摘要
针对巴塞尔新资本协议对银行使用内部评级法的要求,提出了一种新的验证模式对信用风险评估中经常使用的线性判别模型l、ogit模型、probit模型和神经网络模型的预测力进行对比验证。在研究中,采用了ROC曲线对各模型进行全截断点预测力分析,同时,运用自抽样法(Bootstrap)随机抽取子样本进行模型预测力检验,以消除可能存在的样本依赖问题,并且基于上市公司的财务数据进行了实证分析。结果发现,logit模型的预测力较高且比较稳定,优于其他模型。
According to the requirement of the new Basle Accorde that validating on IRB approaches used in banks,this paper presents a methdology to benchmarking the performance of credit risk models. In this method, we introduce ROC curve to evaluate ability of predicting defaults of models firstly, and then use bootstrap method to choose subsample randomly that used in the procedure of validation in order to remove the dependence between models and samples. At last,based on the financial reports of China listed companies,we make an empirical research on linear discriminate model,logit,prohit and artificial neutral network model. The result shows that logit have the best performance among these four modes.
出处
《山西财经大学学报》
CSSCI
2007年第2期86-92,共7页
Journal of Shanxi University of Finance and Economics
