possible bugs with mode_compute=6

[bkjecn]

I believe that the file that governs the behavior of dynare when mode_compute=6 in the estimation command is gmhmaxlik.m. The wiki indicates that the user should be able to control the behavior with options_.Opt6Numb and options_.Opt6Iter (http://www.dynare.org/DynareWiki/MonteCarloOptimization). In versions 4.3.2 and 4.3.3 of dynare, it is impossible to control the number of simulations because the number is hard-coded to be 200000. This appears to be a bug. In the unstable branch, the bug is sort of fixed, but the documentation is incorrect, making the behavior again unexpected. It appears that options.nscale and options.nclimb replace the previously documented options. This can be seen in lines 85 and 86 of the file at https://github.com/DynareTeam/dynare/bl ... mhmaxlik.m. It is unknown to the user if these undocumented options alter other behavior.

[jpfeifer]

UPDATE: The Information in the Wiki is still true. You can only set the number of Iterations used in the estimation of the posterior covariance. The other options (number of iterations in the tuning of the scale parameter and the Hill climbing) have been hardcoded ever since.

The information about the unstable branch (later Dynare 4.4) has been added to the Wiki.

[Aldo]

Dear jpfeifer,
I am using Dynare 4.4.3 and I set

options_.Opt6Numb = 20000;
options_.Opt6Iter = 1;

before estimation command, but is the third time I see the window of "Tuning of the scale parameter", and I have the next question
What do you mean with "You can only set the number of Iterations used in the estimation of the posterior covariance."? Do you refers to options_.Opt6Iter? or to options_.Opt6Numb?

Thanks

[jpfeifer]

In Dynare 4.4.3 you can control pretty much all options of

Code: Select all

mode_compute=6

with the

Code: Select all

optim =(name,value)

option (see the manual). The option governing the iterations in the covariance estimation is

Code: Select all

ncov-mh