n Chapter 6

Algorithms in action

Dark pools and

algorithmic trading 1

The evolution of ‘dark pools’ and how algorithms access and interact
with non-displayed liquidity

George Sofianos*

D espite the title, we will

refrain from using the term
‘dark pools’, which has sinister
The different types of
non-displayed liquidity
We define non-displayed liquidity
connotations! Instead, we will use broadly to include any willingness
the more neutral and more general to buy or sell that is not publicly 
term ‘non-displayed liquidity’. displayed. Exhibit 1 lists the
There is nothing sinister about different types of non-displayed
non-displayed liquidity: most liquidity at execution venues. Non-
liquidity is non-displayed, non- displayed liquidity at execution
displayed liquidity has always been venues consists of reserve orders on
with us, and will always be with electronic limit order books, orders
us. The nature of non-displayed represented by floor brokers on the
liquidity, however, is changing floor-based exchanges, the orders
over time. Moreover, the rise in public crossing networks, and
of algorithmic trading creates non-displayed liquidity at agency
challenges on how algorithms can brokers and broker-dealers.
best interact with non-displayed Electronic limit order books
liquidity. In this chapter we first (ELBs) electronically aggregate and
describe the different types of non- display limit orders (the quantity
displayed liquidity, summarise traders are willing to buy and
*
the evolution of non-displayed sell at different prices). ELB 1 in George Sofianos,
vice president,  
liquidity over time and then discuss Exhibit 2, for example, displays Equity Execution
how algorithms can take advantage traders willingness to sell 1,000 Strategies

of non-displayed liquidity. shares at $25.01 or higher. ELBs Goldman Sachs

1 Also see Kwaku Abrokwah and George Sofianos, ‘Accessing Displayed and Non-Displayed Liquidity’, Journal
of Trading, Vol.1, No. 4, Fall 2006, pages 47 to 57.

n algorithmic trading 2nd edition n a buy-side handbook n the trade 2007

 n Chapter 6

Algorithms in action

Exhibit 1:
Different types of non-displayed liquidity at execution venues

Crossing networks Crossing networks

Reserve orders
Working orders Working orders
Electronic Agency Broker- Dealer capital
limit order brokers dealers
books
(ELBs)

Traditional
floor-based
Stand-alone exchanges
crossing networks Reserve orders
Floor-broker orders
Specialist capital

2 Throughout this
may also have a reserve feature broker-dealer sponsored crossing
chapter we will (in addition to the displayed networks (e.g. Goldman Sachs
refer to electronic
limit order
quantities, they may electronically SIGMASM X). Agency brokers may
2 books as ELBs.
A broader group
aggregate non-displayed liquidity also sponsor crossing network
than just ECNs,
at different prices). At ELB 1 in (e.g. Pulse Trading’s recently
ELBs include Exhibit 2, traders are willing to sell announced BlockCross). The
both registered
exchanges with
an additional 9,000 share at $25.01, term ‘dark pools’ usually refers
ELB platforms but this quantity is not displayed.2 to just crossing networks but
(e.g. Nasdaq
Single Book) and
Crossing networks do non-displayed liquidity is much
ECNs (ELBs not not display any orders. They broader, with crossing networks
registered as an
exchange, e.g.
electronically aggregate accounting for a relatively small
BATS). non-displayed liquidity and part of total non-displayed
3 The SEC defines anonymously match buy and sell liquidity.
crossing networks
as “systems that
orders, usually at the prevailing Agency brokers execute non-
allow participants market mid-quote.3 Once a displayed client orders over
to enter un-
priced orders,
potential match is established time and across liquidity pools.
which are then between two orders, some crossing Agency brokers may also cross
executed with
matching interest
networks allow negotiation over client orders and proactively – but
at a single price, the execution price. Crossing can discreetly – seek counterparties
typically derived
from the primary
be continuous (e.g. Liquidnet) (non-displayed latent liquidity).
public market or at specific times (e.g. Nasdaq Agency brokers do not use
for each crossed
security.” SEC
Crossing Network). Another their own capital to facilitate
Concept Release, distinction is between stand-alone trades. Non-displayed liquidity
Regulation of
Exchanges, 23
independent crossing networks at agency-brokers, therefore,
May 1997 (p.15). (e.g. NYFIX Millennium) and consists of just the client orders

n algorithmic trading 2nd edition n a buy-side handbook n the trade 2007

 n Chapter 6

Algorithms in action

Exhibit 2:
Reserve orders in electronic limit order books (ELB)

