The RATS Software Forum

Attached is an SVAR (A model).

How do I write "compute a21=a31=a32=a41=a42=a43=a51=a52=a53=a54=a57=a61=a62=a63=a64=a65=a71=a72=a73=a74=a75=a76=a81=a82=a83=a84=a85=a87=0.0" i.e declaration of initial values because I am getting errors
"## SX33. Expression is Too Complicated. Break into Parts?"

with the present format of coding?

I tried to write it in two lines:
compute a21=a31=a32=a41=a42=a43=a51=a52=a53=a54=a57=a61=a62=a63=a64=a65=a71=0.0
compute a72=a73=a74=a75=a76=a81=a82=a83=a84=a85=a87=0.0
(I dont know if it is correct to write like this)
I am getting very weird impulses, in fact it does not produce impulses for some shocks.

1. Yes. That's the proper way to break that up.
2. You're missing the (Q) on your CALENDAR so it's being processed as annual data. If you correct that, you'll at least get results. However,
3. Your model violates the Rubio-Ramirez-Waggoner-Zha identification rules so it has multiple modes.

Thank you. Its a b model, I corrected that.

I see that getting error bands is a much more involved procedure under SVAR than under VAR. What is the more reasonable way to obtain the confidence bands with SVAR?

Also if I want to weight some of the coefficients in the SVAR, how do I incorporate that into the matrix?

indrani_5 wrote:Thank you. Its a b model, I corrected that.

I see that getting error bands is a much more involved procedure under SVAR than under VAR. What is the more reasonable way to obtain the confidence bands with SVAR?

Yes. There's no purely mechanical process for doing error bands with an SVAR. However, your model (assuming that you turned the original into B form) is probably fairly well-behaved. ("A" models tend to have more problems than "B"). I would probably start with the Independence Chain M-H and see how that goes.

indrani_5 wrote: Also if I want to weight some of the coefficients in the SVAR, how do I incorporate that into the matrix?

I'm not sure what you mean.