A growing disenchantment with conventional economic models has resulted in increased interest in forecasting with vector autoregressive (VAR) models. In this article, Roy H. Webb develops a statistical procedure for determining the best configuration of explanatory variables in the equations of a VAR model. The resulting model forecasts more accurately than a conventional VAR model and is comparable to VARs improved through other popular methods. In addition, Webb's procedure lets the data determine the form of the model and reduces the role of judgment in specifying equations, consistent with the theoretical spirit of VAR models.
Amanda L. Kramer