Dark Trading Volume at Earnings Announcements

Transcription

1 Dark Trading Volume at Earnings Announcements Xanthi Gkougkousi U.S. Securities and Exchange Commission Wayne R. Landsman University of North Carolina Kenan-Flagler Business School The Securities and Exchange Commission, as a matter of policy, disclaims responsibility for any private publication or statement by any of its employees. The views expressed herein are those of the author and do not necessarily reflect the views of the Commission, of any Commissioners, or of the author s colleagues on the staff of the Commission. This paper is part of the Division of Economic and Risk Analysis Working Paper Series. Papers in this series are the work of the authors and not the work of the Division or the Commission. Inclusion of a paper in this series does not indicate a Division or Commission determination to take any particular action or position. References to this paper should indicate that the paper is a DERA Working Paper. We thank Miguel Ferreira, Abby Kim, Alina Lerman, Salil Pachare, Adam Reed, Eric So, Steve Stubben, Peeyush Taori, Vincent van Kervel, Teri Yohn, Donny Zhao, Wei Zhu, two anonymous reviewers, and the workshop participants at the 2018 American Accounting Association Annual Meeting, the 2018 Colorado Summer Accounting Research Conference, 2018 European Accounting Association Annual Congress, Nova School of Business and Economics, the University of North Carolina at Chapel Hill, and the University of Utah for helpful comments.

2 Dark Trading Volume at Earnings Announcements Abstract We examine how abnormal dark market share changes at earnings announcements and find a statistically and economically significant increase in abnormal dark market share in the weeks prior to, during, and following the earnings announcement. The increase in dark market share is larger for firms with a relatively high quality of information environment and a relatively low level of informed trading. These findings are consistent with informed (uninformed) traders facing lower execution (adverse selection) risk in dark venues for firms with a high quality information environment and for firms with low levels of informed trading. Keywords: dark venue; earnings announcement; trading volume; quality of information environment; informed trading. JEL Classifications: G12, G14, D47, M41.

3 1. INTRODUCTION Dark venues are electronic trading platforms operated by broker/dealers that match buy and sell orders without the intermediation of market makers using prices derived from the exchanges. Dark venues differ from exchanges in terms of pre-trade transparency because, in contrast to exchanges, dark venues do not provide information on order prices, order volumes, and the direction of trade interest. In addition, post-trade dark venue transparency is limited because market participants can only observe trading volume by dark venue with a delay and aggregated over a week. Dark trading has increased in recent years dark trading volume as a percentage of total trading volume has increased from 5 percent in January 2008 to 17 percent by September 2016 (Rosenblatt Securities 2016). Given the rapid growth of dark trading in recent years, it is important to understand what determines the split of trading between dark and lit trading venues both over time and across firms. To this end, we examine three related research questions. First, does the release of information affect the portion of trades executed in dark as opposed to lit trading venues? Understanding where trades are executed around information releases is particularly important in light of regulatory concerns that dark pool trading can detract from market quality and price efficiency (U.S. Securities and Exchange Commission (SEC) Chair White 2014). Although our study does not address directly the market quality and price discovery implications of changes in dark market share, it does so indirectly by providing evidence of changes in dark market share around information releases when trade execution and price discovery likely are more important. Second, is the firm s pre-announcement quality of information environment associated with changes in dark market share around information releases? Understanding the relation between the pre-announcement quality of information environment and the change in dark market share around earnings releases is important because our findings could 1

4 either alleviate or strengthen regulators concerns about the impact of dark trading on market quality and price discovery. Third, is the change in dark market share around earnings releases associated with the level of informed trading around the announcement of earnings? Such an examination can contribute to assessing whether the change in dark market share is attributable to informed traders, uninformed traders, or both, and can provide evidence on whether the relation between the change in dark market share and the quality of information environment is attributable to differences in informed trading for firms with relatively high and low quality information environments. Understanding where informed and uninformed traders trade around information releases has important implications for market quality and price discovery. We measure information releases as the announcement of quarterly earnings because earnings announcements are major corporate news releases that have significant price and volume effects (e.g., Ball and Brown 1968; Beaver 1968). In addition, earnings releases are mandatory announcements and thus are less subject to reverse causality concerns compared to other types of information releases whose occurrence, timing, and/or content is more discretionary. 1 Although there is ample theoretical and empirical evidence that earnings announcements are associated with an increase in aggregate trading volume, there is a paucity of evidence of the impact of earnings announcements on the proportional trading volume between dark and lit trading venues. Whether dark market share increases or decreases at earnings announcements depends on how the relative benefits and costs of dark and lit trading change for informed and uninformed investors around the announcement of earnings. 2 1 For example, Cheng and Lo (2006) shows that insider trading considerations affect the frequency of earnings management forecasts, and Luo (2005) shows that managers learn from the stock market reaction to merger and acquisition (M&A) announcements when making M&A closing decisions (see also Sletten 2012; Li and Zhang 2015; Zuo 2016). 2 Throughout the paper, we use the term uninformed investors to denote all investors, i.e., not just retail investors, who trade for motives other than to exploit their informational advantage. Uninformed investors can trade for motives such as portfolio rebalancing, tax, or liquidity reasons. 2

