`Tis the Season for Earnings! Analysis of Information Spillovers in Earnings Seasons

Transcription

1 `Tis the Season for Earnings! Analysis of Information Spillovers in Earnings Seasons Curtis Hall University of Arizona Jayanthi Sunder University of Arizona Shyam V. Sunder University of Arizona October 2012 We thank Andrew Finley, Ted Goodman, Mindy Kim, Pablo Machado, and seminar participants at the MIT Sloan Economics and Finance Seminar. We thank I/B/E/S for data on analyst forecasts. 0

2 `Tis the Season for Earnings! Analysis of Information Spillovers in Earnings Seasons Abstract A majority of earnings announcements are clustered in calendar time and referred to, in the popular press, as the earnings season. We examine the influence of information spillover during the earnings season on the information contents of earnings announcements of individual firms. We find that there are significant returns in the earnings season prior to a firm s own announcement as the market infers the firm s news from the industry peers news. This spillover effect is more pronounced when the firm announces later in the season and when the firm reports good news. We show that while the 3-day earnings announcement window explains less than 12% of the quarterly CAR, the combined effect of earnings season spillover and announcement reaction explains 26% of the quarter s CAR. We also find that the reaction to earnings news is influenced by the tone set by early announcers. This suggests that assessing the information content of earnings announcements must be placed in the context of the earnings season and announced news needs to be augmented with inferred news. 1

3 1. Introduction `Tis the Season for Earnings! Analysis of Information Spillovers in Earnings Seasons Recent studies have questioned whether earnings announcements convey significant new information to capital markets. Based on an analysis of returns in the 3-day announcement window, these studies conclude that information sources such as management forecasts and analyst forecasts, are more important than earnings announcements (Ball and Shivakumar, 2008; Beyer, Cohen, Lys, and Walther, 2010) 1. However, it is important to note that earnings announcements are not isolated events that convey news only about the announcing firm. Investors make inferences about the impending news of non-announcing firms using the information contained in prior earnings announcements (Foster, 1981; Han and Wild, 1990; Freeman and Tse, 1992; Ramnath, 2002; Thomas and Zhang, 2008). Information spillovers are economically significant since most earnings announcements are tightly clustered in calendar time, a period referred to as earnings season in the business press. We conjecture and find that information spillovers from early announcers in the earnings season significantly pre-empts news contained in earnings announcements of firms that announce later in the earnings season. Our study highlights that conclusions about the information content of earnings announcements must take into account reaction to inferred news while evaluating the news content of announced news. We address two research questions: (a) how is the information content of earnings announcements affected by information spillovers in earnings seasons, and (b) how is the effect 1 Using an alternate methodology, Basu, Duong, Markov, and Tan (2010) conclude that earnings announcements are the dominant source of firm information relative to all other information sources, a conclusion that is also consistent with Francis, Schipper, and Vincent (2002). 1

4 moderated by nature of the news, sequence of the firm s announcement within the earnings season, and tone for the earnings season set by the initial announcer. 2 We use earnings seasons as our setting since it is a time period characterized by rapid fire earnings announcements and a high rate of information arrival about firms in an industry. 3 Earnings seasons have become important information events for a number of reasons. First, the proportion of firms that announce in an earnings seasons has increased over time such that on average 80% of firms in every 3-digit SIC based industry cluster make earnings announcements earnings during the earnings season. Second, clustering of earnings announcements has increased as more firms are adopting the calendar fiscal-year. By 2010, 92% of firms in the median industry had fiscal year end coinciding with the end of calendar year, compared to 83% in Third, as the SEC and stock exchanges have shortened the filing date from the end of the fiscal quarter, the window over which firms can announce earnings has become shorter. For example, the average length of earnings season is 18 days (and 34 days for the fourth quarter) with a decreasing trend over time. Finally, media attention to earnings seasons has also increased with the increase of television coverage and the internet. For example, a search on Google Trends for the phrases earnings or earnings announcements shows a large spike (with scores ranging between 30 and 80) in use of these keywords in online searches coinciding with timing of earnings seasons (see Figure 3). First, we examine whether the reaction to information spillover during the earnings season is consistent investors inferring the firm s earnings news. We find that the firm s abnormal returns during the earnings season leading up to its 3-day earnings announcement window, defined as 2 Information content of earnings announcements is measured using the cumulative abnormal returns (CAR) and earnings response coefficients (ERC)). The nature of news refers to whether the news is good news or bad news. Sequence of the announcement refers to the placement of the announcing firm in calendar time relative to other announcers in the earnings season. Finally, the tone for the earnings season captures the expectations for the announcing firm based on the news contained in prior announcers in the earnings season. 3 The information arrival in this short span of time is so intense that an analyst describes the earnings season as a time when you are drinking [information] from a fire hose. (See, 2

