Problems with identification for a SMOE (As in Gali ch. 7)

[jogr445]

Hi,

I am having trouble identifying any parameters in my model setup. I have generated data from the calibrated model, estimated it on the generated data but I can't seem to get a handle on the identification. I keep getting versions of this error message:

==== Identification analysis ====

Testing prior mean
The number of moments with non-zero derivative is smaller than the number of parameters
up to 10 lags: check your model
Either further increase ar or reduce the list of estimated parameters

----------------------------------------------------------------------------------------------

I have tried reducing the number of paramets to be estimated (and the number of observables), that does not fix my problem. I think I am making some simple but fundamental mistake. However, I am having trouble figuring out what that mistake is. Any guidance on how I can proceed and identify some/any parameters in this model would be most appreciated.

Thank you in advance,

/J

[jpfeifer]

The fundamental issue is the presence of a unit root in your model that is not accounted for. Run model_diagnostics to see this. You have to set

Code: Select all

options_.lik_init=2

But even then there are some issues. We are trying to investigate why this happens.

[jogr445]

Thank you very much. I will change the settings and run diagnostics.

Since this is the baseline setup for a SMOE from Gali Monacelli, is it documented where the UR comes from? Or rather, how do I setup the parameter distributions to avoid a UR?

Again, your input is much appreciated.

/J

[jpfeifer]

With interest rate rules, the price levels become non-stationary (they are not mean-reverting) as there is no nominal anchor.

[jogr445]

I suspected the TR was to blame.

Thank you for clearing that up for me.


/J

[rattoma]

Please try to observe the first difference or the growth rate of observed non-stationary variables (only yhat? pih and pif are inflation rates already ?)

[jogr445]

Yes, pih and pif is home inflation and foreign inflation respectively.

[jogr445]

I get the exakt same error message when I run this code instead. This is for one estimated parameter.

-------------------------------------------------------
The rank condition is verified.


==== Identification analysis ====

Testing prior mean
.
.
Try increasing ar = 10

The number of moments with non-zero derivative is smaller than the number of parameters
up to 10 lags: check your model

Either further increase ar or reduce the list of estimated parameters
Error using identification_analysis (line 102)
IDETooManyParams
======================================

The same model but with a different setup.

I can't seem to get this to work for me.

/J

[jogr445]

This issue percists even after I remove the p_h and cpi_level variables & equations, which are non-stationary. Then the only variable that is non-stationary is e, but it enters as a first difference.

/J

[jpfeifer]

We are still working on this. But if e only enters as a first difference, drop e and define the first difference as a new variable.

[jogr445]

Thank you, I really appreciate your input.

Defining a new variable de=e -e(-1) and dropping e as a variable in the estimation step works, but I don't understand why that should be an issue for identification. I guess this is what you are working on.

Best regards

/J

[jpfeifer]

Exactly. Your definition of e implies that there is a unit root. By only having the first difference of e, the unit root is gone from the model. Dynare should be able to deal with this case, but it is tricky to implement.

[Oriana]

Hi,

It seems to me I have the same problem. I suspect the presence of a unit root associated with my exogenous variables.
My model, however, is quite complex.
I would deeply appreciated if someone could help to find where/what the real problem is.

==== Identification analysis ====

Testing posterior mode
The number of moments with non-zero derivative is smaller than the number of parameters
Try increasing ar = 2
The number of moments with non-zero derivative is smaller than the number of parameters
Try increasing ar = 3
The number of moments with non-zero derivative is smaller than the number of parameters
Try increasing ar = 4
The number of moments with non-zero derivative is smaller than the number of parameters
Try increasing ar = 5
The number of moments with non-zero derivative is smaller than the number of parameters
Try increasing ar = 6
The number of moments with non-zero derivative is smaller than the number of parameters
Try increasing ar = 7
The number of moments with non-zero derivative is smaller than the number of parameters
Try increasing ar = 8
The number of moments with non-zero derivative is smaller than the number of parameters
Try increasing ar = 9
The number of moments with non-zero derivative is smaller than the number of parameters
Try increasing ar = 10
The number of moments with non-zero derivative is smaller than the number of parameters
up to 10 lags: check your model
Either further increase ar or reduce the list of estimated parameters

[jpfeifer]

It seems to me you are simply neglecting the parameter dependence. Your calibration only once updates the other parameters depending on the estimated one. That’s why you should use model-local variables (the ones with the pound operator) or a steady state file. See Remark 4 (Parameter dependence and the use of model-local variables) in Pfeifer(2013): "A Guide to Specifying Observation Equations for the Estimation of DSGE Models" https://sites.google.com/site/pfeiferec ... ations.pdf.

For example, you estimate c1telast, but you set outside of the model or steady state

Code: Select all

c1rhoc = 1/(c1telast -1);
c1rhoo = 1/(c1oelast -1);


Thus, c1rhoc and c1rhoo will not be updated.

[Oriana]

Thank you for the previous observation.

Unfortunately, despite the use of model-local variables for all my dependent parameters I still get the same message.

==== Identification analysis ====

Testing posterior mode
The number of moments with non-zero derivative is smaller than the number of parameters
Try increasing ar = 2
The number of moments with non-zero derivative is smaller than the number of parameters
Try increasing ar = 3
The number of moments with non-zero derivative is smaller than the number of parameters
Try increasing ar = 4
The number of moments with non-zero derivative is smaller than the number of parameters
Try increasing ar = 5
The number of moments with non-zero derivative is smaller than the number of parameters
Try increasing ar = 6
The number of moments with non-zero derivative is smaller than the number of parameters
Try increasing ar = 7
The number of moments with non-zero derivative is smaller than the number of parameters
Try increasing ar = 8
The number of moments with non-zero derivative is smaller than the number of parameters
Try increasing ar = 9
The number of moments with non-zero derivative is smaller than the number of parameters
Try increasing ar = 10
The number of moments with non-zero derivative is smaller than the number of parameters
up to 10 lags: check your model
Either further increase ar or reduce the list of estimated parameters

Could it be the presence of a unit root as suggested before?