Composite and Outlook Forecast Accuracy

This paper investigates whether the accuracy of outlook hog price forecasts can be improved using composite forecasts in an out-of-sample context. Price forecasts from four widely-recognized outlook programs are combined with futures-based forecasts, ARMA, and unrestricted Vector Autoregressive (VAR) models. Quarterly data are available from 1975.I through 2007.IV for Illinois/Purdue and 1975.I-2010.IV for Iowa, Missouri, and USDA forecasts, which allow for a relatively long out-of-sample evaluation after permitting model specification and appropriate composite-weight training periods. Results show that futures and numerous composite procedures outperform outlook forecasts, but no-change forecasts are inferior to outlook forecasts. At intermediate horizons, OLS composite procedures perform well. The superiority of futures and composite forecasts decreases at longer horizons except for an equal-weighted approach. Importantly, with few exceptions, nothing outperforms the equal-weight approach significantly in any program or horizon. In addition, the equal-weight approach as well as other composite approaches can generally produce larger trading profits compared to outlook forecasts. Overall, findings favor the use of equal-weighted composites, consistent with previous empirical findings and recent theoretical papers.