&-#
 &-#

Buy quantity Price Sell quantity Buy quantity Price Sell quantity
$25.03 500 0 $25.03 3,000 0
$25.02 1,000 5,000 4NBSU
SPVUFS $25.02 4,000 0
$25.01 1,000 9,000 $25.01 4,000 0

3,000 1,000 $24.99 0 6,000 $24.99

3FTFSWF
PSEFST
ELB = Electronic Limit Order Book

the brokers are working over liquidity at investors. This non-

time. Broker-dealers, in addition displayed liquidity consists of the
to performing all the functions of orders on the trading desks of
agency brokers, may also use their investment managers and, most
own capital to facilitate trades.4 importantly, latent liquidity. We
Non-displayed liquidity at broker- define latent liquidity as non-
dealers, therefore, in addition displayed liquidity that must be 3
to the client orders worked over proactively solicited. Suppose
time, also includes dealer capital. investment manager AlphaMax
Algorithms at agency brokers and wants to buy 200,000 shares XYZ,
broker-dealers are themselves big the current best ask is $25.01 and
depositories of non-displayed the total amount of displayed and
liquidity. non-displayed liquidity at the
The traditional floor-based various execution venues at this
exchanges combine ELBs (e.g. the price is 50,000 shares. Another
NYSE Display Books), broker- investment manager, BestReturn,
dealers (e.g. the NYSE specialists) is not actively in the market but,
and agency brokers (e.g. the NYSE given the right opportunity, may
floor brokers), all at one physical be interested in selling XYZ shares.
location. The main advantage If AlphaMax could somehow
of floor-based exchanges is the directly indicate to BestReturn its
centralisation of the interaction willingness to buy XYZ, a trade 4 A market-maker
is a broker-dealer
between floor brokers representing may take place at or close to the with special
non-displayed liquidity, specialists best ask. Matching AlphaMax responsibilities:
e.g., must
and ELBs at one physical location. with BestReturn, tapping latent display buy
In addition to non-displayed liquidity, is one of the functions and sell quotes
continuously
liquidity at execution venues, we of traditional agency brokers and during the regular
must also consider non-displayed broker-dealers. Brokers access trading hours.

n algorithmic trading 2nd edition n a buy-side handbook n the trade 2007

 n Chapter 6

Algorithms in action

n “Non-displayed liquidity has been around for a long time

in the form of orders represented by floor brokers on the
trading floors of the NYSE and the other exchanges. In this
respect, therefore, non-displayed liquidity is as old as the
NYSE itself.”
latent liquidity through carefully the NYSE and the other exchanges.
targeted discreet calls to potential In this respect, therefore, non-
counterparties. displayed liquidity is as old as
Most liquidity is not displayed the NYSE itself. Non-displayed
and non-displayed liquidity is liquidity has also always been
fragmented across a large number around for a long time in the form
of broker-dealers, crossing networks of non-displayed orders worked by
and investment managers. In the upstairs brokers and in the form of
next section we discuss how the dealer capital.
 nature of non-displayed liquidity is In the late 1980s, technology
changing over time. began to rapidly change the nature
of non-displayed liquidity with the
The evolution of non-displayed appearance of the first electronic
liquidity crossing networks: ITG’s POSIT
Non-displayed liquidity is a and Instinet’s The Crossing
fundamental feature of financial Network. The 1990s saw the rise
markets. It arises from traders’ of ELBs. Initially, ELBs were only
concern that if they widely electronically displaying liquidity,
advertise their intention to buy a but over time ELBs added reserve
large amount of stock, they will orders, incorporating non-
drive the price of the stock up displayed electronic liquidity. At
ahead of their trade. Non-displayed the same time crossing networks
liquidity has always been around, were innovating and evolving.
but its nature is changing over Liquidnet’s appearance in April
time. Exhibit 3 presents a stylised 2001 was another milestone.
timeline of the evolution of non- Liquidnet’s main innovation was to
displayed liquidity in the US. embed itself in the buy-side order
Non-displayed liquidity has management systems. In this way,
been around for a long time in the buy-side traders could work the
form of orders represented by floor order while simultaneously seeking
brokers on the trading floors of natural counterparties among

