Definition of the Subject
Extreme value theory is concerned with the statistical properties of the extreme events related to a random variable (see Fig. 1), and the understanding and applications of their probability distributions. The methods and the practical use of such a theory have been developed in the last 60 years; though, many complex real-life problems have only recently been tackled. Many disciplines use the tools of extreme value theory including meteorology, hydrology, ocean wave modeling, and finance to name just a few.
Abbreviations
- Random variable:
-
When a coin is tossed two random outcomes are permitted: head or tail. These outcomes can be mapped to numbers in a process which defines a “random variable”: for instance, “head” and “tail” could be mapped respectively to +1 and −1. More generally, any function mapping the outcomes of a random process to real numbers is defined a random variable (Gnedenko 1998). More technically, a random variable is any function from a probability space to some measurable space, i.e., the space of admitted values of the variable, e.g., real numbers with the Borel σ‑algebra. The amount of rainfall in a day or the daily price variation of a stock are two more examples. It is worth to stress that, formally, the outcome of a given random experiment is not a random variable: the random variable is the function describing all the possible outcomes as numbers. Finally, two random variables are said independent when the outcome of either of them has no influence on the other.
- Probability distribution:
-
The probability of either outcomes, “head” and “tail,” in tossing a coin is 50 %. Similarly, a discrete random variable, X, with values {x 1, x 2, …} has an associate discrete probability distribution of occurrence {p 1, p 2, …}. More generally, for a random variable on real numbers, X, the corresponding probability distribution (Gnedenko 1998) is the function returning the probability to find a value of X within a given interval [x 1, x 2] (where x 1 and x 2 are real numbers): \( \Pr\;\left[{x}_1\le X\le {x}_2\right] \). In particular, the random variable, X, is fully characterized by its cumulative distribution function, F(x), which is: \( F(x)= \Pr\;\left[X<x\right] \) for any x in ℛ. The probability distribution density, f(x), can be often defined as the derivative of \( F(x):f(x)=\mathrm{d}F(x)/\mathrm{d}x \).
The probability distribution of two independent random variables, X and Y, is the product of the distributions, F X and F Y , of X and Y: \( F\left(x,y\right)\equiv \Pr\;\left[X<x;Y<y\right]= \Pr\;\left[X<x\right]\cdot \Pr\;\left[Y<y\right]={F}_X(x)\cdot {F}_Y(y) \).
- Expected value:
-
The expected value (Gnedenko 1998) of a random variable is its average outcome over many independent experiments. Consider, for instance, a discrete random variable, X, with values in the set {x 1, x 2, …} and the corresponding probability for each of these values {p 1, p 2, …}. In probability theory, the expected, or average, value of X (denoted E(X)) is just the sum: \( E(X)={\displaystyle \sum {x}_i}{p}_i \). For instance, if you have an asset which can give two returns {x 1, x 2} with probability {p 1, p 2}, its expected return is \( {x}_1{p}_1+{x}_2{p}_2 \).
In case we have a random variable defined on real numbers and F(x) is its probability distribution function, the expected value of X is: \( E(X)={\displaystyle \int X}\mathrm{d}F \). As for some F(x) the above integral may not exist, the “expected value” of a random variable is not always defined.
- Variance and moments:
-
The variance (Gnedenko 1998) of a probability distribution is a measure of the average deviations from the mean of the related random variable. In probability theory, the variance is usually defined as the mean squared deviation, \( E\left({\left(X-E(X)\right)}^2\right) \), i.e., the expected value of \( {\left(X-E(X)\right)}^2 \). The square root of the variance is named the standard deviation and is a more sensible measure of fluctuations of X around E(X). Alike E(X), for some distributions the variance may not exist. In general, the expected value of the kth power of X, E(X k), is called the kth moment of the distribution.
