A large body of evidence shows that combined forecasts typically produce more accurate forecasts (Clemen, 1989). The margin of superiority has been found in many situations, but it has been modest, typically about 7% when two methods are combined (Armstrong, 1986). The conditions under which combined forecasts are most useful have not been clearly identified. The limited prior research on this issue suggests that the gains are greater for short-range than for long-range forecasts. Sanders and Ritzman add useful information. They examined very short-range ex ante forecasts for 260 days. These were one-day-ahead forecasts for 22 daily series on customer shipments from a warehouse. The models were updated after each observation, yielding a validation sample of 5,720 forecasts. Five techniques were examined: single exponential smoothing, adaptive response rate exponential smoothing, Holts exponential smoothing, automatic univariate adaptive estimation procedure, and linear trend by regression. Based on MAPE, the differences among methods were not large. The methods are ordered above according to their accuracy, ranging from a MAPE.,of 70.7 for the single exponential smoothing to 78.7 for the linear trend. Three arbitrarily combined forecasts (using equal weights) proved to be substantially more accurate. The best had a MAPE of 61.9 and the worst was 63.9. On average, the combined forecasts reduced the MAPE by about 16%. In short, combining was an effective strategy for very short-range forecasts.