visibly decreasing function of size, indicating small firms tend to be characterized as having low quality earnings. With AQ smaller firms tend to have more volatile accruals and lower mapping of accruals to cash flows, leading to an overclassification of small firms as having low quality accounting. In contrast, there appears to be very little relation between Ln(ASSETS) and UAF. Thus, UAF should be useful for triangulating research findings that use proxies for earnings quality because it has little association with many of the innate characteristics. Finally, with respect to performance (ROA) in Panel C, both measures exhibit similar patterns. It appears the worst performers are associated with the lowest quality earnings, particularly below the 15th percentile. However, for the rest of the distribution there appears to be no discernable pattern, and it is debatable whether the association with extreme negative performance is a problematic omitted variables issue, or simply showing that extreme negative earnings firms have poor earnings quality. 5.2 Additional Tests Section 2.2 discussed economic bonding between the auditor and the client as a possible alternative explanation for the association between audit fees and accounting quality. In this section, we report the results of additional test that we conducted to rule out bonding as an explanation of our findings. Dhaliwal et al. (2008) show that the relation between non-audit fees and the cost of debt is consistent with non-audit fees capturing economic bonding and audit fees capturing earnings quality. Using data in Audit Analytics, we obtain the amount of non-audit fees, scaled by lagged total assets, for our sample of firms. If economic bonding is responsible

29

for the observed relation between unexpected audit fees and restatements, fraud, or comment letters, then we expect to find similar results when we conduct the same analysis using non-audit fees. Untabulated results show that when both UAF and non-audit fees are included in the restatement prediction model, UAF continues to be significant in all specifications. Non-audit fees are insignificantly related to both restatements and fraud. Overall, we do not find evidence that economic bonding is related to restatements or fraud. In all cases, UAF continues to be significant, indicating that it captures more than just economic bonding between the auditor and the client. We also compute a number of robustness tests to check the sensitivity of our results to different specifications. First, all the current tests use a continuous version of UAF as our measure of accounting quality. Because this is measured as a regression residual, there is the possible concern that extreme realizations of UAF drive the results. To examine this, we reestimate the tests using a ranked version of UAF, where the continuous variable is ranked into ten scaled deciles, ranging from zero to one. We re-estimate the restatement and fraud regressions using the decile ranked UAF instead of the continuous measure. In both settings, the decile ranked UAF measure continues to be statistically significant, and incrementally informative relative to other accounting quality proxies. Our final robustness test removes some variables from the audit fee equation that might be associated with questionable accounting quality. Despite being associated with audit fees, we remove the level of accounts receivable (REC), inventory(INV), and whether the firm experienced a loss in the prior three years (LOSS), because they might capture aspects of audit

30

risk related to accounting quality. In both the restatement and fraud settings, the test statistic on UAF becomes stronger when this modified version of the audit fee model is used.

VI. Conclusion Most existing measures of financial reporting quality focus on financial statement realizations, and are intended to capture earnings quality. In this paper, we introduce a new measure of accounting quality based on unexplained audit fees. Our measure reflects the auditors unobservable private information about the firms underlying accounting systems. A measure based on unexplained audit fees provides a more global assessment of financial statement quality than focusing exclusively on one facet of accounting quality, such as how well accruals map into cash flows. We set out to test whether an audit-fee based measure is incrementally informative relative to existing measures, and the extent to which it is confounded by innate characteristics. We show that our measure of accounting quality is positively correlated with other measures of earnings quality used in the literature, but also contains unique information. When benchmarked against four other accounting quality measures, we find that UAF is incrementally informative for predicting restatements, fraud, and SEC comment letters. Moreover, our measure of unexplained audit fees should help triangulate research questions that are interested in making inferences about accounting quality, because it shows much lower correlation with possible omitted variables, and when bias exists, it tends to be in the opposite direction of the bias in the accrual quality measure. Overall, we believe that unexplained audit fees can help researchers validate or refute theories regarding the economic consequence of financial reporting quality.

