Optimal Length of Moving Averages to Use When Forecasting Basis

Abstract

The previous year's basis and 2 to 5-year historical moving averages were used to forecast preharvest and end of storage period basis for hard wheat, soft wheat, corn, and soybeans. Forecast accuracy was compared within a mixed model, to result in each forecasts mean absolute error. The forecast which produced the minimum mean absolute error provided the optimal forecast. In general, forecasts containing less historical information resulted in optimal basis forecasts. Only the preharvest hard wheat forecasts benefited from longer moving averages. The previous year's basis provided the optimal forecast when considering all crops, periods, and locations. The general rule of thumb resulting from this study is that at locations or during times not likely to undergo structural change, longer moving averages provide the best forecasts, but following a structural change, the previous year's basis provides the optimal forecast.