Forecasting trade

This paper develops a set of time series models to provide short-term forecasts (6 to 18 months ahead) of international trade both at the global level and for selected regions. Our results compare favourably to other forecasts, notably by the International Monetary Fund, as measured by standard evaluation measures, such as the root mean square forecast error. In comparison to other models, our approach offers several methodological advantages, inter alia, a focus on import growth as the core variable, the avoidance of certain difficulties affecting the performance of structural models, the selection of variables and lags on the basis of theoretical considerations and empirical testing as well as a full documentation of the modelling process.