Managerial Risk-Taking Incentives and Earnings Guidance
Tianqi Jiang Florida Atlantic University Zhao Wang University of Rhode Island
Abstract In this paper, we investigate the relationship between managerial risk-taking incentives and management earnings guidance. Using data of forecasts for annual earnings per share, we find that managerial risk preference significantly influences disclosure features. Our evidence suggests that CEO risk-taking incentives are positively related to managers’ propensity to issue annual earnings forecasts and the frequency of the issuance. CFO risk-taking incentives can lower the precision of forecasts and significantly lead to the longer horizon of management earnings forecasts. Our findings are consistent with Chava and Purnanandam’s (2010) explanation that CFOs have more impact than CEOs on decisions involving more specialized knowledge of finance. To some extent that our results provide support to the voice for eliminating short-term earnings guidance (Choi et al., 2011).
Keywords: managerial risk-taking incentives, corporate disclosure, management forecasts, earnings guidance JEL Classification: G32; M41; M52
January 2018
1
Introduction This research is motivated by current debates on management earnings guidance, which
reflects the management voluntary forecasts on the firm’s earnings in the future. Conventional research on earnings guidance typically emphasizes that guidance is a proper channel that enables managers to communicate with outsiders (e.g., Lang and Lundholm, 1996; Healy and Palepu, 2001) and it serves as a tool to mitigate information asymmetry (Coller and Yohn, 1997). However, there is an increasing voice against earnings guidance. Choi at al. (2011) document that some critical institutions, such as the U.S. Chamber of Commerce, the CFA Institute, the Business Roundtable Institute for Corporate Ethics, and The Conference Board, call for the elimination of short-term earnings guidance. Considering preparing costs, potential myopic behavior, and risk-related issues, these institutions believe that providing guidance is detrimental to firms. In this paper, we investigate the impact of managerial equity incentives on providing earnings guidance. There exist at least two mechanisms that executive equity incentives could affect firms’ earnings guidance. First, managers with higher equity incentives may display managerial myopia. One example is provided by Dong, Wang, and Xie (2010). They study managerial decision on capital structure and find evidence showing that CEOs with high risktaking incentives take excessive risks, which are likely to hurt long-term performance. In our case, managers with significant equity incentives may be more likely to issue earnings guidance, which is widely considered as a short-term focus on firms’ operation. The second mechanism is about risk per se. Prior literature suggests that providing earnings guidance is related to litigation risk, stock price crash risk, and potential earnings management risk or say fraud risk (Skinner, 1997; Hamm, Li, and Ng, 2012; Kraft, Lee, and Lopatta, 2014). Thus, higher risk-taking
1
incentives may encourage managers to forecast firms’ earnings, because they are less averse to the increased risks in conjunction with earnings guidance. Inspired by these two mechanisms, we focus on whether executive equity incentives induce managers to issue earnings forecasts and whether Chief Executive Officers (CEOs) and Chief Financial Officers (CFOs) have different impacts on the features of earnings guidance. The reason we investigate these two executives is that CEOs cannot, in general, prepare the forecasts solely by themselves, and they need to cooperate with CFOs. Our empirical results show that CEOs significantly influence the decisions to issue an earnings guidance and how many times to issue earnings guidance in a fiscal year. When the decisions are related to more specialized knowledge, CFOs have more significant impacts. We find that the firm’s actual earnings are less likely to fall within the forecast range and the horizon between earnings forecast date and earnings announcement date will be longer if the firm’s CFO has greater risk-taking incentives. These findings are consistent with Chava and Purnanandam (2010) that CEOs’ risk-taking incentives significantly influence broader corporate decisions, whereas CFOs’ risk-taking incentives are significant determinants of financial judgment. Our study contributes to a growing literature on managerial risk-taking incentives and corporate disclosure policy. First, we provide evidence to establish a connection between managerial equity incentives and earnings guidance issuance. Extensive studies have explored the influence of managerial compensation structure on corporate policies, but few studies discussed the relationship between managerial equity incentives and disclosure decisions. Second, as recent researchers gradually notice the importance of CFOs, we distinguish the different roles played by CEOs and CFOs and contribute to studies on CFOs with empirical evidence. Third, our results in part support the rising voice against short-term earnings guidance.
2
The remainder of this paper is organized as follows. We collect prior research and develop our hypotheses in section 2, and in Section 3 we present data sources, variable definitions, and descriptive statistics. Section 4 presents our empirical results on the association between managerial risk-taking incentives and disclosure behavior. Section 5 concludes.
2
Related literature and hypotheses In this section, we review the literature on corporate disclosure policy and the studies on
managerial equity incentives. We expect issuance of earnings guidance is related to managerial myopia and risk factors, and develop three testable hypotheses. 2.1
Related literature on corporate disclosure policy Prior studies on corporate disclosure policy mainly focus on two fields: (1) the impact of
information contained in disclosures on market characteristics and market participants; (2) the determinants of disclosure quality. Based on the literature from the first field, we can build up a potential channel to link managerial equity incentives to voluntary disclosures and the second stream of literature enable us to isolate the effect of managerial risk-taking incentives. Corporate disclosure can take many forms, quantitative or qualitative, and provide information on many concerns, but the most-studied practice is management earnings forecasts (i.e., earnings guidance) because, in part, they directly reveal performance and are closely related to market reactions. Conventional research on the first field emphasize the benefits of supplying earnings guidance to mitigate information asymmetry. Lang and Lundholm (1996) find that the informativeness of the firm’s disclosure is positively related to analyst coverage and analyst forecast accuracy. Managers care about their outside information environment and decide to
3
release forecasts due to increased information asymmetry (Coller and Yohn, 1997). Healy and Palepu (2001) survey existing empirical research and suggest that disclosure is an essential way for managers to inform outsiders on firm performance and governance. From the perspective of self-interest, Skinner (1997) presents empirical evidence to show that managers have incentives to disclose their bad earnings news to avoid litigation risk and loss of reputation. Recent research (e.g., Houston, Lev, and Tucker, 2010; Chen, Matsumoto, and Rajgopal, 2011) compares the difference between guiding firms and non-guiding firms. These studies believe that firms stop providing earnings guidance because they have unfavorable operating performance, less chance to match or beat analyst forecasts, or fewer informed investors. A guiding firm is more likely to become a non-guiding firm if it does not expect favorable news in the future or long-term investors lower its incentives to release guidance. Thus, one intuitive idea on earnings forecasts is that this practice is favorable for investors and analysts. However, controversial arguments against this practice never stop. In contrast to the research on earnings guidance stoppers, Hu, Hwang, and Jiang (2014) study quarterly guidance stoppers from 2002 to 2011 and find that earnings guidance cessation is positively related to consequential increases in liquidity and declines in information asymmetry. They suggest that decreasing motivation to manipulate earnings is the main reason for these empirical results. If managers do not need to release earnings guidance, they have fewer incentives to engage in earnings management. Similarly, Acito (2010) documents that firms issuing earnings guidance are more likely to manipulate accrual-based earnings to beat their own forecasts, but not to beat analyst forecasts, whereas firms issuing no earnings guidance also engage in earnings manipulation but for beating analyst forecasts. This finding is confirmed by Kraft, Lee, and Lopatta (2014) and their evidence shows that senior managers have a higher
4
likelihood to use accrual-based earnings management to meet their predicted earnings forecasts. Moreover, earnings guidance reveals managerial myopia. Cheng, Subramanyam, and Zhang (2005) document that guiding firms display myopic R&D investment behavior and sacrifice long-term earnings growth to meet a short-term earnings target. Brochet, Loumioti, and Serafeim (2015) show that earning guidance is related to shorter disclosure horizon. Another debate focuses on the informativeness of earnings guidance. Although it is possible that earnings guidance efficiently convey managerial opinions to outsiders, the distortion of information is still a potential problem. Hamm, Li, and Ng (2012) find that the frequency of earning guidance is positively associated with a higher likelihood of crash risk, which measured by a dummy variable that equals to one if there is no less than one extremely low disturbance term. This relationship is more pronounced when the agency problems have potential influence (e.g., when executive stock ownership is high, or the percentage of institutional holdings are low). These findings indicate that if managers have incentives or are merely less monitored, they are possible to distort a disclosure. Another example of distorting information is provided by Cheng, Luo, and Yue (2013). They find that managers intentionally disclose more accurate good news to increase stock prices before insider sales and more precise bad news to decrease stock prices before insider purchases. By investigating audit fees, Krishnan, Pevzner, and Sengupta (2012) document a positive association between issuance of annual or quarterly earnings guidance and audit fees, and lower precision of current annual forecasts leads to higher audit fees for the next year. Their evidence implies that auditors concern such forecast behavior because it represents a higher business risk for auditors. We also notice some current research connecting management earnings guidance to market uncertainty. The empirical findings of these studies are mixed. Rogers, Skinner, and Van
5
Buskirk (2009) investigate implied option volatility dynamics in respond to unbundled earnings guidance and observe that there is a short-term increase in implied volatility around the guidance. Likewise, Agapova and Madura (2016) offer evidence showing that the likelihood of issuing quarterly guidance is lower when market uncertainty, measured by the VIX index, is high. However, these finding is challenged by Billings, Jennings, and Lev (2015) using the samples of bundle guidance with earnings announcements and longer event windows. The second stream of literature discusses the determinants of earnings guidance. Consistent with proprietary costs theory, Ali, Klasa, and Yeung (2014) find that firms in highconcentrated industries provide fewer earnings forecasts and have the shorter horizon. Ittner and Michels (2017) use survey data to show that sophisticated risk-based forecasting and planning procedures are negatively related to earnings forecast errors and range. Larger firms are generally associated with better forecasts and accuracy (Lang and Lundholm, 1996; Yang, 2012) Growth opportunities measured by research and development (R&D) costs make earnings forecasts more difficult (Cheng, Luo, and Yue, 2013). Other factors, such as stability, institutional ownership, and litigation risk, are related to forecasting quality, too. 2.2
Related literature on managerial risk-taking incentives A seminal paper of Coles, Daniel, and Naveen (2006) comprehensively examines the
causal relation between managerial equity incentives and risk-taking behavior. Their two proxies for equity incentives, pay-performance sensitivity (delta) and volatility sensitivity (vega), are widely used after that. Vega is positively associated with more R&D investments, less PPE investments, and higher leverage. Therefore, vega is widely considered as a particular proxy for managerial risk-taking incentives, especially after Armstrong et al. (2013) citing that delta has two countervailing incentive effects. Because delta is a measurement of an option’s price
6
sensitivity to a given change in stock price, the uncertainty of risky projects may result in higher stock volatility and thus rewards managers extra returns on increases in stock price (“reward effect”) or magnifies loss on declined stock price (“risk effect”). However, vega unambiguously measures the magnitude of benefits from taking risky projects. There are extensive related studies examining managerial risk-taking incentives and corporate decisions using CEOs data. But a growing number of researchers start to consider other top executives, such as CFOs, because these executives may play different roles comparing with CEOs. Jiang, Petroni, and Wang (2010) extend Bergstresser and Philippon’s (2006) research and study the association between CFO equity incentives and accrual-based earnings management. They find that CFOs have more influence than CEOs on earnings management, measured by the absolute total accruals, the absolute discretionary accruals, and the likelihood of beating analyst forecasts. In contrast, Feng et al. (2011) criticize that the findings may not be generalized to cases out of GAAP and they believe that CFOs engaging in earnings management merely succumb to the pressure of CEOs. Chava and Purnanandam (2010) also investigate the different effects of CEO risk-taking incentives and CFO risk-taking incentives on capital structure decision, cash holdings, debt maturity, and accrual management. They find that CFO risk-taking incentives are the key determinant of the firm decisions regarding more specialized judgment on the part of the finance team, such as short-term debt, accrual management, and the reverse. Kim, Li, and Zhang (2011) find a positive association between CFO equity incentives and crash risk, whereas CEO equity incentives have negligible impact. This relationship is more pronounced for firms in noncompetitive industries and for firms with more financial leverage. In this paper, we use vega to measure managerial equity incentives because vega captures
7
managerial risk preferences and Dong, Wang, and Xie (2010) suggest that high-vega managers tend to focus on short-term interests. 2.3
Hypothesis development According to prior literature, earnings guidance is a proxy for managers’ myopic
behavior and reflects potential distort information if managers have incentives to do so. The latter mechanism is similar to earnings management. However, strategic decision on earnings guidance is subject to less risk than earnings management or withholding news regarding performance (Cheng, Luo, and Yue, 2013). Moreover, auditors increase audit fees in response to earnings forecast behavior because it increases business risk for auditors. To some extent that distorted guidance results in increasing uncertainty. In conclusion, managers with higher vega should be more likely to engage in earnings forecasts. Accordingly, our first hypothesis predicts:
H1a. Managerial risk-taking incentives are positively related to the propensity to provide management earnings forecasts.
