Acceptance Rate

[Daniel Bendel]

Is it a problem, when the acceptance ratio in a bayesian estimation is below the commly recommended value of about 1/3?
As a relatively low acceptance rate makes it more likely that the MCMC travels through the whole domain, a low value of the acceptance rate shouldn't be a problem. Am I right?

[jpfeifer]

See http://www.dynare.org/phpBB3/viewtopic.php?f=1&t=5307 and the reference therein. Short answer: if you have enough draws, a low acceptance rate is no problem.

[Daniel Bendel]

Thanks again for your answer!!!

Just two other questions:
1. Is it possible to make dynare's search for steady state values faster? I already give dynare very very good starting values. But I was wondering if I should write a steady state file. Where can I find an example for such a file? The example file of dynare was not very helpful for me. Perhaps you know a better example.
2. Where can I find a step by step documentation of what dynare is doing during bayesian estimation? (This documentation should explain such encrypted messages as "Now I am climbing the hill". This would be great for a better understanding.) Do you know any?

[jpfeifer]

1. If you can solve for your steady state analytically, always use a steady state file or the steady_state_model-block. In 4.4 you should also find examples for the latter. Note also that in 4.4. the NK_baseline_steady_state.m has been cleaned up.
2. What you are referring to is not Bayesian estimation, but mode_compute=6. The Bayesian estimation follows e.g. An/Schorfheide(2007). For mode_compute=6, see http://www.dynare.org/DynareWiki/MonteCarloOptimization