Predicting gold prices is crucial for policymakers, investors, and financial planners, especially in commodity-dependent economies like Iraq. This study examines the forecasting performance of Autoregressive Integrated Moving Average (ARIMA) and Seasonal ARIMA (SARIMA) models using daily gold price data in Iraq from (April 30, 2015, to April 30, 2025). After preprocessing and testing for stationary, both models were estimated using Maximum Likelihood Estimation (MLE) and further refined with modern machine learning post-processing, Maximum Likelihood Estimation (MLE) method demonstrated greater stability and interpretability compared to the machine learning-based approach. However, the application of Gradient Boosted Trees (GBDT) to the residuals of the SARIMA model further enhanced short-term predictive performance. Therefore, this paper concludes that the combination of SARIMA and (GBDT) creates a robust hybrid framework suitable for forecasting gold prices in volatile and data-constrained economies such as Iraq Model performance was evaluated using (RMSE, MAE, and MAPE) on both training and testing sets. The results indicate that SARIMA (1,1,1)(1,1,1) captured the seasonal patterns of gold prices more effectively than the baseline ARIMA (1,1,2). While ARIMA served as a benchmark for modeling non-seasonal dynamics, SARIMA provided the final forecasts adjusted for seasonality. Forecasts for (May 2025 to April 2026) show a consistent upward trend with minor seasonal fluctuations, validating the model’s applicability for decision-making in the Iraqi context. These findings demonstrate that SARIMA offers an interpretable and efficient framework for short- and medium-term forecasting, providing a reliable tool in volatile economic environments.