H1b. Managerial risk-taking incentives are positively related to the frequency of releasing management earnings forecasts.
Our second hypothesis is based on the horizon between earnings forecasts and earnings announcement. Uncertainty on the actual earnings increases with the longer horizon. In other words, if the forecast is made earlier, the likelihood of efficiently influencing outsider opinions will be higher. So, our second hypothesis predicts:
8
H2. Managerial risk-taking incentives are positively related to the horizon between management earnings forecasts.
We expect higher managerial risk-taking incentives will result in less accurate forecasts. Further, existing studies suggest that CFOs have more influence on financial decisions. Collectively, our last hypothesis predicts:
H3. CFO risk-taking incentives are negatively related to the precision of earnings forecast, whereas CEO risk-taking incentives have few impact.
3 3.1
Sample and data Sample selection Our empirical analysis requests data on earnings guidance, executive compensation, stock
prices, and financial statements. We obtain earnings guidance data from I/B/E/S guidance for the period of 1992 to 2016. We remove forecasts for quarterly earnings and keep those for annual earnings. We also exclude financial firm based on SIC from 6000 to 6999. Analyst forecasts data is obtained from I/B/E/S detail history. ExecuComp covers top managers’ compensation from 1992 to present, and we use its data to compute managerial risk-taking incentives. Two top managers, CEO and CFO, are examined. Consistent with Jiang, Petroni, and Wang (2010), CEOs are identified using an ExecuComp indicator (CEOANN=”CEO”), and CFOs are recognized using the annual title (TITLEANN). If the annual title contains CFO, chief financial officer, treasurer, controller, finance, and vice president-finance, we regard the corresponding manager as a CFO. However,
9
this lookup process will result in multiple CFOs for some firms. To make sure there is only one CFO for each firm-year, we manually compare those multiple-CFO observations with annual reports and find out who is the real CFO. Our sample of financial data is drawn from annual Compustat. We limit the beginning of financial data to 1992 because the earliest data of earnings guidance and executive compensation is for 1992. Financial data contributes to our control variables. Since some control variables need to be adjusted by stock price and stock returns will be used for calculation, we combine daily and monthly CRSP for our analysis. Finally, we have 16,624 firm-years for analyzing the propensity of forecast issuance and the frequency of forecast issuance, and 11,230 to 11,357 firm-years for studying the informativeness of management earnings forecasts. We winsorize each variable at 1% and 99% of its distribution by years. For panel regressions, year fixed effects are controlled, and tstatistics are based on standard errors clustered at the two-SIC industry-level. 3.2 3.2.1
Variable measurement Management earnings guidance Our primary concern is the relation between managerial risk-taking incentives and
earnings guidance. Following prior literature (Lang and Lundholm, 1996; Cheng, Luo, and Yue, 2013; Ali, Klasa, and Yeung, 2014; Ittner and Michels, 2017), we use three variables to measure the propensity to issue earnings forecasts, the frequency of management earnings forecasts, and the horizon of management earnings forecasts. We also use three variables to measure the precision of management earnings forecasts. By using these six measures, we can explore different roles played by CEOs and CFOs in management forecasts. The propensity to provide earnings forecast, Propensity, is measured as a dummy
10
variable that is equal to one if the firm provides an annual earnings guidance in a fiscal year and zero otherwise. Since our data is collected from I/B/E/S guidance database and firms are not forced to provide earnings guidance, our research suffers a potential selection bias if we do not address the propensity to issue earnings guidance. Meanwhile, making earnings forecasts is an efficient way to convey information and consequently influence analyst forecasts and investor decisions. Our second variable is Frequency, which captures the number of earnings guidance made during a fiscal year for annual earnings. As we discussed in section 2, the frequency of forecasts is related to higher level of risks. Because this variable is truncated at zero, we follow Ali, Klasa, and Yeung (2014) to use Tobit model for the related regressions. Horizon is the average number of days between earnings forecasts and the end date of the fiscal year. We expect increasing uncertainty if the horizon is longer. Our three measures for the precision of management earnings forecasts are Management Errors, Precision, and Width. Following prior literature, we define the first variable as the absolute value of the average difference between actual earnings and management forecasts and the value is adjusted by the share price three days prior to the forecast. However, this measure has one weakness that it cannot faithfully capture precision of those one-bound forecasts. For example, if the actual earning is $2 per share and the average forecast is “$1 per share or more”, the management, in fact, made a correct forecast but the difference is two times of the forecast value. Hence, we develop the second measure, a dummy variable, as a supplement. This dummy variable is equal to one if the actual earning is within the forecast range and zero otherwise. For point forecasts, we consider a forecast is precise if the actual earning is within the range from 95 percent to 105 percent of the forecast value. Consistent with Ittner and Michels (2017), we define
11
that forecast width is equal to the absolute value of forecast range adjusted by stock price and zero is assigned to point forecasts. 3.2.2
Managerial risk-taking incentives We measure managerial risk-taking incentives using portfolio vega rather than delta
because Armstrong et al. (2013) suggest that vega is one unambiguous measure of risk preference. Following Core can Guay (2002), we define vega as the change in dollar value of the executive’s equity portfolio for one percent change in the firm’s stock return volatility. To remove skewness, we follow Chava and Purnanandam (2010) and Armstrong et al. (2013) and use the natural logarithm of one plus vega (ln(1+vega)) as our key explanatory variable in the regression models. 3.2.3
Measures of earnings management Prior literature suggests that management earnings forecasts are potentially associated
with earnings manipulation. So, we control for earnings management measures in the regression models testing the impact of incentives on the precision of forecast. We first use discretionary accruals to replicate those accrual-based studies on earnings management and disclosure accuracy. We follow Bergstresser and Philippon (2006) to estimate discretionary accruals as the residuals of the following regression model:
𝑇𝐴𝑖,𝑡 = 𝛼0 + 𝛼1 × (
1 𝐴𝑖,𝑡−1
) + 𝛼2 × (∆𝑅𝐸𝑉𝑖,𝑡 ) + 𝛼3 × (𝑃𝑃𝐸𝑖,𝑡 ) + 𝜀𝑖,𝑡
(1)
The variable Ai,t-1 is the lagged total assets (Compustat item 6) for firm i at time t-1. ∆REVi,t is the change in sales (Compustat item 12) adjusted by lagged total assets. PPEi,t is gross property, plant and equipment (Compustat item 7) adjusted by lagged total assets. TAi,t refers to
12
the total accruals and is calculated as follows:
𝑇𝐴𝑖,𝑡 = (∆𝐶𝐴𝑖,𝑡 − ∆𝐶𝐿𝑖,𝑡 − ∆𝐶𝑎𝑠ℎ𝑖,𝑡 + ∆𝑆𝑇𝐷𝑖,𝑡 − 𝐷𝑒𝑝𝑖,𝑡 )/𝐴𝑖,𝑡−1
(2)
∆CAi,t is the one-year change in the current assets (Compustat item 4) and ∆CLi,t is the one-year change in the current liabilities (Compustat item 5). ∆Cashi,t represents the one-year change in cash and short-term investments (Compustat item 1). ∆STDi,t is the one-year change in debt in current liabilities (Compustat item 34). The last one, Depi,t, represents the depreciation and amortization expense (Compustat item 14). Because accrual-based earnings management activities decrease after the passage of Sarbanes-Oxley Act (SOX) and real earnings management activities increase after the passage of SOX (Cohen, Dey, and Lys, 2008), we consider three real earnings management methods: abnormal operating cash flows (ROCF), abnormal production costs (RPROD), and abnormal discretionary expenses (RDISX). They are obtained as the residuals from the following models:
𝑂𝐶𝐹𝑖,𝑡 𝐴𝑠𝑠𝑒𝑡𝑠𝑖,𝑡−1
= 𝛽1
1 𝐴𝑠𝑠𝑒𝑡𝑠𝑖,𝑡−1
+ 𝛽2
𝑆𝑎𝑙𝑒𝑠𝑖,𝑡 𝐴𝑠𝑠𝑒𝑡𝑠𝑖,𝑡−1
+ 𝛽3
∆𝑆𝑎𝑙𝑒𝑠𝑖,𝑡 𝐴𝑠𝑠𝑒𝑡𝑠𝑖,𝑡−1
+ 𝜀𝑖,𝑡
(3)
OCFi,t represents the normal operating cash flows for firm i at time t, which is calculated as the difference between the net cash flows from operating activities (Compustat item 308) and cash flows from extraordinary items and discontinued operations (Compustat item 124).
