Over the last decade, numerous papers have investigated the use of Genetic Programming (GP) for creating financial trading strategies. Typically, in the literature, the results are inconclusive but the investigators always suggest the possibility of further improvements, leaving the conclusion regarding the effectiveness of GP undecided. In this paper, we discuss a series of pretests aimed at giving more clear-cut answers as to whether GP can be effective with the training data at hand. Precisely, pretesting allows us to distinguish between a failure due to the market being efficient or due to GP being inefficient. The basic idea here is to compare GP with several variants of random searches and random trading behaviors having well-defined characteristics. In particular, if the outcomes of the pretests reveal no statistical evidence that GP possesses a predictive ability superior to a random search or a random trading behavior, then this suggests to us that there is no point in investing further resources in GP. The analysis is illustrated with GP-evolved strategies for nine markets exhibiting various trends.
Computational Intelligence in Economics and Finance 2007, pp 169-182