Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Kernel Density Estimation and Convolution Revisited

Nicholas Tenkorang, Kwesi Appau Ohene-Obeng, Xiaogang Su

TL;DR

This work recasts kernel density estimation through a principled convolutional lens and introduces SHIDE, a practical density estimator that generates bounded-noise pseudo-data around observations and uses histogram refinement plus natural cubic spline interpolation to estimate densities. SHIDE extends KDE to constrained domains via additive and multiplicative convolution schemes, achieves pointwise consistency and the classical $n^{-4/5}$ convergence rate, and mitigates boundary bias through a transformation-based approach and boundary-preserving splines. Two bandwidth selectors are developed: an AMISE-optimal rule and a percentile-based calibrated spacing rule, which are asymptotically equivalent. Through theoretical analysis and comprehensive simulations, SHIDE demonstrates competitive performance relative to KDE across a range of distributions, with particular strengths for bounded and heavy-tailed data and natural handling of boundaries. The framework opens avenues for multivariate extensions and alternative spline regularization while maintaining computational efficiency.

Abstract

Kernel Density Estimation (KDE) is a cornerstone of nonparametric statistics, yet it remains sensitive to bandwidth choice, boundary bias, and computational inefficiency. This study revisits KDE through a principled convolutional framework, providing an intuitive model-based derivation that naturally extends to constrained domains, such as positive-valued random variables. Building on this perspective, we introduce SHIDE (Simulation and Histogram Interpolation for Density Estimation), a novel and computationally efficient density estimator that generates pseudo-data by adding bounded noise to observations and applies spline interpolation to the resulting histogram. The noise is sampled from a class of bounded polynomial kernel densities, constructed through convolutions of uniform distributions, with a natural bandwidth parameter defined by the kernel's support bound. We establish the theoretical properties of SHIDE, including pointwise consistency, bias-variance decomposition, and asymptotic MISE, showing that SHIDE attains the classical $n^{-4/5}$ convergence rate while mitigating boundary bias. Two data-driven bandwidth selection methods are developed, an AMISE-optimal rule and a percentile-based alternative, which are shown to be asymptotically equivalent. Extensive simulations demonstrate that SHIDE performs comparably to or surpasses KDE across a broad range of models, with particular advantages for bounded and heavy-tailed distributions. These results highlight SHIDE as a theoretically grounded and practically robust alternative to traditional KDE.

Kernel Density Estimation and Convolution Revisited

TL;DR

This work recasts kernel density estimation through a principled convolutional lens and introduces SHIDE, a practical density estimator that generates bounded-noise pseudo-data around observations and uses histogram refinement plus natural cubic spline interpolation to estimate densities. SHIDE extends KDE to constrained domains via additive and multiplicative convolution schemes, achieves pointwise consistency and the classical convergence rate, and mitigates boundary bias through a transformation-based approach and boundary-preserving splines. Two bandwidth selectors are developed: an AMISE-optimal rule and a percentile-based calibrated spacing rule, which are asymptotically equivalent. Through theoretical analysis and comprehensive simulations, SHIDE demonstrates competitive performance relative to KDE across a range of distributions, with particular strengths for bounded and heavy-tailed data and natural handling of boundaries. The framework opens avenues for multivariate extensions and alternative spline regularization while maintaining computational efficiency.

Abstract

Kernel Density Estimation (KDE) is a cornerstone of nonparametric statistics, yet it remains sensitive to bandwidth choice, boundary bias, and computational inefficiency. This study revisits KDE through a principled convolutional framework, providing an intuitive model-based derivation that naturally extends to constrained domains, such as positive-valued random variables. Building on this perspective, we introduce SHIDE (Simulation and Histogram Interpolation for Density Estimation), a novel and computationally efficient density estimator that generates pseudo-data by adding bounded noise to observations and applies spline interpolation to the resulting histogram. The noise is sampled from a class of bounded polynomial kernel densities, constructed through convolutions of uniform distributions, with a natural bandwidth parameter defined by the kernel's support bound. We establish the theoretical properties of SHIDE, including pointwise consistency, bias-variance decomposition, and asymptotic MISE, showing that SHIDE attains the classical convergence rate while mitigating boundary bias. Two data-driven bandwidth selection methods are developed, an AMISE-optimal rule and a percentile-based alternative, which are shown to be asymptotically equivalent. Extensive simulations demonstrate that SHIDE performs comparably to or surpasses KDE across a broad range of models, with particular advantages for bounded and heavy-tailed distributions. These results highlight SHIDE as a theoretically grounded and practically robust alternative to traditional KDE.
Paper Structure (28 sections, 9 theorems, 103 equations, 2 figures, 1 table)

This paper contains 28 sections, 9 theorems, 103 equations, 2 figures, 1 table.

Table of Contents

  1. Introduction
  2. KDE and Convolution
  3. Additive Convolution KDE
  4. Multiplicative Convolution KDE for Positive/Negative Variables
  5. SHIDE
  6. Pseudodata Simulation
  7. Histogram Interpolation with Splines
  8. Extension to Positive/Bounded Variables
  9. Theoretical Analysis
  10. Asymptotic Properties
  11. Comparison with KDE
  12. Bandwidth Selection
  13. AMISE-Optimal Selector
  14. Selector Based on Nearest-Neighbor Spacing
  15. Numerical Results
  16. ...and 13 more sections

Key Result

Proposition 1

Let $U_1, \ldots, U_k$ be independent and identically distributed random variables with the uniform distribution on $[-1/2, 1/2]$, i.e., with PDF $f_U(u) = I(-1/2 < u < 1/2).$ Define $V_k = \sum_{i=1}^k U_i,$ and let $f_k(v)$ denote the probability density function (PDF) of $V_k$. Then:

Figures (2)

  • Figure 1: The PDF and Characteristic functions of $V_m$ for $m=1, 2, 3.$.
  • Figure 2: Illustration of Observed and Simulated Data. The observed data consist of $n = 50$ values generated from a $t(3)$ distribution, exhibiting two apparent outliers. The simulated data comprise 500 values, created by adding random noise to each observed value.

Theorems & Definitions (18)

  • Proposition 1
  • Theorem 1
  • Theorem 2
  • Theorem 3
  • Proposition 2
  • Proposition 3
  • Theorem 4
  • Proposition 4
  • Proposition 5
  • proof
  • ...and 8 more