Commodity Index Investing: S&P GSCI and BCOM Methodology, Roll Yield, and the Evidence

Commodity index investing means holding a rules-based, fully collateralized basket of commodity futures that is rolled forward every month, rather than owning physical commodities. The two main benchmarks, the S&P GSCI and the Bloomberg Commodity Index (BCOM), publish exact rules for which contracts they hold, how they weight them, and on which business days they roll, and their total return equals the futures excess return plus the interest earned on Treasury bill collateral.

In short: a commodity index’s return has three parts. The spot return is the change in the price of the nearby futures contract. The roll return is the gain or loss from replacing an expiring contract with a later one, positive in backwardated markets and negative in contango. The collateral return is the Treasury bill interest on the cash that backs the position. A fourth source, the diversification return from rebalancing, helps explain why a portfolio can earn more than its parts. The evidence on whether long-only indices earn a reliable premium is genuinely contested: Gorton and Rouwenhorst (2006) found an equity-like premium in an equally weighted index; Erb and Harvey (2006) found that the average individual commodity earned roughly nothing, and that index returns depend heavily on weighting and the shape of the futures curve.

This deep dive builds on rolling futures contracts and futures term structure in depth. It is a description of how the indices and the research work, not a view on whether anyone should own them.

The misconception: “a commodity index tracks commodity prices”

An index does not track spot prices. It tracks a strategy: buy futures, hold them until a scheduled roll window, sell them and buy later ones, and keep the full notional value in Treasury bills. Over long periods that strategy can diverge enormously from spot prices, in either direction. Bloomberg’s own methodology describes its spot-price version, BCOMSP, as “not investable,” because its returns “cannot be actually replicated in the underlying futures markets.” Gorton and Rouwenhorst make the same point from the other side: in their data, the cumulative return to collateralized futures “has far exceeded the return to a holder of spot” commodities over 1959-2004.

How the two benchmarks are built

S&P GSCI

The S&P GSCI methodology (S&P Dow Jones Indices, 2026) describes the index as “a production-weighted index that is designed to reflect the relative significance of each of the constituent commodities to the world economy,” with eligibility limited to contracts with enough liquidity.

Weights. Each commodity’s World Production Quantity is a five-year sum of world production, averaged to give a World Production Average, based on data with a three-year lag. These set the contract production weights (CPWs), the number of units of each commodity the index holds.
Liquidity screens. A new commodity’s contract needs annualized total dollar value traded of at least $15 billion; contracts already in the index face lower thresholds. A contract also needs a reference percentage dollar weight (RPDW) of at least 1.00% to enter and 0.10% to remain.
Roll. The index rolls from the 5th to the 9th business day of each month, 20% a day: 80/20 on day 5, then 60/40, 40/60, 20/80, and fully in the new contract on day 9.
Return versions. The Excess Return index (ER) reflects futures returns only; the Total Return index (TR) adds “the daily interest on the funds hypothetically committed,” at a Treasury bill rate.

Because production weights reflect the dollar size of physical markets, energy dominates. Summing the 2026 RPDWs that S&P Dow Jones Indices announced on November 6, 2025 gives:

Sector (24 contracts)	2026 RPDW
Energy (WTI 17.79%, Brent 18.22%, gasoil, heating oil, RBOB, natural gas)	51.78%
Agriculture	17.02%
Industrial metals	12.13%
Livestock	11.19%
Precious metals	7.89%

The 2025 energy weight, by the same sum, was 57.41%. These are reference weights at the annual reset; actual weights drift daily with prices, and new weights are announced each November for the following January.

Bloomberg Commodity Index

BCOM, per Bloomberg’s index methodology, was launched on July 14, 1998, with history calculated back to a base of 100 on January 2, 1991. Its design goals are explicitly different:

Weights. Each commodity’s index percentage combines liquidity and production “using a ratio of 2:1”: two-thirds a five-year average of dollar trading volume, one-third a five-year average of dollar production. Bloomberg argues that production alone understates storable, investment-driven commodities such as gold.
Diversification caps, applied annually: no single commodity above 15%; no commodity together with its derivatives (for example WTI and Brent with ULS diesel, gasoline and gasoil) above 25%; no related group (energy, precious metals, livestock, grains) above 33%; and no commodity below 2% “as liquidity allows.” New commodities need an initial index percentage above 0.4% to qualify; existing ones are removed below 0.36%.
Roll. The 6th to 10th business days of each month, shifting 20% a day.
Total return. BCOMTR adds the return on “cash collateral invested in Treasury Bills,” using the most recent weekly auction high rate for 13-week bills, so that TR(t) = TR(t-1) x (1 + daily excess return + daily T-bill return).

The result is two indices that are both called “commodities” and behave differently: the GSCI is a leveraged view of energy; BCOM is capped and closer to a diversified basket.

