Restatement likelihood

國

立 政 治 大 學

‧

N a tio na

l C h engchi U ni ve rs it y

24

the influence of the board and thus increase CEO power (Lisic et al. 2011). I use four proxies for CEO power: the number of years the executive has served as CEO for the firm (denoted by CEO_Tenure), whether the CEO is the chairman of the board (denoted by CEO_Chair), the ration of CEO bonus to cash compensation (denoted by Bonus to cash) and percentage of firm’s shares owned by the CEO (denoted by CEO_Ownership). I predict their coefficients to be negative because CEO power is likely to reduce the likelihood that firms adopt clawback provisions (Addy et al. 2011; Brown et al. 2011).

Browen et al. (2011) indicates that managerial power, which is measured by the number of directors on the board, is negatively associated with firms’ likelihood of adopting clawback adoptions. Therefore, I include board size (denoted by BSIZE) in the CLAW model and predict its coefficient to be negative. Addy et al. (2011) finds that firms with more independent governance are more likely to adopt clawback provisions. I thus include the percentage of inside directors (denoted by INSIDE_%) in the CLAW model and predict its coefficient to be negative.

3.3 Regression Model

This study extends the audit committee compensation literature by investigating the link between clawback provisions and audit committees’ effectiveness. I use four measures to proxy for audit committees’ oversight failure: the likelihood of restatements, the incidence of ICW, and the level of accruals quality and real earnings management. Each of these oversight failure measures is discussed below.

3.3.1 Restatement likelihood

I use restatement likelihood to proxy for audit committees’ oversight failure because SOX expends audit committees' responsibilities to assure that financial statement accurately portray companies’ economic activities (Laux and Laux 2009). To test whether the clawback provisions enable compensation policy more efficient for audit committees, leading to less likelihood of restatements, I estimate the following logistic model following Archambeault et al. (2008), Efendi et al. (2007), and Palmrose et al. (2004):

‧

where the definitions of all the variables are summarized in Table 2. Note that I include industry fixed effects and year fixed effects as controls for unobserved firm-level heterogeneity over time (Bowen et al. 2010; Linck et al. 2009). The fixed-effect model helps alleviate the endogeneity problem caused by the omitted variables (Campa and Kedia 2002).

[Insert Table 2 here]

Dependent variable

The dependent variable, RESTATED, is a dummy variable that equals 1 if a firm’s year t financial statements are restated and 0 otherwise. Instead of using whether or not firms announce restatements in year t, variable RESTATED provides a more appropriate test of the association between audit committees’ compensation and restatement likelihood because outside directors serving on year t’s audit committees are responsible for overseeing year t’s financial statements and receive year t’s compensation. The use of restatement announcement year will mismatch the year audit committees exercise their oversight responsibility and the year they receive compensation. I thus use RESTATED to proxy for audit committees’ oversight failure and predict that the association between audit committee compensation and financial reporting failure is a moderated by clawback provisions.

Control Variables

In the REST model, I include major firm characteristics that are likely to affect the likelihood of restatements. Similar to previous studies (e.g., Dechow et al. 1996; Richardson et al. 2002; Desai et al. 2006), I control for firm size (denoted by LnASSET) and predict its coefficient to be negative because size might capture firm-specific risk (Fama and French 1995) and larger firms are more likely to be subjected to closer scrutiny by regulators and investors (Balsam et al. 2003; Romanus

‧

立 政 治 大 學

‧

N a tio na

l C h engchi U ni ve rs it y

26

et al. 2008). Also, controlling for size can potentially mitigate the problem of correlated omitted variables (Myers et al., 2005; Ahmed and Goodwin 2007).

Farber (2005) reports a smaller proportion of brand-name audit firms in fraud firms compared with control firms. Therefore, I include Big 4 CPA firms (denoted by BIG4) to control for audit firm quality and predict its coefficient to be negative. In addition, firms receiving going concern opinions are more likely to restate their financial statements afterwards (DeFond and Jiambalvo 1991; Kinney and McDaniel 1989; Sennetti and Turner 1999). Hence, I include going concern opinion as an indicator variable (denoted by GOING) and predict its coefficient to be positive.

Empirical evidence has shown that mergers and acquisitions may increase the probability of restatements due to new, difficult, or contentious accounting issues, and possible business integration problems (e.g., Kinney et al. 2004; Efendi et al. 2007; Stanley and DeZoort 2007;

Carcello et al. 2011). As a result, I control for firms’ merger and acquisition activities (denoted by M&A) and predict its coefficient to be positive. I also consider industry-median-adjusted return on assets (donated by ROA_ind) and predict its coefficient to be negative because prior studies show that more profitable firms are less likely to restate due to weaker incentives of manipulating earnings (e.g., Abbott et al. 2004; DeFond and Jiambalvo 1991; Ettredge et al. 2010; Kinney and McDaniel 1989; Loebbecke et al. 1989; Scholz 2008). I consider firms’ market-to-book ratio (donated by MB) to control for growth opportunities because high-growth firms having less growth opportunities are most likely to adopt aggressive accounting practices (Carcello et al., 2011; Burns and Kedia 2006).

I control for four determinants that may influence the oversight effectiveness of audit committees: ACSIZE, OVERLAPCOM, ACCEXPERT, and MEETING.²⁵ I consider audit committee size (donated by ACSIZE) because larger audit committees are perceived to have increased power (Chen and Zhou 2007; Kalbers and Fogarty 1993) and are more likely to challenge

25The correlation coefficients between these four audit committee characteristics and the compensation variables range from 0.51 to 1.20, which are all insignificant. Also, the variance inflation factors (VIFs) of these variables are all between 4.03 and 7.92. Therefore, the use of these four audit committee characteristics variables shall not lead to multicollinearity problem.

