Goldman Sachs AI head likes technique that 'creates a trader'
Acronyms proliferate in artificial intelligence, but if you thought it was only about AI, ML and NLP, you're missing out. There's also RL, reinforcement learning. In a panel at Quant Strats 2023, Goldman Sachs' EMEA head of applied AI, Francesco Maria Delle Fave, weighed in on where it can be used in trading and what its ramifications are for the future of trading careers.
What is reinforcement learning? Fellow panelist, UBS quant analyst Leila Korbosli, said its "main feature is learning through introduction, through trial and error." Fave says it's about creating a "map, a state of the world," and repeatedly testing courses of action, creating a pseudo "score" which determines the best action you can take.
Goldman and Fave are "exploring different values," with which to use RL. Fave says that "derivatives and executions are the most promising ones." Panelist Aitor Muguraza Gonzalez, head of scientifiic research at hedge fund Kaiju Capital Management, says RL is particularly effective when datasets have "a beginning and an end," which derivatives possess due to their expiration dates.
RL is far from perfect, however. Gonzalez says reinforcement models are "remarkably hard to train," requiring immense amounts of data. This presents a significant barrier to entry and initial investment for firms hoping to implement it. Panel moderator Iman Horavar, director of quant research at Robeco, suggests that the best use case to counter this is high frequency trading, as "there are more seconds than there are months."
Get Morning Coffee in your inbox. Sign up here.
Gonzalez says RL "scares a lot of traders," because it "physically creates an actor that replaces a trader." Fave maintains, however, that its eventual implementation will be a mix of "man and machine." In five to ten years, AI autopilots will be there to assist "a lot of the daily quant work," changing the job but not eliminating it.
Another area RL could be used is as the mythical market predicting robot people assumed ChatGPT could be. What Fave is "very positive about is generating synthetic data and using it to create market predictors." He doesn't expect that avenue to be lucrative within the next decade, however.
Have a confidential story, tip, or comment you’d like to share? Contact: +44 7537 182250 (SMS, Whatsapp or voicemail). Telegram: @SarahButcher. Click here to fill in our anonymous form, or email editortips@efinancialcareers.com. Signal also available.
Bear with us if you leave a comment at the bottom of this article: all our comments are moderated by human beings. Sometimes these humans might be asleep, or away from their desks, so it may take a while for your comment to appear. Eventually it will – unless it’s offensive or libelous (in which case it won’t.)
