Discussion of Charitou, Lambertides and Trigeorgis

Charitou et al. (2007; hereafter CLT), examine the behaviour of accruals for a sample of 859 U.S. firms that filed for bankruptcy over the period 1986–2004. They show that earnings and cash flow from operations drift downwards in the four years prior to bankruptcy, although cash flows from operations drift upwards in period 0 (year of bankruptcy). Total accruals are negative and decreasing over the four-year bankruptcy period. Performance matched current abnormal accruals fluctuate over the period, but are mostly negative (i.e., income decreasing). The performance matched long-term discretionary abnormal accruals decline each year. CLT examine if management turnover, qualified audit opinions and institutional ownership impact the ‘discretionary’ accruals over the four years prior to bankruptcy. They conclude that earnings decreases may be partly due to new managers’ taking earnings baths and that qualified audit opinion may induce managers to employ more conservative financial reporting choices. They also conclude that distressed firms with lower levels of institutional ownership have a greater tendency to manage earnings. Finally CLT undertake market-based tests and conclude that institutional shareholders mitigate the negative abnormal returns. At the Forum, the discussion concentrated on some of the fundamental problems associated with research in the areas of financial distress and accruals management. The particular issues of concern were: what constitutes financial distress?; the reliability of accrual models; and the choice of explanatory variables. After outlining these issues I will then comment on the interpretation of the results. One of the fundamental problems of financial distress research is the lack of underlying theory. This leaves the researcher with the problem of how to define ‘financial distress’ and what variables to put in the model to explain financial distress. CLT describe the prior literature as providing ‘mixed results with regard to earnings behaviour of distressed firms’. This is not surprising, given the multitude of definitions used for ‘financial distress’ in prior literature. Table 1 shows that prior research has used various states of financial distress, such as ‘bankruptcy’, dividend reductions, continuing losses, debt covenant violations (both servicing and technical violations), going concern audit opinions and the change of management. However, it is not clear if there is a hierarchal or time sequence or random relation between these events. CLT use bankruptcy because it is a ‘severe’ or ‘highly distressed’ measure of financial distress. While this is a reasonable approach, the years before bankruptcy are likely to be confounded with other events of ‘financial distress’ such as debt default, management buyouts, takeovers, adverse audit opinions. In studying financial distress the researcher must try to control for events, other than the primary event of interest (e.g., bankruptcy), that might also give rise to financial distress. Another approach would be to further classify the bankrupt firms into various states of distress (e.g., McKeown et al., 1991; Roser, 2003). Another issue with ‘bankruptcy’ is that it is a legal condition as well as a financial one. Hence, an entity that is bankrupt in the U.S. might not be bankrupt in another legal jurisdiction. Furthermore, other jurisdictions have different legal definitions of financial distress (e.g., ‘receivership’). Hence the external validity of CLT is likely to be limited. Following Teoh et al. (1998), CLT decompose accruals into four categories—current, non-current, discretionary and non-discretionary—and use a cross-sectional adaptation of the modified Jones (1991) model. They also provide results using performance matched accruals (Kothari et al., 2005). A key methodological issue in the paper is how well these models are able to partition accruals into normal and abnormal components. There is increasing evidence that these models provide very noisy estimates of ‘abnormal’ accruals.1 The simulation results in Dechow et al. (1995) show that amounts of earnings management under 5 per cent of total assets are likely to go undetected. Thomas and Zhang (2000) show that most models will not outdo a naive model where: total accruals =−5% (last period's total assets). More recently, Jones et al. (2006) find that popular discretionary accrual models fail to capture a substantial amount of extreme earnings management. They find that only models by Kothari et al. (2005), Dechow and Dichev (2002) and McNichols (2002) are able to detect fraudulent earnings beyond a simple measure of total accruals. Existing accrual models make very noisy estimates of accruals, especially for firms with extreme performance. That is, abnormal accruals are more negative (positive) for firms with extreme poor (good) performance. CLT's sample comprises firms that are likely to contain a larger proportion of firms with unusually negative accruals due to reasons other than managers’ accrual discretion. These include asymmetrical accounting practice (e.g., such as the lower of cost of market basis for inventory measurement) and managers’ real decisions (e.g., lowering inventory levels to conserve cash). It is difficult to test hypotheses that predict an increase in negative (income decreasing) accruals, when the experimental bias in the accrual model is in the same direction. The forum noted that accruals are not the only accounting measure that can be used to investigate potential earnings management. Another approach would be to examine the portfolio of accounting policy choices, voluntary changes in accounting policies, and the early adoption of mandatory changes (see Sweeney, 1994). CLT ignore the impact of any accounting policy changes over the period 1986–2004 on their accrual measures, including the effect of mandatory accounting changes. In research of this type the researcher must select the variables of interest they think will be related to abnormal accruals. CLT examine the association between audit qualification, management change and institutional ownership. An alternative approach would be to examine a single explanation in more detail. For example, in addition to further analysing the audit opinion explanation beyond a simply modified/unmodified indicator variable, we might concentrate on the role of auditing and accruals in the years before bankruptcy. Such an analysis would include variables related to audit quality (e.g., Big 4 auditors, industry specialists and ‘superior quality’ audit committees) and the provision of additional analysis on the type of audit qualification. Butler et al. (2004) examine the types of modified opinions that are associated with large accruals. They find that it is the firms with a going concern opinion that drive