Systems and methods for electronic trade order routing
The present application generally relates to electronic trading systems, and more specifically to systems and methods for electronic trade order routing. Specifically, electronic trade orders may be divided into a first set and a second set. The first set of trade orders are executed with ad hoc execution styles. For the second set, execution style recommendations are generated in a form of a lookup table for the second set of trade orders according to a testing routing strategy. The second set of trade orders are executed with recommended execution styles. A difference between performance metrics among the first set of trade orders and the second set of trade orders is computed. A decision on whether to adopt the recommended execution style is generated based at least in part on the computed difference.
1 . A method of electronically routing an incoming electronic trade order at a routing server, the method comprising:
training a machine learning model implemented at the routing server using historical patterns of trade costs from a training dataset of historical trade data to predict an execution style;
in response to receiving, at a communication interface of the routing server, information of incoming electronic trade orders, randomly dividing, at the routing server, the incoming electronic trade orders into a first set and a second set;
routing, via the communication interface of the routing server, the first set of trade orders to one or more trade execution entities each corresponding to a different execution style in an ad hoc manner;
generating, by the trained machine learning model, execution style recommendations for the second set of trade orders according to a testing routing strategy;
routing, via the communication interface of the routing server, the second set of trade orders to corresponding trade execution entities based on the execution style recommendations;
generating a decision to adopt the recommended execution style based at least in part on a performance difference between the first set of trade orders and the second set of trade orders; and
routing, by the routing server, a new incoming trade order to a particular trade execution entity based on machine learning model generated recommendation.
2 . The method of claim 1 , wherein the execution style recommendations are selected from a set of execution styles including an auto-execution style, a request for quote style, and a direct voice-activated style.
3 . The method of claim 1 , wherein the performance metrics may include any of:
a mean execution implementation shortfall associated with a trade order; or
sales of a set of trade orders.
4 . The method of claim 1 , wherein the execution style recommendations are generated by a statistical model based on attributes including any combination of:
a side value,
an original trade size,
a price change in dollar terms in response to change in spread by a single basis point,
a wallet spread,
a coupon rate,
an amount issued, and
a rating.
5 . The method of claim 1 , further comprising:
computing a test statistic based on the difference between mean execution implementation shortfalls and implementation shortfall variances of the first set and the second set.
6 . The method of claim 4 , wherein the first set or the second set has a sample size no smaller than a threshold computed based on a configurable control level of a test corresponding to a different performance quantile level.
7 . The method of claim 1 , wherein the execution style recommendations are generated by combining (i) a respective probability indicating a likelihood that a trade order is to be executed under each execution style from a set of execution styles; and (ii) an estimate of an implementation shortfall metric for the trade order under the respective execution style.
8 . The method of claim 6 , further comprising:
retroactively classifying the first set of trade orders and the second set of trade orders depending on the respective probability or the estimate of the performance metrics; and
computing an impact differential based on a sample average of performance metric measurements of a subset of trade orders that a classification result matches with the respective probability or the estimate of the performance metric.
9 . A system of electronically routing an incoming electronic trade order at a routing server, the system comprising:
a communication interface receiving information of incoming electronic trade orders;
a memory storing a plurality of processor-executable instructions; and
a processor executing the plurality of processor-executable instructions to perform operations comprising:
training a machine learning model implemented at the routing server using historical patterns of trade costs from a training dataset of historical trade data to predict an execution style;
randomly dividing, at the routing server, the incoming electronic trade orders into a first set and a second set;
routing, via the communication interface of the routing server, the first set of trade orders to one or more trade execution entities each corresponding to a different execution style in an ad hoc manner;
generating, by the trained machine learning model trained by historical patterns of trade costs, execution style recommendations for the second set of trade orders according to a testing routing strategy;
routing, via the communication interface of the routing server, the second set of trade orders to corresponding trade execution entities based on the execution style recommendations;
generating a decision to adopt the recommended execution style based at least in part on a performance difference between the first set of trade orders and the second set of trade orders; and
routing, by the routing server, a new incoming trade order to a particular trade execution entity based on machine learning model generated recommendation.
10 . The system of claim 9 , wherein the execution style recommendations are selected from a set of execution styles including an auto-execution style, a request for quote style, and a direct voice-activated style.
11 . The system of claim 9 , wherein the performance metrics may include any of:
a mean execution implementation shortfall associated with a trade order; or
sales of a set of trade orders.
12 . The system of claim 9 , wherein the execution style recommendations are generated by a statistical model based on attributes including any combination of:
a side value,
an original trade size,
a price change in dollar terms in response to change in spread by a single basis point,
a wallet spread,
a coupon rate,
an amount issued, and
a rating.
13 . The system of claim 9 , wherein the operations further comprise:
computing a test statistic based on the difference between mean execution implementation shortfalls and implementation shortfall variances of the first set and the second set.
14 . The system of claim 13 , wherein the first set or the second set has a sample size no smaller than a threshold computed based on a configurable control level of a test corresponding to a different performance quantile level.
15 . The system of claim 9 , wherein the execution style recommendations are generated by combining (i) a respective probability indicating a likelihood that a trade order is to be executed under each execution style from a set of execution styles; and (ii) an estimate of an implementation shortfall metric for the trade order under the respective execution style.
16 . The system of claim 15 , wherein the operations further comprise:
retroactively classifying the first set of trade orders and the second set of trade orders depending on the respective probability or the estimate of the performance metrics; and
computing an impact differential based on a sample average of performance metric measurements of a subset of trade orders that a classification result matches with the respective probability or the estimate of the performance metric.
17 . A non-transitory processor-readable storage medium storing a plurality of processor-executable instructions for electronically routing an incoming electronic trade order at a routing server, the instructions being executed by a processor to perform operations comprising:
training a machine learning model implemented at the routing server using historical patterns of trade costs from a training dataset of historical trade data to predict an execution style;
in response to receiving, at a communication interface of the routing server, information of incoming electronic trade orders, randomly dividing, at the routing server, the incoming electronic trade orders into a first set and a second set;
routing, via the communication interface of the routing server, the first set of trade orders to one or more trade execution entities each corresponding to a different execution style in an ad hoc manner;
generating, by the trained machine learning model, execution style recommendations for the second set of trade orders according to a testing routing strategy;
routing, via the communication interface of the routing server, the second set of trade orders to corresponding trade execution entities based on the execution style recommendations;
generating a decision to adopt the recommended execution style based at least in part on a performance difference between the first set of trade orders and the second set of trade orders; and
routing, by the routing server, a new incoming trade order to a particular trade execution entity based on machine learning model generated recommendation.
18 . The non-transitory processor-readable storage medium of claim 17 , wherein the execution style recommendations are selected from a set of execution styles including an auto-execution style, a request for quote style, and a direct voice-activated style.
19 . The non-transitory processor-readable storage medium of claim 17 , wherein the performance metrics may include any of:
a mean execution implementation shortfall associated with a trade order; or
sales of a set of trade orders.
20 . The non-transitory processor-readable storage medium of claim 17 , wherein the operations further comprise:
computing a test statistic based on the difference between mean execution implementation shortfalls and implementation shortfall variances of the first set and the second set.