Applied Economic Perspectives and Policy (2020) volume 00, number 00, pp. 1–28.

doi:10.1002/aepp.13049

Featured Article
Returns to Investing in Commodity Futures:
Separating the Wheat from the Chaff
Scott H. Irwin*, Dwight R. Sanders, Aaron Smith, and Scott Main

Scott H. Irwin is the Lawrence J. Norton Chair of Agricultural Marketing in the

Department of Agricultural and Consumer Economics at the University of Illinois at
Urbana-Champaign. Dwight R. Sanders is a professor in the Department of Agribusiness
Economics at Southern Illinois University-Carbondale. Aaron Smith is a professor in the
Department of Agricultural and Resource Economics at the University of California,
Davis. Scott Main is a former graduate student in the Department of Agricultural and
Consumer Economics at the University of Illinois at Urbana-Champaign.
Editor in Charge: Craig Gunderson
https://doi.org/10.1002/aepp.13048
*Correspondence may be sent to: E-mail: [email protected]

Submitted 16 November 2018; editorial decision 18 February 2020.

Abstract Commodity futures investment grew rapidly after its popularity exploded in
the mid-2000s. However, real-time performance has been disappointing. Our analysis
shows that the disappointing commodity returns were not driven mechanically by con-
tango or negative “roll yields.” We show that the expected return to individual commod-
ity futures is near zero before expenses, which implies net losses (before interest earnings)
will be equal to order execution and operating costs estimated at 3%–4% per year.
Finally, it is likely that rapid increases in commodity prices during 2004–2008 skewed
investor return expectations upward much like it did in the early 1970s.
Key words: backwardation, commodity, contango, futures price,
investment, risk premium.
JEL codes: D84, G12, G13, G14, Q13, Q41.

“… the question is whether commodity ETFs provide reasonably good

correlation to the commodities (spot?) market with minimal ‘net’ ero-
sion from contango (after considering the positive effects of occasional
backwardation). And, if not, is there anything better for retail investors
to gain exposure to this investment class without betting the farm
through a dedicated commodities trading house? If there’s nothing bet-
ter, then are the deficiencies in the commodity ETFs so chronic that one
should avoid them entirely, and simply stay out of the commodities
space? I can’t find anyone who seems to know the answers, so thank
you for any thoughts you might offer.” —e-mail message sent to one
of the authors by a bewildered commodity investor

© 2020 Agricultural and Applied Economics Association

1
Applied Economic Perspectives and Policy

Introduction
Investment in commodity futures is now regularly included in lists of alter-
native investments that should be considered in discussions about the portfo-
lio mix for investors. These investments include commodity index funds,
exchange traded funds (ETFs), and exchange traded notes (ETNs), all of
which either track broad commodity indices or individual commodities.
Investors find commodity ETFs especially appealing because of the ease of
access. Individuals with standard stock brokerage accounts can gain expo-
sure to commodity futures markets (e.g. WTI crude oil futures) without open-
ing a separate futures brokerage account. Moreover, the structure of ETFs
removes investor concerns about executing trades in the underlying futures
market, margin calls, and trading through delivery periods.
While each type of commodity futures investment is managed in a some-
what different manner, all such investments provide exposure to commodity
prices through the purchase of futures contracts in a specific market or a bas-
ket of markets. As a particular futures contract nears expiration (e.g. June con-
tract), it is sold and a more distant expiration month is simultaneously bought
to replace it (e.g. July contract). This mechanical “rolling” of futures contracts
allows the investment manager to maintain a constant exposure to a particu-
lar commodity futures market or group of markets. The vast majority of
investments provide passive, long-only exposure to commodity futures price
movements.
As shown in figure 1, commodity-linked investments boomed from the
early 2000s through 2012, growing from less than $100 billion to a peak of
almost $450 billion.1 The publication of several influential academic studies
helped fuel the boom by claiming that commodity futures investment gener-
ated “equity-like” returns and provided portfolio diversification benefits
(e.g. Gorton and Rouwenhorst 2006; Gorton, Hayashi, and Rouwenhorst
2007). A flurry of research studies in the private sector around the same time
corroborated and reinforced the findings in the academic studies. These
newer studies built upon earlier work that reported evidence of attractive
returns to commodity futures investment (e.g. Greer 1978, 2000; Bodie and
Rosansky 1980; Fama and French 1987; Ankrim and Hensel 1993).
As quickly as investment in commodity futures boomed, it went bust even
faster. Figure 1 reveals that after peaking during 2010–2012 investment in
commodity futures plummeted by more than half during 2013–2015, to a
low of $160 billion. While investment in commodities recovered, it remained
well below peak levels at the end of the third quarter of 2018. Not surpris-
ingly, there were numerous articles in the financial press chronicling the
demise of commodity futures investments during the bust years. For exam-
ple, a Wall Street Journal article (Dugan 2013) reported, “Pension funds and
other institutions are retreating from popular investments linked to commod-
ities after finding they did little to protect their portfolios against inflation risk
and the unpredictable returns of stocks.” An article in the Washington Post
(Samuelson 2015) entitled “Behind the Commodities Bust” began this way:

1
There is no definitive measure of total investment in commodity futures markets. The series from Barclays
in Figure 1 is the longest and most comprehensive, but it may contain some non-futures commodity invest-
ments, particularly in metals. The best measure is the Index Investment Data (IID) series from the
U.S. Commodity Futures Trading Commission (CFTC), but this series does not begin until December
2007 and the CFTC ceased collecting the data as of October 2015.

2
 Separating the Wheat from the Chaff

Figure 1 Total Global Commodity-Linked Investment, Fourth Quarter 2004 – Third Quarter 2018.
Source: Barclays [Color figure can be viewed at wileyonlinelibrary.com]
500

400

300
Billion Dollars

200

100

0
20 : 2
20 : 1

20 1
20 4

20 2
20 4

20 3

20 : 3

20 1
:4

20 3

20 2

20 1

20 4

20 3

20 2

:4

20 3
:2
:2

:
7:
:

:
16
06
07
04

05

08

10

10

11

12

13

13

14

15

16

17

18
07

09
20

20

20
Year : Quarter

“First was the dot-com bubble, then the housing bubble. Now comes the com-
modities bubble.”
Investors must have been sorely disappointed with the performance of
commodity futures investments or they would not have been so quick to
divest. The experience of the California Public Employees’ Retirement System
(Calpers), the nation’s largest pension fund, is instructive. In October 2012,
Calpers pulled out of 55% of its holdings in commodity indexes after losing
about 8% annually over the previous five years (Dugan 2013). Calpers was
a pioneer in pushing commodity index investing in the pension industry. A
similar but more recent example is the Teucrium Corn Fund (CORN) ETF.
As shown in figure 2, an investor in the Teucrium ETF who bought in January
2015 at a price of $24.98 per share watched the price sink to $18.19 by April
2018 for a loss of 27%. Meanwhile, the nearby futures price for corn rose from
$3.70 to $4.01 per bushel. This was not an isolated example. Table 1 shows
data on returns for twenty-three futures-based ETFs and ETNs that have
five-year track records. These funds had assets of $3.5 billion as of May
7, 2018. Only three of the twenty-three produced a positive return over the
previous five years. Worse, eight of the twenty-three funds lost over half of
their value over that time period, with an average total loss of 37% over the
previous five years and 16% over the previous three years. The puzzling
aspect of this performance is that it occurred during a period of upward price
movements for many commodities, which could lead even an experienced
investor to ask, “What happened to my money?”
One explanation for the disappointing commodity returns is the process of
“financialization,” which is simply a label for the wave of large-scale institu-
tional investment in commodity futures markets that started in the mid-
2000s. Theoretical models (e.g., Acharya, Lochstoer, and Ramadorai 2013;
Hamilton and Wu 2015; Basak and Pavlova 2016) show how buying pressure
associated with financialization can exert downward pressure on risk pre-
miums, or equivalently, upward pressure on commodity futures prices before

3
Applied Economic Perspectives and Policy

Figure 2 Nearby Corn Futures Prices and Teucrium Corn ETF Share Price, January 2015–
May 2018 [Color figure can be viewed at wileyonlinelibrary.com]
30.00 4.50

