hello everybody,

brief question of understanding concerning the filtered variables (1step predictions) in the result structure oo_:

Since oo_.FilteredVariables is reported with respect to the steady state, do I assume correctly that the mean of the prediction errors computed from it should be as close as possible to zero?

Since this is not the case, is there any problem in how I calculate the 1-step ahead prediction error?

```
>> global M_
>> [ys,tchek] = feval([M_.fname '_steadystate'],[zeros(M_.exo_nbr,1);oo_.exo_det_steady_state]);
>> v_pred=y_obs-repmat(ys(3:7)',length(y_obs),1)-y_filtered; % lines 3 to 7 in ys contain the steady state vals of observables!
```

y_obs is a matrix with the “varobs” arranged in columns and y_filtered contains corresponding columns of observed variables from oo_.FilteredVariables.

Is there any problem with this prediction error calculation?

Many thanks for your help

Small update: I seem to be missing the selection matrix M (or mf or Z) to map filtered observables on the true ones (cf. p.4 in presentation of Michael Juillard “estimation with Dynare”). Unfortunately it does not seem to be saved in the results…