Blaming suicide on NASA and divorce on margarine: the hazard of using cointegration to derive inference on spurious correlation

Contrary to common belief, cointegration testing may not distinguish between spurious relations and genuine ones. It is demonstrated that highly correlated series appear as cointegrated, even though common sense tells us that the underlying relation does not make sense. Empirical testing using simulated data, data from daily life and historical data on interest rates shows that cointegration may fail not only to detect spurious correlation but also to capture cointegration in a genuine relation. Cointegration testing to reveal spurious correlation can only be used in conjunction with theory, common sense and intuition.