MANAGERIAL ABILITY AND EARNINGS QUALITY * Peter Demerjian Emory University. Melissa Lewis University of Utah. Baruch Lev New York University

Transcription

1 MANAGERIAL ABILITY AND EARNINGS QUALITY * Peter Demerjian Emory University Melissa Lewis University of Utah Baruch Lev New York University Sarah McVay University of Utah July 28, 2010 ABSTRACT We examine the relation between managerial ability and earnings quality. We find that earnings quality is positively associated with managerial ability. Specifically, more able managers are associated with fewer subsequent restatements, higher earnings and accruals persistence, lower errors in the bad debt provision, and higher quality accrual estimations. The results are consistent with the premise that managers can and do impact the quality of the judgments and estimates used to form earnings. Keywords: Managerial ability, managerial efficiency, earnings quality, accruals quality. Data Availability: Data is publicly available from the sources identified in the text. * We would like to thank Brian Cadman, Ilia Dichev, Weili Ge, Phil Shane, and workshop participants at the 2010 Kapnick Accounting Conference at the University of Michigan, the 2010 Western Region Meeting, Emory University, Florida International University, and the University of Utah for their helpful comments. An earlier version of this paper benefited from comments by Carol Anilowski, Asher Curtis, Patty Dechow, Paul Michaud, Venky Nagar, Larry Seiford, Ram Venkataraman, Norman White, and workshop participants at the 2006 AAA Annual Meeting, Harvard University, the University of California Berkeley, and the University of Indiana.

2 MANAGERIAL ABILITY AND EARNINGS QUALITY I. INTRODUCTION We examine the relation between managerial ability and earnings quality. We anticipate that superior managers are more knowledgeable of their business, leading to better judgments and estimates and thus higher quality earnings. 1 Alternatively, the benefit of higher quality earnings may not be sufficient to warrant the time and attention of skilled management, especially if the variance of feasible estimates is small, in which case we may not find an association between managerial ability and earnings quality. 2 While the empirical literature in the area of earnings quality has largely focused on firmspecific characteristics, such as firm size and board independence (Dechow and Dichev 2002; Klein 2002), we examine the manager-specific aspects of earnings quality. Our study is in the vein of Bertrand and Schoar (2003), who find that managers have an effect on firm choices such as acquisitions or research and development expenditures; Aier et al. (2005), who document that CFOs with more accounting expertise have fewer restatements; and Francis et al. (2008), who document that earnings quality varies inversely with CEO reputation. 3 Our main measure of managerial ability is the score developed in Demerjian et al. (2009), though we perform robustness checks using historical returns, media citations, and manager 1 Following Schipper and Vincent (2003), we consider high quality earnings to be those that are closest to true economic earnings. As they note, reported earnings will deviate from economic earnings because of recognition and measurement rules in U.S. GAAP as well as preparers implementation decisions, including management s judgments and estimates (see also Section 2 of Dechow et al. 2009). Our focus in this paper is on the quality of management s judgments and estimates, though we also consider the effect of GAAP rules on earnings quality when examining the Dechow and Dichev (2002) accruals quality measure in Section IV. 2 Costs to poor earnings quality include higher cost of capital (Francis et al. 2004) and economically significant negative price reactions to the announcement of earnings restatements (Palmrose et al. 2004). 3 Francis et al. (2008) measure CEO reputation with the number of articles mentioning the executive. They find that the number of news articles pertaining to the company s CEO and earnings quality based on the Dechow and Dichev (2002) accruals quality measure are negatively associated. We reconcile the results in this paper with the results in Francis et al. (2008) in Section V. Specifically, when we decompose accruals quality into the estimation component and other components related to real activities and other, unexplained, activities, we find that managerial ability is associated with a higher quality estimation component the component most associated with judgments and estimates made by management. 1

3 fixed effects (e.g., Fee and Hadlock 2003; Milbourn 2003; Francis et al. 2008; Dyreng et al. 2009). Demerjian et al. (2009) first estimate total firm efficiency, where efficient firms are those that generate more revenues for a given set of inputs. Total firm efficiency is influenced by both the manager (i.e., managers can, to varying degrees, predict future demand and understand industry trends) and the firm (i.e., managers in larger firms can negotiate better terms). Thus, Demerjian et al. (2009) then partition total firm efficiency between the firm and the manager, and verify that the component attributed to the manager is associated with a variety of characteristics, including managerial pay and the price reaction to management departures from the firm. 4 Prior research is limited to measures such as media coverage and historical returns, which are difficult to attribute solely to the manager versus the firm (e.g., Francis et al. 2008), or manager fixed effects, where there is evidence of a manager-specific effect, but the quantifiable effect is limited to managers who switch firms (e.g., Bertrand and Schoar 2003; Ge et al. 2008; DeJong and Ling 2009). The main ability measure we use in this study allows us to better separate the manager from the firm and retain an ordinal ranking of quality for a large sample of firms. 5 We expect a more able management team to be better able to estimate accruals accurately. For example, we expect more able managers to have better knowledge of their client base and macro-economic conditions when estimating bad debt expense, or to better assess the 4 As we will discuss in greater detail in the following sections, Demerjian et al. (2009) estimate total firm efficiency using data envelopment analysis, a type of frontier analysis that measures relative efficiency. They then remove identifiable firm characteristics, such as size, that affect the firm s relative efficiency but do not depend upon the quality of the management team. They attribute the unexplained portion of total firm efficiency to the management team. 5 In our setting, we would like to determine the quality of the CFO and their delegates, as we focus on the estimation of accruals, whereas CEOs are more focused on the overall strategy of the firm. Though we cannot disentangle the ability score by CEO and CFO, the ability score does encompass CFOs and their delegates, whereas media citations are, by definition, focused on the CEO. We also identify CFOs switching firms within our sample to correlate the CFOs scores from their old firms with the accruals quality after their arrival in their new firms (see Section V). 2

