I would like to compute the RMSE of out-of-sample forecasts for an estimated model.
My approach: I use the option filter_step_ahead = [1 2 3 4 5 6 7 8], I get forecasts and I subtract the observables data.
My problem is that FilteredVariablesKStepAhead yields forecasts as deviations from the ss (right?).
But to evaluate the RMSE I need the forecasts non-in-deviations from the ss (or the estimated steady state for each Kalman Filter iteration).
Where can I find that / do anybody has an alternative way to compute true RMSE?
Thank you for your answer Johannes. Your help is an invaluable service to the academic community.
About your sidenote yes, I have a 30 years long sample and I am evaluating the RMSE on the last 20.