This paper examines the forecast accuracy of cointegrated vector autoregressive models when confronted with extreme observations at the end of the sample period. It focuses on comparing two outlier correction methods, additive outliers and innovational outliers, within a forecasting framework...