Strongly consistent determination of cointegrating rank via canonical correlations

This article is concerned with the statistical analysis of nonstationary, cointegrated time series. The estimation of the cointegrating structure of such time series is considered, and the problem of identifying the cointegrating rank is addressed. A methodology is presented that leads to strongly consistent estimates of this quantity. The identification is based on a canonical correlation analysis of the original variables and presents an alternative approach to those currently in vogue. The procedures are easily implemented and the practical relevance of the results obtained, which are founded on asymptotic theory, is demonstrated by means of a small simulation study.