The Implied Stock Build Method for China's Crude
China publishes no inventory level, so every headline about its stockpiling rests on a residual: domestic production plus net crude imports minus refinery throughput. The arithmetic is simple, the inputs are free and the answer is published within three weeks of month end. It is also a plug that absorbs every error in both inputs. The EIA puts the average spread between competing estimates at 0.5 million barrels per day, and as much as 1.1 million. This page sets out the method, the inputs, the error sources and what the number is actually good for.
- The identity is: domestic crude production plus crude imports minus refinery crude throughput equals the implied stock change. The careful version also subtracts crude exports and direct industrial burn, which Kemp puts at roughly 2 million tonnes of a 56 million tonne residual for 2025, about 4 per cent.
- The inputs come from two publishers on different clocks. The National Bureau of Statistics publishes crude output and refinery throughput roughly two to three weeks after month end, combining January and February into one release to strip out Lunar New Year distortion. The General Administration of Customs publishes headline trade volumes around the 7th to 9th and the detailed breakdown around the 18th to 20th.
- Both inputs are revised, and one is re scoped annually. Customs states that latest monthly statistics may differ from those previously released because of data verification after dissemination. NBS covers only industrial enterprises above designated size, annual main business income over CNY 20 million, and says the scope changes every year, creating caliber differences versus previously released data. Any published implied build is provisional in both of its inputs.
- China is not entirely silent, which is a commonly repeated error. It publishes no inventory level and no strategic versus commercial split, and Kpler notes the last public update on the locations and availability of China's oil reserves was released in 2017. But NBS does publish an annual crude energy balance containing an inventory change line, two to three years in arrears. The correct statement is that the monthly number does not exist, not that nothing is published.
- The method and the satellites disagree, sometimes enormously. For March 2026 Reuters' supply balance gave a build of 1.74 million barrels per day while Kpler's observed onshore stocks rose by about 175,000 barrels per day. Kpler's own reconciliation found an average oversupply of 780,000 barrels per day between September 2025 and February 2026, roughly 140 million barrels, which it says may largely be explained by barrels withdrawn from underground storage facilities or pipeline systems that are not covered by our monitoring systems.
- The honest reading of that gap is not that the balance is wrong. Satellite tank counting reads external floating roofs. Gabriel Collins of the Baker Institute, testifying to the US China Economic and Security Review Commission in June 2024, states that synthetic aperture radar cannot detect underground cavern storage, fixed roof tank installations or the contents of sealed underground facilities. Collins and Shih Yu Hung costed the evasion in 2018: doming a tank runs about USD 1.0 to 1.2 million and mined caverns are about 60 per cent cheaper per barrel than surface tanks. The two storage modes the cameras cannot read are the cheaper ones to build.
- Use it for sign and order of magnitude over a quarter or a year, not for a monthly point estimate. Three independent approaches converge on 2025 within about 20 per cent: Russell at 1.13 million barrels per day, Kemp at about 1.1 million and the EIA at roughly 0.9 million for January to August. For a single month, publish a range and name the method.
A residual, stated the same way by every practitioner
The identity is implied crude stock change equals domestic crude production plus crude imports minus refinery crude throughput. Reuters columnist Clyde Russell uses an identical formulation in column after column, including on 17 March 2026, 16 April 2026, 17 August 2026 and 15 September 2026: China does not disclose the volumes of crude flowing into or out of its strategic and commercial stockpiles, but an estimate can be made by deducting the amount of oil processed from the total crude available from imports and domestic output.
The Reuters version does not subtract crude exports, because China's are negligible. John Kemp of Base Research publishes the more complete version in tonnes, and his 2025 worked example is the clearest statement of the full chain anywhere. Writing on 15 February 2026 he sets out that China's domestic crude output climbed to 216 million tonnes and the country imported a further 578 million, according to data published by National Bureau of Statistics and the General Administration of Customs, but the country's refineries processed only 738 million tonnes, leaving 56 million unaccounted for, of which perhaps 2 million were probably exported with other small volumes used directly in industry. His conclusion: China's stocks of crude oil apparently increased by 54 million tonnes, about 400 million barrels or 1.1 million barrels per day, during 2025 after a similar increase in 2024.
The EIA states its own variant slightly differently and the difference matters, because it names condensate explicitly. In its 9 October 2025 analysis it says China does not report data on its oil inventories, so we assessed China's stock growth based on imports, exports, refining, and oil inventory data from third party and official sources, balancing crude oil and condensate production reported by NBS with imports, refinery runs and export data from ship tracking and third party sources. Whether condensate and light feedstocks sit inside or outside a given series is exactly the kind of definitional difference that can move a monthly answer by a hundred thousand barrels a day without anybody being wrong.
