ARIMA/AR residual heteroskedasticity


#1

Hi,

I need advice with folowing problem:

I have heteroskedasticity residuals from Arima(1,0,0) according to arch.test. Is this a problem? I need to use residuals&coefficients from this model in order to simulate this AR(1) process n-times. Should I use arima.boot function in order to get correct variance from this model?

Kind Regards
Jan Žemlička


#2

Whether or not violation of the homoscedasticity assumption is a problem depends on your end goal.

If you want to do any kind of inference (not that ARIMA coefficients are especially interesting IMO), then violation of the ARIMA assumptions is a problem and you should use a different model.

GARCH models would be a good next step. I would look into fGarch::garchFit().


#3

Thank you very much!

Kind Regards
Jan Žemlička