On estimation of the dimensionality in linear canonical analysis

Using a penalization of a statistic based on eigenvalues, we introduce a direct estimation method for the dimensionality in linear canonical analysis. Consistency of the resulting estimator is established and simulations show its finite sample performances.