Corporate Use of Social Media

Transcription

1 Corporate Use of Social Media Michael J. Jung,* James P. Naughton, Ahmed Tahoun, and Clare Wang January 2014 Abstract We examine corporate adoption of social media and provide the first large-sample evidence on the determinants and market consequences of the decision to disseminate quarterly earnings news through social media. We find that social media usage for earnings news is distinct from other forms of voluntary disclosure and document a number of interesting attributes of this disclosure mechanism. Social media usage for earnings news is inversely related to the number of social media followers, suggesting that firms with large social media followings are hesitant to use social media for financial information. However, we find that earnings news is more likely to be communicated when the news is positive, suggesting that some firms are opportunistic in their use of social media. Moreover, when we examine the market response to social media communications, we find that trading volume increases and that the primary driver is increases in large rather than small trades. This is inconsistent with the notion that social media primarily benefits small investors. Lastly, we find that the market reaction is stronger for firms that follow a consistent rather than ad hoc social media disclosure policy. Keywords: Commitment to Disclosure, Voluntary Disclosure, Social Media, Facebook, Twitter * Corresponding author. Leonard N. Stern School of Business, New York University, 44 West 4th St., New York, NY 10012, , mjung@stern.nyu.edu; Northwestern University; London Business School. Acknowledgements. We thank workshop participants at New York University and the 2013 UNC/Duke Fall Camp for their comments and suggestions. Naughton and Wang are grateful for the funding of this research by The Kellogg School of Management and the Lawrence Revsine Research Fellowship. We thank Blake Disiere, Sam Faycurry, Michael Licata, Stacy Ni, Alice Yujia Qiu, Borui Xiao, Melody Xu, Rena Xin Xu, David Zeyu Wang and Martin Ying for providing excellent research assistance.

2 1. Introduction Social media has transformed communications in many sectors of the U.S. economy. It is now used for disaster preparation and emergency response (FEMA, 2013), security at major events (PERF, 2011), and public agencies are researching new uses in geolocation (DARPA, 2010), law enforcement, 1 court decisions, 2 and military intelligence (DARPA, 2011). Internationally, social media is credited for organizing political protests across the Middle East (Stone and Cohen, 2009) and a revolution in Egypt (WSJ, 2011; Vargas, 2012). In the business world, social media is commonly considered a revolutionary sales and marketing platform (Forbes, 2013; HBR, 2010; Larcker et al., 2012) and a powerful recruiting and networking channel (Li, 2013). In contrast, there is little to no research on how many firms use social media to communicate financial information to investors and how investors process information provided through social media channels. 3 This omission is notable because, ex ante, it is not clear why a company would adopt a social media platform, nor how investors would respond to financial information disseminated through social media. Furthermore, the need for research is underscored by the recent SEC announcement that firms may use social media to announce key financial information in compliance with Regulation Fair Disclosure (SEC, 2013). Motivated by this omission, we examine a broad set of questions that provide insight into social media as a voluntary disclosure mechanism. More specifically, we evaluate the determinants of and the market reaction to social media usage for reporting quarterly earnings. 1 The State Department sponsored a simulated law enforcement search called the Tag Challenge in March 2012 ( to find five suspects in five cities across North America and Europe using social media. 2 US courts use information posted on social media to determine the appropriate time in jail required for the case ( 3 One exception is Blankespoor et al (2013), who examine the use of social media by 85 small technology firms. 1

3 Our initial analysis focuses on both Facebook and Twitter, as they are the two most prevalent social media platforms. We construct a dataset on the use of both platforms by S&P1500 firms for 2010 through the first quarter of Our dataset identifies which firms have a social media presence, the size of the social media audience in terms of likes and followers, and when the firms used social media to disseminate quarterly earnings news. We first document that, as of July 2013, 47% of firms have adopted a corporate Twitter account and 44% have a Facebook page. However, only about half of the firms with a Twitter account and about a third of the firms with a Facebook page have ever tweeted or posted information about quarterly earnings announcements. 4 Given the choice between the two social media platforms, firms have a stronger preference for Twitter of the firms that disseminate earnings news via social media, 91% use Twitter and 52% use Facebook. Our analysis of the determinants of Twitter adoption reveals that larger firms are more likely to have corporate Twitter accounts and are more likely to use Twitter to disseminate earnings news, contrary to the notion that smaller firms benefit more from using social media (Blankespoor et al., 2013). Surprisingly, we also find that firms with a larger social media audience are less likely to use Twitter for earnings news, which we believe reflects the fact that such firms tend to be retail firms with millions of followers who are primarily customers rather than investors (e.g. McDonald s Corp). We find consistent but statistically weaker results for Facebook. We do not find that factors such as firm performance, growth and leverage, which have been shown to be related to firms traditional disclosure outlets (e.g., conference calls, press 4 Terminology differs slightly between Facebook and Twitter. Firms post information to their Facebook page but tweet information over their Twitter accounts. For brevity, we occasionally use the terms interchangeably. 2

4 releases, etc.), are significant predictors of Twitter (or Facebook) adoption for financial reporting purposes. 5 Drawing on the voluntary disclosure literature, we also consider whether the use of social media to disseminate earnings is partially driven by the direction of the earnings news. We find evidence that firms are more likely to disseminate earnings news through Twitter when the news is good, contrary to the small sample evidence in Blankespoor et al. (2013). 6 We also find that there is significant variation in the consistency (or frequency) of earnings tweets, with about 16% of firms tweeting earnings almost every quarter and 40% of firms doing so rarely. We do not find similar results for Facebook, partially because our Facebook specifications have low power due to the limited number of firms that use Facebook exclusively for reporting quarterly earnings. Overall, these findings suggests that while some firms have made a strong commitment to use social media to disclose or highlight earnings news every quarter, and thus meet the spirit of the SEC s recent announcement, 7 other firms may be more opportunistic in their decision to highlight earnings news on social media. When we examine the capital market consequences of social media usage, we conduct event study tests using daily and intra-day trading data. Using several different market-based measures including abnormal returns, absolute abnormal returns, abnormal volume, abnormal 5 It is beyond the scope of this paper to examine the determinants of general corporate social media usage. Such a model likely requires numerous non-accounting factors such as marketing strategies, social responsibility initiatives, online sales, employee relations, and managers preferences and characteristics. For example, in an interview with a mid-sized commercial bank (which is one of the firms in our sample), managers from the public relations, investor relations, and legal department stated that the bank adopted Facebook and Twitter because its customers, partners, and employees in the field were avid users of social media and preferred that method of communication. This factor would not be captured by traditional accounting measures. 6 Throughout this paper, we compare and contrast several of our results to those found in Blankespoor et al. (2013) because it was one of the first studies to examine the firms use of social media. Any noted differences are likely due to vastly different sample sizes (1500 vs. 85 firms), number of industries, and time periods examined in each study. 7 The Securities and Exchange Commission issued a report on April 2, 2013 that makes clear that companies can use social media outlets like Facebook and Twitter to announce key information in compliance with Regulation Fair Disclosure (Regulation FD) so long as investors have been alerted about which social media will be used to disseminate such information. See SEC release

5 bid-ask spreads, and average trade sizes, we highlight several findings. First, we find corroborating evidence that firms tend to disseminate earnings news over social media when the news is good, as the three-day signed returns are higher and the absolute returns are lower for the quarterly earnings announcements that are disseminated over social media. Second, the absolute market reactions are higher for firms with larger social media audiences, consistent with the notion that a firm s followers do include investors and other capital market participants. Third, in contrast to the early evidence in Blankespoor et al. (2013), we find that the three-day average bid-ask spread is actually higher for firms that announce earnings news via social media, suggesting that such communications increase information asymmetry. For the firms in our sample that tweet during market hours, we use intra-day trading data and hand-collected time stamps of tweets to examine the market reaction associated with three different types of earnings-related tweets: earnings announcement ( EA ), preview and rehash. We categorize a tweet as an EA tweet if it is the first tweet mentioning the firm s earnings announcement and it occurred on the earnings announcement date. A tweet is categorized as a preview tweet if it only mentions the date of the upcoming earnings announcement and a rehash tweet if it mentions highlights from the prior earnings announcement. 8 On average, EA tweets occur several hours after the earnings announcement 9, preview tweets occur 13 days before the earnings announcement date and rehash tweets occur 1 to 2 days after the earnings announcement date. Our intra-day analyses reveal that trading volume increases in response to EA tweets, but not for rehash tweets. We also find higher trading volume following a preview tweet, which is 8 The fact that we routinely observe three types of earnings-related tweets suggests that Twitter, as a disclosure channel, provides unique features relative to other channels. For example, it is less common to observe firms issuing press releases reminding investors of upcoming earnings announcements or rehashing prior earnings highlights. 9 The exact time of the earnings announcement is based on data from I/B/E/S. 4

