Abstract:
The global foreign currency exchange (Forex) market is regarded as one of the most important financial markets in the world, with daily transactions exceeding $4 trillion. In financial market research, forecasting currency rates is a crucial problem. Forex is notorious for being very volatile and difficult to forecast.
In this study, we investigated the use of deep learning approaches in forex forecasting and compared the success of the Long Short-Term Memory (LSTM) model to the performance of AutoRegressive Integrated Moving Average (ARIMA) and Support vector regression (SVR) when predicting forex rates of US Dollar (USD) pair with South African Rand (ZAR) using daily timeframe data obtained from the Metatrader trading platform.
The LSTM outperformed the SVR and ARIMA models according to MSE data. The LSTM is typically good for predicting USDZAR speeds, although being surpassed by the ARIMA model when the Mean Absolute Error (MAE) was assessed.