4 value of inventory (and thereby correctly identify obsolete inventory) based on their knowledge of industry trends and how these trends are expected to affect their firm. We consider four measures of earnings quality: the existence of an earnings restatement (Anderson and Yohn 2002), the persistence of earnings (Lipe 1990; Penman 2001), errors in the bad debt provision (McNichols and Wilson 1988), and the extent to which accruals map into cash flows (Dechow and Dichev 2002; Francis et al. 2008). We include controls for innate characteristics of firms that make it more or less difficult to estimate accruals (e.g., operating cycle and sales volatility) as well as structural characteristics shown to affect the quality of earnings (e.g., board independence and internal control quality). In general, we find that earnings quality is positively associated with managerial ability. This finding is consistent with the premise that capable managers are better able to estimate accruals, which results in a more precise measure of earnings. We contribute to both the earnings quality literature and the managerial ability literature. First, establishing a positive and significant relation between managerial ability and earnings quality suggests a means to improving earnings quality. Many of the factors associated with earnings quality, such as firm size, industry, or operating cycle, cannot be altered to change earnings quality. In contrast, attempts are often made to improve managerial quality. This finding is important for board members considering the costs and benefits of managers; managerial ability affects not only the operations of the firm, but also the quality of its reported earnings, and in turn, its share-price attributes and litigation exposure. 3

5 Second, we identify a real activities-related component of the often-used Dechow and Dichev (2002) accruals quality measure. 6 We find evidence consistent with the expectations of LaFond (2008) that firms undertaking certain real economic activities, such as R&D expenditures or M&As, have lower accruals quality because of these activities. This lower accruals quality is largely a byproduct of the accounting system, and generally does not reflect estimates or judgments. Therefore, it is difficult for management to mitigate this effect. Identifying and measuring this component should allow researchers to enhance their research designs to better identify the desired component of earnings quality (Dechow et al. 2009). In the next section, we develop our hypotheses with a review of the literature. In Section III we describe our sample, test variables, and descriptive statistics. In Section IV we present the main results, and in Section V we consider alternative ability measures and conduct a change analysis for a subset of managers in our sample who switch firms. In the final section we conclude the study. II. PRIOR RESEARCH AND HYPOTHESIS DEVELOPMENT Earnings quality is an important element of financial reports that affects the efficient allocation of resources. Earnings are the main input to investors and analysts valuation models, and firms with poor earnings quality tend to have higher cost of capital (e.g., Francis et al. 2004), while those experiencing restatements or SEC enforcement actions tend to experience an economically significant negative price reaction to the announcement (Feroz et al. 1991; Palmrose et al. 2004). 6 We use the term earnings quality to capture the general construct of higher quality reported earnings, while we use the term accruals quality to discuss the Dechow and Dichev (2002) estimate of earnings quality (the mapping of accruals to cash flows). 4

6 Following Schipper and Vincent (2003), we consider high quality earnings to be those that are closest to true economic earnings. Schipper and Vincent (2003) note that reported earnings deviate from economic earnings because of recognition and measurement rules in U.S. GAAP as well as preparers implementation decisions, including management s judgments and estimates. We expect individual managers to exhibit variation in their abilities to form accurate judgments and estimates. We expect managers with higher innate abilities to be more knowledgeable about the firm and the industry, as well as to be better able to synthesize information into logical forward-looking estimates to report higher quality earnings. Specifically, we expect accruals estimated by high-ability managers to be more accurate. For example, consider the allowance for bad debt estimate. A weaker manager might apply the historical rate of bad debt for the firm, while a stronger manager might adjust the historical rate by considering the macro-economic and industry trends, as well as changes in the firm s customer base. Thus, holding the firm constant, we expect a more able manager to report higher quality earnings. H1: Managerial ability is positively associated with earnings quality. It is possible that the bulk of the variation in earnings quality is driven by the accounting rules, in which case we will not find an association between the ability of managers and the quality of earnings. It is also possible that the benefits to the incremental improvement in earnings quality resulting from the intervention by an able manager (who may, for example, consider additional information) do not exceed the cost of that manager s time, in which case, 5

