Evidence from Socially-tied CEOs and Audit Committee Members
4. EMPIRICAL RESULTS
4.4 Additional Tests
4.4.1 Controls for endogeneity
The results in Tables 3~5 show that auditors perceive that only employment ties in the
4Note that the coefficient on EDU is significant in the S&P 500 sub-sample (p < 0.10), suggesting that auditors will consider education ties to be serious and react by resigning from S&P 500 clients who have low fraud risk. Since this result bears no practical implication and disappears in our subsequent tests of endogeneity, it is a random result and does not deserve further discussion.
23
non-S&P 500 firms will harm AC oversight quality and react accordingly. However, the possibility of endogeneity is a concern in our analysis. Prior studies identify three major sources of endogeneity that may arise in our setting. The first source is time-invariant unobservable heterogeneity, which arises if there are unobservable firm-specific factors that affect both the CEO-AC social ties and auditors' reactions. To address this endogeneity problem, we follow Bruynseels and Cardinaels (2014) by re-estimating all the previous regressions with lagged values of the three social tie measures.
In our audit effort test, Panel A of Table 6 shows that the coefficients on LAGEMPLOY are positive and significant in the full sample (p < 0.01) and the non-S&P 500 sub-sample (p < 0.10). The coefficients on LAGEDU and LAGOTHERACT remain insignificant across different samples. Similarly, in our conservatism test, Panel B indicates that the coefficients on LAGEMPLOY are significant in the full sample and the non-S&P 500 sub-sample at the 0.10 level. Again, the coefficients on LAGEDU and LAGOTHERACT are still insignificant. Finally, Panel C shows that in our auditor change test the coefficients on EMPLOYHIFRAUD are significant in the full sample and the non-S&P 500 sub-sample at the 0.10 one-tailed level. Overall, Table 6 is consistent with Tables 3~5.
[Insert Table 6 here]
Although using lag analyses helps address the time-invariant unobservable heterogeneity problem, they may not mitigate endogeneity arising when an omitted time-variant variable leads to both a change in social ties and a change in auditors' reactions. We use an instrumental variables approach to address this second source of endogeneity. Because our Tables 3~5 consistently show that auditors care about social ties from employment and react accordingly, we follow Bruynseels and Cardinaels (2014) and focus only on EMPLOY. We use the 1-year lag of EMPLOY as the instrumental variable. We implement the instrumental variables approach as follows. In the first stage, we regress the level of EMPLOY on the level of lag EMPLOY and all the control variables used in Tables 3~5 as
24
exogenous variables. In the second stage, we use the predicted level of employment ties from the first stage regressions (denoted by EMPLOY_PRED) in place of the actual values and estimate regressions of auditors' reactions on this predicted EMPLOY measure and all control variables used in Tables 3~5.
The results of the second-stage regressions are reported in the first three columns of Panels A, B, and C in Table 7. The coefficients on EMPLOY_PRED are significant in our full sample and non-S&P 500 sub-sample for the tests of audit effort, conservatism (at 0.10 level or better), and auditor change (at 0.10 one-tailed level). Taken together, our second-stage results produce very similar conclusions.
[Insert Table 7 here]
A third source of endogeneity is simultaneity, which arises when the variable of interest is also a function of the dependent variable. Specifically, dynamic endogeneity may exist in such a way that the values of the variable of interest in year t are a function of the dependent variable in year t-1. This is a potential concern in our setting because auditors’ past reactions could be related to the current levels of CEO-AC social ties. To address this potential endogeneity problem, we follow prior studies (e.g., Blundell and Bond 1998; Kim et al. 2015; Wintoki et al. 2012) and adopt a system of two equations that include a dynamic model transformed into first-differenced form and a dynamic model in levels form. We then use dynamic GMM to estimate the following system of equations using 1-year lagged EMPLOY as the instrument:
EMPLOYt = a0 + a1REACTIONt-1 + Controls, (4-1)
REACTIONt = b0 + b1EMPLOYt + Controls. (4-2)
The last three columns of Panels A ,B, and C in Table 7 report the results from the dynamic GMM estimations. We continue to find significant coefficients on EMPLOY_PRED at the one-tailed 0.10 level or better. This suggests that, after controlling for endogeneity using the dynamic GMM
25
estimator, auditors appear to react to CEO-AC employment ties only in the non-S&P 500 firms by increasing audit effort, becoming more conservative, or even resigning from clients who have high fraud risk. Overall, the results from the above three econometric methods to control for endogeneity indicate that our major findings reported in Tables 3~5 are unlikely to result from endogeneity.5 4.4.2 Audit fee analysis
The regression model
Prior studies show that audit fees reflect audit risk (e.g., Bedard and Johnstone 2004; Chan et al.
