CCC GARCH and sample correlation

[arthurs seat]

Hi everyone, I've estimated a Constant Conditional Correlation Garch Specification and am wondering why the restriction placed on this MGARCH (bi-variate) specification is only that the correlation is constant - my question is, why not also make it equivalent to the sample pair-wise correlation?

In my case I get a CCC of around -0.17 compared to a sample pair-wise correlation of -.25, which do I put my trust in?

[trubador]

The assumption of constant conditional correlation itself is not too realistic and forcing it to equal unconditional correlation will make the model too restrictive than it already is. Therefore, in many applications the generalized version of CCC (e.g. DCC model) is used. If you prefer to use CCC model, then there are several tests available to test the null hypothesis of constant correlation. The difference between the estimated constant conditional correlation and the sample pairwise correlation might also indicate that the correlation should be time dependent (if the difference is statistically significant).

[arthurs seat]

cheers trubador, I asked my teach the question before here and didn't get an answer I liked so appreciate the explanation. Also I don't want to be using it as a test for dynamic correlation, we have others for that, but am i right to visualise that the CCC constraints, I suppose, get "stretched" when the variability in the corr is there -- if we see this we should look at more generalised models.