Problems with Bayesian Estimation？

Estimation
1

Dear Professor Pfeifer,

First of all, Happy New Year !

I’ve got some bad results when doing the bayesian estimation. Would you please give some advise ?

I’ve already done the identification analysis and confirmed that all parameters are identified before doing the estimation. However, after the estimation, I found that MCMC univariate diagnostics of some parameters do not converage in the end (see in the picture) though the mode check plots are all in good shape. What can I do to improve this?
Any advise will be appreciated. Thank you !
1 2 3

1 Like
2

Your phrasing suggests you may be confusing a few things:

  1. Identification suggests model moments are sensitive to changes in the parameters, and hence their best fit values can plausibly be obtained via estimation.

  2. Mode finding is an attempt to provide the estimation routine “good” starting values. It may or may not be consistent with the mode of the posterior.

  3. The chain produced by the RWMH algorithm will converge to the posterior. But this is an asymptotic result. If the chain has not yet converged you should take more draws.

There are quite a few posts on the forum about how to do this using a combination of the mode_compute=0, load_mh, and mode_file estimation options. All are well documented in the manual.

1 Like
3

Also, you may want to have a look at the trace_plots of your chain.

4

Thank you Professor Dombeck and Professor Pfeifer !

Your relies are really helpful .