Voluntary Disclosures and Verification of Financial Statements

The Complementary Roles of Audited Financial Statements and Voluntary Disclosure: A Test of the Confirmation Hypothesis Ray Ball The University of Chicago Booth School of Business 5807 S. Woodlawn Avenue Chicago, IL 60637 Tel. (773) 834 5941 ray.ball@chicagobooth.edu Sudarshan Jayaraman Olin Business School Washington University in St. Louis Campus Box 1133 One Brookings Drive St. Louis, MO 63130 jayaraman@wustl.edu Lakshmanan Shivakumar London Business School Regent’s Park London, NW1 4SA United Kingdom LShivakumar@london.edu July 25, 2009 Abstract We examine the complementarity between voluntary disclosures and audited financial statements. We test the ―confirmation‖ hypothesis, that audited financial statements (which report actual outcomes, and hence are backward-looking) function to discipline and enhance the credibility of managers’ forward-looking disclosures. Using management forecasts as the proxy for voluntary disclosures, we report that committing to higher audit fees (a measure of the extent of financial statement verification), is associated with more frequent and more precise management forecasts, and with a larger market reaction to forecasts. These relations are not driven by litigation risk and exist in regressions that control for time-invariant firm effects. The results are consistent with voluntary disclosure and audited financial statements playing a complementary role, which implies they cannot be evaluated independently. 1. Introduction We test the hypothesis that audited financial statements and voluntary disclosures are complementary mechanisms for managers to communicate information. Gigler and Hemmer (1998) observe that reporting independently audited financial outcomes plays a ―confirmatory role,‖ allowing shareholders to evaluate the informativeness and truthfulness of past discretionary disclosures. In turn, this allows managers to credibly disclose value-relevant information, even if the information is not directly verifiable. Building on this observation, Ball (2001) argues that audited financial statements and voluntary disclosures are complements and cannot be evaluated independently. To the extent that audited financial statements allow managers to commit to disclose non-financial information truthfully, the usefulness of financial reports depends on their contribution to the total information environment and thus is not necessarily an increasing function of their separate ―surprise value.‖ The mechanisms by which managers can commit to a high quality of disclosure are not well understood. An obvious and important mechanism is commitment to a reporting regime: notably, to listing as a public firm (Ball and Shivakumar, 2005, 2008a; Burgstahler et al., 2006; Leuz et al., 2008) and to listing in a particular international jurisdiction (Coffee, 1998; Ball, Kothari and Robin, 2000; Rock, 2002). Commitment to a particular regime incurs the costs of meeting the minimum reporting and disclosure standards of that regime, as well as the costs of failing to meet those standards (litigation costs, fines, jail, etc). However, little is known about how firms within a particular regime can credibly commit to a supra-regime level of disclosure (that is, in excess of minimum requirements). Credibly committing to a high level of disclosure must incur some cost (Spence, 1973). We propose that one mechanism by which firms can credibly commit to informative and truthful 1 disclosure is by committing to a supra-regime level of independent audit of its reported financial outcomes. The costs of managers committing to additional independent audit include additional audit fees, any consequential increases in internal control and internal audit costs, additional management time, and reduced utility from managers restricting their ability to manipulate earnings. We hypothesize that: (1) managers with more forward-looking information to disclose, or with more-precise information to disclose, commit to higher levels of audit verification of actual outcomes in order to enhance the credibility of their disclosures; and (2) commitment to higher levels of audit verification is associated with larger investor responses to managers’ disclosures. We test these hypotheses by focusing on a specific type of voluntary disclosure, namely, management earnings forecasts. Relative to other disclosures (e.g., conference calls, press releases, SEC filings and MD&A reports), earnings forecasts offer several substantive advantages when studying the confirmatory role of audited financials. First, management earnings forecasts are a comparatively homogenous, large-sample disclosure variable. Second, the precision of forecasts can be measured, relative to either the actual earnings outcome or to analysts’ forecasts. Third, the informativeness of forecasts can be measured in terms of the market reaction to their announcement. Fourth, the timing of forecasts and of actual earnings announcements is known, which allows us to measure and control for the forecast horizon. Caution should be exercised in generalizing our results to other types of voluntary disclosures, particularly non-financial disclosures. Earnings forecasts are more directly confirmed by actual earnings realizations than is the case for most other discretionary disclosures. The confirmatory role of audited financial statements most likely is weaker when the disclosure cannot be so directly linked to specific financial statement outcomes. 2 We report evidence consistent with both hypotheses. First, we find a positive association between audit fees and several attributes of the value of management forecasts (viz., forecast frequency, forecast precision and forecast horizon). A one percent increase in forecast frequency is associated with a 0.151 percent increase in audit fees after controlling for all other determinants of audit fees. Similarly, a one percent increase in forecast precision and forecast horizon is associated with a 0.281 percent and a 0.123 percent increase in audit fees respectively. These results are consistent with firms committing to provide greater independent audit of financial outcomes when their managers make more frequent and more informative voluntary disclosures. Second, we document that the market reaction to discretionary forecasts (measured by 3-day abnormal stock returns and abnormal trading volume) increases in firms’ commitments to financial statement verification (measured by audit fees in excess of firm-level determinants). A one standard deviation increase in excess audit fees is associated with an approximately 10% increase in market reaction, suggesting investors perceive the credibility of voluntary disclosures to be a function of independent verification of outcomes. These results are consistent with the confirmation hypothesis of Gigler and Hemmer (1998) and Ball (2001), that voluntary disclosure and audited financial statements play complementary roles. Because firms with greater litigation risk make more disclosures (Skinner, 1994; Field et al., 2005) and also pay more audit fees as compensation for risk (Simunic, 1980; Dye, 1993; Lys and Watts, 1994; Shu 2000), we conduct additional tests to verify whether our results are due to litigation risk as an omitted variable. First, we examine the relation between audit fees and management forecast properties in low versus high litigation risk industries. We find the relation is attenuated in high risk industries, inconsistent with the litigation risk hypothesis. Second, we use data on firms named in securities class-action lawsuits to predict the probability of litigation 3 in a first stage probit regression and, in a second stage OLS regression, use the predicted values of litigation risk to examine its effect on the relation between audit fees and forecast properties. Here too, we find that the relation between audit fees and management forecasts is weaker for firms with greater predicted litigation risk, which does not support a litigation-based explanation of our results. Finally, our finding that the market reaction to management forecasts increases in excess audit fees is inconsistent with the litigation explanation, which implies excess fees are dead-weight costs to investors and could be expected to attenuate the market reaction. We perform several other robustness tests. First, the results are robust to controls for firm-level fixed effects, which mitigates a concern that the results are driven by omitted timeinvariant firm characteristics, including litigation risk and corporate governance characteristics. Second, although the standard errors in all our panel regressions are based on two-way clustering by firm and by year, we also verify that the results are robust to including only one observation per firm. Third, to ensure that the results are not affected by possible outliers, we estimate a robust regression and find that our inferences are unaltered.1 Fourth, we obtain similar results when we use a simpler model to estimate excess audit fees that controls only for firm size. Fifth, we confirm that the results are robust to potential non-linearities in the relation between audit fees and size. We also examine the robustness of our findings to alternative proxies for the level of the firm’s level of commitment to audit verification. First, we employ an indicator for the choice of Big-5 auditors, based on prior literature that finds they provide higher quality audit services, and reach identical conclusions. Second, we add a dummy variable to control for the period subsequent to the Sarbanes Oxley (SOX) Act, which led to increased audit expenditures and 1 A robust regression uses iteratively reweighted least squares and assigns higher weights to better-behaved observations. See Baker and Hall (2004). 4 allegedly to improved auditor independence and audit-verification standards (viz., assessment of internal control weaknesses), to investigate whether the reliability of management forecasts increased as a consequence. Consistent with SOX increasing audit verification standards and with higher audit verification standards increasing the reliability of management forecasts, we find that the market reaction to management forecasts increases in the post-SOX period. Our paper offers four primary contributions to the literature. We believe it is the first to empirically examine the complementarity between audited financial statements and voluntary disclosures, as predicted by Gigler and Hemmer (1998) and Ball (2001). Our results are consistent with complementarity, in that investors attribute greater credibility to management forecasts when firms commit to greater audit verification of financial statement outcomes. This is a central issue, since complementarity implies the usefulness of financial statements depends on their contribution to the total information environment and cannot be evaluated on a stand-alone basis (for example, by the separate ―surprise value‖ of earnings announcements). Second, our evidence is consistent with firms being able to commit within a regime to supra-regime levels of financial reporting and disclosure. Third, our study contributes to the management forecast literature by documenting how financial statement verification affects forecast characteristics as well as forecast consequences.2 Prior research reports that forecasts are associated with favorable market reactions (e.g., Penman (1980), Waymire (1985), Healy et al. (1999)), reduced information asymmetry (e.g., Frankel et al. (1995), Coller and Yohn (1997)) and lower litigation risk (e.g., Skinner (1994), Field et al. (2005)). Our evidence suggests an explanation for these positive outcomes: firms can credibly commit to informative and truthful management forecasts by requiring enhanced auditor authentication of financial statement outcomes. Fourth, our paper contributes to research on the cost-benefit tradeoff of disclosures. While the benefits of voluntary 2 Hirst, Koonce and Venkataraman (2008) review the management forecast literature. 5 disclosures in the form of greater liquidity and lower cost of capital (e.g., Diamond (1985), Diamond and Verrecchia (1991), Leuz and Verrecchia (2000)) are well understood, most studies isolate proprietary costs as the main disclosure cost (e.g., Verrecchia (1983, 1990), Bamber and Cheon (1998), Rogers and Stocken (2005)). We show that voluntary disclosure is associated with an additional cost – the cost of establishing credibility via enhanced financial statement verification. The rest of the paper is arranged as follows: Section 2 presents the hypotheses. Section 3 presents the research design and Section 4 discusses the results of the relation between audit fees and properties of management forecasts. Section 5 presents the relation between market reaction to forecasts and financial statement verification followed by robustness checks in Section 6. Section 7 concludes. 