- The central limit theorem:
-
The central limit theorem (Gnedenko 1998) is a very important result in probability theory stating that the sum of N independent identically distributed random variables, with finite average and variance, has a Gaussian probability distribution in the limit \( N\to \infty \), irrespective of the underlying distributions of the random variables. The domain of attraction of the Gaussian as a limit distribution is, thus, very large and can explain why the Gaussian is so frequently encountered. The theorem is, in practice, very useful since many real random processes have a finite average and variance and are approximately independent.
Bibliography
Antal T, Droz M, Gyrgyi G, Racz Z (2001) Phys Rev Lett 87:240601
Berman SM (1964) Ann Math Stat 35:502
Bouchaud J-P, Mézard M (1997) J Phys A 30:7997
Bramwell ST, Holdsworth PCW, Pinton J-F (1998) Nature (London) 396:552
Bramwell ST, Christensen K, Fortin J-Y, Holdsworth PCW, Jensen HJ, Lise S, Lopez JM, Nicodemi M, Pinton J-F, Sellitto M (2000) Phys Rev Lett 84:3744
Bunde A, Kropp J, Schellnhuber H-J (eds) (2002) The science of disasters-climate disruptions, heart attacks, and market crashes. Springer, Berlin
Castellana M, Decelle A, Zarinelli E (2011) Extreme value statistics distributions in spin glasses. Phys Rev Lett 107:275701
Comtet A, Leboeuf P, Majumdar SN (2007) Phys Rev Lett 98:070404
Dahlstedt K, Jensen HJ (2001) J Phys A 34:11193; [Inspec] [ISI]
Dean DS, Majumdar SN (2001) Phys Rev E 64:046121
Eichner JF, Kantelhardt JW, Bunde A, Havlin S (2006) Phys Rev E 73:016130
Embrechts P, Klüppelberg C, Mikosch T, Karatzas I, Yor M (eds) (1997) Modelling extremal events. Springer, Berlin
Finkenstadt B, Rootzen H (2004) Extreme values in finance, telecommunications, and the environment. Chapman and Hall/CRC Press, London
Galambos J (1978) The asymptotic theory of extreme order statistics. Wiley, New York
Galambos J, Lechner J, Simin E (eds) (1994) Extreme value theory and applications. Kluwer, Dordrecht
Gnedenko BV (1998) Theory of probability. CRC, Boca Raton
Guclu H, Korniss G (2004) Phys Rev E 69:065104(R)
Gumbel EJ (1958) Statistics of extremes. Columbia University Press, New York
Leadbetter MR, Lindgren G, Rootzen H (1983) Extremes and related properties of random sequences and processes. Springer, New York
Majumdar SN, Comtet A (2004) Phys Rev Lett 92:225501; (2005) J Stat Phys 119:777
Raychaudhuri S, Cranston M, Przybyla C, Shapir Y (2001) Phys Rev Lett 87:136101
Reiss RD, Thomas M (2001) Statistical analysis of extreme values: with applications to insurance, finance, hydrology, and other fields. Birkhäuser, Basel
Smith RL (2003) Statistics of extremes, with applications in environment, insurance and finance, chap 1. In: Statistical analysis of extreme values: with applications to insurance, finance, hydrology, and other fields. Birkhäuser, Basel
Smith RL, Tawn JA, Yuen HK (1990) Statistics of multivariate extremes. Int Stat Rev 58:47
von Storch H, Zwiers FW (2001) Statistical analysis in climate research. Cambridge University Press, Cambridge
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2015 Springer Science+Business Media New York
About this entry
Cite this entry
Nicodemi, M. (2015). Extreme Value Statistics. In: Meyers, R. (eds) Encyclopedia of Complexity and Systems Science. Springer, Berlin, Heidelberg. https://doi.org/10.1007/978-3-642-27737-5_197-3
Download citation
DOI: https://doi.org/10.1007/978-3-642-27737-5_197-3
Received:
Accepted:
Published:
Publisher Name: Springer, Berlin, Heidelberg
Online ISBN: 978-3-642-27737-5
eBook Packages: Springer Reference Physics and AstronomyReference Module Physical and Materials ScienceReference Module Chemistry, Materials and Physics