31

References Ashbaugh, H., D.W Collins, W.R. Kinney, and R LaFond 2008. The effect of SOX internal control deficiencies and their remediation on accrual quality. The Accounting Review 83(1): 217-250. Ashbaugh, H., R. LaFond, and B.W. Mayhew. 2003. Do non-audit services compromise auditor independence? The Accounting Review 78: 611-639. Becker, C., M. DeFond, J.Jiambalvo, and K. Subramanyam. 1998. The effect of audit quality on earnings management. Contemporary Accounting Research 15: 1-24. Bell, T.B., W.R. Landsman, and D.A. Shackelford. 2001. Auditors perceived business risk and audit fees: Analysis and evidence. Journal of Accounting Research 39(1): 35-43. Beneish, Messod D. 1999. Incentives and penalties related to earnings overstatements that violate GAAP. The Accounting Review 74(4): 425-457. Biddle, G.C., G. Hilary, and R.S. Verdi, R.S., 2008. How does financial reporting quality improve investment efficiency? MIT Working paper. Carcello, J. and Z.V. Palmrose. 1994. Auditor litigation and modified reporting on bankrupt clients. Journal of Accounting Research 32: 1-30. Chen. R., and R. Johnston. 2008. Securities and exchange commission comment letters: Enforcing accounting quality and disclosure. The Ohio State University working paper. Chen, S., T. Shevlin and Y. Tong. 2007. Are dividend changes associated with changes in the pricing of information risk? Journal of Accounting Research. 45(1): 1-40. Collins, D. W., S.P. Kothari, J. Shanken, and R. Sloan. 1994. Lack of timeliness and noise as explanations for the low contemporaneous return-earnings association. Journal of Accounting and Economics: 18: 289-324. Cronbach, J.L. and P.E. Meehl. 1955. Construct validity in psychological tests. Psychological Bulletin 52(4): 281-302. Dechow, P., Dichev, I., 2002. The quality of accruals and earnings: the role of accrual estimation errors. The Accounting Review 77 (Supplement), 3559. Dechow, P.M., W. Ge, C.R. Larson, and R.G. Sloan. 2008. Predicting material accounting manipulations. Working paper. Defond, M. 2009. Evaluating the quality of earnings quality proxies: A discussion of Dechow, Ge and Schrand. Journal of Accounting and Economics Presentation.

32

Dhaliwal, D.S., C.A. Gleason, S. Heitzman, and K. Melendrez. 2008. Auditor fees and cost of debt. Journal of Accounting, Auditing, and Finance, Forthcoming. Durnev, A., R. Morck, and B. Yeung. 2004. Value-enhancing capital budgeting and firm-specific stock return variation. The Journal of Finance 59(1): 65-105. Ettredge, M.L., S.Y. Kwon, D.B. Smith, and P.A. Zarowin. 2005. The impact of SFAS No. 131 Business Segment Data on the Markets Ability to Anticipate Future Earnings. The Accounting Review 80(3): 773-804. Financial Accounting Standards Board, 1980. Statement of Financial Accounting Concepts No. 2, Qualitative characteristics of accounting information. FASB, Stamford, Connecticut. Francis, J., R. LaFond, P. Olsson, K. Schipper, 2004. Costs of equity and earnings attributes. The Accounting Review 79(4): 9671010. Francis, J., R. LaFond, P. Olsson, K. Schipper, 2005. The market pricing of accruals quality. Journal of Accounting and Economics 39: 295-327. Francis, J., D. Philbrick, and K. Schipper. 1994. Shareholder litigation and corporate disclosures. Journal of Accounting Research 32(2): 137-164. Frankel, R., M.F. Johnson, and K.K. Nelson. 2002. The relation between auditors fees for nonaudit services and earnings management. The Accounting Review 77 (Supplement): 71105. Gompers, P., J. Ishii, and A. Metrick. 2003. Corporation governance and equity prices. Quarterly Journal of Economics, 118: 107-155. Hanlon, M., and G. Krishnan. 2009. Audit fees and book-tax difference Massachusetts Institute of Technology working paper. Hay, D., W.R. Knechel, and H. Ling. 2008. Evidence on the impact of internal control and corporate governance on audit fees. International Journal of Auditing 12: 9 24. Healy, P.M. 1985. The effect of bonus schemes on accounting decisions. Journal of Accounting and Economics 7: 85-107. Hennes, K., A. Leone, and B. Miller. 2008. The importance of distinguishing errors from irregularities in restatement research: The case of restatements and CEO/CFO turnover. The Accounting Review 83(6): 1487-1519. Hribar, P., and C. Nichols. 2007. The use of unsigned earnings quality measures in tests of earnings management. Journal of Accounting Research 45(5): 1017-1053. Jonas, G. J., and J. Blanchet. 2000. Assessing quality of financial reporting. Accounting Horizons

