Semiparametric mixtures of symmetric distributions

TL;DR

This paper introduces a Fourier-based M-estimator for semiparametric mixtures of symmetric distributions, providing consistent parameter estimation without assuming a parametric form for the mixed distribution.

Contribution

It proposes a novel Fourier approach for estimating parameters in semiparametric symmetric mixture models, ensuring identifiability and consistency.

Findings

Estimators are square root consistent under mild conditions.

Finite-sample performance demonstrated via Monte Carlo simulations.

Method successfully applied to real data benchmark.

Abstract

We consider in this paper the semiparametric mixture of two distributions equal up to a shift parameter. The model is said to be semiparametric in the sense that the mixed distribution is not supposed to belong to a parametric family. In order to insure the identifiability of the model it is assumed that the mixed distribution is symmetric, the model being then defined by the mixing proportion, two location parameters, and the probability density function of the mixed distribution. We propose a new class of M-estimators of these parameters based on a Fourier approach, and prove that they are square root consistent under mild regularity conditions. Their finite-sample properties are illustrated by a Monte Carlo study and a benchmark real dataset is also studied with our method.