基于非平衡样本的商业银行客户信用风险评估——以A银行为例

Credit Risk Assessment of Commercial Banks Based on Non-Equilibrium Samples: Taking Bank A as an Example

有效识别和防范信用风险是商业银行稳健经营的生命线。从微观层面出发,以A商业银行海量客户信贷数据为例,首先使用SMOTE算法处理非平衡数据,接着使用随机森林的方法对20个相关变量进行重要性评分与筛选,并对重要性变量展开描述分析,最后建立Logistic模型,得到影响客户信贷风险最主要的五个因素,其中合同期限、银行服务年数与银行信贷风险显著负相关,而贷前6个月月均贷方发生额、贷记卡最近6个月平均使用额度、贷款最近6个月平均应还款与银行信贷风险存在显著的正相关,结论为银行授信、风险预警和防范违约风险提供理论参考和实践指导。

Effectively identifying and preventing credit risks is the lifeblood of the commercial banks well-running. This article starts from the micro level and takes a large amount of customer credit data from A commercial bank as an example. First, through the SMOTE algorithm to deal with unbalanced data, and then by the random forest method to score and filter the importance of 20 related variables as well as further analysis of the important variables. Finally, using Logistic model to empirical analysis the most important factors affecting customer credit risk. The conclusion shows that the contract period and service years of the bank are negative correlation to the risk credit, and the average monthly credit amount in the lately six months of the loan, the average credit limit of the credit card in the lately six months, and the average loan repayment in the lately six months are positive correlation to the bank＇s credit risk. This conclusion provides a theoretical reference and practice guidance for bank credit, risk warning, and risk prevention against default.