33

14 (3): 353-63. Knechel R.W., and J.L. Payne. 2001. Additional evidence on audit report lag. Auditing: A Journal of Practice & Theory 20(1): 137-146. Kothari, S.P., A. Leone, and C. Wasley. 2005. Performance matched unexplained accrual measures. Journal of Accounting and Economics 39: 745-767. Kravet, T., and T. Shevlin. 2009. Accounting restatements and information risk. Review of Accounting Studies, Forthcoming. Larcker, D.F. and S. Richardson. 2004. Fees paid to audit firms, accrual choices, and corporate governance. Journal of Accounting Research 42(3): 625-658. Leuz, C., D. Nanda, and P.D. Wysocki. 2003. Earnings management and investor protection: an international comparison. Journal of Financial Economics 69: 505-527. Levitt, S. 1997. Using electoral cycles in police hiring to estimate the effect of police on crime. The American Economic Review 87(3): 270-290. Liu, M., and P. Wysocki., 2007. Cross-sectional determinants of information quality proxies and cost of capital measures. Pennsylvania State University and MIT Sloan School of Management working paper. Lui, D., S. Markov, and A. Tamayo. 2007. What makes stock risky? Evidence from sell-side analysts risk ratings. Journal of Accounting Research. 43 (3): 629-665. Lundholm, R., and L. Myers. 2002. Bringing the future forward: The effect of disclosure on the returns-earnings relation. Journal of Accounting Research 40(3): 809-839. Magee, R.P., and M.C. Tseng. 1990. Audit pricing and independence. The Accounting Review 65: 315-336. Merced, M. and A.R. Sorkin. 2010. Lehman Brothers hid borrowing, examiner says. The new York Times (March 12). McNichols, M., 2002. Discussion of The quality of accruals and earnings: the role of accrual estimation errors. The Accounting Review 77 (Supplement), 61-69. Palmrose, Z.V., V. J. Richardson, and S. Scholz. 2004. Determinants of the market reactions to restatement announcements. Journal of Accounting and Economics 37: 59-89. Phillips, J., M. Pincus, and S. Rego. 2003. Earnings management: new evidence based on deferred tax expense. The Accounting Review 78: 491-521.

34

Picconi, M. and J.K. Reynolds. 2009. Audit fee theory and estimation: A consideration of the log-log audit fee model. Indiana University working paper. Richardson, S., I. Tuna, and M. Wu. 2002. Predicting earnings management: the case of earnings restatements. University of Pennsylvania working paper. Schipper, K. and L. Vincent. 2003. Earnings quality. Accounting Horizons (Supplement): 97110. Simunic, D. 1980. The pricing of audit services: Theory and evidence. Journal of Accounting Research 18(Spring): 161-190. Valukas, A. 2010. Examiners report of Lehman Brothers Holding, Inc. http://lehmanreport.jenner.com/VOLUME%201.pdf Venkataraman R., J.P. Weber, and M. Willenborg. 2008. Litigation Risk, audit quality and audit fees: Evidence from Initial Public Offerings. The Accounting Review 83(5): 1315-1345.

