«Internal and External Attributions by Managers in Earnings Conference Calls by Zhenhua Chen Business Administration Duke University Date:_ Approved: ...»
Column 3 to 6 present the regression results based on attributions from the presentation and Q&A sections separately. The results indicate that attributions from both the presentation and Q&A sessions are negatively related to earnings persistence.
Untabulated results show that the coefficients are not statistically different between the presentation and Q&A sections.3 To be consistent with prior literature, I also re-estimate the first column of Table 8 using only observations
This table reports ordinary least squares regression estimation of the association between attributions and future ROA. All the variables are defined as Appendix A. T-statistics are shown in parentheses with standard errors clustered at both firm and quarter level. ***/**/* means significance at 0.01, 0.05 and 0.10 level, respectively.
9. Managers’ Attributions and Future Stock Returns Based on the evidence in prior sections, we can conclude that investors react negatively to managers’ attributions, and managerial attributions are negatively related to earnings persistence. It is unclear, however, whether the negative market reaction is complete. Indeed, given that analysts may not fully incorporate the information contained in CEO attributions, it is plausible that capital markets are not completely efficient. Moreover, prior literature documents price drift with respect to quantitative earnings information (Bernard and Thomas 1989) and with respect to qualitative information contained in the linguistic tone of management (Demers and Vega 2008;
Engelberg 2008). In this section I test whether the stock market fully reflects the information in CEO attributions. If the stock market is fully rational and efficient with respect to managers’ attributions, I expect that future returns will not be associated with information contained in managers’ attributions.
Initial examination of this issue is reported in Table 5 Panel A, which reveals that future returns are not significantly correlated with managerial attributions during earnings conference calls. To test this in a multivariate framework, I estimate the following determinant model of future stock returns that controls for factors shown in
prior literature to affect future stock returns:
The results of estimating equation (4) are presented in Table 9. The coefficients on ∆IvsE are not statistically significant in all three columns. This suggests that the stock market appears to be fully efficient with respect to information in managers’ attributions.
The other coefficients are consistent with prior literature. The coefficients on UE are positive and statistically significant, consistent with Doyle et al. (2006), who find that earnings surprises are positively associated with future stock returns in the next three years. I find weak statistical association between POSWORDS and future returns, and no statistical association between NEGWORDS and future returns1. The coefficients on MOM are positive and statistically significant at the 0.10 level, consistent with Chan et al. (1996) who document the association between momentum and future returns. The negative coefficients on ACCRUALS are consistent with the accrual anomaly (Sloan 1996).
Prior literature finds mixed evidence about the associations between management tones and future returns. Engelberg (2008) and Demers and Vega (2008) document associations between future returns and linguistic tones, but Loughran and McDonald (2011) find no association between negative tone and future returns.
Table 9: Managers’ Attributions and Future Stock Returns
This table reports ordinary least squares regression estimation of the association between managers’ attributions and firms’ future stock returns. All the variables are defined in Appendix A. T-stats are shown in parentheses with standard errors clustered at both firm and quarter level.
***/**/* means significance at 0.01, 0.05 and 0.10 level, respectively.
10. Managers’ Attributions and CEO Compensation Given the prior sections suggest that the capital markets recognize and appropriately “punish” self-serving attributions by managers, in this section, I examine whether boards also recognize the negative implications of CEO attributions and react accordingly. Kim (2011) shows that CEOs with self-serving attribution bias are more likely to be fired. In this paper I examine a different punishment mechanism: CEO compensation. Koonce et al. (2011) suggest that boards should adjust compensation depending on the factors managers ascribe to performance. To assess whether boards make any adjustments to compensation in response to self-serving attribution, I regress CEO compensation on the attribution measure, while controlling for other economic
factors that may affect CEO compensation I estimate the following specification:
The dependent variable is the natural logarithm of total compensation obtained from EXECUCOMP. Because the compensation data is only available on an annual basis, I use the mean value of IvsE during the fiscal year and conduct analysis on a firm-year basis. Prior literature on CEO compensation suggests that larger firms with more growth opportunities and more complex operations need higher-ability managers who demand higher compensation (Smith and Watts 1992), therefore I include prior year sales ( ) to proxy for firm size and operating complexity, and prior year book-tomarket ratio ( to proxy for growth opportunities. Classic agency theory and prior empirical results suggest that firm performance is also a determinant of CEO compensation, so I also include current year ROA and stock returns (RET) as additional independent variables. I also control firm risk using standard deviations of returns in the past five years (STDRET). However, because prior literature suggests that firm risks may increase or decrease CEO compensation (Banker and Datar 1989), I do not have a predicted sign for STDRET.
Table 10 presents the regression results of equation (5). The coefficient on ΔIvsE is negative but not statistically significant, suggesting that boards do not adjust CEO compensation in response to managers’ attributions during conference calls. There are two plausible explanations for this finding. One explanation is that boards are irrational with respect to managers’ attributions. However, Kim (2011) suggests that boards are rational in that CEOs with self-serving attribution bias are more likely to be fired. The other possible explanation is that CEOs may communicate with board members through other channels rather than earnings conference calls. A conference call is for managers to communicate with financial analysts and investors, and managers may disclose different information about attributions when they communicate with board members through other avenues such as board meetings.
finds larger firms with more growth opportunities pay their CEOs more. The coefficient on ROA is positive but not statistically significant, while the coefficient on RET is positive and statistically significant, consistent with results in Core et al. (1999). The coefficient on STDRET is negative but not statistically significant.
