8
8

Logarithmic law of large random correlation matrices

Abstract

Consider a random vector y=Σ1/2x\mathbf{y}=\mathbf{\Sigma}^{1/2}\mathbf{x}, where the pp elements of the vector x\mathbf{x} are i.i.d. real-valued random variables with zero mean and finite fourth moment, and Σ1/2\mathbf{\Sigma}^{1/2} is a deterministic p×pp\times p matrix such that the spectral norm of the population correlation matrix R\mathbf{R} of y\mathbf{y} is uniformly bounded. In this paper, we find that the log determinant of the sample correlation matrix R^\hat{\mathbf{R}} based on a sample of size nn from the distribution of y\mathbf{y} satisfies a CLT (central limit theorem) for p/nγ(0,1]p/n\to \gamma\in (0, 1] and pnp\leq n. Explicit formulas for the asymptotic mean and variance are provided. In case the mean of y\mathbf{y} is unknown, we show that after recentering by the empirical mean the obtained CLT holds with a shift in the asymptotic mean. This result is of independent interest in both large dimensional random matrix theory and high-dimensional statistical literature of large sample correlation matrices for non-normal data. At last, the obtained findings are applied for testing of uncorrelatedness of pp random variables. Surprisingly, in the null case R=I\mathbf{R}=\mathbf{I}, the test statistic becomes completely pivotal and the extensive simulations show that the obtained CLT also holds if the moments of order four do not exist at all, which conjectures a promising and robust test statistic for heavy-tailed high-dimensional data.

View on arXiv
Comments on this paper