摘要
首先用Python语言对123家小微企业共约三十七万条交易信息进行分类整理;然后用主成分分析将多种信息归为四类主要因子,并引入定性变量;最后建立多元Logit模型对企业进行由好至差的A、B、C、D四类评级。结果显示:整体评级准确率达到72%;交易信息离评级时点越近,评级的准确率越高;该模型对D级的评级准确率达到100%,说明该方法能有效甄别最差级别企业,为商业银行规避不良贷款发生提供了一种可行性方法。
First, use Python to classify and sort a total of about 370,000 transaction information from 123 small and micro enterprises;then use principal component analysis to classify a variety of infor-mation into four main factors, and introduce qualitative variables;finally, establish Multinomial Logit model to rank companies from good to bad in four categories: A, B, C and D. The result shows: the overall rating accuracy rate reaches 72%;the closer the transaction information is to the rating time point, the higher the accuracy of the rating;the accuracy rate of the D-level rating of this model reaches 100%, it shows that this method can effectively identify the worst-level enterprises, which provides a feasible method for commercial banks to avoid the occurrence of non-performing loans.
出处
《金融》
2022年第6期582-589,共8页
Finance
