computing error bands for impulse responses of AB-type SVAR

[rantueco]

Hello sir,
I ran a SVAR of AB type with five variables. Everything went well but I could not get any idea about computing the error bands for the impulse responses. I modified the simszha.src codes to accommodate the AB model. But the impulse response function for the first variable lies outside the error bands. Is there any simple procedure to compute the error bands for AB type SVAR?

[TomDoan]

A models (as in the Sims-Zha example), B models and A-B models have different behaviors and respond to different types of samplers. If you could post your program and data set, we could take a look at it.

[rantueco]

Many thanks for your quick response to my query.
As per your requirement, I have attached three files containing codes, data and changelog. In the "changelog" pdf file, the changed portion of the simszha.src is highlighted in yellow colour. In anticipation of your kind help!!

[TomDoan]

Your model is very substantially overparameterized (that is, unidentified). With a five variable system, you can have at most ten free parameters if you are concentrating out the diagonal. You have 11 just in the A part alone. Aside from simple counting rules, your A model has far too little actual structure---your second, third and fourth equations are too similar and it will be difficult to separate them statistically.

[rantueco]

Thank you for pointing out the over--parameterized nature of the model. By the way, this model is not the one I am working on. The model was for illustrative purpose only. I just wanted to know whether the codes (written for the model) are correct for computing error bands of AB type SVAR model.

[TomDoan]

It might work; it might not. Some SVAR's are well-identified statistically and are easily handled by importance sampling or independence chain M-H. Some are not as well- behaved (the Sims six variable model being an example, with the money supply and money demand equations having problems) and are multi-modal, which is harder to handle.

[rantueco]

Thanks for your quick response. I also constructed a model similar to Sims and Zha and applied the same codes as in simszha.src (without any modification except the data). Results were found to be quite confusing.

[TomDoan]

The problem with the Sims six variable model is with the first two equations and in particular with the coefficients at the (1,2) and (2,1) positions. Both of those have bimodal distributions, as the model doesn't have strong enough evidence to determine which is actually money supply and which is money demand.

If you have an actual working model that is producing odd behavior you need to post *that*.

[randal_verbrugge]

I can't speak to the technical questions here. I just want to report that I have two "A" models (so I removed "B" references), and the code seems to run very well. Thanks for posting it.