金融时变高阶矩建模及其风险测度研究:基于收益率分解的方法  

Modeling Financial Time-Varying Higher-Order Moments Based on Return Decomposition with Application to Risk Measurement

作　　者：柯睿 郝斌 谭常春 KE Rui;HAO Bin;TAN Chang-chun(School of Economics,Hefei University of Technology,Hefei 230009,China)

机构地区：[1]合肥工业大学经济学院,安徽合肥230009

出　　处：《数理统计与管理》2024年第1期177-190,共14页Journal of Applied Statistics and Management

基　　金：教育部人文社会科学研究青年基金项目(19YJC790052);中央高校基本科研业务费专项资金项目(JZ2021HGTA0137)。

摘　　要：本文基于收益率分解方法,将收益率表示成符号部分和绝对值部分的乘积形式,通过对符号部分、绝对值部分以及这两部分间的相关性分别进行时变性设定,构建了动态收益率分解(DRD)模型来刻画收益率的高阶矩动态特征。该模型能灵活地设置条件偏度和条件峰度的时变演化方式,还能通过对符号部分和绝对值部分间相关性的时变Copula设定来刻画收益率变动的非线性特征,因而使得该模型对收益率的预测具有优势。本文还利用上证综合指数和深证成份指数的收益率数据对DRD模型进行了实证研究,结果表明:两种股指收益率序列均表现出了显著的高阶矩动态特征,条件偏度和条件峰度存在一定程度的波动聚集特征。相较于其他时变高阶矩模型,DRD模型不仅具有更好的样本内模型拟合效果,而且在样本外的风险价值预测和经济价值评价等方面均表现出一定的优越性。Based on the return decomposition method,this paper proposes a dynamic return decomposition(DRD) model to characterize the dynamic higher-order-moment of returns.This approach decomposes the returns into a product of sign and absolute value components and specifies the time-variability of the sign component,absolute value components and the correlation between these two components,respectively.The DRD model can not only flexibly set the time-varying evolution of conditional skewness and kurtosis,but also depict the nonlinear characteristics of returns by setting the time-varying Copula for the correlation between the sign and absolute components.Therefore,this model has an advantage in return prediction.We also conduct an empirical study of the DRD model using daily return data of the Shanghai Composite Index and the Shenzhen Composite Index.The results show that two return series exhibit significant higher-order moment dynamics,and the evolution of conditional skewness and kurtosis have the characteristics of volatility clustering.Compared with other time-varying higher-order moment models,the DRD model not only has a better in-sample model fitting but also shows some superiority in out-of-sample value-at-risk prediction and economic value evaluation.

关 键 词：时变高阶矩 收益率分解 COPULA 风险测度 

分 类 号：F830.92[经济管理—金融学] O212[理学—概率论与数理统计]