Enforcement Actions and Auditor Changes

Abstract

In this paper, we examine whether the uncovering of erroneous financial statements by German enforcement agencies is related to subsequent auditor changes. We argue that enforcement actions are likely to reveal information about the client or its auditor, which might affect auditor choice by initiating an update of mutual expectations. Our empirical findings indicate that firms with erroneous financial statements indeed have an increased probability of subsequent auditor changes. Firms also tend to change from a non-Big4 auditor to a Big4 auditor in this situation, suggesting that clients increasingly seek the reputation and services of Big4 auditors. Big4 auditors in turn do not appear to refrain from taking over error-firms as new clients in the German setting, which is characterized by limited auditor liability. Additionally, auditor changes are more likely to occur before the public announcement of an error, indicating that firms take action as soon as the uncovering of an accounting error becomes sufficiently certain.

Acknowledgments

We thank participants at the 2016 IAS Midyear Meeting in New Orleans, the EAA Annual Congress 2016 in Maastricht, the VHB Annual Meeting 2016 in Munich, the EARNet 2015 Conference in Lausanne, the 2015 GEABA Symposium in Hamburg, as well as workshop participants at Karl-Franzens-University Graz, University of Magdeburg, and University of Mannheim for their valuable comments and suggestions. We are particularly grateful to Luzi Hail (associate editor) and an anonymous reviewer for their extensive and constructive feedback. Comments by Ted Azarmi (GEABA discussant), Pietro Andrea Bianchi (IAS discussant), Ralf Ewert, Joachim Gassen, Dirk Simons, and Stefan Sundgren (EARNet discussant), as well as financial support from the Julius Paul Stiegler Memorial Foundation are also gratefully acknowledged.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1 Throughout the paper, we use the expressions ‘auditor change’ and ‘auditor switch’ interchangeably. We refer to changes of audit firms and not to changes of individual audit engagement partners.

2 In contrast to the German setting, the enforcement system in the United Kingdom (UK), for example, was previously rather informal, with only a low number of error findings being published (Armour, 2009).

3 DeFond and Zhang (2014, p. 278) note that litigation and reputation incentives are difficult to disentangle in a US setting. Our analysis complements studies such as Hope and Langli (2010), who provide evidence on audit quality based on a Norwegian private firm setting with low litigation costs.

4 The General Standard and the Prime Standard represent the entire regulated market in Germany (see Deutsche Börse, 2016, p. 16). However, disclosure requirements for the Prime Standard are higher than for the General Standard.

5 Whereas both Mande and Son (2013) and Hennes et al. (2014) focus on auditor changes after restatement announcements, another stream of literature (e.g. Files et al., 2014; Lennox & Pittman, 2010) uses auditor changes to predict restatements. Note that our setting differs considerably from the one used in these studies. In the US, a newly appointed auditor can use the opportunity to ‘set the record straight’ by restating previous financial reports. In contrast, restatements we are investigating are always initiated by the enforcement agency.

6 Although the FREP slightly adjusted its sampling procedures in 2009, procedures have not changed materially compared with those applied between 2005 and 2009.

7 The latter examinations can be requested by the BaFin or be initiated by the FREP's Pre-Review Committee, which systematically searches media coverage for indications of accounting malfeasance.

8 DAX, MDAX, SDAX, and TecDAX are indices that are part of the Prime Standard of the Frankfurt Stock Exchange. The blue chip index DAX includes the 30 largest German companies listed in the Prime segment in terms of market value. The MDAX consists of 50 mid-cap firms from the non-technology sectors that rank below the companies included in the DAX. The SDAX contains 50 small-cap firms from the non-technology sectors that rank below the companies included in the MDAX. Finally, the TecDAX contains 30 mid-cap firms from the technology sector.

9 Since 2016, the responsibility to sanction a breach of professional obligations by an auditor has been assigned to a newly established governmental body (Abschlussprüferaufsichtsstelle).

