Page 1 of 1

mode_compute=6, stability of acceptance Rate and nclimb

PostPosted: Thu Jun 29, 2017 11:04 pm
by Aldo
I am estimating my model using mode_compute = 6:

mode_compute=6, optim = ('NumberOfMh', 1, 'ncov-mh',50000,'nscale-mh',50000),

and I am seeing that acceptance rate in box "Estimation of the posterior covariance" is increasing very very slowly, the waitbar shows 0.34 y the advance is on half.

1. This question is about stability of acceptance rate. I would like to know if this slowly increasing is normal or if I should modify the priors due to this behavior.

2. By another side, I see in the Manual: "'nclimb' Number of iterations in the last MCMC (climbing mode)" and I would like to know what is Default value of "nclimb"? and How could I modify his default value?

Please.
Thanks in advance

Re: mode_compute=6, stability of acceptance Rate and nclimb

PostPosted: Fri Jun 30, 2017 8:51 am
by jpfeifer
1. During mode-finding, the acceptance rate will not be constant. There is no reason it should be constant. If you move to a region of higher likelihood, you will always accept these draws, leading to a high acceptance rate.

2. There is a typo in the manual. The option can be controlled with
Code: Select all
nclimb-mh

and its default is 200000. See https://github.com/DynareTeam/dynare/pull/1474