U.S. patent application number 15/948430 was filed with the patent office on 2018-08-09 for systems, methods and computer program products for adaptive transaction cost estimation.
This patent application is currently assigned to ITG Software Solutions, Inc.. The applicant listed for this patent is ITG Software Solutions, Inc.. Invention is credited to Milan Borkovec, Ian Domowitz, Mahmoud El-Gamal, Hans G. Heidle, Aaron Schweiger, Konstantine Tyurin.
Application Number | 20180225763 15/948430 |
Document ID | / |
Family ID | 42981740 |
Filed Date | 2018-08-09 |
United States Patent
Application |
20180225763 |
Kind Code |
A1 |
Borkovec; Milan ; et
al. |
August 9, 2018 |
SYSTEMS, METHODS AND COMPUTER PROGRAM PRODUCTS FOR ADAPTIVE
TRANSACTION COST ESTIMATION
Abstract
A system, method and computer program product are provided for
forecasting the transaction costs of a trade using empirical data
and user-defined modeling constraints based on real-time data
regarding changes in market conditions. In preferred embodiments,
the invention acts as a forecaster whereby it accepts inputs from
customers and identifies real-time market analytics, and provides
dynamically adjusted ex ante cost estimates and metrics for the
prevailing market conditions. Specific cost estimation and
optimization algorithms can be provided to model transaction costs
of a specific trade based on empirical data and real-time
variables.
Inventors: |
Borkovec; Milan; (Boston,
MA) ; Domowitz; Ian; (New York, NY) ;
El-Gamal; Mahmoud; (Sugar Land, TX) ; Heidle; Hans
G.; (Quincy, MA) ; Schweiger; Aaron;
(Brookline, MA) ; Tyurin; Konstantine; (Quincy,
MA) |
|
Applicant: |
Name |
City |
State |
Country |
Type |
ITG Software Solutions, Inc. |
Culver city |
CA |
US |
|
|
Assignee: |
ITG Software Solutions,
Inc.
Culver City
CA
|
Family ID: |
42981740 |
Appl. No.: |
15/948430 |
Filed: |
April 9, 2018 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
13423553 |
Mar 19, 2012 |
|
|
|
15948430 |
|
|
|
|
Current U.S.
Class: |
1/1 |
Current CPC
Class: |
G06Q 10/04 20130101;
G06Q 40/06 20130101; G06Q 40/04 20130101 |
International
Class: |
G06Q 40/04 20060101
G06Q040/04; G06Q 40/06 20060101 G06Q040/06; G06Q 10/04 20060101
G06Q010/04 |
Claims
1. A computerized method for adaptive transaction cost estimation,
said computerized method comprising the steps of: a. receiving at a
trade optimization server, electronic information defining a
proposed order to trade one or more assets on an electronic
exchange or marketplace, wherein the electronic information is
stored, by the trade optimization server, in an electronic data
storage facility and comprises, for each of the one or more assets
in the proposed order to trade, information identifying: i.
security name, symbol, or identifier, ii. transaction side, and
iii. total quantity to be traded; b. receiving at the trade
optimization server, electronic market data comprising historical
market data and current market data relating to the one or more
assets; c. determining, through the use of an optimization
algorithm at the trade optimization server, an optimum trading
strategy and a transaction cost estimate for implementing the
proposed order, based upon the proposed trade order, at least one
pre-selected trading parameter, and the electronic market data; and
d. outputting an indication of a quantity of the one or more assets
to be traded in an upcoming time period according to the determined
optimum trading strategy and the transaction cost estimate, the
quantity to be traded being less than or equal to the total
quantity to be traded.
2. The computerized method of claim 1, wherein the method further
includes the step of: e. receiving at the trade optimization
server, a potential trade imbalance of an upcoming time period,
wherein the potential trade imbalance is based on at least the
electronic information reflecting a proposed trade order and the
received electronic market data; wherein the optimum trading
strategy and the transaction cost estimate for implementing the
proposed order are each additionally based on the potential trade
imbalance of an upcoming time period.
3. The computerized method of claim 2, wherein the method further
includes the step of: f. receiving at the trade optimization server
a distribution of realized effective spread and a distribution of
realized permanent price impact of an upcoming time period, wherein
the distribution of realized effective spread and the distribution
of realized permanent price impact are based on at least the
potential trade imbalance of an upcoming time period and the
received electronic market data; wherein the optimum trading
strategy and the transaction cost estimate for implementing the
proposed order are each at least additionally based on the
distribution of realized effective spread and the distribution of
realized permanent price impact of an upcoming time period.
4. The computerized method of claim 1, wherein said step of
receiving at a trade optimization server, electronic information
reflecting a proposed order, further comprises: accessing an order
management system database by at least one computer and retrieving
the electronic information reflecting unplaced orders and creating
the electronic information for the potential order based on the
retrieved information reflecting unplaced orders.
5. The computerized method of claim 2, wherein the at least one
pre-selected trading parameter is selected from the following list:
a. time horizon of a potential order; b. level of discretion of a
potential order; c. expected condition of an upcoming bin; and d.
specification of a function for determining an impact on trade
imbalance by the proposed order, wherein the potential trade
imbalance of an upcoming time period is additionally based on the
function.
6. The computerized method of claim 1, wherein said step of
outputting an indication of a quantity of the one or more assets to
be traded in an upcoming time according to the determined optimum
trading strategy and the transaction cost estimate, further
comprises: communicating to a trading venue via an electronic
network an electronic trade order to trade a quantity of a
corresponding tradable asset equal to the indication of a quantity
output, in the upcoming time period.
7. A system for adaptive transaction cost estimation in an
electronic trading system, comprising: a trading optimization
server configured to receive electronic information defining a
proposed order to trade one or more assets on an electronic
exchange or marketplace, to store the electronic information in an
electronic data storage facility, to receive electronic market data
comprising historical market data and current market data relating
to the one or more assets, to determine, through the use of an
optimization algorithm, an optimum trading strategy and a
transaction cost estimate for implementing the proposed order,
based upon the proposed trade order, at least one pre-selected
trading parameter, and the electronic market data, and to output an
indication of a quantity of the one or more assets to be traded in
an upcoming time period according to the determined optimum trading
strategy and the transaction cost estimate, the quantity to be
traded being less than or equal to the total quantity to be traded;
wherein the electronic information comprises, for each of the one
or more assets in the proposed order to trade, information
identifying: i. security name, symbol, or identifier, ii.
transaction side, and iii. total quantity to be traded.
8. The system of claim 7, wherein the trading optimization server
is further configured to receive a potential trade imbalance of an
upcoming time period, wherein the potential trade imbalance is
based on at least the electronic information reflecting a proposed
trade order and the received electronic market data, and to
determine the optimum trading strategy and the transaction cost
estimate for implementing the proposed order, each additionally
based on the potential trade imbalance of an upcoming time
period.
9. The system of claim 8, wherein the trading optimization server
is further configured to receive a distribution of realized
effective spread and a distribution of realized permanent price
impact of an upcoming time period, wherein the distribution of
realized effective spread and the distribution of realized
permanent price impact are based on at least the potential trade
imbalance of an upcoming time period and the received electronic
market data; wherein the optimum trading strategy and the
transaction cost estimate for implementing the proposed order are
each at least additionally based on the distribution of realized
effective spread and the distribution of realized permanent price
impact of an upcoming time period.
10. The system of claim 7, further comprising a order data
extraction device configured to access an order management system
database by at least one computer and retrieve the electronic
information reflecting unplaced orders and create the electronic
information for the potential order based on the retrieved
information reflecting unplaced orders.
11. The system of claim 8, wherein the at least one pre-selected
trading parameter is selected from the following list: e. time
horizon of a potential order; f. level of discretion of a potential
order; g. expected condition of an upcoming bin; and h.
specification of a function for determining an impact on trade
imbalance by the proposed order, wherein the potential trade
imbalance of an upcoming time period is additionally based on the
function.
12. The system of claim 7, further comprising a trade order
generation device configured to communicate to a trading venue, via
an electronic network, an electronic trade order to trade a
quantity of a corresponding tradable asset equal to the indication
of a quantity output, in the upcoming time period.
