CEO Tenure and Earnings Quality

CEO Tenure and Earnings Quality Weining Zhang School of Management University of Texas at Dallas Email: wxz041000@utdallas.edu December 30 th, 2009 Abstract This study investigates the relation between CEO tenure and earnings quality. I find that, on average, CEOs with long tenures report earnings less aggressively than CEOs with short tenures, both in terms of recognizing economic losses in a timelier manner and reporting lower discretionary accruals. These results are consistent with the notion that at the beginning of their tenures, in order to build reputation of ability, CEOs have incentive to inflate earnings; after the CEOs have established their reputations through their long tenures, they report less aggressively in order to protect their reputations. I also find that long-tenured CEOs report more aggressively in their final year of tenure than in the years leading up to the final year of tenure. These results suggest that just prior to the completion of their tenures, long-tenured CEOs are not as concerned about their reputations because the aggressive reporting may be difficult for investors to detect until the firm s future realizations become known. By then, these CEOs leave their firms.

1. Introduction Fama (1980), Diamond (1989), Holmstrom (1999) and Milbourn (2003) define a CEO s reputation essentially as the market s perception of her ability. CEO reputation is a valuable asset that is associated with many long-term benefits, such as better future compensation, reappointment in the position, and greater managerial autonomy (see, e.g., Fama, 1980; Gibbons and Murphy, 1992; Hermalin and Weisbach, 1998; Ryan et al. 2009). Thus, CEOs have strong incentives to build and protect their reputations. In a survey paper, Graham et al. (2005) note that managers reputational concerns significantly affect their financial reporting decisions. In this study, I empirically investigate how CEOs tendency of building and protecting reputation of ability affects how aggressively earnings are reported. 1 At the beginning of tenure, since a CEO s ability is not well known to the market, the market relies primarily on the CEO s current performance to perceive her ability (see, e.g., Hermalin and Weisbach, 2009). Moreover, Oyer (2008) and Axelson and Bond (2009) argue that at the beginning of her career, if a manager loses a high-profile job due to unsatisfactory performance, even though it may be due to bad market conditions, the market perceives the manager as having low ability, and she suffers significant disadvantages for the rest of her career. Thus, at the beginning of tenure, CEOs have the incentive to report good performance in order to avoid being labeled as low ability CEOs. Accordingly, I predict that at the beginning of their tenure, CEOs tend to inflate reported earnings. 1 Though prior studies, such as Fama (1980), Diamond (1989), Holmstrom (1999) and Milbourn (2003), define reputation as the market s perception of CEO s ability, some other studies, such accounting studies as Desai et al. (2006), Srinwasan (2005), Wilson (2008) and Farber (2005) consider reputation as the market s perception of firm s financial reporting credibility. In this paper, reputation refers to the market s perception of CEO s ability and not the market s perception on firm s financial reporting credibility. 1

A long tenure with her firm helps the CEO establish her reputation since a long tenure is an indicator that the CEO has survived previous retention/dismissal decisions by the board of directors (Milbourn, 2003). After a CEO establishes her reputation, she becomes less concerned with reputation building and more concerned with reputation protection (e.g., Diamond, 1989). Any detection of aggressive reporting can make shareholders doubt the credibility of the CEO s previously reported performance and can substantially impair the CEO s reputation of her ability. Therefore, long-tenured CEOs are likely to refrain from aggressive financial reporting. 2 Thus, I hypothesize that long-tenured CEOs report earnings less aggressively than short-tenured CEOs. The above hypothesis is based on the assumption that CEOs plan to continue on their jobs and thus care about building and protecting their reputation, which is associated with long-term benefits. The above hypothesis is unlikely to hold, however, if CEOs are in the final year of their tenure because they would care more about short-term benefits, such as their current period s compensation and pension which could be a function of final year s compensation (e.g., Dechow and Sloan, 1991). Thus, I hypothesize that in the final year of their tenure, long-tenured CEOs are likely to report more aggressively than in the years leading up to the final year of tenure. I define aggressive reporting of earnings as smaller asymmetric timeliness of loss recognition (e.g., Basu, 1997; Ball and Shivakumar, 2005, 2008) and as higher discretionary accruals (e.g., Ashbaugh et al., 2003; Ball and Shivakumar, 2008; Ali et al., 2007). I measure asymmetric timeliness of loss recognition using the Basu (1997) model and the Ball and Shivakumar (2005, 2008) accruals model, and I measure discretionary 2 Desai et al. (2006) document that detection of previous aggressive reporting increases the likelihood of a CEO s job loss. 2

accruals using the combined Dechow and Dichev (2002) and Jones (1991) model (McNichols, 2002). For the sample period 1993 to 2005, I obtain the following results. Compared with short-tenured CEOs, long-tenured CEOs report earnings less aggressively, in terms of greater asymmetric timeliness of loss recognition and lower discretionary accruals. I also show that CEOs report earnings less aggressively in the second half of their tenures than in the first half. Finally, long-tenured CEOs report earnings more aggressively in their final year than in the years leading up to the final year of tenure. These results are consistent with my hypotheses. This study contributes to the extant literature on both CEO reputation and earnings management. First, theoretical studies (see, e.g., Fama, 1980; Holmstrom, 1999; Diamond, 1989) suggest that CEOs concerned about their reputation loss behave less opportunistically, but there is little empirical evidence on this issue. This study provides supporting evidence by showing that long-tenured CEOs tend to be less opportunistic in their financial reporting decisions. Second, when investigating determinants of earnings quality, prior studies focus on firm characteristics, such as firm size, growth, leverage, and corporate governance (Francis et al., 2008). This paper shows that CEO characteristic, specifically CEO s reputational concern, is also an important determinant of earnings quality. Two prior studies also investigate the relation between CEO reputation and earnings quality. Francis et al. (2008) document a negative correlation between CEO reputation and earnings quality. They use the frequency of media coverage as a measure of CEO reputation, and they use the standard deviation of discretionary accruals and absolute 3