n algorithmic trading 2nd edition n a buy-side handbook n the trade 2007

 n Chapter 6

Algorithms in action

Exhibit 3:
The evolution of non-displayed liquidity

Broker- 2nd generation

dealer Late 1980s: initial crossing networks
capital crossing networks

Reserve orders
Reserve orders Broker-dealer
Exchange crossing facilities
trading
floors Late 1990s

Liquidnet’s other buy-side users. n Technological advances

NYFIX Millennium and Pipeline improved the capabilities
are other examples of second- and performance of crossing
generation crossing networks. systems and made it easier
Liquidnet, Pipeline, NYFIX to create and scale-up new
Millennium are – at least for the systems from scratch. 5
time being – agency brokerage n Commission rates at the main
crossing networks not affiliated crossing networks remained
with any of the major full-service stubbornly high. For example,
broker-dealers. Over the past 2 cents per share at Liquidnet
couple of years broker-dealers (8bps in Europe).6
have started introducing their own n The reduction in the NYSE’s
crossing networks, like Goldman share of trading volume over
Sachs’ SIGMA X. Exchanges are the last couple of years (from
also introducing crossing networks: 84% in 2001 to 54% in March
the ISE Stock Exchange MidPoint 2007) together with the rollout
MatchTM in September 2006, the of more automation as part of
5 The ISE launched
Nasdaq Intraday and Post-Trade the NYSE Hybrid is reducing MidPoint Match
Crosses in June 2007 and the NYSE the importance of the NYSE on September
8, 2006 and
MatchPoint coming up later in floor as a pool of non-displayed Nasdaq launched
2007.5 All combined, currently liquidity. The migration of Crossing
Network on June
there are at least two dozens non-displayed liquidity away 4, 2007. The
crossing networks operating in the from the NYSE creates an NYSE acquired
MatchPoint
US alone. opportunity for new venues for Trading Inc. on
There are several reasons for the non-displayed liquidity. July 17, 2006.
6 The TRADE, Issue
proliferation of crossing networks n The buy-side’s increasing use of 11, Jan – Mar 07,
in the US: low-touch execution channels page 113.

n algorithmic trading 2nd edition n a buy-side handbook n the trade 2007

 n Chapter 6

Algorithms in action

Exhibit 4:
Crossing networks in the long run

-FBSOJOH
GSPN
UIF
FWPMVUJPO
PG
&-#T
/BTEBR

Smart-routers eased 4JOHMF
#PPL
the transition
*OTUJOFU
&$/
5 –10 years

Breakthrough Imitation, experimentation, innovation Consolidation

5IF
FWPMVUJPO
PG
DSPTTJOH
OFUXPSLT

5 –10 years?

Breakthrough *NJUBUJPO
FYQFSJNFOUBUJPO
JOOPWBUJPO Consolidation?

We are here
6
is reducing the average initial breakthrough ELB (the
commission rates received by Instinet ECN in 1997), to at some
full-service broker-dealers. point more than a dozen ECNs
Pressure on commissions (maximum market share less
is forcing broker-dealers to than 20%), and ultimately to the
streamline their execution inevitable consolidation (Nasdaq
process and maximise the Single Book): Instinet merged
internal crossing of client with Island to form INET (2002)
orders. and then Nasdaq acquired both
The current proliferation of the INET and Brut ECNs and
crossing networks mirrors the consolidated them with its own
proliferation of ELBs in the ELB (SuperMontage) into Single
market for Nasdaq stocks in the Book (2006). The consolidated
1990s. This proliferation is part Single Book now accounts for
of the typical evolution process more than 50% of the trading
from major innovation through volume in Nasdaq stocks.
imitation, experimentation ELB consolidation was
and innovation to eventual inevitable because of economies
consolidation. Exhibit 4 illustrates of scale, network externalities,
the evolution of ELBs from the and because ELB innovations

n algorithmic trading 2nd edition n a buy-side handbook n the trade 2007

 n Chapter 6

Algorithms in action

n “Despite their proliferation, crossing networks remain

marginal players in terms of their share of total executed
volume. The independent crossing networks account for
about 3% of the total executed volume.”

