Learning from Masters: Engagement Partners’ Co-Signing Relationships with Non-Engagement Industry Specialist Partners and Audit Quality

Abstract

This paper examines a specific mechanism, partner co-signing relationships, through which knowledge sharing in an audit team may influence audit quality. Specifically, we examine whether an engagement partner’s co-signing relationships with non-engagement industry specialist partners are associated with audit quality. Using a unique setting in which co-signing relationships for each audit engagement is available, we find that audit quality is higher when the engagement partner for the focal client has co-signing relationships with non-engagement partners who are specialists in the focal client’s industry and that the positive association is more likely driven by a learning effect than a consultation effect. Further evidence suggests that attributes of co-signing relationships such as continuity matter and that co-signing relationships with industry specialist partners are more likely to be utilized when engagement partners have limited industry knowledge, clients operate in homogeneous industries, and engagement partners are from large audit firms. Overall, our results suggest that interactions with industry specialist partners facilitate knowledge sharing and hence improve audit quality.

Keywords:

• Individual Audit Partner
• Industry Expertise
• Audit Quality
• Knowledge Sharing

Acknowledgements

We thank Mark DeFond (editor) and two anonymous reviewers for their valuable comments and guidance. For their helpful comments, we thank the workshop participants of the 9th Asia–Pacific Interdisciplinary Research in Accounting Conference, the 42nd Annual Congress of the European Accounting Association, and a research seminar at Griffith University. This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. Errors and omissions are our responsibility.

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.

Supplemental Data and Research Materials

Supplemental data for this article can be accessed on the Taylor & Francis website, https://org/10.1080/09638180.2022.2062409 Appendix A. Sample distribution Appendix B. Descriptive statistics of entropy balanced sample

Notes

1 European Union adopted the Eighth Company Law Directive in 2006, which requires that the engaged partner sign the audit report. Similarly, the PCAOB adopted a new rule in 2015 that requires the engaged partner to disclose their identity in Form AP (PCAOB, 2015a).

2 An audit report in Taiwan is required to be signed by two engagement audit partners, who are explicitly accountable for audit reports issued in their names in cases of sanctions for audit failures and problematic audits.

3 The use of discretionary accruals to proxy for audit quality is supported by DeFond and Zhang (2014) and Francis (2011). Discretionary accruals represent a continuous measure of audit outcomes, consistent with the continuum concept of audit quality (Francis, 2011), and ‘arguably the most widely used proxy for auditor constraints on client reporting decisions in prior literature’ (Chi et al., 2017, p. 1575). In addition, Aobdia (2019) finds a significant positive association between discretionary accruals and PCAOB inspection deficiencies.

4 Aobdia (2019) supports the use of meeting or beating earnings benchmarks to proxy for audit quality, because this measure is significantly related to the measures of audit process deficiencies used by auditors and PCAOB.

5 Our study is related to a growing literature that focuses on knowledge transfers among partners. For example, Bianchi (2018) and Bianchi et al. (2020) examine a unique setting in Italy where private companies’ financial reports are jointly audited by three individual auditors who can come from different audit firms and find that clients engaging better-connected individual auditors exhibit better financial reporting quality and have comparatively lower effective tax rates. Their findings suggest that joint engagements (which could involve multiple audit firms) facilitate knowledge transfer and increase auditor expertise. Huang et al. (2021) show that audit quality is higher when the two engagement partners have more co-working experience in the past, and Hu et al. (2021) find that auditors teamed up with the sanctioned auditors are more likely to issue lenient audit opinions and have accounting irregularities in their audited financial reports during the sanctioned CPAs’ audit-failure period. Our research extends theirs by showing that co-signing relationships with non-engagement industry specialist partners within an audit firm improve an engagement partner’s audits of public clients. Together with these studies, we enlarge the understanding of how social connections influence knowledge transmission among auditors and thus, audit outcomes.

