Can Cattle Basis Forecasts Be Improved? A Bayesian Model Averaging Approach

Basis forecasts aid producers and consumers of agricultural commodities in price risk management. A simple historical moving average of nearby basis on a specific date is the most common forecast approach; however, in previous evaluations of forecast methods, the best prediction of basis has often been inconsistent. The best forecast also differs with respect to commodity and forecast horizon. Given this inconsistency, a Bayesian approach which addresses model uncertainty by combining forecasts from different models is taken. Various regression models are considered for combination, and simple moving averages are evaluated for comparison. We find that model performance differs by location and forecast horizon, but the average model typically performs favorably compared to regression models. However, except for very short-horizon forecasts, the simple moving averages have a lower out of sample forecast error than the regression models. We also examine using a basis series created using a specific month’s futures contract as opposed to the nearby contract and find that basis forecasts calculated this way have lower forecast errors in the month of the contract examined.