Problem about Bayesian estimation

Dear Dynare users:

Recently one strange phenomenon frequently happened in my model. When using Bayesian method to estimate the parameters, I obtain the corresponding posterior irfs. However, when I use the posterior mode as model parameters directly and go through dynare, it comes out different irfs. I am really confused.

Attachment is the estimation code that may useful to identify the reason. Thank you in advance!

I do not think there is anything strange about this finding.
The IRFs at the posterior mode do not have to be the same as the mean of the IRFs obtained using the bayesian_irf option. For the Bayesian IRFs option, Dynare computes the IRFs for many different sets of parameter values, and then the mean and the relevant percentiles at each IRF horizon are presented. This implies that the mean IRF presented does not necessarily have to correspond to a unique set of parameter values, such as the posterior mean or the posterior mode.


Thank you very much for your useful information. I understand that the two IRFs may have slight difference, but in my model, the difference is somewhere large (5%-10%), is this also normal?

Thank you in advance!


You mean the IRFs are away by 0.05 or 0.1 in the two cases?
I would not call that ‘strange’.

I understand, thank you!