Preliminary Comparative Evaluation of Time Series Models in Forecasting Crude Oil Exports in Saudi Arabia

Introduction: Crude oil exports represent more than 60% of the Saudi government’s revenue, so accurate export forecasting is essential for fiscal planning in the context of the Kingdom’s Vision 2030 diversification agenda. While oil price and production forecastings have attracted much attention, export volume forecastings—the most directly related variable to fiscal receipts—have been relatively underexplored, and few studies have compared classical statistical and machine learning models side-by side-for this series.

Methods: A preliminary comparative evaluation of three time series forecasting models, namely Autoregressive Integrated Moving Average (ARIMA), Exponential Smoothing State Space Model (ETS), and Random Forest (RF), is presented for 23 annual observations of Saudi Arabia’s crude oil exports (2002–2024) aggregated from 285 monthly records from the JODI Oil World Database. The models were trained on data from 2002–2019 and evaluated out-of-sample on a 2020–2024 hold-out set using Mean Squared Error (MSE) and Root Mean Squared Error (RMSE) measures.

Results: Random Forest had the lowest RMSE (4,131 kb/d), a 53.8% improvement over ARIMA (8,945 kb/d) and a 43.7% improvement over ETS (7,340 kb/d), confirming the benefit of ensemble averaging in reducing the forecast variance across the structural break induced by the 2020 COVID-19 demand shock. ARIMA and ETS maintained competitiveness in interpretability that is relevant to policy and audit contexts.

Conclusions: Random Forest outperforms ARIMA and ETS in the prediction of annual Saudi crude oil exports, facilitating a more comprehensive study on the subject, including the use of Support Vector Machine and Artificial Neural Network models at a monthly frequency and multi-year forecasts up to 2030, with direct implications for Vision 2030 fiscal and infrastructure planning.