| Review: |
These are the invited papers from a seminar in Gothenburg in December 2001. The aim of the seminar was to acquaint young scientists with the latest research in statistical science. The central paradigm of extreme value theory is semiparametric: you cannot trust standard statistical modelling by normal, lognormal, Weibull, or other distributions all the way to the extreme tails and maxima. However, nonparametric methods cannot be used either, as interest centres on more extreme events than those that one has already encountered. The solution is to use semiparametric models that specify only the distributional shapes of maxima, as the extreme value distributions, or of extreme tails, as the general Pareto distributions. The book begins by surveying how this paradigm answers a variety of questions of interest to applied scientists in climatology, insurance and finance. It goes on to look at how the theory can be applied to extreme events, such as the 1999 rainfall in Venezuela, risk management in finance and its use in data network modelling. The last chapter looks at multivariate extreme value distributions and the problem of measuring extremal dependence. |