5 In contrast to prior literature that examines changes in dark market share in the minutes following the earnings announcement (i.e., Menkveld, Yueshen, and Zhu 2017), we use lower frequency weekly data to examine how dark market share changes around earnings announcements for the following reasons. First, prior studies show that abnormal trading volume remains elevated in the days following the announcement of earnings (e.g., Campbell, Ramadorai, and Schwartz 2009; Kaniel, Liu, Saar, and Titman 2012; Landsman, Maydew, and Thornock 2012). Second, other studies show that different traders choose to trade at different points in time following an earnings announcement. For example, Dey and Radhakrishna (2007) shows that institutions initiate trades immediately after the earnings announcements while individuals initiate trades later on in the earnings announcement day and the day after (see also, Anderson, Harris, and So 2007). Third, the timing of investors trades is associated with the weight investors place on the relative benefits and costs of dark versus lit trading venues. For example, a trader that wants to trade immediately following the earnings announcement likely will place greater weight on the trade immediacy offered by lit venues, while a more patient trader likely will place greater weight on the price improvement and pre-trade opacity offered by dark venues. Hence, the examination of changes in dark market share around earnings announcements during longer time windows using weekly data can shed light on the trading preferences of different types of investors than the examination of changes in dark market share around earnings announcements during shorter time windows. 3 Fourth, extending the event window to include the week prior to the announcement of earnings permits us to determine the extent of anticipatory trading by investors. 3 Related to this point, Comerton-Forde and Putniņš (2015) uses both high and low frequency data to examine the relation between dark trading and informational efficiency because high frequency measures are likely to capture illiquidity as well as informational efficiency. Low frequency measures are less likely to suffer from illiquidity issues. 3

6 We measure abnormal dark market share as the difference between actual dark market share and the average of dark market share during the six weeks starting seven weeks and ending two weeks prior to the week of the earnings announcement scaled by the standard deviation of dark market share measured over the same period. Dark market share is the weekly ratio of dark trading volume divided by consolidated trading volume. Regarding our first research question, our multivariate analysis shows that abnormal dark market share increases by 13%, 40%, and 11% in the weeks prior to, during, and following the earnings announcement, and these increases are both statistically and economically significant. Our finding of an increase in dark market share around earnings announcements is consistent with dark venues becoming more attractive to traders around earnings releases. Regarding our second research question, we examine the relation between the firms pre-announcement quality of information environment and the changes in abnormal dark market share around earnings announcements. We measure the pre-announcement quality of information environment in several ways including analyst forecast error, two composite metrics based on analyst coverage, analyst forecast dispersion, and earnings transparency, and option implied volatility. We find that although dark market share increases both for firms with a relatively low and relatively high quality of information environment, the increase in abnormal dark market share is between 67% and 87% larger depending on the measure of quality of information environment used for firms with a high versus a low quality of information environment. This finding is consistent with execution risk being higher for informed traders and adverse selection risk being higher for uninformed traders in dark venues for firms with a low versus high quality of information environment. Regarding our third research question, we examine the relation between the change in abnormal dark market share around earnings releases and the level of informed trading. We measure informed trading as the ratio of the quarterly absolute changes in holdings by 4

7 informed institutional investors to the sum of the quarterly absolute changes in holdings by informed and uninformed institutional investors. We classify institutional investors into informed and uninformed investors using the Bushee and Goodman (2007) classification. We find an increase in abnormal dark market share both for firms with no informed trading and for firms with no uninformed trading around earnings announcements, but the increase in dark market share is decreasing in the level of informed trading. These results suggest that the positive relation between the changes in dark market share and the quality of information environment is attributable to different levels of informed trading for firms with high and low quality of information environment. These results also suggest that both informed and uninformed traders contribute to the increase in dark market share around earnings announcements. However, as the number of informed relative to uninformed traders increases, informed traders impose higher execution risk on other informed traders and/or higher adverse selection risk on uninformed traders, which leads to a lower increase in abnormal dark market around earnings releases. Finally, these results are consistent with the Zhu (2014) theoretical prediction that when adverse selection is high, an increase in adverse selection could cause a decrease in dark pool market share. Our study contributes to the literature on inter-market competition (e.g., Menkveld et al. 2017) by examining how information releases alter the net benefits of trading in dark versus lit trading venues. Our finding of an increase in dark market share around earnings releases is relevant not only to academics but also to investors because it suggests that dark venues could be platforms where traders seek liquidity around earnings releases. Our finding could also be relevant to exchange and dark venue operators who could be interested in understanding the circumstances under which order flow is diverted from lit to dark venues, which can ultimately affect venue profitability. Finally, our findings are relevant to academic 5