5 CAR Pre, is positively correlated with the actual earnings surprise, Surprise_PreEAS. 4,5 This suggests that investors anticipate earnings news based on peer firms earnings announcements and form an expectation of the firm s surprise that is correlated with the subsequently realized surprise. This spillover effect appears to be more pronounced in case of competitive industries, consistent with the intuition in Balakrishnan and Cohen (2012). Further, the positive relation between CAR Pre and a firm s earnings news is significant only for firms that report good news suggesting that the market is unable to infer the magnitude of a firm s bad news using information spillover. The anticipation of the firm s earnings news by the market in turn has implications for earnings announcement returns (CAR3) when the news actually arrives. If all the relevant firm specific news were contained in the firm s own earnings announcement, we would expect that the entire season s CAR (CAR EAS) is captured by CAR3. It is important to note that unlike raw returns, CARs are not mechanically driven by the length of the return-cumulation window and our expectation of CAR Pre under the assumption of no spillover is zero. However, we find that the explanatory power of CAR3 in explaining the firm s CAR over the entire earnings season (CAR EAS) is lower than CAR Pre. This suggests that markets incorporate earnings news into prices even prior to the firm s announcement and this is stronger for firms that announce later in the season. Taken together, the analysis on CAR Pre and CAR3 provide evidence that there is a shift in market reaction from the 3-day announcement window to the period leading up to the announcement and this is stronger for firms announcing later in the season. Therefore, the CARs in the earnings season leading up to and including the firm s own announcement window can be attributed to the firm s earnings news arising from two sources: 4 See figure 1 for a time line of events in the earnings season. 5 Surprise_PreEAS is computed as the actual earnings announced in that quarter less the last available analyst earnings forecast that was issued prior to the start of the associated earnings season. 3

6 information spillover and the actual earnings announcement. We next examine the relative explanatory power of CAR3 and CAR Pre for the entire quarter s CAR. We find that while CAR3 explains, on average, 12% of the entire quarter s CAR, the joint explanatory power of CAR3 and CAR Pre, is on average 25% of the quarter s CAR. CAR Pre is typically devoid of any news from the firm itself since most firms subject themselves to quiet periods prior to their earnings announcement. 6 Taken together, we conclude that ignoring the inferred news from information spillover understates the information content of the announced news and this effect is stronger for firms that announce later in the season. In additional tests, we examine whether the ERC, i.e. the stock price reaction per unit of earnings news in the 3-day earnings announcement period and over the entire earnings season varies with the sequence in the earnings season and nature of news. We first examine ERCs over the 3-day earnings announcement window. Using the earnings surprise based on the last analyst forecast issued prior to earnings announcement (Surprise), we find that the ERC for the firm s earnings announcement is decreasing as the firm announces later in the earnings season. 7 We assign quartile ranks to announcing firms in the earnings season based on the sequence in which they report their earnings. We find that the ERC is significantly bigger for the first quartile of announcers, in calendar time, compared to those that announce in the last quartile. 8 This effect is 6 NIRI s 2008 survey on earnings release practices of their member firms suggests that across groups of firms with over $100 million market capitalization 0-10% of firms had at least a one-week quiet period prior to earnings release and 59-76% had a quiet period of three weeks or more. For example, Intel Corporation announced in an 8-K filed on December 12, 2011 that Intel's Quiet Period will start from the close of business on Dec. 16 until publication of the company's fourth-quarter earnings release, scheduled for Jan. 19, During the Quiet Period, all of the Business Outlook and other forward-looking statements disclosed in the company's news releases and filings with the SEC should be considered as historical, speaking as of prior to the Quiet Period only and not subject to an update by the company. 7 This result is not merely driven by stale forecasts as discussed in Easton and Zmijewski (1989) since we use the most updated analyst forecast available prior to the earnings announcement. However, it is surprising that despite the significant information spillovers in market returns, we find that only 41% of analysts revise their earnings estimates during the earnings season. 8 Note that our identification of firms that announce later in the earnings season is based on the sequence in which firms within an industry typically announce earnings. This is different from the late and early announcements in Bagnoli, Kross, and Watts (2002) since in their case firms delay announcing with respect to the expected 4

