Financial Planning via Deep Reinforcement Learning AI

Abstract

This paper introduces AIPlanner, a financial planner based upon deep reinforcement learning. AIPlanner provides an investment and consumption strategy intended to optimize lifetime well-being. The results of AIPlanner are very close to the precise analytical solution, as well as to the precise solution computed using stochastic dynamic programming. Deep reinforcement learning is additionally capable of delivering results for far more complicated and realistic financial models that other approaches can't handle. As an example of this capability, a bond model that includes both a yield curve and time varying interest rates is employed. Compared to other popular approaches, in one reasonable scenario, AIPlanner was found to effectively deliver approximately $1,000 to $8,000 of additional consumption per year.