6 surprising considering a preview tweet only contains a mere reminder of an upcoming earnings announcement. 10 When we partition our data into large and small trades, we find that the primary driver of the increased trading volume at the time of the tweet is larger trades. We find corroborating evidence from examining average trade sizes, which increases during the time of the EA tweets. Therefore, while social media is commonly viewed as a disclosure channel that provides timely access to information for all investors, and thus levels the playing field for small investors, our results suggest that larger investors react quicker to earnings-related tweets. In our final set of analyses, we identify firms that use social media for earnings news consistently. We classify this group of approximately 100 firms as committers because once they post earnings news to social media for a particular quarter, they do so again in all subsequent quarters. Our intention is to examine if market reactions differ significantly for committers versus firms that use social media on an ad hoc basis. While we do not find differences across all market reaction variables, we do find that abnormal turnover is higher in the three-day earnings announcement window for committers that post earnings to Facebook and that absolute abnormal returns are higher for committers that tweet earnings over Twitter, relative to the market reactions for uncommitted firms. These results provide some evidence that a commitment to social media usage for earnings news is associated with a greater market reaction. The findings presented in this paper are relevant to firms that have adopted social media, or are considering adoption, and to regulators debating the costs and benefits of firms' use of 10 We conjecture that this increase in volume could be due to investors interpreting a preview tweet as a positive sign for the upcoming earnings announcement. In untabulated tests, we do find that the mean three-day abnormal return is higher for earnings announcements that were previewed versus not previewed (0.4% vs. 0.2%), but the difference is not statistically significant. We also find that the mean abnormal return for earnings announcements that were rehashed is significantly higher than earnings announcements that were not rehashed (1.6% vs. 0.1%), indicating that firms tend to rehash only good earnings news several days after the earnings announcement. 5

7 social media for capital market communications. We document that: 1) adoption of the two most prevalent social media platforms by the largest corporations has exceeded 50%, 2) the propensity to disseminate financial news over social media depends on the direction of the news, 3) there is a market reaction to earnings tweets that is separate from the market reaction to the actual earnings news, and 4) the market reaction differs for firms that use social media on a consistent versus ad hoc basis. In addition, our study reveals how disclosure choices evolve during a time before the SEC approved the use of social media as an official disclosure venue (i.e., April 2013). This setting is interesting because, a priori, it is not clear to what extent firms and investors rely on disclosure venues not yet approved by the SEC, nor it is clear how the market reacts to such disclosures. We show that a subset of firms adopt early, make a commitment, and that the market reaction differs for these firms. Our findings extend prior studies that examined commitments to increased disclosure in international settings (e.g., Leuz and Verrecchia, 2000) by highlighting a setting within the U.S. where there remains cross-sectional variation in firms commitment to increased disclosure using the newest technology. The paper proceeds as follows. In Section 2, we review the literature and develop our hypotheses. In Section 3, we describe the construction of our database and summarize descriptive statistics. We outline the research design in Section 4, describe the empirical results in Section 5 and summarize our conclusions in Section Background and Hypothesis Development Social media provides a unique and revolutionary approach for firms to communicate directly with their investors and interested stakeholders. Conventionally, firms publicize 6

8 earnings announcements by sending a press release to newswire services, to equity research databases, and to individual brokerage firms and financial institutions (Frankel et al., 1999). In this manner, firms send the news once and do not know how many people receive the news. In contrast, social media allows a firm to send multiple messages over time directly to a known number of followers. Typically, this is accomplished by tweeting short messages and including a link to a complete news release found on the corporate website (Blankespoor et al., 2013). The result is a reduction in information dissemination costs, an increase in speed and flexibility for the news dissemination, and a reduction in information acquisition costs for the firm s investors. Despite the difference in how information is provided through social media compared with past alternatives, there is very little research on why some firms use social media to disseminate quarterly earnings news and whether this choice has capital market consequences. Blankespoor et al. (2013) find that firms decision to tweet earnings news is not dependent on the direction (good or bad) of the news, suggesting that firms are not opportunistic in this respect. They also find that that Twitter usage reduces information asymmetry for a sample of 85 technology firms, with the reductions in information asymmetry concentrated in the smaller, less visible firms. Both findings are relevant to the current policy debate on corporate social media usage, highlighted by the SEC s April 2013 statement (SEC, 2013) that firms may use Twitter and Facebook to announce key financial information in compliance with Regulation Fair Disclosure so long as investors have been alerted to look for such announcements. However, the lack of evidence for a broader sample of firms underscores the need for additional research. To develop a prediction about when firms disseminate earnings over social media, we draw from prior studies that have shown that news can be selectively disclosed to benefit either the firm or its managers. For example, Healy and Palepu (1995) find that firms are more likely to 7

9 disclose good news in advance of raising capital or engaging in merger activity. Conversely, Skinner (1994) suggests that firms voluntarily disclose bad news to mitigate litigation risk. Similarly, disclosure choices could affect stock compensation (Aboody and Kasznik, 2000) or the signalling of managerial talent (Trueman, 1986). While these studies have focused on various types of news, including earnings announcements, earnings pre-announcements, and management forecasts, the general finding is that the direction of the news could affect the decision to disseminate each quarter s earnings news via social media. Blankespoor et al. (2013) test this prediction but find that the direction of the earnings news does not affect Twitter usage for a small sample of technology firms. However, this result may not generalize to a large sample of firms; therefore, the first hypothesis we test is whether the decision to disseminate earnings news through social media is related to the direction of the news. H1: Firms decision to disseminate earnings news each quarter on social media is related to the direction of the news. Our remaining predictions focus on the market reactions of news disseminated versus not disseminated over social media. Knowing whether financial disclosures over social media provide information to capital market participants is fundamental to understanding the role of social media in the corporate disclosure process. In fact, understanding which firms provide such disclosures is not valuable if the disclosures themselves do not have capital markets consequences. Analytical work by Holthausen and Verrecchia (1990) demonstrate that returnand volume-based measures are equally valid measures of information content of financial disclosures. Accordingly, empirical studies such as Frankel et al. (1999) conclude that earnings conference calls are informative to stock market participants because of higher levels of return volatility and trading volume during the time of the conference call. Similarly, Bushee et al. 8

10 (2011) find that managerial presentations at investor conferences are informative because they are positively associated with abnormal absolute stock returns and abnormal trading volume. Following the above studies, we also examine market reactions using abnormal return and volume measures. The association with return and volume measures in our social media setting, in which firms may or may not publicize earnings news that is also disseminated over traditional channels, is unclear ex ante. We start with case in which a firm decides to publicize its earnings news over social media, which we assume results in more capital market participants knowing about the news than if the news were not disseminated over social media. Then, depending on the content of the earnings announcement, there could be more or less of a market reaction. Holthausen and Verrecchia (1990) present a model in which the return and volume reactions to an information disclosure depend on both the extent to which investors become more knowledgeable and find consensus (i.e., the informedness and consensus effects). Their work suggests that if an earnings announcement leads to greater informedness and consensus, and dissemination over social media amplifies these effects, then there should be greater return volatility associated with earnings news disseminated over social media. However, greater informedness leads to higher volume reactions while greater consensus leads to lower volume, thus, the ultimate effect on volume depends on which effect dominates. Next, we consider the case in which the firm does not publicize its earnings news over social media. If the reason is that the news is bad, and bad news tends to be associated with larger return and volume reactions (consistent with greater informedness and less consensus in the Holthausen and Verrecchia model and the torpedo effect documented in Skinner and Sloan (2002)), then we should find lower market reactions for the earnings announcement disseminated 9

