LIMITED INVESTOR ATTENTION AND THE EARNINGS ANNOUNCEMENT PREMIUM

Transcription

1 The Pennsylvania State University The Graduate School Smeal College of Business LIMITED INVESTOR ATTENTION AND THE EARNINGS ANNOUNCEMENT PREMIUM A Dissertation in Business Administration by Kimball Chapman 2015 Kimball Chapman Submitted in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy August 2015

2 The dissertation of Kimball Chapman was reviewed and approved* by the following: Orie E. Barron Professor of Accounting Dissertation Advisor Chair of Committee Karl Muller Associate Professor of Accounting Jeremiah Green Assistant Professor of Accounting Vilmos Misangyi Associate Professor of Management & Organization Steven Huddart Professor of Accounting Head of the Department of Accounting *Signatures are on file in the Graduate School

3 iii ABSTRACT This paper explores the extent to which limited investor attention explains positive average stock returns around earnings announcements. I observe positive abnormal returns on days when firms announce the date earnings will be released (earnings notification days) and lower returns around earnings announcements when firms begin providing earnings notifications. I find a similar effect for other highlyvisible news events occurring soon before earnings announcements and higher returns around earnings announcements and when retail investors are more actively acquiring information about the firm. My results suggest that the attention-grabbing effect of earnings announcements provides a partial explanation of positive average returns around earnings announcements.

4 iv TABLE OF CONTENTS List of Figures.... v List of Tables... vi Acknowledgements... vii Chapter 1 Introduction... 1 Chapter 2 Background and hypothesis development... 6 The earnings announcement premium... 6 Limited investor attention... 7 Limited investor attention and investor trading decisions... 8 The passage of time between attention-grabbing events... 9 Earnings notifications Hypothesis development Chapter 3 Sample selection and descriptive data Sample construction Descriptive statistics Chapter 4 Research design & empirical results Abnormal trading patterns and Edgar searches on notification days Earnings notifications and the earnings announcement premium The EAP, Friday announcements, simultaneous earnings announcements and new investors The EAP and other types of news events Chapter 5 Additional considerations Sophisticated or unsophisticated investors Alternative explanations Chapter 6 Summary and Conclusion References Appendix: Examples of Earnings Notifications... 46

5 v LIST OF FIGURES Figure 3-1. Historical Frequency of Earnings Notifications Figure 3-2. Distribution of Notification Lead-time... 19

6 vi LIST OF TABLES Table 3-1. Descriptive Statistics Table 4-1. Abnormal Trading Volume, Returns and Edgar Searches Table 4-2. The Effect of Earnings Notifications Pre/Post Sample Table 4-3. The Effect of Earnings Notifications With Control Sample Table 4-4. Cross-Sectional Tests of the EAP Table 4-5. Effect of Non-Notification News Events on the EAP Table 4-6. The Effect of Earnings Notifications on Price Drift... 40

7 vii ACKNOWLEDGEMENTS I am indebted to my dissertation committee: Orie Barron (Chair), Karl Muller, Jeremiah Green and Vilmos Misangyi for their guidance and helpful suggestions. I thank workshop participants at Penn State University, Washington University, Brigham Young University, Boston College, Indiana University, the Massachusetts Institute of Technology, the University of Rochester, UC Irvine, the University of Pennsylvania, the University of Utah, and Virginia Polytechnic University for their helpful suggestions. I thank Andy Leone for posting Perl code used in analyzing data for this paper. I gratefully acknowledge financial support provided by the Smeal College of Business.

8 Chapter 1 Introduction 1 Merton (1987) observed that awareness of a firm is a necessary condition for trading in its securities and that becoming aware of a firm is costly for investors. Limited time and resources, or self-imposed constraints (such as a geographic or industry-specific investment focus) may prevent investors from becoming aware of firms in the first place. Subsequently, cognitive limitations may cause firms to fall out of the set of awareness if they are crowded-out, overlooked or forgotten. Some investors may be able to overcome constraints leading to limited awareness by hiring staff, subscribing to premium information services or otherwise leveraging human or technological tools to achieve consistent and universal awareness of a large set of firms; however, doing so is likely to be prohibitively costly for many investors. Barber and Odean (2008) argue that the effect of limited firm awareness (hereafter limited investor attention ) is to constrain the set of firms across which investors search when making investment decisions. Ideally, investors have access to the entire universe of stocks for potential investment opportunities, but in practice they only trade in a potentially much smaller set of firms of which they are aware. The purpose of this paper is to explore whether this form of limited investor attention explains positive average stock returns around earnings announcements. Prior research has found a predictable pattern of positive abnormal returns around earnings announcements. This phenomenon, known as the earnings announcement premium, is generally attributed to increased risk when new information is forthcoming (Ball and Kothari, 1991; Savor and Wilson, 2011; Barber, George, Lehavy and Trueman, 2013; Barth and So, 2014), or to transaction costs that limit the extent to which it can be arbitraged away (Cohen, Dey, Lys, Sunder, 2007). However, an alternative explanation based on limited investor attention is that positive average returns are the result of increased awareness caused by the earnings announcement. It is intuitive that earnings announcements increase awareness because firms are required to disseminate earnings information broadly and because these announcements are often the topic of attention-grabbing news headlines and analyst reports. Barber and Odean (2008) explain why increased firm awareness is likely to cause positive stock returns. When