13. A computer readable medium having stored thereon computer
executable instructions for adaptive transaction cost estimation
when executed by performing the following operations: a. receiving
at a trade optimization server, electronic information defining a
proposed order to trade one or more assets on an electronic
exchange or marketplace, wherein the electronic information is
stored, by the trade optimization server, in an electronic data
storage facility and comprises, for each of the one or more assets
in the proposed order to trade, information identifying: i.
security name, symbol, or identifier, ii. transaction side, and
iii. total quantity to be traded; b. receiving at the trade
optimization server, electronic market data comprising historical
market data and current market data relating to the one or more
assets; c. determining, through the use of an optimization
algorithm at the trade optimization server, an optimum trading
strategy and a transaction cost estimate for implementing the
proposed order, based upon the proposed trade order, at least one
pre-selected trading parameter, and the electronic market data; and
d. outputting an indication of a quantity of the one or more assets
to be traded in an upcoming time period according to the determined
optimum trading strategy and the transaction cost estimate, the
quantity to be traded being less than or equal to the total
quantity to be traded.
14. The computer readable medium of claim 13, further including the
executable instructions for performing the operation of: e.
receiving at the trade optimization server, a potential trade
imbalance of an upcoming time period, wherein the potential trade
imbalance is based on at least the electronic information
reflecting a proposed trade order and the received electronic
market data; wherein the optimum trading strategy and the
transaction cost estimate for implementing the proposed order are
each additionally based on the potential trade imbalance of an
upcoming time period.
15. The computer readable medium of claim 14, further including the
executable instructions for performing the operation of: f.
receiving at the trade optimization server a distribution of
realized effective spread and a distribution of realized permanent
price impact of an upcoming time period, wherein the distribution
of realized effective spread and the distribution of realized
permanent price imp act are based on at least the potential trade
imbalance of an upcoming time period and the received electronic
market data; wherein the optimum trading strategy and the
transaction cost estimate for implementing the proposed order are
each at least additionally based on the distribution of realized
effective spread and the distribution of realized permanent price
impact of an upcoming time period.
16. The computer readable medium of claim 13, wherein in the
operation of receiving at a trade optimization server, electronic
information reflecting a proposed order, further comprises:
accessing an order management system database by at least one
computer and retrieving the electronic information reflecting
unplaced orders and creating the electronic information for the
potential order based on the retrieved information reflecting
unplaced orders.
17. The computer readable medium of claim 14, wherein the at least
one pre-selected trading parameter is selected from the following
list: i. time horizon of a potential order; j. level of discretion
of a potential order; k. expected condition of an upcoming bin; and
l. specification of a function for determining an impact on trade
imbalance by the proposed order, wherein the potential trade
imbalance of an upcoming time period is additionally based on the
function.
18. The computer readable medium of claim 13, wherein said
operation of outputting an indication of a quantity of the one or
more assets to be traded in an upcoming time period according to
the determined optimum trading strategy and the transaction cost
estimate, further comprises: communicating to a trading venue via
an electronic network an electronic trade order to trade a quantity
of a corresponding tradable asset equal to the indication of a
quantity output, in the upcoming time period.
19. A system for adaptive transaction cost estimation, comprising:
means for receiving at a trade optimization server, electronic
information defining a proposed order to trade one or more assets
on an electronic exchange or marketplace, wherein the electronic
information is stored, by the trade optimization server, in an
electronic data storage facility and comprises, for each of the one
or more assets in the proposed order to trade, information
identifying: i. security name, symbol, or identifier, ii.
transaction side, and iii. total quantity to be traded; means for
receiving at the trade optimization server, electronic market data
comprising historical market data and current market data relating
to the one or more assets; means for determining, through the use
of an optimization algorithm at the trade optimization server, an
optimum trading strategy and a transaction cost estimate for
implementing the proposed order, based upon the proposed trade
order, at least one pre-selected trading parameter, and the
electronic market data; and means for outputting an indication of a
quantity of the one or more assets to be traded in an upcoming time
period according to the determined optimum trading strategy and the
transaction cost estimate, the quantity to be traded being less
than or equal to the total quantity to be traded.
20. The systems of claim 19, further comprising means for receiving
at the trade optimization server, a potential trade imbalance of an
upcoming time period, wherein the potential trade imbalance is
based on at least the electronic information reflecting a proposed
trade order and the received electronic market data; wherein the
optimum trading strategy and the transaction cost estimate for
implementing the proposed order are each additionally based on the
potential trade imbalance of an upcoming time period.
21. The system of claim 20, further comprising means for receiving
at the trade optimization server a distribution of realized
effective spread and a distribution of realized permanent price
impact of an upcoming time period, wherein the distribution of
realized effective spread and the distribution of realized
permanent price impact are based on at least the potential trade
imbalance of an upcoming time period and the received electronic
market data; wherein the optimum trading strategy and the
transaction cost estimate for implementing the proposed order are
each at least additionally based on the distribution of realized
effective spread and the distribution of realized permanent price
impact of an upcoming time period.
22. The system of claim 19, wherein means for receiving at a trade
optimization server, electronic information reflecting a proposed
order, further comprises: means for accessing an order management
system database by at least one computer and retrieving the
electronic information reflecting unplaced orders and creating the
electronic information for the potential order based on the
retrieved information reflecting unplaced orders.
23. The system of claim 20, wherein the at least one pre-selected
trading parameter is selected from the following list: m. time
horizon of a potential order; n. level of discretion of a potential
order; o. expected condition of an upcoming bin; and p.
specification of a function for determining an impact and trade
imbalance by the proposed order, wherein the potential trade
imbalance of an upcoming time period is additionally based on the
function.
24. The system of claim 19, further comprising means for
communicating to a trading venue via an electronic network an
electronic trade order to trade a quantity of a corresponding
tradable asset equal to the indication of a quantity output, in the
upcoming time period.
Description
BACKGROUND OF THE INVENTION
Field of the Invention
[0001] This invention relates generally to electronic securities
trading. More particularly, the invention relates to systems,
methods, and computer program products for optimizing securities
trading strategies using, inter alia, estimated transaction costs
and market data (both historical and real-time).
Background of the Related Art
[0002] In electronic trading, securities portfolio transactions
typically incur transaction costs, and the minimization of these
costs has been a long-standing aim of securities traders.
Transaction costs may be large, especially when compared to gross
returns, and thus, might substantially reduce or even eliminate the
notional returns of a particular investment. Thus, there is a need
to develop optimal trading strategies that minimize trading costs
and/or some other objective criterion.
[0003] To this end, statistical and mathematical forecasting models
have been developed in an attempt to estimate the transaction costs
of a proposed trade prior to its execution. Such models typically
build upon known empirical facts about trading costs. For example,
empirical studies have established that costs increase with trade
difficulty, a factor systematically related to order size (relative
to average trading volumes), trade direction (BUYS vs. SELLS), firm
size (Market Capitalization), risk (e.g., the volatility of
security returns), liquidity (average daily share volume, spread),
and price level.
[0004] However, existing statistical and mathematical forecasting
models suffer from the inability to perform comprehensive and
accurate analyses of transaction costs because they fail to adapt
to intraday fluctuations in market conditions and rely on the
assumption of market equilibrium, i.e., market neutrality. Also,
many forecasters rely on structured mathematical or econometric
models that require changes to the specification or estimation
techniques to adapt to changes in the statistical properties or
behavior patterns in the market. Further, these models calculate
strategies before an order execution is started and assume that one
follows the strategy independent of any changes in the realized
market conditions.
[0005] Traditional models do not adequately consider the prevailing
market sentiments in assessing the transaction costs of certain
trades. Therefore, there is a need in the field for a forecasting
model that adequately considers real-time data and intraday
fluctuations in market conditions, and is adaptive to user
inputs.
[0006] In particular, there is a need to provide a model that
recommends an optimal trading strategy based on both trader's input
and real-time market conditions. The model should be capable of
updating transaction cost estimates throughout the trade execution
horizon. In order to meet these needs and to overcome deficiencies
in the field, and to provide other non-obvious features and
advantages, the present invention includes systems, methods and
computer program products that forecast the transaction costs of a
proposed trade based on user-selected constraints and real-time
data. The invention can also provide an optimized trading strategy
to satisfy user-defined constraints.
