Lookup NU author(s): Dr Colin Gillespie
Full text for this publication is not currently held within this repository. Alternative links are provided below where available.
Given a set of third- or higher-order moments, not only is the saddlepoint approximation the only realistic 'family-free' technique available for constructing an associated probability distribution, but it is 'optimal' in the sense that it is based on the highly efficient numerical method of steepest descents. However, it suffers from the problem of not always yielding full support, and whilst [S. Wang, General saddlepoint approximations in the bootstrap, Prob. Stat. Lett. 27 (1992) 61.] neat scaling approach provides a solution to this hurdle, it leads to potentially inaccurate and aberrant results. We therefore propose several new ways of surmounting such difficulties, including: extending the inversion of the cumulant generating function to second-order; selecting an appropriate probability structure for higher-order cumulants (the standard moment closure procedure takes them to be zero); and, making subtle changes to the target cumulants and then optimising via the simplex algorithm. © 2006 Elsevier Inc. All rights reserved.
Author(s): Gillespie CS, Renshaw E
Publication type: Article
Publication status: Published
Journal: Mathematical Biosciences
ISSN (print): 0025-5564
ISSN (electronic): 1879-3134
Publisher: Elsevier Inc.
PubMed id: 17306841
Altmetrics provided by Altmetric