2. Hypotheses 2.1. Contracting versus informational roles for financial statements and voluntary disclosures Ball (2001) argues that the primary role of voluntary disclosures is to provide new information to outsiders while that of financial statements is to provide contractible variables that all contracting parties (such as factor owners and customers) can use to transact efficiently with the firm. This separation of the roles of financial statements and voluntary disclosures achieves better efficiency in both contracting and in timely provision of information. In order to provide information that can be used for contracting, financial statements contain data that are predominantly ―historical,‖ i.e., backward-looking, but verifiable by independent auditors. The contractible variables provided by audited financial statements enable lenders to restrict leverage ratios, shareholders to restrict earnings-based bonuses and other 6 stakeholders to receive routine and reliable information about the firm’s ability to fulfill its commitments thus facilitating efficient contracting. However, financial statements are a relatively inefficient mechanism for communicating new information to outsiders as financial statements face greater regulation, standardization and a relatively inflexible structure. These add significant burden in terms of time, costs and relevance, which limit the suitability of financial statements as a means of providing timely information to outsiders (also see, Ball, Robin and Sadka (2008)). Consistent with this view, earnings announcements have been shown to have relatively low informativeness (e.g., Ball and Brown (1968), Beaver (1968) and Ball and Shivakumar (2008b)). In contrast to financial statements, voluntary disclosures by managers primarily have an informational role by providing forward-looking information that is privately known only to managers. Voluntary disclosures are typically characterized by minimal regulatory hurdles, lack of structure and lack of audit. These characteristics accord greater flexibility to voluntary disclosures that makes them a potentially efficient and less expensive mechanism for managers to communicate their private information in a timely manner. However, this potential cannot be unlocked if managers cannot credibly commit to be informative and truthful in their disclosures. 2.2. Verification Demand Hypothesis Although audited financial statements and voluntary disclosures have distinct roles in contracting and in providing information, they are not independent mechanisms for mitigating agency and information asymmetry problems. As observed by Gigler and Hemmer (1998) and Ball (2001), the verifiability of financial statement data enables managers to credibly convey information through voluntary disclosures, since mandated financial statements aid outsiders to 7 evaluate the informativeness and truthfulness of management’s past voluntary disclosures, which then forms the basis upon which outsiders appraise the credibility of management’s subsequent voluntary disclosures. This relation between the audited financial statements and voluntary disclosures creates an additional role for financial statements, namely a confirmatory role, whereby audited financial statements discipline and add credence to voluntary disclosures. By hiring an independent, professional, accounting firm to audit financial reports, managers can commit to limit the extent of their manipulation of financial information.3 However, a finite quantity of auditing reduces, but does not eliminate, the potential for managerial manipulation. We propose that by demanding a higher audit level than mandated by law, managers can voluntarily bind themselves to higher verification levels, and hence limit their discretion over financial reporting and improve the veracity of their financial statements (see Caramanis and Lennox (2008) and Higgs and Skantz (2006)). In turn, committing to greater financial statement veracity in reporting actual outcomes provides a disciplining mechanism for voluntary disclosures, encourages them to be more informative and truthful, and enhances their credibility. Thus, by committing to higher audit verification of their financial statements, managers can improve the reliability and credibility of their voluntary disclosures. Further, managers are likely to commit to more audit costs when they believe the benefits from verification are greater. We conjecture that the benefits from increased verification are greater when managers have more information to disclose or have more informative (more precise) disclosures. Hence, we predict that firms making more frequent, more timely and more precise voluntary disclosures commit to higher audit verification levels. Firms that make few, 3 Titman and Truman (1986) and Datar, Feltham and Hughes (1991) develop models in which managers choose higher audit quality levels to signal favorable private information. Larcker and Richardson (2004) suggest that auditors, primarily due to concerns for their reputation, limit unusual accounting choices of client firms, and show that auditors play a key role in the governance process to limit abnormal accruals. Watts and Zimmerman (1983) trace the evolution of audits to the early stages in the development of business corporations (1200 AC). 8 untimely and vague disclosures are less likely to demand higher audit verification. The above discussion leads us to the Verification Demand hypothesis: H1: Firms making more frequent, more timely and more precise voluntary disclosures commit to higher audit verification levels. Firms that provide voluntary disclosures bear the additional costs of verification in order to increase the reliability of their disclosures to outsiders. Although voluntary disclosures may be relevant to several outside stakeholders, they are expected to be particularly relevant to stock market investors as the residual claimants. Hence, if additional verification of financial statements is intended to improve the reliability of voluntary disclosures, one should observe a greater stock market reaction to voluntary disclosures that are associated with greater financial statement verification levels. This leads us to the following implication of the Verification Demand Hypothesis: H2: The stock market reaction to firms’ voluntary disclosures increases in their level of commitment to financial statement verification. Empirical tests of the Verification Demand Hypothesis require measurement of voluntary disclosures as well as of the audit levels demanded by a firm’s management. We select management forecasts as an important and informative voluntary disclosure variable (Ball and Shivakumar, 2008b). We measure the commitment to an audit level by expenditures on audit 9 fees in excess of firm-level determinants. Audit expenditures are negotiated and approved by audit committees in advance, and are directly linked to the quantity and price of audit activity. 2.3. Litigation Risk Hypothesis Excess audit fees are estimated as the unexplained part of audit fees from a model that controls for firm-level determinants. This variable could reflect the audit effort needed to assess business risk and earnings manipulation risk (Bedard and Johnstone (2004)), or could simply reflect compensation for greater auditor litigation risk.4 We therefore consider an alternative, litigation-risk explanation for an association between management forecasts and excess audit fees. There are potentially two litigation-risk related reasons linking excess audit fees of a firm to its disclosure policy. First, prior studies find that firms with greater litigation risk make more voluntary disclosures (e.g., Skinner (1994), Field et al (2005)) and also pay higher audit fees (e.g., Simunic (1980), Dye (1993), Lys and Watts (1994), Shu (2000)). Second, auditors could view forecasting firms as potentially more risky audits, as these firms are more likely to manipulate stock prices by issuing forecasts and also are more likely to engage in greater earnings management in order to meet their earlier forecasts (Krishnan, Pevzener and Sengupta, (2009)). Moreover, the earnings manipulation risks are likely to be greater when management forecasts are more precise, as more earnings management is generally needed to meet a precise earnings estimate as compared to one that allows for a wide range of earnings numbers. Although we present the litigation risk as an alternative hypothesis to the verification demand hypothesis, it should be pointed out that Field et al (2005) show that voluntary disclosures reduce litigation risk, once the endogeneity between litigation risk and disclosures is 4 Bell, Landsman and Shackelford (2001) show that auditors increase their audit efforts for risky audits, rather than charging higher hourly fees. 10 accounted for. Thus, the empirical evidence contradicts the assumption of a positive relation between litigation risk and voluntary disclosures that underlies the litigation-risk hypothesis. Moreover, it is also important to bear in mind that voluntary disclosures are typically associated with greater litigation risk against the manager rather than the auditor and it is the litigation risk against the auditor which affects audit fees. Nevertheless, we entertain the possibility that litigation risk could lead to a positive relation between excess audit fees and both the quality and quantity of management forecasts. To distinguish between the litigation risk hypothesis and the verification demand hypothesis, we examine how the relation between financial statement verification and voluntary disclosures varies with ex-ante litigation risk of firms. The Litigation Risk hypothesis predicts that the relation between verification and disclosures should be pronounced for firms with high litigation risk. On the other hand, the Verification Demand hypothesis does not predict any difference in the relation between high and low litigation risk firms. If anything, it predicts that the relation would be attenuated for high litigation risk firms as shareholders of these firms are less likely to rely heavily on a single monitoring mechanism, viz., auditing, and, instead, are more likely to employ a wide variety of monitoring mechanisms to discipline managers. Also, the two hypotheses have different implications on the market’s response to management’s earnings forecasts. If excess audit fees represent compensation to auditors for additional litigation risk from voluntary disclosures, then investors would treat the excess audit fees as additional ―costs‖ of voluntary disclosures and consequently, the market reaction to management forecasts should be attenuated by these costs. However, if market perceives excess audit fees as a binding mechanism for higher audit verification, then the market reaction to management forecasts would be enhanced by the increased reliability of the forecasts. 11 To summarize, the Verification Demand Hypothesis predicts an insignificant or negative effect of litigation risk on the relation between management forecasts and excess audit fees, while the Litigation Risk Hypothesis predicts a positive relation. Further, the Verification Demand Hypothesis predicts a positive effect of excess audit fees on investors’ reactions to management forecasts announcements, while the Litigation Risk Hypothesis predicts a negative effect. 3. Research design 3.1. Proxies for frequency and informativeness of management forecasts We follow the voluntary disclosure literature (e.g., Baginski et al (1993), Skinner (1994), Baginski and Hassell (1997), Rogers (2008)) and focus on three attributes of management forecasts to measure the frequency and informativeness of management forecasts – number of forecasts (FREQUENCY), precision of the forecast (PRECISION) and horizon of the forecast (HORIZON). Each proxy is constructed such that larger values indicate either more frequent or more informative disclosures. Consistent with prior studies (e.g., Rogers (2008), Hirst et al (2008)), we exclude forecasts made after the fiscal period end as these are considered early earnings warnings. The first attribute of management forecast that we consider is a count of the number of annual as well as quarterly EPS forecasts made during a year by a firm (FREQUENCY). Firms not making any forecasts in a year are assigned a value of 0 for that firm-year. The next attribute we consider is the precision of the forecast (PRECISION), which is determined by the specificity of the forecasts. More specific forecasts are assigned higher values for PRECISION. Following prior studies (e.g., Baginski et al (1993), Baginski and Hassell 12 (1997), Rogers (2008)), point estimates are considered most precise, followed by range estimates (where the minimum and maximum values are provided); followed by open-ended estimates (where one end of the range is provided but not the other) and finally qualitative estimates. PRECISION takes the value of 4, 3, 2 and 1 for point, range, open-ended and qualitative estimates respectively. Firms that do not make any forecasts are assigned a missing value for precision and so are not considered in the analysis.5 The final attribute of management forecast that we examine is the forecast horizon, which denotes the number of days between the forecast announcement date and the end of the fiscal period for which the forecast is made. HORIZON is computed as the log of the difference between the fiscal period end date and the forecast date, where larger values of HORIZON indicate more timely and, hence, more informative forecasts. HORIZON is computed only for firms making an earnings forecast and with non-missing forecast dates. 