A comparison of price forecasting models for manufacturing beef in the U.S.A

Abstract

This study evaluates price forecasting models for manufacturing beef in the U.S.A., using a range of forecasting methods. Specific objectives were directed at: (1) comparing the ability of a range of forecasting techniques to forecast prices for manufacturing grade beef in the U.S.A.; and (2) determining whether or not any of the selected techniques are capable of providing adequate forecasts. The study considers these issues in relation to the value of the methods used to sectors of the meat industry in New Zealand, one of the major suppliers of manufacturing beef to the U.S. market. The methods evaluated are (1) naive, (2) simple moving averages, (3) double moving averages, (4) single exponential smoothing, (5) double exponential smoothing, (6) Box-Jenkins, (7) single equation econometric, (8) simultaneous equation econometric, and (9) a combined method. Evaluation is based on Mean Absolute Error, Root Mean Square Error, Theil's inequality coefficient, and the ability to forecast turning points-- both cyclical and statistical. The methodology of the methods used and the form of the models developed are described. Results show that the Box-Jenkins, reduced form and combined models performed better than other models, particularly for annual forecasts. For quarterly and monthly forecasts the advantage was not as great, but still worthwhile. The sophisticated models were marginally better than the naive model when forecasting quarterly and monthly prices. When predicting turning points the sophisticated models were considerably better than all other models. The smoothing methods gave poor results under all conditions.