Reproduce “FinXplore: An Adaptive Deep Reinforcement Learning Framework for Balancing and Discovering Investment Opportunities” on OpenResearch

About this paper

Himanshu Choudhary, Arishi Orra, Manoj Thakur

FinXplore, an adaptive deep reinforcement learning framework, employs a dual-agent architecture to dynamically balance asset allocation within an existing investment universe with the exploration of new opportunities. This approach consistently achieved superior cumulative returns (e.g., 127.91% on NIFTY) and higher risk-adjusted returns (e.g., Sharpe Ratio of 1.83 on NIFTY) compared to state-of-the-art DRL and traditional methods across diverse market conditions.

Key insights

• Explicitly integrating the exploration of new investment opportunities within an extended universe alongside the exploitation of existing assets significantly enhances portfolio performance and adaptability.
• A collaborative dual-agent DRL architecture, where PPO handles existing asset allocation and DQL manages new asset discovery, effectively addresses the exploitation-exploration trade-off in dynamic financial markets.
• Utilizing risk-adjusted returns (Sharpe Ratio) as the primary reward function, combined with transaction costs, guides DRL agents towards more robust and financially prudent investment strategies.

Summary by alphaXiv.