Dark Pool DNA: Improving Dark Pool Assessment

January 2012

Dark Pool DNA:

Improving Dark Pool Assessment

Abstract This empirical study based on 517,000 observations from 17 dark pools used by ITG trading systems over a one-month period shows the value in predicting execution quality by fill type rather than simply by dark pool. Using the granular approach in this study, one can better identify low quality trading situations to avoid without the opportunity cost of cutting out an entire market center. A simulated trading strategy using the conclusions in this research is compared to the typical approach for improving market quality. The simulated strategy has better access to liquidity with lower adverse selection and a better mix of likely contra parties. Given the availability of complex order types in modern dark pools, it is possible to constrain orders ahead of the trade such that they may only fill in a certain way. This means that traders and algorithm designers can make practical use of this research to improve market quality for institutions.

Ben Polidore, CFA Director, Algorithmic Trading, ITG Inc. 212.444.6300 info@itg.com www.itg.com

© 2011 Investment Technology Group, Inc. All rights reserved. Not to be reproduced or retransmitted without permission. 21411-27340 These materials are for informational purposes only, and are not intended to be used for trading or investment purposes. The information contained herein has been taken from trade and statistical services and other sources we deem reliable but we do not represent that such information is accurate or complete and it should not be relied upon as such. No guarantee or warranty is made as to the reasonableness of the assumptions or the accuracy of the models or market data used by ITG or the actual results that may be achieved. These materials do not provide any form of advice (investment, tax or legal). ITG Inc. is not a registered investment adviser and does not provide investment advice or recommendations to buy or sell securities, to hire any investment adviser or to pursue any investment or trading strategy.

Broker-dealer products and services are offered by: in the U.S., ITG Inc., member FINRA, SIPC; in Canada, ITG Canada Corp., member Canadian Investor Protection Fund (“CIPF”) and Investment Industry Regulatory Organization of Canada (“IIROC”); in Europe, Investment Technology Group Limited, registered in Ireland No. 283940 (“ITGL”) and/or Investment Technology Group Europe Limited, registered in Ireland No. 283939 (“ITGEL”) (the registered office of ITGL and ITGEL is Georges Court, 54-62 Townsend Street, Dublin 2, Ireland and ITGL is a member of the London Stock Exchange, Euronext and Deutsche Börse). ITGL and ITGEL are authorised and regulated by the Central Bank of Ireland; in Asia, ITG Hong Kong Limited, licensed with the SFC (License No. AHD810), ITG Singapore Pte Limited, licensed with the MAS (CMS Licence No. 100138-1), and ITG Australia Limited (ACN 003 067 409), a market participant of the ASX and Chi-X Australia (AFS License No. 219582). All of the above entities are subsidiaries of Investment Technology Group, Inc. MATCH NowSM is a product offering of TriAct Canada Marketplace LP (“TriAct”), member CIPF and IIROC. TriAct is a wholly owned subsidiary of ITG Canada Corp.

Dark Pool DNA: Improving Dark Pool Assessment

I. Introduction The buy side has more control than ever over market quality. The US marketplace is a decentralized one with many entry points. The market that a trader experiences on a given order is very much at the discretion of that trader and the tools she uses. In a fragmented marketplace, the strategy applied to order routing creates the context of single virtual marketplace.1 For this reason, forward thinking market participants—buy and sell side firms alike—endeavor to measure the quality of each node in the ever expanding network of dark pools and exchanges. The intuition is clear: if one chooses to consolidate only the high quality market centers, then her market composition will be of a higher quality than the unfiltered aggregate of all market centers.

The goal of this paper is to introduce some nuance into the above intuition. We believe that the concept is correct: consolidating a subset of the market can reduce transaction cost.

That said, cutting out an entire dark pool or exchange may have unintended consequences.

Each market center is itself a composition of heterogeneous market participants. We believe that an exchange or dark pool on its own cannot have a quality score—it is simply a piece of software. It is the constituency that matters, and the constituency of most dark pools and exchanges is decidedly mixed.

