Low discrepancy numbers and their use within the ESG
Document ID: 2010-1665
Published on: 15th February 2010
Author: David Redfern
The Barrie & Hibbert ESG (Economic Scenario Generator) uses a Monte Carlo scheme to determine distributions and expectations of various financial and/or economic quantities by running a large number of trials (scenarios). Within the ESG this involves the generation of pseudo-random numbers, which are mapped into shocks and used to represent the stochastic (uncertain) aspects of the simulation. This study considers the effect of replacing the pseudo-random numbers with an alternative source of numbers (low-discrepancy numbers) that are not random, in an effort to improve the efficiency of the ESG.
Under certain conditions the use of low-discrepancy numbers is shown to result in faster convergence of the simulation. This means that fewer trials are necessary to obtain the same level of accuracy, or greater accuracy can be achieved with the same number of trials, when compared to a simulation involving pseudo-random numbers. The disadvantages and/or issues involved are also examined.
