Bernstein-type Inequalities and Nonparametric Estimation under Near-Epoch Dependence

TL;DR

This paper develops Bernstein-type inequalities for irregularly-spaced NED random fields, introduces the concept of 'effective dimension' affecting inequality sharpness, and applies these results to establish optimal convergence rates for various nonparametric estimators.

Contribution

It derives new Bernstein-type inequalities reflecting 'effective dimension' for NED random fields and applies these to prove optimal convergence rates for kernel-based nonparametric estimators.

Findings

Established Bernstein-type inequalities for NED random fields.

Proved optimal convergence rates for local linear estimators.

Derived uniform convergence rates for density estimators.

Abstract

The major contributions of this paper lie in two aspects. Firstly, we focus on deriving Bernstein-type inequalities for both geometric and algebraic irregularly-spaced NED random fields, which contain time series as special case. Furthermore, by introducing the idea of "effective dimension" to the index set of random field, our results reflect that the sharpness of inequalities are only associated with this "effective dimension". Up to the best of our knowledge, our paper may be the first one reflecting this phenomenon. Hence, the first contribution of this paper can be more or less regarded as an update of the pioneering work from \citeA{xu2018sieve}. Additionally, as a corollary of our first contribution, a Bernstein-type inequality for geometric irregularly-spaced $\alpha$-mixing random fields is also obtained. The second aspect of our contributions is that, based on the inequalities mentioned above, we show the $L_{\infty}$ convergence rate of the many interesting kernel-based nonparametric estimators. To do this, two deviation inequalities for the supreme of empirical process are derived under NED and $\alpha$-mixing conditions respectively. Then, for irregularly-spaced NED random fields, we prove the attainability of optimal rate for local linear estimator of nonparametric regression, which refreshes another pioneering work on this topic, \citeA{jenish2012nonparametric}. Subsequently, we analyze the uniform convergence rate of uni-modal regression under the same NED conditions as well. Furthermore, by following the guide of \citeA{rigollet2009optimal}, we also prove that the kernel-based plug-in density level set estimator could be optimal up to a logarithm factor. Meanwhile, when the data is collected from $\alpha$-mixing random fields, we also derive the uniform convergence rate of a simple local polynomial density estimator \cite{cattaneo2020simple}.