No whisper, no value? The effect of analysts earnings preview ban and stock market behavior surrounding an earnings announcement

Transcription

1 No whisper, no value? The effect of analysts earnings preview ban and stock market behavior surrounding an earnings announcement Katsuhiko Okada Institute of Business and Accounting, Kwansei Gakuin University Uegahara Nishinomiya Hyogo , Japan Phone: +81-(0) Hidenori Takahashi Graduate School of Economics, Nagoya University Furo-cho, Chikusa-ku Nagoya , Japan Phone: +81-(0) This version: April 30, 2018 Please do not quote or distribute without the authors permission. 1

2 No whisper, no value? The effect of analysts earnings preview ban and stock market behavior surrounding an earnings announcement Abstract On September 20, 2016, the Japan Securities Dealers Association implemented guidelines prohibiting securities sell-side analysts from obtaining an earnings preview before the earnings official release. We examine the unique impact of these guidelines on market behavior and analyst forecasts in the pre- and post-guideline periods. The results show that in the post-guideline period, stock return volatility around earnings releases increased, suggesting that the regulation improved the information flow. We also find that the guidelines did not impact the accuracy of analysts earnings forecasts. We provide evidence that there is no decrease in forecast accuracy because analysts banned by the regulation rely on alternative information sources that is, forecasts of media analysts who exempt from the regulation. Keywords: Earnings preview; Fair disclosure; Stock market behavior; Analyst forecast JEL classification: G14; G18; G11 2

3 1. Introduction Since the U.S. Securities and Exchange Commission (SEC) introduced Regulation Fair Disclosure (Reg FD), which prohibits firms from selectively disclosing information to certain securities analysts and institutions, 1 a large body of literature has examined whether the regulation has achieved its objective to level the playing field among different classes of investors without chilling the information flow to the market. Although the regulation s impact on the information environment is an important empirical question, the findings reported in prior Reg FD research so far have been mixed and unsettled (see, e.g., Bailey et al., 2003; Campbell et al., 2016; Heflin et al., 2003). One reason behind these conflicting results is the difficulty in controlling for economic events that occurred contemporaneously with the adoption of Reg FD, so-called confounding events (Campbell et al., 2016; Francis et al., 2006; Koch et al., 2013). 2 For example, Bailey et al. (2003) reveal that a seemingly significant decline in return volatility after Reg FD is due to the decimalization of stock trading rather than the adoption of Reg FD. 3 This implies that it is important to control for confounding events in assessing the impact of the regulation. In this paper, we address this issue by using a unique institutional setting in Japan: analysts earnings preview ban. For years, analysts in Japan would obtain preview numbers for a forthcoming earnings announcement from corporate managers and report the information to a subset of investors. Such preview reports were a standard practice, and institutional investors would trade on the basis of this information. 4 However, this practice suddenly became taboo on September 20, 2016, when the Japan Securities Dealers Association (JSDA), a selfregulatory organization, implemented guidelines prohibiting securities analysts from obtaining preview numbers and disseminating them to certain investors. Therefore, this natural experiment presents an excellent opportunity to investigate the impact of disclosure regulation. Specifically, we exploit two key features of the 1 The SEC approved Reg FD on August 2000, and the regulation was enacted on October 23, As examples of confounding events, Francis et al. (2006) point out the U.S. economic recession, the aftermath of the dotcom bubble, decimalization, and accounting fraud and bankruptcy among American companies. 3 Price decimalization is a unit change of stock trading. It was applied to Nasdaq-listed stocks in Instead, the Tokyo Stock Exchange (TSE) requires listed companies to disclose material information in a timely manner rather than withhold it in-house. Thus, it has been thought that the U.S.-like fair disclosure rule is not needed in Japanese stock markets. 3

4 JSDA s guidelines. First and foremost, the regulation does not target how firms should disclose; rather, it restricts sell-side analysts from asking for earnings-related information before an earnings announcement. Secondly, the JSDA s guidelines only apply to its member analysts (i.e., sell-side analysts belonging to securities firms), thereby exempting non-member analysts. The first feature provides an environment where, if the guidelines are effective, the information environment should deteriorate for firms that have analyst coverage, while non-covered firms would remain the same. The second feature creates an environment where member analysts will have less information than non-member analysts after the guidelines are adopted. These settings are ideal for capturing time-series and cross-sectional differences in market behavior and analyst forecasts, enabling us to identify the isolated impact of the analyst regulation on the market and analysts. We first analyze market behavior around earnings releases to examine whether the information flow changes because of the guidelines. We calculate abnormal return volatility around quarterly earnings announcements and compare the volatility between firms with and without analysts coverage before and after the adoption of the guidelines. We predict that if the guidelines as intended prohibit sell-side analysts from obtaining preview numbers and reporting them to a subset of investors, the earnings-related information will be concentrated around the earnings announcement date; thus, the stock price will react surrounding the earnings announcement date. To test for the impact of the guidelines on abnormal return volatility, we use a difference-in-differences approach in which the treatment and control groups consist of firms with and without analysts. Next, we examine changes in analysts forecasts after the implementation of the guidelines. Following the literature, we employ earnings forecast accuracy as the analyst forecasting performance measure. We predict that if the guidelines are effective, sell-side analysts will no longer rely on preview numbers, and this will lead to inaccurate forecasts. Using a fixed effects model with a sample of firms covered by both JSDA member and non-member analysts, we compare forecasts for the same firms before and after the guidelines adoption. The findings of this study can be summarized as follows. Following the adoption of the guidelines, the abnormal return volatility of firms with analysts coverage increases relative to firms without analysts coverage. This result indicates a partial discounting of earnings information in the run-up to the actual announcement preceding the regulation. After the regulation, information is concentrated around the earnings announcement 4