3.2. Proxies for financial statement verification As the demand for financial statement verification is not directly observable, we follow the large auditing literature (starting with Simunic (1980) and Watts and Zimmerman (1983)) and use the amount of excess audit fees paid as the proxy for the extent of financial statement verification, based on the logic that incremental audit efforts demanded by a firm will be priced by auditors. Excess audit fees represent fees that are incremental to those associated with previously identified determinants. We use the log transformation of total audit fees (LN_FEES) to main consistency with prior studies and include previously identified determinants in the 5 Alternatively, we assigned a value of 0 for PRECISION for firm-years without forecasts. qualitative unaffected by this alternative approach. Our results are 13 regression specification to capture the expected or normal level of audit fees in the absence of any management forecasts. 3.3. Measurement of stock reactions to management forecast releases We measure the stock return and volume reactions to management forecasts following the approach of Landsman and Maydew (2002). The cumulative abnormal return (CAR) to management forecasts are defined as the 3-day cumulative stock returns in excess of the valueweighted market returns around the management forecast date, standardized by the standard deviation of the excess returns in the non-announcement period (days -45 to -10 relative to the management forecast date, day 0).6 Similarly, the abnormal volume (ABVOL) reaction to management forecast releases is measured as the average log turnover (i.e., share volume divided by shares outstanding) in days -1 to +1 minus the average log turnover in the non-announcement period, standardized by the standard deviation of log turnover in the non-announcement period. 4. Results 4.1. Sample and descriptive statistics Our sample comprises of data from four main sources. Audit fees data come from Audit Analytics while information on management forecasts of earnings per share (EPS) are from First Call. Stock market data are from CRSP and financial statement related information is compiled from Compustat.7 The final sample covers the period 2000 to 2007 and consists of 51,284 firm- 6 We obtain qualitatively similar results when abnormal returns from a market model are used instead of the excess returns to compute CAR. Our conclusions are also unaffected when CAR is not standardized by the abnormal return volatility. 7 There are 79,270 potential firm-year observations on COMPUSTAT for our sample period. Of these, we lose 17,329 observations when we merge this data with Audit Analytics. A preliminary analysis of the firms that are not covered in the Audit Analytics data reveals these to be relatively smaller firms with a median equity capitalization of 14 year observations for 10,233 unique firms with non-missing data for all variables. Tests regarding market reaction to management forecasts are based on a sample of 27,088 firm-year observations between 2001 and 2007. This sample starts from 2001 because of the requirement of lagged values of audit fees. To ensure that our results are not driven by a few influential observations, we winsorize the continuous variables, in each year, at the top and bottom percentiles.8 Table 1 provides the descriptive statistics for our sample of firm-years. Panel A reports the statistics for management forecast attributes, audit fees, stock market reaction to management forecasts and the control variables used in the audit-fee regressions. The distribution of FREQUENCY ranges from a minimum of 0 to a maximum of 7, with an average value of 0.58. The average value of PRECISION is 2.89, with a standard deviation of 0.64. The mean HORIZON for our sample is 4.79, which corresponds to a forecast made 120 days before the fiscal period end date. The shortest horizon in the sample is 2 days, while the longest horizon is 455 days. The mean and median annual audit fees are $1.11 million and $300,000 respectively, indicating the presence of a few large firms in the sample. A similar skew is observed in the book value of total assets for which the average is $3.6 billion, compared to a median of $240 million. The mean firm earns a negative return on assets during our sample period and 40% of the sample firm-years contain losses. Finally, 70% of our sample firms employ a Big-5 auditor. Panel B of Table 1 presents descriptive statistics on the market reactions to management forecasts releases as well as on control variables in regressions of market reaction to $50 million, compared to a median capitalization of $176 million for firms with data available on both COMPUSTAT and Audit Analytics. 8 Our conclusions are insensitive to winsorization, although the magnitude of the coefficients are in a few regressions sensitive to controls for outliers. To provide a more representative evidence, we focus on analyses that winsorize extreme observations. 15 management forecasts. The analysis in this panel is based on firm-years that have a management forecast. The average CAR, which is not conditional on the direction of the news, is close to zero and statistically insignificant.9 The mean abnormal volume in the 3-day event surrounding management forecast date is 1.109 and is statistically significant at less than the 1% level. On average, nine analysts follow each forecast-issuing firm and about 36% of these firms belong to high-litigation-risk industries, which is defined, following Rogers and Stocken (2005), as biotechnology (SIC code 2833-2836 and 8731-8734), computing (SIC codes 3570-3577 and 73707374), electronics (SIC codes 3600-3674) and retailing (SIC codes 5200-5961). 4.2. Preliminary evidence Figure 1 presents preliminary evidence of the relation between the various disclosure proxies and audit fees. Panel A presents the relation between FREQUENCY and LN_FEES, while panels B and C depict the relation between PRECISION and LN_FEES and HORIZON and LN_FEES respectively. For these figures, we group firm-years based on the values that discrete variables (FREQUENCY or PRECISION) can take or into deciles for the continuous variable, HORIZON, and plot the mean LN_FEES for each of these groups. All three panels show an upward sloping curve suggesting a positive association between the management forecast attributes and audit fees. Audit fees are monotonically increasing in the number of management forecasts (FREQUENCY) and the precision of the forecast (PRECISION). For the horizon of the forecast (HORIZON), audit fees monotonically increases across the lowest eight deciles, but decreases for the last two deciles, indicating a potential non- 9 When the 3-day cumulative abnormal returns around management forecasts are not standardized by the standard deviation of abnormal returns in the non-announcement period, then the average cumulative abnormal returns are 0.14% (t-stat=-2.90). 16 linearity in the relation.10 Although the upward sloping relationship between audit fees and management forecast properties is obvious in Figure 1, this evidence needs to be interpreted with caution as the analysis does not control for other determinants of audit fees. 4.3. Correlations Table 2 reports the Spearman correlation between LN_FEES, the management forecast attribute and the control variables in the regressions. The management forecast properties are positively correlated with each other, but the correlation coefficient is no greater than 0.16 for these variables. More interestingly and consistent with the evidence in figure 1, we find that LN_FEES is significantly positively correlated with FREQUENCY, PRECISION and HORIZON respectively with correlation coefficients ranging between 0.08 and 0.31. As correlations do not control for differences in innate characteristics between firms and over time, they should be interpreted cautiously. Nevertheless, an association between LN_FEES, FREQUENCY, PRECISION and HORIZON in univariate correlations is consistent with a first-order relation between voluntary disclosures and the extent of financial statement verification. In Panel A of Table 2 the correlation coefficients for control variables in the audit fee regressions are generally below 0.3 and, with the exception of the correlation between LN_ASSETS and LN_FEES, are never over 0.5. The high correlation between LN_ASSETS and LN_FEES reflects scale effects. In Table 2, Panel B, the two measures of the magnitude of market reactions to management forecasts releases (viz., absolute value of CAR (ABS_CAR) and abnormal volume (ABVOL)) are positively correlated with EX_FEES, which is the audit fees in excess of the firm-level determinants. This finding provides preliminary evidence consistent 10 In multivariate regressions (discussed later) we examine for non-linearity in the relation between horizon and audit fees by including a squared term for HORIZON in the regressions. There is little support for a non-linear relation in the multivariate setting and our conclusions remain unaffected by the inclusion of the squared term. 17 with the view that market participants perceive additional audit effort as increasing the reliability of management forecasts. 4.4. Multivariate regressions The above univariate analysis provides evidence supportive of the Verification Demand Hypothesis, but this analysis does not control for innate firm characteristics that could affect the audit fees. Hence, to test the relationship between audit fees (LN_FEES) and management forecast properties (DISCLOSE) in a multivariate setting, we estimate the following regression: LN _ FEES i ,t   0   1DISCLOSE i ,t   2 LN _ ASSETS   3 ROA i ,t   4 ACCR i ,t   5 CURRENT i ,t   6 FOREIGN i ,t   7 SEGMENTS i ,t   8 LIAB i ,t   9 LOSS i ,t   10 BIG _ FIVE i ,t   11OPINION i ,t   12 DEC i ,t   t (1) where DISCLOSE refers to either FREQUENCY, PRECISION or HORIZON. Following prior studies, the regression includes several control variables that have been shown to influence audit fees. Since firm size is the primary determinant of audit fees (Simunic (1980), O’Keefe, Simunic and Stein (1994), Bell, Landsman and Shackelford (2001)) we use log of total assets (LN_ASSETS) as the control for firm size. We also include total accruals scaled by total assets (ACCR); the ratio of current assets to total assets (CURRENT); the ratio of foreign segment sales to total sales (FOREIGN); and the number of business segments (SEGMENTS), to control for audit complexity. We use return on assets (ROA) and the ratio of total liabilities to total assets (LIAB) as controls for audit risk. We also include an indicator variable, (LOSS), to capture firms that have negative earnings. As larger and more reputed auditors charge higher audit fees (e.g., Francis and Wilson (1988), DeFond (1992) and Fan and Wong (2005)), we use an indicator variable (BIG_FIVE) to denote the presence of a Big Five auditor. Firms with an unqualified 18 audit opinion are indicated by OPINION. Finally, firms with a December fiscal year-end have a value of 1 for the indicator variable DEC. We expect larger firms (LN_ASSETS), firms with more current assets (CURRENT), more foreign sales (FOREIGN), more segments (SEGMENTS), larger accruals (ACCR) and presence of a Big Five auditor (BIG_FIVE) to be associated with more audit fees. Firms with greater audit risk (LIAB, LOSS) are expected to have higher audit fees, while more profitable firms (ROA) are expected to have lower fees. As qualified opinions require more investigation, we expect a negative coefficient for OPINION. Finally, we expect that DEC firms will have higher audit fees due to peak-season audits. Throughout the paper, we compute t-statistics for panel regressions using standard errors clustered by firm and by year, as suggested by Petersen (2009). Alternatively, we also estimated the standard errors after including industry- and year-fixed-effects to capture common shocks that cause cross-sectional correlation in the errors and firm-level clustering to correct for possible serial correlation due to unobserved firm effects. Since the results are qualitatively similar across the two approaches we report only results based on 2-way clustering. Table 3 presents the estimates for ordinary least squares regression of Equation (1).11 The first three sets of columns present results when each management forecast property is considered separately in the regressions, while the fourth set of columns includes all the management forecast attributes in the same regression. The sample sizes for the FREQUENCY regressions are 51,284 observations while that for the PRECISION and HORIZON regressions falls to 9,033 observations each because the latter regressions are estimated only for firms making management forecasts. 11 In untabulated analysis, we alternatively estimate the regression as a rank regression and obtain qualitatively similar results. 