Decomposing index returns

The identity

Hold the second-month contract, priced F2, until the roll, when it has become the nearby contract, F1 a month later. The monthly excess return is F1(next) / F2(now) - 1. That splits exactly into two factors:

(1 + excess return) = (F1 next / F1 now) x (F1 now / F2 now)

The first factor is the spot return, measured as the change in the nearby price. The second is the roll return, which depends only on today’s curve (the curve shapes are defined in contango and backwardation). Add T-bill interest on the collateral to get total return.

Worked example

Three illustrative 12-month paths, with a T-bill rate of 3.9% a year (credited monthly) and the curve shape held fixed:

Scenario	Spot return	Roll return	Excess return	Collateral	Total return
Contango of 1% a month, nearby rises $70 to $75	+7.14%	-11.26%	-4.92%	+3.97%	-1.12%
Contango of 1% a month, nearby flat at $70	0.00%	-11.26%	-11.26%	+3.97%	-7.70%
Backwardation of 1% a month, nearby flat at $70	0.00%	+12.82%	+12.82%	+3.97%	+17.25%

The arithmetic for the first row: each month the index buys the second contract at 1% over the nearby, a factor of 1/1.01 per month, so the roll return is (1/1.01)^12 - 1 = -11.26%. Spot rose 7.14%. Combined: 1.0714 x 0.8874 - 1 = -4.92%. In a persistent contango, a rising commodity price can still mean a losing index.

Is roll yield a real return?

Here the researchers disagree with the practitioners, and the disagreement matters. Gorton and Rouwenhorst note that “the rolling itself is not a source of return”: a futures contract has zero value after each daily settlement, so exchanging one for another has no cash flow consequence. The roll return in the identity above is accounting for the fact that the futures price was set relative to the expected future spot price. In a market where contango reflects expected spot price increases (full carry), the negative roll return is offset, on average, by spot price drift, and only the risk premium remains.

Gorton, Hayashi and Rouwenhorst (Review of Finance, 2013) make the same point with data: across 31 commodities from 1971 to 2010, the average basis was -1.1% a year (mild average contango) while the equally weighted excess return was +5.75% a year. A positive premium coexisted with negative roll yield. Erb and Harvey’s view is that the curve still matters enormously in practice: in their heating oil and gold comparison the spot components were both close to zero, so almost all of the 11-point gap in returns came from roll, and across commodities roll returns lined up closely with realized excess returns.

The evidence

Gorton and Rouwenhorst (2006): an equity-like premium

“Facts and Fantasies about Commodity Futures” (Financial Analysts Journal, 2006) built an equally weighted, monthly rebalanced index of collateralized commodity futures from July 1959 to December 2004. Its main results:

Risk premium: 5.23% a year over T-bills (t = 2.92), against 5.65% for the S&P 500 and 2.22% for corporate bonds. Sharpe ratio 0.43, against 0.38 for stocks.
Volatility and skew: annualized standard deviation 12.10%, below stocks’ 14.85%; positive skew, where equities were negatively skewed.
Diversification: negatively correlated with stocks and bonds at horizons beyond a month, reaching -0.42 with stocks at five years. In the worst 5% of months for stocks, when equities fell 8.98% on average, the commodity index returned +1.03%.
Inflation: positively correlated with inflation, rising with horizon to 0.45 at five years, while stocks and bonds were negatively correlated.

Erb and Harvey (2006): the premium is in the portfolio, not the commodities

“The Strategic and Tactical Value of Commodity Futures” (Financial Analysts Journal, 2006) reached sharply different conclusions. The figures below are from the working-paper version (NBER Working Paper 11222), covering December 1982 to May 2004:

Individual commodities: the average excess return of twelve GSCI constituents was -1.71% a year, with heating oil at +5.53% and silver at -8.09%.
The index: the GSCI earned a compound excess return of 4.49%, with its energy sector at 7.06% and non-energy at -0.12%. The GSCI’s high return reflected its energy weight.
Roll returns: heating oil’s 5.53% excess return split into roughly 0.93% spot and 4.60% roll; gold’s -5.68% into -0.79% spot and -4.90% roll. Across commodities, “roll returns explain 91% of the cross-sectional variation” in excess returns.
Diversification return: an equally weighted, monthly rebalanced portfolio of the twelve earned 1.01% a year, 2.72 points more than the average of its components, which they attribute to rebalancing rather than a risk premium.
Backfill: the GSCI’s history before trading began is hypothetical. From 1969 to 1991 its backfilled compound return was 15.3%, against 11.6% for the S&P 500; from 1991 to May 2004, 7.0% against 10.4%.

Their conclusion: a long-only strategic allocation “is a bet on the future term structure of commodity prices, in general, and on specific portfolio weighting schemes, in particular.”

