证券市场导报Issue(8)：49-58,10.
基于年报文本主题因子的企业债务违约风险预测
摘要
Abstract
Debt default risk is the result of the combined effect of subjective and objective factors including corporate financial condition,industry cycles,and internal governance.Traditional debt default risk prediction models mainly utilize financial indicators including capital structure,profitability,and operating efficiency,failing to fully reflect the impact of subjective factors such as managerial project planning,internal control,and risk response.Using the text of Management Discussion and Analysis(MD&A)in annual reports,this paper takes Shanghai and Shenzhen A-share listed companies as samples,identifies debt default events based on litigation and arbitration data,extracts 31 text topic factors using the LDA topic model,and employs machine learning models such as XGBoost to predict corporate debt default risk.The study finds that after adding text topic factors into financial indicators,the prediction performance of corporate debt default model improves significantly.The accuracy,AUC value,F1-score,precision,and recall of the XGBoost model increase by 4.1,2.4,4.1,4.7,and 3.5 percentage points,respectively.Further analysis indicates that topic factors related to industry characteristics,project investment,and internal control have relatively large marginal contributions to the model's prediction results,helping the model identify sources of operating pressure,risk transmission paths,and management response logic,thereby enhancing the explanatory power and forward-looking capability of debt default prediction.In addition,text topic factors demonstrate good reusability;after changing machine learning models,using datasets with different time windows,and conducting grouped tests by enterprise size,text topic factors still significantly improve the prediction effect of debt default.This paper provides a new tool for monitoring corporate debt default risk and maintaining financial stability and security.关键词
债务违约预测/管理层讨论与分析/文本主题因子/机器学习/风险预警Key words
debt default prediction/Management Discussion and Analysis(MD&A)/text topic factors/machine learning/risk warning分类
管理科学引用本文复制引用
曹馨予,盛志鸿,李守斐,盛积良..基于年报文本主题因子的企业债务违约风险预测[J].证券市场导报,2026,(8):49-58,10.基金项目
国家自然科学基金面上项目"连续时间金融框架下基金经理基准化激励合同的外部性研究"(批准号:72571120)、江西省自然科学基金重点项目"期权型报酬结构下机构动态投资组合策略研究"(批准号:20232ACB201006)、国家社科基金重大项目"中国粮食安全的统计测度、预警与对策研究"(批准号:23&ZD120) （批准号:72571120）
