Key takeaways
- The Bloomberg Commodity Index total return series compounded at 3.49% a year from its 2 January 1991 base to 31 July 2026. Bloomberg publishes one member of the family on both bases, and the gap between them is 2.79 percentage points a year. That gap is Treasury bills.
- Over the ten years to December 2014, an equally weighted index of commodity spot prices rose 9.42% a year while the collateralised futures version returned 5.09%. Holding the futures cost 4.33 points a year against holding the prices.
- Erb and Harvey measured heating oil and gold from December 1982 to May 2004. The excess returns were 5.53% and -5.68% a year. Only 1.72 points of that 11.22-point gap came from spot prices; 9.50 points came from the roll.
- The collateral leg is not a constant. In 2021 it was worth 0.06 percentage points. In 2024 it was worth 5.29, against an excess return of 0.59%.
- Over the 20 years to November 2011, the Reserve Bank of Australia found that negative roll returns had offset cumulative price increases of around 200 per cent, leaving the excess return close to zero.
A commodity index pays you three things, and the price of the commodity is only one of them
Copper is up, oil is up, and your commodity fund isn't. That's the question people actually type, and the answer is structural rather than bad luck.
A broad commodity fund doesn't own copper. It owns futures contracts, and a futures contract expires. To stay invested you sell the contract that's about to expire and buy a later one, over and over, for as long as you hold. Three separate things then happen to your money.
The first is the change in the futures price itself. The second is what you gain or lose each time you swap one contract for the next. The third is the interest on the cash that isn't tied up in margin.
Bloomberg's methodology names the first two and warns you they don't arrive separately. "The difference between the prices of the two contracts when they are rolled is sometimes referred to as a 'roll yield,' and the change in price that contracts experience while they are components of the Index is sometimes referred to as a 'spot return.'" Then the line that matters: "An investor in the Index cannot receive either the roll yield or the spot return separately."
The third leg is written into the index name. The Bloomberg Commodity Index Total Return "reflects the returns on a fully collateralized investment", which "combines the returns of BCOM with the returns on cash collateral invested in Treasury Bills", priced off "the most recent weekly auction high rate for 13 week (3 Month) U.S. Treasury Bills". An excess-return index is the futures leg alone. A total-return index is the futures leg plus a Treasury bill.
None of this is discretionary. The index shifts from the old contract to the new one "at the rate of 20% per Business Day" across the sixth to tenth business days of each month, whatever the curve looks like that week.
Since 1991 the Treasury bill leg paid 2.79 points a year and the futures leg paid about 0.70
Now put numbers on it, from the index provider's own fact sheets rather than from a fund's marketing.
The headline first. The Bloomberg Commodity Index total return annualised 3.49% from 2 January 1991 to 31 July 2026. It holds 25 exchange-traded futures across 23 commodities, with energy at 37.32% of the index and gold at 11.93%.
Bloomberg publishes only total-return figures for that headline index. It does publish one family member on both bases on the same sheet: the Roll Select variant. Its excess return annualised 3.51% since 2 January 1991 and its total return annualised 6.30%. The difference is 2.79 percentage points a year.
That difference is the collateral leg, and it's the same leg in every index in the family, because the methodology adds the identical Treasury bill return to each of them. Apply it to the headline index and the futures leg of the Bloomberg Commodity Index comes to roughly 0.70 percentage points a year over the whole period. The split isn't exact, because the two legs compound together rather than sitting side by side, but the shape holds.
So the thing most people think they're buying contributed about a fifth of what they got. The cash contributed the rest.
A central bank found the same answer from different data. In its November 2011 Statement on Monetary Policy, the Reserve Bank of Australia wrote that "over the past 20 years, roll returns on the GSCI have on average been negative, which has offset cumulative price increases of around 200 per cent and resulted in a net excess return over this period of close to zero."
Erb and Harvey's figures from a longer window point the same way. From December 1969 to May 2004 the GSCI total return compounded at 12.24% a year, against 6.33% for three-month Treasury bills and 11.20% for the S&P 500. The commodity index beat cash by 5.91 points a year, at an annualised standard deviation of 18.35%. That's the good half of the record, and it ends in 2004.
