The Day Oil Cost Less Than Nothing

At 2:08 p.m. New York time on Monday 20 April 2020, the May WTI crude future traded below zero for the first time in the 37 years since the contract was listed. Twenty minutes later it had fallen to −$39.55. At 2:29 p.m. it printed its all-time low of −$40.32 a barrel, and at 2:30 p.m. it settled at −$37.63: down $55.90, roughly 306%, in a single session. Sellers were not selling oil at any price. They were paying $37.63 a barrel for someone else to take it.

The collapse came out of an unremarkable tape. The contract had closed the previous Friday at $18.27 and opened the Sunday-evening session at $17.73. Nor was it a stampede out of a crowded position: the May contract traded 247,947 contracts that day, 10.4% of the 2.39 million WTI futures that changed hands, on a day the complex traded roughly double its normal 1.2 million. Almost 90% of the volume was in June and later months. The May contract triggered more than 30 dynamic circuit breakers on its way down, and the exchange let it trade.

The strangest part is what did not happen. The June contract, one month further out on the same curve, settled that same afternoon at $20.43. Brent settled at $25.57, down about 9%. Oil, as a commodity, lost a tenth of its value on 20 April. One specific contract, the one that was a day away from becoming an obligation to receive 1,000 physical barrels at Cushing, Oklahoma, lost everything and then $37.63 more. That gap is the whole story.

The price of oil that afternoon was not the price of oil. It was the price of not owning a tank.

The demand side vanished first. As lockdowns spread through March and April 2020, the IEA’s April Oil Market Report estimated global demand that month at 29 million barrels a day below a year earlier, a level last seen in 1995, with the second quarter 23.1 mb/d below and the full year on track for a record 9.3 mb/d decline. In the US, product supplied, the EIA’s proxy for consumption, fell to 13.8 million b/d in the week ending 10 April, the lowest reading in a weekly series that goes back to the early 1990s; the following week averaged 14.1 million b/d, about 31% below the pre-pandemic start of the year.

The supply side, at the worst possible moment, was fighting a price war. OPEC+ talks collapsed in Vienna on 6 March when Russia refused a further 1.5 mb/d of cuts, leaving quotas to expire at the end of the month. On Saturday 7 March, Saudi Aramco cut its official selling prices by the most in at least twenty years, the April Arab Light differential for Asia alone slashed by $6 a barrel. On 10 March it announced it would supply a record 12.3 mb/d in April, above its own 12 mb/d maximum sustained capacity, and a day later the energy ministry directed it to expand capacity to 13 mb/d.

The truce came on 12 April, eight days before the negative print: OPEC+ agreed to cut 9.7 mb/d for May and June. But the cut started on 1 May. Through the last three weeks of April, record supply met a demand hole of nearly 30 mb/d, and the difference had exactly one place to go: into storage. The IEA called the effective May cut 10.7 mb/d against April’s inflated baseline. For the May WTI contract, expiring 21 April, it was ten days too late.

WTI is not an abstraction. The contract’s delivery point is Cushing, Oklahoma, a landlocked pipeline crossroads with, per the EIA’s September 2019 survey, 76.1 million barrels of working storage capacity. Crude that cannot find a buyer, a refinery or an export dock ends up in a tank there, and in April 2020 it was ending up there at industrial speed.

Cushing held 37.9 million barrels in early March 2020. By 10 April it held 55.0 million; by 17 April, the last weekly print before the crash, 59.7 million, filling at roughly 5 million barrels a week. On paper that left a fifth of the hub empty. The EIA’s own commentary that week noted the unfilled space was “likely to have already been leased or otherwise committed”. Reuters, on 21 April, was blunter: five trade sources said the remaining tanks were fully leased out, new leasing had ground to a halt, and one broker put it plainly: “The terminals have already contracted their storage 100%.”

That is the detail that armed the trap. Officially, Cushing was about three-quarters full. Economically, it was sold out. A trader holding the May contract into expiry without pre-arranged tank space could not take delivery at any price, because the thing delivery requires, a place to put 1,000 barrels of oil, no longer existed on the open market.

A WTI future is a physical contract: hold it past its last trading day and NYMEX rules require you to settle by taking delivery of 1,000 barrels at Cushing, or to arrange an exchange-for-physical. In a normal month this is a technicality; the EIA notes only about 1% of contracts go to delivery, because financial holders roll or close well before expiry. The May 2020 contract stopped trading on 21 April. The 20th was the last full day to get out.

The exchange saw the zero bound coming. On 8 April, CME Clearing advisory 20-152 told member firms it had “a tested plan” for negative prices in energy contracts. On 15 April, advisory 20-160 opened a test environment so firms could rehearse negative futures and strike prices in crude. On 21 April, advisory 20-171 switched options valuation to the Bachelier model, which tolerates a negative underlying. When the print came, CME told Reuters its markets had “worked properly”. The impossible number had been wired into the plumbing twelve days in advance.