value of discretionary accruals as measures of earnings quality. LaFond (2008) notes two major concerns with their study. First, he questions the validity of their proxy for reputation because media coverage could be for both good and bad reasons. Second, he questions their use of unsigned discretionary accruals as a measure of earnings quality because it is likely to be driven by the volatility of operating performance. My measure of reputation of CEO s ability, CEO tenure, is not subject to the above criticism. Moreover, CEO tenure has been used in the literature to measure CEO reputation (e.g., Milbourn, 2003). My measures of aggressive reporting, asymmetric timeliness of loss recognition and signed discretionary accruals, are not subject to the above criticism either. Second, Malmendier and Tate (2007) show that superstar CEOs are more likely to inflate reported performance through earnings management because they have to meet market expectation of superstar performance. The results of Malmendier and Tate (2007) suggest that the pressure on celebrity CEOs to consistently report superstar performance dominates their concerns about reputation loss that can result from the detection of opportunistic reporting. My results suggest that most long-tenured CEOs who have built their reputations of ability are not under such a bright limelight as superstar CEOs so their incentives for performance exaggeration are not strong enough to dominate their concerns of reputation loss of their ability from the detection of opportunistic reporting. This paper is organized as follows. Section 2 reviews related studies and develops the hypotheses. Section 3 discusses the sample. Section 4 presents the empirical results, and Section 5 concludes the paper. 4

2. Hypotheses Development Gibbons and Murphy (1992) argue that the market is unaware of a CEO s ability at the beginning of her tenure. First, CEOs rarely leave one firm to join another so at the beginning of their tenures, most CEOs do not have past records of their performance as a CEO. Moreover, firm-specific human capital that a CEO has can be difficult to transfer from one firm to another. Second, it takes time to isolate CEO contribution from other determinants of performance so shareholders are likely to be uncertain about the ability of a newly appointed CEO. Fama (1980) and Holmstrom (1982) suggest that the market evaluates a manager s ability over tenure with her current and past performance. Thus, in order to build reputation with the market and avoid being labeled as low ability, short-tenured CEOs are likely to have strong incentive to report good performance. Thus, if the true performance for a given period is low, short-tenured CEOs are likely to enhance their reported performance through aggressive reporting. Moreover, if the short-tenured CEOs who report earnings aggressively are detected as aggressive reporters in the current period, they have little reputation to lose at the beginning of tenure. 3 If CEOs are aware of their ability level relative to their peers, and since high ability CEOs know they can perform well in the long run, why would they report aggressively at the beginning of tenure and take the risk of being labeled as opportunistic reporters? Oyer (2008) and Axelson and Bond (2009) claim that at the beginning of their careers, if managers report poor outcome, they are labeled as low ability managers and their 3 Diamond (1989) analyzes the process of reputation acquisition in the debt market. He argues that in an adverse selection setting, managers with a short track of repayment record are more likely to invest opportunistically in risky projects. For such managers, reputation loss resulting from repayment default is very low, but risky projects can likely lead to abnormally high investment returns, which helps managers build a reputation of repayment with the market. 5

whole career suffers as a result, even if the temporary unsatisfactory outcome occurs because of bad market conditions rather than lack of managerial ability. Thus, even if uncontrollable events adversely affect firm performance, the high ability short-tenured CEOs are likely to inflate performance through aggressive reporting. 4 There is a chance that after future outcomes are observed, some of the aggressive reporters are detected and possibly face dismissal. However, some aggressive reporters may not be detected because subsequent events may turn out to be favorable, especially for talented CEOs, enabling them to bury the earlier bad performances (Kothari et al., 2009). In such cases, the high-ability CEOs will survive and build their reputations of ability. Once a CEO builds a reputation, his current performance is weighed less when the market updates its belief about his ability (Hermalin and Weisbach, 2009). Therefore, reputable CEOs incentive to report aggressively decreases. In addition, reputable CEOs concerns about reputation loss increase because given any single detection of opportunistic reporting, reputable CEOs have more to lose than CEOs with low reputations. 5 Based on the above arguments, I predict that long-tenured CEOs report less aggressively than short-tenured CEOs. Accordingly, I hypothesize the following: Hypothesis 1: CEOs with long tenures report less aggressively than CEOs with short tenures. 4 If CEOs are not aware of their ability relative to their peers, then in the beginning of their tenure, they have to inflate earnings to build reputation, because in an adverse selection setting everyone needs to demonstrate the best performance to avoid being labeled as low ability. This is also consistent with the prediction based on the assumption that CEOs are aware of their ability relative to their peers. 5 This argument is also consistent with that in Diamond (1989), which he makes in the context of the debt market. After managers have established reputation of repayment over time, they invest less opportunistically in risk projects, because their concerns shift from building their reputation to protecting their reputation and any single default in repayment can substantially impair their reputation. 6

The above hypothesis is unlikely to hold, however, for CEOs who plan to leave their firms after a short period. Since it is not easy for shareholders to detect aggressive reporting until the future outcomes are realized, if these CEOs engage in aggressive reporting, it is not likely to be detected before they leave their firms. Therefore, CEOs reputational concerns would reduce and instead they would more likely focus on reporting aggressively to enhance their current period compensation and pension benefits, which are likely to be based on final year salary. 6 Furthermore, just prior to turnover, it is easy for long-tenured CEOs to report aggressively for two likely reasons. First, long-tenured CEOs are not monitored as much by the board. Second, long-tenured CEOs are not likely to report aggressively in the years leading up to their last year of tenure, thus their balance sheets are unlikely to be bloated, providing them with greater flexibility to report aggressively in their final year. Thus, I hypothesize the following: Hypothesis 2: In the final year of tenure, CEOs with long tenures report more aggressively than in the years prior to their last year of tenure. For Hypothesis 2 to be valid, the following assumption should hold; my test of the hypothesis will be a test of this assumption as well. I assume that for a departing CEO, the private benefits from inflating current earnings exceed the future costs due to potential reputation loss. 7 A CEO can leave the firm, either because of retirement, 6 Dechow and Sloan (1991) find that CEOs tend to cut R&D expenditure during their final years because CEOs with short managerial horizons care much about their short-term benefits from compensation. Kalyta (2009) suggests that CEOs are likely to inflate earnings in their final year when their pensions are based on their final years salary. Watts (2003) also argues that limited horizon plays an important role in managers incentive to inflate earnings. 7 I also assume that a CEO can predict her turnover, and thus have enough time to influence the financial reports. Pourciau (1993) supports this assumption with following arguments. First, voluntary resignation 7