7 again, we will not find an association between the manager s ability and the firm s earnings quality. To date, the bulk of the literature on earnings quality has examined firm-specific characteristics. For example, Dechow and Dichev (2002), examining the mapping of accruals into cash flows, document that earnings quality is poorer for firms that are smaller, are experiencing losses, have greater sales and cash flow volatility, and have longer operating cycles. Each of these innate firm characteristics makes accruals more difficult to estimate. In addition to these innate characteristics, earnings quality has been found to vary with firm infrastructure. Klein (2002) finds that firms with more independent boards and audit committee members have higher quality accruals, consistent with stronger governance constraining earnings management. Doyle et al. (2007) find that earnings quality is poorer in firms that have weaker internal controls over financial reporting, where it is less likely that errors or intentional misstatements will be discovered and corrected, while Ashbaugh-Skaife et al. (2008) document an improvement in earnings quality following the remediation of internal control problems. 7 With respect to the effects of managers on the firm, Bertrand and Schoar (2003) document that managers have a real impact on the firms they manage that firm decisionmaking reflects the style of different managers. In a similar vein, both Ge et al. (2008) and DeJong and Ling (2009) examine manager fixed effects on certain financial reporting policies, and, similar to Bertrand and Schoar, document that individual managers matter: firms accounting and disclosure policies vary with manager fixed effects. 8 Ge et al. (2008) find that 7 Dechow and Dichev (2002) and Doyle et al. (2007) define higher earnings quality as when more accruals are realized as cash, while other studies, such as Klein (2002) and Ashbaugh-Skaife et al. (2008) assess earnings quality with the absolute value of discretionary accruals, where larger absolute discretionary accruals are deemed low quality. We discuss specific earnings quality measures in detail in Section III. 8 Also using fixed effects, Bamber et al. (2010) find that individual managers appear to have preferred styles that are associated with their propensity to issue guidance and the characteristics of the resulting guidance (e.g., the 6

8 CFO-specific factors play a significant role in explaining firms discretionary accruals, offbalance sheet activities, probability of accounting manipulations, financial reporting conservatism, earnings smoothness, and disclosure accuracy and bias. DeJong and Ling (2009) examine manager effects on accruals through both investment and accounting choices. As previously noted, this approach allows researchers to document a manager-specific effect, but it is constrained to managers who switch employers among the sample firms. Most closely related to our study, both Aier et al. (2005) and Francis et al. (2008) examine whether earnings quality varies with managerial characteristics. Aier et al. (2005) examine 456 firm-year observations over and document an association between CFO expertise (e.g., years worked as CFO, experience at another company, advanced degrees and professional certifications) and restatements; they find that firms with CFOs with greater expertise experience fewer restatements. Francis et al. (2008) examine the relation between earnings quality and CEO reputation, measured by the number of business press articles mentioning each CEO. The authors conduct their analysis for a sample of about 2,000 firm-year observations from the S&P 500 over and find a negative relation between CEO reputation and earnings quality. They conclude that boards of directors hire specific managers due to the reputation and expertise these individuals bring to managing the more complex and volatile operating environments of these firms. In other words, they suggest that the volatile operating environments or other innate characteristics of the firm are causing the lower earnings quality, not managerial actions. precision of the guidance), and Richardson et al. (2004) examine board member tastes using a sample of 885 firms with common directors in They find that board member fixed effects are associated with firms governance, financial, disclosure, and strategic policy choices. Finally, Dyreng et al. (2009) quantify the economic effect of specific managers on effective tax rates by comparing the relative fixed effects of managers on effective tax rates. 7

9 Finally, in the banking and insurance industry, respectively, Barr and Siems (1997) and Leverty and Grace (2005) find that managerial ability reduces the likelihood of insurer distress and the amount of time spent in distress. Both studies use data envelopment analysis (which is similar to the method we use to measure managerial ability in this paper) to determine the efficiency of the firm, based on industry-specific inputs and outputs, and they characterize superior managers as those who use inputs efficiently in the production process. In sum, there is mixed evidence on the impact of managers on earnings quality. Though there is some evidence that high quality managers reduce the likelihood of financial distress, and managers with greater expertise are associated with fewer earnings restatements, Francis et al. (2008) document that more reputable managers are associated with lower earnings quality. The latter association is consistent with some firms having low-quality earnings by the nature of their business (e.g., Dechow and Schrand 2004) and these firms hiring better managers. We posit that, after we control for the innate challenges affecting the firm, more able managers will be associated with higher quality earnings, as their judgments and estimations are more informed and accurate. We consider four earnings quality measures. The first is earnings restatements, which are de facto evidence of inaccurate earnings, which offers the least ambiguous proxy for earnings quality. The second is earnings persistence, a measure of the sustainability of earnings; we partition earnings into accrual and cash flow components to examine accruals persistence more directly. Our third earnings quality measure is the accuracy of a specific accrual: the bad debt provision (McNichols and Wilson 1988). Finally, we examine the mapping of working capital accruals into cash from operations, based on Dechow and Dichev (2002). Essentially, if an accrual does not become cash, that accrual is of poor quality. We expand this model by 8

