Audit Pricing of Shared Leadership

ABSTRACT

This study explores audit implications of shared leadership in client firms. Analyzing data from 2002 to 2013 of Korean listed companies, we find that auditors spend fewer audit hours and charge lower audit fees for clients with multiple CEOs. Additional tests reveal that the lower audit fees for co-CEO clients are likely attributable to reduced audit effort rather than to reduced hourly rates. We also document that firms with co-CEOs exhibit better-reporting quality than do firms with a solitary CEO. In sum, this article presents evidence that mutual monitoring via co-CEO appointments assures high-quality financial reporting of audit clients, and thus leads to reduced audit fees.

KEY WORDS:

• audit effort
• audit fees
• audit hours
• co-CEOs
• shared leadership

JEL CLASSIFICATION:

• M42
• G34

Acknowledgments

We are grateful for helpful comments and constructive suggestions from Ali Kutan (editor), two anonymous referees, Jong-Hag Choi, Samuel Cheung (discussant), Lee-Seok Hwang, Eunchul Lee (discussant), seminar participants at Seoul National University, conference participants at Korean Accounting Association International Winter Conference, Korea Accounting Information Association Autumn Conference, and 2017 JCAE Annual Symposium. Woo-Jong Lee appreciates financial support from the Institute of Management Research at Seoul National University.

Notes

1. Throughout the article, we use “multiple CEOs” and “co-CEOs” interchangeably to indicate shared leadership.

2. “The co-CEO model is officially dead”, Business Insider (Feb 2, 2012). See http://www.businessinsider.com/the-co-ceo-model-is-officially-dead-2012-2.

3. “Samsung Co-CEO to take charge of struggling display unit”, Wall Street Journal (April 29, 2016). See http://www.wsj.com/articles/samsung-co-ceo-to-take-charge-of-struggling-display-unit-1461918779.

4. https://geert-hofstede.com/south-korea.html.

5. http://www.businessinsider.com.au/why-major-companies-have-2-ceos-2014-9 .

6. As long as the demands for co-CEO leadership arise from significant changes to organization forms or complexity in operations, we expect that such demands would have a significant impact on audit efforts and fees. In other words, these demands might work as correlated omitted variables in a relation between co-CEO presence and audit variables. We thus construct the propensity-score-matched sample, and conduct additional analyses to address this problem. The results based on the matched sample are further discussed in Section 5.

7. Unlike prior studies based on audit fees models which use the assumption that governance affects auditing but not the reverse, Griffin et al. (2008) state a framework that unifies the joint relation between audit fees, hence audit quality, and corporate governance, and conduct tests wherein increased audit fees create a mechanism for better governance, which in turn offsets audit fees by reducing the price of risk. They provide empirical evidence of a negative relation between governance and audit fees, not because auditing affects governance, but because auditing as a governance mechanism affects the interaction of governance and audit risk.

8. Bae, Choi, and Rho (2016), using the Korean audit fee and audit hour data, show that error terms are significantly heteroscedastic in the raw hourly fee model and becomes homoscedastic when the raw hourly fees are transformed into the natural logarithm. The heteroscedasticity of the error term cause inefficient estimation by the least square model. So, We use the natural log transformation of hourly audit fees, following Bae, Choi, and Rho (2016).

9. In Korea, IFRS adoption has been mandatory since 2011 while 14 companies opted for early adoption of IFRS in 2009.

10. TS2000 does not collect CEO-related information of bankrupted companies while those companies pass through the rehabilitation process. According to Debtor Rehabilitation and Bankruptcy Act (Act No. 12595), “The court selects and appoints persons who are fully qualified to perform duties as custodians after hearing the opinions of the Custodian Committee and the creditors’ consultative council (Article 74).” The custodians perform their duties as CEOs for the period of the rehabilitation procedures. Some companies replace their own CEOs with the custodians while some construct a collective management team where their own CEOs and the custodians work as co-CEOs. Hence, some companies undertaking rehabilitation appear to have two or more CEOs: custodians appointed by the court and their own CEOs appointed by the board. Because the meaning of having multiple CEOs in companies undergoing rehabilitation is different than in other companies, we believe that excluding those companies from our sample is better for testing our research question.