35

Table 1 Audit Fee Model


This table presents the results of estimating unexplained audit fees (UAF). Panel A presents descriptive statistics of the variable used in the audit fee model. Panel B presents the mean results of annual regressions of the audit fee model and t-statistics are calculated based on the standard errors of the seven annual coefficient estimates. Ln(AUDIT FEE)i,t is the natural logarithm of audit fees. BIG4 is a indicator variables that is equal to one if the firms auditor is a member of the Big 4 (or Big 5 before the exit of Arthur Andersen), and zero otherwise (data149). Ln(ASSETS) is the natural logarithm of total assets (data6). BUS SEG is the square root of the number of business segments of the firm as defined by Compustat segment file. FGN is the percentage of foreign sales to total sales and is calculated with Compustat segment data. INV is calculated as inventory (data3) divided by lagged total assets (data6). REC is accounts receivable (data2) divided by lagged total assets (data6). DEBT is long-term (data9) plus short-term debt (data34) divided by lagged total assets (data6). ROA represents return on assets and is defined as operating income after depreciation (data178) divided by lagged total assets (data6). LOSS is a indicator variable equal to one if net income before extraordinary items is negative in any of prior three years (year t, t-1, and t-2), and zero otherwise. AUD OPIN is an indicator variable that is equal to one if a firm receives anything other than a standard unqualified opinion (data149), and zero otherwise. CLIENT is the number of years that the firm has been a client of their current auditor. IPO is an indicator variable set equal to one if the firm underwent an initial public offering in year t, and zero otherwise. SEO is an indicator variable set equal to one if the firm experienced a seasoned equity offering in year t, and zero otherwise. ISSUANCE is an indicator variable that is set equal to one if the firm issued debt during year t, and zero otherwise. LITRISK is an indicator variable that is set equal to one if the firms primary SIC is 2833-2836, 3570-3577, 3600-3674, 5200-5961, 7370-7374, and zero otherwise.

Panel A: Descriptive Statistics for included variables


Lower Quartile -2.198 0.000 3.419 1.000 0.000 0.000 0.047 0.016 -0.095 0.000 0.000 1.732 0.000 0.000 0.000 0.000 Upper Quartile -0.180 1.000 6.921 1.732 0.263 0.163 0.231 0.430 0.112 1.000 1.000 3.000 0.000 0.000 0.000 1.000

Variables Ln(AUDIT FEE) BIG4 LN(ASSETS) BUS SEG FGN INV REC DEBT ROA LOSS AUD OPIN CLIENT IPO SEO ISSUANCE LITRISK

N 38,385 38,385 38,385 38,385 38,385 38,385 38,385 38,385 38,385 38,385 38,385 38,385 38,385 38,385 38,385 38,385

Mean -1.161 0.732 5.125 1.415 0.165 0.110 0.177 0.389 -0.237 0.599 0.409 2.469 0.024 0.052 0.032 0.335

Median -1.267 1.000 5.215 1.000 0.000 0.035 0.125 0.197 0.046 1.000 0.000 2.236 0.000 0.000 0.000 0.000

Std Dev 1.452 0.443 2.635 0.533 0.272 0.160 0.199 0.838 1.266 0.490 0.492 1.107 0.154 0.223 0.175 0.472

36

Table 1 (Continued) Panel B: Audit Fee Model Regression Ln(AUDIT FEE)i,t = 0 + 1BIG4i,t + 2Ln(ASSETS)i,t + 3BUS SEGi,t + 4FGNi,t + 5INVi,t + 6RECi,t + 7DEBTi,t + 8INCOMEi,t + 9LOSSi,t + 10AUD OPINi,t + 11CLIENTi,t + 12IPOi,t + 13SEOi,t + 14ISSUANCEi,t + 15LITRISKi,t + INDi,t + i,t

Pred.Sign INTERCEPT BIG4 Ln(ASSETS) BUS SEG FGN INV REC DEBT INCOME LOSS AUD OPIN CLIENT IPO SEO ISSUANCE LITRISK Industry Fixed Effects Firm-year Observations Mean Adjusted R-squared + + + + + + + ? + + + + + + +

Coeff -4.68 0.27 0.45 0.16 0.40 0.26 0.18 -0.02 -0.04 0.22 0.17 0.00 0.15 0.03 0.00 0.02 Yes 38,385 32.79%

t-stat -38.68*** 12.97*** 24.13*** 15.34*** 11.02*** 5.36*** 4.48*** -1.46 -0.87 22.44*** 16.12*** -0.23 3.62*** 0.83 -0.03 0.97

Asterisks *, **, and *** denote two-tailed (one-tailed when there is a predicted sign) statistical significance at 10%, 5%, and 1% respectively. All variables are winsorized at the 1st and 99th percentile except for indicator variables.

