Forecast Summary
Understanding the Forecast
KEY TERMS USED IN THE ARIMA MODEL AND THIS DASHBOARD
The number of past values (lags) the model uses to predict the next value. A higher p means the model looks further back in history. Think of it as "memory length."
p=2 → uses the last 2 years to predictHow many times the data is differenced to make it stationary (removing trends). A value of d=1 means we look at year-over-year changes instead of raw values.
d=1 → model uses changes, not levelsThe number of past forecast errors the model incorporates. This helps smooth out noise and capture shock effects that ripple through the data.
q=1 → corrects using 1 past errorMeasures forecast accuracy as a percentage. Lower is better. Under 5% is excellent, 5–10% is good, and above 10% suggests the model may struggle with that country's volatility.
MAPE 3.2% → avg error of 3.2% per yearA score used to select the best (p, d, q) combination. It balances model fit against complexity — lower AIC means a better trade-off. The grid search tests many combinations and picks the lowest AIC.
Best model = lowest AIC across gridThe shaded band around forecast lines. There's a 95% probability the true value falls within this range. Wider bands = more uncertainty, which naturally grows the further out you forecast.
Wider band in 2030 vs 2025 = more uncertaintyThe standard unit for measuring oil trade volume. 1 KBD = 1,000 barrels of oil flowing per day. This is the unit used across all values in this dashboard.
5,000 KBD = 5 million barrels/dayThese orders apply to the Exports page only — not imports. Not every country exports oil the same way — some are steady, some are driven by politics, some had sudden booms. We looked at how "jumpy" each country's export data was year-to-year and picked the model that best matches that behaviour.
The US shale revolution made exports grow in a very consistent, momentum-driven way — nearly tripling since 2015. The model only needs to look at past growth (AR), no error-correction needed. Forecasts continue this strong upward trend.
Both had choppy export histories with sharp swings that don't fully settle down on their own. Canada's oil sands output has been uneven; India's refinery exports jumped dramatically in the 2000s then levelled off. The model needs both a longer memory and an error-correction component.
Saudi exports are heavily controlled by OPEC+ production cuts and quota decisions — meaning big sudden changes happen regularly. The 2020 price war, the 2022 output cuts, and ongoing geopolitical tensions all create unpredictable swings that need the most flexible model we used.
These countries showed more momentum in their exports than a simple model could capture. China's refined product surge, Iran's sanctions-driven volatility, and the UAE's rapid ramp-up all benefit from a model that looks back two years.
South Korea's exports have largely plateaued around 1,300–1,400 KBD since the mid-2010s. When they deviate, they tend to snap back. A model that focuses on correcting past errors (MA) rather than chasing trends works best here.
Italy's export data only gave us 12 data points to work with — a more complex model caused numerical errors on such a small window, so we kept it simple and stable.
These are all either in long steady decline or too unpredictable for a complex model to help. Russia's exports have fallen sharply since the 2022 Ukraine invasion. Venezuela has collapsed from over 1,000 KBD in the 1970s to under 200 today. UK North Sea fields are steadily depleting.
Source: JODI Oil World Database · Log-ARIMA models selected via AIC grid search · Imports & Exports · Forecast horizon 2024–2030