基于独立成分分析的多维高频数据波动分析

发布时间：2018-04-16 15:14

本文选题：独立成分分析 + 长记忆性　；参考：《浙江工商大学》2013年硕士论文

【摘要】：在世界经济一体化的背景下,我国金融市场与国际市场之间的联系也越来越紧密,股票市场作为我国金融市场的重要组成部分越来越受到各方的关注。我国股票市场正处于不断发展与完善的阶段,各方参与者必须充分认识到股票市场的风险,同时需具有较强的风险防控意识。资产的风险主要表现在它收益率的波动上,在对资产进行风险管理时往往需要借助对它们收益率波动的相关分析来进行。用于金融资产收益率波动分析的模型主要有GARCH模型和SV模型。多元GARCH模型的提出及发展为多维资产收益率波动的研究提供了一个很好的工具。本论文在对多元GARCH模型进行简单综述的基础上将其引入到高频金融资产收益率的实证研究中,运用DCC-GARCH模型和基于独立成分分析(ICA)的ICA-GARCH模型对白银概念股中的豫光金铅和铜陵有色等8个股票的五分钟的收益率序列进行了估计。实证结果表明这8个概念股之间存在波动相关性,且这种相关性是随着时间改变而改变的,DCC-GARCH模型和ICA-GARCH模型都能很好地对这种相关性进行刻画,而通过比较这两个模型的残差自相关性发现ICA-GARCH模型具有更好的拟合优势,且运行速度更快。同时,我们还将ICA引入到了多维资产的“已实现”协方差矩阵的研究中,构建了具有长记忆性的ICA-ARFIMA模型对多维资产高频收益率序列进行分析。通过ICA处理将多维资产收益率序列转换为几个相互独立的成分,然后分别计算各独立成分的“已实现”波动率并进行相关模型估计,从而达到简化模型参数估计的目的。通过实证发现独立“己实现”波动率和对数独立“已实现”波动率都具有显著的长记忆性,ICA-ARFIMA模型能够很好地对多维资产收益率的“已实现”协方差矩阵进行估计。此外,我们将ICA-ARFIMA模型估计得到的“已实现”波动率引入到了风险价值VaR的计算中。实证结果表明ICA-ARFIMA模型得到的波动率能够很好地对资产收益率的风险进行刻画,比直接利用原收益率序列估计得到的VaR值有更高的估计精度。
[Abstract]:Under the background of world economic integration, the relationship between Chinese financial market and international market is more and more close. As an important part of Chinese financial market, stock market is paid more and more attention by all sides.The stock market of our country is in the stage of continuous development and perfection. All participants must fully realize the risk of the stock market and have a strong awareness of risk prevention and control at the same time.The risk of assets is mainly reflected in the volatility of their return rate. In the risk management of assets, it is often necessary to use the relevant analysis of the volatility of their return rate to carry on the risk management.GARCH model and SV model are the main models used to analyze the volatility of financial asset return.The development of multivariate GARCH model provides a good tool for the study of multi-dimensional asset return volatility.On the basis of a brief review of the multivariate GARCH model, this paper introduces it into the empirical study of the return on high-frequency financial assets.The DCC-GARCH model and the ICA-GARCH model based on independent component analysis (ICA) are used to estimate the five minute yield series of eight stocks including Yuguang gold lead and Tongling nonferrous stock in silver concept stock.The empirical results show that there is a volatility correlation among the eight concept stocks, and this correlation can be described well by both the DCC-GARCH model and the ICA-GARCH model, which change with time.By comparing the residual autocorrelation of the two models, it is found that the ICA-GARCH model has better fitting advantages and faster running speed.At the same time, we introduce ICA into the study of "realized" covariance matrix of multidimensional assets, and construct a long-memory ICA-ARFIMA model to analyze the high-frequency return series of multidimensional assets.The multi-dimensional asset return series is transformed into several independent components by ICA processing, and then the realized volatility of each independent component is calculated and the model estimation is carried out respectively, so as to simplify the parameter estimation of the model.It is found that both independent "self-realized" volatility and logarithmic independent "realized" volatility have significant long-memory properties. ICA-ARFIMA model can estimate the "realized" covariance matrix of multi-dimensional asset return rate.In addition, we introduce the "realized" volatility estimated by ICA-ARFIMA model into the calculation of VaR.The empirical results show that the volatility obtained by the ICA-ARFIMA model can well describe the risk of the return on assets, which is more accurate than the VaR value estimated by using the original return series directly.
【学位授予单位】：浙江工商大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：F224;F832.51

【参考文献】