Project information
Sentiment-Driven Volatility Forecasting (Y. Z)

My research explores the integration of financial market microstructure features and sentiment indicators into a unified predictive framework for stock index futures volatility and return dynamics. The main objective is to enhance the short-term forecasting accuracy of market fluctuations by combining high-frequency trading data with heterogeneous information extracted from investor sentiment indices.

To achieve this, I employ a hybrid machine learning approach that fuses multiple model families—regularized regression, ensemble learning, and SHapley Additive exPlanations (SHAP). The methodological innovation lies in the multi-sentiment fusion, which enables the model to learn nonlinear interactions between market uncertainty, liquidity, and investor sentiments. This approach provides a more adaptive and interpretable framework for real-time financial forecasting compared with traditional econometric models.

The project is closely connected to my doctoral dissertation, which aims to construct an interpretable and data-driven volatility prediction system for commodity and equity futures markets. The study focuses particularly on how information asymmetry and behavioral sentiment propagate through high-frequency price movements and volatility measures.

Through the development and evaluation of advanced machine learning models using high-frequency data, the project contributes both to the methodological field of financial machine learning and to the practical domain of market risk management. It also provides a foundation for designing volatility-managed trading strategies and understanding the behavioral mechanisms behind rapid market risk.