9 deciding which stocks to buy, investors with limited resources are likely to limit their search to firms of 2 which they are already aware; otherwise they face a monumental task of searching across thousands of possible investment options. This is not the case when deciding which stocks to sell because resourceconstrained investors typically own only a small subset of the entire universe of stocks and rarely engage in short-selling. Thus investors are likely to rely more on attention triggers when buying stocks relative to selling stocks, which creates upward pressure on prices and positive average returns around attentiongrabbing events. It is important to note that attention-grabbing events do not necessarily cause investors to trade, but they help define the search criteria for investors searching for stocks to buy. The eventual purchase decision will vary with individual investor preferences. 1 In order to explore the role of limited investor attention on stock prices around earnings announcements, I use a research setting of earnings notifications, which are short announcements of the date and time that the earnings announcement will occur. 2 Earnings notifications provide three helpful characteristics for studying the effects of limited investor attention on stock returns around earnings announcements. First, like earnings announcements, earnings notifications are broadly disseminated and highly visible news events and are therefore likely to increase investor awareness of the firm. 3 Second, earnings notifications contain no explicit information about earnings or about firm performance they simply disclose the date and time of the earnings announcement. This lack of explicit information about firm performance allows me to distinguish the effect of increased awareness from the effect of new information (that is disclosed simultaneously in the case of earnings announcements but not in the case of earnings notifications). Lastly, earnings notifications are a convenient research setting because with the adoption of Regulation Fair Disclosure (hereafter Reg FD ), earnings notifications became mandatory 1 Some subset of investors may trade purely in response to an attention trigger. My results do not attempt to distinguish these traders from rationally-motivated traders with awareness constraints. The point here is that my results are consistent with rational updating in the presence of attention constraints. 2 See Appendix 1 for examples of earnings notifications. 3 The magnitude of the awareness-enhancing effect of earnings notifications is likely to be smaller than that of earnings announcements because earnings notifications are less informative and the subject of secondary news articles less often.

10 for firms hosting earnings conference calls 4, thus the period of time around the adoption of Reg FD 3 provides a setting in which earnings notifications are likely to be exogenous because conference calls were a common practice for many firms before the adoption of Reg FD and because the vast majority of conference call firms continued hosting conference calls after the adoption of Reg FD (Tasker, 1998; Frankel, Johnson and Skinner, 1999; Bushee, Matsumoto and Miller, 2004). To the extent that positive average stock returns result from increased awareness rather than information risk, I expect positive average returns around earnings notifications (in spite of their lack of explicit information about firm performance). My findings are consistent with this prediction. I find that abnormal returns increase approximately 10 basis points on notification days. Similarly, abnormal trading volume increases 20%, and investor searches on the SEC s Edgar website increase over 40% on earnings notification days. I find that these patterns are muted for notifications provided on Fridays when overall investor attention is likely to be lower, but that they otherwise persist in spite of no significant change in stock option implied volatility on notification days and after controlling for implicit information contained in earnings notifications (such as whether earnings will be earlier or later than expected). Next, I analyze average price reactions around earnings announcements when firms begin providing earnings notifications. I predict that the awareness-increasing effect of earnings announcements is attenuated in the presence of an earnings notification, implying that the magnitude of stock returns around earnings announcements should be lower when firms provide earnings notifications. My findings are consistent with this prediction - I find that the earnings announcement premium decreases when firms begin providing earnings notifications. I also find a similar reduction in the earnings announcement premium for other types of highly visible news events in the weeks leading up to earnings announcements, suggesting that my results are generally applicable to attention-grabbing news events before earnings announcements and not the unique characteristics of earnings notifications. 4 Guidelines provided by the SEC state that firms must provide advance notice and access to earnings conference calls and that such notice must be disseminated broadly to the public: Figure 1 illustrates a significant increase in the frequency of earnings notifications around the adoption of Reg FD.