dismissal, or to take a job in another company. If the departing CEO retires after leaving the firm, then she may care more about current benefits from inflated earnings, such as compensation in the final year of tenure and pension plan benefits which are likely to be tied to the final year s salary (Kalyta, 2009). Also, she would be less concerned about reputation loss because the benefits from reputation after retirement are minimal. In the case where the departing CEO is dismissed by the board, then her reputation is impaired due to the dismissal so she is less concerned about reputation loss and is more likely to inflate earnings to extract private benefits such as current period s compensation. If the CEO transfers to another company, she would be concerned about maintaining her reputation and hence she is unlikely to report aggressively in the last year of her previous job. However, CEOs job transfers are not frequent (Gibbon and Murphy, 1992), hence, the effect from these cases is unlikely to dominate and, on average, the prediction in Hypothesis 2 would hold. 8 3. Data Sample and Descriptive Statistics For a sample period from 1993 to 2006, I obtain data of CEO tenure, CEO age, and CEO ownership from the ExecuComp database, accounting data from Compustat Annual Tape File, return data from CRSP database, G-index data from IRRC Corporate Governance database, and institutional holding data from Thomson Reuters 13f File. 9 and retirement can be predicted by CEOs. Second, it is very hard for the boards to ask CEOs to leave. Even if CEOs are asked to leave, they are always given time to prove themselves. Third, CEOs are not isolated from the events related to dismissal decisions, such as information gathering, feedbacks from colleagues, and coalition formation; and therefore, they can judge the seriousness of the situation. 8 Based on the ExecuComp database, I find that only 5 percent of CEOs join other firms as CEOs after their departure. 9 The G-Index is a variable of the corporate governance index based on the measure developed by Gompers et al. (2003). This index is comprised of 24 indicators that reflect the shareholder rights and is reverse ordered, i.e., higher index values indicate weaker shareholder rights. While the G-Index measure is 8

My final sample contains 14,234 observations to estimate the Basu model. Panel A of Table 1 presents the descriptive statistics of the variables that are used to estimate the Basu model. 10 The variables include TENURE it (CEO tenure), NI it (net income before extraordinary items scaled by the market value of equity at the beginning of fiscal year), RET it (annual return), NEG it (an indicator variable for negative annual return), MVEBOOK it-1 (market-to-book ratio), LEVERAGE it-1 (total debt divided by total assets), LNMVE it-1 (log of market value of equity), LIT it-1 (an indicator variable for a litigious industry), OWN it-1 (the percentage of CEO s ownership), AGE it (CEO s age), FIRMAGE it (firm age), GINDEX it (governance index), and INSTPERC it-1 (institutional holding). In my sample, CEOs have an average tenure of about 6.7 years, age of 56 years, and 2.1 percent ownership. These statistics are comparable to the statistics reported in prior studies. 11 The mean (median) NI it is 0.048 (0.056); the mean (median) RET it is 0.158 (0.108); the mean (median) market-to-book ratio is 3.833 (2.254); the mean (median) leverage is 0.231 (0.227); and the mean (median) log market capitalization is 7.521 (7.424). About 20.5 percent of the sample belongs to the high-litigation-risk industries. The mean (median) corporate governance index (G-Index) is 9.33 (9). The mean (median) institutional holding of firm s shares is 43.8 percent (51.8 percent). The average age of firms is about 37 years. 12 Panel B of Table 1 compares the mean of these variables between highest and lowest available for 1990, 1993, 1995, 1998, 2000, 2002, and 2004, I use the latest available G-Index value for the years where data is not available. 10 All the continuous variables are winsorized into the top and bottom 1% throughout the paper. 11 For example, in Coles et al. (2008), the mean of CEO tenure and age is 6.6 and 55.3, respectively; in Ryan et al. (2009), the mean of CEO tenure is 8.31; in Berger et al. (1997), the mean of CEO tenure is 7; the mean of CEO ownership in Cadman, Klasa and Matsunaga (2006) is 2.8%; and the mean of CEO ownership in LaFond and Roychowdhury (2008) is 3%. 12 Firm age is calculated based on the first year when a firm appears in the Compustat Database. 9

decile rank of CEO tenure. I observe that long-tenured CEOs are older and have higher percentage of ownership. Since managerial ownership is commonly used to measure the extent of interest alignment, which also affects the demand on asymmetric timeliness of loss recognition (LaFond and Roychowdhury, 2008), it is important to control for managerial ownership when investigating the relation between CEO tenure and earnings quality. The comparison also shows that the firms with long-tenured CEOs are more profitable and larger than those with short-tenured CEOs. Table 2 presents the Pearson and Spearman correlations. Since the results from the Pearson correlation and the Spearman correlation are consistent, I focus on the Pearson correlation to facilitate discussion. Consistent with Panel B of Table 1, CEO tenure is positively correlated with net income, CEO ownership, and CEO age. Net income is positively correlated with return (coefficient = 0.171) and negatively correlated with NEG it (coefficient = -0.207), suggesting that reported earnings reflect at least a portion of the information impounded in returns. 4. Empirical Results I measure asymmetric timeliness of loss recognition with the Basu model and the Ball and Shivakumar (2005, 2008) accruals model, and I measure signed discretionary accruals with the combined Dechow and Dichev (2002) and Jones (1991) model (McNichols, 2002). 13 To test the hypotheses, I identify long-tenured CEOs with two dimensions. One dimension is to compare tenures among different CEOs. Long-tenured 13 After considering all the popular measures of conditional conservatism, Ryan (2006) notes that asymmetric timeliness recognition is the most reasonable measure to capture conservatism. Ball et al. (2009) also validate the Basu model econometrically, and they show that conditional conservatism measures based on both asymmetric reversion in earnings changes and on accruals can also capture the asymmetric timeliness of loss recognition. 10

