[R] vetor autoregressions and BVARs

[Paul Gilbert]

Nirav Mehta wrote:

> I have not been able to find any programs for running vector 
> autoregressions with R. I am interested in running Bayesian VARs and 
> also running VARs that run all combinations of variables in the vector.

If you mean linear combinations of the variables, you cannot "run" these 
because there are an infinite number. However, there are several 
techniques related to PCA that find the the "best" linear combination, 
according to some criteria. Much of the early development in this area 
was done by Akaike. Some of these are implemented in the dse bundle of 
packages and there are examples in the Users' Guide. My favorite is the 
bft method (but I am biased).  Avoid the technique promoted by Aoki as 
his main theorem has an error and his approach does not work. (He 
claimed it should only fail occasionally, but it fails often in my 
experience.)

If you really mean to restrict yourself to VAR models there is a 
problem. These techniques are typically based on the state space 
representation. You can find an equivalent ARMA representation, but not 
necessarily an equivalent VAR representation. Even though you may start 
with a VAR model you can end up with no (finite) VAR representation.

As for the Bayesian part, I have not worked on that.

Paul Gilbert

> Is anyone currently developing this?
> 
> -Nirav Mehta
> 
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://www.stat.math.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide! 
> http://www.R-project.org/posting-guide.html
>