CHAPTER 4 EMPIRICAL RESULTS
4.3 Additional Tests
國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
48
failure, particularly in those firms with low CEO equity-based compensation. The control variables are consistent with the predicted signs.
Overall, I find that larger amounts and portions of stocks and options are associated with higher restatement and ICW likelihood and greater real earnings management. Therefore, equity-based compensation appears to harm audit committees' oversight effectiveness. However, the adoption of the clawback provisions significantly mitigates such negative effect. The findings in this paper alert regulators and researchers that mandating incentive clawback provisions could potentially affect audit committees’ effectiveness, a benefit that has not been documented in prior research.
4.3 Additional Tests 4.3.1 Trigger effects
While SOX specifies misconduct as the trigger of recoupment, Dodd Frank Act does not limit recovery to restatements where there has been misconduct. When firms voluntarily adopt clawback provisions, they have the discretion to determine the type of action that triggers the clawback as well as the extent of the amount recovered. Chrry and Wong (2009) analyzes the common trigger events and indicates that clawback provisions could be triggered under three circumstances:
misconduct or fraud, the restatement of financial results, or the event of employees’ bad faith. For voluntarily adopters, clawback provisions could be arranged according to firms’ objectives. I therefore analyze what clawback trigger causes the observed improvement in audit committees’
oversight effectiveness. I decompose CLAWBACK into four types: fraud/misconduct (denoted by TRIGGER1), all restatements (denoted by TRIGGER2), bad faith (denoted by TRIGGER3), and multiple triggers (denoted by TRIGGER4).36 Because firms with prior frauds or restatements may tend to use frauds or restatements as the triggers, I exclude 54 observations with fraud or restatements prior to the initial adoption of clawback provisions and eliminate their matched
36Due to a lack of theory and empirical evidence that can explain/predict the relative efficacy of these four triggers in implementing the clawback provisions, I use observations with and without clawback provisions to examine the trigger effects.
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
49
sample.
The differential effects of triggers are reported in Table 12. The results reveal that the coefficients on TRIGGER1 and TRIGGER3 are significantly negative across four oversight failure measures under both the magnitude and percentage approaches. Note that the association between TRIGGER3 and oversight failure measures is stronger, suggesting that specifying bad faith as the trigger appears to be more desirable if regulators attempt to increase the benefits of the clawback provisions. The significant coefficient on TRIGGER1 may possibly due to the fact that, firms simply follow SOX (which explicitly specifies misconduct as the trigger) and choose fraud/misconduct as the trigger. This paper is the first empirical research showing the different trigger effects. Since the observed benefit of adopting clawback provisions is driven by bad faith trigger, regulator shall pay more attention to the terms of firms’ recoupment policies.
[Insert Table 12 here]
4.3.2 Excluding special restatement period
It is possible that the restatement period may span firms’ initial adoption of clawback provisions. For example, a firm restates its 2005-2008 financial statements in 2009 but adopts the clawback provisions in 2007. Under this situation, this firm will be classified as a non-adopter in 2005 and 2006 but will be classified as an adopter in 2007 and 2008. To eliminate the potential impact of this classification issue on my empirical results, I exclude restatements which span the initial adoption of the clawback provisions and report the results in Table 13. My previous results and conclusions remain unchanged.
[Insert Table 13 here]
4.3.3 Panel data results for restatements
From the yearly distribution of restatements shown in Panel A of Table 6, it appears that restatements substantially decreased after 2006. To ensure that the association between audit committee compensation and restatements likelihood is not subjected to time series biases, I re-run REST model using panel data analysis. I also use a Hausman test to examine whether a fixed or
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
50
random effects model is appropriate. The significant Hausman statistics suggest that the use of the fixed-effect model is adequate. The panel data results are reported in Table 14. This alternative research design does not change the findings.
[Insert Table 14 here]
4.3.4 Excluding certain industries
In my main analyses, I include industry fixed effects and year fixed effects to controls for unobserved firm-level heterogeneity. However, according to industry distribution shown in Panel B of Table 6, more restatements, ICW, and real earnings management seems to exist in certain industries. Accruals quality appears to be consistent during the sample period. To control for the potential biases due to some specific industry characteristics, I exclude particular industries which contain unusual frequencies of restatements/ICW and level of real earnings management. I re-run REST model by excluding firms in the pharmaceuticals and utilities industries. I also eliminate firms in the food industry and re-run the ICW model. Finally, firms in the food, durable manufactories, transportation, services, and computer industries are excluded when I re-run my EM model. The results are reported in Table 15. Even though the coefficients of few variables are not significant, their directions are correct. Therefore, the overall results documented in section 4.2 are generally unchanged.
[Insert Table 15 here]
4.3.5 Excluding cash compensation ratio variable
In the regression model, I use three compensation ratios to test the differential percentage effects of cash, stocks, and options. Because these three compensation ratios sum up to one, a potential multicollinearity may exist. To address this issue, I exclude ACCASH% and use ACSTOCK% and ACOPTION% to re-run all regression models. As reported in Table 16, this alternative research design does not alter the results.
