A Comparative Study Between Multivariate GARCH Model And Stochastic Volatility Model

The rapid integration of financial markets makes the pricemovements of different markets and assets spread and influence eachother quickly, and financial markets are more dependent on each other. Soit is necessary and useful to extend the single variable volatility model tomultivariate volatility model and then study characteristics of volatilitybetween multiple variables. Multivariate volatility model mainly includemultivariate generalized autoregressive conditional heteroskedasticitymodel and multivariate stochastic volatility model, the former one makesvolatility as deterministic function of past information, in other words,volatility is the function of lagged squared residuals of observations andthe former conditional variance. The latter one think that the volatility isdetermined by unobservable stochastic process. Each model can dividedinto two models, constant conditional correlation model and dynamicconditional correlation model, on the basis that conditional correlationcoefficient whether or not change over time.This article first gives a brief analysis on how to set the order ofreturn rate series and how to estimate the model parameter in order to getsome useful information. This paper mainly discusses the difference inparameter estimation, model stability and prediction based inout-of-sample,expected to find the proper model to precisely describe the characteristics of financial time series.Lastly,we compare some series ofValue at Risk based on MGARCH model and MSV model.The final results about empirical analysis showed that Clustering andmemory characters of volatility series described by MSV model are betterthan MGARCH model,and more consistent with the actual situation.Butfor this article,the predictive ability of MSV model are no better thanMGARCH model based in out-of-sample.