Forex Price Trend Prediction using Convolutional Neural Network

Abstract

Foreign exchange (Forex) currency trading is an attractive investment. Real profits from the Forex trading, like stock trading, come from differences between buying prices and selling prices, which can be recognized by exchange rate trends. Our goal is to predict a trend direction from the most recent set of exchange rates using a simple deep learning model. The Forex price time series is converted to an image which is used as an input of a convolution neural network. Uptrend/downtrend labels of Forex prices were recognized using Martingale strategy. The model was trained using EUR/USD 2013 and 2015 open price data. For testing performances, initial 26-minute samples of uptrend/ downtrend events from EUR/USD 2016-2018 and 2020 open price data were used and 93% accuracy is achieved. To show trading benefits, a simple trading algorithm was simulated using our model and common trend indicators for EUR/USD 2013/2015-2018/2020 and GBP/USD 2020. In most cases, the trading algorithm using our model gained profits about 6% – 422% higher than those using other techniques. Therefore, our proposed model, if used correctly, may provide substantial profits in Forex trading.