Exploration of ANNs for the economic purposes is described and empirically examined with the foreign exchange market data. For the experiments, panel data of the exchange rates (USD/EUR, JPN/USD, USD/GBP) are examined and optimized to be used for time-series predictions with neural networks. In this stage the input selection, in which the processing steps to prepare the raw data to a suitable input for the models are investigated. The best neural network is found with the best forecasting abilities, based on a certain performance measure. A visual graphs on the experiments data set is presented after processing steps, to illustrate that particular results. The out-of-sample results are compared with training ones.