11 4 As additional evidence, I test the prediction in Barber and Odean (2008) that the asymmetric trading response by attention-constrained investors to attention-grabbing events is driven primarily by retail investors who are not current owners of the firm. Consistent with this prediction, I find that the earnings announcement premium is larger when the proportion of Edgar searches from new investors is high. I also conduct cross-sectional tests of the earnings announcement premium under conditions that prior literature finds to be associated with reduced levels of investor attention. Consistent with reduced attention on Fridays (DellaVigna and Pollet, 2009) and when investors are distracted by multiple earnings announcements on the same day (Hirshleifer, Lim, and Teoh, 2009), I find that the earnings announcement premium is lower for Friday announcements and when there are multiple same-day earnings announcements. Lastly, I do not find changes in pre or post earnings announcement drift in the presence of earnings notifications, which I interpret as evidence that my results are not likely to be explained by earnings notifications alleviating limited attention by sophisticated investors. I use empirical specifications with firm fixed effects and a difference-in-difference design with a control sample of firms not subject to rules requiring earnings notifications. My results are robust to controls for changes in risk and other factors likely to explain returns around earnings announcements. I address several alternative explanations of my results including (1) the possible effects of opening conference calls to the public, (2) news events that are simultaneous with earnings notifications, and (3) changes in the information environment associated with the adoption of Reg FD 5. This paper contributes to prior research by distinguishing the effect of new information disclosed in earnings announcements (and the related risk premium demanded by investors) from the awarenessenhancing nature of the announcement itself. While prior research documents higher returns when new information is forthcoming (Kalay and Loewenstein, 1986; Ball and Kothari, 1991; Savor and Wilson, 2011) and higher returns around attention-grabbing events (Lee 1992; Hirshleifer, J. Myers, L. Myers, 5 Firms not providing earnings conference calls are not subject to rules requiring earnings notifications. I use these firms as a control sample. Although earnings notifications contain no explicit information about earnings or firm performance, the timing of when earnings is announced has been shown to be informative (Chambers and Penman, 1984; Bagnoli, Kross, and Watts, 2002). I control for this form of implicit information in several ways.

12 Teoh, 2008), prior literature has largely overlooked the possibility that both factors could be relevant in 5 explaining the earnings announcement premium.

13 Chapter 2 Background and hypothesis development 6 The earnings announcement premium The earnings announcement premium is a long-standing and economically significant stock price anomaly that is defined as positive abnormal returns around earnings announcements. Prior research suggests that it is caused by risk arising from the disclosure of new information in earnings announcements. Chari, Jagannathan, and Ofer (1988) find that the seasonality of stock returns is consistent with risk premia increasing around earnings announcements. Ball and Kothari (1991) find that the earnings announcement premium is partially explained by changes in firm-specific risk around earnings announcements. Barber, George, Lehavy and Trueman (2013) find evidence of the earnings announcement premium in a global context consistent with higher firm-specific risk around earnings announcements. Barth and So (2014) provide evidence that investors are exposed to non-diversifiable volatility risk around earnings announcements. 6 Alternatively, the possible role of limited investor attention in explaining the earnings announcement premium has been largely overlooked in prior research. Two notable exceptions are Frazzini and Lamont (2007), who observe that firms with larger earnings announcement premia tend to have higher trading volume, which they interpret as evidence of a link between attention-motivated trading and the earnings announcement premium and, Aboody, Lehavy, and Truman (2010), who find that firms with high prior stock returns (and are therefore more likely to be attention-grabbing stocks) have positive abnormal returns and subsequent stock price reversals around earnings announcements. 7 6 In related work, Johnson and So (2014) show that when earnings dates are correctly identified, much of the positive abnormal returns observed on the day of the earnings announcement actually occur the previous day. This finding does not necessarily contradict my results because earnings announcements dates are highly predictable and the expectation of an announcement may create attention in a way similar to the announcement itself. 7 My approach differs from these papers by using an identification strategy (earnings notifications) that cleanly distinguishes between changes in information and changes in investor attention around earnings announcements.

14 Limited investor attention 7 Prior literature describes two related but distinct manifestations of limited investor attention. Under the first, investors face attention constraints that inhibit or delay the processing of new information. Such constraints may be self-imposed (as in the case of an under-reaction to information revealed on Fridays, consistent with DellaVigna and Pollet, 2009) or simply the result of information overload (as in the case of an under-reaction to information revealed at times when a lot of new information is forthcoming, consistent with Hirshleifer, Lim, and Teoh, 2009). Stated simply, these papers show that investors must be paying attention to new information for that information to be incorporated into stock prices. A second manifestation of limited investor attention (which provides the theoretical motivation for this paper) is that particular investors may be simply unaware of entire sets of firms. As stated concisely by Merton (1987), If an investor does not follow a particular firm, then an earnings or other specific announcement about that firm is not likely to cause that investor to take a position in the firm. Merton (1987) argues that becoming aware of a firm is costly for investors, and the extent to which investors are aware of a firm changes the effective size of the market for shares of the firm. While generally speaking all investors everywhere may trade the shares of a firm, in practice only those investors who are aware of the firm actually trade its shares. This manifestation of limited investor attention implies a lack of universal, simultaneous awareness among investors of all available stocks. Prior research documents two ways in which this form of limited investor attention affects stock prices. First, Merton (1987) shows that when fewer investors are aware of a firm, its value is lower in equilibrium because risk sharing is less efficient across a smaller set of possible investors for neglected stocks. Second, a lack of awareness of all available stocks affects prices by changing how investors decide to buy or sell securities.

