On the estimation of the structure parameter of a normal distribution of order p

Abstract

In this paper we compare four different approaches to estimate the structure parameter of a normal distribution of order p (often called exponential power distribution). In particular, we have considered the maximization of the log-likelihood, of the profile log-likelihood, of the conditional profile log-likelihood and a method based on an index of kurtosis. The results of a simulation study seem to indicate the latter approach as the best.