10 partitioning accruals quality into three components (estimation difficulty, real operating activities, and the unexplained portion). Each of these measures is affected by both intentional and unintentional errors, and more able managers may be more likely to introduce intentional errors, either to signal their private information about the firm or to extract perquisites from the firm and the shareholders. Because we examine a broad sample of firms across many years, our sample firms are not expected to have incentives to misreport, on average, and thus our focus is on conscientious judgments and estimates made by management. 9 III. DATA, VARIABLE DEFINITIONS, AND DESCRIPTIVE STATISTICS We obtain our data from the 2008 Annual Compustat File (to estimate our main measure of managerial ability and calculate the bulk of our earnings quality variables and controls), CRSP (to form historical returns, an alternate managerial ability measure), Execucomp (to track CFOs across firms), IRRC (to obtain board independence data), and Audit Analytics (for recent years of restatements and internal control opinions). We also obtain several datasets made available by researchers to obtain media citations (from Baik et al and Francis et al. 2008), restatements (from Hennes et al. 2008) and internal control quality data (from Doyle et al. 2007). Our main sample includes all firms with available data to calculate managerial ability and at least one of our earnings quality variables, resulting in a maximum of 86,303 firm-year observations from The period begins with 1989 because 1988 is the first year for which firms widely reported cash flow statements, and the Dechow and Dichev (2002) earnings 9 In the event of earnings management, however, we expect more able managers to be better able to manage earnings successfully, for example, accelerating sales only if they know there will be sufficient sales in the next period to cover the acceleration, thereby avoiding large accruals reversals and restatements. We leave a more direct examination of the interaction between managerial ability and intentional earnings management for future research. 9

11 quality variable requires one year of historical cash flow data. The sample ends in 2007, as our earnings quality variables (described in the following section) require at least one year of future realizations. Variable Definitions Managerial Ability Measure We estimate managerial ability following Demerjian et al. (2009). Their measure of managerial ability generates an estimate of how efficiently managers use their firms resources. All firms use common resources capital, labor, innovative assets to generate output: revenues and earnings. High quality managers apply superior business systems and processes (supply chains, compensation systems) to generate a higher rate of output from given inputs than lower quality managers. Following Demerjian et al. (2009), we implement data envelopment analysis (DEA) within industries, comparing the sales generated by each firm conditional on the inputs used by the firm (Cost of Goods Sold, Selling and Administrative Expenses, Net PP&E, Net Operating Leases, Net Research and Development, Purchased Goodwill, and Other Intangible Assets). 10 Our measured resources thus reflect assets (tangible and intangible), innovative capital (R&D), and other inputs that are not reported separately in the financial statement (labor, consulting services) but whose costs are included in cost of sales and SG&A. We provide the motivation for, and definition of, each of these variables in the Appendix. For our implementation of DEA, we solve the following optimization problem: 10 DEA is a form of frontier analysis. DEA calculates efficiency as the ratio of weighted outputs to weighted inputs. Unlike other measures of efficiency (e.g., ROA or ROE), DEA does not require an explicit set of weights. Rather, DEA uses an optimization program to determine the firm-specific optimal weights (termed implicit weights ) on the inputs and outputs. The implicit weights capture the efficiency of the firm based on the selected inputs and outputs, allowing the optimal mix of inputs and outputs to vary by firm. 10

12 & & The optimization finds the firm-specific vector of optimal weights on the seven inputs, v, by comparing each of the input choices of the firm under study to those of the other firms in its estimation group. 11 The efficiency measure that DEA produces, θ, can take a value between zero and one (due to constraints in the optimization program). Observations with a value of one are most efficient; the set of firms with efficiency equal to one trace a frontier through the efficient set of possible input combinations. Observations with efficiency measures less than one fall below the frontier. Their score indicates the degree to which they are inefficient. For example, a firm with an efficiency score of 0.85 would need to reduce the costs (in some combination) by about 15 percent to achieve efficiency. The efficiency measure generated by the DEA estimation is attributable to both the firm and the manager, similar to other measures of managerial ability such as historical returns and media coverage. For example, a more able manager will be better able to predict trends, regardless of the size of the firm, while a manager in a larger firm will be better able to negotiate terms with suppliers, regardless of his or her quality. Accordingly, in the second stage of estimating managerial ability, Demerjian et al. (2009) purge the first-stage estimate of total firm efficiency from known efficiency drivers unrelated to current managers. This is done in Equation (1), where the first-stage efficiency estimate is regressed on firm size, market share, free cash flows, number of segments, and an indicator for global operations, by year and industry Since sales is the only output, its weight is standardized to one across observations. For the general DEA model, please see the Appendix. 12 To the extent that managers also affect some of the independent variables in Equation (1), such as free cash flows, the final managerial ability score is a conservative (understated) measure of managerial efficiency. 11

