Forecasting Basis Levels in the Soybean Complex: A Comparison of Time Series Methods

A battery of time series methods are compared for forecasting basis levels in the soybean futures complex: soybeans, soybean meal, and soybean oil. Specifically, nearby basis forecasts are generated with exponential smoothing techniques, autoregression moving average (ARMA), and vector autoregression (VAR) models. The forecasts are compared to those of the 5-year average, year ago, and no change methods. Using the 5-year average as the benchmark method, the forecast evaluation results suggest that alternative naive techniques may produce better forecasts, and the improvement gained by time series modeling is relatively small. In this sample, there is little evidence that the basis has become systematically more difficult to forecast in recent years.