2012; Gul and Tsui 2001). If auditors observe significant audit risk, they are likely to charge higher fees to cover the extra costs arising from additional audit effort and/or the assignment of more experienced staff to the audit (Simunic and Stein 1996). Previous research also documents a positive relation between the aggressiveness of financial reporting and audit fees (e.g., Gul et al. 2003; Hogan and Wilkins 2008; Hoitash et al. 2008). We posit that the auditors will charge higher audit fees if clients’ AC members are socially tied with the CEOs. This is because these ties may impair AC independence and facilitate earnings management, which increases the likelihood of restatements and in turn audit risk. In their review paper, DeFond and Zhang (2014) identify that auditors are more likely to respond to litigation risk by reducing or bearing risk through audit fee adjustments.
Following prior audit fee studies (e.g., Chaney et al. 2004; Ferguson et al. 2003; Francis et al.
2005; Whisenant et al. 2003), we employ the following models (5-1) and (5-2) to test whether auditors adjust higher audit fees (LNAF) in response to the existence of social ties between CEOs and AC members. Similar to our main analyses, we use ALLTIES in model (5-1) and replace it with EDU,
5We also try to use a within-firm sample to test whether there are significant changes in audit effort, conservatism, and auditor changes before and after a firm has an individual social tie. To do this test, we need to make sure that a firm does not have any social tie for a certain number of years and then has a social tie in subsequent years. In other words, each firm should have complete social tie data during our sample period. Missing data in any single year cannot be our observations.
Due to this reason, our within-firm samples for three types of social ties are too small to provide any meaningful regression results. Therefore, we cannot use this method to control for endogeneity.
26
EMPLOY, and OTHERACT in model (5-2). Note that we include the same set of control variables as in Bruynseels and Cardinaels (2014) so that we can make comparisons between our results and theirs.
LNAF = 0 + 1ALLTIES + 2CEOCHAIR + 3CEOTENURE
+ 4BOARDSIZE + 5INDEP + 6FINEXP + 7AUDSIZE + 8LOGTA + 9INITIAL + 10BIG4+ 11LEV+ 12CURR + 13ROA+ 14LOSS+ 15INVREC+ 16BM+ 17FOREIGN
+ 18GROWTH + 19NEWFIN + Fixed effects + , (5-1) LNAF = 0 + 1EDU + 2EMPLOY + 3OTHERACT + 4CEOCHAIR
+ 5CEOTENURE + 6BOARDSIZE + 7INDEP + 8FINEXP + 9AUDSIZE + 10LOGTA + 11INITIAL + 12BIG4+ 13LEV + 14CURR + 15ROA+ 16LOSS+ 17INVREC+ 18BM
+ 19FOREIGN + 20GROWTH + 21NEWFIN + Fixed effects + , (5-2) where all the dependent and independent variables are defined in the appendix.
The empirical results
Panel A of Table 8 presents the descriptive statistics of all variables used in models (5-1) and (5-2). The mean and median values of audit fees (LNAF) are 13.69, consistent with Bruynseels and Cardinaels (2014) and other prior studies. About 17% of all AC members are socially tied with the CEOs in our audit fee sample. In addition, the average board size is about 8 members and 75% of the firms are audited by the Big 4. Note that the means of ALLTIES, EDU, EMPLOY, and OTHERACT (which are 17%, 4%, 11%, and 5%, respectively) are very close to those reported in Table 4 of Bruynseels and Cardinaels (2014), indicating that our sample is comparable to their audit fee sample.
[Insert Table 8 here]
Panel B shows that, when we use the full sample, the coefficient on ALLTIES is negative and significant at the 0.10 level. This implies that auditors charge lower audit fees to clients with CEO-AC social ties. This result, which is consistent with Bruynseels and Cardinaels (2014), does not support our H1. The second and the third columns of Panel B indicate that the coefficient on ALLTIES is negative
27
and significant at the 0.05 level only in the non-S&P 500 sub-sample. These results are consistent with H3. Panel C further indicates that the coefficients on OTHERACT are negative and significant in both the full sample (p < 0.05) and non-S&P 500 sub-sample (p < 0.01) while the coefficients on EDU and EMPLOY are not significant in any sample. These results suggest that auditors react to CEO-AC non-professional ties by charging lower audit fees, consistent with H2. Taken together, our audit fee analyses provide supply-side evidence supporting Bruynseels and Cardinaels’s (2014) findings.6
Discussions of the inconsistent findings
DeFond and Zhang (2014) point out that audit fees are jointly determined by the demand-side firms and the supply-side auditors. From the demand-side, Bruynseels and Cardinaels (2014) show that CEO-AC social ties resulting from OTHERACT will harm AC independence, leading to lower audit fees and audit quality. From the supply-side, our results confirm Bruynseels and Cardinaels’ (2014) demand-side results. The evidence obtained from both sides not only contradicts with prior literature’s conclusion that auditors will raise audit fees to compensate increased audit risk, but also contradicts with our main findings that auditors regard EMPLOY ties as the most critical and react. Notably, given Bruynseels and Cardinaels’ (2014) results that audit fees will be lower, the auditors shall have strong incentive to reduce audit effort. However, we do not find negative and significant coefficients on OTHERTACT in our audit effort model across different samples. Given these results, the critical question we want to ask is: Why are the auditors willing to charge a lower level of audit fees but still put more effort when they identify EMPLOY ties between the CEOs and AC members?
