Bayesian Estimation of Indeterminate Model in DYNARE

Hello,

I have trying to follow the procedure of Farmer, Khramov and Nicolo (forthcoming JEDC) for estimating an indeterminate model in DYNARE by redefining a non-fundamental error as a newly defined fundamental. The model I am trying to replicate is the small scale New Keynesian model in Lubic and Schorfeide (2004, AER). Since the likelihood function is discontinuous at the boundary of the determinacy-indeterminacy region, they conduct the computations for the two regions of the parameter space separately. I believe what they do is draw from the posterior and then check whether the parameter draw comes from a determinacy region or indeterminacy region and then decide whether to keep that draw or not depending on whether their focus is on the determinate region or the indeterminate region.

I am currently stuck in doing so in DYNARE. Is there any way to implement the above procedure in DYNARE while estimating the Lubic and Schorfeide (2004) small scale NK model.

Your response will be highly appreciated.

Thanking in advance.

Regards,

Qazi.

Dear Qazi,

I do not kwow what L&S are doing, but there is no interface to do what you want in Dynare. You have to hack the code. You will first have to change the routines returning the likelihood (or posterior kernel) because they return infinity if BK conditions are not satisfied and you probably do not want that, then you will have to adapt the routines related to the MCMC (the main one being random_walk_metropolis_hastings_core.m).

Best,
St├ęphane.

Dear Stephane,

Thank a bunch for your prompt reply. Much appreciated.

Cheers,

Qazi.

Dear Stephane,

Just to confirm one thing, when you said that when the BK condition is not satisfied the posterior kernel is set to infinity. To be specific, did you mean that the log of the posterior kernel is set to NEGATIVE infinity (ie, log(posterior Kernel) = -100000000) in the code? Which then implies that the posterior kernel is actually set to zero. Is that what you meant?

Thanks once again.

Cheers,
Qazi.