A note on boundary kernels for distribution function estimation

TL;DR

This paper explores an expanded class of boundary kernels for distribution function estimation, demonstrating improved performance over classical methods especially near boundaries where traditional estimators struggle.

Contribution

It introduces a broader class of boundary kernels that enhance distribution function estimation accuracy at boundaries, addressing limitations of classical kernel methods.

Findings

Enhanced boundary kernel class improves estimation accuracy

Better performance near distribution boundaries

Addresses limitations of classical kernel estimators

Abstract

The use of second order boundary kernels for distribution function estimation was recently addressed in the literature (C. Tenreiro, 2013, Boundary kernels for distribution function estimation, REVSTAT-Statistical Journal, 11, 169-190). In this note we return to the subject by considering an enlarged class of boundary kernels that shows it self to be especially performing when the classical kernel distribution function estimator suffers from severe boundary problems.