Authors: Thomas Gkourlas

Title: Forecasting foreign exchange rates with artificial neural networks

Abstract

Understanding exchange rates movement has long been an extremely, challenging and important task because these rates are impacted by a variety of factors including different economic, political and psychological factors. These factors are highly correlated and interact each other in a very unstable and volatile manner making the forecast a very difficult case. In that way Neural Networks give us an alternative to linear regression models approach for developing accurate forecast models. NN are algorithms and techniques that can be used for statistical modeling and designed for detecting knowledge data without human interruptions. This survey aims to analyze the back-propagation NN on foreign currency exchange rates and describes how to build an artificial BBNN step by step. We try two network structures with 5 and 9 input variables and 3 and 5 hidden layers. Training 10 different algorithms on different kind of input data, we compare their performance to ascertain what type of training algorithm is the best fit in each case. We found that both models using the Levenberg – Marquardt as the training algorithm can be used to correctly predict one step ahead fx rates, minimizing – in that way – the risk and the relative uncertainties.