37

Table 2 Comparison of Firms in Bottom and Top Quartile of UAF Distribution


This table presents a comparison of financial variables for firms in the bottom and top quartiles of the distribution of UAF. UAF is calculated as the residual from annual cross-sectional regressions of the audit fee model. Assets and sales are the total assets (data6) and sales revenue (data12), respectively. ROA represents return on assets and is defined as operating income after depreciation (data178) divided by beginning total assets (data6). LEV is long-term (data9) plus short-term debt (data34) divided by beginning total assets (data6). BTM is the book to market ratio calculated as book value of equity (data60) divided by market value (data25*data199). All variables are winsorized at the 1st and 99th percentile.

UAF Distribution
Full Sample (N = 35,492) Variables UAF Assets Sales ROA LEV BM Mean 2,259 1,525 -0.34 0.30 0.44 Median 175 128 0.02 0.17 0.43 Stdev 7,081 4,697 1.57 0.56 1.28 Bottom Quartile (high quality) (N = 8,873) Mean 2,119 1,189 -0.39 0.28 0.52 Median 165 93 0.01 0.15 0.46 Stdev 6,693 4,054 1.80 0.53 1.14 Top Quartile (low quality) (N=8,873) Mean 2,464 1,772 -0.39 0.34 0.28 Median 190 161 0.01 0.19 0.36 Stdev 7,542 5,016 1.61 0.63 1.43

38

Table 3 Association between Unexplained Audit Fees and Other Measures of Accounting Quality
This table compares unexplained audit fees to other measures of earnings quality. Panel A presents descriptive statistics for the various proxies of accounting quality. Panel B presents the Pearson (top) and Spearman (bottom) correlation coefficients and bolded amounts represent significance at .05 or less. Panel C presents the results from regressions of UAF on variables representing accounting quality proxies. UAF is calculated as the residual from annual cross-sectional regressions of the audit fee model. TACC is total accruals scaled by total assets. AQ is the standard deviation of firm is residuals from a regression of current accruals on lag, current, and future cash flows from operations. ABS(DA) is the absolute value of discretionary accruals from the performance-adjusted modified cross-sectional Jones model. SMOOTH is the ratio of firms is standard deviation of earnings before extraordinary items (scaled by total assets) to the standard deviation of cash flows from operations (scaled by total assets). CFO is the standard deviation of the firms rolling ten-year cash flow from operations. All regression variables are winsorized at the 1st and 99th percentile.

Panel A: Descriptive Statistics


Lower Quartile -0.32 0.04 -0.11 0.02 0.62 0.04 Upper Quartile 0.34 0.14 -0.01 0.10 1.43 0.14

Variables UAF AQ TACC ABS(DA) SMOOTH CFO

N 25,415 25,415 25,415 25,415 25,415 25,415

Mean 0.01 0.11 -0.09 0.08 1.16 0.13

Std.Dev. 0.53 0.07 -0.06 0.05 0.97 0.08

Median 0.00 0.12 0.23 0.10 0.84 0.17

Panel B: Pearson (top) and Spearman (bottom) Correlation Coefficients


UAF UAF AQ TACC ABS(DA) SMOOTH CFO 0.08 -0.06 0.03 0.08 0.04 AQ 0.08 -0.16 0.47 0.37 0.69 TACC -0.05 -0.34 -0.12 -0.15 -0.15 ABS(DA) 0.05 0.54 -0.26 0.14 0.44 SMOOTH 0.08 0.39 -0.16 0.14 0.10 CFO 0.05 0.62 -0.35 0.47 0.04

39

Table 3 (Continued) Panel C: Regression of UAF on Other Accounting Quality Proxies


Model: UAFi,t = 1 + 2AQi,t + 3TACCi,t + 4ABS(DA)i,t + 5SMOOTHi,t + 6CFOi,t + i,t

Pred. Sign INTERCEPT AQ TACC ABS(DA) SMOOTH CFO ? + + + + ?

