Discretionary-accruals models and audit qualifications

Abstract

The primary goal of this study is to evaluate the ability of the Cross-sectional Jones Model and the Cross-sectional Modified Jones Model to detect earnings management vis-à-vis their time-series counterparts by examining the association between discretionary accruals and audit qualifications. These two cross-sectional models have not been formally evaluated by prior research, and their use may offer certain advantages to investors and researchers over their time-series counterparts. A sample of 173 distinct firms with qualified audit reports and a matched-pair control sample with clean audit reports are used. Only the two cross-sectional models are consistently able to detect earnings management. One limitation of this study is that its findings merely indicate the superiority of the cross-sectional models vis-à-vis their time-series counterparts in an audit qualification setting, not validate either the former or the latter.

Introduction

Many earnings management studies examine managers’ use of discretionary accruals to shift reported income among fiscal periods, which entails specification of a model to estimate discretionary accruals. The models range from the simple, in which the change in total accruals is used as a measure of discretionary accruals to the relatively sophisticated, which decompose accruals into discretionary and nondiscretionary components using regression analysis. The most popular six models are the DeAngelo (1986) Model, the Healy (1985) Model, the Jones (1991) Model, the Modified Jones Model (Dechow et al., 1995), the Industry Model (Dechow et al., 1995), and the Cross-sectional Jones Model (DeFond and Jiambalvo, 1994).

Dechow et al. (1995) evaluate the relative performance of five of these models in detecting earnings management by comparing the specification and power of commonly used tests across discretionary accruals generated by the models. They evaluate the specification of the test statistics by examining the frequency with which the statistics generate type I errors and the power of the tests by examining the frequency with which the statistics generate type II errors. Using various samples and assumptions, they demonstrate that all models appear well specified for random samples, generate tests of low power for earnings management, and reject the null hypothesis of no earnings management at rates exceeding the specified test-levels when applied to samples of firms with extreme financial performance. Additionally, they show that the Modified Jones Model provides the most powerful test of earnings management.

Guay et al. (1996) argue that comparisons of discretionary-accruals models in Dechow et al. (1995) critically hinge on such important (implicit) assumptions as the behavior of earnings absent discretion and how management exercises discretion over accruals conditional on nondiscretionary earnings. Evaluations of time-series discretionary-accruals models using stock returns depend, additionally, on assumptions about the relation between accounting numbers and stock prices (e.g., market efficiency with respect to earnings information, and stock prices lead earnings). Guay et al. also show that attempts to increase statistical power by using non-random samples (e.g., firms with extreme financial performance as in Dechow et al. (1995)) cloud the findings, as they increase the likelihood that correlated omitted variables confound the results. Their findings thus cast doubts on the ability of various time-series models to separate accruals into discretionary and nondiscretionary components. Healy (1996), however, points out that the study of Guay et al. relies on strong assumptions such as strong-form stock market efficiency, and that its tests examine the aggregate relation between stock returns, discretionary accruals, and nondiscretionary earnings, rather than relations for a specific sample where earnings management is expected. Thus, whether these discretionary-accruals models are able to separate accruals into discretionary and nondiscretionary components and thereby detect earnings management is still an open empirical question.

The primary goal of this study is to evaluate empirically the ability of the cross-sectional version of two discretionary-accruals model, the Cross-sectional Jones Model and the Cross-sectional Modified Jones Model, to detect earnings management vis-à-vis their time series counterparts. We are motivated to undertake this evaluation for two reasons. First, the two cross-sectional models have not been evaluated by prior research. Second, each type relies on a different set of assumptions and it is an empirical question which set is more descriptively valid. For completeness, we utilize our research method to also evaluate three other models used by prior studies: the Industry Model, the DeAngelo Model, and the Healy Model.

Prior earnings management studies (see, e.g., Healy 1985; DeAngelo, 1986; Jones, 1991) have found that high discretionary accruals indicate earnings manipulations. Thus, high discretionary accruals should be associated with audit qualifications. However, there are other factors that can lead to audit qualifications. Prior research has identified a number of such variables covering operational complexity and various types of risks. We incorporate the implications of these studies’ findings into our research design in two ways. First we use a matched-pair design in which we evaluate the ability of each model to distinguish between 173 distinct firms with qualified audit opinions and 173 matched-pair firms with clean reports. Second, we perform sensitivity tests to evaluate the effects of book-to-market ratios, leverage, profitability, firm size, total accruals, and mergers and acquisitions on our findings.