7 stronger when the firm has good news suggesting that they get a bigger pop in their returns to good news if they announce ahead of their peers. On the other hand, the reaction to bad news is unaffected by the firm s quartile rank. Next, we test whether the market ERC over the entire earnings season is systematically different based on its quartile rank of announcement timing. While the spillover hypothesis would predict no difference across firms based on timing since all the news is public by the end of the season, anecdotal evidence from the media suggest that managers perceive an advantage to announcing ahead of peers. We refer to this phenomena as setting the tone for the earnings season. 9 We find that, on average, the reaction to news, primarily good news, is weaker when the firms announce later in the season. This highlights the importance of recognizing earnings announcements that set the tone while evaluating the news content of subsequent announcers. An interesting implication of the prior results is that a firm has incentives to jump the queue in the sequence of announcers in order to improve the market reaction to their earnings announcement. We examine if such a change in the sequence of earnings announcements has an effect on the ERC. We construct a transition matrix of the quartile ranks over successive earnings seasons. We find that the relative position of a firm in the sequence of announcers in an industry is fairly sticky. However, the increases (decreases) in the quartile rank are associated with bad (good) news, similar in spirit to Bagnoli, Kross, and Watts (2002). Focusing on the effect of changes in the firm s quartile rank on the reaction to firm news, we find that, only firms announcement date. While some of the later announcers could be delaying their announcements relative to an expected date, our results hold in the sample of firms that do not change their relative rank in the announcement sequence. 9 For example, in the for the first quarter of 2012 as the banking industry is recovering from the financial crisis, individual banks have been competing with each other to come out earlier with their better-than-expected performance. The competition to set the tone for the industry is capture by the following quote Wells Fargo joined JPMorgan Chase and Co. as the first of the major banks to report results. Wells moved up its report this quarter to be on the same day with its New York peer, which traditionally had been the first to disclose its results and set the tone for the earnings season. [ 5

8 announcing good news experience lower ERCs in their earnings announcement window when they announce later in the earnings season (i.e. increase their quartile rank). Together, the results suggest that firms with good news have strong incentives to report earlier since they get a bigger announcement reaction. However, it does not appear to be easy to break into the first quartile of announcers who tend to be larger firms with greater analyst following. This in turn leads to the question of whether the tone, i.e. bad (good) news earnings announcements of the first announcer(s) in an industry, influences the reaction to earnings announcements of other firms. We find that when the first announcer(s) set a bad tone (i.e. report bad news), subsequent announcers with bad news experience less negative returns. There is no significant effect on subsequent announcers when the initial tone is good. Overall, our results suggest that informativeness of individual firm earnings announcements in earnings seasons is moderated by earnings announcements of intra-industry peer firms, timing of the firm s announcement in the sequence of announcers within the season, content of the news being disclosed, and degree of product market competition within the industry. Collectively our results provide a new perspective on the puzzle of low informativeness of earnings. While other researchers have suggested that low information content of earnings announcements is due to other information announcements by individual firms, there could other explanations too. We provide evidence that earnings announcement news for a large number of firms is preempted by information spillovers from other firms. Our results do not reject the view that management forecasts and analyst forecasts are important relative to earnings announcements. Instead investors should be aware that earnings announcements are important news events but there is significant amount of news that is inferred before it is announced. This also emphasizes the importance viewing earnings news in context of earnings seasons as an extended information window. 6

9 The rest of the paper is organized as follows: section 2 provides the motivation and review of the related literature relevant to our study, section 3 describes the data, section 4 contains the research design, analyses, and discussion of results, and section 5 concludes. 2. Literature and Predictions 2.1 Earnings announcements and earnings seasons Earnings announcements are clustered in calendar time and a majority of these are made in a relatively short span of time. The cluster of announcements in calendar time is referred to in the media and by analysts as the earnings season. The earnings season starts typically within 10 days from the end of the fiscal quarter. The clustering is to be expected since the SEC requires that firms file earnings reports within 60 to 90 days for the annual form (Form 10-K) and 40 to 45 days for the quarterly form (Form 10-Q) from the fiscal quarter end date. 10 However the actual clustering of earnings announcements is far tighter than the allowed days under SEC requirements. For example, in the earnings season for the fourth quarter of 2004, over 80% of firms by market capitalization (numbers) announced their earnings within 28 (43) trading days from the last day of the end of the fiscal year (see Figure 2). An explanation of this clustering could be the exchange listing rules that require that firms disseminate the information as soon as it s available. 11 Thus, in 10 Over the sample period we have used in this study, the requirements have changed. Effective December 2005, large accelerated filers (those with $700 million or more public float) have 60 days for the 10-K and 40 days for the 10-Q. Accelerated filers (those with more than $75 million and less than $700 million public float) must file 10-K within 75 days and 10-Q within 40 days. Non-accelerated filers (those with less than $75 million public float) must file 10-K within 90 days and 10-Q within 45 days. Prior to these changes, until, 2002 most companies were required to file 10-K within 90 days and 10-Q within 60 days. 11 According to the NYSE manual Sec , A listed company is expected to release quickly to the public any news or information which might reasonably be expected to materially affect the market for its securities. This is one of the most important and fundamental purposes of the listing agreement which the company enters into with the Exchange. Further Sec specifies that Information required to be released quickly to the public under Section above should be disclosed by means of any Regulation FD compliant method (or combination of methods). Finally Sec specifies that, Any company with voting or non-voting common securities listed on the Exchange that is required to file interim financial statements with the SEC is required to disseminate in a manner 7

