Dual $\phi$-divergences estimation in normal models

TL;DR

This paper investigates a class of robust estimators derived from dual $\,\phi$-divergences in normal models, demonstrating their efficiency and robustness as alternatives to maximum likelihood estimators.

Contribution

It introduces and empirically studies a new class of dual $\,\phi$-divergence estimators for normal models, highlighting their robustness and efficiency.

Findings

Estimators are efficient at the true model.

They offer robustness advantages over maximum likelihood.

Empirical comparisons show competitive performance.

Abstract

A class of robust estimators which are obtained from dual representation of $\phi$-divergences, are studied empirically for the normal location model. Members of this class of estimators are compared, and it is found that they are efficient at the true model and offer an attractive alternative to the maximum likelihood, in term of robustness .