‧

立 政 治 大 學

‧

N a tio na

l C h engchi U ni ve rs it y

27

top management and internal control personnel in fulfilling their monitoring responsibilities (Goh 2009; Krishnan 2005). I also control for membership overlapping (denoted by OVERLAPCOM) because there has been a tendency of significant overlapping between the audit committees and other committees (Hoitash and Hoitash 2009; Lorsch and MacIver 1989; Wall Street Journal 2011).²⁶ Some research shows that overlapping compensation and audit committees creates the conflict of interests, and compensation committee members sitting on the audit committees will result in less effective CEO compensation contracts (e.g., Laux and Laux 2009; Hoitash and Hoitash 2009). CEO power continues to have an impact on audit committees’ effectiveness in the post-SOX era (Lisic et al. 2011). I thus predict the coefficient on OVERLAPCOM to be positive.

Recent studies examine whether narrowly-defined accounting and finance expertise individually contributes to audit committees’ monitoring activities (e.g., Archambeault and DeZoort 2001; Bédard et al. 2004; Goh 2009; Krishnan 2005; Krishnan and Visvanathan 2008;

Raghunandan et al. 2001; Dhaliwal et al. 2010; Engel et al. 2010). Following DeFond et al. (2005), I measure ACCEXPERT by the percentage of audit committee members having accounting expertise only. ²⁷ Accounting experts are members who have CPA licenses or with accounting-related experience (e.g., accountants, auditors, controllers, or chief accounting officers).

Since more specialized skills in accounting contribute more to audit committees’ oversight effectiveness (Agrawal and Chadha 2005; DeFond et al. 2005; McDaniel et al. 2002), I predict the coefficient of ACCEXPERT to be negative. I also consider annual meeting times (denoted by MEETING) to capture audit committees' effort (Engel et al. 2010) because more diligent audit committees are more likely to effectively exercise their oversight duties (DeZoort et al. 2002) so that they can remain informed of accounting and auditing issues (Raghunandan et al. 2001).

26Laux and Laux (2009) reports that, based on the 2006 proxy statements of the S&P 100 firms, 23 percent have at least one member of the compensation committees sitting on the audit committees. In about 20 percent of the cases, the chairs of the compensation committees serve on the audit committees as well.

27Engel et al. (2010) categorizes four types of financial reporting expertise for the selection of audit committee chairs:

non-financial director, finance financial expert, general accounting financial expert, and accounting expert with Big4 employment experience. Their classifications of finance and accounting expertise are narrower than those of the proposed and final SEC rules. Because accounting expertise is not a major test variable in this study, I adopt a simplified classification used by DeFond et al. (2005).

‧

立 政 治 大 學

‧

N a tio na

l C h engchi U ni ve rs it y

28

In REST model, three major audit committee’s compensation components are examined: cash retainer, which does not include meeting fees; stock awards, which include common stock with and without restrictions, deferred stock units, and phantom stock units; option grants, which include short-term and long-term stock options. The value of compensation is measured using firms’

disclosures in the proxy statement. For firms disclosing the number of stock or option compensation only, the value of stocks is determined by multiplying the number of shares awarded by the closing price. Following Brick et al. (2006) and Core et al. (1999), I compute the value of options using the 25 percent of their exercise price or the closing market price on the annual meeting date if exercise price is not available. I exclude meeting fees because they are often viewed as an opportunity cost of attending a meeting and, thus, are not similar to annual compensation (Adams and Ferreirs 2008). Also, an exclusion of meeting fees avoids a potential mechanical relation with the meeting times (Engel et al. 2010), one of the key control variables in the regression model.

I use the dollar amounts and relative weights of cash, stocks and options in an audit committees’ compensation package to test how the amounts and portions of various compensation components affect restatement likelihood. I refer to these two constructs as the magnitude and percentage approaches, respectively. Under the magnitude approach, I use the natural logs of cash (denoted by ACCASH), stocks (denoted by ACSTOCK), and options (denoted by ACOPTION) paid to the audit committees as my test variables in the REST model to examine the associations between individual compensation components and restatement likelihood.²⁸ Hypothesis H1 predicts the coefficients on ACSTOCK and ACOPTION to be positive. I include an indicator variable CLAWBACK, which is equal to one if firms adopt the clawback provisions in the year t, and 0 otherwise into the REST model. Following Chan et al. (2012a), I predict that the coefficient of CLAWBACK to be negative. Hypothesis H2 predicts the coefficient on ACSTOCK×CLAWBACK

28I use skewness and kurtosis statistics to test the normality of the compensation data. Both tests reject the null hypothesis that compensation amounts are normally distributed (p < 0.01), and this result is robust to total compensation as well as to various compensation components. To ensure normality for the regression analyses, I us the natural logs to transform the dollar amounts of cash, stocks, and options.

‧

Under the percentage approach, variables ACCASH%, ACSTOCK%, and ACOPTION% are the ratios of cash, stocks, and options to total compensation, respectively. I thus replace ACCASH, ACSTOCK, and ACOPTION by these three percentage variables in the REST model. H1 predicts the coefficient of ACSTOCK% and ACOPTION% to be positive. Furthermore, according to hypothesis H2, I predict the coefficients on ACSTOCK%×CLAWBACK and ACOPTION%×CLAWBACK to be negative.