Quantitative money managers are exploring how to apply generative artificial intelligence and large language models (LLMs) in their day-to-day work, but many are skeptical about AI's uses in the investment process to generate alpha.
Instead, mundane tasks are proving more fruitful, said panelists and attendees at the Quant Strats conference, held in New York on March 12.
The idea of AI completely taking over and the human touch becoming irrelevant is doubtful, said Stacie Mintz, head of quantitative equity at PGIM Quantitative Solutions.
“Human beings still create the models, so even though we use models for stock selections and portfolio construction, the humans are the ones creating and evolving the models,” she said.
Investors will always be faced with situations and geopolitical risks where models need thoughtful adjustments, Mintz said.
“Really with any new technology, you need to understand what the potential pitfalls are,” she said. “And as investors, we are always going to come into situations that we've never seen before, so COVID is a great example.”
AI needs human supervision, said Tim Liu, a senior researcher at Neo Ivy Capital, a quantitative hedge fund.
“In my view, AI has some potential, but it’s not the key for everything,” he said.
It would take 100% confidence that generative AI can replicate humans for it revolutionize investment decision-making and replace humans, Liu said, adding, “I don’t see the day coming anytime soon.”
A true leap could come if something closer to artificial general intelligence — a field of AI that is trying to create software with human cognitive abilities so it can solve unfamiliar tasks — develops, said Gordon Ritter, CIO of Ritter Alpha, an investment adviser that runs systematic absolute-return strategies.
Currently, quants need to be cautious around using AI to generate trading models, said Michael Weinberg, an adjunct professor of finance and economics at the Columbia Business School and special adviser to the Tokyo University of Science endowment.
“You have to be uber careful that you’re not overfitting or looking at spurious correlations or relationships,” he said. “It could be that that the AI or LLM finds a relationship where certain stocks or industries or markets outperform or underperform while something else happens, but that could be entirely coincidental, not causal.”
Weinberg uses a checklist with over 80 data points to ascertain what alternatives managers are really doing with artificial intelligence.
For now, generative AI is great at automating routine tasks, he said.
Many quants are seeing applications and productivity gains outside of investing.
“[Y]ou can automate a lot of the simpler stuff, like docs and legal … and operations,” said Milind Sharma, CEO of the hedge fund QuantZ Capital Management. “I think there is tremendous and obvious cost savings and productivity that can be enhanced on in the middle and back office.”
Yesim Tokat-Acikel, a managing director and portfolio manager at Principal Asset Management, said she is finding generative AI use cases experimenting with automating and writing responses for requests for proposals, creating initial drafts of documents and commentary.
Using a large language model that costs $30 a month is helpful for automation work and debugging code and can replace some work that was previously done by entry-level quants or software engineers, said Revant Nayar, CIO of the hedge fund FMI Technologies.
But for alpha generation, generative AI shouldn’t have an impact, Nayar said. It’s an “extreme statement” he said, but with everyone having access to models, a consensus opinion will take hold. The best quants will be able to think of things that LLMs have not, he said.