Genetic Algorithms: A Heuristic Approach to Multi-Dimensional Problems

Abstract

Evolutionary algorithms are not new and have been developed, both their concepts and framework, since around the 1950’s based on the idea that the evolutionary process could be used as a general-purpose optimization tool. The goal of this paper is to propose an alternative to classical optimization techniques that can handle systems of a very high dimension. With the rapid rise of computing power, as well as the augmentation of alternative sources of data, quantitative analysts are confronted by numerical challenges that didn’t exist a decade ago. In this paper, we show that a Genetic Algorithm (GAs) is a simple process based on the evolution paradigm that is well adapted to very large portfolios, increasing the execution speed; an optimization of a portfolio of more than 100’000 times series of 5’000 daily returns takes less than 5 minutes. Finally, we illustrate that, although GAs are a random process that generates a different solution every time it is run on the same data, it is remarkably stable.

Keywords: genetic algorithms, dimensionality reduction, portfolio optimisation

JEL Classification: G0, C6, C8