The RATS Software Forum

Hi,
When fitting a bivariate Garch-m model where the mean is estimated like follow:

nonlin(parmset=meanparms) ar1 ar2 b1 b2
frml resid(1) = (y(1)-b1-ar1*y(1){1}-c1*(h(1,1){1}))
frml resid(2) = (y(2)-b2-ar2*y(2){1}-c2*(h(2,2){1}))
compute b1=b2=c1=c2=ar1=ar2=0.05


Is it important to vary the initial value here or should I only vary the initial value for the GARCH equations?

(I'm using maximise because I can't estimate the model I need using the inbuilt functions)

Thanks.

Anja wrote:Hi,
When fitting a bivariate Garch-m model where the mean is estimated like follow:

nonlin(parmset=meanparms) ar1 ar2 b1 b2
frml resid(1) = (y(1)-b1-ar1*y(1){1}-c1*(h(1,1){1}))
frml resid(2) = (y(2)-b2-ar2*y(2){1}-c2*(h(2,2){1}))
compute b1=b2=c1=c2=ar1=ar2=0.05


Is it important to vary the initial value here or should I only vary the initial value for the GARCH equations?

(I'm using maximise because I can't estimate the model I need using the inbuilt functions)

Thanks.

The GARCH estimates tend to be quite insensitive to the initial guesses for the mean function.