Observed variable's decomposition of shocks and FEVD

Dear Johannes,
First thank you very much for your previous helpful guidance, I am grateful.
I have a question, I have decomposed observed variable’s historical dynamics into shocks (shock_decomposition command in Dynare), I mean my graph measures how these shocks contribute to the an observed variable’s deviation from its steady state over time.
I have also plot forecast error variance decomposition for the same observed variable.
I find some shocks contribute only a little to the observed variable’s historical deviations from its steady state, but contributes a lot to forecast error variance decomposition. I am wondering is this normal? if not, what kind of problems this may indicate?
Thank you very much and look forward to hearing from you.

Kind regards,

I guess the most obvious issue is model misspecification that leads the shocks in the historical decomposition to be correlated, while the FEVD is based on orthogonalization that may be violated in sample.