Developing A Multi-Agent and Self-Adaptive Framework with Deep Reinforcement Learning for Dynamic Portfolio Risk Management

  • 2024-02-03 15:11:23
  • Zhenglong Li, Vincent Tam, Kwan L. Yeung
  • 0


Deep or reinforcement learning (RL) approaches have been adapted as reactiveagents to quickly learn and respond with new investment strategies forportfolio management under the highly turbulent financial market environmentsin recent years. In many cases, due to the very complex correlations amongvarious financial sectors, and the fluctuating trends in different financialmarkets, a deep or reinforcement learning based agent can be biased inmaximising the total returns of the newly formulated investment portfolio whileneglecting its potential risks under the turmoil of various market conditionsin the global or regional sectors. Accordingly, a multi-agent and self-adaptiveframework namely the MASA is proposed in which a sophisticated multi-agentreinforcement learning (RL) approach is adopted through two cooperating andreactive agents to carefully and dynamically balance the trade-off between theoverall portfolio returns and their potential risks. Besides, a very flexibleand proactive agent as the market observer is integrated into the MASAframework to provide some additional information on the estimated market trendsas valuable feedbacks for multi-agent RL approach to quickly adapt to theever-changing market conditions. The obtained empirical results clearly revealthe potential strengths of our proposed MASA framework based on the multi-agentRL approach against many well-known RL-based approaches on the challenging datasets of the CSI 300, Dow Jones Industrial Average and S&P 500 indexes over thepast 10 years. More importantly, our proposed MASA framework shed lights onmany possible directions for future investigation.


Quick Read (beta)

loading the full paper ...