𝑃𝑅𝑂𝐷𝑖,𝑡 𝐴𝑠𝑠𝑒𝑡𝑠𝑖,𝑡−1
= 𝛽1
1 𝐴𝑠𝑠𝑒𝑡𝑠𝑖,𝑡−1
+ 𝛽2
𝑆𝑎𝑙𝑒𝑠𝑖,𝑡 𝐴𝑠𝑠𝑒𝑡𝑠𝑖,𝑡−1
13
+ 𝛽3
∆𝑆𝑎𝑙𝑒𝑠𝑖,𝑡 𝐴𝑠𝑠𝑒𝑡𝑠𝑖,𝑡−1
+ 𝛽4
∆𝑆𝑎𝑙𝑒𝑠𝑖,𝑡−1 𝐴𝑠𝑠𝑒𝑡𝑠𝑖,𝑡−1
+ 𝜀𝑖,𝑡 (4)
PRODi,t refers to production costs, which are defined as the sum of GOGS (Compustat item 41) and change in inventory (Compustat item 3).
𝐷𝐼𝑆𝑋𝑖,𝑡 𝐴𝑠𝑠𝑒𝑡𝑠𝑖,𝑡−1
= 𝛽1
1 𝐴𝑠𝑠𝑒𝑡𝑠𝑖,𝑡−1
+ 𝛽2
𝑆𝑎𝑙𝑒𝑠𝑖,𝑡 𝐴𝑠𝑠𝑒𝑡𝑠𝑖,𝑡−1
+ 𝜀𝑖,𝑡
(5)
DISXi,t is the normal level of discretionary expenses, and it is defined as the sum of advertising expenses (Compustat item 45), R&D expenses (Compustat item 46), and SG&A (Compustat item 189). 3.2.4
Other control variables A variety of variables that prior research finds to influence earnings guidance quality is
controlled. Variables measuring firm-level characteristics include Size, which is calculated as the natural logarithm of firm’s total assets, ROA, which is return on assets and defined as income before extraordinary items divided by total assets, MB, which refers to market-to-book ratio and is calculated as the sum of the book value of debts and the market value of common equity divided by total assets, R&D, which represents R&D expenses, and Leverage, which is total liabilities scaled by total assets. Following Jiang, Petroni, and Wang (2010), we consider Age as the number of years that a firm has been in annual Compustat database since the first occurrence. The firm’s stock returns volatility, Volatility, is the annualized standard deviation of monthly stock returns over a fiscal year. EPS volatility is the standard deviation of the firm’s actual earnings per share over the prior five years, and we require that there are at least three-year observations. ∆EPS is the absolute change in annual earnings per share deflated by stock price. Consistent with Ali, Klasa, and Yeung (2014), we control for the industry Herfindahl-Hirschman index, HHI, which is calculated for the firm’s two-digit SIC industry by summing up the squared
14
market shares of each individual firm in the industry and the market share is defined as the percentage of an industry total sales earned by a particular firm over a fiscal year. Three analystrelated variables are controlled. Analyst error is the absolute value of the difference between analyst forecast mean and the actual number, scaled by stock price. Analyst volatility is the standard deviation of analyst forecast mean for each time over a fiscal year. Analyst coverage refers to the average number of analysts making earnings forecasts for a particular firm over a fiscal year. A dummy variable, SOX, is equal to one if the year for the observations is 2002 or later and zero otherwise. This dummy variable is designed to capture the effect of SOX act. Litigation risk is also a dummy variable taking a value of one if the firm is in SIC industries 2833-3836, 3507-3577, 3600-3674, 5200-5961, or 7370-7374, which are industries with high litigation risk and zero otherwise. [Insert Table 1 Here] Table 1 provides descriptive statistics of the variables with the mean, median, and standard deviation. The average adjusted vega of CEOs is significantly greater than that of CFOs and the difference is 1.063, which is significant at 1% level.
4
Empirical results We have an unbalanced panel data from 1992 to 2016. Because data of 1992 is
insufficient in most of the specifications, it is automatically omitted. Therefore, our efficient sample starts from 1993. We include year fixed effects to remove year-specific unobservable factors, and standard errors are clustered at the level of two-digit SIC codes to account for the possibility of correlations across observations of the same industry in different years. We estimate three models for each outcome. In the first model, we include CEO vega and
15
control variables, and in the second model, we replace CEO vega with CFO vega. In the third model, we include both CEO vega and CFO vega. Given the high correlation coefficient between CEO vega and CFO vega, our approach is helpful to avoid multicollinearity problem. [Insert Table 2 Here] Table 2 reports Pearson correlation coefficients between explanatory variables and dependent variables. The significantly positive coefficients on both CEO vega and CFO vega in column 1 and column 2 indicate that the propensity to issue earnings guidance and the frequency of earnings guidance are positively associated with managerial risk-taking incentives. The significantly negative coefficients on both CEO and CFO risk preferences through column 3 to column 5 suggest that managerial risk-taking incentives are negatively related to the precision of forecast. Column 6 confirms that if managers are more sensitive to the increase in volatility, the horizon between earnings guidance date and earnings announcement date will be longer. Overall, this evidence is consistent with our predictions. To compare the heterogeneous impacts of CEO vega and CFO vega, we first split our sample firm into four groups: the firms whose CEO vega is high and CFO vega is low (HELF), the firms whose CEO vega is low and CFO vega is high (LEHF), the firms whose CEO vega is high and CFO vega is high (HEHF), and the firms whose CEO vega is low and CFO vega is low (LELF). For each year, we consider a CEO vega (CFO vega) is high if CEO vega (CFO vega) is greater than its industry median; a low vega means that it is less than the industry median. [Insert Table 3 Here] In table 3, we compare HELF firms and LELF firms year by year. Different from lowCEO-vega firms, high-CEO-vega firms have significantly higher likelihood to issue earnings guidance and issue more earnings guidance. However, these two groups of firms have no
16
significant difference in the precision of forecast and horizon. [Insert Table 4 Here] Table 4 reports the comparison between LEHF firms and LELF firms. Besides the influence on the first two outcomes, high CFO vega firms have less precise forecasts and longer horizon than low CFO vega firms. So, this alternative approach also provides evidence to support our hypotheses. 4.1
The likelihood of issuance We start with a Probit regression to analyze the relationship between managerial risk-
taking incentives and the propensity to issue earnings guidance. Table 5 reports regression results. After controlling for firm-specific characteristics, risk factors, and analyst factors, the significantly positive coefficients on CEO vega and CFO vega are reported in column 1 and column 2, respectively. This indicates that both CEO and CFO risk preferences are positively associated with issuance of earnings guidance. The marginal effect of CEO vega is close to that of CFO vega. However, when two managers’ risk-taking incentives are jointly included in column 3, the coefficient on CFO vega becomes insignificant. This result suggests that in a firm CEO risk preference is the determinant of issuing earnings guidance. It seems contradictory to the findings reported in table 4. However, this inconsistency may result from the CEO power over the CFO. Because CEOs have significantly higher vega than CFOs, it is possible that CEO vega is still higher than CFO vega even the former is less than the industry median. Thus, CEOs are likely to force their CFOs to generate earnings forecasts to realize their own risk preferences. We believe these regression results are consistent with our hypotheses. [Insert Table 5 Here] 4.2
The frequency of earnings guidance
17
Next, we investigate the relationship between the frequency of earnings guidance issuance and managerial risk preferences. Because the number of earnings forecasts in a year is truncated at zero, we follow Ali, Klasa, and Yeung (2014) to use Tobit model for the regressions and control for same variables in section 4.1 to compare the independent influence of CEO vega and CFO vega. Table 6 reports regression results. When CEO vega and CFO vega are considered simultaneously, the relation between CFO vega and the frequency becomes statistically insignificant. [Insert Table 6 Here] Our findings on the propensity to issue earnings guidance and the frequency of earnings guidance support Chava and Purnanandam (2010) that CEOs’ risk-taking incentives significantly influence broader corporate decisions. 4.3
The horizon of issuance Table 7 represents results from using the days between issuing an earnings guidance and
announcing actual earnings to capture the horizon of management earnings forecasts. We find a significant and positive relation between CFO vega and the horizon. Different from columns 1 to 3, columns 4 to 6 cluster standard errors at two dimensions – industry level and year level. Then, even the coefficient on CEO vega is negative and significant at 10% level in column 3, it becomes insignificant in column 6, and the coefficient on CFO vega is still positive and significant at 1% level from being it is significant at 5% level in column 3. [Insert Table 7 Here] These results suggest that CFO risk-taking incentives are closely associated with issuance timing. If CFOs are more sensitive to the change in volatility, they will choose to issue guidance as early as possible. This action will result in adjustments on investor expectations or analyst
18
expectations, and, in turn, increase the market uncertainty. 4.4
The precision of forecasts We use three variables to measure the precision of earnings guidance and introduce four
lagged proxies for earnings management in the regressions. In table 8, we define precision of forecasts as the absolute value of the average difference between forecast values and the actual values. We do not find any significant relation between forecast precision and managerial risktaking incentives. Further, the coefficients on CEO vega and CFO vega are almost zero. One interesting finding is the coefficient on abnormal operating cash flows (ROCF) are significantly negative, indicating that real earnings management is related to less precise earnings forecasts for the next year. [Insert Table 8 Here] Our second measure is a dummy variable that is equal to one if the actual earnings fall within the forecast range and zero otherwise. Table 9 represents Probit regression results. We find a weak negative association between CFO vega and forecast precision. This weak relation also exists in table 10 when we use the forecast range as the proxy for forecast precision. [Insert Table 9 Here] [Insert Table 10 Here] Collectively, our empirical results in part support Chava and Purnanandam (2010) that CFOs play an essential role when decisions are related to specialized knowledge of finance. However, it also raises a question whether CFO risk-taking incentives, in practice, are strong enough to affect forecast precision given the weak relationship we observed.
5
Conclusion
19
In this paper, we investigate the relationship between managerial risk-taking incentives and management earnings guidance. Using data of forecasts for annual earnings per share, we find that managerial risk preference significantly influences disclosure features. Our evidence suggests that CEO risk-taking incentives are positively related to managers’ propensity to issue annual earnings forecasts and the frequency of the issuance. CFO risk-taking incentives can lower the precision of forecasts and significantly lead to the longer horizon of management earnings forecasts. Our findings are consistent with Chava and Purnanandam’s (2010) explanation that CFOs have more impact than CEOs on decisions involving more specialized knowledge of finance. To some extent that our results provide support to the voice for eliminating short-term earnings guidance (Choi et al., 2011). Further research can be done with the deep investigation of other factors, such as the balance between CEOs and CFOs. According to a recent study on managerial balance and earnings management (Wu, 2017), the effects of the shifts in power between CEOs and CFOs may also influence corporate disclosure policy.