4.25
27.50
Nearby Corn
4.00
Futures Price
(right axis)
25.00 3.75

3.50
$/Share

22.50

$/Bushel
3.25

20.00 3.00
Teucrium Corn
ETF Share Price
2.75
17.50
2.50

15.00 2.25
Jan-15
Mar-15
May-15
Jul-15
Sep-15
Nov-15
Jan-16
Mar-16
May-16
Jul-16
Sep-16
Nov-16
Jan-17
Mar-17
May-17
Jul-17
Sep-17
Nov-17
Jan-18
Mar-18
May-18
Date

expiration. Main et al. (2018) test this prediction and find that the average
unconditional return for nineteen individual commodity futures markets is
approximately the same before and after financialization. Therefore, despite
the logical appeal of the aforementioned theoretical models, average returns
in commodity futures market appear to have been largely unaffected by the
process of financialization.
Disappointing returns to long-only commodity futures investments have
also been blamed on the “carry” structure of futures prices. The two basic
types of carry are contango and backwardation, where contango (backwarda-
tion) occurs when futures contracts further from expiration on a given date
have a higher (lower) price than those contracts closer to expiration. The pro-
cess of rolling long positions from lower-priced nearby to higher-priced
deferred contracts in a contango term structure is said to create a negative
“roll yield.” (e.g., Moskowitz, Ooi, and Pedersen 2012). It is widely accepted
in the investment industry that roll yield is a key part of the return generating
process for commodity investments. For example, a Wall Street Journal article
(Shumsky 2014) explains, “Most commodities ETFs get their exposure by
buying futures contracts, and over time they typically shift, or roll, their posi-
tions from nearby to later-dated contracts. Sometimes, they have to pay more
for the new contracts, which eats into returns.” Several researchers have dis-
sented from this conventional wisdom about roll yield and commodity
futures returns (e.g., Sanders and Irwin 2012; Bhardwaj, Gorton, and Rou-
wenhorst 2015; Bessembinder et al. 2016; Bessembinder 2018).
This discussion suggests there is a major puzzle regarding returns to com-
modity futures investments. On one hand, both academic and private-sector
research offers the prospect of commodity futures investments earning
“equity-like” returns. On the other hand, actual investor experience with
commodity futures investments, at least in the last decade, has clearly disap-
pointed. The purpose of this article is to analyze the reasons for the puzzling
returns of commodity futures investments in the last decade. In particular, are

4
 Table 1 Recent Single Commodity Exchange Traded Fund (ETF) and Note (ETN) Performance
Total Return (annualized)
($ millions) Expense
Symbol Fund Name Commodity Total Assets 3 Year 5 Year Ratio

USO United States Oil Fund WTI Crude Oil 1,976.5 −10.76% −11.74% 0.77%
DBO PowerShares DB Oil Fund WTI Crude Oil 372.9 −6.86% −10.86% 0.75%
UNG United States Natural Gas Fund Natural Gas 364.1 −19.74% −14.88% 1.30%
DGL PowerShares DB Gold Fund Gold 199.3 1.87% −3.40% 0.75%
BNO United States Brent Oil Fund Brent Crude Oil 98.7 −4.88% −9.48% 0.90%
USL United States 12 Month Oil Fund WTI Crude Oil 88.1 −4.21% −7.82% 0.86%
CORN Teucrium Corn Fund Corn 79.5 −7.57% −10.97% 1.00%
WEAT Teucrium Wheat Wheat 67.1 −11.34% −12.78% 1.00%
UGA United States Gasoline Fund Gasoline 45.9 −5.10% −8.10% 0.75%

5
NIB iPath Dow Jones-UBS Cocoa ETN Cocoa 42.0 −2.90% 1.55% 0.55%
DBS PowerShares DB Silver Fund Silver 20.1 −1.79% −7.69% 0.75%
SOYB Teucrium Soybean Soybeans 16.6 −1.92% −3.99% 1.00%
CANE Teucrium Sugar Sugar 12.8 −8.11% −10.59% 0.50%
CPER United States Copper Index Fund Copper 10.8 −0.19% −2.79% 0.80%
PTM E-TRACS UBS Bloomberg CMCITR Long Platinum ETN Platinum 10.1 −7.49% −8.73% 0.65%
UHN United States Diesel-Heating Oil Fund Diesel-Heating Oil 8.3 −5.84% −7.05% 0.75%
OLO DB Crude Oil Long ETN Crude Oil 8.0 −7.07% −10.54% 0.75%
UNL United States 12 Month Natural Gas Fund Natural Gas 5.9 −10.19% −10.71% 0.90%
UBG E-TRACS UBS Bloomberg CMCI Gold ETN Gold 3.7 2.94% −2.66% 0.30%
OLEM iPath Pure Beta Crude Oil ETN Crude Oil 2.8 −6.32% −9.34% 0.45%
UBC E-TRACS UBS Bloomberg CMCI Livestock ETN Cattle and Hogs 2.4 −5.34% 0.40% 0.65%
USV E-TRACS UBS Bloomberg CMCI Silver ETN Silver 2.1 −1.92% −7.45% 0.40%
LD iPath Dow Jones-UBS Lead ETN Lead 0.4 1.64% 0.34% 0.75%
Average −5.35% −7.36% 0.75%
Total 3,438.2

Notes: All data are from the ETF Database (etfdb.com) as of May 7, 2018 and includes long-only, futures-based, mechanical, unlevered funds that focus on single market or a very narrow market mix.
Funds had to have a five-year track record to be included.
Separating the Wheat from the Chaff
Applied Economic Perspectives and Policy

lackluster returns the norm or anomalous? Our main focus is on long-only

returns at the individual market level in order to: (i) reflect the passive,
long-only exposure desired by the vast majority of commodity futures inves-
tors; and (ii) investigate the fundamentals of commodity futures returns with-
out the complications introduced by portfolio strategies that are inherently
active in nature rather than passive.
In the first part of the article we use the theory of storable commodity mar-
kets to show that the disappointing returns of the last decade were not driven
by contango or negative roll yields, as is widely purported. Next, we show
how contango markets may have actually inflated investor return expecta-
tions due to misperceptions about spot versus futures investing. Then, we
use six decades of daily return data for nineteen storable commodity futures
markets to document that the expected excess return in individual commod-
ity futures markets is near zero before expenses. This implies that net losses to
commodity futures investments (before interest earnings) will be equal to
order execution and operating costs estimated at 3%–4% per year. This mir-
rors the actual experience of investors in commodity futures during the last
decade and is also consistent with the trading losses of commodity index
traders (CITs) in agricultural futures markets documented by Moran, Irwin,
and Garcia (2020). It is likely that rapid increases in commodity prices during
2004–2008 skewed investor return expectations upward much like it did in
the early 1970s, with investors disappointed by subsequent returns nearer
the longer-run average of zero before expenses. Finally, there is limited evi-
dence that the slope of the futures term structure (roll returns) provides a reli-
able signal about expected returns for individual commodity futures markets.

Basic Theory of Storable Commodity Futures Returns

We begin our analysis commodity futures investment with a graphical pre-
sentation of the basic theory of returns in storable commodity futures mar-
kets. Figure 3 shows hypothetical pricing curves for a commodity futures

Figure 3 Pricing of Storable Commodities under Contango and Backwardation Term Structure.
Panel A: Contango Term Structure. Panel B: Backwardation Term Structure

