Rugarch:garch estimates

Any probability model is unlikely to be the true DGP, but rather an 
approximation to it...which is why White (1982) suggested the use
of the "robust" standard errors since they give asymptotically valid
confidence intervals for the 'approximate-model' parameters, minimizing
the information distance between the true probability measure and the
quasi-likelihood...all of which is available in any standard 
econometrics textbook...as to what to consider, it depends on what
you want to do. There are a host of other tests which are displayed
in the summary which should provide further guidance about which model
to choose (e.g. sign bias) or distribution (gof test).

Note that h/w questions should not be posted to this forum, and
posting the same message more than once is considered rude.

-Alexios

Dear All,
Please, I am using the rugarch package for a univariate garch
estimation. I learnt the optimal parameter estimation is based on MLE
and the Robust standard error is based on QMLE. However, I have problem
with choice of model since the MLE yields significant estimates and the
QMLE yields insignificant estimates for some models but not all.
Actually, I consider sGARCH, EGARCH and TGARCH.
Please, any assistance as to how to proceed further to select the best
models? Should I consider the result based on the MLE only or QMLE.
Hope to hear from you all
regards,
Evelyn