(e.g. reserve orders) were easily Despite their proliferation,

replicated by competitors. As the crossing networks remain marginal
ELBs evolved it became harder to players in terms of their share
distinguish among them, forcing of total executed volume. The
them to compete on price. Under independent crossing networks
pure price competition, taking account for about 3% of the
advantage of economies of scale total executed volume.7 Broker-
is critical, and the lowest-cost dealer crossing networks account
provider captures market share and for another 3%, bringing the
consolidates. Network externalities combined total to 6% of total
accelerate the process – the ELB executed volume. If fill rates differ 
that has the biggest volume has across venues, however, executed
better ability to match order and volume is a poor indicator for
deliver better execution quality. comparing posted non-displayed
It took between five to ten years liquidity. Crossing networks, in
for the market for Nasdaq stocks to particular, have relatively low fill
go from ELB proliferation to ELB rates compared to ELBs and the
consolidation. During these years, traditional exchanges. The crossing
smart-routers eased the transition networks’ share of executed
from proliferation to consolidation volume, therefore, underestimates 7 Estimating
the crossing
by efficiently aggregating the the amount of non-displayed networks share of
electronically displayed liquidity liquidity at these venues. Assuming executed volume
is complicated
across the competing ELBs an average fill rate on crossing because we
into essentially a single virtual networks of 10%, suggests that the must rely on
spotty advertised
electronic book. non-displayed liquidity residing volume by the
Just like ELB consolidation, in these systems is ten-times the various venues
that may be
crossing network consolidation amount actually crossed (6%), double-counted
is also inevitable, and for the or the equivalent to 60% of the and may include
both matched
same reasons. The only questions executed volume in NYSE stocks. and volume
are how long will it take, and The large amount of non- merely routed
through. All our
what venue will emerge as the displayed liquidity sitting in the estimates are
consolidator. crossing networks and the low approximations.

n algorithmic trading 2nd edition n a buy-side handbook n the trade 2007

 n Chapter 6

Algorithms in action

Exhibit 5:
The three main components of an algorithm

Parent order

-JRVJEJUZ
BDDFTT
$IJMEPSEFS

1
PWFS
UJNF HFOFSBUJPO
MPHJD

A B C D E

Child order

A
$IJMEPSEFS

2
QMBDFNFOU
MPHJD

-JRVJEJUZ
BDDFTT

BU
B
QPJOU
PG

8 UJNF

Smart router

3 4NBSUSPVUFS

MPHJD

Execution venues

fill rate create a challenge for displayed liquidity. We set the

algorithms: how to more effectively stage by describing the three main
access and interact with this non- components of a typical algorithm
displayed liquidity. (Exhibit 5).

Algorithms and non-displayed 1. Child-order generation logic:

liquidity This can be defined as the logic
We will now focus on the the algorithm uses to slice-up a
interaction of algorithms and non- parent order and distribute child

n algorithmic trading 2nd edition n a buy-side handbook n the trade 2007

 n Chapter 6

Algorithms in action

n “The proliferation of crossing networks creates a

challenge for algorithms: where to optimally post orders?”