6
 Separating the Wheat from the Chaff

market with a spot (cash) price Pt equal to $100 on January 1. The figure shows
contango (Panel A) and backwardation (Panel B) term structures for com-
modity futures contracts that expire on December 31 of the same year. The
cost-of-carry, C, in the contango chart is $10 per annum. The components of
the carry include forgone interest, physical storage costs, and convenience
yield. The latter is the implicit benefit to owners of physical commodity inven-
tories for immediate use and it enters C as a negative value. The three possible
spot prices,PT,on December 31 are $105, $110, and $115 depending on the risk
premium (points A, B, and D on the chart in Panel A).
It is important to note that in this basic model the cost-of-carry is assumed
to be exogenous, and therefore, unaffected by the trading volume of any indi-
vidual trader or groups of traders. In other words, the model does not account
for order flow impacts, which, as we discuss later, may be an important com-
ponent of the cost of commodity futures investments.
The middle line in Panel A of figure 3 represents the case with no risk pre-
mium. This means that a trader could buy the commodity on January 1, pay
the storage fee, and be willing to sell on December 31 for $100 + $10 = $110.
Thus, the price on January 1 of a futures contract that expires on December
31 (Ft, T) is also $110. In equivalent terms, the current futures price equals
the expected value of the December 31 spot price (Ft, T = E(PT) = $110).In the
course of a year the spot price will rise to the futures price of $110 to compen-
sate owners of the physical commodity for storage costs. With no risk pre-
mium, the net return to a holder of the spot commodity over January 1 to
December 31 is zero ($110−$100−$10 = 0) after accounting for the cost C paid
to physically store the spot commodity. Likewise, the return to a futures
investor is zero because the futures contract purchased on January 1 at $110
is worth $110 at the expiration of the contract on December 31.2 No-arbitrage
conditions force convergence of spot and futures prices at contract expiration.
If a positive risk premium π is introduced then the returns to both spot and
futures are altered. As demonstrated by the upper line in Panel A of figure 3,
spot prices appreciate above the level dictated by cost-of-carry and holders of
the spot commodity earn a positive return after paying the cost associated
with storing the physical commodity. In this specific example, a spot investor
purchasing the commodity on January 1 at $100 earns a return of $115−$100−
$10 = $5 to compensate for the risk of owning the commodity. The futures
investor realizes the same return by purchasing the futures contract at $110
on January 1 and then selling the contract at $115 on December 31. The futures
investor is able to purchase the contract at $110 on January 1 because that is
the risk-free price of the commodity on December 31. The risk premium is
paid to the futures investor in the form of a downward-biased futures price (Ft,
T = $110 < E(PT) = $115). Since no-arbitrage conditions force convergence of
spot and futures prices at contract expiration, the same risk premium is
earned by spot and futures investors. In sum, the spot investor earns a profit
over and above the cost-of-carry and the futures investor realizes the same
return by purchasing the futures contract at a price less than the expected spot
price.
The upper line in Panel A of figure 3 represents the classic Keynesian theory
of normal backwardation in a storable commodity futures market. Since there
is a long for every short in the commodity futures market, someone must earn

2
The examples considered here do not have a stochastic (error) component. This is considered in the formal
mathematical presentation of the cost-of-carry model in the Appendix.

7
Applied Economic Perspectives and Policy

a negative return if the commodity futures investor earns a positive return.

The Keynesian theory of normal backwardation assumes the risk premium
is compensation paid by hedgers (shorts) to speculators/investors (longs)
for bearing price risk.
The lower line in the Panel A of figure 3 represents the case of a negative
risk premium. Here, spot price appreciation is below the level dictated by
cost-of-carry and the holder of the spot commodity will earn a negative return
($105−$100−$10 = −$5) after paying the cost associated with storing the phys-
ical commodity. Because the current futures price is biased upward compared
to the expected spot price (Ft, T = $110 > E(PT) = $105), a futures investor earns
the same negative return as the spot holder. A negative risk premium reverses
the logic of the Keynesian theory of normal backwardation. That is, hedgers
(shorts) earn the risk premium paid by speculators/investors.
Panel B in figure 3 represents a futures market with a backwardation term
structure.3 The three possible spot prices PT on December 31 are $95, $90, and
$85 depending on the risk premium (points A, B, and D on the chart in Panel
B). The cost-of-carry C is assumed to be -$10 per year, representing a market
situation where the convenience yield (negative value) is larger than the other
components of storage costs. This situation is not typical in most storable
commodity futures markets and tends to occur when the supply/demand
balance is “tight” and inventories are relatively low. Under such market con-
ditions, inventories will only be held by those firms for which running out of
inventory would be very costly. They pay a relatively high price to buy the
commodity and then reap the $10 convenience yield through maintaining
operations when others may not. Once again, the middle line in Panel B rep-
resents the case of no risk premium, so a spot investor purchases at $100 on
January 1 and sells at $90 on December 31. Somewhat counterintuitively,
the holder of the spot commodity earns a net return of zero ($90−$100 +
$10), reflecting the offset of warehouse, insurance, and interest costs by conve-
nience yield.
The return to a futures investor with no risk premium is also zero because
the purchase price on January 1 (Ft, T = E(PT) = $90) is the same as the selling
price at expiration on December 31 ($90). The introduction of a positive risk
premium, represented again by the upper line in Panel B, benefits the spot
holder who purchases at $100 and sells at $95. This offsets part of the depre-
ciation in spot prices as determined by the (negative) cost-of-carry, so the
net return is $100−$95 + $10 = $5. A futures investor purchasing on January
1 at $90 can expect the futures price to rise to the expected spot price at $95,
and thereby, earns the same return as the spot holder. Results are simply
reversed for the case of a negative risk premium with a backwardated term
structure (the lower line in Panel B).
The analysis based on figure 3 suggests two important results. First, the
carry structure of the futures market, whether contango or backwardation,
has no direct impact on the long return to holding a commodity futures con-
tract in terms of cash flows to investors. Second, the return earned by a long
futures investor is solely determined by the risk premium embedded in the

3
“Backwardation term structure” and “Keynesian theory of normal backwardation” are distinct concepts
that are often confused. The former refers to the difference between spot and futures on a particular trading
date. The latter refers to the expected change in the price of a particular futures contract between trading
dates.

8
 Separating the Wheat from the Chaff

futures price prior to expiration. If the risk premium is positive (negative)

then the long investor will earn a positive (negative) average profit.
A formal mathematical development the no-arbitrage cost-of-carry model
for storable futures markets can be found in the Appendix. There, we derive
expressions for continuously compounded returns to holding spot and
futures positions and show how these are related to risk premiums and carry-
ing costs.

Decomposing Commodity Futures Returns

A popular approach to characterizing long-only commodity futures
returns is to decompose returns into the sum of spot and “roll” returns (e.-
g. Moskowitz, Ooi, and Pedersen 2012) as follows,

Futures Return = Spot Return + Roll Return ð1Þ

We show in the Appendix how this accounting identity can be derived for-
mally in the cost-of-carry model and that it is identical to stating that the
futures return equals the spot return plus the carry. In other words, the carry
discussed in the previous section is the same thing as the roll return (but
sometimes stated with opposite signs).
The standard decomposition given by equation (1) is frequently used to
explain high and low returns to long-only commodity futures investments
based on the carry structure of prices in futures markets. Examples in the
financial press abound. For example, a recent Bloomberg.com article
(Nussbaum and Javier 2016) stated, “Part of the problem is how fund man-
agers try to mimic price changes. Rather than buy raw materials that have
to be stored, they use futures contracts. However, when those expire—
sometimes every month—returns suffer if contracts are replaced at higher
cost. That occurs when markets are in contango, meaning that commodities
for immediate delivery are cheaper than in the future, as they are now for
everything from corn to crude.”
The problem with the conventional wisdom regarding roll returns as a
driver of future returns is that no such link is implied by the standard cost-
of-carry model. The crucial observation is that no causal direction among
the components is implied by this representation of futures returns. In the
rational theory of storage (e.g. Williams and Wright 1991), the futures price,
spot price, and price of storage (carry or roll returns) are determined simulta-
neously. That is, a market in contango (negative roll return) or backwardation
(positive roll return) is determined by the supply and demand for storage in
that particular commodity market in conjunction with the simultaneous
determination of spot and futures prices. This means that the variable on
the left-hand side of the standard decomposition can be stated with equal the-
oretical validity as,

Spot Return = Futures Return −Roll Return ð2aÞ

Roll Return = Futures Return −Spot Return ð2bÞ

There simply is no theoretical basis for asserting a particular direction of

causality among the components of (1) in the standard cost-of-carry model.
Moreover, if a contango market caused a negative futures return in the sense
used by most investment practitioners, it would imply a gross violation of

9
Applied Economic Perspectives and Policy

market efficiency in futures markets, in that an extremely simple and highly

profitable trading rule could be easily implemented for a given market
(Sanders and Irwin 2012).
In sum, these results conclusively demonstrate that the popular discussion
around roll yield as a source of commodity futures returns is mistaken. Bes-
sembinder (2018) makes the point in even stronger terms, arguing that “…
the roll yield as an actual cash gain or loss to a futures investor is a myth.”

