Forecasting Gulf Exchange Rates with Artificial Intelligence: A Comparative Study of Tree-Based Models for OMR, SAR, AED, KWD, QAR, and BHD

Abstract

Accurate exchange rate forecasting is critical for international finance, yet it remains notoriously challenging. This challenge is uniquely framed in the Gulf Cooperation Council (GCC) region, where most currencies operate under formal or de facto pegs to the US dollar. This study investigates the efficacy of two advanced tree-based machine learning algorithms—Random Forest (RF) and eXtreme Gradient Boosting (XGBoost)—in forecasting the daily US dollar exchange rates of six Gulf currencies: the Omani rial (OMR), Saudi riyal (SAR), UAE dirham (AED), Kuwaiti dinar (KWD), Qatari riyal (QAR), and Bahraini dinar (BHD). Utilizing daily data from January 1, 2010, to January 31, 2025, we train models on an in-sample period (2010-2023) and evaluate their out-of-sample performance (2023-2025) using R², Mean Squared Error (MSE), and Mean Absolute Error (MAE). A key contribution is the inclusion of a naive persistence model as a benchmark to assess the marginal value added by machine learning. Our findings indicate that while the pegged currencies (OMR, SAR, AED, QAR, BHD) exhibit minimal forecastable variation, making sophisticated models only marginally better than the naive benchmark, the Kuwaiti dinar (KWD), with its basket peg, presents a more fruitful case for ML application. The study provides a fully reproducible Python framework, underscoring the importance of model interpretability and economic significance over mere statistical fit in highly stable, policy-constrained financial environments. The results offer practical insights for corporate treasurers, FX risk managers, and policymakers in the GCC.