【作者】 申敏；

【导师】 江孝感；

【作者基本信息】 东南大学 ， 管理科学与工程， 2005， 硕士

【摘要】 波动持续性是广泛存在于金融时间序列的一类普遍现象,波动持续性建模方法是从动态的角度研究风险变化的一种有效方法。随着世界各国资本市场开放程度的加强,各国的资本市场的联系越来越紧密,相关性呈上升趋势,因此研究时间序列的波动持续性和不同时间序列间波动的协同持续性对投资者的风险管理具有重要意义。由于分形理论能准确地描述经济行为本身的非线性结构特点,因此有必要将分形方法引入波动持续和协同持续的研究中。本文在总结目前国际流行的条件异方差的模型基础上,指出了各模型的特点和局限性。进一步介绍了体现长记忆性的分形市场理论。在单整GARCH模型持续性及检验基础上重点介绍了FIGARCH模型,给出了FIGARCH的长记忆持续性检验方法,并对已有方法作了改进,同时将谱似然估计方法引入模型的参数估计中。为研究多变量FIGARCH,本文将协同持续概念进行推广,提出了向量FIGARCH的协同持续性,并给出向量FIGARCH协同持续估计和检验的谱方法以及向量FIGARCH协同持续性的性质。此外,本文将普通协整的概念推广到存在异方差及方差持续的协整,同时考察了相应的协同持续关系。最后将本文所提出的理论方法,应用到实际金融时间序列中去,考察了沪深股市的波动持续性和协同持续性。通过实证分析,证实了沪深股市价格向量存在具有异方差的协整关系,同时存在持续性和协同持续关系,但协整组合不能消除持续性。实证结果还表明沪、深两股市收益率序列均具有明显的分形特征,波动表现出明显的FIGARCH特性,且两股市波动的长期记忆程度相同,并存在分数协同持续关系。更多还原

【Abstract】 Volatility persistence is found in many financial time series, the method of modeling volatility persistence is an effective instrument from the point of dynamic. With the opening of the capital markets of the world ,the relation among capital markets in different countries is more and more closer ,and the relativity is on rising trend, accordingly ,the research on the persistence and co-persistence among multivariate time series is of great importance to risk management of investors. Because fractal theory can accruately describe the nonlinear dynamic of economic behavior , it is necessary to introduce fractal theory into research of persistence and co-persistence of volatility.In this dissertation, the writer concluds the popular conditional heteroskedasticity models in the world, points out the characters and limits of these methods.. Based on fractal market theory which represents long memory, the writer makes an improvement on existed method of testing long memory persistence and presents spectral-likelihood estimation method of the FIGARCH model. In order to research multivariate FIGARCH ,the writer extends the concept of co-persistence ,brings forward the method of estimating univariate and multivariate FIGARCH, and the spectral method of testing co-persistence, which is the core of this dissertation. Furthermore , the writer extends common concept of cointegeration to one with conditional heteroskedasticity. At last ,the writer tests the model and method discussed above with date of Shanghai and Shenzhen stock markets to show the effectivenes.更多还原