Computation intelligence based daily algorithmic strategies for trading in the foreign exchange market

Abstract

Successful trading in financial markets is not possible without a support system that manages the preparation of the data, prediction system, and risk management and evaluates the trading efficiency. Selected orthogonal data was used to predict exchange rates by applying recurrent neural network (RNN) software based on the open source framework Keras and the graphical processing unit (GPU) NVIDIA GTX1070 to accelerate RNN learning. The newly developed software on the GPU predicted ten high-low distributions in approximately 90 minutes. This paper compares different daily algorithmic trading strategies based on four methods of portfolio creation: split equally, optimisation, orthogonality, and maximal expectations. Each investigated portfolio has opportunities and limitations dependent on market state and behaviour of investors, and the efficiencies of the trading support systems for investors in foreign exchange market were tested in a demo FOREX market in real time and compared with similar results obtained for risk-free rates.