UAF Coeff t-stat -0.06 0.18 -0.06 0.01 0.04 0.04 25,415 1.05% -5.89 3.02 -2.81 0.30 5.09 1.08
*** *** *** ***

UAF Coeff t-stat -0.06 0.19 -0.06 -5.90 3.16 -2.82


*** *** ***

UAF Coeff t-stat -0.06 -6.10


***

-0.06 0.07

-3.10 1.53 6.78 3.00

*** * *** ***

0.04 0.04 25,415 1.05%

5.08 1.14

***

0.05 0.10 25,415 0.98%

Firm-year Observations Adjusted R-squared

Asterisks *, **, and *** denote two-tailed (one-tailed when there is a predicted sign) statistical significance at 10%, 5%, and 1% respectively. All regressions include heteroskedasticity-consistent standard errors that have been adjusted for clustering at the firm level.

40

Table 4 Association between Unexplained Audit Fees and Internal Control Deficiencies
This table examines the association between unexplained audit fees and internal control deficiencies. Panel A presents descriptive statistics. Panel B presents the results from a regression of UAF on ICDt-1. UAF is calculated as the residual from annual cross-sectional regressions of the audit fee model. ICDt-1 is an indicator variable set equal to 1 for firms indicating an internal control deficiency in period t-1. All regressions include heteroskedasticity-consistent standard errors that have been adjusted for clustering at the firm level.

Panel A: Descriptive Statistics (N = 4,883)


Variables UAFt ICDt-1 Mean 0.10 0.13 Std.Dev. 0.51 0.34 Median 0.08 0.00 Lower Quartile -0.22 0.00 Upper Quartile 0.42 0.00

Panel B: Regression of UAF on Internal Control Deficiencies


Pred. Sign INTERCEPT ICDt-1 Firm-year Observations Adjusted R-squared ? + UAFt Coeff 0.05 0.39 4,883 6.67% t-stat 5.98*** 14.42***

The t-statistics are White adjusted. Asterisks *, **, and *** denote two-tailed (one-tailed when there is a predicted sign) statistical significance at 10%, 5%, and 1% respectively.

41

Table 5 Association between Unexplained Audit Fees and Restatements


This table examines if unexplained audit fees are predictive of accounting restatements and fraud in the year of the transgression. Table 5 presents logisitic regressions predicting restatements and fraud, both conditionally and unconditionally, with varying levels of controls. RESTATEMENTi,t is an indicator variable that is equal to one if any period in firm is fiscal year t financial restatements are subsequently restated, and zero otherwise. FRAUD is an indicator variable that is equal to one if the restatement is the result of fraud, and zero otherwise. UAF is calculated as the residual from annual cross-sectional regressions of the audit fee model. TACC is total accruals scaled by total assets. AQ is the standard deviation of firm is residuals from a regression of current accruals on lag, current, and future cash flows from operations. ABS(DA) is the absolute value of discretionary accruals from the performanceadjusted modified cross-sectional Jones model. SMOOTH is the ratio of firms is standard deviation of earnings before extraordinary items (scaled by total assets) to the standard deviation of cash flows from operations (scaled by total assets). CFO is the standard deviation of the firms rolling five-year cash flow from operations. REC is the change in REC where REC is defined as receivables (data2) divided by lagged total assets. INV is the change in INV where INV is defined as inventory (data3) divided by lagged total assets. CASH SALES is the percentage change in cash sales where cash sales is calculated as sales (data12) minus REC. NIBE is the change in ROA where ROA is defined as net income before extraordinary items (data18) divided by lagged total assets. EMPLOYEES is the abnormal change in employees and is defined as the percentage change in the number of employees (data29) minus the percentage change in total assets (data6). BTM is the book to market ratio calculated as book value of equity (data60) divided by market value (data25*data199). All regression variables are winsorized at the 1st and 99th percentile except for indicator variables.

Logistic Regression of Restatement Probability on UAF, Other Accounting Quality Proxies, Audit Effort Variables, and Other Control Variable
Intercept -0.006 (0.924) Accounting Quality Proxies 0.169 UAF + (0.053) * AQ + ABS(DA) SMOOTH TACC
+ + +

RESTATEMENT -0.250 -0.362 (0.048) ** (0.016) 0.261 (0.023) 2.210 (0.010) -0.440 (0.350) 0.078 (0.200) 0.334 (0.277) 0.284 (0.016) 1.731 (0.068) -0.598 (0.304) 0.102 (0.151) 0.076 (0.451) 0.756 (0.396) 0.238 (0.850) 1.823 (0.352) -0.026 (0.859) -0.251 (0.625) -0.374

