Ignore the hype and focus on machine learning objectives, say buy-side tech experts

Industry buzz around machine learning and AI must be set aside in favour of transparency and interpretability according to TradeTech FX speakers.

Advancements in machine learning and artificial intelligence (AI) may be improving trading strategies, but at the cost of transparency and interpretability, according to panellists at this TradeTech FX Europe.

Ian McWilliams, investment analyst at Aberdeen Asset Management, detailed how the understanding of what machine learning technologies are capable of is being distorted by a lack of understanding and exaggeration.

“I joke that when you are advertising externally you say AI, but inside you say machine learning and actually you are just doing logistic regression and things like that,” he said. “I don’t think that’s disingenuous, maybe it’s a bit of hyperbole, but it’s not wrong in terms of definitions, because when we talk about machine learning it really is anything where you are getting an algorithm to learn from data.”

“We’re taking a lot of market signals and sentiment signals, forecasting what markets are going to do in the future and using those to build trading strategies.”

McWilliams explained that the hype around elements of machine learning such as deep learning, image recognition and natural language processing (NLP) are distorting expectations around what are essentially tools to better model data for trading strategy decisions, particularly when it comes to conversations with fund managers.

“The interesting thing we need to think about as an industry and maybe where attitudes need to change is around interpretability of the models, which is a big question in a lot of areas, not just finance,” he said.

“Whenever we come out with a trade a question we get asked by the traditional fund managers is ‘Why is it making that trade?’ and they generally expect a very causal, A to B explanation, but that often defeats the point of these very complex algorithms. The middle ground is not good enough to just say that the algorithm says to do it, so we are doing it, but there needs to be more conversation between the quant people and more traditional people to understand there is a trade-off there.”

Saeed Amen, founder of trading consultancy Cuemacro and veteran of developed systematic trading strategies for Lehman Brothers and Nomura, extolled the virtues of simplicity when approaching the adoption of trading strategies based on machine learning technologies.

“The question to ask when you are thinking about using machine learning is: What are you trying to achieve, and can you use logistic regression or linear regression, for example, as sufficient for your task?” Amen said. “I would always try to go for the simplest tool necessary. There needs to be a rationale for the complexity involved in the trading strategy.”