Robust estimation of economic indicators from survey samples based on Pareto tail modeling

Motivated by a practical application, this paper investigates robust estimation of economic indicators from survey samples based on a semiparametric Pareto tail model. Economic performance is typically measured by a set of indicators, which are often estimated from survey data – the motivating example being the European indicators on social exclusion and poverty computed from the well known EU-SILC survey. Since economic data typically contain variables with heavily tailed distributions and additional extreme outliers, the idea is to use robust Pareto tail modeling to detect the extreme outliers and reduce their influence on the indicators. In the survey context, however, sample weights need to be considered when modeling the tail with a Pareto distribution such that the true distribution on the population level is accurately reflected. Therefore, the main methodological contribution is to adapt commonly used robust estimators for the parameters of the Pareto distribution to take sample weights into account. The resulting approach for robust estimation of indicators is then evaluated by means of a simulation study and applied in the context of estimating the Gini coefficient from EU-SILC data.