19 The coefficient of each management forecast attribute is positive and significant at less than the 1% level. In particular, the coefficient of FREQUENCY is 0.060 and the t-statistic is 4.97. Similarly, the coefficients of PRECISION and HORIZON are 0.122 and 0.123 and the associated t-statistics are 3.36 and 6.91 respectively. Thus, even after controlling for known determinants, audit fees are higher in firms with more frequent forecasts, more precise forecasts and longer horizon forecasts. To assess the economic significance of our results, we regress audit fees on log values of our management forecast proxies (except HORIZON, which is already in log). In unreported results, we find that the coefficient of the log values of FREQUENCY and PRECISION is 0.151 and 0.281 respectively and the t-statistics are 4.32 and 3.31 respectively. The coefficients suggest that a 1% increase in forecast frequency and in forecast precision is associated with a 0.151 percent and a 0.281 percent increase in audit fees respectively. Further, the coefficient of 0.123 of HORIZON suggests that a 1% increase in forecast horizon is associated with a 0.123% percent increase in audit fees. The relations between audit fees and the control variables are broadly consistent with prior studies. The explanatory power of the regressions varies between 64% and 73%, which is consistent with the adjusted r-squares reported in prior studies. When all three attributes of management forecasts are included in the same regression, all the coefficients continue to remain significantly positive. The above results supports our Hypothesis H1 that firms where managers make more frequent and more informative forecasts are associated with a greater level of financial statement verification. 4.5. Effect of financial statement verification levels on market reaction to forecasts We examine Hypothesis H2 by regressing stock price and volume reactions to management forecasts on audit fees in excess of firm-level determinants. It is worth noting that 20 the Verification demand hypothesis, as stated in Hypothesis H2, predicts a positive relation between the market reaction to management forecasts and excess audit verification levels, while the litigation risk hypothesis predicts a negative relation between the two variables. We measure the price and volume reactions to management forecasts as the absolute value of CAR (ABS_CAR) and abnormal volume (ABVOL) respectively. We compute excess audit fees (EX_FEES) as the residual from the regression of log audit fees (LN_FEES) on the customary firm-level determinants of audit fees excluding the disclosure proxy (from equation (1)). In this analysis, we focus on excess audit-fees rather than the total audit fees to ensure that the relation between financial statement verification and market reaction to management forecasts is not driven by firm characteristics. We lag EX_FEES by one year so that they capture the ex-ante level of financial statement verification. One concern with the use of EX_FEES is that it averages zero across the sample. Consequently, by construction, some firm-years will have negative EX_FEES, while others will have positive EX_FEES. Our contention is not that a negative EX_FEES implies a fee discount or that a positive EX_FEES measures the fee premium for additional audit verification associated with management forecasts. We are more interested in the ordinal ranking of EX_FEES across firms rather than their actual values, even though we use the actual values in the regressions. In untabulated analysis, we confirm that our results are robust to using ranks of EX_FEES in the regressions. Figure 2 presents the preliminary relation between market reaction to management forecasts and excess audit fees. The x-axes plot increasing deciles of excess audit fees with the rightmost decile indicating most excess fees. The y-axes plot mean values of absolute CAR 21 (ABS_CAR) and abnormal trading volume (ABVOL) for each decile.12 The plots for ABS_CAR and ABVOL are presented in panels A and B respectively. Both panels depict upward sloping lines which indicate a positive association between market reaction to management forecasts and excess audit fees. To control for other determinants of market reactions to forecasts, we estimate the following multivariate regressions: ABS _ CARi ,t   0   1EX _ FEESi ,t   2 MVEi ,t   3 LEVi ,t   4 MBi ,t   5 ANALYST,t i   6 RETSTDi ,t   7 LITi ,t   t ABVOLi ,t   0   1EX _ FEESi ,t   2 MVEi ,t   3 LEVi ,t   4 MBi ,t   5 ANALYST,t i   6 RETSTDi ,t   7 LITi ,t   t (2) Following prior studies, we include the log of market value of equity (MVE), leverage (LEV), the market-to-book ratio (MB), analyst following defined as the log of number of analysts (ANALYST), volatility of stock returns (RETSTD) to capture firm uncertainty and the litigation indicator (LIT) as controls. Table 4 presents the results of the above regressions. The first set of columns in Panel A of the table present the relation between absolute value of CAR (ABS_CAR) and excess audit fees (EX_FEES) while the next set of columns present the relation between abnormal trading volume (ABVOL) and EX_FEES. Consistent with Hypothesis H2, the coefficient of EX_FEES is positive (coefficients are 0.388 and 0.148) and significant (t-statistics are 6.36 and 5.19) in both regressions. A one standard deviation (0.733) increase in excess audit fees is associated with 9.6% increase in stock price reaction and a 9.78% increase in trading volume reaction. This supports Hypothesis H2 that the market reaction to management forecasts is greater for firms 12 We draw identical conclusions when we focus on median values of ABS_CAR and ABVOL. 22 with higher levels of financial statement verification, consistent with the prediction in Ball (2001). This result is inconsistent with the Litigation Risk hypothesis. The above regressions show that the magnitudes of market reactions to management forecasts are greater for firms with higher excess audit fees. However, the higher magnitude of the reaction could be due to larger forecast surprises rather than greater reliance by investors on forecasts associated with higher financial statement verification. To test this alternative, we estimate the following regression of CAR on CARi ,t  0  1SURPi ,t  2 SURP,t * EX _ FEESi ,t  3 EX _ FEESi ,t  4 MVEi ,t  5 LEVi ,t  6 MBi ,t i  7 ANALYST,t  8 RETSTD,t  9 LITi ,t   t i i (3) where CAR is the cumulative abnormal return described in section 3.3. SURPi,t is the new information in the management forecast. It is computed as the difference between point management forecast (or midpoint if forecast is provided as a range) of earnings per share (EPS) and the most recent consensus analysts forecast of EPS, measured either as the mean or as the median analysts’ forecasts, scaled by the average pre-announcement period stock price (days -45 to -10 relative to the management forecast date, day 0).13 In order to compute SURP, we retain only point and range forecasts in this analysis. The other variables are as defined earlier. A positive coefficient for the interactive term (β2) in the above equation would indicate that, even after controlling for the magnitude of the forecast surprise, investors place greater reliance on management forecasts that are associated with higher financial statement verification. The results in Panel B of Table 4 indicate a significantly positive coefficient on SURP, which suggests that management forecasts surprises are informative to stock market investors. Further, the coefficient on β2 is also significantly positive. The results are qualitatively similar 13 The consensus analysts forecast data are obtained from Details file of IBES. 23 across regressions based on median and mean analysts’ forecasts. The coefficient of 4.96 on the interactive term SURP*EX_FEES suggests that the market reaction to the median earnings surprise increases by 10.8% for a firm that moves from the first quartile of EX_FEES to the third quartile of EX_FEES. These results support Hypothesis H2 of the verification demand hypothesis, but are inconsistent with the Litigation Risk hypothesis. 5. Examining the alternative Litigation Risk hypothesis Although the results from analysis of market reactions to management forecasts do not support the Litigation Risk hypothesis, we conduct additional tests to examine the validity of this hypothesis. These additional tests exploit the cross-sectional implication of the Litigation Risk hypothesis, which is that, if auditors require a fee premium to cover greater litigation risks associated with management forecasts, then the premium (per unit of risk) will be larger for firms with higher litigation risks. We examine this cross-sectional implication by interacting DISCLSOURE variable in Equation (1) with a proxy for litigation risk. The Litigation Risk hypothesis predicts a positive coefficient on this interactive term. On the other hand, the Verification Demand hypothesis does not have any direct prediction for the interactive term. However, to the extent shareholders of high-litigation-risk firms rely on a variety of alternative mechanisms to discipline managers, in addition to relying on auditors, the coefficient on the interactive term would be negative. We employ two alternative proxies for litigation risks, namely, an indicator variable for industries prone to greater litigation and probability of litigation that is estimated from a prediction model. The following two sub-sections present results from analyses based on each of these alternative proxies for litigation risks. 24 5.1. High litigation risk industries Prior studies have identified firms that operate in the bio-technology (SIC code 28332836 and 8731-8734), computing (SIC codes 3570-3577 and 7370-7374), electronics (SIC codes 3600-3674) and retailing (SIC codes 5200-5961) industries to be more prone to litigation risk than other firms (e.g., Francis et al (1994), Shu (2000), Johnson et al (2001), Field et al (2005), Rogers and Stocken (2005)). We follow these studies and define LIT as an indicator variable to denote firms that operate in these high litigation risk industries. To examine the role of litigation risk in the relation between audit fees and voluntary disclosures, we include an additional explanatory variable in Equation (1) which interacts LIT with each of the management forecast properties (FREQUENCY, PRECISION, and HORIZON). From Table 5, which tabulates the regression estimates, we observe a negative coefficient of the interaction term for all three management forecast attributes((FREQUENCY, PRECISION, and HORIZON), although the coefficient is statistically significant only in the FREQUENCY regression. This indicates that the relation between management forecasts and audit fees is weaker in high litigation risk industries. Further, the stand-alone coefficients of FREQUENCY, PRECISION, and HORIZON continue to be positive and significant (t-statistics of 6.44, 3.75 and 6.84 respectively). The negative coefficient on the interactive variable is contrary to the prediction from Litigation Risk Hypothesis and shows that the relation between audit fees and management forecasts is not driven by litigation risk. These results are more consistent with the Verification Demand hypothesis. 25 5.2. Probability of Litigation based on securities class-action lawsuits To further explore the litigation hypothesis, we use a sample of firms that have been named in securities class-action lawsuits. The Securities Class Action Clearinghouse database is maintained by Stanford Law School (in co-operation with Cornerstone Research) and has been used in recent studies such as Rogers and Stocken (2005) and Rogers and Van Buskirk (2009).14 We first build a prediction model of the probability of litigation based on firm-level determinants of litigation risk identified in prior studies (e.g., Lys and Watts (1994), Shu (2000), Johnson et al (2001)). We then use the predicted probabilities from this model to examine the role of litigation risk in the relation between audit fees and management forecasts (see Rogers and Stocken (2005) for a similar methodology). The prediction model in the first stage is as follows: Pr(Law1)  G( 0   1SIZEi ,t   2 RETi ,t  3 RETSTDi ,t   4TURNi ,t  5 MBi ,t   6CLIENT _ VOLi ,t  7 NBR _ CLIENTSi ,t  8 DOL _ AUDITi ,t  9OPINIONi ,t  10 LITi ,t   i ,t (4) where, the variables are as defined below. LAW SIZE ANNRET RETSTD TURN MB CLIENT_VOL NBR_CLIENTS DOL_AUDIT OPINION LIT Indicator denoting a securities class action lawsuit during the year Log of market value of equity Annual stock return during the year Standard deviation of stock returns Stock turnover Market to book ratio Variance of the auditor’s client size Number of clients of the auditor The total sales dollars that the auditor audits Indicator variable, taking a value of 1 for unqualified audit opinions High litigation risk industry (defined in Section 5.1) Table 6, Panel A, presents results of the above prediction model. We estimate the model using Probit, although our results are qualitatively similar when we estimate it using Logit. Our results are consistent with those in prior studies and suggest that the likelihood of litigation is 14 http://securities.stanford.edu/ 26 greater for larger firms, firms that experienced poor stock performance during the year, more volatile firms, firms with greater turnover and those that operate in the high risk industries (biotechnology, computing, electronics and retailing). Further, several of the auditor related variables are also significant, suggesting that the prediction model potentially captures auditor litigation risk as well. In particular, firms are more likely to be sued when the auditor audits a wide cross-section of clients (greater CLIENT_VOL), audits more clients, audits fewer sales dollars, and issues an unqualified audit opinion. The explanatory power of the model is 0.14, which is slightly higher than the 0.12 of Rogers and Stocken (2005). We estimate the predicted probability of litigation (LIT_PRED) from the probit model and interact it with the forecast properties. As per the Litigation Risk hypothesis, we expect the relation between audit fees and voluntary disclosures to be stronger for firms with high litigation risk, i.e., we expect the coefficient of the interaction terms to be positive. Panel B of Table 6 presents results of the interactions of LIT_PRED with the forecast proxies. Similar to the results in Table 5, the coefficient of FREQUENCY*LIT_PRED is negative and significant. Further, the coefficient of PRECISION*LIT_PRED is also negative and significant while that of HORIZON*LIT_PRED is insignificant. The coefficients on FREQUENCY, PRECISION, and HORIZON continue to be significantly positive (t-statistics of 6.59, 2.92 and 5.49 respectively). These show that the relation between voluntary disclosures and audit fees is not pronounced for firms with higher predicted probabilities of litigation risk. This result does not support the Litigation Risk hypothesis, but is consistent with the Verification Demand Hypothesis.15 15 In untabulated results, we examine whether market reactions to management forecasts vary across firms depending on their litigation risks since it is possible that potential litigation, rather than audit verification, acts as a disciplining mechanism for manager’s voluntary disclosures. Contrary to the above expectation, we find that market reaction to management forecasts, irrespective of whether or not we control for the magnitude of the forecast surprise, are 27 6. Robustness tests We conduct a battery of robustness checks of our results by using alternative empirical specifications, employing alternative proxies for financial statement verification and verifying that the relation between audit fees and management forecasts exists across various firm size groups. 