opinion/accrual relation. They also conclude that there is no evidence that firms receiving modified audit opinions manage earnings more than those receiving clean audit opinions. In contrast, CLT conclude that their results ‘are consistent with the increased conservatism hypothesis induced by audit opinion’. We can only speculate how robust the results would be if an in-depth analysis of the role of the audit function were undertaken. While the choice of institutional shareholding as an explanatory variable is novel, it does not have a strong conceptual basis. In my opinion, there is large gap between the level of institutional ownership and the manipulation of accruals. Furthermore, the arguments provided by CLT apply to all investors (e.g., ‘earnings behaviour may adversely affect share prices’) not just institutional investors. The link would be strengthened if the institutional investors were represented by a director, especially on the audit committee. CLT motivate the use of institutional ownership by invoking the relation between institutional investors and ‘investor protection’. However, it is difficult to follow the implications of this relation for this study. Prior research has examined investor protection across jurisdictions. CLT is a study within one jurisdiction, so it is not clear that there is any variation in ‘investor protection’. There are several features that cause me to doubt the interpretation of the results in CLT. My concerns relate to the terminology used, the discussion, and the sample. One of the unfortunate features of the ‘earnings management’ literature, in my opinion, is that the papers move too quickly from the label ‘abnormal accruals’ to infer ‘discretionary accruals’ and then to infer ‘earnings management’. CLT is no exception. My preference is to use the ‘abnormal’ label while describing the results of the study and leave the terms ‘discretionary’ and ‘earnings management’ until the discussion of the results. The discussion should also include other possible interpretations, such as poor modelling of ‘normal’ accruals; the impact of events, conditions and firm performance on ‘abnormal accruals’; and the (in)ability to distinguish between discretionary and nondiscretionary abnormal accruals. The results in the paper are not strong. For example, in Table 4 the ‘performance matched discretionary accruals’ for both current and non-current accruals are only significant in year −1. In Table 9 none of the univariate results are significant and only one of the multivariate regressions has a significant F test at the 0.01 level. The reader needs to be careful because statements are made about results that are not significant. Furthermore, in many instances the discussion of the results is selective. For example, in discussing Table 2, CLT claim that the ‘median net income (and cash flow) measure is lowest one year prior to bankruptcy filing’. This is true only for cash flows but not net income. Given the large positive cash flow in year 0 (Table 2), I find it difficult to agree that ‘cash flows appear more stable’. In Table 5, Panels B1 and B2, the results are similar. They are significant only for periods −1 and 0 and not for periods −3 and −2. This suggests that the results, while possibly dependent on management change, are independent of when management changes (otherwise the results in Panel B would be a period earlier). Similar comments could be made about Table 6. I do not agree that the results ‘suggest that management changes may affect subsequent earnings choices’ or ‘are consistent with the increased conservatism hypothesis induced by audit opinions’. There is an association but there is nothing to suggest a causal relation. The number of observations differs in each panel of each table. My concern is the variation in accruals from year to year before bankruptcy might arise from different samples, as much as from the hypothesized cause. Another concern I have is that the institutional ownership variable is measured at period −2, rather than being allowed to vary in each period. If a sample self-selects (e.g., bankrupt firms), the possibility of omitted correlated variables is a major concern when designing the study and interpreting the results. One of the main research design features to mitigate this issue is the use of a ‘healthy’ or performance matched sample. CLT match each bankrupt observation with a firm in the same industry, same calendar year, with a similar return on assets in year −3. However, Table 2 of the paper reveals that in period −3 the performance adjusted cash flows and net income are non-zero in period −3. That is, despite the best efforts of the authors, the matching process is not effective. Furthermore, it is not clear in all cases why the matched pair test is the most interesting comparison in a time-series study. For example, the main conclusion from Tables 5 and 6 is that management turnover and modified audit opinion are (significantly) greater for distressed firms than for ‘healthy’ firms. We also know from Table 4 that bankrupt firms have increasingly negative earnings. Therefore another approach would be to analyse the marginal impact on negative accruals of management turnover and modified opinions for the sample of bankrupt firms only. On the other hand, the institutional investor analysis seems to assess the marginal contribution. For example, ‘institutional investors of distressed firms have incentives to mitigate earnings-decreasing (accruals) behaviour whether due to management turnover or conservatism imposed by auditors’. Also Table 9 assesses the marginal contribution of INSOWN. While Table 8 assesses the marginal contribution of QAUD and MNGCH on INSOWN, the story is built around the marginal contribution of INSOWN on QUAD and MNGCH. Trying to uncover an empirical regularity between accruals and financial distress is an interesting endeavor. In this type of research there are a number of methodological choices, including the choice of the primary financial distress event and the need to control for other potential events that might impact accruals. Given our state of knowledge of the effect of auditing or management change on bankruptcy, it is not clear that we can use simple dummy variables to control for these events, while looking for more exotic effects, such as institutional ownership. A fundamental research issue is the partitioning of accruals into discretionary and nondiscretionary components. This is fraught with difficulties. Untangling the effects of earnings manipulation, real decisions, and abnormal non-discretionary accrual on total accruals is a major barrier to interpreting the results. Interpretation is even more difficult when extreme negative performance creates extreme negative abnormal accruals and the events of interest also hypothesize negative abnormal accruals.