Measuring the market response to going concern modifications: the importance of disclosure timing

Furthermore, reporting rules require management to disclose going concern issues when they exist. Guidance from the PCAOB (in AU 341 (AS 2415) paragraphs 10 and 14) and from the U.S. Securities and Exchange Commission (SEC) (Division of Corporate Finance Financial Reporting Manual 4230.1b) is explicit that financial statements include “appropriate and prominent disclosure of the financial difficulties giving rise to that uncertainty.” Thus, when an auditor issues a GCM, management must address this uncertainty in the financial statements. To the extent that these disclosures represent new information, they are likely to induce a negative market reaction at the GCM disclosure date.

Note that for our sample, any market reaction at the EA date is unlikely to be attributable solely to earnings news because EAs for troubled companies typically disclose adverse events such as debt covenant violations or goodwill impairments. Because our objective is to differentiate between the market reaction to GCMs and to other information disclosed in EAs (earnings news or otherwise), it is not important which EA information investors are reacting to; it is only important that EA-related information be considered when evaluating the market response to GCMs.

We identified whether the company received a prior GCM using data from Audit Analytics and by searching for the term “going concern” in all prior year filings. We exclude any company that filed for bankruptcy prior to the disclosure of the GCM. Menon and Williams (2010) include only those companies for which there is no record of a prior GCM in any observable year (i.e., “first-time” GCMs). In contrast, we include all “new” GCMs because the history of audit reports is limited and because this allows for a larger sample and more powerful tests. Nonetheless, following Menon and Williams (2010), we restrict our sample to companies with “first-time” GCMs and all of our inferences are unaffected. We tabulate our primary findings for this restricted sample in Appendix 2 Table 9.

We do not attempt to identify a control sample of non-GCM companies because we argue that reasonable counterfactuals (i.e., companies that should have received GCMs but did not) do not exist or are unidentifiable. Because many factors contributing to GCMs are difficult to identify or measure (e.g., violation of debt contracts, supply chain issues, regulatory problems, management’s remediation plans, ability to secure financing), any findings would be endogenous since these unmodeled factors would contribute to both the decision to issue a GCM and the magnitude of any market reaction. To illustrate, if we use propensity score matching to match GCM companies to non-GCM companies on profitability, leverage, and distress measures, we find that the GCM sample exhibits bankruptcy rates of approximately 9.0% while the counterfactual matched sample exhibits bankruptcy rates of approximately 0.5%. This suggests that common observable factors do not adequately capture the level of distress. Because identifying a suitable control sample of non-GCM companies is infeasible, we instead leverage variation in the timing of disclosures to compare the market reaction around EAs with and without concurrent GCMs. That is, our counterfactuals also receive GCMs, but differ from our treatment sample with respect to the timing of the GCM disclosure relative to the EA. Supporting the use of these counterfactuals, we do not find a significant difference in earnings surprise (Table 1) or in subsequent bankruptcy rates between the treatment sample and this control sample.

We winsorize SURPRISE at −1 and + 1 to reduce the effect of outliers, but our inferences are unchanged if we do not winsorize SURPRISE. Furthermore, our inferences are unchanged if we calculate SURPRISE scaling by prior quarter ending stock price rather than by the median consensus analyst forecast (see online Appendix O-2).

The returns for “EAs with GCMs” differ slightly from the returns for “GCMs with EAs” because there are 56 instances where the EA date precedes the GCM date but the GCM is disclosed in the EA return window (i.e., day +1 or + 2). If we exclude these 56 observations from our tests, our inferences are unchanged (see online Appendix O-4).

To address potential issues related to non-linearities in the market response to our earnings surprise variables (Freeman and Tse 1992), we add higher order (squared and cubed) terms of our earnings surprise variables to all regressions and use non-parametric (decile indicator) specifications; the inferences are similar using either of these alternative specifications (see online Appendix O-5).

We standardize EA_DELAY so that the uninteracted base variables can be interpreted as the relation at the mean EA_DELAY.

We also note the significant, negative intercept (in columns (1) through (3)), which represents the estimated market reaction for sample observations when all variables in the model take a value of zero; this suggests that EAs for companies receiving first-time GCMs contain confounding negative disclosures beyond the negative information in earnings announcements and management forecasts.

As noted in Figure 2, the vast majority (90.7%) of EAs that include the GCM (GCMwEA = 1) occur concurrently with the filing of the annual report. We also perform tests from Table 4 with a subsample where the EA date is not the annual report date (see online Appendix O-6). Although the estimated coefficient on GCMwEA is negative and significant for this limited subsample, we cannot draw strong inferences because this subsample represents only a small fraction (9.3%) of EAs that include GCMs and the finding is not robust in tests that use the analyst-based earnings surprise control variable. Furthermore, for the subsample of EAs that do not include GCMs, we do not observe a significant market reaction when the GCM is eventually disclosed (see Table 2); this suggests that the difference in returns at the EA date for this subsample is likely attributable to something other than the GCM disclosure.

Following Landsman et al. (2012), we define abnormal trading volume as the natural log of actual trading volume scaled by expected trading volume. We measure actual trading volume (the numerator) as the mean daily trading volume in the event window (days [0, +2]), where daily trading volume is calculated as the number of shares traded scaled by the number of shares outstanding, and we use two different event windows to estimate the expected trading volume (the denominator) (i.e., [−60, −10] and [−224, −75] from the EA date). For the shorter (longer) estimation window, we require observations to have at least 20 (50) days of trading information available from CRSP. We assume that when trading days are available in CRSP but are missing volume data, the trading volume is zero, and we treat trading days that are unavailable in CRSP as missing. We winsorize AVOL at the 1st and 99th percentiles to reduce the effects of outliers, but our inferences are not sensitive to this design choice.

Four (4) of the 7 significant responses occur in the [0, +3] window. Given that all companies in our sample experience substantial distress, the longer [0, +3] window increases the likelihood that the market reaction is related to other news. However, in all cases, we observe no significant response using a more traditional [−1, +1] window (see Online Appendix O-10).