We show how one can use fill type to improve the prediction of dark fill quality. Using our granular approach, one can better identify and avoid low quality trading situations ahead of the trade without the opportunity cost of cutting out an entire market center. A simulated trading strategy using the conclusions in this research is compared to the typical approach for improving market quality. The simulated strategy has better access to liquidity with lower adverse selection and a better mix of likely contra parties. Given the availability of complex order types in modern dark pools, it is possible to guarantee certain fill types which means that traders and algorithm designers can make practical use of this research to improve market quality for institutions.

II. Adverse Selection—Isolating Dark Pool Performance Our hypothesis says that fill type is an important predictor of adverse selection, which is a systematic byproduct of sourcing liquidity from heterogeneous contra parties, some of which may be better informed.2 We base this hypothesis on the intuition that a given market participant has an objective function that causes him to trade in a certain way, which manifests in a particular type of trade or fill type. For example, typically it is only institutions that are willing to trade blocks over a certain size, so if one uses a minimum fill quantity constraint, he is very likely to trade exclusively with institutions. We test our hypothesis as follows:

1 2

Investment Technology Group, Inc. (ITG)

O’Hara and Ye. 2011. Is Market Fragmentation Harming Market Quality? Journal of Financial Economics (June) Saraiya and Mittal. 2009. Understanding and Avoiding Adverse Selection in Dark Pools. ITG white paper.

2

Dark Pool DNA: Improving Dark Pool Assessment

1.

Measure short-term adverse selection for a set of dark pool fills

2.

Aggregate these executions by venue identity and fill type

3.

Compare these groups across and within venues using common statistical tests

Benchmark Let adverse selection take the form:

Where: •

Q is the quantity executed

•

Pt is the midpoint price at the time of the fill

•

Pt+x is the midpoint price at some fixed time interval after the fill

• •

x is 10 seconds

Thus, a favorable post-trade price movement will yield a negative value in AS. We set x to ten seconds rather than a longer duration due to higher signal to noise ratio. A short-term measurement like this results in a measure of systematic, rather than natural, adverse selection.

This means it is based on the short-term alpha of contra parties, rather than long term price trends caused by broad market moves or stock impact.

Liquidity Groups To rule out differences explained only by changes in liquidity, we ran all of our tests and analysis in three liquidity buckets. Inside these liquidity buckets, we also changed the definition of a block trade as follows:

Table 1: Breakpoints for separating block executions based on ADV and Executed Shares

Illiquid

Liquid

Very Liquid

21 Day ADV

< 1mm

1mm to 4mm

> 4mm

Block Threshold

> 1,000

> 1,800

> 2,500

These breaks are based on standard ITG TCA thresholds.

Investment Technology Group, Inc. (ITG)

3

Dark Pool DNA: Improving Dark Pool Assessment

Sample This study draws upon a random sample of ITG trades in 17 U.S. dark pools between 6/20/11 and 7/22/11 resulting in 517,000 observations. Each observation represents an individual dark pool fill on a child order from an ITG trading algorithm. The parent orders are from a variety of strategy types, including VWAP, implementation shortfall, and dark aggregation algorithms. These strategies access dark pools in a passive manner, i.e. posting at midpoint to move ahead in the order without directly creating impact. The most aggressive liquidity seeking algorithms have been removed from the sample, to avoid distortion caused by shortterm price impact. Table 1 is a set of descriptive statistics for each liquidity group.

Table 2: Descriptive statistics

Investment Technology Group, Inc. (ITG)

4

Dark Pool DNA: Improving Dark Pool Assessment

Fill Types Once we group orders by stock liquidity, we measure the difference in our adverse selection score among various fill type categories. Statistically significant differences among these groups are the test of our hypothesis that fill type is a proxy for contraparty, which can help predict the quality of dark pool executions.3 We use the following partitions: •

Midpoint Taker fills remove liquidity at midpoint4 from a dark pool, either by filling an IOC order type or immediately filling a posted order

•

Midpoint Supplier fills provision liquidity at midpoint to a dark pool. They are the other contra to a taker fill at midpoint

•

Passive fills provision liquidity at the near side of the spread: a passive buy supplies liquidity at the bid price to aggressive sellers while a passive sell supplies liquidity at the offer to aggressive buyers

•

Block fills are executions above a certain number of shares.5 Block trading is the clearest indication of contra party identity, so if an observation meets the size criterion, we always define it as a block regardless of whether the fill was taker or supplier.

