Would evolutionary computation help in designs of ANNs in forecasting foreign exchange rates?

This paper evaluates the relevance of evolutionary artificial neural nets to forecasting the tick-by-tick DEM/USD exchange rate. With an analysis based on modern econometric techniques, this time series is shown to be a complex nonlinear series, and is qualified to be a challenge for ANNs and EANNs. Based on the five criteria, including the Sharpe ratio and a risk-adjusted profit rate, we compare the performance of 8 ANNs, 8 EANNs and the random-walk model. By the Granger-Newbold test, it is found that all neural network models can statistically beat the RW model in all criteria at the 1% significance level. In addition, among the 16 NN models generated in different designs, the best model is the EANN with the largest search space.