Application of the LSTM - Deep Neural Networks - in Forecasting Foreign Currency Exchange rates

Abstract

The global foreign currency exchange market is among the top financial markets worldwide. Predicting forex rates is a concern because of forex rates volatility. We explore the use of deep learning approaches in the forecasting of forex rates and measure the success of, in particular, the LSTM model against the performance of ARIMA and SVR when predicting forex rates of USD pairs with ZAR. The MSE results indicated that the LSTM outperformed the SVR and ARIMA. However, despite being outperformed by the ARIMA model when the MAE was tested, the LSTM is generally excellent at predicting USDZAR speeds.