|
Article Excerpt
I. INTRODUCTION
Recent studies infer the increasing usefulness of earnings announcements from overtime increases in the absolute or squared abnormal stock returns or abnormal trading volume at earnings announcement dates (e.g., Kross and Kim 2000; Landsman and Maydew 2002; Francis et al. 2002); the increase appears to be concentrated among large firms (Buchheit and Kohlbeck 2002; Lo and Lys 2001). We examine three explanations for this trend using a sample of 426 large firms: (1) over-time increases in the amount of unexpected earnings news; (2) over-time increases in investors' reaction to a unit of unexpected earnings news; and (3) over-time increases in the amount of other (besides "bottom line" earnings) information disclosed in the earnings announcement press release. (1)
It is important to distinguish among these three explanations because of their differing implications about the financial reporting information environment. The first explanation speaks to whether earnings are increasingly bringing news to investors (as opposed to summarizing or aggregating it); evidence in support of this explanation would suggest reconsidering claims that the summary earnings number is increasingly irrelevant to investors. (2) The second explanation speaks to behavioral changes among investors such that they respond more intensively to a given piece of earnings information. Such changes might result from properties of earnings (e.g., increasing persistence) or from economic or psychological factors unrelated to financial reporting practices, per se. The third explanation speaks to voluntary disclosure practices that are related to but distinct from financial reporting as it is regulated by generally accepted accounting principles. Evidence that the earnings announcement but not the summary earnings number itself is conveying increasing amounts of information to investors would suggest that discussions of changes in the usefulness of financial reporting should consider the package of information that the earnings announcement press release reveals to investors, including disaggregated and forward-looking earnings information.
The first explanation we consider is based on prior studies' findings that the magnitude of the market reaction to earnings announcements increases with the magnitude of unexpected earnings (e.g., Beaver et al. 1979), and that the relation is not linear (e.g., Freeman and Tse 1992; Basu 1997; Lipe et al. 1998). If the amount of absolute unexpected earnings revealed by earnings announcements has increased over time, then we expect to observe increases in the magnitude of the absolute market reactions to these announcements.
The second explanation is that the intensity of investors' reactions to a unit of unexpected earnings news, as measured by the earnings response coefficient (ERC), has increased over time. We investigate two sources of possible changes in ERCs: (1) increases in ERCs resulting from changes in factors known to affect ERCs--such as an increase in earnings persistence, or declines in systematic risk or risk-free interest rates (Easton and Zmijewski 1989; Collins and Kothari 1989); and (2) increases in ERCs resulting from shifts in the distribution of unexpected earnings to more steeply sloped regions of the unexpected returns-unexpected earnings curve. In terms of the latter, prior research shows that the ERC for extreme values of unexpected earnings (either positive or negative) is less than the ERC for moderate values of unexpected earnings, that the ERC for negative earnings surprises is less than the ERC for positive earnings surprises, and that the ERC for losses is smaller still. Consequently, if the distribution of unexpected earnings has become less extreme and/or characterized by fewer loss observations over time, then this might explain over-time increases in the absolute reactions to earnings announcements. (3) Our analysis of over-time changes in ERCs as an explanation for the increased market reactions to earnings announcements differs from Lev and Zarowin's (1999) analysis, which examines changes in ERCs from long-window association tests and does not investigate the sources of these changes.
The third explanation is based on the Hoskins et al. (1986) finding that earnings announcement press releases often convey other information (besides bottom line earnings) that explains a significant portion of the market reaction to earnings announcements. Our analysis of concurrent disclosures in earnings announcement press releases is consistent with Lo and Lys' (2001) conjecture that the disparity between results showing increasing information content of earnings announcements (as measured by absolute value or squared return responses and/or volume responses) and decreasing value-relevance of earnings (as measured, for example, by the [R.sup.2] from a regression of returns on earnings levels and changes) may be due to other disclosures released concurrently with the summary earnings number. Specifically, we examine whether concurrent disclosures have increased over time, and if so, whether this increase explains the increased price reactions to earnings announcements. Relative to Hoskins et al. (1986), we examine an expanded set of information items and we focus on trends in the disclosures over the past 20 years.
We begin by documenting the increased usefulness of earnings announcements for the 426 firms for which Compustat reports earnings announcement dates for every quarter from 1980 to 1999. Requiring a complete series of 80 Compustat earnings announcement dates ensures that the results are not due to changes in sample composition. This requirement also biases the sample toward long-surviving firms, which tend to be large and stable. Given prior studies' evidence that over-time increases in market reactions to earnings announcements are concentrated in large firms (Lo and Lys 2001; Buchheit and Kohlbeck 2002), this selection bias increases the likelihood that our sample exhibits the phenomenon under investigation. The focus on large firms also suggests that the results may not be generalizable to broader samples of firms, particularly if the propensity to include concurrent disclosures in earnings announcements is primarily a large-firm phenomenon.
To test the first explanation, we analyze the trend in the absolute magnitudes of unexpected earnings conveyed by earnings announcements. Results examining the absolute value of unexpected earnings (measured using a seasonal random walk [SRW] model and, separately, using analysts' forecasts) show a decline in the SRW measure and no pattern in the analyst measure. Both sets of results are inconsistent with the first possible explanation--that the increase in the magnitude of the market reaction is attributable to an increase in the magnitude of unexpected (bottom line) earnings. Tests of the second explanation--that the intensity of investors' reactions to a unit of unexpected earnings news has increased over time--examine the trend in investors' response to a unit of unexpected earnings. Regardless of whether we use SRW or analysts' forecasts to measure unexpected earnings, we find a significant decline in ERCs, inconsistent with the second explanation for the increased usefulness of earnings announcements.
