多元GARCH模型与多元SV模型的比较研究

发布时间：2018-04-25 23:17

本文选题：MGARCH + MSV　；参考：《兰州商学院》2013年硕士论文

【摘要】：金融市场的快速一体化，使得不同市场、不同资产间的价格运动迅速扩散，相互影响。金融市场间彼此更加依赖。因此，将一元波动率模型推广到多元波动率模型，研究多个变量之间的波动特性，是件很有必要且有意义的事情。多元波动率模型主要包括多元广义自回归条件异方差(MGARCH)模型与多元随机波动率(MSV)模型,前者将波动率当做过去信息集的确定函数，即波动率是其滞后残差平方观测值和前期条件方差的函数；后者则认为波动率由不可观测的潜在的随机过程所决定。根据条件相关系数是否时变，每类模型又可分为常相关模型与动态相关模型，基于此，本文做了MGARCH模型与MSV模型的比较研究。本文首先对收益率序列阶数如何判定以及参数估计等方面做了分析，以期观察到有用的基本信息。然后重点探讨了MGARCH模型与MSV模型波动率序列在参数估计、模型稳定性的不同之处，并做了基于模型的样本外预测能力比较模型优劣，以期找到更适合的模型来准确描述金融市场时间序列的相关特性。最后将基于MSV模型与MGARCH模型下的多个在险价值(VaR)序列做了比较。实证分析的最终结果表明，MSV类模型描述的波动率序列更具集聚性与记忆性，，更加符合实际情况，但就本文而言，MSV类模型外推预测能力并没有比MGARCH类模型更强。
[Abstract]:The rapid integration of financial markets makes the price movement between different markets and assets spread rapidly and affect each other. Financial markets are more dependent on each other. Therefore, it is necessary and meaningful to extend the univariate volatility model to the multivariate volatility model and to study the volatility characteristics among multiple variables. The multivariate volatility model mainly includes the multivariate generalized autoregressive conditional heteroscedasticity (MGARCH) model and the multivariate random volatility model (MSV). The former regards volatility as the deterministic function of the past information set. That is, volatility is a function of the square observations of lag residuals and pre-conditional variance, while the latter holds that volatility is determined by unobservable potential stochastic processes. According to whether the conditional correlation coefficient is time-varying, each kind of model can be divided into normal correlation model and dynamic correlation model. Based on this, this paper makes a comparative study of MGARCH model and MSV model. In this paper, we first analyze how to judge the order of return sequence and how to estimate the parameters, in order to observe useful basic information. Then, the differences between MGARCH model and MSV model in parameter estimation and model stability are discussed. In order to find a more suitable model to accurately describe the relevant characteristics of financial market time series. Finally, a comparison between MSV model and MGARCH model is made. The final results of empirical analysis show that the volatility series described by MSV model are more agglomeration and memory, more in line with the actual situation, but the extrapolation and prediction ability of MSV model is not stronger than that of MGARCH model in this paper.
【学位授予单位】：兰州商学院
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：F224;F830.91

【参考文献】