Imposing cointegration on a forecasting system, if cointegration is present, is believed to improve long horizon forecasts. Contrary to this belief, at long horizons nothing is lost by ignoring cointegration when the forecasts are evaluated using standard multivariate forecast accuracy measures. In fact simple univariate Box-Jenkins forecasts are just as accurate. These results potentially highlight an important deficiency of standard forecast accuracy measures – they fail to value the maintenance of cointegration relationships among variables. The paper suggests alternatives that explicitly do so.