Is There a Relation Between Residual Audit Fees and Analysts’ Forecasts?

Abstract

We examine the relationship between residual audit fees and the ability to predict future earnings. Recent research suggests that residual audit fees contain information about accounting quality. However, residual audit fees could either represent high accounting quality or a risk premium for low accounting quality. We extend this literature by providing evidence that residual audit fees are indicative of a lower quality information environment which has a negative impact on investors’ ability to anticipate future earnings. Specifically, we first show that residual audit fees are negatively associated with the ability of current earnings to predict future earnings. Furthermore, residual audit fees are negatively associated with analyst forecast accuracy and positively associated with the dispersion in analyst forecasts. Overall, our results are consistent with the notion that residual audit fees are indicative of poor earnings quality, and that this lower quality manifests itself in a lower quality information environment for investors and analysts.

Keywords

audit fees earnings quality analysts’ forecasts forecast dispersion forecast accuracy

Introduction

The purpose of this study is to examine whether residual audit fees are associated with the ability to anticipate future earnings. Specifically, we first examine the relation between residual audit fees and earnings persistence. We then investigate the association between residual audit fees and two properties of financial analysts’ earnings forecasts—forecast accuracy and forecast dispersion. Despite the fact that auditors and financial analysts are the two primary information intermediaries that play a critical role in the functioning of the capital markets by enhancing the credibility or by analyzing the quality of information in financial statements, there is very little empirical evidence on how auditing affects financial analysts’ decision making.¹ In other words, there is a paucity of evidence on how the relationship between audit fees and financial information influences the properties of analysts’ forecasts. Theoretical literature argues that high residual fees could reflect both a premium paid to auditors by risky firms and the extra effort that auditors would exert if they perceive a firm to be high risk (e.g., Dontoh, Ronen, & Sarath, 2013). Consequently, higher residual fees could, depending on the context, translate into either higher or lower quality audits.

To the extent that residual audit fees measure auditor effort, audit quality should increase. However, Hribar, Kravet, and Wilson (2014) argue that while auditors increase audit effort when they perceive the client to be high risk, such effort may improve accounting quality at the margin but does not necessarily transform low accounting quality to high accounting quality. If true, then residual audit fees would be associated with a lower quality information environment and would result in the decreased ability of investors to predict future earnings. Both theoretical arguments and the results of Hribar et al. (2014) suggest that higher residual audit fees are associated with greater uncertainty about audit quality and the financial report.

Our results suggest that higher residual audit fees are associated with a lower quality information environment, consistent with the finding of Hribar et al. (2014). We document several key findings. First, the ability of current earnings to predict future earnings, that is, earnings persistence, is significantly lower for observations with high residual audit fees. Second, residual audit fees are negatively associated with analyst forecast accuracy. This association between residual audit fees and analyst forecast accuracy appears to be economically significant. For a 1 standard deviation increase in residual audit fees, analyst forecast accuracy declines by 7.56%. Third, residual audit fees are positively related to analyst forecast dispersion. For a 1 standard deviation increase in residual audit fees, analyst forecast dispersion increases by 5.79%. Overall, our results are consistent with those of Hribar et al. (2014) and suggest that residual audit fees proxy for greater uncertainty about the quality of audits.

Our study has several potential implications for capital market participants. First, persistence is a particularly important attribute of earnings from a valuation perspective. Specifically, the time-series property of earnings is a critical element of various theoretical valuation models (Ohlson, 2007). Thus, empirical evidence on the relation between residual audit fees and earnings persistence is potentially important to capital market participants. Second, as indicated earlier, the linkages between auditing and financial analysts’ decision making remain largely unexplored in accounting research. Prior research suggests that financial analysts serve as a key provider of information to the capital markets (Schipper, 1991), and that earnings quality and audit quality² are expected to affect analysts’ earnings forecasts (Behn, Choi, & Kang, 2008; Healy, Hutton, & Palepu, 1999; Lang & Lundholm, 1996; Williams, 1996). While prior research has documented a link between residual audit fees and financial reporting quality (Doogar, Sivadasan, & Solomon, 2015; Hribar et al., 2014), prior research has not investigated a direct link with users of financial statement information. We attempt to build upon this stream of research by providing empirical evidence of whether and how residual audit fees are related to analysts’ forecast accuracy and forecast dispersion.

We also contribute to the growing literature on the relation between audit fees and financial reporting quality. There is some controversy as to whether the residual audit fee proxies for incremental auditor effort demanded by the audit client that is not fully controlled for in the audit fee model (e.g., Knechel & Willekens, 2006) or alternatively, proxies for poor accounting quality (Hribar et al., 2014).³ We address this issue by investigating the relation between residual audit fees and earnings predictability. Specifically, we provide consistent evidence that residual audit fees are associated with not only less persistent earnings but also analyst forecasts that are less accurate and more dispersed.

In addition, we contribute to the literature on analyst forecasts by extending prior studies which suggest that financial analysts use audited financial statement information in predicting future earnings (e.g., Abarbanell & Bushee, 1997; Barth & Hutton, 2004; Behn et al., 2008). Specifically, we document that residual audit fees are negatively associated with properties of analyst forecasts. This finding suggests that poor earnings quality leads to a poor information environment which affects the publicly available information that financial analysts use in formulating their forecasts.

The rest of this article is organized as follows. The next section provides background and develops the hypotheses. Section “Variable Measurement and Empirical Design” explains the empirical models. Section “Data and Empirical Results” describes the sample selection and presents the results, and Section “Conclusion” concludes the article.

Background and Hypotheses Development

Audit fees have received considerable amount of attention from standard setters, investors, and academics, particularly because the well-known audit failures have led to a number of important regulatory changes such as the Sarbanes–Oxley Act (SOX). The seminal work by Simunic (1980) spawned an expanding body of literature that has examined a large number of drivers of audit fees. In general, research has shown that audit fees are determined by the amount of audit effort and the level of the auditor’s litigation risk. However, DeFond and Zhang (2014) note that this relationship limits the interpretability of many audit fee studies because most studies are unable to address whether higher fees are a result of increased effort or increased risk. For example, on one hand, Caramanis and Lennox (2008) provide evidence that increased effort increases financial reporting quality, which would imply that an increase in audit fees should be interpreted as an increase in audit quality. However, Lyon and Maher (2005) provide evidence that audit firms charge higher audit fees to clients with higher business risk, which is consistent with higher fees signaling risk and in turn, lower quality earnings. Hogan and Wilkins (2008) provide evidence that riskier firms pay higher audit fees, but acknowledge that they cannot disentangle whether this higher fee is a result of increased effort or the existence of a risk premium.

More recently, Hribar et al. (2014) argue that audit fees incorporate the expected cost of poor earnings. Therefore, in response to the risk of reputation and litigation costs associated with low-quality accounting, auditors could increase the risk premium, increase the scope of their audit, or both. All of these actions will lead to higher audit fees. As a result, the level of unexplained audit fees based on the determinants in the audit fee model will provide information about the auditor’s assessment of the auditee’s accounting system. That is, when an audit firm assesses a client’s accounting quality as low, it will increase both audit hours and audit fees. However, Hribar et al. (2014) argue that in such a setting, increased audit effort only improves accounting quality at the margin but will not necessarily transform accounting quality from low to high. They predict and find a negative relation between residual audit fees and accounting quality as proxied by accruals quality, earnings smoothness, fraud, restatements, and Securities and Exchange Commission (SEC) comment letters.

Our study builds upon this prior research by examining the relation between residual audit fees and accounting quality according to the broad definition of Dechow and Schrand (2004):

Accounting quality is the extent to which accounting information accurately reflects the company’s current operating performance, is useful in predicting future performance, and helps assess firm value. (Hribar et al., 2014, p. 511)

Specifically, we investigate whether unexplained audit fees, as a signal of accounting quality, are associated with the ability to anticipate future performance. We examine the associations between current earnings and future earnings because they are considered to be an indicator of earnings quality (e.g., Dechow & Schrand, 2004; Schipper & Vincent, 2003). Furthermore, recent survey of CFOs also indicates that sustainable and persistent earnings is the most common idea of earnings quality (Dichev, Graham, Harvey, & Rajgopal, 2013).

If the residual audit fee reflects increased audit effort that increases audit quality, we expect a positive relation between residual audit fees and earnings persistence. However, to the extent that residual audit fee reflects the auditors’ unobservable private information about a firm’s underlying accounting system, there will be a negative relation between residual auditor fees and earnings persistence, holding other things constant. Given the contrasting views above, we formulate our first hypothesis in a non-directional manner:

Hypothesis 1: Residual audit fee is related to the correlation between current earnings and future earnings.