SUMMARY OF THE INVENTION
[0007] The present invention provides systems, methods and computer
program products for adaptive transaction cost estimation. In one
embodiment of the present invention, the systems, methods and/or
computer program products are seamlessly integrated into existing
trading technology architectures, such that outputs may be accessed
by other systems or products, and such that the trading strategies
may be executed automatically or manually through one or more
electronically accessible trade venues.
[0008] One embodiment of the current invention is a computerized
method for adaptive transaction cost estimation. The computerized
method is performed by executing a number of steps. First, at a
trade optimization server, electronic information is received that
defines a proposed order to trade one or more assets on an
electronic exchange or marketplace. The electronic information is
stored, by the trade optimization server, in an electronic data
storage facility. The electronic information, for each of the one
or more assets in the proposed order to trade, identifies: security
name, symbol, or identifier; transaction side; and total quantity
to be traded. Second, at the trade optimization server, electronic
market data is received that includes historical market data and
current market data relating to the one or more assets. Third,
through the use of an optimization algorithm at the trade
optimization server, an optimum trading strategy and a transaction
cost estimate for implementing the proposed order are determined
based upon the proposed trade order, at least one pre-selected
trading parameter, and the electronic market data. Fourth, an
indication of a quantity of the one or more assets to be traded in
an upcoming time period according to the determined optimum trading
strategy and the transaction cost estimate are outputted. The
quantity to be traded is less than or equal to the total quantity
to be traded. It is contemplated that in other embodiments of the
current invention, the individual steps, listed above, may be
performed in various orders or not at all.
[0009] One embodiment of the current invention is a system for
adaptive transaction cost estimation in an electronic trading
system. The system includes a trading optimization server
configured to receive electronic information defining a proposed
order to trade one or more assets on an electronic exchange or
marketplace, to store the electronic information in an electronic
data storage facility, to receive electronic market data
(comprising historical market data and current market data)
relating to the one or more assets, to determine (through the use
of an optimization algorithm) an optimum trading strategy and a
transaction cost estimate for implementing the proposed order
(based upon the proposed trade order, at least one pre-selected
trading parameter, and the electronic market data) and to output an
indication of a quantity of the one or more assets to be traded in
an upcoming time period according to the determined optimum trading
strategy and the transaction cost estimate (the quantity to be
traded being less than or equal to the total quantity to be
traded). The electronic information includes, for each of the one
or more assets in the proposed order to trade, information
identifying: security name, symbol, or identifier; transaction
side; and total quantity to be traded.
[0010] One embodiment of the current invention is a computer
readable medium having stored thereon computer executable
instructions for adaptive transaction cost estimation when executed
by performing the following operations. First, at a trade
optimization server, electronic information is received that
defines a proposed order to trade one or more assets on an
electronic exchange or marketplace. The electronic information is
stored, by the trade optimization server, in an electronic data
storage facility. The electronic information, for each of the one
or more assets in the proposed order to trade, identifies: security
name, symbol, or identifier; transaction side; and total quantity
to be traded. Second, at the trade optimization server, electronic
market data is received that includes historical market data and
current market data relating to the one or more assets. Third,
through the use of an optimization algorithm at the trade
optimization server, an optimum trading strategy and a transaction
cost estimate for implementing the proposed order are determined
based upon the proposed trade order, at least one pre-selected
trading parameter, and the electronic market data. Fourth, an
indication of a quantity of the one or more assets to be traded in
an upcoming time period according to the determined optimum trading
strategy and the transaction cost estimate are outputted. The
quantity to be traded is less than or equal to the total quantity
to be traded. It is contemplated that in other embodiments of the
current invention, the individual operations, listed above, may be
performed in various orders or not at all.
BRIEF DESCRIPTION OF THE DRAWINGS
[0011] FIG. 1 is a block diagram of an exemplary trading system in
which aspects of the present invention may be applied.
[0012] FIG. 2 is a flow diagram of an exemplary method for
performing the steps of the present invention.
[0013] FIGS. 3-5 are screen shots of exemplary user interfaces
according to embodiments of the present invention.
DETAILED DESCRIPTION OF THE INVENTION
[0014] The present invention provides systems, methods and computer
program products for adaptive transaction cost estimation. The
present invention may include features allowing users to adapt the
transaction cost model based upon their own beliefs regarding the
market conditions that would prevail in the near future. Because
market conditions affect trading costs, and therefore impact market
participant trading strategies, effective cost estimates should be
adaptive to prevailing market conditions. The present invention may
be integrated into or be coupled with known trading systems,
trading venues, and other systems and applications used in
effecting and managing securities trading and trade data.
Embodiments of the present invention may be used in connection with
many different types of asset classes, including, but not limited
to: securities, index futures, options, F/X, etc. Embodiments of
the present invention may be used in trading exchanges across the
world, such as, but not limited to: New York Stock Exchange,
Australian Securities Exchange, Hong Kong Stock Exchange, Shanghai
Stock Exchange, and London Stock Exchange.
[0015] Unlike most pre-trade models, the transaction cost model of
the present invention (the "TC model") does not rely on any
assumption of market neutrality or market equilibrium. That is, the
invented TC model does not assume that estimated pre-trade costs
are entirely based on one's own trading strategy and the associated
price impact. The assumption of most pre-trade models is that the
trade imbalance without one's own trading would be zero.
[0016] The present invention eliminates the necessity of the market
neutrality assumption. The function/distribution H, utilized in the
present invention considers the effects of one's trading strategy
by relating a given trade size to the trade imbalance and other
market conditions in the market over a certain time period. The
present invention allows for a market to be out of equilibrium even
in the absence of one's own trading, and also allows for the market
to potentially react to one's own trading, e.g., due to herding
behavior. Additionally, the present invention allows a user to
input one's own beliefs into the model, for example, through the
function/distribution H. These inputs change the empirical
distributions underlying the model and in turn change the
calculated optimal solution (strategy).
[0017] According to a preferred embodiment, the present invention
includes special purpose electronic facilities for modeling two
cost components: (1) costs within a time period or bin, (effective
spread, often referred to as temporary price impact), and (2) costs
across bins (law of motion, often referred to as permanent price
impact). In contrast to most transaction cost models, the invented
TC model does not rely exclusively on modeling the first and second
moment of the cost components, but rather, models the entire
distribution of F (the law of price motion) and G (the effective
spread within a bin) using a non- or semi-parametric approach.
[0018] Further, the invented TC model does not model the effect of
one's own trading on price and costs directly. The direct price
impact of one's own trade size on prices and costs is actually
removed. Instead, there is an indirect impact on prices/costs
through the effect of one's own trading on the market trade
imbalance and/or other market attributes. The function/distribution
H of the trade imbalances and other market attributes is estimated
directly from execution data, while F and G can be estimated using
publicly available Level 1 or Level 2 data. The
function/distribution H can include not only trade imbalances but
also volume, volatility, spread, depth, to capture the effect of a
client's trading even better and, most importantly, how the market
participant affects various aspects of the stock dynamics.
[0019] When provided with an order size and the existing market
sentiment, the present invention is configured to calculate a cost
distribution of possible outcomes including the mean, E, and the
expected average execution price, avg, over the specified time
frame. The invented TC model can also suggest optimal bounds for
the number of shares to be traded in the specified time interval.
Thus, the invented TC model can provide not only the ex-ante
optimal strategy but also the simulated ex-post strategies for
different outcomes. This feature of the model allows a trader to
understand and assess the acceptable magnitude of deviation without
facing the risk of executing a trade and incurring an unacceptable
adverse affect due to opportunity costs from normal market
conditions.
[0020] The present invention can be utilized to trade more or less
aggressively, depending on whether the stock is moving up or down.
This can be particularly useful in volatile markets where rapid
price movements can result in significantly higher costs and
traders must complete orders at the best price possible.