13 Firm Efficiency i = β 0 + β 1 Ln(Total Assets i ) + β 2 Market Share i + β 3 Positive Free Cash Flow i + β 4 Ln(Segments i ) + β 5 Foreign Currency Indicator i ) + Year i + ε i (1) The estimation is intended to abstract away from firm-specific benefits or challenges, where larger firms with more market share and cash are expected to be able to generate more sales for a given level of inputs, and firms with additional complexities, such as multiple segments or foreign operations, are expected to generate fewer sales for a given level of inputs. The residual from the estimation is our main measure of managerial ability. Demerjian et al. (2009) conduct a series of validity tests on this measure, documenting that more able managers are paid more and generate higher returns, and that this measure outperforms alternative ability measures, such as media citations and historical returns, in explaining stock price reactions to managerial turnovers. In other words, they document that the departure of a more (less) efficient manager is associated with a negative (positive) price reaction to the turnover announcement. Nonetheless, we also consider media citations, historical stock returns, and manager fixed effects in Section V. We decile rank Managerial Ability by year and industry to make the score more comparable across time and industries and to mitigate the influence of extreme observations. Results are similar using a continuous variable (results not tabulated). We have also added firm fixed effects to Equation (1). Although this reduces the comparability across firms (see Demerjian et al. 2009), it mitigates the concern that unidentified firm characteristics are driving the association. Results are similar using this alternate specification of managerial ability (not tabulated). Earnings Quality Measures Overview As noted in the review of earnings quality by Dechow et al. (2009), among others, a multitude of earnings quality measures are used in the literature, including persistence, abnormal 12

14 accruals, earnings smoothness, asymmetric timeliness and timely loss recognition, benchmarking (in which just meeting a benchmark is viewed as lower quality), earnings response coefficients, AAERs, restatements and internal control procedure deficiencies (see their page 3). Our goal in this study is to examine the impact of managers on accruals, so we select earnings restatements, earnings persistence, errors in the bad debt provision, and the mapping of accruals into cash flows as our four measures of earnings quality. We select these measures because increased correspondence between accruals and the associated economic activity likely reduces earnings restatements, increases earnings persistence, and lowers the likelihood of errors in accruals. We expect that better managers are able to report accruals that more closely correspond to the underlying economic activity; thus we expect the earnings quality metrics that are impacted by accrual estimation to vary with managerial ability. We discuss each earnings quality metric in greater detail in Section IV. We eliminate the absolute value of discretionary accruals, earnings smoothness and benchmarking, as the relation between improved accruals estimation and these metrics is not clear. 13 We do not consider timely loss recognition, as this is largely an artifact of the accounting system (Dechow et al. 2009, p. 13), and it is not clear whether more or less timely loss recognition is more associated with the underlying economics of the firm, the focus of our analysis. As noted in Dechow et al. (2009), ERCs are a poor measure of earnings quality because much of the earnings information can be voluntarily disclosed prior to the earnings announcement. Finally, of the three external indicators of earnings quality restatements, AAERs and internal control disclosures we consider only restatements. We do not consider 13 For example, abnormally high accruals may be high quality accruals that are associated with cash flows, while abnormally low accruals may simply reflect extremely negative performance, which also reflects the underlying economics of the firm. Neither of these abnormal accruals provides information on the manager s ability to appropriately estimate accruals, as the measure does not incorporate ex post realizations. 13

15 AAERs due to data constraints (the data is not readily available in machine readable form) and because these tend to be more fraudulent than basic errors in estimation (Hennes et al. 2008). Furthermore, we do not consider internal control deficiencies as an outcome because the determinants of internal control problems are largely firm-specific, such as having adequate resources to establish and maintain these controls. The role of an able manager in the determination of strong internal controls is less clear, and does not speak to management s ability to estimate accruals. 14 Control Variables Our main set of control variables is based on the firm-specific determinants of earnings quality noted in Dechow and Dichev (2002) and Hribar and Nichols (2007): firm size, proportion of losses, sales volatility, cash flow volatility, and operating cycle. 15 Firm Size is the log of total assets of the firm at the end of year t (XFN = AT). The remainder of the controls are estimated over at least three out of five years, from year t 4 through year t (except when paired with Aggregate AQ, when they are estimated over the same estimation period as Aggregate AQ). Loss Proportion is the ratio of the number of years of losses (IBC < 0); Sales Volatility is the standard deviation of sales scaled by average assets σ(xfn = SALE / XFN = AT); Cash Flow Volatility is the standard deviation of cash from operations scaled by average assets σ(xfn = OANCF / XFN = AT); and Operating Cycle is the log of the average of [(Sales/360) / (Average 14 Future research might consider, however, whether more able managers are better able to sustain high quality earnings in the face of weak internal controls. 15 Dechow and Dichev (2002) also consider earnings volatility and the magnitude of working capital accruals. However, they consider each determinant individually, while our goal is to estimate a multivariate regression. Including earnings volatility along with the other control variables in our regressions introduces a relatively high degree of multicollinearity. Thus, we do not include earnings volatility or the magnitude of working capital accruals in our tabulated results. 14

