zero variance decomposition

[earsmall]

I modified the model proposed by Justiniano and Preston (2010) and have been trying to estimate the new model.
Actually, I wanted to do with mode_compute=4 but it was hard to find the initial posterior mode.
So, I use mode_compute=6.

Eventually, estimation was done. However, results of posterior variance decomposition is strange.
Mostly, they are zero...

I hope to have possible reasons for this.
I attached my mod file and data.

[jpfeifer]

Please upload everything including your mode-file as a zip-file

[earsmall]

I uploaded zip file which includes the mod file and data.

Thank you

[jpfeifer]

You still did not provide your _mode.mat-file

[earsmall]

Sorry
I uploaded the mode file as well

[jpfeifer]

The mode-file does not match the mod-file:

Code: Select all

The posterior mode file fx_est_post_z_mode has been generated using another specification of the model or another model!
--> Estimated parameter eps_gg is not present in the loaded mode file.
--> Estimated parameter rho_gg is not present in the loaded mode file.

[earsmall]

Sorry, I uploaded data, program (*.mod) and mode file (*_mode.mat).
Also, I uploaded the result (*.mat) in whch fevd mean of most variables are zero.

I think there might be a stationary issue. But in my opinion it has to be with some variables only
such as z, e, p, pf, ph. Not c or y.

[jpfeifer]

I can confirm that the problem relates to nonstationary variables. For those variables, we are displaying 0 for all shocks, which is incorrect and should be NaN. We will fix this.

[earsmall]

I am wondering if you let me know which variable or part of the model incurs stationary problems.
In fact, c (consumption), pi (inflation) and y (domestic output) are definitely stationary.

[jpfeifer]

My guess is that

Code: Select all

// (7) Terms of Trade
s = pf - ph ;


causes this, because you have two non-stationary variables on the RHS and from there is spills over to the real side of the model.