When conducting Bayesian estimation, is it advisable to linearize the model beforehand, especially considering that some parameters might consolidate into a single new parameter following linearization since the estimation process assumes a linearized model? If this is indeed the case, what determines which parameters can be estimated within the Bayesian framework and is there a rationale behind it?
You do not need to linearize yourself the model, dynare will do it for you if you provide a stationary non linear model. The estimated parameters are the parameters appearing in the non linear version of the model (the deep parameters) not what you call the “new parameters” in the linearised version of the model. Doing so we can take care of all the non linear restrictions between the “new parameters” in the linearised model.
Often we do not estimate all the deep parameters, some are calibrated (typically, if you know that a parameter will not affect the likelihood or moments).
Best,
Stéphane.