Contango and backwardation are the slope of a curve, and the slope is a storage bill
The roll leg has a name depending on which way the futures curve tilts, and the names get taught as definitions when they're really prices.
The Reserve Bank of Australia states the mechanism cleanly. When the next contract costs more than the spot price, a situation known as contango, "the roll return will be negative". When it costs less, which is backwardation, "the investor receives a positive roll return". You're not forecasting anything. You're paying or collecting the difference, every month, mechanically.
What sets that difference is physical. Someone has to hold the actual barrel between now and delivery, and holding costs money. The U.S. Energy Information Administration reads the curve as a storage gauge: "a steep contango indicating both high demand for and limited availably of storage". In April 2016 the crude market was working exactly that way. Between 1 and 22 April the WTI spread between the front month and one year out "dropped from $6/b to $3.57/b", as supply disruptions cut the incentive to store.
Now watch what that slope did to two real commodities over 21 years. Erb and Harvey measured December 1982 to May 2004. Heating oil delivered an excess return of 5.53% a year: 0.93% of spot and 4.60% of roll. Gold delivered -5.68%: -0.79% of spot and -4.90% of roll.
Read the spot columns again. Over two decades, the price of heating oil and the price of gold went almost nowhere in futures terms, and they went almost nowhere together. The 11.22-point gap in what an investor earned was 1.72 points of price and 9.50 points of curve. Across the individual commodities they tested, "roll returns explain 91% of the cross-sectional variation of commodity futures returns".
Bloomberg's own methodology flags the gold case in its risk section: "Certain commodities included in the Index, such as gold, have historically traded in contango markets." Gold is 11.93% of the index by weight, and a futures position in it has been structurally different from a bar in a vault. The separate question of how much gold belongs in a portfolio, and what its own long record shows, is argued in our piece on the 5-10% gold case and its 45-year problem.
The decade when spot rose 9.42% a year and the futures investor got 5.09%
Roll yield doesn't sit at a constant level. There are long stretches where it's the dominant term, and one of them is recent.
Bhardwaj, Gorton and Rouwenhorst maintain the longest commodity futures data set there is, an equally weighted index running from July 1959. Their Table 1 shows the collateralised futures return beside the matching spot price index, decade by decade, as arithmetic average annual returns.
From January 2005 to December 2014, spot prices rose 9.42% a year and the collateralised futures index returned 5.09%. Owning the prices would have beaten owning the futures by 4.33 points a year, for a decade.
Every earlier period runs the other way. July 1959 to December 1964: futures 3.88% against spot 2.98%. January 1965 to December 1974: 19.23% against 13.18%. January 1975 to December 1984: 9.01% against 7.08%. January 1985 to December 1994: 9.69% against 6.66%. January 1995 to December 2004: 8.55% against 8.20%. The chart below plots that wedge for each period.
Two cautions on reading the wedge. It contains the Treasury bill collateral as well as the roll, because the futures series is collateralised and the table doesn't separate the two. And the index is rebalanced monthly, so some of the wedge is a rebalancing effect rather than a curve effect.
Even so, the sign change is the story, and the Reserve Bank of Australia dates it to the same moment: "since the mid 2000s, aggregate roll returns have been negative and have accounted for much of the divergence between price and excess return indices". Bloomberg's own research desk described the steady state the same way in May 2021, writing that "over the long term, broad benchmarks tend to deliver a negative roll return which can dampen portfolio returns".
When the policy rate is zero, a third of the answer quietly disappears
The collateral leg is the least discussed and the easiest to see, because Bloomberg prints both versions of the Roll Select index for every calendar year.
In 2015 the excess return was -23.49% and the total return was -23.45%. Collateral was worth 0.04 percentage points. In 2021 the excess return was 27.87% and the total return 27.93%, a collateral contribution of 0.06 points. Cash paid nothing, so the two lines were the same line.