8 researchers because they highlight the importance of accounting for different venues when studying and interpreting trading volume around information releases. Our results on the relation between the firms pre-announcement quality of information environment and the change in dark market share around earnings releases contribute to the literature on the role of information asymmetry on price and volume reactions to information releases (e.g., Kim and Verrecchia 1991) and the role of information asymmetry in the choice of trading venues (e.g., Zhu 2014). One advantage of our research methodology relative to prior studies that examine the relation between information asymmetry and the choice of trading venues is that we examine the relation between the change in dark market share and information asymmetry, which potentially mitigates omitted variables concerns that are more problematic for studies that examine the relation between the level of dark market share and information asymmetry. In addition, assuming that any concerns about detrimental effects of dark market share are more pronounced for firms with low quality of information environment, our finding that the increase in dark market share around earnings releases is lower for firms with low quality partially alleviates these concerns. Our paper also contributes to the literature that examines the trading venue choices of informed and uninformed traders (e.g., Comerton-Forde and Putniņš 2015) by providing indirect evidence that the increase in dark market share around earnings releases is attributable to both informed and uninformed trading. 4 Finally, our findings on the interaction between information releases, quality of information environment, and informed trading could be relevant to regulators who could consider these factors when making policy decisions, such as requiring increased pre-trade transparency for securities that exceed a 4 See section for a discussion of the relation of our study to Comerton-Forde and Putniņš (2015) and Menkveld et al. (2017), both of which also study empirical aspects of dark trading. 6

9 certain threshold of dark trading. For example, the U.S. SEC has considered altering the trading volume threshold that triggers public display obligations of order information for dark venues from 5 percent to 0.25 percent for each stock (U.S. SEC, Regulation of Non-Public Trading Interest, 2009). 2. BACKGROUND AND PREDICTIONS 2.1 Institutional Background Equity trading can be classified broadly into on- and off-exchange trading. Both onand off-exchange trading is supervised and regulated by the U.S. SEC. On-exchange trading comprises trading on venues that provide both pre- and post-trade transparency: order prices and order volumes are publicly displayed prior to trade execution, and trade details are publicly disseminated in real time. As of March 2012, there were 13 active U.S. stock exchanges that executed approximately 67 percent of the consolidated equity trading volume (Chartered Financial Analyst Institute (CFA) 2012). Off-exchange trades are all trades executed away from exchanges, and can be classified into non-alternative Trading System (ATS) and ATS trades. Non-ATS trades, also known as internalization trades, are trades that broker/dealers execute internally either as agents or principals. ATS trades are trades executed on electronic trading platforms that are registered as broker/dealers and match buy and sell orders without the intermediation of market makers. Off-exchange trading accounted for approximately 33 percent of the consolidated equity trading volume as of March 2012; 18 percent is regarded as non-ats trades and 15 percent is regarded as ATS trades (CFA 2012). Off-exchange trading mainly differs from on-exchange trading in terms of trade transparency. Regarding pre-trade transparency, there is no order book information i.e., order prices, order volumes, and the direction of trade interest for the vast majority of offexchange trades. Regarding post-trade transparency, prior to May 2014, all off-exchange 7

10 trades were reported on an aggregate basis on the consolidated tape. Hence, prior to May 2014, market participants were unable to distinguish between non-ats and ATS trades, and they were also unable to identify in which ATS venue or by which broker/dealer each trade was executed. As of May 2014 (April 2016), Financial Industry Reporting Authority (FINRA) publicly disseminates information on ATS (non-ats) trades by security and electronic trading venue (broker/dealer). The level of post-trade transparency for offexchange trades, however, still remains lower than the level of post-trade transparency for on-exchange trades because market participants have access to off-exchange trade information with a delay FINRA reports trades on Tier 1 NMS stocks with a two-week delay and all other stocks with a four-week delay and the trade information is only available at a weekly frequency. 5 In contrast, market participants have access to trade-by-trade information for trades executed on exchanges almost instantaneously (i.e., within 90 seconds from the time of the trade execution). There are two types of ATS trades: ATS trades executed on dark venues and ATS trades executed on Electronic Communication Networks (ECNs). The main difference between dark venues and ECNs is that dark venues do not provide pre-trade transparency, while ECNs publicly display their order book. The majority of ATS trades relate to dark trades as of March 2012, 88 percent of the ATS trades were trades executed on dark venues, while the remaining 12 percent were trades executed on ECNs (CFA 2012). Dark venues offer the following three main advantages over lit venues. First, dark venues allow investors to execute trades with minimum information leakage and price 5 There are three types of equity securities: NMS Tier 1, NMS Tier 2, and OTC securities. NMS securities are the exchange-listed equity securities and OTC are the over-the-counter securities. NMS Tier 1 stocks are stocks included in the S&P 500, Russell 1000, and selected exchange-traded products, and NMS Tier 2 stocks are all other NMS stocks. 8

