Filing speed, information leakage, and price formation

We use the terms “reporting lag” and “filing lag” interchangeably.

A concurrent study by Ben-Rephael et al. (2017) examines institutional and retail investor attention around 8 K event and filing dates. They measure institutional (retail) attention by the number of searches on Bloomberg terminals (google). They find heightened attention by institutional investors around the event and filing dates, and no change in attention by retail investors. They then provide evidence that price discovery increases with attention. Our paper differs from Ben-Rephael et al. (2017) on the following dimensions: (1) They do not test the same hypotheses as this study and, in particular, the implications of the “buy on the rumor, sell on the news” strategy of Hirshleifer et al. (1994) and Brunnermeier (2005) on price formation. (2) Because of data limitations, they do not provide evidence related to the extent, magnitude, and economic benefits associated with strategic trading. (3) Our research methodology allows us to provide direct evidence on price discovery on each day during the event window. (4) Our sample is much larger, covering all 8 K items whereas their analysis is restricted to five news items only. Excluding all filings with a filing gap of less than two days, their sample comprises around 17,000 8 K filings, whereas our sample comprises over 95,000 filings.

Lerman and Livnat (2010) show that 8Ks often trigger market reactions in the form of abnormal equity returns and trading volumes. Segal and Segal (2016) show that managers disclose negative non-earnings information from the 8 K report strategically when investor attention is low. Rubin et al. (2017) examine analysts’ reaction to 8 K information. Bird and Karolyi (2016) examine the effect of institutional ownership on the quality of the 8 K disclosure.

With the exception of auditor changes (resignation, firing, hiring) and director resignations that had to be reported within five business days

While clearly some of the events are predictable, the overall evidence in the literature suggests that 8 K forms are by and large unpredictable, as their filing results in significant abnormal returns and analysts’ revisions.

Indeed, Lerman and Livnat (2010) provide evidence of heightened trading around both the 8 K event and the filing date.

Abel Noser stopped reporting unique institutional identifiers after 2011 to further protect their clients’ privacy. Hence, it is impossible to track an institution’s trades across stocks and time after 2011. Also, note that the 8 K sample period is longer than the institutional data. Therefore, we elect to use the expanded 8 K sample whenever institutional data are not required, in order to increase the power of the tests.

Hu et al. (2018) provide a comprehensive review of the data and note that the data have been widely used in the accounting and finance literatures. Their analysis indicates that that the database contains 12 years of data covering 233 million transactions with $37 trillion traded. They find that institutional trade sizes decline dramatically over time, rendering institutional trade size–based inferences problematic, and that the data cover 12% of CRSP volume.

The data include separate codes for the fund family (e.g., Fidelity) and the specific fund within the fund family (e.g., Fidelity’s Magellan Fund). We examine the trading around 8 K filings at the specific fund level.

Although companies indicate the “event date” on the 8 K form, it is possible that the event occurs after trading hours. Hence, if the company files the 8 K report on the day after the event date, we cannot determine whether the informed trader was aware of the news as early as on the event date.

The event date is typically the date on which the information first becomes known. However, in many cases, information about the impending event can leak out. For example, consider director resignations, where the event date is the date on which the director resigns. Information about the director’s intention to resign is likely to leak out prior to the actual resignation date as the director contemplates whether and when to resign.

Total volume transacted during the day is an alternative potential deflator. However, results are quite sensitive to scaling by total volume. The reason is that total volume, which includes institutional investor volume not accounted for in Ancerno, increases considerably during the event and filing windows—on average, we observe an increase of roughly 22% in total volume. Consequently, when scaling (the increasing) daily trading volume of institutional traders by (the increasing) total volume, we do not observe any change during the aforementioned periods.

We define non-news days as the days on which the company did not issue preliminary earnings or 10-Q or 10-K reports. In other words, we exclude, from the sample, the days on which the company reported preliminary earnings and the 10-Q and 10-K reports.

The differences between the coefficients on the days within the event window are highly significant (p value<0.01). Also, untabulated tests show a spike in the coefficient on the first filing day, relative to the last day of the event window.

We use a three day window to allow for the possibility that 8 K forms are reported on the last trading day of the week, and to give institutional investors sufficient time to reverse their trading position

The ostensibly small number of institutions that engage in strategic trading is explained by the observation that for 75% of the 8 K forms, there is no strategic trading. There are two reasons for this: First, not all 8Ks contain significant economic news. Second, institutional investors do not always reverse their position (i.e., they trade according to the sign of the news but do not reverse their position post filing); hence, they are not captured by our sample of “strategic traders.” To further ascertain the magnitude of strategic trading, we exclude 8 K reports that did not yield strategic trading. We find that the mean number of institutional investors that engage in strategic trading is 7.4.

With the exception of the Average Filing Volume for five- to seven-day filing lags, which is significant at the 5% level, all figures in the table are significant at the 1% level. Further analysis indicates that the difference between the long (five- to seven-day) and short (two- to four-day) filing lags is also significant.

We elected to focus on the larger sample and not control for news reports in most of our analyses for two reasons. First, given the large volume of data, we cannot determine whether the media content is indeed related to the event itself. Second, Raven Pack data cover less than 50% of our sample firms (6193 of 12,601). Nevertheless, our results hold for the smaller sample as well.

Note that the average filing lag in the pre period of 2001–2003 (3.3 days) is significantly greater (p value<0.01) than the average filing lag in the post period of 2005–2007 (2.85 days).

Note that since both the numerator and denominator are scaled by the same constant (number of shares outstanding in the case of the volume ratio, and average daily volatility in the case of the volatility ratio), the volume ratio is equivalent to the ratio of total volume during the filing window scaled by total volume from the event date through one day after the filing date. Similarly, the volatility ratio is equivalent to the ratio of total equity volatility during the filing window scaled by equity volatility from the event date through one day after the filing date.

Untabulated results indicate that the rate of the incorporation of the information related to the event is independent of the reporting lag. This suggests that while institutions may learn about the filing lag, they do not seem to have superior information regarding the firm’s strategy as to when to file; hence, they trade immediately as they learn about the event.

The Ancerno data are based on the client-manager level, where the manager is typically an investment house. A client can have investments with different managers. Because most investment houses have funds with different investment characteristics (e.g., indexers, hedge funds, mutual funds) and because the data fail to include a description of fund type, we treat each client-manager as an independent observation, although it is likely that the same fund is represented more than once in the data (with different clients). To alleviate this concern, we examine the results at the investment house level, but the downside here is that we aggregate funds with different objectives.

We obtain similar results when we examine retail trades from 2004 to 2013, where 2013 is the end year of the 8 K sample.

Collin-Dufresne and Fos (2015) explain their findings by showing that informed traders trade during times of high liquidity and use limit orders.

Interestingly, the coefficient on the negative news indicator is positive but not significant post 2004, indicating that negative news is not reported later than positive news. However, the negative aspect of the news is captured by CMAR, and its coefficient is negative and significant at the 1% level. Furthermore, excluding CMAR and its interaction with the negative news indicator from the regression yields a positive and highly significant coefficient (p value<0.01) for the negative news indicator.

In a sensitivity analysis we measure ROA as operating income scaled by total assets. Results are very similar.