**

-0.026 (0.794) 0.527 (0.003)

***

FRAUD 0.204 (0.403) 1.023 (<.001) 1.136 (0.225) -2.347 (0.122) -0.103 (0.261) 1.289 (0.141)

-0.044 (0.887) 1.084 (<.001) -0.181 (0.462) -3.301 (0.065) -0.015 (0.466) 1.655 (0.125) 2.934 (0.170) -1.467 (0.532) 1.876 (0.585) -0.231 (0.299) 0.136 (0.916) -0.108

**

**

***

***

***

***

Other Controls CFO REC INV CSALES NIBE EMP


? ? ? ? ? ?

42

Table 5 (Continued) (0.092) 0.026 (0.739)


*

BTM

(0.845) 0.176 (0.254) 280 20.56 0.110 140 140

17.14% -0.30% -4.65% -0.31% 3.67% P-values are presented in parentheses below coefficients. Asterisks *, **, and *** denote two-tailed (one-tailed when there is a predicted sign) statistical significance at 10%, 5%, and 1% respectively.

N 1,168 790 790 402 280 Wald 2 2.60 14.39 19.12 7.48 16.78 Pseudo-R2 0.003 0.026 0.035 0.025 0.088 Number of Observations where: DV = 0 584 395 395 201 140 DV = 1 584 395 395 201 140 Change in the Probability that DV=1 When Variables Change from Quartile 1 to Quartile 3 UAF 2.95% 4.55% 4.85% 8.79% 16.19% AQ 3.84% 3.02% 1.86% ABS(DA) -0.65% -0.88% -3.31% SMOOTH 1.62% 2.07% -2.08% TACC 0.74% 0.17% 2.86%

43

Table 6 Association between Unexplained Audit Fees and Comment Letters


This table examines if unexplained audit fees are predictive of comment letters. Panel A presents descriptive statistics for the variables used in these tests. Panels B examine logisitic regressions predicting comment letters, both conditionally and unconditionally, with varying levels of controls. COMMENTi,t is an indicator variable that is equal to one if firm is fiscal year t financial restatements are the subject of an SEC comment letter, and zero otherwise. UAF is calculated as the residual from annual cross-sectional regressions of the audit fee model. AQ is the standard deviation of firm is residuals from a regression of current accruals on lag, current, and future cash flows from operations. ABS(DA) is the absolute value of discretionary accruals from the performance-adjusted modified cross-sectional Jones model. SMOOTH is the ratio of firms is standard deviation of earnings before extraordinary items (scaled by total assets) to the standard deviation of cash flows from operations (scaled by total assets). TACC is total accruals scaled by total assets. Log(Market_Value) is the natural logarithm of the firms market value in the current year. Salesvol is the standard deviation of sales over the prior five years. Sync is the logistic transformation relative idiosyncratic volatility estimated from the market model over the prior year. Rating is the transformed S&P credit rating for the current year which equals 0 for firms who do not have ratings and ranges from 1 for AAA to 22 for D for firms with ratings. No_Rating is an indicator variable set equal to one if the firm does not have an S&P credit rating in the prior year, and zero otherwise. Restatement is an indicator variable set equal to one if the firm has filed restatements before, and zero otherwise. Proportion is the firms share of industry revenue in the current year. All regression variables are winsorized at the 1st and 99th percentile except for indicator variables.

Logistic Regression of Comment Letters on UAF, Other Accounting Quality Proxies, and Other Control Variables
Intercept Accounting Quality Proxies UAF + AQ ABS(DA) SMOOTH TACC Other Controls EP log(Market_Value) Salesvol Sync Rating
+ + + +

-0.009 (0.829) 0.259 (<.001)

COMMENT LETTER -0.058 (0.46) 0.254 (<.001) 0.122 (0.401) -0.106 (0.435) 0.019 (0.371) -0.545 (0.105)

0.132 (0.77) 0.224 (0.01) 0.166 (0.375) -0.019 (0.489) 0.023 (0.355) -0.584 (0.098) 0.128 (0.333) -0.003 (0.465) 0.364 (0.131) -0.016 (0.231) -0.009 (0.331)

