EIA Crude Oil Price Forecast

In the September 2026 Short-Term Energy Outlook, EIA projects WTI at $88 a barrel in September 2026, easing to $58 by December 2027. Across the next twelve projected months that averages $78.50 for WTI and $82.75 for Brent. EIA’s own published calendar-year figures — the ones that end up quoted in price decks — are $84.65 for 2026 and $69.74 for 2027. For reference, the last month with a realised spot average is August 2026, when WTI came in at $83.90 and Brent at $91.08. Every projected figure is EIA’s, reproduced without adjustment, and EIA publishes them rounded to whole dollars.

Two Short-Term Energy Outlook series from Table 2, Energy Prices: WTIPUUS for the West Texas Intermediate spot average and BREPUUS for Brent, both monthly and in dollars per barrel, retrieved through EIA API v2. The realised prices every forecast is scored against are separate series from the Petroleum and Other Liquids dataset: PET.RWTC.M for Cushing WTI spot and PET.RBRTE.M for Brent.

A STEO normally projects about eighteen months forward. The January edition is the exception: it extends through December of the following year, so it reaches roughly twenty-four months out. That is why the forecast horizon on this page is longest in January and shortens month by month until the next January restores it.

By the line style on the chart and by the Basis column in the table. Solid lines are realised monthly spot averages; dashed lines beginning at the boundary are EIA’s projection. The two live in separate arrays in the payload rather than in one series with a marker, so a projected month cannot be drawn as history by accident. The last observed month is stated in words at the top of the page as well.

STEO publishes history and projection in a single column with no flag separating them. This page finds the split by comparing the STEO series against EIA’s own published monthly spot averages and taking the newest month where the two still agree to within two cents. STEO copies the realised average into its history rows, so an observed month matches exactly and a projected month does not. The scan runs newest-first deliberately: an oldest-first walk would stop at the first restated month and wrongly treat everything after it as forecast.

Because that is how EIA publishes them. Historical months in the same series carry cents, projected months are rounded to the nearest dollar. It is a useful reminder of the intended precision: the monthly path is a shape, not a set of point estimates accurate to the cent, and the rounding is EIA saying so.

EIA documents three inputs. A pooling model averaging five linear regressions that use futures-spot spreads, non-oil commodity prices, product crack spreads and US crude inventory changes; a separate regression on month-to-month changes in US petroleum inventories, OECD inventories relative to their four-year average, and global GDP growth; and analyst judgment, which iterates a price path against the model output, the previous month’s forecast and prevailing futures prices until the implied global balance is consistent. The last of those three is why the forecast is not simply a mechanical function of the futures curve.

They answer to different physical markets and the spread between them moves. WTI is the US benchmark, priced at Cushing, Oklahoma and tied to domestic pipeline and export economics. Brent is the waterborne international benchmark against which most globally traded crude is priced. A model built on one and quoted against the other will drift, so both projections are published here side by side.

Measured against realised spot prices across 1,320 matured WTI forecasts since the January 2020 edition: one month ahead the mean absolute error is $2.99 and 81% of calls land within five dollars of the outturn. Twelve months ahead the mean absolute error is $18.75, only 14% land within five dollars, and the average signed error is -$8.46 — meaning the forecast has on balance run below what the market delivered. The comparison that gives those numbers meaning is the crudest possible alternative: carrying the last observed monthly average forward unchanged scores $18.88 at the same twelve-month horizon. Near-term the forecast is clearly informative; at a year out, on this record, it is not distinguishable from assuming nothing changes.

The simplest forecast anybody can make: take the last observed monthly average and assume every future month is the same. It is included because an error figure on its own is unreadable — $18.75 at twelve months is impressive or embarrassing depending entirely on what the alternative was. On the current record the published forecast beats no-change through the first few months and then stops: from 5 months out the benchmark is the more accurate of the two ($12.85 against $13.05), and it stays ahead from there. Those rows are marked in the table above.

Forecast minus outturn, in dollars per barrel, for every projected month that has since closed. A negative error means EIA projected a lower price than the market delivered. Mean absolute error averages the size of the miss regardless of direction; average error keeps the sign, so it shows whether the misses cancel out or lean one way. Errors are grouped by horizon, meaning the number of months between the last observed month of that release and the month being projected.

Over the whole window the average twelve-month error is -$8.46, so on balance the projection has come in below what printed. That single figure hides more than it shows. Grouped by the year the forecast was made, the sign flips outright: 2021 vintages missed low by $35.77 at twelve months, 2024 vintages missed high by $11.61, and 2023 was the most accurate year of the record at $6.45 mean absolute error. A persistent one-directional bias is not what this shows. What it shows is a forecast that struggles at turning points, in whichever direction the turn happens to go.