Behn et al. (2008) suggest that higher quality audits lead to more reliable accounting earnings, which improves users’ decision making, that is, the properties of earnings forecasts issued by financial analysts. Previous literature suggests that analysts, a sophisticated group of financial statement users, use historical earnings information in predicting firms’ future earnings. For example, Schipper (1991) notes that analysts assimilate and process publicly available information such as past earnings and prices. Abarbanell and Bushee (1997) find that historical earnings explain variation in analysts’ forecast revisions. These studies suggest that analysts’ forecasting ability likely increases with the quality (i.e., reliability) of financial information that they use to predict future earnings; the more the historical earnings information contains errors that are not indicative of the underlying economic situation or future firm performance, the less likely that analysts issue accurate forecasts.

Taken together, to the extent that the residual audit fee reflects the quality of earnings used for forecasting, it is likely to influence analyst forecast properties. However, given the uncertainty about the association between residual audit fees and audit quality, it is unclear how the properties of analyst forecasts will be affected. Specifically, if higher residual audit fees represent higher effort leading to increased audit quality, then analyst forecast accuracy will be higher for firms with higher residual audit fees. However, if residual audit fees represent a lower quality information environment, then analyst forecast accuracy would be lower for firms with higher residual audit fees. We thus present the following hypothesis on the relation between audit fees and analyst forecast accuracy:

Hypothesis 2: Residual audit fee is associated with analyst earnings forecast accuracy.

Audit fees might also relate to analysts’ forecast dispersion, which reflects uncertainty about the firm’s information environment (e.g., Imhoff & Lobo, 1992; Payne & Robb, 2000). Imhoff and Lobo (1992) interpret forecast dispersion as a proxy for ex ante earnings uncertainty (i.e., uncertainty about earnings before they are announced).⁴ Similar to the discussion above, to the extent that residual audit fee relates to higher financial reporting quality, analyst forecast dispersion will be lower. However, to the extent that residual audit fee relates to lower financial reporting reliability, which can increase the uncertainty in the firm’s information environment, analyst forecast dispersion may be higher for firms with greater residual fees.⁵ Accordingly, we formulate our forecast dispersion hypothesis as follows:

Hypothesis 3: Residual audit fee is associated with analyst earnings forecast dispersion.

Variable Measurement and Empirical Design

Variable Measurement

Residual audit fees

The seminal study of Simunic (1980) posits that audit fees are a positive function of three client-specific factors: client size, client complexity, and client-specific risk. Prior empirical studies generally provide supporting evidence (Chaney, Jeter, & Shivakumar, 2004; DeFond, Raghunandan, & Subramanyam, 2002; Whisenant, Sankaraguruswamy, & Raghunandan, 2003). Subsequent studies argue and provide empirical evidence that auditor characteristics, for example, Big N auditors, auditor specialization (Craswell, Francis, & Taylor, 1995; DeAngelo, 1981b; Francis, 1984; Francis, Reichelt, & Wang, 2005; Francis & Stokes, 1986); auditor–client relation, for example, initial audit engagement and auditor tenure (Craswell & Francis, 1999; Davis, Soo, & Trompeter, 2009; DeAngelo, 1981a; Ettredge & Greenberg, 1990); regulation, for example, SOX Act (Ghosh & Pawlewicz, 2009; Raghunandan & Rama, 2006), affect the audit fees. We follow prior studies (e.g., Hribar et al., 2014; Sankaraguruswamy & Whisenant, 2009) to model audit fee as follows (firm and time subscripts are omitted)⁶:

\begin{matrix} LNAUDFEE = & β_{0} + β_{1} \times LNTA + β_{2} \times SQSEGS + β_{3} \times INVREC + β_{4} \times FOROPS \\ + β_{5} \times CR + β_{6} \times BTM + β_{7} \times LEV + β_{8} \times SQEMPLS \\ + β_{9} \times M & A + β_{10} \times DECEMBER + β_{11} \times ROA + β_{12} \times LOSS \\ + β_{13} \times GC + β_{14} \times INITIAL + β_{15} \times SHORT + β_{16} \times MEDIUM \\ + β_{17} \times AGE_SEC + β_{18} \times NEW_FIN + β_{19} \times RISK_IND \\ + β_{20} \times IC_AUDIT + β_{21} \times IC_ADV + β_{22} \times RESTATE \\ + Industry Effects + ε . \end{matrix}

The variables are defined in the appendix.

Specifically, we regress the natural logarithm of total auditing fees on the determinants of audit fees, including industry fixed effects by year. For each firm, the predicted value is obtained by applying the estimated coefficient to its corresponding determinant. The predicted value is referred to as the normal fees charged by the auditor for a given engagement. Correspondingly, the residual from this regression serves as our proxy for residual audit fees. It is possible that there could be potential measurement error given the increasing trend in the fees paid to audit firms between 2003 and 2010 because there has been a significant structural change in audit fees since the passage of SOX. However, we mitigate this concern because our residual audit fees are based on audit fees estimated by year, which allows the intercept and coefficients to vary by year and control for any year-specific events.

For our set of firm characteristics, we include firm size (LNTA), measured as the natural logarithm of total assets. To account for firm complexity, we include SQSEGS, the square root of the number of segments; INVREC, measured as the inventory plus accounts receivable deflated by total assets at the beginning of the year; FOROPS, an indicator variable equal to 1 if the audit client has foreign operations; SQEMPLS, the square root of the number of employees; M&A, indicator variable equal to 1 if the client engaged in a merger and acquisition and 0 otherwise; NEW_FIN, an indicator variable equal to 1 if the audit client has material new equity or debt issue; and BTM, book value of equity divided by market value of equity. For firm risk, we include CR, the ratio of current assets to current liabilities; LOSS, an indicator variable equal to 1 if the audit client reported a loss in the current fiscal year; ROA, return on assets; LEV, the sum of short-term debt and long-term debt scaled by total assets; RESTATE, an indicator variable equal to 1 if the audit client announced a restatement in the previous year; RISK_IND, equal to 1 if the firm is in the high technology industry; AGE_SEC, number of fiscal years since the company’s initial public offering.

We use indicator variables, where the variable is equal to 1 if the condition exists, and 0 otherwise, to proxy for auditor–client relation, special audit engagement characteristics, and regulation requirements. Specifically, we include whether the firm’s auditor has been with the client for less than 1 year (INITIAL); if the firm’s auditor has been with the client for 2 or 3 years (SHORT); if the firm’s auditor has been with the client for 4 to 15 years (MEDIUM); if the annual report includes an opinion on management’s assessment of internal control (IC_AUDIT); if the firm’s internal control is ineffective (IC_ADV); if the audit opinion includes a going concern modification (GC). Furthermore, we control for industry fixed effects. Our measure of residual audit fees (AB_AUD) is the residual of Model 1 estimated by year.

Analyst forecast accuracy and dispersion

Institutional Brokers Estimate Service (I/B/E/S) has U.S. Detail History and Summary History files for properties of analyst forecasts. The Summary History file contains the actual data and summary statistics, such as mean, median, and standard deviation values for analyst forecasts. The Detail History file contains actual data and individual analysts’ forecasts. I/B/E/S adjusts forecasts and actual data for stock splits and rounds them to the nearest cent in the summary file, which could artificially reduce forecast errors and forecast dispersion.⁷ By comparison, in the detailed file, I/B/E/S adjusts individual forecasts and actual data for stock splits and rounds them to one hundredth of 1 cent. Thus, the measurement errors are smaller when using the detail file. Therefore, we obtain analysts’ earnings forecast and actual earnings per share data from I/B/E/S detailed historical file to measure analyst forecast accuracy and forecast dispersion. Specifically, we obtain the annual earnings forecasts issued within 90 days before the year t earnings announcement and ending 3 days before the announcement.⁸ If an analyst makes more than one forecast during this period, we use the most recent earnings forecast prior to the announcement of earnings. These individual annual earnings forecasts are further adjusted to the basis of the number of shares outstanding as of the earnings announcement date for the calculations of mean, median, and standard deviation of analysts’ earnings forecast. The obtained mean of annual forecasts is denoted as FORECAST.⁹ Forecast accuracy is defined as follows:

ACCY = (- 1) \times \frac{| ACTUAL - FORECAST |}{PRICE} .

ACTUAL is the actual earnings per share reported in the I/B/E/S detail file. PRICE is the stock price of the fiscal year-end date, adjusted to the basis of the number of shares outstanding as of the earnings announcement date. We multiply this absolute forecast error by (−1) to make the forecast accuracy (ACCY) increase with greater forecast accuracy.

We measure forecast dispersion (DISP) as the standard deviation of individual analysts’ most recent earnings forecasts issued during the period starting 90 days before the corresponding earnings announcement and ending 3 days before the announcement, scaled by the stock price at the end of fiscal year-end.

