Simulated Methods of Moments

Posted:
Fri Apr 24, 2015 3:13 pm
by xtsong
I found the function [param,sigma] = simulated_moments_estimation(dataset,options,parallel) for dynare.
But I don't know how to do to run the SMM estimation in dynare? Should this be include in a dynare script?
I would like to know if anyone has an example include this function.
Thanks~
Re: Simulated Methods of Moments

Posted:
Sat Apr 25, 2015 7:42 am
by jpfeifer
This function is experimental and not supported by us yet. If you are looking for a simulated method of moments, please have a look at the replication files for Born/Pfeifer (2014): "Risk Matters: A comment" at
https://sites.google.com/site/pfeiferecon/20140525_replication_codes.zip?attredirects=0. There, we use SMM to estimate a third-order approximated model.
Re: Simulated Methods of Moments

Posted:
Tue Apr 28, 2015 7:57 pm
by xtsong
Appreciate to your answer.
Re: Simulated Methods of Moments

Posted:
Thu Apr 30, 2015 1:46 pm
by Daniel Bendel
As far as I understand your code right, you do not use a weighting matrix. If I want to use a weighting matrix, how do I have to change the code to insert one?
Re: Simulated Methods of Moments

Posted:
Thu Apr 30, 2015 2:39 pm
by jpfeifer
At the last line, we have
- Code: Select all
% compute objective function
fval=(simulated_moments-target)*(simulated_moments-target)'
Implicitly, this is
- Code: Select all
% compute objective function
fval=(simulated_moments-target)/eye(n,n)*(simulated_moments-target)'
You would need to replace the identity in the middle with your weighting matrix, derived e.g. using a Newey-West type estimator. See Ruge-Murcia's codes for an example.