JPMorgan will soon be using a first-of-its-kind robot to execute trades across its global equities algorithms business, after a European trial of the bank’s new artificial intelligence (AI) programme showed it was much more efficient than traditional methods of buying and selling. The AI — known internally as LOXM — has been used in the bank’s European equities algorithms business since the first quarter and will be launched across Asia and the US in the fourth quarter, Daniel Ciment, JPMorgan’s head of global equities electronic trading, told the Financial Times. LOXM’s job is to execute client orders with maximum speed at the best price, by using lessons it has learnt from billions of past trades — both real and simulated — to tackle problems such as how best to offload big equity stakes without moving market prices.
“Such customisation was previously implemented by humans, but now the AI machine is able to do it on a much larger and more efficient scale,” said David Fellah, of JPMorgan’s European Equity Quant Research team. Mr Ciment said that, so far, the European trials showed that the pricing achieved by LOXM was “significantly better” than its benchmark. Investment banks have been trying to use AI, automation and robotics to help cut costs and eliminate time-consuming routine work. For example, UBS’s recent deployment of AI to deal with client post-trade allocation requests, which saves as much as 45 minutes of human labour per task. UBS has also brought in AI to help clients trade volatility. JPMorgan, which is the world’s biggest investment bank by revenue, believes it is the first on Wall Street to use AI with trade execution and said it would take rivals 18 to 24 months and an investment of “multiple millions” to come up with similar technology.
“Best execution is becoming more and more important to clients,” said Mr Ciment of JPMorgan’s decision to invest in the pioneering technology, adding that it could become part of the marketing pitch the bank makes to clients. The AI was developed using “Deep Reinforcement Learning” methods, which are able to learn from millions of historic scenarios. Mr Fellah said DRL had “many other potential uses in banking, such as in automatic hedging and market making”. One possible evolution of LOXM is teaching the machine how to get to know individual clients, so that it could consider their behaviour and reaction as it decides how to trade. “Any customisation would only be if the client agrees to that,” Mr Ciment added. Unlike the robo advisers offered by some private banks, JPMorgan’s AI has no decision-making capabilities around what is bought and sold, its role is solely to decide how things are bought and sold.