11 impact. 6 For this reason, dark venues were initially primarily used by institutional investors to execute block trades. Presently, however, dark venues are used not only by institutional investors to execute block trades but also by other investors to execute smaller trades (Tuttle 2013). Second, dark venues provide price improvements relative to lit venues because the majority of dark trades are executed at mid-point. 7 Third, dark venues, in contrast to exchanges, are allowed to restrict access to certain traders, such as institutional investors or high-frequency traders, and these access restrictions can provide protection to certain traders from informed or predatory trading. The main disadvantage of dark venues is the trade execution uncertainty liquidity is low and there is no market maker on dark venues to absorb the excess order flow. As a result, fill rates differ substantially for dark and lit venues. For example, Tuttle (2013) shows that the fill rate for orders submitted on dark venues ranged from 0.03% to 8.38% in In contrast, Battalio, Corwin, and Jennings (2016) shows that the fill rates for at-the-quote limit orders, i.e., limit orders with prices equal to the NBBO, submitted on exchanges ranged from 48.77% to 74.38% in Related Literature and Empirical Predictions Abnormal Dark Market Share around Earnings Announcements Empirical literature provides evidence of increased trading volume around earnings announcements (e.g., Beaver 1968; Bamber 1987; Landsman and Maydew 2002). Empirical and theoretical studies suggest that trading volume increases around earnings releases because both informed trading (e.g., Demski and Feltham 1994; McNichols and Trueman 6 Investors can also hide their trade intention by submitting hidden orders on exchanges. The SEC staff estimates that hidden volume, i.e., the total volume of trades executed against hidden orders divided by the total volume of all trades, was between 5 and 25% for the various exchanges as of March 14, 2017 (U.S. SEC 2017). 7 The operators of dark venues that match trades at the mid-price do not earn a bid-ask spread and thus rely on commissions and fees to generate revenue. Dark venue operators can generate additional revenue by engaging in proprietary trading using the information in the observed dark order flow. Dark venue operators also enjoy cost savings because they can match buy and sell client orders and thus can avoid paying fees to exchanges when routing orders for execution. 9

12 1994; Christophe, Ferri, and Angel 2004; Hendershott, Livdan, and Schurhoff 2015) and uninformed trading increase (e.g., Admati and Pfleiderer 1988; Kross, Ha, and Heflin 1994; Bushee and Goodman 2007). Informed trading increases around the announcement of earnings because of differential belief revision caused by differences in investors preannouncement beliefs (e.g., Kim and Verrecchia 1991) and because of differences in investors interpretation of the public announcement (e.g., Kandel and Pearson 1995; Kim and Verrecchia 1997). Uninformed trading increases around the announcement of earnings primarily as a result of portfolio rebalancing motives earnings announcements reveal information about a firm s systematic risk (Patton and Verardo 2012), which may lead uninformed traders to trade more around earnings releases to realign their portfolio with their risk preferences. 8 The impact of earnings announcements on aggregate trading volume is relatively well studied. The impact of earnings announcements on the proportional trading volume between dark and lit venues (i.e., dark market share), however, is less well understood. The change in dark market share around earnings releases depends on how the relative benefits and costs of trading in dark and lit trading venues change for informed and uninformed traders around the earnings release. 9 Dark market share can increase around earnings releases for several reasons. First, at earnings announcements, spreads widen in anticipation of the news release (e.g., Lee, 8 Tax incentives also can affect trading volume around earnings releases. In particular, investors might sell stocks following bad earnings news to realize capital losses, and they might delay selling stocks to rebalance their portfolio following good earnings news because investors face higher tax rates on short-term relative to long-term capital gains (e.g., Blouin, Raedy, and Shackelford 2003). Nevertheless, according to Blouin et al. (2003, p. 613) capital gains tax effects, if they matter at all, are not dominant determinants of equity trading. 9 Retail investors often delegate the choice of trading venue to their broker/dealer. Literature suggests that broker/dealers may route retail orders to certain venues to maximize order flow payments while negatively affecting client order execution quality (e.g., Battalio et al. 2016). We have no data available that would allow us to infer whether the choice of dark venues around earnings announcements is related to dark venues order flow payment schedule. Also, to the best of our knowledge, there is no evidence that dark venues offer different order flow payment schedules compared to lit venues or non-ats broker/dealers. In addition, broker/dealers owe to customers the duty of best execution and the broker/dealer market is competitive. For all these reasons, we do not consider the role of broker/dealers in venue selection in our hypothesis development, and we assume that broker/dealer routing decisions reflect primarily investor preferences. 10

