Hi, dynare experts,
I've got a question on the use of marginal likelihood. Seems like it's very common in the literature to use the marginal likelihood to compare model fit, and to decide which model would be the best one. My question is that, when I want to compare two models, when is it appropriate to use marginal likelihood to decide one is 'better' than the other?
To be more specific, I'm trying to estimate two models: the first one is one with determinacy; the second one is the one which falls into the indeterminacy region, so I use the method by Farmer, et.al. (2014) to capture (loosely) the expectation error and include the sunspot shock. So basically, compared to the first model, the second one has an additional shock as well as more parameters to estimate. So is it appropriate to use two marginal likelihoods provided in Dynare to conclude which one is better?
Thanks very much!