GeneralizedHyperbolicDistribution: Generalized Hyperbolic Distribution

Description

Density function, distribution function, quantiles and random number generation for the generalized hyperbolic distribution, with parameters $\alpha$ (tail), $\beta$ (skewness), $\delta$ (peakness), $\mu$ (location) and $\lambda$ (shape).

Usage

dghyp(x, mu = 0, delta = 1, alpha = 1, beta = 0, lambda = 1,
      param = c(mu, delta, alpha, beta, lambda))
pghyp(q, mu = 0, delta = 1, alpha = 1, beta = 0, lambda = 1,
      param = c(mu, delta, alpha, beta, lambda),
      lower.tail = TRUE, subdivisions = 100,
      intTol = .Machine$double.eps^0.25, valueOnly = TRUE, ...)
qghyp(p, mu = 0, delta = 1, alpha = 1, beta = 0, lambda = 1,
      param = c(mu, delta, alpha, beta, lambda),
      lower.tail = TRUE, method = c("spline","integrate"),
      nInterpol = 501, uniTol = .Machine$double.eps^0.25,
      subdivisions = 100, intTol = uniTol, ...)
rghyp(n, mu = 0, delta = 1, alpha = 1, beta = 0, lambda = 1,
      param = c(mu, delta, alpha, beta, lambda))
ddghyp(x, mu = 0, delta = 1, alpha = 1, beta = 0, lambda = 1,
       param = c(mu, delta, alpha, beta, lambda))

Value

dghyp gives the density function, pghyp gives the distribution function, qghyp gives the quantile function and

rghyp generates random variates.

An estimate of the accuracy of the approximation to the distribution function can be found by setting valueOnly = FALSE in the call to

pghyp which returns a list with components value and

error.

ddghyp gives the derivative of dghyp.

Arguments

x,q: Vector of quantiles.
p: Vector of probabilities.
n: Number of random variates to be generated.
mu: Location parameter $\mu$, default is 0.
delta: Scale parameter $\delta$, default is 1.
alpha: Tail parameter $\alpha$, default is 1.
beta: Skewness parameter $\beta$, default is 0.
lambda: Shape parameter $\lambda$, default is 1.
param: Specifying the parameters as a vector of the form
c(mu,delta,alpha,beta,lambda).
method: Character. If "spline" quantiles are found from a spline approximation to the distribution function. If "integrate", the distribution function used is always obtained by integration.
lower.tail: Logical. If TRUE, probabilities are $P(X\leq x)$, otherwise they are $P(X>x)$.
subdivisions: The maximum number of subdivisions used to integrate the density and determine the accuracy of the distribution function calculation.
intTol: Value of rel.tol and hence abs.tol in calls to integrate. See integrate.
valueOnly: Logical. If valueOnly = TRUE calls to pghyp only return the value obtained for the integral. If valueOnly = FALSE an estimate of the accuracy of the numerical integration is also returned.
nInterpol: Number of points used in qghyp for cubic spline interpolation of the distribution function.
uniTol: Value of tol in calls to uniroot. See uniroot.
...: Passes additional arguments to integrate in pghyp and qghyp, and to uniroot in qghyp.

Author

David Scott d.scott@auckland.ac.nz

Details

Users may either specify the values of the parameters individually or as a vector. If both forms are specified, then the values specified by the vector param will overwrite the other ones. In addition the parameter values are examined by calling the function ghypCheckPars to see if they are valid.

The density function is

$$f(x)=c(\lambda,\alpha,\beta,\delta)\times% \frac{K_{\lambda-1/2}(\alpha\sqrt{\delta^2+(x-\mu)^2})}% {(\frac{\sqrt{\delta^2+(x-\mu)^2}}{\alpha})^{1/2-\lambda}}% e^{\beta(x-\mu)}$$

where $K_\nu()$ is the modified Bessel function of the third kind with order $\nu$, and $$c(\lambda,\alpha,\beta,\delta)=% \frac{(\frac{\sqrt{\alpha^2-\beta^2}}{\delta})^\lambda}% {\sqrt{2\pi}K_\lambda(\delta\sqrt{\alpha^2-\beta^2})}$$

Use ghypChangePars to convert from the $(\rho, \zeta)$, $(\xi, \chi)$, $(\bar\alpha, \bar\beta)$, or $(\pi, \zeta)$ parameterizations to the $(\alpha, \beta)$ parameterization used above.

pghyp uses the function integrate to numerically integrate the density function. The integration is from -Inf to x if x is to the left of the mode, and from x to Inf if x is to the right of the mode. The probability calculated this way is subtracted from 1 if required. Integration in this manner appears to make calculation of the quantile function more stable in extreme cases.

