How Important are Earnings Announcements as an Information Source?

Abstract

In a competitive information market, a single information source can only dominate other sources individually, not collectively. We explore whether earnings announcements constitute such a dominant source using Ball and Shivakumar's (2008) [How much new information is there in earnings?, Journal of Accounting Research, 2008, 46(5), pp. 975–1016] R ² metric: the proportion of the variation in annual returns explained by the four quarterly earnings announcement returns. We find that the earnings announcement days' R ² is 11% – higher than the corresponding R ² of days with dividend announcements, management forecasts, preannouncements, and 10-K and 10-Q filings and their amendments, and comparable to that of the four days with the largest realised absolute returns in a year. Additional analysis reveals that earnings announcements convey extreme bad news as often as management forecasts and preannouncements; for any other type of news, earnings announcements are much more frequent. We conclude that earnings announcements are an important source of new information in the equity market.

Acknowledgements

We thank Bill Cready, Salma Ibrahim, Todd Kravet, Lakshmanan Shivakumar, Jerry Zimmermann, and workshop participants at the University of Alberta, City University of Hong Kong, Florida State University, Keio Business School, London Business School, Nanjing University, National Taiwan University, Temple University, the 2010 AAA Mid-Atlantic Regional meeting, and the 2010 AAA Annual meeting for helpful comments and suggestions.

Notes

We partition company-issued guidelines from First Call based on whether they occur before or after the end of the fiscal quarter to which the guidance pertains; we call the former management forecasts and the latter preannouncements (Baginski et al., 1994; Soffer et al., 2000). 10-K and 10-Q amendments include earnings restatements but also include other minor corrections.

Beyer et al. (2010) reach a conclusion similar to Ball and Shivakumar (2008) using a parallel research design. However, Beyer et al. (2010) analyse only 70,700 firm-quarter observations for 2747 firms over the period from 1994 to 2007 while we analyse 276,180 firm-quarters for 11,823 firms over the same period, i.e. four times as many. Because Beyer et al. (2010) study only firms followed by financial analysts, their inferences likely apply primarily to large firms (e.g. Bhushan, 1989).

Managers forecast and preannounce earnings primarily when their information differs sufficiently from the market's expectation, whereas they announce earnings regularly regardless of market expectations (e.g. Ball and Shivakumar, 2008). Thus, managers forecast and preannounce earnings less frequently than they announce earnings, but this raises the possibility that extreme good news and bad news preannouncements and management forecasts occur more often than similarly extreme good news and bad news earnings announcements.

Intuitively, one can view management forecasts as conveying only timely information and earnings announcements as conveying sometimes timely information and sometimes stale information. Our findings suggest that the number of earnings announcements conveying timely information is at least as high as the number of management forecasts conveying timely information.

Our evidence is consistent with Francis et al.'s (2002) evidence that the majority of absolute earnings announcement returns fall in the top two deciles of daily absolute returns. We extend Francis et al. (2002) by using Ball and Shivakumar's (2008) approach to measure the proportions of the variation in annual returns explained by earnings announcement returns and the four most extreme returns in a year.

The pricing effect of earnings announcements may be overstated under this approach if earnings releases are associated with increased production of information by market participants or increased dissemination of other financial or non-financial information by firms.

EAD returns and non-EAD returns could be negatively correlated if investors overreact to earnings announcements (e.g. DeBondt and Thaler, 1985) or if transaction costs such as bid-ask spreads constrain daily price movements (e.g. French and Roll, 1986).

In passing, we note that variance-based approaches in the tradition of Beaver (1968) do not require assumptions about how the market forms expectations about earnings – Earnings Response Coefficient-based approaches, however, require such assumptions.

Ball and Shivakumar (2008) report that their results are robust to (a) return computation method (adding vs. multiplying), (b) outlier exclusion, (c) December fiscal year-end firms only, (d) annual return sign, (e) concurrent management earnings forecasts, (f) controls for size, market-to-book, leverage, and industry, (g) different announcement return windows, and (h) trading volume instead of returns.

However, Abarbanell and Kim (2011) find that returns' ability to predict future earnings is concentrated disproportionately on EADs.

Bathke and Lorek (1984, p. 175) estimate that unexpected quarterly earnings account for only 15% of the information that reaches the market on EADs and conclude ‘the results also underscore a lack of monopolistic control by the accounting profession over the dissemination of firm-specific financial information’.

Examples include quality of financial reporting (Lev, 1989), information production in the preannouncement period by investors (Atiase, 1985), market efficiency (Bernard and Thomas, 1989, 1990), differential interpretation of information (Kandel and Pearson, 1995), disclosure practices in different countries (Bailey et al., 2003), and effects of changes in disclosure regulation (Heflin et al., 2003).

Their market shares are much smaller if one considers weekly and monthly magazines, radio, television, the Internet, and gossip as competing suppliers in a more inclusive market for news.

Ball and Shivakumar (2008) also examine the first (last) analyst earnings forecast revision after (before) each quarterly earnings announcement for the subsample of firms covered by I/B/E/S. They note that an advantage of their approach is that they can examine all forecast types, not just point and range forecasts.

In contrast to our approach, Beyer et al. (2010) impute an abnormal stock return of exactly zero for every missing observation, which in our view artificially restricts the mean and variance of the signal on these days.

If earnings are announced on a non-trading day, day 0 is the first trading day after the announcement. Day +1 is often included to account for earnings announcements made after trading hours, while day −1 is included to account for potential leakage of information. Earnings announcement dates are extracted from the COMPUSTAT quarterly file and return data from the CRSP monthly and daily files.

