Restricting parameters in Dynare

Dynare help (legacy posts)
1

Hello,

I am trying to optimize the parameters of a simple Taylor rule in a neo-keynisian enviroment.

The parameters turn out to be non-sensical. I was wondering if you could give me some insight:

  1. Is there a possibility to impose a restriction to the parameters.
  2. If not, what do you recommend?

Thank you,
Germán

2

Which context are we talking about? In OSR, this is not easily possible yet.
In estimation, you can use the estimated_params-block to set upper and lower bound for your parameters or specify a prior that does not allow the parameters to be bigger than certain bounds.

3

Thank you for your reply.

Actually, I was thinking of implementing it on an OSR…could you help me providing some guidance in order to do this?

Thank you again.

Kind Regards,
Germán

4

This is somewhat complicated. Please provide your current codes and indicate the bounds you want to set and I will try post an example.

5

Thank you again for your reply.

This is my mod file.

Kind Regards,
Germán
sinramsey.mod (4.21 KB)

6

Dear Germán,

you need to use the current unstable version (to be Dynare 4.5) with the attached files. In osr_optimizer_function_wrapper.m you can manually change the bounds (and the optimizer to use). It currently requires a Matlab Toolbox as I am using fmincon as an example. Note also that there is a bug in the current unstable that will soon be fixed. You need to replace in dynare_minimize_objective.m the calls to

by

See github.com/JohannesPfeifer/dynare/commit/be58d739d4b2338977395bceee6780955915310c
sinramsey.mod (4.25 KB)
osr_optimizer_function_wrapper.m (689 Bytes)

7

Dear Dr. Pfeifer,

does this solution already apply to the current stable version of Dynare? Thanks

8

No, it only works with the unstable, to be published as Dynare 4.5

9

Hello Johannes,

After reading your code, I’m wondering whether osr now works with second order approximation to the model’s policy function? (I’m asking because of the “order=2” in the osr command).

Best,
gauthier

10

No, because a first order approximation already delivers second-order accuracy in the unconditional moments, which osr minimizes. Going to second order would yield fourth-order accuracy.