orders across brief time buckets the smart-router logic determines

over execution horizon. A full-day where to place each child or
VWAP algorithm, for example, grandchild order across the various
will slice up the parent order in execution venues. The smart-router
proportion to the historical volume logic should optimise execution
smile over the day. Ideally, the child across as many displayed and
order generation logic should have non-displayed pools of liquidity
flexibility to re-allocate child orders as possible, while minimising
across time buckets and its market information leakage.
footprint should not be predictable.
Non-displayed liquidity comes
2. Child-order placement logic into play in all three components
This is the logic the algorithm uses of an algorithm. As an algorithm’s
to determine how to execute each child order generation logic slices 
child order within its designated up the parent order, the child
time bucket. For example, to order placement logic immediately
execute the child order as a market begins actively working the first
order or as a limit order and try to child order (A in Exhibit 5).
capture the spread? If a limit order, The residual parent order (child
at what limit price and when and orders B through E) remains non-
how to adjust this price? When to displayed within the algorithmic
switch a limit order to a market engine waiting for the scheduled
order to guarantee execution? The trigger to place the next child
child order placement logic may order. The child generation logic
further subdivide a child order itself, therefore, creates a large pool
into grandchild orders. Just like of resident non-displayed liquidity
the child order generation logic, that can potentially interact with
the placement logic should have other orders.
flexibility to re-allocate child orders At GSAT® (Goldman Sachs
across time buckets and should not Algorithmic Trading), the extent to
leave a predictable footprint. which the residual non-displayed
parent orders take advantage of
3. Smart-router logic non-displayed liquidity ahead
Once the algorithm determines of their scheduled time depends
how to place each child order, on how rigidly each particular

n algorithmic trading 2nd edition n a buy-side handbook n the trade 2007

 n Chapter 6

Algorithms in action

n “To avoid the risk of double-execution, algorithms

distribute the quantity to post across the various crossing
networks… This allocation is sub-optimal and this is exactly
one of the problems the fragmentation of non-displayed
liquidity creates for optimal execution in general.”

algorithm prescribes a time-size a minimum quantity to respond to.

schedule. Algorithms that rigidly The GSAT child order placement
prescribe a time schedule (e.g. logic also dynamically adapts to
TWAP or VWAP algorithms) bring forward or delay executions
limit the extent the child order depending on the liquidity
generation logic can take advantage experience of the current child
of non-displayed liquidity outside order executing in the market. The
the prescribed schedule. Even for child order generation logic of the
10 these algorithms, however, the GSAT portfolio algorithm (PORT
GSAT child order generation logic X), in taking advantage of non-
retains some flexibility to interact displayed liquidity is subject to a
with non-displayed liquidity at further constraint: that the overall
favourable prices, consistent with portfolio risk exposure does not
not getting too far ahead of the get out of line over the execution
prescribed time-size schedule. horizon.
Algorithms that do not prescribe The proliferation of crossing
a rigid time-size schedule (e.g. networks creates a challenge for
shortfall algorithms) allow greater algorithms: where to optimally
flexibility for the parent order post orders? We can use a simple
to interact with non-displayed example to illustrate one of the
liquidity. The GSAT 4Cast and challenges. Suppose there are only
Sonar algorithms, for example, two crossing networks, CN-A and
while slicing up and sequentially CN-B, and the algorithm wants
routing child orders for execution, to post a 100,000-share buy order
post without displaying a large in symbol XYZ. There is a non-
part of the residual parent order displayed 200,000-share XYZ sell
on the various crossing networks. order in CN-A and nothing in
To avoid information leakage to CN-B. Since the liquidity is non-
predators ‘pinging’ the crossing displayed, to the algorithms CN-A
networks, GSAT algorithms specify and CN-B look identical. Should