[0021] Regarding modeling constraints, the present invention can be
configured to allow for modeling tail behavior (via the expected
shortfall) and accommodate the possibility that the order does not
have to be fully completed, by specifying a threshold for the
expected costs below which order execution is possible, and
estimating the expected amount that can be filled.
[0022] The present invention includes a dynamic and adaptive model.
Real-time data information can be utilized as soon as it becomes
available and, as described in further detail below, optimizations
can be run repeatedly over specified time intervals, thereby
adapting an optimal trading strategy to the most up-to-date
information available. It is not assumed that one would follow a
pre-determined trading strategy identified prior to initiating the
order execution without considering changes in the market
conditions. Any strategy that may be optimal and appropriate at the
moment when the order is initiated will no longer be optimal even a
short time later as market conditions fluctuate and one observes
the realized executions, which can be more or less than what the
strategy called for, resulting in the need for modification of the
execution strategy.
[0023] The present invention can explicitly account for signals
about the prevailing market conditions during the execution of an
order. Signals may be constructed simultaneously for many variables
including volatility, volume, spread, and depth of the market. Some
of these real-time variables can be specialized in the form of
"smart indicators." Smart indicators are discrete variables that
provide indication of abnormal market conditions.
Co-owned/co-pending U.S. patent application Ser. No. 11/476,895,
titled System and Method for Generating Real-Time Smart Indicators
to Indicate Abnormal Conditions in a Trading List or Portfolio, and
filed on Jun. 29, 2006, includes a detailed discussion of smart
indicators, the entire contents of which are herein incorporated by
reference. The present invention can combine the smart indicator
values with the set of definitions describing how this information
can be embedded into the optimization problem. For instance, the TC
model suggests that abnormally high volatility surprises will often
lead to higher than normal costs in the following short time frame,
suggesting that trading should be slowed down.
[0024] Unlike models utilized in the prior art, which define an
objective function as a trade-off between expected costs and the
standard deviation of costs, the present invention allows for
flexibility in the formulation of the objective function.
[0025] As can be readily understood by a person of ordinary skill
in the art, in alternative embodiments of the present invention,
proposed trade executions can be automatically transferred within
the network from one server operating according to a first trade
strategy algorithm to another server having a second different
trade strategy algorithm.
[0026] FIG. 1 is a block diagram of a simplified trading system in
which embodiments of the present invention may be applied.
[0027] Referring to FIG. 1, system 100 may include a plurality of
specially programmed computers at various locations, which may be
coupled with an electronic data network 120. For example, computer
client 102 can be a networked computer configured to communicate
desired trades or trade orders. In one embodiment of the present
invention, trading desk 104 might be associated, for example, with
a buy-side trading desk at a buy-side trading institution. The
trading desk may include one or more computers coupled with
electronic data network 120 and configured to execute various trade
desk programs, such as, order management systems (OMS), execution
management systems (EMS), and to generate trade analytics. Trade
systems may be used to manage long-term and short-term trading
goals, and to connect electronically to electronic trade venues,
such as the New York Stock Exchange (NYSE) 108, ITG's POSIT.RTM.
110, over-the-counter (OTC) venues 112 (including, but not limited
to, NASDAQ), or electronic communications networks (ECNs) 114. The
aforementioned list of electronic trade venues is not all
encompassing, and in no way limits the range of trade venues that
the current invention might integrate or communicate with. For
example, the current invention contemplates the use of
indications-of-interest, and thus might integrate or communicate
with an indication-of-interest venue, such as ITG's POSIT
ALERT.
[0028] The electronic data network 120 may include public or
private networks, electronic data networks, packeted networks,
wireless or other communication services, etc. In one embodiment of
the present invention, the electronic data network 120 might
include the Internet and/or a local area network. Server 118 may be
coupled with the electronic data network 120 and may be configured
to perform adaptive transaction cost estimation, as described below
in further detail. The server 118 may have access to various
electronic trading venues (108, 110, 112, 114) through the
electronic data network 120. Server 118 may also be in
communication with a database 106. The database 106 may be housed
on server 118 or on a separate computer, i.e., a database server.
The database 106 may be a relational database, such as an ORACLE
database.
[0029] System 100 can include an information provider 116, which
may feed market data to the server 118 though the electronic data
network 120. Both historical and real-time trade data may be
provided to the server 118. Additionally, trade data fed from the
information provider 116 may also be stored in the database 106,
and thus be readily accessible via server 118.
[0030] Even though FIG. 1 shows only one server 118, it is
contemplated that multiple servers may be provided on the
electronic network 120, with each server running a variety of
applications, which may include transaction cost analysis programs.
Likewise, these additional servers may have access to various
trading venues (108, 110, 112, 114).
[0031] A user may electronically submit a proposed portfolio trade
order, through computer client 102, for analysis, optimization,
and/or reconstruction to server 118. This submission may occur
automatically, that is, without explicit action by the user at the
time of the order submission. One way this submission might be
implemented is through the use of defaults. For example, when a
user sends a trade to OTC 112, via their OMS, it is contemplated
that the order might first be transmitted electronically by
default, to server 118 as a proposed order with the goal of
identifying or creating the best strategy for executing the order.
Once the optimal trading strategy has been identified or created,
it can be displayed or otherwise communicated to the user of the
system. Further, the user may configure the system not to display
the results of the optimization. Furthermore, it is contemplated
that once the optimal strategy has been identified, the system may,
if designated to do so, automatically carry out the execution of
the order by creating and submitting electronic orders according to
the created/identified optimal strategy.
[0032] FIG. 2 is a flow diagram detailing steps of an embodiment of
the present invention. Other embodiments of the present invention
may utilize all or some of the steps listed in FIG. 2, in the same
or a different order. FIG. 2 in no way limits the scope of the
present invention to the steps shown therein and described
below.
[0033] The steps shown in FIG. 2 may be executed iteratively during
a trading period. The trading period can be broken up into smaller
sub-periods called bins. In one embodiment, the present invention
contemplates that the trading period may be a trading day that is
broken into 13 thirty-minute bins. Thus, the present invention
contemplates that the steps shown in FIG. 2 may be run prior to
each bin. However, because the trading circumstances may deviate
during the trading day from what the user expects, the steps shown
in FIG. 2 may run on demand (i.e., at any time or continuously) or
be automatically triggered by some event, such as a larger than
normal trade.
[0034] At step 202, the user of the system (i.e., a portfolio
manager, trader, or other market participant) may input variables
regarding the proposed trades that the user intends to execute. For
example, a user could input variables relating to a desire to BUY
1,000,000 shares of IBM with a LIMIT price of $85.50.
Alternatively, an order may be defined by the following parameter
list: ticker, date, order size, starting bin (s), ending bin (e).
These inputs may be entered all at once or multiple times, and may
be changed or updated at any time prior to, during, or after the
trading period, as shown in FIG. 2 by the line connecting step 216
to steps 202 and 204.
[0035] Additionally, the user may input other parameters that
define factors or preferences that affect the way that the trading
will occur. For example, the user may choose: various levels of
acceptable risk, time horizons for trades to be completed, urgency
levels, predicted market conditions, and/or level of discretion
(i.e., what types of orders, market or limit, the user is trading
with, and what percentage of each type of order the user expects to
execute). While step 202 is described above in terms of a user
"inputting" information, it is contemplated that current trading
technology may allow for integrations between various systems to
allow for information in one system, such as a user's OMS, to be
transmitted into another system, such as the system of the current
invention, with or without manual intervention by the user. For
example, it is contemplated that a user may set defaults directed
to which orders should be accessed and submitted to the system of
the present invention. Likewise, a user may set defaults for the
input variables that affect the way in which the trading will
occur, as discussed above. For example, if a particular risk averse
user always prefers the lowest level of risk, the current invention
contemplates that for this user the lowest level of risk may be set
explicitly as a default value.
