[Paper Review] The alchemy of probability distributions: beyond Gram-Charlier expansions, and a skew-kurtotic-normal distribution from a rank transmutation map

WT Shaw, Irc Buckley|ArXiv.org|Jan 5, 2009

Financial Risk and Volatility Modeling18 references308 citations

TL;DR

This paper introduces a novel rank transmutation map (RTM) technique to generate flexible, tractable probability distributions with controlled skewness and kurtosis by composing the cumulative distribution function (CDF) of a base distribution with the quantile function of a target distribution. The method avoids the pathologies of Gram-Charlier expansions by providing exact, non-asymptotic transformations, enabling the derivation of a skew-kurtotic-normal distribution with simple, linear moment expressions in terms of transmutation parameters.

ABSTRACT

Motivated by the need for parametric families of rich and yet tractable distributions in financial mathematics, both in pricing and risk management settings, but also considering wider statistical applications, we investigate a novel technique for introducing skewness or kurtosis into a symmetric or other distribution. We use a "transmutation" map, which is the functional composition of the cumulative distribution function of one distribution with the inverse cumulative distribution (quantile) function of another. In contrast to the Gram-Charlier approach, this is done without resorting to an asymptotic expansion, and so avoids the pathologies that are often associated with it. Examples of parametric distributions that we can generate in this way include the skew-uniform, skew-exponential, skew-normal, and skew-kurtotic-normal.

Motivation & Objective

• To develop a robust, non-asymptotic method for introducing skewness and kurtosis into symmetric or base distributions, particularly for financial modeling and risk management.
• To overcome the limitations of Gram-Charlier and Cornish-Fisher expansions, such as negative densities and moment convergence issues.
• To provide a tractable simulation framework using quantile functions and closed-form transmutation maps.
• To derive a new family of distributions—specifically the skew-kurtotic-normal—featuring simple, linear moment expressions in terms of transmutation parameters.
• To enable practical application in Monte Carlo simulation and copula modeling through efficient sampling algorithms.

Proposed method

• The method employs a rank transmutation map (RTM) defined as $ G_{R_{12}}(u) = F_2(F_1^{-1}(u)) $, where $ F_1 $ is the CDF of the base distribution and $ F_2 $ is the CDF of the target distribution.
• A quadratic RTM is used as a starting point: $ G_{R_{12}}(u) = u + \lambda u(1-u) $, which introduces skewness and allows closed-form inversion.
• For higher-order modulation, a cubic polynomial RTM is introduced: $ P(z, \alpha_1, \alpha_2) = z + \alpha_1 z(1-z) + \alpha_2 z(1-z)^2 $, enabling simultaneous control of skewness and kurtosis.
• The transmuted CDF is constructed as $ F_2(x) = F_1^{-1}(G_{R_{12}}^{-1}(u)) $, with sampling performed via inverse transform sampling using the base distribution’s quantile function.
• The method ensures valid probability density functions by restricting parameters to a region in $ (\alpha_1, \alpha_2) $-space where the density remains non-negative and continuous.
• Monte Carlo sampling is achieved by solving the cubic equation $ P(z, \alpha_1, \alpha_2) = u $ for $ z $, then applying the base distribution’s quantile function to obtain $ X = F_1^{-1}(z) $.

Experimental results

Research questions

• RQ1 Can a non-asymptotic, exact method be developed to introduce skewness and kurtosis into a base distribution without relying on Gram-Charlier or Cornish-Fisher expansions?
• RQ2 What is the mathematical structure of a rank transmutation map that allows for simultaneous control of skewness and kurtosis in a base distribution?
• RQ3 How can the moments of the resulting transmuted distribution be expressed in a simple, closed-form, and linear manner with respect to transmutation parameters?
• RQ4 What are the admissible parameter regions in $ (\alpha_1, \alpha_2) $-space that ensure a valid, non-negative probability density function?
• RQ5 How does the proposed method compare in tractability and accuracy to existing skew-normal and Azzalini-type distributions?

Key findings

• The skew-kurtotic-normal distribution is derived via a cubic rank transmutation map, with its CDF expressed as $ F_2(x) = \phi(x) P'(\Phi(x), \alpha_1, \alpha_2) $, where $ \phi $ and $ \Phi $ are the standard normal PDF and CDF.
• The first five moments of the skew-kurtotic-normal distribution are linear functions of the transmutation parameters $ \alpha_1 $ and $ \alpha_2 $, with explicit expressions provided in Table 2.
• For $ \alpha_1 = 1, \alpha_2 = 0 $, the transmuted distribution corresponds to the maximum of two i.i.d. standard normal variates.
• For $ \alpha_1 = 0, \alpha_2 = 1 $, the distribution corresponds to the minimum of two i.i.d. standard normal variates.
• The method successfully generates special cases such as the maximum and minimum of three i.i.d. normals, and the middle order statistic of three, via specific $ (\alpha_1, \alpha_2) $ values.
• The admissible parameter region in $ (\alpha_1, \alpha_2) $-space is bounded but contains a large open set around the origin, enabling practical modeling of modest skewness and kurtosis.