Does an upward-sloping curve mean the market expects higher prices?
No. And the reason is worth more than the answer, because the correct version of this idea is subtle enough that a lot of professional commentary gets it wrong.
Two quantities that are not the same thing
- F(t,T) — today's futures price for delivery at date T. Observable, published, cleared.
- E[S(T)] — the market's expected spot price at date T. Not observable by anybody, ever.
The relationship between them is:
F(t,T) = E[S(T)] − risk premium
If the people who must transfer price risk (typically producers, who sell forward) outweigh the people who want it, then whoever takes the other side has to be paid. They get paid by buying the future below the expected spot price and watching it drift up toward realisation. That is a risk premium flowing to the buyer, and it means the futures price sits below expected spot even in a market where nobody expects prices to fall.
Keynes's term, and the trap in it
Keynes called that condition normal backwardation. Here is the trap, and it catches almost everyone:
Normal backwardation is a statement about F versus expected spot. Ordinary backwardation is a statement about F versus today's spot. They are different comparisons, and a market can be in contango and normal backwardation at the same time.
A curve can slope upward (contango, because storage is expensive) while the futures price still sits below the expected spot price (normal backwardation, because producers are paying to hedge). Nothing is contradictory. Two different reference points.
If the hedging pressure runs the other way — a market dominated by consumers buying forward, an airline fixing jet fuel, a utility fixing gas — the premium flips sign and the futures price sits above expected spot. Keynes's word for that is, confusingly, contango as well. This is why practitioners generally avoid the Keynesian usage and reserve contango and backwardation for the observable slope.
The three things mixed into one number
Any deferred futures price contains, inseparably:
- Carry — financing, storage, insurance. Partly observable.
- Convenience yield — physical tightness. Not observable; a residual.
- Risk premium — who needs to hedge, and what they pay. Not observable; contested even in the academic literature.
You have one equation and three unknowns. Which is why the curve cannot be decoded into an expectation.
Watch it fail with numbers
Spot $78.00, six-month contract $80.30 — a clear contango.
- If the six-month risk premium is +2%, then E[S] ≈ 80.30 × 1.02 ≈ $81.90. The market "expects" more than the futures price says.
- If the premium is −2%, then E[S] ≈ 80.30 × 0.98 ≈ $78.70. The market expects roughly nothing to happen.
- If the premium is zero, E[S] = $80.30.
All three are consistent with the same observed curve. Anyone who reads "$80.30" as "the market expects $80.30" has silently assumed the third case — and has assumed away one of the most contested quantities in the field.
Empirical estimates of commodity risk premia vary by commodity, by era, and by method. They have been found positive, negative and indistinguishable from zero in different studies of different markets. Treat any specific number the way this Academy treats a term-premium estimate or a DCF output: a made-visible assumption, never a measured fact.
What the curve does tell you
It tells you the price of storage, the price of prompt physical, and the terms on which risk is currently being transferred between hedgers and everyone else. That is a great deal of genuine information about the present. It is not information about the future, and treating it as such is the most reliable way to sound authoritative while saying nothing.
For who is actually on each side, the Futures & Forwards course covers open interest and the Commitments of Traders report — with the same warning attached.
Try it now
- Take a commodity's current six-month contract price from an exchange settlement page and write it down as a single number. Then write down what you would have to assume about the risk premium for that number to be a forecast. Notice that you cannot check your assumption against anything.
- Five years of the same commodity's front-month price is below. Pick a starting date, Measure six months forward from it, and write down what actually happened over that window.
- Do it again from five more starting dates, spread across the window. Some will be up, some down, and the spread between them will be wide. That spread is your answer: a curve that were a forecast would not produce it.