Roll Returns and the Performance Gap

The discussion in the previous section suggests there is considerable confu-
sion about the nature of returns to long-only commodity futures investments.
In particular, the belief that roll returns drive futures returns can lead to the
perception of a “performance gap” in commodity futures investments. We
demonstrate how the perception of a performance gap can develop by com-
paring (implied) spot prices and the value of the three largest single-
commodity ETFs— U.S. Oil Fund (USO), PowerShares DB Gold Fund
(DGL), and U.S. Natural Gas Fund (UNG). Our procedures are similar to
those found in Bessembinder et al. (2016). The USO series begins on April
10, 2006; the DGL series begins on January 5, 2007; and the UNG series begins
on April 18, 2007. All three ETF series end on June 30, 2017. We compute
implied spot price prices using a two-step procedure. First, the daily cost-of-
carry is estimated as Ct = [1/(Z − T)]  ln[Ft, Z/Ft, T], where Ft, T is the price
of the nearby futures contract with delivery date T, Ft, Z is the price of the
next-to-expire futures contract with delivery date Z, and Z > T. Second, the
daily implied spot price is estimated by discounting the nearby futures price
as Pt = Ft,T =eCt ðT − tÞ . Implied spot prices are estimated in order to provide a
consistent measurement of spot prices across all three commodities.4
Figures 4–6 plot the path of the three ETFs over their lifetime compared to
the corresponding (implied) spot price. We also plot the path of a simulated
ETF implied by holding a long position in nearby futures and rolling to the
next contract 21 days before first delivery day.5 We also subtract a daily fee
equal to the reported expense ratios for each of the three ETFs, i.e. annualized
0.76% for USO, 0.77% for DGL, and 1.7% for UNG.6 The USO ETF was
launched April 10, 2006 at a price of $68.02 per share, roughly equivalent to
the crude oil implied spot price on the same date of $68.31/barrel. By June
2014, the crude oil implied spot price had risen to over $100/barrel, but the
USO price had fallen to $38. In the latter part of 2014, the futures market went
into backwardation, both funds fell sharply, and the gap closed somewhat.
Figure 4 shows that the simulated ETF closely matches fluctuations in the
USO fund. In particular, the simulated and actual funds did not rebound with

4
Because crude oil, gold, and natural gas futures have contract expirations every month, the implied spot
prices are very close to the nearby futures throughout the sample, i.e., the results that follow would be very
similar if we used Ft,T in place of the implied spot price.
5
The simulated ETF prices could also be computed by subtracting carrying costs from the return on the
implied spot price.
6
Source http://finance.yahoo.com. These expense ratios are in line with ETFs on commodity indexes. We
collected the annualized expense ratio of four ETFs over 2008–2012. The average expense ratios for these
funds were: iShares S&P GSCI Commodity-Indexed Trust (GSG, expense ratio = 0.75), iPath DJ-UBS
Commodity Index Trust, Exchange Traded Note (DJP, expense ratio = 0.75), GreenHaven Continuous
Commodity Index (GCC, expense ratio = 0.85) and GS Connect S&P GSCI Enhanced Commodity Trust,
Exchange Traded (GSC, sxpense ratio = 1.25).

10
 Separating the Wheat from the Chaff

Figure 4 Daily U.S. Oil Fund (USO) Share Price Compared to WTI Crude Oil Price and a Simu-
lated ETF, April 10, 2006–June 30, 2017 [Color figure can be viewed at wileyonlinelibrary.com]
160

140
WTI USO Simulated ETF

120

100
$/barrel

80

60

40

20

0
2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017

the spot price in 2009 because the price of storage was very high in that
period.
Perhaps the most vivid example of the performance gap is found in gold,
where storage costs are very stable (essentially just interest costs) at 1.3%
per year. Although both the spot and futures show a positive average return
over this sample, the difference between the implied spot price and the
futures return is a very consistent 1.8% per year. The difference manifests
itself as a performance gap that accumulates in a stable fashion, as shown in
figure 5 for a simulated gold ETF that almost exactly matches the DGL.

Figure 5 Daily Powershares Gold Fund (DGL) Share Price Compared to LBMA Gold Price and a
Simulated ETF, January 5, 2007–June 30, 2017. [Color figure can be viewed at
wileyonlinelibrary.com]
2,000

1,800

1,600

1,400

1,200
$/ounce

1,000

800

600
Gold DGL Simulated ETF
400

200

0
2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017

11
Applied Economic Perspectives and Policy

Figure 6 Daily U.S. Natural Gas Fund (UNG) Share Price Compared to Henry Hub Natural Gas
Price and a Simulated ETF, April 18, 2007–June 30, 2017 [Color figure can be viewed at
wileyonlinelibrary.com]
16

14
Henry Hub UNG Simulated ETF

12

10
$/Million BTU

0
2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017

Natural gas prices decreased significantly during the sample period, as

shown in figure 6. High storage costs further compounded the misery for
investors in UNG. Figure 6 shows that, as the Henry Hub spot price declined
from $7.66/MMBtu in April 2007 to $2.89 in December 2014, the UNG fund
and the simulated ETF dropped to $0.28 and $0.30, respectively. In contrast
to gold, the simulated ETF returns do not match the UNG returns exactly.
We also saw in figure 4 that simulated crude oil returns did not match the
USO return exactly. These differences likely reflect differences in the trading
strategies used by the USO fund and those employed in our estimation proce-
dures. Trading strategies can matter because the implied price of storage can
differ between nearby and more distant futures contracts, especially for natu-
ral gas, which has a seasonal storage pattern.
There is a straightforward explanation of the ETF underperformance when
compared to spot commodity prices, i.e. the performance gap in figures 4–6.
The expected difference between spot and futures returns equals the carrying
cost (see equation (A5) in the appendix). Thus, if carrying costs are positive,
spot prices will appreciate relative to futures prices to cover the cost of stor-
age. As a result, an apparent performance “gap” will develop between the
price of the spot commodity and the value of a commodity futures invest-
ment. If carrying costs are negative—an inverted market structure—then this
apparent performance gap will narrow or possibly even reverse in sign. How-
ever, the narrowing only occurs because the spot price declines relative to the
futures price as a reflection of the convenience yield. The difference between
spot price returns and futures market returns is simply a function of the cost
of storage, or roll returns, which drives a wedge between the two price series.
Indeed, the price series would only coincide in the absence of carrying costs.
In sum, comparing long-only commodity futures investment performance
to a spot price benchmark for a market with a contango term structure will
always be frustrating to investors as the spot price performance represents
an unattainable strategy of buying-and-holding the spot commodity without

12
 Separating the Wheat from the Chaff

Table 2 Commodity Futures Markets and Sample Periods

Exchange Start Date End Date No. Daily Obs.

WTI Crude Oil NYMEX/CME 30-03-1983 30-06-2017 8,548

Heating Oil NYMEX/CME 05-06-1979 30-06-2017 9,504
RBOB (Gasoline) NYMEX/CME 03-12-1984 30-06-2017 8,095
Natural Gas NYMEX/CME 01-06-1990 30-06-2017 6,772
Gold COMEX/CME 02-01-1975 30-06-2017 10,622
Silver COMEX/CME 05-08-1963 30-06-2017 13,375
Copper COMEX/CME 01-07-1959 30-06-2017 14,390
Corn CBOT/CME 01-07-1959 30-06-2017 14,613
SRW Wheat CBOT/CME 01-07-1959 30-06-2017 14,601
HRW Wheat KCBOT/CME 09-04-1976 30-06-2017 10,378
Soybeans CBOT/CME 01-07-1959 30-06-2017 14,597
Soybean Meal CBOT/CME 01-07-1959 30-06-2017 14,565
Soybean Oil CBOT/CME 01-07-1959 30-06-2017 14,525
Rough Rice CBOT/CME 20-08-1986 30-06-2017 7,760
Oats CBOT/CME 01-07-1959 30-06-2017 14,564
Cotton NYBOT/ICE 04-12-1959 30-06-2017 14,247
Cocoa NYBOT/ICE 04-08-1969 30-06-2017 11,845
Coffee NYBOT/ICE 17-08-1973 30-06-2017 10,859
Sugar NYBOT/ICE 04-01-1961 30-06-2017 13,927

Notes: CBOT, Chicago Board of Trade; CME, Chicago Mercantile Exchange; COMEX, Commodity
Exchange; ICE, Intercontinental Exchange; KCBOT, Kansas City Board of Trade; NYMEX, New York
Mercantile Exchange.

paying storage. Conversely, comparing long-only commodity futures invest-

ment performance to a spot price benchmark in a backwardated term struc-
ture will always show the futures investment outperforming the spot. A
trading strategy in futures does not entail holding the physical commodity
and its return includes no compensation for storage costs as these costs are
already built into the term structure of futures prices. Since contango is the
norm in storable commodity markets, these examples highlight that a “per-
formance gap” between spot prices and the value of commodity investments
is also the norm. To the degree that commodity investors base return expecta-
tions on spot prices, this will lead to disappointment about “poor” returns
when the actual problem is the use of a wrong performance benchmark.