6 Although there may be competition among partners and ‘each partner bears the full cost of her own effort but reaps only 1/N of the benefit of an N-partner firm’ (Lennox & Wu, 2018, p. 3), knowledge sharing is motivated by the potential liability arising from audit failures by other partners in the same audit firm. Audit firms in Taiwan are required to be structured as unlimited liability partnerships or proprietorships, under which structure, individual partners face high levels of potential liability (Chen et al., 2008).

7 For further information on control systems in audit firms, see also Bedard et al. (2008), Dowling (2009), Epps and Messier (2007), Francis et al. (2014), Nelson and Tan (2005), and Rich et al. (1997).

8 All the companies in our sample have their fiscal year end at December 31.

9 To illustrate, assume that partner A and partner B sign the audit report of client 1, and partner A and partner C sign the audit report of client 2, and assume also that partners C is an industry specialist partner of the client industry and clients 1 and 2 operate in the same industry. For client 1, CON equals 1 (a co-signing relationship with industry specialist partners of the focal client industry exists: partner C). If clients 1 and 2 operate in different industries or if client 1 is audited by partner A and partner C, then CON equals 0 for client 1.

10 Our inferences remain similar, but the results become weaker when we use a longer period to calculate the measure of co-signing relationships.

11 Although there are differences in the responsibilities and the level of involvement between lead and concurring partners, it doesn’t mean that concurring partners add no value to audit engagements.

12 This approach is validated by Aobdia et al. (2015, p. 2152), in their private conversations with the Big 4 audit firms that ‘the lead partner tends to be the one that has the longer tenure with the client prior to the adoption of the partner rotation rules and the practice changed around the adoption of the new rules (2004/2005)’ and that ‘now, it is common practice that the lead partner signs the audit report first.’ Empirically, Aobdia et al. (2015) further validate this approach to identifying lead partners by showing that including lead partner fixed effects substantially increase the adjusted R² of the audit quality model but including concurring partner fixed effects does not and that market participants respond to the quality of lead partners but not concurring partners.

13 Industry is classified based on the Taiwan Stock Exchange industry classification.

14 We thank the anonymous reviewer for suggesting this approach.

15 Please see Audousset-Coulier et al. (2016) for a review on the measures of industry specialist auditors.

16 We thank the anonymous reviewers for this suggestion.

17 We do not use audit fees as the base because audit fee data are not available for many years during our sample period.

18 Although discretionary accruals are not directly associated with violations of accounting standards (e.g., International Financial Reporting Standards or Generally Accepted Accounting Principles), they capture within-standards earnings management, which is likely to reduce financial reporting quality (DeFond & Zhang, 2014). In addition, given that discretionary accruals are a continuous variable, they better capture the nuanced difference in audit quality and are more in line with the continuum concept of audit quality (Francis, 2011).

19 As in Chi and Chin (2011), we focus on working capital accruals because previous studies suggest that managers have the greatest discretion over working capital accruals (Ashbaugh et al., 2003; Becker et al., 1998; Chin et al., 2005).

20 The results remain unchanged when we replace LAGROA with current ROA.

21 At least 10 observations are required for each year–industry grouping to run model (1).

22 Anecdotal and academic evidence suggests that managers and the market participants focus on whether firms are able to meet or beat estimates of earnings per share. For example, Facebook experienced a 20 percent decrease in stock price after it announced earnings of $3.67 per share, which was lower than the consensus earnings estimate of $3.84 per share (Feiner, 2022). Fox (1997) suggests that ‘the simplest, most visible, most merciless measure of corporate success in the 1990s has become this one: Did you make your earnings last quarter?’ Consistently, Graham et al. (2005, Table 7)’s survey results indicate that managers are likely to give up a profitable project if pursuing the project will make the company fail to meet the estimate of earnings per share, and Jiang (2008) shows that beating last year’s earnings per share is associated with credit rating upgrades in the following year. These suggest that managers have strong incentives to meet or beat last year’s earnings per share. A similar specification has been used in Chin et al. (2009) which also examine the setting in Taiwan and in several studies on meeting or beating earnings benchmarks (e.g., Chapman & Steenburgh, 2011; Davis et al., 2009; Koh et al., 2008; Myers et al., 2007).

