Forecasting aggregated Euro area inflation rate with space-time models

Economic variables are typically observed over time and across different but likely correlated areas. When interested in forecasting the aggregate across the various areas, a question that naturally arises is whether gains in efficiency can be obtained using a direct approach or an indirect approach. This issue has been recently considered in Giacomini and Granger (2004), where it is shown that stationary space-time AR(1, 1) models are relatively more efficient than traditional ARMAs and V ARs models in terms of forecasting accuracy. We extend these findings by considering a more general and realistic non-stationary context, where cointegration constraints in time are allowed to exist. A concrete application with monthly inflation rate for Euro-zone economies is presented