Historical Commodity Futures Returns

It is certainly reasonable for investors to ask why they made or lost money
in a long-only commodity futures investment, but this is a function of risk pre-
miums rather than roll returns (see figure 3 and related discussion). As Bhard-
waj, Gorton, and Rouwenhorst (2015) state, “The source of value to an
investor in commodity futures is the risk premium received for bearing future
spot price risk.” To that end, we estimate historic average returns for nineteen
storable commodity futures markets over July 1959 through June 2017. The
markets, sample starting dates, and number of daily observations are listed
in table 2. The commodities include New York Mercantile Exchange
(NYMEX) energy and metals markets, Intercontinental Exchange (ICE) softs
markets, Chicago Board of Trade (CBOT) grain markets, and the Kansas City
Board of Trade (KCBT) wheat market. We do not include any livestock mar-
kets in our sample because these are nonstorable commodities and the cost-of-

13
Applied Economic Perspectives and Policy

carry model does not strictly apply. Over 10,000 daily observations are avail-
able in fourteen of the nineteen markets and the remaining five markets have
at least 6,700 daily observations. Therefore, the sample is broad in terms of the
number and types of commodities and deep in terms of the number of obser-
vations available for each market. This should provide a comprehensive per-
spective on the long-only returns available to investors in storable commodity
futures markets.
Daily (annualized) futures returns are computed as the log difference in
nearby futures prices. In all computations, returns are for the same nearby
futures contracts and contracts are rolled to the next contract twenty-one days
before first delivery day. The daily averages are annualized by multiplying by
250. Finally, the futures return that we compute is technically considered to be
an “excess” return because it does not add the interest (Treasury Bill) return
associated with a fully collateralized long-only commodity investment.7 In
order to simplify the exposition in the remainder of the article, we use the
terms “futures return” and “excess futures return” interchangeably.
The full sample estimates in table 3 show that average futures returns are
negative in twelve of the nineteen markets. Three markets show evidence of
statistically significant negative returns—natural gas, corn, and rough rice.
Only the crude oil and soybean complexes show consistently positive average
returns. The one statistically significant and positive return is copper. The
average futures return across the nineteen markets is −0.3% per year and it
is statistically indistinguishable from zero. Note that average returns are
reported in annualized form, which implies that daily average returns are
very small. Consider the largest annualized positive return of 7.1% for copper.
This translates into slightly less than three basis points per day. Most of the
average returns for individual markets are less than two basis points per day.
When the small size of the average daily futures returns in table 3 is com-
bined with the high variability of futures returns, it is not surprising there is
so little evidence of statistically significant futures returns. With daily stan-
dard deviations for the nineteen markets ranging from 1.2 to 2.9%, it is natural
to be concerned about low statistical power in such “noisy” return series
(e.g. Summers 1986). This issue is less of a concern when one considers the fact
that point estimates for average returns are negative for almost two-thirds of
the markets. In addition, the point estimate of average returns across the nine-
teen markets is negative. Hence, it is unlikely that average returns are actually
positive but we fail to detect significant positive returns due to low power
tests. We conclude that the full sample results in table 3 provide strong evi-
dence that realized excess returns for individual futures markets are
near zero.
An important fact to keep in mind is that the “paper” returns reported in
table 3 do not account for the costs of long-only commodity investments. Ear-
lier in table 1, we reported that the direct operational expenses for long-only
ETFs average 0.75% per year. Previous studies show that the order flow cost
of roll trades for long-only commodity futures investments is much larger
than direct operational expenses. Mou (2011) estimates the order flow cost

7
The addition of interest returns to excess futures returns to create a fully collateralized futures return will
create a return series that may have quite different statistical properties than excess returns. The addition of
interest returns will not only change the mean of the return series but also the volatility. For this reason,
means, standard deviations, or statistical test results for excess futures returns should not be compared
to those derived from fully collateralized returns.

14
 Table 3 Average Daily Returns (Annualized) for Commodity Futures Over Full Sample and by Decade, July 1959–June 2017
Full Sample Average Return By Decade
Average Return t-statistic 1960s 1970s 1980s 1990s 2000s 2010s

WTI Crude Oil 1.8% 0.30 5.6% 8.0% −18.0%

Heating Oil 1.4% 0.28 4.9% −0.7% 10.2% −8.7%
RBOB (Gasoline) 7.0% 1.18 6.0% 12.4% −4.0%
Natural Gas −23.7% −2.66 −16.2% −24.6% −32.3%
Gold −0.8% −0.28 −12.5% −6.8% 9.0% 1.0%
Silver −2.1% −0.53 17.9% −27.9% −4.2% 5.3% −2.5%
Copper 7.1% 2.11 0.6% 0.4% 3.3% 11.3% −3.6%
Corn −5.3% −1.90 −5.3% 5.8% −7.0% −8.9% −11.3% −5.0%
CBOT Wheat −5.0% −1.59 −5.7% 11.9% −6.8% −10.6% −8.8% −12.2%

15
KCBT Wheat −4.1% −1.18 −1.3% −3.5% −3.3% −2.1% −10.2%
Soybeans 2.5% 0.88 4.3% 10.5% −8.8% −3.8% 9.7% 4.0%
Soybean Meal 1.9% 0.59 4.1% 26.8% −10.0% −6.8% 3.5% −6.9%
Soybean Oil 6.2% 1.88 9.0% 8.4% −5.4% −1.9% 15.3% 12.2%
Rough Rice −7.8% −1.90 −9.1% −6.7% −13.7%
Oats −3.4% −0.98 −7.9% 4.8% −7.3% −18.2% 5.2% 3.6%
Cotton −1.1% −0.38 −2.8% 8.3% 3.6% −2.1% −15.7% 3.0%
Cocoa −0.6% −0.15 −29.3% 29.2% −18.9% −14.1% 9.0% −8.9%
Coffee −2.9% −0.57 −5.0% 1.6% −18.8% −9.5%
Sugar −6.3% −1.23 13.4% −23.9% −5.9% 6.1% −13.0%
All −0.3% −0.19 0.4% 13.6% −7.1% −5.4% 1.3% −6.6%

Notes: All entries are daily annualized returns (log price changes). Interest earnings are not included, so the reported returns are technically “excess” returns. Not all markets have data for the full
July 1959 through June 2017 sample period. See Table 2 for details on number of observations for each market.
Separating the Wheat from the Chaff
Applied Economic Perspectives and Policy