5 day. This finding is inconsistent with the empirical findings reported in prior Reg FD studies, but consistent with the intuition of market participants. 5 We find no evidence that the accuracy of JSDA analysts decreases relative to non-member analysts. This means that the deteriorated information environment has little effect on the member analysts. We further explore why there is no observable difference in forecast performance between the member analysts, who are subject to the guidelines, and non-member analysts, who are outside the realm of the guidelines. As a possible explanation of this result, we find evidence that after the guidelines adoption, member analysts tend to rely on the forecasts of non-member analysts, who are likely to have an information advantage. This paper s contribution to the literature is twofold. First, we contribute to the literature on the evaluation of securities regulation specifically, Reg FD s impact on the information environment. While the argument in prior studies is that Reg FD limited selective disclosure (e.g., Ahmed and Schneible, 2007), Campbell et al. (2016) argue that selective disclosure has continued even after the implementation of Reg FD. Thus, the impact of Reg FD is unsettled and has been debated. Although many studies have been conducted on Reg FD (see Koch et al., 2013 for a review), to the best of our knowledge, ours is the first study to examine the impact of securities regulation by exploiting the adoption of the JSDA s guidelines in Japanese financial markets. We provide new evidence that the implementation of these guidelines is related to higher abnormal return volatility around earnings announcements, which is consistent with a decrease in information asymmetry among investors and no information chill. These findings can be useful for evaluating whether such regulations are effective for improving the information environment. Our paper is related to, but different from, the work of Francis et al. (2006). Although they examine the impact of Reg FD by comparing U.S. firms and foreign firms listed on U.S. exchanges, we conduct a cross-sectional comparison of the impact, exploiting the fact that non-covered firms are beyond the guidelines reach. Second, our paper contributes to the literature investigating the source of analysts information. Green et al. (2014) argue that access to management at broker-hosted investor conferences leads to informational advantage. 5 On July 21, 2016, Bloomberg News reported, The lack of preview notes on corporate earnings has resulted in more volatile stock moves after earnings are formally published, and a wider dispersion of analyst forecasts. 5

6 Our results imply that access to management is a crucial source of information for securities analysts. We provide evidence that the regulation on sell-side analysts does not impair the analysts performance because they tend to rely on alternative information sources. We document how the regulation encourages analysts herding behavior because it creates information imbalances between member and non-member analysts, providing insights for regulators. Although Huang et al. (2014) employ a text analysis, we employ an institutional change and identify the information source. These findings call into question securities analysts raison d être as playing a price discovery role in financial markets. In the European Union (EU), the Markets in Financial Instruments Directive II (MiFID II) has been in effect since January This new legislative framework requires transparency from securities firms. 6 Analysts are required to provide clearer, more costeffective means. The rest of this paper is organized as follows. Section 2 clarifies the background of the guidelines and our data set. Section 3 describes our data set. Section 4 provides the empirical results on market behavior and analyst forecasts. Section 5 presents the robustness. Section 6 concludes the paper. 2. The JSDA s guidelines, their impact on market behavior around earnings releases, and analyst forecasts 2.1. Institutional background In 2015 and in the following years, a few scandals emerging from brokerage firms caught the public s attention in Japan. Sell-side analysts tipped off their clients about some non-public material information obtained from corporate managers. This incident triggered a debate on how the sell-side analysts should behave in a fair manner in the Japanese stock markets; then, on September 20, 2016, the JSDA implemented new guidelines. The objective of these guidelines was to prohibit sell-side analysts from obtaining preview numbers from corporate managers and disseminating them to their clients before their official release, in the hope of creating a level playing field among the different classes of investors. Surprisingly, before the guidelines were 6 European Securities and Markets Authority: 6

