shock_decomposition differs from simult_

[zhanshuo]

Hi, dear professor, I find that shocks decomposition result is different from what I get from using simult_ with the certain smoothed shock series.
Is it normal?
how to explain the difference?
Thanks!

[jpfeifer]

No, that is not normal. Using the smoothed states at the beginning with the smoothed shocks should give the series exactly. Are you sure the variable and shock ordering is correct?

[zhanshuo]

The order is right, and the result is sometimes slightly different, or alike in shape but different in degree.
So what do you think possibly is the reason?
I wonder maybe the calibrated model may not rightly use the estimated parameters result.
I just manually draw the estimated results out into a calibrated model.
Is there a easier and more reliable way to directly use the estimated parameters to do IRF experiments repeatedly?

[zhanshuo]

I look into the m files in dynare and I find out that shock decomposition call for the file of "evaluate_smoother", which I doubt may different from the simult command.
So is it possible that cause the difference?

[jpfeifer]

The call to

Code: Select all

evaluate_smoother

will generate the smoothed shocks and observables given the parameter set provided. If the parameter set is the same as from estimation, the results should be the same as with simult_ given the smoother estimates.