Using a genetic algorithm to determine an index of leading economic indicators

In this paper, we report results of experiments investigating use of a genetic algorithm to select an index of leading economic indicators. Genetic algorithms apply operations of mutation, reproduction, and crossover to candidate solutions according to their relative fitness scores in successive populations of candidates. For our problem, a candidate solution is a subset of the publicly available economic indicators, considered at varying temporal offsets. We use several methods to focus search for an index, including reusing economic indicators from best solution candidates found during previous runs of the algorithm. Indices of leading economic indicators were found that were able to predict, with reasonable accuracy, previously observed troughs in economic activity.