23 We do not use analyst forecasts as earnings benchmarks because of the scarcity of analyst forecasts in Taiwan.

24 We test two alternative MBE measures. MBE_A equals one if the changes in earnings per share between year t and t−1 divided by the stock price at year t−1 are between 0 and 0.05, and zero otherwise. MBE_B equals one if the changes in earnings per share between year t and t−1 divided by the stock price at year t−1 are between 0 and 0.005, and zero otherwise. The mean value of MBE_A is 0.187, substantially higher than that of our original MBE measure (0.019), and the mean value of MBE_B is 0.032. Our results remain consistent when using MBE_B as the dependent variable, but become weaker when using MBE_A as the dependent variable.

25 Our measures of audit quality are output based, which is well suited for audit quality research (DeFond & Zhang, 2014). This is because discretionary accruals ‘capture within-GAAP manipulation and quality variations for a large number of firms’ (Goodwin & Wu, 2016, p. 350). Using discretionary accruals is consistent with the concept of audit quality because accruals models attempt to capture whether financial reporting reflects the underlying economic condition of a company, which is an important goal of audit engagements (Aobdia et al., 2015; DeFond & Zhang, 2014). Further, using earnings decrease avoidance avoids the estimation errors that result from the discretionary accruals models. Importantly, Aobdia (2019) shows a significant association between meeting or beating earnings benchmarks and measures of audit process deficiencies used by auditors and the PCAOB, which validates the use of this variable as a proxy for audit quality.

26 Throughout this study, we use robust standard errors, with clustering along the client firm dimension.

27 Chen et al. (2018) suggest that incorrect inferences may occur when using residuals as dependent variables. It should be noted that Chen et al. (2018) do not directly speak to studies using transformed residuals. Nevertheless, we include all the first-step regressors in model (2) when the dependent variable is |DACC| or +DACC (Chen et al., 2018, p. 782). This approach is more suitable to our study because it allows us to have a larger sample size when estimating discretionary accruals and allows us to study income-increasing discretionary accruals. Results are consistent when removing the first-step regressors from model (2).

28 Our research is fundamentally different from Bianchi (2018) and Bianchi et al. (2020) in that in our sample, the two engagement partners of a client come from a same audit firm.

29 We control for lead and concurring partner fixed effects in our models. We thank the anonymous reviewer for this suggestion.

30 In our sample, audit partners rarely switch audit firms. Therefore, audit partners fixed effects subsume audit firm fixed effects.

31 Our results remain unchanged when we include year and industry fixed effects and estimate an ordinary least squares regression when the dependent variable is |DACC| or +DACC and a logistic regression when the dependent variable is MBE.

32 On December 31, 2019, TWD1 equaled USD0.0334.

33 All the continuous variables are winsorized at the 1% and 99% percentiles.

34 The results are available upon request.

35 0.008 / 0.019 = 0.4211.

36 We thank the anonymous reviewer for suggesting this test.

37 Gaver and Utke (2019) indicate that ‘entropy balancing is an equal percent bias reducing matching method, which ensures that covariate imbalance improves after matching’ and that entropy balancing increases testing power because no observations are discarded and there are no random matches (King et al., 2011). In contrast, although propensity score matching (PSM) is often used in the literature, PSM does not ensure the similarity of the treatment and control firms, fails to solve the functional form misspecification (covariate imbalance) due to random matches, and reduces testing power due to discarding observations (DeFond et al., 2017; Gaver & Utke, 2019; King & Nielsen, 2019; King et al., 2011; McMullin & Schonberger, 2020; Shipman et al., 2017). The problem of low-power tests is particularly relevant for our study because our variable of interest (CON) has low frequency. In their response, Gaver and Utke (2019) conclude that entropy balancing is preferred ‘because it accounts for observable differences across clients without random matching or discarding data (Hainmueller, 2012; McMullin & Schonberger, 2020)’ and suggest that PSM ‘should be used with caution or not at all.’

38 We do not use the number of clients as an alternative base because an untabulated analysis shows that the measure of industry specialist partners based on the number of clients is not significantly associated with our audit quality measures.

