«Charles A. Dice Center for Research in Financial Economics Diving Into Dark Pools Sabrina Buti, Rotman School of Management, University of Toronto, ...»
Fisher College of Business
Working Paper Series
Charles A. Dice Center for
Research in Financial Economics
Diving Into Dark Pools
Sabrina Buti, Rotman School of Management,
University of Toronto,
Barbara Rindi, Bocconi University,
Ingrid M. Werner, Department of Finance,
The Ohio State University
Dice Center WP 2010-10
Fisher College of Business WP 2010-03-010
Revision: November 2011
This paper can be downloaded without charge from:
http://www.ssrn.com/abstract=1630499 An index to the working paper in the Fisher College of
Business Working Paper Series is located at:
http://www.ssrn.com/link/Fisher-College-of-Business.html fisher.osu.edu Electronic copy available at: http://ssrn.com/abstract=1630499 Diving Into Dark Pools * Sabrina Buti† Barbara Rindi‡ and Ingrid M. Werner§ Date: November 17, 2011 * We are grateful to SIFMA for assisting us in collecting the dark pool share volume data that forms the basis for this study. We also want to thank Jamie Selway for making this study possible, Kewei Hou for helpful comments on the empirical design, and participants at the Hong Kong University of Science and Technology Symposium, the Notre Dame Conference on Current Topics in Financial Regulation, the University of Cyprus academic conference, and seminar participants at Columbia University for comments.
† Rotman School of Management, University of Toronto, email@example.com ‡ Bocconi University and IGIER, firstname.lastname@example.org § Fisher College of Business, The Ohio State University, email@example.com Electronic copy available at: http://ssrn.com/abstract=1630499 Diving Into Dark Pools
ABSTRACTThis paper examines unique data on dark pool activity for a large cross-section of US stocks in
2009. Dark pool activity isconcentrated in liquid stocks. Nasdaq (AMEX) stocks have significantly higher (lower) dark pool activity than NYSE stocks controlling for liquidity. For a given stock, dark pool activity is significantly higher on days with high share volume, high depth, low intraday volatility, low order imbalances relative to share volume, and low absolute returns. Results show that increased dark pool activity improves market quality measures such as spreads, depth, and short-term volatility. The relationship between dark pool activity and measures of price-efficiency is more complex.
There are several reasons for why institutional traders may want to avoid displaying their orders in the continuous limit order market. Order display invites imitation, potentially reducing the alpha of the underlying investment strategy. Displayed orders also invite front running and quote matching by broker-dealers as well as by opportunistic traders, resulting in higher trading costs. Further, traditional order display is associated with direct broker involvement, generating significant commission costs.
Institutional traders worry about counterparty risk, i.e. the risk of trading against informed order flow especially order flow from proprietary trading desks. Institutional sized orders also face another problem: average trade and order sizes have fallen dramatically in recent years, making it virtually impossible to trade in size in the continuous limit order market.
It is therefore not surprising that there is a growing demand for trading venues that make it possible for institutions to keep their orders secret, offer low commission rates, maximize the chances of trading with other institutions (as naturals), and allow institutions to trade in size at the mid-quote.
Such non-displayed pools of liquidity have been present in US equity markets for a very long time.
Examples include reserve and hidden orders within exchanges’ and Electronic Communication Networks’ (ECNs) trading systems, floor broker orders and specialist capital on floor-based exchanges, working orders handled by agency brokers or broker-dealers, dealer capital and stand-alone as well as broker and exchange/ECN operated crossing networks.1 More recently, non-displayed liquidity pools such as internalization pools and ping destinations have been added to the list. Nowadays opaque sources of liquidity are often grouped together under a single label (with unfortunate nefarious connotations): dark pools.
