Page 1 of 1
2nd Order Estimation - Particle Filter Question
Posted:
Wed Mar 05, 2014 4:03 pmby pcb
I am doing a second order bayesian estimation and would like to know what is the default number of particles used when evaluating the likelihood with the particle filter? Can the number of particles be changed and how?
Thanks!
Re: 2nd Order Estimation - Particle Filter Question
Posted:
Wed Mar 05, 2014 4:07 pmby federico
before the estimation command add this line
options_.particle.number_of_particles = 500;
I think the default number of particles is 500.
Re: 2nd Order Estimation - Particle Filter Question
Posted:
Wed Mar 05, 2014 11:28 pmby pcb
Thanks. Related question. Is the particle filter estimation going to run faster on Dynare++? Is there something that can be done to increase speed? It seems to take quite a long time.
Re: 2nd Order Estimation - Particle Filter Question
Posted:
Thu Mar 06, 2014 10:26 amby federico
Maybe, but I'm not sure about the gain. In general the PF could take a very long time depending on the magnitude of your model
Re: 2nd Order Estimation - Particle Filter Question
Posted:
Sat Mar 08, 2014 11:34 amby jpfeifer
Particle filters generally take a long time. Dynare++ does not support them.
Re: 2nd Order Estimation - Particle Filter Question
Posted:
Thu Mar 13, 2014 3:31 amby pcb
Thanks a lot, it is very helpful...
Re: 2nd Order Estimation - Particle Filter Question
Posted:
Fri Mar 28, 2014 5:32 amby pcb
Related question:
It seems that if I run a 2nd order estimation in different platforms PC/OS-X/Linux(Redhat) the MCMC acceptance rate seems to differ (at least initially). Is this something to be expected?
Re: 2nd Order Estimation - Particle Filter Question
Posted:
Fri Mar 28, 2014 8:07 amby jpfeifer
I don't know. If the same random number generator is used by all platforms with the same seed, this is not expected. But I cannot say anything regarding the RNGs on other platforms. Do the final results also differ?