Central limit theorem for Hotelling's \(T^{2}\) statistic under large dimension (Q655585)

For a sample \(X_1, \ldots, X_n\) from a distribution on \(\mathbb{R}^p\) with mean \(\mu\), denote by \(\overline{x}\) the sample mean and by \(S\) the sample covariance matrix. Hotelling's statistic \( T^2 = n (\overline{x} - \mu_0)^T S^{-1} (\overline{x} - \mu_0)\) is a classical object in multivariate statistics, useful, e.g., for testing the hypothesis that \(\mu = \mu_0\) against \(\mu \neq \mu_0\). Nowadays, e.g., in a context of data mining, one is often interested in letting the number \(p\) of variables grow proportionally to the sample size \(n\). The paper under review shows that in this asymptotic regime, if the components of \(X_1\) are iid and have a finite fourth moment, then \(T^2\), suitably shifted and scaled, tends to a standard normal limit as \(n \to \infty\).