摘要
将Lasso-logistic模型引入个人信用评估,通过模拟实验发现,逐步回归法倾向于保留一些不重要的变量,而且选出正确模型的概率较低,而Lasso不仅计算更加快捷,可以同时进行变量选择和参数估计,而且能更准确地筛选出重要的变量。以信用卡消费信贷违约数据为例对我国个人信用评估进行实证分析发现,相对于全变量Logistic模型和逐步回归Logistic模型,Lasso-logistic模型更能抓住影响消费信用风险的关键因素,而且预测准确率也更高。
This article applies Lasso-logistic model to the individual credit risk prediction, the simulation suggests that stepwise method is apt to keep nonzero var- iable and the probability to select correct model is low, while Lasso has the advan- tages over other methods in computation, variable selection, parameter estimation simultaneously and grasping key factors. In additional, this article gives an empiri- cal analysis of individual credit evaluation using credit cards data from a famous bank in China, finding that Lasso-logistic model relative to full logistic model and stepwise logistic model can select key factors and the predict accuracy is much higher.
出处
《数量经济技术经济研究》
CSSCI
北大核心
2014年第2期125-136,共12页
Journal of Quantitative & Technological Economics
基金
国家自然科学基金青年项目(71201139
71303200)
国家社科基金青年项目(13CTJ001)
教育部人文社科项目(12YJC790263)的资助
