Is historical shock decomposition based on posterior mode or posterior mean in Dynare?

Estimation ML/Bayesian estimation
1

Dear Johannes,
First thank you very much for your help and guidance. I am grateful.
I have a question:
Is historical shock decomposition based on posterior mode or posterior mean in Dynare?
Because some literature is based on posterior mode or posterior mean?
Thank you again and look forward to hearing from you?
Best wishes,
Jesse

2

By default dynare uses the posterior mean for the shock decomposition. See the manual, p. 89 under Options the first command.

If you specify

parameter_set = posterior_mode

dynare performs a shock decomposition at the mode. This is helpful if you don’t want to fully estimate the model. Therefore you can just perform a shock decomposition right after the mode finding procedure.

And if you use

parameter_set = calibration

dynare just uses the calibrated parameters specified in the mod file to perform a shock decomposition. In this case you would have to provide a data file and the observables via varobs.

1 Like
3

Thank you very much!

4

Actually, @Robert’s answer is not fully correct. If you run the MCMC, it will be the posterior mean. If you only ran mode-finding (or ML), it will be the mode.

1 Like
5

I have a doubt related to this question: after an MCMC, for unobserved variables the shock decomposition command plots the mean smoothed variables and their historical shock decomposition.

Is the mean smoothed variable (the black line) used for the shock decomposition the same one would get with the calib_smoother command, with a model calibrated at the posterior means obtained with the MCMC?

Thank you!

6

Thank you very much!

7

@ShortRun No, due to Jensen’s Inequality. See Shock decomposition - shocks names display

8

Thank you.

I am probably missing something.

Citing your reply in the link you posted:

But if I “calibrate” the model with the posterior means obtained through the MCMC and run the calib_smoother command I am not drawing from the different parameter vectors from estimation, I am only using the mean values of the parameters. To be clear, I could even remove the Metropolis folder with the chains. What I should get is the “smoothed object at the mean parameter vector” which I guess is the same object used for the shock decomposition (the black line in the shock demposition).

Where am I wrong?