[0036] At step 204, market data may be input into the system of the
present invention. This market data may include both historical
market/trading data and real-time market data. Historical
market/trading data can include all market data from trading that
has occurred prior to the current trading period, e.g., the current
trading day. By utilizing real-time market optimizing data,
functions (e.g., H, G, and F, as described in greater detail below)
may be changed dynamically. Real-time market data can include data
obtained in real time during the current trading period, e.g., the
current trading day or the recent history. While the system of the
present invention may have access to a vast bank of market data,
not all of the accessible market data must be utilized. For
example, the system may utilize information pertaining to the
security that the user intends to trade and/or other securities
that have similar characteristics and/or trading patterns to the
security that the user intends to trade. It is contemplated that
associations between securities may be generated in real-time or be
pre-formed, and based upon characteristics and/or trading patterns
of the securities. These relationships may be stored in one or more
electronic data storing facilities connected to the electronic data
network of the present invention. Additionally, it is contemplated
that historical data may be weighted depending on the age/relevance
of the trade data. Embodiments of the present invention may
utilize, but are not limited to, market data including: the
real-time spread, volume, and volatility, as well as the historical
spread, volume, and volatility, for the particular security under
consideration.
[0037] At step 206, the system computes the relationship between
the remaining order size and the market imbalance. This
relationship may be computed utilizing the H Function. The H
Function describes the stochastic relationship between the user's
remaining order size and its impact on the market trade imbalance
(the difference between buyer and seller initiated volume) for a
certain sub-period (bin) of the trading period. The remaining order
size should be used because this variable will continually decrease
due to the iterative nature of the present invention.
[0038] In one embodiment of the present invention, the H Function
is empirically estimated from peer group data. This data may
include information on trade executions, and may be stored in a
database that may be integrated with or connected to the system of
the present invention, such as ITG's Peer Group Database, as
described in the co-owned/co-pending U.S. patent application Ser.
No. 10/674,432 titled System and Method for Estimating Transaction
Costs Related to Trading a Security, and filed on Oct. 1, 2003, the
entire content of which is herein incorporated by reference into
the present application. In order to minimize the influence of data
that could improperly skew the results of the H Function, i.e.,
noise, the securities in the peer group may be categorized into
liquidity groups. For example, the universe of all securities may
be split into four liquidity groups, where the liquidity groups can
be defined via the average daily local currency volume of the
universe of all securities.
[0039] For each order, the relative order size is determined and
compared with the associated market trade imbalance. For example,
when analyzing a one-day order, the relative order size (S/ADV) is
determined by dividing the remaining order size with the previous
day's 21 day median daily dollar volume. This value is compared
with the associated trade imbalance (TI/ADV) determined by finding
the difference between the net BUYS and SELLS volume divided by the
previous day's 21 day median daily dollar volume. Next, the pairs
(S.sub.i/ADV.sub.i, TI.sub.i/ADV.sub.i).sub.i may be segmented into
(1) the corresponding liquidity group, as described above, and then
(2) the relative remaining order size.
[0040] In one embodiment of the present invention, the relative
remaining order size may be determined for each liquidity group
separately. For each liquidity group, the securities universe may
be divided into 100 order size buckets. The buckets may be
constructed so that there are enough observations per bucket. For
each bucket, all percentiles (1 though 99) of the conditioned
empirical distribution of (TI/ADV) given the value of x=(S/ADV) may
be determined. Additionally, these percentiles may be used to
approximate the conditional mean E(TI/ADV|x) and conditional
standard deviation SDV(TI/ADV|x) of each bucket.
[0041] Next, the parametric forms of functions e(x)=E(TI/ADV|x) and
sdv(x)=SDV(TI/ADV|x) may be estimated. These parameterizations may
be approximated using the following functions:
e ( x ) = E ( TI / ADV | x ) = { .beta. e x , x .ltoreq. x e .beta.
e x e ( 1 + ( x - x e ) .alpha. e x e ) .alpha. e , else sdx ( x )
= SDV ( TI / ADV | x ) = { y sdv , x .ltoreq. x sdv , 1 y sdv +
.beta. sdv ( x - x sdv , 1 ) .alpha. sdv , 1 , x sdv , 1 < x
.ltoreq. x sdv , 2 c sdv ( 1 + .beta. sdv .alpha. sdv , 1 ( x sdv ,
2 - x sdv , 1 ) .alpha. sdv , 1 - 1 ( x - x sdv , 2 ) .alpha. sdv ,
2 c sdv ) .alpha. sdv , 2 , else ##EQU00001##
where
[0042]
C.sub.sdv=SDV(TI/ADV|x.sub.sdv,2)=y.sub.sdv+.beta..sub.sdv(x.sub.sd-
v,2-x.sub.sdv,1).sup..alpha..sup.sdv,1;
[0043] .beta..sub.e.di-elect cons.(0,1), .alpha..sub.e.di-elect
cons.[0.1,1] and x.sub.e.di-elect cons.{avgSize.sub.1, . . . ,
avgSize.sub.100} are the parameters that need to be estimated for
the function e(x) (the average sizes are the mean relative order
sizes in each bucket and thus ordered), and .beta..sub.sdv.di-elect
cons.(0,1), .alpha..sub.sdv,1.di-elect cons.[0.1,1],
.alpha..sub.sdv,2.di-elect cons.[0.1,1], x.sub.sdv,1,
x.sub.sdv,2.di-elect cons.{avgSize.sub.1, . . . , avgSize.sub.100}
and y.sub.sdv are the parameters that need to be estimated for the
function sdv(x) (y.sub.sdv is the average trade imbalance value
associated to x.sub.sdv,1). All parameters can be estimated using
standard estimation techniques, e.g., grid searches.
[0044] In one embodiment of the present invention, once the
computation of the H Function, i.e. the stochastic relationship
between the user's remaining order size and the market trade
imbalance, is completed, a potential trade imbalance is fed into
step 210. This allows the G and F Functions to reflect the
relationship determined in step 206, and thus alleviate the
assumption of market neutrality and equilibrium in the G and F
Functions. Additionally, the H Function may be used to determine
the potential volume, potential volatility, potential spread,
potential depth, and other qualities characterizing the market
conditions relevant for execution of the potential order.
[0045] At step 210, the system, computes the relationship between
the market trade imbalance and the Law of Motion of Prices. Whereas
step 208 is related to the temporary price impact, step 210 is
related to the permanent price impact. The F Function determines
the relationship between the market trade imbalance and the Law of
Motion of Prices, i.e., the permanent price impact. The F Function
depends on the market trade imbalance, and (through the data output
by the H Function) on the remaining order size. Additionally, the F
Function may utilize historical stock-specific characteristics and
real-time market conditions.
[0046] The F Function can be is defined as:
F i , j a = P t , j a - P t , j - 1 a P i , j - 1 a 1 .sigma. _ i ,
j a , ##EQU00002##
where {tilde over (P)}.sub.i,j.sup.a is the volume-weighted average
execution price in bin j of day t of security a and
.sigma..sub.i,j.sup.a is the historical volatility of stock a in
bin j. The F Function, of the current embodiment, may utilize the
following historical stock-specific characteristics: average
volume, average volatility, average spread, and the previous day's
closing price. The F Function also may take into consideration and
compensate for real-time market conditions, such as: surprises in a
security's volume, volatility, and/or spread. The output of this
step is the distribution (per bin or per trading period) of the
permanent price impact.
[0047] At step 208, the system computes the relationship between
the market trade imbalance, and, in some embodiments, computes
other market conditions and the effective spread. One purpose of
this step is to determine the temporary price impact of trading the
remaining order size. The G Function can be utilized to determine
the relationship between the market trade imbalance and the
effective spread. The G Function depends on the market trade
imbalance, and (through the data output by the H Function) on the
remaining order size. Additionally, the G Function may utilize
historical stock-specific characteristics and real-time market
conditions.
[0048] The G Function can be defined as:
G i , j a = P ~ i , j a - P i , j - 1 a P i , j - 1 a 1 .sigma. _ i
, j a - 1 2 F i , j d , ##EQU00003##
where {tilde over (P)}.sub.i,j.sup.a is the volume-weighted average
execution price in bin j of day i of security a and {tilde over
(.sigma.)}.sub.i,j.sup.a is the historical volatility of stock a in
bin j. The G Function may utilize the following historical
stock-specific characteristics: average volume, average volatility,
average spread, and/or the previous day's closing price. The G
Function may also take into consideration and compensate for
real-time market conditions, such as: unexpected security volume
(i.e. lower than average, higher than average, etc. . . . ),
volatility, and/or spread. The output of this step is the
distribution (per bin or per trading period) of the realized
effective spread. However, when compared to the G Function, the F
Function may rely less on the real-time spread. For both the G and
F Functions, the real-time volatility and volume are typically the
most important of the real-time market conditions.
