Variables included in data of Bayesian estimation

How to decide whether a variable should be incorporated in data of Bayesian estimation. Depending on availability or anything else? I know the rule that we can not have more observed variables than shocks in our models. What else?


This is a very tricky question. It depends a lot on the properties of your model and on the parameters and shocks you want to estimate. You have to make sure that you have the information in the data to estimate the parameters and shocks you are interested in. For instance it would probably be more difficult to estimate the TFP shock, if you do not have data on output and labour. So before launching any estimation you should study the properties of your model, and check, looking at the IRFs, that the shocks you want to estimate have (different) effects on the set of variables you will consider in your sample. In Dynare we have tools for performing identification analysis, this may help in this regard.


I see. Thank you, Professor.
All bests,


A more detailed discussion with some references is in Remark 7 (Which Observables for Estimation?) in Pfeifer(2013): “A Guide to Specifying Observation Equations for the Estimation of DSGE Models”

Thank you, Professor. I will read it.
Best regards,