In broad brush terms, dark pools are characterized by limited or no pre-trade transparency, anonymity, and derivative (almost exclusively mid-quote) pricing. However, they differ in terms of whether or not they attract order flow through Indications of Interests (IOIs)/advertising and whether or not they allow interaction with proprietary and black box order flow.2 It is difficult to accurately measure the amount of volume that is actually matched through dark pools but estimates range from 8of share volume.3 In its recent Concept Release on Equity Market Structure (SEC, 2010), the SEC raises concerns about the consequences of a rising dark pool market share on public order execution quality and price discovery. In Congressional testimony, Dr. Hatheway (Nasdaq OMX) speaks to this issue and argues that when stocks experience “dark” trading in excess of 40 percent of total volume, execution quality begins to deteriorate. Weaver (2011) studies broader measures of market fragmentation and also argues that dark trading is associated with a reduction in market quality. In contrast O’Hara and Ye (2011) find that fragmentation of trading generally reduces transactions costs and increases execution speed. These contradictory results are not surprising as the researchers rely on very imprecise proxies for dark trading. The O’Hara and Ye (2011) study focuses on the effect of fragmentation on market quality during 2008 and uses volume reported to the Trade Reporting Facilities (TRFs) as a proxy without even netting out fully transparent venues such as BATS and DirectEdge. The same strategy is used by Weaver (2011), but his sample is more recent, from October 2009. The Nasdaq OMX study uses TRF volume minus BATS and DirectEdge as a proxy for dark pools, but this data still includes internalized order flow.
To better inform the regulatory debate, we use more granular data to empirically assess the effects of dark pools on market quality and price discovery. Specifically, the Securities Industry and See Mittal (2008) for a discussion of dark pool characteristics.
Rosenblatt Securities, Inc. started tabulating monthly share volume for dark pools of liquidity in its Trading Talk publication in March 2008 and TABB Group started its Liquidity Matrix publication in April 2007. Efforts to track volume in these venues are problematic due to a lack of uniform dark pool reporting standards.
Financial Market Association (SIFMA) solicited daily stock-level dark pool share volume data for the 2009 calendar year from all their members operating dark pools. The reporting was completely voluntary, and in the end SIFMA collected data on daily single-counted share volume from eleven dark pools on our behalf. The data is anonymous, and no attempt to study the data by individual dark pools will be made.
This study will focus on answering three questions:
1. How does dark pool market share vary across stocks and time?
2. Is dark pool activity associated with lower market quality?
3. Is dark pool activity associated with impaired price efficiency?
There is very limited empirical evidence on dark pool activity in the cross-section and the timeseries. A few studies have focused on crossing networks. Gresse (2006) finds that crossing networks have a very limited market share and do not have a detrimental effect on the liquidity of the continuous market. Conrad, Johnson, and Wahal (2003) find that institutional orders executed in crossing networks have significantly lower realized execution costs and that traders use the continuous market to trade their exhaust. Naes and Odegaard (2006) find that institutional orders sent first to crossing networks and then to the continuous market obtain lower realized execution costs for the crossed component, but not necessarily for the entire order. Fong, Madhavan, and Swan (2004) find no evidence of a liquidity drain away from the continuous market when traders can trade in a crossing network. Ready (2010) studies monthly volume by stock in three dark pools: Liquidnet, POSIT, and Pipeline during June 2005-September 2007. He finds that the market share of these dark pools is less than one percent of consolidated volume, and that dark pool volume is concentrated in liquid stocks (low spreads, high share volume). Two more recent papers by Brandes and Domowitz (2010) and Buchanan et al. (2011) study dark pool trading in Europe and find that increased participation of dark pools enhances the price discovery process. In contrast, Degryse, de Jong, and van Kervel (2011) find that fragmentation is beneficial for the liquidity of 52 Dutch stocks as long as trading is transparent, but that opaque trading on the local exchange (dark pools and OTC) has a detrimental effect on global liquidity.
Our sample has several advantages compared to the Ready (2010) sample: it covers more dark pools, includes daily share volume data, and is more recent. It also has several advantages relative to the data used in Degryse, de Jong, and van Kervel (2011) in that we have a much broader sample and more importantly our dark pool data excludes internalized trades. Nevertheless, several caveats apply.
