BVAR IRFs and Cholesky ordering

[bcfc82]

Hi, I am estimating some simple Litterman-prior BVARs. When I generate the impulse responses, the Cholesky ordering seems not to matter to the output (the other variables always responds with a one period lag, regardless of the ordering). Is this normal behaviour?

[EViews Gareth]

Which version of EViews, and which type of IRF?

[bcfc82]

Eviews 13 and Litterman/ Minnesota prior BVARs. IRFs using Bayesian sampling, but I get the same results using the posterior/ empirical residual covariances.

[EViews Kai]

Yes, impulse response results should be invariant to Cholesky ordering under impulse methods that use a diagonal error covariance matrix.

[bcfc82]

OK, thanks. So the Bayesian IRF methods impose a diagonal cov matrix?

[EViews Gareth]

The Litterman prior does, yes.