Do Joint Audits Improve Audit Quality? Evidence from Voluntary Joint Audits

Abstract

This study examines whether the decision to voluntarily (i.e. without a statutory obligation) employ two audit firms to conduct a joint audit is related to audit quality. We use separate samples and empirical designs for public and privately held companies in Sweden, where a sufficient number of companies have a joint audit on a voluntary basis. Our empirical findings suggest that companies opting to employ joint audits have a higher degree of earnings conservatism, lower abnormal accruals, better credit ratings and lower perceived risk of becoming insolvent within the next year than other firms. These findings are robust to the use of a propensity score matching technique to control for the differences in client characteristics between firms that employ joint audits and those that use single Big 4 auditors (i.e. auditor self-selection). We also find evidence that the choice of a joint audit is associated with substantial increases in the fees paid by the client firm, suggesting a higher perceived level of quality. Collectively, our analyses support the view that voluntary joint audits are positively associated with audit quality in a relatively low litigious setting both for public and private firms.

Acknowledgements

We appreciate the comments received from Ann Vanstraelen (the associate editor), two anonymous reviewers, Jean C. Bedard, Kris Hardies, Henry Jarva, Robert Knechel and Petri Sahlström. We would also like to thank the participants at the Third Workshop on Audit Quality in Lake Como, Italy (2010), the 34th European Accounting Association Annual Congress in Rome, Italy (2011), the 17th Annual International Symposium on Audit Research in Québec, Canada (2011) and a research seminar at the University of Vaasa, Finland (2010). Elina Haapamäki gratefully acknowledges the financial support received from the Jenny and Antti Wihuri Foundation, the Marcus Wallenberg Foundation, the Foundation for Economic Education and Nordea Bank Foundation. Mikko Zerni also gratefully acknowledges the financial support received from the Finnish Cultural Foundation, the Finnish Foundation for Economic Education, and the Marcus Wallenberg Foundation. This research is part of the research project of the Academy of Finland, Grant Numbers 140000 and 126630. Any remaining errors are our own.

Notes

Several recent high-profile reports in the US, UK and European Union have raised similar concerns. See for the US: United States General Accounting Office (2003, 2008), Advisory Committee on the Auditing Profession (2008), Center for Audit Quality (AICPA) (2008), The American Assembly (2005) and US Chamber of Commerce (2006, 2007). For the UK: Oxera Consulting (2006), Audit and Assurance Faculty (ICAEW) (2005) and Office of Fair Trading (2004). For the European Union: London Economics (2006), Oxera Consulting (2007) and Commission of the European Communities - Directorate General for Internal Market and Services (2008) (cf. Sirois and Simunic, 2010).

An easier and more cost-effective solution could be to mandate joint audits for only some specific class of client firms (e.g. financial intermediaries and very large companies). For instance, joint audits in Sweden are mandated only in the finance sector. As another safeguard to protect the (perceived) integrity of the audit process, the second auditor for the finance sector firms is appointed by the Swedish Financial Supervisory Authority (Finansinspektionen) and not by the client firms themselves. These regulations may be interpreted as meaning that the legislature in Sweden has taken the perspective that joint audits would provide more independent and higher quality assurance on the fairness of the client firms' financial statements. The current credit crisis has further amplified discussion regarding the effectiveness of corporate governance devices. Effective monitoring mechanisms are of crucial importance, especially with respect to the finance sector, in which corporate governance failure can lead to major negative market-wide externalities such as the drying up of liquidity in the financial markets.

It is important to note that in our analyses, there are no pairings of non-Big 4 firms.

In general, the research on the effects of joint audits on audit quality and fees is scarce because joint audits are unusual. In Europe, the only country currently requiring mandatory joint audits is France, which has done so since 1966 (though Denmark also did so from 1930 until 2004). Other countries with a joint audit requirement include Algeria, Morocco, the Ivory Coast, Tunisia, Congo, Saudi Arabia and Kuwait. The main argument for abolishing the joint audit requirement in Denmark was that it was a burden on Danish companies (Hasselager et al., 1997). Consistent with this view, Thinggaard and Kiertzner (2008) report that after the abolition of two-auditor system in 2004, 15 of the 63 (23.8%) companies investigated retained two auditors in the next year. However, it is noteworthy that more than one-fifth of the companies still decided to retain two audit firms. Exploring the determinants of this decision after regime change would potentially provide some interesting insights into audit demand. An interesting concurrent working paper by Holm and Thinggaard (2011) addresses these issues, among others.

