摘要
地方政府隐性债务已经成为系统性金融风险的潜在触发点,为了防范隐性债务风险,文章构建了地方政府隐性债务风险先导预警模型。使用修正的KMV模型测度隐性债务违约概率,依据违约概率设计预警模型输出指标,构建了包含17个指标的输入指标体系,使用多种机器学习方法对预警模型进行训练,选取最优方法构建先导预警模型。实证结果表明:使用随机森林方法可训练出最优的地方政府隐性债务风险先导预警模型,其准确率在91.94%~99.35%,具有较强的精确性和先导预警性。预警模型输入指标体系不仅可以预测隐性债务绝对风险等级,还可以预测隐性债务风险动态变化。进一步分析发现,地方政府隐性债务风险主要来源于隐性债务规模的快速扩张,而非地方政府偿债能力的弱化,其中以城投债、PPP项目以及地方国有企业的相关指标对隐性债务风险的预测能力最强。
The implicit debt of local government has become the potential trigger point of systemic financial risk. In order to prevent the implicit debt risk, this paper constructs the leading early warning model of the implicit debt risk of local government.By using the modified KMV model to measure the default probability of implicit debt, the output index of the early warning model is designed, and the input index system containing 17 indexes is constructed. A variety of machine learning methods are used to train the early warning model, and the best method is selected to construct the leading early warning model. The empirical results are shown as follows: The optimal leading early warning model of implicit debt risk of local government can be trained by random forest method, and its accuracy is between 91.94% and 99.35%, which has strong accuracy and leading early warning. The input index system of the designed early warning model can predict not only the absolute risk grade of implicit debt, but also the dynamic change of implicit debt risk. Further analysis shows that the implicit debt risk of local government mainly comes from the rapid expansion of the implicit debt scale rather than the weakening of the solvency of local government. Relevant indicators of urban investment bonds, PPP projects and local state-owned enterprises have the strongest ability to predict implicit debt risk.
作者
苏振兴
扈文秀
夏元婷
Su Zhenxing;Hu Wenxiu;Xia Yuanting(School of Economics and Man agement,Xi'an University of Technology,Xi'an 710054,China)
出处
《统计与决策》
CSSCI
北大核心
2022年第7期20-25,共6页
Statistics & Decision
基金
国家自然科学基金资助项目(71971169)
陕西省哲学社会科学领军人才特支计划资助项目(2020063005SX)
河南省软科学研究项目(212400410533)。
关键词
隐性债务风险
先导预警
机器学习
随机森林
implicit debt risk
leading early warning
machine learning
random forest
