Discretionary disclosure and the market reaction to restatements

For example, the market reacts more negatively when: core earnings are restated, a greater number of accounts are restated, and the restatement is fraud-related (Palmrose et al. 2004; GAO 2002).

In our sample, 86 % of restatements decrease past reported earnings. In Sect. 5.6.1 we separately examine downward versus upward restatements.

However, as PRS note, the restatements initiated by management are more frequently fraud-related and income-reducing than their overall sample and thus the relative severity of the restatements may outweigh any benefits otherwise accruing to management’s self-reporting.

Concurrent research by Myers et al. (2009) examines a similar issue of where a restatement is disclosed focusing on the transparency of 8-K filings. They find that the initial market reaction is smaller when restatements not made initially in an 8-K filing, consistent with our finding that the market reaction to restatements announced in an earnings announcement (rather than a standalone announcement) is less negative.

Anderson and Yohn’s (2002) measure of specificity is a dichotomous variable taking the value of one if the restatement effect is quantified and zero otherwise. As discussed in Sect. 3.3, our measure of the level of specificity is a three-state discrete variable based on whether detailed and complete information about the effects of restatement on the financial statements is presented at the time of the announcement.

We also estimate the model including industry dummy control variables and our results (untabulated) remain the same.

We use the positive and negative wordlist specific to the domain of financial disclosure because this list has been shown to result in a measure of the tone of earnings announcements that has greater predictive validity than alternative wordlists, particularly in smaller sample sizes (Henry and Leone 2009).

In examining tone of earnings announcements prior to restatement, a possible concern is that tone may be a proxy for past earnings. Including market returns in the 120 days prior to the restatement in our regression controls for the market's evaluation of the firm's performance and expectations. In Sect. 5.5.2, we also present additional analyses of Prior Tone and firm's past performance.

The prior market reaction has the potential to also capture some element of prior disclosure. When we exclude it from the primary regressions, we find similar results.

Even though the GAO database states that it excludes technical restatements in its restatement list, we identified a total of 19 restatements related to technical changes, which is about 5 % of our restatement sample. These technical changes are related to EITF 00-10, EITF 00-14, EITF 00-22, EITF 00-25, SFAS 132, SFAS 133, and SFAS 142 (similar to Baber et al. 2009). Our results do not change when we exclude these 19 restatements from the sample.

We exclude financial firms from our sample for two main reasons: (1) differences in disclosure environment because of banks' regulatory disclosure requirements (Johnson and Schwartz 2005) and (2) differences in market reaction to earnings announcements due to industry structure (Biddle and Seow 1991).

The GAO (2002) database does not cover the full years of 1996 and 2002. Thus, the numbers of restatements for 1996 and 2002 are comparable to the other years.

Additional analysis (untabulated) suggests that the impact of prior disclosure may differ for upward versus downward restatements, with Amount having more market impact for upward restatements and Tone having a greater impact for downward restatements. We conjecture that the greater impact of Tone for negative restatements results from its increasing the surprise of negative restatements; we leave further investigation of this topic for future research.

We perform additional analyses, in which we classify the content of officer quotes into positive or negative. While restatement related quotes do not contain unambiguously positive quotes, we segregate non-restatement related quotes into positive and negative quotes, using the approach in Francis et al. (2002). We find that 86 % of the non-restatement quotes are positive. Our regression analysis shows that positive non-restatement quotes are positively and significantly related to the market reaction, but negative non-restatement related quotes are not.

We also considered the classification approach in Bowen et al. (2005) for those restatements announced in an earnings press release. Specifically, we explored measuring emphasis by whether the mention occurs first in the headline, the first/second paragraph, paragraph three or later, or in the financial statement only. We develop three indicator variables to represent these four locations. However, untabulated results do not reveal any significance on the three indicator variables.

For clarification, we identify the regressions models estimated using the composite variables by adding a “C” after the model number.

Additionally, we further probe the interaction between prior and concurrent disclosure by interacting composite prior disclosure with individual restatement announcement variables. Results (untabulated) indicate market reaction is more positively associated with restatement quotes and headlined restatements for higher prior disclosure, but more negatively associated with restatement detail. Specifically, we find: (1) the interaction of prior disclosure with Rest Quote is significantly positive with a p value of 0.07 (one-sided), implying that restatement quotes made by higher disclosures are more positively viewed; (2) the interactive term with Nonrest Quote is not significant; (3) the interactive term with Headline is significantly positive with a p value of 0.01 (one-sided), suggesting that emphasis on the restatement by higher disclosures is more positively viewed, and (4) the interactive term with Rest Detail is significantly negative with a p value of 0.00 (one-sided) implying that additional restatement details are less important when prior disclosure is high.

In addition to using the composite measures, we also estimate the four regressions models including each prior disclosure variable, each concurrent disclosure variables, and both groups of variables. The findings and implications are similar to those using the composite measures.

Lexis-Nexis covers more than 350 newspapers, more than 400 magazines and journals, over 600 newsletters, and wire services (including the Associated Press, Business Wire and PR Newswire).

The change in earnings is calculated as quarterly earnings per share for the quarter announced minus the earnings per share for the same quarter in the prior year scaled by the average of sales for the two quarters.

The separation of restatements into downward and upward ones has been shown to be meaningfully related to outside director turnover. For instance, Srinivasan (2005) examines the consequences of a restatement for outside directors finding greater turnover in companies with downward restatements than upward restatements. He also finds greater turnover in companies with upward restatements than those with technical restatements. As mentioned in footnote 10, we find 19 technical restatements in our sample.

Classification of the 49 no-effect restatements as upward versus downward does not change our results.

Among the 256 common observations, 39 restatements are identified as frauds in our sample and irregularities in Hennes et al. (2008) sample; 183 restatements are identified as non-frauds in our sample and non-irregularities in Hennes’ sample; 32 restatements are non-frauds but irregularities; 2 restatements are fraud but non-irregularity.

We also examine three different definitions of high institutional holding based on whether a company falls above the median in each of the categories yielding no additional insights: (1) percentage of institutional ownership of the company, (2) number of institutional investors the company had in the restatement quarter and (3) number of long-term institutional investors.