Let $X_1,\ldots,X_n$ be an i.i.d sample from a distribution on $\mathbb R^p$ with mean $\mu = 0 \in \mathbb R^p$ and $p$-by-$p$ covariance matrix $\Sigma$ of rank $r \le p$. Consider the centered Hotelling $T^2$ statistic defined by $T_n^2 := n\hat{\mu}_n^T\Sigma_n^{\dagger}\hat{\mu}_n$, $\hat{\mu}_n$ and $\hat{\Sigma}_n$ are the empirical mean vector and covariance matrix, respectively. It is well-known that $H_n$ converges in law to $\chi^2_{(r)}$, a chi-squared distribution with $r$ degrees of freedom (Sepanski 1994).


What are good non-asymptotic bounds for $P(T_n^2 \ge \epsilon)$ ?

I'm hoping for a Berry-Esseen-type bound of the form $\sup_{\epsilon > 0}|P(T^2_n \ge \epsilon) - P(\chi^2_{(r)} \ge \epsilon)| \le C/n^\delta$, for some absolute constants $C > 0$ and $\delta > 0$ (rough guess: $\delta \in [1/6,1/2]$).


A non-asymptotic tail-bound is available in the noncentral case $\mu \ne 0$. See Hotelling non-centralisé, e.g Theorem 4.24 of this paper.


Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Browse other questions tagged or ask your own question.