Abstract: A scientific prediction of stock volatility is of significant theoretical and practical importance for risk measurement and management by investors and regulatory authorities. This paper incorporates investor sentiment, time-varying fluctuation, asymmetry, fat-tailed distribution, and long-term memory properties to develop the SMA-Realized AHAR GARCH CICSI model under the skewed-t distribution framework for volatility forecasting and risk assessment. Using high-frequency data of the CSI 300 Index, we estimate the model parameters through the maximum likelihood method. VaR predition and posterior test, MCS test, out-of-sample R² test, and the economic value evaluation are employed to comprehensively compare the volatility forecasting performance and economic utility of different models. The empirical results indicate that accounting for investor sentiment, time-varying fluctuation, asymmetry, fat-tailed characteristics, and long-term memory significantly enhances the model’s fitting performance and predictive accuracy. Among the tested models, the SMA-Realized AHAR GARCH CICSI model demonstrates superior volatility forecasting ability and economic utility.