correct interpretation VAR IRFs to 1 st. dev. shocks

[nadja123]

Hi,

I would appreciate feedback on correct data specification to compute VAR impulse response to a 1 standard deviation shock & on correct units to interpret such response.

When recalculated EViews IRFs in Matlab, I noticed that "residual 1 std. deviation" option coefficients are "residual 1 unit" option coefficients multiplied by st. dev. of the 'shocked' variable residual.

I had a discussion with a prof on how to compute IRFs to custom size shocks. We also hit on IRFs to 1 st. dev. shocks. The prof said for such IRFs the underlying data to estimate the VAR should be normalized - each endogenous variable (series) divided by st. dev. However, I do not see any such caution in the EViews helpfile.

I think that under "Cholesky no-/dof adjustment" option EViews computes responses to 1 standard deviation shocks, too, but under triangular ordering. So if I directly want to use such IRFs in my paper, I need to be sure about the interpretation. I run the VAR and IRFs estimation in Eviews using original, non-normalized (not divided by std. dev.) series. Wrt interpretation, I have a macroeconomic context with one of endogenous variables being the GDP. Can I report how much per cent do estimated residual standard deviations do on GDP? Or, shall I eventually multiply the estimated IRFs by a coefficient to obtain responses to 1 GDP-percentage point shock? I must report graphs with IRFs in the paper, so I need to get them in a correct unit.

I hope I explained the problem in a more or less understandable way.
Thanks for any feedback!