Stable distribution

Levy skew alpha-stable

Probability density function Symmetric centered Lévy distributions with unit scale factor Skewed centered Lévy distributions with unit scale factor
Cumulative distribution function CDFs for symmetric centered Lévy distributions CDFs for skewed centered Levy distributions
Parameters	$\alpha \in (0,2]\,$ exponent (real) $\beta \in [-1,1]\,$ skewness (real) $c\in [0,\infty )\,$ scale (real) $\mu \in (-\infty ,\infty )\,$ location (real)
Support	$x\in (-\infty ,+\infty )\!$ (real)
PDF	usually not analytically expressible (see text)
CDF	usually not analytically expressible (see text)
Mean	undefined when α≤1, otherwise μ
Median	usually not analytically expressible (see text). Equal to μ when β=0
Mode	usually not analytically expressible. Equal to μ when β=0
Variance	infinite except when α=2, when it is 2c2
Skewness	undefined
Excess kurtosis	undefined
Entropy	not analytically expressible (see text)
MGF	undefined
CF	$\exp \left[~it\mu -|ct|^{\alpha }\,(1-i\beta \,{\mbox{sgn}}(t)\Phi )~\right]$ $\Phi =\tan(\pi \alpha /2)\,$ for $\alpha \neq 1\,$ $\Phi =-(2/\pi )\log |t|\,$ for $\alpha =1\,$

In probability theory, a Lévy skew alpha-stable distribution or just stable distribution, developed by Paul Lévy, is actually a family of probability distributions which are characterized by four parameters: α, β, μ and c , as well as the distributed value, x . The μ and c  are shift and scale parameters which do not determine the shape of the distribution. The stable distribution has the important property of stability: If a number of independent identically distributed (iid) random variables have a stable distribution, then a linear combination of these variables will have the same distribution, except for possibly different shift and scale parameters. To be more precise:

If $X_{1}$ and $X_{2}$ are distributed according to a stable distribution $L(x;\alpha ,\beta ,\mu ,c)$, and if $Y=AX_{1}+BX_{2}+C$ is a linear combination of the two, then there exist values of D  and E  such that $DY+E$ is distributed according to a stable distribution $L(DY+E;\alpha ,\beta ,\mu ,c)$ or, equivalently, $Y$ is distributed according to a stable distribution $L(Y;\alpha ,\beta ,(\mu -E)/D,c/D)$. If $E=0$ for all $A$, $B$, and $C$ then Y is said to have a strictly stable distribution. Since the normal distribution, the Cauchy distribution, and the Lévy distribution all have the above property, it follows that they are special cases of the stable distribution.Template:Ref harvard

Stable distributions owe their importance in both theory and practice to the generalization of the Central Limit Theorem to random variables without second (and possibly first) order moments and the concomitant self-similarity of the stable family. It was the demand for self-similarity and the seeming departure from normality of financial data that led Benoît Mandelbrot to propose that cotton prices followed a Lévy skew alpha-stable distribution with α equal to 1.7. Levy skew alpha-stable distributions are frequently found in analysis of critical behavior and financial data. Template:Ref harvard Lévy skew alpha-stable distributions are also found in spectroscopy as a general expression for a quasistatically pressure-broadened spectral line. Template:Ref harvard All stable distributions are infinitely divisible and with the exception of the normal distribution for which α=2, stable distributions are Heavy-tailed distributions.

The distribution

A Lévy skew stable distribution is specified by scale $c$, exponent $\alpha$, shift $\mu$ and skewness parameter $\beta$. The skewness parameter must lie in the range [−1, 1] and when it is zero, the distribution is symmetric and is referred to as a Lévy symmetric alpha-stable distribution. The exponent $\alpha$ must lie in the range [0, 2].

