摘要
自2014年我国出现首例债券实质违约以来,我国的债券违约数量不断增多,特别是在2018年,债券市场更是频繁爆雷,违约金额达到1205.61亿元,给中国投资者和整个市场带来不良影响.随着我国经济增速的放缓,地方政府隐性担保能力下降,企业资金链紧张,债券的刚性兑付被打破,企业债券违约不断出现.对于市场和投资者来说,准确预测中国债券违约成为众所关注的问题.与以前的研究不同,本文使用机器学习方法中的随机森林模型进行违约预测,并将其预测结果与传统的logit回归模型进行对比.我们的研究样本包括2014年1月1日-2019年8月31日的87家债券违约主体和870家非违约主体,选取政府隐性担保以及样本公司违约滞后两年的财务数据、公司特征变量等作为预测因子,实证结果表明,随机森林模型能够有效地预测中国债券违约,随机森林模型的预测准确率非常高,样本内训练时模型的总体准确率达到99.45%,样本外预测时模型的总体准确率也达到94.84%.无论是样本内训练还是样本外预测,随机森林模型的准确率均高于logit模型.这表明随机森林模型可能更适用于预测企业债券违约问题.
Since the first bond default in China in 2014,the number of bond defaults has been increasing.Especially the year 2018,the amount of default reached120.56 billion yuan,which had a very bad effect on investors and the whole market.With China’s economic growth slowing down,the implicit guarantee ability of local governments declines,the rigid payment is broken,the capital chain of enterprises is tense,and the default of corporate bonds is constantly emerging.For the market and investors,accurately predicting Chinese bond defaults has become a matter of public concern.Different from previous studies,this paper used random forest model which is representative in machine learning methods to predict bond default.And compared the prediction results with the traditional logit model.In this paper,we collected 87 bond default companies and 870 companies which had not defaulted from January 1,2014 to August 31,2019 as samples.We selected lag of two years of implicit government guarantee、financial indexes and corporate identity variables as predictive factors.Empirical results show that random forest model can predict bond default efficiently and has high accuracy.In-sample training,the accuracy of random forest is 99.45%.When on out-sample test,the accuracy of random forest is 94.84%.Whether on insample training or on out-sample test,the accuracy of random forest is higher than logit model.This shows random forest model can be used to predict bond default and it can play a very important role.
作者
戴雅榕
沈艺峰
DAI Yarong;SHEN Yifeng(Siming District of Xiamen City Development and Reform Commission,Xiamen 361005,China;School of Management,Xiamen University,Xiamen 361005,China)
出处
《计量经济学报》
2022年第2期418-440,共23页
China Journal of Econometrics
