26 August 2014
AUTHORS
Ian Domowitz
Managing Director
Head of Analytics
[email protected]
Kristi Reitnauer,
Assistant Vice President
Analytics Research
[email protected]
Colleen Ruane
Director
Analytics Research
[email protected]
CONTACT
Asia Pacific
+852.2846.3500
Canada
+1.416.874.0900
EMEA
+44.20.7670.4000
United States
+1.212.588.4000
[email protected]
www.itg.com
Garbage In, Garbage Out:
An Optical Tour of the Role of Strategy
in Venue Analysis
ABSTRACT
We address a single question in this paper: is consideration of trading strategy an
essential component in assessing venue performance? The answer is, yes. We arrive
at this conclusion through comparisons of strategy use across venues and
performance metrics, by venue, venue type, and strategy.
INTRODUCTION
Venue analysis is all the rage these days, but an old theme. Nasdaq and the NYSE
once fought to advertise the relative execution quality on their venues, motivating
academic work purporting to make the comparison. With the arrival of electronic
trading venues, initial efforts concentrated on explaining the mechanics of different
venues, followed by empirical studies, which attempted to contrast execution on
electronic and floor systems.1 The analogue in the past few months has been the
race to publish Reg ATS filings in the interest of transparency. Venue quality
reporting via regulatory mandate appeared with Reg NMS. Closely following, the
interaction of dark pools with routing schemes generated work on information
leakage.2 Comparisons of lit and dark markets began to appear.3 In the midst of all
this activity, sponsors of alternative execution venues began to publish material on
the relative benefits of their own dark pools.4 Flash forward to present day, and the
furor surrounding the recent Michael Lewis book has generated collective amnesia
with respect to evolution of our knowledge of venues over the past twenty years.
The focus is on microseconds. We reject that line of inquiry here, and we are not
alone. Firms such as Pragma, whose business is arguably in tiny fractions of a
second, note that drawing conclusions from microsecond snaps of fills is a fruitless
undertaking.5 After investigating several popular methods of venue analysis, Pragma
concludes that such methods do not help understanding or improve performance;
they turn their attention instead to venue fee structures and the patterns of routing
behavior created by them.
Ian Domowitz, “A Taxonomy of Automated Trade Execution Systems,”Journal of International Money and Finance 12, 1993; for a review of
papers comparing electronic to floor trading, see Ian Domowitz and Benn Steil, “Automation, Trading Costs, and the Structure of the
Securities Trading Industry,” Brookings-Wharton Papers on Financial Services, 1999.
2
Ian Domowitz, Ilya Finkelstein, and Henry Yegerman, “Cul de Sacs and Highways: An Optical Tour of Dark Pool Trading Performance,” Journal
of Trading, 2008.
3
Ian Domowitz, Ilya Finkelstein, and Henry Yegerman, “Cul de Sacs and Highways: An Optical Tour of Dark Pool Trading Performance,” Journal
of Trading, 2008; Yossi Brandes and Ian Domowitz, “Alternative Trading Systems in Europe: Trading Performance by European Venues Post
MIFID, Journal of Trading, Summer 2010; Yossi Brandes and Ian Domowitz, “Alternative Trading Systems in Europe: Trading Performance by
European Venues Post MIFID, An Update for 2010, Journal of Trading, Spring 2011; Quantitative Services Group, “QSG Shines a Light on a
Dark Pool,” Research Note, June 2008.
4
See, for example, Hitesh Mittal, “Are You Playing in a Toxic Dark Pool? A Guide to Preventing information Leakage,” Journal of Trading,
Summer 2008; and George Sofianos and David Jeria, “Quantifying the SIGMA X Crossing Benefit” in Street Smart, Goldman Sachs, March,
2008.
5
“Venue Analysis: What is it Good for?”, Pragma Research Note No. 7, May 2014.
1
26 August 2014
2
Venue performance cannot be separated from trading strategy, market conditions, or
even constraints on traders’ workflows. We have made this point before, in the
context of comparisons of algorithmic trading strategies.6 More broadly, venue
analysis is most useful when viewed in the larger context of overall performance.
Understood in that light, a better grasp of routing practices and performance permit
traders to use the full suite of electronic tools at their disposal effectively, as well as
generating more useful conversations with their brokers.
We address a single question in this paper: is consideration of trading strategy an
essential component in assessing venue performance? Before this issue is settled
empirically, it is impossible to construct comparisons of individual venues, or even
contrast dark versus lit markets in the aggregate.