16 Accounts Receivable) + (Cost of Goods Sold/360) / Average Inventory)], averaged over years t 4 to t. 16 We also consider two infrastructure-related control variables that have been shown to be associated with earnings quality governance and internal control quality. With respect to governance, Klein (2002) finds that audit committee independence and board independence are negatively associated with the absolute value of abnormal accruals. We consider the percentage of independent board members, obtained from IRRC from , ranked by year and industry (Board Independence). With respect to internal control quality, Doyle et al. (2007) find that earnings quality is relatively poor in firms with internal control problems. We proxy for internal control quality with the disclosure of material weaknesses in internal control, where firms reporting material weaknesses are considered to have poor internal control quality. We obtain internal control data from Doyle et al. (2007) for and from Audit Analytics from As we have data for only the last few years of our study for each of these variables, we do not include the variables in our main analysis, but rather consider them in supplemental analyses. Descriptive Statistics We present descriptive statistics in Table 1; for each of our transformed variables (MgrlAbility, Historical Ret, Media Count, Firm-Specific Earnings Persistence, Aggregate AQ, Annual AQ, Firm Size, and Operating Cycle), we present the untransformed variable for ease of interpretation. Managerial ability has a mean and median close to zero, by construction, as this is a residual from Equation (1). The five-year historical return has a mean of approximately Specifically, operating cycle is the log of the average of (SALE/360) / Average RECT) + (COGS/360) / Average INVT). 17 The Doyle et al. (2007) data is available at 15

17 percent, and on average, CEOs are cited by the media approximately 40 times a year (214 times over five years). Approximately 11 percent of firms experience a restatement in the next three years, and firm-specific earnings persistence is approximately 0.33, on average. The error in the provision for bad debt as a percentage of sales (BDE Error) has a mean (median) of (0.006). Mean (median) Aggregate AQ is 0.03 ( 0.02), similar to that in Francis et al. (2005) and Dechow and Dichev (2002) (we have multiplied the standard deviation by negative one). Interestingly, Annual AQ appears to have a very similar distribution to the four-year standard deviation, with a mean of 0.05 and a similar median and first and third quartile to the aggregated variables. We compare these two measures further in Section IV. Our control variables are consistent with those in the literature. For example, the average firm has $1,139 million in assets, and 38 percent of the firm-years examined experience a loss. In Panel B of Table 1 we partition our main earnings quality measures by managerial ability, where low-quality (high-quality) managers are those in the bottom (top) quartile of managerial ability, where quartiles are formed by industry-year. Historical returns are significantly larger among high quality managers, consistent with Fee and Hadlock (2003) and Demerjian et al. (2009), though media counts are significantly lower for managers with higher ability. 18 We explore the relations between these ability measures more completely in Section V. Restatements are more prevalent among low-quality managers, firm-specific earnings persistence is significantly higher among high-quality managers, and errors in the provision for bad debt are larger among low-quality managers, providing initial support for our hypothesis. We do not, however, find consistent evidence when examining earnings quality based on the Dechow and Dichev (2002) measure; we explore this further in our multivariate analysis. 18 Note that Demerjian et al. (2009) document a U shaped relation between their ability score and media citations. Specifically, they note that the best and worst managers have a greater number of media citations. 16

18 We present a correlation matrix in Table 2. In general, the results support our ability measure: better managers are negatively correlated with losses and positively correlated with future earnings. Managerial ability is negatively correlated with restatements and errors in the provision for bad debt, and positively correlated with firm-specific earnings persistence, consistent with the univariate analysis presented in Table 1, Panel B, though both of the Dechow and Dichev (2002) accruals quality measures are negatively associated with managerial ability, consistent with Francis et al. (2008). Because aggregate accruals quality is positively correlated with restatements, we investigate the attributes of these earnings quality measures when testing our hypothesis in a multivariate setting. IV. TEST DESIGN AND RESULTS Earnings Restatements The first earnings quality measure we consider is earnings restatements, which are ex post evidence of erroneous earnings and thus have been used as a signal of poor earnings quality (e.g., Anderson and Yohn 2002; Aier et al. 2005; Doyle et al. 2007; Dechow et al. 2009). Though restatements can occur for reasons other than errors in accrual estimation (our focus), this earnings quality measure is the least reliant on an estimation procedure and thus provides a relatively unambiguous signal of earnings quality. Moreover, we do expect restatements to be associated with errors in accrual estimation, as most restatements impact an accrual account (Palmrose and Shultz 2004). We use the restatement data from Hennes et al. (2008) for restatements from and from Audit Analytics from Hennes et al. (2008) form their restatement sample 17

