Dear all,
I have just estimated a model by giving the command:
estimation(datafile = newdata, xls_sheet=Feuil1,xls_range = A1:E35, mh_replic=2000,mode_compute=6,mh_nblocks=2, mh_drop=0.45,bayesian_irf, irf=20,mh_jscale=0.65) dataYdifU dataFXdifU dataPicU dataRdU;

Dynare gives me all the figures (Priors, MCMC Univariate, Multivariate diagnostics, Priors and posteriors, Orthogonalized shocks, smoothed shocks, historical and smoothed variables)

I was expecting Dynare to produce Tabular results of posterior distributions with mode, standard deviation, mean and confidence interval.

Did I omit an option in the estimation command above? If yes, are those Tabular results stored somewhere in Dynare files and can they be recovered (I am not posting the code… yet because it took Dynare 14 hours and 38mns to run it).

I thought the estimated parameters obtained after a MCMC estimation would show before the figures. Thanks.
Now, I have three more issues

I have read somewhere that it is better to use the posterior mode for stoch_simul after a MCMC estimation. The posterior mode of the estimated parameters are not in the log-file though the mode is shown as a vertical green line on each figure produced by Dynare. Where can I find them?

There are two blocks of parameters in the log-file because of the option “mh nblocks = 2” in the “estimation” command. I guess I can use either for any purpose.

I was expecting to see standard-deviations and confidence intervals for posterior parameters’ estimates. Where are they?

A final note: I wish I could share my figures with you to have some hint on the acceptability of the results because some of them look strange to me.

I was referring to the log file in the main folder with the same name as the mod-file, not the mh_history.log

The parameters after mode-finding are stored in the _mode.mat file. There is no theoretical reason to prefer one over the other. See my last post at [Variance Decomposition)

You should not use anything from the log-file.

In oo_ there should be all results saved in fields like oo_.posterior_std, oo_.posterior_hpdinf, and oo_.posterior_hpdsup

For sharing figures, put everything into a zip-file and then attach them to your post.

I take note for the suitability of the posterior mean or the posterior mode for simulation purpose.

After a MCMC estimation, Matlab creates in its Work folder a subfolder named like the .mod file which contains 3 sub-subfolders (metropolis, Output, priors). I guess it is where I am supposed to find my model’s estimated results.

I have checked but none of those sub-subfolders includes the mode.mat file or files with fields like oo.posterior_std, oo_.posterior_hpdinf, and oo_.posterior_hpdsup. I wonder if an option is missing in my estimation command or if my software is outdated (I am using Dynare 4.3.3 with Matlab 7 R14) or perhaps if all estimated parameters are not identified.

I have enclosed the ‘output’ subfolder for some hint.

The files I am talking about are located in the main folder, not a subfolder. oo_ is the name of a variable stored in the workspace after running Dynare and also saved into the _results.mat file in the main folder.

Thank you very much for your invaluable help. The Tabular results were sitting in Matlab Workspace. So, I have saved and zipped the Workspace after estimation. That file is attached to the post (mylogfile.zip).

Going back to the graphs I told you about (included in output.zip file), I am concerned with the types of posterior graphs that can be found for example in PriorsAndPosteriors2.eps panel. In that panel, SE_sleveps is bi-modal (the second mode far from the first), SE_geps is bi-modal (the second mode close to the first) and, the parameter “ha” posterior distribution is a vertical line. All this is disturbing because at the same time the multivariate diagnostic seems OK (Figure 17: Multivariate Diagnostic).

What should I do ? Increase the number of draws? Change the prior distribution? Keep the prior distribution and tighten priors by stating a minimum and maximum priors for those parameters? Decrease the number of calibrated (fixed) parameters in the model?

I would be very grateful if you could give me your take.