Empirical Models

Earnings persistence

We use the following models to test our first hypothesis regarding the impact of residual audit fees on the association between current earnings and 1-year-ahead earnings:

\begin{matrix} EARN_LEAD = β_{0} + β_{1} EARN + β_{2} LOSS + β_{3} LOSS \times EARN + β_{4} STDROA \\ + β_{5} STDROA \times EARN + β_{6} SIZE + β_{7} SIZE \times EARN \\ + β_{8} R_AB_AUD + β_{9} R_AB_AUD \times EARN + Year dummies + ε, \end{matrix}

where EARN_LEAD is the income available to common shareholders before extraordinary items in year t+ 1 scaled by market value of equity at the beginning of year t+ 1; EARN is the income available to common shareholders before extraordinary items in year t scaled by market value of equity at the beginning of year t; ACCR is accrual component of earning at year t; CFO is cash flow component of earnings at year t; SIZE is the natural log market value of equity; STDROA is standard deviation of ROA (return on asset) for the period of 5 years (year t− 4 to year t); and LOSS is 1 if the company had negative ROA, 0 otherwise.

In this regression, we use R_AB_AUD, decile ranked measure of AB_AUD by year to make the interpretation of coefficient on the interaction term straightforward. The results are qualitatively similar when we use the raw measure as of AB_AUD. Firm and time subscripts are omitted from models. Model 2 estimates 1-year-ahead earnings persistence. R_AB_AUD×EARN allows earnings persistence to vary across the level of residual audit fees. If higher residual audit fees indicate lower quality earnings, we predict the coefficient on R_AB_AUD×EARN to be negative.

In Model 2, we control for several variables that have been found to affect the earnings persistence and the association between current earnings and future cash flows. Prior studies find that losses tend to be transitory (Basu, 1997; Hayn, 1995). Therefore, we include LOSS to control for the transitory effect of operating loss on earnings persistence and the association between earnings and future cash flows. Dichev and Tang (2009) find that earnings volatility is related to earnings persistence. Therefore, we also control for earnings variability (STDROA). Finally, we control for SIZE and year fixed effects.

The standard errors from regression models using panel data may be correlated across firms and across time. Therefore, the t statistics are computed based on White (1980) heteroscedasticity-adjusted robust variance estimates that are adjusted for residual correlation arising from pooling cross-sectional observations across time. Furthermore, we follow Petersen (2009) and adjust for within-cluster correlation where the individual firm comprises the cluster (Rogers, 1993).

Forecast accuracy

For the analyst forecast accuracy test, we follow Duru and Reeb (2002) and Behn et al. (2008) to use the following model (firm and time subscripts are omitted).

\begin{matrix} ACCY = β_{0} + β_{1} AB_AUD + β_{2} INDSP_AT + β_{3} SIZE + β_{4} SURPRISE \\ + β_{5} LOSS + β_{6} ZMIJ + β_{7} HORIZON + β_{8} STDROE + β_{9} NANA \\ + β_{10} EL + Industry dummies + Year dummies + ε, \end{matrix}

where ACCY is the negative of the absolute value of forecast error scaled by stock price at time t− 1; INDSP_AT is industry specialist auditor measure, defined as the sum of the square root of the total assets of the clients of an auditor in a specific industry divided by total sum of the square root of the total assets of entire clients of the auditor; SIZE is the natural logarithm of market value of equity; SURPRISE is the absolute value of the difference between the current year’s EPS and last year’s EPS, deflated by the price at the beginning of the year; LOSS is 1 if the company had negative ROA, 0 otherwise; ZMIJ is Zmijewski’s financial distress score; HORIZON is the natural logarithm of the average of the number of calendar days between mean forecast announcement date and subsequent actual earnings announcement date; STDROE is the historical standard deviation of return on equity computed over the preceding 5 years; NANA is the natural logarithm of the number of analysts providing an annual earnings forecast; and EL is earnings per share.

We control for auditor industry specialization (INDSP_AT) because Behn et al. (2008) provide evidence that specialized auditors provide higher quality accounting information that can facilitate analyst forecast accuracy. We control for firm size (SIZE) because larger firms tend to have more information available that can facilitate analyst forecasts (Lang & Lundholm, 1996). We control for the absolute value of the earnings changes from that of last year (SURPRISE) because larger changes in earnings are found to be associated with lower forecast accuracy (Lang & Lundholm, 1996). Firms that report a loss and financially distressed firms are associated with less accurate forecasts (Hwang, Jan, & Basu, 1996). Therefore, we control for loss using an indicator variable for whether the firm reported a loss (LOSS) and Zmijewski’s (1984) financial distress score (ZMIJ). We control for forecast horizon (HORIZON) because previous research finds that longer forecast horizons are associated with lower analyst forecast accuracy (e.g., Brown, 2001; Brown, Richardson, & Schwager, 1987; O’Brien & Bhushan, 1990). We control for earnings volatility (STDROE) because long-term earnings volatility negatively affects the analyst forecast accuracy (Kross, Ro, & Schroeder, 1990). Prior studies document that forecast accuracy increases with analyst following (Lang & Lundholm, 1996; Lys & Soo, 1995). Therefore, we control for the number of analysts who issue a forecast during the period of 90 days prior to the earnings announcement (NANA). In addition to the changes of earnings, we control for earnings per share (EL) because Eames and Glover (2003) find that earnings level is related to forecast accuracy. Finally, we include indicator variables to control for industry and fiscal year because forecasting ability might differ across industries and time periods.

Forecast dispersion

For the dispersion test, we follow Behn et al. (2008) and use the following model (firm and time subscripts are omitted):

\begin{matrix} DISP = β_{0} + β_{1} AB_AUD + β_{2} INDSP_AT + β_{3} SIZE + β_{4} SURPRISE \\ + β_{5} LOSS + β_{6} ZMIJ + β_{7} HORIZON + β_{8} STDROE \\ + Industry dummies + Year dummies + ε, \end{matrix}

where DISP is the standard deviation of analysts’ earnings forecasts scaled by the stock price at the end of the prior fiscal year. The other variables are defined as in Model 3.

Data and Empirical Results

Data

The audit fees, audit opinion, internal control assessment, firm restatements, and audit firm identification are obtained from the Audit Analytics database. The sample covers the period 2000 through 2010 based on the Compustat fiscal year. We obtain the financial statement data from the Compustat XPressfeed (XPF) annual data file. After merging the data from Audit Analytics sample with Compustat, we retain firms that have Big N auditors and observations in Compustat with non-missing financial data. Big N is defined as the audit firms of Arthur Andersen, Deloitte & Touche, Ernst & Young, KPMG, or PricewaterhouseCoopers. Prior studies document that large (Big N) audit firms charge a fee premium (e.g., Choi, Kim, Liu, & Simunic, 2008; Craswell et al., 1995; DeAngelo, 1981b). Therefore, restricting the sample to Big N auditors avoids potential confounding effects and enables us to study observable auditor characteristics that vary within each client group. In addition, we exclude utilities (Standard Industrial Classification [SIC] codes 4900 to 4999) and financial services firms (SIC codes 6000-6999). The sample for our audit fee models consists of 28,837 firm-year observations and 5,115 unique firms for Fiscal Years 2000 through 2010 with the required auditor data from Audit Analytics and Compustat financial data for the audit fee model.

For our analyst forecast accuracy and dispersion tests, we obtain the 1-year-ahead annual earnings forecasts and the actual earnings per share from the I/B/E/S detailed historical file. The stock price and cumulative adjusting factors are obtained from the Center for Research in Security Prices (CRSP) daily file. For our earnings and cash flow persistence tests, we obtain the financial statement data from the Compustat XPF annual file. Our sample sizes differ for different tests due to various research designs and data requirements. To mitigate the influence of outliers, we winsorize all continuous dependent and independent variables at the top and bottom 1 percentile of their distributions. Table 1 presents the sample selection procedure and the number of observations used for each test.

Table 1.

Sample Selection Criteria.

Panel A: Sample Selection Criteria for the Audit Fee Model Estimation.
	Observations
Observations from Audit Analytics from 2000 through 2010	140,951
Less: Observations with missing or zero audit fees	10,975
Less: Firms located in other countries	15,421
Less: Firms not identified in Compustat database	58,064
Less: Financial institutions (SIC code: 6000-6999) and utilities companies (SIC code: 4900-4999)	15,312
Less: Observations with insufficient data for audit fee model	1,864
Less: Firms audited by non-Big 5 auditors	10,478
Number of observations for the audit fee model	28,837
Unique firms for the audit fee model	5,115

Panel B: Sample Selection Criteria for the Final Analyses.
	Earnings persistence	Forecast accuracy	Forecast dispersion
Initial sample with residual audit fee measure	28,837	28,837	28,837
Less: Observation with insufficient data for persistence model	5,514
Less: Firms not identified on I/B/E/S or without stock returns from CRSP		11,622	11,622
Less: Observation with insufficient data for accuracy model		4,294	4,294
Less: Observation with less than three analysts following			4,761
Number of observations in the final analysis	23,323	12,921	8,160
Unique firms	4,241	2,882	2,121

Note. SIC = Standard Industrial Classification; CRSP = Center for Research in Security Prices; I/B/E/S = Institutional Brokers Estimate Service.

