摘要
准确识别跨境资本流动影响因素,前瞻性地防范化解资本异常流动风险具有重要政策意义。本文借助XGBoost-SHAP可解释机器学习方法对2008—2023年中国跨境资本流动关键预警因子进行识别,并进一步从机器学习视角出发拓展“在险资本流动”的宏观研究新范式,通过建立“深度在险资本流动”框架实现对跨境资本异常流动风险的监测预警。本文合成的四类跨境资本流动关键预警因子分别从不同侧面对中国跨境资本流动的阶段性、不稳定性变动特征进行了捕捉,国际收支状况与外部风险因素是近年来影响中国跨境资本流动的主导因素。此外,中国跨境资本异常流动风险非对称性特征显著,资本流动撤离风险相对激增风险波动更为剧烈,在预警因子震荡时期跨境资本流动的波动性扩张态势通常表现为资本撤离和资本涌入概率空间的双重增加。本文将机器学习方法贯穿研究始末,一定程度突破了传统计量手段在相关研究上的局限,为后续研究开展提供了较为坚实的方法论基础和数据支撑。
It is of great policy significance to accurately identify the influencing factors of cross-border capital flow and proactively prevent and resolve abnormal capital flow risks.By leveraging the XGBoost-SHAP interpretable machine learning method,this paper identifies the key early-warning factors of China's cross-border capital flows from 2008-2023,and further expands the research paradigm of“Capital Flows at Risk”from the perspective of machine learning,and realizes the monitoring and early warning of the risks of abnormal cross-border capital flows by establishing a framework of“Deep Capital Flows at Risk”.The results show that the four types of key early-warning factors of cross-border capital flows capture the characteristics of staginess and instability of China's cross-border capital flows from different aspects,and that the balance of payments and external risk factors have been the dominant factors affecting China's cross-border capital flows in recent years.In addition,the asymmetric characteristics of China's abnormal capital flow risks are significant,with the risk of capital flow withdrawal fluctuating sharply relative to the risk of surges.The volatility of cross-border capital flows expanding during the period of the early warning factor oscillations is usually manifested as a double increase in the probability space of capital withdrawal and inflow.This paper integrates machine learning methods throughout the research process,breaking through the limitations of traditional econometric approaches in relevant studies to a certain extent,and offers a relatively solid methodological foundation and data support for subsequent research.
作者
谢敬轩
杨晨龙
邓创
XIE Jingxuan;YANG Chenlong;DENG Chuang
出处
《经济问题探索》
北大核心
2025年第4期160-175,共16页
Inquiry Into Economic Issues
基金
教育部人文社会科学重点研究基地重大项目“跨周期和逆周期结合下的金融安全维护研究”(22JJD790066),项目负责人:邓创
国家自然科学基金面上项目“不确定性冲击的分类识别、传导机制及其对中国经济金融稳定的影响研究”(72473051),项目负责人:邓创。
关键词
跨境资本流动
异常流动风险
风险预警
机器学习
Cross-border capital flows
Abnormal capital flow risk
Risk warning
Machine learning
作者简介
谢敬轩,河南师范大学商学院讲师,经济学博士,研究方向:跨境资本流动与金融稳定;通讯作者:杨晨龙,吉林大学商学与管理学院博士研究生,研究方向:宏观经济计量分析;邓创,吉林大学商学与管理学院教授、博士生导师,研究方向:宏观经济计量分析。