15 Limited investor attention and investor trading decisions 8 When attention constraints are binding, a lack of universal firm awareness will change how investors make trading decisions (in particular stock purchase decisions) because investors will be more likely to rely on some attention trigger to narrow the set of possible stocks to a more manageable size when searching for a stock to purchase. As articulated by Odean (1999), Unable to evaluate each security, investors are likely to consider purchasing securities to which their attention has been drawn. It is important to note that while some form of an attention trigger is likely to limit the set of consideration, the actual purchase decision is not necessarily caused by attention per se, but rather by individual investor preferences. In this sense, trading decisions made under binding attention constraints may be consistent with rational updating after a firm has been added to a smaller subset of possible investments because of an attention trigger. It is important to note that when making the decision of which stocks to sell, investors face a much easier search process because individual investors typically own a very small subset of the entire universe of stocks and because individual investors usually avoid short selling (Barber and Odean, 2008). Thus, with relatively few stocks to choose among when making the decision of which stocks to sell, investors have little need to rely upon attention triggers to narrow the search set. Evidence in prior literature suggests that this asymmetric role of attention triggers in trading decisions (with investors relying more upon attention triggers for stock purchases than for stock sales) causes positive average price changes around attention-grabbing events because attention-constrained investors are likely to be net buyers of attention-grabbing stocks. Similarly, Lee (1992) observes an anomalous proclivity of small investors to buy stocks around significant news events, regardless of the direction of the news. Similarly, Hirshleifer, Myers, Myers, Teoh (2008) show that retail investors are net buyers of both positive and negative extreme earnings stocks around earnings announcements. Given the evidence in prior literature that limited investor attention may cause positive average price reactions to attention grabbing events, it seems plausible that the same mechanism may explain

16 positive average returns around earnings announcements. Relative to other types of events, earnings 9 announcements are likely to be attention-grabbing because they are information-rich disclosures and because firms provide earnings announcements in highly-visible ways, such as with press releases and conference calls. Thus, under the limited attention hypothesis, the earnings announcement premium may be caused as the attention-grabbing effect of earnings announcements narrows the search for investors looking for stocks to purchase without similarly narrowing the search criteria for investors looking to sell stocks. This may create a positive average price effect independent of any risk premium associated with the disclosure of new information. The passage of time between attention-grabbing events A central argument in Barber and Odean (2008) is that investors with attention constraints will not actively pay attention to stocks that they do not own. Thus when a stock grabs the attention of an investor who is not a current owner of the firm, that investor is likely to revert back to not actively monitoring that firm if they do not purchase the stock in response to the attention-grabbing event. Therefore, the response to recurring attention-grabbing events should be consistent with the response to a single attentiongrabbing event because attention-grabbing events do not necessarily cause investors to trade (although for attention-constrained investors, attention is a necessary condition for trading). However, the magnitude of the attention-motivated response is likely to be different for recurring events. As the length of time between attention-grabbing events increases, conditions at the firm will have changed such that attention-constrained investors will have different opinions about the firm when their attention is focused on the firm. These differential opinions are likely to trigger more trading, which may increase the magnitude of the asymmetric response under Barber and Odean (2008). In contrast, when attention-grabbing events recur over a relatively short period of time, the magnitude of attentionmotivated trading is likely to be muted because little has changed at the firm and investors opinions are

17 less likely to have shifted since a broader set of investors last paid attention. A muted response to a 10 subsequent attention-grabbing event is consistent with earnings notifications preempting the attentionmotivated trading that would otherwise occur around earnings announcements because earnings notifications occur 12 days (on average) before earnings announcements. Earnings notifications Under Reg FD, firms are required to provide advance notice and public access to quarterly earnings conference calls. Guidelines provided by the SEC state that the advance notice must include the date, time, subject matter and call-in information for the conference call, and be disseminated broadly to the public. 8 Firms provide this information in a press release distributed through newswire services. I label these announcements earnings notifications to distinguish them from the actual earnings announcement and from earnings pre-announcements, which differ from earnings notifications because preannouncements contain earnings forecasts and because pre-announcements are provided voluntarily (while notifications are mandatory for conference call firms). While earnings notifications have received relatively little attention in prior literature, So (2014) shows that firm-initiated revisions of expected earnings dates (via earnings notifications) are informative signals about future firm performance and that investors fail to fully unravel these signals in a timely manner. Boulland and Dessaint (2014) provide evidence consistent with increased investor attention when there is more time between the notification date and the earnings date and that managers use this attention effect in strategic communication decisions. 9 8 See the SEC s Reg FD compliance and disclosure interpretations at: 9 Their focus is on the effects of within-firm variation in notification lead-time. I focus on the effect of the presence of earnings notifications using the natural experiment of Reg FD when notifications were exogenously mandated for conference call firms.

18 My approach assumes earnings notifications reach investors with attention constraints. While 11 newswire services are paid by firms to distribute earnings notifications to a broad set of local, regional and national media outlets, those media outlets are under no obligation to disseminate the earnings notification to their audience once they have received it from the news wire service. One limitation in my data is that I only observe when the notification has been sent from the news wire service and not whether each individual local, regional or national media outlet has chosen to disseminate the notification to their audience. However, firms usage of newswire services suggests that at least some firms believe this distribution method is effective at reaching individual investors as evidenced by the fact that this is the method that firms adopted when required by the SEC to disseminate the earnings notifications broadly. Moreover, to the extent that individual media outlets do not choose to disseminate earnings notifications, this will most likely bias against my predicted results because I assume every notification carried by the newswire service is noticed by investors and I predict an associated response. Hypothesis development While the primary purpose of this research is to analyze the effect of limited investor attention on the earnings announcement premium, I start by analyzing the market reaction to earnings notifications. By so doing, I analyze whether the abnormal trading patterns and Edgar searches observed on notification days are attributable to limited investor attention. This analysis provides a necessary foundation for subsequent tests of the effect of earnings notifications on the earnings announcement premium because subsequent tests assume that earnings notifications increase investor attention toward the firm in the period before earnings are announced. Although earnings notifications do not contain explicit information about firm performance, it is possible that they are informative because they contain implicit information. For example, earnings notifications may provide implicit information about firm performance by indicating whether the

