A re-examination of the fragility of evidence from cointegration-based tests of foreign exchange market efficiency

We re-examine Sephton and Larsen's (1991) conclusion that cointegration-based tests for market efficiency suffer from temporal instability. We improve upon their research by i) including a drift term in the vector error correction model (VECM) in the Johansen procedure, ii) correcting the likelihood ratio test statistic for finite-sample bias, and iii) fitting the model over longer data sets. We show that instability of the Johansen cointegration tests mostly disappears after accounting for these two factors. The evidence is even more stable in favour of no cointegration when we apply our analysis to longer data sets