19 from the General Accounting Office (GAO) report of restatements. 19 Restate is an indicator variable that is equal to one if the firm announces a restatement in year t, t+1, or t+2, and zero otherwise; it is available from Our subsample of restatement firm-years excludes restatements classified as duplicate restatements per Hennes et al. (2008). To determine whether managerial ability varies with earnings restatements, we estimate the following equation using a pooled logistic regression: Restatement t = β 0 + β 1 MgrlAbility t + β 2 Firm Size t + β 3 Sales Volatilty t + β 4 Cash Flow Volatility t + β 5 Operating Cycle t + β 6 Losses t + β 7 Earnings t + ε t (2) We include each of the control variables discussed above; we also include earnings, as firm profitability has been shown to be associated with restatements (e.g., Kinney and McDaniel 1989). Because our tests rely on panel data, standard errors may be correlated within years and across time by firm. Thus, we cluster our standard errors by firm and year following Petersen (2009). 20 Results are presented in Table 3. As in our univariate analysis, we document a negative relation between managerial ability and restatements, supporting our hypothesis that more able managers are associated with higher quality earnings. The more efficient the manager, the less likely the firm is to restate (β 1 = 0.32; p < 0.01). The marginal effect is 2.9 percent (not tabulated). Given that the unconditional likelihood of having a restatement is 11 percent, this 19 We thank the authors for this dataset, which is available at In addition to the restatement indicator variable used in our main tests, we utilize the partitioning of restatements into irregularities (intentional misstatements) and errors (unintentional misstatements) by Hennes et al. (2008) to determine if managerial ability differentially impacts these two types of restatements. We find that ability is associated with both types of restatements (not tabulated). 20 We do not tabulate results from the estimation of Equation (2) with firm fixed effects because the sample size is reduced severely for firm fixed effects models with binary dependent variables, as the model can be estimated only for firms with variation in the dependent variable (i.e., firms that have restated at some point during our sample). Nonetheless, when we estimate Equation (2) using firm fixed effects, we continue to observe a significant and negative relation between managerial ability and restatements. 18

20 indicates that moving from the worst to the best decile of managerial ability reduces the odds of having a restatement by about one-third. We re-estimate Equation (2) after adding board independence and internal control weakness (ICW) to control for the known infrastructure effects of these variables (not tabulated). We continue to observe a significant and negative coefficient on managerial ability, consistent with our hypothesis. The coefficient on ICW is significant and positive, consistent with Doyle et al. (2007), while the coefficient on board independence does not differ significantly from zero, consistent with Agrawal and Chadha (2005). Earnings Persistence Our second measure of earnings quality is earnings persistence, which is frequently discussed as a measure of earnings quality (e.g., Penman 2001, 623; Revsine et al. 2002, 245). Though clearly earnings persistence will vary with the underlying economics of the individual firm, such as earnings volatility, we expect more able managers to report the highest quality earnings after we control for these innate effects on persistence. We calculate earnings as earnings before extraordinary items (Xpressfeed variable name (hereafter XFN ) XFN=IBC) scaled by average total assets (XFN=AT), and estimate earnings persistence using the following model: Earnings t+1 = β 0 + β 1 Earnings t + β 2 Earnings t MgrlAbility t + β 3 Earnings t NegEarn t + β 4 Earnings t MgrlAbility t NegEarn t + β 5 MgrlAbility t + β 6 NegEarn t + + NegEarn t ) +ε t (3) The coefficient β 1 is earnings persistence. 21 We allow the coefficient on earnings to vary with managerial ability and expect earnings persistence to increase with managerial ability for 21 Though we consider firm-specific earnings persistence for descriptive purposes in Tables 1 and 2, we examine earnings persistence in the cross-section to formally test our hypothesis, as we are interested in how earnings persistence varies with managerial ability, which is measured annually. 19

21 firms with non-negative earnings (β 2 > 0). Because earnings persistence is not desirable for loss firms, we include a negative earnings indicator variable (NegEarn) in Equation (3) and interact NegEarn with all of the variables in the model. We expect earnings persistence among loss firms to be lower in the presence of better managers (β 4 < 0), as more able managers are expected to return their firms to profitability more quickly (e.g., Barr and Siems 1997; Leverty and Grace 2005). Recent work suggests that performance metrics that are more smooth or persistent are less value relevant because they conceal underlying changes in cash flows (Barton et al. 2010). We posit that higher earnings persistence is a sign of better earnings quality when driven by improved accruals estimation that strengthens the relation between reported earnings and cash flows. Thus, we next partition Earnings in year t into Accruals and Cash from Operations (CFO). This allows us to determine if any differential earnings persistence in Equation (3) is, at least in part, due to more persistent accruals. Earnings t+1 = α 0 + α 1 Accruals t + α 2 Accruals t MgrlAbility t + α 3 Accruals t NegEarn t + α 4 Accruals t MgrlAbility t NegEarn t +α 5 CFO t MgrlAbility t + α 6 CFO t NegEarn t + α 7 CFO t MgrlAbility t NegEarn t + α 8 MgrlAbility t + α 9 NegEarn t + α + α NegEarn t ) +ε t (4) In Equation (4) we again include NegEarn and interact NegEarn with all of the variables in the model. After partitioning earnings in year t into the accrual and cash flow components, we expect more able managers to be better able to estimate accruals, and thus expect these accruals to be of higher quality (i.e., more persistent). Thus, in Equation (4), we expect to observe a positive coefficient on Accruals MgrlAbility (α 2 > 0). We present the estimates from these two estimations in Table 4. Referring to the first column of estimates, which does not include managerial ability, we see that earnings have a core 20