Interestingly, the descriptive statistics presented in Panel B of Table 7 for publicly-listed client companies actually show that some of the clients that employ joint audits purchase all of their non-audit services from the auditor that has a smaller share of the audit fee revenues. In this regard, note that neither the Swedish law nor the Swedish Code of Corporate Governance impose any specific guidance regarding the extent and type of consulting services that would not be within acceptable limits. Nevertheless, in some rare occasions the provision of advisory services might cause an auditor to resign from an audit engagement.

We wish to thank an anonymous reviewer for making this point.

FAR SRS (the professional institute for authorised public accountants, approved public accountants and other highly qualified professionals in the accountancy sector in Sweden) offers group insurance to all of its members. The minimum coverage per claim is approximately €900,000.

Even though lawsuits against auditors are rare in Sweden, the legal liability risk exists. Specifically, there are out-of-court settlements and court cases. For an example of a court decision awarding damages to investors (banks) that purchased shares in a company (granted credit) on the basis of inaccurate accounts certified by the statutory auditor, see: NJA 1994:63 (NJA 1996:224). There is no legal liability cap in Sweden, and third parties can sue the auditor within ten years from the time that the damage occurred. A statutory audit is considered to be in the interest of not only the company but also the public. As a result, any third party may recover damages from the statutory auditor upon providing proof for the elements of the liability claim, which usually focuses on fault, damages and causation.

Entrenchment problems arise from the possibility that large shareholders opt to use their power to expropriate minority shareholders by taking actions and investment decisions serving their own interests leading to suboptimal outcomes for outside shareholders (see for instance, Claessens et al., 2000; La Porta et al., 1999). Separation of control (voting rights) from ownership (cash-flow rights) further exacerbates the agency problem because the controlling shareholder will not bear the full cost resulting from his/her decisions.

Note that the low number of publicly listed Swedish companies does not allow us to employ propensity score-based matching for the subsample of public firms. This is simply because sufficiently close peer firms are not available to enable a successful matching process.

The initial sample of observations includes all of the auditors and their clients, appointed either as an audit partner in-charge or as a deputy auditor for at least one publicly listed company from 2001 to 2008. For more details, we refer the reader to Knechel et al. (2012), who employ a similar data set from the same source.

Propensity score matching aims to simultaneously isolate a wide set of client characteristics from the treatment effects. The major advantages of the propensity score matching method (or any other peer-based matching approach) are that it does not rely on a specific functional form and does not require appropriate exogenous instrumental variables (exclusion restrictions in the first stage, a difficult condition to meet for models predicting auditor choice) (Lawrence et al., 2011; Lennox et al., 2012; Li and Prabhala, 2007).

Inferences remain qualitatively similar if the replacement is matched. Distance is the difference between the predicted probabilities of choosing the treatment according to the estimated logit model between matched pairs (Dehejia and Wahba, 2002; Lawrence et al., 2011).

In untabulated results, we estimate and match the propensity scores within each economic sector rather than using indicator variables using the following industry sectors: manufacturing, retail, services, hotel and restaurants, transport, real estate and other/miscellaneous. Results from these analyses yield similar inferences to those reported in all relevant aspects. Note that because of the relatively low number of unique firms employing joint audits, we cannot use more precise industry categories.

Note that in each test, missing information for one observation leads to the exclusion of two observations: the treatment observation and its propensity-score-matched counterpart.

Note that critiques of the Basu (1997) earnings conservatism model also exist (see, for example, Dietrich et al., 2007; Gigler and Hemmer, 2001; Givoly et al., 2007; Jarva, 2010).

