基于XGBoost的上市企业财务违约预测研究

公司年度报告中的管理层讨论与分析部分是企业信息披露的重要组成.构建表面情感语调STONE和隐含违约倾向IPD两个文本特征指标对年报中管理层讨论与分析的定性文本数据进行量化，并提出了一种基于XGBoost的上市公司财务违约预测模型，该方法对上市公司财务违约实现了较好的预测效果.根据特征重要性排序对特征与财务违约之间的关系进行挖掘，进一步利用敏感性分析验证了表面情感语调和隐含违约倾向指标的有效性.

XGBoost-based Algorithm for Predicting Financial Default of Listed Companies

The management discussion and analysis(MD&A) in the annual report is an important part of corporate information disclosure. This paper constructs two textual feature indexes, superficial tone (STONE) and implicit propensity for default (IPD) to quantify MD&A text in the annual report, and proposes a financial default prediction method for listed companies based on XGBoost model. This method has achieved a good prediction effect on the financial default of listed companies. According to feature importance ranking, the relationship between features and financial defaults is mined, and the sensitivity analysis is used to verify the effectiveness of STONE and IPD.