Extreme value prediction: a computational approach

Abstract

In this paper we faced the problem of constructing prediction limits for series of extreme values, comparing procedures proposed by Fonseca et al (2012) and Hall et al (1999). Both prediction methods were applied to generalize extreme value distribution. A simulation study was performed for comparing the previous methods to the estimative approach applied to Generalized Extreme Value distribution and Generalized Pareto distribution. Forecasting methodologies were finally applied to series of real data related to rainfall in the UK, taken from the Met Office Integrated Data Archive System (MIDAS) database. Despite the presence of some computational problems due to the calculation accuracy required, results of the simulations joined to what was seen on the real data have shown the improvement of the considered prediction methods on the estimative approaches.