Related Experiment Videos
AHRL-PM: Asynchronous Hierarchical Reinforcement Learning Framework for Enhanced Portfolio Management
Summary
This study introduces an asynchronous hierarchical reinforcement learning for portfolio management (AHRL-PM). The AHRL-PM framework enhances active portfolio management by outperforming traditional methods in global markets.
Area of Science:
- Quantitative Finance
- Machine Learning
- Financial Engineering
Background:
- Effective portfolio management (PM) faces challenges from market data uncertainty and high dimensionality.
- Traditional PM techniques often result in suboptimal asset allocation due to these complexities.
Purpose of the Study:
- To introduce an asynchronous hierarchical reinforcement learning for portfolio management (AHRL-PM) framework to enhance active PM.
- To address the limitations of traditional PM techniques in navigating market complexities and uncertainties.
Main Methods:
- A two-tiered agent system: a first-layer agent selects stocks monthly using alpha factors.
- A second-layer agent adjusts asset weights daily using market fundamentals and technical indicators for dynamic rebalancing.
Main Results:
- The AHRL-PM framework consistently outperformed traditional benchmarks in annualized return (AR) and Sharpe ratio (SR) across six global markets.
- Ablation studies confirmed superior profitability, risk-adjusted returns, and reduced training time compared to synchronous and single-layer RL models.
Conclusions:
- The AHRL-PM model demonstrates robustness and versatility in portfolio management.
- The framework offers efficient and practical applicability for responding to market dynamics and uncertainties.