How to access numbers for bootstrap-based confidence bands after VAR, IRF

For technical questions regarding estimation of single equations, systems, VARs, Factor analysis and State Space Models in EViews. General econometric questions and advice should go in the Econometric Discussions forum.

Moderators: EViews Gareth, EViews Moderator

p2021
Posts: 8
Joined: Thu Jul 01, 2021 3:35 pm

How to access numbers for bootstrap-based confidence bands after VAR, IRF

Postby p2021 » Fri Jul 02, 2021 8:12 am

After estimating a VAR and using a certain Cholesky ordering assumption, I’d like to access the numbers for Kilian’s unbiased bootstrap-based standard errors for the IRFs for different cilevels.

To do so, I first estimate a VAR and name it myvar. Then I run, for example:

myvar.impulse(30, m, se=boot, bs=ku, fdb, rep=999, smat=my1, cilevels=0.95) [and click through the Cholesky options]

This produces two matrices, my1 and my1_se.
The matrix my1 contains the expected numbers for the IRF point estimates.
But the matrix my1_se contains zeros only.
This must be a bug...
(If I instead run, for example, myvar.impulse(30, m, se=a, smat=my2, cilevels=0.95), my2_se contains the expected entries, not zeros.)

Given that the above procedure fails, is there currently any other way to access what should have been in my1_se?

Peter

EViews Mirza
Posts: 61
Joined: Sat Apr 22, 2017 8:23 pm

Re: How to access numbers for bootstrap-based confidence bands after VAR, IRF

Postby EViews Mirza » Tue Oct 05, 2021 8:42 am

Hi Peter!

This has now been fixed. It should go out in the next patch.

Now that the all zero-matrix has been fixed, it's important to mention here that the command you're using to retrieve bootstrap CIs would not do the intended operation. In particular, issuing smat=mymat will produce a matrix of bootstrap standard errors and NOT bootstrap CIs, and I believe what you're after are the CIs and not the SEs.

Hence, in the next patch, there will be another set of option keywords:

1) cimat
2) rcimat

The first will produce a matrix of consisting only of CIs in (lower, upper) pairs for each impulse response combination. The second is similar to cimat but will prepend to each (lower, upper) pair the impulse response combination to which it is associated. In fact, the rcimat will work even if you didn't perform a boootstrap, but did obtain the SEs. In the latter case, the (lower, upper) combination will be the equivalent of a lower and upper SE band. In any case, the idea here is that rcimat is be better suited if you wish to produce your own graphs.

As I said, all of this will be going out in the next patch and you'll be able to play with it then.


Return to “Estimation”

Who is online

Users browsing this forum: No registered users and 24 guests