摘要
文章综合机器学习可解释性方法与深度在险资本流动框架,对中国跨境资本异常流动关键风险因子进行识别,并进行异常流动风险监测,在此基础上运用局部投影模型深入探讨宏观审慎政策工具的占优调控策略。研究表明:中国跨境资本异常流动关键风险因子呈现显著阶段性和非稳态特征,外部风险因素与国际收支状况对资本波动的解释力持续强化;中国跨境资本异常流动风险异质分化特征突出,跨境资本撤离风险相较于激增风险波幅更大且自2015年以来始终保持在较高水平;宏观审慎政策立场的整体收紧能够有效缓解中国跨境资本大幅流入或流出的压力,不同类型宏观审慎政策工具对中国跨境资本异常流动风险的调控效果存在显著差异,“资本类政策工具收紧+信贷类和流动性政策工具保持稳健”是当前占优调控模式。
This paper combines the explainable machine learning method and the framework of Deep Capital Flows at Risk to identify the key risk factors of abnormal cross-border capital flows in China and to monitor the risk of abnormal flows.Based on this,a local projection model is adopted to explore in depth the dominant regulatory strategies of macroprudential policy tools.The research shows that the key risk factors of abnormal cross-border capital flows in China exhibit significant periodic and non-stationary characteristics,and the explanatory power of external risk factors and balance of payments on capital volatility continues to strengthen;the risk of abnormal cross-border capital flows in China is characterized by heterogeneity and differentiation,with the risk of cross-border capital withdrawal being more volatile than the risk of cross-border capital surge and remaining at a high level since 2015;the overall tightening of macroprudential policy stance can effectively alleviate the pressure of significant inflows or outflows of cross-border capital in China,and there are significant differences in the regulatory effects of different types of macroprudential policy tools on the risk of abnormal cross-border capital flows in China with the current dominant regulatory mode being"tightening of capital policy tools+maintenance in stability of credit and liquidity policy tools".
作者
谢敬轩
杨晨龙
邓创
XIE Jingxuan;YANG Chenlong;DENG Chuang
出处
《国际经贸探索》
北大核心
2025年第8期72-87,共16页
International Economics and Trade Research
基金
教育部人文社会科学重点研究基地重大项目(22JJD790066)
国家自然科学基金面上项目(72473051)。
关键词
跨境资本异常流动
撤离风险
激增风险
宏观审慎政策
机器学习
abnormal cross-border capital flows
withdrawal risk
surge risk
macroprudential policy
machine learning
