A comparison of alternative price forecasting models for slaughter hogs

The purpose of this study was to develop a price forecasting model capable of accurately predicting the price of slaughter hogs one quarter in advance. Both time series and econometric models were hypothesized, estimated, and tested. In an attempt to improve fore casting accuracy, composite models were developed which utilized the price forecasts from both time series and econometric models as inputs. Two types of time series models were estimated. The simplest was an ARIMA model, which assumes current price is related to past price plus a random error term. Also estimated was a transfer function which is similar to the ARIMA model, but which includes an input variable. Three econometric models were hypothesized and estimated. The first was a recursive model which consisted of supply and demand equations separated by a change-in-storage equation. A reduced form model was derived from the recursive system. The final econometric model was a five-equation simultaneous system based upon equations and variables specified in a previous study. The simultaneous system was dropped from consideration due to poor estimation results. Three composite models were estimated by regressing 30 forecasted prices from two individual models against actual price. The estimated beta coefficients became the weights assigned to each individual forecast. Each model was evaluated over an eight-quarter out-of-sample period in terms of Theil's U₂ coefficient, root mean square error, and how well it could predict direction of price change. Results indicated that no model was able to predict better than a naive no-price-change model (calculated U₂ coefficient greater than one). Theil's U₂ coefficient ranged from a low of 1.11 for the transfer function to a high of 2.36 for the recursive system. Corresponding root mean square errors were $6.43 and $13.66, respectively. The best predictor of the direction of the change in price was the recursive model, which was correct 100 percent of the time.