Optimal mean-variance portfolio selection with mean-field reinforcement learning

• Main Author: Cheng, Zhengxing
• Other Authors: Patrick Pun Chi Seng
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Science::Mathematics
• Online Access: https://hdl.handle.net/10356/166475
• Institution: Nanyang Technological University

We study the mean-variance portfolio selection problem which is important in the finance field. The objective of the mean-variance portfolio selection problem is to find an optimal allocation strategy that achieves a great balance between expected return and risk. Because of the non-separable variance term, it is challenging to directly utilize dynamic programming or standard reinforcement learning to solve the problem. In this work, we construct a novel mean-field reinforcement learning framework to find the optimal strategy of the multi-period mean-variance portfolio problem in the discrete time-space. We first build a mean-field formulation of the mean-variance portfolio selection problem for mean-field reinforcement learning. After that, we propose and implement the multiple-period mean-field Q-learning with function approximation algorithm to obtain the optimal strategies. We design the linear quadratic Q-functions that fit the objective function and discrete time-space of the problem. we also per- form evaluations in various parameter settings to demonstrate the effectiveness of our proposed mean-field reinforcement learning framework.