| Vose Software

Industry: Oil and Gas
Product: ModelRisk
Application: Market demand analysis


Will Oil Demand Peak Before 2030? The Decision Lives in the Tail

Global oil demand sat at 102 Mb/d in 2024 — almost exactly the pre-pandemic peak. A deterministic forecast draws a single line forward from there. The chart below draws what the model actually produces: a fan of 50,000 demand paths to 2034, with the median, the P25–P75 band, and the P10–P90 envelope, against the dashed deterministic baseline that grows at a steady +0.5%/year.

10-year global oil demand — P10/P25/P50/P75/P90 fan

By 2034 the median lands near 106 Mb/d, but the P10–P90 envelope runs from roughly 98 to 113 Mb/d — a 15 Mb/d spread that brackets both "demand still rising" and "demand has rolled over." IEA's central case projects 105 Mb/d by 2030; BP's Energy Outlook puts the 2035 range from 90 to 110 Mb/d depending on policy scenario; OPEC sees demand still rising into the 2040s. The decisions that depend on which of those is right — multi-billion-dollar refining capex, upstream sanctioning, midstream pipeline commitments — turn on which scenario, with what probability, and what the joint distribution looks like. An integrated oil and gas major rebuilt its 10-year demand model in ModelRisk around region-decomposed growth, energy-intensity decline, and a discrete scenario weighting of the substitution curve that is the single most important uncertainty in the entire forecast.

Three drivers, two of them continuous and one discrete

Regional GDP growth. OECD economies are modelled with annual log-Normal returns: μ = 1.6%, σ = 1.0%. Non-OECD with μ = 4.2%, σ = 1.5%. The decomposition matters because the elasticities of oil demand to GDP are very different — OECD is at 0.1, non-OECD closer to 0.6 — and a single global GDP collapses that into noise.

Energy intensity (oil/GDP). Continues its multi-decade decline at 2.0% ± 0.7%/year in OECD (electrification, efficiency standards), 1.0% ± 0.5%/year in non-OECD (slower structural transition). These trends are themselves probabilistic; an aggressive policy shift could pull either toward the lower bound.

Substitution / EV adoption. This is the variable for which a continuous Normal distribution would be wrong. The substitution curve is logistic, parameterised by the asymptote L and rate k, and the policy environment determines which logistic. The model uses a three-point discrete scenario distribution:

  • Slow transition (25% probability): k = 0.18, L = 0.20 — substitution stalls below 20% of original demand
  • Central transition (55%): k = 0.25, L = 0.32 — IEA STEPS-like
  • Fast transition (20%): k = 0.45, L = 0.55 — IEA APS / accelerated-policy world

Substitution is applied 2× faster in OECD than in non-OECD, reflecting where the electric-vehicle stocks live and where the policy is. The total demand at year t is then:

\[ D_t = D^{OECD}_t \cdot (1 - 2 s_t) + D^{non-OECD}_t \cdot (1 - 0.6 s_t) \]

with s_t the cumulative substitution share at year t under the drawn scenario.

The fan that the deterministic forecast cannot draw

The deterministic baseline (+0.5%/year trend) is the dashed line in the opening chart. The probabilistic forecast is the fan around it, which widens to a P10–P90 spread of roughly 15 Mb/d by 2034 — about 15% of 2024 demand. The envelope is close to symmetric around the 106 Mb/d median: non-OECD GDP growth keeps the upper paths rising even against modest substitution, while the fast-transition scenarios pull the lower paths back toward and below the 2024 level.

Will demand peak before 2034?

This is the question the forecast actually exists to answer. Defining a structural peak the way a capex decision cares about it — demand in 2034 ending up below the 2024 level of 102 Mb/d, the curve having turned over and net declined — demand peaks before 2034 in approximately 25% of paths, almost entirely the fast-transition scenarios where OECD substitution outruns non-OECD growth. That is the probabilistic version of "yes, demand might peak by 2030." It is not 100% and it is not 0%, and a capital plan based on either extreme is mis-spent.

Global oil demand in 2034 — distribution across scenarios

The 2034 demand distribution is single-peaked but broad, centred near 106 Mb/d with a long left shoulder, reflecting the discrete substitution-scenario structure: the fast-transition mass sits around 99 Mb/d, the central scenario around 106 Mb/d, and the slow-transition mass around 112 Mb/d. The three overlap into one wide hump rather than separating cleanly — which is exactly why the mean is a misleading planning number: roughly a quarter of that mass sits below the 2024 starting level even though the average is comfortably above it.

What actually drives 2034 demand uncertainty

Tornado: what drives 2034 demand uncertainty

The substitution scenario dominates by a wide margin — the single most important input is which discrete trajectory the world is on. Non-OECD GDP growth is second. Energy-intensity decline rates rank third and fourth. The implication for the forecasting function is operational: invest in monitoring substitution leading indicators (EV penetration, policy signals, ICE-vehicle scrappage rates) rather than in higher-precision GDP nowcasting.

What changed

  • A $1.2B refining capex deferred by two years. The simulation showed a 32% probability that refining margins collapse below sanctioning thresholds by 2030 under the fast-transition tail. The capital was retained as optionality.
  • Upstream sanctioning hurdle rate raised for long-cycle deepwater projects by 1.5 percentage points, reflecting the price-collapse probability under demand-peak scenarios.
  • Portfolio tilt toward natural gas and LNG: the same model run on natural gas shows demand peaking later and with a narrower range, supporting reallocation of $3B of capex toward gas-leveraged projects.
  • Scenario-narrative reports replace the single-line forecast in board materials: every demand presentation now leads with P10/P50/P90 and the substitution-scenario probability assumption.
  • Joint-venture partner alignment: the same probabilistic forecast is now the agreed planning input across two upstream JVs, replacing the previous practice of each partner running their own deterministic case.

ModelRisk Functionality Used

  • Region-decomposed demand model (OECD vs non-OECD) with separate GDP, energy-intensity, and substitution parameters — replacing the single-global-trend approach that collapses the elasticity heterogeneity into noise.
  • Discrete scenario distribution for the substitution-curve parameters (slow / central / fast) with explicit probabilities — the right structure for a fundamentally discontinuous policy-driven variable, where a continuous distribution would be indefensible.
  • Logistic substitution curve rather than a linear or exponential trend — capturing the inflection-point dynamics of technology adoption that linear models miss completely.
  • Fan-chart visualization with P10/P25/P50/P75/P90 envelopes and overlaid sample paths, replacing the single-line deterministic forecast that hides the substitution-scenario bimodality.
  • Tornado-ranked driver attribution identifying the substitution scenario as the dominant uncertainty — directly funding the leading-indicator monitoring programme rather than the GDP nowcasting team.
  • Peak-demand probability as a primary output, not just the 2034 level — because the question that drives capex sanctioning is when not how much.

The question is not whether oil demand peaks before 2034 — it is the probability it does, conditional on policy and macro paths. Monte Carlo simulation in ModelRisk turns that probability into a number the capital committee can act on, and a $1.2B capex deferral is the price-tag of getting that probability right.