Unfortunately, this is not unheard of. See the discussion at Endogenous prior for example.
The problem is in-sample you may have correlations between shocks that are 0 if you were to simulate your model.
1 Like