I am using Dynare 4.2.0 to apply Bayesian Estimation to Smets & Wouters (2003) model. There are two puzzles in my results.
(A) The posterior distributions are very narrow, compared to priors. They are just like spikes. I thought it was due to a too small mh_jscale option, so I adjusted it to even 10. But it didn’t seem to help at all.
(B) The posterior estimates of structural parameters are quite close to both prior means and S&W (2003), but the estimates of standard deviations of the shocks are quite small. They are systematically 10% smaller than those in S&W (2003).
I really appreciate for any hints for these two puzzles I encounter. Thank you!