I estimate a Bayesian VAR and get different results with different Matlab versions. While the response is the same, the confidence bands are not. Using Matlab 2010b the results look ok, but using 2016b the responses are outside of the confidence band (which in addition is much tighter).
As an example, I run the VAR with only three variables (in log-differences): oil prices, GDP, and interest rates. And the same happens. I attached the impulse responses to a oil price shock.
I am puzzled that results look ok with Matlab 2010b but not with 2016b (and 2014b as well). What could be the reason?
Thank you very much indeed!
IRFs.pdf (56.6 KB)