A1205
Title: Using EVT to test for outliers
Authors: Nathaniel Gbenro - Ensea-Abidjan (Cote d'Ivoire) [presenting]
Abdou Ka Diongue - Gaston Berger (Senegal)
El hadji Deme - Gaston Berger University (Senegal)
Abstract: The focus is on the identification of aberrant values by using extreme value theory (EVT) techniques. A new approach is proposed in the identification process of outliers. In this framework, the algorithm suggested by a prior study is extended to Gaussian or non-Gaussian distributions. Two empirical applications have been established to illustrate the efficiency of the approach. First, simulated data is used from Gaussian distributions, and the methodology is compared to the one proposed by a past study. In the second application, simulated data from various non-Gaussian distributions is also used to study the performance of our approach. The results suggest that the EVT outliers' test has good power when the sample size is large.