6.1. Cross-sectional regression Although the standard errors in our regressions are clustered by firm and by year, to further mitigate concerns about serial-correlation due to unobserved firm-specific effects, we use a cross-sectional regression that uses one observation per firm. For each firm, we compute the time-series average of the variables and include these averages in the regression. The averages also alleviate concerns that our results are driven by lack of controls for year fixed-effects. Panel A of Table 7 presents the results of the cross-sectional regression of audit fees on the various management forecast proxies while panels B and C present the regression of market reactions to management forecasts on excess audit fees. The number of observations in Panel A reduces to 10,233 for FREQUENCY and to 2,322 for PRECISION and HORIZON, which corresponds to the number of unique firms in the sample. In regressions that separately includes each forecast attribute, the coefficients of FREQUENCY and HORIZON are positive and significant (t-statistics are 10.60 and 5.17 respectively), while that on PRECISION is significantly only at the 10% level (t-statistic is 1.65). When all forecasts attributes are included in the same regression, FREQUENCY and HORIZON continue to be statistically significant, significantly lower for firms with higher ex-ante probability of litigation in analysis based on LIT_PRED variable and insignificantly different across high and low litigation industries in analysis based on LIT indicator. These provide further evidence that our earlier results are unlikely to be explained by litigation risks. 28 while the coefficient on PRECISION is insignificant. An F-test that all three variables are jointly zero is rejected at less than 1% significance level. In panel B, the relation of absolute CAR and abnormal volume with excess audit fees is robust to the cross-sectional specification. The coefficient of EX_FEES is positive (0.324 and 0.123) and significant (t-statistics = 5.12 and 5.10 respectively) in the ABS_CAR and the ABVOL regressions respectively. These results suggest that the relation between voluntary disclosures and financial statement verification is not affected by possible serial correlation or year fixed effects. Finally, we also confirm in Panel C of Table 7, that the effect of excess audit fees on market reactions to forecast surprises (SURP) is robust to the cross-sectional specification. The coefficient on SUPR*EX_FEES is a significantly positive 4.972 (t-stat=2.01) in regressions based on median forecasts and 4.726 (t-stat=1.69) in regressions based on mean analysts forecasts. 6.2. Firm-fixed effects regressions It is possible that our results from the levels regression of audit fees on management forecast attributes might be confounded by uncontrolled firm-specific factors that are associated with management forecasts as well as with audit fees. Recent studies such as Ajinkya et al (2005) and Karamanou and Vafeas (2005) find that the corporate governance structure of the firm (in particular board structure, institutional ownership and audit committee structure) affects firm’s voluntary disclosure behavior. Further, Carcello et al (2002) find that board structure is associated with audit fees. To ensure that our results are not driven by such omitted factors, we estimate the audit-fee regressions using a firm fixed-effects specification. This test is also relevant to distinguishing between Litigation Risk hypothesis and the Verification Demand 29 hypothesis, since one motivation for the Litigation Risk hypothesis is that riskier firms increase voluntary disclosures and because of their riskiness (rather than increased voluntary disclosures) face greater audit fees. However, it should be noted that, to the extent that a firm’s policy on voluntary disclosures is also relatively constant over time, the firm fixed effects can lead to an over-correction. Hence, the firm-fixed effect regression is likely to yield a conservatively biased estimate of the relation between audit verification and management forecasts. We implement the firm fixed effect regressions by differencing the variable values for a given firm-year with their averages over time for the firm. This approach allows us to, not only control for firm fixed effects, but also to estimate standard errors that are clustered by firm and by year. The results are presented in Table 8. The coefficients of FREQUENCY, PRECISION and HORIZON are all positive and significant (t-statistics are 2.32, 3.43 and 4.88 respectively) in the audit fee regressions, reported in Panel A. Further, when all three proxies are introduced in the same regression, the coefficients of all three variables remain positive and significant. These results provide evidence that firm fixed-effects do not drive the observed relation between management forecast properties and audit fees. This evidence is also inconsistent with the omitted correlated variables arguments of the Litigation Risk hypothesis. In Panel B of table 8, which presents results from regression of market reactions to forecasts, the coefficient on excess audit fees (EX_FEES) is positive (0.501 and 0.182) and significant (t-statistics are 6.98 and 6.41) in the ABS_CAR and the ABVOL regressions respectively. Thus, even after controlling for firm fixed-effects, excess audit fees are significantly related to market reactions to management forecasts. These results provide 30 reassurance that our results are not driven by omitted time-invariant factors such as the corporate governance structure or litigation risks of the firm. 6.3. Robust regression We estimate a robust regression to verify that our results are not driven by outliers. To conserve space, we discuss these results without tabulating them. The coefficients of FREQUENCY, PRECISION and HORIZON in audit fee regressions are positive and highly significantly in these regressions with t-statistics exceeding 9.5 for all the variables. In regressions of absolute CAR and abnormal volume, the coefficient on EX_FEES continues to be positive and significant. The results from regression of CAR on forecasts surprises (SURP) and SURP interacted with EX_FEES also are consistent with the earlier reported results. The coefficient on the SURP*EX_FEES is 3.83 and the t-statistics is3.02 (2.92) when median (mean) analysts forecasts are employed to compute SURP, Overall, the robust regressions show that outliers do not affect the documented relation between voluntary disclosures and audit fees as well as between market reaction to management forecasts and excess audit fees. 6.4. Audit fee model In order to investigate the relation between audit verification and management forecasts, we rely on the audit fee model in Equation (1) based on prior audit literature. However, we recognize that this model is relatively ad-hoc and combines both scaled and unscaled variables. Since firm size tends to be one of the most important explanatory variables in determining the level of audit fees, we repeat our analysis using a very simple audit fee model that regresses log 31 of audit fees on log of total assets. The untabulated results from this model yield identical conclusions to those derived from Tables 3 and 4. 6.5. Firm size terciles We ensure that our results are not driven by possible non-linearities in the relation between audit fees and firm size. We do so by splitting our sample based on terciles of firm size and estimating the relation between audit fees and voluntary disclosures in each of the terciles. We only discuss the results for the sake of brevity. We find a positive and significant relation between LN_FEES and FREQUENCY and LN_FEES and PRECISION in all three terciles. As far as HORIZON is concerned, the number of observations in the smallest tercile falls to 581. We find a positive but insignificant relation between HORIZON and audit fees for these firms. The relation between HORIZON and audit fees is positive and significant (at the 10% and 1% respectively) for the second and third terciles respectively. Overall, the relation between voluntary disclosures and audit fees exists across all three firm size terciles. 6.6. Choice of auditor to proxy for financial statement verification While our empirical specifications in Sections 4 and 5 include the presence of a Big-5 (or Big-4) auditor as a control variable, we use this variable in this section as an alternate proxy for financial statement verification as prior studies have observed that Big Five auditors provide superior audit quality (e.g., DeAngelo (1981), Willenborg (1999)). Hence, to increase the reliability of management forecasts, a firm’s management could employ Big-5 auditors to ensure a higher verification standard for its financial statements. 32 Table 9 reports results from a Probit regression of the choice of Big-5 auditor on FREQUENCY and other control variables. The control variables included in the regression are firm size (LN_ASSETS), profitability (ROA), leverage (LEV), asset turnover (ASSET_TURN), the market-to-book ratio (MB) and the extent of capital expenditure scaled by total assets (CAPEX). In this analysis, we report the results only for FREQUENCY since the other two attributes of management forecasts, namely PRECISION and HORIZON as well as regressions of market reactions to management forecasts are limited to firms issuing forecasts and among these firms 93% of the firms employ Big-5 auditors. In comparison, among firms not making management forecasts, less than 70% employ Big Five auditor. While the larger percentage of Big Five auditors in the sample of firms issuing management forecasts (before controlling for innate firm characteristics) provides preliminary evidence in support of the Verification Demand Hypothesis, the relatively large percentage of Big-5 auditors leaves very little variation in the dependent variables when analysis is restricted to the sample of firms issuing forecasts. The coefficient on FREQUENCY is a significant 0.218 (t-statistics=7.41). This is consistent with our earlier findings based on audit fees and supports the contention that firms issuing management forecasts are more likely to employ higher quality Big-Five auditors. The controls variables in the regression are generally significant, with the exception of MB. The results indicate that larger firms as well as firms with greater turnover and capital expenditures are more likely to hire a Big-5 auditor, while more profitable firms and more levered firms are less likely to hire a Big-5 auditor. These results are broadly in line with those reported in Lai (2009) among others. 