of roll trades for investments tracking the Standard & Poor’s Goldman Sachs
Commodity Index (SP-GSCI) to be 3.6% per year. In a similar vein, Bessem-
binder et al. (2016) estimate that the order flow cost of roll trades for the
USO ETF to be about 3% per year. The order flow impact arises because the
large relative size of roll trades pushes prices for expiring futures contracts
lower than they otherwise would be and the price of next to mature contracts
higher than they otherwise would be. This order flow impact, while costly to
commodity investors, is temporary because the term structure of futures
prices reverts to its starting shape once the rolling activities are completed.8
We can combine the above information to estimate total expenses for long-
only commodity futures to be in the range of 3%–4% per year. Total expenses
of this size would substantially increase net losses in the twelve markets with
negative gross returns and flip positive gross returns to negative returns in
four of the seven markets with positive gross returns. Average returns are
large enough to remain positive with expenses of 3%–4% per year in only
three markets—RBOB gasoline, copper, and soybean oil. We also note that
an average net loss of 3%–4% for buying-and-holding storable commodity
futures is consistent with the annualized net losses reported earlier for com-
modity ETFs during the last decade in table 1.
Further perspective is provided by examining the pattern of commodity
returns by decade. Returns for each decade since the 1960s are presented in
table 3. We present estimates only for complete decades. For example, a mar-
ket that started trading in 1978 does not show any data in the 1970s but shows
a complete return history for the 1980s. The data are arranged in this manner
to keep the markets within each decade consistent and complete. The decade
averages in table 3 reveal that positive futures returns have really only
marked two decades, the 1970s (13.6%, with eleven of twelve markets posi-
tive) and the 2000s (1.3%, with twelve of nineteen markets positive). The pos-
itive futures returns in the 1970s were mostly driven by the rapid price
adjustments in the early part of the decade. Sanders and Irwin (2012) note that
“…the three tumultuous years from 1972-1974, when the commodity markets
underwent dramatic structural shifts, accounts for 96% of the decade’s
returns and 68% of all returns during 1961–2010.” (p. 524) It is not surprising
that returns of this magnitude were accompanied by an increased focus on
commodities by investors. For example, Labys and Thomas (1975) observe
that there was a “switch of speculative funds away from traditional asset
placements and toward commodity futures contracts” following the spike
in prices during 1972–1974.
In a similar fashion, the upheaval in commodity prices from 2004–2010—
which was accompanied by popular themes such as “peak oil” and “food
for fuel”—piqued the interest of investors in commodity-related instruments.
Investment firms met this demand with index funds, ETFs, and other vehicles
that allowed investors to have convenient access to those markets. However,
much like what happened in the 1970s, it appears that the performance of

8
There is considerable confusion among market analysts and in the financial press about the difference
between order flow impacts of roll trades and whether commodity returns are mechanically related to the
term structure of futures prices. Order flow costs are incurred regardless of the shape of the term structure
of commodity futures prices at any particular point in time. The key is that the order flow pressure of roll
trades forces commodity investors to sell expiring contracts at lower prices than they would otherwise and
buy next to mature contracts at higher prices than they would otherwise. This order flow impact does not
depend on whether the term structure is in contango or backwardation. See Bessembinder (2018) for further
discussion along these lines.

16
 Separating the Wheat from the Chaff

Figure 7 Bootstrap Simulation of the Probability of the Average Futures Return over a 10-Year
Horizon Being Greater than Zero for 19 Storable Futures Markets, July 1959–June 2017 [Color fig-
ure can be viewed at wileyonlinelibrary.com]
90%

Probability of Average Futures Return > 0 80%

70%

60%

50%

40%

30%

20%

10%

0%
C d

er

SR W r
tin eal

gh n
B ean r

C n
C a

ur ce
l
So rud ns

as
TI yb )

W t
as il

il

at
ea Oi

W ga
BO oyb ppe

W So line

W hea
ol

at
o

ou or
to

fe
(G O

lv

at Ri

G
C ea

oc

he
H nM

O
of

R Su
ot

R C
Si
yb e

g
S Co

al
ea

N
H
R

Commodity Futures Market

those investments was overhyped and that investors anchored expectations

too much on recent high returns. Not surprisingly, subsequent years were
characterized by poor performance, with the 1980s averaging −7.1% and
the 2010s averaging −6.6%. Undoubtedly, commodity investors in both
periods were likely disappointed—and perhaps puzzled—by the lackluster
returns they earned. In this context, it is interesting to note there is well-
documented precedent of investment returns in the commodity space being
overhyped (e.g. Elton, Gruber, and Rentzler 1987, 1989; Edwards and Ma
1988; Irwin 1994; Bhardwaj, Gorton, and Rouwenhorst 2014).
The historical averages by decade provide a useful perspective on the pat-
tern of expected returns in commodity futures markets since the 1960s. How-
ever, these comparisons are limited by the fact that most of the markets have
different starting years, which means that all but the most recent decades con-
tain different sets of markets. In order to standardize the results across mar-
kets we conduct a bootstrap simulation of ten-year average returns where:
(i) 2,500 daily returns are randomly sampled from the full sample of observed
daily returns for each market, (ii) the average return over the 2,500 randomly
drawn returns is computed, and (iii) the simulation is repeated 1,000 times
and the proportion of ten-year average returns greater than zero are counted.
The results of the bootstrap simulation reported in figure 7 are striking. Only
seven of the nineteen markets have more than a 50% probability of positive
average returns over a ten-year horizon and only three markets have more
than a 70% chance of positive average returns. Equally interesting is the fact
that eight of the nineteen markets have a 40% or smaller probability of posi-
tive average returns over a ten-year horizon. These simulation results high-
light the limited potential for earning positive returns in long-only
commodity futures over a ten-year investment horizon.
Finally, the results reported in this section are consistent with the findings
in a number of previous studies that the average unconditional return to

17
Applied Economic Perspectives and Policy

individual commodity futures markets is approximately equal to zero. For

example, Kolb (1992) examines twenty-nine commodity futures markets
and finds evidence that eighteen had positive returns and eleven had negative
returns. He concludes that most markets exhibit no risk premium. Gorton and
Rouwenhorst (2006) analyze thirty-six individual markets and find that eigh-
teen had positive returns and eighteen had negative returns, with none of the
individual markets having statistically significant positive returns. Erb and
Harvey (2006) find that the average return to twelve commodity futures mar-
kets from 1982–2004 is −1.71% with no individual market producing statisti-
cally positive returns. Sanders and Irwin (2012) examine twenty commodity
futures markets and report that, outside of the 1970s, the number of markets
with negative returns roughly equals the number with positive returns and
the average return is relatively close to zero. Most recently, Bakshi, Gao,
and Rossi (2019) report that twenty of twenty-nine commodity futures mar-
kets over 1970–2011 have return–risk ratios (Sharpe ratios) below 0.25, a level
that they argue makes long-only commodity investment unattractive on a
stand-alone basis.
In sum, we agree with Kolb (1992) that “normal backwardation is not nor-
mal” in commodity futures markets. This conclusion is not as surprising
when one considers the fact that commodity futures contracts are not
included in the market portfolio because there is a short position for every
long position (Black 1976). In other words, futures markets are simply side
bets on commodity prices, not capital assets that earn a structural risk pre-
mium. An alternative explanation is that the supply of speculative services
in commodity futures markets is perfectly elastic (e.g. Telser 1958, 1967).

Risk Premiums and Roll Returns

While the average unconditional return for individual commodity futures
markets is approximately equal to zero, this does not necessarily imply that
the conditional return is also equal to zero. There is a long strand of the liter-
ature that reports evidence of time-varying risk premiums (e.g. Cootner 1960;
Carter, Rausser, and Schmitz 1983; Bjornson and Carter 1997; Moskowitz,
Ooi, and Pedersen 2012; Szymankowska et al. 2014; Bakshi, Gao, and Rossi
2019), where expected returns in commodity futures markets change in rela-
tion to the position of hedgers or financial market conditions. It is crucial to
recognize that any commodity futures investment based on conditional
returns requires some type of underlying dynamic trading strategy. That is,
earning conditional returns may require investors to follow a trading strategy
that goes both long and short in commodity futures markets. A less dynamic
strategy entails investors reducing their long exposure to a commodity based
on some type of signal, but not taking outright short positions. Regardless,
following any type of dynamic strategy is quite different from the passive,
long-only strategy we have analyzed up to this point.
A strategy that has attracted considerable attention in recent years is one of
buying commodities with low roll returns (“low basis/backwardation”) and
selling commodities with high roll returns (“high basis/contango”). In this
case, conditioning on the level of roll returns serves as a signal for the magni-
tude of risk premiums. Several studies show that this strategy produces pos-
itive and statistically significant returns (e.g., Erb and Harvey 2006; Gorton
and Rouwenhorst 2006; Bhardwaj, Gorton, and Rouwenhorst 2015). From a
conceptual standpoint, this strategy implies that the risk premium in the

