When I set number of Metropolis-Hasting replications on 800000 and discard the first 200000 of them, multivariate convergence of Brooks-Gelman(1998) test shows the results of interval,
m2 and m3 based on 800000 sample.
I want to know that dynare shows the posterior mean and posterior standard deviation and posterior mode based on 600000 sample or 800000 ?? although we can see 800000 in Brooks-Gelman(1998) multivariate convergence.The other question is the in the section of Posterior Results after model estimation in Dynare, mode report value is prior mode or posterior mode? I think it’s the posterior mode value.
Thank you so much.
We compute the posterior moments by discarding the first draws of the MCMC, the number of discarded draws is controlled by the value of
mh_drop. In your case we consider only the last 600000 draws from the Metropolis.
I don’t think I understand your second question. By default Dynare produces two tables. A first one after the estimation of the posterior mode (maximizing the posterior kernel), in this table we report an estimate of the posterior mode (second column) and the prior mean (first column). A second one after the MCMC, where we report the posterior mean (second column) and the prior mean (first column).
Thank you so much. Your answer was so helpful but my main question was that if Dynare calculates the posterior moments based on the after discard sample, for example in my case this number is 600000 why we see before discard replications number in Brooks-Gelman(1998) graph? for example in my case this number is 800000.
Your answer to my second question was complete and helpful.
I don’t really remember all that stuff, but I suppose it’s because in principle you should choose the number of draws to be discarded, i.e. the value of
mh_drop based on the outcome of the convergence diagnostics.