multivariate garch time-trend in correlatio​n equation

[bkoksal]

Hi,

I have winrats 6 and I was wondering how the following task can be accomplished? I'd appreciate if somebody could provide an example.
Thanks for any help.

I would like to calculate 90-days rolling-window correlations between conditional volatilities of daily returns for two stocks by using constant conditional correlation model. That is, by using the returns for days 1-90, I will estimate a bivariate-garch model and calculate the correlation and save it. Then, I will calculate the correlations for days 2-91 and so on. This bivariate process will be done for each pair of more than 2 stocks.

One complication, though is to include time trends in both variance equations and the correlation equation. The purpose of including a time-trend in the correlation equation is to test whether correlations between volatilites of stock pairs are increasing over time. Thank you.

[TomDoan]

First off, a 90 data point window for fitting a bivariate GARCH model sounds too short. We found that there is often a problem with unstable estimates for windows even longer than that due to quiet periods followed by noisy ones or vice versa. Wouldn't a DCC make more sense if you think the correlations may be changing? If you really want to do a non-standard CC model like that, you would need to use a MAXIMIZE set up, as the GARCH instruction won't do that.