A Simulation Exercise

Engle and Yoo (1987) use a small simulation exercise to compare the forecasting ability of a bivariate unrestricted VAR including just one lag with that of an ECM estimated by the two-step procedure proposed by Engle and Granger (1987). The restricted ECM outperforms, in terms of mean squared error, its unrestricted opponent if the number of steps ahead predicted is six or greater. They argue that the better performance of the unrestricted VAR in the short run might be due to the long run character of the constraints. Hence, according to their argument, the restricted ECM would be misspecified in the short run. Whereas, it would be correctly specified once the forecasting horizon were such that the cointegrating restrictions became true.