10 practice, most firms announce their earnings synchronously with their industry peers leading to the clustering in calendar time. Another driver of the earnings season is a trend among firms to adopt the calendar year as the fiscal year. Over time, the proportion of firms choosing the December month end as their end of fiscal year has been increasing. For example in 1994 while 83% of firms in the median industry had their fiscal year end coinciding with the end of calendar year by 2010, the percentage increased to 92% of firms in the median industry. 12 As a result, the prominence of the earnings season has increased with greater number of firms announcing in this period resulting in intensive coverage in media and the internet. As an example an analysis of online search expression, earnings announcement using Google Insights for Search, shows large spikes around earnings seasons reflecting the investor and media attention to earnings announcements in these time periods (see Figure 3). This is consistent with increased information demand and processing prior to earnings announcements highlighted by Drake et al. (2012). Based on the preceding discussion, we expect that the reaction to the firm s earnings news is not just restricted to the narrow 3-day firm announcement window. Recently, a number of studies have addressed the issue of the importance of earnings announcements as a source of news about the firm relative to other sources such as management forecasts, earnings pre-announcements, SEC filings and analyst forecasts. Beyer et al. (2010) show that of the accounting based information provided in quarter for a given firm, earnings announcements contribute to only 8%. Using a different approach, Ball and Shivakumar (2008) show that the average quarterly earnings announcement is associated with about 1% to 2% of total annual information conveyed about an consistent with the Exchange's immediate release policy an interim earnings release as soon as its interim financial statements are available. 12 There are some notable exceptions to the December-end fiscal year that tend to be in the technology industry. For example Apple, Microsoft, and Cisco have fiscal years that do not end in December. These firms are not in our sample since we limit our sample to December year end firms. By excluding some such large firms we could be underestimating the information spillover effects for some industries. 8

11 individual firm. These results have been re- examined by Basu et al. (2010) more recently by modifying the methodology of Ball and Shivakumar (2008) and they conclude that earnings announcements convey more information than indicated in the earlier studies. In fact, as a benchmark, they find that earnings announcement returns are comparable to returns in the four most extreme return days in a year. Common to all these studies, and widely accepted as the methodology for assessing informativeness, is the use of a window of 3 days including the earnings announcement day (-1, 0 and +1 days). The short window has the advantage of isolating the effects of the market reaction of earnings announcement without contaminating information from other events. However, the short window in turn excludes the reactions of the firm that are preempted by information spillovers from other firms announcing ahead in calendar time (Foster, 1981; Han and Wild, 1990; Freeman and Tse, 1992). There is disagreement in the literature about whether such inferences are subject to over or under reaction (Ramnath, 2002; Thomas and Zhang 2008). However it is a robust result that there are information spillovers from earnings announcements for firms that have not yet made earnings announcements. We expect that there are significant CARs in the earnings season leading up to the firm s own earnings announcement and that the returns anticipate the actual earnings surprise of the firm using information from announcing peer firms. The information spillover within the earnings season would affect the measured informativeness of individual firms earnings in the 3-day earnings announcement window. 2.2 Timing and nature of news and earnings seasons Market Reactions to earnings announcements are sensitive to the timing of the information release. Prior research has shown that firms with good news tend to announce early (McNichols, 9

12 1988; Begley and Fischer, 1998). Skinner (1994) suggests that firms with bad news will announce early by way of voluntary disclosures to avoid litigation. In contrast to these approaches, we examine the timing of the release of earnings announcement by individual firms in the sequence in which all firms announce within an industry s earnings season. In our setting, earlier and later announcements refers to the timing of the announcement within the earnings season. Anecdotally, the business press has accounts of firms setting the tone by announcing earnings ahead of their normally scheduled announcement date in the earnings season. Due to information spillover from news announced by peer firms earlier in the earnings season, firms that report later in the season will have more of their announcement news preempted. We expect that this will reduce the informativeness of the firm s own earnings announcement. Another dimension on which earnings announcement reactions systemically vary is based on whether the firm s announcement contains good news or bad news. According to Roychowdhury and Sletten (2011), earnings information usefulness lies in the fact that bad news information cannot be delayed any more. Thus if the managers have already not disclosed bad news through voluntary disclosures, the news has to be released in the earnings announcement. Thus informativeness of bad news earnings announcements would, ceteris paribus be expected to be more informative. We examine whether the information spillover and the relative importance of the earnings announcement varies by the nature of the firm s news. Finally, we examine whether the timing and nature of news interact to influence the information spillover effect. As part of the analysis, we also examine whether the timing in the earnings season has an effect on the tone that is set for an industry and therefore, on the reaction to earnings news of subsequent announcers. 10

