In this paper,18 million farmers' petty loan records based on real samples from 2007 to 2015 of a region at northern Jiangsu province were used,combined with the local and national CPI index,GDP index,the first industry production index,agricultural production material price index and other macroeconomic indicators.By using the stepwise enter method and logistic model based on maximum likelihood estimation,we selected the indexes which have significant impact on the probability of farmers'petty loan default,then explained the economic meaning of these indexes and analyzed the robust of our model,and finally put forward some policy suggestions.Our study found: The relationship between the credit level index and the farmers'real default is not significant,which means the internal credit rating process of the local credit institutions before the loan can't predict the farmers' credit risk effectively;The micro indicators like interest rate,gender,marital status,occupation and education have great influence on credit risk;National GDP,national CPI,Jiangsu agricultural production price index,one-year lag of the local first industry GDP and other macro indicators also have predictive effects on credit risk;Logistic model can still maintain a good classification accuracy on imbalanced data sets.%基于江苏北部某地区2007-2015年18万条真实农户大样本小额贷款记录,结合当地及全国CPI、GDP、第一产业产值、地区农业生产资料价格指数等宏观经济指标,采用基于最大似然估计逐步进入法的Logistic模型,筛选对农户违约概率影响较为显著的指标,之后对各指标经济含义进行了解释并对模型进行稳健性分析.研究发现:信用水平指标与农户的真实违约情况关联不显著,意味着当地信贷机构对贷款农户的贷前内部信用评级不能有效地预测农户的信用风险;利率、性别、婚姻状况、职业、教育等微观指标对信用风险有较大影响;全国范围的GDP和CPI、江苏农业生产资料价格指数、滞后一期的当地第一产业GDP等宏观指标也对信用风险具有预测作用;Logistic模型在不平衡数据集上,依然能保持较好的分类精度.
展开▼
