Error with Conditional FEVD


I’m requesting help with an error that is generated when attempting to do the conditional FEVD for my model on Dynare v. 4.5.7. I’m encountering the following errors:

  1. The FEVD is identical at all horizons.
  2. The FEVD is only computed when I enter one period. I enter multiple periods as follows:

estimation(datafile = K2021_DATA_EXP, nograph, nodiagnostic, mode_compute=0, mode_file = K2021_BASE_EXP_mode, mh_replic = 0, mh_nblocks = 2, mh_jscale = 0.3, mh_drop = 0.25, load_mh_file, moments_varendo, conditional_variance_decomposition = [1 2 4]) Y PI_OBS i_OBS;

The following error is generated:

Any advice on how I can rectify these two errors? I’ve attached my mod file below. Please let me know if I can provide any further information.


K2021_BASE_EXP.mod (8.5 KB)

Is the problem still there in Dynare 4.6.4? You did not provide all files needed to run the file.

Apologies. I’ve added the link to the folder here. Hope you can access it; if not I can fix accordingly. The code can be run via the “main” script file. This project began with a replication of another paper. Those authors coded the steady state in an older version of Dynare which is why I stuck to Dynare 4.5.7. I haven’t tried running this in the newer version as it will require me to update the steady state file. If there is no fix to this issue in 4.5.7 then I’ll look to switch over.

Here is a fixed steady state file for Dynare 4.7. There I cannot reproduce issues.
K2021_BASE_EXP_steadystate.m (4.9 KB)

My hunch is that the problem you experienced derives from a bug fixed in Dynare 4.6.2.

Note that you have to rerun the MCMC unless you incorporate Restore backward-compatibility when using load_mh_file (!1944) · Merge requests · Dynare / dynare · GitLab

1 Like

Thank you very much! Running the MCMC now. This was incredibly helpful. I’ll post on this forum again if I encounter any further issues.

A more general question: my experience has been that in the case of complex models such as this one, the analytical solver methods seem to not work when computing the mode. As such, I have relied on the numerical solver: mode_compute = 6. It takes quite a while to compute the mode this way. Any advice on how to speed the process along? Would it be better to try a different numerical solver? If so, which one?

mode_compute=5 often works well. If it does not, it often makes sense to debug the issue instead of going brute force. The mode_check-plots are an important tool for that.

1 Like