volatility

[h-godarzi]

hi all
suppose in mean equation for modeling garch model we come to know that ARMA(4,1) is best fit based on SBIC.then we will estimate several garch model from 0 to 5 orders.suppose among estimated garch model according to SBIC garch(1,1) is best fit for variance equation but one of the coefficient in the mean equation for example ar(4) is insignificant.what would we do.are mean equation coefficients important in our estimation or not.thanks

[trubador]

Estimation of the GARCH model is final, and therefore you can drop the insignificant terms from the mean equation if the post-hoc analysis tells you so.

[h-godarzi]

thank you, trubador
in garch(1,1) all variance equation are significant but in mean equation ar(4) is insignificant at 5% level of significant.you mean insted of using ARMA(4,1) I should use MA(1) even if the SBC tells the adequacy of ARMA(4,1)? cant we expand the significance level?
thanks again

[trubador]

Of course, you can prefer to work with a higher significance level (say 10%) if you have a priori knowledge or reason to maintain the ARMA(4,1) structure. Significant MA lag may be an indication of an ill-defined GARCH specification. You can also take into account the in-sample and out-sample forecast performances to evaluate the different models.

[h-godarzi]

hi sir
thanks a lot trubador.your comments are always helpful.
regards
hojat