Convergence verification

[Nice]

Hello Jpfeifer,

I am not sure if my estimation has converged for ALL parameters. The univariate diagnostics depict that the red and blue lines are not completely flat but that there is still a little bit of movement. Does this mean that the MH has not converged? If so how do I fix this, as this estimation is based on 5 million replications and took over 100 hours to run. Is there something other than increasing the number of replications that I can try?

Secondly, for the impulse responses the bands around the mean response ( bolded black line) is not smooth, I am using the latest unstable version. I am aware that with the older version this was a bug and a bug fix file was provided on the forum. Do I need to replace that file even in the version that I am using?

Many Thanks!

[jpfeifer]

I would recommend looking at the traceplots of some parameters. If you are using the unstable version, you could also use the use_tarb option available since the end of June 2015. Note that it currently still requires the statistics toolbox of Matlab.

Regarding the IRFs, the bands look OK. But I would still look at the IRFs at the posterior mean after estimation using stoch_simul-

[Nice]

Dear jpfeifer,

Many thanks for your response. By traceplots are you referring to the mode_check plots generated after mode finding?

[jpfeifer]

No, I am referring to the trace_plot command. See Pfeifer (2014): An Introduction to Graphs in Dynare at https://sites.google.com/site/pfeiferecon/dynare