7 issued, there had been no restrictions on private communications between analysts and companies in Japan. The JSDA is a self-regulatory organization, but it is an association authorized by the Financial Services Agency. Therefore, although the JSDA s guidelines are merely self-regulations, not laws, they exert a significant influence on sell-side analysts. If analysts break the JSDA s regulations, they receive a fine. In fact, securities analysts in brokerages fear the guidelines. Anecdotal evidence suggests that in response to the enactment of the guidelines, some major brokerages banned preview reports. According to a report by Bloomberg News (May 27, 2016), Japanese securities firms, including Nomura Holdings Inc., no longer allow analysts to interview companies before earnings releases and instruct their analysts to use publicly available information, not preview numbers. The JSDA s guidelines are like the SEC s Reg FD in terms of their goal that is, to level the playing field among investors. There are, however, major differences between the JSDA s guidelines and Reg FD. While Reg FD prohibits managers from distributing material information to certain analysts and institutional investors, the JSDA s guidelines prohibit sell-side analysts from obtaining material information and releasing it to certain investors. Furthermore, the JSDA s guidelines can be applied to its member analysts, but not its non-member analysts. The JSDA s members include almost all securities firms operating in Japan (e.g., Nomura Securities Co., Ltd., Daiwa Securities Co., Ltd., and SMBC Nikko Securities Inc.). As of September 8, 2017, 260 securities firms were registered on the members list. 7 Non-member analysts include buy-side analysts and analysts belonging to publishers. In Japan, there are three types of analysts sell-side, buy-side, and media analysts who belong to major publishers, such as Nikkei and Toyo Keizai Inc. 8 Buy-side analysts do not visit corporations regularly, but media analysts do. Since media analysts are not members of the JSDA, they are exempt from the guidelines. The guidelines do not necessarily prohibit communication with corporate managers before an earnings announcement. The guidelines permit analysts to interview corporate managers and obtain mosaic information that is, a piece of information that is valuable when combined with other information. Analysts 7 The members are listed on the JSDA s website: 8 Nikkei is a major newspaper publisher specializing in business and economics. 7

8 can also use all public information to predict the company s future prospects. 9 Thus, if analysts play their intended role, the guidelines will not affect their performance. Critics of the guidelines claim that the regulation could chill communication between analysts and management and reduce the flow of information to the market; consequently, the capital market would not fully incorporate private information and, thus, be inefficient. If the guidelines prohibit analysts from performing their role with respect to price discovery, the stock prices would deviate from their fundamental value. But how have the guidelines affected the stock markets? A Bloomberg News article states that The impact so far has been striking: The proportion of companies missing profit estimates has climbed to the highest since 2011, earnings-day stock swings are near the biggest in 30 quarters and analysts forecasts are the most widely dispersed in two years. 10 However, these Japanese phenomena are inconsistent with the empirical findings reported by prior Reg FD research (Bailey et al., 2003; Eleswarapu et al., 2004; Heflin et al., 2003; Shane et al., 2001). Shane et al., Heflin et al., and Eleswarapu et al. observe a decrease in stock return volatility around earnings announcements after Reg FD, whereas Bailey et al. find no change. Further, although market participants feel that analysts forecasts have become more dispersed since the introduction of the guidelines, the previous authors have obtained mixed results on analyst forecast dispersion (see, e.g., Agrawal et al., 2006; Heflin et al., 2003) Stock price reactions The primary focus of prior research on Reg FD is on whether the regulation has brought about a level playing field for all investors without chilling the information flow. The empirical findings are mixed. Heflin et al. (2003) find that the absolute cumulative return around earnings announcements, which is a proxy for the information gap, is smaller post- compared to pre-reg FD, showing that Reg FD has improved the information flow. In contrast, some studies provide evidence that Reg FD has not improved the information flow. Bailey et al. (2003) and Eleswarapu et al. (2004) find no evidence of an increase in stock return volatility after Reg FD. 9 In May 2017, the Financial Services Agency approved the enactment of the Japanese fair disclosure rule (Revision of Financial Instruments and Exchange Act). The rule has been in force since April Toshiro Hasegawa, Yuko Takeo, and Phil Kuntz, Whisper numbers that tipped off Japanese traders are suddenly taboo, Bloomberg News, May 27,

9 Using a cross-sectional approach, Francis et al. (2006) compare the changes in abnormal return volatility and trading volume for U.S. firms and foreign listed firms traded in the United States (i.e., American Depository Receipts (ADRs)) that are exempt from Reg FD to control for confounding events. They find that these public information metrics of U.S. firms are not significantly different from ADR firms. Francis et al. conclude that even after controlling for confounding events, Reg FD evinces no unique impact on the information environment. Ahmed and Schneible (2007) find that Reg FD has succeeded in eliminating selective disclosure for a particular type of firm: Reg FD was effective in limiting selective disclosure practices for small firms and high-tech firms whose costs for selective disclosure are higher. To reconcile these conflicts, we empirically assess the impact of the JSDA s guidelines on market behavior surrounding earnings announcements. We take advantage of a unique feature of the JSDA s guidelines namely, that the guidelines are applied to analysts, not companies. If the guidelines are effective in preventing analysts earnings preview practices, this will result in a reduction of information disparities between individual investors and institutions. We can consider that the guidelines would only affect firms covered by analysts, not firms without coverage. Non-public information about actual figures should be incorporated into stock prices only when this information is released. Therefore, we predict that after the implementation of the guidelines, the analysts following stock price reactions will be larger than that of firms outside the analyst coverage Forecast accuracy There is also debate about the impact of Reg FD on analyst forecasts. Prior studies show mixed results. Some studies find a decrease in forecast accuracy or dispersion; others do not. Bailey et al. (2003) find no evidence that forecast accuracy increases, but forecast dispersion does after Reg FD. Heflin et al. (2003) find that in univariate results, analysts forecasts are less accurate and more dispersed after Reg FD; however, regression analyses have found no such evidence. Mohanram and Sunder (2006) examine the information that individual analysts generate through their own efforts of price discovery and find that after Reg FD, the precision of idiosyncratic information increases even after controlling for the change in the information environment in which analysts operate. Similarly, Shane et al. (2001) identify an improvement in forecast accuracy because of analysts efforts to gather uncertainty-reducing information after Reg FD. Agrawal et al. (2006) find that analyst 9