Because the errors are not spread evenly across time. They cluster around the moments when crude re-rated, and a smooth projection made months before a re-rating misses the whole of it — which is why 2021 vintages are -$35.77 at twelve months while 2023 vintages sit at $6.45 mean absolute error. That is not a peculiarity of EIA. It is the general problem with forecasting a price set by a market that has already priced in everything currently known: the large moves come from what was not known.

This page does not answer that, because it does not carry futures settlement data. It answers a smaller question — how the forecast compares with assuming no change — and that comparison is on the chart. Anyone wanting the futures benchmark would need the historical forward curve at each release date, which is a different dataset from the two published EIA series used here.

The vintage store begins with the January 2020 edition, which gives a window long enough to contain several distinct price regimes rather than one, and which is as far back as EIA’s archived workbooks are read here. It also means the sample includes the dislocation of 2020, which flatters the no-change benchmark in some months and punishes it in others. The by-year table exists precisely so that no single stretch has to stand in for the whole record.

Because past eighteen months the only forecasts in the sample come from January releases, which are the only ones that reach that far. A twenty-four-month error figure would therefore describe a handful of particular Januaries rather than the forecast in general, and would sit in the same table as horizons drawn from every month of the year. Cutting the table where the sample stops being comparable is more honest than extending it.

One monthly edition of the forecast, named for its first projected month. The August 2026 vintage is the release whose history stops in July 2026 and whose projection begins in August. Comparing vintages is how a revision becomes visible: the same target month, forecast repeatedly, by the same organisation, with more information each time.

From EIA’s own archive. The API returns one vintage and overwrites the rest, so the earlier forecasts are read from the workbooks EIA publishes for every past release at eia.gov/outlooks/steo/archives/, which include an explicit Last Historical Month flag stating where that edition stopped reporting and started projecting. StorageCurve stores the projected months from each of those releases and adds one new vintage each month. No number in the revision history is estimated or interpolated.

Considerably more than the confident presentation of a single number suggests. EIA’s call on the 2026 WTI average has moved across a $40.20 range over the 21 releases that have projected it: first published at $62.50 in the January 2025 edition, ranging between $47.75 and $87.95, and standing at $84.46 today. A single target month is starker still — forecasts for August 2026 ranged from $49 to $95 across successive releases before the month closed at $83.90. Reading a forecast next to its own revision path is a different experience from reading it alone, which is why both are on one page.

Because the inputs do. A month brings new inventory data, revised demand and supply estimates, a moved futures curve, and sometimes a supply event that changes the balance outright. A forecast that did not respond to that would be worse, not better. The point of showing the revisions is not that revising is wrong; it is that a number being revised by tens of dollars within a year should not be quoted as though it were settled.

Two definitions are in use and they are labelled separately. The headline annual figures are EIA’s own published calendar-year averages, computed from daily prices. The revision path is a mean of each vintage’s twelve monthly values, because archived releases publish monthly columns and no annual one. The two differ by a few cents. The revisions being tracked are tens of dollars, so the distinction matters for accuracy of description rather than for the conclusion.

Once a month, after EIA publishes the new Short-Term Energy Outlook, normally in the first half of the month. Each update replaces the forecast path, appends that release to the vintage store, and rescores every forecast whose month has since closed. The release date and the modelling cut-off of the edition in use are shown beneath the chart.

Equity and credit models for producers, refiners and oilfield services need a crude price deck, and a free, monthly, methodologically documented government projection is a common starting point for one. StorageCurve cannot measure how often it is used or how much weight any particular model gives it, and does not claim to. What this page does claim is narrower and checkable: the number that circulates has a revision history and a track record, and both are reproduced here.

Yes. The STEO report itself carries a 95% confidence interval around the crude price path derived from the options market, published as a chart in the release rather than as a series in the API. It is not reproduced here because it is not available through the data route this page uses. Readers who want the market-implied band should take it from the STEO release directly; the error record on this page is a different thing, being what happened rather than what was expected to be possible.

No, and nothing here is investment advice. This page reproduces a public forecast and measures it against public outturns. That a forecast has missed by a given amount in the past is a fact about the past; it is not a prediction about the next release, and an error distribution is not a trading edge. StorageCurve publishes no price target of its own and takes no position on where crude goes next.

No. Every projected number on this page is EIA’s, reproduced unchanged. Everything StorageCurve adds is arithmetic on two published series — subtract the outturn from the forecast, group by horizon, average. There is no model, no adjustment, no correction factor and no view. If EIA revises its history, these numbers change with it.

The physical balance the forecast is trying to describe. The global oil demand and supply page shows the monthly consumption and production flows and the implied stock build or draw; OPEC spare capacity shows how much idle production sits behind the market; and the US crude inventory curve shows where the balance is actually landing week to week. A forecast is a claim about those quantities, and they are observable.