HCIproject

This project presents a novel human-centric reinforcement learning (RL) system for equity portfolio allocation that enhances traditional and existing AI-based trading strategies. Utilizing a customized RL environment based on OpenAI Gym and historical data from Dow 30 companies, our system integrates Deep Q-Networks (DQN) and Deep Deterministic Policy Gradient (DDPG) methodologies. These frameworks allow for dynamic adaptation to market changes with an emphasis on risk mitigation and capital preservation. Our comprehensive backtesting demonstrates the system’s ability to outperform conventional trading strategies, achieving an annual return of 22.29% with a Sharpe ratio of 1.72. Insights from clus- tering analysis on user preferences and behavior underline the system’s versatility across varying trader expertise, from novices to experts. Challenges related to incentive function design, computa- tional demands, and data dependency are discussed. Moreover, the importance of addressing emotional aspects of trading decisions is emphasized, illustrating the system’s potential to revolutionize financial portfolio management through a blend of AI and human judgment.