Optimal reparametrization and large sample likelihood inference for the location-scale skew-normal model (Q452837)

The skew normal distribution plays an important role in statistical modelling since its introduction by \textit{A. Azzalini} [Scand. J. Stat., Theory Appl. 12, 171--178 (1985; Zbl 0581.62014); Statistica 45, 199--208 (1986; Zbl 0606.62013)]. The distribution can be reparametrized in different ways. In the original deviation an asymmetry parameter was introduced but this parametrization presents some theoretical difficulties, especially with respect to the asymptotic properties of ML estimates. While other parameterizations proposed can solve partially this problem, it remains an interesting problem to consider parameterizations that allow for proving a certain range of properties for ML estimates and the related hypothesis testing procedures. This paper concerns asymptotic likelihood estimation and hypothesis testing for this univariate location-scale skew normal distribution. In particular, based on the special representation of the skew normal distribution a new parametrization is constructed under which the symmetry hypothesis is invariant. Then it is exploited how this parametrization is related to the centered parametrization of Azzalini. Finding such an optimal parametrization one can find simple limit behavior of the maximum likelihood estimator and also to derive the asymptotic distribution of the likelihood ratio test statistic associated with the symmetry hypothesis which is a composite one. Similar results for non-optimal representations are discussed.