基于LightGBM和SHAP值的企业信用预警模型和实证分析

Enterprise Credit Early Warning Model and Empirical Analysis Based on LightGBM and SHAP Values

随着信用拓展成为政府实现社会治理现代化的重要手段,对企业的信用评估预警变得尤为重要。以进入浙江省信用联合惩戒“黑名单”的严重失信企业为预测目标,构建行业通用的企业信用预警指标体系,并使用LightGBM模型进行预测,取得了较好的预测效果。实证表明,企业是否严重失信与企业历史履约情况的相关性最大,其次是行业因素以及企业基本因素,区域因素的影响相对较小。最后,对区域因素、行业因素、企业因素三个维度的重要特征进行分析,进一步解释不同特征下企业失信概率的变化情况。

With the development of credit as an important means for the government to realize the modernization of social governance,credit assessment and early warning of enterprises have become particularly important.Taking the seriously untrustworthy enterprises that entered the“blacklist”of Zhejiang Province's joint credit punishment as the prediction target,a common enterprise credit early warning index system is constructed,and the LightGBM model is built for prediction which has achieved accurate results.Empirical evidence shows that the correlation between whether an enterprise is seriously untrustworthy and several factors were different:the enterprise's historical performance was the strongest,followed by industry factors and basic enterprise factors,and regional factors was weakest.Finally,the important characteristics of regional factors,industry factors,and enterprise factors are analyzed to further explain the changes in the probability of enterprise untrustworthiness under different characteristics.