\(\newcommand{\R}{\mathbb{R}}\) \(\newcommand{\N}{\mathbb{N}}\) \(\newcommand{\E}{\mathbb{E}}\) \(\newcommand{\P}{\mathbb{P}}\) \(\newcommand{\var}{\text{var}}\) \(\newcommand{\sd}{\text{sd}}\) \(\newcommand{\skew}{\text{skew}}\) \(\newcommand{\kurt}{\text{kurt}}\)
  1. Virtual Laboratories
  2. 4. Special Distributions
  3. 1
  4. 2
  5. 3
  6. 4
  7. 5
  8. 6
  9. 7
  10. 8
  11. 9
  12. 10
  13. 11
  14. 12
  15. 13
  16. 14
  17. 15
  18. Answers

14. The Extreme Value Distribution

Extreme value distributions arise as limiting distributions for maximums or minimums (extreme values) of a sample of independent, identically distributed random variables, as the sample size increases. Thus, these distributions are important in statistics.

The Standard Distribution for Maximums

Distribution and Denisty Functions

The function given below is a distribution function for a continuous distribution on \(\R\).

\[ G(v) = e^{-e^{-v}}, \quad v \in \R \]

The distribution defined by the distribution function in Exercise 1 is the type 1 extreme value distribution for maximums. It is also known as the Gumbel distribution in honor of Emil Gumbel. This distribution arises as the limit of the maximum of \(n\) independent random variables, each with the standard exponential distribution (when this maximum is appropriately scaled and centered). This is the main reason that the distribution is special, and is the reason for the name.

The probability density function is given by

\[ g(v) = e^{-v} e^{-e^{-v}}, \quad v \in \R \]

The distribution is unimodal and skewed right. The probability density function satisfies the following properties:

  1. \(g\) is increasing on \((-\infty, 0)\) and decreasing on \((0, \infty)\)
  2. The mode occurs at 0
  3. \(g\) is concave upward on \((-\infty, -c)\) and on \((c, \infty)\), and concave downward on \((-c, c)\), where \( c = \ln [(3 + \sqrt{5}) / 2)] \).

In the special distribution simulator, select the extreme value distribution and note the shape and location of the density function. Run the simulation 1000 times and note the apparent convergence of the empirical density function to the probability density function.

The quantile function is

\[ G^{-1}(p) = -\ln[-\ln(p)], \quad p \in (0, 1) \]

In particular,

  1. the first quartile is \(-\ln[-\ln(4)] \approx -0.3266\).
  2. the median is \(-\ln[-\ln(2)] \approx 0.3665\)
  3. the third quartile is \(-\ln[\ln(4) - \ln(3)] \approx 1.2459\)

In the special distribution calculator, select the extreme value distribution and note the shape and location of the density function and the distribution function. Compute the quantiles of order 0.1, 0.3, 0.6, and 0.9

Moments

The moment generating function of the standard extreme value distribution has a simple expression in terms of the gamma function.

Suppose that \(V\) has the extreme value distribution for maximums. The moment generating function is given by

\[ m(t) = \E(e^{t \, V}) = \Gamma(1 - t), \quad t \lt 1 \]

We can now compute the mean and variance. First, recall that the Euler constant, named for Leonhard Euler is defined by

\[ \gamma = -\Gamma^\prime(1) = -\int_0^\infty e^{-x} \ln(x) \, dx \approx 0.5772156649 \]

If \(V\) has the extreme value distribution for maximums then

  1. \(\E(V) = \gamma\)
  2. \(\var(V) = \frac{\pi^2}{6}\)

In the special distribution simulator, select the extreme value distribution and note the shape and location of the mean and standard deviation bar. Run the simulation 1000 times and note the apparent convergence of the empirical moments to the true moments.

The General Extreme Value Distribution

As with many other distributions we have studied, the standard extreme value distribution can be generalized by applying a linear transformation to the standard variable. Thus, suppose that \(V\) has the type 1 extreme value distribution for maximums, discussed above. First, \(U = -V\) has the type 1 extreme value distribution for minimums. More generally, we can form the location-scale family associated with these standard distributions. If \(a \in \R\) and \(b \in (0, \infty)\), then

Distribution Functions

\(X = a + b \, V\) has distribution function

\[ F(x) = \exp\left[-\exp\left(-\frac{x - a}{b}\right)\right], \quad x \in \R \]

\(X = a - b \, V\) has distribution function

\[ F(x) = 1 - \exp\left[-\exp\left(\frac{x - a}{b}\right)\right], \quad x \in \R \].

Probability Density Functions

\(X = a + b \, V\) has probability density function

\[ f(x) = \frac{1}{b} \exp\left(-\frac{x - a}{b}\right) \exp\left[-\exp\left(-\frac{x - a}{b}\right)\right], \quad x \in \R \]

\(X = a - b \, V\) has probability density function

\[ f(x) = \frac{1}{b} \exp\left(\frac{x - a}{b}\right) \exp\left[-\exp\left(\frac{x - a}{b}\right)\right], \quad x \in \R \]

Quantile Functions

\(X = a + b \, V\) has quantile function

\[ F^{-1}(p) = a - b \ln[-\ln(p)], \quad p \in (0, 1) \]

Show that \(X = a - b \, V\) has quantile function

\[ F^{-1}(p) = a + b \ln[-\ln(1 - p)], \quad p \in (0, 1) \]

Moments

\(X = a + b \, V\) has moment generating function

\[ M(t) = e^{a \, t} \Gamma(1 - b \, t), \quad t \lt \frac{1}{b} \]

Show that \(X = a - b \, V\) has moment generating function

\[ M(t) = e^{a \, t} \Gamma(1 + b \, t), \quad t \gt -\frac{1}{b} \]

The means and variances are

  1. \(\E(a + b \, V) = a + b \, \gamma\)
  2. \(\E(a - b \, V) = a - b \, \gamma\)
  3. \(\var(a + b \, V) = \var(a - b \, V) = b^2 \frac{\pi}{6}\)

Transformations

The exponential and extreme value distributions are related as follows:

  1. If \(X\) has the standard exponential distribution then \(U = \ln(X)\) has the standard extreme value distribution for minimums.
  2. If \(U\) has the standard extreme value distribution for minimums then \(X = e^U\) has the standard exponential distribution.

More generally,

  1. If \(X\) has the Weibull distribution with shape parameter \(k\) and scale parameter \(b\) then \(U = \ln(X)\) has the extreme value distribution for minimums, with location parameter \(\ln(b)\) and scale parameter \(\frac{1}{k}\).
  2. If \(U\) has the extreme value distribution for minimums, with location parameter \(a\) and scale parameter \(b\), then \(X = e^U\) has the Weibull distribution with shape parameter \(\frac{1}{b}\) and scale parameter \(e^a\).