Cointegration and predictability of asset prices

This article argues that whatever concerns one might have about the identification of a cointegrating relationship with market inefficiency, cointegration tests can still be usefully employed to investigate the predictability of asset prices. We examine the case of n-dimensional systems, and show that the standard assumption made in the literature that cointegration implies predictability of all n asset prices is not valid. In the presence of r cointegrating vectors, only r prices are predictable, and standard Wald or LM tests carried out within Johansen's Maximum Likelihood (ML) framework can be used to establish for which prices unpredictability does not hold. Similarly, one can distinguish between (joint) unpredictability in the short- and long-run when asset prices are driven by fundamentals. (C) 1998 Elsevier Science Ltd. All rights reserved.