posterior distribution

[Brigitte]

Dear Dynare creators ,

first of all, I would like to appreciate that you come up with this amazing program which allows so many awesome things concerning DSGE which would be difficult and time consuming to program them by ourselfs. Moreover, thank you very much for this forum where you instantly give us advice and solution to our problems even if they are completely stupid.

And I have also one problem. I am estimating my model. I used the identification command and the parameters are completely identified. The mode checks seem fine. The estimated parameters are also reasonable. However, the posterior distributions have very small variance. For many parametres, they look like a line, not a distributrion. What could be wrong? Thank you very much for any suggestion.

[jpfeifer]

Do a traceplot to see the Markov chain (See my Introduction to Graphs in Dynare). Does it mix sufficiently?

Also, what does the convergence diagnostics say?

[Brigitte]

Dear professor Pfeifer,

thank you for your response. The traceplots suggest that it did not mix sufficiently even after 2000 000 draws in two blocks. Should I just add more draws and blocks? Thank you very much.

[jpfeifer]

Just adding draws is often not sufficient if it does not work after 2 million draws. Was the acceptance rate in the suggested range 23-30%? If not, increase mh_jscale.
You could also try the unstable version with the use_tarb option. This requires fewer draws, but takes longer.

[Brigitte]

Thank you very much. The acceptance ratio was 26%. I'll try the use_tarb option.