Extreme Value Theory (EVT) and Applications

The existing estimators in extreme value theory suffer from high bias and variance. In some cases a reduction of these values was attained.

We expect to continue developing and studying, both asymptotically and by Monte Carlo methods, estimators of parameters and other related quantities in modeling rare events – e.g. the tail index, the extremal index, high quantiles and probability of failure sets – both in an independence and in a dependence framework. We also aim to apply the theoretical results to real data, in a joint work with researchers belonging to other Units in the university.

The application of the generalized jackknife methodology has proved to result in the bias reduction of estimators of the tail index in heavy tailed models. We aim to extend the application of this methodology to estimators defined in a broader class of distributions. The kernel estimators used in density estimation and in smoothing may also be applied as a flexible tool to reduce bias and/or volatility.
The analysis of dependence regarded in a temporal or spatial approach in univariate and multivariate EVT is also of interest. We are particularly interested in problems like the estimation of the extremal index, involving bootstrap techniques, and modeling dependence and asymptotic dependence.

Possible applications are e.g. in environmental problems, like large fires and hydrology data, as well as in finance and insurance. 

Members:

Ana Ferreira Henriques
M. João Martins
M. Manuela Neves Figueiredo