Descriptive Statistics and Results of Estimating Audit Fee Model

Table 2, Panel A, provides descriptive statistics for the key variables used in the audit fee model. The sample size is 28,837 firm-year observations. The mean (median) audit fee in our sample is US$1,513,908 (US$670,000). With respect to firm characteristics, the untabulated mean of unlogged assets is about US$457 million, and about 36% of the observations report a loss.

Table 2.

Panel A: Descriptive Statistics for the Variables in the Audit Fee Model and Estimated Residual Audit Fees.
Variable	n	M	Median	SD	Q1	Q3
LNAUDFEE	28,837	13.419	13.415	1.265	12.461	14.270
AUDIT_FEES	28,837	1,531,908	670,000	2,669,410	257,968	1,575,000
AB_AUD	28,837	0	0.004	0.494	−0.316	0.319
LNTA	28,837	6.133	6.090	1.872	4.812	7.389
SQSEGS	28,837	1.349	1.000	0.458	1.000	1.732
INVREC	28,837	0.247	0.217	0.180	0.100	0.354
FOROPS	28,837	0.245	0.000	0.430	0.000	0.000
CR	28,837	2.937	2.077	2.795	1.381	3.344
BTM	28,837	0.640	0.458	0.718	0.258	0.766
LEV	28,837	0.196	0.158	0.193	0.008	0.322
SQEMPLS	28,837	2.080	1.282	2.295	0.602	2.595
M&A	28,837	0.209	0.000	0.407	0.000	0.000
DECEMBER	28,837	0.696	1.000	0.460	0.000	1.000
ROA	28,837	−0.049	0.029	0.276	−0.052	0.073
LOSS	28,837	0.359	0.000	0.480	0.000	1.000
GC	28,837	0.026	0.000	0.160	0.000	0.000
INITIAL	28,837	0.066	0.000	0.249	0.000	0.000
SHORT	28,837	0.146	0.000	0.353	0.000	0.000
MEDIUM	28,837	0.519	1.000	0.500	0.000	1.000
AGE_SEC	28,837	17.381	12.000	14.672	7.000	24.000
NEW_FIN	28,837	0.912	1.000	0.283	1.000	1.000
RISK_IND	28,837	0.281	0.000	0.449	0.000	1.000
IC_AUDIT	28,837	0.474	0.000	0.499	0.000	1.000
IC_ADV	28,837	0.033	0.000	0.179	0.000	0.000
RESTATE	28,837	0.127	0.000	0.334	0.000	0.000

Panel B: The Estimate of the Audit Fee Model by Year.
Parameter	2000		2001		2002		2003		2004		2005		2006		2007		2008		2009		2010
	Est.	p value	Est.	p value	Est.	p value	Est.	p value	Est.	p value	Est.	p value	Est.	p value	Est.	p value	Est.	p value	Est.	p value	Est.	p value
Intercept	9.476	<.0001	9.675	<.0001	9.530	<.0001	9.672	<.000	9.935	<.0001	10.246	<.0001	10.569	<.0001	10.673	<.0001	10.729	<.0001	10.786	<.0001	11.275	<.0001
LNTA	0.392	<.0001	0.383	<.0001	0.430	<.0001	0.438	<.000	0.462	<.0001	0.442	<.0001	0.423	<.0001	0.446	<.0001	0.442	<.0001	0.421	<.0001	0.414	<.0001
SQSEGS	0.122	<.0001	0.138	<.0001	0.159	<.0001	0.150	<.000	0.148	<.0001	0.167	<.0001	0.174	<.0001	0.093	.0002	0.120	<.0001	0.112	<.0001	0.142	<.0001
INVREC	0.442	<.0001	0.335	<.0001	0.365	<.0001	0.321	<.000	0.414	<.0001	0.470	<.0001	0.556	<.0001	0.601	<.0001	0.575	<.0001	0.665	<.0001	0.415	<.0001
FOROPS	0.275	<.0001	0.229	<.0001	0.233	<.0001	0.225	<.000	0.195	<.0001	0.222	<.0001	0.204	<.0001	0.172	<.0001	0.196	<.0001	0.199	<.0001	0.166	<.0001
CR	−0.031	<.0001	−0.037	<.0001	−0.041	<.0001	−0.041	<.000	−0.026	<.0001	−0.032	<.0001	−0.023	<.0001	−0.022	.0001	−0.020	.0006	−0.020	.0005	−0.022	.0005
BTM	0.013	.1388	0.015	.1836	0.009	.4906	−0.038	.1151	−0.099	.0069	−0.027	.4342	−0.005	.8968	−0.025	.4313	0.017	.1125	0.003	.8931	−0.043	.1810
LEV	0.065	.2818	0.041	.4452	0.035	.5477	−0.032	.6004	−0.193	.0056	−0.162	.0182	−0.141	.0469	−0.164	.0147	−0.213	.0010	−0.185	.0038	−0.180	.0090
SQEMPLS	0.088	<.0001	0.100	<.0001	0.086	<.0001	0.077	<.000	0.064	<.0001	0.059	<.0001	0.064	<.0001	0.054	<.0001	0.057	<.0001	0.072	<.0001	0.082	<.0001
M&A	0.043	.0744	0.036	.0950	0.063	.0146	0.068	.0078	0.014	.5976	0.040	.1226	0.034	.1833	0.011	.6729	0.079	.0028	0.046	.1078	0.068	.0103
DECEMBER	0.036	.1772	0.085	<.0001	0.128	<.0001	0.112	<.000	0.044	.1881	0.000	.9961	0.029	.2430	0.027	.2791	0.049	.0516	0.082	.0005	0.026	.3530
ROA	−0.336	<.0001	−0.204	<.0001	−0.307	<.0001	−0.285	<.000	−0.545	<.0001	−0.399	<.0001	−0.200	.0201	−0.386	<.0001	−0.245	<.0001	−0.359	<.0001	−0.257	.0059
LOSS	0.060	.0217	0.078	.0002	0.108	<.0001	0.165	<.000	0.106	.0012	0.093	.0059	0.140	<.0001	0.083	.0131	0.064	.0315	0.071	.0142	0.104	.0014
GC	0.147	.0044	0.130	.0025	0.098	.0698	0.182	.0053	0.214	.0093	0.092	.2777	0.120	.2153	−0.084	.3490	−0.068	.3386	−0.039	.6469	0.030	.7703
INITIAL	−0.001	.9766	−0.189	<.0001	−0.156	<.0001	−0.116 0	.024	−0.098	.0774	−0.026	.6194	−0.042	.4490	−0.021	.7283	−0.097	.1146	−0.160	.0085	−0.257	.0004
SHORT	−0.044	.3284	−0.026	.4722	0.092	.0195	−0.034	.2892	−0.003	.9234	0.138	.0027	0.168	.0002	0.108	.0130	0.017	.7049	0.011	.8070	0.015	.7439
MEDIUM	−0.097	.0037	−0.066	.0150	−0.005	.8802	−0.009	.7745	−0.032	.3173	−0.030	.2989	−0.033	.2546	−0.013	.6478	0.018	.4767	0.035	.1524	0.032	.2115
AGE_SEC	0.027	.1587	−0.024	.1399	−0.014	.4141	−0.001	.9733	−0.049	.0079	−0.058	.0015	−0.066	.0002	−0.040	.0276	−0.023	.1733	−0.001	.9576	−0.007	.6961
NEW_FIN	0.016	.6624	−0.036	.1950	−0.056	.0704	0.026	.4431	0.168	.0001	0.146	.0017	0.056	.2556	0.167	.0004	0.050	.1702	0.073	.0247	0.033	.3790
RISK_IND	0.054	.0767	0.057	.0300	0.088	.0031	0.092	.0030	0.066	.0518	0.090	.0079	0.107	.0022	0.048	.1602	0.072	.0381	0.055	.0985	0.010	.7612
IC_AUDIT	0.000		0.000		0.000		0.000		0.416	<.0001	0.328	<.0001	0.314	<.0001	0.025	.6023	−0.075	.3508	−0.116	.1080	−0.404	<.0001
IC_ADV	0.000		0.000		0.000		0.000		0.352	<.0001	0.423	<.0001	0.374	<.0001	0.391	<.0001	0.454	<.0001	0.388	<.0001	0.317	<.0001
RESTATE	0.000		0.277	<.0001	0.112	<.0001	−0.004	.8896	0.096	.0007	0.092	.0020	0.074	.0173	0.042	.1864	−0.002	.9587	−0.051	.1670	−0.086	.5633
Industry effects	Yes		Yes		Yes		Yes		Yes		Yes		Yes		Yes		Yes		Yes		Yes
Adjusted R²	.746		.786		.764		.766		.767		.764		.755		.76		.77		.784		.803
n	2,744		3,295		3,240		3,148		2,916		2,641		2,455		2,295		2,163		2,135		1,805

Note. The variables are defined in the appendix. The audit fees model is estimated by year with industry-dummy variables.

