基于贝叶斯极端分位数回归的金融风险度量理论及应用研究

The project takes the return data of financial assets, financial institutions and financial market as the research object. Based on the time series description model of the return, the project propose the corresponding Bayesian extreme quantile regression model by utilizing Bayesian statistical inference theory and quantile regression modeling technology. Furthermore, the project constructs Bayesian financial risk measurement theories and methods. The main content of the project is composed of three parts. First, the extreme quantile regression model of Bayesian GARCH and MS-AR/SETAR are constructed to make a estimation of VaR and ES. Second, the Bayesian extreme quantile regression model with one-way dependency relationship is constructed to estimate the CoVaR, Risk Spillover Effect and CoES, what’s more, the structural relationship between financial risk and exogenous factors is studied. Third, the averaging methods of Bayesian model is applied to estimate ES and CoES, then the inconsistency problem of VaR and CoVaR can be solved. Last but no least, the Bayesian VAR-VaR extreme quantile regression model is constructed to reveal the interaction dependence on financial risk and to settle the problem of impulse response and risk transfer..The Bayesian extreme quantile regression model based on the the return data of financial assets, financial institutions and financial market contributes to modeling of financial risk measurement while the model parameters are uncertain. In addition, this project proposes the Bayesian average estimation method of ES and CoES, and solve the inconsistent problem of VaR and CoVaR.

本项目以金融资产/机构/市场的收益率为研究对象，根据收益率的时间序列描述模型，利用贝叶斯统计推断理论和分位数回归建模技术，构建金融资产/机构/市场收益率的贝叶斯极端分位数回归模型，建立贝叶斯金融风险度量理论与方法。构建贝叶斯GARCH、MS-AR/SETAR极端分位数回归模型，进行VaR等金融风险度量；构建具有单向相依关系的贝叶斯极端分位数回归模型，估计CoVaR和风险溢出效应，研究金融风险与外生因素之间的结构关系；并运用贝叶斯模型平均方法估计ES和CoES；构建贝叶斯VAR-VAR极端分位数回归模型，研究金融风险的交互相依关系，脉冲响应函数和风险转染问题。.本项目通过金融资产/机构/市场收益率的贝叶斯极端分位数回归分析，提供金融风险度量建模工具，解决参数不确定性条件下VaR和CoVaR金融风险度量问题，提出ES和CoES的贝叶斯模型平均估计方法，据此解决VaR和CoVaR的非一致性问题。

金融是现代经济的核心，金融市场是整个市场经济体系的动脉，防范化解金融市场风险，确保经济稳定发展，是我国实现经济高质量发展必须跨越的重大关口。金融风险度量是金融风险化解的前提和基本依据。.项目以金融资产/机构/市场的收益率为研究对象，根据收益率的时间序列描述模型，利用贝叶斯统计推断理论和分位数回归建模技术，进行金融风险度量建模与应用研究。项目针对传统均值理论模型的分析工具已难以满足对金融经济变量之间复杂性特征问题，构建了极端分位数边际期望损失模型、分位协整模型、Cross-quantilogram模型，刻画金融市场相依关系。设计了分位数之间格兰杰因果检验的非参数方法，提出了金融时间序列分位因果关系检验方法，实证研究金融风险溢出问题；构建了MS-AR/SETAR极端分位数回归模型和单向相依关系的贝叶斯极端分位数回归模型，提供了金融风险度量建模工具，进行VaR金融风险度量和CoVaR风险溢出效应分析；运用贝叶斯模型平均方法估计ES和CoES，构建了贝叶斯VAR-VAR极端分位数回归模型，研究金融风险的交互相依关系，脉冲响应函数和风险转染问题。.在Konker的分位数回归建模的基础上，通过协变量的分位数估计，构建了金融时间序列分位数-分位数回归模型、贝叶斯分位数-分位数回归模型，拓展了金融时间序列的分位数回归模型范畴；在小波时频变换框架下，根据金融风险与分位数之间的内在联系，以及频率与时间尺度的对应关系，定义了多尺度金融风险涵义，构建了小波极端分位数回归模型，从时域与频域联合分析的视角，实证检验了大宗商品市场与国际金融市场之间的相依关系。.项目的相关研究成果发表在《International Review of Finance Anlaysis》、《North American Journal of Economics and Finance》、《Energy Economics》、《European Journal of Operational Research》、《Economic Modelling》和《财经理论与实践》等学术期刊；其中，SSCI期刊论文15篇，SCIE期刊论文7篇，CSSCI期刊论文7篇；培养了6名硕士研究生、6博士研究生，其中，湖南省优秀博士学位论文1篇。