Page 1 of 1

Posterior Mean and Confidence Intervals

PostPosted: Tue May 01, 2012 11:18 am
by sashe86
Hi there,

This is probably very simple, but none the less, I am struggling, so any help is greatly appreciated, as always.

When I run the model, the output in my command window displays the results from the posterior maximization for each of my parameters: It shows the prior mean and standard deviation, plus the mode, the t-stat and the distribution type.

I want this table to also show the posterior mean and the confidence interval results. I have tried including the moments_varendo and the conf_sig commands in the model, but I just can't seem to find the results. I understand they should be stored in the oo_ structure, but I can't find them.

Re: Posterior Mean and Confidence Intervals

PostPosted: Tue May 01, 2012 3:15 pm
by jpfeifer
Which version of Dynare are you using? Could you post the mod-file and the data-file?

Re: Posterior Mean and Confidence Intervals

PostPosted: Tue May 01, 2012 3:22 pm
by sashe86
Hi jpfeifer,

I'm using the latest version - Dynare 4.2.5

The .mod file and data set I'm using is attached.

Thanks for the help.

Re: Posterior Mean and Confidence Intervals

PostPosted: Tue May 01, 2012 7:00 pm
by jpfeifer
You are not actually running a Metropolis-Hastings algorithm (mh_replic=0). Your posterior mode is the value that maximizes the posterior likelihood using a certain optimizer. The displayed t-stats follow from a normality assumption and using the inverse hessian to approximate the standard errors. In that sense, the mean is identical to the mode (due to the normality assumption) and confidence intervals can be easily computed as you know the mean and the standard deviation. If you want to have an actual posterior distribution of the parameters, you have to run a MCMC.

Re: Posterior Mean and Confidence Intervals

PostPosted: Tue May 01, 2012 9:34 pm
by sashe86
Hi jpfeifer,

That's great, thanks for pointing that out. I have amended the mh_replic and re-ran the model and now I can get the posterior mean results.

Thanks again!!

S