Common nonstationary components of asset prices (Q1102850)

Portfolio separation in a multi-period general equilibrium context implies that asset prices are collinear. In economies that do not exhibit separation, but that move close to separation, `approximate collinearity' or cointegration emerges as a statistical model of actual data. A new test for cointegration is proposed. It is based on time-series canonical correlation analysis and solves the problem of unidentified parameters under the null hypothesis and of identification of the cointegrating vectors when more than two time series are investigated. An empirical analysis of prices of five size-based portfolios and five industry-based portfolios of common stock reveals substantial evidence of cointegration.