J.P. Morgan has started applying tech that enables machine-trading programs to learn from previous trades and search for the most profitable way to execute them. The broker told Reuters its new algorithm — named DNA or Deep Neural Network for Algo Execution — effectively combined what a multitude of algos currently do, into a single strategy and allowed the framework to judge how a client order should be executed.
For example, a typical time-weighted average price order executed by an algorithm may aim to buy a particular amount of currency over a few minutes or hours. But if the order is not executed within that time frame, the machine will trade aggressively toward the end to buy the required amount. The new algo aims to take that decision-making process one step further by determining on how best to execute the transaction, based on results of past trades.
The algo has already been deployed for trading G7 currencies such as the euro, dollar and sterling, where it has access to data from thousands of past trades. “The objective of an algo is to minimize market impact by executing in an efficient and timely manner,” Chi Nzelu, head of macro eCommerce at JP Morgan, told Reuters. “What we have done is establish a neural network using a machine learning technique which determines how to place the order, at what price and execution style.”