19 upcoming earnings announcement date is earlier or later than expected (Givoly and Palmon, 1982; 12 Chambers and Penman, 1984; Begley and Fisher, 1998; Bagnoli, Kross, and Watts, 2002; Kothari, Shu, Wysocki, 2009; So, 2014). A second possible form of implicit information contained in earnings notifications is the resolution of uncertainty about upcoming earnings dates (independent of whether earnings will arrive earlier or later than expected), particularly for those firms whose earnings dates are inconsistent. 10 My first hypothesis predicts that attention-motivated behavior offers the best explanation for trading patterns and Edgar searches on notification days. Stated in the alternative form: H1: Abnormal trading volumes, returns and Edgar searches are higher on notification days. Next, I consider the effect of earnings notifications on the earnings announcement premium. With the passage of time, it is likely that there will be a gradual shift in the mix of investors with attention constraints that would purchase a stock if they were paying attention to it. Thus, when more time has passed between attention grabbing events, it is likely that there will be a larger attention-motivated response because there is a larger proportion of investors who would be inclined to purchase the stock once it grabs their attention. Empirical evidence on how investors respond when stock price levels cross highly-visible thresholds is consistent with the prediction that the passage of time magnifies the response to attention grabbing events. Huddart, Lang, and Yetman (2009) find that abnormal volume when stock prices achieve a new 52 week high is greater when more time has passed since the most recent 52 week high price was achieved. 11 Consistent with this evidence, I predict that earnings notifications provided soon before 10 A final possible cause of trading patterns on earnings notification days is informed trade because a highly visible reminder that the earnings announcement is forthcoming may increase the incentive for informed investors to trade on their private information about future earnings. Such a trade may be easier to disguise or less costly to execute on notification days because of higher trading volumes. I control for this possibility in my tests of the market reaction to earnings notifications. 11 Two notable differences between my setting and the setting used by Huddart, Lang and Yetman (2009) potentially limit the extent to which their findings offer predictions in my setting. First, the time scales differ, with my setting focuses on weeks and months (the length of time between earnings announcements) while theirs encompasses longer periods of time. Second, in Huddart, Lang and Yetman (2009), investors respond to the same attention-grabbing

20 earnings are announced will reduce the magnitude of the attention-motivated response to the earnings 13 announcement, thereby reducing the earnings announcement premium. This discussion leads to my second hypothesis, stated in the alternative form: H2: The earnings announcement premium is reduced in the presence of prior earnings notifications. Next, I consider other evidence that the earnings announcement premium is attention-driven. Prior research identifies several settings in which investors have demonstrated reduced attention. DellaVigna and Pollet (2009) find that investors under-react to Friday earnings announcements. To the extent that the earnings announcement premium is attributable to attention-motivated trading, this evidence would predict a lower earnings announcement premium for Friday announcers. Similarly, Hirshleifer, Lim, and Teoh (2009) identify a similar effect when there are a large number of same-day earnings announcements. To the extent that the earnings announcement premium is attention-driven, this result would also predict a lower earnings announcement premium on days with simultaneous earnings announcements. Prior research also suggests that attention-motivated stock purchases on highly visible news events are likely to be attributable to new retail investors (Barber and Odean, 2008). Thus, I expect a positive relationship between the earnings announcement premium and the proportion of new investors conducting Edgar searches around earnings announcements. This logic leads to my third hypothesis, which consists of the following predictions, stated in the alternative form; H3a: The earnings announcement premium is lower for earnings announcements made on Fridays. H3b: The earnings announcement premium is lower when there are a large number of same-day earnings announcements. H3c: There is a positive relationship between the proportion of Edgar searches conducted by new investors and the earnings announcement premium. event (an extreme price level), while in my setting investors are responding to two different events (earnings notifications and earnings announcements).

21 To the extent that earnings notifications attract investor attention and reduce attention-motivated 14 trading around earnings announcements, I would expect a similar effect for other types of highly visible news events. This type of evidence would suggest that the attenuation effect of a previous attentiongrabbing event on a subsequent event is a general phenomenon and not attributable to the unique characteristics of earnings notifications. This discussion leads to my fourth hypothesis, stated in the alternative form; H4: The earnings announcement premium is lower when highly-visible news events occur soon before earnings announcements.

