Fragmentation in Financial Markets: The Rise of Dark Liquidity

Sabrina Buti

Global Risk Institute – April 7th 2016 Where do U.S. trade?

Market shares in -listed securities shares in NYSE-listed securities

Other SRO Other SRO NASDAQ 2% 2% Direct Edge Direct Edge 11% BX 11% NASDAQ 10% 3% 25% BATS PSX BATS Nasdaq Venue 10% 1% NYSE 10% 0% TRF BX NYSE Venue ARCA ARCA 2% 11% 8% PSX BATS 1% NASDAQ Direct Edge TRF NYSE TRF 33% 3% Other Venues NYSE 17%

NYSE TRF NASDAQ TRF 2% 38%

Off-exchange trading is reported through the two Trade Reporting Facilities(TRF).

NASDAQ data represents shares matched on the NASDAQ book plus shares reported to the Consolidated Tape. Uses September 2013 data. What about Canadian stocks?

Source: of Canada Review, Autumn 2013. And European stocks?

Source: Fioravanti and Gentile, 2011, The impact of fragmentation on European exchanges, Consob. Evolution of US market structure

2007: Reg NMS 2009/2010: Increased 1994: Nasdaq collusion implemented. focus on HFT and Increased competition latency in markets

1997: handling 2005: Reg NMS 2010: Flash crash rules approved

2001/2: Autoquote on 2010-2013: Response NYSE. Beginning of to flash crash. 2000: Decimalization Equity market structure 2002: ECN reviews. consolidation State of the market

Source: research. A new world: Trading complexity

16

13

50 16 29

10

2002 2009 2012

Lit Venues

Dark Venues

Source: Nasdaq Dark liquidity

 Internalized flows  Dark Pools  Hidden liquidity on lit exchanges

Source: SEC MARKET SHARE Total other venues 18%

Total dark pools 13%

Total ECNs Lit exchanges 2% 67%

Source: CFA Institute 2012 Dark trading

40

35

30

25 Number of Securities with over 40% Dark Market Share Total Dark Liquidity 20

15

Market Share(%) 3,855 10

5

0 2009 2010 2011 2012 2013 What are dark pools?

 Alternative Trading Systems that do not provide their best-priced orders for inclusion in the consolidated quotation data  Provide anonymity and opacity pre- trade to institutions  Derivative pricing, mid-quote, inside, at-quotes (controversial) Dark pools relevance Dark pools classification (I)

 Dark Pools classification based on:  market model  [periodic vs continuous crossing, blind vs advertisement based, ...]  ownership  [traditional exchange vs single or a group of broker- dealers]  access  [buy or sell side, both retail vs institutional, ...] Dark pools classification (II)

 Independent/Agency pools

 Bank/Brokers pools

pools

 Consortium-Sponsored pools

 Exchange-Based dark pools Dark pools volume: Dec 2012 Percentage of Consolidated US Volume

Independent / Agency Pools Bank / Broker Pools ITG POSIT 0.77% CITI MATCH 0.54%

INSTINET CBX 0.58% Consortium-Sponsored Pools GOLD. SACHS SIGMA X 1.41% VWAP 0.08% BIDS TRADING 0.6% UBS PIN ATS 1.09% CVGX VORTEX 0.23% LEVEL 0.38% CS CROSSFINDER 1.88%

CVGX MILLENNIUM 0.26% BARCLAYS LX 1.41% KNIGHT MATCH 0.8% DB SUPER X 1.01% LIQUIDNET 0.24% MS-POOL 0.76% LIQUIDNET H20 0.13% 0.98% 3.09% 8.1% Market Maker Pools 2.17% GETMATCHED 0.77% KNIGHT LINK 1.4%

Total USA: 14.34% Source: Rosenblatt Securities Inc. A few questions about dark pools

 Who wants to hide in the dark? Some theory: The starting point

INFORMED KYLE (ECTA, 85) TRADERS The informed faces a trade-off between VOLUME Price uncertainty: and price impact: MARKET ORDER the higher the volume, the larger Lit market the impact. One Market single maker price UNINFORMED TRADERS Ye (WP, 2011): Informed go dark

The informed trader still INFORMED faces a trade-off TRADERS between VOLUME and Execution uncertainty: price impact. UNKNOWN IMBALANCE Incentives to move to the DP, where there is NO PRICE IMPACT. BUT the more you Lit market trade on the DP, the One (DP): Market lower is the probability single Crossing maker Derivative of execution. price Network Price UNINFORMED UNINFORMED Go dark as as it is TRADERS TRADERS profitable to do so! Zhu (RFS, 2013): Informed go lit

INFORMED TRADERS Execution uncertainty: Informed traders tend UNKNOWN IMBALANCE to trade in the same direction.

