Bayesian IRFs bug

Hello users!

I’m having a problem that I can’t figure out what it is. I’m working with a DSGE model, I’ve done some tests, run the model in the calibrated way, I’ve estimated it and everything seems to be going well.

Until I went to perform a new test and this time I chose to use the loglinear option in the estimation function, and the Bayesian IRFs were all wrong: all IRFs are practically identical, variables that are stochastic processes are reacting to other stochastic processes, and all these processes have the same IRF with the same confidence intervals.

When I ran the same model in a calibrated version with the loglinear option, it doesn’t give the same problem.

At first, I thought it was a problem with the old dynare version of Linux, which I used to run on, so I tried to run it using the load_mh_file option on Windows and a newer version of dynare and I got the same problem. I tried to force it to run using the load_mh_file option, making few new extractions, but without the loglinear option and then the IRFs were ok.

I don’t know if it’s a dynare bug, or if I’m declaring the model in a wrong way. Does anyone have any idea what could be going on?

Attached below are some files.

Thanks to anyone who can help me.

Thiago Abreu

Thanks for reporting this. A bug fix is at Fix posterior IRF generation with loglinear option (!2083) · Merge requests · Dynare / dynare · GitLab
It will be part of the next stable release.

Thank you, Pfeifer