Variance Decomposition for MONTEZHA.PRC

[bonilla]

I am using the procedure MONTEZHA.PRC which shows how to implement Zha's procedure for overidentified BVAR. I work with a 7 variables model and, after some modifications, it works well, the problem is that I cannot compute variance decomposition in typical way (known by me).

When I add the following lines to the procedure:

errors(impulse) 7 15
#1
#2
#3
#4
#5
#6
#7

The output is one and when I use:

errors(model=VARMODEL,impulse,steps=15)

it changes.

I guess that it is because the procedure computes impulses in a different manner but I don't know how to obtain the "true" variance decomposition results. My code is attached if someone can help me, thanks a lot!! I am debutant in RATS and I need your help! Thanks !

[TomDoan]

The ERRORS instruction needs to be inside the loop, not outside. You want to use the newer MODEL=... rather than the older supplementary card form for ERRORS. The reason you get different answers from the two is that the MODEL form brings the covariance matrix with it, while the supplementary cards don't. Note, however, that neither of those is what you want anyway, because if you don't use a DECOMP option (it looks like you're using version 6), the error decomposition is based upon a Cholesky factorization, not the structural model that you created.

However, you have a much more serious problem. Your 2nd and 3rd equations aren't identified (they are identical except for normalization), and neither are your 4th, 5th, 6th and 7th which are all also effectively identical. You need to work on getting a better formed model.

[bonilla]

Thanks a lot Tom! I am working on that!

[TomDoan]

The importance sampling done in that can often fail to work well. You might want to look instead at the Sims six variable models in the Sims and Zha replications (http://www.estima.com/forum/viewtopic.php?f=8&t=335) which do random walk Metropolis-Hastings instead.