20
References Acito, A., 2010. Does quarterly earnings guidance increase or reduce earnings management? Dissertation. Agapova, A., Madura, J., 2016. Market uncertainty and earnings guidance. The Quarterly Review of Economics and Finance 61, 97-111. Ali, A., Klasa, S., Yeung, E., 2014. Industry concentration and corporate disclosure policy. Journal of Accounting and Economics 58, 240-264. Armstrong, C., Larcker, D., Ormazabal, G., and Taylor, D, 2013. The relation between equity incentives and misreporting: The role of risk-taking incentives. Journal of Financial Economics 109, 327-350. Bergstresser, D., Philippon, T., 2006. CEO incentives and earnings management. Journal of Financial Economics 80, 511-529. Billings, M., Jennings, R., Lev, B., 2015. On guidance and volatility. Journal of Accounting and Economics 60, 161-180. Brochet, F., Loumioti, M. and Serafeim, G., 2015. Speaking of the short-term: disclosure horizon and managerial myopia. Review of Accounting Studies20(3), 1122-1163. Chava, S., Purnanandam, A., 2010. CEOs versus CFOs: incentives and corporate policies. Journal of Financial Economics 97, 263-278. Chen, S., Matsumoto, D., Rajgopal, S., 2011. Is silence golden? An empirical analysis of firms that stop giving quarterly earnings guidance. Journal of Accounting and Economics 51, 134150. Cheng, Q., Luo, T., Yue, H., 2013. Managerial incentives and management forecast precision. The Accounting Review 88(5), 1575-1602. Cheng, M., Subramanyam, K. R., Zhang, Y. 2005. Earnings guidance and managerial myopia. Working paper. Choi, J., Myers, L., Zang, Y., Ziebart, D., 2011. Do management EPS forecasts allow returns to reflect future earnings? Implications for the continuation of management’s quarterly earnings guidance. Review of Accounting Studies 16, 143–182. Cohen, D., Dey, A., Lys, T., 2008. Real and accrual-based earnings management in the pre- and post-Sarbanes Oxley periods. The Accounting Review 83, 757-787. Cohen, D., Zarowin, P., 2010. Accrual-based and real earnings management activities around seasoned equity offerings. Journal of Accounting and Economics 50(1), 2-19.
21
Coles, J., Daniel, N., Naveen, L., 2006. Managerial incentives and risk-taking. Journal of Financial Economics 79, 431-468. Coller, M., Yohn, T., 1997. Management forecasts and information asymmetry: an examination of bid-ask spreads. Journal of Accounting Research 35(2), 181–191. Core, J., Guay, W., 2002. Estimating the value of employee stock option portfolios and their sensitivities to price and volatility. Journal of Accounting Research 40, 613-630. Dechow, P., Richardson, S., Tuna, A., 2003. Why are earnings kinky? An examination of the earnings management explanation. Review of Accounting Studies 8 (2-3), 355-384. DeFond, M., Subramanyam, K., 1998. Auditor Changes and Discretionary Accruals. Journal of Accounting and Economics 25, 35–67. Dong, Z., Wang, C., Xie, F., 2010. Do executive stock options induce excessive risk taking?. Journal of Banking & Finance 34(10), 2518-2529. Fama, E., French, K., 1997. Industry costs of equity. Journal of Financial Economics 43, 153193. Feng, M., Ge, W., Luo, S., Shevlin, T., 2011. Why do CFOs become involved in material accounting manipulations? Journal of Accounting and Economics 51, 21-36. Hamm, S., Li, E., Ng, J., 2012. Management earnings guidance and stock price crash risk. Working paper. Healy, P., Palepu, K., 2001. Information asymmetry, corporate disclosure, and the capital markets: are view of the empirical disclosure literature. Journal of Accounting and Economics. 31, 405–440. Houston, J., Lev, B., Tucker, J., 2010. To guide or not to guide? Causes and consequences of stopping quarterly earnings guidance. Contemporary Accounting Research 27(1), 143-185. Hu, B., Hwang, J., Jiang, C., 2014. The impact of earnings guidance cessation on information asymmetry. Journal of Business Finance and Accounting 41, 73-99. Ittner, C., Michel, J., 2017. Risk-based forecasting and planning and management earnings forecasts. Review of Accounting Studies 22, 1005-1047. Jiang, J., Petroni, K., Wang, I., 2010. CFOs and CEOs: who have the most influence on earnings management? Journal of Financial Economics 96, 513-526. Kim, JB., Li Y., Zhang, L., 2011. CFOs versus CEOs: Equity incentives and crashes. Journal of Financial Economics 101, 713-730.
22
Kim, K., Patro S., Pereira, R., 2017. Option incentives, leverage, and risk-taking. Journal of Corporate Finance 43, 1-18. Koh, K., Matsumoto, D., Rajgopal, S., 2008. Meeting or beating analyst expectations in the postscandals world: changes in stock market rewards and managerial actions. Contemporary Accounting Research 25(4), 1067-1098. Kraft, A., Lee, B., Lopatta, K., 2014. Management earnings forecasts, insider trading, and information asymmetry. Journal of Corporate Finance 26, 96-123. Krishnan, G., Pevzner, M., Sengupta, P., 2012. How do auditors view managers’ voluntary disclosure strategy? The effect of earnings guidance on audit fees. Journal of Accounting and Public Policy 31, 492-515. Lang, M., Lundholm, R., 1996. Corporate disclosure policy and analyst behavior. The Accounting Review 71(4), 467-492. Rogers, J., Skinner, D., Van Buskirk, A., 2009. Earnings guidance and market uncertainty. Journal of Accounting and Economics 48(1), 90-109. Skinner, D., 1997. Earnings disclosures and stockholder lawsuits. Journal of Accounting and Economics. 23, 249–282. Wu, Q., 2017. Managing earnings for my boss? Financial reporting and the balance of power between CEOs and CFOs. Working paper. Yang, H., 2012. Capital market consequences of managers’ voluntary disclosure styles. Journal of Accounting and Economics 53, 167-184.
23
Table 1. Descriptive statistics Variable
N
Mean
Std. Dev
Q25
Median
Q75
Propensity
31,165
0.364
0.481
0
1
1
Frequency
31,165
1.655
2.740
0
0
3.000
Width
11,352
0.003
0.006
0.001
0.002
0.004
Management error
11,230
0.013
0.043
0.001
0.003
0.009
Precision
11,357
0.266
0.442
0.000
0.000
1.000
Horizon
11,357
205
75
176
201
237
CEO vega
27,000
3.507
1.853
2.456
3.757
4.836
CFO vega
27,097
2.444
1.461
1.437
2.528
3.483
CEO vega - CFO vega
26,655
1.063***
1.113
Size
28,225
7.286
1.579
6.146
7.186
8.327
ROA
28,214
0.036
0.114
0.016
0.047
0.085
MB
27,758
1.984
1.358
1.203
1.559
2.237
R&D
16,624
0.051
0.067
0.004
0.025
0.074
Leverage
28,161
0.533
0.234
0.372
0.534
0.676
Age
28,238
3.081
0.733
2.565
3.135
3.738
Volatility
27,628
0.391
0.231
0.235
0.334
0.480
Analyst error
24,869
0.034
0.197
0.002
0.004
0.012
Analyst volatility
24,974
0.249
0.658
0.048
0.103
0.227
Analyst coverage
25,167
2.337
0.810
1.792
2.398
2.944
EPS volatility
25,161
0.846
1.960
0.199
0.380
0.754
ΔEPS
24,398
0.069
0.369
0.004
0.010
0.026
HHI
31,165
0.066
0.062
0.032
0.043
0.075
ROCF
28,015
0.162
0.482
-0.009
0.076
0.247
RPROD
27,587
-0.076
0.516
-0.178
-0.035
0.060
RDISX
25,673
-0.186
2.252
-0.456
-0.109
0.064
DA
27,800
1.051
5.764
0.037
0.118
0.499
SOX
31,165
0.696
0.460
0
1
1
Litigation Risk
31,165
0.317
0.465
0
0
1
Panel A: disclosure variables
Panel B: key variables
Panel C: control variables
This table provides detailed descriptive statistics information for a full sample from 1992 to 2016. Propensity is a dummy variable that equals one if the firm provides an annual earnings guidance in a fiscal year and zero otherwise. Frequency is the number of earnings guidance made during a fiscal year for annual earnings of that fiscal year. Horizon is the average number of days between earnings forecasts and the end date of the fiscal year. Management error is the absolute value of the average difference between actual earnings and management forecasts, and the value is adjusted by the share price three days before the
24
forecast. Precision is a dummy variable that equals one if the actual earnings are within the forecast range. Width refers to the absolute value of forecast range adjusted by stock price, and zero is assigned to point forecasts. CEO vega and CFO vega are the natural logs of one plus Core and Guay (2002) vega. Measurements for earnings management include discretionary accruals (DA), abnormal operating cash flows (ROCF), abnormal production cost (RPROD), and abnormal discretionary expenses (RDISX). Size is calculated as the natural logarithm of firm’s total assets. ROA is return on assets and defined as income before extraordinary items divided by total assets. MB refers to market-to-book ratio and is calculated as the sum of the book value of debts and the market value of common equity divided by total assets. R&D represents R&D expenses, and Leverage is total liabilities scaled by total assets. Age is the number of years that a firm has been in annual Compustat database since the first occurrence. Volatility is the annualized standard deviation of monthly stock returns over a fiscal year. EPS volatility is the standard deviation of firm’s actual earnings per share over the prior five years, and we require that there are at least three-year observations. ∆EPS is the absolute change in annual earnings per share deflated by stock price. The industry HerfindahlHirschman index, HHI, which is calculated for the firm’s two-digit SIC industry by summing up the squared market shares of each individual firm in the industry and the market share is defined as the percentage of an industry total sales earned by a particular firm over a fiscal year. Analyst error is the absolute value of the difference between analyst forecast mean and actual number, scaled by stock price. Analyst volatility is the standard deviation of analyst forecast mean for each time over a fiscal year. Analyst coverage refers to the average number of analysts making earnings forecasts for a particular firm over a fiscal year. A dummy variable, SOX, is equal to one if the year of the observation is 2002 or later and zero otherwise. Litigation risk is also a dummy variable taking a value of one if the firm is in SIC industries 2833-3836, 3507-3577, 3600-3674, 5200-5961, or 73707374, which are industries with high litigation risk and zero otherwise. We winsorize each variable at 1% and 99% of its distribution by years. ***, **, and *** indicate that the coefficient is statistically different from zero at 1%, 5%, and 10% significance level, respectively.
