小微企业信用评级模型及比较研究  

Comparative study of credit rating models for small and micro enterprises

作　　者：肖斌卿[1] 杨旸[2] 余哲 沈才胜[1,4] Xiao Binqing Yang Yang Yu Zhe Shen Caisheng(School of Engineering and Management, Nanjing University, Nanjing 210093, China School of Business, Nanjing University, Nanjing 210093, China Zhengzhou Commodity Exchange, Zhengzhou 450008, China Zijin Rural Commercial Bank, Nanjing 210019, China)

机构地区：[1]南京大学工程管理学院,江苏南京210093 [2]南京大学商学院,江苏南京210093 [3]郑州商品交易所,河南郑州450008 [4]紫金农商银行,江苏南京210019

出　　处：《系统工程学报》2016年第6期798-807,830,共11页Journal of Systems Engineering

基　　金：国家自然科学基金重点资助项目(70932003);国家自然科学基金资助项目(71271109;71201074;70901037;71271-110;71501131);教育部科技创新工程重大项目培育资金资助项目(708044);教育部人文社会科学研究青年资助项目(13YJC790174)

摘　　要：在调查和文献基础上,进行信用风险来源识别、评级指标分类和评级方法选择,构建商业银行内部信用评级模型,以期在授信审批环节提高信用风险管理水平.基于某商业银行2008—2013年小微企业实际信贷数据,运用线性判别分析、二项逻辑回归和十种基于不同学习算法的BP神经网络模型构建内部信用评级模型,并在评级指标体系中加入宏观经济变量,使度量风险的稳健性进一步得到提升.最后通过四种方法对不同模型的结果和评级有效性进行了对比分析,认为基于Levenbery-Marquardt学习算法的NN10模型具有最优的评级有效性.On the basis of investigation and literature research, conducting risk source identification, rating indicators classification and rating methods assessment, the paper constructs commercial bank＇s internal credit rating models to improve the credit risk management in the credit approval procedures. Based on the credit data of small and micro enterprises in a commercial hank from 2008 to 2013, using the linear discriminant analysis, logistic regression and 10 types of BP neural network relying on different learning algorithms, internal credit rating models are constructed with macroeconomic variables, which may further improve the robustness of risk measurement. Finally, results and rating effectiveness of different models are analyzed and compared, and show that the NN10 model based on Levenbery-Marquardt learning algorithm performs optimal rating effectiveness.

关 键 词：信用评级 小微企业 模型比较 学习算法 

分 类 号：F832.59[经济管理—金融学]