10 forecasts become less accurate and more dispersed following Reg FD; this result is more pronounced in early forecasts compared to late forecasts. Although the early works of Bailey et al. (2003) and Heflin et al. (2003) analyze a relatively short-term period (three quarters after Reg FD) and Agrawal et al. (2006) analyze a mediumterm period (15 quarters in post-reg FD), Keskek et al. (2017) use a longer post-regulation period of and find that analyst forecast accuracy decreases after several regulations implemented from 2000 to 2003, including Reg FD, the Sarbanes-Oxley Act, and the Global Settlement Act. In sum, the mixed results of prior research are sensitive and caused by differences in sample periods, type of firms (i.e., large or small firms), and type of forecasts (i.e., early or late forecasts). Thus, it is unclear whether Reg FD has been significantly effective. In the context of the JSDA s guidelines, if they have been effective in prohibiting analysts private access to managers, analysts will no longer be able to rely on preview numbers from a company s executive. In this new environment, analysts forecasts should be less accurate, ceteris paribus, after the implementation of the guidelines. However, it is possible that the accuracy has remained unchanged for the following reasons. First, analysts might increase their information search efforts because they have to rely more on independent research using publicly available information (Bailey et al., 2003; Mohanram and Sunder, 2006; Shane et al., 2001). Second, analysts may rely more on alternative information sources (e.g., Bushee et al., 2017; Green et al., 2014). Third, it is possible that the guidelines have been ineffective and the practice of previews remains prevalent Although we cannot directly observe private communication between analysts and companies, we can examine whether analyst forecasts have changed since the enactment of the guidelines. To examine the impact of the JSDA s guidelines on analyst forecasts, we focus on a second unique feature of the guidelines that is, they can only be applied to member analysts, while non-member analysts are exempt. In other words, non-member analysts can freely access corporate executives and obtain any information before 11 There is evidence that Reg FD has been ineffective (e.g., Campbell et al., 2016) and that private communication with management or selective access has prevailed, even since the introduction of Reg FD. Bushee et al. (2017) find that selective access, which is the opportunity for investors to meet privately, is a valuable information source, even in the post-reg FD period. Green et al. (2014) find that superior access to management at broker-hosted investor conferences is an essential information source for analysts in the post-reg FD period. 12 In the context of Reg FD, previous studies expected that companies would increase voluntary disclosure without reducing the amount of information available to analysts (Heflin et al., 2003). However, the JSDA s guidelines do not prohibit companies from selective disclosure; thus, this is not expected. 10

11 an earnings announcement. Since the guidelines are meant for analysts, not companies, there is nothing impeding conversations between non-member analysts and companies. By comparing the forecasts of JSDA member and non-member analysts for the same firms, we obtain a good estimate of member analysts behaved after the regulation. This allows us to control the firms information environment, because we compare forecasts for the same firms. If the JSDA s guidelines have been effective, non-member analysts will enjoy superior information access than member analysts. The member analysts no longer have access to material information before its public release, while non-members analysts have full access. Indeed, one media company officially declares in its publication, We are not analysts, but journalists; thus, outside the realm of the guidelines. If companies refuse our interview due to the guidelines, we intend to critically confront them. 13 Therefore, we predict that JSDA member analysts forecasts become less accurate compared to non-members forecasts. 3. Data and descriptive statistics 3.1. Data sources and sample selection Our primary sample consists of all Japanese firm-quarter observations in the Nikkei NEEDS database for the three-year period between the first quarter of 2015 and the last quarter of Following prior studies (e.g., Bailey et al., 2003; Heflin et al., 2003), we confine our sample to firms with a particular fiscal year-end to align the timing of forecast and earnings releases. Although previous studies using U.S. firms limit their samples to December fiscal year-end firms, we use March fiscal year-end firms because a large portion of Japanese firms employ March fiscal year-end. 14 We obtain data on earnings announcement dates, actual earnings, and firm characteristics from Nikkei NEEDS. We obtain analyst forecast data from IFIS Consensus Data provided by IFIS. IFIS collects analyst reports and provides statistics (e.g., means and standard deviation of analysts forecasts) and rating consensus ranging from 2 (strong sell) to +2 (strong buy). Using the IFIS data, we identify whether the firm is followed by analysts for each quarter. We exclude firms that have analysts only before or after the guidelines because a loss (gain) of 13 Kaisha Shikiho (the English edition is The Japan Company Handbook) (2016) 14 March fiscal year-end firms dominate, on average, more than 70% of listed firms during our sample period. 11

