Research on Stock Realized Volatility Prediction Using Multi-Models Incorporating International Macro Factors

Abstract

Predicting realized volatility (RV) is crucial for asset allocation and risk management. Existing multi-factor research on predicting realized volatility in stocks has rarely incorporated international macroeconomic factors, and the comparison of machine learning models remains limited. This paper constructs a multi-factor framework combining technical and international macroeconomic factors using monthly data from the 50 most liquid stocks in the CSI 300 Index from 2018 to 2024. Three models-ridge regression, extreme gradient boosting (XGBoost), and a multilayer perceptron (MLP)-are used to predict the next period's RV. Results show that incorporating international macroeconomic factors improves the forecast accuracy of all models, with machine learning performing even better on the test set (R²≈0.43). The study finds that incorporating international macroeconomic factors significantly improves model forecast accuracy, with machine learning outperforming linear regression models. This research provides a reference for selecting methods for volatility forecasting in different market scenarios and offers theoretical insights for investors, asset pricing, and risk management.