For the sake of brevity, we do not report the correlation matrices.

Approximately 27% of the joint audit observations have negative operative cash flow, compared with 26% for the propensity-score-matched clients of the single Big 4 auditors.

However, while the differences in estimated parameters appear to be economically significant, the LR-test statistic for equality of the parameters R*DR*JOINT and R*DR*BIG cannot be statistically rejected. Furthermore, VIF values for some of the interaction effects are high because of the common underlying variable R. However, all these values are below 11. According to Belsley et al. (2005), collinearity is a potential problem in a regression when the condition number is above 20 and a severe problem if it is above 30.

The problem with using the Jones (1991) model and its extensions is that there are limited numbers of Swedish firms in specific industries, resulting in substantially fewer observations and less precise parameter estimates. Small industry samples have been cited as one reason why the Jones model does not perform well on non-US data (Meuwissen et al. 2007). The DeFond and Park (2001) model is not curtailed by the number of within-industry observations and has been employed in several other recent audit research papers (e.g. Carey and Simnett, 2006; Francis and Wang, 2008; Francis et al., 2009; Maijoor and Vanstraelen, 2006).

We have also tested whether the level of abnormal accruals or our other proxies for audit quality are significantly different for the joint audits conducted by two Big 4 audit firms compared with a pairing of Big 4 and non-Big 4 audit firms (untabulated). More specifically, we allowed the coefficient on our test variables to vary between joint audits conducted by two Big 4 audit firms. The results of estimating these model specifications show that the additional variable(s) is (are) not significant throughout the estimations while the inferences from the joint audit indicator (JOINT) and its interactions remain unchanged. Hence, the empirical evidence suggests that joint audits per se are better in terms of audit quality than the audits conducted by a single Big 4 or a single non-Big 4 auditor. However, because of the relatively small number of unique firms employing joint audits, especially in the public firm sample, these additional tests may suffer from low statistical power. Note that to avoid severe multicollinearity problems, we cannot conduct a similar model extension regarding Basu's earnings conservatism tests reported in panel A of Table 3. In particular, in these analyses the variance inflation factors (VIF) would be over 30 because of the common underlying variable stock return (R).

The UC AB stated, ‘Risk Forecast states how great the probability is that a company will become insolvent within the next year, i.e., it states with great precision the risk that the company will be unable to fulfill its payment obligations’. In forming the rating (risk forecast) for a limited liability company, UC AB considers the following aspects, among others: accounting information, key ratios, payment complaints, board information and ownership structure. For more details about the credit ratings and risk forecasts, see https://www.uc.se/en/source.php/1084640/UC%20Riskf%F6retag%20-%20Description%20eng.pdf.

Maximum likelihood methods do not compute the sums of squares. However, by calculating the residual and total sums of squares, we can report the adjusted R-square statistics to facilitate comparability between related studies.

Although privately-held Swedish clients are required to report information about their audit fees in the notes of their financial statements, there is no database from which this information can be readily extracted. To economise the data collection process, we rely on our sample of publicly listed companies when examining the effect of joint audit selection on audit fees.

The audit firm that receives the largest amount of revenue through audit fees is designated as the first auditor.

We also estimate the model by including a random firm intercept instead of firm-fixed effects (assuming that the random intercept is normally distributed and treating its non-zero mean and variance as parameters). The results that we obtain from this analysis are qualitatively similar to those reported in Table 8. However, the LR-test statistics strongly support the use of a firm-fixed effect extension over random effects (χ ² = 240.5, p-value < 0.0001).

However, it should be noted that with respect to largely time-invariant variables, the use of firm-fixed effects in regressions may remove true (cross-sectional) variation and thereby fail to detect a relationship in the data even when it exists (e.g. Zhou 2001).

To confirm that the results are not caused by a few outlying observations, we either trimmed or winsorised all of the continuous variables at the 1% and 99% levels and re-estimated all of our models. In these analyses, the empirical findings remain qualitatively similar to the main results reported in this study.

We wish to thank an anonymous Reviewer for noting this implication.