基于文本分析和机器学习的汽车上市公司信用风险识别研究  

Research on Credit Risk Identification of China's Listed Automotive Companies-Based on Text Intonation and Machine Learning Perspectives

作　　者：陈友余 赵嘉仪 邓骞杰 刘纯霞 CHEN Youyu;ZHAO Jiayi;DENG Qianjie;LIU Chunxia(School of Accounting,Hunan University of Finance and Economics,Changsha 410205)

机构地区：[1]湖南财政经济学院会计学院,长沙410205

出　　处：《系统科学与数学》2025年第2期456-469,共14页Journal of Systems Science and Mathematical Sciences

基　　金：国家社会科学基金项目(20BJY006)资助课题。

摘　　要：企业信用风险事件已深刻影响到企业、行业与市场的健康运行.为更全面反映企业信用风险特性,在传统信用风险指标体系的基础上,引入由管理层语调和语调操纵组成的文本披露指标,并选取机器学习方法,对企业信用风险进行识别.研究发现:1)将管理层语调和语调操纵纳入企业信用风险指标体系,均能提升信用风险识别准确率,且两者同时加入时对应的信用风险识别准确率最高;2)文本信息具备信息增量与信息操纵功能,引入语调操纵能更客观地反映中国真实情境;3)通过对现有信用风险识别模型进行优化,发现基于AdaBoost算法、随机森林和支持向量机的组合模型的预测能力最强,能显著提升信用风险识别能力,能为小样本下机器学习方法的应用提供方法指引.该研究能为企业信用风险识别指标体系设计和信用风险识别方法优化提供经验证据和决策支持.Enterprise credit risk events have profoundly affected the healthy operation of companies,industries and markets.In order to more comprehensively reflect the characteristics of enterprise credit risk,we introduce text disclosure indicators consisting of management tone and tone manipulation based on the traditional credit risk indicator system,and select machine learning methods to identify enterprise credit risk.The conclusions of the study are as follows:1)The inclusion of both management tone and tone manipulation in a firm's credit risk indicator system can improve the accuracy of credit risk identification,and the accuracy of credit risk identification is highest when both are included simultaneously;2)Textual information has the functions of information increment and information manipulation,and the introduction of intonation manipulation can more objectively reflect the real situation in China;3)By optimizing the existing credit risk identification models,it is found that the combined model based on AdaBoost algorithm,random forest and support vector machines has the strongest predictive ability,which can significantly improve the credit risk identification ability,and can provide methodological guidelines for the application of machine learning methods under small samples.This study can provide empirical evidence and decision support for the design of corporate credit risk identification index system and the optimization of credit risk identification methods.

关 键 词：信用风险 管理层语调 语调操纵 机器学习 供应链金融 

分 类 号：F426.471[经济管理—产业经济] F832.51