基于金融高频数据的波动性实证研究

发布时间：2018-07-21 15:01

【摘要】：金融高频数据包含着大量的市场信息，因而对高频数据的研究变得越来越重要，高频数据成为金融领域的研究热点。为了能够更深入的研究市场结构，关于金融高频数据的波动性研究己经成为国内外计量经济学者的焦点问题。本文主要是研究金融高频数据的波动率及其建模分析。首先选取上证综指和深圳成指的一分钟数据进行统计特征分析，验证了我国股市高频收益率序列并不是正态分布，而是呈现出高峰厚尾性，且有着显著的‘旧历效应”。其次，介绍了“已实现”波动率和“已实现”双幂次变差，通过分析得出“己实现”双幂次变差不仅具备“己实现”波动率的所有优点，并且具有稳健性，在一定条件下，比“已实现”波动率更有效，同时，在考虑微观结构噪声的条件下，构造均方误差，通过分析得出“己实现”波动率和“已实现”双幂次变差的最优抽样频率都是5分钟。再次，利用ADF-KPSS联合检验、R/S分析方法，修正的R/S分析方法对对5min上证综指的日收益率和“己实现”双幂次变差的长记忆性进行实证分析，实证结果表明上证综指日收益不具有显著的长记忆性特征，而“己实现”双幂次变差序列存在长记忆性特征。最后，选取上证综指的5min数据进行实证分析，采用跳跃性显著检验方法分离连续性波动和跳跃性波动，并且这两种波动成分具有显著的自相关性，在HA.R-RV模型的基础上，本文分别对这两种波动成分建立LHAR-CV模型和LHAR-LJ模型，实证结果表明LHAR-CV能够捕捉连续性波动的杠杆效应，能较好的拟合和预测连续性波动的走势，并且中长期的波动起主要作用。在LHAR-LJ模型中，前期周跳跃性波动对当期的跳跃性波动的影响最为显著，而前期日跳跃性波动和月跳跃性波动对当前的跳跃性波动的影响不显著，而负收益率对跳跃性波动的杠杆效应不显著。本文图20幅，表22个，，参考文献70篇。
[Abstract]:Financial high-frequency data contains a lot of market information, so the research of high-frequency data becomes more and more important, and high-frequency data becomes a hot research topic in the field of finance. In order to study the market structure more deeply, the volatility of financial high-frequency data has become the focus of econometrics at home and abroad. This paper mainly studies the volatility of financial high-frequency data and its modeling analysis. Firstly, the one-minute data of Shanghai Composite Index and Shenzhen Composite Index are selected to analyze the statistical characteristics. It is verified that the high-frequency yield series of China's stock market is not a normal distribution, but a peak and thick tail, and has a significant 'old calendar effect'. Secondly, the paper introduces "realized" volatility and "realized" double power variation. Through analysis, it is concluded that "self-realized" double power variation not only has all the advantages of "self-realized" volatility, but also has robustness, under certain conditions, It is more effective than "realized" volatility. At the same time, under the condition of microstructure noise, the mean square error is constructed. By analyzing the mean square error, it is concluded that the optimal sampling frequency of both "realized" volatility and "realized" power variation is 5 minutes. Thirdly, using ADF-KPSS combined test / R / S analysis method, the modified R / S analysis method is used to empirically analyze the daily return rate of 5min Shanghai Composite Index and the long memory of "self realized" double power variation. The empirical results show that the daily returns of the Shanghai Composite Index do not have significant long-memory characteristics, but the "self-realization" dipotent variation series have long-memory characteristics. Finally, the 5min data of Shanghai Composite Index are selected for empirical analysis, and the method of jumping significance test is used to separate the continuous volatility from the jump volatility, and these two volatility components have significant autocorrelation, based on the HA.R-RV model. In this paper, LHAR-CV model and LHAR-LJ model are established for these two volatility components respectively. The empirical results show that LHAR-CV can capture the leverage effect of continuous volatility, can better fit and predict the trend of continuous volatility, and the medium and long term volatility plays a major role. In the LHAR-LJ model, the early week jump fluctuation has the most significant effect on the current jump fluctuation, while the previous day jump fluctuation and the monthly jump fluctuation have no significant effect on the current jump fluctuation. However, the leverage effect of negative yield on jumping volatility is not significant. There are 20 figures, 22 tables and 70 references in this paper.
【学位授予单位】：中南大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：F832.51

【相似文献】