The answer to the question is yes. Even examination of the distribution of trading
strategies across venue types suggests that different venues are used to implement
disparate strategy choices. We begin with some comments on our data and a few
definitions, which are required to make the analysis transparent.
DATA AND TERMINOLOGY
A blizzard of numbers and charts is avoided by sharply limiting the universe of data
for this analysis. Trading activity is for the U.S. only, and restricted to large cap
stocks. High frequency trading is concentrated in this capitalization subset. All
activity took place over the 2013 calendar year. Small orders are analyzed, up to one
percent of median daily volume. This last restriction is not terribly binding, since the
vast majority of orders submitted for algorithmic trading fall into this category. The
data are not limited to ITG executions; all buy-side participants represented in the
sample trade with a variety of brokers. This means, in particular, that we study order
routing and venues in the large, and the results do not reflect directly on ITG’s
methodology.7
There sometimes is confusion over terminology, and we want to be as clear in this
regard as we are with the data. An algorithm or trading strategy determines how
many shares to be traded at any particular point in time. The router takes these
instructions and determines venues for the execution. A trade or execution is a
single fill, and is typically only a partial fill of the order. Venues are characterized as
being either lit or dark. In terms of pricing models, we restrict most of the analysis to
maker/taker venues. An inverted model is taker/maker, and comprises about five
percent of the data. We present differences between the two as suits the context; if
a chart does not differentiate between the two, it is based on maker/taker only.
Algorithms are aggregated across brokers into several categories.8 We refer to
scheduled strategies, which include VWAP, TWAP, and Participation strategies. Dark
denotes a liquidity-seeking strategy concentrating on dark pools, while opportunistic
is general liquidity-seeking in nature. IS is shorthand for implementation shortfall,
creating execution patterns based on cost and risk minimization. Non-algo in the
charts is generally desk trading, while Other is dominated by direct market access
and and includes some pairs trading.
Ian Domowitz and Henry Yegerman, “Algorithmic Trading Usage Patterns and their Costs,” Journal of Trading, Summer 2011.
We also exclude extreme cost outliers, defined by plus or minus 100 basis points (bps) versus the order decision price, as well as any orders
for which an implementation shortfall calculation is not possible for any reason; similarly, individual trades for which t+1 and t+5 second
reversion cannot be computed are excluded, as well as orders with limits greater than 100 bps away from next price post decision price.
8
See Ian Domowitz and Henry Yegerman, “Algorithmic Trading Usage Patterns and their Costs,” Journal of Trading, Summer 2011, for further
detail and additional characterizations of strategies from broker algorithms.
6
7
3
26 August 2014
FIGURE 1
Strategy Distribution by Venue Type
100%
80%
60%
40%
20%
0%
Dark
DARK
Maker/Taker
IMPLEMENTATION
OPPORTUNISTIC
Taker/Maker
OTHER
SCHEDULED
Source: ITG
SOME VENUE CHARACTERISTICS BY STRATEGY TYPE
Trading strategies can affect venue analysis in several ways, given the profile of fills
flowing back to the algorithm from routing information, and disparities in how
strategies cycle child orders into the markets. Routers also may be tuned to favor
certain venues depending on strategy, a more direct approach. Venue types are
characterized by different mixes of strategies which feed them.
Some disparities in venues across strategies are simply intuitive. Dark pool trading is
dominated by dark aggregation strategies, no surprise there. Scheduled strategies
favor lit markets, but even within that category, the taker/maker model exhibits
proportionately higher activity, possibly due to the interaction between strategy and
fee structure. Implementation shortfall algorithms make up about a quarter of lit
market activity, but account for less than 10 percent of what we see in dark pools.
Similarly, a small fraction of DMA orders hit the dark pools directly, compared to the
lit markets.
Fill size typically is considered a performance metric, but may also be a characteristic
stemming from segmentation of participant types in any given venue. We cannot
observe segmentation directly. Fill size is traceable to the nature of strategies,
however, given the interaction of trade size with sizes determined by the strategy
itself. The disparity across venue types is illustrated in Figure 2.
FIGURE 2
Average Fill Size by Venue Type and Strategy
250
Fill Size (sh)
200
150
100
50
0
Dark
DARK
Source: ITG
Maker/Taker
OPPORTUNISTIC
IMPLEMENTATION
Taker/Maker
SCHEDULED
4
26 August 2014
At the lower levels of the percentage of average daily volume, there are similarities
across scheduled strategies and implementation shortfall, but also differences.
