Hi, I am trying to estimate a BVAR model but before displaying the impulse responses functions I receive the following message: ”Some of the VAR models sampled from the posterior distribution

were found to be explosive”. Is there any alternative to solve this problem?

Thank you very much!

Is this only a warning? Can you judge how large the problem is? Rejecting some draws from the posterior is not too problematic as long as it not a big share of the posterior mass.

Otherwise, you might have to adjust your prior in order to assign more mass to the stability region.

Thank you very much. Yes. Indeed is a warning. I have noticed that when I use stationary data (difference of log) there are many ”var models sampled from the posterior distribution that are found to be explosive” but when I use HP-detrended data the number of samples found to be explosive significantly reduces and the irf plots look better.

I hopy you are using a one-sided HP-filter. See Remark 12 (Non-Causal Filters) in sites.google.com/site/pfeiferecon/Pfeifer_2013_Observation_Equations.pdf