12 analyst coverage increases (decreases) the information asymmetry of the firm (Derrien and Kecskés, 2013). For firms with analyst coverage, we obtain all earnings forecasts for the first quarter of 2014 to the last quarter of We then retain only the last forecast for a given firm s year-quarter, as long as the forecast is released more than 90 days before earnings announcement dates and within five days of an earnings announcement. We require our sample firms to have data from the first quarter of 2015 to the last quarter of We exclude (1) financial firms (Nikkei industry codes 47, 49, and 51), because there is no data available on the ownership of these firms, and (2) firms with no consensus data in any single quarter. Our final sample consists of 17,099 firmquarter observations for 1,585 unique firms. In the analysis of analyst forecasts, we use a sample of firms both covered by JSDA member and non-member analysts (i.e., Nikkei analysts). Nikkei analysts forecasts are obtained from Nikkei NEEDS Measures of market behavior and analyst forecast performance We investigate the impacts of the guidelines on the information environment by using abnormal return volatility. We define abnormal return volatility (Abnormal Return Volatility ) as the sum of the absolute abnormal return over the three days ( 1, +1) around the earnings announcement date (day 0). Stock returns are calculated as closing price at trading day t minus closing price at trading day t 1, divided by closing price at time t 1; the closing price is adjusted for dividend and stock splits. Daily abnormal returns during the event window are defined as the raw return minus the expected return, which is estimated using the single-factor model with TOPIX serving as a proxy for market portfolio. An estimation window is ( 252, 31) trading days prior to the earnings reporting day. To obtain a reliable estimate of the expected returns, we do not estimate the expected returns when the estimation window does not have more than 30 observations; thus, abnormal returns are not calculated in this case. Further, to reduce the impact of extreme values, we winsorize abnormal returns at the 1st and 99th percentiles. For analysts forecast performance, we use forecast accuracy (Forecast Accuracy), which is the absolute value of the firm s actual earnings minus consensus (i.e., the mean of all analyst forecasts). This measure is scaled by market capitalization at the end of the pre-guideline quarter to prevent changes due to changing stock 12

13 price, 15 and it is winsorized at the 99th percentile to mitigate the effect of outliers. This metric becomes smaller as the forecast becomes more accurate Timing of the adoption of the guidelines and its impacts Although the guidelines went into effect in September 2016, anecdotal evidence suggests that some major brokerages had already banned their analysts from interviewing companies even before the implementation of the guidelines. According to Bloomberg News (2016; Japanese edition), Japanese securities firms, including Nomura Holdings Inc., proscribed analysts from interviewing companies before earnings releases starting from the third quarter of Other securities firms, such as Daiwa Securities Group Inc., SMBC Nikko Securities Inc., Mizuho Securities Co., Ltd., and Citigroup Inc., also prohibited their analysts from interviewing corporate managers as of the release of this news article. If this is the case, the impact of the guidelines should be observed from the third quarter of Thus, we define the pre-guideline period as the six quarters from the first quarter of 2015 to the second quarter of 2016 and the post-guideline period as six quarters from the third quarter of 2016 to the fourth quarter of We create a Post indicator variable that equals one if the quarter is after the third quarter of 2016 and zero otherwise. To identify the timing of the guidelines adoption and its impact, it is useful to provide an example of the fiscal-quarter timeline and earnings announcement day for the quarter. Appendix A illustrates the example of Toyota Motor Corporation, a March fiscal-year-end company Descriptive statistics Table 1 provides descriptive statistics for our sample of 17,099 firm-quarters. Panel A reports the summary statistics for firm characteristics, abnormal return volatility, and abnormal trading volume. Firm characteristics include firm size, floating shares, and foreign ownership. We provide detailed definitions of the variables in Appendix B. We winsorize firm size at the 1st and 99th percentiles. The mean and median firm sizes are larger for firms with coverage than for firms without coverage. The average proportion of floating shares is 12% for firms with analysts and 22% for firms without analysts. Average foreign ownership is 23% and 5% for firms 15 In several prior studies (e.g., Agrawal et al., 2006), forecast earnings per share (EPS) has been used for analyst forecast accuracy. In Japan, forecast EPS is less likely to be reported. Thus, we use earnings forecasts. 13