The Lévy skew stable probability distribution is defined by the Fourier transform of its characteristic function $\varphi (t)$ Template:Ref harvardTemplate:Ref harvard

$f(x;\alpha ,\beta ,c,\mu )={1 \over 2\pi }\int _{-\infty }^{+\infty }\varphi (t)e^{-itx}\,dt$

where $\varphi (t)$ is given by:

$\varphi (t)=\exp \left[~it\mu \!-\!|ct|^{\alpha }\,(1\!-\!i\beta \,{\textrm {sgn}}(t)\Phi )~\right]$

where sgn(t)  is just the sign of t and $\Phi$ is given by

$\Phi =\tan(\pi \alpha /2)\,$

for all $\alpha$ except $\alpha =1$ in which case:

$\Phi =-(2/\pi )\log |t|.\,$

$\mu$ is a shift parameter, $\beta$ is a measure of asymmetry, with $\beta$=0 yielding a distribution symmetric about $\mu$. $c$ is a scale factor which is a measure of the width of the distribution and $\alpha$ is the exponent or index of the distribution and specifies the asymptotic behavior of the distribution for $\alpha <2$. Note that this is only one of the parameterizations in use for stable distributions; it is the most common but is not continuous in the parameters.

The asymptotic behavior is described, for α<2, by: Template:Ref harvard

$f(x)\sim {\frac {\alpha c^{\alpha }(1+\beta )\sin(\pi \alpha /2)\Gamma (\alpha )/\pi }{|x|^{1+\alpha }}}$

where Γ is the Gamma function (except that when α<1 and β=1 or -1, the tail vanishes to the left or right, resp., of μ). This "heavy tail" behavior causes the variance of Lévy distributions to be infinite for all $\alpha <2$. This property is illustrated in the log-log plots below.

Log-log plot of symmetric centered Levy distribution PDF's showing the power law behavior for large x. The power law behavior is evidenced by the straight-line appearance of the PDF for large x, with the slope equal to -(α+1). (The only exception is for α=2, in black, which is a normal distribution.)

Log-log plot of skewed centered Levy distribution PDF's showing the power law behavior for large x. Again the slope of the linear portions is equal to -(α+1)

When α=2, the distribution is Gaussian (see below), with tails asymptotic to exp(-x²/4c²)/(2c√π).

Special cases

There is no general analytic solution for the form of $p(x)$. There are, however three special cases which can be analytically expressed as can be seen by inspection of the characteristic function.

• For $\alpha =2$ the distribution reduces to a Gaussian distribution with variance $\sigma ^{2}=2c^{2}$ and mean $\mu$ and the skewness parameter $\beta$ has no effect. Template:Ref harvardTemplate:Ref harvard
• For $\alpha =1$ and $\beta =0$ the distribution reduces to a Cauchy distribution with scale parameter $c$ and shift parameter $\mu$. Template:Ref harvardTemplate:Ref harvard
• For $\alpha =1/2$ and $\beta =1$ the distribution reduces to a Lévy distribution with scale parameter $c$ and shift parameter $\mu$. Template:Ref harvardTemplate:Ref harvard

Note that the above three distributions are also connected, in the following way: A standard Cauchy random variable can be viewed as a mixture of Gaussian random variables (all with mean zero), with the variance being drawn from a standard Lévy distribution. And in fact this is a special case of a more general theorem which allows any symmetric alpha-stable distribution to be viewed in this way (with the alpha parameter of the mixture distribution equal to twice the alpha parameter of the mixing distribution -- and the beta parameter of the mixing distribution always equal to unity).

Other special cases are:

• In the limit as $c$ approaches zero or as $\alpha$ approaches zero the distribution will approach a Dirac delta function $\delta (x-\mu )$.

An interactive tutorial of Stable Laws may be found at http://mathestate.com/tools/power

Stability property

(See Template:Ref harvard and Template:Ref harvard

The Lévy alpha-stable distributions have the "stability" property that if $N$ alpha-stable variates $X_{i}$ are drawn from the distribution

$X_{i}\sim f(x;\alpha ,\beta ,c,\mu )\,$

then the sum

$Y=\sum _{i=1}^{N}k_{i}(X_{i}-\mu )\,$

will also be distributed as an alpha-stable variate,

$Y\sim {\frac {1}{s}}\,\,f(y/s;\alpha ,\beta ,c,0).\,$

where

$s=\left(\sum _{i=1}^{N}|k_{i}|^{\alpha }\right)^{1/\alpha }.\,$

This can be easily proven using the properties of characteristic functions.

