Hello together,

I am trying to estimate pairwise dynamic correlations between country stock index returns via the ddc mgarch model.

According to Stata manual on DCC Garch, the software uses the approach developed by R. Engle (2002), who suggests:
".... a two-stage procedure with each variable first being modelled separately as a univariate GARCH process. A joint log-likelihood function would then simply be the sum of the two log-likelihoods for the individual GARCH models. In the second stage, the conditional likelihood is maximized with respect to any unknown parameters in the correlation matrix."

Question 1:
In a first step I fitted univariate ARIMA-GARCH processes for each country index return series. Is there any chance to include those specifications in the dcc function?
E.g. an ARIMA(1,0,1)GARCH(1,1) process for series A and an ARIMA(1,0,0)Garch(1,1) process for series B?

Question 2:
If i ignore the univariate specifications and run the following code:

mgarch dcc (fra uk =), arch(1/1) garch(1/1) nolog
predict corr*, corr

and then plot the predicted "corr_fra_uk" variable (see attachement) , obviously there is something wrong in the first few periods. I think it has to do with a misspecification of the initial value. Since Garch always includes past periods variances i am wondering how the values for t=1 are calculated when there is no data available before that.

Array

Thanks in advance!
Jano