13 Mucklow, and Ready 1993; Engelberg, Reed, and Ringgenberg 2012; Rogers, Skinner, and Zechman 2017). This occurs because information asymmetry increases around earnings releases, and liquidity providers, i.e., market makers and limit order traders, widen spreads to actively manage this increase in risk. Dark venues become more attractive to both informed and uninformed traders as spreads widen because trades on dark venues usually are executed at mid-point (see, e.g., Zhu 2014). Second, in contrast to exchanges, dark venues are allowed to restrict access to certain traders as long as the restrictions are fair and non-discriminatory (Code of Federal Regulations (b)(5)). Indeed, many dark pools seek to restrict access to certain trader types, such as institutional investors, broker/dealers, or high-frequency traders (Boni, Brown, and Leach 2013). Depending on the type of access restriction, dark venues can become more attractive to uninformed and informed traders around earnings releases. In particular, dark venues that restrict access to institutional investors can become more attractive to uninformed traders around earnings releases because the cost of trading against an informed counterparty increases around the announcement of earnings. In addition, dark venues that restrict access to high-frequency traders can become more attractive to informed institutional investors because the cost of being front run by high-frequency traders increases around the announcement of earnings. 10 Third, dark market share can increase around earnings announcements because of a clientele effect: institutional investors frequently trade in the dark (e.g., Buti et al. 2016) and these investors trade relatively more compared to other types of investors around earnings releases (e.g., Cready 1988; Lee 1992; Dey and Radhakrishna 2007). Institutional investors 10 According to some industry insiders, dark pools are in practice open to most investors (Ye 2016). In addition, the SEC has settled enforcement actions against dark pool operators for secretly allowing affiliated highfrequency traders and broker/dealers to access their venues (U.S. SEC Commissioner Aguilar 2015). Hence, it is unclear whether access restrictions can alter the perceived benefits of trading in dark venues around earnings releases. 11

14 prefer to trade in the dark rather than on exchanges because dark venues allow institutional investors to trade large blocks of stock with limited price impact. Fourth, the lack of pre-trade transparency in dark venues can become more valuable for uninformed investors that trade for portfolio rebalancing purposes around earnings releases. The reason is that the lack of pre-trade transparency protects uninformed investors from being front-run on their trades. Hence, uninformed investors that trade for portfolio rebalancing purposes around earnings releases can increase their dark pool participation, and thus increase dark market share. Fifth, the lack of pre-trade transparency in dark venues likely becomes more valuable for informed traders around earnings releases, because the benefits of hiding private information increase around the announcement of earnings (see, e.g., Ye 2011). Hence, informed traders might choose to increase their participation in dark venues around earnings releases, and thus increase dark market share. However, there are several reasons why dark market share can decrease around earnings releases. First, the increase in informed trading around earnings announcements increases information asymmetry and adverse selection risk for uninformed investors in both dark venues and exchanges. The increase in adverse selection risk, however, can become more costly for uninformed investors in dark venues rather than exchanges because dark venues are less transparent than exchanges (Zhu 2014). The increase in adverse selection risk is more costly when pre-trade transparency is low because uninformed investors cannot learn about the trading intentions of informed investors by observing the order book. Hence, uninformed traders might decrease their participation in dark venues around earnings releases, and thus decrease dark market share. Second, informed traders might be more concerned about execution risk in dark venues around earnings releases compared to non-announcement periods because their 12

15 urgency to trade is higher (Menkveld et al. 2017). Trade urgency increases around information releases because the upcoming release of information can eliminate the traders informational advantage. Hence, informed investors that are more concerned about execution risk around earnings releases compared to non-announcement periods might decrease their participation in dark venues, which can result in a decrease in dark market share. Third, informed trading increases around the announcement of earnings, and this increase in informed trading can increase execution risk in both dark venues and exchanges because informed investors tend to trade in the same direction creating greater order imbalances (see also, Zhu 2014). The increase in execution risk, however, likely is more pronounced in dark venues rather than exchanges because dark venues are less liquid and there is no market maker to absorb the excess order flow. Hence, informed traders might decrease their participation in dark venues around earnings releases, which can result in a decrease in dark market share. Fourth, investors can mitigate execution risk in dark venues by sequentially placing orders in various venues until their order is fully filled. This sequential placement of orders, however, can increase the risk of information leakage. Informed investors might be more concerned about information leakage around earnings releases compared to nonannouncement periods, which might deter them from seeking liquidity in dark venues around earnings releases, thereby decreasing dark market share. 11 Because there are compelling reasons for increases and decreases in abnormal dark market share around earnings announcements, we make no prediction regarding the direction of the change in abnormal dark market share around earnings announcements. 11 ATS generally operate during the regular trading hours set by the primary market for each security. As a result, any change in abnormal dark market share around earnings announcements is likely not attributable to different hours of operations for lit and dark venues (see, e.g., Form ATS for JPM-X dated January 30, 2018 and Form ATS for XE dated March 5, 2018). 13

16 H1: There is no change in abnormal dark market share around earnings announcements. An increase in dark market share is also consistent with a contemporaneous study, Balakrishnan and Taori (2018), which finds an increase in dark market share around scheduled and unscheduled news releases, measured using earnings announcements and analyst forecast revisions, respectively. 12 Finally, an increase is consistent with the findings of Chakrabarty and Shaw (2008), which documents an increase in hidden orders on an ECN around earnings announcements. 13 A decrease in dark market share around earnings announcements is consistent with Menkveld et al. (2017), which finds a decrease in dark trading volume and an increase in lit trading volume in the minutes following VIX shocks, macroeconomic announcements, and earnings releases Quality of Information Environment, Informed Trading, and Abnormal Dark Market Share around Earnings Announcements Next, we examine the relation between changes in dark market share around earnings announcements and the pre-announcement quality of firms information environment and the level of informed trading Pre-announcement quality of information environment Quality of information environment is typically defined as the precision of information that is available to investors about the firm s value (e.g., Holthausen and Verrecchia 1988), with greater precision of information implying higher quality of information environment. The relation between the pre-announcement quality of information 12 Two contemporaneous related studies, Pan (2017) and Thomas, Zhang, and Zhu (2018), examine the relation between dark trading and price discovery. 13 Chakrabarty and Shaw (2008) examines hidden trades executed in lit venues rather than dark trades. Prior research shows that hidden trades are not perfect substitutes for trades executed on dark venues and are affected by different market conditions than dark trades (Degryse, Karagiannis, Tombeur, and Wuyts 2015). In addition, the presence and magnitude of hidden orders can be predicted to a significant degree (Bessembinder, Panayides, and Venkataraman 2009). 14