11 over social media because the news tends to be good. Such a finding would also be consistent with our first hypothesis. In either case, we would expect differential market reactions for earnings announcements publicized over social media relative to earnings announcements not publicized over social media. We state our second hypothesis as follows. H2a: The market reaction to earnings announcements disseminated over social media differs from the market reaction to earnings announcements not disseminated over social media. In addition to examining measures of information content, prior studies have examined measures of information asymmetry associated with financial disclosures. Analytical work by Diamond and Verrecchia (1991) show that information disclosures can reduce information asymmetry and lead to greater liquidity in the stock, while Kim and Verrecchia (1994) consider how an earnings announcement may allow certain traders to make superior judgements over other traders that lead to greater information asymmetry and less liquidity. In their study of 85 technology firms that used Twitter from March 2007 to September 2009, Blankespoor et al (2013) find increased stock market liquidity (lower abnormal bid-ask spreads and greater abnormal depths) when smaller, less visible firms disseminate earnings news via Twitter, consistent with a reduction in information asymmetry. But the authors also note that their results may not generalize to other firms, industries, or time periods. Therefore, we also test the hypothesis that earnings news publicized over social media for a broad set of firms is associated with a reduction in information asymmetry. H2b: Information asymmetry is reduced when firms disseminate earnings news over social media. A unique aspect of disseminating earnings news over Twitter is that firms may tweet multiple earnings-related messages over time. We indeed find not only tweets about a firm s 10

12 earnings announcements, which we refer to as EA tweets, but also tweets reminding followers of upcoming earnings announcements and tweets rehashing highlights from past earnings announcements. We refer to these types of tweets as preview tweets and rehash tweets, respectively, and we examine all three types of tweet separately in our intra-day analyses. We believe that examining the reactions to preview and rehash tweets is important to understanding the overall disclosure strategy that firms employ to communicate with investors via social media and how those communications are interpreted by the market. While one may not expect preview and rehash tweets to generate any market reaction because they do not contain any new information, it is plausible that a preview tweet could prompt some investors to trade in advance of the actual earnings announcement and rehash tweets could prompt some investors to trade after the earnings announcement period. For example, in the case of preview tweets, some investors may believe that firms are more likely to tweet reminders of upcoming earnings announcements if the earnings news is expected to be positive. Investors who take this view will purchase the firm s stock after receiving the preview tweet. Therefore, we test whether the market reactions of preview and rehash tweets differ from earnings announcement tweets. We state our next hypothesis in the null form. H3: Preview and rehash tweets generate the same market reaction as EA tweets. Lastly, we examine variation in firms consistency, or level of commitment, to use social media for financial disclosures. Specifically, we investigate whether the determinants and capital market consequences are different between firms that disseminate earnings news over social media every quarter and firms that do so on an ad hoc basis. The consequences of a commitment to a social media disclosure policy is especially interesting since some firms presumably made this commitment before the SEC endorsed the use of social media as an 11

13 official disclosure venue. We define firms to be committers if, after they have disseminated earnings news over social media in one quarter, they do so again each and every subsequent quarter. 11 A firm must have disclosed earnings news on social media for at least two consecutive quarters before we designate it as a committer. We conjecture that a firm s decision to commit to disseminate earnings news over social media every quarter is related to the market reaction to the firm s earnings announcements prior to the commitment. That is, a firm that perceives the market does not react to its earnings announcements is more likely to make a commitment to disseminate earnings announcements over social media in an effort to increase the market reaction. There are a number of reasons why market participants may not react to a given firm s earnings announcement. First, the firm may be neglected in the sense that very few investors and analysts follow the firm. Second, investors may deem the firm s earnings news as less than credible. Third, the issue with credibility may be exacerbated if the market perceives that the firm will only disseminate earnings news over social media when the news is good and suppress the news when it is bad. We expect these issues to be mitigated for firms that have shown a commitment to disclose earnings news over social media on a consistent basis. H4: There is a greater market reaction for firms that disseminate earnings news over social media every quarter than for firms that do so on an ad hoc basis. 3. Data We begin with all firms included in the S&P1500 index as of January 2013, based on data from Compustat. We then collect data from each firm s Facebook and Twitter sites (if they 11 Leuz and Verrecchia (2000) note that the distinction between a commitment and a voluntary disclosure is that the former is a decision by the firm about what it will disclose before it knows the content of the information (i.e., ex ante), whereas the latter is a decision by the firm made after it observes the content (i.e., ex post). In our setting, we assume that committed firms disseminate earnings news over social regardless of whether their earnings announcements reveal good or bad news. 12

14 exist) using the following procedure. 12 First, we visit each firm s corporate website and look for icons or links to its social media sites. This step ensures that we find the firm s true corporate Facebook and Twitter sites, as opposed to sites that may be managed by communities or user groups associated with the firm. If we do not find social media links on the corporate website, then we manually search for the firm s Facebook and Twitter pages on the respective social media sites, taking care to use only the official corporate pages if they exist. Once we have found the corporate Facebook and Twitter sites, we proceed as follows. If the firm has a Facebook page, then we collect information on when the firm joined Facebook, the number of Likes (as of July 2013), and whether the firm created any posts concerning earnings announcements for any quarter from the first quarter of 2010 through the first quarter of We accomplish this last step by scrolling through the entire timeline and searching for terms such as quarter, fiscal, earnings, results, and their variants. If a firm has a Twitter page, then we collect information on the number of tweets and followers. We then use a web utility at to retrieve all the firm s tweets (up to a maximum of 3,200 tweets), record the date of the first tweet, and search for tweets about earnings announcements for any quarter from the first quarter of 2010 through the first quarter of For firms that had more than 3,200 tweets (214 out of the 708 firms that use Twitter), we used Twitter s advanced search feature to manually retrieve all tweets containing the earnings-related terms. Appendix B provides examples of earnings news posted to Facebook and tweeted over Twitter. We collected data on the first 250 firms and then hired four research assistants (RAs) to collect data on the remaining 1,250 firms. To ensure the accuracy of the data, each RA was 12 All of our empirical tests focus on the corporate use of Facebook and Twitter because those were the only outlets explicitly identified by the SEC in its April 2013 announcement concerning social media usage to comply with Regulation Fair Disclosure. We explored the possibility that other social media platforms such as LinkedIn, Pinterest, YouTube, and Google+ could be used by firms, but these other platforms are not conducive to disseminating earnings news. 13

15 responsible for collecting data for 625 firms, resulting in two RAs collecting data for each firm. We then cross-checked the data from each pair of RAs for consistency and manually checked firms Facebook and Twitter sites to correct any inconsistencies in the data. For the subsample of firms that have ever used Twitter to disseminate earnings-related news during market trading hours (9:30AM to 4:00PM Eastern Standard Time), we record the time stamp of all earnings-related tweets, consisting of EA tweets, preview tweets, and rehash tweets. We employ six additional research assistants and again ensure accuracy by assigning pairs of RAs to collect time stamps for approximately 400 firms each and cross-checking any inconsistencies ourselves. Tweets are categorized as EA tweets if they were the first tweet mentioning the firm s earnings announcement and they occurred on the same date as the earnings announcement date. Tweets are categorized as preview tweets if they only mention the date of the next earnings announcement and are categorized as rehash tweets if they mention highlights from the most recent earnings announcement. On average, preview tweets occur 13 days before the earnings announcement date and rehash tweets occur 1 to 2 days after the earnings announcement date. To examine the intra-day market reactions to earnings-related tweets, we only use the tweets that occur between 9:45AM and 3:45PM to insure 15 minutes of trading before and after the tweet. Trading data comes from the Trade and Quote (TAQ) database through Wharton Research Data Services (WRDS). 3.1 Descriptive Statistics An overview of the corporate use of social media is illustrated in Figure 1, Panel A. Slightly over half (52%) of S&P1500 firms have adopted social media as of July The majority of these firms have both a Facebook page and a Twitter account; the remainder is split 14

