Almost Commodities
The markets that never quite formed, and the ones that traded and then died. The failures are more instructive than the successes, because they show what a market actually needs.
A working commodity market is a rare and slightly improbable thing. For every crude oil, gold, or soybean that trades billions of dollars a day, there is a list of products that someone, usually a serious exchange with serious money behind it, tried to turn into a tradeable contract and could not. And there is a second list: contracts that traded actively for decades and then quietly went dark. Put the two lists together and a pattern emerges. The same handful of conditions decides every time whether a market lives.
The ones that almost formed
The most spectacular near-miss was bandwidth. In 1999, at the height of the telecom boom, Enron Broadband Services set out to do for internet capacity what the gas desk had done for natural gas: standardized forward contracts settled at physical “pooling points” that it explicitly pitched as bandwidth’s Henry Hub. The first trade, in December 1999, was a forward for DS-3 capacity on the New York to Los Angeles route, sold to Global Crossing. Williams, Dynegy, and a dozen others piled into a 2000 bandwidth-trading bubble. It collapsed completely. A circuit on one route at one quality of service was nothing like a circuit on another, so there was no fungible thing to deliver; the late-1990s fiber build then produced a glut that sent capacity prices toward zero; and the 2001 telecom bust and Enron’s own implosion finished it. Bandwidth never became a commodity.
Memory chipshave tempted exchanges for even longer. DRAM is the most commodity-like product the semiconductor industry makes, nearly fungible, violently cyclical, and the Chicago Board of Trade began designing a DRAM futures contract as far back as 1980. It surveyed Silicon Valley, found no appetite, and gave up. Enron tried a DRAM forward-trading service in 2001; it died with the company. Singapore’s exchange shelved a planned contract over the sheer difficulty of writing a deliverable standard for a product whose specification is obsolete within a year or two. (The thing people call “DRAMeXchange” is a spot-price reporting service, not a futures market.) Memory has a spot price, a famous boom-bust “silicon cycle,” and to this day no futures market at all.
The same itch keeps returning. CME launched California waterfutures in December 2020, the first of their kind, settled against an index of water-rights prices across five California markets; a UN human-rights official condemned the idea of trading water, but the contract’s real problem was prosaic: with no two water rights alike and almost no one needing to hedge, trading has run at a trickle of a few dozen contracts a day. CME’s Case-Shiller home-price futures, launched in 2006, have limped along near zero for the same reason: no house is fungible with another. And in the early 2000s Goldman Sachs and CME ran auctions in economic-data derivatives, contracts on the next jobs report or inflation print, which drew headlines, never drew liquidity, and were wound down within a few years.
The two markets the law actually bans
Cantor Fitzgerald and a rival won CFTC approval in 2010 to trade futures on Hollywood box-office receipts, “popcorn futures.” The studios lobbied furiously, and within weeks the Dodd-Frank Act amended federal law to make them illegal. That is how box-office receipts joined onions, banned in 1958, as the only two products the United States bans from futures trading by name, side by side in the same section of the statute.
The ones that traded and died
Death by obsolescence is just as common as failure to launch. The frozen pork belly, the contract that made the trading pit famous in Trading Places, traded on CME for some fifty years and was delisted in 2011. Bacon had simply gone year-round: with less product frozen and stored, the seasonal swings that gave packers a reason to hedge disappeared, and volume fell to a handful of contracts a month. The egg contract has an even longer arc. CME began life in 1898 as the Chicago Butter and Egg Board; standardized egg futures traded from 1919 and were once second only to grain. Then egg production industrialized into a few integrated giants, price volatility collapsed, the hedging need vanished, and the shell-egg contract was delisted in 1982 after more than sixty years.
Even big modern launches die. The London Metal Exchange listed a steel billetcontract in 2008 to great fanfare, more than a billion dollars of notional in its first year, but the large producers and merchants refused to use it, the futures drifted away from real billet prices, and it was delisted in 2017. Sometimes a contract survives only by reinventing itself: CME’s classic random-length lumber future, after the wild limit moves of 2020 to 2022 scared everyone off, was delisted in 2023 and replaced by a contract a quarter of the size. And in the 1990s an exchange even tried frozen shrimp futures, which failed within a few years for want of volume. Catastrophe-insurance futures, launched on the CBOT in 1992, met the same fate by 2000, even as catastrophe bonds, the same risk in a different wrapper, went on to thrive.
