I and my friend are doing a study involving Bayesian-DSGE. However, we have convergence problem. I present the graphs of the Bayesian distributions of our study as an attachment. We could not make an inference about the reason for the convergence problem from the graphs. We would be very happy if you could help with this.
Did you have a look at the convergence diagnostics? And what do the trace plots show? How many draws did you use and why 6 chains?
Thanks for your quick response.
we use 2 million draws and 6 chains (what is your suggestion regarding draw and chain?).
We add trace plots and diagnostics results as an attachment in the pdf
Estimation.pdf (504.2 KB)
Have a healthy day
The trace plot looks like the chain wandered to a parameter bound. Did you set
prior_trunc=0? And how do the other chains look like?
Thanks for your suggestion. when we set prior_trunc=0 our results come close to convergence. However, posterior distribution graph of some variables made double peak (we tried 100.000 and 2000000 draws). Alternatively, we put some bounds for these variable who have double peak graph for posterior distributions;however, results became worse. If you can give suggestions to us we will be very glad. Sincerely yours.
prior_trunc=0.pdf|attachment (451.0 KB)
Dear Johannes. I hope you are well. I wonder whether you saw my mail regarding results of version (prior_trunc=0). Have a nice day
I was away for a while. You need to check the trace plots for the different chains. My hunch is that one of chains did not converge yet, causing the multimodality.