16 between firms that have only one or the other. Among firms that have ever disseminated earnings announcements over social media, there is a stronger preference to do so using Twitter rather than Facebook. There are 214 firms that have ever tweeted earnings news on Twitter (but not Facebook), 192 firms that have used both Twitter and Facebook, and only 40 firms that have posted earnings to Facebook (but not Twitter). The data suggests that Twitter is now the preferred platform for firms that choose to disseminate earnings news over social media. One reason may be that, despite Twitter s 140 character limit per tweet, a firm can send multiple tweets about different aspects of the earnings news. As illustrated in Panel B of Appendix B, Alcoa sent 24 tweets regarding its 2013 first quarter earnings. The time trend in corporate social media adoption is illustrated in Figure 1, Panel B. The earliest adopters of Facebook joined in November 2007 and the first set of firms to create a Twitter account did so in May By early 2013, the corporate adoption rate of Twitter surpassed the rate for Facebook. By the end of our data collection period, approximately 47% of S&P1500 firms had a Twitter account, 44% had a Facebook page, and 52% had adopted one or the other. The time trend data also suggests that Twitter is becoming the preferred social media platform for companies. The breakdown of our sample of 1,500 firms across industries (Fama-French 30), and their adoption of social media, is provided in Table 1. Facebook and Twitter adoption is highest for customer facing industries such as Meals, Retail, Books and Services (each over 60%), while adoption is lowest for industrial sectors such as Oil and Steel (roughly 20%). This evidence is consistent with surveys indicating the potential of social media as a sales and marketing platform (e.g. Larcker et al, 2012). However, a high percentage of social media adoption within an industry does not translate well into social media usage to disseminate earnings news. For 15

17 example, the Meals industry contains 26 firms, of which 19 have a Facebook site and 18 have a Twitter account. However, only 1 firm uses either Facebook or Twitter to disseminate quarterly earnings. The pattern is similar for the Retail industry, in which only 5 out of 65 firms that use Facebook and 10 out of 61 firms that use Twitter, ever disseminate earnings news over the respective channels. In contrast, the industries with the highest percentage of firms that use social media to disseminate earnings, conditional on having a social media presence, are Oil and Steel. Descriptive statistics of the variables used in our empirical tests, spanning 1,452 firms with requisite Compustat, CRSP, and IBES data over a maximum of 13 quarters (Q to Q1 2013), are provided in Table 2. Continuous variables are winsorized at the 1 st and 99 th percentiles. Of the 18,820 firm-quarters in our sample, a corporate Facebook page exists 35.5% of the time, compared with only 5.7% of firm-quarters in which quarterly earnings are posted on Facebook. This gap highlights the significant difference between identifying Facebook usage and identifying Facebook usage for financial information. A similar, but not as disparate, pattern holds for Twitter. A corporate Twitter account exists for 31.7% of the sample firm-quarters, while earnings are tweeted in 11.7% of those firm-quarters. Even when the units of analyses are firm-quarters, the data suggests that Twitter is the more prevalent social media platform to disseminate earnings news. 4. Research Design This section proceeds by first outlining how we identify the firm attributes that are correlated with a firm s choice to disseminate earnings news via social media. We use the results of this estimation to not only provide insights into the drivers of social media usage to announce 16

18 earnings, but also to provide a basis for the propensity score approach to selecting control firms for our tests of the market consequences of social media usage. 4.1 Determinants of Social Media Usage for Earnings Announcements Our determinants tests proceed in two steps. First, we estimate a firm-level, crosssectional regression using firms attributes for the last fiscal quarter on or before March 31, We use a set of probit regressions that estimate the determinants of: 1) whether the firm disseminated earnings news on Facebook (Twitter) in any quarter during the sample period, and 2) whether the firm committed to using Facebook (Twitter) to disseminate earnings news. 13 As noted in Section 2, a committed firm is one that uses Facebook (Twitter) to disseminate earnings news each and every quarter once it starts to disseminate earnings news on Facebook (Twitter). The specifications are as follows: FB_EA i = α 0a + α 1a LOG_FB_LIKES i + α 2a TW_EA i + α ia Disclosure Factors i + ϵ 1a (1a) TW_EA i = α 0b + α 1b LOG_TW_FOLWRS i + α 2b FB_EA i + α ib Disclosure Factors i + ϵ 1b (1b) All variables are defined in Appendix A. All specifications include industry fixed effects (Fama French 10). The dependent variables FB_EA i and TW_EA i take the value of 1 (0 otherwise) if firm i posted earnings news to Facebook and Twitter, respectively, at least once during the sample period. We include LOG_FB_LIKES i (LOG_TW_FOLWRS i ) to proxy for the size of firm i s Facebook (Twitter) audience, as it could be related to the propensity of the firm to 13 We do not model the decision to use social media for non-financial purposes because doing so would require a model that incorporates numerous non-accounting factors. For example, one of the primary benefits of social media highlighted in the popular press is as a sales and marketing tool. Therefore, it is likely that marketing expense, corporate social responsibility performance, online retail presence and several other non-accounting factors would drive social media usage. Other than through changing the incentives to use social media, there is little prior research to suggest that these variables would affect a firm s voluntary disclosure choice. 17

19 disseminate earnings news over social media. Similarly, we include TW_EA i (FB_EA i ) in equation 1a (1b), to control for the fact that there is a competing social media platform. We first run the regression using all firms in the sample, including those that do not use either social media platform, and then we re-estimate the regression including only those firms that have a Facebook Page (Twitter Account). This procedure allows us to identify more precisely whether the firms that do not use social media are good controls for the firms that use social media. The Disclosure Factors we use follow related research that has examined other voluntary disclosure outlets, including conference calls (Frankel et al. 1999), corporate websites (Ettredge et al., 2002), and conference presentations (Bushee et al. 2011). More specifically, we include a set of variables that reflect the size, analyst coverage, performance, and risk of the firm. These variables are defined in Panel D of Appendix A. We also re-estimate (1a) and (1b) focusing on firms that have committed to use social media for earnings news each and every quarter. In that specification, the independent variables are unchanged, and the dependant variables are FB_EA_COMMIT and TW_EA_COMMIT. FB_EA_COMMIT (TW_EA_COMMIT) takes the value of 1 (zero otherwise) if firm i announces its earnings consistently each and every quarter on Facebook (Twitter) once it starts using that social media platform. A firm must have disclosed earnings news on social media for at least two consecutive quarters before it is designated as a committer. In other words, no firm will be designated as committed for the quarter in which it disseminates earnings news through social media for the first time. In the second step of our determinants tests, we estimate a panel regression using firmquarter observations. This increases the number of observations from 1,452 firms in equation (1) to as many as 18,820 firm-quarters. More importantly, it allows us to test whether there are timevarying firm characteristics that are correlated with social media disclosures. We test our first 18

20 hypothesis (H1) by expanding the set of independent variables in equation (1) to include a variable MEETBEAT i,q, which takes the value of 1 if firm i meets or beats the consensus analyst forecast in quarter q. Including this variable allows us to examine whether firms opportunistically disseminate earnings news using social media on a quarter-by-quarter basis. The specifications we employ are as follows: FB_EA_Q i,q = γ 0a + γ 1a MEETBEAT i,q + γ 2a FB_PAGE_Q i,q-1 + γ 3a LOG_FB_LIKES i + γ 4a TW_EA_Q i,q + γ ia Disclosure Factors i,q + ϵ 2a (2a) TW_EA_Q i,q = γ 0b + γ 1b MEETBEAT i,q + γ 2b TW_ACCOUNT_Q i,q-1 + γ 3b LOG_TW_FOLWRS i + γ 4b FB_EA_Q i,q + γ ib Disclosure Factors i,q + ϵ 2b (2b) All variables are defined in Appendix A. All specifications include industry fixed effects (Fama French 10) and quarter fixed effects. 14 The dependent variables FB_EA_Q i,q and TW_EA_Q i,q, take the value of 1 (0 otherwise) if firm i posted earnings news to Facebook and Twitter, respectively, for fiscal quarter q. If firms only post good earnings news to social media, then the coefficient on MEETBEAT will be positive (i.e. γ 1 >0). But if firms do not distinguish good news from bad when deciding to disseminate earnings over social media, then the coefficient will not be significantly different from zero (i.e. γ 1 =0). We include FB_PAGE_Q (TW_ACCOUNT_Q), an indicator variable set to 1 if firm i has a Facebook page (Twitter account) as of the end of quarter q-1 and zero otherwise, because the likelihood that a firm releases earnings on Facebook (Twitter) is dependent on whether the firm has an existing Facebook page (Twitter account). We use lagged values for these variables because we are capturing whether the firm has an existing social media presence at the time it 14 In robustness tests, we include firm fixed effects because it is possible that there is an unobserved firm-specific factor whose omission from our multivariate analysis is material. This is a very conservative approach in our setting, since our dataset only covers 13 quarters. However, none of our inferences are affected by this change. 19

