RL Summit

Reinforcement learning-based simulation of markets

Tuesday, March 29, 4:30PM UTC

Sumitra Ganesh from J.P. Morgan talks about using RL for agent-based simulation of markets.

Agent-based simulation of markets provides a useful tool for policy optimization, counterfactual analysis, and market mechanism design. In this talk, we will present our work in modeling complex economic systems as a network of heterogeneous, utility maximizing agents with partial observability. We demonstrate how reinforcement learning (RL) can be used to solve two primary challenges in agent-based modeling — finding the equilibrium with multiple strategic agents and calibrating the model using real data. These techniques have been useful in enabling the practical application of agent-based modeling.



Sumitra Ganesh

Sumitra Ganesh

Research Director, J.P. Morgan AI Research

Sumitra Ganesh leads the Multi-agent Learning & Simulation group at JPMorgan AI Research. Her team’s research focuses on modeling complex economic systems and efficient policy learning. Sumitra has led the development of a multi-agent simulation platform that uses reinforcement learning to learn agent behaviors in a scalable manner. The simulation platform developed by her team is being used across multiple use cases (market simulation, operational processes, consumer loan markets) for counterfactual analysis and strategy optimization. Prior to joining AI Research, Sumitra led the X-asset Client Intelligence team in the Corporate & Investment Bank at J.P.Morgan where she worked with sales and product teams to improve client experience. Her team developed the first personalization engine for JPMorgan Markets and machine learning products to improve workflow for Equities sales. Prior to joining JPMorgan in 2016, Sumitra was part of the Franchise Analytics Strats team at Goldman Sachs. Sumitra has a PhD in Electrical Engineering and Computer Science from U.C. Berkeley.