GTC 2020: An Approach to Using Reinforcement Learning to Mimic Portfolio Behavior
Yigal Jhirad, Cohen & Steers
We'll present the application of reinforcement learning to create factor-mimicking portfolios. Factor portfolios have become a more prominent part of the investment landscape. We'll discuss alternatives to derive factor-mimicking portfolios with multi-period constraints using a model free reinforcement-learning approach. The reinforcement-learning implementation utilizes Q Learning to update the Q table using the Bellman equation. Effectively, the agent determines an appropriate policy that can replicate a portfolio conditional on multi-period historical constraints, creating a more robust framework to build out a tracking portfolio.