18
 Separating the Wheat from the Chaff

cost-of-carry model (see figure 3) is negative for markets in contango and pos-
itive for markets in backwardation. The theoretical model in Gorton, Hayashi,
and Rouwenhorst (2013) provides rigorous justification for this type of strat-
egy. Both storage costs and the risk premium depend on inventories and price
volatility in their model. Under certain conditions, a market in contango will
tend to have less volatile prices, which implies that risk-averse investors
accept a lower risk premium than when the market is in backwardation. It
is important to emphasize that saying a market in contango is a signal that
the realized return in a subsequent period will be negative is quite different
from saying that rolling commodity futures positions in a contango market
creates a contemporaneous negative realized futures return.
We investigate whether roll returns serve as a signal for risk premiums in
the nineteen commodity futures markets included in our study. If such rela-
tionships exist, then it would imply that, even though there is no mechanical
connection between the futures term structure (roll returns) and commodity
futures returns, traders can profit by using roll returns as a trading signal.
We use a two-step procedure to decompose futures returns into spot and roll
returns. The first step involves computing daily returns as follows:



Ft + 1,T Ft + 1,T
Nonroll Days : ln = ln +0 ð3aÞ
Ft,T Ft,T




Ft + 1,Z Ft + 1,Z Ft,T
Roll Days : ln = ln + ln , ð3bÞ
Ft,Z Ft,T Ft,Z
|fflfflfflfflfflfflffl{zfflfflfflfflfflfflffl} |fflfflfflfflfflfflffl{zfflfflfflfflfflfflffl} |fflfflfflfflffl{zfflfflfflfflffl}
Futures Spot Roll
Return Return Return

Equation (3a) is used to compute futures and spot returns on nonroll days.
Note that futures and spot returns are the same on nonroll days because we
account for all the storage costs on roll days. Equation (3b) is used to compute
returns on roll days, or days when the contract in the nearby position changes
between day t and t + 1. On those days, the trader rolls out of the old nearby
contract that expires on date T and into the new nearby contract that expires
on date Z, with Z > T.
We assume that a trader holding the spot commodity pays for storage on
the roll day. As shown on the right-hand side of (3b), the roll return, ln[Ft,
T/Ft, Z], is the log difference between the nearby contract and next-to-expire
contract on day t. This is a standard way to compute storage costs for a com-
modity on a given date based on the term structure of commodity futures
prices. Note that this formulation results in negative roll returns when a mar-
ket has a contango term structure and positive roll returns when a market has
a backwardation term structure. Since the nearby contract changes on roll
days,the spot return shown on the right-hand side of (3b) includes the roll
return. Specifically, the spot return on roll days is the log difference between
the price of the nearby contract at t + 1, Ft + 1, Z, and the nearby contract at t, Ft,
T. This, of course, means the spot return on roll days is computed across dif-
ferent futures contracts. The futures return on roll days, ln[Ft + 1, Z/Ft, Z], is
computed using the “new” nearby futures contract at t + 1.
The second step in the procedure is to compute an average of the daily
returns generated by equations (3a) and (3b) for futures, spot, and roll returns.
In all three cases, a simple average is computed across all nonroll and roll

19
Applied Economic Perspectives and Policy

Figure 8 Average Daily Roll and Futures Returns (annualized) for 18 Storable Futures Markets
(natural gas excluded), July 1959 – June 2017. Notes: The observations denoted with filled circles
represent full sample averages for daily futures and roll returns (annualized). The error bars cor-
respond to 95% confidence intervals for the mean of each futures return series [Color figure can be
viewed at wileyonlinelibrary.com]
20%

15%
Average Daily Futures Return (annualized)

10%

5%

0%

-5%

-10%

-15%

-20%
-15% -10% -5% 0% 5% 10% 15%

Average Roll Return (annualized)

days. The averaging process is straightforward except perhaps in the case of

roll returns, which are zero except on roll days. By averaging across zero roll
returns on nonroll days and nonzero roll returns on roll days, the roll return
(storage cost) is allocated across all days in the sample. Equivalently, storage
costs are paid on a single day—the roll day—and then averaged across all
days.9
Figure 8 plots the average roll and futures returns, excluding natural gas
since it distorts the scales. Across all markets, the spot return averages a sta-
tistically insignificant 2.9%. The roll return is consistently negative (sixteen
of the nineteen markets) and the average roll return is −3.3% and statistically
significant, indicating that contango is the normal market structure. Impor-
tantly, roll returns and futures returns have the same sign in fifteen of the
nineteen markets. As suggested by visual inspection of figure 8, there is a
strong positive correlation in average roll and futures returns across markets
(correlation = 0.96), consistent with the evidence in previous studies that com-
modities with higher roll returns have lower futures returns and vice versa
(e.g. Erb and Harvey 2006; Gorton and Rouwenhorst 2006; Gorton, Hayashi,
and Rouwenhorst 2013; Bhardwaj, Gorton, and Rouwenhorst 2015). In other
words, storable commodity futures tend to exhibit higher returns when in
backwardation (higher roll returns) than in contango (lower roll returns).
While there is undoubtedly a high cross-sectional correlation between aver-
age futures and roll returns, “connecting the dots” between the cross-
sectional averages obscures the inherent variability in the time-series average

9
This method of computing futures, spot, and roll returns is generalizable to any holding period. Specifi-
cally, the method is equally applicable to monthly returns. In contracts which are rolled each month (like
crude oil), equations (A3a) and (A3b) will correctly calculate each component. In contracts which are rolled
at irregular intervals (like grains), (3a) is used for non-roll months and (3b) is used for roll months. Then,
similar to daily data, averaging the returns essentially prorates the roll return across all months.

20
 Separating the Wheat from the Chaff

for each market. Figure 8 include 95% error bars for average futures returns,
which illustrates the very substantial time-series variability obscured by only
focusing on cross-sectional averages.
These results lead naturally to the fundamental question of whether a
time-series relationship exists between roll returns and futures returns that
commodity futures investors can use to reliably increase returns over that of
a passive long-only strategy. To test this formally, we regress annualized
daily futures returns for each of the nineteen markets in this study on the roll
return (term spread) as of the most recent roll day. If the coefficient on the roll
return is statistically different from zero, then the roll returns provide infor-
mation about subsequent returns in the time series.
Table 4 shows that thirteen of the nineteen commodities exhibit higher
returns in backwardation periods (higher roll returns) than in contango
periods (lower roll returns) over the entire sample period. The coefficients
indicate statistically significant coefficients for six markets, with the estimates
providing evidence of higher returns in five of six markets during backwarda-
tion and lower returns in one market (rough rice). To further investigate the
reliability of these signals, we split the sample at the end of 1988. In the first
half of the sample, four of the nineteen markets have statistically higher
returns during backwardation. In the second half of the sample the relation-
ship seems to fade. That is, only two of the nineteen markets have statistically
significant differences between returns in contango markets and returns in
backwardated markets. One of those has higher returns in backwardation
(oats) and one has lower returns (rough rice).
In spite of the unevenness of the statistical significance of the results in
table 4, some of the point estimates are quite large. For instance, across
the entire sample the annualized return to crude oil increased by 3.45%
for every 1% increase in the roll return (term spread). However, even that
large of a return difference only generates a t-statistic of 1.27 across the
entire sample. The variability is further illustrated in silver where in the sec-
ond half of the sample a trader only holding long positions would have
experienced an annualized return that increased by 9.85% for every 1%
increase in the roll return, but it would essentially be indistinguishable
from luck (t-statistic = 0.88).
The combination of large coefficient estimates and low statistical signifi-
cance arises because time-series returns at the individual market level are
highly volatile. This volatility means that, even if the differences in returns
are stable at the values displayed in table 4, a trader could hold long positions
for years in a given commodity futures market and still not realize a benefit
from only buying in backwardated market conditions. These results do not
provide for high confidence in investment strategies based on changing con-
tango and backwardation. Moreover, the value of this signal varies widely
across markets and is much less prevalent in recent years, when eleven of
the nineteen markets show lower returns during backwardation. An investor
would have to be fortunate enough to pick the right market over the right
time period to realize any benefit from roll return signals. It is certainly debat-
able whether investing in commodity futures based on roll returns is a reliable
strategy or one that most investors in commodities would be willing to
follow.
In sum, picking up risk premiums based on the level of roll returns is highly
uncertain and requires following a dynamic trading strategy that has an
unclear foundation. In addition, the cost of implementing and managing

