Stock index direction forecasting using an explainable eXtreme Gradient Boosting and investor sentiments

•Multiple types of investor sentiment in Chinese security market are utilized for forecasting.•Eighteen sentiment features are employed as inputs for the forecasting models.•XGBoost produces the best forecasting results among all the machine learning methods.•Relative importance of the sentiment features and their changes over time are analyzed. In this article, an explainable eXtreme Gradient Boosting (XGBoost) method is proposed for stock index prediction and trading simulation in the Chinese security market. Sentiment features of three types of investors, including institutional, individual, and foreign investors, are utilized as explanatory variables, and a binary classification model based on XGBoost is constructed to predict the direction of the Shanghai composite index and Shenzhen composite index movements. Additionally, the Gain function of XGBoost and SHapley Additive exPlanations (SHAP) are employed to estimate the importance of sentimental factors affecting index direction forecasting. Experimental results demonstrate that the XGBoost-based approach using multiple investor sentiments achieved the best forecasting accuracy, and sentiment features of the institutional investor were relatively more essential than the individual investor and foreign investor sentiments for index direction prediction in most out-of-sample periods. It demonstrates that the method proposed in this research can provide useful references for market participants to support their investment in the Chinese security market.