Nonparametric cointegration analysis

In this paper we propose consistent cointegration tests, and estimators of a basis of the space of cointegrating vectors, that do not used specification of the data-generating process, apart from some mild regularity conditions, or estimation of structural and/or nuisance parameters. This nonparametric approach is in the same spirit as Johansenʹs LR method in that the test statistics involved are obtained from the solutions of a generalized eigenvalue problem, and the hypotheses to be tested are the same, but in our case the two matrices in the generalized eigenvalue problem involved are constructed independently of the data-generating process. We compare our approach empirically as well as by a limited Monte Carlo simulation with Johansenʹs approach, using the series for In(wages) and In(GNP) from the extended Nelson-Plosser data.