Why there is no GMM/SMM functions in Dynare

Hi everyone,

I need to match my model to a set data from 2004-2015 (12 years). My supervisor thinks the data is too short to use the Bayesian estimation method. Instead, he suggests GMM/SMM.

However, after I looked through, I found no GMM/SMM functions in Dynare. Why Dynare team gives it up? Does it imply that GMM/SMM is strictly infeiror to Bays method?

And is the quarterly data from 2004 to 2015 (48 periods) really too short to use Bays method?

1 Like


The choice is not a matter of absolute ranking between full information, full information + prior, and partial information approaches. It depends a lot on what you are trying to achieve. If you believe that your model is sensible only on a limited set of moments, a partial information approach is better suited. Using a full information in this case, you may end up with crazy estimates if the parameters are identified by matching moments for which your model does a poor job (you do not have control over the moments used to identify the parameters in a full information approach).

It is not obvious to me that {G,S}MM methods perform better on small samples.

These partial information methods will eventually come in Dynare, but it takes time to code this in a general manner. if you are proficient with Matlab, it is not very difficult to write a code specific to your model by calling the matlab routines shipped with Dynare (we have routines returning the theoretical moments at first order or simulating samples, you will easily find them by reading the stoch_simul.m in the matlab folder).