Calculation of quantiles using qghyp permits the use of two different methods. Both methods use uniroot to find the value of $x$ for which a given $q$ is equal $F(x)$ where $F$ denotes the cumulative distribution function. The difference is in how the numerical approximation to $F$ is obtained. The obvious and more accurate method is to calculate the value of $F(x)$ whenever it is required using a call to pghyp. This is what is done if the method is specified as "integrate". It is clear that the time required for this approach is roughly linear in the number of quantiles being calculated. A Q-Q plot of a large data set will clearly take some time. The alternative (and default) method is that for the major part of the distribution a spline approximation to $F(x)$ is calculated and quantiles found using uniroot with this approximation. For extreme values (for which the tail probability is less than $10^{-7}$), the integration method is still used even when the method specifed is "spline".

If accurate probabilities or quantiles are required, tolerances (intTol and uniTol) should be set to small values, say $10^{-10}$ or $10^{-12}$ with method = "integrate". Generally then accuracy might be expected to be at least $10^{-9}$. If the default values of the functions are used, accuracy can only be expected to be around $10^{-4}$. Note that on 32-bit systems .Machine$double.eps^0.25 = 0.0001220703 is a typical value.

References

Barndorff-Nielsen, O., and Blæsild, P (1983). Hyperbolic distributions. In Encyclopedia of Statistical Sciences, eds., Johnson, N. L., Kotz, S., and Read, C. B., Vol. 3, pp. 700--707.-New York: Wiley.

Bibby, B. M., and Sörenson,M. (2003). Hyperbolic processes in finance. In Handbook of Heavy Tailed Distributions in Finance,ed., Rachev, S. T. pp. 212--248. Elsevier Science B.~V.

Dagpunar, J.S. (1989). An easily implemented generalised inverse Gaussian generator Commun. Statist.-Simula., 18, 703--710.

Prause, K. (1999) The generalized hyperbolic models: Estimation, financial derivatives and risk measurement. PhD Thesis, Mathematics Faculty, University of Freiburg.

Examples

Run this code

param <- c(0, 1, 3, 1, 1/2)
ghypRange <- ghypCalcRange(param = param, tol = 10^(-3))
par(mfrow = c(1, 2))

### curves of density and distribution
curve(dghyp(x, param = param), ghypRange[1], ghypRange[2], n = 1000)
title("Density of the \n Generalized  Hyperbolic Distribution")
curve(pghyp(x, param = param), ghypRange[1], ghypRange[2], n = 500)
title("Distribution Function of the \n Generalized Hyperbolic Distribution")

### curves of density and log density
par(mfrow = c(1, 2))
data <- rghyp(1000, param = param)
curve(dghyp(x, param = param), range(data)[1], range(data)[2],
      n = 1000, col = 2)
hist(data, freq = FALSE, add = TRUE)
title("Density and Histogram of the\n Generalized Hyperbolic Distribution")
DistributionUtils::logHist(data,
    main = "Log-Density and Log-Histogram of\n the Generalized Hyperbolic Distribution")
curve(log(dghyp(x, param = param)),
      range(data)[1], range(data)[2],
      n = 500, add = TRUE, col = 2)

### plots of density and derivative
par(mfrow = c(2, 1))
curve(dghyp(x, param = param), ghypRange[1], ghypRange[2], n = 1000)
title("Density of the\n Generalized  Hyperbolic Distribution")
curve(ddghyp(x, param = param), ghypRange[1], ghypRange[2], n = 1000)
title("Derivative of the Density of the\n Generalized Hyperbolic Distribution")

Run the code above in your browser using DataLab