A Genetic Algorithm System For Predicting The OEX by Deniz Yuret and Michael de la Maza
Since artificial intelligence made its debut within the pages of Stocks & Commodities, various and sundry methods by which neural networks can be used in designing trading systems have been proposed. Here, Yuret and Maza explain the development of a genetic algorithm system with which to forecast the OEX.
Designing a successful automatic trading system, as many have discovered, is a task of daunting proportions. But powerful new artificial intelligence methods such as neural networks have the potential to simplify this task. To that end, we decided to apply genetic algorithms to predicting the OEX.
Genetic algorithms were invented more than 20 years ago by mathematician and psychologist John Holland, who drew on sources in biology and economics to develop a general optimization algorithm that proved to be adept at finding good solutions to problems ranging from jet propeller design to protein folding. It is only during the last five years, however, that genetic algorithms have achieved widespread popularity within the computer science community and only now are being applied to stock market prediction.
The basic genetic algorithm maintains a set of individuals, which can be thought of as potential solutions to a problem. The best individuals in this set undergo crossover and mutation to produce new individuals that in turn are expected to be better solutions to the problem. After this process is repeated several times, the algorithm stops.