17 environment and the changes in dark market share around earnings announcements is unclear a priori. There are at least two reasons to expect a negative relation between the quality of a firm s information environment and the change in dark market share around earnings announcements. First, the quality of a firm s information environment is negatively related to firm-specific information asymmetry and adverse selection risk, and thus negatively related to bid-ask spreads (e.g., Welker 1995; Healy, Hutton, and Palepu 1999). Both informed and uninformed traders prefer trading in dark venues over exchanges when spreads are wide because dark venues offer price improvements relative to exchanges (Zhu 2014). Second, the quality of a firm s information environment is negatively related to the probability of informed trading because it is more likely that investors will discover and trade on private information when the quality of information environment is low (e.g., Verrecchia 1982; Diamond 1985; Brown and Hillegeist 2007). Informed traders might prefer trading on dark venues rather than exchanges because dark venues offer lower pre-trade transparency compared to exchanges, and thus allow informed investors to better hide their private information (Ye 2011). The higher probability of informed trading coupled with informed traders potential preference for dark venues rather than exchanges can result in higher dark market share for firms with a low quality of information environment. However, there are at least two reasons to expect a positive relation between the quality of a firm s information environment and the change in dark market share around earnings announcements. First, the probability of informed trading is higher for firms with a low quality of information environment. The higher probability of informed trading for firms with a low quality of information environment can increase execution risk for informed traders in dark venues because informed investors tend to trade in the same direction creating 15

18 greater order imbalances (Zhu 2014). As a result, participation rates of informed traders in dark venues can decrease. Second, as the probability of informed trading increases with decreases in the quality of information environment, adverse selection risk increases for uninformed traders in both dark venues and exchanges, because it becomes more likely that uninformed traders will trade against an informed counterparty. This increase in adverse selection risk, however, is more costly in dark venues rather than exchanges, because dark venues are less transparent than exchanges. As a result, uninformed traders might prefer trading in exchanges rather than dark venues for firms with low quality of information environment (Zhu 2014). Because there are compelling reasons for there to be positive and negative relations between the quality of a firm s information environment and the change in abnormal dark market share around earnings announcements, we make no prediction regarding the relation between the firm s pre-announcement quality of information environment and the change in abnormal dark market share around earnings announcements. H2: There is no relation between the firms pre-announcement quality of information environment and the change in abnormal dark market share around earnings announcements Informed trading The relation between the level of informed trading and the changes in dark market share around earnings announcements is difficult to determine a priori. As with the relation between dark market share and the quality of information environment, there are at least two reasons to expect a positive relation between the level of informed trading and the change in abnormal dark market share around earnings announcements. First, informed trading is positively related to adverse selection risk, and thus positively related to bid-ask spreads (e.g., Glosten and Milgrom 1985; Easley and O Hara 1987). Both informed and uninformed traders prefer trading in dark venues over exchanges when spreads are wide because dark 16

19 venues offer price improvements relative to exchanges (Zhu 2014). Second, informed traders might prefer trading on dark venues rather than exchanges because dark venues offer lower pre-trade transparency compared to exchanges (Ye 2011). Hence, there could be a positive relation between changes in abnormal dark market share and informed trading around earnings releases. However, there are at least two reasons to expect a negative relation between informed trading and the change in abnormal dark market share around earnings releases. Informed trading can increase execution risk for informed traders in dark venues because informed investors tend to trade in the same direction creating greater order imbalances (Zhu 2014). As a result, participation rates of informed traders in dark venues could decrease and thus abnormal dark market share could decrease as informed trading increases. Second, as informed trading increases, adverse selection risk increases for uninformed traders in both dark venues and exchanges, because it becomes more likely that uninformed traders will trade against an informed counterparty. This increase in adverse selection risk, however, is more costly in dark venues rather than exchanges, because dark venues are less transparent than exchanges. As a result, uninformed traders might prefer trading in exchanges rather than dark venues for firms with low quality of information environment, which ultimately would decrease abnormal dark market share (Zhu 2014). Because there are compelling reasons for there to be positive and negative relations between informed trading and the change in abnormal dark market share around earnings announcements, we make no prediction regarding the relation between informed trading and the change in abnormal dark market share. H3: There is no relation between the firms level of informed trading and the change in dark market share around earnings announcements Related Empirical Studies Examining Dark Trading 17