[Insert Table 16 here]
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
51
4.3.6 Corporate governance effects
It is possible that the moderating effect of clawback provisions on the unfavorable influence of audit committee equity-based compensation is spuriously caused by a lack of control for variations of corporate governance mechanisms. Therefore, I include a corporate governance indicator variable (Gov_Index) and their interactions with the compensation-related variables (i.e., ACCASH× Gov_Index, ACSTOCK× Gov_Index and ACOPTION× Gov_Index) as additional control variables in the regression model to test the robustness of the relation among equity-based compensation, clawback provision and audit committees’ monitoring effectiveness.
Following Vyas (2011), I measure corporate governance for the sample period using the index-adjusted Corporate Governance Quotient (CGQ) computed by Institutional Shareholder Services (ISS) for the same period. The CGQ comprises 63 variables in the following eight categories: board of directors, audit, charter and bylaw provisions, laws of the state of incorporation, executive and director compensation, qualitative factors, ownership, and director education. I then classify observation into two subsamples. Gov_Index is coded one if firms’ CGO is more than the sample median, and 0 otherwise. The results are reported in Table 17. I find a significantly positive coefficient of ACSTOCK× Gov_Index and ACOPTION× Gov_Index, suggesting that sound corporate governance does not moderate the detrimental effects of equity-based compensation.
Further, I find significant coefficients on ACSTOCK× CLAWBACK and ACOPTION× CLAWBACK, even after controlling for the quality of governance. Thus, the results reported in the previous sections remain the same.
[Insert Table 17 here]
4.3.7 Effects of board compensation
Although audit committees assume the ultimate responsibility in overseeing the quality of firms’ internal controls and financial reporting, the entire boards are also charged with the duty to monitor firms’ overall financial reporting process (Cullinan et al. 2010). Linck et al. (2009) examines the changes in board compensation structure after SOX and shows that director
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
52
compensation increased substantially post-SOX due to the changes in directors’ workload. Some research also emphasizes that directors should be sufficiently compensated to increase boards’
governance effectiveness (e.g., Ryan and Wiggins 2004). Therefore, it is possible that the clawback provisions affect not only the audit committees, but also other members in the boards. To address this concern, I replace all audit committee compensation variables by compensation paid to board members other than audit committees and CEOs (denoted by OtherCASH, OtherSTOCK, OtherOPTION, OtherCASH%, OtherSTOCK%, and OtherOPTION) and re-run all regressions using the High-Low CEO Equity Groups. Table 18 shows that all coefficients on these compensation variables and their interactions with the clawback provisions are insignificant except for few that are only marginally significant. These results imply that the adverse effect of equity-based compensation and the mitigating effect of the clawback provisions affect the audit committees rather than other board members.
[Insert Table 18 here]
4.3.8 Heckman two stage results
I also employ Heckman’s (1979) two-stage model to control for endogenous biases. At the first stage, I use the voluntary adoption model to estimate the inverse Mill’s ratio (denoted by IMR).
At the second stage, I include IMR as an additional explanatory variable in the second-stage regression model. Lennox et al. (2012) suggests that the implementations of the Heckman model in accounting research should satisfy three requirements: variable exclusion restrictions, tests for multicollinearity, and significant inverse Mill’s ratio. Because only one variable (i.e., LnASSET) in the first stage model is also used in the second stage, I roughly meet the exclusion restrictions. In addition, I use variance inflation factors (VIFs) to evaluate the multicollinearity of IMR and find that its VIFs are between 2.98 ~ 4.04, which are far below the 10.0 threshold suggested by Neter et al. (1996). Finally, Table 19 reports that all coefficients on IMR are significant. Note that, Table 19 shows some differences in the empirical results between the Heckman model and the propensity score matching. Specifically, my hypotheses H1a, H1b, and H1c are not fully supported under the
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
53
Heckman model, even though my H2 is supported.
[Insert Table 19 here]
4.3.9 Alternative measures in firm performances
To ensure that my results are not sensitive to measures of firm performance, I control for firm performances in the four regression models. There are many measures to proxy for firm performances. Following Chhaochharia and Grinstein (2009), I use the natural log of the gross annual stock return of the firms (donted by lnRETURN) to replace firm performance variables (i.e., ROA_Ind and LOSS) and add it in all regression models. As reported in Table 20, the different measures in the firm performance do not alter the results.
[Insert Table 20 here]
4.3.10 Excluding financial crisis period
Recent studies argue that the financial crisis after 2008 has a significant impact on firms' risk management, financing policies and firm performance (Brunnermeier 2009; Erkens et al., 2012), an important concern is that observed clawback provision effects could be biased during the crisis period. To control for the economic impact, I exclude the 2008-2009 sample and conduct the related tests described in section 3 of this paper, with no substantive change in the results. The results are reported in Table 21. However, the negative coefficients on ACSTOCK× CLAWBACK and ACOPTION× CLAWBACK are insignificant in ICW model. A possible explanation for this limited evidence is the bias under initial stage of Section 404 when I use ICW as the dependent variable. Because Section 404 reports are disclosure from the fiscal year ends of 2004, AuditAnalytics compiles data from public disclosures of material weaknesses in the 10-K reports of accelerated filers and suggests that, among the 2,451 companies, the number disclosing ICW fell from 337 in 2004 to 179 in 2005. A likely explanation is that internal control quality is improving overall. The tremendous ICW reported in the first two or three years could cause the weak coefficients on the interactions. Overall, the results are robust to the variation in sample period.
[Insert Table 21 here]
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
54