A Jarque-Bera test for sphericity of a large-dimensional covariance matrix

This article provides a new test for sphericity of the covariance matrix of a d-dimensional multinormal population X Nd. This test is applicable if the sample size, n + 1, and d both go to infinity while d/n ! y 2 (0,1), provided that the limits of tr(k)/d, k = 1, . . . , 8, are finite. The main idea of this test is to check whether the empirical eigenvalue distribution of a suitably standardized sample covariance matrix obeys the semicircle law. Due to similarities of the semicircle law to the normal distribution, the proposed test statistic is of the type of the Jarque-Bera test statistic. Simulation results show that the new sphericity test outperforms the tests from the current literature for certain local alternatives if y is small. -- Test for covariance matrix ; High-dimensional data ; Spectral distribution ; Semicircle law ; Free cumulant ; Jarque-Bera test