DCC GARCH and unit roots

[hashem]

Dear Tom,

I have used the DCC GARCH model as it is programed in RATS and I had a couplde of queries about the results.
Frstly the DCC model used is supposed (I guess) to be mean reverting with a+b less than one however i get values that are very close to one. I tested for unit roots on the fitted vales of the DCC equation and the null of unit roots can not be rejected. I need to test for a level change around the 2007 financial crises and I didnt know if I should apply this on the fitted values without worrying about unit roots or not.

Secondly if my alpha paramter is insignificant does this mean that I cant rely on this particular model to get the dynamic correlations?

Regards
Hashem

[TomDoan]

I'm a bit confused about what you're testing for unit roots. The DCC parameters govern the evolution of the covariance matrix, not the data itself.

[hashem]

I was referring to the time dynamic conditional correlation series we obatian from the DCC estimation rho(t). I wanted to use it to test for level shift following Chiang et al (2007): Dynamic correlation analysis of financial contagion: Evidence from Asian markets, Journal of International Money and Finance 26 (2007) 1206e1228.

[TomDoan]

It's true that distribution on the test statistics on a mean break will be affected by whether the process has unit roots or not. But what's important is the behavior of the series after you allow for the break. If you have a stationary process with a mean break, it's quite possible that a unit root that doesn't allow for the break will give a false acceptance of unit root behavior. (This is Perron's point---series with breaks tend to look persistent).