Out-of-sample predictions are based on the assumption that the discovered relationship holds beyond the observation range, and should be applied with caution.