The generalized central limit theorem

Another important property of Lévy distributions is the role that they play in a generalized central limit theorem. The central limit theorem states that the sum of a number of random variables with finite variances will tend to a normal distribution as the number of variables grows. A generalization due to Gnedenko and Kolmogorov states that the sum of a number of random variables with power-law tail distributions decreasing as $1/|x|^{\alpha +1}$ (and therefore having infinite variance) will tend to a stable Levy distribution $f(x;\alpha ,0,c,0)$ as the number of variables grows. Template:Ref harvard

Series representation

The stable distribution can be restated as the real part of simpler integral:Template:Ref harvard

$f(x;\alpha ,\beta ,c,\mu )={\frac {1}{\pi }}\Re \left[\int _{0}^{\infty }e^{it(x-\mu )}e^{-(ct)^{\alpha }(1-i\beta \Phi )}\,dt\right]$

Expressing the second exponential as a Taylor series, we have:

$f(x;\alpha ,\beta ,c,\mu )={\frac {1}{\pi }}\Re \left[\int _{0}^{\infty }e^{it(x-\mu )}\sum _{n=0}^{\infty }{\frac {(-qt^{\alpha })^{n}}{n!}}\right]$

where $q=c^{\alpha }(1-i\beta \Phi )$. Reversing the order of integration and summation, and carrying out the integration yields:

$f(x;\alpha ,\beta ,c,\mu )={\frac {1}{\pi }}\Re \left[\sum _{n=1}^{\infty }{\frac {(-q)^{n}}{n!}}\left({\frac {1}{i(x-\mu )}}\right)^{\alpha n+1}\Gamma (\alpha n+1)\right]$

which will be valid for $x\neq \mu$ and will converge for appropriate values of the parameters. (Note that the n=0 term which yields a delta function in $x-\mu$ has therefore been dropped.) Expressing the first exponential as a series will yield another series in positive powers of $x-\mu$ which is generally less useful.

See also

• Lévy flight
• Lévy process
• Other "power law" distributions
  • Cauchy distribution
  • Pareto distribution
  • Zeta distribution
  • Zipf distribution
  • Zipf-Mandelbrot distribution

References

• GNU Scientific Library - Reference Manual Edition 1.6, for GSL Version 1.6, 27 December 2004
  • "The Levy alpha-Stable Distributions". GNU Scientific Library - Reference Manual. Retrieved December 22. {{cite web}}: Check date values in: |accessdate= (help); Unknown parameter |accessyear= ignored (|access-date= suggested) (help)
  • "The Levy skew alpha-Stable Distribution". GNU Scientific Library - Reference Manual. Retrieved December 22. {{cite web}}: Check date values in: |accessdate= (help); Unknown parameter |accessyear= ignored (|access-date= suggested) (help)
• B. V. Gnedenko and A. N. Kolmogorov (1954). Limit Distributions for Sums of Independent Random Variables. Addison-Wesley.
• Johannes Voit (2003). The Statistical Mechanics of Financial Markets (Texts and Monographs in Physics). Springer-Verlag. ISBN 3-540-00978-7.
• "Some improvements in numerical evaluation of symmetric stable density and its derivatives" (PDF). CIRGE Discussion paper. Retrieved July 13. {{cite web}}: Check date values in: |accessdate= (help); Unknown parameter |accessyear= ignored (|access-date= suggested) (help)
• John P. Nolan. "Information on stable distributions". Retrieved December 22. {{cite web}}: Check date values in: |accessdate= (help); Unknown parameter |accessyear= ignored (|access-date= suggested) (help)
  • John P. Nolan (January 11, 2005). "Stable Distributions Models for Heavy Tailed Data" (PDF). Retrieved December 22. {{cite web}}: Check date values in: |accessdate= (help); Unknown parameter |accessyear= ignored (|access-date= suggested) (help)
  • John P. Nolan (November 29, 2005). "Bibliography on stable distributions, processes and related topics" (PDF). Retrieved December 22. {{cite web}}: Check date values in: |accessdate= (help); Unknown parameter |accessyear= ignored (|access-date= suggested) (help)
• I. Ibragimov, Yu. Linnik (1971). Independent and Stationary Sequences of Random Variables. Wolters-Noordhoff Publishing Groningen, The Netherlands.
• Peach, G. (1981). "Theory of the pressure broadening and shift of spectral lines". Advances in Physics. 30 (3): 367–474.
• V.M. Zolotarev (1986). One-dimensional Stable Distributions. American Mathematical Society.