III. The Relationship between Adverse Selection and Fill Types Our results show that fill type is a good predictor of adverse selection, along with venue identity. These findings are consistent among our three DNA charts. Figures 1, 2, and 3 demonstrate our results visually for three different stock liquidity buckets. The first four series in each chart show the adverse selection averaged by fill type. Each point in the series represents the average adverse selection for a single dark pool. The fifth series shows the average adverse selection across all fill types for each dark pool, and represents an aggregate view of the data when we do not take the type of an execution into account.

The granularity of our groupings—17 dark pools and 4 fill type subcategories—opens us up to the risk of finding false statistical significance through multiple testing. To mitigate this concern, we performed an F-test on the following groupings to test whether the differences across the groupings were statistically significant:

3

1.

Fill type plus dark pool combination: 68 groups

2.

Fill type: 4 groups

3.

Dark pool: 17 groups

We found the differences to be significant at 1% for each of the above.

Aggressive taker fills (e.g., buy at the offer) were excluded. We decided that these fills require special attention that is outside the scope of this paper.

4

5

Investment Technology Group, Inc. (ITG)

The size thresholds for blocks are in Table 2

5

Dark Pool DNA: Improving Dark Pool Assessment

Midpoint Supplier These fills have the worst average adverse selection score. This type of execution gives up both the timing of the trade and half the spread. Given these disadvantages, the risk of supplying liquidity to aggressive or information-based contra party is high. Also, since this type of execution has the least protection, the performance of small, midpoint supplier trades may be the clearest measure of a dark poolâ&#x20AC;&#x2122;s natural constituency.

Midpoint Taker Liquidity taking (e.g., IOC) fills at midpoint show the least adverse selection for small fills in our study and the least dispersion across dark pools in all three liquidity groups. This is likely due to the institutional nature of the average midpoint supplier who represents the only possible contra party for a liquidity taker. Consider: if one is willing to give up the timing and half the spread on a fill by supplying at midpoint, he is likely working down a position, not making a market. As we describe in detail later, taker fills may be a good substitute for midpoint supplier fills, especially when trading in dark pools with low aggregate quality scores.

Passive Passive fills showed similar adverse selection to midpoint supplier fills, but this data point ignores a key fact: our study measures adverse selection from midpoint to midpoint but passive fills earn more of the spread than midpoint fills. After accounting for spread gains, we find that that small, passive fills better compensate traders for adverse selection than any kind of midpoint order, but the fill rate is much lower.

Block We had the fewest observations in the block category, which is to be expected. We found that block trading had the least adverse selection of any of the groupings, but there was wide variation across pools as sample size declined. Previous studies show that block trading reduces transaction cost, especially for difficult orders where mild adverse selection is offset by impact savings.6

Mittal, Sharpe, and Saraiya. 2011. Capturing Block Liquidity to Improve Algorithm Performance. The Journal of Trading (Spring)

6

Investment Technology Group, Inc. (ITG)

6

Dark Pool DNA: Improving Dark Pool Assessment

Figure 1: Very liquid group

Left of black line: average adverse selection for various dark pools grouped by fill type. Right of black line: global average adverse selection for various dark pools.

Figure 2: Liquid group

Left of black line: average adverse selection for various dark pools grouped by fill type. Right of black line: global average adverse selection for various dark pools.

Figure 3: Illiquid group

Left of black line: average adverse selection for various dark pools grouped by fill type. Right of black line: global average adverse selection for various dark pools.

Investment Technology Group, Inc. (ITG)

7

Dark Pool DNA: Improving Dark Pool Assessment

IV. Application—More Efficient Routing Decisions One of the practical goals of this paper is to apply measured differences among dark pools and fill types to design a better order routing strategy for algorithms. This means improving upon what we describe as the “standard approach” to improving market quality: measuring the average performance of the dark pools and excluding some bottom quantile going forward. This paper suggests a different approach: excluding trading situations, i.e. dark pool plus fill type, rather than entire market centers. Below, we apply each method, and measure the change in average performance. Then we test the statistical significance of the difference between the mean of the excluded group versus what remains in each routing strategy via a t-test.