21 decides to disseminate earnings. We include LOG_FB_LIKES and LOG_TW_FOLWRS to proxy for the size of the social media presence. We do not have historical data on the number of likes on Facebook or the number of followers on Twitter, so the value we use is based on an examination of each social media platform as of the end of our sample period. As with equation (1), we control for the alternate social media platform, and as an additional specification, we reestimate equation (2) by dropping firms that do not have a social media presence. 4.2 Capital Market Consequences of Social Media Usage When we examine the capital market consequences of social media usage, we conduct event study tests using daily and intra-day trading data. Using daily data provided by CRSP, we test for a market reaction using three proxies measured over a three-day window around the date of the earnings announcement: absolute abnormal returns (ABS_CAR), abnormal turnover (ABN_TURN), and abnormal bid-ask spread (ABN_SPREAD). Because the choice to disseminate financial information via social media is endogenous, we employ a propensity score approach (Rosenbaum, 2005; Rosenbaum and Rubin, 1983) where we match two firms that are equally likely to disseminate earnings via social media based on equation (2), but where only one firm actually disseminates earnings via social media. The advantage of this approach is that it allows us to match firms along all the observable disclosure factors used in our determinants model discussed in Section 4.1. The specifications we employ are as follows: EA_RESPONSE i,q = ρ 0a + ρ 1a MEETBEAT i,q + ρ 2a FB_EA_Q i,q +ρ 3a LOG_FB_LIKES i + ρ 4a TW_EA_Q i,q + ρ ia Disclosure Factors i,q + ϵ 3a (3a) EA_RESPONSE i,q = ρ 0b + ρ 1b MEETBEAT i,q + ρ 2b TW_EA_Q i,q +ρ 3b LOG_TW_FOLWRS i + ρ 4b FB_EA_Q i,q + ρ ib Disclosure Factors i,q + ϵ 3b (3b) 20

22 Where EA_RESPONSE is either the three-day abnormal absolute returns (ABS_CAR), abnormal trading turnover (ABN_TURN), or abnormal bid-ask spread (ABN_SPREAD) surrounding the release of quarterly earnings on Facebook in equation 3a and Twitter in 3b. All variables are defined in Appendix A. Our second hypothesis (H2a and H2b) is that there is a different market reaction when earnings news is disseminated over social media (i.e., ρ 2 0). A potential issue with equation (3) is that any documented market reaction may be driven by another event that occurs around the same time as the dissemination of the earnings news over social media. We address this concern by performing a set of tests that utilize intra-day data. We focus on tweets because we can identify the specific time of the tweet. We measure the market reaction during five 5-minute intervals: t 0 is the first 5 minutes after the tweet, t 1 is between 5 to 10 minutes after the tweet, t 2 is between 10 to 15 minutes after the tweet, t -1 is 5 minutes before the tweet, and t -2 is between 5 and 10 minutes before the tweet. We compute four volume-based measures to examine the market response during the minutes surrounding the tweets. Abnormal volume (ABN_VOL) is defined as the trading volume during the 5-minute time interval scaled by a measure of normal trading volume. Since each time interval is relatively short, our abnormal volume measure may be distorted for firms with normally low trading volume. Therefore, we choose the scalar to be the average trading volume for a 5-minute time interval from the trading day one week prior (i.e., the entire day s volume divided by 78 5-minute intervals within a 6.5 hour trading day). In addition, to examine potential differences in the market response due to large and small investors, we compute ABN_VOL_LG and ABN_VOL_SM as the abnormal volume from large and small trades, respectively, where large trades are defined as those for $50,000 or more (Lee, 1992; Bushee et al., 2003). Finally, we compute the change in the average trade size (TRADE_SIZE) 21

23 as the mean size of all trades during the 5-minute time interval divided by the mean size of all trades from the trading day one week prior. Lastly, we extend the specifications in equation (3) to investigate whether a commitment to social media affects the capital market response. We include two additional variables FB_EA_COMMIT is an indicator set to 1 (0 otherwise) for firms that have committed to disseminate earnings news over Facebook and FB_EA_COMMIT_Q is an indicator set to 1 (0 otherwise) for the quarters in which the committed firm has used Facebook for earnings news. We define similar variables for firms that have committed to using Twitter for earnings news (TW_EA_COMMIT and TW_EA_COMMIT_Q). We note that for each committed firm, there are quarters in which the firm has not yet used social media for earnings news, which we refer to as the pre-commitment period. We expect that the coefficient on FB_EA_COMMIT (TW_EA_COMMIT) to be negative, indicating that there was relatively less of a market reaction in the pre-period for firms who became committers. Our fourth hypothesis (H4) states that the coefficient on FB_EA_COMMIT_Q (TW_EA_COMMIT_Q) should be positive, indicating that committed firms experienced a larger capital market response in the post-period. 5. Results This section proceeds in two parts. In the first subsection, we summarize our findings on the determinants of social media usage. Next, we provide evidence on the capital market response to social media usage. 5.1 Determinants of Social Media Usage for Earnings Announcements The results of estimating firm-level regressions using equation (1) are shown in Table 3, Panel A. The Facebook specifications are shown in columns (1) (3) and the Twitter 22

24 specifications in columns (4) (6). A comparison of column (1) and (4) highlights the differences between the firms that use Facebook to disseminate earnings announcements (at least once) versus those that use Twitter. As noted earlier, there is a significant overlap in Facebook and Twitter usage. However, these specifications include controls for the alternative social media outlet, and therefore the coefficients can be interpreted as the incremental effect of Facebook and Twitter, respectively. In terms of the differences, we find that firm size is negatively associated with Facebook usage but positively associated with Twitter usage. In other words, we find that large firms are more likely to use Twitter and less likely to use Facebook to disseminate earnings news. It is somewhat surprising that large firms are more likely to use Twitter for disseminating earnings announcements since the benefits of improved dissemination have been argued to be greater for smaller firms (Blankespoor et al. 2013). We also find that analyst coverage is negatively associated with Twitter and Facebook. This finding is consistent with both social media platforms being used by firms with low analyst coverage to disseminate information to investors. In both specifications, the size of the firms social media presence (proxied by either likes on Facebook or followers on Twitter) is positively associated with whether earnings are released via social media. In addition, the statistical significance of this relationship is very strong. We do not find that other traditional determinants of voluntary disclosure, such as market-to-book (MTB), firm performance (ROA) or growth (GROWTH), are associated with the decision to use social media to disseminate earnings news. Despite the lack of significance on these variables, the explanatory power of our model is relatively strong. The pseudo-r 2 is 40% in column (1) and 48% in column (4). When we restrict the sample to the firms that have a Facebook page or Twitter account, the results change somewhat, as shown in columns (2) and (5). Because these specifications only 23

25 include firms that use social media, the coefficients are capturing the difference between firms that disseminate and do not disseminate earnings through social media. For the Facebook specifications, there is no longer a statistically significant association between firm size and Facebook usage for earnings news. In addition, there is now a negative association between the size of the firm s Facebook audience (LOG_FB_LIKES) and Facebook usage for earnings news, in contrast with the results in column (1). For the Twitter specifications, the positive association between firm size and Twitter usage for earnings dissemination is stronger. However, as with the Facebook specifications, there is now a negative association between Twitter usage and the size of the firm s Twitter audience (LOG_TW_FOLWRS). Overall, these results suggest while larger firms are more likely to use Twitter to disseminate earnings news, firms with more followers are not are more reluctant to use Twitter for earnings news. While we have not fully investigated the reasons for this apparent inconsistency, it is plausible that firms with more followers are using social media primarily for reaching customers rather than investors. In our summary statistics, we found that retail firms are more likely to use social media and have more followers. We next focus on firms that have committed to disseminating earnings news on Facebook and Twitter in columns (3) and (6), respectively. For these specifications, we only include firms that use social media. As a result, the coefficients capture the difference between a committed and non-committed firm, conditional on the firm using social media during our sample period. There are 406 (232) firms who have used Twitter (Facebook) to disseminate earnings news; 108 firms commit to disclose their quarterly earnings on social media (either on Twitter or Facebook), 90 firms commit to disclose on Twitter, 43 firms commit to disclose on Facebook, and 25 firms commit to disclose on both Facebook and Twitter. 24

