The potential of a small model
We run a number of experiments with a small two variable VAR model of GDP growth and unemployment with both quarterly and yearly data. We compare the forecasts of these simple models with the published forecasts of the CPB and we conclude that there is not much di erence. We then show how easy it is to evaluate the usefulness of a given variable for forecasting by extending the model to include world trade. Perfect knowledge of future world trade growth would help considerably but is obviously not available at the time the forecasts were made. The available world trade data doesn't improve the forecasts. Finally we also show how quick and exible measures of the output gap can be constructed.