基于LR-RF-XGBoost的债券违约风险预警
Early Warning Research on Bond Default Risk Based on LR-RF-XGBoost
陈湘州 1刘佳2
作者信息
- 1. 湖南科技大学 商学院,湖南 湘潭 411201;湖南省新型工业化研究基地,湖南 湘潭 411201
- 2. 湖南科技大学 商学院,湖南 湘潭 411201
- 折叠
摘要
随着市场经济的迅猛发展,各国的债券市场也相继成长,并趋向于多元化发展.然而,在这一发展过程中,中国的债券违约事件屡见不鲜且愈演愈烈,极大地阻碍了市场活力.以发行企业债券、公司债券、短期融资债券以及中期债券的公司为研究主体,提出LR-RF-XGBoost债券违约预警模型,该模型基于软投票法将逻辑回归(Logistic Regression)、随机森林(Random Forest)、极端梯度提升算法(Extreme Gradient Boosting)相融合,对样本的财务指标及非财务指标数据进行研究.研究结果发现:LR-RF-XGBoost融合模型相比于其他单一预警模型泛化能力更强,准确率高达 95.3%.该方法有利于为投资者以及债券市场监督部门提供可靠的预测信息,帮助企业及早识别风险,为债券市场的健康发展提供保障.
Abstract
With the rapid development of the market economy,the Chinese bond market has also grown and gradually developed into an integral part of the market economy.However,in the course of this development,bond defaults have been common and increasing,greatly hampering market dynamics.This paper takes companies issuing corporate bonds,corporate bonds,short-term financing bonds and medium-term bonds as the research subjects,and proposes the LR-RF-XGBoost bond default warning model,which is based on a soft voting method combining Logistic Regression,Random Forest and Extreme Gradient Boosting algorithm,study of financial indicators as well as non-financial indicators for the sample.Results show that the LR-RF-XGBoost fusion model has a higher generalisation capability than other single warning models,with an accuracy of 95.3%.The method is useful in providing investors and bond market supervisory authorities with reliable predictive information,and can better help companies themselves to identify risks early,providing protection for the healthy development of the bond market.
关键词
债券违约/逻辑回归/随机森林/极端梯度提升Key words
bond defaults/logistic regression/random forest/extreme gradient boosting引用本文复制引用
出版年
2024
NETL
NSTL
维普
万方数据