One step is almost never disclosed and is worth flagging. NBS publishes in tonnes and the barrel figures in every headline are the analyst's own conversion. Reuters' 2025 full year figures imply about 7.30 barrels per tonne and Kemp's 54 million tonnes to roughly 400 million barrels implies about 7.35. That is our own arithmetic on published totals, not a stated factor from either source, and the 0.7 per cent gap between the two is worth roughly 100,000 barrels per day against a throughput base near 14.75 million. It is larger than several of the error terms analysts do discuss.
المشروع 54Production is one of three measured inputs. The inventory number nobody publishes is what is left when the other two are subtracted.Two publishers, two clocks, two revision regimes
Crude output and refinery throughput come from the National Bureau of Statistics, in its monthly energy production release, roughly two to three weeks after month end. January is never published alone: NBS combines January and February into a single release in early to mid March to remove Lunar New Year distortion, which is why the first column of any year covers a two month block.
The coverage caveat is in the NBS explanatory notes and it is the single most useful primary citation on the method's limits. Coverage is restricted to industrial enterprises with annual main business income of more than 20 million yuan, the above designated size universe, and NBS states that the scope of industrial enterprises above the designated size changes every year, which creates caliber differences versus previously released data. Two consequences follow: throughput at the smallest plants sits outside the reported universe entirely, and the reporting universe is reconstituted annually, so monthly series are not cleanly comparable across years and the months need not sum to the annual total.
Crude imports and exports come from the General Administration of Customs, with headline trade volumes typically around the 7th to 9th of the following month and the detailed commodity and origin breakdown around the 18th to 20th. Customs publishes its revision policy plainly: latest monthly statistics may differ from those previously released because of data verification after dissemination, and the latest release is of more accuracy.
What China does not publish is the level. There is no commercial or strategic crude inventory figure and no split between the two. Kpler records that the last public update on the locations and availability of China's oil reserves was released in 2017. Kemp notes the reason bluntly: stocks of crude and refined products are treated as a state secret, and there is no benefit sharing inventory levels with potential adversaries. The partial exception, almost never mentioned, is that NBS does publish an annual energy balance for crude oil containing an explicit inventory change line, covering production, imports, exports, transformation, final consumption and stock change. It is annual and runs two to three years in arrears, which makes it useless for trading and is precisely why the monthly residual exists.
| Input | Publisher | إيقاع | Typical lag | Revision behaviour |
|---|---|---|---|---|
| Crude oil output | National Bureau of Statistics | Monthly, January and February combined | About 2 to 3 weeks after month end | Above designated size scope, CNY 20 million revenue, is reset annually, creating caliber differences versus previously released data |
| Refinery crude throughput | National Bureau of Statistics | Monthly, January and February combined | About 2 to 3 weeks after month end | Same scope and annual re scoping; excludes plants below the threshold |
| Crude imports | General Administration of Customs | Monthly, in two releases | About 7 to 9 days headline, 18 to 20 days detail | Latest monthly statistics may differ from those previously released because of data verification after dissemination |
| Crude exports | General Administration of Customs | Monthly | About 7 to 9 days | As above. Volumes are small, about 2 of a 56 million tonne residual in 2025 |
| Seaborne arrivals, cross check | Kpler, Vortexa | Daily to near real time | 0 to 3 days | Continuously revised, and excludes pipeline crude, so not a substitute for customs |
| Observed onshore tank stocks, cross check | Kayrros, Ursa Space, Vortexa, Kpler | Daily to weekly | 1 to 7 days | External floating roof tanks only; blind to fixed roof tanks and underground caverns |
| National crude inventory change | NBS annual energy balance | Annual | Two to three years | No strategic or commercial split, no monthly detail |
| Strategic reserve level | Not published | غير قابل للتطبيق | غير قابل للتطبيق | Last public update on locations and availability was 2017, per Kpler |
A large build that turned into draws after the Gulf disruption
The 2026 series computed by Russell on NBS and customs data, all in millions of barrels per day, runs: January and February combined a build of 1.24, March a build of 1.74, April a build of 0.43, May a draw of 0.50, June a draw of 0.94, July a build of 0.21 and August a draw of 0.64. The stated year to date averages from the same columns are a build of 1.41 for the first quarter, 1.16 for the first four months, 0.48 for the first seven months and 0.33 for the first eight. Those averages reconcile with the monthly figures to within rounding, which is a useful internal check on any series reproduced second hand.