n algorithmic trading 2nd edition n a buy-side handbook n the trade 2007

 n Chapter 6

Algorithms in action

the algorithm post the whole of The GSAT Sonar algorithm

the 100,000 shares at each venue posts limit orders in the Goldman
or distribute the 100,000 shares Sachs SIGMA X crossing network
between the two venues? The first and in several external crossing
strategy increases the probability networks. GSAT is also upgrading
of crossing the most quantity: the all its algorithms to post ‘pegged’
algorithm will cross all 100,000 at limit orders in multiple crossing
CN-A. This strategy, however, runs networks.
the risk of double execution and The proliferation of crossing
overfilling the order: what if part of networks also creates challenges
the order also crosses at CN-B? for the smart-router logic
To avoid the risk of double- of algorithms: where to seek
execution, algorithms distribute the liquidity? In seeking liquidity, the
quantity to post across the various GSAT Sonar algorithm takes full
crossing networks. For example, advantage of the Goldman Sachs
post 50,000 shares at CN-A and SIGMA smart-router and SIGMA
50,000 shares at CN-A. This X crossing network. SIGMA X
allocation is sub-optimal and this is was launched by GSEC (Goldman 11
exactly one of the problems the Sachs Execution & Clearing, L.P.),
fragmentation of non-displayed the Goldman Sachs electronic
liquidity creates for optimal executions affiliate, in July 2005,
execution in general. One way to complementing its SIGMA smart-
get around this problem and router introduced in 2002. SIGMA
increase quantity crossed is to learn X combines a host of external
from experience, posting more size liquidity providers (XLPs) together
in crossing networks with higher with limit orders clients post
fill rates. Another way to increase directly into SIGMA X to create a
quantity crossed is to quickly virtual consolidated pool of private
redistribute posted orders towards non-displayed liquidity. The
the venue that just executed. external liquidity providers consist
Continuing our example, as soon of:
as the 50,000 shares execute at CN- n Crossing networks, e.g.
A, quickly reallocate the remaining Liquidnet H2O and NYFIX
50,000 shares across the two Millennium.
venues, so another 25,000 shares n Electronic market makers, e.g.
execute at CN-A, and so on. One Knight and ATD (Automated
downside of this strategy is that the Trading Desk).
redistributed orders may lose time n Goldman Sachs (GSCO)
priority. broker-dealer liquidity

n algorithmic trading 2nd edition n a buy-side handbook n the trade 2007

 n Chapter 6

Algorithms in action

Exhibit 6:
SIGMA smart router and SIGMA X crossing network
Liquidity-seeking -JRVJEJUZTFFLJOH
PSEFST
TUSFBNJOH

(market orders) OP
EFMBZ
(4&$
4*(." /POEJTQMBZFE
MJRVJEJUZ %JTQMBZFE
MJRVJEJUZ
Client orders
FIX direct
REDIPlus
GSAT Sonar SIGMA X External Electronic GSCO
4*(."
9 posted crossing market Exchanges ELBs Broker-
liquidity
liquidity networks makers dealer
Liquidity-posting Opt-out quotes
(limit orders)

GSEC &MFDUSPOJD
FYFDVUJPO
TFSWJDFT GSCO 'VMMTFSWJDF
CSPLFSEFBMFS

1. Goldman Sachs Execution & Clearing, L.P., an affiliate of Goldman, Sachs & Co., is a separately registered broker-dealer that offers
its clients order entry and routing capabilities to a variety of US and non-US market centres either on an agency basis or purely as a
technology provider.

consisting of the orders logic of GSAT algorithms like

2 executed by the Goldman Sonar also enters liquidity-seeking
Sachs full-service broker-dealer orders into SIGMA. SIGMA first
affiliate. SIGMA X clients instantaneously streams these
may opt-out from interacting orders through the SIGMA X
with the GSCO pool of non- liquidity pool to find a match or
displayed liquidity. inside the consolidated best bid
Through REDIPlus® or FIX and offer. In this way, liquidity-
connections, GSEC clients can post seeking orders access all the various
block-sized orders in SIGMA X. pools of non-displayed liquidity
These orders are non-displayed aggregated by SIGMA X. If no
top-of-book limit orders and match is made within SIGMA
interact with incoming flow while X, SIGMA immediately accesses
waiting for a natural cross. GSEC the public markets and routes
clients posting orders in SIGMA X and executes orders against the
include the GSAT Sonar algorithm. displayed liquidity and ELB reserve
Exhibit 6 summarises how the orders, or against the broker-dealer
SIGMA smart-router and SIGMA public quotes. Every day over 300
X interact to maximise crossing million shares pass through the
opportunities. Through REDIPlus SIGMA smart-router.
and FIX connections, GSEC clients Reserve orders on multiple
enter liquidity-seeking orders into ELBs, just like multiple crossing
SIGMA. The child order placement networks, also challenge