First of all, the SIFMA dark pool data covers only those eleven dark pools that voluntarily responded to the data request. According to the SEC 2010 Concept Release on Market Structure, there are approximately 32 active dark pools during our sample period. Hence, our sample of eleven dark pools captures only roughly 1/3rd of dark pools operating in the US equity market. Second, to our knowledge there is no publicly available data on dark pools which makes it difficult to check the SIFMA data for accuracy. To gauge the coverage of our data, we compare it to monthly data reported by Rosenblatt, Inc. However, we note that this source is based on a combination of self-reported data and Rosenblatt estimates. Third, while our data permits a study of both time-series and cross-sectional variation in dark pool activity for the SIFMA sample of dark pools, we have no way of knowing if these eleven dark pools represent the same fraction of dark pool activity over stocks and over time. Therefore, we cannot claim that the variation in dark pool activity within our sample is representative of the entire population of dark pools. These caveats should all be kept in mind when drawing conclusions based on the SIFMA data.
We describe our sample construction in Section 2, and provide a univariate analysis of dark pool activity in Section 3. Descriptive statistics for our explanatory variables are in Section 4. Our analysis of the dark pool activity in the cross-section and in the time series is in Section 5. In Section 6, we study the relationship between dark pool activity and measures of market quality. Section 7 explores the relationship between dark pool activity and price efficiency. Section 8 concludes.
2. SAMPLE CONSTRUCTIONWe first benchmark the raw SIFMA data against the monthly total share volume in dark pools as reported by Rosenblatt, Inc. in their monthly Let There Be Light publication. Figure 1 shows that the SIFMA data mirrors the monthly time series variation in the Rosenblatt share volume pretty closely.
Figure 2 shows that dark pool share volume as reported in the SIFMA (Rosenblatt) data represents 3.65 (7.74) percent of consolidated volume in January, and 6.10 (10.15) percent of consolidated volume in December. Finally, Figure 3 shows that the SIFMA data covers roughly half of the Rosenblatt share volume. Specifically, the market share of the dark pools submitting data for our study increases from 47% in January to 60% in December.
The raw SIFMA data covers 10,178 unique securities and the coverage by individual dark pools ranges from a low of 5,646 to a high of 8,251 securities. In order to merge the SIFMA data with data from TAQ, CRSP, etc., we screen the data following standard practice as summarized in Table 1. We first exclude 1,525 ticker symbols with suffixes (e.g., preferred, warrants, non-voting, etc.) and the ticker symbols with a fifth character (unless also in CRSP as A, B, or K). Second, we exclude 4,035 securities that are not common stocks (SHRCD 10 or 11) covered by CRSP. As we need to merge CRSP with the SIFMA data, we also exclude 87 stocks with missing ticker symbols in CRSP and 49 stocks with duplicate stock identifiers (permno or cusip) for the same ticker symbol. Our SIFMA sample consists of 4,482 stocks with non-zero dark pool volume for at least one day in 2009. Finally, we add the CRSP common stocks that do not have any SIFMA reported dark pool volume, setting daily dark pool volume to zero.4 We also create subsamples that are similar to the samples used by Weaver (2011) and O’Hara and Ye (2011) to benchmark our data against previous samples. Weaver (2011) excludes stocks with price above $1,000 and O’Hara
3. UNIVARIATE STATISTICSTo examine the cross-sectional distribution of dark pool activity, we compute dark pool volume (DPVOL) as the number of shares per stock per day (single-counted) that execute in one of our eleven dark pools. We also compute the fraction of daily consolidated share volume (VOL) as reported in CRSP that was executed in one of the dark pools as 100*DPVOL/VOL for every stock in our sample. This variable will be labeled RELDP. Further, we count the number of different dark pools that are active in a stock on a given day and call this variable COUNTDP. To get a better sense of the degree of competition among dark pools, we compute the inverse Herfindahl index (IHERF) based daily stock-level dark pool market shares. Recall that if the market shares are evenly distributed across dark pools, IHERF will be equal to COUNTDP. IHERF will be lower than COUNTDP the more concentrated dark pool trading activity is for a given stock day.
We report the overall results in Table 2, Panel A. Dark pool volume represents on average 4.51 percent of consolidated volume. Dark pool activity is skewed as the median is lower, at 3.05 percent.
On average almost half the SIFMA reporting dark pools (5.27) are active in a stock on any given day.