The Impact of CEO/CFO Outside Directorships on Auditor Selection and Audit Quality

Abstract

We examine whether outside directorships of chief executive officer/chief financial officer (CEO/CFO) and resulting network ties to auditors affect auditor selection decisions and subsequent audit quality. The network ties arise when the CEO/CFO of a firm (home firm) serves as an outside director of another firm that hires an auditor (connected auditor). Using a sample of firms that switch auditors in the post-Sarbanes-Oxley Act period, we find that home firms are more likely to appoint connected auditors. We also find that home firms hiring connected auditors experience a significant decline in subsequent audit quality, compared to those hiring non-connected auditors. Specifically, the increases in the likelihood of misstatements, the magnitude of absolute discretionary accruals, and the propensity to meet or beat earnings benchmarks after home firms appoint connected auditors are significantly greater, compared to those for other firms switching to non-connected auditors. We further find that the decline in audit quality is more pronounced when the network is established at the local office level.

Keywords:

• CEO/CFO outside directorship
• Auditor selection
• Audit quality
• Auditor independence

JEL codes:

• G34
• M40
• M42

Acknowledgements

We thank Juha-Pekka Kallunki (Associate Editor) and two anonymous referees. We also thank Pietro Bianchi, Qiang Cheng, Ah-rum Choi, Jong-Hag Choi, Inder Khurana, Jeffrey Ng, Hai Lu, Claudine Mangen, Marleen Willekens, Teri Yohn, Xin Zheng, and workshop participants at Erasmus University Rotterdam, Frankfurt School of Finance and Management, KAIST, Sungkyunkwan University, Tilburg University, WHU, Yonsei University, 2015 American Accounting Association Annual Meeting, 2018 BAFA Auditing Conference, 2018 Canadian Accounting Association Annual Conference, 2016 European Accounting Association Annual Congress, 2015 European Auditing Research Network Ph.D. Workshop, 2018 International Symposium on Audit Research Annual Conference, 2015 Korean Accounting Association Annual Meeting, 2019 MedAR / AF Joint Conference, and 2019 MIT Asia Conference in Accounting for their constructive comments. Zang acknowledges financial support from the School of Accountancy Research Center (SOAR) at Singapore Management University.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data Availability

Data are available from the public sources cited in the text.

Notes

1 Geletkanycz and Boyd (2011) and Ruigrok et al. (2006) call these two views ‘the embeddedness view’ and ‘the agency view,’ respectively. We follow these studies and use the same terms.

2 Among senior executives, we focus on the interlocking of CEO/CFOs because they play the most important roles in financial reporting and auditor selection. We discuss this in detail in section 2.1.

3 Furthermore, the likelihood of appointing a connected auditor can be higher when the auditor exhibits superior audit quality for the connected firm. We examine this possibility in a later section.

4 According to the U.S. Census Bureau Office for Management and Budget, a metropolitan statistical area (MSA) refers to a geographical area normally with a large city and its neighboring areas in the U.S. Prior auditing studies often use MSAs to identify the geographic location of a local audit practice office.

5 Consistent with economic agents with pre-existing relationships enjoying better information flow, Cohen et al. (2008) and L. Cohen et al. (2010) document that Wall Street money managers and financial analysts benefit from their social ties with managers of public firms. Similarly, Engelberg et al. (2012) find that the presence of interpersonal links between firm and bank managers improves monitoring by facilitating the exchange of information between lenders and borrowers.

6 Regulators also recognize that familiarity or trust can be a threat to auditor independence. Specifically, Guide to Professional Ethics of the Institute of Chartered Accountants in England and Wales (ICAEW) recommends that auditors avoid situations that may lead them to become over-influenced or to be too trusting of the client’s directors and management which could consequently lead to audit staff being too sympathetic to the client interest (para 2.5 of Integrity, Objectivity, and Independence).

