基于文本挖掘的公司股价崩盘风险预警——来自年报MD&A的经验证据

Risk Warning of Stock Price Collapse Based on Text Mining——Empirical Evidence from Annual Report MD&A

以2010—2021年沪深A股上市公司作为研究样本,根据年报中管理层讨论与分析(MD&A)章节,利用文本挖掘技术提取信息披露特征与文本特征,采用机器学习包括决策树模型、梯度提升树模型、极端梯度提升树与前馈神经网络模型预测公司股价崩盘风险。研究发现,MD&A信息披露特征与文本特征作为重要的非财务信息对预测股价崩盘风险起着至关重要的作用,能够显著提升模型精度,同时利用偏依赖图也验证了管理层捂盘假说对于解释股价崩盘风险的科学性与合理性。应规范年报文本信息披露,加强监督力度,促进资本市场健康发展。

Taking Shanghai and Shenzhen A-share listed companies from 2010 to 2021 as the research samples,the information disclosure features and text features are extracted using text mining technology according to the MD&A chapter of the annual report,and the company's share price collapse risk is predicted using machine learning including Decision Tree Model,Gradient Boosting Decision Tree Model,Extreme Gradient Boosting Tree and Feed-forward Neural Network Model.The research results show that MD&A's information disclosure features and text features as important non-financial information play a vital role in predicting the risk of stock price collapse,and can significantly improve the accuracy of the model.At the same time,the partial dependency graph also verifies the scientific validity and rationality of the management's cover hypothesis in explaining the risk of stock price collapse.The disclosure of information in the annual report should be standardized,and the regulatory authorities should strengthen supervision to promote the healthy development of the capital market.