13 3. Data In this section we describe the construction of our sample. We begin our sample construction with all firm-quarter observations from the COMPUSTAT-CRSP combined data with information on earnings announcement dates and with available information to compute the 3- day abnormal returns around the earnings window. We then require the firm to have available information on I/B/E/S to construct the analysts expectation of the firm s earnings for the period 1994 through We begin our sample in 1994 because the coverage and accuracy of I/B/E/S data improved significantly from This yields a sample of 139,040 firm quarters. We exclude 10,831 firm-quarters in SICs and as firms in these industries, which are in the Retail sector cluster with January fiscal year-ends. All other industry groups have December as the modal fiscal year end. We then construct industry groups based on 3-digit SIC codes since we define the earnings season for each 3-digit SIC group. Since our definition of earnings season and most of our analysis is at the 3-digit SIC level, we then require at least 4 firm-quarter observations for each industry group. This yields a sample of 104,817 firms. On average over 80% of firms in a given 3-digit SIC code group have December as their fiscal year end and therefore, the clustering in earnings announcements or seasons arises because of firms with December fiscal year-ends. Consequently, we only retain firms with December year-ends. This allows us to align firms in both calendar time and fiscal quarter ends. This yields the final sample, 82,683 firm-quarters represent firms that announce during the industry s earnings season. However, when we study the abnormal returns during the earnings season but prior to the firm s announcement window (CAR Pre), this variable is not defined for firms that announce in the first two days of the earnings season and therefore the sample reduces further to 73,172 firm-quarters. 11

14 4. Empirical Tests and Results 4.1 Earnings Seasons We first examine the extent to which firms cluster with respect to their earnings announcements within the earnings season. The earnings season can be viewed either at the market-wide level or the industry level. At the market-wide level, the earnings announcement by Alcoa Inc. is traditionally viewed as the start of the earnings season as described in section 2. In Figure 2, we plot the cumulative distribution of announcers by numbers and market capitalization starting with the earnings announcement by Alcoa Inc. for a representative quarter, December The earnings season in this case began on January 10, 2005, which is 6 trading days after the fiscal year-end and within the next 17 trading days, more than half of the firms by market capitalization announced their earnings for the fiscal year However, not all industries report over the same time interval and there are noticeable industry-clusters of earnings announcement dates. Since we focus on intra-industry information spillovers, we define earnings seasons at a 3-digit SIC level. Specifically, for each 3-digit industry, the earnings season starts from the first earnings announcement by any firm in that 3- digit SIC and ends when the last announcer in the same 3-digit SIC industry announces its earnings. Firms are ranked based on the number of trading days between the firm s own earnings announcement date and the start of the season. A firm is considered early if the firm announces its earnings in the first quartile of announcers within its season and a firm is considered late if it announces in the last quartile. We drop firms that announce earnings after the SEC filing date deadline for 10-Qs and 10-Ks since these are outlier firms with some kind of a reporting problem. We measure the length of the earnings season in two ways. First, we measure the number of trading days that elapse from the first to the last announcer in the 3-digit SIC industry group. 12

15 Second, we construct a measure, labeled as Duration of the earnings season. Duration is the average time to announcement of a given firm weighted by its market capitalization, scaled by the total market capitalization of the industry (. The length of the earnings season using trading days and duration are reported in Table 1. For fiscal quarter 4 announcements, while the earnings season lasts around 34 trading days on average, the number of days using duration is less than 10 days. This suggests that larger firms in any given industry tend to systematically announce earlier in the earnings season. This trend holds across our sample period and across all other quarters too. 4.2 Reaction to earnings news in the earnings season We first examine the descriptive statistics of firms that announce in the earnings season by the placement of the firm in the sequence of all announcers and by the nature of the earnings news reported. The results are in Table 2. Panel A shows firm characteristics of announcers across three groups based on the sequence of announcers within the earnings season. Those that report early in the season (first quartile of announcers), those that report in the late in the season (last quartile) and the rest are ones that report in the middle of the season. The tests of differences in means and medians of the firm characteristics suggests that the announcers who report in the first quartile are significantly larger, more profitable and have lower B/M, leverage, and incidence of losses. They also have greater analyst coverage. Interestingly, they have more positive earnings news on average compared with the later announcers and consequently higher abnormal returns. This is consistent with larger firms having better systems in place to get the earnings information out faster and these firms might also enjoy some priority from audit firms trying to allocate their resources across various clients at the busy time. Finally, firms with 13

16 negative earnings surprises and losses potentially take longer to compile their earnings numbers since they may want to recheck their numbers to ensure their accuracy before reporting bad news. Finally, CAR3 is greater for firms that report earlier in the season than those that report in the middle of or late in the season. In Panel B, we report the descriptive statistics for the sub-samples of good-news and badnews firms. We find that firms with bad news (negative earnings surprise) tend to be smaller and less profitable than firms with good news. The abnormal returns to bad news are significantly higher in the earnings announcement window (CAR3) and the total earnings season window (CAR_EAS) with the magnitudes typically double (triple) that of the mean (median) return for good news, consistent with evidence in Roychowdhury and Sletten (2011) who find that the earnings announcement CARs are bigger in magnitude for bad news firms. We find preliminary evidence of pre-announcement information spillover when we examine the differences in CAR Pre. The mean and median CAR Pre is significantly greater for good news relative to bad news suggesting that investors are at a minimum able to anticipate the direction of the earnings surprise. We explore this result in more detail in the following section Information spillovers prior to earnings announcements To examine the hypothesis that markets attempt to infer the firm s earnings news from peer firm announcements during the earnings season, we examine the abnormal returns during the earnings season but prior to the firm s own announcement. CARPre is measured as the sizeadjusted buy and hold returns over the window from the day when the first peer firm announces in the same 3-digit SIC code for a given quarter through day -2 relative to the firm s own earnings announcement, where day 0 is the firm s own earnings announcement day. This analysis excludes firms that report in the first 2 days of the earnings season as they do not have a distinct earnings season period prior to their own announcement. We examine whether the 14

