Exploring the predictability in Technical Trading rules

Genetic algorithms are applied to several simulated data series to study the predicting power of technical trading rules. The two kinds of rules considered are moving averages and filter rules. Genetic algorithms simulate how the best rule is uncovered by analysts over time. The analysis supports the predicting power of technical rules. The predictability is especially strong for data with moving average and nonlinear structures. For these data, the return difference between buy and sell signals are much higher than that can be explained by risk differences. The result suggests that technical rules are exploiting high order nonlinearity in the data that cannot be modeled with linear projection frameworks. ∗I thank William Barnett, James Morley for their helpful comments. I am also grateful to Frank Allen, Risto Karjalainen, Melvin Hinich for their generosity in sharing the programs used in this paper. All errors remain mine.