Stock Tail Risk and Investor Sentiment: A Machine Learning Approach to Multi-Factor Pricing

Abstract

Using China's Α-share market as the research object, this study examines the predictive effect of investor sentiment on stock tail risk and its threshold effect. First, principal component analysis is used to extract a composite sentiment index from proxy variables such as turnover, the growth rate of new investor account openings, and the margin financing and securities lending ratio, and tail risk is measured using Conditional Value at Risk ( CVaR a ). Second, an eXtreme Gradient Boosting ( XGBoost ) model is constructed to conduct out-of-sample prediction of tail risk, and its performance is compared with that of linear regression and random forest. The results show that the prediction error of the XGBoost model is significantly lower than that of the benchmark models, indicating a nonlinear relationship between sentiment and tail risk. Further identification based on SHapley Additive exPlanations (SHAP) value decomposition shows that the sentiment threshold is approximately 0.0838. When the sentiment index exceeds this threshold, its marginal contribution to tail risk becomes significantly stronger. This study provides empirical evidence for understanding the formation mechanism of extreme risk driven by investor sentiment and offers a quantitative reference for regulators to conduct dynamic risk early warning.