Opportunistic trading may look similar to dark aggregation, venue type by venue type,
but it is quite different from the other two categories in the chart.
Such results are not figments of the aggregation process. Results on fill size,
disaggregated by venue, appear in Figure 3.
FIGURE 3
Distribution of Fill Sizes for Selected Strategy Types and Venues
SCHEDULED ALGORITHMS
BARCLAYS ATS
BATS YEXCHANGE, INC.
EDGA EXCHANGE
NASDAQ OMX BX
JPMX
NYSE ARCA
BATS EXCHANGE
EDGX EXCHANGE
NEW YORK STOCK EXCHANGE, INC.
NASDAQ ALL MARKETS
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
OPPORTUNISTIC ALGORITHMS
DEUTSCHE BANK SUPER X
EDGA EXCHANGE
BARCLAYS ATS
CREDIT SUISSE AES CROSSFINDER
JPMX
EDGX EXCHANGE
NEW YORK STOCK EXCHANGE, INC.
NYSE ARCA
BATS EXCHANGE
NASDAQ ALL MARKETS
0%
A)1-199
B)200-499
C)500-999
D)>1000
Source: ITG
Although it is possible to compare these charts line by line, that is not the goal of the
exercise. Rather, the two charts, representing two strategy choices, should be viewed
as heat maps, or mosaiques, with respect to underlying activity. In that sense, the
two graphs look very different. Strategy type clearly matters with respect to fill size,
regardless of whether it is viewed as a performance metric or as simply a
characteristic of venues.
5
26 August 2014
COMMON PERFORMANCE METRICS
If venue characteristics vary across strategies, performance comparisons also must
be affected. Even the simplest performance statistic, average spread capture, varies
across venues by strategy.
FIGURE 4
Spread Capture by Venue Type and Strategy
0.6
Spread Capture
0.5
0.4
0.3
0.2
0.1
0
Dark
DARK
Maker/Taker
OPPORTUNISTIC
IMPLEMENTATION
Taker/Maker
SCHEDULED
Source: ITG
Differences in spread capture for maker/taker markets vary by as much as 16
percent, for example. That figure rises to 45 percent for the taker/maker paradigm.
Dark market spread capture can vary by 46 percent across strategy types.
In terms of the “classic” measure of trading performance, these differences translate
into variation across venues, but traceable to trading strategies, as opposed to
distinct properties of the pools themselves. Figure 5 contains results on
implementation shortfall cost at the order level.
FIGURE 5
Implementation Shortfall Cost by Venue and Strategy
0
-1
Cost (bps)
-2
-3
-4
-5
-6
-7
DARK
Source: ITG
Dark
OPPORTUNISTIC
Maker/Taker
IMPLEMENTATION
SCHEDULED
Taker/Maker
NON-ALGO
OTHER
6
26 August 2014
We expand the strategy set in this chart to include DMA activity (Other) and trading
desk activity (Non-Algo). The difference between these categories for the maker/
taker venues is as much as 80 percent. Even a comparison between implementation
shortfall and opportunistic strategies for lit markets differs by 29 percent. Dark
aggregation juxtaposed with scheduled strategies yields a 49 percent difference in
cost for the same venue type. Any comparison of dark pools to lit markets fails, if
strategy is not taken into account.
Given the idiosyncratic nature of time stamps in trading systems, calculation of
implementation shortfall cost is typically possible only when a broker strategy is
considered. For this reason, many have resorted to reversion metrics, which depend
only on a snapshot of price, at and after, an individual fill. We now turn to the effect of
strategy on such analysis.
REVERSION METRICS
Reversion measures differ by the time after the fill. The idea is that if a great deal of
reversion is exhibited in price over very short time horizons, then the fill is bad
relative to what might have happened in the absence of toxicity in a venue. We
illustrate one second and five second timeframes in Figure 6.