This table reports ordinary least squares regression estimation of the association between managers’ attributions and CEO compensation. All the variables are defined in Appendix A. Tstatistics are shown in parentheses with standard errors clustered at both firm and quarter level.
***/**/* means significance at 0.01, 0.05 and 0.10 level, respectively.
11. Further Empirical Tests Thus far the results suggest that CEOs exhibit self-serving attribution bias, and that the market rationally incorporates its ramifications. However, it is not clear why CEOs still make self-serving attributions even though the market reacts unfavorably. To provide some exploratory insight, I examine whether the CEO’s attribution behavior varies in a systematic manner. More specifically, I test whether CEOs attribution behavior differs in four different settings: when CEOs are less established as proxied by tenure, when CEOs attach less weight in attributions as proxied by high growth, when the earnings conference calls are in the financial crisis period of 2007, and when CEOs are overconfident.
In a model relating managers’ career concerns to disclosures, Hermalin and Weisbach (2011) suggest that owners of a firm seek to assess a CEO’s ability based on the information available to them, and to replace him or her if the assessment is too low.
Kim (2011) finds that self-serving CEOs are more likely to be fired, suggesting that the labor market does evaluate CEOs based on their attribution styles.
If CEOs’ abilities have not been revealed to the market, the labor market reaction to their self-serving attributions may be stronger. For a CEO who has been with the firm for a shorter time (lower tenure), his or her ability is less likely to have been revealed to the labor market, and hence he or she is more subject to the negative consequences of self-serving attributions. As a result I predict newer CEOs will be less likely to make self-serving attributions.
To test this prediction, I include LOWTENURE and its interaction term with ROA as additional independent variables in model (1). LOWTENURE is an indicator variable that equals 1 if a CEO’s tenure is below the median of the sample. In column (1) of Table 11, the coefficient on the interactive term ROA*HIGHTENURE is negative and statistically significant, suggesting newer CEOs are less likely to make self-serving attributions.
The second type of attenuation that I consider is how sensitive a firm’s stock price is to information. Given the capital market reacts negatively to attributions, the cost of supplying a self-serving attribution should be higher in settings where the stock market reacts more strongly to signals. Extant literature suggests that stock markets react more strongly and asymmetrically negatively to unfavorable earnings signals (Skinner and Sloan 2002). I posit that CEOs at high growth firms will be less likely to engage in self-serving attributions in order to avoid the relatively severe stock market response to signals. To test this assertion, I include HIGHGROWTH and its interaction term with ROA in equation (1), where HIGHGROWTH equals 1 if a firm’s sales growth rate is above the median. Column (2) of table 11 shows that the coefficient on ROA*HIGHGROWTH is negative and statistically significant, consistent with my prediction.
This table presents the ordinary least squares regression between management attributions and firm performance in different tenure groups. All the variables are defined in Appendix A. T-statistics are shown in parentheses with standard errors clustered at both firm and quarter level. ***/**/* means significance at 0.01, 0.05 and 0.10 level, respectively.
The third type of attenuation that I consider is financial crisis. My sample period includes 2007, which is during the U.S. financial crisis. External economic conditions give managers more extensive and credible leeway to blame financial crisis as the performance driver. Keusch et al. (2012) find that CEOs of large European companies are more self-serving in their letters to shareholders during the financial crisis. Therefore, I examine whether CEOs are more self-serving during the 2007 financial crisis.
To be consistent with the predicted sign, I include an indicator variable, NOCRISIS, and its interactive term with ROA as additional independent variables in equation 1. NOCRISIS equals 1 if the calendar year of a quarter is between 2002 and 2006 and 0 otherwise. The regression results are presented in column (3). I find no association between financial crisis and managers’ self-serving bias, as the coefficient on ROA*CRISIS is not statistically significant.
One plausible explanation for my finding is that some managers behave in the opposite way during crisis. Indeed, managers during a financial crisis may find it to be an opportunity to demonstrate leadership (Halverson et al. 2004). Lee et al. (2004) argue that managers take more responsibility for unfavorable outcomes in order to show that they are capable of influencing the company. For instance, Jeff Immelt, the CEO of GE, offered a “Mea Culpa” attribution during the 2008 financial crisis. The two effects may work against each other during the financial crisis and result in a muted association. The other possible explanation is that CEO attributions during the financial crisis are not captured by third-person pronouns, as I recognized in the validity test section of this paper.
The fourth type of attenuation that I examine is the effect of CEO overconfidence.
Libby and Rennekamp (2010) suggest that self-serving attribution bias is closely related to managerial overconfidence. Li (2010) uses self-attribution bias to capture CEO overconfidence and finds that overconfidence is related to corporate financial policies.
Overconfident CEOs may make more self-serving attributions than others. Therefore I include overconfidence and its interactive term with ROA to examine the effects of overconfidence on CEO attributions.
Following Malmendier and Tate (2005, 2008), I measure CEO overconfidence based on the CEOs’ beliefs about future stock returns. To be consistent with the predicted sign, I use the negative value of the overconfidence indicators. I use two measures: NEG_LONGHOLDER, which equals -1 if a CEO holds an option until the year of expiration even though the option is at least 40% in the money, and NEG_HOLDER67, which equals to -1 once a CEO fails to exercise options with 5 years remaining duration despite a stock price increase of at least 67% since grant date. 1 Because the overconfidence measures require outstanding options data from EXECCOMP, observations before 2006 were dropped in this analysis.
Results presented in column (4) and (5) do not support the hypothesis that overconfident CEOs are more self-serving than other managers. The coefficients on the interactive terms of overconfidence and ROA are not statistically significant.
12. Robustness Tests