Some problems about the bayesian estimation irf

Hello, everyone. I am a beginner learning dynare for DSGE. When I generate the irf by bayesian estimation, I find the gray shade area is too narrow. My shocks are all inverse gamma distribution with mean of 0.5 and inf std.ev. Why the gray shade area is so narrow. How can I enlarge the gray shade area?

Thank you very much for any help!

That means the uncertainty about the parameters is tiny. How do the prior/posterior plots look like?

Hi, Prof. Johannes Pfeifer.
Thank you for your reply.
The distributions of my shocks are:
I try 5000 draws and the graphs of prior and the posterior of the shocks are:

So there is a problem with your MCMC, probably a too low acceptance rate and too few draws.

Yes, the accepting ratio is too low. I try 250000 draws this time and the accepting ratio is around 11%.
image. Can I set a threshold for the accepting ratio? Thanks

But do the plots look better with the higher number of draws?

Maybe no? They are very similar. I don’t know how to tell the improvements.

There is something strange going on. I would need to see the files.

Thank you, Prof. Pfeifer.
Here is my mod file. I have tried 250000 draws and 0.05 mh_jscale. Then the gray shade area of the irf of other shocks is enlarged, but the gray shade area of the monetary policy shock is still narrow. Is that due to my taylor feed back rule?
LIANG_Junjie_Problematic_file20220501.mod (10.4 KB)

Please provide all necessary files. There is no data file.

Yes, Prof. Pfeifer. Here is the data file.
data.xlsx (24.4 KB)

Can you also provide your mode-file?

Excuse me, Prof. Where is the mode-file? Or what is its name?

In Dynare 5.x, it should be in the Output-subfolder as a _mode.mat-file

Is this the mode-file?
LIANG_Junjie20220503_problematic_file_mode.mat (4.1 KB)

You can verify in your mode_check-plots that you did not find a proper mode. Using


finds a much higher mode, but with a problematic Hessian.

Hi Prof. Pfeifer,
Thank you for your kind assistance. Are there any methods I can try to find a proper mode? I want to spend time trying it on my own. :pray:

Loading the last mode-file and rerunning mode-finding with a different optimizer often works.

Thank you, Prof. Pfeifer. :pray: I will try it out.

Hi Prof. Pfeifer,
I try mode_compute=9. And I demean the i and not to observe hc. It works.
Many thanks for your assistance!:pray: