信用风险模型的贝叶斯改进研究

A Study on Bayesian Improvement on Credit Risk Models

基于小样本数据和外部先验信息,本文运用贝叶斯(Bayes)估计量来改进信用风险模型的违约预测力。同时,运用中国上市公司财务数据,分别对贝叶斯估计量和标准Logit估计量进行了模拟估计,并通过统计量AUC值和布莱尔分数(Brier Score)对其预测精度进行比较。结果表明,贝叶斯估计量具有更高的预测精度和稳定性。

Based on the prior information of external credit risk models and the internal models built for predicting default by banks, this paper suggests two Bayesian estimators to improve the predictability of credit risk models. Therefore, we make a bootstrap simulation to compare the accuracy of Bayesian estimators and straight Logit estimators by AUC and Brier score. The result shows Bayesian estimators have a higher accuracy than straight Logit estimator.