I've been using J-test for model selection but apparently it's not a good measure when models have different degree of freedoms. To my understanding, Akaike information criterion (AIC) or Bayesian information criterion (BIC) could be used as they consider the degree of freedom as well. I believe either of these two criteria can simply be used in the Indirect Inference method as we have the maximized value of the likelihood function, but how about in the Method of Simulated Moments? Can I just use the negative of the Argmin function or it's more sophisticated than that?
I’m looking forward to hearing your suggestions for model selection (on the use of AIC, BIC, or any other techniques) in Indirect Inference and Method of Simulated Moments.
Hi.
AIC and BIC are Information criteria for model selection. There exist other method maned algebraic model order selection like MEV (minimum eigen value) or SVD (singular value decomposition) based rank test to model selection.
Before you choose one method, at first, a description on the relative variables should have been made, for example, graph or table, at least you could find the tendency, and then you try to test with different models;
maybe you could find good model to fit the data.
It is better to look through all different relative literature and the relative model.
If you have the same data when trying to chose the best model, you may use any suitable GoF method to select the best model, as the degrees of freedom do not change much from one model to another. I would use some statistics of the residual errors, which helps to avoid serial bias as well. (E.g. Durbin-Watson statistics.)
- Badges
- Science topic
- Similar topics
- Engineering
- Modeling