25
Table 2. Pearson correlation coefficients Variable
Propensity
Number
Management Precision Error
Width
Horizon
Panel A: key variables CEO vega CFO vega
0.204 0.186
***
0.179 0.158 0.037 -0.160 0.042 0.088 -0.179 -0.103 -0.127 0.133 -0.128 -0.106 0.006 0.056 -0.035 -0.021 0.016 0.274 0.026
***
***
0.215 0.196
***
0.226 0.137 0.028 -0.137 0.063 0.120 -0.208 -0.087 -0.108 0.163 -0.102 -0.089 0.004 0.063 -0.034 -0.027 0.023 0.321 0.014
***
***
-0.095 -0.130
***
-0.041 -0.256 -0.216 0.025 0.098 0.028 0.189 0.376 0.400 -0.126 0.353 0.327 0.049 -0.080 0.056 0.042 0.006 0.110 -0.033
***
***
-0.039 -0.058
***
-0.058 -0.232 -0.105 0.066 0.030 -0.040 0.168 0.386 0.402 -0.070 0.338 0.298 0.032 -0.051 0.030 0.010 -0.007 -0.048 0.007
***
***
-0.014 -0.017
*
0.060 0.074
***
0.052 -0.032 0.033 0.049 0.008 -0.008 0.043 0.001 0.020 0.093 0.007 -0.007 -0.022 0.037 -0.004 -0.020 -0.001 0.034 -0.011
***
***
Panel B: control variables Size ROA MB R&D Leverage Age Volatility Analyst error Analyst volatility Analyst coverage EPS volatility ΔEPS HHI ROCF RPROD RDISX DA SOX Litigation Risk
*** *** *** *** *** *** *** *** *** *** ***
*** *** *** *** *** ***
*** *** *** *** *** *** *** *** *** *** ***
*** *** *** *** *** ***
*** *** ** *** *** *** *** *** *** *** *** *** *** *** ***
*** ***
*** *** *** *** *** *** *** *** *** *** *** *** *** ***
***
0.099 0.016 0.003 -0.004 0.055 0.076 -0.077 -0.023 -0.042 0.036 -0.033 -0.016 -0.040 -0.012 0.006 -0.009 -0.025 -0.080 -0.040
*** *
*** *** *** ** *** *** *** * ***
*** *** ***
*** *** ***
***
** ***
** ***
*
***
This table describes correlations for a full sample from 1992 to 2016. Propensity is a dummy variable that equals one if the firm provides an annual earnings guidance in a fiscal year and zero otherwise. Frequency is the number of earnings guidance made during a fiscal year for annual earnings of that fiscal year. Horizon is the average number of days between earnings forecasts and the end date of the fiscal year. Management error is the absolute value of the average difference between actual earnings and management forecasts, and the value is adjusted by the share price three days before the forecast. Precision is a dummy variable that equals one if the actual earnings are within the forecast range. Width refers to the absolute value of forecast range adjusted by stock price, and zero is assigned to point forecasts. CEO vega and CFO vega are the natural logs of one plus Core and Guay (2002) vega. Measurements for earnings management include discretionary accruals (DA), abnormal operating cash flows (ROCF), abnormal production cost (RPROD), and abnormal discretionary expenses (RDISX). Size is calculated as the natural logarithm of firm’s total assets. ROA is return on assets and defined as income before extraordinary items divided by total assets. MB refers to market-to-book ratio and is calculated as the sum of the book value of debts and the market value of common equity divided by total assets. R&D represents R&D expenses, and Leverage is total liabilities scaled by total assets. Age is the number of years that a firm has been in annual Compustat database since the first occurrence. Volatility is the annualized standard deviation of monthly stock returns over a fiscal year. EPS volatility is the standard deviation of firm’s actual earnings per share over the prior five years, and we require that there are at least three-year observations. ∆EPS is the absolute change in annual earnings per share deflated by stock price. The industry Herfindahl-Hirschman index, HHI, which is calculated for the firm’s two-digit SIC industry by summing up the squared market shares of each individual firm in the industry and the market share is defined as the percentage of an industry total sales earned by a particular firm over a fiscal year. Analyst error is the absolute value of the difference between analyst forecast mean and actual number, scaled by stock price. Analyst volatility is the standard deviation of analyst forecast mean for each time over a fiscal year. Analyst coverage refers to the average number of analysts making earnings forecasts for a particular firm over a fiscal year. A dummy variable, SOX, is equal to one if the year of the observation is 2002 or later and zero otherwise. Litigation risk is also a dummy variable taking a value of one if the firm is in SIC industries 2833-3836, 3507-3577, 3600-3674, 5200-5961, or 7370-7374, which are industries with high litigation risk and zero otherwise. We winsorize each variable at 1% and 99% of its distribution by years. ***, **, and *** indicate that the coefficient is statistically different from zero at 1%, 5%, and 10% significance level, respectively.
26
27
0.035
0.023
0.107
0.245
0.198
0.293
0.419
0.516
0.534
0.571
0.579
0.594
0.577
0.172
0.421
0.409
0.476
0.489
0.514
0.468
0.504
0.427
1995
1996
1997
1998
1999
20 0 0
20 0 1
20 0 2
20 0 3
20 0 4
20 0 5
20 0 6
20 0 7
20 0 8
20 0 9
20 10
20 11
20 12
20 13
20 14
20 15
20 16
0.341
0.300
0.313
0.298
0.337
0.294
0.284
0.282
0.353
0.391
0.410
0.383
0.409
0.403
0.355
0.382
0.191
0.160
0.127
0.070
0.073
0.028
0.014
0.806*
0.204***
0.155***
0.216***
0.152***
0.182***
0.125***
0.139***
0.119**
0.185***
0.184***
0.196***
0.162***
0.131**
0.161***
0.037
0.102**
0.038
0.118**
0.037
0.050**
0.007
0.025
HELF-LELF
2.127
2.451
2.459
2.458
2.305
2.303
2.106
2.016
2.323
3.321
3.376
3.017
2.262
2.029
1.887
1.103
0.500
0.323
0.330
0.136
0.023
0.047
0.039
1.513
1.429
1.420
1.359
1.516
1.346
1.432
1.310
1.725
1.883
2.126
1.810
1.655
1.535
1.196
0.953
0.297
0.245
0.164
0.076
0.087
0.028
0.014
LELF
Frequency HELF
0.614**
1.022***
1.039***
1.099***
0.789***
0.957***
0.674**
0.706***
0.598**
1.438***
1.250***
1.207***
0.607**
0.494**
0.691***
0.150
0.203**
0.0779
0.166**
0.06
0.064***
0.019
0.025
HELF-LELF
0.003
0.003
0.003
0.003
0.004
0.004
0.005
0.009
0.005
0.003
0.003
0.005
0.005
0.004
0.003
0.003
0.001
0.001
0.001
0.000
0.000
0.000
0.000
0.004
0.003
0.004
0.005
0.006
0.004
0.005
0.008
0.005
0.004
0.003
0.004
0.004
0.005
0.004
0.004
0.002
0.001
0.002
0.001
0.001
0.001
0.013
LELF
Width HELF
-0.001
0.000
0.000
-0.002**
-0.002*
0.000
0.000
0.002
0.001
0.000
-0.001
0.001
0.000
-0.002
-0.001
-0.001
-0.001
0.000
-0.001
-0.001
-0.001**
-0.001
-0.013
HELF-LELF
0.006
0.005
0.009
0.007
0.009
0.008
0.006
0.043
0.012
0.014
0.012
0.032
0.018
0.023
0.040
0.039
0.017
0.009
0.007
0.001
0.001
0.001
0.006
HELF
0.007
0.006
0.008
0.012
0.016
0.009
0.009
0.029
0.015
0.017
0.014
0.023
0.022
0.025
0.019
0.034
0.018
0.017
0.013
0.006
0.009
0.016
0.068
LELF
Management Error
-0.002
-0.001
0.001
0.005*
-0.007**
-0.001
-0.003**
0.014
-0.003
-0.002
-0.002
0.009
-0.004
-0.002
0.021
0.005
-0.001
-0.008
-0.006
-0.004**
-0.009**
-0.015
-0.062
HELF-LELF
0.298
0.246
0.216
0.346
0.266
0.246
0.259
0.208
0.183
0.127
0.203
0.157
0.292
0.273
0.280
0.245
0.387
0.368
0.304
0.636
0.500
0.667
0.000
0.243
0.244
0.237
0.209
0.304
0.256
0.277
0.197
0.247
0.170
0.162
0.177
0.221
0.234
0.287
0.273
0.336
0.385
0.420
0.611
0.389
0.462
0.500
LELF
Precision HELF
0.055
0.001
-0.021
0.136**
-0.038
-0.010
-0.017
0.011
-0.063
-0.043
0.041
-0.020
0.071
0.039
-0.006
-0.028
0.051
-0.017
-0.116
0.025
0.111
0.205
0.500**
HELF-LELF
208.000
208.500
197.900
194.800
199.000
193.500
196.800
195.900
210.700
207.200
210.100
199.400
223.200
219.800
202.300
186.500
216.800
185.400
147.700
153.000
51.000
76.167
120.500
195.700
202.500
190.100
199.000
184.600
195.300
185.700
197.300
204.100
202.200
195.700
209.800
217.400
226.000
231.400
205.800
187.000
213.700
154.300
161.900
201.300
124.000
71.125
LELF
Horizon HELF
12.300
6.000
7.800
-4.200
14.400
-1.800
11.100
-1.400
6.600
5.000
14.400
-10.400
5.800
-6.200
-29.100
-19.300
29.800
-28.300
-6.600
-8.900
-150.300
-47.833
49.375
HELF-LELF
1.700 201.400 199.700 0.002 0.247 0.249 0.000 0.017 0.017 -0.001* 0.004 0.004 0.739*** 1.105 1.844 0.146*** 0.267 0.412 Total This table provides mean difference test results for full sample by years. Propensity is a dummy variable that equals one if the firm provides an annual earnings guidance in a fiscal year and zero otherwise. Frequency is the number of earnings guidance made during a fiscal year for annual earnings of that fiscal year. Horizon is the average number of days between earnings forecasts and the end date of the fiscal year. Management error is the absolute value of average difference between actual earnings and management forecasts and the value is adjusted by the share price three days before the forecast. Precision is a dummy variable that equals one if the actual earnings are within the forecast range. Width refers to the absolute value of forecast range adjusted by stock price and zero is assigned to point forecasts. HELF represents the firms with highvega CEOs (vega is higher than annual median by industries) and low-vega CFOs (vega is lower than annual median by industries). LELF represents the firms with low-vega CEOs (vega is lower than annual median by industries) and low-vega CFOs (vega is lower than annual median by industries). ***, **, and *** indicate that the coefficient is statistically different from zero at 1%, 5%, and 10% significance level, respectively.
