Neural Network - Predictions with data which is out of range

Question

I’ve set up a neural network model to experiment with predicting foreign exchange rates based on various economic data. The model learned fine and test data is OK (R^2 = 0.