FIGURE 6
Reversion Seconds After the Fill
ONE SECOND POST TRADE REVERSION BY VENUE TYPE AND STRATEGY
0
-0.1
-0.2
Cost (bps)
-0.3
-0.4
-0.5
-0.6
-0.7
-0.8
Dark
Maker/Taker
Taker/Maker
FIVE SECOND POST TRADE REVERSION BY VENUE TYPE AND STRATEGY
0
-0.1
Cost (bps)
-0.2
-0.3
-0.4
-0.5
-0.6
-0.7
-0.8
Dark
DARK
Source: ITG
IMPLEMENTATION
Maker/Taker
NON-ALGO
OPPORTUNISTIC
Taker/Maker
OTHER
SCHEDULED
7
26 August 2014
The differences across strategies, within and between types of venue, are clearly
evident. Within maker/taker markets, at the five second level, the variation between
scheduled and opportunistic strategies is over 60 percent. For 1 second reversion,
the difference in cost for dark pools across strategy types ranges up to 73 percent.
Most of such deviations jump right off the chart.
Disaggregated results appear in Figure 7 below. We limit these charts to 1-second
reversion, but the five-second graphs tell the same story.
FIGURE 7
T+1 Second Reversion
DARK ALGORITHMS
NASDAQ ALL MARKETS
JPMX
NYSE ARCA
NASDAQ/NMS (GLOBAL MARKET)
MS POOL ATS
ITG POSIT
BATS EXCHANGE
DEUTSCHE BANK SUPER X
BARCLAYS ATS
GOLDMAN SACH MTF
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
IS ALGORITHMS
BATS YEXCHANGE, INC.
NASDAQ OMX BX
EDGA EXCHANGE
BARCLAYS ATS
EDGX EXCHANGE
NYSE ARCA
ITG POSIT
BATS EXCHANGE
NEW YORK STOCK EXCHANGE, INC.
NASDAQ ALL MARKETS
A)<-1 bps
Source: ITG
B)-1 to -0.5 bps
C)-0.5 to 0.5 bps
D)0.5 to 1 bps
E)> 1 bps
8
26 August 2014
FIGURE 7 (CONT.):
Distribution of T+1 Second Reversion Opportunistic Algorithms
OPPORTUNISTIC ALGORITHMS
DEUTSCHE BANK SUPER X
EDGA EXCHANGE
BARCLAYS ATS
CREDIT SUISSE AES CROSSFINDER
JPMX
EDGX EXCHANGE
NEW YORK STOCK EXCHANGE, INC.
NYSE ARCA
BATS EXCHANGE
NASDAQ ALL MARKETS
0%
10%
20%
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
30%
40%
50%
60%
70%
80%
90%
100%
SCHEDULED ALGORITHMS
BARCLAYS ATS
BATS YEXCHANGE, INC.
EDGA EXCHANGE
NASDAQ OMX BX
JPMX
NYSE ARCA
BATS EXCHANGE
EDGX EXCHANGE
NEW YORK STOCK EXCHANGE, INC.
NASDAQ ALL MARKETS
A)<-1 bps
Source: ITG
B)-1 to -0.5 bps
C)-0.5 to 0.5 bps
D)0.5 to 1 bps
E)> 1 bps
9
26 August 2014
Viewed as heat maps, the pictures are obviously different. Within each strategy
category, one might now attempt a comparison across individual venues. We give a
flavor of this approach through an examination of lit and dark markets along the
dimension of price reversion in Figure 8.
FIGURE 8
Reversion in Dark and Lit Markets for Implementation Shortfall Strategies
SELECTED DARK POOLS - IS ALGORITHMS
BANK OF AMERICA MERRILL LYNCH ATS
BARCLAYS ATS
CITI MATCH
CREDIT SUISSE AES CROSSFINDER
DEUTSCHE BANK SUPER X
GOLDMAN SACH MTF
ITG POSIT
JPMX
KNIGHT MATCH ATS
MS POOL ATS
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
20%
30%
40%
50%
60%
70%
80%
90%
100%
SELECTED LIT VENUES - IS ALGORITHMS
BATS EXCHANGE
BATS YEXCHANGE, INC.
EDGA EXCHANGE
EDGX EXCHANGE
NASDAQ ALL MARKETS
NASDAQ OMX BX
NASDAQ/NMS (GLOBAL MARKET)
NEW YORK STOCK EXCHANGE, INC.
NYSE ARCA
0%
A)<-1 bps
10%
B)-1 to -0.5 bps
C)-0.5 to 0 bps
D)0 to 0.5 bps
E)0.5 to 1 bps
F)> 1 bps
Source: ITG
Comparisons across charts illustrate performance in terms of price reversion
between dark and lit markets. Within each chart, individual venues may be
compared, controlling for strategy. We are not at the point of “league tables,”
however. Controlling for market conditions should be a part of such an exercise.