0.039
1994
LELF
Propensity
HELF
Table 3. Mean difference tests by years for high-vega CEO/low-vega CFO firms and low-vega CEO/low-vega CFO firms
28
73.200 -40.900
154.300 213.700
227.500 172.800
0.125 -0.052
0.420 0.385
0.546 0.333
-0.009** -0.009**
0.017
0.008
0.000
0.001
0.001
0.042
0.245
0.287
0.000
0.160
0.160
0.187
0.398
0.454
0.433
0.452
0.464
0.504
0.503
0.510
0.423
0.396
0.440
0.418
0.493
0.488
0.387
0.290
1999
20 0 0
20 0 1
20 0 2
20 0 3
20 0 4
20 0 5
20 0 6
20 0 7
20 0 8
20 0 9
20 10
20 11
20 12
20 13
20 14
20 15
20 16 -0.052
0.088*
0.174***
0.195***
0.081*
0.146***
0.112***
0.141***
0.157***
-0.112**
-0.094**
-0.081*
0.043
0.030
0.099*
0.017
-0.004
1.280
1.847
2.457
2.389
1.754
1.968
1.639
2.023
2.600
2.748
3.221
2.288
1.895
1.279
1.464
0.956
0.318
1.513
1.429
1.420
1.359
1.516
1.346
1.432
1.310
1.725
1.883
2.126
1.810
1.655
1.535
1.196
0.953
0.297
-0.233
0.418
1.037***
1.030***
0.237
0.622***
0.207
0.714**
0.875***
0.866***
1.096***
-0.478*
0.241
-0.256
0.268
0.003
0.020
0.003
0.003
0.003
0.003
0.004
0.004
0.005
0.006
0.005
0.003
0.002
0.003
0.002
0.004
0.003
0.004
0.001
0.004
0.003
0.004
0.005
0.006
0.004
0.005
0.008
0.005
0.004
0.003
0.004
0.004
0.005
0.004
0.004
0.002
-0.001
0.000
-0.001**
-0.001***
-0.002*
-0.001
0.001
-0.002
0.000
-0.001
-0.01***
-0.001
-0.002***
-0.002
-0.001*
0.000
-0.001*
0.007
0.005
0.005
0.006
0.008
0.007
0.014
0.019
0.021
0.011
0.011
0.011
0.006
0.013
0.010
0.034
0.032
0.007
0.006
0.008
0.012
0.016
0.010
0.009
0.029
0.015
0.017
0.014
0.023
0.022
0.025
0.019
0.034
0.018
0.000
-0.001
-0.003**
-0.006***
-0.008**
-0.002
0.005
-0.010
0.006
-0.005
-0.003
-0.012*
-0.016***
-0.012*
-0.009**
0.001
0.014
0.276
0.279
0.304
0.394
0.304
0.261
0.263
0.291
0.257
0.197
0.121
0.172
0.179
0.289
0.205
0.133
0.350
0.243
0.244
0.237
0.209
0.304
0.256
0.277
0.197
0.247
0.170
0.162
0.177
0.221
0.234
0.287
0.273
0.336
0.033
0.035
0.067
0.185***
0.000
0.005
-0.014
0.094
0.010
0.028
-0.041
-0.004
-0.043
0.055
-0.082
-0.139**
0.014
-0.111
-0.189
212.100
202.700
209.000
207.600
194.700
196.600
198.700
210.400
203.600
200.500
214.400
202.100
211.400
205.900
239.400
216.500
219.500
162.900
207.300
195.700
202.500
190.100
199.000
184.600
195.300
185.700
197.300
204.100
202.200
195.700
209.800
217.400
226.000
213.400
205.800
187.000
161.900
201.300
16.400
0.200
18.900**
8.600
10.100
1.300
13.000
13.100
-0.500
-1.700
18.700**
-7.700
-6.000
-20.100
26.000*
10.700
32.500
1.000
6.000
6.412** 206.100 199.700 0.0102 0.247 0.258 -0.005*** 0.017 0.012 -0.001*** 0.004 0.003 0.511*** 1.105 1.616 0.103*** 0.267 0.370 Total This table provides mean difference test results for full sample by years. Propensity is a dummy variable that equals one if the firm provides an annual earnings guidance in a fiscal year and zero otherwise. Frequency is the number of earnings guidance made during a fiscal year for annual earnings of that fiscal year. Horizon is the average number of days between earnings forecasts and the end date of the fiscal year. Management error is the absolute value of average difference between actual earnings and management forecasts and the value is adjusted by the share price three days before the forecast. Precision is a dummy variable that equals one if the actual earnings are within the forecast range. Width refers to the absolute value of forecast range adjusted by stock price and zero is assigned to point forecasts. LEHF represents the firms with lowvega CEOs (vega is higher than annual median by industries) and high-vega CFOs (vega is higher than annual median by industries). LELF represents the firms with low-vega CEOs (vega is lower than annual median by industries) and low-vega CFOs (vega is lower than annual median by industries). ***, **, and *** indicate that the coefficient is statistically different from zero at 1%, 5%, and 10% significance level, respectively.
0.342
0.300
0.313
0.298
0.337
0.294
0.284
0.282
0.353
0.391
0.410
0.383
0.409
0.403
0.355
0.382
0.191
0.611
0.389
44.300
0.013
0.500
0.200
124.000
0.004
0.008
-0.003
168.300
-
LEHF-LELF
-0.001**
0.006
0.009
-0.128
-
LELF
0.002
0.014
0.006
0.462
-
0.000
0.001
0.000
0.333
-
-0.006
0.001
0.001
-0.014
-
0.164
0.002
0.001
0.016
-
0.158
-0.014 0.099*
0.003
-
-0.011
0.076
0.087
0.000
-
Horizon LEHF
0.127
0.175
0.073
0.001
-
LEHF-LELF
0.116
-0.012 0.066*
0.001
-
LELF
Precision LEHF
1998
0.070
0.073
0.105
-
LEHF-LELF
0.136
0.028
-
LELF
0.061
0.133
-0.014***
LEHF
Management Error
1997
0.045
0.014
LEHF-LELF
1996
0.028
0.000
LELF
Width LEHF
0.072
-0.014***
LEHF-LELF
1995
0.014
LELF
Frequency LEHF
0.000
LEHF-LELF
1994
LELF
Propensity
LEHF
Table 4. Mean difference tests by years for low-vega CEO/high-vega CFO firms and low-vega CEO/low-vega CFO firms
Table 5. Managerial risk-taking incentives and the propensity to release management forecast of earnings Model Intercept CEO vega
(1) -0.083 (-0.293) 0.059 (3.214)
(2) -0.051 (-0.172)
***
CFO vega Size ROA MB R&D Leverage Age Volatility Analyst error Analyst volatility Analyst coverage EPS volatility ΔEPS HHI SOX Litigation Risk Pseudo R2 N
0.002 (0.053) 1.144 (5.786) 0.014 (0.573) -2.558 (-2.367) 0.574 (2.883) -0.059 (-1.080) -1.039 (-10.392) -0.620 (-1.430) -0.245 (-1.691) 0.181 (2.714) -0.047 (-1.573) 0.050 (0.245) 0.077 (0.123) -0.151 (-1.923) 0.052 (0.357) 0.158
0.064 (2.597) -0.002 (-0.065) 1.132 (5.784) 0.007 (0.293) -2.493 (-2.363) 0.608 (3.012) -0.046 (-0.853) -1.028 (-10.494) -0.641 (-1.450) -0.257 (-1.775) 0.184 (2.710) -0.042 (-1.402) 0.040 (0.207) -0.022 (-0.035) -0.164 (-2.115) 0.051 (0.350) 0.154
***
**
***
***
*
***
*
14,232
14,256
***
***
**
***
***
*
***
**
(3) -0.087 (-0.301) 0.051 (3.023) 0.016 (0.614) -0.001 (-0.017) 1.144 (5.875) 0.014 (0.568) -2.564 (-2.370) 0.573 (2.805) -0.055 (-0.994) -1.031 (-10.340) -0.618 (-1.420) -0.245 (-1.699) 0.176 (2.608) -0.044 (-1.483) 0.052 (0.265) 0.060 (0.096) -0.138 (-1.780) 0.056 (0.380) 0.157
***
***
**
***
***
*
***
*
14,055
This table provides Probit regression results for a full sample from 1992 to 2016. The dependent variable is Propensity, which is a dummy variable that equals one if the firm provides an annual earnings guidance in a fiscal year and zero otherwise. CEO vega and CFO vega are the natural logs of one plus Core and Guay (2002) vega. Measurements for earnings management include discretionary accruals (DA), abnormal operating cash flows (ROCF), abnormal production cost (RPROD), and abnormal discretionary expenses (RDISX). Size is calculated as the natural logarithm of firm’s total assets. ROA is return on assets and defined as income before extraordinary items divided by total assets. MB refers to market-to-book ratio and is calculated as the sum of the book value of debts and the market value of common equity divided by total assets. R&D represents R&D expenses, and Leverage is total liabilities scaled by total assets. Age is the number of years that a firm has been in annual Compustat
29
database since the first occurrence. Volatility is the annualized standard deviation of monthly stock returns over a fiscal year. EPS volatility is the standard deviation of firm’s actual earnings per share over the prior five years, and we require that there are at least three-year observations. ∆EPS is the absolute change in annual earnings per share deflated by stock price. The industry Herfindahl-Hirschman index, HHI, which is calculated for the firm’s two-digit SIC industry by summing up the squared market shares of each individual firm in the industry and the market share is defined as the percentage of an industry total sales earned by a particular firm over a fiscal year. Analyst error is the absolute value of the difference between analyst forecast mean and actual number, scaled by stock price. Analyst volatility is the standard deviation of analyst forecast mean for each time over a fiscal year. Analyst coverage refers to the average number of analysts making earnings forecasts for a particular firm over a fiscal year. A dummy variable, SOX, is equal to one if the year of the observation is 2002 or later and zero otherwise. Litigation risk is also a dummy variable taking a value of one if the firm is in SIC industries 2833-3836, 3507-3577, 3600-3674, 5200-5961, or 7370-7374, which are industries with high litigation risk and zero otherwise. We winsorize each variable at 1% and 99% of its distribution by years. Year fixed effects are controlled, and robust Z-statistics are reported in the parentheses. ***, **, and *** indicate that the coefficient is statistically different from zero at 1%, 5%, and 10% significance level, respectively.