[0049] In one embodiment of the present invention, once steps 208
and 210 have been completed, i.e., the joint distribution of the F
and G functions is obtained, the outputs are fed into step 212,
where an optimal trading strategy may be computed prior to each
bin. The optimal trading strategy may or may not be a projection
that extends past the upcoming bin. That is, because the remaining
order size and trading environment may change during the trading
period, it is not necessarily efficient to output trading
strategies that will need to be subsequently revised based on the
activity during the upcoming bins. However, situations may arise
that call for projected trading strategies extending past the
upcoming bin.
[0050] In one embodiment of the present invention, the objective
function used in the optimization of step 212, may be configured to
solve for the optimal trading strategy in the most efficient
fashion (for example, in the smallest possible number of
iterations). The objective function should also ensure stability,
such that optimal trading strategy does not change drastically when
the inputs are changed slightly.
[0051] In the discussion of the optimization functions we use the
notation: e=T, i.e., the close of the terminal bin, and remaining
order size, N.sub.s and s=0, 1, 2, , , , , T. Let
(x.sub.j).sub.j=s, . . . ,T be a given specified strategy from
starting bin (s) to ending bin (e), (or T if the ending bin is the
close of trading period). MC.sub.s is the set of parameters
characterizing the actual market conditions observed at time s
(which is the start time of bin (s)), and SC.sub.s is the set of
parameters containing the stock-specific characteristics in bin (s)
(e.g., liquidity group, spread, ADV, volatility). The market
conditions at the starting bin (MC.sub.s) are given by the
previously observed volume, volatility, spread, and previous day's
volatility. (MC.sub.s) are dynamically updated while (SC.sub.s) are
pre-determined. In one embodiment of the current invention, the
market conditions may be summarized using the following scale:
TABLE-US-00001 -1: unconditioned 0: low volatility, low volume 1:
low volatility, medium volume 2: low volatility, high volume 3:
medium volatility, low volume 4: medium volatility, medium volume
5: medium volatility, high volume 6: high volatility, low volume 7:
high volatility, medium volume 8: high volatility, high volume
[0052] In one embodiment of the present invention, the system can
compute the mean and the projected percentiles of the associated
distribution of costs, by combining the inputs from step 202 (e.g.,
starting bin (s), ending bin (e) (or T if e is the closing bin of
the trading period, i.e., e=T=12 for U.S. trading if the bins have
30-minute interval lengths), order size N.sub.s, ticker or
stock-specific historical characteristics SC.sub.s, and the
prevailing market conditions MC.sub.s) at the start of each bin (s)
with the trading strategy (x.sub.j).sub.j=s . . . T, where the cost
is defined as
Cost s = ( t = s T x t P ~ t + 1 ) - N s P s N s P s .
##EQU00004##
Thus, the system can compute the mean and the percentiles using the
following equations:
E(Cost.sub.s|(x.sub.j).sub.j=s, . . .
,T,N.sub.s,SC.sub.s,MC.sub.s)
and (Pct.sub.s,d(Cost.sub.s|(x.sub.j).sub.j=s, . . .
,T,N.sub.s,SC.sub.s,MC.sub.s)).sub.d=1, . . . ,99.
[0053] The current invention contemplates a number of alternative
objective functions that may be used in step 212. However, none of
the functions described below are intended to limit the scope of
the current invention in any way.
[0054] A basic optimization function may be utilized in finding the
optimal trading strategy at step 212. One basic optimization
function contemplated in the current inventions is:
( x t * ) arg min ( x j ) j = s , , T { E ( Cost s | ( x j ) j = s
, , T , N s , SC s , MC s ) } . ##EQU00005##
While this solution provides a suitable answer, more precise
solutions may be derived using more complex optimization functions.
The following five optimization functions are more complex than the
basic optimization function, and yield a richer framework of
analysis. Each of the functions listed below have been contemplated
as suitable for performing step 212.
[0055] A two step optimization procedure may be used in step 212.
The following is one embodiment of the two step optimization:
Step 1:
[0056] ( x j * ) arg min ( x j ) j = s , , T { E ( Cost s | ( x j )
j = s , , T , N s , SC s , MC s ) } , C s * = E ( Cost s | ( x j *
) j = s , , T , N s , SC s , MC s ) ##EQU00006##
Step 2: Find solutions (x.sub.j.sup.(low)).sub.j=s, . . . ,T and
(x.sub.j.sup.(up)).sub.j=s, . . . ,T such that:
E ( Cost s | ( x j * ) j = s , , T , N s , SC s , MC s ) + .DELTA.
= E ( Cost s | ( x j ( low ) ) j = s , , T , N s , SC s , MC s ) =
E ( Cost s | ( x j ( up ) ) j = s , , T , N s , SC s , MC s )
##EQU00007##
and x.sub.s.sup.(low)<x.sub.s*<x.sub.s.sup.(up) for some
.DELTA.>0 that is supplied to the optimization problem as well.
It is likely that users of the system of the current invention will
have the best intuition when viewing the results of the
optimization program presented in total dollars, and thus .DELTA.
would probably vary with the order size (more precisely,
.DELTA.(N.sub.s).dwnarw. as N.sub.s.uparw..
[0057] Alternatively, risk averse objective functions may be
utilized in the optimization program at step 212. One contemplated
objective function may be as follows:
( x j * ) j = s , , T arg min ( x j ) j = s , , T { E ( u ( Cost s
) | ( x j ) j = s , , T , N s , SC s , MC s ) } , ##EQU00008##
where u(z) is supposed to be a convex function. The following two
functions are examples:
u ( z ) = 1 .gamma. ( e .gamma. z - 1 ) , a ) u ( z ) = { z , if z
< .alpha. z + .delta. z , else b ) ##EQU00009##
[0058] Alternatively, an optimization function may include
value-at-risk (VaR) constraints. These constraints allow a user to
strike a balance between minimizing cost and risk. One contemplated
approach in quantifying risk is to specify a maximum or critical
level of value-at-risk (VaR.sub.s.sup.crit) that should not be
exceeded with more than probability .alpha.. Both
VaR.sub.s.sup.crit and a may be provided by the user; however,
VaR.sub.s.sup.crit and a can be set as defaults. For example, in
one embodiment a will be set to a default value of 0.05 if not user
specified. While VaR.sub.s.sup.crit may be represented in various
forms, it is likely most useful when it is measured in dollars and
translated to basis points by the system. The following is an
example of the optimization function that includes VaR
constraints:
min ( x j ) j = s , , T { E ( Cost s | ( x j ) j = s , , T , N s ,
SC s , MC s ) } ##EQU00010##
given that P(Cost.sub.s>VaR.sub.s.sup.crit).ltoreq..alpha..
[0059] In order to evaluate the probability,
P(Cost.sub.s>VaR.sub.s.sup.crit).ltoreq..alpha., it is necessary
to use (Pct.sub.s,d(Cost.sub.s|(x.sub.j).sub.j=s, . . .
,T,N.sub.s,SC.sub.s,MC.sub.s)).sub.d=1, . . . ,99. Unfortunately,
for some values of VaR.sub.s.sup.crit and .alpha., the optimization
problem listed above has no solution. When a non-solution event
occurs, the system at step 212 may follow any of the following
courses of action: (1) do nothing and provide an error message to
the user, (2) allow the user to change (increase) the VaR parameter
VaR.sub.s.sup.crit, or (3) inform the user what the maximal number
of shares N.sub.s.sup.filled=N.sub.s.sup.filled(VaR.sub.s.sup.crit)
(order size) that can be filled, and for which an optimal strategy
can be provided would be. Obviously, when the third option is
chosen, N.sub.s.sup.filled.ltoreq.N.sub.s.