To be precise, we add 4.8% to the abnormal R ² numbers reported in Table 2 of Ball and Shivakumar (2008). They define abnormal R ² as the regression adjusted R ² value minus its expectation assuming i.i.d. daily returns, which they separately estimate as 4.8%.

Dellavigna and Pollet (2009) find that, where I/B/E/S and COMPUSTAT earnings announcement dates differ, the earlier date is usually the actual date of the announcement. For the 1983–2011 subperiod, replacing the earnings announcement date from COMPUSTAT with an earlier one from I/B/E/S, when available, yields a marginally higher mean R ² of 11.58%.

We did not impose these requirements because their effect is to produce an atypical sample. The sample would contain few trading days in proximity to each other, e.g. days that occur in the same calendar week or month; it would also likely be skewed toward certain days of the week.

These results are based on samples that do not exclude observations with overlapping dividend announcement days and EADs. In untabulated analyses, on average, 4.39% (10.51%) of firm-quarters (firm-years) have overlapping earnings and dividend announcement windows; excluding observations with overlapping windows results in a lower mean dividend announcement days' R ² of 6.44%, so the likely effect of not excluding these observations is to make it more difficult to detect a difference in the R ²s of EADs and dividend announcement days.

Put differently, the conditional information content of dividend announcements by firms that pay dividends more frequently seems to be less than that for firms that pay dividends less frequently. Note that we ‘fill in’ missing quarterly dividend announcement days with a randomly chosen trading day, which typically has a low R ² (see Table 1). Including such random trading days in the second sample should have reduced the average R ² unless the actual dividend announcement days were more informative on average than those in the first sample.

On average, 4.86% (16.77%) of firm-quarters (firm-years) have overlapping earnings announcement and high-information arrival windows.

Moreover, the return distributions of low-information arrival days with and without the inclusion of EADs are likely to be very similar as the overlap between EADs and low-information arrival days is likely minimal (Francis et al., 2002).

In addition to management guidance for interim quarter earnings, we include annual earnings guidance issued during or after the fiscal fourth quarter as such guidance is equivalent to fourth quarter earnings guidance.

However, when annual earnings announcements were voluntary before the SEC was instituted, they were not very informative to the stock market (Sivakumar and Waymire, 1993). A possible reconciliation is that earnings reports have become more credible over the last century. Similarly, Butler et al. (2007) find that the earnings timeliness of firms that voluntarily reported quarterly was no greater than that of those that only reported at the mandatory semi-annual frequency during 1955–1969. The latter result is consistent with a self-selection argument in that only firms that would otherwise experience low earnings timeliness find it desirable to report earnings more frequently.

Foster et al. (1983), Stice (1991), and Easton and Zmijewski (1993), among others, failed to find significant market reactions to 10-K and 10-Q filings with the SEC. However, more recent studies of the EDGAR filing regime find significant market reactions (e.g. Qi et al., 2000; Asthana and Balsam, 2001; Griffin, 2003). The differing results are likely due to at least two factors: potential problems with identifying effective filing dates for the paper filing regime (i.e. when investors could only get paper copies of the filings) and limited access to the paper filings relative to electronic filings due to a cumbersome process for obtaining hard copies.

We also perform the same analyses on observations with above-median institutional ownership (from Thomson Financial) and below-median number of quarterly losses in the eight quarters prior to the event quarter. The results are similar and are available from the authors upon request.

Because of this infrequency, Ball and Shivakumar (2008) modify their approach and estimate a regression of calendar-quarter returns on three-day management forecast or preannouncement returns. We continue with an annual regression to facilitate combining the annual estimates for different earnings information channels later.

Panel D of Table 4 presents results for an analysis of filing amendments using the same procedure that generated panels A and B. Filing amendments appear to be less informative than the initial filings, with a conditional mean R ² of only 3.75%. For robustness, we also assess the informativeness of restatements using announcement dates procured from the US Government Accountability Office (GAO) for the 1997–2006 period. Compared with the results in panel D, the conditional mean R ² using US GAO data is much higher at 10% for samples of 134 firms on average, but the unconditional mean R ² is very similar at 3.17%. For brevity, these restatement results are not reported, but are available from the authors upon request.

Announcement returns are adjusted for the holding period return of the corresponding CRSP size decile portfolio. Also, we exclude the top and bottom 0.5% of the combined size-adjusted announcement return distribution.

We compute the distributional parameters for each announcement type quarterly, and use the time-series distributions of the parameters to test for differences.

Ball and Shivakumar (2008) report that the R ² for a sample of EADs without contemporaneous management forecasts during 1994–2006 is slightly smaller than the R ² for their main sample of EADs during 1972–2006. Because they do not report an R ² comparison between all EADs and EADs excluding contemporaneous management forecasts for the same time period, their reported differences could reflect environmental changes during their sample period.

We are likely understating the contribution of SEC filing days because we do not account for overlaps between SEC filing days and management earnings forecast days; i.e. to the extent they overlap, we assign all the explanatory power to management forecasts.

Firms make other announcements that convey earnings information, such as write-offs and restructurings (e.g. Bartov et al., 1998), mergers and acquisitions (e.g. Haw et al., 1990), 8-K filings (e.g. Lerman and Livnat, 2010), but these are relatively infrequent. In a few industries such as automobiles, firms release weekly sales numbers, while in a few other industries such as banking, railroads, and public utilities, there are additional mandatory regulatory filings. We expect these other announcements and filings to have minimal incremental impact on the total R ² in Table 6.

Additional information

Notes on contributors

Sudipta Basu

Paper accepted by Laurence van Lent.

Truong Xuan Duong

Paper accepted by Laurence van Lent.

Stanimir Markov

Paper accepted by Laurence van Lent.

Eng-Joo Tan

Paper accepted by Laurence van Lent.