Title: Forecasting agricultural product and energy prices: A simulation-based model selection approach
Authors: Robert Kunst - Institute for Advanced Studies (Austria) [presenting]
Adusei Jumah - Central University Accra (Ghana)
Abstract: The aim is twofold. First, we study whether and to what degree the dynamic interaction between commodity prices and energy prices can be exploited for forecasting. Second, we present informative examples for the simulation-based forecast-model selection procedure. Apart from prediction by competing specifications to be selected from a small choice set, we also explore forecast combinations based on Bates-Granger weights constructed from a continuum in the same framework. The simulation-based method explicitly permits letting the forecast model choice depend on the intended time horizon of the forecast. With regard to classical Granger causality, the evidence supports a causal direction from food prices to fuel prices, without feedback and somewhat in contrast to our expectations. This causal link, however, only benefits forecasting accuracy at relatively large sample sizes. Similarly, clear evidence on considerable seasonal patterns cannot be fused to a seasonal time-series model that outperforms non-seasonal rivals. The simulation experiments generally favor the handling of all price series in first differences. Ultimately, the forecast combination experiments indicate a window of opportunity at a specific horizon, whereas pure strategies dominate at smaller and larger horizons.