基于Block-Bootstrap的银行内部评级系统区分力度量

Measuring the Discriminatory Power of Internal Rating Systems Based on the Block-Bootstrap Approach

内部评级法允许合格银行自行计算其资本要求,评级质量因而就显得至关重要。本文应用ROC曲线及其AUC量度检验评级系统的区分力,并针对多数银行违约数据不足和现有验证均假设违约独立的现实问题,引入Block-Bootstrap方法,在保持样本原有违约相关结构的同时,扩充检验样本规模;然后,通过具体实例计算、比较原样本与Block-Bootstrap方法扩充样本两种情况得出的评级系统ROC曲线和AUC量度值的准确性。

The internal ratings-based approach allows the qualified banks to calculate their capital requirements on their own,so the quality of their rating is crucial. For solving the problems that the majority of banks don't have sufficient default data and the existing verification are all assumed as independent default,this paper brings Block-Bootstrap approach to extend the sample size with maintaining the original default correlation structure. And then,it uses the ROC curve and its AUC to measure the discriminatory power of the internal rating system. Finally,the accuracy of the ROC curve and the AUC measurement of the internal rating system are obtained through an example calculation and comparison of the original samples and the Block-Bootstrap ones.