API vs EIA Inventory Reports, Explained

1. The verdict first: two reports, side by side

Every week, the US oil market gets the same story twice. On Tuesday afternoon the American Petroleum Institute publishes its Weekly Statistical Bulletin. On Wednesday morning the Energy Information Administration publishes the Weekly Petroleum Status Report. Both describe the same week ending the previous Friday. They frequently disagree, and the disagreement is not an error in either one. They are two separate surveys with two separate tallies, and the table below is the whole comparison in one view.

Read the rest of this page as the footnotes to that table. Why two surveys of the same week disagree, how large the disagreement can get, and how to use a Tuesday preview without mistaking it for the Wednesday verdict.

2. Same forms, same week, different tally

The strangest fact in this comparison is how much the two surveys share. The API states, in its own frequently asked questions, that it collects an exact copy of the data submitted to the EIA, that it uses the same EIA weekly survey forms, and that it follows the EIA's collection schedule and submission deadline. The respondents are the same population: US refiners, operators of bulk storage and blending terminals, pipeline companies, and importers, reporting stocks and operations for crude oil and the major products. If the inputs are that similar, why do the outputs differ?

Four reasons, in rough order of importance. First, participation: reporting to the EIA is mandatory, reporting to the API is voluntary, so the respondent sets are close but not identical, and a large terminal that reports to one and not the other shifts that week's tally. Second, timing: a submission that misses one organization's internal cutoff can still land in the other's, and in a survey assembled over a weekend that margin matters. Third, estimation: neither survey hears from the whole industry. The API states that both it and the EIA cover roughly 90 percent of the industry and publish estimates, and two different estimates of the missing tenth produce two different totals. Fourth, the frame around the numbers: the EIA grosses up its weekly sample with ratios taken from its own monthly surveys, and its fuller monthly record appears later as a separate series, while the API's way of estimating the share it does not hear from is its own. The two reports are best understood as two measurements of the same physical week, taken with slightly different instruments.

3. Inside the Wednesday number: how the EIA builds its estimate

The EIA publishes its method in the explanatory notes appended to the Weekly Petroleum Status Report (Appendix B, checked October 2026), and the notes answer most of the questions people ask about the gap with the API. The weekly report draws on seven weekly supply surveys, Forms EIA-800 through EIA-805 and EIA-809, covering refineries and gas plants, bulk terminals, product pipelines, crude oil stockholders, importers, blenders, and fuel ethanol plants. Those weekly respondents are a sample of the companies that file the longer monthly forms. The EIA builds the sample by ranking companies from largest to smallest on recent volumes and adding them until the sample covers approximately 90 percent of the total for each item and each region, and it reviews the sample every month. The report week runs from 7:01 a.m. on a Friday to 7:00 a.m. the following Friday, and respondents must submit by 5:00 p.m. on the Monday after.

The missing tenth is not ignored; it is estimated by ratio. For each product and region, the EIA sums what this week's sample companies reported, then scales that sum by how large the whole industry was relative to the same sample companies in the most recent monthly survey. In the EIA's notation the estimate is Wt = (Mt / Ms) x Ws, where Ws is the weekly sample sum, Ms the same companies' latest monthly total, and Mt the latest monthly total for all companies. Companies that miss the deadline, or whose figures fail editing and cannot be confirmed, are filled in with an exponentially smoothed mean of their own recent weekly reports. The notes put the weekly response rate at generally 95 to 100 percent and the nonresponse share of a published estimate usually between 1 and 2 percent.

A worked illustration with stated, hypothetical inputs shows why two surveys of nearly the same companies can disagree. Suppose the weekly sample of distillate stockholders in one region reports 30.0 million barrels (Ws). Last month the same companies reported 31.5 million, while the full monthly survey for the region counted 34.0 million (so Mt / Ms = 34.0 / 31.5 = 1.079). The weekly estimate is 30.0 x 1.079 = 32.4 million barrels. Now let one terminal holding 0.6 million barrels report to the EIA but not to the API that week. A survey that scales a sample short of that terminal by its own factor publishes roughly 31.7 million instead, a 0.7 million barrel difference in the level, and the weekly change differs by the same amount if the terminal reported in the prior week. Nothing in that arithmetic is an error. It is what grossing up two slightly different samples produces, and it is largest where pools are small and concentrated, which is why distillates and regional gasoline stocks are where the two reports diverge most.

The second thing the notes make clear is that several headline lines in the EIA report are not survey counts at all, so the API has nothing to match them against:

Revisions follow a stated rule as well. The EIA disseminates a revised weekly figure only when it expects the revision to substantively change the understanding of US supplies, and then in the next regular release; otherwise late or corrected submissions feed into the following weeks' editing and imputation. So the weekly series a reader downloads is mostly the series as first published, with the definitive record arriving later in the monthly data.

