Every week someone publishes a chart of ETF holdings or futures positioning and attaches a conclusion to it. Most of those conclusions do not survive contact with how the instruments actually work. This edition is a careful read of the paper market: what the series measure, how they connect to the metal, and where the popular interpretations break. Educational content only — not investment advice.
What an ETF holding actually is
A physically backed gold ETF is a wrapper. When demand for shares exceeds supply, authorised participants create new shares by delivering metal to the trust; when shares are sold persistently, they redeem shares and take metal out. Tonnage therefore rises and falls as a consequence of investor demand expressed through the share price relative to the metal.
That single mechanical fact disposes of the most common error in gold commentary. ETF flows are not an independent buyer arriving in the market; they are a record of investment demand already expressed. Tonnage is a thermometer. Reading it as a cause is reading the thermometer as the weather.
It remains a genuinely useful series, for two reasons. It is daily, published and unusually clean compared with almost everything else in this market. And because Western investment demand is the component most sensitive to real yields, ETF tonnage is the best available proxy for the driver that sets gold's ceiling.
Where the price is actually made
Spot gold does not print in a single place. In practice the price is discovered where the volume is: the futures market, the over-the-counter loco-London market, and the twice-daily auction that produces the benchmark used for settlement in physical contracts. Futures dominate visible volume by a wide margin; OTC dominates institutional size and is far less visible.
The consequence is that on any given day the price is set overwhelmingly by participants who have no intention of taking delivery. This is not a scandal — it is true of every commodity with a functioning derivatives market, and the same mechanism is what lets a miner hedge production and a jeweller fix input costs. But it explains why physical demand can be strong for months while the price is flat, and why the two only converge when delivery pressure or arbitrage forces them to.
Positioning data, without the conspiracy layer
Positioning reports show how different categories of participant are positioned. They are useful and they are systematically over-read. Three cautions travel with every use.
First, they are a snapshot with a reporting lag, so a large position may already be gone by the time it is discussed. Second, the categories are legal classifications rather than motives: a large short can be a directional bet, a producer hedge, or the offsetting leg of a long OTC position, and the report cannot distinguish them. Third, extremes are contrarian signals with a terrible sense of timing — crowded positioning can persist for months and become more crowded before it resolves.
What positioning is genuinely good for is describing the fuel available for a move. A market with heavily one-sided positioning has more forced flow to add to a reversal. That is a statement about the shape of a move, not a forecast of its date.
Lease rates, EFPs and the boring plumbing
The series that actually indicate stress in the paper-to-physical link get almost no coverage: gold lease rates, the exchange-for-physical spread between futures and loco-London metal, and the level and direction of registered exchange inventories.
When the EFP spread widens abnormally and lease rates rise together, the market is telling you that immediately deliverable metal in a particular location is harder to obtain than the flat price implies. That combination has preceded most of the genuine dislocations of the past two decades, and it is checkable by anyone who bothers to look. "Backwardation" repeated without either series is a slogan.
How to read a divergence
The recurring puzzle of this cycle is spot strength alongside flat or falling ETF tonnage. The framework resolves it cleanly: official-sector and Asian physical demand can absorb supply and support the floor without any of it appearing in Western ETF data, while Western investment demand — the ETF component — stays suppressed by positive real yields.
That divergence is informative rather than contradictory. It says the marginal buyer is price-insensitive and the price-sensitive buyer has not yet returned. The historical pattern is that when real yields turn and Western demand does return, it arrives into a market whose floor has already been raised. Whether that repeats is not knowable in advance, and anyone telling you the date is guessing.
Creation, redemption and the arbitrage that holds it together
It is worth being concrete about the link, because the vagueness is where bad conclusions grow. If ETF shares trade above the value of the metal per share, an authorised participant can buy metal in the loco-London market, deliver it to the trust, receive shares and sell them, pocketing the difference. If shares trade below, the trade runs the other way and metal leaves the trust. That arbitrage is what keeps the wrapper tracking the metal, and it is also why tonnage changes lag the price move that caused them rather than leading it.
Two implications follow. A day of heavy share turnover with no change in tonnage means buyers and sellers matched in the secondary market and no metal moved at all — most days look like this. And a tonnage change without a corresponding price move usually reflects a basket adjustment or a share-class transfer rather than fresh investment demand. Anyone quoting weekly tonnage as a demand signal without knowing which of these applies is quoting noise.
The four venues, and why they disagree
Loco-London over the counter settles unallocated credits in an account and dominates institutional size while publishing almost nothing in real time. The futures market publishes everything and dominates visible volume. The benchmark auction produces the reference price that physical contracts settle against. The Asian physical exchanges price locally deliverable metal against local demand and local policy.
Divergence between those four is information, not malfunction. A futures market at a wide premium to loco-London says deliverable metal is scarce in a particular vault system; a Shanghai premium with a soft London price says physical is being pulled east while paper is indifferent; a benchmark auction repeatedly clearing away from the surrounding spot prints says one side of the physical market is transacting with urgency.
The four checks before quoting a flow number
Is it tonnage or dollar value? Dollar flows rise with the price even when tonnage is falling, and the two get quoted interchangeably. Is it a single fund or the global aggregate? Regional flows frequently offset. Is the period a week, a month or year-to-date? Weekly ETF data is mostly noise. And does the source publish its methodology? If not, the number is an assertion.
What would falsify this framework
Two observations would force a revision. If ETF tonnage rose sustainably while the price fell, the wrapper mechanics described here would not be operating as stated. And if the EFP spread and lease rates stayed elevated for many months without either an inventory response or a price response, the arbitrage link between paper and physical would be weaker than we currently believe.
Two divergences worth learning to name
Price up, tonnage flat. Marginal demand is coming from somewhere that does not touch Western wrappers — official-sector accumulation, Asian physical, or futures-led flow. The move is real but the constituency behind it is price-insensitive, which historically means shallower corrections and less follow-through from momentum.
Price flat, tonnage up. Western portfolio demand is rebuilding into supply from another source, frequently recycling, which rises quickly after a price move because scrap has typically supplied a quarter to a third of annual supply. This configuration tends to precede the more durable advances, because the absorbing supply is finite.
Series to pull yourself
Exchange positioning reports and registered inventory data from the futures exchanges, fund-level tonnage from the ETF issuers' own daily disclosures, benchmark auction prices from the administrator, and published lease-rate curves. All are public. Any claim in this edition you cannot reconstruct from those is our interpretation, and should be treated as such.
On the programme
Market structure, delivery mechanics and the paper-physical link are the subject of the market-plumbing block on 10 October 2026. General Admission is free and includes ninety days of replay access. Editorial and educational only — not investment advice.