A Consulting System for Portfolio Assets Allocation by Selecting the Best Agent in the Short Term Based on Cumulative Returns with Deep Reinforcement Learning

This paper proposes a sophisticated consulting system designed to optimize investment portfolios using deep reinforcement learning (DRL). The system includes a unique decision-making mechanism that selects the best-performing agent in the short term based on cumulative returns. We employ five different DRL agents: Advantage Actor-Critic (A2C), Soft Actor-Critic (SAC), Twin Delayed Deep Deterministic Policy Gradient (TD3), Deep Deterministic Policy Gradient (DDPG), and Proximal Policy Optimization (PPO), to allocate portfolio weights dynamically. By evaluating each agent’s performance over the previous ten days, our system adaptively selects the agent most likely to maximize returns in the immediate future. This approach enhances portfolio performance by leveraging the strengths of multiple DRL agents. The effectiveness of our proposed method is validated using historical data from the Dow Jones index, demonstrating significant improvements in cumulative returns and risk-adjusted performance metrics. Our results show that the main proposed method achieved the highest annual return of 11.43%, cumulative returns of 38.29%, and a Sharpe ratio of 0.832, outperforming individual DRL agents and highlighting the potential of DRL in advancing portfolio management strategies.