16
30

Asymptotically Efficient Estimation of Smooth Functionals of Covariance Operators

Abstract

Let XX be a centered Gaussian random variable in a separable Hilbert space H{\mathbb H} with covariance operator Σ.\Sigma. We study a problem of estimation of a smooth functional of Σ\Sigma based on a sample X1,,XnX_1,\dots ,X_n of nn independent observations of X.X. More specifically, we are interested in functionals of the form f(Σ),B,\langle f(\Sigma), B\rangle, where f:RRf:{\mathbb R}\mapsto {\mathbb R} is a smooth function and BB is a nuclear operator in H.{\mathbb H}. We prove concentration and normal approximation bounds for plug-in estimator f(Σ^),B,\langle f(\hat \Sigma),B\rangle, Σ^:=n1j=1nXjXj\hat \Sigma:=n^{-1}\sum_{j=1}^n X_j\otimes X_j being the sample covariance based on X1,,Xn.X_1,\dots, X_n. These bounds show that f(Σ^),B\langle f(\hat \Sigma),B\rangle is an asymptotically normal estimator of its expectation EΣf(Σ^),B{\mathbb E}_{\Sigma} \langle f(\hat \Sigma),B\rangle (rather than of parameter of interest f(Σ),B\langle f(\Sigma),B\rangle) with a parametric convergence rate O(n1/2)O(n^{-1/2}) provided that the effective rank r(Σ):=tr(Σ)Σ{\bf r}(\Sigma):= \frac{{\bf tr}(\Sigma)}{\|\Sigma\|} (tr(Σ){\rm tr}(\Sigma) being the trace and Σ\|\Sigma\| being the operator norm of Σ\Sigma) satisfies the assumption r(Σ)=o(n).{\bf r}(\Sigma)=o(n). At the same time, we show that the bias of this estimator is typically as large as r(Σ)n\frac{{\bf r}(\Sigma)}{n} (which is larger than n1/2n^{-1/2} if r(Σ)n1/2{\bf r}(\Sigma)\geq n^{1/2}). In the case when H{\mathbb H} is finite-dimensional space of dimension d=o(n),d=o(n), we develop a method of bias reduction and construct an estimator h(Σ^),B\langle h(\hat \Sigma),B\rangle of f(Σ),B\langle f(\Sigma),B\rangle that is asymptotically normal with convergence rate O(n1/2).O(n^{-1/2}). Moreover, we study asymptotic properties of the risk of this estimator and prove minimax lower bounds for arbitrary estimators showing the asymptotic efficiency of h(Σ^),B\langle h(\hat \Sigma),B\rangle in a semi-parametric sense.

View on arXiv
Comments on this paper