基于支持向量机的集团信用风险预警研究  

Research of group credit risk early-warning model based on support vector machine

作　　者：冯一宁[1] 邵元海[1] 陈静[1] 王来生[1] 邓乃扬[1] 

机构地区：[1]中国农业大学理学院,北京100083

出　　处：《中国农业大学学报》2008年第2期94-98,共5页Journal of China Agricultural University

基　　金：国家自然科学基金资助项目(106310706057315810771213)

摘　　要：针对我国现有信用风险预警模型普遍以单一企业为样本,无法反映集团信用风险状况的问题,采用支持向量机算法,建立了企业集团信用风险预警指标体系,利用全国大额贷款和授信企业集团客户样本数据进行训练,提出了基于支持向量机的集团信用风险预警模型。理论分析及预警数值试验结果表明:基于支持向量机的集团信用风险预警模型与传统的基于逻辑回归算法预警模型相比,具有更好的泛化能力;在相同预警敏感水平下,采用前者预警的假负率为16.67%,与后者的23.45%相比具有更高的预警精度。基于支持向量机的集团信用风险预警模型可较好的应用于企业集团信用风险预警领域中。In order to forecast the group credit risk that currently couldn＇t be reflected in the credit risk early-warning system based on the sample of a single enterprise, an index system for a group credit risk early-warning was set up based on support vector machine （SVM） and trained using the national loan and confer group dataset. Finally a group credit risk early-warning system based on SVM was established. The results of theoretical analysis and numerical experiment indicated that the early-warning system based on SVM could reduce the false negative rate from 23.45% to 16.67% compared with the early-warning system based on logistic regression under the same level of sensitivity, which means this model has an obvious advantage of generalization and a higher early-warning precision.

关 键 词：支持向量机 巴塞尔新资本协议 企业集团 信用风险预警 逻辑回归 

分 类 号：F830.9[经济管理—金融学]