CEOs are those whose tenures are longer than other CEOs. The other dimension is to compare different stages of a CEO s tenure. For each CEO, the second half of tenure is regarded as longer tenured than the first half of tenure. 4.1. Test of Hypothesis 1: Comparing Tenures among Different CEOs 4.1.1. Asymmetric Timeliness of Loss Recognition Estimated Using the Basu Model with Market Return as a Proxy for Economic Gains and Losses Basu (1997) proposes a asymmetric timeliness of loss recognition model in which economic gains and losses are proxied by positive and negative market returns. Asymmetric timeliness of loss recognition is measured as the extent to which negative returns are reflected in reported earnings more rapidly than positive returns. NI it = α 0 + α 1 NEG it + α 2 RET it + α 3 RET it NEG it + ε it (1) NI it is year t net income before extraordinary items (Compustat #18) scaled by the market value of equity (Compustat #199 Compustat #25) at the beginning of fiscal year t. RET it is annual return, calculated as returns from three months after fiscal year end date in year t-1 to three months after fiscal year end date in year t. NEG it is an indicator variable that equals one if RET it is negative, and zero otherwise. α 2 measures the timeliness of earnings for economic gains recognition, and (α 2 + α 3 ) measures the timeliness of earnings for economic losses recognition. The coefficient α 3 captures asymmetric timeliness of earnings for economic losses recognition. 14 14 Though the Basu model assumes market informational efficiency, accounting information is still useful in order for shareholders to evaluate a manager s performance. Watts (2003) argues that accounting 11

Similar to LaFond and Roychowdhury (2008) and LaFond and Watts (2008), I allow for the variation of asymmetric timeliness of loss recognition with firm size, leverage, market-to-book ratio, and firm s litigation risk. In addition, CEO s tenure is positively correlated with CEO ownership in a firm (e.g., Coles et al., 2008). LaFond and Roychowdhury (2008) document that CEO ownership is an important determinant of asymmetric timeliness of loss recognition so I control for the effect of CEO ownership in the following analyses. I also control for CEO age because CEO age is a proxy for career concerns and is highly correlated with CEO tenure. In addition, CEO tenure is also considered to capture CEO entrenchment (e.g., Berger et al., 1997). To rule out the argument that long-tenured CEOs report more aggressively just because shareholders have greater demand on higher quality earnings for entrenched CEOs, I control for the corporate governance factors that are related to CEO entrenchment. Ahmed and Duellman (2006) suggest that the corporate governance index and institutional holding have an impact on asymmetric timeliness of loss recognition. Coles et al. (2008) document that firm age is correlated with corporate governance. I include firm age, the governance index, and institutional holding as controls. NI it = α 0 + α 1 NEG it + α 2 RET it + α 3 RET it NEG it + α 4 TENR it + α 5 TENR it NEG it + α 6 TENR it RET it + α 7 TENR it RET it NEG it + α 8 CEOOWNR it-1 + α 9 CEOOWNR it-1 NEG it + α 10 CEOOWNR it-1 RET it + α 11 CEOOWNR it-1 RET it NEG it + α 12 MTB it-1 + α 13 MTB it-1 NEG it + α 14 MTB it-1 RET it + α 15 MTB it-1 RET it NEG it + α 16 LEV it-1 + α 17 LEV it-1 NEG it + α 18 LEV it-1 RET it + α 19 LEV it-1 RET it NEG it + α 20 SIZE it-1 + α 21 SIZE it-1 NEG it + α 22 SIZE it-1 RET it + information is the hard benchmark to verify other sources of information. Sloan (1993) also argues that earnings reflect firm-specific changes in value, but they are less sensitive to market-wide movements in equity values so earnings are useful in evaluating firm-specific performance, which is more related to the manager s ability. In other words, even if shareholders have other information sources, they still rely heavily on the earnings number to evaluate a manager s ability. Therefore, even under the assumption of market efficiency, earnings performance still helps managers develop reputation. 12

α 23 SIZE it-1 RET it NEG it + α 24 LIT it-1 + α 25 LIT it-1 NEG it + α 26 LIT it-1 RET it + α 27 LIT it-1 RET it NEG it + α 28 AGER it + α 29 AGER it NEG it + α 30 AGER it RET it + α 31 AGER it RET it NEG it + α 32 LFIRMAGE it + α 33 LFIRMAGE it NEG it + α 34 LFIRMAGE it RET it + α 35 LFIRMAGE it RET it NEG it + α 36 GINDEX it + α 37 GINDEX it NEG it + α 38 GINDEX it RET it + α 39 GINDEX it RET it NEG it + α 40 INST it-1 + α 41 INST it-1 NEG it + α 42 INST it-1 RET it + α 43 INST it-1 RET it NEG it + ε it (2) TENR it is the scaled decile rank of CEO tenure. 15 CEOOWNR it-1 is the scaled decile rank of CEO s ownership percentage at the beginning of fiscal year t. MTB it-1 is the scaled decile rank of the market-to-book ratio (Compustat #199 Compustat #25/Compustat #60) at the beginning of the fiscal year t. LEV it-1 is the scaled decile rank of total debt (Compustat #9 + Compustat #34) divided by total assets (Compustat #6) at the beginning of the fiscal year t. SIZE it-1 is the scaled decile rank of the market value of equity (Compustat #199 Compustat #25) at the beginning of the fiscal year t. LIT it-1 is an indicator variable that equals one if a firm was in a litigious industry (SIC codes 2833 to 2836; 3570 to 3577; 3600 to 3674; 5200 to 5961, and 7370) at the beginning of year t, and zero otherwise. AGER it is the scaled decile rank of the CEO s age in year t. LFIRMAGE it is natural log of firm age in year t. GINDEX it is the governance index in year t (G-index) from Gompers et al. (2003). INST it-1 is the scaled decile rank of institutional holding at the beginning of the fiscal year t. In Equation (2), α 2 captures earnings timeliness with respect to economic gains, and α 3 captures asymmetric timeliness with respect to economic losses. α 6, α 10, α 14, α 18, α 22, α 26, α 30, α 34, α 38, and α 42 capture the association of earnings recognition timeliness with respect to economic gains with TENR it, CEOOWNR it-1, MTB it-1, LEV it-1, SIZE it-1, LIT it-1, 15 I use CEO tenure in year t, instead of year t-1, because, conceptually, accrued tenure and reputation level can be regarded as contemporaneous. Moreover, since tenure is consecutive and I use decile ranking in the analysis, all observations of tenure minus one should not change the results. 13