Then rates moved. In 2024 the excess return was 0.59% and the total return was 5.88%. The collateral leg paid 5.29 points, and it was the entire result. An investor who held a commodity fund through 2024 and read a mid-single-digit return as a commodity story had misread it. The commodities did almost nothing.
The same pattern shows in the annualised figures. Over the ten years to 31 July 2026, Roll Select returned 4.98% on an excess-return basis and 7.56% on a total-return basis, a 2.58-point collateral contribution. Whether a commodity allocation looks respectable therefore depends partly on what short-dated government paper happens to be paying, which is a fact about the money market rather than about metals or grain.
The strongest objection: roll yield isn't a fee, it's the absence of a storage bill
There's a serious case against the framing above, and it comes from the people with the longest data set.
Bhardwaj, Gorton and Rouwenhorst argue that comparing a futures return to a spot return is not a fair comparison at all. "The spot price appreciation is not a feasible investment return, because the holder of physical oil would have incurred storage costs which are not subtracted." Correct for carrying costs, they write, and "a fair comparison of collateralized futures to spot returns net of carrying cost would indeed show that both tracked closely following 2008."
Bloomberg concedes the point in its own methodology. The spot version of the index "provides a general estimate of the trend in commodity prices, without the positive or negative return effects caused by rolling futures or the costs involved in actually holding physical commodities", and it "is not 'investable'".
That objection reframes the decomposition without dissolving it. Negative roll yield isn't money taken from you by the index. It's the storage cost you avoided, priced by the market and charged through the curve. The practical consequence for a holder is identical: the spot chart in the news is not the return series you own.
The same team's out-of-sample verdict is more sympathetic to commodities than the 2005-2014 wedge suggests. "Between 2005 and 2014, the average risk premium of the EW index was 3.67% per annum, which is somewhat lower but not significantly different from the in-sample average of 5.23% during the 1959-2004 period." A decade of bad roll did not, on their measure, kill the risk premium.
And index construction can attack the roll directly. Roll Select "rolls into the futures contract showing the most backwardation or least contango, selecting from those contracts with nine months or fewer until expiration". The catch is on the same page: it was introduced on 18 July 2011, so most of its long record is a back-test, and Bloomberg's own disclaimer says "back-tested performance is not actual performance".
What this decomposition can't tell you
Start with the limitation in the headline split. The 2.79-point collateral figure is a difference between two annualised series, not a clean accounting line, because the futures and cash legs compound on each other. Treat it as the right order of magnitude rather than an exact attribution.
The 3.49% and the 3.51% belong to two different indices. One is the standard Bloomberg Commodity Index, the other the Roll Select variant, and the second was designed after the contango years it back-tests through. Comparing the two directly would flatter the design.
The Bhardwaj, Gorton and Rouwenhorst table is arithmetic rather than compound. Their own note says the returns are annualised "by multiplying by 12", which overstates what a holder actually kept in volatile decades.
All of this is a US futures record priced in dollars. A sterling investor's outcome also depends on the currency, and none of these series carry a fund's fees. Both matter, and neither is in the sample.
Above all, a decomposition is history, not a forecast. The wedge was positive in every period back to 1959 and negative only in the most recent one. Nothing in the arithmetic says which regime you're standing in.
What would change the conclusion
If short rates return to zero, the collateral leg goes with them, and a commodity index has to earn its return from the futures alone. The 2015 and 2021 columns are what that looks like: collateral worth 0.04 and 0.06 percentage points.
If curves flip back to persistent backwardation, the roll leg turns from a cost into a payment, and the 2005 to 2014 wedge reverses sign. Physical scarcity is what would do it. The EIA reads a steep contango as a shortage of storage rather than of the commodity, and reads a backwardated curve as inventories drawing down.
If the reason for holding commodities is inflation protection, the more direct instruments have their own documented failure mode, set out in our piece on the uplift, the lag, and 2022.
And if you find yourself comparing a commodity fund to a spot price chart, the question worth asking is which of the three legs that chart is showing. It is usually one of them.