我国中小银行信用风险基础数据缺陷处理方法研究

Research on Solving Defects in Basic Credit Data of Small and Medium-sized Banks in China

本文结合案例银行实际数据发现我国中小银行信用风险基础数据主要表现为正态分布、F分布和X2分布三种形态，伴有小样本量下的厚尾、平峰或有偏以及非小样本量下的厚尾、尖峰或异常值等多种缺陷。通过使用蒙特卡洛模拟技术对比自举法、线性插补法、均值法和拟合分布法四种简易处理方法的效果，本文得到针对不同类型缺陷，拟合分布法、均值法和自举法分别最优的结论。通过案例银行数据的实证分析，验证了该结论的实践效果。

This paper finds that the basic data in credit risk management in Chinese small and medium-sized banks always follow three kinds of distributions like normal, F and X2, ac- companied by kinds of detects including fat tail, smooth peak and bias with small sample size, and fat tail, shark peak and outliers with normal sample size, according to the analysis on case bank＇s real data. Through comparing the effect of four simple methods such as boot- strap method, linear-interpolation method, mean method and fitting-distribution method by monte carlo simulation, this paper gets a conclusion that fitting-distribution method, mean method and bootstrap method are the most effective methods on dealing with different de- tects respectively. At last, this paper makes an analysis on the case banks＇ real data with the method conclusion, which showed the conclusion＇s practice effects.