22 Chapter 3 Sample selection and descriptive data 15 Sample construction Earnings notifications are identified from headlines of firm-initiated press releases collected from PR Newsire, MarketWire, BusinessWire, and Globe Newswire using the Factset news search application 12. Consistent with Bushee and Miller (2012) and Core, Guay, and Larcker (2008), I assume that press release articles carried on news wires are firm-initiated disclosures. Data regarding daily stock return and trading volumes, quarterly firm characteristics and accounting information, analyst coverage and news events are collected from the Center for Research in Security Prices (CRSP), the Compustat quarterly file, the I/B/E/S detailed recommendation database and the RavenPack events database, respectively. Notification-day Edgar searches are collected from Edgar server logs provided by the SEC through a Freedom of Information Act request. 13 Measures of the frequency of firm disclosures are collected from the SEC s Edgar website. Earnings notifications are distinguished from other press releases through an analysis of the title and date of the press release. To be classified as an earnings notification, the title of the press release must include language indicating an upcoming earnings release or an upcoming conference call and the date of the press release must be between 2 and 21 days before the earnings announcement. 14 Notifications with a lead-time of one day or less are omitted to avoid an overlap with the earnings announcement period. Notifications with a lead-time of more than 21 days are omitted because these are most likely to be 12 The UBM annual market survey of newswire services lists these providers as market leaders. A search of PRWeb for earnings notifications provided zero results. See: 13 I follow several recent papers that use SEC server log data acquired through a Freedom of Information Act Request. Drake, Roulstone, and Thornock (2013) analyze the determinants and consequences of investors searches on the SEC s Edgar website. I follow their approach for excluding bots from server log data. Lee, Ma and Wang (2014) use the same data to propose an improved methodology for identifying competitive firms based on investor search activity. 14 The search string used in the Factset news application is ((hd="conference call" or hd="results call" or hd="webcast" or hd="earnings call") and (hd="earnings" or hd="results" or hd="quarter" or hd="fiscal" or hd="year") or (hd="to announce" or hd="to report" or hd="to release" or hd="date" or hd="timing") and (hd="earnings" or hd="results" or hd="quarter" or hd="fiscal" or hd="year")).

23 erroneous observations. 15 Observations with titles consistent with bundling of earnings notifications 16 with other information about the firm are also excluded in order to remove observations with explicit information about firm performance. 16 Using this process, 179,507 earnings notifications are collected between the years 1998 and 2012 (inclusive). Earnings notifications are matched to CRSP and Compustat using firm identifiers (ticker and cusip) provided by Factset and a proximity test to earnings announcement dates in COMPUSTAT and I/B/E/S. 135,177 firm/quarter observations of earnings notifications are available after omitting observations that cannot be matched to CRSP, Compustat and IBES in order to collect necessary data. Tests of both notification and non-notification quarters use 386,600 firm/quarter observations available from Compustat between 1998 and 2012 where necessary data from CRSP, IBES and other data sources are also available. Descriptive statistics Table 1 provides descriptive statistics. Firms provide notifications in 29% of observations in my sample (Notify). This percentage understates the percentage of firms that provide notifications subsequent to the passage of Reg FD because my sample contains several years before notifications were mandated. The earnings announcement premium averages approximately 26 basis points across my sample (EAP) In reading a random sample of earnings notifications by hand, I observe that excluding notifications arriving more than 21 days before the earnings announcement are more likely to be erroneous observations for one of two reasons; (1) occasionally, the same word patterns I use to identify earnings notifications are other news events (these are more common outside the 21-day window because earnings notifications are more likely soon before earnings announcements), (2) earnings notifications dates are occasionally incorrect in Factset (meaning the date of the press release observed in Factset is inconsistent with the date provided by the newswire service when I manually confirm the dates). 16 Observations of earnings notifications bundled with other types of information are highly unusual in my sample. In reading through a sample of those that exist, I find no clear trend of either positive or negative information bundled with earnings notifications. 17 EAP is the cumulative abnormal return (adjusted by the value-weighted marker return) for each firm/quarter during the three-day window (-1, 0, +1) centered on the earnings announcement date identified by COMPUSTAT.

24 Table 3-1. Descriptive Statistics 17 Variable Mean 25% Median 75% n AbnormalVolume ,496 AbnormalReturn (0.013) (0.000) ,619 AbnormalSearches ,565 ImpliedVolatility ,419 DailyNews ,484 NotPriorYearReturn (0.256) (0.013) ,266 Notify ,820 StdAnnDelay ,202 AnnouncementDelayInc (0.414) (2.000) ,912 EarningSurprise (0.003) (0.008) ,563 EAP (3.819) (0.062) ,986 LnReturnVolatility (3.476) (4.303) (3.317) (2.450) 408,529 LnMveq ,820 MarketToBook ,380 PriorYearReturn (0.003) (0.351) (0.074) ,776 EarningsVolatility ,439 YearReturnVolatility ,658 NotifyDayNews ,820 NumAnalysts ,820 NumSecFilings ,911 InstOwnership ,076 FirmAge ,813 Forecasts ,820 FridayAnnouncement ,820 SimultaneousAnnouncements ,820 PropNewSearchers ,909 PressReleaseBeforeEarnings ,820 CAR [1,5] (0.003) (0.038) (0.004) ,062 CAR [1,15] (0.002) (0.062) (0.006) ,138 CAR [-5,-1] (0.025) (0.000) ,155 CAR [-10,-1] (0.037) (0.001) ,204 Figure 1-1 illustrates the historic frequency of earnings notifications, with a sharp and sudden increase visible around the year 2000 with the adoption of Reg FD. This period is the primary setting in which I test the effect of earnings notifications on the earnings announcement premium because firms that

