受经济全球化影响，各国金融市场间关联程度进一步加深，市场间风险传染频 繁发生。中国金融市场起步较晚，在市场开放过程中将受到较大风险溢出效应影响。 因此，准确测度其他国家对中国金融市场的风险溢出效应，对于风险投资者和金融 管理者等至关重要。如今，CoVaR成为测度风险溢出效应的重要工具，其关键在 于多元条件联合分布的刻画，copula-GARCH模型为此奠定了基础。但是，传统 copula-GARCH模型存在两方面缺陷：一方面未考虑低频宏观经济变量的影响；另 一方面存在维数灾难问题。对此，本文改进传统copula-GARCH模型，进而给出 CoVaR测度方法，用以检测金融风险溢出效应，主要开展了以下两个方面的研究 工作。 首先，为了更准确地测度“一对一”情景下两市场间风险溢出效应，本文提出 二元copula-GARCH-MIDAS模型。该模型在GARCH的长期波动因子中引入低频 宏观经济变量构成GARCH-MIDAS模型对单个金融市场的边缘分布进行拟合；然 后用copula方法描述“一对一”情景下两市场间的相依结构，实现条件联合分布 建模；最后基于条件联合分布推导CoVaR类风险测度结果。在实证研究中，运用 二元copula-GARCH-MIDAS模型，检测国际原油市场对中国金融市场的风险溢出 效应。实证结果表明，与传统二元copula-GARCH模型相比，本文提出的二元copula- GARCH-MIDAS模型在CoVaR类风险测度方面更优，意味着在GARCH的长期波 动因子中引入低频宏观经济变量能够提高风险溢出度量的准确性。 其次，为了改进二元copula-GARCH-MIDAS模型无法满足多个市场下风险溢 出测度的需要，本文使用vine-copula方法，构建vine-copula-GARCH-MIDAS模 型，能实现“多对一”情景下三个及以上金融市场间风险溢出效应的测度。实证研 究美国、英国、日本三个发达国家股票市场对中国股票市场的风险溢出效应，发现 vine-copula-GARCH-MIDAS模型优于vine-copula-GARCH模型和DCC-GARCH模 型，再次证实在GARCH的长期波动因子中引入宏观经济变量能提高CoVaR类风 险测度的精度；多个股票市场同时陷入危机情景对中国市场存在显著风险溢出效 应，监管者需要同时关注多个市场而非单一市场对中国市场的影响。 本文研究工作在理论上，建立了copula-GARCH-MIDAS模型，通过GARCH- MIDAS能够挖掘混频数据信息，改善边缘分布拟合效果；通过copula技术，能够 刻画相关结构，实现多元条件联合分布建模。在实际应用中，引入低频宏观经济变 量，提高了系统性风险溢出效应测度准确性，为相关决策提供参考。 关键词：金融风险；风险溢出；混频数据；CoVaR；GARCH-MIDAS；Copula II ABSTRACT With the influence of economic globalization, the relationship between various financial markets becomes closer, and the risk contagion among markets occurs frequently. Due to the late start of the Chinese financial market, China will suffer greater risk spillover effects in the process of financial market opening. Therefore, it is very important for venture capitalists and financial managers to accurately measure the risk spillovers of other countries on the Chinese financial market. To date, CoVaR has become an important tool to measure the risk spillovers among financial markets, which depends heavily on the joint distribution modeling. Regarding this, copula-GARCH model provides an important tool. However, the conventional copula-GARCH model has two defects: the neglect of low frequency macroeconomic variables and the problem of dimension disaster. To this end, this disertation improves the conventional copula- GARCH model with mixed frequency data, and proposes the CoVaR measure to detect the financial risk spillovers. The study devotes to the following two aspects. First, in order to accurately measure the risk spillovers between two markets in a “one- to-one” pattern, this disertation proposes a bivariate copula-GARCH-MIDAS model. A GARCH-MIDAS framework with long-run volatility component driven by low frequency macroeconomic fundamentals is applied to fit the marginal distribution of a single market. Then, the copula technique is used to model dependence structure between two markets and derive the joint distribution. Finally, the CoVaR-type risk measures are calculated with the estimated joint distribution. In empirical research, the bivariate copula-GARCH-MIDAS model is applied to estimate risk spillovers from international crude oil market to the Chinese financial market. The empirical results show that bivariate copula-GARCH-MIDAS model outperforms the bivariate copula-GARCH model in terms of CoVaR measure, which means incorporating low-frequency macroeconomic fundamentals in long-run volatility component do help improve the accuracy of measuring risk spillovers. Second, in order to solve the problem that bivariate copula-GARCH-MIDAS is unable to measure risk spillovers among multiple markets, vine-copula technique is used to construct a vine-copula-GARCH-MIDAS model, which enables measure risk spillovers among financial markets in a “multiple-to-one” pattern. Through the empirical study on investigating risk spillovers from multiple developed stock markets to the Chinese stock market, the results show that vine-copula-GARCH-MIDAS model outperforms the vine- III copula-GARCH model and the DCC-GARCH model, which once again confirms incorporating low-frequency macroeconomic fundamentals in long-run volatility component do help improve the accuracy of CoVaR measure. What’s more, there is a significant risk spillovers on the Chinese stock market when multiple stock markets fall into crisis at the same time. Regulators should pay attention to the impact of multiple markets rather than that of a single market on the Chinese market. Theoretically, this disertation constructs a copula-GARCH-MIDAS model, which improves the goodness-of-fit on the marginal distribution. Additionally, copulae is used to model the dependence structure between two markets and derive the joint distribution. Practically, incorporating low-frequency macroeconomic fundamentals improves the accuracy of measuring systematic risk spillovers, which is beneficial for relevant decisions. KEYWORDS: Financial risk; Risk spillovers; Mixed frequency data; CoVaR; GARCH- MIDAS; Copula IV 目 录 第一章 绪论 ......................................................... 1 1.1 研究背景及意义 ....................................................................................... 1 1.1.1 研究背景 ........................................................................................ 1 1.1.2 研究意义 ........................................................................................ 2 1.2 研究内容和方法 ....................................................................................... 3 1.2.1 研究内容 ........................................................................................ 3 1.2.2 研究方法 ........................................................................................ 3 1.3 论文主要创新和结构安排 ....................................................................... 4 1.3.1 主要创新 ........................................................................................ 4 1.3.2 结构安排 ........................................................................................ 4 第二章 国内外研究综述 ............................................... 6 2.1 金融风险溢出计量 ................................................................................... 6 2.2 GARCH-MIDAS理论与方法 ................................................................... 7 2.3 copula理论与方法 .................................................................................... 9 2.4文献述评....................................................................................................11 第三章 基于二元copula-GARCH-MIDAS的两市场间金融风险溢出效应研究 . 12 3.1 模型构建与求解 ..................................................................................... 12 3.1.1 边缘分布拟合：GARCH-MIDAS .............................................. 12 3.1.2 关联结构刻画：copula................................................................ 13 3.1.3 模型估计：copula-GARCH-MIDAS .......................................... 13 3.2 二元copula-GARCH-MIDAS模型的CoVaR度量与返回测试 ......... 14 3.2.1 风险指标 ...................................................................................... 14 3.2.2 计算方法 ...................................................................................... 14 3.2.3 返回测试 .............................................