61
8

Estimation of smooth functionals in high-dimensional models: bootstrap chains and Gaussian approximation

Abstract

Let X(n)X^{(n)} be an observation sampled from a distribution Pθ(n)P_{\theta}^{(n)} with an unknown parameter θ,\theta, θ\theta being a vector in a Banach space EE (most often, a high-dimensional space of dimension dd). We study the problem of estimation of f(θ)f(\theta) for a functional f:ERf:E\mapsto {\mathbb R} of some smoothness s>0s>0 based on an observation X(n)Pθ(n).X^{(n)}\sim P_{\theta}^{(n)}. Assuming that there exists an estimator θ^n=θ^n(X(n))\hat \theta_n=\hat \theta_n(X^{(n)}) of parameter θ\theta such that n(θ^nθ)\sqrt{n}(\hat \theta_n-\theta) is sufficiently close in distribution to a mean zero Gaussian random vector in E,E, we construct a functional g:ERg:E\mapsto {\mathbb R} such that g(θ^n)g(\hat \theta_n) is an asymptotically normal estimator of f(θ)f(\theta) with n\sqrt{n} rate provided that s>11αs>\frac{1}{1-\alpha} and dnαd\leq n^{\alpha} for some α(0,1).\alpha\in (0,1). We also derive general upper bounds on Orlicz norm error rates for estimator g(θ^)g(\hat \theta) depending on smoothness s,s, dimension d,d, sample size nn and the accuracy of normal approximation of n(θ^nθ).\sqrt{n}(\hat \theta_n-\theta). In particular, this approach yields asymptotically efficient estimators in some high-dimensional exponential models.

View on arXiv
Comments on this paper