26 None of the traditional measures of the incentives for voluntary disclosure are statistically significant in the Facebook specifications. Similarly, only size is statistically significant in the Twitter specifications. This result suggests that other than size, traditional voluntary disclosure models do not explain the choice to commit to social media. We do find a significantly negative coefficient for LOG_FB_LIKES, indicating that firms with a larger Facebook following are less likely to commit to posting earnings news consistently. Alternatively, the result could indicate that firms with a small Facebook following are trying to attract a larger audience by demonstrating a commitment to use social media for earnings news. The results of estimating firm-quarter-level regressions using equation (2) are shown in Table 3, Panel B. The Facebook specifications are shown in columns (1) (3) and the Twitter specifications in columns (4) (6). The first specification for both Facebook and Twitter use our entire sample of 18,820 firm-quarters. Under this specification, we find that book leverage (LEVERAGE) is positively associated with disseminating earnings news on Facebook, and that firm size (SIZE) has a strong positive association with disseminating earnings news on Twitter. These specifications include a single indicator variable for lagged social media presence. As noted earlier, we expected the decision to disseminate earnings news through social media to be dependent on having a social media presence. This expectation is clearly satisfied by our data, based on the extremely high levels of significance on the lagged values of FB_PAGE_Q and TW_ACCOUNT_Q. In column (2) and (5) we focus on the subset of the firm-quarters in which firms have a social media presence. We do this because it is possible that non-social media firms are not good controls due to unobserved differences. Under this approach, we find that several of the traditional disclosure factors are associated with social media usage. We find that Facebook 25

27 usage is positively associated with SIZE and ROA, indicating that larger and more profitable firms are more likely to use Facebook to disseminate earnings news. Similarly, we find that Twitter usage is positively associated with SIZE and MTB. These results suggest that Twitter usage is also more prevalent in larger and better performing firms. Our first hypothesis (H1) is that the decision to disseminate earnings news each quarter on social media is related to the direction of the news. We test H1 by including the variable MEETBEAT, as shown in columns (3) and (6). We find that firms are more likely to announce quarterly earnings through social media when they meet or beat the consensus analyst forecast for the quarter. However, this result only holds for Twitter the coefficient on MEETBEAT in column (6) is positive and highly significant. For Facebook, the coefficient in column (3) is positive but insignificant. This lack of a result for Facebook may, in part, be due to the statistical power of our tests. As noted earlier, there are only 40 firms that disseminate earnings news on Facebook exclusively, compared with 214 firms that use Twitter exclusively. Overall, our findings suggest that firms are more likely to tweet earnings when the news is positive. 5.2 Capital Market Consequences of Social Media Usage We first present univariate evidence in Table 4 on the market reaction of quarterly earnings announcements that are disseminated versus not disseminated over social media, conditional on social media usage. We include market-based measures Size Adjusted Return (SAR), Cumulative Abnormal Return (CAR), the absolute value of CAR (ABS_CAR), abnormal turnover (ABN_TURN) and abnormal spread (ABN_SPREAD). Each variable is defined in Panel B of Appendix A. For each variable, we calculate the mean for quarters in which earnings news is on social media and not on social media (i.e., the Yes and No columns), and then we test for differences in these means. Hypothesis H2a is that there is a significant difference in the 26

28 market reactions. The results for earnings on Facebook are provided in Panel A and the results for earnings on Twitter are provided in Panel B. Panel A reveals that the mean ABS_CAR and ABN_TURN are both significantly lower for earnings announcements that are posted to Facebook, indicating that the average market reaction is lower in absolute terms. There is no significant difference for SAR, CAR or ABN_SPREAD. The results for earnings that are tweeted on Twitter (Panel B) show a similar pattern in direction, but with higher levels of statistical significance. The mean ABS_CAR, ABN_TURN and ABN_SPREAD are all significantly lower for earnings announcements tweeted on Twitter. In addition, both SAR and CAR are significantly higher for earnings announcements tweeted on Twitter. Together, these results indicate that firms tend to tweet more good than bad news earnings, consistent with the evidence from our determinants tests from the prior section. Assuming that this market reaction is primarily driven by the information content of the earnings release itself, these results indicate that firms are not disseminating negative earnings surprises on Twitter. Our results using abnormal spreads (ABN_SPREAD) indicate that the average bid-ask spread is lower when earnings announcements are tweeted via Twitter, consistent with hypothesis H2b and the findings from Blankespoor et al. (2013) for small technology firms. We selected the sample used in our subsequent analyses using the propensity scores (i.e., the predicted probability of FB_EA_Q i,q = 1 or TW_EA_Q i,q = 1) computed from our determinants model. We matched each of the 1,038 (2,071) treatment observations 15 (i.e., firms who disseminated earnings news on Facebook (Twitter)) to a control observation (i.e., firms that have a Facebook (Twitter) presence but did not disseminate earnings news on Facebook 15 The number of Facebook (Twitter) earnings news treatment observations (i.e., 1,038 (2,071)) is less than the total in Table 4 (i.e., 1,066 (2,136) ) because at the time we collected data some firms in our sample had not yet announced first quarter 2013 earnings. 27

29 (Twitter)), using the same fiscal quarter and with the smallest propensity score difference. To assess the effectiveness of the matching procedure, we evaluate the covariate balance between the two samples; i.e., whether the treatment and control samples are similar along the determinants variables included in our model. In untabulated results, we find only a few statistically significant differences between the two groups. Firms using Facebook to disseminate earnings news are smaller; firms using Twitter to disseminate earnings have less analyst coverage, lower ROA, lower sales growth and higher leverage. The results of estimating equation (3) are provided in Table 5. The coefficients on FB_EA_Q and TW_EA_Q are insignificant for the ABS_CAR and ABN_TURN variables, but positive and significant for the ABN_SPREAD. This result indicates that the endogenous nature of a firm s choice of social media communications has a significant effect on the regression results. Our multivariate analysis suggests that the average bid-ask spread is actually higher for firms that announce earnings via social media, inconsistent with H2b. Under the Kim and Verrecchia (1994) model, this result suggests that the wide-spread dissemination of an earnings announcement over social media allows some investors to make judgements about the firm s performance that are superior to the judgments of other investors, resulting in greater information asymmetry and less stock liquidity. The coefficients on both measures of social media audience size LOG_FB_LIKES for Facebook and LOG_TW_FOLWRS for Twitter are positive and highly significant for the ABS_CAR and ABN_TURN specifications (columns (1), (2), (4) and (5)). These results indicate that the larger the social media audience, the larger the market reaction for earnings news disseminated over social media, suggesting that a firm s followers includes many capital market participants. Moreover, the negative and significant coefficients on MEETBEAT suggest 28

30 that there is a greater market response to firms that report negative earnings surprises, consistent with the torpedo effect (Sloan and Skinner, 2002). The results of our intra-day analyses are illustrated in Figure 2 and Table 6. Among the firms that have a Twitter account and use it for disseminating earnings news, only a subset sends earnings-related tweets during market hours and only for some quarters. Therefore, our intra-day analyses are not without limitations and possible self-selection bias. There are 292 firm-quarters for which we observe an EA tweet during market hours, 380 firm-quarters for rehash tweets, and 607 firm quarters for preview tweets. In each panel of Figure 2, we plot the mean market response for EA tweets, preview tweets, and rehash tweets for each time interval. Interval t 0 is the first 5 minutes after the tweet, t 1 is between 5 to 10 minutes after the tweet, t 2 is between 10 to 15 minutes after the tweet, t -1 is 5 minutes before the tweet, and t -2 is between 5 and 10 minutes before the tweet. In each case, we scale by the average 5-minute trading volume from the same day one week prior to the tweet. We note that we do not solely compare volume around the tweet relative to volume during a control period, but rather, we compare volume across the five time intervals. The scaling produces values in excess of 100% for both the EA and rehash tweets, suggesting that these tweets are generally disseminated during periods of unusually high trading volume. In contrast, the preview tweets are typically disseminated when trading volume is low. Figure 2, Panel A shows that volume increases most for EA tweets during interval t 0, increases slightly for preview tweets during interval t 1, but does not change much for rehash tweets across the five time periods. Subject to the above mentioned limitations of our intra-day analyses, we conclude that EA tweets, preview tweets and rehash tweets generate different market reactions. 29

