Dear colleagues,
Is it the expected behavior in Dynare v5.4 that if smoother option is used in Bayesian estimation with mh_replic>0 then oo_.SmoothedVariables structure contains Mean and other substructures.
However, if shock_decomposition command is added after the above-mentioned estimation command, then oo_.SmoothedVariables structure contains the names of endogenous variables; so there is no Mean, HPDinf, HPDsup or other substructures.
Ginters
Yes, that is the intended behavior because you compute the shock_decomposition for only one parameter vector.
Thanks, Johannes.
In my case, shock decomposition is yet another output, and I am still using other oo_ substructures, whose content suddenly changes. Besides removing the distributional information in oo_.SmoothedVariables, also the ordering of variables in oo_.FilteredVariablesKStepAhead is changed, so that I have to modify the code for e.g. plotting the forecasting behavior, depending on whether shock_decomposition is switched on or off. This is a minor issue though, so the standardization of output depends mainly on the developers’ preferences.
Ginters
Thanks for the feedback. That indeed seems like undesired behavior. I opened a ticket. See Discuss the expected output of shock_decomposition (#1897) · Issues · Dynare / dynare · GitLab