They crowd on the heavy side of the Lit market Dark Pool market, and face a (DP): higher execution risk Crossing Bid-ask Dealer Derivative in the DP, relative to Network Price uninformed traders.

Informed traders prefer lit exchanges! To summarize…

 Ye (WP, 2011):  DPs attract informed traders, harm and decrease adverse selection on the lit market  DPs have incentives to get rid of informed traders

 Zhu (RFS, 2013):  Exchanges are more attractive to informed traders, and DPs are more attractive to uninformed traders  Adding a DP could concentrate price-relevant info on the exchange and improve price discovery, but reduce exchange liquidity (spread)

Now to the data!

 Buti, Rindi and Werner (WP, 2011):  DP activity in the U.S. results in improved price efficiency based on -term measures

 Comerton-Forde and Putnins (JFE, 2015):  in Australia, orders executed in the dark are less informed that orders executed in the lit  dark trades increase adverse selection on the main market, worsening market quality  low levels of dark trading are benign or even beneficial for informational efficiency, but high levels are harmful A few questions about dark pools

 Who wants to hide in the dark?  Price improvement in dark pools?  Should we be afraid of the dark? Better prices in dark pools?

Source: Nasdaq analysis SEC Reg NMS Rule 612

 Introduced on August 29, 2005

 Established minimum price variation for lit markets:  $0.01 for stocks priced higher than $1  $0.0001 for stocks priced at $1 and below

 Dark markets are exempt from Rule 612 provided:  execute less than 5% of the volume  do not display their orders

 So Rule 612 allows for:  broker-dealer internalization  dark trading in sub-penny Queue jumping… Exchange Dark pool or Internalization 퐴2 푎5 푎4 푎3 퐴1 푎2

푎1 푣 푏1 Sub-penny trade 푏 퐵1 2 Market order 푏3 푏4 퐵2 푏5 How does sub-penny trading work?

In theory… Rule 612 states that no market participant can accept, rank, or display orders priced in sub-pennies.

In practice… On Jan. 2015 the SEC fined a dark pool operator $14.4 million for accepting and ranking hundreds of millions of orders priced in increments smaller than one cent that were submitted to his dark pool. Growth in sub-penny trades (US)

Source: Nanex Canadian experiment

 On October 15, 2012, Canadian regulators made providing liquidity in the dark more expensive  The new rule specifies the improvement that dark liquidity providing orders must offer relative to the best lit bid and offer:  At least 1 cent (1/2 if the lit spread is 1 cent)  Applies to marketable orders below 5,000 shares or $100,000 in value  Does not affect midpoint orders and block trading  Dark markets moved from fractional pricing (90/10 and 80/20) to midpoint

Source: IIROC Canadian experiment: Outcome

Dark pool volume dropped by 42%.

Source: IIROC What is going on?

Dark trading activity is significantly reduced:

 Volume on IntraSpread, on which dark trading interacts with a segregated flow of retail orders, is most strongly impacted when these market-making opportunities are reduced

 Volume on TCM, in which dark participants trade mainly to minimize information leakage and market impact, is less impacted

 Consistently, active retail traders and passive High Frequency Traders show the greatest reduction in dark trading

Successful experiment?

 Reduction in dark volume without meaningful price improvement

 Minimal market-wide impact as most measures of market quality showed no deterioration

 However… Talis and Putnins (JFE, forthcoming) analyze the same experiment and find that midpoint crossing systems do not benefit market quality, but dark limit order markets are beneficial to market quality Why separate?

