Forecasting fed beef production: an evaluation of systems forecasting by parts for optimal survey history

Abstract

The following study examines forecasts of fed beef production through a by-parts estimation framework to provide a more practical alternative to more complex simultaneous systems of equations. A disconnect exists between the academic literature and the methods commonly used in private industry decision making. Practical techniques employed by industry professionals hold promise in strengthening conversations around forecasting research. Recent shifts in cattle production, including elevated contributions of heifers in the slaughter mix and larger than anticipated weights in fed cattle, provide an appropriate case study that underscores the need to revisit past events in the development of new forecasting strategies. By analyzing survey history in the data selection process, examining analog time periods, and considering concerns of autocorrelated errors within Deterministic Trend / Deterministic Seasonality models, this study highlights that practical enhancements to forecast accuracy can be sustained by simple remedial measures. The results of this study demonstrate that (1) historical data selection can significantly impact forecast quality; (2) methods that allow for autocorrelated error corrections can improve model performance; (3) the effectiveness of different forecast estimation methods varies by the selected horizon; and (4) the use of more simplistic assumptions underlying a forecast model can produce competitive and accurate results.