I have been trying to estimate a DSGE model using Bayesian methods. But I encounter some difficulties, and thus, the following question.
I have invoked the csminwel as the algorithm for the maxlik procedure. However, it has been running for more than 300 iterations, and the improvement on each iteration is still some distance from the tolerance accuracy.
I’d like to get insights on: 1) the meaning of this; 2) solutions on how to get over this problem.
Thanks so much for your attention.
If the mode value for the parameters is far from the initial values that you are providing (the prior mean), the optimization step may be very complicated.
The problem may come from a conflict between your model and your data. Calibrate your model with the priod mean and run stoch_simul. Then compare the theoretical moments reported by Dynare with the corresponding empirical moments of the observed variables. If you see huge differences (one being 50 times bigger than the other or even more) you should wonder where this comes from.
The next thing that you can try is to diminish the number of observed variables until the problem disappears. Then you add the observed variables back one by one. When you hit the problem again, you know which variable (and which mechanism in the model) creates problems.