The RATS Software Forum

Hi,

I would Like to know, how could I determine the lag length of a VAR model that contains change in the volatility?

I depict this change by visual inspection, and I don't know if I should test it before determining the lag length? But if I conduct the test in a multivariate form, I have to detremine the lag length first!!

So do you have any guidance or coded Rats' programs?

Thank you in advance.

Ahmed Sahloul

I think most people just pick the lag length based upon the standard VAR.

Thank you for your reply.

However, I read some recent papers that mention that the classical informations criteria are no longer valid valid when there is a change in the volatility,

see for example "LAG LENGTH IDENTIFICATION FOR VAR MODELS WITH NON-CONSTANT VARIANCE", that is why I was asking if you programmed such adaptive information criteria.

Regards.

I think most people just pick the lag length based upon the standard VAR.

The lag length is tangential to the analysis. If you're doing a Lanne-Lutkepohl type break in volatility, then the semi-parametric estimator that is used in the paper you cite is also "wrong". And the only way to "estimate" the lag length is to fit the LL model with each lag length to get the true "AIC" or whatever criterion appropriate for the model. So, in fact, you do have to estimate the model before you know the lag length.