Uhlig JME 2005 Sign-Restricted IRF's for VAR

[TomDoan]

Aren't they going to look the same other than scale? 72 virtually identical graphs will look rather uninteresting no matter how you arrange them. You can't combine them on a single graph or they'll just smear together. Maybe just graph a few representatives.

[RK2509]

Hi Tom,

Is this correct?

Code: Select all

dec rect[series] Z(N,1)
do i=1,N
set Z(i,1) = lambda(i,1)*goodresp(t)(k,3)
end do i

Thanks a ton for all your help.
Kind Regards

[RK2509]

I know they'll pretty much look the same. But that would always be the case when we try to analyse the shock on 72 food prices through the latent factor lambda. I would perhaps go with a few graphs as you advised.

[umee]

Would it be incorrect to try to interpret Uhilg's shock in terms of the federal funds rate? In other words, how does shock generated by sign restriction compare to a 50 basis point fed funds shock?

[TomDoan]

I guess you could do that. However, note that the "size" of a shock in the Uhlig procedure is determined by it being an impulse vector (column out of a factorization of the covariance matrix) and not any specific impact on any variable.

[umee]

One follow up, which is not necessarily related to code. What would be the implication of Uhlig remaining agnostic about the price level (GDP deflator) as well?

[TomDoan]

If you're talking about multiple shocks, have you looked at Mountford and Uhlig?

[umee]

Not multiple shocks, less restrictions. Being agnostic about output and the price level.

[TomDoan]

Fewer restrictions? Then you might end up with a smear of very different shocks. There's a fine line between having enough restrictions to actually "identify" the desired behavior and so many that no shocks actually satisfy them.

[umee]

In regards to code you posted for Uhlig 2005, is the behavior of the federal funds rate in the impulse vector consistent with a 50 basis point federal funds rate shock under recursive estimation?

In other words, if I want to compare the results generated by this code to results from the recursive approach. Can I make straight forward comparisons? Would the one standard deviation shock to the impulse vector coincide with a 50 basis point increase in the federal funds rate?

[TomDoan]

Didn't you already ask that?

[umee]

Sort of, but is it not possible to make sure that the federal fund rate change is consistent across both methods?

[TomDoan]

It's a sign restriction, not a size restriction. Although you can generate a sign-restricted impulse vector and then scale it to have a specific impact, that completely alters the probability. For instance, if you have an accepted impulse vector with a .05 impact on the FFR and you scale it up to .50, you're now taking a 10 standard deviation shock which should have probability effectively zero.

[tasnim223]

Dear Tom,

I have bit of confusion in Uhligh1.rpf.

First , if I want to give shock to the real GDP , that is the first variable, do I make the following changes,
*****
if atest(1)<0
compute a=-1.0*a

******

also if I want to want to give it a negative shock , do I change the following things;
*******
if atest(1)<0
compute a=1.0*a
*********

Many thanks
Rosen

[TomDoan]

I'm not sure what you mean by "give a shock to real GDP". That's not how sign restrictions work.