To answer the above question, we first posit that other non-professional activities are private personal actions that may not be observable to the auditors. In contrast, prior employment experience could be relatively easier to observe from auditors’ own social networking. Therefore, auditors are able
6We also separate our sample into 2004-2008 and 2009-2012 sub-samples and re-estimate the audit fee model separately.
The results are still consistent with Bruynseels and Cardinaels (2014).
28
to identify EMPLOY ties and perceive them as a potential threat to AC independence.7 We then posit that the demand-side firms may have stronger bargaining power than the supply-side auditors in determining audit fees. To test this conjecture, we adopt industry expert to capture auditors’ bargaining power because previous research shows that industry specialization leads to a fee premium (e.g., Casterella et al. 2004) and this audit fee premium is driven by city-level industry expertise (e.g., Francis et al. 2005). Following prior studies (e.g., Balsam et al. 2003; Kim et al. 2015; Lim and Tan 2008), we use clients’ sales to estimate industry market share (MS) of an audit firm as follows:
∑
∑ ∑
The numerator is the sum of the sales of all clients of audit firm i in industry k. The denominator is the sales of all clients in industry k, summed over all audit firms in the industry.
We require a minimum of 20 clients in each industry to estimate industry market share for an audit firm in a given year. Auditor with the top two industry market shares are defined as industry experts.
We then separate our sample into Industry Expert vs. Non-industry Expert sub-samples and re-estimate our audit fee and audit effort models. Panels A and B of Table 9 indicate that only in the non-industry Expert sub-sample that the coefficient on OTHERACT is negative (p < 0.05) and the coefficient on EMPLOY is positive (p < 0.01). In other words, Bruynseels and Cardinaels’ (2014) results exist only
7We obtained this information from personal interviews with managing partners, CEOs, chairmen, and senior partners at the Big 4’s headquarters. They kindly provide several reasons. First, because the auditors have to maintain independence in fact, partners at the Big 4 are not allowed to participate in private parties or banquets held by clients’ executives. Therefore, it is relatively difficult for the partners to observe CEOs’ and audit committee members’ private connections. Second, only employment ties involve “professional knowledge” interactions from which the CEOs and audit committee members (especially the chairs) can figure out how to manage earnings and find acceptable reasons to convince audit partners that a reduction in audit fees is necessary. Third, CEOs’ non-professional activities are so diversified that it’s almost impossible to identify them. Also, the frequencies of these activities are usually high. In contrast, CEOs’ prior and current employment ties are highly focused and, once the employment ties are established, they will be stable over time. Finally, to show their appreciation, people who get promoted in a company are more likely to cooperate with their supervisors who promote them. It is relatively easier for the audit partners to know who have been promoted in their clients (especially the S&P 500) from partners’ personal social networking.
29
when auditors are non-experts. Because non-experts have only few clients in each industry, the importance of these clients increases. Also, a lack of industry expertise prevents non-experts from earning audit fee premium. These factors jointly motivate firms with OTHERACT ties to cut audit fees.
Since non-experts may not observe OTHERACT ties but can observe EMPLOY ties, they will react by exerting more audit effort to reduce potential audit risk.
[Insert Table 9 here]
Panels A and B of Table 9 also indicates that the coefficients on EDU, EMPLOY, and OTHERACT are all insignificant in the Industry Experts sub-sample. One possible reason underlying these results is that industry experts have already charged high audit fees. Therefore, when they observe social ties between clients’ CEOs and AC members, they tend to take other actions such as conservatism rather than increasing audit fees. To test whether our conjecture is correct, we estimate the abnormal audit fees for industry experts and non-experts. Untabulated results indicate that the average normal audit fees (in log values) for industry experts and non-experts are 14.04 and 13.05, respectively, and the difference is significant (t = 45.53; p < 0.0001). Similarly, there is a significant difference in average abnormal audit fees (t = 4.31; p < 0.0001) for industry experts (i.e., 0.421) and non-experts (i.e., 0.392). The median tests between these two sub-samples provide the same results.