Nevertheless, examination of results for implementation shortfall strategies yields a
couple of discussion points. The first point is the appropriate comparison, which
depends upon one’s perspective. If the issue at hand is the proportion of activity
which exhibits little reversion, then the correct comparison within and across charts
might be a range around zero, between -0.5 bps and 0.5 bps. If the probability of a
bad outcome is of interest, the tails of the distribution matter, and the correct item
of attention might be the unlimited range of less than -1 bps.
10
26 August 2014
On the basis of ‘little reversion,’ and for implementation shortfall strategies only, the
New York Stock Exchange exhibits only about 13 percent of activity within the range
around zero. The range between best and worst performers is rather narrow in the lit
markets, however. BATS, for example, exhibits about 16 percent of activity around
zero reversion. There are more significant differences in the extreme tail behavior.
The difference between best and worst performance across lit venues with respect to
the probability of a bad outcome is roughly 44 percent.
Comparisons across dark pools exhibit more variation in the distribution of
outcomes. The difference between pools with respect to straddling zero reversion is
as high as 65 percent. For ITG’s POSIT®, 43 percent of fills are close to zero
reversion, and few venues fall below 30 percent in that category. On that basis, dark
pools, controlling for this particular strategy type, dominate lit markets in terms of
performance.
THE ANSWER TO THE QUESTION: A LAST LOOK
A complete set of comparisons across lit venues, dark pools, and comparisons
between the two types, is beyond the simple goal of this paper; we will return to that
exercise in the near future. We emphasize our main point one more time before
closing, however. Figure 9 illustrates dark pool performance in reversion terms, now
for dark aggregator strategies only.
FIGURE 9
Distribution of Price Reversion: Dark Pools – Dark Aggregator Strategies
BANK OF AMERICA MERRILL LYNCH ATS
BARCLAYS ATS
CREDIT SUISSE AES CROSSFINDER
DEUTSCHE BANK SUPER X
GOLDMAN SACH MTF
ITG POSIT
JPMX
KNIGHT MATCH ATS
MS POOL ATS
UBS PIN
(UBS PRICEIMPROVEMENT NETWORK)
0%
A)<-1 bps
10%
B)-1 to -0.5 bps
20%
30%
C)-0.5 to 0 bps
40%
50%
60%
D)0 to 0.5 bps
70%
80%
E)0.5 to 1 bps
90%
100%
F)> 1 bps
Source: ITG
This chart should be compared to that of dark pool reversion in Figure 8. The
percentage of fills for which there is reversion around zero is now much higher than
observed for implementation shortfall strategies. The range for the latter is roughly
between 15 and 45 percent. The use of a dark algorithm, which by its nature is tuned
to dark pools, yields a range between 25 and over 60 percent; in fact, for nine of ten
pools, the bottom of the range is above 40 percent. The percentage of zeroreversion fills for ITG’s POSIT®, for example, is 50 percent higher for dark aggregation
strategies, as opposed to implementation shortfall.
26 August 2014
11
The difference in comparisons, and the potential advantage to using an algorithm
tuned to the venue type, are also evident in an examination of the tails of the price
reversion distribution. We noted large percentages of poor outcomes, in the case of
the implementation shortfall strategies. The tail of the distribution for reversion in
the dark, using dark aggregation, is much smaller, overall and for nine of the ten
individual venues. Performance around zero reversion improves, and the probability
of high reversion numbers falls.
The answer to our original question is now easily stated: consideration of trading
strategy is an essential component in assessing venue performance. This is
particularly true, when attempting to compare and contrast different market
structures, such as dark pools versus lit markets. It also is important in assessing
venue quality on a disaggregated basis, within market structure categories. The
results suggest that proper “tuning” of routing functionality to strategy may improve
performance as well. A proper treatment of this suggestion is a natural next, and
constructive, step in venue research.
© 2014 Investment Technology Group, Inc. All rights reserved. Not to be reproduced or retransmitted without permission. 81214- 14098
These materials are for informational purposes only and are not intended to be used for trading or investment purposes or as an offer to sell
or the solicitation of an offer to buy any security or financial product. 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. These materials do not provide any form of advice (investment, tax or legal). All trademarks, service marks, and
trade names not owned by ITG are the property of their respective owners. The positions taken in this document reflect the judgement of the
individual author(s) and are not necessarily those of ITG.
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 First Floor, Block A, Georges Quay, 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 (CMS Licence No. 100138-1), and ITG Australia Limited
(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.