[0060] Alternatively, the optimization problem may be specified to
have a Markov decision process structure. Rather than provide one
and only one optimal number of shares to be traded, it may be
beneficial to provide an interval in which the "actual" optimal
number of shares to be traded in bin (s) can be found (with a very
large probability).
[0061] In utilizing the optimization function may have a Markov
decision process structure, let F.sub.j(
|TI.sub.j,SC.sub.s,MC.sub.s) and G.sub.j(
|TI.sub.j,SC.sub.s,MC.sub.s) be the F and G functions, as described
above, for j=s+i, T. The following describes a way to provide a
confidence interval of optimal number of shares to trade in bin (s)
assuming that the future outcome of market conditions in bin (s) is
known.
[0062] In order to simplify the problem and speed up performance,
the system may only sample the volume and volatility surprises
vol.sub.s.sup.surprise and vola.sub.s.sup.surprise, while the
remaining market conditions are held fixed at their prevailing
values. Therefore,
MC.sub.s=(otherMC.sub.s,vol.sub.s.sup.surprise,vola.sub.s.sup.surprise).
[0063] When using this optimization technique, the user may be
allowed to select a confidence interval for volume and volatility
surprises in bin (s) (or alternatively use plus or minus one
standard deviation based on the ITG proprietary smart indicator
distributions). Thus, vol.sub.s+1.sup.surprise.di-elect
cons.[vol.sub.s+1.sup.suprise,low, vol.sub.s+1.sup.surprise,up] and
vola.sub.s+1.sup.surprise.di-elect
cons.[vol.sub.s+1.sup.asurprise,low,vola.sub.s+1.sup.surprise,up],
where
vol.sub.s+1.sup.surprise,low=vol.sub.s.sup.surprise-.DELTA..sub.s,low,
vol.sub.s+1.sup.surprise,up=vol.sub.s.sup.surprise+.DELTA..sub.s,up,
vola.sub.s+1.sup.surprise,low=vola.sub.s.sup.surprise-.PSI..sub.s,low,
and
vola.sub.s+1.sup.surprise,up=vola.sub.s.sup.surprise+.PSI..sub.s,up.
[0064] Next, the following five optimization programs are solved:
[0065] use the regular values vol.sub.s.sup.surprise and
vola.sub.s.sup.surprise, to determine (x.sub.j*).sub.j=s, . . .
,T=arg.sub.x min E(Cost.sub.s|(x.sub.j*).sub.j=s, . . . ,T,
N.sub.s, SC.sub.s, otherMC.sub.s, vol.sub.s.sup.surprise,
vola.sub.s.sup.surprise) [0066] use vol.sub.s+1.sup.surprise,low
and vola.sub.s+1.sup.surprise,low and use F.sub.j(
|TI.sub.j,SC.sub.s,MC.sub.s+1), and G.sub.j(
|TI.sub.j,SC.sub.s,MC.sub.s+1) instead of F.sub.j(
|TI.sub.j,SC.sub.s,MC.sub.s) and G.sub.j(
|TI.sub.j,SC.sub.s,MC.sub.s) for j=s+1, s+2, . . . , T, [0067]
(x.sub.j.sup.*,low,low).sub.j=s, . . . ,T, [0068]
C.sub.*,low,low=E(Cost.sub.s|(x.sub.j.sup.*,low,low).sub.j=s, . . .
,T,N.sub.s,SC.sub.s,otherMC.sub.s,vol.sub.s+1.sup.surprise,low,vola-
.sub.s+1.sup.surprise,low) [0069] use vol.sub.s+1.sup.surprise,low
and vola.sub.s+1.sup.surprise,up and use F.sub.j(
|TI.sub.j,SC.sub.s,MC.sub.s+1) and G.sub.j(
|TI.sub.j,SC.sub.s,MC.sub.s+1) instead of F.sub.j(
|TI.sub.j,SC.sub.s,MC.sub.s) and G.sub.j(
|TI.sub.j,SC.sub.s,MC.sub.s) for j=s+1, s+2, . . . , T, [0070]
(x.sub.j.sup.*,low,up).sub.j=s, . . . ,T, [0071]
C.sup.*,low,up=E(Cost.sub.s|x.sub.j.sup.*,low,up).sub.j=s, . . .
,T,N.sub.s,SC.sub.s,otherMC.sub.s,vol.sub.s+1.sup.surprise,low,vola.s-
ub.s+1.sup.surprise,up) [0072] use vol.sub.s+1.sup.surprise,up and
vola.sub.s+1.sup.surprise,low and use F.sub.j(
|TI.sub.j,SC.sub.s,MC.sub.s+1) and G.sub.j(
|TI.sub.j,SC.sub.s,MC.sub.s+1) instead of F.sub.j(
|TI.sub.j,SC.sub.s,MC.sub.s) and G.sub.j(
|TI.sub.j,SC.sub.s,MC.sub.s) for j=s+1, s+2, . . . , T, [0073]
(x.sub.j.sup.*,up,low).sub.j=s, . . . , T, [0074]
C.sup.*,up,low=E(Cost.sub.s|x.sub.j.sup.*,up,low).sub.j=s, . . .
,T,N.sub.s,SC.sub.s,otherMC.sub.s,vol.sub.s+1.sup.surprise,up,vola.su-
b.s+1.sup.surprise,up) [0075] use vol.sub.s+1.sup.surprise,up and
vola.sub.s+1.sup.surprise,up and use F.sub.j(
|TI.sub.j,SC.sub.s,MC.sub.s+1) and G.sub.j(
|TI.sub.j,SC.sub.s,MC.sub.s+1) instead of F.sub.j(
|TI.sub.j,SC.sub.s,MC.sub.s) and G.sub.j(
|TI.sub.j,SC.sub.s,MC.sub.s) for j=s+1, s+2, . . . , T, [0076]
(x.sub.j.sup.*,up,up).sub.j=s, . . . , T, [0077]
C.sup.*,up,up=E(Cost.sub.s|x.sub.j.sup.*,up,up).sub.j=s, . . .
,T,N.sub.s,SC.sub.s,otherMC.sub.s,vol.sub.s+1.sup.surprise,up,vola.sub.s+-
1.sup.surprise,up)
[0078] Alternatively, step 212 may utilize an optimization
technique that is conditioned on the market trade imbalances in the
start bin (s). Again, rather than provide one and only one optimal
number of shares to be traded, it may be beneficial to provide an
interval in which the "actual" optimal number of shares to be
traded in bin (s) can be found (with a very large probability).
However, instead of varying volume and volatility surprises, as
discussed above with regard to optimization functions that include
Markov decision process structures, it is possible to optimize by
varying the market trade imbalances.
[0079] The functions are as follows:
E ( Cost s | ( x j ) t = s , , T , N s , SC s , MC s ) = E ( E (
Cost s ( x t ) t = s , , T , N s , SC s , MC s , TI s + 1 ) ( x j )
j = s , , T , N s , SC s , MC s ) = E ( E ( Cost s ( x t ) t = s ,
, T , SC s , MC s , TI s + 1 ) ( x j ) j = s , , T , N s , SC s ,
MC s ) ##EQU00011##
Thus, it is possible to compare how the optimal strategy changes
based on changes in TI.sub.s+1. More precisely, following five
optimizations: [0080] run regular optimization, i.e. [0081]
(x.sub.j*).sub.j=s, . . . ,T=arg.sub.x
min{E(Cost.sub.s|(x.sub.j).sub.j=s, . . .
,T,N.sub.s,SC.sub.s,MC.sub.s)}(x.sub.j*).sub.j=s, . . . ,T [0082]
(x.sub.j.sup.10).sub.j=s, . . . ,T=arg.sub.x
min{E(Cost.sub.s|(x.sub.j).sub.j=s, . . .
,T,N.sub.s,SC.sub.s,MC.sub.s,TI.sub.s+1=PcH.sub.10)}(x.sub.j*.sup.10).sub-
.j=s, . . . ,T [0083] (x.sub.j.sup.25).sub.j=s, . . . ,T=arg.sub.x
min{E(Cost.sub.s|(x.sub.j).sub.j=s, . . .