7 We note that Lennox and Yu (2016) examines the network ties of (both inside and outside) directors and executives to auditors, similar to our study. They find that firms are more likely to appoint auditors with whom directors and executives are acquainted through external directorships and that hiring those auditors is positively associated with auditor tenure and audit quality. Since the roles and incentives of outside directors are quite different from those of executives, our focus on CEO/CFO-auditor ties establishes a clearer setting to examine top executives’ motives for hiring connected auditors (i.e., embeddedness view vs. agency view). Lennox and Yu (2016)’s inconsistent results on audit quality may come from examining the interlocks of executives and outside directors together, despite their different roles and incentives, and/or using a different research design. Our additional analysis with a DID research design in a subsequent section indicates that hiring auditors connected to home firms’ AC members through external directorship does not significantly affect subsequent audit quality. Unlike Lennox and Yu (2016), we also document that the effect of CEO/CFO-auditor ties on auditor selection and audit quality is more pronounced when the tie is established at the local office level.

8 Using audit office data in Audit Analytics, we check whether home firms hire an audit office to which CEO/CFOs have a connection. In alumni affiliation research, it is almost impossible to identify an audit office where an affiliated officer worked in the past. Studies on school ties use non-U.S. data and thus generalizability to the U.S. is uncertain.

9 Since we use a DID research design in a non-random setting as in prior literature (e.g., Francis et al., 2017; Jiang et al., 2018). Nevertheless, our research design may not completely solve the endogeneity concerns because auditor change does not occur randomly. To further mitigate the endogeneity concerns particularly related to the analysis of subsequent audit quality, we combine our DID research design with propensity-score matching technique, as discussed later. It is also noteworthy that firms switch their auditor at a different point in time, so our DID research design is staggered. Hence, we do not believe that our results are driven by macroeconomic factors we are unable to observe.

10 The AICPA’s auditing standards in the U.S. (AU section 325) states that auditors are required to directly report to the board of directors if they become aware that ‘the oversight of the company’s external financial reporting and internal control over financial reporting by the company’s audit committee is ineffective.’ (https://pcaobus.org/Standards/Auditing/Pages/AU325b.aspx)

11 McCracken et al. (2008) document that, when audit firms assign their audit partners, they consider client CFOs’ preferences for certain partners, suggesting that the relationship between client CFO and audit partner is important for auditing. One interviewee of Dodgson et al. (2020) states, ‘Management can express a preference to the AC, because management wants to make sure that they get somebody they can work with and that knows their business and that can deal with issues in a timely manner.’ Another interview participant says, ‘You’re generally not going to see an AC insist on engagement partner that the management team objects to. I think the AC understands the working relationship aspects of this too.’ The evidence indicates the importance of the relationship between client executives and auditors.

12 It is possible that the connected auditor is unwilling to compromise independence, despite the CEO/CFO’s bargaining power, given that SOX implemented numerous steps to improve audit quality and auditor independence. Moreover, the newly created Public Company Accounting Oversight Board (PCAOB) increased both oversight and penalties for audit-related deficiencies. Under this possibility, the CEO/CFO’s great bargaining power may not result in lowered audit quality.

13 The effect of the network ties on auditor selection and audit quality can be even stronger if the ties are developed with the same engagement audit partner for both home and connected firms. Since the disclosure of engagement audit partner only came into effect in 2017, we do not have enough data to perform meaningful analyses for this possibility.

14 From the database, CEOs are identified based on the following titles: CEO, interim CEO, co-CEO, group CEO, chief executive (officer), group chief executive (officer), company leader, and group leader. Similarly, CFOs are identified based on the following titles: CFO, co-CFO, interim CFO, group CFO, CFO (part-time), chief financial/finance (officer), and principal financial/finance (officer).

15 BoardEx provides biographical information about senior managers and board members. The database started to collect the information in 2003, backfilling data to 2000. In 2005, BoardEx carried out a major extension of its coverage, backfilling data to 2003, which substantially increased the coverage. Our exploration of the database reveals that the number of U.S. firms covered by BoardEx increased from 2,028 in 2002 to 4,154 in 2003. Its coverage gradually increases in subsequent years, providing annual data for more than 5,000 firms in recent years. Despite the extended coverage, we might fail to identify some CEO/CFOs’ external directorships because BoardEx does not cover all public firms in the U.S. However, this failure is likely to bias against our findings.

16 Among 757 auditor switching firms in the final sample, we find that the CEO/CFOs of 513 firms do not serve as outsider directors of any firms covered by BoardEx. The CEO/CFOs of 162 firms serve as outside directors of only one firm in the BoardEx universe. The CEO/CFOs of 56 (17, 8, 1) firms have two (three, four, five) external directorships, so some have connections to more than one audit firm.

