Solution techniques under zero bound

[kwaner]

I am looking to estimate a DSGE imposing the zero bound on nominal interest rates. I am just going to be doing simulations (not looking at optimal policy). What will be the best and computationally feasible technique?

I saw augmenting the Taylor Rule with anticipatory monetary shocks which is easy to implement when estimating log-linearized models. Calibration seems doable for global/non-linear methods. But are there any alternatives for estimating DSGE models?

[StephaneAdjemian]

Hi, if you are able to simulate your model with a ZLB on nominal interest rate, then you can estimate this model by the Simulated Method of Moments. Likelihood based approaches are not yet feasible in dynare for this kind of problem.

Best, Stéphane.

[kwaner]

Thanks. I saw one your joint papers that does simulated method of moments, but it looks like that the set of parameters estimated is fairly small.

Just a quick question to follow up on my earlier note. If I adopt the Bayesian framework in a linearized model, will the following work?

(1) in an MCMC scheme, draw a set of parameters.
(2) simulate the model for this particular set for T periods.
(2) retain the draw if it does not violate zero bound, else draw another set.

[StephaneAdjemian]

Hi,

Thanks. I saw one your joint papers that does simulated method of moments, but it looks like that the set of parameters estimated is fairly small.

Yes, the number of estimated parameters is small, but it's just a matter of time...

Just a quick question to follow up on my earlier note. If I adopt the Bayesian framework in a linearized model, will the following work?

(1) in an MCMC scheme, draw a set of parameters.
(2) simulate the model for this particular set for T periods.
(2) retain the draw if it does not violate zero bound, else draw another set.

I don't think that we should skip the draws such that the economy hits the ZLB (as we do for the parameters such that Blanchard and Kahn conditions are violated) even if it is possible to do what you suggest. Also the problem is that it is not possible to linearize the model if ZLB (or any occasionally binding constraint) is an issue.

Best,
Stéphane.