A PLS Based Approach to Cointegration Analysis

Abstract

This paper addresses the testing for cointegrating vectors and the estimation
of cointegrating relations using Partial Least Squares. Together with Harris
(1997) and Bossaerts (1988), the PLS approach relies on a method of
multivariate statistics and thus does not require identifying restrictions on
the cointegrating vectors or of a full specification of the short-run dynamics
of the process. The PLS estimator for the cointegrating vectors is found to be
super consistent and robust to heavy-tailed innovations. A test is provided for
the rank of cointegration which is assessed by means of Monte Carlo
simulation. A brief application to Mexican inflation data is also provided.