22 persistence of 0.72, along the lines of prior research. Turning to the second column of estimates, which includes managerial ability, the base persistence is lower at 0.45, and this persistence is increasing with managerial ability. Positive earnings persistence is expected to increase from 0.45 to 0.83 ( ) when moving from the lowest to the highest decile of managerial ability. This higher persistence is clearly economically significant and supports our hypothesis. Also as predicted, negative earnings are less persistent when reported by higher ability managers (Earnings MgrlAbility NegEarn = 0.24, p = 0.001). When we partition earnings into accruals and cash flows, managerial ability increases the persistence of both components when firms report positive earnings. The accruals reported by positive earnings firms have a base persistence of The incremental coefficient on accruals for firms with higher ability managers is 0.26 (p = 0.002), suggesting that the increased earnings persistence of firms with higher ability managers, in part, is due to more persistent accruals. These findings support our hypothesis that higher quality managers are better able to estimate accruals, resulting in higher earnings quality. 22 In our second set of estimations, we include firm fixed effects. Results are similar to those without fixed effects. McNichols and Wilson (1988) Error in the Provision for Bad Debt We examine a specific accrual as our third measure of earnings quality: the provision for bad debt, modeled in McNichols and Wilson (1988), assuming a balance sheet perspective to estimating bad debt, adherence to GAAP, and perfect foresight of future write-offs, as follows: Bad Debt Expense t = β 0 +β 1 Allowance for Doubtful Accounts t 1 + β 2 Write-offs t + β 3 Write-offs t+1 + φ t (5) 22 That more able managers are also associated with more persistent cash flows is consistent with these managers making more efficient operating decisions. 21

23 where Bad Debt Expense and Write-offs are hand-collected from the firm s SEC filings and Allowance for Doubtful Accounts is available from Xpressfeed (XFN = RECD). All variables are deflated by sales in year t. The above model expects bad debt expense to increase by the amount of current period write-offs that exceed the beginning balance in allowance for doubtful accounts, plus the write-offs anticipated in the coming year. The error (φ t ) has two components: a discretionary earnings management component, and a forecast error component. McNichols and Wilson (1988) investigate subsets of firms with specific earnings management incentives and thus assume that errors in accrual estimates average to zero. In contrast, we assume that, on average, the error due to earnings management is near zero for the full sample, but we expect errors in accrual estimates to vary with managerial ability. We expect φ t to decrease with managerial ability. As illustrated in Equation (5), managers must estimate accounts receivable write-offs they expect to occur in year t+1. We expect that that more able managers judgments and estimates regarding anticipated write-offs will be more accurate (compared to realized write-offs in year t+1, which are included in Equation (5)) than those of less able managers. Because the data for this analysis must be hand-collected from SEC filings, we limit the analysis to firms with managers in the top or bottom quintiles of managerial ability and in industries where accounts receivable (relative to assets) and bad debt expense (relative to earnings) are large. Following McNichols and Wilson (1988), we consider firms from three industries: (1) printing and publishing, (2) nondurable wholesale goods, and (3) business services, and we estimate: BDE Error t = β 0 + β 1 High Ability Indicator t + β 2 Firm Size t + β 3 Losses t + β 4 Sales Volatilty t + β 5 Cash Flow Volatility t + β 6 Operating Cycle t + ε t (6) 22

24 where BDE Error is the absolute value of the residual from Equation (5), and High Ability Indicator is an indicator variable that is equal to one (zero) if the managerial ability score in year t is in the top (bottom) quintile relative to industry-year peers. A negative coefficient on High Ability Indicator is consistent with more able managers forming better estimates of bad debt provisions. Results are presented in Table 5. In support of our hypothesis, β 1 is (p = 0.05). Managers with higher ability scores produce higher quality bad debt provisions. 23 The Dechow and Dichev (2002) Measure of Earnings Quality The Relation between Accruals Quality and Managerial Ability Our fourth and final measure of earnings quality follows Dechow and Dichev (2002), who posit that high quality accruals are eventually realized as cash flows. Incorrectly estimated accruals are less likely to be realized as cash flows. We hypothesize that the better managers know their business, the less likely they are to have erroneous accruals. We determine how well a firm s accruals map into cash flows by estimating the following regression by industry-year. WC t = β 0 + β 1 CFO t 1 + β 2 CFO t + β 3 CFO t+1 + β 4 REV t + β 5 PPE t + ε t (7) The residual from the regression measures the extent to which current accruals ( WC) map into past, present, or future cash flows (CFO), with smaller residuals (in absolute terms) indicating superior mapping. 24 Following both McNichols (2002) and Francis et al. (2005), we include the current year change in sales ( REV; XFN = SALE) and the current year level of 23 Because our sample selection procedures for this analysis result in a small sample of firm-years that do not necessarily contain the same firm over multiple years (i.e., we do not have panel data), we do not cluster standard errors by firm and year, nor do we estimate a firm fixed-effects specification. In untabulated results, however, we continue to observe the negative relation between managerial ability and BDE Error when we cluster standard errors by year or include year fixed effects. 24 We define the change in working capital from year t 1 to t as WC = Accounts Receivable + Inventory Accounts Payable Taxes Payable + Other Assets, or WC = (RECCH + INVCH + APALCH + TXACH + AOLOCH). CFO is cash flow from operations (OANCF). We use information from the statement of cash flows, rather than the balance sheet, to estimate current accruals because the balance sheet approach can lead to noisy estimates (Hribar and Collins 2002). All variables in Equation (7) are scaled by average total assets (AT) and winsorized at the 1 st and 99 th percentiles, by year. 23