Patience is a virtue when it comes to machine learning say experts

Industry experts talk down immediate results from machine learning and that there are no shortcuts to success.

Capital markets firms that are looking to implement machine learning and artificial intelligence (AI) systems within their trading processes must be prepared to undertake a multi-year project that requires significant patience before seeing results.

Speakers taking part in a keynote address at this year’s TradeTech conference outlined how their firms had approached machine learning projects and warned that those expecting immediate results from such endeavours would most likely be disappointed.

Antish Manna, head of execution research at MAN GLG, said that the firm went live with a machine learning-based framework for order flow and broker allocation last year.

“This framework effectively takes away the need for human to set an arbitrary target for ‘my first three brokers are going to get this amount of flow’ and continuously updating that target to having a machine that automatically does that”, Manna explained.

“The beauty of it is that it becomes a very clean conversation with our brokers; they know how we are doing things and that they will get more flow, and this machinery also adapts to changing market conditions.”

Manna explained that although the hype around machine learning has grown to a point where expectations are now becoming unrealistic as to what the technology can achieve, starting with a relatively simple element such as broker allocation means the firm can build out the framework take on more expansive and intuitive projects in future.

Representing the sell-side was Shary Mudassir, co-head of global equities execution at RBC Capital Markets, who agreed that industry perceptions around machine learning were often false, particularly around how long such developments take to complete and the amount of time it can take to build the required expertise.

“There is a perception that you can hire people and have meaningful, AI-based outcomes…it doesn’t work that way. Real success with AI requires very large teams,” he said.

“At RBC, we’ve got in our AI research team over 100 AI scientists. Within the applied AI space we have over 300 data scientists on the bank side. On the equities execution team now for the most recent product we will be rolling out at some point over the year, we have a team of over 60 people, and these 60 people are not all AI scientists – these are sales traders, traders, quants, execution consultants, technologists…they have all come together over that period of time to deliver on one big outcome.”

Addressing those unrealistic expectations, Manna said that the majority of time spent on machine learning projects is used to clean data before research and development can take place, and that those firms that are only now starting their journey with machine learning should be not expect to see results in the short-term.

“The truth is, it is a fallacy and it takes a huge amount of time to build a framework where you can deliver things at scale that work,” he said. “On the machine learning and AI side of things, problems are best solved by teams of people, because you need the challenge, rigour and time to learn and fail, learn and try again; that process takes a lot of time.”