Who was left holding May? Open interest at the session open, 6:00 p.m. on Sunday 19 April, was 108,593 contracts, about 69% more than the average for a penultimate trading day over the prior year. It was not that nobody had left, May open interest had peaked at 634,727 contracts on 2 April and the big index funds and ETFs had already rolled, but that an unusually large tail remained: 173 of the day’s 448 reportable traders still held May positions. Roughly 95,000 contracts, twice the usual penultimate-day compression, had to find the exit in one session.

One mechanical footnote matters for reading that tape. June had become the exchange’s “active month” on Friday 17 April, so Monday’s May settlement was derived from the June settlement plus the volume-weighted price of May-June spread trades in the final two minutes, 2:28 to 2:30 p.m. The machinery functioned exactly as designed. It just cleared at a price nobody had ever seen.

Put the two contracts side by side and the crash resolves into a spread. On Friday 17 April, May traded $6.76 under June, already deep contango. At Monday’s settle the gap was $58.06: −$37.63 against $20.43. By Tuesday’s close it was $1.56, the blowout gone, as if nothing had happened. The market was not saying oil is worthless. It was saying, for one afternoon, that carrying a barrel from 21 April to 21 May, which requires a tank at Cushing, was worth $58, and the marginal holder of a May contract had no tank.

A futures curve in contango is a rental market for time: the spread is what the market pays whoever can store the commodity. Normally that rent barely registers in the spread. On 20 April it spiked to $58, because the one input the spread prices, tank space at the delivery hub, had gone unobtainable. Traders with storage were handed the trade of a lifetime; traders without it were the trade.

The next day showed the distress was about the calendar, not just one doomed contract. The expiring May future traded as low as −$16.74 before settling its final session at $10.01. June, now carrying the front of the curve, crashed 43% to $11.57, its lowest close in 21 years, after touching $6.50. July fell 31%. The negative print was May’s alone, but the storage problem belonged to the whole curve, and it took the OPEC+ cuts, from 1 May, and reopening demand to unwind it.

A futures curve is a rental market for time. On 20 April 2020, one month of tank space at Cushing cleared at $58 a barrel.

The purest losses landed on people who never knew they were short a tank. Bank of China’s “Yuan You Bao” (“Crude Oil Treasure”), a paper-crude product sold to retail savers, froze client accounts at 10 p.m. Beijing time on 20 April, while WTI still traded around $11, then settled the product at the exchange price of −$37.63 reached hours later. More than 60,000 investors lost their entire 4.2 billion yuan of principal and owed the bank a further 5.8 billion: a wipeout of roughly 10 billion yuan, about $1.4 billion. The banking regulator later fined Bank of China 50.5 million yuan and four of its staff a combined 1.8 million.

Interactive Brokers’ clients found the same trapdoor in cash-settled form. Several customers were long cash-settled WTI derivatives on CME and ICE Europe that marked to the −$37.63 settle; their losses blew through their account equity, and the broker had to make the clearinghouses whole. IBKR announced an $88 million provisionary loss the next day and ultimately booked $103 million compensating affected customers, per its SEC filing.

USO, the largest oil ETF, was the dog that did not bite, quite. It held no May contracts on 20 April, having rolled into June and beyond the week before. But it entered the week owning about 158,000 June contracts, roughly 27% of that contract’s open interest, while retail buyers poured $1.6 billion into the fund betting on a rebound. What followed was a forced retreat in public: five 8-K filings in eleven days, CME-imposed position limits, a sale of 110,000 June contracts in a single day alongside a Samsung ETF, and an 8-for-1 reverse split. The fund survived by becoming something else: a diversified-calendar product that no longer simply owned the front month.

And someone collected. Bloomberg reported that a group of traders at Vega Capital London, working from home in Essex, made on the order of $500 million to $660 million in a single afternoon. A US class action, Mish v. Vega Capital, alleges they did it by selling May futures aggressively into that final settlement window, and that the selling amounted to manipulation; the claims survived dismissal, a class was certified in June 2025, and the case is ongoing, with no court finding of wrongdoing to date. The CFTC’s own interim report, published that November, walked through the mechanics and declined to attribute the print to any single cause.

Oil was never worth less than nothing. The June contract said so, at $20.43, the same afternoon May settled at −$37.63. What went negative was an obligation: to take delivery of 1,000 barrels in a town whose tanks were spoken for. At expiry a future stops being exposure to a price and becomes the thing itself, and on 20 April 2020 the thing itself had nowhere to go.

Three durable lessons sit in the wreckage. Products that track the front month, however they are wrapped for retail, own roll risk and storage economics whether their buyers know it or not; that is what Yuan You Bao’s savers and USO’s dip-buyers actually purchased. The curve is information: the spread, not the headline price, was where the stress lived, visible for weeks before the print. And plumbing is destiny: the exchange had rewired itself for negative prices twelve days early, so the “unthinkable” number arrived with its paperwork already filed.

−$37.63 was not the price of a barrel of oil. It was the price of a promise to take one, with nowhere to put it.