Hessian estimation with user provided function

[Dr. En Ami]

How does one return (to dynare_estimation_1) an estimate of the Hessian with the user defined mode estimator?

For what is the mh_posterior_mode_estimation option?

Where may I post bug reports or request revisions to the documentation?

[jpfeifer]

Currently this is not possible, but if you could provide details on what you are trying to do and what you would need, we could try to accomodate this.

Code: Select all

mh_posterior_mode_estimation

is for example used when loading a pre-existing mode file and not wanting to start again using this mode as starting value.

You can post bug reports and request revisions to the documentation either here or directly at https://github.com/DynareTeam/dynare/issues?page=1&state=open

[Dr. En Ami]

I need a positive definite Hessian which none of the documented built-in methods provide. Actually it would be sufficient if I could use the undocumented 'prior' value for mode_compute.

[jpfeifer]

I am not sure to be following. Problems with optimizers are rarely responsible for the non-positive definite Hessian. Usually there is an issue with the model or finding the mode. If neither mode_compute=6,8,9,10 (or a sequence thereof) provide a satsifactory results, there is usually a more fundamental problem hidden in the model.

[Dr. En Ami]

Yes, I am having trouble finding the mode with any of the built in optimizers, but I don't need the mode to run Metropolis-Hastings. I merely need a positive definite matrix which dynare calls the Hessian. I posted an issue about how to accomplish this when the optimizers fail.

[jpfeifer]

A quick and dirty hack that would solve your problem can be found here: http://www.dynare.org/phpBB3/viewtopic.php?f=1&t=4841

[Dr. En Ami]

That does not seem to work either, presumably due to limited precision.

[jpfeifer]

You could also simple comment out the error after the chol().command and set hh to an identity.