Foreign exchange prediction based on indices and commodities price using convolutional neural network

Abstract

The level of accuracy in predicting is the key in conducting forex trading activities in gaining profits. Some predictions are made only by using historical currency data to be predicted, this makes predictions less accurate because they do not consider external influences. This study examines external factors that can influence the results of predictions, by looking for the relationship between the value of indices such as NTFSE and S & P 500 and the value of commodities such as gold and silver to the prediction process of EUR / USD. Prediction carried out using a deep learning algorithm with the Convolutional Neural Network method uses 2 1-dimensional convolution layers with ReL activation. The data used is the value of Open, High, Low and Close prices on forex, indices and commodities which are combined into one with the close forex value target for the next 5 days. Testing of EUR / USD test data only gets MSE results of 0.00081894. While the results of testing of the combined test data between EUR / USD, indices and commodities producing MSE vary between 0.00068717 to 0.0109606 where the best combination is a combination of FTSE 100 and Natural Gas values. So it can be concluded that other factors included in predicting have an influence on the results obtained.