引入WFCM算法能提高信用违约测度模型准确率吗?——以沪深A股制造业上市公司为样本的实证研究  

Is WFCM Algorithm a Good Tool to Improve the Accuracy of the Default Measuring Model?——Empirical Research on Chinese Listed Manufacturing Companies in Shanghai and Shenzhen A Shares

作　　者：熊正德[1] 张帆 熊一鹏[1] 

机构地区：[1]湖南大学工商管理学院,湖南长沙410082

出　　处：《财经理论与实践》2018年第1期147-153,共7页The Theory and Practice of Finance and Economics

基　　金：国家社科基金项目(17BJY002);国家自然科学基金项目(71373072)

摘　　要：选取沪深A股上市的制造业公司财务变量构建信用风险评价体系,在利用因子分析法对其进行维数约简后,采用数据挖掘技术和统计学方法对信用违约概率测度作了有价值的探索。模型包含两个阶段,聚类阶段采用加权模糊C均值聚类(WFCM)算法将样本聚成同质的类,使同簇样本更具代表性;违约测度阶段应用Logistic回归方法分别对不同组样本进行测度。实证结果表明:在Logistic模型中引入WFCM算法能显著提高预测样本的违约概率测度准确率;对于样本总体与ST企业而言,其违约预测准确率比Logistic模型分别提高了10.7%和20%;ROC检验结果也说明WFCM-Logistic模型具有更强适用性。This paper builds a credit risk evaluation model based on financial indicators of Ashare listed manufacturing companies,then the factor analysis is used as a dimension reduction method,and data mining and statistics have been employed to find out key proof eventually in order to elevating the predictive accuracy.Herein,weighting fuzzy c-means clustering(WFCM)algorithm has been used to classify the sample data by aggregating homogeneous samples into clusters,and secondly,logistic regression is introduced to measure the default probability of each sample.The empirical results show that the WFCM algorithm can significantly improve the accuracy of the model to predict the default probability of the prediction sample in the Logistic model.The results show that the WFCM-Logistic model is more suitable for the whole sample and the ST enterprise,and the prediction accuracy of the model is 10.7%and 20% higher than the Logistic model.The ROC test results also show that the WFCM-Logistic model is more applicable.

关 键 词：违约预测 加权模糊C均值聚类 LOGISTIC模型 信用评级 

分 类 号：F064.1[经济管理—政治经济学]