20 This study is related to empirical studies of dark trading, particularly Comerton-Forde and Putniņš (2015) and Menkveld et al. (2017). Comerton-Forde and Putniņš (2015) uses trade data stamped to the millisecond to examine whether dark trades are more or less informed than lit trades and to examine the impact of dark trading on adverse selection risk, price discovery, and informational efficiency. Menkveld et al. (2017) uses trade data stamped to the millisecond to examine the change in dark market share following VIX shocks, macroeconomic news, and firms earnings surprises. Our study differs from the aforementioned studies in several ways. First, our study uses lower frequency data compared to Menkveld et al. (2017) because our study examines dark trading behavior over longer event windows surrounding earnings releases. Therefore, by design, our study addresses a different research question that Menkveld et al. (2017) does not address. Whereas Menkveld et al. (2017) finds a decrease in dark market share in the seconds following the earnings announcement, which supports that study s prediction that traders prefer lit over dark venues following the earnings announcement to ensure trade execution, we find an increase in dark market share in the weeks surrounding the earnings announcement, which supports our prediction that traders value the benefits offered by dark venues following the earnings announcements, such as price improvement and lower information leakage. Second, both our study and Comerton-Forde and Putniņš (2015) contribute to the debate of where informed and uninformed traders trade by using an alternative measure of informed trading the ratio of absolute changes in holdings of informed institutional investors divided by the sum of the absolute changes in holdings of informed and uninformed institutional investors. Measuring informed trading in dark venues is difficult because prices of trades executed in the dark are derived from exchanges. As a result, traditional measures of informed trading are less informative when applied in dark trading data. In addition, 18

21 related studies provide conflicting evidence on whether dark trades are informed (e.g., Reed et al. 2018; Balakrishnan and Taori 2018). Our results contribute to this debate by showing that both informed and uninformed investors contribute to the increase in dark market share around earnings announcements, but the increase in dark market share decreases with increases in informed trading. Third, our study uses data relating to a broad sample of U.S. firms. In contrast, Comerton-Forde and Putniņš (2015) uses a sample relating to Australian firms. The operation of dark venues in Australia differs from the operation of dark venues in the U.S. Most notably, there is no payment for order flow in Australia, and Australian dark trades can be executed either at mid or at NBBO so there is no competition based on tick size in Australian dark venues. In addition, whereas Comerton-Forde and Putniņš (2015) examines dark trading for the 500 largest stocks by market capitalization, our sample comprises the CRSP universe, and the trading of larger stocks can be fundamentally different than smaller stocks. Although Menkveld et al. (2017) uses a U.S. sample, it comprises only 117 stocks in October In contrast, our sample captures the CRSP universe between May 2014 and December 2016, and hence the inferences we draw apply to a wider cross-section of stocks and period of time. Fourth, in contrast to Comerton-Forde and Putniņš (2015) and Menkveld et al. (2017), our study examines the association between changes in dark market share and earnings releases and the firm s quality of information environment. This analysis permits us to provide a richer understanding of how trading behavior changes when earnings are released. Finally, our study uses a different measure of dark trading than that used in Comerton-Forde and Putniņš (2015) and Menkveld et al. (2017). Comerton-Forde and Putniņš (2015) groups together dark and internalization trades in the empirical analysis. Dark and internalization trades have several fundamental differences, which potentially 19

22 contaminate inferences regarding changes in dark vs. lit trading volume in reaction to news events. 14 Although Menkveld et al. (2017) separately examines dark and internalization trades, the analysis potentially is subject to classification error because it is difficult to identify dark from internalization trades in the study s proprietary database. 3. RESEARCH DESIGN 3.1 Dark Market Share around Earnings Announcements To examine the change in dark market share around earnings announcements, we estimate the following equation: Abnornal_dark_market_share i,t = α 0 + a 1 Prior i,t + a 2 At i,t + a 3 Post i,t + a n Controls n,i,t + FE i + FE t + ε i,t (1) We define abnormal dark market share, Abnormal_dark_market_share, according to the following formula (Landsman and Maydew 2002): Abnormal_dark_market_share i,t = (Dark_market_share i,t 15 n=4 Dark_market_share i,t )/σ i,t, where Dark_market_share is the ratio of weekly dark trading volume divided by the weekly consolidated trading volume for firm i in trading week t. Dark_market_share i,t and σ i,t are the mean and standard deviation of Dark_market_share for firm i between weeks t 7 and t 2. Prior is an indicator variable that equals one for observations in the week prior to the earnings announcement and zero otherwise. At is an indicator variable that equals one for observations in the week of the earnings announcement and zero otherwise. Post is an indicator variable that equals one for observations in the week following the earnings 14 For example, whereas dark trades are executed electronically, internalization trades usually are executed over the phone or using dealers chat rooms. In addition, whereas dark trades involve a match between buyer and seller, internationalization trades usually are executed against the dealer s order book. 20