We estimate Model 1 by year with 72 industry-dummy variables. Table 2, Panel B, presents the annual regression results of the audit fee model in Equation 1. The adjusted R² values of annual estimations range from .746 to .803. The coefficients on the determinants of audit fees generally have the predicted signs and are consistent with the results reported in prior studies.

The Association Between Residual Audit Fees and Earnings Persistence

Table 3, Panel A, provides descriptive statistics for the key variables used in earnings prediction tests. There are 23,323 firm-year observations representing 4,241 firms in this sample. Our sample firms report mean (median) earnings (EARN) of −0.016 (0.035). The mean (median) residual audit fee (AB_AUD) for the sample firms is −0.004 (0.002). Panel B provides descriptive statistics for the key variables for analyst forecast accuracy tests. By construction, the mean and median ACCY values are negative with the values of −0.008 and −0.002, respectively. There are 12,921 firm-year observations representing 2,882 firms in this sample. The mean (median) residual audit fee (AB_AUD) for this sample of firms is −0.014 (0.001). Panel C provides descriptive statistics for the key variables for analyst forecast dispersion tests. For the dispersion test, we require the firm to be followed by at least three analysts. Thus, the sample size is smaller than for the accuracy tests, and the final sample for this analysis is 8,160 observations from 2,121 unique firms. The mean and median values of DISP are 0.004 and 0.001, respectively, while the mean (median) residual audit fee (AB_AUD) for the sample firms is −0.013 (0.003).

Table 3.

Descriptive Statistics.

Panel A: Earnings Persistence Sample.
Variable	n	M	Median	SD	Q1	Q3
EARN_LEAD	23,323	−0.016	0.035	0.186	−0.032	0.078
EARN	23,323	−0.015	0.035	0.187	−0.032	0.079
SIZE	23,323	6.205	6.223	1.993	4.886	7.484
STDROA	23,323	0.109	0.051	0.178	0.024	0.116
LOSS	23,323	0.324	0.000	0.468	0.000	1.000
AB_AUD	23,323	−0.004	0.002	0.466	−0.311	0.308

Panel B: Analyst Forecast Accuracy Sample.
Variable	n	M	Median	SD	Q1	Q3
ACCY	12,921	−0.008	−0.002	0.022	−0.005	−0.001
INDSP_AT	12,921	0.103	0.097	0.061	0.052	0.139
SIZE	12,921	7.215	7.098	1.646	6.088	8.221
SURPRISE	12,921	0.046	0.015	0.095	0.007	0.041
LOSS	12,921	0.192	0.000	0.394	0.000	0.000
ZMIJ	12,921	−3.133	−3.318	1.627	−4.195	−2.423
HORIZON	12,921	47.484	46.000	20.434	32.286	62.000
STDROE	12,921	0.313	0.092	0.791	0.043	0.222
NANA	12,921	5.484	4.000	5.385	2.000	7.000
EL	12,921	1.198	1.020	1.977	0.220	1.960
AB_AUD	12,921	−0.014	0.001	0.465	−0.320	0.301

Panel C: Analyst Forecast Dispersion Sample.
Variable	n	M	Median	SD	Q1	Q3
DISP	8,160	0.004	0.001	0.010	0.001	0.004
INDSP_AT	8,160	0.104	0.098	0.061	0.052	0.140
SIZE	8,160	7.700	7.545	1.596	6.567	8.714
SURPRISE	8,160	0.037	0.014	0.065	0.006	0.037
LOSS	8,160	0.154	0.000	0.361	0.000	0.000
ZMIJ	8,160	−3.172	−3.318	1.424	−4.181	−2.451
HORIZON	8,160	46.509	44.667	17.971	33.333	58.500
STDROE	8,160	0.285	0.088	0.728	0.042	0.201
AB_AUD	8,160	−0.013	0.003	0.463	−0.318	0.305

Note. All variables are defined in the appendix.

Table 4, Panel A, presents Pearson’s correlation coefficients among the variables used in our earnings persistence test. The correlation coefficients in bold are significant at the .01 level or better, whereas correlation coefficients in italics are significant at the .01 to .10 level. Panel B shows the Pearson correlation matrix for the variables used in the analyst forecast accuracy regression analysis. Consistent with prior literature, firm size, number of analyst following, and earnings level are positively associated with accuracy, whereas the rest of the control variables correlate negatively with accuracy. Our variable of interest, residual audit fee (AB_AUD), also has a negative univariate correlation with accuracy. Table 4, Panel C, shows the Pearson correlation matrix for the variables used in the analyst forecast dispersion regression analysis. Variables that are positively related to DISP are associated with higher forecast dispersion. Residual audit fee (AB_AUD) is positively correlated with forecast dispersion. These univariate results suggest that on average, residual audit fees are associated with less accurate and more dispersed analyst forecasts.¹⁰

Table 4.

Pearson’s Correlations.

Panel A: Earnings Persistence Sample.
	EARN_LEAD	EARN	SIZE	STDROA	LOSS
EARN	.731
SIZE	.336	.374
STDROA	−.382	−.451	−.219
LOSS	−.527	−.691	−.378	.344
AB_AUD	−.017	−.008	.012	.039	.001

Panel B: Analyst Forecast Accuracy Sample.
	ACCY	DISP	INDSP	SIZE	SURPRISE	LOSS	ZMIJ	HORIZON	STDROE	NANA	EL
DISP	−.655
INDSP_AT	−.009	.018
SIZE	.313	−.256	.100
SURPRISE	−.610	.526	.038	−.358
LOSS	−.372	.338	.121	−.387	.436
ZMIJ	−.262	.260	.063	−.158	.315	.360
HORIZON	−.041	−.016	.030	−.085	.028	.066	.018
STDROE	−.167	.193	.050	−.154	.234	.242	.317	.016
NANA	.115	−.052	.056	.492	−.089	−.147	−.056	−.068	−.059
EL	.295	−.248	−.097	.470	−.332	−.541	−.246	−.076	−.181	.222
AB_AUD	−.036	.033	.001	.038	.026	−.043	.001	.013	.049	−.023	.046

Panel C: Analyst Forecast Dispersion Sample.
	ACCY	DISP	INDSP	SIZE	SURPRISE	LOSS	ZMIJ	HORIZON	STDROE
DISP	−.368
INDSP_AT	.008	.017
SIZE	.123	−.259	.160
SURPRISE	−.340	.529	.022	−.320
LOSS	−.161	.356	.104	−.340	.435
ZMIJ	−.144	.251	.084	−.142	.283	.352
HORIZON	.001	−.024	.050	−.058	−.008	.060	−.007
STDROE	−.114	.179	.062	−.140	.204	.238	.281	.008
AB_AUD	−.004	.024	.012	.046	.018	−.048	−.002	.019	.032

Note. Correlation coefficients in bold are significant at the .01 level or better. Correlation coefficients in italics are significant at the .01 to .10 level. All variables are defined in the appendix.

Multivariate Analysis

Table 5 reports the estimation results of Model 2 that examines the impact of residual audit fees on earnings persistence. Here, our variable of interest is the interaction term between earnings and decile ranked residual audit fees (R_AB_AUD×EARN). The coefficient on EARN is positive and significant, suggesting that earnings are persistent. The coefficient on R_AB_AUD is negative and significant. The coefficient on the interaction term loads negatively and significantly (−.0060; t = −1.90), which suggests that residual audit fees are associated with less persistent earnings. The explanatory power as measured by the adjusted R² is .556 for the earnings persistence model. This result is consistent with that of Hribar et al. (2014) and suggests that residual audit fee is indicative of poor accounting quality which diminishes investors’ ability to predict future earnings.

Table 5.

The Association Between Residual Audit Fees and the Ability of Current Earnings to Predict Future Earnings.

\begin{matrix} EARN_LEAD = β_{0} + β_{1} EARN + β_{2} LOSS + β_{3} LOSS \times EARN + β_{4} STDROA \\ + β_{5} STDROA \times EARN + β_{6} SIZE + β_{7} SIZE \times EARN \\ + β_{8} R_AB_AUD + β_{9} R_AB_AUD \times EARN + Year dummies + ε . \end{matrix} (2)

Parameter	EARN_LEAD
	Estimate	t statistics
Intercept	−0.0639	−14.18***
EARN	0.6757	13.53***
LOSS	−0.0029	−0.96
LOSS×EARN	0.0365	1.16
STDROA	−0.0983	−8.72***
STDROA×EARN	−0.1712	−5.23***
SIZE	0.0056	12.30***
SIZE×EARN	0.0082	1.42
R_AB_AUD	−0.0006	−2.55**
R_AB_AUD×EARN	−0.0060	−1.90*
Year dummies	Yes
Adjusted R²	.556
N	23,323

Note. All variables are defined in the appendix. R_AB_AUD is the decile ranked measure of AB_AUD by year. All the t statistics are based on White’s (1980) heteroscedasticity-corrected standard errors and clustering procedure by each firm.

