An introduction to Genetic Algorithms

 

In computing, the genetic algorithm (GA) is understood as a really useful optimization technique. It is associated to the bigger category of evolutionary algorithms, which produces solutions to optimization problems using techniques inspired by natural evolution, like inheritance, mutation, selection, and crossover.

Genetic Algorithms, originally developed by John Holland (1975) as a simple model of genetic evolution, have swiftly evolved to be used in lots of different areas, including some economic models as well. Up to now, there have been quite a number of publications in this area . In the majority of the papers GA are applied to well known standard economic models, such as cobweb-type models, the prisoner's dilemma, or industrial organization problems. One reason for GA being an attractive tool for economic research has to do with the assumption of a normative behavioural foundation of individual action. In economics this is a very common assumption, and it makes models analytically tractable. The downside of this procedure is that it demands other strong assumptions like homogeneity, unbounded rationality and convexity. However, actually one will hardly find economic agents perfectly behaving like economic models want them to behave. Therefore, instead of a normative behavioural foundation, it appears to be a promising alternative among others to derive individual behaviour from artificial intelligence methods, of which GA are one. Because they always involve a number of strategies competing against each other GA-based economic models can be interpreted particularly well in a game theoretic context.

GA have been developed in analogy to the concepts of biological evolution and even the terminology is quite similar. Even though there is no standard GA but many variations of GA, there are some basic elements common to all GA.  GA consists in a number of strings containing information about how to behave in their environment and some operators, changing the strings. After “behaving“, the strings are evaluated by a fitness function, representing their environment, and the better adapted strings get higher scores. These in turn are important for the probability to be chosen by a selection operator, that determines which strings are allowed to reproduce. The chosen strings then undergo a procedure of crossing-over and mutation, and therefore the so built “offspring” forms next periods generation that undergoes an equivalent operations.

Because of their notable power to improve and find good solutions even in confusing or changing environments, the GA, initially developed as a descriptive tool, were often used as an optimization procedure for complex technical or logistic problems also .