基于异质市场假说的中国股市已实现波动率研究
发布时间：2018-11-02 09:37
【摘要】：随着全球经济一体化的发展与中国股票市场运作机制的日趋完善,我国股票市场已成为全球金融市场中一个重要组成部分,在金融市场研究中,波动率是一直为热点课题,学者们对波动率领域的大量研究得出了求解波动率的三种主流方法：一是由Black-Scholes方程求解到的隐含波动率；二是以(?)RCH类和SV类模型为代表的历史波动率；三是基于高频数据研究的已实现波动率,也就是本文的研究对象。随着高频数据可以越来越方便的获得,以前的模型波动率方法已不再适合高频数据研究需要。相比于模型波动率,已实现波动率能够更直接、准确地描述股市波动率的特征。金融研究文献表明,以往的GARCH类和SV类模型的日收益低频数据会损失很多对投资者波动预期很有用的信息,而日内高频数据的可得性则弥补了这一缺陷,高频率的采样数据包含了尽可能全面的信息。另外,已实现波动率方法是一种非参数方法,没有模型波动率方法带来的参数估计问题,更没有因为复杂参数估计带来的“维数灾难”,是近年来比较新颖的波动率研究方法。 对已实现波动率的研究理论中,Muller(1993)等人的异质市场假说的理论,Peters(1994)的分形市场假说理论,Lux and Marchesi(1999)的混合市场理论,市场交易的异质性特征在这三种理论中都有所体现,与同质市场相反,异质市场假说认为,市场波动率与市场活跃程度成正比,即参与者越多,交易行为的多样化会导致波动的正向变化。本文选择基于异质市场假说,在考虑了市场微观结构噪声和跳跃基础上,采用HAR-RV以及相关模型(HAR-RV-GARCH, HAR-RV-J,HAR-RV-CJ)对我国股票市场中的已实现波动率进行了有效的估计,分析研究了短期,中期,长期交易类型对股市已实现波动率的影响。 本文选取从2005年4月8日至2010年4月22日上证综指共计59816个5分钟高频收盘价数据,计算得到1227个已实现波动率数据,2005年1月4日至2011年8月23日深证综指共计77424个5分钟高频收盘价数据,计算获得1613个已实现波动率作为研究数据,基于二次幂变差理论,将波动率中的跳跃成分进行有效的分离,通过HAR-RV-J(?)HAR-RV-CJ模型对短中长期已实现波动率进行了有效的估计,本论文实证表明,对于日已实现波动率,短期投资类型对其影响最大,而中长期交易类型的边际贡献率偏低,但两者相差不大,而三个解释变量系数均显著,表明我国股票市场异质性特征的存在,同时也验证了异质市场假说中的理论,即短期已实现波动率受自身前期值的影响,同时也受中长期交易的影响,而对于长期交易者而言,主要受其自身前期值的影响。这与于小蕾在《基于HAR模型对中国股票市场已实现波动率研究》中的结论类似。 而后对HAR-RV模型进行最小二乘回归后的残差进行分析,发现残差序列具有群聚现象,因此我们检验了误差项的ARCH效应,并添加了GARCH项,通过HAR-RV-GARCH模型进一步研究了三种交易类型对已实现波动率的影响,消除了残差项的ARCH效应,HAR-RV-GARCH模型实证表明,添加GARCH项后的模型与原来的HAR-RV模型几乎没有什么差异。在研究跳跃影响方面,根据二次幂变差理论,将已实现波动率分解为连续样本路径方差和离散跳跃方差两部分,考虑到市场微观结构噪声对已实现波动率的影响,用错列的收益率绝对值乘积(原始的二次幂变差公式是相邻收益率绝对值的乘积)修正了二次幂变差的计算方式,以减小市场微观结构的影响,这样通过修正的Z统计量重新得到考虑市场微观结构噪声影响后的跳跃方差的序列,在显著水平为0.01条件下,由修正的Z统计量检验得到的显著性到达的跳跃共308次,比之前的276次多了32次,捕捉率也由修正前的22.49%增强为25.1%,因此,修正的Z统计量检验的跳跃变差序列更为精确一些,我们根据Huang和Tanchen(2005)的研究内容,先是对HAR-RV-J模型实证,结果表明,跳跃项对已实现波动率具有显著的负影响,通过HAR-RV-J和HAR-RV-CJ模型对波动率的估计结果,对比HAR-RV模型,我们可以知道,考虑跳跃项的HAR-RV-J和HAR-RV-CJ模型有更好的拟合效果和预测性能。 本文创新点即充分考虑市场微观结构噪声的条件下量化了跳跃成分对不同已实现波动率的影响,采用修正的Z统计量有效的检测出显著性跳跃次数,对波动率的捕捉程度有明显的提高,从而更有效的分析影响已实现波动率的各个因子。
[Abstract]:With the development of global economic integration and the perfection of the market operation mechanism of China's stock market, our stock market has become an important part of the global financial market. Scholars have obtained three main methods for solving the fluctuation rate: one is the implicit fluctuation rate obtained by Black-Scholes equation; 2 is (?) The RCH class and SV class model represent the historical fluctuation rate; three are the realized fluctuation rate based on high frequency data research, which is the research object of this paper. As high frequency data can be obtained more and more conveniently, previous model wave rate methods are no longer suitable for high frequency data research. Compared with the model fluctuation rate, the fluctuation rate can be more directly and accurately described. The literature of financial research shows that the daily gains and low frequency data of GARCH and SV models have lost many useful information about investor's fluctuation, while the availability of high frequency data in Japan makes up for this defect, and high frequency sampling data contains as much information as possible. In addition, the method of fluctuation rate is a kind of non-parametric method, there is no parameter estimation problem brought by model fluctuation rate method, and there is no 鈥渄imension disaster鈥，
本文编号：2305654
[Abstract]:With the development of global economic integration and the perfection of the market operation mechanism of China's stock market, our stock market has become an important part of the global financial market. Scholars have obtained three main methods for solving the fluctuation rate: one is the implicit fluctuation rate obtained by Black-Scholes equation; 2 is (?) The RCH class and SV class model represent the historical fluctuation rate; three are the realized fluctuation rate based on high frequency data research, which is the research object of this paper. As high frequency data can be obtained more and more conveniently, previous model wave rate methods are no longer suitable for high frequency data research. Compared with the model fluctuation rate, the fluctuation rate can be more directly and accurately described. The literature of financial research shows that the daily gains and low frequency data of GARCH and SV models have lost many useful information about investor's fluctuation, while the availability of high frequency data in Japan makes up for this defect, and high frequency sampling data contains as much information as possible. In addition, the method of fluctuation rate is a kind of non-parametric method, there is no parameter estimation problem brought by model fluctuation rate method, and there is no 鈥渄imension disaster鈥，
本文编号：2305654
本文链接：https://www.wllwen.com/guanlilunwen/zhqtouz/2305654.html
最近更新
教材专著