31 Figure 2, Panel B and Panel C partition the volume data from Panel A into large and small trades, respectively. The existing literature has noted that large traders should be analysed separately from small traders (e.g. Lee, 1992; Bushee, Matsumoto, Miller, 2003). We follow this literature and identify large trades as those that involved in excess of $50,000. These panels show that the increases in trading volume for EA tweets at t 0 and preview tweets at t 1 is due to increases in volume from large trades. Therefore, while social media is commonly viewed as a disclosure channel that provides timely access to information for all investors, and thus level the playing field for small investors, our results suggest that larger investors react quicker to earnings-related tweets. This finding is further illustrated in Figure 2, Panel D, which shows how the average trade size changes over the testing period. The average trade size increases substantially in response to EA tweets. The 6% increase in the average trade size is a result of the average trade size increasing from 160 shares in the control period to 170 shares in the testing period. 16 The results in Table 6 show that the trading volume associated with large trades increased in a statistically significant way relative to small trades for each type of tweet. These results suggest that larger traders are more likely to follow tweets. Moreover, it shows that larger trades respond not only to earnings announcements tweets, but also preview and rehash tweets as well. The final set of results focuses on firms that commit to using social media for earnings news. The results in Table 7 add two additional variables that allow us to identify whether there is a differential capital market reaction for firms that are committed to social media disclosures. Column (2) shows that the coefficient on FB_EA_COMMIT is significantly negative when the 16 While an increase in average trade size from 160 to 170 shares may appear nominal, the vast majority of trades are for one round lot (100 shares). Thus, it requires a substantial increase in large lot trades to raise the average to 170. For example, if there are 1,000 trades at 100 shares and 71 trades at 1,000 shares, the average trade size would be 160 shares (171,000/1,071). Holding the number of one lot trades constant, there would need to be 84 ten-lot trades (an 18% increase) in order to raise the average trade size to 170 shares (184,000/1,084). 30

32 dependent variable is abnormal turnover (ABN_TURN), indicating that there was relatively less trading volume in the pre-period for firms who became committers on Facebook. Similarly, column (4) shows that the coefficient on TW_EA_COMMIT is significantly negative when the dependent variable is absolute abnormal return (ABS_CAR), indicating less market reaction prior to the commitment to use Twitter. For the post-commitment period, we find results consistent with our fourth hypothesis (H4). The coefficients on FB_EA_COMMIT_Q and TW_EA_COMMIT_Q are positive and significant in columns (2) and (4), indicating that committed firms experienced a larger capital market response in the post-commitment period. Collectively, these results indicate that even though the overall market reaction is modest across all firms in our sample, the market reaction is relatively strong for firms that committed to disseminating earnings news over social media consistently. 6. Conclusion This study is the first to document the adoption of social media by the largest publiclytraded companies in the U.S. and their specific use of social media to disseminate financial information. Using hand-collected data on the use of social media by S&P1500 firms from 2010 to early 2013, we conclude that corporate adoption of social media has surpassed 50% and that Twitter is the preferred platform to tweet quarterly earnings news. However, our evidence also indicates that firms are more likely to tweet only the good news and not the bad earnings news, suggesting some opportunism in the decision to use social media for financial information. This finding is relevant to policy debates concerning emerging disclosure technologies and their impact on firms, investors, and capital markets. 31

33 This study also provides evidence on how the market responds to earnings news provided through social media. Using intra-day data, we find that trading volume increases in response to the initial announcement of earnings on Twitter and that the primary driver of the increased trading volume is larger rather than smaller trades. In addition, for firms that have shown a commitment to using social media for financial information by disseminating earnings news each and every quarter regardless of the direction of the news, there is a larger market reaction as reflected in greater information content and trading volume during the earnings announcement window. This finding is relevant for firms, their managers, and their board of directors that may be considering or establishing their social media disclosure policies. 32

34 References Aboody, D., and R. Kasznik CEO stock options awards and the timing of corporate voluntary disclosures. Journal of Accounting and Economics 29: Blankespoor, E., G.S. Miller, and H.D. White The Role of Dissemination in Market Liquidity: Evidence From Firms Use of Twitter. The Accounting Review, forthcoming. Bushee, B.J., M.J. Jung, and G.S. Miller Conference Presentations and the Disclosure Milieu. Journal of Accounting Research 49 (5): Bushee, B.J., D. Matsumoto, and G.S. Miller Open versus Closed Conference Calls: the Determinants and Effects of Broadening Access to Disclosure. Journal of Accounting and Economics 34: DARPA (Defense Advanced Research Projects Agency) DARPA Network Challenge Project Report, February 16. Available online at: DARPA (Defense Advanced Research Projects Agency) Broad Agency Announcement: Social Media in Strategic Communication (SMISC). DARPA-BAA-11-64, July 14. Available online at: Diamond, D.W., and R.E. Verrecchia Disclosure, liquidity, and the cost of capital. The Journal of Finance, 46(4): Ettredge, M., V. Richardson, and S. Scholz Dissemination of information for investors at corporate Web sites. Journal of Accounting and Public Policy 21 (4): Forbes, Behind the Scenes of Oreo s Real-Time Super Bowl Slam Dunk. Frankel, R., M. Johnson, and D.J. Skinner An Empirical Examination of Conference Calls as a Voluntary Disclosure Medium. Journal of Accounting Research 37 (1): Harvard Business Review (BHR) The New Conversation: Taking Social Media from Talk to Action. Available at: Healy, P. and K. Palepu The challenges of investor communications: the case of CUC International, Inc. Journal of Financial Economics 38, Holthausen, R., and R.E. Verrecchia. E The Effect of Informedness and Consensus on Price and Volume Behavior. The Accounting Review 65(1): Kim, O., and R.E. Verrecchia Market liquidity and volume around earnings announcements. Journal of Accounting and Economics 17,

35 Federal Emergency Management Agency (FEMA) National Preparedness Report (March 30). Available online at: Larcker, D.F., S.M. Larcker, and B. Tayan What Do Corporate Directors and Senior Managers Know about Social Media? The Conference Board. Available at: Leuz, C. and R.E. Verrecchia The Economic Consequences of Increased Disclosure. Journal of Accounting Research 38: Lee, C Earnings News and Small Traders: an Intra-day Analysis. Journal of Accounting and Economics 15: Li, K. and N. Prabhala Self-selection models in corporate finance. In: Eckbo, B.E. (Ed.), Handbook of Corporate Finance, Empirical Corporate Finance, Elsevier, North-Holland Vol. 1, Li, N Labor Market Peer Firms. Working paper, University of Chicago. Police Executive Research Forum (PERF) Managing Major Events: Best Practices from the Field (June). Critical Issues in Policing Series. Available online at: Rosenbaum, P.R., Rubin, D.B., The central role of the propensity score in observational studies for causal effects. Biometrika 70, Rosenbaum, P.R., Observational study. Encyclopedia of Statistics in Behavioural Science 3, Securities and Exchange Commission (SEC) SEC Says Social Media OK for Company Announcements if Investors Are Alerted. Skinner, D.J Why firms voluntarily disclose bad news. Journal of Accounting Research 32: Skinner, D.J., and R.G. Sloan Earnings Surprises, Growth Expectations, and Stock Returns or Don't Let an Earnings Torpedo Sink Your Portfolio. Review of Accounting Studies 7: Stone, B., and N. Cohen Social Networks Spread Defiance Online. The New York Times (June 15). Trueman, B Why do managers voluntarily release earnings forecasts? Journal of Accounting and Economics 8: Vargas, J.A., Spring Awakening: How an Egyptian Revolution Began on Facebook. New York Times (February 17). 34

