Mode_compute = 6 works for Bayesian estimation but not SMM/GMM estimation

I run a mod file successfully with mode_compute = 6 under Bayesian estimation. However, I get the error: Error using chol when estimating the parameters under SMM estimation (mode_compute = 6).

Under SMM estimation, I changed the optimizer to mode_compute = 4…and I get the message solver stopped prematurely, so the moments do not match.

I tried using simulated data, but the same problems under SMM estimation (Error using chol).

Is it a common problem when order=1 in method_of_moments() and the model is linear?

Finding the optimum is generally hard. mode_compute=6 is a very inefficient optimizer and within the context of SMM/GMM it does not have the advantage of always providing a positive definite Hessian.

Also, I thought specifying prior distributions in Bayesian estimation maybe sort of help the optimizer. But specifying prior distributions (prior shapes, for example) is not available in SMM/GMM estimation yet.

It’s a quadratic objective function, so any prior will be considered approximately normal. But generally, priors are supported.

1 Like