Predicting Forex Prices: An Evaluation of Long Short-Term Memory, XGBoost and Transformer Architectures

This research explores the effectiveness of various machine learning and deep learning models in predicting Forex prices using technical and fundamental indicators. Given the Forex market's unpredictability, accurate models are crucial for informed trading decisions and academic research. We use technical, fundamental, and combined indicators to compare models like Long Short-Term Memory, XGBoost, and Transformer-based architectures across different settings. An ensemble model is also introduced to balance accuracy and reliability. The analysis focuses on currency pairs like EUR/USD, NZD/USD, EUR/NZD, and AUD/NZD over a daily timeframe. We use metrics such as Mean Absolute Error, Mean Absolute Percentage Error, Root Mean Square Error, and R-squared to evaluate performance. Findings suggest that only some models excel in all scenarios, underscoring the need for model selection based on market conditions. This paper provides valuable insights into the strengths and weaknesses of different models for Forex price prediction, benefiting both academics and practitioners in finance.