Our analysis of earnings announcement press releases suggests that the missing piece to this puzzle is an over-time increase in the amount of concurrent information that firms disclose in these press releases. Specifically, for a sample of 2,190 press releases made by a subset of 30 (randomly selected from the 426) firms over 1980-1999, we find a significant expansion in the number of concurrent disclosures. For our sample, earnings announcement press releases are increasingly likely to include detailed financial statements, information about current operating data (such as shipments, orders, and margins), and information about nonrecurring earnings components (such as restructuring charges and foreign exchange gains or losses). These trends in concurrent disclosures explain the over-time increase in the absolute price reactions to earnings announcements for our sample firms. The most important concurrent disclosure, in terms of explaining the increasing reactions to earnings announcements, is a detailed income statement. We conclude that managers' voluntary decisions to expand concurrent disclosures in earnings announcement press releases (especially by including detailed income statements) drive the increased usefulness of those announcements to investors.
Long-window association tests reported in Francis and Schipper (1999) and Collins et al. (1997) suggest that the earnings number, as a single aggregate performance measure, declined in usefulness over 1980-1999. In contrast, our results suggest that the disclosure of earnings components--that is, the disaggregated elements that make up the earnings number--has increased the usefulness of the information package (the press release) whose central element is the summary earnings number. This latter finding, which is consistent with long-window research showing that earnings components provide information beyond that contained in the summary earnings number (e.g., Lipe 1986), suggests that discussions of changes in the usefulness of financial reports should approach the issue from the perspective of the bundle of information made available to investors in the financial reporting process, and not from the narrow perspective of bottom line earnings.
The rest of the paper is organized as follows. Section II details the sample and documents patterns in market reactions to earnings announcements similar to those shown in prior research (e.g., increasing usefulness and a positive association between unexpected returns and unexpected earnings for large, long-lived firms). Section III describes our investigations of over-time changes in unexpected earnings and in the intensity of investors' responses to unexpected earnings. Section IV reports the analysis of trends in the concurrent disclosures in earnings announcement press releases. Section V summarizes the results and concludes.
II. SAMPLE DESCRIPTION AND REPLICATION OF PRIOR RESULTS ON THE INCREASING USEFULNESS OF EARNINGS ANNOUNCEMENTS
Sample and Descriptive Statistics
Our sample includes all firms with four quarterly earnings announcements on Compustat (primary, secondary, tertiary, and research databases) for each calendar year t = 1980-1999 and with common stock returns on CRSP for the entire 20-year period. In total, 426 firms with 34,080 quarterly announcements meet these criteria. The benefits of this design choice include control over sample composition, and the increased likelihood of finding the phenomenon we wish to study because the increased usefulness of earnings announcements appears to be concentrated among larger firms (Buchheit and Kohlbeck 2002; Lo and Lys 2001). The cost is the inability to generalize results beyond the arguably large, surviving firms meeting our sample selection criteria.
Table 1 reports descriptive information about the average values of the sample firms' assets, market values, market-to-book ratios, and return-on-asset values for each year t = 1980-1999; we also report mean values calculated for the pooled sample (row labeled "All years") as well as for all Compustat firms with any available data (row labeled "All Compustat"). The rows labeled TREND in Table 1 show coefficient estimates and p-values from regressing the annual means for the sample and for the Compustat population on [TREND.sub.tq] = t - 1979. Given the selection criteria, it is not surprising that the reported asset and market values indicate that, relative to the population of Compustat firms, our sample firms are larger and grew at a faster rate over 1980-1999 (differences, not reported, are significant at the 0.001 level). As we note in subsequent tests, controlling for firm size and for changes in firm size does not affect our inferences. However, these controls for within-sample size effects do not shed light on whether the results we report would generalize to other samples composed of significantly smaller firms.
Table 1 also shows that sample firms' market-to-book ratios increased over 1980-1999 (the trend slope coefficient is 0.08, significant at the 0.001 level), consistent with intuition that surviving firms perform well. The population of Compustat firms also experienced an increasing trend in market-to-book ratios, but at a smaller rate than the sample firms (0.06 per year vs. 0.08 per year, difference significant at the 0.05 level). Return-on-asset comparisons provide further evidence of our sample firms' superior performance. Our sample firms enjoy a substantially higher return on assets than the Compustat population (1.39 percent vs. -5.52 percent, difference significant at the 0.001 level). (4) Return on assets decreased over the sample period for both the sample firms and the Compustat population (the trend for the sample is -0.02 percent per year compared to -0.43 percent for the Compustat population, both trends significant at the 0.001 level). Givoly and Hayn (2000, Tables 1 and 2) report similar decreases in other earnings-based performance measures and attribute the trend to an increasing conservatism in financial reporting.
Because some of the subsequent analyses examine over-time changes in the relation between unexpected returns and unexpected earnings, we examine three factors prior research has shown to affect the intensity of this relation. Table 1 provides descriptive evidence on risk-free rates (measured as the compounded monthly short-term Treasury bill rate in year t), systematic risk or beta (measured by the slope coefficient from firm-specific market model regressions), and persistence (measured as the slope coefficient, [[phi].sub.1], from the Foster (1977) seasonal first-order autoregressive model, estimated in cross-section for each year t). Annual estimates of risk free rates, betas and [[phi].sub.1] are reported in the three right-most columns of Table 1; the last row of Table 1 shows the trend in each variable. The significant (at the 0.001 level) declines in risk-free rates, and betas (5) suggest an increase in ERCs over 1980-1999; however, the significant (at the 0.003 level) decline in persistence suggests a decrease in ERCs. Given this mixed evidence, we do not predict whether the ERCs for our sample firms likely increased or decreased during 1980-1999.
Market Reactions to Earnings Announcements
We calculate beta-adjusted abnormal returns on each of the three days surrounding...
|