25 Number of Notifications started providing earnings notifications at this time likely did so in response to the mandate under Reg 18 FD. Figure 2 illustrates when earnings notifications are provided relative to earnings announcements. On average, earnings notifications are provided 12 days before earnings announcements, consistent with the role of earnings notifications in providing advance notice and access to earnings conference calls. 20,000 18,000 16,000 14,000 12,000 10,000 8,000 6,000 4,000 2, Year Figure 3-1. Historical Frequency of Earnings Notifications. Figure 1-2 shows the distribution of notification lead-time for all notifications in the sample. Lead-time is defined as the number of days between the notification and the earnings announcement. Notifications with lead-times of less than 2 days or more than 21 days are excluded.

26 18% 19 16% 14% 12% 10% 8% 6% 4% 2% 0% Figure 3-2. Distribution of Notification Lead-time

27 Chapter 4 Research design & empirical results 20 Abnormal trading patterns and Edgar searches on notification days To test the extent to which abnormal trading volume, internet searches of firm disclosures on the SEC s Edgar website, and abnormal returns increase around notification days, I estimate the following OLS regression with year and industry fixed effects and standard errors clustered by day. (1) Where I use three proxies for abnormal activity: AbnormalVolume, AbnormalReturn and AbnormalSearches and where: AbnormalVolume Is the ratio of daily trading volume to average trading volume in the 10 trading-day window [-12,-2] relative to the earnings notification date. AbnormalReturn Is the daily return to the firm minus the value-weighted market return times 100. AbnormalSearches Is the ratio of daily Edgar searches to average searches in the 10 trading-day window [- 12,-2] relative to the earnings notification date. Notify Is an indicator variable equal to one for the notification date. Is the standard deviation of the announcement delay over the prior 8 quarters where StdAnnDelay announcement delay is defined as the number of trading days between the last day of the fiscal quarter and the earnings announcement date. AnnouncementDelayInc Is the increase in the announcement delay compared to the same quarter in the prior year. Is earnings before unusual items less earnings before unusual items in the same quarter of EarningSurprise the prior year scaled by the market value of equity for the firm at the end of the fiscal quarter. DailyNews Is an indicator variable equal to one if there was a non-notification news event about the firm on the date. NotPriorYearReturn Is the cumulative abnormal return (benchmarked by the value-weighted return) over the previous year, measured at 5 days before the earnings notification date. LnMveq Is the log of the market value of equity, measured on the last day of the fiscal quarter. ImpliedVolatility Is the average implied volatility of call options on each trading day.

28 21 Equation 1 is estimated over a two-day period for each firm/quarter in my sample, including the day before the notification date and the notification date 18. Notify is my variable of interest and is equal to one for notification-day observations. I expect that abnormal trading and Edgar search patterns cannot be fully explained by implicit signals contained in earnings notifications, leading to a prediction of positive coefficient estimate on Notify ( ) where the dependent variable is AbnormalVolume, AbnormalReturn, or AbnormalSearches. I control for several potential forms of implicit information contained in earnings announcements. AnnouncementDelayInc measures the extent to which the earnings notification provides information about a delay in the upcoming earnings announcement. StdAnnDelay measures the extent to which the earnings notification resolves uncertainty about the date of the upcoming earnings announcement. Because abnormal trading and Edgar search patterns on earnings notifications days may also be attributable to informed traders who have been reminded that earnings is forthcoming, I control for unexpected earnings (EarningSurprise). I control for other factors likely to explain abnormal returns, volume, and Edgar searches on earnings notification days. DailyNews controls for the possibility that my results are driven by news events simultaneous or closely preceding notifications. PriorYearReturn controls for prior stock price momentum in explaining abnormal returns. LnMveq controls for firm size. Panel A of Table 4-1 reports the results of Equation 1 when estimated across the entire available sample. Consistent with my prediction, coefficient estimates for are positive and significant at conventional levels in the first three models (t-stats = 20.92, 8.18, for dependent variables AbnormalVolume, AbnormalReturn, and AbnormalSearches, respectively). The coefficient estimates for Notify suggest that trading volume, abnormal stock returns and Edgar searches are higher on earnings notification days. 18 Equation 1 is intended to test for abnormal stock return, volume and Edgar trading activity on notification days. The dependent variables are constructed to compare notification-day observations along these dimensions with the recent past. Equation 1 is estimated over a two-day window because volumes and Edgar searches tend to increase on the day after earnings notifications when the notification lead-time is short.