14 with and without analysts, respectively. The mean value of abnormal return volatility is 0.07 for firms with analysts. Panel B reports the summary statistics on the firm and analyst reports over time. The number of firms following analysts during the sample period ranges from 507 to 605. Approximately 40% of our sample of listed firms received analyst coverage. The median number of analysts covering the firm is two and remains constant during our sample period. The average rating consensus is also similar during the sample period, ranging from 0.67 to The median rating consensus is also more than 0.67 over the sample period, suggesting that analysts have an optimistic bias. 16 This optimistic recommendation bias is consistent with previous literature (e.g., Ljungqvist et al., 2007). These results suggest that analysts behavior might not have been affected by the guidelines adoption in terms of coverage and recommendations. However, the number of days between analyst reporting date and the actual earnings announcement date gradually increases during our sample period. The median value jumps at the fourth quarter of This could be because the regulations drive analysts to carefully distance themselves from the management prior to earnings releases Univariate tests Panel A of Table 2 presents the results of the univariate tests of the cross-sectional difference between firms with and without analysts. In addition to the abnormal return volatility and trading volume, we report the threeday market-adjusted cumulative abnormal return responses to the positive and negative earnings surprises. Using the standard event study approach, we calculate cumulative abnormal returns (CARs) over the three-day event window ( 1, +1). In Panel A, the mean abnormal return volatility for firms with analysts is higher than without analysts during our sample period; the differences are statistically significant (except for 2015Q2, 2015Q4, and 2017Q4). However, the result of abnormal volume is inconclusive. Panel B of Table 2 presents the changes in market behavior before versus after the guidelines. Abnormal return volatility increases for firms with analysts, and the increase in return volatility is significant (p = 0.00), while that for firms without analysts remains unchanged. The changes in mean CARs are statistically significant, 16 For instance, three analysts propose ratings of 1 (indicates Sell ), 0 ( Hold ), and +2 ( Strong buy ), so the rating consensus is calculated as 0.33 (i.e., the average of 1, 0, and +2). 14

15 suggesting an increase in the magnitude of stock price reaction after the adoption of the guidelines. Panel A of Table 3 reports the results of comparisons of earnings forecasts between JSDA member and nonmember analysts from the second quarter of 2015 through the last quarter of The forecast horizon of the second quarters is a half-year forecast; the fourth quarters is the annual forecast. The table presents the mean and median forecast accuracy and dispersion. Analyst forecast dispersion is the standard deviation of the most recent analysts earnings forecasts. We note that for Nikkei analysts, forecast dispersion is not defined because Nikkei analyst is the only non-member analyst in our sample. The mean value of JSDA analysts forecast accuracy is larger than that of Nikkei analysts, but the difference is statistically insignificant. The median forecast accuracy takes a similar value; the difference is significant at the 10% level. In Panel B of Table 3, we compare the forecasts of member and non-member analysts in the pre- and postguideline periods. The mean forecast accuracy decreases after the introduction of the guidelines for both JSDA and Nikkei analysts, but the decrease in forecast accuracy is statistically insignificant. The result for forecast dispersion is not uniform: the mean of forecast dispersion increases after the implementation of the guidelines and is statistically significant (p-value = 0.081), whereas the median value decreases, but is insignificant (pvalue = 0.134). Overall, indirect evidence shows that there is no significant deterioration in analyst forecast accuracy after the adoption of the guidelines, suggesting that the guidelines may not have adversely affected the information available. 4. Methodology and results In this section, we employ a regression framework to formally test whether the guidelines changed the information environment. We begin this section by analyzing the impact of the guidelines on the return volatility around earnings releases. In the second part, we investigate the impact of the guidelines adoption on analyst forecasts Comparisons between firms with and without sell-side analysts in the pre- and post-guideline periods An earnings preview ban makes it difficult for analysts to obtain preview numbers and report the information 15

16 to their clients. If this is the case, the value-relevant information should be concentrated around the earnings announcement date after the guidelines adoption. To test for the impact of the guidelines adoption on return volatility, we employ a difference-in-differences estimation framework (Roberts and Whited, 2013). Specifically, using the combined sample of firms with and without analysts, we estimate the following model: Abnormal Return Volatility i,t = β 0 + β 1 Analyst Following i + β 2 Post t + β 3 Analyst Following i Post t γ Controls i,t + δ q Quarter q + θ d Industry d + ε i,t, q=1 d=1 (1) where Abnormal Return Volatility i,t is the abnormal return volatility around firm i s quarter t earnings announcement. Analyst Following i is an indicator variable equal to one if firm i is covered by analysts and zero otherwise, Post t is an indicator variable equal to one if quarter t is after the third quarter of 2016 and zero otherwise, and Analyst Following i Post t is an interaction variable created from these two dummy variables. The permanent cross-sectional differences in abnormal return volatility between firms with and without analysts are differenced away by the inclusion of the Analyst Following i dummy. The time-series differences in abnormal return volatility of all firms before and after the guidelines are captured through the Post t dummy. Our main variable of interest is the difference-in-differences estimator, Analyst Following Post. The coefficient β 3 is our estimate of the unique impact of the guidelines adoption. If the guidelines reduce sell-side analysts information advantage by preventing them from previewing numbers, the valuerelevant information is delayed until the earnings announcement date; thus, the coefficient of the interaction term should be positive after controlling for several factors. Controls i,t represents a set of control variables including observed firm characteristics firm size and ownership structure. As shown in Table 1, there is a substantial difference in size between firms with and without analysts. Hence, in the regression, we control for firm size, which is the log of the market capitalization at the end of quarter (Firm Size), which is also a crucial determinant of analyst coverage, as Hong et al. (2000) show. 16

