EViews.com

Hi everyone,

I would like to inquire about the way asymptotic confidence bands for generalized impulse response functions (VAR) are calculated in Eviews.
I need to understand in more detail in order to analyze my result. For this reason, I am trying to reconstruct the outcome obtained through eviews manually.

On the paper by Warne (2008) available on the following link: http://www.texlips.net/download/general ... sponse.pdf, I found the formula in attachment.

This formula worked perfectly for estimating confidence bands for responses on the first period (simultaneous response), since the first part of the equation is null at t=1.

But on the second period, I have not been able to successfully find the same figures as eviews.
Since every element of the equation is clearly defined on the paper, except the (∂vec(C_h))/∂θ', I suspect that the mistake is related to this term. (I replaced it by the identity matrix Ip on the second period, using the formula for confidence bands in Hamilton 94, chapter 11. Maybe this is wrong)

Could you please help me with this issue?

Thank you.

I've looked in depth at the EViews code and believe that there are no issues (but can be convinced otherwise).

I haven't gone through the paper you reference carefully, but the asymptotic analytic standard errors for the Generalized impulses follow directly from the results for the Cholesky standard errors outlined in, say Lutkepohl.

In the example that you tried to replicate, what is the order of your VAR?
Is it a first-order VAR?