Forecasting downstream petroleum production in South Africa using Bayesian SARIMA

Modelling downstream petroleum production in South Africa requires an appropriate model to generate reliable forecasts over the desired time horizon. Therefore, this study evaluates ARIMA, SARIMA, and Bayesian SARIMA models for forecasting downstream petroleum production in South Africa. The models were evaluated using simulated and real downstream data and model selection criteria such as the Akaike information criterion (AIC), Bayesian information criterion (BIC), and corrected Akaike information criterion (AICc). To determine models with minimal forecast error, metrics such as root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE) were also employed. The results show that Bayesian SARIMA outperformed the competing models with the simulated data; meanwhile, for the real data, SARIMA ( 1,2,0 ) 1,2,0 12 model produced the best out-of-sample point forecasts, with RMSE, MAE, and MAPE values of 139.97, 108.83, and 37.19%, respectively. Bayesian SARIMA ranked second, with RMSE, MAE, and MAPE values of 179.00, 146.46, and 51.67%. Although Bayesian SARIMA did not provide the most accurate point forecasts, its posterior predictive distributions provide useful information on uncertainty surrounding future production. These results have implications for domestic supply planning, import requirements, infrastructure investment, and longer-term energy policy in South Africa.