17 market infers the earnings news of firm i based on earnings announcements by other firms in the same industry that precede its announcement. We use the realized news of firm i as a proxy for the market s expectation of firm i s news during the earnings season but prior to its earnings announcement. We measure the earnings surprise of firm i at the start of the earnings season as firm i s actual earnings less the last issued analyst forecast for firm i s earnings and define this as Surprise_PreEAS. The mean (median) length of this pre-announcement period in the earnings season is 10 (9) trading days. It is important to note here that the reaction during the earnings season is unlikely to be a reaction to new information from the firm because most firms have a voluntary quiet period of typically 4 weeks prior to the earnings announcement (see footnote 7). Therefore any reaction can be attributed to information from peer firms earnings announcements. We therefore estimate the following model: CAR Pre i = α + β 1 Surprise_PreEAS i + β 2 Industry Surprise Pre + β 3-8 Controls + ε (1) where Industry Surprise Pre represents the average earnings news of all peer firms in the same 3-digit SIC that announce earnings prior to a firm s own earning announcement. Specifically, for each peer firm that has already announced earnings, we compute earnings news as their reported earnings less the last issued analyst forecast. High (Low) Competition is an indicator variable that takes the value 1 for industries with below- (above-) median Herfindahl Index. If the returns of firm i respond to the industry news during the earnings season, we expect a positive coefficient on Industry Surprise Pre. A positive coefficient on Surprise i implies that the market s expectation of firm i s news based on information spillovers is, on average, positively correlated with the actual news. The results are reported in Table 3. We find that the coefficient on Surprise_PreEAS i is always positive and statistically significant in all specifications except the sub-sample of firms announcing bad news. This suggests that the market is able to at least partially infer firm i's earnings news from the announcements by other 15

18 industry firms and investors find it difficult to preempt firm-specific bad news. This provides strong evidence that the information spillover during the earnings season preempts part of the earnings news of the firm even before its announcement. Interestingly, Industry Surprise Pre is on average insignificant. However, within the sample of highly competitive firms, the coefficient is positive and significant. This suggests that the peer firms earnings surprises are themselves valuable when the industry is highly competitive, consistent with Balakrishnan and Cohen (2012) who find that there is greater comparability of earnings in competitive industries. The control variables have the expected sign consistent with the prior literature Relative importance of the 3-day reaction and pre-announcement reaction Next we examine the relative importance of the 3-day CAR centered on firm i's earnings announcement (CAR3) and the CAR Pre in explaining the overall earnings season abnormal returns (CAR EAS). In the absence of information spillover from other firms, CAR3 would be the primary driver of CAR EAS and there may be some delayed processing of earnings information after the 3-day event window. However, the abnormal returns prior to the actual earnings announcement should not explain the overall earnings season returns for firm i unless of course the market anticipates the firm s news in CAR Pre. We estimate the following equation: CAR EAS(entire season CAR) = α + β 1 CAR Pre+ β 2 CAR3 + β 3-8 Controls + ε (3) The results are reported in Table 4. We estimate equation (6) in the full sample as well as sub-samples of earlier announcers, later announcers, announcers with good and bad news. In all samples, the results indicate that the CAR Pre contributes significantly to the overall abnormal returns during the earning season. The striking differences in the relative importance of CAR3 and CAR Pre in explaining the season s CAR is across the subsamples based on the placement in the sequence of industry announcements, i.e Early versus Late announcers in the earnings season. Consistent with information spillover from other firms announcements, we find that 16