n algorithmic trading 2nd edition n a buy-side handbook n the trade 2007

 n Chapter 6

Algorithms in action

algorithms. Because reserve orders and latent liquidity. Some venues

cannot be seen, they are difficult restrict access to their liquidity
to optimally consolidate across pool. Liquidnet, for example,
venues. In Exhibit 2, for example, only allows buy-side firms to
the displayed best ask at ELB 1 is participate in its crossing network.
1,000 shares at $25.01 with 9,000 The proliferation of broker-dealer
additional non-displayed shares. crossing networks also creates an
The displayed ask at ELB 2 is 4,000 access issue: will a broker-dealer
shares at $25.01 but with no reserve allow competing broker-dealers
orders. The smart-router receives a to access its crossing network?
market order to buy 10,000 shares. And if access were allowed, will a
For optimal execution, the smart- broker-dealer provide liquidity to a
router should send at least 6,000 competing broker-dealer’s crossing
shares to ELB 1 and the rest (if network? The recently launched
any) to ELB 2, buying all 10,000 BIDS crossing network, formed by
shares at $25.01. The smart-router, a consortium of broker-dealers is
however, cannot see the 9,000 an attempt to solve the problem
reserve shares at ELB 1. Taking of fragmentation of non-displayed 13
just the displayed liquidity into liquidity across competing broker-
account, the smart-router will only dealers. The consortium includes
send 2,000 shares to ELB 1 and Citi, Goldman Sachs, Lehman
the remaining 8,000 to ELB 2. In Brothers, Merrill Lynch, Morgan
this case, the smart-router will buy Stanley and UBS.
6,000 shares at $25.01 and 4,000 Perhaps the greatest challenge
shares at the higher price $25.02, a for algorithms is how to access
sub-optimal execution. latent liquidity. What is the
The SIGMA smart-router uses algorithmic equivalent of the
a feature called MemoryMatch ‘discreet, well-placed’ telephone
to alleviate this problem. call of traditional brokers? What
MemoryMatch keeps track where it is the best way of pulling latent
places reserve orders and optimally liquidity into the market? Latent
uses this information to route liquidity raises fundamental issues
market orders where they are most over how liquidity is generated
likely to fill at the best price. and dynamically replenished.
Execution strategies and algorithms
The way forward that minimise liquidity impact
In interacting with non-displayed throughout an order’s execution
liquidity, algorithms face two horizon may not always be optimal
other challenges: restricted access because they may fail to elicit

n algorithmic trading 2nd edition n a buy-side handbook n the trade 2007

 n Chapter 6

Algorithms in action

sufficient latent liquidity and best accessed the traditional

prolong execution. Aggressively way. And algorithms will never
executing a larger amount upfront, completely replace flesh-and-
8 See, for example,
with higher liquidity impact, may blood brokers. n
Anna Obizhaeva bring more latent liquidity into the
and Jiang Wang,
‘Optimal Trading
market, speeding up execution.8
Strategy and The initial high-impact child
Supply/Demand
Dynamics’,
execution, essentially acts as a
working paper, wake-up call to latent liquidity.
http://mit.edu/
people/obizheav/
It is possible, however, that
OW060408.pdf latent liquidity will always be

This material was prepared by the Goldman Sachs Equity Execution Strategies Group and is not the product of
Goldman Sachs Global Investment Research. It is not a research report and should not be construed as such.
14 The information in this article has been taken from trade data 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. This information
is indicative, based on among other things, market conditions at the time of writing and is subject to change
without notice. Goldman Sachs’ algorithmic models derive pricing and trading estimates based on historical
volume patterns, real-time market data and parameters selected by the GSAT user. The ability of Goldman Sachs’
algorithmic models to achieve the performance described in this article may be impacted by changes in market
conditions, systems or communications failures, etc. Finally, factors such as order quantity, liquidity and the
parameters selected by the GSAT user may impact the performance results.

The opinions expressed in this article are the authors and do not necessarily represent the views of Goldman,
Sachs & Co. These opinions represent the authors’ judgment at this date and are subject to change.

Goldman, Sachs & Co. is not soliciting any action based on this paper. It is for general information and does
not constitute a personal recommendation or take into account the particular investment objectives, financial
situations, or needs of individual users. Before acting on any advice or recommendation in this paper, users should
consider whether it is suitable for their particular circumstances.

GSAT®, SIGMASM, and REDIPlus® are registered service marks of Goldman Sachs Execution & Clearing, L.P.

Copyright: Summer 2007 by Goldman, Sachs & Co.

n algorithmic trading 2nd edition n a buy-side handbook n the trade 2007