Problem with Steady-state and Estimation

[matheus_silva92]

Hello all,

I'm trying to estimate the model in Chen, Cúrdia and Ferrero (2012). I am, however, getting the following error:

==== Identification analysis ====

Testing prior mean

-----------
Parameter error:
The model does not solve for prior_mean with error code info = 22

info==22 %! The steady has NaNs.
-----------

Try sampling up to 50 parameter sets from the prior.
Evaluating simulated moment uncertainty ... please wait
Doing 252 replicas of length 300 periods.
Simulated moment uncertainty ... done!

All parameters are identified in the model (rank of H).


WARNING !!!
The rank of J (moments) is deficient!

However, the steady-state is calculated and I obtain the correct impulse response functions if I delete the estimation lines (estimated_params block and estimate command). I don't know what is happening, can someone please help?

Thanks in advance

[jpfeifer]

When using the estimated_params-block, the prior mean will be used instead of the calibrated parameters you used. Nothing guarantees that the steady state can be computed for these different parameter values. That is exactly what happens here. In particular, some calibrated values are very unlikely given your prior distribution. For example, adf is 15 prior standard deviations away from the prior mean.

[matheus_silva92]

Hi Johannes,

Thanks for noticing this. I have updated my priors (they're now correct), and this new version is now uploaded here.

Now a new problem happens.
I have tested the prior means as parameters and they solve the model when I exclude the observation equations. When I do keep these, however, the model solution fails.

So, in this new code, estimation begins, but due to this I cannot solve for the posterior mode (it even starts the optimization algorithm but it stops at the initial guess).

Do you have any idea of why this is going on and how to solve it?

Thanks in advance,
Matheus

[jpfeifer]

That happens because your observation equations are wrong. The data means for the observables do not match the model variable means. For example, L_obs in the data has a mean around 50, while the model variable l_obs has a steady state of 0.5102