,T,N.sub.s,SC.sub.s,MC.sub.s,TI.sub.s+1=PcH.sub.25)}(x.sub.j*.sup.25).sub-
.j=s, . . . ,T [0084] (x.sub.j.sup.75).sub.j=s, . . . ,T=arg.sub.x
min{E(Cost.sub.s|(x.sub.j).sub.j=s, . . .
,T,N.sub.s,SC.sub.s,MC.sub.s,TI.sub.s+1=PcH.sub.75)}(x.sub.j*.sup.75).sub-
.j=s, . . . ,T [0085] (x.sub.j.sup.90).sub.j=s, . . . ,T=arg.sub.x
min{E(Cost.sub.s|(x.sub.j).sub.j=s, . . .
,T,N.sub.s,SC.sub.s,MC.sub.s,TI.sub.s+1=PcH.sub.90)}(x.sub.j*.sup.90).sub-
.j=s, . . . ,T
[0086] Regardless of the optimization technique used, at step 214,
the optimal trading strategy for the upcoming bin (or the multiple
upcoming bins, depending on the configuration of the system) may
be, due in part to the possibility of non-solution events, an
output to the user (or alternatively to another system for manual
or automatic execution of the optimal trading strategy). Once the
optimal trading strategy has been displayed to the user, the user
may execute the strategy utilizing known trading systems and
techniques. The output of the system may be a clear description of
how many shares should be traded in the next bin to best achieve
the overall trade objective. For example, one output could be an
instruction: "BUY 100,000 IBM during the next bin." Additionally,
confidence levels can be provided for the trade strategy output.
Additionally, the system may output the transaction cost for
executing the trade, as being determined previously at step 212.
Further, an additional strategy output may be provided by the
system, such as at what minimum or maximum prices the trades should
be executed.
[0087] At step 216, the system determines if all of the shares of
the order have been traded. If there are no remaining shares to be
traded, the algorithm terminates at step 218. However, if at step
216 it is determined that there are remaining shares to be traded,
the system continues to steps 202 and 204 and completes another
iteration for the upcoming bin. The inputs at step 202 may be
adjusted at any time, including the time prior to, during, or after
a bin.
[0088] Provided the order size and the existing market sentiment
have been specified, the model of the current invention may
calculate certain characteristics of the cost distribution of
possible outcomes, including the mean, E, and the expected average
execution price, avgP, over the specified time frame. This model
may also suggest optimal bounds for the number of shares to be
traded in the specified time frame, as discussed above in step 214.
Thus, this model can provide not only an ex-ante optimal strategy
but also the simulated ex-post strategies for different outcomes.
This would allow a trader to understand how much deviation from the
proposed optimal strategy can be tolerated without risking an
unacceptable adverse affect due to opportunity costs.
[0089] The present invention may be used to trade more or less
aggressively, depending on whether the stock is moving up or down.
This can be particularly useful in volatile markets, during which
rapid price movements can result in significantly higher costs, and
traders must complete their orders at the best price possible.
[0090] Some embodiments of the present invention may account for
signals about market conditions during the execution of an order.
Signals may be constructed for many variables including volatility,
volume, and spread. Some of these real-time variables are "smart
indicators." The smart indicators are analytics that provide
indication of abnormal market conditions. The present invention can
combine smart indicator-based analytics with a set of definitions
describing how to use the information, such that the information
may be used during the optimization, as described above in step
212.
[0091] FIGS. 3-5 are screen shots of exemplary graphical user
interfaces (GUIs) of embodiments of the present invention. GUIs of
the invention can be configured to control and perform aspects of
the invention as described above. The results of the above
described embodiments of the present invention may be communicated
or displayed to users in graphical formats, e.g., charts, graphs,
etc. FIGS. 3-5 are the property of ITG Inc., and are protected
under copyright laws of the United States.
[0092] As illustrated in FIGS. 3-5, embodiments of the present
invention may include an interface, which could be, for example, a
chart or spreadsheet based interface for submitting information
regarding orders, preferences, etc., and for displaying results to
a user, consistent with the above-described embodiments. These
types of interfaces are common to systems used in financial
trading, e.g., OMS and EMS. Thus, the user interface of the present
invention may be incorporated into the GUI of an OMS or EMS, and
can be implemented with known hardware and software components.
Additionally, the user interface of the present invention may be
web enabled and accessible through a browser, such as: Microsoft's
Internet Explorer, Google's Chrome, or Mozilla's Firefox. Further,
because the current invention contemplates integration with other
financial trading systems, it is likewise contemplated that
implementations of user interface will vary and are in no way
limited to the examples illustrated in FIGS. 3-5.
[0093] FIG. 3 is an example of how an embodiment of the invention
could be accessed and displayed in an EMS, such as ITG Triton.RTM..
In the current example, an EMS screen 302 includes a graphical
representation of the projected trade schedule generated by an
embodiment of the current invention 304. The graphical
representation of the projected trade schedule 304 may be
interactive, allowing the user to view an enhanced projected trade
schedule 306. In one embodiment, a configuration screen 308 may be
accessed either from the EMS screen 302, the enhanced projected
trade schedule 306, or some other menu. The configuration screen
308, in one embodiment, allows users to change the assumptions and
inputs used by embodiments of the current invention. Some of these
assumptions and inputs, as discussed above, include: horizon,
urgency, risk tolerance, volume, spread, and price trend.
[0094] FIG. 4 is an example of how one embodiment of the current
invention might integrate a graphical representation of smart
indicators, as described earlier in the application. In one
embodiment, the smart indicators are accessed in a browser, through
a "widget-style" graphical interface 402, or through other such
display mechanisms. In one embodiment, the widget-style graphical
interface 402 is interactive. For example, in FIG. 4 a detailed
smart indicator screen 404 is accessed through the widget-style
graphical interface 402. Co-owned/co-pending U.S. Provisional
Patent Application No. 61/103,719, titled Systems, Methods and
Computer Products for Providing Widgets for Performing Dynamic
Trading Analytics in a Financial Trading System, and filed on Oct.
8, 2008, includes a detailed discussion of widget-style GUIs, the
entire contents of which are herein incorporated by reference.
[0095] FIG. 5 is an example of how an embodiment of the invention
could be accessed and displayed. In one embodiment of the present
invention, the information used by and outputted from the current
invention can be shown in tabular or graphical format 502. This
interface allows for the user to interact with the values and
underlying settings of the model. The interface also allows for
interaction with an EMS, as discussed above, or an execution server
to facilitate the execution of trades suggested by the invention.
In this example, a user, by interacting with or clicking on the
field Shares Opt 504, can bring up or otherwise access the Dynamic
TC Widget 506. Using the Dynamic TC Widget 506 a user can change
the underlying trade information and reconfigure the trade strategy
by pressing the confirm button 512. Additionally, a user can view a
graphical representation of the trade strategy 508. In some
embodiments, a user can execute the trade strategy. This can be
done by pressing the Send to EMS button 510.
[0096] Various embodiments of the current invention could be
implemented using a combination of both hardware and software
components. For example, embodiments of the current invention could
be implemented using a Hewlett-Packard DL380 G6 with an Intel Xeon
Processor x5560, 72 gigabytes of random access memory, a graphics
processor, input/output devices (e.g., mouse, keyboard, monitor),
and storage (e.g., one or more hard disk drives) as a trade
optimization server. Additionally, in the current example, the
trade optimization server could be executing Red Hat Linux as an
operating system. Additionally, in the current example, an
electronic data storage facility comprises a combination of a
server, either the afore mentioned trade optimization server or a
different server, and a database software application, such as
Sybase ASE 15. Additionally, in the current example, the embodiment
of the invention includes an application server comprising a
combination of a server, either the afore mentioned trade
optimization server or a different server, and an application
server software application, such as JBOSS 4.3. One of ordinary
skill in the art will readily understand that embodiments of the
current invention could be implemented on other hardware and
software combinations.
[0097] One of ordinary skill in the art will readily understand
that the system can be based on architectures such as the Internet,
an intranet, a client server, a centralized server, distributed
servers, etc. The invention being thus described, it will be
apparent to those skilled in the art that the same may be varied in
many ways without departing from the spirit and scope of the
invention. Any and all such modifications are intended to be
included within the scope of the invention.
* * * * *