What do you mean with “deviation”? IRFs in Dynare are always relative to a pre-shock concept.
I mean by deviation the difference relative to steady state (C - C_ss).
From your response prof , I infer that in the case where variables in FOC are deviations from state (as is log-linearization of NK models) i.e c = C - C_ss and the case where variable are in logarithmic scale then IRF in these two cases is percentage deviation and this applies for both linear or Non linear models. Is my understanding correct ?
The type of model does not matter. Ultimately, the only question is whether the model variables are recentered or not. With manually linearized models, they are typically recentered to be mean 0 and are measured in (log)-deviations from steady state.
Thanks a million prof.