Asymptotic locations of bounded and unbounded eigenvalues of sample correlation matrices of certain factor models -- application to a components retention rule

Let the dimension of data and the sample size tend to with . The spectral properties of a sample correlation matrix and a sample covariance matrix are asymptotically equal whenever the population correlation matrix is bounded (El Karoui 2009). We demonstrate this also for general linear models for unbounded , by examining the behavior of the singular values of multiplicatively perturbed matrices. By this, we establish: Given a factor model of an idiosyncratic noise variance and a rank- factor loading matrix which rows all have common Euclidean norm . Then, the th largest eigenvalues of satisfy almost surely: (1) diverges, (2) for the th largest singular value of , and (3) for . Whenever is much larger than , then broken-stick rule (Frontier 1976, Jackson 1993), which estimates by a random partition (Holst 1980) of , tends to (a.s.). We also provide a natural factor model where the rule tends to "essential rank" of (a.s.) which is smaller than .
View on arXiv