Problems with mode_check plots?

Estimation ML/Bayesian estimation
1

Dear Professor Pfeifer,

After my bayesian estimation, I’ve got mode check plots as attached. I think they are in good shape ( would you please take a look and help me confirm this ? thank you !).

Then I try to get posterior distribution by Metropolis-Hasting. I find that when I increase the number of draws from 20000 to 200000, the acceptance rate decrease from 30% to 20%. Is this common and reasonable ? And also, do I need to set smaller value of mh_jscale ?

The other probelm is that the variance decomposition result I got seems a little strange ( e.g. the explanatory power of technology shock to the output is only 4% or so ). Is there any way to improve the result of variance decomposition ? Or the variance decomposition result is determined by the modes so that I can not improve it under these modes ?

Any advice will be appreciated. Thanks again !
mode check plots.rar (18.3 KB)

2
  1. The mode-check plots look indeed good.
  2. The variance decomposition is an immediate function of the estimation results. So there is no way to change it, given your estimates.