Which number tells you a market is tight before the price does?
Flows are hard to see. Nobody publishes a reliable real-time figure for world consumption. But the difference between production and consumption piles up — or drains — somewhere physical, and that somewhere is measurable. Inventories are the single most informative observable in a commodity market, because they are the accumulated record of every imbalance so far.
Days of cover, not tonnes
A raw inventory number is meaningless without a denominator. The standard normalisation:
days of cover = inventory ÷ daily consumption
For agricultural markets the same idea is expressed as the stocks-to-use ratio — closing stocks divided by a year's usage. Both answer the same question: if supply stopped, how long could the world keep going?
Why inventories change the elasticity itself
This is the part worth carrying. Inventory is the buffer that lets an inelastic market absorb a shock without a price move — so the same shortfall produces wildly different prices depending on how full the tanks are.
- Comfortable cover. 60 days of stock, a 2% shortfall for a month. Draw down about 0.6 days of cover. Nobody has to consume less. The price barely notices.
- Thin cover. 20 days of stock and falling, the same shortfall. Now buyers compete for the last barrels, and rationing by price is the only mechanism left. This is where the elasticity arithmetic from lesson 1 bites at full strength.
So a market is not simply "inelastic". It is inelastic once the buffer is gone — and inventories tell you where you are on that curve before the price does.
A worked example
A metal market: global consumption 60,000 tonnes a day, exchange warehouse stocks 300,000 tonnes.
- Days of cover = 300,000 ÷ 60,000 = 5 days.
- A single smelter outage removing 3% of supply for two months drains 3% × 60 days × 60,000 = 108,000 tonnes — more than a third of the entire visible buffer.
Now run the same outage against a market carrying 40 days of cover: it consumes about 4.5% of the buffer. Identical physical event, completely different price consequence, and the inventory number told you which world you were in before the outage happened.
The honest caveats
- Inventories are reported with lags — weekly for some energy series, monthly or quarterly for others.
- Only visible stock is counted. Exchange warehouses, reported commercial tanks and strategic reserves are published; private inventories, oil on ships and state stockpiles in several large countries are not. A "drawdown" is sometimes just metal moving somewhere unreported.
- The right denominator changes. Days of cover computed against last year's consumption misleads if consumption itself has shifted.
Treat inventories as the best available gauge, not as truth.
The curve reports the stocks nobody publishes. The calendar spread between the first two contract months is the price the market pays today to hold physical for one more month, and the people setting it include the owners of the private tanks, the cargoes at sea and the state stockpiles that the published series leaves out. A published draw with a prompt spread still at full contango means the barrels moved somewhere unreported; a flat inventory print with a spread moving into backwardation means the missing stock is being drawn. Read days of cover from the statistics and read the spread the same day; Why do traders watch the front spread more closely than the price? turns the spread into an implied storage number.
Try it now
- Pick a commodity and write down, before looking at anything, whether you would guess the world carries days, weeks or months of cover in it. Energy, industrial metals and grains give very different answers.
- Five years of natural gas is below — a market that ran genuinely tight in the middle of that window. Measure the sharpest run-up you can find, then ask the question that matters: was the physical shortfall behind it anywhere near that large in percentage terms? It never is. What you are measuring is the buffer emptying, not the shortfall.
- State the rule: the same shortfall is a rounding error at 60 days of cover and a crisis at 5.