Revisiting cointegration tests : power versus frequency - some further Monte Carlo results

Abstract

We study the effect of increasing the frequency of observations and the data span on alternative cointegration tests (Engle-Granger, Phillips-Ouliaris and Johansen), we consider systems with two, three and four variables via Monte Carlo simulations. We find that when both the data length and the frequency vary, the power of the tests depends more on the sample length. In addition, we explore the behaviour of unit root tests and the Engle-Granger and Johansen methods when explosive processes are included. The results show that the performance of the tests depends on the kind of the type of the explosion.