Actuarial Matrix Computations with Fewer Tears- Rehabilitating Correlations, Avoiding Inversion, and Extracting Roots
Registration: 17:00 - 17:30
Programme: 17:30 - 19:00
Drinks reception: 19:00 - 20:00
The presentation will be given by the Numerical Algorithms Group (NAG) and the University of Manchester.
Linear algebra is an important part of many actuarial calculations. But the standard textbook formulae do not always provide the best way to carry out computations with large or ill conditioned matrices. Moreover, when computations are carried out in finite precision arithmetic, and when the data has large uncertainties, the computed results are not always satisfactory. In the worst cases poor results can be obtained without any indication that something has gone wrong.
John Holden will talk briefly about:
- Numerical software and tools for the actuarial community
- NAG and its collaboration with the University of Manchester
Professor Nicholas Higham will deliver the keynote talk. He will discuss three topics:
- How to compute the nearest correlation matrix to an "almost correlation matrix", one having some small negative eigenvalues due to incomplete underlying data.
- Why matrix inversion can and should usually be avoided.
- How to compute a p'th root of a matrix for some positive integer p. with particular attention to stochastic matrices.
Who should attend?
The seminar is aimed at actuaries, whose roles include a major component of developing, supervising or using numerical and statistical software within or outside the Financial Services industry.