Currency Exchange Rate Prediction Using Gated Recurrent Unit (GRU) with Historical Data and Economic Factor

Abstract

This study presents a currency exchange rate prediction model using a Gated Recurrent Unit (GRU) with historical price data and selected economic factors. Historical data, including Open, High, Low, and Close (OHLC) prices, were obtained from Yahoo Finance. Economic factor data, including Non-Farm Payrolls (NFP), Gross Domestic Product (GDP), Purchasing Managers Index (PMI), Retail Sales, and Durable Goods Orders, were collected from Trading View. Data preprocessing involved chronological sorting, missing value handling, feature scaling, and sequence generation. Multiple experiment cases were evaluated: historical data alone, historical data combined with all economic factors, and historical data combined with each individual factor. The GRU model achieved its best performance when incorporating historical data with Durable Goods Orders, indicating that this economic indicator provides significant predictive value, as reflected by the lowest RMSE (0.0076) and MAPE (0.0054), and the highest R² (0.9764) indicating that this economic factor provides significant predictive value. These findings highlight the importance of integrating selected economic factors into exchange rate prediction models to enhance forecasting accuracy.