On Fri, 11 Nov 2022 at 5:28 AM, Agustín Alonso Rodríguez <aalonso@rcumariacristina.com> wrote:

Dear Sirs:

 Dear Augustinian, one way  is the Hendry’s General to specific criteria to select the model.  you retain the lagged variables , which ,usually , have the  p value less than or equal to .05 in your model.

How to refine or simplify a VAR model suppressing the insignificant estimated coefs?

 

Hi, I don't think we currently have a ready-made procedure for this yet:

- Some time ago I wrote some hansl code that takes a VAR, throws out the nominally insignificant regressors, and re-estimates the resulting SUR system (with classical GLS). This basically sounds like what you want. The problem is, I never found the time to package it up properly, and this kind of pre-testing procedure is also frowned upon by some people.

- Fellow gretl warrior Luca and I have been thinking about producing a function package with some basic BVAR stuff (VARs with Bayesian shrinkage), which would be a related thing. But this hasn't progressed yet, either.

- At some point in the future gretl's Lasso implementation (regls addon) might also cover multivariate systems / VARs, but again: not yet.

People to develop more function packages with gretl's econometric scripting language hansl are always welcome, no knowledge of C is required!

cheers

sven