Statistical Modelling of Stock-Market Volatility Forecasting with Investor Sentiment

Abstract

Volatility forecasting plays a fundamental role in risk management, asset allocation and derivative pricing. In the Chinese equity markets, where retail participation is high and arbitrage is limited, the volatility of returns is well documented to be driven by investor sentiment. We construct and document an investor-sentiment index and embed it into an easily reproduced framework for forecasting volatility. The framework combines the GARCH family (GARCH, EGARCH and GJR-GARCH) with a sentiment-augmented mixed-frequency extension (GARCH-MIDAS) and a recurrent-neural-network benchmark (LSTM). Evaluation is carried out out-of-sample using RMSE, MAE and the Diebold–Mariano test. Each step is made concrete and auditable by generating the figures and tables through code that runs on a transparent dataset calibrated to the documented stylized facts of the CSI 300; the same code accepts real index and sentiment series as drop-in replacements. The paper’s contribution is therefore the methodology and its tooling: the modelling pipeline, the recipe for constructing the sentiment index, and a complete, scripted experimental design that an applied researcher can run on live data to obtain empirical estimates.