A large shift in a time series that is expected to persist. The effect is usually a change in extrapolation (Armstrong and Collopy 1992).