Our results suggest that a more granular process can outperform the standard approach.

Rather than cutting out, in our example, the bottom 30% of market centers, one could cut out the bottom 30% of trading situations, where a trading situation is a dark pool plus fill type combination. Recall Figures 1, 2 and 3: the performance of a single dark pool changes substantially when you control for fill type. Using our method, one is closer to targeting contra parties, which we think is the underlying goal of this process. The results in Table 3 show that this approach outperforms the standard approach even if we use the same threshold: the total amount of liquidity available is increased and the adverse selection score is reduced.7 Table 4 demonstrates that the mix of fill types is more favorable with our granular approach: fewer information leaking midpoint supplier trades and more taker and block trades. In practice, we recommend using a threshold that removes less liquidity given the increased dispersion of quality scores when we group by fill type.

In Figure 4, the red triangles represent errors made by the standard approach, which our more granular approach avoids. These are either fills that one wishes he avoided, but did not (labeled “1”) or fills that one avoided that he wishes he hadn’t (labeled “2”). More detail: 1. These are trading situations that are below the aggregate threshold, but are not avoided using the standard method. By using our method, one can avoid these fills while still trading to these dark pools when they are beneficial by using complex order types to restrict orders to a certain type of execution. For example, most pools have poor midpoint supplier performance, but very good midpoint taker performance. We can cut out the former without missing the latter by using IOC orders at midpoint.

2. These are trading situations that are above the threshold but mistakenly avoided when one cuts out an entire dark pool. The standard approach can only make a binary decision about an entire market center using an average score that hides much detail. As we have discussed, most dark pools have a heterogeneous constituency. Using our more granular approach, one can remove the fills that contribute to a poor quality score for a dark pool without missing out on whatever quality liquidity remains in that pool.

Note that the results in this section calculate adverse selection using the average price of the fill instead of the midpoint at the time of the fill. This results in a global shift upward for passive supplier fills in all dark pools, which earn more of the spread. Other groups are not affected since they execute at midpoint.

7

Investment Technology Group, Inc. (ITG)

8

Dark Pool DNA: Improving Dark Pool Assessment

Figure 4: Comparing two methodologies for controlling market quality

Table 3: Standard vs. granular approach, quality Sample All

% Liquidity Average 10s Adverse Available Selection (bps)

t-value of in vs. out group

100%

-0.46

--

Standard Approach

68%

-0.37

5.6

More Granular Approach

72%

-0.28

17.1

Compared to the standard approach, the more granular approach has a better quality score and the liquidity that is removed is more differentiated from what remains.

Investment Technology Group, Inc. (ITG)

9

Dark Pool DNA: Improving Dark Pool Assessment

Table 4: Standard vs. granular approach, fill types Sample

% Supplier

% Taker

% Block

% Passive

All

35%

46%

16%

3%

Standard Approach

33%

51%

12%

4%

More Granular Approach

12%

64%

19%

4%

Compared to the standard approach, the more granular approach results in a more desirable breakdown of fill types

V. Conclusion Demand for data on market centers and routing efficiency is increasing. Many firms are attempting to use these data to define a list of approved market centers based on some sort of venue quality score. Given that so few market centers have a pure constituency, this research provides vital context for algorithm designers and institutional traders trying to source quality liquidity. We show that using fill type as a proxy for contra party, one can improve liquidity seeking algorithms by placing orders that are unlikely to execute in ways known to have low average quality scores. Using a simulation, we demonstrate that this approach substantially outperforms the standard approach that removes entire market centers both in terms of average adverse selection and access to liquidity. Going forward, we think the concept of granular contra party â&#x20AC;&#x153;proxyingâ&#x20AC;? is an interesting area of research that has the potential to improve all classes of liquidity seeking algorithms by informing routing strategies that closely align with trading objectives.

Investment Technology Group, Inc. (ITG)

10