In: Journal of Time Series Analysis, 19, No. 3, May 1998, pp. 377-378

Contents: ...

This book is complemented by an excellent statistical package on CD ROM, and much of the text is directed towards the facilities available in the program, which is called XTREMES. It has a Windows front end, which is at times very slick, and it is certainly fast when using the full 32-bit implementation under Windows 95. The package also comes with its own programming language, XPL, similiar to Pascal.

The book is both strengthened and limited by its close association to the program. One would of course expect to find very little on the underlying limit theorems - of which there are naturally very many appertaining to this field - in the program. Correspondingly, there is very little about them in the book. The various extreme value distributions are largely accepted by face value and then the practicalities of statistical inference for data arising from such distributions is carefully described and implemented in an easy to use manner in the computer package. This package is perhaps too self-contained, in that it is very much centred on its own excellent facilities for simulating data. Indeed the new user is recommended to start with such data rather than use their own.

The book is however much more than a manual for the program and it would provide an excellent introduction to the subject for any statistician looking for a practical introduction to the subject of extreme values. It has some material on multivariate extreme distributions, but very little on multivariate data in the sense of covariates associated with univariate extremes. Regression is dealt with solely as an exercise in de-trending data, so as to produce numbers which can then be analysed using extreme value techniques.

P.J. Laycock, Manchester