中国家庭债务风险测度及其预警研究

Measurement and Early Warning of Household Debt Risk in China

家庭债务风险的预警对于防范系统性金融风险具有重要意义。本文基于2013年、2015年、2017年和2019年CHFS数据,采用因子分析法分别构建了中国城镇和农村家庭债务风险指标,结果显示城镇和农村高风险率分别约为25.06%和12.29%。进一步地,使用基于LightGBM算法的机器学习模型对城镇和农村家庭债务风险分别进行了预警,预警模型的混淆矩阵显示正确率分别为98.30%和98.18%,AUC值分别为0.9725和0.9599。SHAP框架下的机器学习模型可解释性分析显示,家庭债务因素、人口结构以及流动性等因素对城镇和农村家庭债务风险均存在明显的非线性影响。经过更换机器学习参数、机器学习算法以及预警模型的可解释方法,显示本文研究结论具有稳健性。本文为中国家庭债务风险提供了相对客观的测度方法,对家庭债务风险起到预警作用。

The early warning of household debt risk is of great significance to prevent systemic financial risks.Based on CHFS data in 2013,2015,2017 and 2019,this paper uses factor analysis method to construct urban and rural household debt risk indicators in China. The results show that the high risk rate of urban and rural households are about 25.06% and 12.29% respectively. Further,an early warning model based on the LightGBM algorithm is used for urban and rural households’ debt risk respectively,and the confusion matrix of the early warning model shows the correct judgment rates of 98.30% and 98.18%,with the AUC values of 0.9725 and 0.9599,respectively. Explainability analysis of machine learning models under the SHAP framework shows that there are significant nonlinear effects of household debt factors,demographic structure,and mobility on debt risk for both urban and rural households. After replacing the machine learning parameters,machine learning algorithms,and interpretable methods of the early warning model,the results of this paper are shown to be robust. This paper provides a relatively objective measure of household debt risk in China and serves as an early warning for household debt risk.