Asymptotic theory for canonical correlation analysis

Abstract

The asymptotic distribution of the sample canonical correlations and coefficients of the canonical variates is obtained when the nonzero population canonical correlations are distinct and sampling is from the normal distribution. The asymptotic distributions are also obtained for reduced rank regression when one set of variables is treated as independent (stochastic or nonstochastic) and the other set as dependent. Earlier work is corrected. © 1999 Academic Press.