Prediction of Crude Oil and Condensate Resources in Malaysia Using the ARIMA Approach

Abstract

Crude oil and condensate resources are hard to plan around without a reasonable forecast, and that matters most in economies like Malaysia, where upstream petroleum production sits across three distinct regions: Peninsular Malaysia, Sabah, and Sarawak. This study uses the Box–Jenkins ARIMA approach on Malaysia's annual Total crude oil and condensate resources series (1990–2021, n = 32), taken from the Malaysia Energy Statistics Handbook 2023. After a literature review covering 20 Q1/Q2-indexed studies on ARIMA-based petroleum forecasting, the series was checked for stationarity, differenced once (d = 1), and sixteen candidate ARIMA(p,1,q) models (p, q = 0–3) were built from the ACF/PACF diagnostics and ranked by AIC and BIC. Rather than stopping at the AIC/BIC ranking, the final model choice came down to which specification actually forecast well against a held-out 2017–2021 test period (RMSE, MAPE). ARIMA(0,1,2) and ARIMA(2,1,0) had the lowest AIC/BIC, but it was ARIMA(0,1,1) that gave the best out-of-sample result (RMSE = 0.400, MAPE = 8.47%), so that is the model carried forward. Refit on the full series, it forecasts resources holding roughly flat at about 4.42 billion barrels through 2026, with a 95% interval that widens from [3.07, 5.77] in 2022 out to [1.15, 7.69] by 2026. The main takeaway is that for a short annual series like this, out-of-sample testing should outrank information criteria when picking a final model. Condensate-specific, region-level forecasting is still an open gap worth more attention.