Some questions about Bayesian estimation


I have two basic technique questions about Bayesian estimation, and hope I can find assistance here.

  1. I am trying to compute the impulse response function with confidence interval after estimation as most papers presented. I checked the forum, but still get confused. What are the differences between having Bayesian_irf with estimation (which I did, but the figures do not look like what we normally see?) and running stoch_simul after estimation? If we do the latter one, it will run with the estimated mean of parameters, and do we need to specify the shock as we purely run stoch_simul?

  2. I used set_dynare_seed(‘default’) command for estimation routine, but I still got different estimated values from different runs. Do I need to specify in a different way?

I have checked the forum, but am still not clear about these questions. If somehow I missed any post with clear answer, please let me know.

Thank you so much for any answer/clarification/help in advance!


The “bayesian_irf” option computes IRFs for several parameter sets: posterior mean, posterior median, confidence intervals… whereas “stoch_simul” only does the posterior mean. Note that “bayesian_irf” does not display the IRFs by default, you have to retrieve them from oo_ (see the relevant entries in the reference manual).

This is strange. Which version of Dynare are you using? Also note that there is no need to explicitly call “set_dynare_seed”.