Chew-Seng Chee

Department of Statistics, University of Auckland

Semiparametric mixture models for symmetric density estimation

We present a general semiparametric framework based on mixtures for univariate symmetric density estimation and propose a semiparametric mixture symmetric density estimator (spsym). The performance of the estimator hinges on appropriate choice of the tuning parameter. To this end, we introduce a simple strategy for selecting the tuning parameter in practical implementations. Since the spsym is essentially a semiparametric mixture model, this allows us to take advantage of the CNM-MS algorithm used to fit semiparametric mixture models and with minor modification the algorithm can be used for the spsym computation. A simple real example shows that the mixture-based method provides an attractive complement to the traditional kernel-based method. The performances of the mixture-based and kernel-based methods are illustrated through a simulation study.