Exchange rate forecasting based on neural network with revised weight

In this study, firstly, the authors provide a brief introduction to studies of exchange rate forecasting in recent years and focus on the neural network methods. Secondly, they study five data samples by four neural network models, which include Genetic Algorithm neural network (GA-ANN), Quasi Newton neural network (QN-ANN), a common BP-neural network (ANN) and a BP-neural network given 10 times random initialized weights (ANN10). They compare the predictive ability of the GA-ANN model to other models in terms of forecast accuracy. They find that the GA-ANN model tend to improve the forecasting accuracy of ANN model and ANN10 model at all time horizon, and the GA-ANN model also performance better than QN-ANN in general.