考虑定性指标及误判损失的企业违约判别神经网络模型

识别和度量企业的违约风险是银行风险管理中很重要的一项工作.目前企业违约判别模型离实际应用还具有一定差距,表现在:1)模型所使用的样本基本都是配对模式,不能代表整体样本;2)很少直接引入影响违约的定性指标,如行业,地区和规模;3)没有考虑到误判损失的非对称性.针对上述问题,本文应用前向BP网络针对某国有商业银行的2003年全部有效的短期贷款企业的财务数据,引入了定性指标,采用全样本进行训练,最后确定使误判损失最小的切割点,这样就得到优化的神经网络模型.

Building Default Discriminating NN Model on Firm''''s Short-Term-Loan Data-Considering Qualitative Indexes and Misclassification Loss

To date,using models to predict whether firm's default is still a problem.It shows as follows: a. most models using pair wise pattern;b. lack of qualitative indexes that affect firm's default;c. asymmetric misclassification loss between normal firm and default firm.So,introducing qualitative indexes,using all samples and considering misclassification loss,this paper builds a neural network model on short-term-loan data.Though training,and testing,its performance is good.