Asymptotic Distribution of Restricted Canonical Correlations and Relevant Resampling Methods

As restricted canonical correlation with a nonnegativity condition on the coefficients depend only on the covariance matrix, their sample counterparts can be obtained from the sample covariance matrix. For such estimators, asymptotic normality results are established, and the role of resampling methods in this context is critically examined. The effectiveness of the usual jackknife and bootstrap methods is studied analytically, and the findings are supplemented by numerical studies.