【作者】 徐正国；

【导师】 张世英；

【作者基本信息】 天津大学 ， 技术经济与管理， 2004， 硕士

【摘要】 金融市场高频/超高频时间序列的分析与建模是金融计量学一个全新的研究领域。高频时间序列通常是指以小时、分钟甚至秒为频率所采集的金融类数据;而超高频时间序列是记录金融市场每笔交易的数据。本论文主要研究了金融市场高频时间序列和超高频时间序列的特性、建模以及应用问题。本文的主要创新工作如下: 1、通过对上证指数和深圳成份指数的高频时间序列的实证研究发现:与低频时间序列不同,金融市场高频时间序列往往具有高的偏度和峰度、具有负的一阶自相关性、具有稳定的“日历效应”现象; 2、利用弹性傅立叶形式回归(FFF)技术对上海股市高频时间序列的“日历效应”进行了定量研究,发现与日本股市不同,上海股市的波动呈现日内单“U”型走势。接着,在弹性傅立叶回归(FFF)的基础上对上海股市冬夏两季的波动水平和模式的异同进行了检验。最后,在使用弹性傅立叶形式(FFF)回归技术对“日历效应”滤波的基础上,对上证综指的高频时间序列建立了长记忆 SV 模型,发现与低频时间序列不同,高频时间序列往往具有低的波动持续性; 3、主要对基于高频时间序列的“已实现”波动进行了深入地研究。针对国际文献中“已实现”波动的计算中测量误差过大的问题,本文对它的计算方法进行了修正,提出调整“已实现”波动率。而且在综合考虑微观结构误差和测量误差的基础上,定义了最优抽样频率。接着,实证研究了上海股市的“已实现”波动和调整“已实现”波动的特性,并针对调整“已实现”波动的长记忆性和“杠杠”效应建立了 ARFIMAX 模型。然后,通过设定一些标准比较了基于调整“已实现”波动的 ARFIMAX 模型,GARCH 模型以及 SV 模型的预测能力,发现基于调整“已实现”波动的 ARFIMAX 模型给出最好的预测。最后,把一维情形的“已实现”波动率扩展到了多维情况的“已实现”协方差阵,对“已实现”协方差阵建立了 FIVAR 模型; 4、主要研究了超高频时间序列的建模问题,通过对上海股市的指数和个股的超高频时间序列建立 ACD 模型和 UHF-GARCH 模型,研究了上海股市的微观结构; 5、提出了用小波神经网络(WNN)来定量研究高频金融时间序列“日历效应”,通过比较发现 WNN 是比弹性傅立叶形式(FFF)回归技术更具优势的方法。 最后总结全文,最后展望了该领域今后的研究趋势。 本文的研究内容是国家自然科学基金项目“多变量时间序列波动持续性及其在金融系统中的应用研究”(70171001)的组成部分。更多还原

【Abstract】 High-frequency and ultra-high-frequency time series’ analysis and modeling is a new research field in financial econometrics. High-frequency time series is referred to financial data which is sampled with interval of one hour, one minute even one second; And ultra-high-frequency time series is tick-by-tick data. The paper studies high-frequency / ultra-high-frequency time series’ analysis and modeling. The main work and innovations of the dissertation include: 1、Through empirical research of high-frequency time series of Shanghai Composite Index and Shenzhen Component Index, the paper discovers that high-frequency time series usually has the characteristic of high Skewness and Kurtosis, negative first order autocorrelation, calendar effects. 2、The paper studies calendar effects of Shanghai Stock Market quantitatively with Flexible Fourier Form regression based on high-frequency time series, and find out its volatility is a single “U” shape, which is different from volatility of Japanese Stock Market. Then the chapter tests winter and summer volatility. At last, uses Flexible Fourier Form Regression to fit the calendar effects, and constructs long-memory SV model for Shanghai Composite Index, and find high-frequency time series usually has low volatility persistence. 3、The paper studies realized volatility, then puts forward a more efficient approach which is the adjusted realized volatility. And the paper defines optimal frequency based on microstructure error and measure error. This paper studies the characteristics of the adjusted realized volatility of Shanghai Stock Market, and estimates ARFIMAX model. Then through a variety of criterions the paper studies prediction ability of adjusted realized volatility, GARCH model and SV model. At last, extends realized volatility based on high-frequency data to realized covariance matrix based on multivariate high-frequency data. Then the chapter studies the characteristics of the realized covariance matrix of Shanghai Composite Index and Shenzhen Component Index, and constructs FIVAR model to its long memory characteristic. 4、The paper studies the modeling of ultra-high-frequency time series, then through constructing the ACD model and UHF-GARCH model of index and single stock data in Shanghai Stock Market, the chapter researches its market microstructure. 5、The paper proposes application of Wavelet Neural Network in high-frequency time series calendar effects’ study. At last, the paper proves that WNN is better than classical FFF regression. At last, the paper reviews the research of this paper and points out its’ research trend. The research is sponsored by National Natural Science Foundation of China: Persistence in Volatility of Multivariate Time Series and Its Applications in Financial System (No. 70171001).更多还原