Statistical Analysis of Genetic Algorithms in Discovering Technical Trading Strategies

Abstract

In this study, the performance of ordinal GA-based trading strategies is\r\nevaluated under six classes of time series model, namely, the linear ARMA\r\nmodel, the bilinear model, the ARCH model, the GARCH model, the\r\nthreshold model and the chaotic model. The performance criteria employed\r\nare the winning probability, accumulated returns, Sharpe ratio and luck\r\ncoefficient. Asymptotic test statistics for these criteria are derived. The\r\nhypothesis as to the superiority of GA over a benchmark, say, buy-and-hold,\r\ncan then be tested using Monte Carlo simulation. From this rigorouslyestablished\r\nevaluation process, we find that simple genetic algorithms\r\ncan work very well in linear stochastic environments, and that they also\r\nwork very well in nonlinear deterministic (chaotic) environments. However,\r\nthey may perform much worse in pure nonlinear stochastic cases. These\r\nresults shed light on the superior performance of GA when it is applied\r\nto the two tick-by-tick time series of foreign exchange rates: EUR/USD\r\nand USD/JPY.