17 Further, we capture the cross-sectional differences in ownership structure by including Floating Shares and Foreign Ownership. To difference away the quarterly trend, we include fiscal-quarter dummies. In addition, we include industry dummies to control for cross-industry differences. ε i,t is the error term. We estimate equation (1) using ordinary least squares (OLS) regression with robust standard errors clustered at the firm. Before turning to the regression analysis, it is important to confirm a crucial assumption behind the differencein-differences estimator: the parallel trends assumption, which requires that treatment and control groups have parallel trends in outcome variables if the treatment is absent (see Roberts and Whited, 2013). In our context, if the guidelines adoption is absent, the average difference in changes between outcome variables across firms with and without analysts would be the same for the post-guideline period as the pre-guideline period. The parallel trends assumption, however, is formally untestable because we cannot observe the counterfactual of the guidelines not having been introduced. Instead, we check the internal validity of the difference-in-differences estimates by plotting the average outcomes for the treatment and control groups over time. Figure 1 illustrates the average level of abnormal return volatility around the guidelines adoption. The abnormal return volatility of firms with analysts (treated firms) is constantly higher than that of firms without analysts (control firms). The outcome variables between these two groups exhibit similar trends before the guidelines adoption. 17 Thus, we conclude that our sample will be used in the regression analysis because it seems to satisfy the necessary condition of the similar trends in outcomes in the pre-treatment period. The figure also shows that the increased volatility is temporal: the abnormal return volatility increases from the third quarter of 2016 to the first quarter of 2017 (2016Q3 2017Q1) and decreases following the first quarter of This pattern is consistent with Mathew et al. (2004), who document that the impact of Reg FD on information flow is temporal. Interestingly, we find that non-analyst coverage firms volatility also increases in the third quarter of Firms without analysts are not related to the guidelines; thus, the increase in return volatility might be caused by economic shocks that could affect the return volatility of firms with analysts as well. 17 We perform a formal statistical test of the difference in change across firms with and without analysts in the pre-guideline period. We find that the differences are not significantly different from zero. 17

18 4.2. Empirical results: stock price reactions We now turn to our multivariate regressions. Table 4 presents the difference-in-differences estimates. Panel A presents the results of the pooled OLS. We provide the results for several comparison periods. Column 1 reports the empirical results for 2016Q3 2017Q4 versus 2015Q1 2016Q2; column 2 reports the results for 2017Q1 Q4 versus 2015Q1 Q4; column 3 reports the results for 2017Q1 Q2 versus 2015Q1 Q2; column 4 reports the results for 2017Q3 Q4 versus 2015Q3 Q4. Given any comparison period, the estimated coefficients for the Analyst Following dummy are positive and statistically significant, suggesting that the return volatility of firms with analyst coverage is higher than that of firms without analyst coverage during our sample period. The estimated coefficients for the Post dummy are positive and statistically significant in columns 1 and 4, suggesting that abnormal return volatility of all firms increases after the adoption of the guidelines. More importantly, we find that the coefficient on the interaction term Analyst Following i Post t is positive and statistically significant in all regressions. These results indicate that the abnormal return volatility of firms with analysts significantly jumps after the guidelines adoption relative to firms outside the coverage. The impact is economically large and conspicuous. For instance, in column 1 of Panel A, we find that firms with analysts experience a 0.7% increase in return volatility after the introduction of the guidelines. Turning to the control variables, a significant negative coefficient on firm size in all columns indicates that a larger firm has a lower abnormal return volatility; this result is intuitive and consistent with the result in Bailey et al. (2003). The results show that the ownership structure of firms can influence stock return volatility. The coefficient estimates on Floating Shares are significantly negative, suggesting that the stock return is less volatile when the stock is held by small shareholders. The coefficient estimates on Foreign Ownership are significantly positive, suggesting that foreign investor holding leads to higher stock return volatility. To confirm whether the result in Panel A is caused by specification problems, we consider a panel specification. Panel B presents the results of the firm fixed effects model. Note that the analyst-following dummy, Analyst Following i, is absorbed by the firm fixed effects. We find that the results are essentially unchanged: the interaction term between the analyst-following and post dummies is significantly positive. Overall, the above results indicate that the guidelines drive a concentration of information flows on the day 18

