F measurement errors always exist and combining makes them smaller;
F unstable or changing patterns or relationships cause single models to be unreliable, but combining several models improves their accuracy; and
F models usually minimize past data errors that were poor predictors of future sales.