Audit-Firm Portfolio Characteristics and Client Financial Reporting Quality

Abstract

This paper contributes to the audit quality literature by defining continuous measures of expected future audit-firm losses and testing their association with proxies of financial reporting quality. In prior studies audit-firm size has been used as a proxy for expected future audit-firm losses and hence – following theoretical arguments in DeAngelo (Journal of Accounting and Economics, 3(3), pp. 183–199, 1981) and Dye (Journal of Political Economy, 101(5), pp. 887–914, 1993) – for audit quality. In particular, the Big 8/6/5/4 indicator variable has been tested empirically against various measures of client financial reporting quality. In this paper, we focus on testing various characteristics of an audit firm's client portfolio as drivers of audit quality (and therefore subsequently also client financial reporting quality), including measures to proxy size, visibility and financial health characteristics of an audit-firm portfolio. We test both a disclosure and earnings quality model for that purpose, and find for a sample of Belgian companies in financial distress, that audit-firm portfolio characteristics better explain variations in client financial reporting quality than the traditionally used Big N indicator variable. In particular, we find that – ceteris paribus – the size of an audit-firm portfolio is irrelevant in explaining the variation in financial reporting quality amongst companies. Next, we find that client visibility characteristics of an audit-firm portfolio have a constraining impact on earnings management, but no impact on the disclosure quality in the notes. Third, we find that solvency characteristics of an audit-firm portfolio (but not liquidity and profitability characteristics), are positively associated with the financial reporting quality. Taken together, the evidence suggests that not so much the size of an audit-firm portfolio (and the audit firm) but other portfolio and client characteristics drive audit and financial reporting quality.

Acknowledgements

The authors are listed in alphabetical order. We are grateful to J. Francis, N. Huyghebaert, R. Knechel, T. Laitinen, C. Lennox, B. Rees, S. Sundgren and L. Van de Gucht for suggestions and remarks on previous versions of this paper. We acknowledge the comments of participants at the Midyear Auditing Conference of the AAA and the Accounting Workshop Day at KULeuven.

Notes

1. Other size proxies suggested in the literature but less frequently tested include: (1) the number of audit-firm partners, (2) the number of professionals, (3) the number of CPAs, (4) the number of clients and (5) total personnel (see, for example, Colbert and Murray, 1998; Krishnan and Schauer, 2000).

2.  The impact of the auditor size variable has also been tested in other settings. A significant relationship exists between the auditor size and audit fees, which is then explained as indirect evidence of product differentiation by the large audit firms (see, for example, Simunic, 1980; Palmrose, 1986; Francis and Simon, 1987; Gist, 1992; Craswell et al., 1995). In capital market research, larger earnings response coefficients (see Teoh and Wong, 1993) are registered, while the amount of underpricing at the time of an IPO is smaller when a Big A auditor audits the financial statements (see, for example, Balvers et al., 1988; Beatty, 1989).

3. A firm is obliged to prepare and publish financial statements in case of limited liability of the owners. The amount of financial information provided depends on the size of the firm. Large firms must publish the ‘full scheme’ of the financial statements and are required to appoint a statutory auditor. At that time, large firms were defined as companies that met at least two of the following size criteria: total assets > 3,125,000 euros; turnover > 6,250,000 euros; number of employees > 50. Firms that employ more than 100 employees are always classified as large firms irrespective of the size of the total assets or turnover.

4. Note that there is NO mandatory rotation after three years.

5. This implies an average of 12.6 initiated cases per year. Note that the average number of chartered auditors in Belgium between 1990 and 1999 was about 800. So, on average, annually less than 2% of the auditors were subject to initiation of a disciplinary case.

