Is world trade data useful when forecasting a small open economy?

By using pooled VAR forecasts we have attempted to minimise a number of issues that make inference difficult regarding the importance of individual variables for forecast accuracy using traditional methods. We find that the most accurate forecasts for GDP growth can typically be made without using the world trade data that would have been available at the time the forecasts were made. Furthermore, this paper has shown that world trade data also doesn't improve forecast accuracy for export growth, the component of GDP most closely linked to world trade. As a robustness test we also repeated the forecasting competition using direct multi-period forecasts (see Pesaran et al. (2011) and Marcellino et al. (2006) for recent examples) but they were statistically no more accurate than the standard VAR approach. We examine why world trade data doesn't help and find a non- linear relationship between the accuracy of world trade forecasts and GDP forecasts. Whilst perfect foresight of world trade would improve forecast accuracy for both GDP and export growth, the improvement in world trade forecasts from the pooled VAR models relative to a 'no knowledge' benchmark does not.