Russell's own description of the August figure, published 15 September 2026: China dipped into its massive crude oil stockpile in August for the third month in the past four as increased refinery processing exceeded a modest increase in oil imports, with a surplus of 330,000 barrels per day for the first eight months. The comparison points from 2025 show how far conditions have moved: a build of 1.88 million barrels per day in November 2025, 2.67 million in December 2025 which Russell described as a record, and 1.13 million for the full year.
The swing is driven by supply disruption rather than by a change in policy. Monthly crude imports in the series fall from 11.99 million barrels per day in the January and February block to 8.41 million in July, a reduction of roughly 3.6 million barrels per day, and refiners have been meeting the shortfall from inventory. Our note on whether China is still stockpiling in 2026 tracks the policy question; this page is about the instrument.
At the time of writing the most recent month for which the full input set exists is August 2026. September NBS and customs data are due around the middle of October, so any figure quoted here should be read as through August.
Everything the identity cannot see lands in the residual
The most important error is unmeasured throughput. Russell attaches a standing caveat to every column: not all the surplus crude was likely to have been added to storage, with some being processed in plants not captured by the official data. Because the NBS universe excludes plants below the revenue threshold and is re scoped each year, any barrel run through an unreported plant inflates the residual and overstates the build, one for one. Joseph Webster of the Atlantic Council adds a harder version, writing in March 2023 that Chinese refineries are notorious for misclassifying production with recent instances of refineries outright underreporting production to evade taxes.
Non crude feedstocks cut the other way. Independent refiners run fuel oil and bitumen blend when crude import quotas are exhausted. Those volumes never enter the supply side of the identity because they are not crude imports, but if the processing is reported as crude processing then throughput is inflated and the build is understated. The scale is not trivial: China imported 21.6 million tonnes of fuel oil in 2025, about 376,000 barrels per day, down 10.4 per cent from 2024's record. The EIA addresses the problem by assumption rather than measurement, stating that although we know some portion of China's refinery capacity can utilize heavy fuel oil, we assume refinery run numbers are strictly crude oil for our estimates.
Then there are the errors that do not change the answer so much as change what the answer means. A residual counts every barrel in the system, including barrels that are not recoverable inventory. Filling new tank farms and new pipelines to minimum operating level consumes real crude that is not available stock, and China has been commissioning large new capacity. Pipeline fill and tank bottoms are material even in a transparent market: the EIA calculates that removing pipeline fill and lease stocks takes about 120 million barrels out of United States crude inventories, almost 30 per cent of the national total. Timing is another: customs records a cargo at clearance while tanker trackers record it at discharge, so the same barrels land in different months and the noise washes out over a year but not within one.
A common misunderstanding is worth correcting directly. Net product trade does not affect the crude identity at all. It matters for the adjacent apparent demand calculation, and it explains why a large crude surplus does not imply a collapse in domestic consumption. China exported 4.65 million tonnes of refined fuels in July 2026. Finally, two structural limits: the method cannot separate strategic from commercial stocks, and the EIA notes that since 2024 Chinese national oil companies have been directed to add oil to commercial stockpiles which function as a secondary strategic reserve, blurring traditional distinctions between commercial and government controlled reserves. Our note on why published reserve estimates vary by source covers how that plays out across publishers.
Reliable for sign and scale, and it may see more than the cameras do
The quantified answer on dispersion comes from the EIA, writing on 9 October 2025: depending on the source used and assumptions made, the range between different stock build estimates is 0.5 million barrels per day on average and can be as large as 1.1 million. Set against a build running at roughly 0.9 to 1.1 million barrels per day, an average dispersion of 0.5 million is about half the signal.
The spread is visible in a single month. For April 2026, Russell's supply balance gave a build of about 430,000 barrels per day, Emma Li of Vortexa put observed aboveground stocks rising at around 580,000 barrels per day to a record 1.24 billion barrels, and Muyu Xu of Kpler recorded an onshore build of 8 million barrels, roughly 267,000 barrels per day, taking onshore stocks to 1,232 million barrels. That is a factor of more than two across three credible estimators for the same month. For March 2026 the divergence was far larger, 1.74 million barrels per day on the balance against about 175,000 observed by Kpler.
The instinctive reading is that the balance is wrong and the cameras are right. Kpler's own reconciliation suggests otherwise. Its note of 13 May 2026 found an average oversupply of 780,000 barrels per day between September 2025 and February 2026, implying roughly 140 million barrels that did not appear in observed onshore stocks, and explains the gap as barrels that may largely be explained by barrels withdrawn from underground storage facilities or pipeline systems that are not covered by our monitoring systems. Satellite methods read external floating roofs. Gabriel Collins of Rice University's Baker Institute, testifying to the US China Economic and Security Review Commission in June 2024, states that radar cannot detect underground cavern storage, fixed roof tank installations or the contents of sealed underground facilities, naming Huangdao, Jinzhou, Zhanjiang and Huizhou. Collins and Shih Yu Hung costed the concealment in 2018: doming a tank costs roughly USD 1.0 to 1.2 million, concealing the Zhoushan facility about USD 60 million, and mined underground caverns are about 60 per cent cheaper per barrel than surface tanks. The modes the cameras cannot read are the cheaper ones to build, which means the residual may be the more complete measure and the census the incomplete one. Our page on whether satellites can see China's oil storage goes into the imaging limits.