6. This hypothesis is motivated taking the specific features of the Belgian institutional environment into account. As only a limited number of companies are listed on the Belgian stock exchange (see, Knechel and Willekens, 2006), we do not link visibility to the listing status of the client firms. However, we do link visibility to an auditor's exposure towards the client firm's employees, as the auditor is required to clarify the financial situation of the firm and explain the issued audit opinion to the employees in the works council. Note also that employees are powerful stakeholders, as they have to coincide with the choice of auditor suggested by the board of directors before the auditor can be appointed. It is likely that employees and their unions will oppose the re-appointment of an auditor who is not very informative during the works councils.

7. Given this definition, a lot of firms meet these criteria, which implies that this study covers a very significant portion of the Belgian corporate scene.

8. In the robustness checks, we investigated how sensitive the results are to the use of median values to characterize the audit-firm portfolio.

9. The computation of these variables is based on audit-client level data (thus individual company data). For the collection of client-level data we used the Belgian National Bank (BNB) database, and select all firms that appointed a statutory auditor in 1997. Next, we identified all audit firms in which these statutory auditors were active through inspection of the membership lists of the Belgian Institute of Auditors. We find that 165 different audit firms were active in the Belgian statutory audit market in 1997, and that together they audited 15,450 client firms. Finally, we collect financial statement data for these 15,450 client firms from the BNB database that serve as a base to compute the audit-firm portfolio observations for each of the 165 audit firms.

10. Different approaches exist for resolving the issue of how many factors to extract from a data-set. However, the two criteria that are widely used to assist in deciding how many factors to extract are Kaiser's eigenvalue rule and Catell's scree test (DeVellis, 1991; Hair et al., 1998).

11. We conclude on the five-factor solution based on the results from the exploratory factor analysis. To test how well the data fit this hypothesized five-factor model, we additionally performed a confirmatory factor analysis. In assessing the model fit, we found a Goodness of Fit Index (GFI) of 0.84 and a GFI Adjusted for Degrees of Freedom (AGFI) of 0.75.

12. In the sensitivity analysis, we used other alternatives than summated scales and results are similar.

13. In previous disclosure studies market-based control variables (trading volume; Scott, 1994) are also added. However, as all firms in the sample are privately held, those are relevant in the current study. There exists also evidence that managers manipulate earnings during periods when they or their companies are selling shares to capital markets (Beneish and Vargus, 2002).

14. The probability of bankruptcy is computed using the model developed by Gaeremynck and Willekens (2003). This model is preferred above other bankruptcy models as it uses Belgian companies to estimate the likelihood of business termination and therefore takes account of institutional features typical to the Belgian environment. The Bust model predicts the likelihood of business termination using the following model:

where AR_c is a dummy which equals 1 if the firm c receives a non-clean report; ROE_c is return on equity for company c; STDEBT_c the ratio of short-term debt divided by total debt for company c; OPROFIT_c is a dummy equalling 1 when the firm realizes operating income and 0 otherwise; NCASH_c = (Cash + Short-term investments − Short-term bank debt)/Total assets; EQFIN_c = (Retained earnings + Reserves)/Total assets for company c; LIQ_c = Cash/Current assets for company c.

15. The impact of CEO incentives on accounting choices are not considered in this paper (compensation contracts, e.g. Healy et al., 1985; Leung and Horwitz, 2004; Bergstresser and Philippon, 2006) also influence financial reporting choices but are not studied in the current context due to the lack of data. Privately held firms do not have to disclose any information about the relation between management and the firm. As all firms in the sample are privately held, no market-based measures of performance or stock exchange listing variables are introduced in the analysis as control variables.

16. Companies need to file their financial statements with the National Bank within 30 days after the approval by the general assembly.

17. The entire population of firms in difficulties in 1997 amounts to 3,846 firms. However, due to incomplete data, we could only include 3,278 firms in this analysis.

18. The multivariate results are not distorted by those extreme values. When those observations are deleted from the sample, multivariate results remain the same.

19. From the correlation matrix (reported in the Appendix) it is clear that results are not distorted by multicollinearity problems.

20. In developing these models, we preferred to use the same variables as in the basic model. We do not re-run the factor analysis as the current approach offers the advantage that the models can be directly compared to the basic model.