Generalized inverse Gaussian distribution

Generalized inverse Gaussian
Probability density function
Probability density plots of GIG distributions
Parameters a > 0, b > 0, p real
Support x > 0
PDF f(x)=(a/b)p/22Kp(ab)x(p−1)e−(ax+b/x)/2
Mean E⁡[x]=b Kp+1(ab)a Kp(ab)
E⁡[x−1]=a Kp+1(ab)b Kp(ab)−2pb
E⁡[ln⁡x]=ln⁡ba+∂∂pln⁡Kp(ab)
Mode (p−1)+(p−1)2+aba
Variance (ba)[Kp+2(ab)Kp(ab)−(Kp+1(ab)Kp(ab))2]
MGF (aa−2t)p2Kp(b(a−2t))Kp(ab)
CF (aa−2it)p2Kp(b(a−2it))Kp(ab)

In probability theory and statistics, the generalized inverse Gaussian distribution (GIG) is a three-parameter family of continuous probability distributions with probability density function

f(x)=(a/b)p/22Kp(ab)x(p−1)e−(ax+b/x)/2,x>0,

where Kp is a modified Bessel function of the second kind, a > 0, b > 0 and p a real parameter. It is used extensively in geostatistics, statistical linguistics, finance, etc. This distribution was first proposed by Étienne Halphen. It was rediscovered and popularised by Ole Barndorff-Nielsen, who called it the generalized inverse Gaussian distribution. Its statistical properties are discussed in Bent Jørgensen's lecture notes.

Properties

Alternative parametrization

By setting θ=ab and η=b/a, we can alternatively express the GIG distribution as

f(x)=12ηKp(θ)(xη)p−1e−θ(x/η+η/x)/2,

where θ is the concentration parameter while η is the scaling parameter.

Summation

Barndorff-Nielsen and Halgreen proved that the GIG distribution is infinitely divisible.

Entropy

The entropy of the generalized inverse Gaussian distribution is given as[citation needed]

H=12log⁡(ba)+log⁡(2Kp(ab))−(p−1)[ddνKν(ab)]ν=pKp(ab)+ab2Kp(ab)(Kp+1(ab)+Kp−1(ab))

where [ddνKν(ab)]ν=p is a derivative of the modified Bessel function of the second kind with respect to the order ν evaluated at ν=p

Characteristic function

The characteristic of a random variable X∼GIG(p,a,b) is given as

E(eitX)=(aa−2it)p2Kp((a−2it)b)Kp(ab)

for t∈ℝ where i denotes the imaginary unit.

Special cases

The inverse Gaussian and gamma distributions are special cases of the generalized inverse Gaussian distribution for p = −1/2 and b = 0, respectively. Specifically, an inverse Gaussian distribution of the form

f(x;μ,λ)=[λ2πx3]1/2exp⁡(−λ(x−μ)22μ2x)

is a GIG with a=λ/μ2, b=λ, and p=−1/2. A gamma distribution of the form

g(x;α,β)=βα1Γ(α)xα−1e−βx is a GIG with a=2β, b=0, and p=α.

Other special cases include the inverse-gamma distribution, for a = 0.

Conjugate prior for Gaussian

The GIG distribution is conjugate to the normal distribution when serving as the mixing distribution in a normal variance-mean mixture. Let the prior distribution for some hidden variable, say z, be GIG: P(z∣a,b,p)=GIG⁡(z∣a,b,p) and let there be T observed data points, X=x1,…,xT, with normal likelihood function, conditioned on z:

P(X∣z,α,β)=∏i=1TN(xi∣α+βz,z)

where N(x∣μ,v) is the normal distribution, with mean μ and variance v. Then the posterior for z, given the data is also GIG: P(z∣X,a,b,p,α,β)=GIG(z∣a+Tβ2,b+S,p−T2) where S=∑i=1T(xi−α)2.

Sichel distribution

The Sichel distribution results when the GIG is used as the mixing distribution for the Poisson parameter λ.

See also