11. In December 2001, Financial Supervisory Service (FSS, the Korean equivalent of the SEC) adopted a rule mandating that companies disclose audit fees and audit hours in annual reports. However, audit fees and audit hours disclosed in 2001, particularly audit hours, are less reliable because 2001 is the initial year of the rule. In addition, large portion of companies reported audit hours in days rather than hours. So, TS2000 does not provide audit fee and hour information in 2001.

12. We exclude companies reporting the number of audit days, not the number of audit hours. Results of all of our analysis do not change when we exclude the fiscal year 2002.

13. For the sake of brevity, descriptive statistics of full sample are presented in Panel A of Table S1, available online.

14. In all regression specification, we adjust the standard errors by firm clusters to correct for heteroscedasticity and serial dependence and include year and industry indicator variables.

15. The univariate analysis of audit fees between co-CEO firms and solo-CEO firms in Table 2 shows higher audit fees for co-CEO firms (151,380 thousands Korean won for co-CEO firms vs. 119,166 thousands Korean won for solo-CEO firms at the 1% significance level).

16. The positive coefficient of CHANGE is not consistent with initial audit fee discounts (a.k.a., low-balling) documented in the literature. However, when we do not include auditor tenure (TENURE) in the regression model, the coefficient becomes negative. We thus conclude that the positive coefficient of TENURE incorporates initial fee discounts, leaving the coefficient on CHANGE positive. Moreover, the positive coefficient on CHANGE might be attributable to high audit fees on mandatory auditor rotation or in auditor designation cases. From 2006 to 2011, Korea had adopted mandatory audit firm rotation system which requires firms to rotate their audit firms periodically. Because of high bargaining power of the incoming auditors in such cases, initial audit fees tend to be higher than they do in other cases (Kwon, Lim, and Simnett 2010).

17. When we use fees per hour (FEE) instead of the natural logarithm of the fees per hour (ln(FEE)), we also find an insignificant coefficient on co_CEO.

18. In addition, we conduct a Fama–MacBeth (Fama and MacBeth 1973) regression analysis to control for cross-sectional dependence. We conduct the regression analysis year by year and untabulated results show a significantly negative mean coefficient on co_CEO for audit fees, audit hours, and hourly rate of audit fees, consistent with our main analysis.

19. We include return on assets (ROA and LOSS), leverage (LEV) and liquidity (LIQ) in our model to control for business risk. Even after we control for additional variables such as cash flow volatility and sales volatility, our inferences remain unchanged.

20. We do not employ the Heckman 2-stage approach because it is hard to find good exclusion restrictions (i.e., valid Z variables) that have a direct impact on co-CEO selection but no impact on audit fees, hours and hourly rates. With arbitrary exclusion restrictions, Heckman 2-stage model cannot be better than OLS (Lennox et al. 2012). We therefore report results based on the PSM method despite its limitation, namely that the selection is only based on observable characteristics. However, the results based on the Heckman 2SLS are qualitatively similar.

21. To reflect Korean-unique governance mechanism, we include an indicator for chaebol affiliation.

22. All continuous variables are winsorized at the top and bottom 1% of their distribution.

23. The matched sample also does not show significant differences in audit fees and audit hours between the two groups.

24. Shipman, Swanquist, and Whited (2017) point out that to address the concern of potential confounding or misspecified covariates between treatment and control groups, the first-stage model and the second-stage model should be similar in their variables in general.

25. As aforementioned, we use a one-year lagged independent variable in the matching process. The audit pricing model, Equation (1), includes some of those independent variables in the matching process, but uses their current value. For example, lagged value of ln(SEG) is used in the matching process while current value of ln(SEG) is included in the audit pricing model. In this case, we use the current value of those variables for the second stage. But, even when we use all of prior and current year’s value in the second stage, the results remain unchanged.

26. Even when the propensity-score-based matching does not fully address the potential endogeneity problem, it might not be of a serious concern in making inferences. Note that the demands for co-CEO leadership typically arise in firms with complex organizational forms in which we expect further efforts of auditors are required with additional fee charges. Therefore, any unidentified and uncontrolled motives for co-CEO leadership would work against us finding the negative relation between co-CEO structure and audit fees.

