The problem with one-step-ahead predictions