A distinguishing feature of our research method is our simultaneous effort to maximize power (by selecting a sample where earnings management is expected) while minimizing potential biases arising from using a non-random sample that may lead to erroneous inferences (by adding controls for confounding variables). For example, Dechow et al. (1995, pp. 208–209) report that for firms experiencing extreme financial performance, the discretionary-accruals models they evaluate are unable to completely extract the low (high) non-discretionary accruals associated with the low (high) earnings performance. We thus evaluate the association between discretionary accruals and audit qualifications after controlling for earnings performance and total accruals.

Contingency-table tests for the association between high discretionary accruals and audit qualifications show significant results for the Modified Jones Models, and the two cross-sectional models. The contingency table results for the other four models are not significant. Univariate logistic-regression tests show a significant relation between discretionary accruals and the likelihood of receiving qualified reports for all models, except the DeAngelo Model. Thus, our univariate tests’ results show that the Cross-sectional Jones Model and the Cross-sectional Modified-Jones Model, not evaluated by prior research, are able to detect earnings management. Additionally, like Dechow et al. (1995), using univariate logistic regressions that do not control for potential confounding variables, we provide evidence suggesting that the Jones Model, the Modified Jones, the Healy Model, and the Industry Model are able to detect earnings management. However, with respect to the DeAngelo Model, their findings differ from ours. While they conclude that this model is also successful in detecting earnings management, our findings do not support the ability of the DeAngelo Model to detect earnings management.

While our matched-pair design alleviates concerns regarding the role of potential confounding effects, it does not eliminate them entirely as the control firms differ from the test firms with respect to certain firm characteristics. In an effort to assess the effect of potential confounding variables on our findings, we perform a number of sensitivity tests. One sensitivity test concerns replicating the logistic regression tests after including explanatory variables shown by prior research to have power in explaining audit qualifications: earnings performance, firm-size, debt-to-equity ratios, and book-to-market ratios. A second sensitivity analysis involves replicating the regression tests after matching the test and control samples on total accruals. A third type evaluates the effects of mergers and acquisitions on our findings due to concerns in Collins and Hribar (1999) that the balance sheet approach used by prior research and this study to estimate total accruals can lead to serious errors for firms with mergers or acquisitions. The results show that only the two cross-sectional models survive all sensitivity tests.

This study finds that the Cross-sectional Jones Model and the Cross-sectional Modified Jones Model, not evaluated by prior research, perform better than their time-series counterparts in detecting earnings management at least among firms with extreme earnings management (i.e. those with qualified audit reports). This result is important for future earnings management research particularly because using a cross-sectional model, rather than its time-series counterpart, should result in a larger sample size that is less subject to a survivorship bias arising from requiring long time-series data (Subramanyam, 1996, p. 254). The reliance on time-series models also limits their usefulness. For example, only the cross-sectional models allow investigations of firms with a short history, e.g., new startups engaging in initial public offerings.

However, our results must be interpreted with caution for two reasons. First, as is the case with any study evaluating empirically the relative performance of existing models, our findings are relative, not absolute. That is, they merely indicate the superiority of the cross-sectional models vis-à-vis their time-series counterparts, not validate either the former or the latter. Thus, it is left for future research to develop a better model of non-discretionary accruals. Still, our results may be helpful for this research in both the model-development phase and the model-testing phase. Given our findings, in the model-development phase future research should consider time-series models as well as their cross-sectional counterparts. In the model-testing phase, audit qualifications may be used to indicate accruals management. Second, since our tests evaluate the ability of discretionary-accruals models to identify firms engaging in an extreme form of earnings management, our findings may not be generalized to firms with moderate levels of earnings management, e.g., firms engaging in earnings management within generally accepted accounting principles (GAAP).

The next section describes the seven competing discretionary-accruals models we evaluate and outlines the theoretical background underlying our investigation. Section 3 reports the sample selection procedure and describes the data. Section 4 outlines the tests and discusses the results, and the final section concludes the study.