29 Table 4-1. Abnormal Trading Volume, Returns and Edgar Searches 22 PANEL A: Full Sample Abnormal Volume Dependent Variable Abnormal Returns Abnormal Edgar Searches Implied Volatility Intercept 1.428*** *** 1.036*** (13.76) (-0.91) (22.39) (42.70) Notify 0.064*** 0.094*** 0.233*** (20.92) (8.18) (27.38) (0.63) StdAnnDelay 0.001*** *** (4.69) (-1.49) (-0.11) (7.73) AnnouncementDelayInc (0.31) (-1.61) (-0.34) (-0.08) EarningSurprise *** *** (1.57) (6.15) (0.86) (-10.39) DailyNews 0.224*** 0.052** 0.186*** 0.025*** (31.40) (2.08) (12.69) (3.64) PriorYearReturn *** *** (-2.91) (1.02) (-0.35) (-11.61) LnMveq *** *** *** (-4.11) (-4.98) (-1.26) (-61.24) N 255, , ,167 19,766 Adj 1.1% 0.2% 0.9% 50.6% Fixed Effects Year, Industry Year, Industry Year, Industry Year, Industry SE Clustered by: Day Day Day Day

30 PANEL B: Sample partitioned by notifications on Fridays and non-fridays 23 Dependent Variables Abnormal Volume Abnormal Edgar Searches Abnormal Returns Fridays Non- Non- Non- Fridays Fridays Fridays Fridays Fridays Intercept 1.027*** 1.505*** *** 1.254*** (4.58) (12.46) (-1.05) (-0.16) (7.19) (10.64) Notify 0.043*** 0.068*** *** *** (3.64) (16.65) (1.09) (7.98) (-0.97) (24.01) StdAnnDelay 0.001** 0.001*** ** (1.82) (4.44) (-2.08) (-0.65) (-0.43) (-0.04) AnnouncementDelayInc * ** 0.001* (-1.34) (0.95) (-0.38) (-1.72) (1.57) (-1.10) EarningSurprise 0.066* *** 0.696*** (1.48) (1.02) (3.05) (5.41) (-0.75) (1.17) DailyNews 0.227*** 0.223*** *** 0.207*** 0.179*** (12.90) (29.16) (-0.74) (2.65) (6.87) (10.89) PriorYearReturn *** * * (-4.13) (-1.46) (-0.06) (1.14) (-1.31) (0.08) LnMveq 0.004* *** * *** *** (1.48) (-5.12) (-1.55) (-4.88) (-2.57) (-0.68) N 45, ,881 45, ,215 26, ,459 Adj 1.1% 1.1% 0.3% 0.1% 0.7% 1.3% P-value of test of difference in Notify on Friday vs. non- Friday Year, Year, Year, Year, Year, Fixed Effects Industry Industry Industry Industry Industry SE Clustered by: Day Day Day Day Day Day Year, Industry As an additional test of whether earnings notifications are informative in ways not captured with my control variables, I also estimate Equation 1 using implied stock option volatility as the dependent variable. Column four of Table 4-1 Panel A presents this analysis. I find no significant change in stock price implied volatility on earnings notification days (t-stat = 0.63). Not surprisingly, DailyNews is strongly associated with AbnormalVolume, AbnormalSearches, and ImpliedVolatility, consistent with news events causing trading, information acquisition and increasing uncertainty. The magnitude of the coefficient estimate on DailyNews is larger than that of the coefficient estimate on the Notify when AbnormalVolume is the dependent variable. One interpretation of this result

31 24 is that earnings notifications may be less attention-grabbing than other news events, at least with respect to the impact on trading volumes. The possibility that notifications are less attention-grabbing than other news events is not problematic in interpreting my results because it is not the magnitude of the attention effect that makes earnings notifications a good research setting, but rather the fact that they do not contain explicit earnings-relevant information. It may well be that other types of news events are more attentiongrabbing than earnings notifications, but identifying the effect of attention, distinct from the effect of new information, is much easier with earnings notifications relative to other types of news events. In Panel B of Table 4-1 I interact Notify with Friday, which is an indicator variable equal to one if the notification occurred on a Friday and zero for any other day. To the extent that the market reaction to earnings notifications is primarily attention-driven, I expect a reduced market reaction on Fridays, consistent with prior literature suggesting lower levels of investor attention on Fridays (DellaVigna and Pollet, 2009). Results in Panel B of Table 4-1 are consistent with this prediction. For Friday notifications, coefficient estimates for are significant only for abnormal volume (t-stats = 3.64, 1.09, for dependent variables AbnormalVolume, AbnormalReturn, and AbnormalSearches, respectively). I interpret these findings as evidence that the market reaction to earnings notifications reduced when investors are less attentive. The finding of positive abnormal returns on earnings notification days raises the question of whether there is a subsequent price reversal. Arguments can be made both for and against an expectation of a subsequent price reversal. On one hand, prior findings of stock returns around attention-grabbing events leads to an expectation of a price reversal because the initial positive average price reaction is caused by investors overreacting to the event. This is consistent with Aboody, Lehavy, and Truman (2010), and Engelberg, Sasseville and Williams (2012), both of whom document subsequent price reversals after initial positive returns around attention-grabbing events. I describe these findings of price reversals as consistent with a behavioral mechanism driving the initial positive average price reaction; in both cases the authors describe the initial price reaction as an over-reaction to the attention-grabbing event.