So the verdict is narrow and usable. The supply balance is the only estimate of Chinese crude stock change that is free, timely and reproducible by anyone, and it captures barrels in places no camera reaches. It is reliable for sign and order of magnitude over a quarter or a year: three independent approaches put 2025 at 1.13, about 1.1 and roughly 0.9 million barrels per day. It is not reliable to better than roughly half a million barrels per day in any single month. The discipline that follows is simple. Quote the method, quote the period, quote a competing estimate, and never publish a single bare figure as if it were a measurement.
استمع وخذها معك
هل تفضل الاستماع إلى التسجيل الصوتي، أم تحتاج إلى العرض التقديمي للمراجعة الداخلية؟ يتوفر العرض التقديمي الكامل كحلقة بودكاست وعرض شرائح قابل للتنزيل.
When you see a headline China stockpiling figure, what do you check first?
الأسئلة المتكررة
Domestic crude production plus crude imports minus refinery crude throughput. The more careful version also subtracts crude exports and small volumes burned directly in industry. John Kemp's 2025 worked example runs 216 million tonnes of domestic output plus 578 million tonnes of imports, giving 794 million tonnes available, less 738 million tonnes processed, leaving a 56 million tonne residual, less about 2 million tonnes exported and small direct industrial use, giving roughly 54 million tonnes, about 400 million barrels or 1.1 million barrels per day.
Two publishers. The National Bureau of Statistics publishes crude output and refinery throughput in its monthly energy production release, roughly two to three weeks after month end, combining January and February into a single release. The General Administration of Customs publishes headline crude import and export volumes around the 7th to 9th of the following month, with the detailed origin breakdown around the 18th to 20th. Both revise. Customs states that latest monthly statistics may differ from those previously released because of data verification after dissemination, and NBS states that its above designated size reporting scope changes every year.
Not for the level, and not monthly. There is no published commercial or strategic crude inventory figure and no split between the two, and Kpler records that the last public update on the locations and availability of China's oil reserves was in 2017. NBS does publish an annual energy balance for crude oil containing an inventory change line, but it is annual, carries a two to three year publication lag and gives no strategic or commercial split. So the accurate statement is that the number traders need does not exist, rather than that China publishes nothing.
The EIA's published figure is the best citation: depending on the source used and assumptions made, the range between different stock build estimates is 0.5 million barrels per day on average and can be as large as 1.1 million. For April 2026, three credible estimators spanned roughly 267,000 to 580,000 barrels per day for the same month. For March 2026 the supply balance gave 1.74 million barrels per day against about 175,000 observed onshore by Kpler. Over a full year the convergence is much better: 2025 estimates cluster around 0.9 to 1.13 million barrels per day.
It is more direct but not more complete, and the two measure different things. Satellite methods infer volume from the position of external floating roofs, so they give a level rather than a change and offer site level detail. But Gabriel Collins of the Baker Institute states in testimony that radar cannot detect underground cavern storage, fixed roof tank installations or the contents of sealed underground facilities, and mined caverns are roughly 60 per cent cheaper per barrel than surface tanks. Kpler's own reconciliation attributes an unexplained 780,000 barrels per day gap over six months to barrels withdrawn from underground storage facilities or pipeline systems not covered by its monitoring. The serious practitioners use both, as the EIA does in building its China estimate.
Direction, turning points and annual magnitude. The 2026 series flipped from a build of 1.74 million barrels per day in March to a draw of 0.94 million in June on identical inputs computed identically, and that sign change was the market relevant signal. For annual totals three independent approaches land within about 20 per cent of each other. What it cannot give you is a stock level, a strategic versus commercial split, or any way to tell real inventory from unmeasured refinery runs, pipeline fill and statistical error. Treat a single month as a range, not a measurement.
احصل على التالي إسقاط معلومات استخباراتية
انضم إلى قادة الطاقة والصناعة واحصل على معلوماتنا التسويقية، ونمو الذكاء الاصطناعي، وهيكلة الإيرادات، مباشرة وبدون حشو.
اسمك مدرج في القائمة
أهلاً بكم في موجز نمو الطاقة، تابعوا بريدكم الوارد للاطلاع على النشرة القادمة.