商业银行个人信用风险评价的投影寻踪建模及其实证研究

Individual Credit Risk Evaluation Model for Commercial Banks Based on Projection Pursuit Clustering Technique and Its Positive Research

根据已有的商业银行个人信用风险评价指标体系和数据,应用投影寻踪分类(PPC)技术进行建模,通过改变指标归一化方式前后权重是否互为相反数等性质,以确保PPC建模时求得真正的全局最优解.实证研究表明:对于商业银行个人信用分类问题,PPC模型的识别正确率高于判别分析模型;在评价指标的四个方面中,借贷人与本行关系最重要,其次是其基本情况,而偿债能力和稳定性的影响很小;而且,删除偿债能力和稳定性两个均值差异不存在显著性的指标后建立的PPC模型,不仅不会降低识别正确率,还有利于银行降低采集数据的成本和节约时间,简化流程,对提高PPC模型的实用性等具有重要理论意义和实践价值.

Applying the established individual credit risk evaluation （ICRE） system for commercial banks （CB） and the collected sample data, we established the ICRE model for CB based on the projection pursuit clustering （PPC） technique. By changing the standardization mode for the indexes and judging whether the weights of the indexes are being their opposites, we can conclude that the optimizing process of establishing the PPC model reaches the real global optimum. The positive research shows that the PPC model possesses a higher accuracy than that of linear discriminant analysis for the evaluation and classifica- tion of ICRE. Of the four indexes, the relationship between the borrower and the bank is most important, the next is the basic information of the borrower, and the debt paying ability and the stability of the borrower＇s has minor significance on the indi vidual credit risk. The PPC model was established deleting the two indexes （namely simplified PPC model）. This paper firstly sifted two kinds of indices：the debt paying ability and the stability of the borrower＇s, whose mean value differences are not sig nificant, i. e. , has an equivalent accuracy with the PPC model using all indexes. In this way, the commercial banks can reduce the cost and save the time of collecting data, and simplify the evaluation and classification process. The simplified PPC model isin favor in improving the practicability of the PPC model.