A field guide to markets that didn’t make it
| Market | The attempt | Why it didn’t hold |
|---|---|---|
| Bandwidth | Enron Broadband and others, 1999–2001 | No fungible unit (route, quality of service); fiber glut; telecom bust |
| Memory chips (DRAM) | Attempted from 1980 (CBOT), Enron 2001, never traded | Spec obsolete within a year or two; oligopoly, few independent hedgers |
| Box-office receipts | Cantor and Media Derivatives, CFTC-approved 2010 | Banned by the Dodd-Frank Act after studio lobbying |
| California water | CME NQH2O index futures, 2020 | No two water rights alike; almost no hedging demand |
| Home prices | CME Case-Shiller futures, 2006 | No house is fungible; near-zero liquidity |
| Economic data | Goldman / CME auctions on payrolls, CPI, ~2002–07 | No natural two-sided hedging need; thin, wound down |
| Catastrophe insurance | CBOT cat futures and PCS options, 1992–2000 | Cat bonds did the job better; exchange form delisted |
| Frozen shrimp | Minneapolis Grain Exchange, 1993–94 | Too little volume; poor hedging fit |
| Pork bellies | CME, ~1961, delisted 2011 | Bacon went year-round; seasonality and hedging need vanished |
| Shell eggs | CME, 1919, delisted 1982 | Industry integrated; volatility and hedging need collapsed |
| Steel billet | LME, 2008, delisted 2017 | Producers wouldn’t use it; futures diverged from physical |
| AI compute (GPU-hours) | SF Compute and others, 2024 onward | Live experiment; the unit keeps changing (see below) |
Dates and details verified against exchange notices, CFTC releases, and contemporaneous reporting. Onions (1958) and box-office receipts (2010) are the two products banned from US futures trading by statute; the rest failed or faded on their own.
The common thread
A market forms only when six things line up, and it dies when one of them gives way. The product has to be standardizable, a homogeneous, gradeable thing where one unit is genuinely interchangeable with another. It needs to be deliverable or cleanly cash-settled, either storable or backed by a reference price robust enough to settle against. There must be many independent buyers and sellers: a fragmented industry where lots of parties hold opposite risks. Those parties need a real, two-sided reason to hedge and enough price volatility to make hedging worth it. The reference price has to be hard to manipulate, with no single player big enough to corner it. And, trivially but decisively, it must be legal.
Run the failures through that checklist and each one trips a specific wire. Bandwidth, water, and houses were never fungible. Eggs, broadband, and DRAM were owned by integrated giants or an oligopoly, so there were too few independent hands needing to trade. Pork bellies lost their volatility when bacon went year-round. Economic-data contracts had no natural two-sided hedger. Onions and the old Maine potato contract were small enough to corner and manipulate, which is exactly why one ended in a famous default and the other in a federal ban. Box-office futures cleared every test but the last one and were simply outlawed.
The tech near-misses, bandwidth then memory and now AI compute, share one extra curse: the unit keeps changing. A megabit on a particular route, a specific DRAM generation, an hour on this year’s top GPU, none of them stays the same thing long enough to anchor a forward contract. The San Francisco Compute Company and a few rivals are now building marketplaces where buyers rent GPU clusters and resell what they don’t use, which is a real and growing business; but it is a rental market with a resale desk, not a cleared futures market, and it runs straight into the same wall that stopped bandwidth: by the time you could write a standard GPU-hour contract, the chip underneath has moved on. Whether AI compute becomes a true commodity will turn on whether anyone can define a unit that holds still.
The encouraging mirror image is that when the six conditions are met, even unpromising products trade beautifully. Electricity cannot be stored at all, yet power markets thrive on cash settlement against hub prices. Iron ore traded on fixed annual benchmarks for decades until the system broke in 2008, and the instant that many independent buyers suddenly needed to hedge a floating price, a deep swaps-and-futures market grew up in just a few years. The lesson of the graveyard is not that these markets were bad ideas. It is that a tradeable commodity is a specific, demanding structure, and you cannot will one into existence simply by wanting to trade the thing.