AGER it, LFIRMAGE it, GINDEX it, and INST it-1. α 7, α 11, α 15, α 19, α 23, α 27, α 31, α 35, α 39, and α 43 capture the association of asymmetric timeliness of earnings recognition with respect to economic losses with TENR it, CEOOWNR it-1, MTB it-1, LEV it-1, SIZE it-1, LIT it-1, AGER it, LFIRMAGE it, GINDEX it, and INST it-1. Table 3 presents the Fama-MacBeth regression estimates of Equation (2). 16 Panel A reports the results without controlling for AGER it, LFIRMAGE it, GINDEX it, and INST it-1 ; and Panel B reports the results with all control variables in Equation (2). In Panel A, the coefficient on RET it NEG it is 0.208 (t-statistics = 3.63), suggesting that, in general, firms financial reporting is conservative. The coefficient on TENR it RET it NEG it is 0.087 (t-statistics = 2.19), suggesting that long-tenured CEOs recognize loss in a timelier manner than short-tenured CEOs. These results are consistent with the prediction of Hypothesis 1. 17 The coefficients on other variables are similar to those in prior studies. In Panel B, the coefficient on TENR it RET it NEG it is 0.045 (t-statistics = 2.27), consistent with Hypothesis 1. The effect of CEO tenure on asymmetric timeliness of loss recognition is robust to considering other related corporate governance factors. In this model, except for the fact that institutional holding is negatively correlated with asymmetric timeliness of loss recognition, other corporate governance factors do not take effect. Thus, in the test of Hypothesis 2, for brevity, I will omit these corporate governance factors in my regressions. 4.1.2. Asymmetric Timeliness of Loss Recognition Estimated Using the Accrual Model 16 Throughout the paper, all the results are based on the Fama-MacBeth regressions. The conclusions based on these results are not sensitive to estimating the models using the pooled time-series cross-sectional data with the Huber-White procedure and clustering by both firm and year.. 17 In a sensitivity test, I also include board characteristics, namely, board composition, board size, CEO duality, and CEO power, as control variables. My conclusions continue to hold. 14

with Cash Flow as a Proxy for Economic Gains and Losses As Ball and Shivakumar (2005, 2006, 2008) specify, timeliness of loss recognition requires that economic losses are recognized on a timelier basis as unrealized accrued charges against earnings, while economic gains are recognized when gains are realized. Therefore, the positive relation between accruals and negative cash flow (a proxy for economic loss) is more pronounced for firms with greater timeliness of loss recognition than for firms with less timeliness of loss recognition. Ball and Shivakumar (2008) propose an accrual model by including variables used in the models of Jones (1991) and Ball and Shivakumar (2008). ACC it = α 0 + α 1 CFO it + α 2 CHSALES it + α 3 FASSET it + α 4 DCFO it + α 5 DCFO it CFO it + ε it (3) ACC it is the accruals in year t, measured as earnings before extraordinary items (Compustat data #18) minus cash flow from operations (Compustat data #308), scaled by total assets (Compustat data #6) at the beginning of year t. 18 CFO it is cash flow from operations in year t, scaled by total assets at the beginning of year t. DCFO it is an indicator variable that takes the value of one if CFO it is negative, and zero otherwise. CHSALES it is the change in revenue (Compustat data #12) in year t, scaled by total assets at the beginning of year t. FASSET it is gross property, plant, and equipment (Compustat data #7) in year t, scaled by total assets at the beginning of year t. Ball and Shivakumar (2005, 2006) discuss two roles of accruals. The first role of accruals is to mitigate noise in operating cash flow, implying that accruals and operating 18 According to Collins and Hribar (2002), estimating accruals with the balance sheet approach has measurement errors so I measure accruals directly from statement of cash flow throughout this paper. 15

cash flows are negatively correlated. α 1 in Equation (3) is predicted to be negative, capturing the noise mitigation effect. The second role of accruals is asymmetric timeliness of the recognition of economic losses, implying α 5 in Equation (3) positive. I allow for the asymmetric timeliness of loss recognition to vary with CEO tenure, as well as with several control variables, including those suggested by Ball and Shivakumar (2008) and obtain the following model. ACC it = α 0 + α 1 CFO it + α 2 CHSALES it + α 3 FASSET it + α 4 DCFO it + α 5 DCFO it CFO it + α 6 TENR it + α 7 TENR it CFO it + α 8 TENR it CHSALES it + α 9 TENR it FASSET it + α 10 TENR it DCFO it + α 11 TENR it DCFO it CFO it + α 12 CEOOWNR it-1 + α 13 CEOOENR it-1 CFO it + α 14 CEOOWNR it-1 CHSALES it + α 15 CEOOWNR it-1 FASSET it + α 16 CEOOWNR it-1 DCFO it + α 17 CEOOWNR it-1 DCFO it CFO it + α 18 AGER it + α 19 AGER it CFO it + α 20 AGER it CHSALES it + α 21 AGER it FASSET it + α 22 AGER it DCFO it + α 23 AGER it DCFO it CFO it + α 24 LFIRMAGE it + α 25 LFIRMAGE it CFO it + α 26 LFIRMAGE it CHSALES it + α 27 LFIRMAGE it FASSET it + α 28 LFIRMAGE it DCFO it + α 29 LFIRMAGE it DCFO it CFO it + α 30 GINDEX it + α 31 GINDEX it CFO it + α 32 GINDEX it CHSALES it + α 33 GINDEX it FASSET it + α 34 GINDEX it DCFO it + α 35 GINDEX it DCFO it CFO it + α 36 INST it-1 + α 37 INST it-1 CFO it + α 38 INST it-1 CHSALES it + α 39 INST it-1 FASSET it + α 40 INST it-1 DCFO it + α 41 INST it-1 DCFO it CFO it + ε it (4) Similar to Equation (2), TENR it is the scaled decile rank of CEO tenure. CEOOWNR it-1 is the scaled decile rank of CEO s ownership percentage at the beginning of the fiscal year t. AGER it is the scaled decile rank of CEO age. LFIRMAGE it is the natural log of firm age. GINDEX it is the governance index (G-index) from Gompers et al. (2003). INST it-1 is the scaled decile rank of institutional holding at the beginning of fiscal year t. Table 4 presents the Fama-MacBeth regression estimates of Equation (4). In Panel A, the coefficient on TENR it DCFO it CFO it is 1.176 (t-statistics = 2.37), suggesting that 16

