ANN for FOREX Forecasting and Trading

• Published in: 2015 20th International Conference on Control Systems and Computer Science pp. 322 - 328
• Main Authors: Czekalski, Piotr, Niezabitowski, Michal, Styblinski, Rafal
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.05.2015
• Subjects: activation function; ANN; Artificial intelligence; Artificial neural networks; Chapters; Forecasting; FOREX; network training; neural network; Neurons; Perceptron; Training
• ISSN: 2379-0474
• DOI: 10.1109/CSCS.2015.51

Modern approach to the FOREX currency exchange market requires support from the computer algorithms to manage huge volumes of the transactions and to find opportunities in a vast number of currency pairs traded daily. There are many well known techniques used by market participants on both FOREX and stock-exchange markets (i.e. Fundamental and technical analysis) but nowadays AI based techniques seem to play key role in the automated transaction and decision supporting systems. This paper presents the comprehensive analysis over Feed Forward Multilayer Perceptron (ANN) parameters and their impact to accurately forecast FOREX trend of the selected currency pair. The goal of this paper is to provide information on how to construct an ANN with particular respect to its parameters and training method to obtain the best possible forecasting capabilities. The ANN parameters investigated in this paper include: number of hidden layers, number of neurons in hidden layers, use of constant/bias neurons, activation functions, but also reviews the impact of the training methods in the process of the creating reliable and valuable ANN, useful to predict the market trends. The experimental part has been performed on the historical data of the EUR/USD pair.