Reproduce “Deep Graph Convolutional Reinforcement Learning for Financial Portfolio Management -- DeepPocket” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Farzan Soleymani, Eric Paquet
A framework named DeepPocket combines Graph Convolutional Networks with an Actor-Critic Reinforcement Learning agent for dynamic financial portfolio management, demonstrating superior risk-adjusted returns and profitability even during the COVID-19 market downturn. Developed by researchers at the National Research Council, this system adapts to evolving market conditions through a hybrid offline-online learning strategy.
Key insights
- Explicitly modeling the dynamic interrelations among financial assets as a graph using GCNs is crucial for accurately capturing the non-Euclidean complexities of market data.
- The Actor-Critic reinforcement learning framework is effective for continuous action spaces like portfolio weight allocation, balancing exploration and exploitation to achieve stable policy optimization.
- A two-stage offline-online learning strategy provides adaptability to 'concept drift' by integrating foundational learning from extensive historical data with continuous updates from recent market conditions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.