基于大样本数据模型的汽车贷款违约预测研究

发布时间：2018-04-23 03:05

本文选题：汽车贷款 + 违约预测　；参考：《管理评论》2017年09期

【摘要】：本文运用国内某知名汽车金融公司2014年12月的47138条客户数据,首先运用ROC曲线检验逐步回归功效,再分别建立二值选择模型和计数模型对贷款客户违约状况进行预测,并运用遗传算法对不平衡样本进行一对一匹配,最终得到预测结果。结果表明现存违约评估体系不够有效,客户基本信息、区位、贷款信息、车型、信用状况、房产、贷款期间冲击事件等均会对违约状况产生相应影响。另外,我们得出匹配后的平衡样本预测准确率仍然很高,Logistic模型最适用于客户是否违约的预测,而负二项模型在违约时长的预测中效果更佳的结论。
[Abstract]:In this paper, using 47138 customer data of a well-known domestic auto finance company in December 2014, the ROC curve is used to test the stepwise regression function first, and then the two value selection model and counting model are established to predict the default status of the loan customers, and the genetic algorithm is used to match the unbalanced sample. Finally, the prediction knot is obtained. The results show that the existing default evaluation system is not effective. The customer basic information, location, loan information, model, credit status, property, and loan shock event will have a corresponding impact on the default situation. In addition, we conclude that the prediction accuracy of the balanced sample after matching is still high. The Logistic model is most applicable to the customer or not. About the prediction, the negative two models have better results in the prediction of default time.

【作者单位】：华中科技大学经济学院;
【基金】：中央高校基本科研业务费(2015AC007) 华中科技大学研究生创新训练项目(2015650011)
【分类号】：F832.479

【相似文献】