17 Our results are qualitatively similar when we use unbalanced panel data without this restriction.

18 Due to the smaller coverage of I/B/E/S, the sample for the analysis of meeting/beating analysts’ consensus forecasts is limited to 596 (149 unique firms). The sample size for this analysis is commonly smaller than for other audit quality analyses such as misstatements or discretionary accruals (e.g., Reichelt & Wang, 2010).

19 We measure variables in the year immediately before auditor changes, consistent with Lennox and Park (2007) and Dhaliwal et al. (2015). For conciseness, we omit firm and year subscripts.

20 Note that firms currently hiring XX (e.g., PwC) are not able to switch to XX (e.g., PwC). Thus, we estimate Eq. (1) after dropping firms whose predecessor auditor corresponds to XX.

21 Executives’ former working experiences in the audit profession are manually collected from proxy statements of our sample firms. Based on this, we construct AlumniXX. Similarly, we construct two other alumni-related variables, AlumniConn and AlumniAud, which serve as a control in regression equation (2)(2) $\begin{array}{rl}HiringConn& ={\text{β}}_0+{\text{β}}_{1}ConnAQ+{\text{β}}_{2}GovIndex+{\text{β}}_{3}SameMSA+{\text{β}}_{4}SameInd+{\text{β}}_{6}MatchConn\\ & +{\text{β}}_{7}AlumniConn+{\text{β}}_{8}SpecConn+{\text{β}}_{9}FBig4+{\text{β}}_{10}LogTA\\ & +{\text{β}}_{11}BankruptcyScore+{\text{β}}_{12}LitIndustry+{\text{β}}_{13}AudDismissal+\epsilon \end{array}$(2) and (3), respectively.

22 The most straightforward way to examine whether home firms are more likely to hire auditors from the same connected office is to estimate equation (1)(1) $XX={\alpha }_0+{\alpha }_{1}ConnXX+{\alpha }_{2}SpecXX+{\alpha }_{3}MatchXX+{\alpha }_{4}AlumniXX+{\alpha }_{5}FBig4+\epsilon$(1) while treating each of the audit offices as a distinct auditor. However, this approach is not feasible because it requires running numerous regressions (for each of the Big 4 audit firms’ audit offices, which total more than 250) with only few observations hiring a specific audit office. Alternatively, we use the location of a home firm’s incoming auditor’s office, which is available ex post, to infer where the home firm looks for its incoming auditor. Using this information, we test whether the likelihood of hiring a connected auditor is stronger when the incoming auditor’s office and the connected auditor’s office are located in the same MSA.

23 To construct GovIndex, we first calculate the sum of the following indicator variables: I (the CEO does not hold the position of chairman), I (the CEO/CFO is externally hired), I (the home firm’s board independence is equal to or greater than its median of our sample), I (the home firm’s co-opted directors is lower than its median of our sample), and I (the home firm’s AC includes at least one accounting expert) where I (.) is the operator to return one if the condition of the argument is satisfied, and zero otherwise. For each of five dimensions, if I (.) yields one, then the firm has a strong corporate governance for the dimension. We then define GovIndex as one if the sum is equal to or greater than the median of our sample.

24 Another popular measure of audit quality is the auditor’s propensity to issue going-concern opinions. We are unable to employ this measure because all firms switching to a connected auditor in our sample receive a clean audit opinion for both pre- and post-auditor-switch periods.

25 Among 757 auditor switching firms, 90 (608) firms switched from a non-connected auditor to a connected (non-connected) auditor, forming our treatment (control) group. These sample sizes are greater than those of Dhaliwal et al. (2015), who find that, among 420 post-SOX Big 4 appointments, 52 (368) firms switched to an affiliated (non-affiliated) auditor. Note that we exclude 48 (11) firms that switched from a connected auditor to a non-connected (another connected) auditor from our audit quality test samples to obtain clean treatment and control firms. Since the number of these firms is too small, we could not implement meaningful tests for the changes in audit quality.

26 As described in our sample selection process earlier, our sample for audit quality analyses includes two-year observations immediately before and after auditor change respectively. All variables are measured at their-fiscal-year end, so they are time-varying. For conciseness, we omit firm and year subscripts.

