Rank-based Optimal Testing for Semiparametric Cointegration Models (with Marc Hallin and Ramon van den Akker).
Abstract:
This paper discusses asymptotically efficient testing for hypotheses about the cointegrating rank or about the cointegrating vectors in a cointegration model with elliptically distributed innovations. The model is semiparametric in the sense that the radial density and scatter matrix of the innovations are unknown. The tests developed use a multivariate notion of ranks and are asymptotically distribution-free. The tests are build on a reference density that can be chosen freely. Validity of the test in terms of asymptotic size is guaranteed irrespective of the reference density. The asymptotic power of the test improves when the chosen reference density happens to be closer to the actual innovation density. A suitably estimated reference density leads to fully semiparametrically efficient tests.