Moment_calibration: Which options? Curvature?


#1

Hi,
I’m trying to exploit the moment_calibration option to give extra weight on particular moments of interest in the estimation of a DSGE model.

The Dynare manual does not appear to explain which kind of options can be passed to this command.

It also occurs to me that a uniform prior on the endogenous moments is most likely leading to corner solutions (unless the prior is not binding/likelihood and exogenous priors are flat…) This appears to imply that the Hessian at the mode would not be positive definite, impeding the use of the mode and hessian as starting value for the MCMC.

Questions:

  1. What are the options of moment_calibration?
  2. Is there a way to impose curvature on the endogenous priors?
  3. Am I totally off and this is not the way moment_calibration is supposed to be used?

Anybody?

Thanks

Gianni


#2
  1. Have a look at https://github.com/DynareTeam/dynare/commit/c7290a5f0341748672836b6827ac4a9175df8fb0
  2. No, currently you can only specify to reject parameter draws that do not satisfy the restriction. That is, they get assigned a prior density of 0. Systems priors are still on our todo list.
  3. The point about mode-finding is a bit strange. Internally, a penalty function is used that penalizes violations of the restriction in a continuous way. That should go a long way to alleviating cliffs and associated corner solutions.

#3

thanks!
gianni