36 Wall Street Journal (February 14). Egypt s Revolution by Social Media: Facebook and Twitter let the people keep ahead of the regime. 35

37 Appendix A: Variable Description and Data Sources Specifications of variables used throughout the paper. Table A1 Panels A, B, C and D describe the social media variables, the market reaction variables, the intra-day market reaction variables and the control variables, respectively. Variable Description Data Source Panel A: Social Media Variables FB_PAGE_Q FB_EA FB_EA_Q FB_EA_COMMIT FB_EA_COMMIT_Q FB_LIKES Indicator variable set to 1 if the firm has a Facebook page at the end of the quarter Indicator variable set to 1 if the firm posted news of its earnings on Facebook (i.e., FB_EA_Q = 1) at least once during our sample period Indicator variable set to 1 if the firm posted news of its earnings on Facebook on the actual date of its earnings announcement or one day afterwards for the quarter Indicator variable set to 1 if the firm posted news of its earnings on Facebook (i.e., FB_EA_Q = 1) each and every quarter after the first time Indicator variable set to 1 (0 otherwise) for the quarters in which the committed firm (i.e., FB_EA_COMMIT = 1) has used Facebook for earnings news The number of Facebook likes that a firm had at the end of July 2013 Facebook Facebook Facebook Facebook Facebook Facebook LOG_FB_LIKES The natural logarithm of the number of Facebook likes Facebook TW_ACCOUNT_Q TW_EA TW_EA_Q TW_EA_COMMIT TW_EA_COMMIT_Q TW_FOLWRS Indicator variable set to 1 if the firm has a Twitter account at the end of the quarter Indicator variable set to 1 if the firm tweeted news of its earnings (i.e., TW_EA_Q = 1) at least once during our sample period Indicator variable set to 1 if the firm tweeted news of its earnings on the actual date of its earnings announcement or one day afterwards for the quarter Indicator variable set to 1 if the firm tweeted news of its earnings (i.e., TW_EA_Q = 1) each and every quarter after the first time Indicator variable set to 1 (0 otherwise) for the quarters in which the committed firm (i.e., TW_EA_COMMIT = 1) has used Twitter for earnings news The number of Twitter followers that a firm had at the end of July 2013 Twitter Twitter Twitter Twitter Twitter Twitter LOG_TW_FOLWRS The natural logarithm of the number of Twitter followers Twitter 36

38 Panel B: Market Reaction Variables SAR CAR Three-day raw return minus the return of the corresponding size-decile index centered at the dates of the quarterly earnings announcement ( QEA ) Three-day cumulative abnormal return CAR measured as the residual from a market model. The market model parameters are estimated over the period from 11 to 265 days before the QEA using returns from a value-weighted market portfolio CRSP CRSP ABS_CAR Absolute value of CAR CRSP ABN_TURN ABN_SPREAD Three-day average volume divided by shares outstanding, less the average turnover in the estimation period. The estimation period beings 61 days prior to the QEA and ends 2 days prior to the QEA Constructed as above using the bid-ask spread, defined as the difference between the bid and ask price divided by the average of the bid and ask price, multiplied by 100 CRSP CRSP Panel C: Intra-day Market Reaction Variables ABN_VOL ABN_VOL_LG ABN_VOL_SM TRADE_SIZE Panel D: Control Variables SIZE MTB LOGANALYST ROA Trading volume during the 5-minute time interval divided by the trading volume during a control period. The control period is the average trading volume for a 5-minute time interval from the trading day one week prior. Constructed as above for trades that were for $50,000 or more Constructed as above for trades that were for less than $50,000 The mean size of all trades during the 5-minute time interval divided by the mean size of all trades from the trading day one week prior Natural logarithm of total assets, measured at the end of the quarter Market value of equity divided by common equity, measured at the end of the quarter Natural logarithm of number of analysts with an EPS forecast for the quarter Income before extraordinary items divided by total assets, measured at the end of the quarter TAQ TAQ TAQ TAQ Compustat Compustat IBES Compustat GROWTH Year-over-year percentage change in quarterly sales Compustat LEVERAGE MEETBEAT Sum of long-term debt and debt in current liabilities divided by total assets, measured at the end of the quarter Indicator variable set to 1 if the firm s actual EPS meet or beat the consensus analyst forecast for the quarter Compustat IBES 37

39 Appendix B: Examples of Earnings News Posted to Facebook and Tweeted over Twitter Panel A: An Earnings Post from Alcoa on April 8,

40 Panel B: Multiple Earnings Tweets from Alcoa on April 8, 2013 $AA Listen to Replay of Alcoa's 1Q13 Earnings Available From 04/08/ :00 PM ET To: 04/15/ :59 PM ET Apr 08, 2013 Alcoa Reports First Quarter Net Income of $0.13 Per Share; Income of $0.11 Per Share Excluding Special Items Apr 08, 2013 $AA Reports 1Q13: Alcoa s aluminum helps airplanes and autos increase energy efficiency Apr 08, 2013 *HAPPENING NOW* Tune-in now for $AA Alcoa's 1Q13 earnings webcast. Get the slides, listen-in, and get replay info. Apr 08, 2013 $AA: Global end market growth strong in Aero, Auto, Truck, Packaging, Building & Construction, Industrial Gas Turbine Apr 08, 2013 $AA Download Alcoa's 1Q13 earnings presentation Apr 08, 2013 $AA CEO Kleinfeld: Alcoa achieved these results by pressing our innovation edge, scale and strength in end markets Apr 08, 2013 $AA Alcoa's 1Q13 earnings presentation webcast about to begin at 5:00pm ET. Tune in at Apr 08, 2013 (2/2) $AA CEO Kleinfeld: while our upstream business continues to move down the cost curve. Apr 08, 2013 (1/2) CEO Kleinfeld: Our mid & downstream businesses now account for 72% of our total after-tax operating income Apr 08, 2013 (3/3) CEO Kleinfeld: and remarkable upstream performance in the face of weak metal prices. Apr 08, 2013 (2/3) $AA CEO Kleinfeld: improved results in our midstream business Apr 08, 2013 (1/3) $AA CEO Kleinfeld: This was a strong quarter led by record profitability in our downstream business Apr 08, 2013 $AA Reports 1Q13: Alcoa s 1Q13 net income excluding special items was the best since the third quarter of Apr 08, 2013 $AA Reports 1Q13: Alcoa delivered solid first quarter results across all business segments Apr 08, 2013 $AA Reports 1Q13: Value-added businesses now account for 72% of total after tax operating income Apr 08, 2013 $AA Reports: Global end market growth remains solid, forecast of 7% global aluminum demand growth in 2013 reaffirmed Apr 08, 2013 $AA Reports 1Q13: Debt-to-capital ratio 35 percent Apr 08, 2013 $AA Reports 1Q13: Strong liquidity with cash on hand of $1.6 billion Apr 08, 2013 $AA Reports 1Q13: Record low first quarter days working capital Apr 08, 2013 $AA Reports 1Q13: Improved performance in Alumina and Primary Metals year-over-year, despite lower metal prices Apr 08, 2013 $AA Reminder: Tune in to Alcoa's Webcast of 1Q13 Results Beginning at 5:00pm ET at Release: Apr 08, 2013 $AA Reports 1Q13: Record after-tax operating income in Engineered Products and Solutions Apr 08, 2013 $AA Reports 1Q13: Net income $0.13 per share Apr 08,

41 Nov-07 Feb-08 May-08 Aug-08 Nov-08 Feb-09 May-09 Aug-09 Nov-09 Feb-10 May-10 Aug-10 Nov-10 Feb-11 May-11 Aug-11 Nov-11 Feb-12 May-12 Aug-12 Nov-12 Feb-13 May-13 Figure 1: Social Media Usage by S&P1500 Firms Panel A: Corporate Use of Social Media Panel B: Adoption of Social Media over Time 55% 50% 45% 40% 35% 30% 25% 20% 15% 10% 5% 0% Either 52% Twitter 47% Facebook 44% 40