4. A worked week: October 2026, when distillates flipped sign

The week ending October 16, 2026 shows every element of the comparison in a single episode. The API published first, on Tuesday, October 20, and its figures, as reported in market coverage that evening, were a draw across the board: crude stocks down 4.93 million barrels, gasoline down 2.0 million barrels, distillates down 2.6 million barrels. A reader who stopped there had a uniformly bullish picture: inventories falling everywhere.

The EIA published the next morning, Wednesday, October 21, for the same week. Crude drew 3.060 million barrels, against a consensus expectation of about 0.82 million. Gasoline drew 2.952 million barrels, a larger draw than the API had shown. And distillates, the product the API had down 2.6 million, came in as a build of 1.675 million barrels, against a consensus expectation of a draw of about 1.162 million. The arithmetic of the disagreement:

Same week, same forms, same industry: a 1.87 million barrel difference on the crude change, and on distillates a disagreement about whether stocks rose or fell at all, 4.28 million barrels wide. Neither figure is a fabrication; they are two estimates built from respondent sets that overlap almost completely but not entirely, each grossing up its own sample to a national total. Distillate stocks are a smaller, more regionally concentrated pool than crude, so a coverage difference shows up there first. The episode is the standing answer to the reader who asks why the two reports cannot simply be reconciled: there is nothing to reconcile. There are two surveys, and the official one publishes second.

The EIA's own highlights for that week (summary text checked October 2026) add the context the Tuesday headlines lacked. Commercial crude stocks fell to 412.5 million barrels, about 3 percent below the five-year average for the time of year. Gasoline stocks sat about 5 percent below their five-year average and distillate stocks about 9 percent below, even after the build, and total commercial petroleum stocks fell 8.5 million barrels in the week. Refineries ran 16.6 million barrels a day at 93.4 percent of capacity, with distillate output up to 5.2 million barrels a day, which is the plain mechanical reason a distillate build was possible while crude and gasoline drew. The API headline carried none of that: no levels against the seasonal range, no refinery detail, no demand line. A reader judging the week on Tuesday night had the direction of three numbers; a reader on Wednesday had the balance sheet that explained them.

5. The line where the EIA admits its own gap: the crude oil adjustment

Comparing the two reports invites a question about the official one: if the API's tally can miss the EIA's by a couple of million barrels, how close is the EIA's weekly estimate to itself? The EIA answers that in public, every Wednesday, on a line below the headline numbers. Table 1 of the weekly report carries a crude oil "Adjustment", and its footnote says what it is: formerly known as unaccounted-for crude oil, "this is a balancing item." It exists because the crude balance is assembled from pieces that are measured separately; the EIA says they should ideally balance perfectly but in practice rarely do.

The balance is simple. Crude that enters the US system in a week (domestic production, transfers of other liquids blended into the crude stream, and imports) has to go somewhere: into refineries, out as exports, or into stocks. Each piece comes from a different place. Stocks, refinery inputs and imports come from the weekly surveys, production from a model, exports from customs filings. When the pieces disagree, the EIA does not force the stock number to match the flows. It publishes the survey stocks as counted and puts the difference, disposition minus supply, on the adjustment line. In a 2015 explainer the agency described the line as reflecting "the combined uncertainty around each of the crude oil data elements."

The worked week, closed to the barrel

Take the same week as section 4, the week ending October 16, 2026, in thousand barrels a day from the EIA's Table 1 history file. Supply: production 13,269, plus transfers to crude oil supply 537, plus imports 5,758, makes 19,564. Disposition: refinery net input of crude 16,649, plus exports 3,442, plus the change in crude stocks including the Strategic Petroleum Reserve, minus 480, makes 19,611. Disposition minus supply is 47, and 47 is exactly the adjustment the EIA printed. That week the line was small: 47 thousand barrels a day is about 0.3 percent of refinery crude input, and the flows and the tank counts told nearly the same story.

The week before shows the other case. For the week ending October 9, 2026, total crude stocks fell 1,222 thousand barrels a day, a draw of about 8.6 million barrels over the week, and the adjustment was minus 308. Remove the adjustment and the flows alone imply a fall of 914 thousand barrels a day, about 6.4 million barrels. Roughly a quarter of the published draw that week was the balancing item rather than anything the flow estimates could account for. Nothing was wrong with the tank survey. The point is that a weekly draw is two numbers at once: the counted change in stocks, and the flows that are supposed to explain it, and the adjustment is the distance between them.