30
Table 6. Managerial risk-taking incentives and the frequency to release management forecast of earnings (1) (2) (3) Model Intercept CEO vega
-32.054 (-120.508) 0.230 (4.408)
***
ROA MB R&D Leverage Age Volatility Analyst error Analyst volatility Analyst coverage EPS volatility ΔEPS HHI SOX Litigation Risk Pseudo R2 N
0.185 (5.486) 5.063 (6.125) 0.120 (1.476) -10.231 (-5.166) 2.401 (6.260) -0.260 (-3.353) -4.474 (-9.189) -3.110 (-2.838) -1.177 (-3.661) 0.738 (7.573) -0.177 (-1.695) -0.019 (-0.032) -0.788 (-0.576) 30.374 (113.317) 0.085 (0.391) 0.081
***
0.261 (3.773) 0.16 (4.668) 5.072 (5.997) 0.089 (1.079) -10.081 (-4.995) 2.572 (6.604) -0.208 (-2.640) -4.456 (-9.054) -3.254 (-2.864) -1.225 (-3.710) 0.752 (7.594) -0.160 (-1.507) -0.013 (-0.022) -1.121 (-0.803) 30.501 (112.310) 0.090 (0.408) 0.079
***
***
CFO vega Size
-32.116 (-118.976)
***
***
***
***
***
***
***
***
***
*
***
14,248
14,272
***
***
***
***
***
***
***
***
***
***
-32.047 *** (-118.332) 0.194 *** (3.512) 0.073 (1.003) 0.177 *** (5.127) 5.074 *** (5.904) 0.119 (1.448) -10.232 *** (-5.059) 2.399 *** (6.141) -0.245 *** (-3.084) -4.426 *** (-9.005) -3.132 *** (-2.867) -1.161 *** (-3.575) 0.709 *** (7.077) -0.172 (-1.642) 0.011 (0.018) -0.832 (-0.600) 30.381 *** (111.414) 0.105 (0.476) 0.081 14,071
This table provides Tobit regression results for a full sample from 1992 to 2016. The dependent variable is Frequency is the number of earnings guidance made during a fiscal year for annual earnings of that fiscal year. CEO vega and CFO vega are the natural logs of one plus Core and Guay (2002) vega. Measurements for earnings management include discretionary accruals (DA), abnormal operating cash flows (ROCF), abnormal production cost (RPROD), and abnormal discretionary expenses (RDISX). Size is calculated as the natural logarithm of firm’s total assets. ROA is return on assets and defined as income before extraordinary items divided by total assets. MB refers to market-to-book ratio and is calculated as the sum of the book value of debts and the market value of common equity divided by total assets. R&D represents R&D expenses, and Leverage is total liabilities scaled by total assets. Age is the number of years that a firm has been in annual Compustat database since the first
31
occurrence. Volatility is the annualized standard deviation of monthly stock returns over a fiscal year. EPS volatility is the standard deviation of firm’s actual earnings per share over the prior five years, and we require that there are at least three-year observations. ∆EPS is the absolute change in annual earnings per share deflated by stock price. The industry HerfindahlHirschman index, HHI, which is calculated for the firm’s two-digit SIC industry by summing up the squared market shares of each individual firm in the industry and the market share is defined as the percentage of an industry total sales earned by a particular firm over a fiscal year. Analyst error is the absolute value of the difference between analyst forecast mean and actual number, scaled by stock price. Analyst volatility is the standard deviation of analyst forecast mean for each time over a fiscal year. Analyst coverage refers to the average number of analysts making earnings forecasts for a particular firm over a fiscal year. A dummy variable, SOX, is equal to one if the year of the observation is 2002 or later and zero otherwise. Litigation risk is also a dummy variable taking a value of one if the firm is in SIC industries 2833-3836, 3507-3577, 3600-3674, 5200-5961, or 73707374, which are industries with high litigation risk and zero otherwise. We winsorize each variable at 1% and 99% of its distribution by years. Robust t-statistics are reported in parentheses. ***, **, and *** indicate that the coefficient is statistically different from zero at 1%, 5%, and 10% significance level, respectively.
32
Model Intercept
Table 7. Managerial risk-taking incentives and the horizon of management forecast of earnings (1) (2) (3) (4) (5)
CEO vega
134.209 *** (4.665) 0.047 (0.067)
CFO vega Size ROA MB R&D Leverage Age Volatility Analyst error Analyst volatility Analyst coverage EPS volatility ΔEPS HHI SOX Litigation Risk R2 N
0.419 (0.303) -50.546 ** (-2.055) 1.949 ** (2.429) 41.496 (0.830) -1.819 (-0.368) -0.232 (-0.095) 18.336 * (1.719) 4.145 (0.108) 8.307 (1.500) 11.078 *** (4.024) -0.026 (-0.016) -42.893 ** (-2.263) -26.304 (-1.506) 32.095 (1.128) -4.397 (-0.898) 0.044 6,578
135.356 (4.815)
***
1.634 (1.802) 0.052 (0.036) -53.218 (-2.276) 1.838 (2.316) 43.729 (0.873) -0.890 (-0.168) -0.427 (-0.180) 19.042 (1.763) 4.802 (0.128) 8.626 (1.543) 10.402 (3.792) -0.192 (-0.120) -43.028 (-2.399) -22.476 (-1.277) 31.613 (1.113) -4.526 (-0.954) 0.045
*
6,572
**
**
*
***
**
137.094 *** (4.816) -1.735 * (-1.753) 3.102 ** (2.625) 0.029 (0.021) -52.094 ** (-2.177) 1.770 ** (2.260) 42.852 (0.867) -0.225 (-0.043) -0.277 (-0.115) 19.956 * (1.870) 1.944 (0.052) 8.917 (1.672) 10.800 *** (3.987) -0.218 (-0.138) -42.651 ** (-2.363) -22.942 (-1.316) 30.991 (1.086) -4.441 (-0.941) 0.046 6,509
163.512 *** (8.980) 0.796 (1.140)
***
(6)
1.128 (0.761) -57.730 *** (-2.956) 2.836 *** (3.388) 31.993 (0.673) -1.660 (-0.287) -1.562 (-0.706) 25.613 (1.630) 19.796 (0.407) 5.383 (0.920) 8.387 *** (3.494) 0.034 (0.020) -47.778 ** (-2.299) -28.317 * (-1.873) 7.692 (0.919) -4.011 (-0.883) 0.021
2.535 *** (2.595) 0.583 (0.369) -59.938 *** (-3.148) 2.692 *** (3.229) 34.960 (0.733) 0.185 (0.031) -1.552 (-0.746) 24.945 (1.596) 20.369 (0.428) 5.787 (0.967) 7.924 *** (3.257) -0.077 (-0.044) -48.324 ** (-2.506) -23.892 (-1.499) 7.444 (0.897) -4.045 (-0.927) 0.023
165.078 *** (9.218) -1.459 (-1.616) 3.770 *** (3.215) 0.554 (0.363) -59.462 *** (-3.110) 2.636 *** (3.216) 33.466 (0.710) 0.613 (0.107) -1.494 (-0.700) 25.823 (1.642) 17.520 (0.368) 6.051 (1.059) 8.259 *** (3.450) -0.093 (-0.053) -47.936 ** (-2.422) -24.480 (-1.568) 8.049 (0.966) -3.984 (-0.916) 0.023
6,578
6,572
6,509
164.114 (9.139)
This table reports OLS regression results for a full sample from 1992 to 2016. Horizon, the dependent variable, is the average number of days between earnings forecasts and the end date of the fiscal year. CEO vega and CFO vega are the natural logs of one plus Core and Guay (2002) vega. Measurements for earnings management include discretionary accruals (DA), abnormal operating cash flows (ROCF), abnormal production cost (RPROD), and abnormal discretionary expenses (RDISX). Size is calculated as the natural logarithm of firm’s total assets. ROA is return on assets and defined as income before extraordinary items divided by total assets. MB refers to market-to-book ratio and is calculated as the sum of the book value of debts and the market value of common equity divided by total assets. R&D represents R&D expenses, and Leverage is total liabilities scaled by total assets. Age is the number of years that a firm has been in annual Compustat database since the first occurrence. Volatility is the annualized standard deviation of monthly stock returns over a fiscal year. EPS volatility is the standard deviation of firm’s actual earnings per share over the prior five years, and we require that there are at least three-year observations. ∆EPS is the absolute change in annual earnings per share deflated by stock price. The industry Herfindahl-Hirschman index, HHI, which is calculated for the firm’s two-digit SIC industry by summing up the squared market shares of each individual firm in the industry and the market share is defined as the percentage of an industry total sales earned by a particular firm over a fiscal year. Analyst error is the absolute value of the difference between analyst forecast mean and actual number, scaled by stock price. Analyst
33
volatility is the standard deviation of analyst forecast mean for each time over a fiscal year. Analyst coverage refers to the average number of analysts making earnings forecasts for a particular firm over a fiscal year. A dummy variable, SOX, is equal to one if the year of the observation is 2002 or later and zero otherwise. Litigation risk is also a dummy variable taking a value of one if the firm is in SIC industries 2833-3836, 3507-3577, 3600-3674, 5200-5961, or 7370-7374, which are industries with high litigation risk and zero otherwise. We winsorize each variable at 1% and 99% of its distribution by years. Year fixed effects are controlled, and t-statistics (reported in the parentheses) for column 1 to 3 are based on standard errors clustered at the two-SIC industry levels and for column 4 to 6 standard errors are clustered at year and industry levels. ***, **, and *** indicate that the coefficient is statistically different from zero at 1%, 5%, and 10% significance level, respectively.
34
Model Intercept
Table 8. Managerial risk-taking incentives and management forecast errors (1) (2) 0.014
**
0.014
(2.649) CEO vega
**
(2.444)
-0.000
0.000 (0.051) 0.000
RDISX DA
-0.002
**
-0.002
**
R&D Leverage Age Volatility Analyst error Analyst volatility Analyst coverage EPS volatility ΔEPS HHI SOX Litigation Risk R2 N
-0.002 (-2.209)
0.001
0.001
0.001
(1.111)
(0.993)
(1.075)
-0.000
-0.000
-0.000
(-0.414)
(-0.405)
(-0.442)
-0.000
-0.000
-0.000
-0.002
(-0.956) ***
-0.002
-0.079
-0.079
-0.001
-0.001
-0.002 -0.080 -0.001
(-2.112)
(-1.967)
0.013 (0.779) 0.002 (0.597) 0.001 (0.398) 0.005 (1.367) 0.160 (1.500) 0.027 (3.204) 0.001 (0.640) 0.006 (1.780) -0.021 (-0.383) 0.002 (0.385) -0.002 (-0.680) 0.002 (1.460) 0.231
0.015 (0.868) 0.002 (0.553) 0.001 (0.383) 0.005 (1.489) 0.160 (1.497) 0.027 (3.156) 0.001 (0.738) 0.006 (1.762) -0.020 (-0.374) 0.002 (0.277) -0.001 (-0.629) 0.002 (1.469) 0.232
0.015 (0.838) 0.002 (0.537) 0.001 (0.373) 0.005 (1.354) 0.160 (1.496) 0.028 (3.183) 0.001 (0.651) 0.006 (1.749) -0.021 (-0.384) 0.002 (0.358) -0.001 (-0.596) 0.002 (1.417) 0.231
*
6,494
6,489
35
***
(-4.489) **
(-2.101)
***
***
(-2.722) ***
(-4.470) **
**
(-0.908) **
(-2.637) ***
(-4.603) MB
**
(-2.240)
(-3.116) ROA
-0.000 (-0.904)
(-2.076)
(-0.883) Size
**
(2.459)
(-2.335)
RPROD
0.014
(-1.005) CFO vega ROCF
(3)
***
*
6,427
*
***
*
This table reports OLS regression results for a full sample from 1992 to 2016. The dependent variable, Management error, is the absolute value of the average difference between actual earnings and management forecasts and the value is adjusted by the share price three days before the forecast. CEO vega and CFO vega are the natural logs of one plus Core and Guay (2002) vega. Measurements for earnings management include discretionary accruals (DA), abnormal operating cash flows (ROCF), abnormal production cost (RPROD), and abnormal discretionary expenses (RDISX). Size is calculated as the natural logarithm of firm’s total assets. ROA is return on assets and defined as income before extraordinary items divided by total assets. MB refers to market-to-book ratio and is calculated as the sum of the book value of debts and the market value of common equity divided by total assets. R&D represents R&D expenses, and Leverage is total liabilities scaled by total assets. Age is the number of years that a firm has been in annual Compustat database since the first occurrence. Volatility is the annualized standard deviation of monthly stock returns over a fiscal year. EPS volatility is the standard deviation of firm’s actual earnings per share over the prior five years, and we require that there are at least three-year observations. ∆EPS is the absolute change in annual earnings per share deflated by stock price. The industry Herfindahl-Hirschman index, HHI, which is calculated for the firm’s two-digit SIC industry by summing up the squared market shares of each individual firm in the industry and the market share is defined as the percentage of an industry total sales earned by a particular firm over a fiscal year. Analyst error is the absolute value of the difference between analyst forecast mean and actual number, scaled by stock price. Analyst volatility is the standard deviation of analyst forecast mean for each time over a fiscal year. Analyst coverage refers to the average number of analysts making earnings forecasts for a particular firm over a fiscal year. A dummy variable, SOX, is equal to one if the year of the observation is 2002 or later and zero otherwise. Litigation risk is also a dummy variable taking a value of one if the firm is in SIC industries 2833-3836, 3507-3577, 3600-3674, 5200-5961, or 7370-7374, which are industries with high litigation risk and zero otherwise. We winsorize each variable at 1% and 99% of its distribution by years. Year fixed effects are controlled, and t-statistics are reported in the parentheses and are based on standard errors clustered at the two-SIC industry levels. ***, **, and *** indicate that the coefficient is statistically different from zero at 1%, 5%, and 10% significance level, respectively.
