It is argued that size distortions and power properties of likelihood based tests for cointegration are so poor in many small sample situstions that alternative are necessary for analysing typical macroeconomic relationships.