long-tenured CEOs recognize losses in a timelier manner than short-tenured CEOs, consistent with the prediction of Hypothesis 1. The result from Panel B leads to the same conclusion on the relation between CEO tenure and reporting aggressiveness. 4.1.3. Asymmetric Timeliness of Loss Recognition Estimated Using the Earnings Change Model with Earnings Change as a Proxy for Economic Gains and Losses Ball and Shivakumar (2005) propose that timely recognition of economic gains and losses implies that transitory changes in earnings components tend to reverse in the next period. For firms with greater timeliness of loss recognition, the reversal is more pronounced for earnings decreases in the current period. Consider the following basic model: ΔNI it+1 = α 0 + α 1 DΔNI it + α 2 ΔNI it + α 3 DΔNI it ΔNI it + ε it (5) ΔNI it is the change in net income before extraordinary items (Compustat data #18) from year t-1 to year t, scaled by total assets at the beginning of year t. DΔNI it is an indicator variable that equals one if ΔNI it <0, and zero otherwise. According to Ball and Shivakumar (2005), asymmetric timeliness of economic losses implies α 3 negative. Similar to Ball and Shivakumar (2005), I extend Equation (5) by including the variable of interest, CEO tenure, and control variables. ΔNI it+1 = α 0 + α 1 DΔNI it + α 2 ΔNI it + α 3 DΔNI it ΔNI it + α 4 TENR it + α 5 TENR it DΔNI it + α 6 TENR it ΔNI it + α 7 TENR it DΔNI it ΔNI it + α 8 CEOOENR it-1 + α 9 CEOOWNR it-1 DΔNI it + α 10 CEOOENR it-1 ΔNI it + α 11 CEOOWNR it-1 DΔNI it ΔNI it + α 12 SIZED it + α 13 SIZED it DΔNI it + α 14 SIZED it ΔNI it + 17

α 15 SIZED it DΔNI it ΔNI it + α 16 AGER it + α 17 AGER it DΔNI it + α 18 AGER it ΔNI it + α 19 AGER it DΔNI it ΔNI it + α 20 LFIRMAGE it + α 21 LFIRMAGE it DΔNI it + α 22 LFIRMAGE it ΔNI it + α 23 LFIRMAGE it DΔNI it ΔNI it + α 24 GINDEX it + α 25 GINDEX it DΔNI it + α 26 GINDEX it ΔNI it + α 27 GINDEX it DΔNI it ΔNI it + α 28 INST it-1 + α 29 INST it-1 DΔNI it + α 30 INST it-1 ΔNI it + α 31 INST it-1 DΔNI it ΔNI it + ε it (6) TENR it is the scaled decile rank of CEO tenure. CEOOWNR it-1 is the scaled decile rank of CEO s ownership percentage at the beginning of the fiscal year t. AGER it is the scaled decile rank of CEO s age. SIZED it is the percentile rank of total assets at end of the fiscal year t, standardized to vary between zero and one. 19 LFIRMAGE it is the natural log of firm age. GINDEX it is the governance index. INST it-1 is the scaled decile rank of institutional holding at the beginning of the fiscal year t. Hypothesis 1 predicts α 7 to be negative. Table 5 reports the Fama-MacBeth regression estimates of Equation (6). In Panel A, the coefficient on TENR it DΔNI it ΔNI it is -0.328 (t-statistics = -2.74), suggesting that long-tenured CEOs recognize the economic losses in a timelier manner than short-tenured CEOs, consistent with the prediction of Hypothesis 1. These results are not sensitive to including corporate governance control variables, as shown in Panel B. 4.1.4. Discretionary Accruals Following McNichols (2002), I estimate nondiscretionary accruals with the following model: 20 19 To be consistent with the specifications in Ball and Shivakumar (2005), I use percentile rank of total assets to measure firm size in (6). This is different from the firm size measure used in Equation (2). 20 Ball and Shivakumar (2006) propose an accruals model to estimate nondiscretionary accruals; that is similar to Equation (3). I use the alternative accruals model and assure at similar conclusions. 18

ACC it = λ 0 + λ 1 CFO it-1 + λ 2 CFO it + λ 3 CFO it+1 + λ 4 ΔREV it + λ 5 PPE it + ε it (7) ACC it is the accruals in year t, measured as earnings before extraordinary items minus cash flow from operations, scaled by total assets at the beginning of year t. CFO it is cash flow from operations in year t, scaled by total assets at the beginning of year t. REV it is the change in revenue in year t, scaled by total assets at the beginning of year t. PPE it is gross property, plant, and equipment in year t, scaled by total assets at the beginning of year t. I estimate Model (7) separately for each industry-year group and use residuals, ε it, from these regressions as a proxy for discretionary accruals (DA). 21 I estimate the following model to examine whether long-tenured CEOs report lower discretionary accruals than short-tenured CEOs. The control variables are from Ashbaugh et al. (2003), Ali et al. (2007), and Cheng and Warfield (2005). The significance level is based on standard errors, adjusted using the Huber-White procedure with clustering by firms. DA it = α 0 + α 1 TENR it + α 2 CEOOWN it-1 + α 3 LNMVE it + α 4 MVEBOOK it +α 5 LIT it +α 6 LEVERAGE it +α 7 ROA it + α 8 LOSS it + α 9 MA it + α 10 LTACC it + α 11 LNOA it + α 12 CFO it + α 13 FIN it + α 14 INSTPERC it-1 + Year Indicators + Industry Indicators + δ it (8) TENR it is the scaled decile rank of CEO tenure. CEOOWN it-1 is CEO s ownership percentage at the beginning of fiscal year t. LNMVE it is the log of market value of equity, controlling for firm size. MVEBOOK it is market-to-book ratio, defined as market value of equity divided by the book value of equity at the end of year t, controlling for growth. LIT it is an indicator variable that equals one if the firm operates in a high-litigation 21 I estimate model (7), requiring that each one-digit SIC-year group has at least 10 observations. 19