33 6.7. Effect of Sarbanes Oxley Act on reliability of management forecasts. The Sarbanes Oxley Act (SOX) introduced several requirements aimed at improving audit verification standards, such as enhanced auditor independence, monitoring of auditors, requiring auditors to assess internal control weaknesses, etc. The introduction of this Act provides a natural experiment to examine the effect of enhanced audit verification standards on forecast reliability and also allows the use of an alternative proxy for verification standards. We study the effect of SOX on investors’ responses to management forecasts by reestimating regression equations (2) and (3) after replacing EX_FEES with an indicator variable, POST that takes the value 1 for the years 2004 to 2007. We also exclude forecasts made in 2002 and 2003 from this analysis to clearly distinguish the periods before and after the SOX regulation became effective. The results from this analysis are reported in Table 10. In Panel A, which reports results from regression of absolute values of abnormal returns and regression of abnormal volume, we observe a significantly positive coefficient on POSTit. Panel B reports results from regression of CAR on forecast surprise (SURP). Interestingly, even in this regression that controls for the magnitude of the forecast surprise, we find a significantly positive coefficient on the interactive variable SURP*POST. These results indicate that investors respond more to forecast surprises in the post-SOX period than in the pre-SOX period, consistent with increased audit verification standards increasing reliability of management forecasts. While this provides corroborative evidence for the Verification Demand Hypothesis, we recognize that the introduction of SOX is endogenous to the relatively unique events during that period, which could explain our results. Also, it is possible that SOX affected investors’ responses to management forecasts in ways 34 other than through audit verification, such as through its effects on corporate governance mechanisms. 7. Conclusion This paper examines the complementarity between audited financial statements and voluntary managerial disclosures, as hypothesized by Gigler and Hemmer (1998) and Ball (2001). First, we find a positive association between measures of the value of voluntary forecasts (forecast frequency, forecast precision and forecast horizon) and the amount of audit fees. This suggests that firms commit to greater auditor financial statement authentication when their managers make more frequent and more informative voluntary forecasts. Thus, voluntary disclosures bring about greater financial statement verification thereby improving the contracting value of financial statements. Second, we find that investor reaction to management forecasts is greater in firms with greater financial statement verification (as measured by audit fees). This suggests that in an environment where financial statements are subjected to a greater degree of verification, investors perceive voluntary management forecasts to be more credible. Thus, financial statement verification enhances the information value of management forecasts. Additional tests suggest that the relations are not driven by litigation risk and are robust to alternative empirical specifications. In particular, we employ changes regressions and control for firm fixed-effects. These results provide reassurance that our results are not driven by correlated omitted firm factors, such as governance structure and litigation risk, which could affect the characteristics of management forecasts, the market reaction to forecasts and also audit 35 fees. Our results are also robust to the use of a Big Five auditor as an alternative proxy for financial statement verification. Our paper offers four primary contributions to the literature. We believe it is the first to empirically examine the complementarity between audited financial statements and voluntary disclosures, as predicted by Gigler and Hemmer (1998) and Ball (2001). This is a central issue, since complementarity implies the usefulness of financial statements depends on their contribution to the total information environment and cannot be evaluated on a stand-alone basis (for example, by the separate ―surprise value‖ of earnings announcements). Second, our evidence is consistent with firms being able to commit within a regime to supra-regime levels of financial reporting and disclosure. Third, our study contributes to the management forecast literature by documenting how financial statement verification affects forecast characteristics as well as forecast consequences. Fourth, our paper contributes to research on the cost-benefit tradeoff of disclosures. Prior research views disclosure of proprietary information as the primary cost of disclosure, whereas we identify an additional disclosure cost, the cost of establishing credibility via enhanced financial statement verification. 36 References Ajinkya, B.B., S. Bhojraj and P. Sengupta. 2005, The association between outside directors, institutional investors and the properties of management earnings forecasts. Journal of Accounting Research 43 (3): 343-376 Baginski, S., E. Conrad and J. Hassell. 1993, The effects of management forecast precision on equity pricing and on the assessment of earnings uncertainty. The Accounting Review 68: 913–27 Baginski, S., and J. Hassell. 1997, Determinants of Management Forecast Precision. The Accounting Review 72: 303–12 Baker, G. and B. Hall, 2004, CEO incentives and firm size, Journal of Labor Economics 22(4), 767-798 Ball, R., 2001, Infrastructure Requirements for an Economically Efficient System of Public Financial Reporting and Disclosure, Brookings-Wharton Papers on Financial Services: 127–69 Ball, R. and P. Brown, 1968, An empirical evaluation of accounting income numbers, Journal of Accounting Research 6: 159-178. Ball, R., Kothari, S.P., Robin, A., 2000. The effect of international institutional factors on properties of accounting earnings, Journal of Accounting & Economics 29, 1-51. Ball, R., A. Robin and G. Sadka, 2008, Is Financial Reporting Shaped by Equity Markets or by Debt Markets? An International Study of Timeliness and Conservatism, Review of Accounting Studies 13 (2008): 168-205. Ball, R., Shivakumar, L., 2005. Earnings quality in U.K. private firms: comparative loss recognition timeliness, Journal of Accounting & Economics 39, 83-128. Ball, R. and L. Shivakumar, 2008a, Earnings Quality at Initial Public Offerings, Journal of Accounting and Economics 45, 24-49. Ball, R. and L.Shivakumar, 2008b, How Much New Information Is There in Earnings ? Journal of Accounting Research, Vol. 46 (5), pp. 975-1016 Bamber, L., and Y. S. Cheon. 1998, Discretionary Management Earnings Forecast Disclosures: Antecedents and Outcomes Associated with Forecast Venue and Forecast Specificity Choices. Journal of Accounting Research 36: 167–90. Beaver, W. H., 1968, The Information Content of Annual Earnings Announcements. Journal of Accounting Research, (Supplement) 6: 67-92. 37 Bedard, J. C., and K. M. Johnstone., 2004, Earnings Manipulation Risk, Corporate Governance Risk, and Auditors' Planning and Pricing Decisions. The Accounting Review 79, 277-304. Bell, T., W. Landsman and D. Shackelford, D. 2001, Auditors’ perceived business risk and audit fees: Analysis and evidence. Journal of Accounting Research 39, 35–43. Burgstahler, David C., Hail, Luzi and Leuz, Christian, 2006, The Importance of Reporting Incentives: Earnings Management in European Private and Public Firms. The Accounting Review 81, 983-1016. Caramanis,C. and C. Lennox, 2008, Audit effort and earnings management, Journal of Accounting and Economics 45: 116-138. Carcello, J. V., Hermanson, D. R., Neal and T. L., Riley, R. A., 2002, Board characteristics and audit fees. Contemporary Accounting Research, Vol. 19, pp. 365–384. Coller, M., and T. Yohn. 1997, Management Forecasts and Information Asymmetry: An Examination of Bid-Ask Spreads.‖ Journal of Accounting Research 35 (1997): 181–91 Datar , S.M., G.A. Feltham and J.S. Hughes, 1991, The role of audits and audit quality in valuing new issues, Journal of Accounting and Economics 14,3-49. DeAngelo, L., 1981, Auditor size and auditor quality, Journal of Accounting and Economics 3(3):183-199. DeFond, M. L., 1992. The Associations between Changes in Client Firm Agency Cost and Auditor Switching. Auditing: A Journal Of Practice And Theory 11, 16–31. Diamond, D., 1985, Optimal Release of Information by Firms Journal of Finance 40: 1071– 1094. Diamond, D., and R. Verrecchia. 1991, Disclosure, Liquidity, and the Cost of Capital. Journal of Finance 66: 1325–55 Dye, Ronald A., 1993, Auditing Standards, Legal Liability, and Auditor Wealth, The Journal of Political Economy, Vol. 101, No. 5 Fan, J. and T.J. Wong, 2005, Do external auditors perform a corporate governance role in emerging markets? Evidence from East Asia. Journal of Accounting Research, Vol. 43, pp. 35–72. Field, L., M. Lowry and S. Shu, 2005, Does disclosure deter or trigger litigation? Journal of Accounting and Economics 39: 487-507 Francis, J.; D. Philbrick; and K. Schipper, 1994, Shareholder Litigation and Corporate Disclosures. Journal of Accounting Research 32, 137–64. 38 Francis, J.R. and E. Wilson, 1988. Auditor changes: a test of theories relating to agency costs and auditor differentiation. The Accounting Review 63, 663–682 Frankel, R., M. McNichols, and G. P. Wilson., 1995, Discretionary Disclosure and External Financing Accounting Review 70, no. 1: 135-50. Gigler, F., and T. Hemmer. 1998, On the Frequency, Quality, and Informational Role of Mandatory Financial Reports Journal of Accounting Research 36: 117–47. Healy, P.; A. Hutton; and K. Palepu. 1999, Stock Performance and Intermediation Changes Surrounding Increases in Disclosure. Contemporary Accounting Research 16: 485–520 Higgs, J.L. and T. R. Skantz, 2006, Audit and non-audit fees and the market’s reaction to earnings announcements. Auditing: A journal of practice and theory 25: 1-26. Hirst, E. D; L. Koonce and S. Venkataraman, 2008, Management Earnings Forecasts: A Review and Framework, Accounting Horizons ; 22, 3 Johnson, M., R. Kasznik and K. Nelson, 2001, The impact of securities litigation reform on the disclosure of forward-looking information by high technology firms, Journal of Accounting Research 39, pp. 297–327 Karamanou, R., and N. Vafeas, 2005, The Association between Corporate Boards, Audit Committees, and Management Earnings Forecasts: An Empirical Analysis. Journal of Accounting Research 43: 453–73 Krishnan, G.V, M. Pevzner and P. Sengupta, 2009, How do auditors view managers’ disclosure strategy? The effect of earnings guidance on audit fees. Working paper, Lehigh University. Lai, K., 2009, Does audit quality matter more for firms with high investment opportunities?, Journal of Accounting and Public Policy, 28(1):33-50. Landsman, W. and E.L. Maydew, 2002, Has the information content of quarterly earnings announcements declined in the past three decades?, Journal of Accounting Research 40: 797-808. Larcker, D.F. and S.A. Richardson, 2004, Fees paid to audit firms, accrual choices and corporate governance, Journal of Accounting Research 42(3): 625-658. Leuz, Christian, Triantis, Alexander J. and Wang, Tracy Yue, 2008. Why Do Firms Go Dark? Causes and Economic Consequences of Voluntary SEC Deregistrations. Journal of Accounting & Economics. Leuz, C., and R. Verrecchia. 2000, The Economic Consequences of Increased Disclosure. Journal of Accounting Research 38, 91–124. 39 Lys, T. and Watts, R.L., 1994, Lawsuits against auditors, Journal of Accounting Research 32, 65–93. O'Keefe, Terrence B., D. Simunic and M. Stein, 1994, The Production of Audit Services: Evidence from a major public accounting firm, Journal of Accounting Research 32, Penman, S. H., 1980, An Empirical-Investigation of the Voluntary Disclosure of CorporateEarnings Forecasts Journal of Accounting Research 18, 132-60. Petersen, M., 2009, Estimating standard errors in finance panel data sets: Comparing approaches. Review of Financial Studies 22, 435-480. Rock, Edward B., 2002, Securities Regulation as Lobster Trap: A Credible Commitment Theory of Mandatory Disclosure. Cardozo Law Review 23, 675-704. Rogers, J., 2008, Disclosure quality and management trading incentives. Journal of Accounting Research 46, 1209-1240. Rogers, J. L., and Andrew Van Buskirk, 2009, Shareholder litigation and changes in disclosure behavior, Journal of Accounting and Economics 47, 136-156. Rogers, J., and P. Stocken. 2005, Credibility of Management Forecasts. The Accounting Review 80, 1233–60. Shu, S. Z., 2000, Auditor Resignations: Clientele Effects and Legal Liability. Journal of Accounting & Economics 29, 173–205 Simunic, D. A., 1980, The pricing of audit services: Theory and evidence. Journal of Accounting Research 18: 161–190. Skinner, D. 1994, Why Firms Voluntarily Disclose Bad News. Journal of Accounting Research 32: 38–60. Spence, M., 1973, Job market signaling, Quarterly Journal of Economics 87, 355-374. Titman, S. and B. Trueman, 1986, Information quality and the valuation of new issues, Journal of Accounting and Economics 8, 159-172. Verrecchia, R. E., 1983, Discretionary Disclosure Journal of Accounting & Economics 5, 17994. ———, 1990, Endogenous Proprietary Costs through Firm Interdependence Journal of Accounting & Economics 12, no. 1-3: 245-50. 40 Watts, R. and Zimmerman, J., 1983, Agency Problems, Auditing, and the Theory of the Firm: Some Evidence, Journal of Law and Economics 26, 613–633. Waymire, G. "Earnings Volatility and Voluntary Management Forecast Disclosure." Journal of Accounting Research 23, no. 1 (1985): 268-95. Willenborg, M., 1999, Empirical analysis of the economic demand for auditing in the initial public offerings market, Journal of Accounting Research 37(1): 225-238. 