基于高维非线性广义分位数回归的系统性金融风险计量

Financial crisis has raised the revolution for the ideas of systemic risk management. The risk management needs to care about not only the financial system itself, but also the effects from the external factors, such as the real economy and the international environment, in order to investigate the evolution laws of systemic risk. However, the existing systemic risk measures have not provided enough consideration, which has brought serious underestimation phenomena...This project aims to investigate systemic financial risk measurement issues, take account of the complex influence of many factors on the systemic risk and also improve the accuracy and stability of the systemic risk measurement. First, through the generalized quantile regression method, CoGQRisk-type measurements are proposed, with their properties being studied, such as the inclusiveness, etc. Then, the generalized quantile regression model for high dimensional data is established to identify the key factors that influence the systemic risk, and also to provide the risk marginal contribution measurement. Next, the establishment of nonlinear generalized quantile regression model is utilized to describe the nonlinear connectedness between financial institutions, which help to identify systemically important financial institutions (SIFIs). Finally, the generalized quantile vector autoregressive distributed lag model is extended to investigate the stability and risk tolerance of the financial system...This project sets up an unified analysis framework for systemic risk measurement, which not only contains the current mainstream methods, but also puts forward a promising new method theoretically. In practice, the project is able to reveal the mechanism and the fluctuation rule of systemic risk, which offers decision support for systemic financial risk management.

金融危机引发了系统性风险管理理念变革，不仅需要考虑金融系统自身，还要考虑实体经济、国际环境等外围因素影响来探究系统性风险演变规律。已有系统性风险测度方法，对此考虑不足，存在严重低估问题。本项目研究系统性金融风险计量，考虑众多因素对系统性风险的复杂影响，提高系统性风险测度的准确性与稳定性。首先，通过广义分位数回归，给出CoGQR-isk类风险测度方法，研究其包容性等；其次，建立高维数据广义分位数回归模型，识别关键影响因素，给出边际风险贡献测度；再次，建立非线性广义分位数回归模型，刻画金融机构之间非线性关联关系，判别系统重要性金融机构；最后，建立广义分位数向量自回归分布滞后模型，研究金融系统稳定性与风险承载能力。本项目研究在理论上，建立系统性风险测度统一分析框架，不仅包容了现行主流方法，还提出了新测度方法；在实践上，能够有效地揭示系统性金融风险的发生机理与变动规律，为金融风险管理提供决策支持。

金融危机引发了系统性风险管理理念变革，不仅需要考虑金融系统自身，还需要考虑实体经济、国际环境等外围因素的影响。已有系统性风险测度方法，对此考虑不足，存在严重低估问题。.基于此，本项目基于高维非线性广义分位数回归，研究系统性金融风险计量问题，重点研究系统性金融风险测度、系统性金融风险关键影响因素识别、系统性金融风险非线性关联关系刻画与系统性金融风险溢出效应等。.本项目圆满完成了既定目标，取得了系列研究成果。主要创新有：第一，在广义分位数回归框架下，基于小波分析提出W-QR-CoVaR方法，基于混频数据抽样模型构建MIDAS-QR和MIDAS-ER模型，基于联合可导出损失构建JE-MIDAS模型，实证结果表明这些模型能够更加准确并及时地测度系统性金融风险。第二，在高维大规模数据分位数回归框架下，构建带有弹性网约束的Expectile回归模型和带有LASSO惩罚项的分位数回归模型，结果表明该类模型综合考虑多个方面的影响，有助于识别系统性金融风险的关键影响因素。第三，在非线性分位数回归框架下，基于神经网络结构，分别构建了QARNN、QRNN-MIDAS、ERNN和NCARE模型，结果表明这类模型可以有效刻画系统性金融风险非线性关联关系。第四，在多变量分位数回归框架下，基于copula构建藤copula-CAViaR和vine-copula-GARCH-MIDAS模型，基于向量自回归构建QVARDL模型，实证结果表明这类模型准确刻画了系统性金融风险溢出效应。.本项目研究成果主要以学术论文形式，在《管理科学学报》《系统工程理论与实践》《中国管理科学》《Knowledge-Based Systems》《Pacific-Basin Finance Journal》等主流刊物共发表（含录用）标注基金资助论文42篇，其中SCI/SSCI收录26篇。本项目获得省部级社科优秀成果一等奖1项。本项目共培养研究生19名，其中2名博士生和6名硕士生获得国家奖学金，2名硕士生被评选为安徽省优秀毕业生。.本项目研究在理论上，建立起系统性风险测度统一分析框架，开发了新的系统性金融风险管理工具与方法；在实践上，有助于揭示系统性金融风险的发生机理、传染方式与演变规律等，对于制定系统性金融风险防范措施具有决策参考价值。