Statistika: Statistics and Economy Journal (Mar 2017)
Statistical Inference Based on L-Moments
Abstract
To overcome drawbacks of central moments and comoment matrices usually used to characterize univariate and multivariate distributions, respectively, their generalization, termed L-moments, has been proposed. L-moments of all orders are defined for any random variable or vector with finite mean. L-moments have been widely employed in the past 20 years in statistical inference. The aim of the paper is to present the review of the theory of L-moments and to illustrate their application in parameter estimating and hypothesis testing. The problem of estimating the three-parameter generalized Pareto distribution’s (GPD) parameters that is generally used in modelling extreme events is considered. A small simulation study is performed to show the superiority of the L-moment method in some cases. Because nowadays L-moments are often employed in estimating extreme events by regional approaches, the focus is on the key assumption of index-flood based regional frequency analysis (RFA), that is homogeneity testing. The benefits of the nonparametric L-moment homogeneity test are implemented on extreme meteorological events observed in the Czech Republic.