Reconciling the two

The papers are not as contradictory as they look. Both find a positive return to a diversified, rebalanced portfolio. They disagree about where it comes from: Gorton and Rouwenhorst attribute it to a risk premium in the spirit of Keynes; Erb and Harvey to rebalancing and to the curve. Gorton, Hayashi and Rouwenhorst (2013) add a third piece: the premium is not constant but varies with inventories, highest when stocks are scarce, which is also when curves are backwardated. That is consistent with both camps’ data.

Worked example: the diversification return

Erb and Harvey approximate the diversification return as half the difference between the average variance of the constituents and the average covariance with the portfolio, which for an equally weighted portfolio equals the portfolio’s own variance. Take 20 commodities, each with 25% volatility:

Average variance: 0.25 squared = 0.0625.
With an average pairwise correlation of 0.09 (the figure Erb and Harvey report for their twelve commodities), portfolio variance = 0.0625 x (1/20 + 19/20 x 0.09) = 0.00847, a portfolio volatility of 9.2%.
Diversification return = (0.0625 - 0.00847) / 2 = 2.70% a year.
With an average correlation of 0.30, portfolio volatility rises to 14.5% and the diversification return falls to 2.08%.

That is a large number relative to the premiums being debated, and it depends on low correlations persisting. It is the main reason equal weighting and frequent rebalancing look good in historical commodity data.

The 2000s index boom

Index investing grew rapidly in the 2000s, and its role in the 2007-2008 commodity price spike became a policy fight. In testimony to the US Senate in May 2008, hedge fund manager Michael Masters argued that index buying had inflated prices. The CFTC’s response was a special call to swap dealers and index traders in June 2008, published as the “Staff Report on Commodity Swap Dealers & Index Traders” in September 2008. Its findings:

Net commodity index investment was estimated at $146 billion on December 31, 2007, $168 billion on March 31, 2008 and $200 billion on June 30, 2008, of which $161 billion was tied to US-regulated markets.
That $161 billion was about 17% of the $945 billion notional open interest in the 33 US markets in major indices, though as high as 47% in CBOT wheat.
In NYMEX crude oil, index notional rose from about $39 billion to $51 billion over the first half of 2008, but measured in futures-equivalent contracts, index traders’ long positions fell about 11%, from roughly 408,000 to 363,000, while oil rose from about $96 to $140. The dollar growth reflected prices, not new buying.

Scott Irwin and Dwight Sanders surveyed the subsequent research (“Index Funds, Financialization, and Commodity Futures Markets,” Applied Economic Perspectives and Policy, 2011) and concluded that “the lack of a direct empirical link between index fund trading and commodity futures prices casts considerable doubt on the belief that index funds fueled a price bubble.” The debate about financialization’s subtler effects, such as on correlations across commodities, continued after that.

How professionals run index exposure

Swaps and notes as well as futures. Much index exposure is held through total return swaps with dealers, who hedge in futures. That is why the CFTC had to survey swap dealers to measure it.
Roll timing. The GSCI and BCOM roll windows are public, so the market knows when billions of dollars will sell nearby contracts and buy deferred ones. Index providers publish variants with different schedules; the GSCI methodology lists a Pre-Roll family that rolls on the 1st to 5th business days.
Curve-aware variants. “Enhanced,” roll-select and longer-dated versions choose contracts further out, or based on the curve, to reduce contango drag. They are bets on the curve, not free improvements.
Collateral matters. The T-bill component is a real part of total return. When bill rates were near zero, as in 2009-2015 and 2020-2021, that leg contributed almost nothing.

What can go wrong

Persistent contango. As the worked example shows, a steep contango can turn rising prices into losses. Retail products that roll mechanically are the most exposed; see futures ETFs explained for the 2020 oil episode.
Concentration. With energy over half of the GSCI’s 2026 reference weight, a “commodity” allocation can be mostly a crude oil position.
Correlation drift. The diversification return and the stock-commodity diversification both depend on low correlations, which can rise in a crisis or under heavy index flows.
Backfilled history. Erb and Harvey’s comparison shows how much performance can depend on pre-launch, hypothetical data.
Regime dependence. Gorton and Rouwenhorst’s index outperformed stocks in the 1970s, but “this performance was reversed during the 1990s.” Any long-run average contains long stretches of poor results.

The short version

A commodity index is a rules-based strategy: hold futures, roll them on fixed business days, and collateralize with Treasury bills. The S&P GSCI weights by world production and ends up more than half energy; BCOM weights two-thirds by liquidity and one-third by production, then caps groups at 33% and single commodities at 15%. Returns decompose into spot, roll and collateral, with rebalancing adding a diversification return. Gorton and Rouwenhorst found an equity-like premium in a diversified index; Erb and Harvey found the average commodity earned nothing and that weighting and the curve drive results; inventory-based research suggests both are seeing a premium that is real but time-varying. The CFTC’s 2008 data and later research did not support the claim that index money drove the 2008 spike. The mechanics are simple. The evidence on what to expect from them is not.