41 Table 1: Descriptive statistics Panel A of this table presents descriptive statistics for variables used in regressions of audit fees on properties of management forecasts. The sample comprises of firms with audit fee data available on Audit Analytics and covers the period 2000 to 2007. FREQUENCY indicates the number of management forecasts made during the year. PRECISION captures the precision of the forecast and takes the value from 0 to 4. Point estimates are assigned a value of 4, while range estimates, open-ended estimates and qualitative estimates are assigned values of 3, 2 and 1 respectively. Firms not making a forecast are given a value of 0. HORIZON denotes the time period between the date of the forecast and the date of the fiscal period end. It is populated only for firms making a forecast. The amount of annual audit fees in $millions is indicated by AUDIT_FEES. LN_FEES is the log of audit fees. ASSETS represents total assets in $millions. ROA is net income divided by total assets. ACCR indicates the absolute value of accruals (measured as the difference between cash flow from operations and net income) divided by total assets. CURRENT denotes current assets as a proportion of total assets. FOREIGN is the ratio of foreign sales to total sales. The number of business segments is captured by SEG. LIAB represents total liabilities divided by total assets. LOSS, BIG_FIVE, OPINION and DEC are dummy variables that indicate whether or not the firm-year reported a loss, has one of the Big Five as its auditor, received an unqualified audit opinion and has a December year end respectively. Panel B presents descriptive statistics for variables used in regressions of stock market reaction to management forecast announcements. For inclusion in the analysis in this panel, firm-years should have audit fee data on audit analytics as well as should have management forecasts for that firm-year. CAR represents cumulative abnormal returns and ABVOL denotes abnormal trading volume in the 3-day window surrounding the management forecast date respectively. CAR is defined as the cumulative stock returns in excess of value-weighted market returns in event window (days -1 to +1) around the management forecast date (day 0), standardized by the standard deviation of excess returns during the non-announcement period (days -45 to -10). ABVOL is the difference in average log turnover in the event window and the average log turnover in the non-announcement period, standardized by the standard deviation of log turnover in the non-announcement period. MVE denotes the market value of equity in $millions. LEV indicates leverage defined as total debt divided by total assets. MB is the market-to-book ratio defined as the market value of assets divided by the book value of assets. ANALYST represents the number of financial analysts following the firm. RETSTD indicates stock return volatility and is measured as the standard deviation of daily returns. LIT indicates industries with high litigation risk (i.e., in SIC codes 2833-2836, 3570-3577, 3600-3674, 5200-5961, 7370-7374 and 8731-8734). 42 Table 1 (contd) Panel A: Audit fees sample Variables FREQUENCY PRECISION HORIZON AUDIT_FEES LN_FEES ASSETS ROA ACCR CURRENT FOREIGN SEG LIAB LOSS BIG_FIVE OPINION DEC Obs. 51,284 9,033 9,033 51,284 51,284 51,284 51,284 51,284 51,284 51,284 51,284 51,284 51,284 51,284 51,284 51,284 Mean 0.58 2.89 4.79 1.11 12.72 3,653.07 -0.34 0.29 0.48 0.26 4.57 0.81 0.40 0.70 0.57 0.73 Median 0.00 3.00 5.11 0.30 12.62 240.60 0.01 0.06 0.47 0.00 3.00 0.55 0.00 1.00 1.00 1.00 Std dev 1.47 0.64 0.86 2.48 1.53 14,350.15 1.68 1.12 0.27 0.39 4.17 1.77 0.49 0.46 0.50 0.44 Min 0.00 1.00 0.69 0.01 8.92 0.05 -19.58 0.00 0.00 0.00 1.00 0.03 0.00 0.00 0.00 0.00 Max 7.00 4.00 6.12 23.20 16.96 176,205.60 0.46 14.19 1.00 1.00 21.00 21.45 1.00 1.00 1.00 1.00 Panel B: Market reaction sample Variables CAR ABVOL MVE LEV MB ANALYST RETSTD LIT Obs. 27,088 27,088 27,088 27,088 27,088 27,088 27,088 27,088 Mean -0.001 1.109 7,934.41 0.214 2.009 9.015 0.107 0.356 Median 0.062 1.010 1,458.19 0.193 1.671 7.375 0.088 0.00 Std dev 4.309 1.145 26,860.73 0.182 1.127 6.805 0.069 0.479 Min -15.463 -1.554 0.033 0.00 0.626 0.000 0.025 0.00 Max 14.793 4.757 397,831.600 0.874 8.284 38.330 0.649 1.00 43 Table 2: Correlations This table presents correlations across variables employed in the audit fee regression (Panel A) and in regressions of stock market reactions to management forecasts (Panel B). FREQUENCY indicates the number of management forecasts made during the year. PRECISION captures the precision of the forecast. HORIZON denotes the time period between the date of the forecast and the date of the fiscal period end. The log of annual audit fees is indicated by LN_FEES. ASSETS represents total assets in $millions. ROA is net income divided by total assets. ACCR indicates the absolute value of accruals divided by total assets. CURRENT denotes current assets as a proportion of total assets. FOREIGN is the ratio of foreign sales to total sales. The number of business segments is captured by SEG. LIAB represents total liabilities divided by total assets. LOSS, BIG_FIVE, OPINION and DEC are dummy variables that indicate whether or not the firm incurred a loss, has one of the Big Five as its auditor, received an unqualified audit opinion and has a December year end respectively. EX_FEES is the residual of the regression of audit fees on firm-level determinants identified in equation (1). ABS_CAR and ABVOL measure the absolute value of the cumulative abnormal returns and abnormal trading volume respectively in the 3-day window surrounding the management forecast date. MVE denotes the log of market value of equity. LEV indicates leverage defined as total debt divided by total assets. MB is the market-to-book ratio defined as the market value of assets divided by the book value of assets. ANALYST represents the number of financial analysts following the firm. RETSTD indicates stock return volatility and is measured as the standard deviation of daily returns. LIT indicates industries with high litigation risk (i.e., in SIC codes 2833-2836, 3570-3577, 3600-3674, 5200-5961, 7370-7374 and 8731-8734). Detailed variable definitions are in Table 1. Correlations indicated by ***, **, * are significant at the 1%, 5% and 10% levels respectively. 44 Table 2 (contd) Panel A: Audit fees sample FREQUENCY PRECISION LN_ASSETS CURRENT HORIZON BIG_FIVE FOREIGN LN_FEES OPINION Variables ACCR LOSS LIAB DEC ROA SEG FREQUENCY PRECISION HORIZON LN_FEES LN_ASSETS ROA ACCR CURRENT FOREIGN SEG LIAB LOSS BIG_FIVE OPINION DEC 1.00 0.11*** 0.16 0.31 0.30 0.23 *** *** *** *** *** *** 1.00 0.06*** 0.08 0.03 0.08 *** *** *** *** * * 1.00 0.16*** 0.14 0.07 *** *** *** *** *** 1.00 0.80*** 0.33 *** *** *** 1.00 0.41*** -0.42 -0.31 0.20 0.15 0.16 *** *** 1.00 -0.38*** -0.09 0.14 0.13 *** *** *** *** *** -0.10 -0.07 0.18 0.14 -0.06 -0.09 -0.11 -0.06 -0.25 -0.22 0.34 0.17 0.09 1.00 -0.02*** -0.03 *** -0.02 -0.02 -0.04 1.00 0.05*** -0.17 -0.17 0.09 *** *** *** *** *** *** *** *** *** *** *** *** *** *** 1.00 0.20*** -0.14 -0.06 0.23 *** *** *** 0.01 0.12 *** *** 0.00 0.05 0.45 *** *** *** *** 1.00 -0.08*** -0.08 0.21 0.06 *** *** *** -0.06 -0.21 0.23 0.00 -0.11 *** -0.20 -0.82 0.24 0.12 1.00 0.04*** -0.10 -0.19 *** *** -0.12 -0.32 0.55 -0.47 0.54 *** *** 1.00 -0.21*** -0.09 *** *** *** 0.01 -0.06 *** -0.01 -0.06 *** *** *** *** *** *** *** -0.15 -0.13 -0.16 0.10 *** *** 1.00 0.01 0.06*** 1.00 -0.05*** 1.00 -0.04 -0.12 -0.03 *** -0.02 -0.07*** -0.04*** 0.06*** 0.09*** 0.13*** -0.03*** -0.05*** -0.10*** -0.04*** -0.02*** 0.09*** -0.01** 45 Table 2 (contd) Panel B: Market reaction sample Variables EX_FEES ABS_CAR ABVOL MVE LEV MB ANALYST RETSTD LIT EX_FEES 1.00 0.08*** 0.11 0.02 0.08 *** ABS_CAR 1.00 0.48*** 0.01 -0.05 0.03 0.03 *** ABVOL MVE LEV MB ANALYST RETSTD LIT 1.00 0.10*** -0.07 0.13 0.07 *** 0.18*** *** 1.00 0.18*** 0.30*** 0.72 *** 1.00 -0.29*** -0.01 -0.23*** -0.28 *** 0.09*** *** 0.05*** *** 0.12*** *** 1.00 0.28*** -0.04*** 0.22 *** 1.00 -0.25*** 0.16 *** -0.24*** -0.06 *** 0.02*** *** 0.00 *** -0.51*** -0.04 *** 1.00 0.27*** 1.00 46 Table 3: Relation between voluntary disclosures and audit fees The table presents estimates from the following regression: LN _ FEES i , t   0   1DISCLOSE i , t   2 LN _ ASSETS   3 ROAi , t   4 ACCRi , t   5CURRENT i , t   6 FOREIGN i , t   7 SEGMENTS i , t  8 LIAB i , t   9 LOSS i , t  10 BIG _ FIVE i , t  11OPINION i , t  12 DEC i , t   t where, DISCLOSE is the respective disclosure proxy (FREQUENCY, PRECISION or HORIZON), LN_FEES is log of annual audit fees. FREQUENCY indicates the number of management forecasts made during the year. PRECISION captures the precision of the forecast. HORIZON denotes the time period between the date of the forecast and the date of the fiscal period end. LN_ASSETS represents the log of total assets in $millions. ROA is net income divided by total assets. ACCR indicates the absolute value of accruals divided by total assets. CURRENT denotes current assets as a proportion of total assets. FOREIGN is the ratio of foreign sales to total sales. The number of business segments is captured by SEG. LIAB represents total liabilities divided by total assets. LOSS, BIG_FIVE, OPINION and DEC are dummy variables that indicate whether or not the firm incurred a loss, has one of the Big Five as its auditor, received an unqualified audit opinion and has a December year end respectively. The t-statistics are based on robust standard errors clustered by firm and by year. Detailed variable definitions are in Table 1. Pred. sign Intercept FREQUENCY PRECISION HORIZON LN_ASSETS ROA ACCR CURRENT FOREIGN SEG LIAB LOSS BIG_FIVE OPINION DEC Adj. R2 Observations + + + + – + + + + + + + – + FREQUENCY Coeff. 9.914 0.060 t-stat. 55.59 4.97 PRECISION Coeff. 9.541 0.122 t-stat. 39.02 3.36 HORIZON Coeff. 9.318 t-stat. 35.20 All Coeff. 9.100 0.065 0.087 t-stat. 42.61 4.48 2.80 6.78 26.33 -0.98 -1.98 3.08 11.19 0.14 1.57 1.34 -0.46 -1.83 0.97 0.123 0.445 -0.039 0.014 0.144 0.538 0.011 0.032 0.199 0.295 -0.323 -0.008 0.73 51,284 33.25 -3.81 1.29 1.57 11.49 0.99 4.61 5.20 2.39 -2.56 -0.12 0.502 -0.042 -0.158 0.312 0.590 0.001 0.175 0.054 -0.068 -0.296 0.078 0.64 9,033 27.61 -0.32 -1.36 2.74 10.87 0.07 1.29 0.77 -0.61 -1.91 0.69 0.499 -0.040 -0.137 0.323 0.597 0.000 0.153 0.054 -0.046 -0.292 0.074 0.64 9,033 6.91 28.58 -0.29 -1.10 2.92 10.85 0.03 1.07 0.77 -0.41 -1.87 0.66 0.087 0.479 -0.124 -0.234 0.320 0.600 0.001 0.204 0.100 -0.052 -0.267 0.103 0.65 9,033 47 Table 4: Relation between market reaction to disclosures and excess audit fees The table presents estimates from the following regressions: ABS _ CARi ,t  0   1EX _ FEESi ,t   2 MVEi ,t  3 LEVi ,t   4 MBi ,t  5 ANALYST,t  6 RETSTD,t  7 LITi ,t   t i i ABVOL,t  0   1EX _ FEESi ,t   2 MVEi ,t  3 LEVi ,t   4 MBi ,t  5 ANALYST,t  6 RETSTD,t  7 LITi ,t   t i i i CARi ,t  0  1SURPi ,t   2 SURP,t * EX _ FEESi ,t  3 EX _ FEESi ,t   4 MVEi ,t  5 LEVi ,t  6 MBi ,t  7 ANALYST,t i i  8 RETSTD,t  9 LITi ,t   t i Panel A presents the results for the first two regressions, while Panel B presents the results for the last regression. ABS_CAR and ABVOL measure the absolute value of the cumulative abnormal returns and abnormal trading volume respectively in the 3-day window surrounding the management forecast date. In Panel B, the dependent variable is the signed cumulative abnormal returns (CAR). CAR is defined as firm return minus value weighted market return in the 3-day window surrounding the management forecast date, standardized by stock’s return volatility in the preforecast-announcement period (i.e., days -45 to day -10 relative to the forecast announcement date). EX_FEES is excess audit fees defined as the residual of the regression of audit fees on firm-level determinants. MVE denotes the log of market value of equity. LEV indicates leverage defined as total debt divided by total assets. MB is the marketto-book ratio defined as the market value of assets divided by the book value of assets. ANALYST represents the number of financial analysts following the firm. RETSTD indicates stock return volatility and is measured as the standard deviation of daily returns. SURP denotes forecast surprise and is defined as the difference between the management forecast value and the consensus analyst forecast scaled by the pre-announcement period price. LIT indicates industries with high litigation risk (i.e., in SIC codes 2833-2836, 3570-3577, 3600-3674, 5200-5961, 73707374 and 8731-8734). In Panel B, the first set of columns uses the median forecast and the next uses the mean forecast. Detailed variable definitions are in Table 1. All regressions report t-statistics based on robust standard errors clustered by firm and by year. Panel A: Regression of absolute CAR and abnormal volume Pred. sign Intercept EX_FEES MVE LEV MB ANALYST RETSTD LIT Adj. R2 Observations ABS_CAR Coeff. 3.381 0.388 -0.100 -0.512 0.069 0.188 -0.080 0.060 0.01 27,088 t-stat. 15.19 6.36 -4.98 -3.07 2.56 4.03 -0.08 0.77 ABVOL Coeff. 0.683 0.148 0.003 -0.261 0.060 0.147 -0.049 0.088 0.04 27,088 t-stat. 6.32 5.19 0.27 -3.12 6.22 6.49 -0.11 2.57 + ? + + – ? ? 48 Table 4 (contd). Panel B: Regression of signed CAR on forecast news and excess audit fees Pred. sign Intercept SURP SURP*EX_FEES EX_FEES MVE LEV MB ANALYST RETSTD LIT Adj. R2 Observations Median forecast Coeff. -0.960 11.994 4.960 -0.105 0.143 -0.646 0.244 -0.341 -0.696 0.059 0.01 22,116 t-stat. -3.90 5.17 2.05 -1.56 5.31 -4.62 4.08 -6.08 -0.92 0.50 Mean forecast Coeff. -0.948 12.900 4.910 -0.095 0.138 -0.663 0.245 -0.333 -0.714 0.060 0.01 22,116 t-stat. -3.86 5.20 1.90 -1.40 5.20 -4.68 4.12 -6.05 -0.96 0.51 + + ? – + + – ? ? 