[R-SIG-Finance] Violation of underlying assumptions - ARMA-GARCH

[Lasse Thorst]

I am estimating an ARMA-GARCH model using QMLE
(Quasi-Maxiumum-likelihood-estimation). However, I have some trouble - when
I do test to see if the residuals are uncorrelated - a Weighted Ljung-Box
test - this is rejected. So, it looks like my residuals have serial
correlation.

I am unsure if this is violation of the underlying assumptions - and what
effects it would have on my results.

For example, what effect does this have on my estimates? And the strong
consistency of the estimates? And asymptotic normality of the QMLE?

I am using GARCH-models by Francq & Zakoian - I think it might be in here,
but I am having some trouble understanding it.

Regards,

Note: I've posted this on stats.exchange as well, but haven't had any luck
yet.

	[[alternative HTML version deleted]]