What a decade of the line shows

The definition changed partway through the record, and the change is the most useful thing in it. Starting with the November 15, 2023 publication, the footnote says, the crude oil adjustment includes transfers to crude oil supply: natural gas liquids and unfinished oils that the EIA identified as blended into crude in pipelines and storage. The EIA's monthly notes explain the direction of the effect. Crude oil adjustment quantities are normally greater than zero, and transfers, which normally add to supply, tend to bring the adjustment closer to zero. The weekly record bears that out. In the 461 weeks from January 2015 through the week ending October 27, 2023, the adjustment averaged plus 354 thousand barrels a day and was positive in 346 of them, 75 percent. In the 155 weeks from the week ending November 10, 2023 through October 23, 2026, it averaged plus 37 and was positive in 82, 53 percent. The persistent one-way gap is gone.

The noise is not. The typical size of the adjustment, the median of its absolute value, was 456 thousand barrels a day in the earlier period and 441 in the later one. Since the change it has ranged from minus 1,569 (week ending February 2, 2024) to plus 1,376 (week ending May 17, 2024). In its 2015 explainer the EIA wrote: "Typically, the adjustment is less than 2% of refinery crude oil inputs." On the weekly data since November 2023, the adjustment was under 2 percent of refinery crude input in 63 of the 155 weeks, 41 percent, and the median week sat at 2.7 percent. The same explainer noted that a four-week or eight-week average is often smaller, and that holds: the median size of the four-week average since the change is about 200 thousand barrels a day, less than half the weekly figure, because positive and negative weeks cancel.

Set that against the number the market trades. Since November 2023, the median size of the weekly change in total US crude stocks, up or down, has been 512 thousand barrels a day. The adjustment was larger than the stock change itself in 61 of the 155 weeks, 39 percent. And in 34 weeks, 22 percent, the stock change implied by the flow estimates alone pointed the other way from the published one: the tanks said build and the flows said draw, or the reverse. That is the honest scale of uncertainty in a single Wednesday crude figure, and it is published by the agency itself.

What this means for the API comparison

Start with what the bulletin is built to hold. The scope the API lists for it covers stocks, refinery operations, production and imports, and does not list exports, one of the pieces a crude balance needs, so it is not set up to close a crude balance the way Table 1 does. Its headline is a stock tally, and stocks are the part of the EIA balance that is counted rather than modelled, so the two reports compete on the hardest data in the release, not the softest. An API-EIA gap of one or two million barrels on the crude change is the same order as the EIA's own weekly balancing item: 441 thousand barrels a day is about 3.1 million barrels over seven days. A gap that size is a measure of how hard one week is to pin down, not evidence that either survey is broken. And commentary that corrects a weekly draw by pointing to the adjustment is doing something the EIA deliberately does not do. The agency keeps the tank count and shows the residual beside it. A reader who wants the flows' version of the week can compute it, as above, but it is a second estimate, not a truer one. The four-week average remains the best single defence against both kinds of noise.

6. Which one moves the market, and why the preview still matters

The EIA report moves prices more, and market coverage of the weekly releases says so in nearly those words: the EIA figures are treated as the official benchmark and typically drive the bigger move. The reasons are structural. The survey behind the EIA report is mandatory, so its coverage does not depend on goodwill. The report carries the detail the market's models actually consume: crude production estimates, product supplied as the demand proxy, stocks by refining district, Cushing inventories for the WTI delivery point, and the Strategic Petroleum Reserve, none of which the API headline provides. And the release is free, simultaneous, and fixed at 10:30 a.m. eastern in the middle of the trading morning.

The API bulletin matters for a different reason: it arrives 18 hours earlier, after the Tuesday close, and the market treats it as an early read on direction. Prices move on it in the evening session, positioning builds around it, and Wednesday morning then becomes a second trade: the EIA figure against both the consensus and the API preview. When the EIA confirms the API, the move extends. When the EIA contradicts it, as it did on distillates in October 2026, the reversal is sharp precisely because the preview had concentrated the positioning. The two reports are complements in the market's week, which is why the EIA report schedule treats them as appointments in a single calendar.

7. How to use both reports: a reading protocol

• Tuesday evening: read the API as a direction, not a number. Note which way crude and each product moved, and how far that is from the published consensus for Wednesday. Do not carry the API's levels into any calculation; they will not match the EIA's levels even when the changes agree.
• Wednesday 10:30: judge the EIA against consensus, not against the API. The price reaction is driven by actual minus consensus. An EIA draw smaller than the API preview is not bearish unless it is also smaller than the market expected.
• Where they disagree, slow down. A product where the two surveys flip sign, like distillates in October 2026, is a product where the week's data is genuinely uncertain. The honest read is the range between the two prints, and the four week average, not either single figure.
• Wednesday 10:31: check the adjustment line before trusting the crude change. Table 1 prints the crude oil adjustment beside the stock change. If the adjustment is larger than the stock change, as it was in 39 percent of weeks since November 2023, the flows and the tank count disagree about the week, and the crude headline deserves the same caution as an API-EIA sign flip. Compare it with its own four-week average, whose median size since November 2023 is less than half the weekly figure's, before reading anything into it.
• For anything kept, keep the EIA series. Histories, averages, days of supply, and seasonal comparisons must be built from one consistent survey. The EIA series is the one that is public, benchmarked to monthly data, and archived, which is why the data on our analysis page and in our guide to reading the weekly petroleum report is EIA data throughout. Demand-side lines need the same care: the product supplied explainer shows why one week's implied demand swing is often a filing artifact, and the Cushing stock line is the sub-total the WTI contract reacts to first.