These statistics suggest that industry experts already charge higher normal and abnormal audit fees than non-experts. Because prior studies indicate that the increase in audit fees is attributable mainly to the increase in audit hours rather than higher rate per hour (e.g., Bae et al. 2016), industry experts have already devoted more effort in their audit work. Therefore, these could be the reasons that industry experts will not charge higher audit fees and put more effort as reactions to the existence of CEO-AC EMPLOY ties.
4.4.3 Effects of Accounting Expertise and Social Ties
30
Regulators regard accounting expertise as one of the most important characteristics of ACs.
Prior studies find strong support for the notion that AC members’ accounting expertise is positively associated with firms’ financial reporting quality (e.g., Agrawal and Chadha 2005; Bedard et al. 2004;
DeFond et al. 2005; Dhaliwal et al. 2010; Krishnan and Visvanathan 2008; Xie et al. 2003; Zhang et al.
2007). However, because AC oversight quality is jointly determined by AC members’ professional knowledge and integrity, accounting expertise by itself cannot maintain or improve AC effectiveness without AC independence. Recent studies find that AC independence may be compromised due to option compensation (e.g., Archambeault et al. 2008; Campbell et al. 2015; Keune and Johnstone 2015), presence of former audit partners (e.g., Naiker and Sharma 2009; Naiker et al. 2013), and social ties between CEOs and AC members (e.g., Bruynseels and Cardinaels 2014). In this study, we strictly define accounting experts as those who have bachelor or master degrees in accounting, prior working experience as an auditor, controller, or internal auditor, and hold CPA licenses. We first separate our full sample into sub-samples with and without at least one accounting expert on the AC. We then estimate our regression models separately on these two sub-samples to examine how auditors view the relative efficacy of expertise and independence on AC oversight quality.
Panels A of Table 10 shows that, if there is at least one accounting expert on the AC, the coefficients on EDU, EMPLOY, and OTHERACT are not significant in our full sample, S&P 500, and non-S&P 500 sub-samples. However, if there is no accounting expert on the AC, only the coefficients on EMPLOY become significant in our full sample and non-S&P 500 sub-sample at the 0.01 level.
These results suggest that the auditors will not exert more audit effort for any social ties when both S&P 500 and non-S&P 500 firms have at least one accounting expert on their ACs. If there is no accounting expert, the auditors only react to employment ties in the non-S&P 500 firms by exerting more audit effort. In contrast, Panel B shows that, if there is at least one accounting expert on the AC,
31
only the coefficients on EMPLOY are significant in our full sample and non-S&P 500 sub-sample at the 0.10 level or better. These results imply that auditors will be more conservative in auditing non-S&P 500 firms’ financial statements even though these firms have accounting experts on their ACs.
Finally, Panel C shows that, for firms who have accounting experts on their ACs, the coefficients on EDU, EMPLOY, OTHERACT, and EMPLOYHIFRAUD are significant at 0.05 level or better.
However, we test and find that only the sum of the coefficients on EMPLOY and EMPLOYHIFRAUD is positive and significant in the S&P 500 sub-sample (p < 0.0346). One possible reason of this result is that the auditors regard accounting experts who have prior or current employment connection with the CEOs to be more capable of managing earnings and more willing to cooperate with the CEOs in hiding earnings management. Therefore, the auditors tend to react by resignation.8 While recent studies have found that larger portion of accounting experts on ACs can effectively enhance AC oversight quality (e.g., Dhaliwal et al. 2010; Krishnan and Visvanathan 2008), our results show that the negative effect of CEO-AC employment ties may outweigh the benefit of having accounting experts on ACs. In contrast, we test and find that the sums of the coefficients on EMPLOY and EMPLOYHIFRAUD are positive but insignificant in the full sample (p < 0.1007) and non-S&P 500 sub-sample (p < 0.1329).
[Insert Table 10 here]
Taken together, Table 10 indicates that, while auditors consistently react to EMPLOY ties only, their reactions vary depending on firm types and whether firms’ ACs have accounting experts. If there is at least one accounting expert on AC, the auditors will react by (1) not exerting more effort to either S&P 500 or non-S&P500 firms, (2) being more conservative to non-S&P 500 only, and (3) resigning from S&P 500. If there is no accounting expert on AC, the auditors will react by exerting more effort
8We check our data and find that, in our full sample for auditor change analyses, there are 2441 S&P 500 firms. Nearly 55%
8We check our data and find that, in our full sample for auditor change analyses, there are 2441 S&P 500 firms. Nearly 55%