*, **, and *** indicate significance at the 10%, 5%, and 1% levels in a two-tailed test, respectively.

We next examine the relation between residual audit fees and properties of analyst forecasts. Table 6 reports the multivariate regression results examining the association between residual audit fees and analyst forecast accuracy after controlling for other determinants of forecast accuracy. The coefficient on AB_AUD is −0.0013 and is statistically significant at the .01 level (t = −3.72). To gauge the economic significance of this result, we multiply −0.0013 by the standard deviation of AB_AUD in the sample, which is 0.465, and divide it by the mean of ACCY, which is −0.008. The result of 0.0756 suggests that when residual audit fees increase by 1 standard deviation, the analyst forecast accuracy reduces by 7.56%, which is economically significant using the standard 5% rule of thumb for materiality. This result is also consistent with our previous findings and suggests that residual audit fee indicates lower accounting quality. This lower quality impairs the ability of analysts to accurately predict future earnings, which has negative consequences on capital market participants.

Table 6.

The Association Between Residual Audit Fees and Analysts’ Forecast Accuracy.

\begin{matrix} ACCY = β_{0} + β_{1} AB_AUD + β_{2} INDSP_AT + β_{3} SIZE + β_{4} SURPRISE + β_{5} LOSS \\ + β_{6} ZMIJ + β_{7} HORIZON + β_{8} STDROE + β_{9} NANA + β_{10} EL \\ + Industry dummies + Year dummies + ε . \end{matrix} (3)

Parameter	Dependent variable = ACCY
	Estimate	t statistics
Intercept	−0.0131	−3.40***
AB_AUD	−0.0013	−3.72***
INDSP_AT	0.0053	1.23
SIZE	0.0008	5.14***
SURPRISE	−0.1191	−19.49***
LOSS	−0.0055	−7.60***
ZMIJ	−0.0006	−2.54**
HORIZON	−0.0009	−2.82***
STDROE	0.0002	0.59
NANA	0.0001	3.55***
EL	0.0006	2.85***
Industry dummies	Yes
Year dummies	Yes
Adjusted R²	.416
n	12,921

Note. All variables are defined in the appendix. All the t statistics are based on White’s (1980) heteroscedasticity-corrected standard errors and clustering procedure by each firm.

*, **, and *** indicate significance at the 10%, 5%, and 1% levels in a two-tailed test, respectively.

We also estimate the economic magnitude of the lower forecast accuracy using only the top quartile of residual fees. As expected, the coefficient is much larger with the value of −0.0038 and statistically significant at 5% level. For observations in the top quartile of residual fees, when residual audit fee increases by 1 standard deviation, the analyst forecast accuracy reduces by 9.63%.¹¹

Overall, the results reported in Tables 5 and 6 suggest that residual audit fees are associated with lower financial reporting quality which manifests in lower quality information environment.

Table 7 presents the multivariate regression results examining the association between residual audit fees and dispersion in analysts’ forecasts after controlling for other determinants of analyst forecast dispersion. The coefficient on AB_AUD is 0.0005 and is statistically significant at the .05 level with a t statistic of 2.47. To gauge its economic significance, we multiply 0.0005 by the standard deviation of AB_AUD which is 0.463, and divide by the mean of DISP, which is 0.004. This result (0.0579) suggests that when residual audit fees increase by 1 standard deviation, the analyst forecast dispersion increases by 5.79%, which is also economically significant.

Table 7.

The Association Between Residual Audit Fees and Analysts’ Forecast Dispersion.

\begin{matrix} DISP = β_{0} + β_{1} AB_AUD + β_{2} INDSP_AT + β_{3} SIZE + β_{4} SURPRISE \\ + β_{5} LOSS + β_{6} ZMIJ + β_{7} HORIZON + β_{8} STDROE \\ + Industry dummies + Year dummies + ε . \end{matrix} (4)

	Dependent variable = DISP
Parameter	Estimate	t statistics
Intercept	0.0063	6.40***
AB_AUD	0.0005	2.47**
INDSP_AT	−0.0007	−0.26
SIZE	−0.0006	−8.69***
SURPRISE	0.0591	15.34***
LOSS	0.0036	8.06***
ZMIJ	0.0004	3.68***
HORIZON	−0.0004	−1.95*
STDROE	0.0005	2.11**
Industry dummies	Yes
Year dummies	Yes
Adjusted R²	.353
n	8,160

Note. All variables are defined in the appendix. All the t statistics are based on White’s (1980) heteroscedasticity-corrected standard errors and clustering procedure by each firm.

*, **, and *** indicate significance at the 10%, 5%, and 1% levels in a two-tailed test, respectively.

Additional Tests

Does the analyst’s experience mitigate the impact of residual audit fees on forecast accuracy?

Prior research finds that analysts’ performance improves with experience (Markov & Tamayo, 2006; Mikhail, Walther, & Willis, 1997). We examine whether an analyst’s experience in following a firm mitigates the negative relation between residual audit fees and forecast accuracy. We retain the most recent 1-year-ahead earnings forecast for each analyst during the period (−90, −3) relative to earnings announcement date.¹² Then, we measure the forecast accuracy for each analyst. We measure the analyst experience (EXP_ijt) as the number of years that the analyst j has provided annual earnings forecast up to year t for firm i. We construct a decile ranked variable, R_EXP_ijt, which takes the value of 1 to 10.¹³ Untabulated results indicate that accuracy and analyst experience is positively correlated.

To examine the impact of analyst experience with the firm on the association between residual audit fees and analyst forecast accuracy, we further control for some firm characteristics such as size, volatility, performance, and financial distress; firm information environment such as the number of analyst following; and analyst characteristics such as the number of firms for which the analyst provides annual forecasts (COVER). The results are presented in Table 8. The number of observations available is much higher than the sample in Table 6 because our test is an analyst-level test. Model 1 is similar to Table 6, that is, AB_AUD is not interacted with R_EXP. As before, AB_AUD is negative (−0.0013) and statistically significant at the .01 level, which is consistent with our primary results. In Model 2, when AB_AUD is interacted with R_EXP, the coefficient on AB_AUD continues to be negative (−0.0018) and statistically significant at the .01 level. However, the coefficient on AB_AUD×R_EXP is positive (0.0001) and statistically significant at the .05 level, indicating that analyst experience mitigates the negative effect of residual audit fees on forecast accuracy.

Table 8.

The Impact of Analyst Experience on the Relation Between Residual Audit Fees and Analyst Forecast Accuracy.

Parameter	Dependent variable = ACCY
	(1)		(2)
	Estimate	t statistics	Estimate	t statistics
Intercept	−0.0195	−3.23***	−0.0195	−3.23***
AB_AUD	−0.0013	−5.11***	−0.0018	−5.08***
R_EXP			0.0000	0.00
AB_AUD×R_EXP			0.0001	2.07**
INDSP_AT	0.0072	2.67**	0.0072	2.66***
SIZE	0.0003	2.39**	0.0003	2.37**
NANA	0.0001	4.31***	0.0001	4.32***
LOSS	−0.0111	−18.30***	−0.0111	−18.29***
HORIZON	−0.0005	−5.42***	−0.0005	−5.43***
ZMIJ	−0.0015	−9.83***	−0.0015	−9.83***
STDROE	−0.0007	−3.10***	−0.0007	−3.08***
EL	0.0007	5.03***	0.0007	5.03***
COVER	0.000	0.22	0.0000	0.21
Year dummies	Yes		Yes
Industry dummies	Yes		Yes
Adjusted R²	.211		.211
n	81,353		81,353

Note. We measure the analyst experience (EXP_ijt) as the number of years that the analyst j has provided annual earnings forecast up to year t for firm i. Then, we construct a decile ranked variable of R_EXP_ijt from 1 to 10. COVER is the number of firms that analyst j provides annual earnings forecast in year t. All the other variables are defined in the appendix.

*, **, and *** indicate significance at the 10%, 5%, and 1% levels in a two-tailed test, respectively.

Control for stock return volatility

To mitigate the possibility that our residual audit fees capture unexplained risk or complexity, we also control for firms’ stock return volatility of 12 months preceding to firms’ fiscal year-end. In untabulated results, both forecast accuracy and dispersion results remain robust after controlling for stock return volatility.

