Forecasting the daily dynamic hedge ratios by GARCH models: evidence from the agricultural futures markets

Abstract

This paper investigates the forecasting ability of six different generalized autoregressive conditional heteroskedasticity (GARCH) models; bivariate GARCH, BEKK GARCH, GARCH-X, BEKK-X, Q-GARCH and GARCH–GJR based on two different distributions (normal and student-t). Forecast errors based on four agricultural commodities’ futures portfolio return forecasts (based on forecasted hedge ratio) are employed to evaluate the out-of-sample forecasting ability of the six GARCH models. The four commodities under investigation are two storable commodities: wheat and soybean, and two non-storable commodities: live cattle and live hogs. We apply the rolling forecasting method and the Model Confidence Set approach to evaluate and compare the forecasting ability of the six GARCH models. Our results show that the forecasting performances of the six GARCH models are different for storable and non-storable agricultural commodities. We find that the BEKK-type models perform the best in the case of storable products, while the asymmetric GARCH models dominate in the case of non-storable commodities. These results are regardless of the forecast horizon and residual distributions.