industry, and zero otherwise, controlling for ex-ante litigation risk. LEVERAGE it equals the total debt divided by total assets at the end of year t, controlling for financial distress. ROA it is earnings before extraordinary items divided by total assets, controlling for firm performance. LOSS it is an indicator variable that equals one if the firm reports a net loss, and zero otherwise. MA it is an indicator variable that equals one if the firm has engaged in a merger and acquisition, and zero otherwise. LTACC it is last year s total accruals scaled by the total assets at the beginning of the fiscal year, controlling for the accrual reversal. LNOA it is the net operating asset at the end of last year, calculated as shareholders equity minus cash and marketable securities, plus total debt, divided by sales, controlling for previous periods earnings management (Barton and Simko, 2002). CFO it is cash flow from operations scaled by the total assets at the beginning of the fiscal year. FIN it is an indicator variable that equals one if MA it is not equal to one and if the number of outstanding shares increased by at least 10%, or long-term debts increased at least 20%, or the firm first appears on the CRSP monthly returns database during the fiscal year, and zero otherwise. INSTPERC it-1 is the percentage of stocks held by institutional investors at the beginning of year t, controlling for institutional monitoring. Industry indictors are based on the Fama-French 48 industry groups. Hypothesis 1 predicts α 1 to be negative. Table 6 presents the regression results of Equation (8). In Panel A, the coefficient on TENR it is -0.0035 (t-statistics = -2.25); and in Panel B, the coefficient on TENR it is -0.0038 (t-statistics = -2.18). These results suggest that long-tenured CEOs report lower discretionary accruals, consistent with Hypothesis 1 that long-tenured CEOs report less aggressively. The coefficients on other variables are similar to those in prior studies. 20

4.2. Test of Hypothesis 1: Comparing Different Stages of a CEO s Tenure 4.2.1. Asymmetric Timeliness of Loss Recognition Estimated Using the Basu Model with Market Return as a Proxy for Economic Gains and Losses I divide each CEO s tenure into two equal parts and use the indictor variable, LATTER it, to identify the second half of CEO tenure. For this test, I select CEOs who have at least six years of tenure. I run a pooled regression and apply the Huber-White procedure with clustering by firm (Petersen, 2009). NI it = α 0 + α 1 NEG it + α 2 RET it + α 3 RET it NEG it + α 4 LATTER it + α 5 LATTER it NEG it + α 6 LATTER it RET it + α 7 LATTER it RET it NEG it + α 8 CEOOWNR it-1 + α 9 CEOOWNR it-1 NEG it + α 10 CEOOWNR it-1 RET it + α 11 CEOOWNR it-1 RET it NEG it + α 12 MTB it-1 + α 13 MTB it-1 NEG it + α 14 MTB it-1 RET it + α 15 MTB it-1 RET it NEG it + α 16 LEV it-1 + α 17 LEV it-1 NEG it + α 18 LEV it-1 RET it + α 19 LEV it-1 RET it NEG it + α 20 SIZE it-1 + α 21 SIZE it-1 NEG it + α 22 SIZE it-1 RET it + α 23 SIZE it-1 RET it NEG it + α 24 LIT it-1 + α 25 LIT it-1 NEG it + α 26 LIT it-1 RET it + α 27 LIT it-1 RET it NEG it + α 28 TREND t + α 29 TREND t NEG it + α 30 TREND t RET it + α 31 TREND t RET it NEG it + δ it (9) TREND t is the time trend, defined as year minus 1992. Other variables are defined as in Equation (2). Table 7 reports the regression estimates of Equation (9). In panel A, the coefficient on LATTER it RET it NEG it is 0.041 (t-statistics = 2.88), suggesting that CEOs recognize losses in a timelier manner in the second half of their tenures than in the first half, consistent with the prediction of Hypothesis 1. Panel B shows that these results are not sensitive to the exclusion of the control variable of time trend. 4.2.2. Discretionary Accruals I also examine whether discretionary accruals are lower during the second half of CEOs tenures than their first half. For this purpose, I use a model similar to Equation (8). 21