10 We describe the mutual updating process from an auditor and a client perspective, because the auditor and the supervisory board dominate the auditor selection process. In Germany, the supervisory board proposes an auditor that is then formally elected by the general assembly. However, a proposal is typically not substantiated in detail and is usually confirmed by the shareholders.

11 In our analysis, we compare firms for which an error has been identified with the remaining firms that were subject to FREP enforcement in a respective year. Please note that FREP investigations remain unknown if no accounting error is detected. The full list of examinations is not publicly available.

12 In addition to informal interviews with auditors from the enforcement competence centers of two Big4 auditors, we conducted structured interviews with a managing director and a partner responsible for the enforcement services of a Big4 company.

13 We provide further details on the construction of the variables in the Appendix (Table 8).

14 For simplification, we drop firm and time subscripts if not needed.

15 The impending publication that is mandatory for all accounting errors uncovered under the German enforcement system might affect the client's decision-making, for example by providing increased incentives to commit to a higher accounting quality in the future. We use the publication date as an alternative reference point in Section 5.4.

16 We use lagged explanatory variables in our auditor change models to avoid potential endogeneity.

17 In our tests, Acquirer and Target are not included in lagged form because they relate to more than one year by definition.

18 For further construction details of the misalignment proxy, refer to Appendix B.

19 Industries are classified into 10 groups based on Fama and French (1997).

20 To avoid double counting of individual errors, we eliminate error announcements that are mere rephrased or amended versions of original announcements.

21 We repeat our analyses without dropping these observations and include an indicator for overlapping periods as an explanatory variable instead. Our results are virtually unaffected by this alternative approach.

22 Additionally, the number of publications declines toward 2014 because we cover only financial reports through the end of 2012.

23 Böcking et al. (2015) report an average time span of an enforcement investigation of 236 days with a standard deviation of 131 days, using proprietary data provided by the BaFin. The authors define the investigation period to be from the date of the notification letter to the date of the decision letter.

24 We use a Wald test because firm-clustered standard errors make a likelihood ratio test unfeasible (Long & Freese, 2014). Our approach generally biases against supporting a good fit of the specification compared with the standard approach, which compares the full model specification with a model containing only the intercept. We additionally use a Hosmer–Lemeshow test to assess the goodness of fit of our logistic models and the multinomial logistic models (Fagerland, Hosmer, & Bofin, 2008; Hosmer, Lemeshow, & Sturdivant, 2013). Associated p-values are well above 10% for all our tests indicating no evidence of misspecification despite a relatively low number of auditor changes.

25 Our results are robust to the inclusion of additional control variables such as a variable capturing changes in audit fees and a Big4 dummy variable. We also conduct separate regressions for firms with a Big4 and a non-Big4 incumbent auditor. The results suggest that significance levels with respect to our variables of interest are higher for clients with an incumbent non-Big4 auditor. However, the increase in the probability of an auditor change is also marginally significant for Big4 clients.

26 The results are not affected by the inclusion of various additional variables that we use in our main specification in the first-stage regression.

27 First-stage regression results are available upon request.

28 We keep using the control variables because our matching approach might not completely eliminate differences with respect to these variables. However, the results for our variables of interest are robust to excluding the control variables from the second stage. The results also persist if we base our matching on a parsimonious model that is based only on those control variables that are significant in the full sample regressions.

29 We further exclude the fixed-effects dummies in the first stage to reduce potential noise. Results are similar if we use the full specification in the first and the second stage.

30 We additionally run firm fixed-effects regressions to control for unobserved time-invariant characteristics of each firm. Results are similar.

31 The number of top-level issues ranges from 1–10, with a median of two.

32 We repeat our analysis in a logistic framework in which we use a binary indicator for each category of , respectively. Untabulated results confirm our findings from the multinomial logistic model.

33 We exclude variables that by construction resemble the change in other control variables, for example, the variable Growth resembles the change in Size. Furthermore, we do not consider ΔMismatch because it specifically refers to auditor changes.

34 The respective group contains one firm-year observation for which the audit report lag decreases by more than 55%.