Online Program

A Comparison of the Forecast of Pork Carcasses Futures by Three Methods: A SETAR Model, a Seasonal ARIMA Model, and Holt-Winters Smoothing (303048)

Keywords: Forecast, SETAR, SARIMA, Holt Winters smoothing, future pork carcass

We investigate a SETAR (Self-Exciting Threshold Autoregressive) model, a SARIMA (0,1,1) model, and Holt-Winters smoothing to weekly pork carcasses futures, with information from 2003–2011 in the agricultural futures market (Chicago Mercantile Exchange). Point forecasts were performed using these three methods for the weeks between January 2012 and May 2014, considering the prediction horizons ranging from one week to eight weeks. The accuracy of these forecasts was compared using a test proposed by Francis X. Diebold and Roberto S. Mariano (Comparing Predictive Accuracy, Journal of Business & Economic Statistics (1995), Vol. 13, pp. 253–263). It was found that when the prediction horizon is one week, the predictions made by Holt-Winters smoothing and the SARIMA model are equally accurate and that the forecasts made by the SETAR model (p1 = 18, p2 = 52, d = 17 ) are more accurate than those made with the SARIMA. When the prediction horizon goes from two to eight weeks, the Holt-Winters smoothing and those made with the SETAR model are more accurate than those made with the SARIMA model.