The association between earnings persistence and analyst following and forecast properties

The same client characteristics that are associated with higher residual audit fees could be associated with analyst coverage and/or analyst forecast properties such as forecast accuracy and dispersion. To address this endogeneity concern, we attempt to triangulate the results for analyst forecast accuracy and dispersion with those results for earnings persistence, a non-analyst related measure. Specifically, we examine whether there is an association between earnings persistence and analyst coverage, forecast accuracy, and forecast dispersion. Table 9 reports the results. We find that in a pooled model, that is, with all analyst characteristics in the model, there is no significant relation between analyst coverage or forecast accuracy and earnings persistence; however, forecast dispersion is negatively associated with earnings persistence. These results offer some comfort but do not completely eliminate the concern of endogeneity.

Table 9.

Additional Test: The Relation Between Earnings Persistence and Forecast Characteristics.

Parameter	EARN_LEAD		EARN_LEAD		EARN_LEAD		EARN_LEAD
	Estimate	t statistics	Estimate	t statistics	Estimate	t statistics	Estimate	t statistics
Intercept	−0.1116	−13.31***	−0.1070	−13.21***	−0.1303	−13.73***	−0.1324	−13.18***
EARN	0.7387	7.86***	0.8625	9.49***	1.0587	10.11***	1.0130	8.89***
LOSS	−0.0003	−0.08	0.0017	0.40	0.0024	0.51	0.0032	0.69
EARN×LOSS	−0.0592	−1.32	−0.0602	−1.35	−0.1480	−2.92***	−0.1409	−2.76***
STDROA	−0.0972	−6.72***	−0.0977	−6.69***	−0.1100	−6.63***	−0.1079	−6.50***
EARN×STDROA	−0.1774	−3.86***	−0.1699	−3.68***	−0.1377	−2.40**	−0.1317	−2.25**
SIZE	0.0099	9.91***	0.0087	9.81***	0.0104	10.03***	0.0113	9.32***
EARN×SIZE	0.0103	0.84	−0.0100	−0.91	−0.0277	−2.22**	−0.0187	−1.25
R_AB_AUD	0.0001	0.20	0.0002	0.50	0.0009	2.09**	0.0008	1.92*
EARN×R_AB_AUD	−0.0108	−2.27**	−0.0095	−1.98**	−0.0152	−2.82***	−0.0159	−2.95***
NANA	−0.0007	−2.59**					−0.0007	−2.20**
EARN_NANA	−0.0051	−1.66*					−0.0032	−0.95
ACCY			0.2537	2.45**			0.1252	0.70
EARN_ACCY			1.7731	3.32***			−0.3193	−0.41
DISP					−0.4164	−2.95***	−0.2899	−1.38
EARN_DISP					−4.7090	−4.53***	−5.0729	−3.90***
Year dummies	Yes		Yes		Yes		Yes
Adjusted R²	.530		.531		.528		.529
n	11,862		11,862		9,449		9,449

*, **, and *** indicate significance at the 10%, 5%, and 1% levels in a two-tailed test, respectively.

Conclusion

The purpose of this study was to provide direct evidence on the association between residual audit fees and investors’ ability to anticipate future earnings. The issue of whether audit fees, particularly residual audit fees, signal improved usefulness through increased audit effort or lower usefulness through lower financial reporting quality is a question of fundamental interest to various participants in the capital markets (Doogar et al., 2015). Recent research by Hribar et al. (2014) suggests that residual audit fees represent a broad indicator of audit quality because unexplained fees capture the auditor’s unobservable private information about the audit client’s accounting quality. To address this issue, we first examine the relation between residual audit fees and earnings persistence. We also examine how residual audit fees are associated with the forecasts of financial analysts, who are often regarded as sophisticated and primary users of financial statements (e.g., Schipper, 1991). Our research design controls for potential cross-sectional variation in analyst forecast properties associated with not only various firms but also analyst characteristics such as firm size, profitability, financial distress, forecast horizon, and analyst experience (Markov & Tamayo, 2006; Mikhail et al., 1997).

Our evidence shows that the residual or unexplained audit fee is negatively associated with the ability of current earnings to predict future earnings. Furthermore, the residual audit fee is negatively associated with analyst forecast accuracy and positively associated with the dispersion in analyst forecasts. Our results are consistent with the notion that residual audit fees proxy for poor financial reporting quality which has an adverse effect on a firm’s information environment.

Our findings have important implications for analysts, audit committees, investors, and others. For example, earnings of client firms with high levels of unexplained audit fees paid to their auditors might be more difficult to forecast than earnings of other firms. Thus, analysts need to devote more effort and attention in forecasting earnings of firms that pay higher residual audit fees. The finding has implications for boards of directors and the audit committees because residual audit fees might signal greater future variability in stock prices if analysts’ forecasts are less accurate and more dispersed.

Footnotes

Appendix

Variable Definitions.

Panel A: Variables for the Audit Fee Model.
LNAUDFEE	The natural logarithm of audit fees paid to the external auditor
LNTA	Natural logarithm of total assets
SQSEGS	The square root of the number of business segments
INVREC	Sum of accounts receivable and inventory divided by total assets
FOROPS	1 if the firm has foreign operations, 0 otherwise
CR	The ratio of current assets to current liabilities
BTM	Book value of equity divided by market value of equity
LEV	The sum of short-term debt and long-term debt scaled by total assets
SQEMPLS	The square root of the number of employees
M&A	1 for mergers and acquisitions (Sale_fn = “AA,”“AB,”“AR,”“AS,”“FA,”“FB,”“FC,”“FD,”“FE,”“FF,”“FW”) and 0 otherwise
DECEMBER	1 if the firm’s fiscal year-end month is December, 0 otherwise
ROA	Return on assets
LOSS	1 if the company had negative ROA, 0 otherwise
GC	1 if firm receives a going concern opinion, 0 otherwise
INITIAL	1 if the firm’s auditor has been with the client for less than 1 year, 0 otherwise
SHORT	1 if the firm’s auditor has been with the client for 2 or 3 years, 0 otherwise
MEDIUM	1 if the firm’s auditor has been with the client for 4 to 14 years, 0 otherwise
AGE_SEC	Number of fiscal years since the company’s initial appearance in Compustat
NEW_FIN	1 if the firm was involved in any financing, 0 otherwise
RISK_IND	1 if the firm is in the high technology industry
IC_AUDIT	1 if the firm is required to assess the internal control effectiveness, 0 otherwise
IC_ADV	1 if the firm has internal control problem disclosure in that year, 0 otherwise
RESTATE	1 if the firm restated its financial statements, 0 otherwise
Industry effects	1 if the firm falls in that industry (2-digit SIC), 0 otherwise

Panel B: Variables for the Earnings Persistence Model.
EARN_LEAD	The income available to common shareholders before extraordinary items in year t+ 1 scaled by market value of equity at the beginning of year t+ 1
EARN	The income available to common shareholders before extraordinary items in year t scaled by market value of equity at the beginning of year t
ACCR	Accrual component of earnings at year t
CFO	Cash flow component of earnings at year t
SIZE	The natural log of market value of equity
STDROA	Standard deviation of ROA (return on asset) for the period of 5 years (year t− 4 to year t)
LOSS	1 if the company had negative ROA, 0 otherwise

Panel C: Variables for Analyst Forecast Accuracy and Dispersion Model.
ACCY	The negative of the absolute value of forecast error scaled by stock price at time t− 1
INDSP_AT	Industry specialist auditor measure, defined as the sum of the square root of the total assets of the clients of an auditor in a specific industry divided by total sum of the square root of the total assets of entire clients of the auditor
SIZE	The natural logarithm of market value of equity
SURPRISE	The absolute value of the difference between the current year’s EPS and last year’s EPS, deflated by the price at the beginning of the year
LOSS	1 if the company had negative ROA, 0 otherwise
ZMIJ	Zmijewski’s financial distress score
HORIZON	The natural logarithm of the average of the number of calendar days between mean forecast announcement date and subsequent actual earnings announcement date
STDROE	The historical standard deviation of return on equity computed over the preceding 5 years
NANA	The natural logarithm of number of analysts providing an annual earnings forecast
EL	Earnings per share
FORECAST	The mean I/B/E/S consensus forecast of period made during the period starting 90 days before the corresponding actual earnings announcement and ending 3 days before the announcement

Panel D: Test Variable.
AB_AUD	The residual audit fees obtained from the audit fee model

Note. SIC = Standard Industrial Classification; I/B/E/S = Institutional Brokers Estimate Service.

Acknowledgements

The authors thank Bharat Sarath (editor), an anonymous reviewer, and seminar participants at London Business School and 34th Annual Congress of the European Accounting Association for their helpful comments and suggestions.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Notes

References

Abarbanell

J. S.

Bushee

B. J.

(1997). Fundamental analysis, future earnings, and stock prices. Journal of Accounting Research, 35, 1-24.

