Forecasting seasonally cointegrated systems: supply response of the Austrian breeding sow herd

This paper examines the relevance of incorporating seasonality in agricultural supply models. Former studies have eliminated the problem of seasonality by using seasonally adjusted data. Recent developments in cointegration techniques allow the comprehensive modelling of error-correcting structures in the presence of seasonality. We consider a four-variable model for Austrian agriculture. Series on the producer price for soy beans, bulls and pigs, as well as the stock of breeding sows are included. A vector autoregression incorporating seasonal cointegration is estimated. A tentative interpretation of long-run and seasonal features is considered. Forecasting experiments are reported. The results of our experiments indicate that models that do not account for seasonal cointegration may yield better forecasts at short prediction horizons, but the seasonally cointegrated model tends to dominate at larger step sizes.