Page 1 of 1

Bayesian estimation: tolerance

PostPosted: Thu Oct 02, 2008 1:53 pm
by tanya
Deal all

How can I change the default tolerance in the main algorithm of Bayesian estimation?

I have an iterative block INSIDE the mod-file with extremely slow convergence to, say, E^-10. I cannot optimise it further than it is currently done. As a result, computing posterior distributions takes ages. I need to decrease the default tolerance to, say E^-6, or something less that internal convergence tolerance. I use MATLAB. I undertsand Dynare uses some standard MATLAB minsearch or something like this, how can I change the default parameters in the correct place?

Many thanks
Tanya

PostPosted: Fri Oct 03, 2008 4:04 pm
by bigbigben
Hi, Tanya:
If you use Dynare, the default optimizer is csminwel, not any built-in function in Matlab. Try add the following lines in your mod file before the estimation line

options_.dynatol=1e-5;
options_.cutoff=1e-8;

If you use fminunc or something else in Matlab, add the line before using the minimizing algorithm

options = optimset('TolFun',1e-5,'TolX',1e-3);