Estimation of DCC-GARCH model

[ines_morais]

Dear all,
I am new in RATS and I have to estimate a DCC-GARCH model for my thesis to conclude my master degree. But to do the estimation, I have many problems and I don't know how to do it. Please, someone can help me?
I wanna examine if there is contagion between two stock market and I have already the returns series of each market, but I don't know what to do next?
Could someone explain to me from the beginning? All the steps to follow?

Thanks for all.
I really need help.

[TomDoan]

Could you post what you've done? The GARCHMV.RPF example includes many forms of MV GARCH models, including DCC.

[ines_morais]

Hi Tom!
I only made the estimation of returns using ARMA models, to eliminate correlation in residuals (I made this on Eviews). And now, I want to know what I have to do next. For using DCC-GARCH I will need to use the residuals of ARMA models, right?
Thanks for your help

[TomDoan]

The usual procedure is to estimate the mean model and variance model in one go, rather than as a two-step process. The first-step ARMA estimates will be computed based upon homoscedastic residuals, which you don't have.

If you're trying to analyze cross effects between the series, wouldn't you need something other than ARMA models for the means? A common choice is a VAR(1) model for the mean. In the GARCHMV.RPF example, one of the samples is

Code: Select all

*
* VAR(1) model for the mean, BEKK for the variance
*
system(model=var1)
variables xjpn xfra xsui
lags 1
det constant
end(system)
*
garch(p=1,q=1,model=var1,mv=bekk,pmethod=simplex,piters=10)

For your case, you would want to replace MV=BEKK with MV=DCC, and use your two variables rather than the three in this example.

[ines_morais]

Hello!
Thank's for the help.
I already have the ARMA models for returns and I want to use that residuals in estimation of DCC-GARCH model. I do that, but when I test for univariate ARCH effects, they disappear, bur if I test for @mvarchtest, p-value is always very very small (p=0.0001) what means that is ARCH effects in multivariate terms. Is it possible or the estimation is misspecified?
I really need help and I will appreciate

[TomDoan]

DCC fits a univariate GARCH model to each series, so the fact that it passes univariate diagnostics isn't a surprise. If your univariate ARMA residuals leave cross serial correlation then that could help to force the MVARCHTEST to be significant---a maintained assumption in that test is that the residuals have no serial correlation, both univariate and multivariate. That's why it's not a good idea to use the ARMA residuals rather than incorporating a multivariate mean model into the GARCH estimation.