Machines are less susceptible to irrational exuberance, meaning a growing reliance on AI-powered investment advice could produce fewer Dutch tulip manias and dot-com bubbles, suggests a new paper from the U.S. Federal Reserve.
In a staff working paper examining the potential implications for financial stability from increased use of AI tools to inform trading decisions, researchers used laboratory experiments to study the behaviour of large language models (LLMs) in classic studies of herding in financial markets.
“Our results show that AI agents make more rational decisions than humans, relying predominantly on private information over market trends,” it reported.
In the experiment, researchers found AI agents made rational decisions between 61% and 97% of the time, compared with 46% to 51% for humans.
The AI agents were also less susceptible to “information cascades,” when investors trade based on the actions of others rather than their own information. Information cascades occurred in 0–9% of AI decisions, versus 20% for humans.
The research also found that “when AI agents did engage in cascade trading behaviour, they traded against market trends (contrarian behaviour) rather than following the herd.”
Unlike human traders, AI agents don’t “exhibit completely irrational behaviour (or make trading errors).”
“Our results suggest that AI agents exhibit less herd behaviour than human financial professionals, a finding with significant implications for future financial stability as generative AI gains traction in market decision making,” the paper said.
Less herding behaviour could mean less market volatility and fewer asset price bubbles, the paper said, “contributing to greater overall financial market stability.”
That said, the researchers noted “the introduction of AI agents could fundamentally alter market dynamics in ways that are not yet fully understood.”