36
Model Intercept
Table 9. Managerial risk-taking incentives and management forecast precision (1) (2) -0.986
***
-1.023
(-2.826) CEO vega
-0.025
***
(-2.952)
-0.002 -0.035
**
(-2.173)
RDISX
-0.018
-0.020
(-0.571)
(-0.658)
-0.022
-0.025
-0.023
(-1.303)
(-1.533)
(-1.436)
0.000
0.000
-0.000
-0.006
(0.039) ***
-0.006
ROA
0.088
0.09
R&D Leverage Age Volatility Analyst error Analyst volatility Analyst coverage EPS volatility ΔEPS HHI SOX Litigation Risk Pseudo R2 N
-0.006 0.093
(4.392)
(4.375)
0.035
0.012
0.017
0.039
(0.045) *
0.042
0.043
(1.972)
(2.119)
0.313 (0.672) -0.044 (-0.355) 0.003 (0.063) -0.465 (-3.882) 0.273 (0.488) -0.230 (-2.957) -0.067 (-1.760) 0.003 (0.097) 0.507 (1.254) -0.235 (-0.666) 0.077 (0.172) -0.012 (-0.368) 0.032
0.251 (0.520) -0.066 (-0.513) -0.000 (-0.012) -0.445 (-3.723) 0.244 (0.416) -0.219 (-2.692) -0.061 (-1.577) -0.002 (-0.074) 0.527 (1.286) -0.237 (-0.672) 0.092 (0.205) -0.022 (-0.614) 0.032
0.230 (0.505) -0.067 (-0.530) 0.001 (0.022) -0.451 (-3.809) 0.236 (0.410) -0.225 (-2.879) -0.065 (-1.680) 0.001 (0.027) 0.527 (1.296) -0.246 (-0.691) 0.085 (0.192) -0.014 (-0.412) 0.032
***
*
6,528
6,523
37
***
(0.065) **
(1.893)
***
***
(-2.844) ***
(4.026) (0.133) MB
(-0.022) ***
(-2.825) ***
*
(-1.834)
-0.020
(-3.072) Size
-0.036
(-0.692)
(0.083) DA
***
(-0.138)
CFO vega
RPROD
-1.028 (-2.955)
*
(-1.772)
ROCF
(3)
***
***
6,461
**
***
***
*
This table reports Probit regression results for a full sample from 1992 to 2016. Precision is the dependent variable and is a dummy variable that equals one if the actual earnings are within the forecast range. CEO vega and CFO vega are the natural logs of one plus Core and Guay (2002) vega. Measurements for earnings management include discretionary accruals (DA), abnormal operating cash flows (ROCF), abnormal production cost (RPROD), and abnormal discretionary expenses (RDISX). Size is calculated as the natural logarithm of firm’s total assets. ROA is return on assets and defined as income before extraordinary items divided by total assets. MB refers to market-to-book ratio and is calculated as the sum of the book value of debts and the market value of common equity divided by total assets. R&D represents R&D expenses, and Leverage is total liabilities scaled by total assets. Age is the number of years that a firm has been in annual Compustat database since the first occurrence. Volatility is the annualized standard deviation of monthly stock returns over a fiscal year. EPS volatility is the standard deviation of firm’s actual earnings per share over the prior five years, and we require that there are at least three-year observations. ∆EPS is the absolute change in annual earnings per share deflated by stock price. The industry Herfindahl-Hirschman index, HHI, which is calculated for the firm’s two-digit SIC industry by summing up the squared market shares of each individual firm in the industry and the market share is defined as the percentage of an industry total sales earned by a particular firm over a fiscal year. Analyst error is the absolute value of the difference between analyst forecast mean and actual number, scaled by stock price. Analyst volatility is the standard deviation of analyst forecast mean for each time over a fiscal year. Analyst coverage refers to the average number of analysts making earnings forecasts for a particular firm over a fiscal year. A dummy variable, SOX, is equal to one if the year of the observation is 2002 or later and zero otherwise. Litigation risk is also a dummy variable taking a value of one if the firm is in SIC industries 2833-3836, 3507-3577, 3600-3674, 5200-5961, or 7370-7374, which are industries with high litigation risk and zero otherwise. We winsorize each variable at 1% and 99% of its distribution by years. Year fixed effects are controlled, and robust z-statistics are reported in the parentheses. ***, **, and *** indicate that the coefficient is statistically different from zero at 1%, 5%, and 10% significance level, respectively.
38
Model Intercept
Table 10. Managerial risk-taking incentives and management earnings forecast width (1) (2)
CEO vega
0.001
0.001
0.001
(1.168)
(0.728)
(0.892)
-0.130
***
-0.000
(-3.146)
(-0.638)
CFO vega
-0.022
***
(-3.756) ROCF RPROD RDISX DA
-0.000
-0.000
(-0.314)
(-0.277)
0.000
-0.000
-0.000
(0.157)
(-0.066)
(-0.018)
0.000
0.000
0.000
(1.065)
(1.074)
(1.068)
-0.000
-0.000
-0.000
-0.026
(-1.190) **
-0.023
-0.005
-0.005
(-2.914) MB R&D Leverage Age Volatility Analyst error Analyst volatility Analyst coverage EPS volatility ΔEPS HHI SOX Litigation Risk R2 N
-0.048
(-1.088) **
(-2.116) ***
-0.047
-0.023 -0.005 -0.046
(-6.740)
(-6.743)
0.003 (0.925) 0.001 (2.473) 0.041 (2.407) 0.003 (2.836) 0.020 (2.948) 0.004 (4.168) -0.000 (-1.669) 0.001 (1.505) -0.003 (-0.736) 0.001 (0.896) 0.003 (3.520) 0.000 (0.240) 0.274
0.002 (0.771) 0.001 (2.388) 0.040 (2.339) 0.003 (2.905) 0.020 (2.939) 0.004 (4.161) -0.000 (-1.484) 0.001 (1.512) -0.003 (-0.766) 0.001 (0.786) 0.003 (3.636) 0.000 (0.205) 0.278
0.002 (0.850) 0.001 (2.338) 0.040 (2.334) 0.003 (2.746) 0.020 (2.965) 0.004 (4.140) -0.000 (-1.624) 0.001 (1.480) -0.003 (-0.761) 0.001 (0.858) 0.003 (3.616) 0.000 (0.181) 0.275
**
***
***
***
***
6,526
6,521
39
***
(-2.690) ***
(-6.506)
**
**
(-2.065) **
(-2.633) ***
*
(-1.704)
-0.000
(-2.664) ROA
-0.017
(-0.268)
(-1.011) Size
(3)
**
**
***
***
***
***
6,459
***
**
**
***
***
***
***
This table reports OLS regression results for a full sample from 1992 to 2016. The dependent variable, Width, refers to the absolute value of forecast range adjusted by stock price and zero is assigned to point forecasts. CEO vega and CFO vega are the natural logs of one plus Core and Guay (2002) vega. Measurements for earnings management include discretionary accruals (DA), abnormal operating cash flows (ROCF), abnormal production cost (RPROD), and abnormal discretionary expenses (RDISX). Size is calculated as the natural logarithm of firm’s total assets. ROA is return on assets and defined as income before extraordinary items divided by total assets. MB refers to market-to-book ratio and is calculated as the sum of the book value of debts and the market value of common equity divided by total assets. R&D represents R&D expenses, and Leverage is total liabilities scaled by total assets. Age is the number of years that a firm has been in annual Compustat database since the first occurrence. Volatility is the annualized standard deviation of monthly stock returns over a fiscal year. EPS volatility is the standard deviation of firm’s actual earnings per share over the prior five years, and we require that there are at least three-year observations. ∆EPS is the absolute change in annual earnings per share deflated by stock price. The industry HerfindahlHirschman index, HHI, which is calculated for the firm’s two-digit SIC industry by summing up the squared market shares of each individual firm in the industry and the market share is defined as the percentage of an industry total sales earned by a particular firm over a fiscal year. Analyst error is the absolute value of the difference between analyst forecast mean and actual number, scaled by stock price. Analyst volatility is the standard deviation of analyst forecast mean for each time over a fiscal year. Analyst coverage refers to the average number of analysts making earnings forecasts for a particular firm over a fiscal year. A dummy variable, SOX, is equal to one if the year of the observation is 2002 or later and zero otherwise. Litigation risk is also a dummy variable taking a value of one if the firm is in SIC industries 2833-3836, 3507-3577, 3600-3674, 5200-5961, or 73707374, which are industries with high litigation risk and zero otherwise. We winsorize each variable at 1% and 99% of its distribution by years. Year fixed effects are controlled, and t-statistics are reported in the parentheses and are based on standard errors clustered at the two-SIC industry levels. The coefficients on CEO vega, CFO vega, Size, MB, and Age are multiplied by 100 to make them readable. ***, **, and *** indicate that the coefficient is statistically different from zero at 1%, 5%, and 10% significance level, respectively.
40
