Research
My current line of research investigates the different ways in which AI-based trading agents may learn non-normative (or simply unlawful) behaviors when tasked with profit maximization in a complex financial market. To this end, I have created two realistic multi-agent market simulations: ABIDES and minABIDES.
Using my simulations, I have explored whether a reinforcement-learning based trader will learn to pump-and-dump a market using LLM-enabled social media posts, how to stop a similar trader from learning to spoof the stock market, or whether multiple AI trading agents within the same firm might accidentally learn cooperative price fixing. The simulations have proven flexible enough to also empirically validate theoretical impovements in privacy-preserving federated learning protocols and to study the effect of AI traders on bubble formation. My work has thus far been cited in over one thousand peer reviewed publications.
In the future, I hope to broaden the application scope of my research to incorporate normative reinforcement learning in other contexts, and to broaden the lens of misbehavior I consider within the financial space. Of particular interest are my ongoing efforts to positively influence regulation and governance practices around autonomous AI agents introduced into complex real-world systems.
Selected Peer-Reviewed Articles
-
"The Accidental Pump and Dump:
When Agentic AI Meets Autonomous Trading"
-
"Learning to Trade with Preferences:
Interpretable Execution via Mixture-of-Experts"
-
"Multi-Agent Market Simulation for Deep
Reinforcement Learning with High-Frequency Historical Order Streams"
-
"Learning Not to Spoof"
-
"Differentially private secure multi-party
computation for federated learning in financial applications"
Further Information
Please see my Google Scholar for a complete publication listing with citation metrics and my CV for further information.