49 Table 5: Role of litigation risk – High versus low litigation risk industries The table presents estimates from the following regression: LN _ FEES i , t   0   1DISCLOSE i , t   2 DISCLOSE i , t * LIT i , t   3 LIT i , t   4 LN _ ASSETS   5 ROAi , t   6 ACCRi , t   7CURRENT i , t   8 FOREIGN i , t   9 SEGMENTS i , t  10 LIAB i , t  11LOSS i , t  12 BIG _ FIVE i , t  13OPINION i , t  14 DEC i , t   t where, LIT denotes high litigation risk industries (i.e., in SIC codes 2833-2836, 3570-3577, 3600-3674, 5200-5961, 7370-7374 and 8731-8734) and the other variables are as defined in Table 3. The t-statistics are based on robust standard errors clustered by firm and by year Pred. sign Intercept FREQUENCY FREQUENCY*LIT PRECISION PRECISION*LIT HORIZON HORIZON*LIT LIT LN_ASSETS ROA ACCR CURRENT FOREIGN SEG LIAB LOSS BIG_FIVE OPINION DEC Adj. R 2 FREQUENCY Coeff. 9.876 t-stat. 57.42 6.44 -6.96 PRECISION Coeff. 9.474 t-stat. 39.26 HORIZON Coeff. 9.282 t-stat. 34.17 All Coeff. 9.045 0.083 -0.056 t-stat. 40.92 5.26 -4.28 3.37 -0.44 6.94 0.68 1.21 26.46 -0.67 -1.82 2.91 11.13 0.13 1.72 1.32 -0.48 -1.83 0.92 + + + 0.077 -0.058 0.139 -0.041 3.75 -1.08 0.127 -0.013 6.84 -0.47 0.82 27.99 -0.27 -1.24 2.71 10.75 0.09 1.21 0.69 -0.43 -1.87 0.71 0.096 -0.016 0.078 0.024 0.142 0.480 -0.086 -0.209 0.311 0.593 0.001 0.216 0.094 -0.053 -0.265 0.098 0.65 9,033 0.183 + – + + + + + + + – + 0.451 -0.035 0.018 0.099 0.516 0.011 0.036 0.175 0.269 -0.320 -0.006 0.73 51,284 4.18 35.69 -3.52 1.56 1.00 12.28 1.05 5.50 5.06 2.30 -2.58 -0.08 0.165 0.502 -0.036 -0.166 0.297 0.588 0.001 0.195 0.046 -0.069 -0.296 0.084 0.67 9,033 1.43 27.16 -0.28 -1.51 2.54 10.81 0.13 1.49 0.68 -0.62 -1.91 0.74 0.104 0.498 -0.036 -0.146 0.310 0.595 0.001 0.169 0.048 -0.049 -0.291 0.080 0.64 9,033 Observations 50 Table 6: Role of the litigation risk based on class-action lawsuits Panels A and B of the table presents estimates from the following regressions respectively: Pr(Law 1)  G( 0   1SIZEi ,t   2 RETi ,t   3 RETSTDi ,t   4TURNi ,t   5 MBi ,t   6CLIENT _ VOLi ,t  7 NBR _ CLIENTSi ,t  8 DOL _ AUDITi ,t   9OPINIONi ,t  10 LITi ,t   i ,t LN _ FEES i ,t   0   1DISCLOSE i ,t   2 DISCLOSE i ,t * LIT _ PRED i , t   2 RETi ,t   3 RETSTD i ,t   4TURN i ,t   3 LIT _ PRED i , t   4 LN _ ASSETS   5 ROAi , t   6 ACCRi ,t   7CURRENT i , t  8 FOREIGN i ,t   9 SEGMENTS i , t  10 LIAB i ,t  11LOSS i ,t  12 BIG _ FIVE i , t  13OPINION i ,t  14 DEC i , t   t where, LAW indicates a securities class action lawsuit during the year, SIZE indicates the log of market value of equity. RET is annual stock return. RETSTD indicates stock return volatility and is measured as the standard deviation of daily returns. TURN represents stock turnover. The market-to-book ratio is denoted by MB. CLIENT_VOL is the variance of the auditor’s client size. The number of clients audited by the auditor is captured by NBR_CLIENTS. DOL_AUDIT indicates the total sales dollars that the auditor audits. OPINION indicates an unqualified audit opinion. LIT represents industries with high litigation risk (i.e., in SIC codes 2833-2836, 35703577, 3600-3674, 5200-5961, 7370-7374 and 8731-8734), LIT_PRED denotes the predicted value of litigation risk from panel A. The other variables are as defined in Table 3. The t-statistics are based on robust standard errors clustered by firm and by year Panel A: Litigation prediction model Predicted sign Intercept SIZE RET RETSTD TURN MB CLIENT_VOL NBR_CLIENTS DOL_AUDIT OPINION LIT Pseudo Adj. R2 Observations + – + + + + ? ? ? + Probit Coeff. -4.286 0.198 -0.290 2.213 0.696 0.000 0.392 0.002 0.000 0.079 0.110 0.14 37,837 z-stat. -14.18 7.72 -5.74 4.70 5.77 0.05 2.87 3.07 -3.47 1.78 1.86 51 Table 6 (contd) Panel B: Relation between audit fees and voluntary disclosures interacted with litigation probability Sign FREQUENCY Coeff. Intercept FREQUENCY FREQUENCY*LIT_PRED PRECISION PRECISION*LIT_PRED HORIZON HORIZON*LIT_PRED LIT_PRED LN_ASSETS ROA ACCR CURRENT FOREIGN SEG LIAB LOSS BIG_FIVE OPINION DEC Adj. R2 Observations + ? + ? + ? ? + – + + + + + + + – + -1.585 0.484 -0.136 0.128 0.156 0.567 0.009 -0.031 0.197 0.219 -0.312 -0.008 0.67 37,837 -2.55 37.07 -4.14 1.82 1.17 11.52 0.89 -0.43 5.36 1.84 -2.08 -0.09 -1.834 0.556 -0.074 0.078 0.393 0.627 -0.001 -0.008 0.118 -0.128 -0.234 0.070 0.66 8,970 -1.57 36.01 -0.47 0.47 3.46 12.35 -0.12 -0.05 2.09 -1.36 -1.66 0.64 9.806 0.074 -0.505 t-stat. 38.59 6.59 -3.39 0.127 -0.784 2.92 -1.94 0.122 -0.352 -2.225 0.552 -0.068 0.088 0.403 0.632 -0.001 -0.027 0.123 -0.107 -0.230 0.071 0.66 8,970 5.49 -1.22 -1.42 36.12 -0.44 0.51 3.61 12.09 -0.13 -0.18 2.26 -1.13 -1.63 0.66 PRECISION Coeff. 9.356 t-stat. 36.08 HORIZON Coeff. 9.143 t-stat. 33.16 All Coeff. 8.918 0.075 -0.244 0.093 -0.686 0.081 -0.154 -0.467 0.531 -0.139 -0.009 0.397 0.635 0.000 0.034 0.164 -0.118 -0.208 0.097 0.67 8,970 t-stat. 35.52 5.13 -2.22 2.69 -2.36 6.19 -0.77 -0.25 34.83 -0.97 -0.06 3.77 12.63 -0.04 0.26 2.82 -1.24 -1.57 0.92 52 Table 7: Cross-sectional regressions The table presents results from cross-sectional regressions. The regression equations and variable definitions are given in Tables 3 and 4. The below regressions include only one observation per firm. For each firm, the variables are averaged over time and the cross-sectional regressions are estimated using these averages. The t-statistics are based on robust standard errors. Panel A: Relation between audit fees and management forecast properties. Pred. sign Intercept FREQUENCY PRECISION HORIZON LN_ASSETS ROA ACCR CURRENT FOREIGN SEG LIAB LOSS BIG_FIVE OPINION DEC Adj. R 2 FREQUENCY Coeff. 9.736 t-stat. 277.96 10.60 PRECISION Coeff. 9.790 0.029 t-stat. 99.53 1.65 HORIZON Coeff. 9.553 t-stat. 86.33 Coeff. All t-stat. 82.02 3.59 0.50 4.07 54.29 2.50 2.61 4.93 23.21 10.26 3.98 5.46 2.37 -10.01 1.67 9.542 0.028 0.009 + + + + – + + + + + + + – + 0.055 0.066 0.415 -0.012 0.060 0.102 0.605 0.040 0.025 0.328 0.509 -0.354 -0.018 0.80 10,233 101.42 -0.99 3.19 3.79 31.45 20.71 3.50 15.85 26.10 -15.88 -1.23 0.455 0.286 0.484 0.279 0.682 0.029 0.165 0.216 0.128 -0.495 0.024 0.79 2,322 56.91 2.43 2.62 4.77 22.93 10.28 3.65 4.50 2.34 -10.34 1.03 0.454 0.312 0.533 0.286 0.686 0.029 0.155 0.235 0.140 -0.492 0.023 0.79 2,322 5.17 57.41 2.52 2.65 4.88 23.30 10.36 3.45 4.97 2.60 -10.30 1.00 0.054 0.448 0.301 0.507 0.287 0.683 0.028 0.175 0.264 0.128 -0.477 0.040 0.79 2,322 Observations 53 Table 7 (contd) Panel B: Relation between absolute value of cumulative abnormal returns and excess audit fees Pred. sign Intercept EX_FEES MVE LEV MB ANALYST RETSTD LIT Adj. R2 Observations ABS_CAR Coeff. 2.996 0.324 -0.068 -0.438 0.077 0.306 -0.579 0.021 0.04 2,095 t-stat. 14.26 5.12 -1.86 -1.86 1.75 4.20 -1.01 0.26 ABVOL Coeff. 0.419 0.123 0.026 -0.282 0.074 0.181 0.119 0.044 0.16 2,095 t-stat. 5.14 5.10 1.87 -3.41 4.67 6.55 0.47 1.37 ? ? + + 54 Table 7 (contd) Panel C: Relation between cumulative abnormal returns, forecast surprise and excess audit fees Pred. sign Intercept SURP SURP*EX_FEES EX_FEES MVE LEV MB ANALYST RETSTD LIT Adj. R2 Observations Median forecast Coeff. -1.070 6.033 4.972 0.037 0.190 -0.501 0.197 -0.393 -1.848 0.130 0.06 1,745 t-stat. -3.61 3.45 2.01 0.33 4.19 -1.49 3.49 -4.44 -2.33 1.20 Mean forecast Coeff. -1.067 7.420 4.726 0.044 0.183 -0.511 0.201 -0.386 -1.813 0.133 0.07 1,745 t-stat. -3.59 4.05 1.69 0.39 4.07 -1.53 3.56 -4.38 -2.27 1.22 + + ? ? + + – ? ? 55 Table 8: Firm fixed-effects regressions The table presents the results from the regressions presented in Tables 3 and 4, but additionally control for firm-fixed effects. The firm-fixed effects regression is implemented by differencing from each variable its time-series mean, estimated across all years for that firm. The standard errors are clustered by firm and by year. Panel A: Relation between management forecast attributes and audit fees Sign FREQUENCY Coeff. Intercept FREQUENCY PRECISION HORIZON LN_ASSETS ROA ACCR CURRENT FOREIGN SEG LIAB LOSS BIG_FIVE OPINION DEC Adj. R 2 PRECISION Coeff. -0.015 0.095 t-stat. -0.13 3.43 HORIZON Coeff. -0.015 t-stat. -0.14 Coeff. -0.029 0.032 0.083 0.088 4.88 11.34 -0.51 -0.93 5.31 -0.18 -3.87 7.02 0.23 -3.38 -1.04 3.07 0.071 0.962 -0.063 -0.196 0.828 -0.012 -0.051 0.654 0.028 -0.259 -0.122 0.417 0.41 9,033 1.000 -0.054 -0.178 0.830 -0.022 -0.052 0.647 0.011 -0.260 -0.131 0.415 0.40 9,033 All t-stat. -0.27 2.05 3.54 5.49 10.58 -0.61 -1.04 5.36 -0.10 -3.82 7.37 0.53 -3.34 -0.99 3.17 t-stat. 0.02 2.32 0.002 + + + + – + + + + + + + – + 0.594 -0.071 -0.023 0.099 -0.012 -0.060 0.078 0.001 -0.113 -0.167 0.221 0.31 51,284 0.047 26.09 -4.97 -2.02 1.85 -0.12 -5.14 6.52 0.03 -1.93 -1.68 3.46 0.999 -0.055 -0.187 0.839 -0.014 -0.052 0.640 0.020 -0.268 -0.133 0.384 0.40 9,033 11.24 -0.53 -0.97 5.36 -0.11 -3.90 6.86 0.41 -3.51 -1.06 2.85 Observations 56 Table 8 (contd.) Panel B: Relation between market reaction to management forecasts and excess audit fees Pred. sign Intercept EX_FEES MVE LEV MB ANALYST RETSTD Adj. R2 Observations ABS_CAR Coeff. -0.252 0.501 0.175 -0.199 -0.115 0.384 1.107 0.01 27,088 t-stat. -0.42 6.98 3.80 -1.66 -2.19 3.45 0.93 ABVOL Coeff. -0.013 0.182 0.103 -0.065 -0.023 0.176 -0.412 0.02 27,088 t-stat. -0.66 6.41 6.85 -0.50 -1.19 5.20 -1.52 ? ? + + ? ? Panel C: Effect of excess audit fees on relation between cumulative abnormal returns and forecast surprises. Pred. sign Intercept SURP SURP*EX_FEES EX_FEES MVE LEV MB ANALYST RETSTD Adj. R2 Observations + + ? – + + – ? Median forecast Coeff. -0.038 22.374 9.617 -0.130 0.412 0.336 0.273 -0.856 -1.812 0.01 22,116 t-stat. -0.83 5.54 1.62 -2.12 3.91 1.22 5.73 -7.79 -1.79 Mean forecast Coeff. -0.037 24.369 10.762 -0.113 0.416 0.330 0.269 -0.844 -1.799 0.01 22,116 t-stat. -0.79 6.07 1.83 -1.83 3.90 1.20 5.81 -7.66 -1.78 57 Table 9: Relation between voluntary disclosures and BIG_FIVE The dependent variable is BIG_FIVE which denotes the presence of a Big Five auditor. FREQUENCY indicates the number of forecasts. LN_ASSETS represents the log of total assets in $millions. ROA is net income divided by total assets. LEV indicates the leverage ratio. ASSET_TURN indicates the asset-turnover ratio. MB is the market-to-book ratio and CAPEX denotes capital expenditures. Detailed variable definitions are in Table 1. Pr ( BIG _ FIVE  1)  G( 0  1FREQUENCY   2 LN _ ASSETS  3 ROA   4 LEV  5 ASSET _ TURN   6 MB   7CAPEX   ) Pred. sign 2-way clustering Coeff. t-stat. -7.81 7.41 24.80 -3.94 -5.69 3.19 -0.51 2.68 Intercept FREQUENCY LN_ASSETS ROA LEV ASSET_TURN MB CAPEX Pseudo R2 Observations + + + + -2.415 0.218 0.640 -0.150 -0.441 0.196 -0.003 1.324 0.30 48,020 58 Table 10: Effect of Sarbanes Oxley Act on Market Reactions to management forecast surprises The table presents estimates from the following regressions: ABS _ CARi , t   0   1POSTi , t   2 MVEi , t   3 LEVi , t   4 MBi , t   5 ANALYST, t i   6 RETSTD, t   7 LITi , t   t i ABVOL, t   0   1POSTi , t   2 MVEi , t   3 LEVi , t   4 MBi , t   5 ANALYST, t i i   6 RETSTD, t   7 LITi , t   t i CARi , t   0  1SURPi , t   2 SURP, t * POSTi , t   3 POSTi , t   4 MVEi , t   5 LEVi , t   6 MBi , t   7 ANALYST, t i i   8 RETSTD, t   9 LITi , t   t i POSTi,t takes the value 1 for forecasts made during the years 2004 to 2007, 0 otherwise. All other variables are as defined in Table 4. Forecasts made in the years 2002 and 2003 are excluded from the sample. All regressions report t-statistics based on robust standard errors clustered by firm and by year. Panel A: Regression of absolute CAR and abnormal volume Pred. sign Intercept POST MVE LEV MB ANALYST RETSTD LIT Adj. R2 Observations ABS_CAR Coeff. 1.688 1.451 -0.064 -0.432 0.061 0.204 3.067 -0.050 0.02 19,294 t-stat. 3.79 8.24 -2.81 -2.71 3.06 2.60 2.03 -0.70 ABVOL Coeff. -0.041 0.611 0.018 -0.252 0.048 0.164 1.266 0.045 0.05 19,294 t-stat. -0.23 7.89 1.43 -3.63 5.41 5.10 1.97 1.34 + ? + + – ? ? 59 Table 10 (contd). Panel B: Regression of signed CAR on forecast news and excess audit fees Pred. sign Intercept SURP SURP*POST POST MVE LEV MB ANALYST RETSTD LIT Adj. R 2 Median forecast Coeff. -1.180 5.273 12.750 -0.143 0.159 -0.727 0.251 -0.307 -0.741 0.088 0.02 16,411 t-stat. -5.89 10.66 4.70 -1.45 6.07 -4.22 3.21 -4.61 -0.64 0.58 Mean forecast Coeff. -1.173 5.953 13.501 -0.135 0.152 -0.742 0.251 -0.297 -0.698 0.088 0.02 16,411 t-stat. -6.33 14.01 5.11 -1.45 5.93 -4.28 3.25 -4.55 -0.63 0.58 + ? ? – + + – ? ? Observations 60 Figure 1: Relation between voluntary disclosure frequency and audit fees Panels A, B and C plot the average of LN_FEES across firm-years for firm-years grouped based on FREQUENCY, PRECISION and HORIZON respectively. The sample firm-years are first grouped based on the possible values for the discrete variables, FREQUENCY and PRECISION, and into equal quartiles for the continuous variable, HORIZON. The below figures plot the mean LN_FEES for each of the groups. Panel A: FREQUENCY and LN_FEES 61 Figure 1 (cotnd) Panel B: PRECISION and LN_FEES Panel C: HORIZON and LN_FEES 62 Figure 2: Relation between market reaction to disclosures and excess audit fees ABS_CAR and ABVOL measure absolute cumulative abnormal returns and abnormal trading volume in the 3-day window surrounding the management forecast date respectively. Excess audit fees (EX_FEES) is defined as the residual of the regression of audit fees on firm-level determinants. The x-axes plot deciles of EX_FEES while the yaxes plot mean values of ABS_CAR (in panel A) and ABVOL (in panel B). Panel A: Absolute cumulative abnormal returns (ABS_CAR) Panel B: Abnormal trading volume (ABVOL) 63