27 Prior studies (e.g., Kim et al., 2003) argue that auditors tend to be more concerned about their clients’ income-increasing misstatements which are more likely intentional and egregious. From the entire population from Audit Analytics, we confirm that about 86% of the restatements are income-decreasing ones that resulted from income-increasing misstatements. While we exclude income-decreasing misstatements from the sample, untabulated results reveal that our results are qualitatively similar irrespective of whether we classify income-decreasing misstatements to misstatement sample or not. When we further limit our misstatement sample to those with accounting-related misstatement, we find that our results remain qualitatively similar.

28 Our untabulated analyses show that the results are qualitatively similar when MeetConsensus is defined as one if earnings meet or beat the latest analysts' consensus earnings forecasts by one cent per share or less, and zero otherwise, and MeetLast as one if the firm’s earnings in this year meet or beat its last year earnings by one percent of the market capitalization at the beginning of the year, and zero otherwise.

29 Since our auditor switches occur in Compustat fiscal years 2003–2015, the Post variable captures years up to 2017.

30 Ai and Norton (2003) show that, in a logit model with interaction terms, the effect of the interaction term on expected probability can be different in sign from the coefficient loading on the interaction term. However, Puhani (2012) shows that, when the interaction term is simply the product of a treatment group dummy variable (e.g., Treat) and a treatment period dummy variable (e.g., Post), the sign of the treatment effect is equal to the sign of the coefficient of the interaction term. Based on insights derived from this study, we believe that it is appropriate to infer the sign of the treatment effect based on the sign of the Post * Treat coefficient, as we have done.

31 In all models for audit quality tests, continuous variables are winsorized at 1% and 99%, and the p-values are calculated with client firm-clustered standard errors.

32 When we employ 33.3% as an alternative benchmark, which is a random probability that a Big 4 auditor is switched to one of the other three Big 4 auditors, the difference is still significant at p < 0.001.

33 To evaluate the economic importance of ConnXX against that of AlumniXX, we check whether the coefficient on ConnXX is statistically different from that on AlumniXX in each of columns (1)–(4). We find that the differences are statistically insignificant in all columns, suggesting that the economic magnitude of the effect of ConnXX on auditor selection decisions is comparable to that of AlumniXX.

34 Our results are robust when we add a bankruptcy score, leverage, an indicator for the issuance of debt and equity, board independence, and an indicator for CEO-chairperson duality, following Lennox and Park (2007).

35 When we estimate the misstatement regression in column (1) without the interaction term, Post * Treat, we continue to find that the coefficient on Post is negative and statistically significant.

36 A negative coefficient on Treat implies that, in the pre-auditor switch period, treatment firms are less likely to misstate their financial statements, relative to control firms. This outcome could derive from differences in firm characteristics between two groups, such as firm size. To mitigate the concern about differences in firm characteristics, we replicate our analysis using the propensity score matched sample in a subsequent section.

37 Shipman et al. (2017) state that ‘PSM should not include variables in the matching stage that are excluded from MR’ [multiple regression].’ We follow this guideline in performing PSM.

38 To keep all treatment firms, we do not require a maximum caliper distance in this PSM matching. Our inferences, however, remain similar when we enforce maximum caliper widths of 0.05, 0.1, and 0.3, although the sample size in each analysis becomes smaller. In addition, the results are qualitatively similar when we replicate the analyses without replacement.

39 One disadvantage of our PSM model is a small sample size relative to the number of predictors, which reduces the statistical power of our tests. To alleviate this concern, we alternatively estimate each model using bootstrap. For each analysis, we generate 100 datasets from the original sample. The number of observations in each of the 100 samples is the same as the number for the original sample. Untabulated results reveal that the results are qualitatively similar.

40 A possible reason for insignificant differences is small size of the sample that switched to connected auditors, which could lower statistical powers in our models. Among 757 firms switching to Big 4 auditors, the number of firms with CEO/CFO-auditor ties via AC membership is 143, while that with CEO/CFO-auditor ties via board membership is 244. Moreover, the number of firms switching to connected auditors based on AC memberships is only 46, compared to 101 based on board membership.

41 To examine this possibility, we adopt a DID research design in which we compare the changes in audit quality from the pre- to post-auditor-switch periods for the connected firms of our treatment firms, with the changes in audit quality for other connected firms of our control firms, using the same audit quality proxies.