Conditional quantile estimation using optimal quantization: a numerical study

Abstract : We construct a nonparametric estimator of conditional quantiles of Y given X = x using optimal quantization. Conditional quantiles are particularly of interest when the condi-tional mean is not representative of the impact of the covariable X on the dependent variable Y . L p -norm optimal quantization is a discretizing method used since the 1950's in engineer-ing. It allows to construct the best approximation of a continuous law with a discrete law with support of size N . The aim of this work is then to use optimal quantization to construct con-ditional quantile estimators. We study the convergence of the approximation (N → ∞) and the consistency of the resulting estimator for this fixed-N approximation. This estimator was implemented in R in order to evaluate the numerical behavior and to compare it with existing methods.
Document type :
Conference papers
Complete list of metadatas

Cited literature [6 references]  Display  Hide  Download

https://hal.inria.fr/hal-01109009
Contributor : Isabelle Charlier <>
Submitted on : Tuesday, January 27, 2015 - 3:00:31 PM
Last modification on : Thursday, January 11, 2018 - 6:22:11 AM
Long-term archiving on : Friday, September 11, 2015 - 9:02:18 AM

File

CharlierIsabellesubmission.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : hal-01109009, version 1

Collections

Citation

Isabelle Charlier, Davy Paindaveine, Jérôme Saracco. Conditional quantile estimation using optimal quantization: a numerical study. International Conference on Computational Statistics (COMPSTAT'2014), Aug 2014, Genève, Switzerland. ⟨hal-01109009⟩

Share

Metrics

Record views

486

Files downloads

113