Assessing AI in Trading

How far have artificial intelligence applications in capital markets come, and where is the technology headed?

That was a discussion topic at the FIX Americas Trading Conference 2024, held Oct. 16 in New York.  

The AI Insights: Revolutionizing Financial Trading panel, opened with the simple, straightforward question of what has changed in the past year. 

It was noted that conversations in October 2023 were more about pilot programs and proofs of concepts, whereas now with more development now there is more AI actually in use.

Still, some unrealistic expectations of a “magical trading robot” or the like remain out there, and remain far from being realized.  

AI goes hand-in-hand with data, and the panel noted that AI is making strides in giving traders more useful knowledge while reducing the noise that comes from data overload. For example, an AI-powered cluster model can screen stocks for characteristics such as capitalization, liquidity, and spread, telling the trader whether a given stock is relatively easy or difficult to trade. 

Even if that works well, one panelist said “there is always information on the trading desk that is not captured in data, and you need a trader to still make human decisions.” And in fact, areas of progress over the past year have been in better human decision-making and challenging of the data. 

Risk management is fertile ground for applying AI, including the newer generative AI, the panel noted. AI won’t itself solve a risk problem, but “it will give a human expert a head start on where it’s best to apply efforts” to solve the problem. 

The panel stressed the importance of guardrails and education in rolling out AI. The technology “isn’t really making decisions in these early days. But it is freeing up humans to do more interesting and important work.”

One rough benchmark to strive for is AI freeing up 90% of human trader and technologist time, so they can focus on the most important 10% of their work. 

The future of AI is potentially boundless, as it was noted that today’s AI models “are the worst you’ll ever see” when compared with what’s to come.