Quantile Functions
(aka Inverse Cumulative Distribution Functions)
I
have some research projects running on the determination of quantile functions
of distributions relevant to mathematical finance applications. Quantile
functions are formally the inverse (when it exists) of the cumulative
distribution function, but when people talk about quantiles they are usually
talking in one of two (overlapping) contexts:
A.
Simply-determined key percentage points of distributions for statistical
applications;
B.
The (fast) application of quantile functions to random samples from the unit
interval to produce samples from non-uniform distributions.
My
work is on topics related to item B, and applications to Monte Carlo sampling,
low-discrepancy and copula methods. If you have come here looking for (A)
statistical methodology, you should go here for a very quick introduction
and see the book by Gilchrist for an extensive discussion.
Topics:
Recursive
quantiles
With
Gyorgy Steinbrecher I am investigating recursive algorithms for key
distributions. The work looks at the creation of very-high-precision benchmark
algorithms (with applications to calibrating new and fast rational and
polynomial approximations), and faster methods for real-time use. The first
version of our paper on this is downloadable here,
and prototype F95 and C++ code is available now for the normal case and a first
version in F95 for the Student is also available covering the entire unit
interval. The quad precision F95 code is typically returning errors better than
1.0 E-28! A C++ code for the Student is under development.
C++ code
for the normal quantile on [e, 1-e], e=0.0005
F95 code
for the normal quantile on [e,1-e], e=0.0007, error < 1.0E-29
F95 code
for the Student distribution on (0,1), (prototype, error < 1.0E-24)
Benchmarking
Rational Approximations
I
took a look at the existing methods for the normal case (Beasley-Springer-Moro,
Acklam, AS241) and my results are in the working paper and related material
below (Refinement of the Normal Quantile).
Adding
skewness and kurtosis to an existing distribution
With
Ian Buckley I have been looking at methods for modulating the skewness and
kurtosis of an existing distribution in a way that retains tractable forms for
the CDF and quantile functions. This has resulted in new definitions of a
skew-normal and kurtotic-normal distribution, and indeed a skew-kurtotic-normal
distribution. See Shaw and Buckley, 2007.
Analytical
Methods
My
work on the quantile function for the Student distribution appeared in the
Journal of Computational Finance (Shaw, 2006). The power series developed there
can be extended massively, and without the use of symbolic computation, using
the recursive methods developed with Steinbrecher.
Publications and working papers
Steinbrecher, G.
and Shaw, W.T., 2007, Differential Equations for Quantile Functions [PDF
of May 1 07 working paper]. [Key Words: Inverse CDF, quantile function, Normal,
Student, Beta, T-distribution, Simulation, Monte Carlo, Inverse Cumulative
Distribution Function, non linear ordinary differential equations, recurrence
relations]
Shaw, W.T.,
2007, Refinement of the Normal Quantile, Simple improvements to the
Beasley-Springer-Moro method of simulating the Normal Distribution, and a
comparison with Acklam's method and Wichura’s AS241, [PDF],
[Mathematica Notebook]
[Key Words, Rational Approximation, Beasley Springer, Moro, Acklam, AS241,
Wichura, Inverse Error function, Normal Quantile, Inverse Cumulative
Distribution Function] (Working paper, updated 20 Feb 07). Also available – (a)
AS241 (Wichura’s method alone) Notebook
and PDF;
(b) Acklam’s method (Notebook
and PDF);
(c) first version of C++ code
for non-linear recursion.
Shaw, W.T., and
Buckley, I.R.C. 2007, The Alchemy of Probability Distributions: Beyond
Gram-Charlier & Cornish-Fisher Expansions, and Skew-Normal or
Kurtotic-Normal Distributions. (Submitted, Feb 07) [PDF] [Key
Words: Distributional Alchemy, Gram Charlier, Cornish Fisher, Skew Uniform,
Skew Normal, Skew T, Skew Student, Student Distribution, T Distribution,
Kurtotic Uniform, Kurtotic Normal, Skew Kurtotic Normal, Skew Exponential]
Sydney QMF 2006
presentation on distributional alchemy and related applications of computer
algebra to Monte Carlo methods (Sydney, Australia, Dec 2006) [PDF
of presentation] [Mathematica
Notebook of presentation] [Key
Words: Student Distribution, Distributional alchemy, Cornish Fisher
expansion, Gram Charlier expansion, Skew Normal Distribution]
Shaw,
W.T., 2006, Sampling Student’s T distribution – use of the inverse cumulative
distribution function. Journal of Computational Finance, Vol 9 Issue 4, pp
37-73, Summer 2006 Journal
Link. On-line supplements to this article are available here. [Key
words: Student, Student’s T Distribution, T-Distribution, Inverse CDF, Inverse
Cumulative Distribution Function, Quantile, T-Quantile, Simulation, Monte
Carlo, Copula]
Other
Refs
W. Gilchrist,
2000, Statistical
Modelling with Quantile Functions, CRC Press
P. Jaeckel, 2002, Monte
Carlo Methods in Finance, Wiley