39 Instead of the portfolio share approach, we also construct measures of industry specialist audit partners using the market share approach. However, none of these alternative measures is significantly associated with our audit quality measures in an untabulated analysis. This supports our conjecture that the portfolio share approach better suits our study because it focuses on the ‘relative distribution of audit services’ across various industries for each audit partners considered individually and because it captures the extent to which an individual audit partners build their industry knowledge which indicates the ‘degree of industry fit between an individual auditor and a client’ (see Neal & Riley, 2004; Numan & Wilekens, 2012), whereas the market share approach ‘does not take into account the size of the industries’ and ‘fails to recognize that some industries are too small to merit the development of industry specialization’ or that some industries are so large that most auditors are ‘prompted to make major investments in the development of industry specialization through technologies and expertise’ (Audousset-Coulier et al., 2016, p. 145).

40 We do not examine the issuance of modified audit opinions, which is common in studies using data from jurisdictions other than the U.S., for two reasons. First, co-signing relationships with industry specialist partners are more likely to capture audit quality from the auditor competence perspective than the independence perspective. As indicated by DeFond and Zhang (2014, Table 2), audit opinions uniquely capture ‘auditor independence’ and do not capture ‘subtle quality variation.’ It is unclear how co-signing relationships with industry specialist partners affect auditor independence and hence audit opinions. Second, modified audit opinions are very common in Taiwan (e.g., almost 80 percent after 2006), which prevents us from finding a meaningful relationship between co-signing relationships with industry specialist partners and modified audit opinions. Similarly, we do not examine the issuances of going-concern opinions because they capture ‘auditor independence’ and because they are rare and more likely to be issued to distressed clients, leading to low statistical power (DeFond & Zhang, 2014). In addition, prior studies suggest that going-concern opinions are not associated with earnings management and are frequently issued in error, raising concerns about their validity as a proxy for audit quality (Butler et al., 2004; Chu et al., 2020). Consistent with expectations, our untabulated analyses suggest that CON is not associated with modified audit opinions or going-concern opinions.

41 For example, assume that Joe, David and Cathy are partners in an audit firm and Cathy is an industry specialist partner in the airline industry. If Joe and David audit a semiconductor company TSMC (focal client) in the current year, and Joe and Cathy audit another semiconductor company, MTK, during the same year, CON_OTHER equals 1. However, it should be noted that Cathy is not an industry specialist in the semiconductor industry.

42 In an untabulated analysis, we further decompose CON based on the timing when the co-signing relationships with industry specialists are formed. Specifically, CON_ t equals 1 if a lead partner had co-signing relationships with industry specialists in year t, and 0 otherwise. CON_ t−1 equals 1 if a lead partner had co-signing relationships with industry specialists in year t−1 but such relationships do not continue in year t. CON_ t−2 equals 1 if a lead partner had co-signing relationships with industry specialists in year t−2 but such relationships do not continue in years t−1 to t. CON_ t−3 equals 1 if a lead partner had co-signing relationships with industry specialists in year t−3 but such relationships do not continue in years t−2 to t. CON_ t−4 equals 1 if a lead partner had co-signing relationships with industry specialists in year t−4 but such relationships do not continue in years t−3 to t. CON_ t−5 equals 1 if a lead partner had co-signing relationships with industry specialists in year t−5 but such relationships do not continue in years t−4 to t. CON_ t−6 equals 1 if a lead partner had co-signing relationships with industry specialists in year t−6 or earlier but such relationships do not continue in years t−5 to t. We find some evidence that in general the coefficients on the co-signing relationships are more significant when such relationships are formed between year t and year t−2 and insignificant when such relationships are formed before year t−2 and discontinue in recent years. This suggests that the value of co-signing relationships will vanish when co-signing relationships with industry specialists are formed previously and do not continue in the recent years and thus supports our focus on the recent years when measuring co-signing relationships.

43 For example, assume that partner A and partner B sign the audit report of client 1, and partner A and partner C sign the audit report of client 2, and assume also that partners C is an industry specialist partner of client 1 industry but clients 1 and 2 operate in different industries.