Optimal artificial neural network topology for foreign exchange forecasting

Foreign exchange market is one of the highest investments markets the average daily trade volume is 1.8 trillion USD. Foreign exchange rate forecasting has been always one of the most challenging subject and area of researches. Trader around the world is relying on the technical indicators which just following the price and has emerged a lag results. When the currency market has a random move (when the market is not trending) most of the indicators gets confused because of the fact that classical linear methods are unable to react with the non linearity in the data and hence with the market behavior. This research reports empirical results that tend to confirm the applicability of a neural network model to the prediction of the foreign exchange rates market. Artificial neural networks have proven to be efficient and profitable in forecasting financial time series in particular, feed forwarded back propagation. It is important to use an optimal ANN topology that emerged great results in short term prediction and the daily predication results showed that ANN model learns well and most likely to generalize well. Weekly predication results demonstrate good results in the low prediction while failed to have a good results on the high and the close prediction while the monthly prediction did not give a satisfactory results due to a very few data samples.