AI-Enhanced Portfolio Management in China’s A-Share Market: A Dynamic Factor Integration Framework

This study develops and validates an integrated AI-driven asset management framework tailored for the Chinese A-share market. By synergistically combining Gradient Boosting Decision Trees (GBDT) for return forecasting, Reinforcement Learning (RL) for dynamic portfolio allocation, and Graph Neural Networks (GNNs) for cross-asset relationship modeling, we construct a multi-factor strategy that adapts to evolving market regimes. Rigorous event-study analysis is employed to quantify and exploit stock price reactions to corporate announcements, capturing both short-term anomalies and medium-term trends. Our backtesting results (2021–2025) demonstrate that this integrated framework achieves a Sharpe ratio of 1.34 and a cumulative return of + 901.7%, significantly outperforming static benchmarks like the CSI 300. The findings underscore the importance of adaptive factor timing, integration of alternative data sources, and event-driven insights for generating persistent alpha in complex and volatile equity markets. These results suggest that AI-driven multi-factor frameworks can enhance investment decision-making, improve portfolio robustness, and offer practical guidance for asset managers seeking systematic and data-driven approaches in emerging market environments.