Can machines learn Chinese mutual funds?

This study applies machine learning techniques to predict mutual fund performance and construct high-performing portfolios in the Chinese mutual fund market. Using Random Forest, XGBoost, and LightGBM models, we frame fund selection as a classification problem, estimating each fund's probability of outperforming the market benchmark. Filtered fund characteristics serve as predictors, allowing us to build investable portfolios ranked by predicted outperformance likelihood. Empirical results reveal that machine learning-based portfolios consistently outperform the benchmark, with LightGBM delivering the strongest performance across key metrics. Furthermore, a flexible machine learning combination framework is introduced to improve predictive robustness and portfolio stability. These findings highlight the practical value of machine learning for fund selection and strategy design in complex financial environments.