problem with identification command

[reluginbuhl]

Ithink I may have found a problem with Dynare 4.3.0! Something seems to be wrong with the identification command.

Here is an example of a mod file which leads to an error in the identification command (based on code given in http://www.lkf.vwl.uni-muenchen.de/down ... nesian.pdf). The model is a simple New Keynesian model with a dynamic IS equation, a new Keynesian Phillips curve, and a Taylor rule.

When the identification argument max_dim_cova_group=1, then there is no error and Dynare completes the command successfully. However, with max_dim_cova_group=2 as in the uploaded mod file, Dynare crashes.

I would appreciate help on this one.

Thank you.

Sincerely,
Rob Luginbuhl

[rattoma]

Thanks for pointing to this problem. I have fixed it and it will be available for the next release 4.3.1.
For your model, I think the chains of co-linearities between the parameters in your model are already clear form what is provided by the code.

[reluginbuhl]

Thank you for the quick reply.

I am quite surprised that identification tells me that the Taylor Rule coefficients on inflation and the output gap are not identified in this simple New Keynesian Model. Is this well known? Strangely, when I estimate this simple model using 15 EU countries for the period 1987-2009 and use mode=5, the Hessian of the posterior mode is of full rank! This suggests that these parameters ARE identified!

I am confused.

Any elucidating thoughts on this matter would be greatly appreciated!

Sincerely,
Rob Luginbuhl

[jpfeifer]

I am also a bit puzzled. Isn't one of the points of Cochrane's recent JPE paper that in contrast to single equation estimation the cross-equation restrictions in FIML can be used to estimate a Taylor rule?

[rattoma]

Hi there,

I guess this is due to the fact that in the mod file there is no monetary shock.
If you move the ei shock to the taylor rule and give a non-zero value for that shock (currently that shock is null), you should get identification there. Also the ey shock should be set non-zero to avoid stochastic singularity.
best
Marco

[reluginbuhl]

We seem to have determined that a permanent technology shock does not contribute to the identification of the Taylor rule. Identification requires a temporary shock in technology as well. This makes economic sense in that the output gap in the New Keynesian Phillips curve will not be effected by a permanent shock to technology, and so the permanent shock will have no effect on the Taylor rule. Only a temporary shock will effect the output gap and thereby contribure to the identification of the Taylor rule.

Sincerely,
Rob Luginbuhl

ps. This model has been made stationary by dividing output by permanent technology: yf = output/permanent technology