Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.

# st: Units of measurement in IRFs in structural VAR models

 From jdasmith@gmx.com To statalist@hsphsun2.harvard.edu Subject st: Units of measurement in IRFs in structural VAR models Date Fri, 13 Dec 2013 16:00:37 -0500

```I estimated a structural VAR (SVAR) model and obtained the simple impulse response functions (IRF), the orthogonal impulse response functions (OIRF), and the structural impulse response functions (SIRF).  I have a question about their interpretations.

To fix the idea, assume this is a bivariate VAR model between I and C (both variables are stationary - forget about nonstationarity or cointegration; suppose both are measured in dollars).

Suppose that in period 1, from C to I we find IRF=0.2, OIRF=0.3 (assuming order of C I), SIRF=0.4. My question is, how do we interpret these numbers, especially the SIRF? I know we usually just draw a graph but I need a little more than that.

IRF=0.2, it means that if C increases by 1 dollar, I increases by \$0.2 (but because the shocks are related, this interpretation is questionable - I got that). But for SIRF=0.4, which one of the following statements would be correct (or neither of them)?
(1) If C increases by \$1, I increases by \$0.4.
(2) If C increases by 1 standard deviation, I increases by 0.4 sd.

I guess my question is really about the units of measurement of the SIRF. In EViews it appears that one can choose a one-unit shock (\$1) or a one standard deviation shock. In Stata there is no such option. My question is, which one Stata uses, based on the SIRF reported?