19 of an earnings announcement, because they prevent analysts from diffusing information through earnings preview reports. This leads to the media report that JSDA guidelines cause higher return volatility surrounding an announcement A falsification test A potential concern regarding the results in Table 4 is that the change in abnormal return volatility after the guidelines could be caused by a macro shock occurring in the pre-guideline period, rather than the JSDA guidelines. To confirm this possibility, we conduct a falsification test. Specifically, we falsely assume that the onset of treatment occurs several quarters before it actually does. We create dummy variables for placebo periods occurring several quarters prior to the adoption of the guidelines: Post Placebo p equals one if quarter t is after the quarter p, where p = 2015Q2, 2015Q3, and 2015Q4, and zero otherwise. Table 5 presents the results of the falsification test. In the regression, we use the subsample of the period between the first quarter of 2015 and the second quarter of The interaction term between the analystfollowing and post-placebo dummies is statistically insignificant. This result suggests that an increase in return volatility is not caused by macro-economic shock in the pre-guideline period but by the guidelines adoption Comparisons between member and non-member analyst forecasts in the pre- and post-guideline periods To assess the guidelines impact on analyst forecasts, we compare JSDA member and non-member analysts forecasts for the same firms before and after the introduction of the guidelines. For non-members forecasts, we use the forecasts of analysts belonging to Nikkei (i.e., Nikkei analysts). 18 Nikkei releases half-year or annual earnings forecasts in the Nikkei newspaper prior to a company s official release. In the analysis, it is important to consider the possibility that companies would change their behavior in response to the guidelines. For instance, companies may increase information quality or quantity (see, e.g., Bailey et al., 2003). Mohanram and Sunder (2006) show that the information environment in which analysts operate has changed in the post-reg FD period: absolute change in gross domestic product and the proportion of firms disclosing bad news have increased. To control for this change in corporate behavior, we use the sample of firms with both JSDA and Nikkei analyst 18 In this paper, we use Nikkei analyst forecasts as non-member analysts because Toyo Keizai s The Japan Company Handbook is a quarterly publication, and it is not available for examining the stock reactions to forecasts. 19

20 forecasts and compare the difference in forecast accuracy between these analysts. Specifically, we estimate the following model: Diff. Forecast Accuracy i,t 3 32 = β 0 + β 1 Post t + γ Controls i,t + δ q Quarter q + θ d Industry d + ε i,t, q=1 d=1 (2) where Diff. Forecast Accuracy i,t is the difference in forecast accuracy between securities firms and Nikkei analysts for firm i at quarter t. Member and non-member analysts might be expected to differ in forecast ability. We capture the difference in ability through an intercept, β 0. The estimated coefficient, β 1, captures the change in the JSDA analysts forecasts relative to non-member analysts forecasts associated with the guidelines. If the guidelines could reduce the information advantage of sell-side analysts, we expect that β 1 should be positive. Controls i,t is a vector of the control variables. It includes firm size (Firm Size i,t ), forecasting horizon (Forecast Horizon i,t ), the days covered between JSDA analysts forecast release date and Nikkei analysts forecast release date (Forecast Lag i,t ), an after-nikkei-analyst forecast dummy (After Nikkei Forecast i,t ), and the number of analysts following (No. of Analysts i,t ). Firm size is a proxy for the firm s information environment. 19 As we expect analysts forecasts to worsen as the forecast horizon increases, we control for forecast horizon. As before, we include quarter and industry dummies, allowing us to control for possible seasonality in analyst forecasts (i.e., half-year or annual forecast) and cross-industry differences. ε i,t is the error term. We cluster the standard errors at the firm level Empirical results: analyst forecast accuracy Panel A of Table 6 presents the OLS regression results when the difference in forecast accuracy is the dependent variable. We find that the intercept β 0 is positive and statistically significant, suggesting that the 19 We re-estimate equation (2) after replacing firm size with the number of analysts, which serve as a proxy for the firm s information environment, and obtain similar results. The correlation of firm size and number of analysts is 0.67, indicating that there is a positive relationship between these variables. 20

21 forecast accuracy of Nikkei analysts is better than JSDA member analysts. We find that the estimated coefficients of the post-guideline dummy are positive but insignificant. This result suggests that member analysts forecast accuracy is not impaired relative to non-member analysts forecast accuracy after the adoption of the guidelines. We find significant negative coefficients for After Nikkei Forecasts, indicating that when member analysts forecasts are released after the Nikkei forecast, the difference in forecast accuracy between these analysts narrows. Panel B of Table 6 reports the time series of estimated coefficients and standard errors on the intercept. The estimated coefficient on the intercept is positive after the fourth quarter of This result suggests that the forecast accuracy of member analysts becomes worse than that of Nikkei analysts. Overall, we find no evidence of analyst forecasts becoming less accurate, on average, after the guidelines adoption. This is either because analysts use alternative information sources or because companies increase the disclosure of public information. Our empirical findings, however, oppose the latter possibility, as we compare forecasts for the same companies. If companies change their behavior because of the guidelines, the change in the information environment should affect all analysts unilaterally. Therefore, our results show that sell-side analysts, who are subject to the guidelines, rely on alternative information sources such as media analyst forecasts. 5. Additional analyses In this section, we investigate how member analysts, who are subject to the guidelines, are able to maintain their forecast accuracy after the implementation of the guidelines. We then examine the informativeness of analyst reports. Finally, we examine the impact of the guidelines on analyst forecast dispersion Information source of sell-side analysts after the introduction of the guidelines We find evidence that sell-side analysts rely on alternative information sources after the guidelines adoption. Figure 2 indicates the distribution of the forecast release timing by the sell-side analysts around the Nikkei analysts forecast release day. The horizontal axis is the difference between the dates when the Nikkei analysts 21