Author:
Deng Shangkun,Zhu Yingke,Duan Shuangyang,Fu Zhe,Liu Zonghua
Abstract
Stock price crashes have occurred frequently in the Chinese security market during the last three decades. They have not only caused substantial economic losses to market investors but also seriously threatened the stability and financial safety of the security market. To protect against the price crash risk of individual stocks, a prediction and explanation approach has been proposed by combining eXtreme Gradient Boosting (XGBoost), the Non-dominated Sorting Genetic Algorithm II (NSGA-II), and SHapley Additive exPlanations (SHAP). We assume that financial indicators can be adopted for stock crash risk prediction, and they are utilized as prediction variables. In the proposed method, XGBoost is used to classify the stock crash and non-crash samples, while NSGA-II is employed to optimize the hyperparameters of XGBoost. To obtain the essential features for stock crash prediction, the importance of each financial indicator is calculated, and the outputs of the prediction model are explained by SHAP. Compared with the results of benchmarks using traditional machine learning methods, we found that the proposed method performed best in terms of both prediction accuracy and efficiency. Especially for the small market capitalization samples, the accuracy of classifying all samples reached 78.41%, and the accuracy of identifying the crash samples was up to 81.31%. In summary, the performance of the proposed method demonstrates that it could be employed as a valuable reference for market regulators engaged in the Chinese security market.
Funder
Natural Science Foundation of Hubei Province
Philosophy and Social Science Research Project of the Department of Education of Hubei Province
Natural Science Foundation of Yichang City
Subject
Information Systems and Management,Computer Networks and Communications,Modeling and Simulation,Control and Systems Engineering,Software
Cited by
8 articles.
订阅此论文施引文献
订阅此论文施引文献,注册后可以免费订阅5篇论文的施引文献,订阅后可以查看论文全部施引文献