Portfolio optimization is crucial for achieving a balance between risk and return in financial markets. Traditional deep reinforcement learning (DRL) approaches are limited as they often work within a fixed investment universe, which restricts the exploration of potentially profitable opportunities. To overcome this limitation, the proposed methodology introduces an innovative investment landscape that employs two DRL agents. One agent is tasked with managing assets within the existing universe while the other focuses on exploring new investment possibilities in an extended universe.
The experiments conducted using two real-world market datasets provide compelling evidence of the proposed approach’s effectiveness. By dynamically shifting focus between exploiting known assets and exploring new opportunities, the framework enhances overall portfolio performance compared to state-of-the-art strategies. The implications are significant, suggesting that employing a dual-agent system can lead to better risk-adjusted returns. However, risks remain, including the potential for overfitting to historical data and the challenges of managing exploration efficiently.
👉 Pročitaj original: arXiv AI Papers
