Measuring audit quality

See Appendix Table 13 for the most frequently cited audit deficiencies.

St. Pierre and Anderson (1984) describe audit deficiencies found in 129 lawsuits against accountants in the 1960s and ‘70s, but this classification predates much of GAAS. Beasley et al. (1999) and Beasley et al. (2013), in separate reports commissioned by the American Institute of CPAs (AICPA) and the Center for Audit Quality (CAQ), respectively, also report descriptive data on audit deficiencies identified by the SEC for 56 and 81 AAERs for the periods 1987–1997 and 1998–2010. Our sample is more comprehensive in that we also cover 153 nondismissed lawsuits against auditors over the period 1996–2016. Moreover, there are substantial differences in the nature of deficiencies identified by the SEC when compared with the class action lawyers, as detailed later in the paper.

We focus on convergent validity, which we test empirically as the association between each audit quality proxy and audit deficiencies raised in AAERs and lawsuits. These deficiencies provide us with unique information about process-level audit failures, which ideally should have predictable associations with widely used proxies for audit quality.

Moorthy and Sarath (2017) examine the likelihood of auditor lawsuits and settlement in securities class action lawsuits. They find that auditors are more likely to be sued and settle the suit in lawsuits involving alleged earnings manipulation and GAAP violations. They, however, do not include AAER samples nor examine the validity of audit quality proxies.

Some of these cases related to the company’s audit report, but the SEC did not pursue the auditor directly. For example, in AAER 3063 (SEC vs. China Holdings, Inc. and its CEO), the CEO forged the audit report and the auditor resigned. The SEC sued the company and its CEO, but it did not sue its auditor.

This could mean that either (i) there could be data errors in the ISS database or (ii) the auditor was dismissed while other defendant(s) remained in the lawsuit. Regardless, we exclude cases where auditor’s name does not appear on the complaint.

When coding the audit deficiencies in lawsuits, we look for sections where the complaint lists all the audit standard violations for defendants. Usually this section is named as “Defendant Auditor’s Violation of Auditing Standards” or something similar. If this section is missing, we go through the entire document to look for alleged audit deficiencies. Specifically, we look for terms such as “the auditor violated a certain GAAS standard.” We exclude cases where no concrete violations of auditing standards are alleged.

A skeptic might worry that AAERs and lawsuits have different objectives and that it is unclear whether these cases can be pooled. We pooled both the public and private lawsuits for analyses, because they conceptually set precedents for future lawsuits against auditors. As a robustness check, in untabulated work, we re-estimated Tables 4, 5, 6, 7, 8, 9 and 10 after including the lawsuit indicator as an independent variable. The findings are similar to the main results reported in the paper, where we excluded the lawsuit indicator variable.

We also validate discretionary accruals, DA, and absolute value of discretionary accruals, AbsDA. However, we only report results of Total Accruals for brevity, since it is well known that these variables are highly correlated. We estimate DA using the cross-sectional modified Jones model, following the literature (e.g., Jones 1991; Kothari et al. 2005; DeFond and Zhang 2014). We subtract the derived nondiscretionary accruals from accruals to obtain signed discretionary accruals. AbsDA is the absolute value of DA. In untabulated results, we find that DA and AbsDA are not associated with any individual allegations. Additionally, both variables do not load in combined regressions if we use either to replace Total Accruals

We define a city as a Metropolitan Statistical Area (MSA) for all variables at the city level, consistent with Reichelt and Wang (2010).

There are costs and benefits associated with using constant versus varying sample. We allow the sample size to vary for each regression model because of data availability and our attempt to best preserve the power of our tests, consistent with research design choices in the literature (e.g., Aobdia 2019). We also use a constant sample to re-estimate models in Tables 4, 5, 6, 7, 8, and 9. The untabulated results remain similar.

For completeness, we report correlation matrix in Appendix Table 14.

Lennox and Li (2019) and Moorthy and Sarath (2017) examine the determinants of an auditor involved in a class action lawsuit. Lennox and Li (2019) find no association between restatements and the likelihood of an auditor being sued. But Moorthy and Sarath (2017) find that auditors are more likely to be involved in lawsuits involving restatements of earnings. The different conclusions are likely due to research design choices (i.e., different sample periods and control variables). Our study has different research objectives and finds that restatements are associated with several types of alleged audit deficiencies found in lawsuits against auditors. Note that the work of Lennox and Li (2019) and ours differ in sample composition, unit of analysis in regressions, sample period, and control variables. First, Lennox and Li (2019) consider class action lawsuits and material federal civil lawsuits identified in the litigation database in Audit Analytics. We study AAERs and class action lawsuits. That is, their sample does not include all the AAERs. Our sample does not include civil lawsuits other than class action lawsuits. Second, their sample period covers 2000 to 2015. Our violation years span 1978 to 2016. Moreover, their sample includes both dismissed and contested lawsuits. However, we only include nondismissed class action lawsuits to eliminate frivolous allegations. Third, the unit of analysis differs. Lennox and Li (2019) analyze at the lawsuit level. We analyze at the violation firm-year level, because each AAER and nondismissed lawsuit may involve multiple violation years.

According to the literature, the city-level industry specialization measure (City Specialist) suffers from a few limitations. First, City Specialist could proxy for city-level industry specialization or client characteristics (e.g., Minutti-Meza 2014). Second, the literature has various proxies for auditor industry specialization. These in general exhibit low internal and external construct validity, according to Audousset-Coulier et al. (2016). We therefore suggest that readers exercise caution when interpreting our results of City Specialist.