27. In an additional test, all regression models include year and industry indicator variables and are estimated with firm-clustered standard errors to correct for heteroscedasticity and serial dependence.

28. Alternatively, we also use the absolute value of the residuals from the Modified Jones (1991) model.

29. Our measure of accounting quality is the absolute value of the residuals from the McNichols (2002) model. So, smaller values indicate less deviation from the nondiscretionary accrual driven by firms’ business operations and indicate better financial reporting quality.

30. Control variables for the accrual quality model are as follows: The natural logarithm of total assets (ln(AT)); sales growth (SGROWTH); sales volatility over the last five years (SVOL); cash flow from operations (CFO); cash flow volatility over the last five years (CFVOL); leverage (LEV); a dummy variables that equals one if a firm has negative net income (LOSS); Zmijewski’s (1984) bankruptcy score (ZSCORE); ratio of net-operating assets to sales (NOA_t-1); operating cycle (OCYCLE); a lagged dummy variables that equals one if an auditor is one of the Big N auditors (BIGN); Annual stock returns (RET). Among control variables for the opinion model, ln(AT), SGROWTH, SVOL, CFO, CFVOL, LEV, LOSS, ZSCORE, OCYCLE, BIGN and RET have the same definition as variables in the accrual quality model. Other control variables for the opinion model are as follows: change in leverage (CHLEV); a lagged dummy variable for a loss (LOSS_t-1); ratio of cash holding to total assets (CASH); lagged accrual quality (AQ_t-1) a firm’s market beta (BETA); and stock return volatility during a fiscal year (VOL).

31. Using a dummy variable for receiving a modified opinion as a dependent variable, we find that co-CEO firms have a lower likelihood of doing so. Specifically, we classify qualified opinions, disclaimers, and adverse opinions as “modified opinions.” The lower likelihood of issuing a modified opinion is attributable to clients’ high-quality reporting per se or impairment of auditor independence. We thus draw caution of readers in interpreting this result…

32. We adjust ROA based on the SIC 2-digit industry classification.

33. However, controversies exist on whether and how board sizes and the chaebol structure are related to the quality of corporate governance. For example, prior studies indicate that the implication of chaebol affiliation to governance quality is not clear (e.g., Zeff 2007). In other countries; however, family ownership might represent good quality in corporate governance (e.g., Khan, Muttakin, and Siddiqui 2015; Setia-Atmajaa, Hamanb, Tanewskib, 2011). We then do not emphasize the positive relation between board size (chaebol structure) and audit variables since it is beyond the scope of our article.

34. According to Oliver, Marwell, and Teixeira (1985), in more general production function, i.e., S-shaped curve, the collective action will “top out” when marginal returns to a participant’s contribution change to a decreasing slope from an increasing slope.

35. The majority of empirical articles citing the critical mass theory (Oliver, Marwell, and Teixeira 1985) consider the “critical mass” as a threshold, which is about having enough participants so that meaningful collective actions can be achieved. But, this understanding only applies to accelerating production. Oliver, Marwell, and Teixeira (1985) provide their argument in various ways, adopting several cases involving production: Accelerating function (i.e., convex production function), decelerating function (i.e., concave production function), and third-order S-shaped curve (i.e., production function of having convex curve at initial stage and concave curve at later stage).

36. This is reminiscent of a longstanding debate about the relation between board size and firm value. For example, Yermack (1996) shows that smaller and more independent boards are better boards. In contrast, Coles, Daniel, and Naveen (2008) document that the relation between Tobin’s Q and board size is on average U-shaped.

37. The number of firm-year observations with solo-CEO is 3,818 (i.e., 6,186 – (1,906 + 399 + 63) = 3,818). CEO1 is omitted because it is perfectly predicted by the combination of CEO2, CEO3, and CEO>3.

38. Out of 6,186 observations, we identify 413 (405) observations appointing (terminating) co-CEOs. When we run change regressions where all regression variables take change forms, we do not find evidence that such changes in management structure is related to the change in audit fees. The insignificant results are attributable to the limited number of co-CEO changes as well as limited variations in audit fee changes.