A new confidence interval for all characteristic roots of a covariance matrix (Q2271699)

An approximate confidence interval is constructed which covers all eigenvalues of the covariance matrix of i.i.d. Gaussian observations. It is based on the normal approximation to the true distribution of the cube roots of the sample covariance matrix eigenvalues. Results of simulations demonstrate that the proposed interval is quite conservative and can be used for sample sizes \(n\geq 100\). An asymptotic expansion for the correlation coefficient of the largest and smallest eigenvalues of a sample covariance matrix is given.