8. Mistakes the two-report system invites

• Treating the API bulletin as a leak of the EIA report. It is a separate survey of the same respondents, published earlier by design. Nothing in it comes from the EIA's compiled data.
• Assuming one of them must be wrong when they differ. Both are estimates covering roughly 90 percent of the industry. A gap of one to two million barrels on a weekly crude change is the same order as the EIA's own weekly balancing item (section 5); the October 2026 distillate flip is a reminder of how uncertain one week is, not a scandal.
• Comparing an API change to an EIA level. Headlines report changes; databases store levels. Mixing the API's change with the EIA's level produces stock figures that exist in neither survey.
• Reading the API as the consensus. The API figure is one survey's estimate, published before the official one. Consensus is the analyst survey expectation for the EIA print. The two are different numbers with different jobs.
• Ignoring the products. Crude gets the headline, but the recurring disagreements live in gasoline and distillates, where the pools are smaller and regional coverage differences bite.

9. Frequently asked questions

Data and sources: American Petroleum Institute, Weekly Statistical Bulletin frequently asked questions, read October 21, 2026 (publication schedule; survey of US refiners, bulk storage and blending terminals, pipeline companies, and importers; an exact copy of the data submitted to the EIA, the same EIA weekly survey forms, and the EIA collection schedule and deadline; coverage of crude oil, motor gasoline, kerosene jet fuel, distillate fuel oil, and residual fuel oil; neither the API nor the EIA collects 100 percent of the data, both cover roughly 90 percent of the industry, and both publish estimates). U.S. Energy Information Administration, Weekly Petroleum Status Report and its release schedule page, read October 21, 2026 (standard release Wednesday 10:30 a.m. eastern; holiday exceptions). Market coverage of the October 20, 2026 API bulletin (crude -4.93 million barrels, gasoline -2.0 million, distillates -2.6 million) and of the October 21, 2026 EIA report for the week ending October 16, 2026 (crude -3.060 million against consensus of about -0.82 million; gasoline -2.952 million; distillates +1.675 million against consensus of about -1.162 million), as reported by market news outlets on those dates and read October 21, 2026; the EIA side of that week was rechecked against the EIA's own Weekly Petroleum Status Report highlights, checked October 22, 2026 (commercial crude stocks down 3.1 million barrels to 412.5 million, about 3 percent below the five-year average; gasoline down 3.0 million, about 5 percent below; distillate up 1.7 million, about 9 percent below; total commercial petroleum stocks down 8.5 million; refinery inputs 16.6 million barrels a day at 93.4 percent utilization; distillate output 5.2 million barrels a day). EIA, Weekly Petroleum Status Report explanatory notes (Appendix B), checked October 22, 2026 (seven weekly supply surveys, Forms EIA-800 through EIA-805 and EIA-809; cutoff sample built to cover approximately 90 percent of volumes by item and region and reviewed monthly; report week 7:01 a.m. Friday to 7:00 a.m. Friday; filing deadline 5:00 p.m. Monday; ratio estimator Wt = (Mt / Ms) x Ws; imputation by exponentially smoothed means; response rate generally 95 to 100 percent and nonresponse usually 1 to 2 percent; crude production estimated from the STEO model and Alaska data; exports from Customs and Border Protection ACE filings; revision policy). API, Weekly Statistical Bulletin page and FAQ, checked October 29, 2026 (scope: crude inventories, refinery operations, and production, imports and inventories of the major products, with no exports listed; within 1% of the EIA about 100% of the time for crude, gasoline and distillate stocks in the first quarter of 2025, per the API). For section 5, checked October 29, 2026: EIA, WPSR Table 1, U.S. Petroleum Balance Sheet (footnote: formerly known as Unaccounted-for Crude Oil, a balancing item; includes transfers to crude oil supply starting with the November 15, 2023 publication); EIA, WPSR historical Table 1 data (psw01.xls) (weekly production, transfers, imports, exports, refinery net input, crude stock change and adjustment, January 2015 to the week ending October 23, 2026); EIA Today in Energy, "Crude oil adjustment balances independently developed supply and disposition components", June 2, 2015; EIA, Petroleum Supply Monthly and Annual explanatory notes (crude oil adjustment and transfers to crude oil supply). Period averages, counts and percentages in section 5 are PetroEyes arithmetic on the weekly file. All other figures are as published by the sources named, for the periods stated.