Barth

Hutton

(2004). Analyst earnings forecast revisions and the pricing of accruals. Review of Accounting Studies, 9, 59-96.

Basu

(1997). The conservatism principle and the asymmetric timeliness of earnings. Journal of Accounting & Economics, 24, 3-37.

Behn

B. K.

Choi

J. H.

Kang

(2008). Audit quality and properties of analysts’ earnings forecasts. The Accounting Review, 83, 327-349.

Brown

L. D.

(2001). A temporal analysis of earnings surprises: Profits and losses. Journal of Accounting Research, 39, 221-241.

Brown

L. D.

Richardson

Schwager

(1987). An information interpretation of financial analyst superiority in forecasting earnings. Journal of Accounting Research, 25, 49-67.

Caramanis

Lennox

(2008). Audit effort and earnings management. Journal of Accounting & Economics, 45, 116-138.

Chaney

Jeter

D. C.

Shivakumar

(2004). Self-selection of auditors and audit pricing in private firms. The Accounting Review, 79, 51-72.

Choi

Kim

Liu

Simunic

(2008). Audit pricing, legal liability regimes, and Big 4 premiums: Theory and cross-country evidence. Contemporary Accounting Research, 25, 55-99.

10.

Craswell

Francis

(1999). Pricing of initial audit engagements. The Accounting Review, 74, 201-216.

11.

Craswell

Francis

Taylor

(1995). Auditor brand name reputations and industry specializations. Journal of Accounting & Economics, 20, 297-322.

12.

Davis

Soo

Trompeter

(2009). Auditor tenure and the ability to meet or beat earnings forecasts. Contemporary Accounting Research, 26, 517-548.

13.

DeAngelo

L. E.

(1981a). Auditor independence, “low balling,” and disclosure regulation. Journal of Accounting & Economics, 3, 113-127.

14.

DeAngelo

L. E.

(1981b). Auditor size and audit quality. Journal of Accounting & Economics, 13, 183-199.

15.

Dechow

P. M.

Schrand

C. M.

(2004). Earnings quality. Charlottesville, VA: Research Foundation of CFA Institute.

16.

DeFond

M. L.

Raghunandan

L. K.

Subramanyam

K. R.

(2002). Do non-audit service fees impair auditor independence? Evidence from going concern audit opinions. Journal of Accounting Research, 40, 1247-1274.

17.

DeFond

M. L.

Zhang

(2014). A review of archival auditing research. Journal of Accounting & Economics, 58, 275-326.

18.

Dichev

Graham

Harvey

Rajgopal

(2013). Earnings quality: Evidence from the field. Journal of Accounting & Economics, 56, 1-33.

19.

Dichev

Tang

(2009). Earnings volatility and earnings predictability. Journal of Accounting & Economics, 47, 160-181.

20.

Diether

Malloy

Scherbina

(2002). Differences of opinion and the cross section of stock returns. Journal of Finance, 57, 2113-2141.

21.

Dontoh

Ronen

Sarath

(2013). Financial statements insurance. Abacus, 49, 269-307.

22.

Doogar

Sivadasan

Solomon

(2015). Audit fee residuals: Costs or rents?Review of Accounting Studies, 20, 1247-1286.

23.

Duru

Reeb

(2002). International diversification and analysts’ forecast accuracy and bias. The Accounting Review, 77, 415-433.

24.

Eames

Glover

(2003). Earnings predictability and the directing of analysts’ earnings forecast errors. The Accounting Review, 78, 707-724.

25.

Ettredge

Greenberg

(1990). Determinants of fee cutting on initial audit engagements. Journal of Accounting Research, 28, 198-210.

26.

Francis

(1984). The effect of audit firm size on audit prices: A study of the Australian market. Journal of Accounting & Economics, 6, 133-151.

27.

Francis

Reichelt

Wang

(2005). The pricing of national and city-specific reputations for industry expertise in the U.S. audit market. The Accounting Review, 80, 113-136.

28.

Francis

Stokes

(1986). Audit prices, product differentiation, and scale economies: Further evidence. Journal of Accounting Research, 24, 383-393.

29.

Ghosh

Pawlewicz

(2009). The impact of regulation on auditor fees: Evidence from the Sarbanes-Oxley Act. Auditing: A Journal of Practice and Theory, 28, 171-197.

30.

Hayn

(1995). The information content of losses. Journal of Accounting & Economics, 20, 123-153.

31.

Healy

Hutton

Palepu

(1999). Stock performance and intermediation changes surrounding sustained increases in disclosure. Contemporary Accounting Research, 16, 485-520.

32.

Hogan

C. E.

Wilkins

M. S.

(2008). Evidence on the audit risk model: Do auditors increase audit fees in the presence of internal control deficiencies?Contemporary Accounting Research, 25, 219-242.

33.

Hribar

Kravet

Wilson

(2014). A new measure of accounting quality. Review of Accounting Studies, 19, 506-538.

34.

Hwang

Jan

Basu

(1996). Loss firms and analysts’ earnings forecast errors. Journal of Financial Statement Analysis, 1, 18-31.

35.

Imhoff

E. A.

Lobo

G. J.

(1992). The effect of ex ante earnings uncertainty on earnings response coefficients. The Accounting Review, 67, 427-439.

36.

Knechel

W. R.

Willekens

(2006). The role of risk management and governance in determining audit demand. Journal of Business Finance & Accounting, 33, 1344-1367.

37.

Kross

B. T.

Schroeder

(1990). Earnings expectations: The analysts’ information advantage. The Accounting Review, 65, 461-476.

38.

Lang

Lundholm

(1996). Corporate disclosure policy and analyst behavior. The Accounting Review, 71, 467-492.

39.

Lipe

(1990). The relation between stock returns and accounting earnings given alternative information. The Accounting Review, 65, 49-71.

40.

Lyon

J. D.

Maher

M. W.

(2005). The importance of business risk in setting audit fees: Evidence from cases of client misconduct. Journal of Accounting Research, 43, 133-151.

41.

Lys

Soo

(1995). Analysts’ forecast precision as a response to competition. Journal of Accounting, Auditing and Finance, 10, 751-765.

42.

Markov

Tamayo

(2006). Predictability in financial analyst forecast errors: Learning or irrationality?Journal of Accounting Research, 44, 725-761.

43.

Mikhail

Walther

Willis

(1997). Do security analysts improve their performance with experience?. Journal of Accounting Research, 35, 131-157.

44.

O’Brien

Bhushan

(1990). Analyst following and institutional ownership. Journal of Accounting Research, 28, 55-76.

45.

Ohlson

J. A.

(2009). Accounting data and value: The basic results. Contemporary Accounting Research, 26, 231-259.

46.

Payne

Robb

(2000). Earnings management: The effect of ex ante earnings uncertainty. Journal of Accounting, Auditing & Finance, 15, 371-392.

47.

Payne

Thomas

(2003). The implications of using stock-split adjusted I/B/E/S data in empirical research. The Accounting Review, 78, 1049-1067.

48.

Petersen

(2009). Estimating standard errors in finance panel data sets: Comparing approaches. Review of Financial Studies, 22, 435-480.

49.

Raghunandan

Rama

(2006). SOX section 404 material weakness disclosures and audit fees. Auditing: A Journal of Practice & Theory, 25, 99-115.

50.

Rogers

(1993). Regression standard errors in clustered samples (Stata Technical Bulletin Reprints No. 13). College Station, TX: Stata Press.

51.

Sankaraguruswamy

Whisenant

(2009, November). Pricing initial audit engagements: Empirical evidence following public disclosure of audit fees (Working paper). Houston, TX: University of Houston.

52.

Schipper

(1991). Commentary on analysts’ forecasts. Accounting Horizons, 5, 105-121.

53.

Schipper

Vincent

(2003). Earnings quality. Accounting Horizons, 17, 97-110.

54.

Simunic

D. A.

(1980). The pricing of audit services: Theory and Evidence. Journal of Accounting Research, 18, 161-190.

55.

Wallace

W. A.

(1987). The economic role of the audit in free and regulated markets. Research in Accounting Regulation1, 7-34.

56.

Whisenant

Sankaraguruswamy

Raghunandan

(2003). Evidence on the joint determination of audit and non-audit fees. Journal of Accounting Research, 41, 721-744.

57.

White

(1980). A heteroskedasticity-consistent covariance matrix estimator and a direct test for heteroskedasticity. Econometrica, 48, 817-838.

58.

Williams

(1996). The relation between a prior earnings forecast by management and analyst response to a current management forecast. The Accounting Review, 71, 103-113.

59.

Ziebart

(1990). The association between consensus of beliefs and trading activity surrounding earnings announcements. The Accounting Review, 65, 477-488.

60.

Zmijewski

M. E.

(1984). Methodological issues related to the estimation of financial distress prediction models. Journal of Accounting Research, 22, 59-82.