This document describes a genetic learning algorithm called GLOWER that is designed for financial prediction problems. It discusses how financial prediction is difficult due to high dimensionality, weak nonlinear relationships between variables, and important variable interactions. Standard algorithms like decision trees are limited by their greedy search approach, which can miss complex patterns. Genetic algorithms can perform a more thorough search without biases. The paper evaluates GLOWER on several datasets and finds it uncovers more effective patterns than other algorithms for difficult problems with weak structure.