LOCALIRFS

[trubador]

This thread is about localirfs add-in for VAR-type models. It is a complementary tool for VAR object just as hdecomp and svarpatterns add-ins.

The add-in allows you to produce impulse responses by local projection method introduced by Jordà (2005) and generate associated confidence bands as detailed in Jordà (2009). For further instruction, please see the documentation that comes with the add-in.

There is also an example file available, which outlines use of add-in through Jordà (2009) study.

[dakila]

Good job. Thank you very much, Erin for the very useful, interesting add-in.

[sunrrcn]

Thanks for this add-in. I installed it on my eviews 9. I estimated a VAR model and then used this add-in. In so doing, I followed http://blog.eviews.com/2016/06/impulse- ... ns_43.html
However, the local projections IRFGRPHs are those without any responses reported. Only VAR IRFs are reported. Those individual graphs for LP IRFs are simply graphs with a zero line plus frames, with response-matrix of zeros.

I tried different datasets and different VARs, but ended with the same LP IRFs.

Did I do something wrong? Can somebody help? I really need it for my ongoing project.

Thanks a lot.

[sunrrcn]

Dear Trubador,
Have you tried out this add-in? Any hint on how it works would be of my great appreciation.
I tried the example file as well and got the localirf of all zeros.
Thanks.

[Piergiorgio]

Hi all

it is a great tool indeed, but I'm having troubles with the "order" option.

In theory this should allow you to change the (Cholesky) ordering used to calculate the responses. In practice, it doesnt seem to work: i tried to assign "order" using a row vector, a column vector or a matrix but I always get an error saying that "the number of variables does not match" (so the default ordering is used instead).

The number of variables does match though: for a 4-variable VAR, for instance, I'd set order=[1 4 2 3].
I even tried to drop intercept and constant from the VAR (to avoid confusion between variables and deterministic parts), but that doesn't help.

Any ideas on how to fix this?

[cyp74]

Hi,
This is a great add-in and not sure why EViews is still not considering adding this method to their core product!
I have a question on the results I'm getting though. The impulse responses seem to change when I increase the impulse horizon.
I would expect the impulse responses to be identical up to a horizon regardless of the horizon which is the case with the R package.

Can you please let me know if I'm doing something wrong with the syntax?
Here's what I'm using with a saved VAR object and the respmat1 and repmat2 are very different for the first 12 responses.
var02.localirfs(horizon=12,charts,respmat="respmat1")
var02.localirfs(horizon=24,charts,respmat="respmat2")

Thanks
Troy

[asalgadotorres1982]

Hi!

Actually I am using localirfs to estimate several models of investment vs several drivers. I need to save the matrices of waldtest for each estimation. Has someone idea about the option of localirfs to save waldtest matrices?

Thanks a lot!!!!

Alfredo