19 firms that go later in the season have realized more of the overall season returns even before their earnings announcement while firms that go early in the season have more of their news impounded into returns during their 3-day announcement window. The incremental importance of CAR Pre relative to CAR3 holds regardless of the type of news. Overall, this analysis suggests that the importance of the 3-day earnings announcement window depends on when the firm reports in the sequence of announcers in an industry Revisiting the importance of earnings in explaining the quarter s CARs Collectively, our results provide strong evidence of information spillover during the earnings season, which in turn reduces the informativeness of the earnings announcement. To quantify the magnitude of the information spillover, we regress CAR3 and CAR Pre on the total quarter s CAR, measured as the size-adjusted CAR over the entire announcement quarter, and assess the incremental R 2 of each component of reaction to firm s earnings news. The results are reported in Tables 5. We find that while CAR3 by itself explains, on average, 12% of the entire quarter s CAR, the joint explanatory power of CAR3 and CAR Pre, is on average 25% of the quarter s CAR. CAR Pre is unlikely to contain any news from the firm itself since most firms subject themselves to quiet periods prior to their earnings announcement (see footnote 7). Further, since we are studying abnormal returns rather than raw returns, the expectation of CAR in the absence of any news is zero. This suggests that restricting the analysis of earnings announcement reactions to the 3-day announcement window as is conventionally done in the literature underestimates the importance of the earnings news. One solution is to extend the earnings window to cover 10 days prior to the earnings announcement since, on average, the length of the prior period is 10 days in our sample. We include an analysis of this pseudo window (CAR[- 10,1]) and find that the explanatory power of the returns is 19%, which is similar order of magnitude to the combined explanatory power of CAR3 and CAR Pre. We note that the pseudo 17

20 window also has additional explanatory power beyond CAR3 alone, which is also consistent with information spillover Implications of information spillover for ERCs in the 3-day earnings window If earnings news is pre-empted in the days of earnings season that precede the firm s announcement, then the sensitivity of returns to the earnings news in the 3-day announcement window should be decreasing in the extent of spillover. This implies that firms which announce later in the season have a greater opportunity for information spillover and should have lower ERCs in their own earnings window. We also examine whether the market pre-empts the firm s news differently depending on the nature of the firm s news. We therefore estimate the following two equations with standard errors clustered at the firm and year level: CAR3 = α + β 1 Surprise + β 2 Surprise*Days Since Start of Season + β 3-8 Controls + ε (2a) CAR3 = α + β 1 Surprise + β 2 Surprise*Early + β 3 Suprise*Late + β 4 Early + β 5 Late + β 6-11 Controls + ε (2b) Where Surprise is measured as the reported earnings less the last issued analyst forecast of earnings prior to the firm s earnings announcement. We find that analysts revise earnings forecasts during the earnings season for 41% of firms while in the case of the remaining 59% Surprise takes on the same value as Surprise_PreEAS. The results of the regressions are reported in Table 6. The first column reports the estimates from equation (2a) above. While the coefficient on Surprise is positive and significant, the sensitivity of the announcement window returns to Surprise drops significantly for each day that elapses since the start of the season and firm i's announcement. Column (2) reports the estimates from equation (2b) on the entire sample. The baseline case in equation (2B) is the group of firms that announce in the middle of the season (quartiles 2 and 3, when firms are ranked in calendar time based on the sequence in which they report earnings). The coefficient on Surprise * Early is positive significant suggesting that firms that report early in the season have significantly higher ERCs than firms that announce in 18

21 the middle of the season. Consistent with these results, the coefficient on Surprise * Late is negative and significant implying that firms that announce later in the season have a lower ERC consistent with most of the earnings news being pre-empted. We also test for difference in ERCs between the firms that announce earlier in the season and those that announce later and find that difference is significant at the 1% level (F-statistics of 51.7). One potential concern in interpreting this result is that firms that announce later in the earnings season may have chosen to delay their earnings announcement in relation to their expected announcement date, in the spirit of Bagnoli et al (2002). We would like to separate this effect from the habitual sequence in which firms tend to report wherein some firms tend to announce earlier in the season while other announce later in calendar time. To check this, we restrict the analysis to the sample of firms with no quarter-on-quarter change in quartile ranks of the announcement sequence. The results of this estimation are reported in column (3) of Table 6. We continue to find that the sensitivity to earnings news is decreasing as the firm reports later in the season even in this sub-sample, consistent with greater preemption of news for later announcers. Therefore, we conclude that our results are not subsumed by the effect documented in Bagnoli et al. We then estimate (4b) separately on sub-samples of firms based on their earnings news. The difference between earlier and later announcers is bigger when the firms report good news compared with firms that report bad news. Overall, the CAR3 results are consistent with the results on CAR Pre in that when news can be pre-empted from earnings announcement of other firms, the 3-day abnormal returns are less sensitive to the earnings news. This suggests that focusing exclusively on the 3-day earnings announcement window to estimate the importance of earnings announcements understates the importance of the information. This is because in competitive information markets, market participants infer firm-specific news from the earnings season information spillover. 19

