Dear Prof Johannes,
I often face with failing to compute mode without specifying measurement error. That is the augmenting RBC model with some standard frictions and I do estimate without difficulty if I have at least one measurement error, say investment or similar.
Alternatively, if I add more shocks, hence, I have more shocks than observables then most of the time Dynare cannot find posterior modes though I try all “mode compute” options. However, adding back the measurement error the problem seems to be disappeared.
My question is: how important is measurement error in the Bayesian estimation as implemented in Dynare? Since with or without measurement error may affect parameter estimations how can one defense the choice of measurement error even when the model is rich of shocks? Actually, I am doubting on the estimation of dsge model since it is highly sensitive with user settings such as priors or the case above.
My last question is should we demean one-sided filtered data since I have seen that one-sided cyclical components have non-zero means?
Hope to receive your support.
Thank you and best regards,