23 announcement and zero otherwise. We include Prior to allow for the possibility of trading behavior changing in advance of the week of the earnings announcement, and Post to allow for the possibility that changes in trading behavior extend beyond the announcement week if price discovery is not immediate or if investors with different information processing capabilities continue to process the public information release into private information beyond the earnings announcement week (Kim and Verrecchia 1994, 1997; Brennan, Huh, and Subrahmanyam 2018). Assuming there is an increase (decrease) in dark venue market share, then a1, a2, and a3 will be positive (negative). If trading behavior does not change until the earnings announcement week, then a1 will be zero, and if the earnings-related trading does not continue in the week after the earnings announcement then a3 will be zero. Following prior literature (Buti, Rindi, and Werner 2016; Tuttle 2013; Ready 2014; Menkveld et al. 2017), we include a number of variables as controls in the estimating equations. The Appendix provides definitions of all regression variables. FEi and FEt are firm- and week-fixed effect. All continuous regression variables except for the variables that are bound by construction are winsorized at the 1 percent level. We estimate the regression coefficients using ordinary least squares and we cluster standard errors by firm and week (Petersen 2009). 3.2 Quality of Information Environment and Dark Market Share around Earnings Announcements To test H2 we estimate the equation (2), which modifies equation (1) by including variables that are interactions of all regression variables, including the earnings announcement indicator variables, Prior, At, and Post, with an indicator variable, 21

24 High_quality, which equals one if the quality of the firm s information environment is above the median of sample firms and zero otherwise: 15 Abnormal_dark_market_share i,t = a 0 + a 1 Prior i,t + a 2 At i,t + a 3 Post i,t + a 4 Prior i,t High_quality i,t + a 5 At i,t High_quality i,t + a 6 Post i,t High_quality i,t + a 7 High_quality i,t + a n Controls n,i,t 19 n= a n Controls n,i,t High_quality i,t + FE i + FE t n=20 + ε i,t (2) We use five proxies for the firms pre-announcement quality of information environment: Abs(FE), PCA, Composite, and Implied_volatility. The first proxy of information environment, Abs(FE), is the absolute magnitude of the analysts forecast error. The idea behind this proxy is that analysts better anticipate earnings news when the preannouncement quality of information environment is high. Hence, lower values of analyst forecast errors imply higher quality of information environment (e.g., Lang and Lundholm 1996). We measure the absolute analyst forecast error as the absolute value of the difference between actual earnings less the most recent median analyst forecast, with the difference scaled by stock price in the week prior to the earnings announcement week. 16 The second and third proxies of information environment, PCA and Composite, are composite measures based on analyst following, Analyst_coverage, analyst forecast 15 We also estimated a version of equation (2) that includes interactions of the earnings announcement indicator variables, Prior, At, and Post, only with the High_quality indicator variable. Inferences based on untabulated findings are similar to those based on tabulated findings relating to equation (2). 16 Untabulated statistics reveal that the average period between the analyst forecast announcement date and the earnings announcement date is 19 calendar days and the standard deviation is 24 calendar days. 22

25 dispersion, Forecast_dispersion, and earnings transparency, Earnings_transparency. 17 We use composite measures of the firms quality of information environment to reduce the measurement error associated with each of the individual information environment proxies. 18 The first of the two composite measures, i.e., PCA, is constructed as the first principal component of analyst following, analyst forecast dispersion, and earnings transparency. The first principal component of analyst following, analyst forecast dispersion, and earnings transparency explains 37 percent of the variation in the three information environment proxies, and it is negatively related to forecast dispersion ( 0.68) and positively related to analyst following (0.72) and earnings transparency (0.17). Hence, the first principal component of forecast dispersion, analyst following, and earnings transparency seems to capture variability in the three components related to the firm s information environment, with higher values of the first principal component indicating higher quality of the firm s information environment. The second of the two composite measures, i.e., Composite, is constructed as the sum of indicator variables based on the median values of analyst following, analyst forecast dispersion, and earnings transparency. Lower values of analyst forecast dispersion, higher values of analyst following, and higher values of earnings transparency imply higher quality of information environment (e.g., Heflin, Subramanyam, and Zhang 2003). The first 17 We examine the relation between the analyst forecast error and the changes in dark market share separately from the relation between analyst following, analyst forecast dispersion, and earnings transparency and changes in dark market share because whereas the analyst forecast error depends on the earnings announced, analyst following, analyst forecast dispersion, and earnings transparency are measured before the earnings announcement. Untabulated findings reveal that inferences are the same as those based on tabulated findings if we use a composite measure of the analyst forecast error, analyst following, analyst forecast dispersion, and earnings transparency as a measure of firms quality of information environment. 18 We follow two different approaches, typically adopted in the accounting literature in combining the different variables into new latent variables (Lang, Lins, and Maffett 2012). The first approach employs a principal component analysis of the variables and the second approach is to weight equally the sum of ranks for each of the group s variables. The fundamental difference between the two approaches is that the former method gives greater weight to those variables with greater variability, and the latter approach equally weights all variables. Untabulated findings from estimations of equation (2) that use the separate measures of information quality rather than the composite measures generally yield similar inferences as those based on tabulated findings. 23