Research in the statistical analysis of extreme values hasflourished over the past decade: new probability models, inferenceand data analysis. : Statistics of Extremes: Theory and Applications (): Jan Beirlant, Yuri Goegebeur, Johan Segers, Jozef L. Teugels, Daniel De Waal. Statistics of Extremes Theory and ApplicationsJan Beirlant, Yuri Goegebeur, and Jozef Teugels University Center of Sta.
|Published (Last):||1 April 2007|
|PDF File Size:||8.40 Mb|
|ePub File Size:||7.56 Mb|
|Price:||Free* [*Free Regsitration Required]|
The semi-parametric point of view then results in the appearance of an asymptotic bias in the results. An insurance company will always safeguard itself against portfolio contamination caused by claims that should be considered as extreme rather than average.
Statistics of Extremes: Theory and Applications
The ordered residuals scatter quite well around the first diagonal, giving evidence of a good fit of the GP regression model to the Ca exceedances. It will turn out that the sign of the EVI is the dominating factor in the description of the tail of the underlying distribution F. For the Secura data, the generalized quantile plot is given in Figure 6. If G is a max-stable staistics function with spectral measure S as in 8.
The ultimate goal of this case study is to provide the participating reinsurance companies beielant an objective statistical analysis in order to assist in the pricing of the unlimited excess-loss layer above an operational priority R.
The general linear model links the dependent variable to the covariates in an approximate linear way: The statistical aspects had to wait even longer before they received the necessary attention. Let us pause to prove the following basic limiting result.
Apart from being a volume of worked exercises in its own right, it is also a solutions manual for exercises and problems appering in textbook Grimmett G. In other areas of application of extreme value statistics such as hydrology, the use of Pareto models appears to be much beirlannt systematic.
These plots are given in Figures 6. For this purpose, the usual bootstrap does not work properly, especially because it seriously underestimates bias. The methods based on the regression models above such beiralnt ii and iii ask most computing effort.
In practice, this often turns out to be a hard job. A typical way to establish convergence of point processes is via convergence of Laplace functionals. The fit of the exponential distribution can be further evaluated on the basis of the exponential QQ-plot given in Figure 1.
Claim sizes are expressed as a fraction of the sum insured. In this situation, the observations will be first transformed to i. Indeed, limit and integral can be interchanged there using the dominated convergence theorem heirlant on the Potter bounds 2.
These generalized residuals then form the basis for applying 7. In general, better approximations can be obtained by the profile likelihood function. In this way, for each value of k0an estimator of kopt is obtained.
We felt that, for practical purposes, we could safely restrict ourselves to the case where the underlying random variables are sufficiently continuous. The sequential method appears to be the fastest overall.
From the exponential quantile plot, a point of inflection with different slopes to the left and to the right can be detected. If k oc chosen too small, the resulting estimators will have a high variance.
By the property in the previous statkstics, what is needed is convergence of the dependence structure, to be specified next. The beitlant common type of point processes are Poisson processes. For a detailed description of these data, we refer to Goegebeur et al.
To keep a smooth flow in the development, technical details within a chapter are deferred to the last section of that chapter. This problem can be circumvented by taking resamples of smaller size than the original one and linking the bootstrap estimates for the optimal subsample fraction to kn,opt for the full sample. The covariates are assumed non-random. The above-described maximum likelihood procedure is then applied to a sample of maxima. Now we should turn to the domain of attraction problem: We evaluate the fit of both GP regression models based on an exponential quantile plot of the generalized residuals 7.
However, by transforming the dependent variables, the dependence on the covariates may be at least partly removed. As in the previous chapter, one can now establish an estimation procedure analogous to that induced by the Hill estimator.
Richard L. Smith
The method is fundamental in the study of limit theorems and generally in assessing the quality of statisgics to a given probabilistic model. It is essentially due to de Haan and Resnickwho considered the special case where the two norms are equal to the Euclidean norm.
This can be found in the second-order theory discussed in Chapter 3. After rewriting the limiting result in its historical form, we give some sufficient conditions. This parameter is important in modelling water course systems, urban drainage and water runoff. The penalized log-likelihood function is a difference of two terms.
An alternative solution is to summarize the main properties of the dependence structure in a number of well-chosen coefficients that give a rough but representative picture of the full dependence structure. The proofs of the results in this section are probably among the most technical points in the discussion of the attraction problem for the maximum.
In Chapter 3, this will be shown not to be the case overall. These are given in Figure 6. The measure S is called the spectral measure.