22 4.3.5 Role of Announcer Sequence on how news is interpreted over the entire season The results so far suggest that the degree of spillover and the reaction to the earnings news is a function of the placement of the firm in the sequence of announcers within the 3-digit SIC group. The information spillover hypothesis has predictions for timing of the reaction within the season by not the overall reaction once the season is over. However, we do not expect any differences in the total reaction to a firm s news over the entire earnings season, unless firms who report early in the season get more attention than the later announcers. There is some anecdotal evidence that some firms believe that this may be the case. 13 Alternatively, firms that report early might face a greater distraction problem (as suggested by Hirshleifer et. al., 2009) because the number of announcements (i.e. intensity) is typically higher in the early part of the season. We therefore estimate the following equation: CAR EAS = α + β 1 Surprise_PreEAS + β 2 Surprise_PreEAS*Early + β 3 Surprise_PreEAS*Late + β 4 Early + β 5 Late + β 6 Industry Surprise Pre + β 7-12 Controls + ε (4a) CAR EAS = α + β 1 Surprise_PreEAS + β 2 Surprise_PreEAS * Early + β 3 Surprise_PreEAS*Late + β 4 Early + β 5 Late + β 6 Industry Surprise Pre*High Compet. + β 7 Industry Surprise Pre* Low Compet + β 8-13 Controls + ε (4b) We expect no differences in the coefficients on Surprise_PreEAS, the earnings surprise as measured at the start of the earnings season, that are systematically related to when in the season the firm announced its earnings. This is because, over the entire earnings season, the returns should incorporate the entire earnings news such that by the end of the season, the sequence and spillover should not matter. Further, we examine whether the firm i s returns incorporate the industry news per se or whether the reaction to the industry news in the CAR Pre was merely part 13 Wells Fargo seems to have some good news to share this quarter in order to justify its decision to move up its earnings release date to this Friday so that it coincides with competitor JPMorgan Chase traditionally the first bank to report earnings. This appears to be yet another way the two banking giants are competing for attention. Forbes magazine dated April 10,

23 of the process of inferring firm i s news. The difference between (4a) and (4b) is that we allow the importance of industry news to vary based on the nature of industry competition. The results are reported in Table 7 Panel A. First, we note that on average firms that report earlier in the earnings season have higher returns than those who report later in the season. While part of this result may be driven by firms that miss their expected date getting penalized by the market (Bagnoli and Watts, 2002), that is unlikely to be a full explanation for this result. As we discuss later, there is a fair amount of stickiness in the sequencing of firms in the industry. Second, we find that in three out of the four specifications, the coefficient on Surprise_PreEAS is significantly lower for firms who announce later in the season compared with those who announce earlier, i.e. the ERCs seem to be decreasing as the earnings season unfolds. However, in the case of bad news firms the overall earnings season CAR is significantly negative with later announcers facing more negative returns, but the timing does not affect the overall season ERC. Overall, this suggests that firms that report later not only face lower CARs over the earnings season but also a lower reaction to their news when they report good news later in the season. We then examine whether the earlier result from Table 3 that the CAR Pre incorporates the industry news persists when all the information is released over the entire earnings season. Interestingly, once firm i s own earnings news in announced, the abnormal returns do not incorporate the surprise from other firms in competitive industries. This suggests that the reaction to peer firms was merely part of the process of forming expectations about firm i s earnings news rather than a reaction to industry news per se. Surprisingly, there remains some effect of the peer firm news in not so competitive industries. Next, we examine whether firms actively pick their announcement timing in the earnings season sequence. The transition matrix is presented in Panel B of Table 7. First, focusing on the diagonal of the transition matrix, we find that the largest percentage of firms stays on the 21

24 diagonal, i.e. they maintain their quartile rank in the industry. This is particularly true for the early announcers. Given the sophistication in internal controls and systems and access to auditors required to announce early, only about 4% (8%) of firms announcing in the last (fourth) quartile in a given quarter, announce early in the next quarter. Conversely, only 3% of early announcers in a given quarter announce in the last quartile in the next quarter. Despite the stickiness of the relative timing within an industry, the earnings surprise of the current quarter in decreasing in the change in rank across consecutive quarters, consistent with bad news firms reporting late with respect to both their own expected announcement date and other firms in the industry. We then examine whether a change in the relative quartile rank in an earnings season affects the market reaction to earnings surprises. We define a variable called Change in Quartile Rank which is the firm s current season quartile rank minus its season rank from last quarter. For example a firm moving from early to late has a Change in Quartile Rank of 3 while a firm moving from late to early has a Change in Quartile Rank of -3. The results are reported in Panel C of Table 7. We find that in the short 3-day earnings announcement window, the reaction to good news is decreasing in the change in rank. This suggests that firms who announce their good news early in the season get a bigger reaction to the firm s own earnings announcement. However, over the entire earnings season, this difference disappears implying that while the good news is impounded into the CAR EAS, the change in quartile rank does not matter for the overall reaction. The change in quartile ranks does not appear to affect the reaction to bad news in either the short or long windows. Finally, we provide evidence on the effect of setting the initial tone for the rest of earnings season within an industry. We construct Good Tone (Bad Tone) as indicator variables that takes the value 1 if the average earnings surprise of all firms announcing on the first day of the season is greater than or equal to (less than) zero. Conditional on the tone for the entire industry for a 22