Value investing with machine learning: China market

Abstract

This study investigates the use of machine learning models—XGBoost, LightGBM, and Support Vector Regression (SVR)—for stock price prediction in the Chinese market, leveraging financial indicators such as EPS, ROE, revenue growth rate, PE ratio, and PB ratio. Using a dataset spanning 2019 to 2023 with 2,200 stocks, the models were evaluated using Mean Squared Error (MSE) and the Coefficient of Determination (R²). XGBoost excelled in multi-feature scenarios, achieving an R² of 0.81, while SVR demonstrated strong performance with single-feature data (R² = 0.60), effectively capturing non-linear relationships. LightGBM closely followed, achieving an R² of 0.80, showcasing its efficiency in handling high-dimensional data. Stocks recommended by all three models, identified based on predicted prices exceeding actual prices by over 100%, represent high-confidence investment opportunities. This study underscores the po- tential of machine learning to enhance financial forecasting in the volatile and policy-driven Chinese stock market, where high-confidence recommendations are ensured through the combined strengths and consensus of multiple models. Future research could incorporate macroeconomic indicators, market sentiment, and advanced hybrid modeling approaches to further improve predictive accuracy and robustness in dynamic financial environments.