 There are two main reasons why trades execute on alternative trading systems at fractions of penny:  Undercutting orders posted at the top of public limit order books (queue-jumping& mid-crossing)  The execution system of some dark pools and broker- dealers internalization systems follows a derivative pricing rule: trades execute at the midpoint of the inside spread (mid-crossing)

 Mid-crossing therefore does not necessarily include only undercutting and could potentially mix different trading strategies Back to the U.S.

 October 1st – November 30th, 2010 (42 trading days)  Stratified sample of 90 Nasdaq and 90 NYSE listed common stocks, sorted into terciles by and price as of the end of 2009  Sub-penny trading is further divided into:  Mid-crossing  Queue-jumping Example in TAQ data: IBM

ID Symbol Date Time Exchang Price Size Rounded Price Price Type of Sub- e Improvement Penny 1 IBM 20101001 10:00:00 K 135,76 100 135,76 0 None 2 IBM 20101001 10:00:00 K 135,77 100 135,77 0 None 3 IBM 20101001 10:00:00 K 135,84 100 135,84 0 None … IBM 20101001 10:00:00 … … … … … … 17 IBM 20101001 10:00:00 T 135,84 200 135,84 0 None 18 IBM 20101001 10:00:01 D 135,64 196 135,64 0 None 19 IBM 20101001 10:00:01 D 135,7375 200 135,74 0,0025 Queue-Jumping 20 IBM 20101001 10:00:01 N 135,67 100 135,67 0 None … IBM 20101001 10:00:01 … … … … … … 28 IBM 20101001 10:00:03 B 135,76 100 135,76 0 None 29 IBM 20101001 10:00:03 D 135,76 100 135,76 0 None 30 IBM 20101001 10:00:03 D 135,825 300 135,83 0,005 Mid-Crossing

Mid-crossing: Queue-jumping: price improvement = 0.005 price improvement ≠ 0.005 Mid-crossing vs. queue-jumping

 Mid-crossing (MID) for stock i and day t is defined as:

푉표푙푢푚푒 푒푥푒푐푢푡푒푑 푖푛 푀퐼퐷

푇표푡푎푙 푐표푛푠표푙푖푑푎푡푒푑 푣표푙푢푚푒

 Queue-jumping (QJ) for stock i and day t is defined as: 푉표푙푢푚푒 푒푥푒푐푢푡푒푑 푖푛 푄퐽

푇표푡푎푙 푐표푛푠표푙푖푑푎푡푒푑 푣표푙푢푚푒 Sub-penny trading and market quality

 Simultaneous equations (2SLS) to address endogeneity following Hasbrouck and Saar (JFM, 2013):

 Where:

 푁푂푇푖,푡 are other stocks listed on the same exchange and in the same market capitalization group as stock i  SP is QJ or MID and MQM can be either bid depth, share volume, quoted, or relative spread

Empirical analysis

Bid depth and QJ Bid depth and MID

Full sample Small Large Full sample Small Large

푎1 0.342*** 0.044 0.281* 푎1 0.202 -0.041 0.253 (3.358) (0.240) (2.338) (0.940) (-0.198) (0.790)

푎2 0.523*** 0.371*** 0.645*** 푎2 0.602*** 0.368*** 0.719*** (10.266) (4.294) (8.665) (13.768) (4.279) (9.500) Observations 7,560 2,520 2,520 Observations 7,560 2,520 2,520

Relative spread and QJ Relative spread and MID

Full sample Small Large Full sample Small Large

푎1 -0.493*** -0.551 -0.288* 푎1 -0.321 -0.076 -0.581 (-3.749) (-1.837) (-2.339) (-1.360) (-0.209) (-1.283)

푎2 0.631*** 0.582*** 0.684*** 푎2 0.742*** 0.719*** 0.669*** (11.720) (5.614) (9.592) (17.162) (12.300) (5.874) Observations 7,560 2,520 2,520 Observations 7,560 2,520 2,520 Should we be afraid of the dark?

 Dark pools seem to be used mainly by uninformed traders and not to have a negative effect on price efficiency

 No evidence that SPT harms liquidity: QJ seems to improve market quality at least for liquid stocks and MID shows no significant effect, both in US and Canada

 But too early to draw final conclusions… we still need better data!