Using Genetic Algorithms to Find Technical Trading Rules

Abstract

A genetic algorithm is used to learn technical trading rules for Standard and Poor's composite stock index using data from 1963-69. In the out-of-sample test period 1970-1989 the rules are able to identify periods to be in the indexwhen returns are positive and volatility is low and out when the reverse is true. Compared to a simple buy-and-hold strategy, they lead to positive excess returns after transaction costs in the period of 1970-89. Using data for other periods since 1929, the rules can identify high returns and low volatility but do not lead to excess returns after transaction costs. The results are compared to benchmark models of a random walk, an autoregressive model, and a GARCH-AR model. Bootstrapping simulations indicate that none of these models of stock returns can explain the findings.

JEL Classification: C63