***

***

***

+ + + -

44

Table 6 (Continued) No_Rating Restatement Proportion


? + +

-0.298 (0.146) 0.132 (0.091) -0.244 (0.210) 2,276 13.27 0.78% 1,138 1,138 2,276 20.84 1.23% 1,138 1,138

N 2,676 Wald 2 11.43 Pseudo-R2 0.68% Number of Observations where: DV = 0 1,338 DV = 1 1,338

P-values are presented in parentheses below coefficients. Asterisks *, **, and *** denote two-tailed (one-tailed when there is a predicted sign) statistical significance at 10%, 5%, and 1% respectively.

45

Table 7 Association between the Forward Earnings Response Coefficient (FERC) and UAF
This table examines whether UAF is associated with the ability of the market to price future earnings. Panel A presents descriptive statistics of the variables used in tests of the FERC. Panel B presents the results from the pooled cross-sectional regression of the model indicated. Rit is the annual stock return for year t, measured over the 12-month period ending three months after the firms fiscal year end. Ei,t-1, Ei,t, and Ei,t+1 are the income available to common shareholders before extraordinary items for the year t-1, t, and t+1, respectively, and all are deflated by the market value of equity three months after the year t-1 fiscal year end. UAF is calculated as the residual from annual cross-sectional regressions of the audit fee model.

Panel A: Descriptive Statistics (N = 19,713)


Variables Et + 1 Eit Eit-1 Rit+1 Rit UAFit Mean -0.03 -0.07 -0.06 0.24 0.22 0.01 Median 0.04 0.03 0.03 0.11 0.10 0.00 Std. Dev. 0.23 0.35 0.36 0.73 0.76 0.53 Lower Quartile -0.04 -0.05 -0.04 -0.17 -0.22 -0.31 Upper Quartile 0.07 0.06 0.06 0.44 0.45 0.34

Panel B: Regression Results with UAF


Pred. INTERCEPT Eit-1 Eit Eit+1 Rit+1 UAFit UAFit Eit-1 UAFit Eit UAFit Eit+1 UAFit Rit+1 Firm-year Observations Adjusted R2 Sign ? + + ? ? ? ? Coeff 0.30 -0.35 0.33 0.75 -0.21 -0.01 0.16 -0.01 -0.15 0.03 19,713 10.86% Rit t-stat 52.20*** -13.93*** 14.88*** 23.62*** -25.87*** 1.19 3.81*** -0.37 -2.70*** 1.74*

46

Figure 1 Accounting Quality and Innate Characteristics

Auditors Identify Low Quality Accounting

Higher Engagement Risk

Increased testing to reduce overall engagement risk to acceptable levels

Higher Audit Fees

Fee premium in exchange for accepting higher engagement risk

Higher Audit Fees

47

Figure 2 Values of UAF and AQ relative to innate firm characteristics


These figures show values of UAF and AQ, relative to percentiles of different innate firm characteristics. Panel A plots UAF and AQ against percentiles of the standard deviation of operating cash flows, measured over the past 5 years. Panel B plots UAF and AQ against percentiles of size, measure as the log of total assets. Panel C plots UAF and AQ against performance, measured as net income divided by lagged total assets.

Panel A: Standard Deviation of Cash Flows and UAF/AQ

0.25

0.12 0.1

0.2 0.08 0.06 UAF 0.04 0.1 0.02 0 0.05 -0.02 0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100 Percentiles of Standard Deviation of Cash Flows -0.04
AQ UAF

0.15 AQ

48

Figure 3 (continued) Panel B: Log of total assets and UAF/AQ


0.2 0.18 0.16 0.14 0.12 AQ 0.1 0.08 0.06 0.04 0.02 0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95100 Percentiles of Firm Size 0.08 0.07 0.06 0.05 0.04 0.03 0.02 0.01 0 -0.01 -0.02
AQ UAF

Panel C: Return on Assets and UAF/AQ


0.25 0.15

0.2

0.1

0.15
AQ

0.05
UAF
AQ UAF

0.1

0.05

-0.05

0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100 Percentiles of ROA

-0.1

UAF

49

You might also like