The Use of Deep Reinforcement Learning in Tactical Asset Allocation

Abstract

The Tactical Asset Allocation (TAA) problem is a problem to accurately capture short to medium term market trends and anomalies in order to allocate the assets in a portfolio so as to optimize its performance by increasing the risk adjusted returns. This project seeks to address the Tactical Asset Allocation problem by employing Deep Reinforcement Learning (DRL) Algorithms in a Machine Learning Environment as well as employing Neural Network Autoencoders for selection of portfolio assets. This paper presents the implementation of this proposed methodology applied to 30 stocks of the Dow Jones Industrial Average (DJIA). In (1), the Introduction to the project objectives is done with the Problem Description presented in (2). Part (3) presents the literature review of similar studies in the subject area. The methodology used for our implementation is presented in (4) whilst (5) and (6) presents the benchmark portfolios and the DRL portfolios development respectively. The evaluation of the performance of the models is presented in (7) and we present our conclusions and the future works in (8).

Keywords: Tactical Asset Allocation, Reinforcement Learning, Neural Networks, Markov Decision Process, Asset Portfolio

JEL Classification: G11, C61