DA it = α 0 + α 1 LATTER it + α 2 CEOOWN it-1 + α 3 LNMVE it + α 4 MVEBOOK it +α 5 LIT it +α 6 LEVERAGE it +α 7 ROA it + α 8 LOSS it + α 9 MA it + α 10 LTACC it + α 11 LNOA it + α 12 CFO it + α 13 FIN it + α 14 INSTPERC it-1 + Year Indicators + Industry Indicators + δ it (10) Table 8 presents the regression estimates of Equation (10). The coefficient on LATTER it is -0.0033 (t-statistics = -2.26), which is consistent with Hypothesis 2 that CEOs report less aggressively in the latter half of their tenure than in the first half. 4.3. Test of Hypothesis 2 4.3.1. Asymmetric Asymmetric Timeliness of Loss Recognition Estimated Using Basu Model with Market Return as a Proxy for Economic Gains and Losses To test Hypothesis 2, I include in Equation (2) a new variable, FINALYEAR it, defined as one if the year is one year prior to CEO s turnover, and as zero otherwise. 22 NI it = α 0 + α 1 NEG it + α 2 RET it + α 3 RET it NEG it + α 4 TENR it + α 5 TENR it NEG it + α 6 TENR it RET it + α 7 TENR it RET it NEG it +α 8 FINALYEAR it + α 9 FINALYEAR it NEG it + α 10 FINALYEAR it RET it + α 11 FINALYEAR it RET it NEG it + α 12 TENR it FINALYEAR it + α 13 TENR it FINALYEAR it NEG it + α 14 TENR it FINALYEAR it RET it + α 15 TENR it FINALYEAR it RET it NEG it + α 16 CEOOWNR it-1 + α 17 CEOOWNR it-1 NEG it + α 18 CEOOWNR it-1 RET it + α 19 CEOOWNR it-1 RET it NEG it + α 20 MTB it-1 + α 21 MTB it-1 NEG it + α 22 MTB it-1 RET it + α 23 MTB it-1 RET it NEG it + α 24 LEV it-1 + α 25 LEV it-1 NEG it + α 26 LEV it-1 RET it + α 27 LEV it-1 RET it NEG it + α 28 SIZE it-1 + α 29 SIZE it-1 NEG it + α 30 SIZE it-1 RET it + α 31 SIZE it-1 RET it NEG it + α 32 LIT it-1 + α 33 LIT it-1 NEG it + α 34 LIT it-1 RET it + α 35 LIT it-1 RET it NEG it + ε it (11) Table 9 reports the Fama-MacBeth regression estimates of Equation (11). The coefficient on TENR it RET it NEG it is 0.062 (t-statistics = 2.58), consistent with the prediction of Hypothesis 1. The coefficient on TENR it FINALYEAR it RET it NEG it is 22 Since the sample for analysis is till 2005, I can use the CEO information data till 2006 in ExecuComp to identify the CEO turnover before 2005. 22

-0.178 (t-statistics = -2.32), suggesting that long-tenured CEOs report more aggressively in their final year than in the years leading up to the final year. This finding is consistent with Hypothesis 2. 4.3.2. Discretionary Accruals I estimate the following model to examine whether long-tenured CEOs increase discretionary accruals when they are in their final year of tenure. DA it = α 0 + α 1 TENR it + α 2 FINALYEAR it + α 3 TENR it FINALYEAR it + α 4 CEOOWN it-1 + α 5 LNMVE it + α 6 MVEBOOK it +α 7 LIT it +α 8 LEVERAGE it +α 9 ROA it + α 10 LOSS it + α 11 MA it + α 12 LTACC it + α 13 LNOA it + α 14 CFO it + α 15 FIN it + α 16 INSTPERC it-1 + Year Indicators + Industry Indicators + δ it (12) Hypothesis 2 predicts α 3 to be positive. Table 10 presents the regression results of Equation (12). The coefficient on TENR it is -0.0034 (t-statistics = -2.43), consistent with Hypothesis 1. The coefficient on TENR it FINALYEAR it is 0.0082 (t-statistics = 2.25). These results suggest that long-tenured CEOs report more aggressively in the final year of their tenure, consistent with the notion that at that stage, their reputational concerns reduce and the benefits from aggressive reporting in that period dominates. 4.4. Additional Analyses The above analyses show that long-tenured CEOs in general report less aggressively than short-tenured CEOs. I argue that it is because long-tenured CEOs care more about protecting their reputation with higher quality reporting, while short-tenured CEOs care more about establishing their reputation with more aggressive reporting. An alternative explanation could be that conservative reporting is a personal trait. Specifically, 23

long-tenured CEOs are the type who has greater innate ability so they enjoy long tenure because of excellent performance and they do not need to report opportunistically. On the other hand, short-tenured CEOs are the type who has less innate ability so they cannot survive for long in their position even with opportunistic reporting in their early tenure. This explanation suggests that long-tenured CEOs would report less aggressively than short-tenured CEOs. This argument has two other predictions, however. First, for long-tenured CEOs, their level of aggressive reporting would not change over their tenure. Second, at the beginning of their tenure, long-tenured CEOs would report less aggressively than short-tenured CEOs. The results and discussions in the Section 4.2. show that CEOs who have stayed in office at least six years report less aggressively in the second half of their tenure than in the first half. This is inconsistent with the first prediction of the above argument. I select six years as a cutoff to define long-tenured CEOs because the median of CEO tenure in my sample is five years. The results are robust when I use a seven- or eight-year of tenure as the cutoff. I test the second prediction of the above argument with the modified models of Equation (2) and Equation (8). I restrict the subsample to include only the first three years of CEO tenure, and I define the indicator variable, LONG it, to identify long-tenured CEOs. LONG it is equal to one if the CEO has a total of at least six years of tenure in the firm, and zero otherwise. Table 11 shows that the coefficient on LONG it RET it NEG it is insignificant, and Table 12 shows that the coefficient on LONG it is insignificant. These results suggest that in the early years of tenure of CEOs who eventually had long tenures, the level of aggressive reporting is not significantly less than that of CEOs who eventually 24

had short tenures. These results are inconsistent with the alternative explanation that higher ability CEOs report less aggressively through their career. 5. Conclusion This study investigates the relation between CEO tenure and earnings quality. I find that, on average, long-tenured CEOs report earnings less aggressively than short-tenured CEOs, both in terms of recognizing economic losses in a timelier manner and reporting lower discretionary accruals. These results are consistent with the notion that at the beginning of their tenure, in order to build reputation of ability, CEOs have incentives to inflate earnings; however, after the CEOs establish reputation, as proxied by long tenure, they report conservatively to protect their reputation. I also find that in the last year of their tenure, long-tenured CEOs report more aggressively than in the other years of their tenure. This result suggests that CEOs are not concerned about protecting their reputation in the final year of their tenure. Moreover, these CEOs have the incentive to report aggressively to extract short-term benefits associated with inflated current period s earnings, such as higher current period s compensation or higher pension annuity. This aggressive behavior may be difficult for investors to detect until the firm s future realizations become known. However, by then, these CEOs have left their firms. 25