21
 Table 4 Results for Regression of Daily Futures Return (Annualized) on Daily Roll Return for Commodity Futures Markets, July 1959–June 2017
Jul. 1959–Jun. 2017 Jul. 1959–Dec. 1988 Jan. 1989–Jun. 2017
coef. t-stat. % contango coef. t-stat. % contango coef. t-stat. % contango
Applied Economic Perspectives and Policy

WTI Crude Oil 3.45 1.27 54 6.63 0.71 24 3.39 1.17 61

Heating Oil 0.49 0.22 68 1.59 0.46 56 0.12 0.04 71
RBOB (Gasoline) −3.01 1.79 43 0.93 0.16 31 −3.39 1.91 44
Natural Gas −0.98 0.56 82 −0.98 0.56 82
Gold 7.50 1.88 99 100 19.00 1.95 98
Silver 6.32 1.40 99 9.90 1.39 99 9.85 0.88 99
Copper 4.76 3.75 57 5.28 4.02 56 −0.09 0.02 57
Corn −0.47 0.55 79 0.67 0.68 70 −2.35 1.58 88

22
CBOT Wheat 0.95 1.11 82 1.01 1.15 76 −0.28 0.14 87
KCBT Wheat 2.26 1.87 76 2.36 1.56 68 2.18 1.30 80
Soybeans −0.13 0.11 70 3.31 1.80 72 −3.04 1.74 69
Soybean Meal 4.48 3.23 72 4.67 2.83 56 −1.64 0.38 89
Soybean Oil 1.49 1.08 56 4.81 2.37 64 −1.83 0.96 48
Rough Rice −4.02 3.23 88 −11.13 2.42 85 −3.09 2.39 88
Oats 1.92 2.52 69 −0.13 0.11 64 2.99 2.86 74
Cotton −0.60 0.82 68 −0.69 0.74 64 −0.77 0.69 73
Cocoa 3.84 3.22 74 4.12 2.91 59 1.41 0.49 85
Coffee 3.49 2.66 70 2.70 1.42 47 3.05 1.58 82
Sugar 0.29 0.39 62 0.25 0.24 71 −0.15 0.12 54

Notes: Daily roll return is the term spread on the most recent roll day as written in equation (3b). Not all markets have data for the full July 1959 through June 2017 sample period. See table 2 for
details on number of observations for each market.
 Separating the Wheat from the Chaff

dynamic strategies undoubtedly is higher than the 3%–4% per year estimated
earlier for passive long-only strategies. This only raises the performance bar
further. We agree with Erb and Harvey (2006): “There is, of course, reason
to doubt how broad based the demand for commodity TAA (tactical
asset allocation) might be. Many, if not most, investors interested in investing
in commodities are interested only in a long-only exposure to commodity
futures. A TAA approach is unacceptable to these investors because they
want to know that they will always have a well-defined long exposure to
the commodities market. Tactical strategies that allocate among commodities,
or go long or short commodity futures, will naturally leave these investors
wondering about what sort of portfolio exposure they happen to have at
any point in time.” (p. 91) The bottom line is that the return to a strategy based
on roll returns is not one that can be earned by passive investing, which to
date has been strongly preferred by investors.

Summary and Conclusions

The returns to long-only commodity futures investments have generally
disappointed since the explosion in their popularity during the mid-2000s.
The puzzling aspect of the poor performance is that it occurred when the
overall trend in commodity prices was upward. The purpose of this article
is to analyze the reasons for the lackluster returns of commodity futures
investments in the last decade.
The “carry,” or term structure, of futures prices is often blamed for disap-
pointing returns to commodity futures investments. We use the cost-of-carry
model for storable commodity prices to show that the carry in a commodity
futures market has no direct effect on the returns to long futures positions.
As Bessembinder (2018, p. 51) states, “…the roll yield as an actual cash gain
or loss to a futures investor is a myth.” We then demonstrate how this com-
mon misconception about carry and commodity futures returns can create
an apparent “performance gap” in the minds of investors. Using data for
the three largest single commodity ETFs, we show that the apparent perfor-
mance gap between spot price levels and long-only commodity ETF invest-
ments is due to an “apples and oranges” comparison problem. In particular,
comparing long-only commodity futures investment performance to a spot
price benchmark for a market with a contango term structure will always be
frustrating to investors as the spot price performance represents an unattain-
able strategy of buying-and-holding the spot commodity without paying
storage.
We next investigate daily returns for nineteen storable commodity futures
markets over July 1959 through June 2017 in order to provide historic evi-
dence about expected returns. Unconditional futures returns are very close
to zero on average, with twelve of the nineteen markets showing a negative
return. Daily average returns are very small, at most two to three basis points
per day. The average annual futures return is −0.3% per year across the nine-
teen markets and it is statistically indistinguishable from zero. Overall, the
results strongly suggest that the unconditional return in commodity futures
markets is near zero before expenses. We concur with Erb and Harvey
(2006) that the average commodity futures contract does not have an
“equity-like” return.
We estimate that direct operating costs and order flow costs for commodity
futures investments are 3%–4% per year, which combined with the result that

23
Applied Economic Perspectives and Policy

unconditional average returns are zero, implies that the expected net loss for
investment in long-only commodity futures (before interest earnings) is
around 3–4% per year. This mirrors the actual experience of investors in com-
modity ETFs during the last decade and is also consistent with the trading
losses of commodity index traders (CITs) in agricultural futures markets
documented by Moran, Irwin, and Garcia (2020). Hence, the attraction of pas-
sive, long-only commodity investments is rather elusive.
So, why the rush to commodities starting in the mid-2000s? The pattern of
historical commodity returns is instructive. Positive commodity futures
returns have only marked two decades—the 1970s and the 2000s. The positive
futures returns in the 1970s were mostly driven by the rapid price adjust-
ments in the early part of the decade and this led to an increased focus on com-
modities by investors. In a similar fashion, the upheaval in commodity prices
from 2004−2008—which was accompanied by popular themes such as “peak
oil” and “food for fuel”—piqued the interest of investors in commodity-
related instruments. Investment firms met this demand with ETFs and other
vehicles that allowed investors convenient access to those markets. However,
much like what happened in the 1970s, the performance of those investments
was overhyped, and investors anchored their expectations too much on
recent peak returns. There is certainly precedent for investment returns in
the commodity space being overhyped (e.g. Elton, Gruber, and Rentzler
1987, 1989; Edwards and Ma 1988; Irwin 1994; Bhardwaj, Gorton, and Rou-
wenhorst 2014).
Lastly, we investigate whether the slope of the futures term structure (roll
returns) provides a reliable signal about expected returns in storable com-
modity markets. The roll return is consistently negative (sixteen of the nine-
teen markets) and the average roll return is −3.3% per year and statistically
significant, indicating that contango is the normal market structure. We find
some evidence of a relationship between the slope of the term structure of
commodity futures markets and subsequent returns. While the difference in
returns between contango markets and backwardated markets can be rela-
tively large, the differences are highly variable with somewhat limited statis-
tical significance. Moreover, the strength of this finding varies widely across
markets and fades in the second half of the sample period, leaving investors
with little confidence in the value of this signal. In addition, following the sig-
nal requires that investors go both long and short in commodity futures mar-
kets, something quite different from the passive, long-only exposure desired
by the vast majority of commodity investors to date.
We purposely did not examine returns to portfolios of commodity futures
in order to focus on the fundamentals of returns at the individual market
level. Due to a “rebalancing bonus” or “diversification return,” it is possible
for portfolios of commodity futures positions to generate a positive return
despite unconditional returns of individual commodity futures being near
zero. However, the diversification return depends entirely on the strategy
used to rebalance the portfolio (Willenbrock 2011; Chambers and Zdanowicz
2014), and hence, portfolios of commodity futures represent a form of active
rather than passive investment. It is not clear whether investors will find com-
modities attractive if the case depends solely on a portfolio diversification
return. Of course, there is also the possibility that sophisticated dynamic strat-
egies can be used to earn “time-varying” risk premiums in commodity futures
markets (e.g. Bakshi, Gao, and Rossi 2019; Boons and Prado 2019). But the
new wave of basis and momentum-based strategies that purport to track such

24
 Separating the Wheat from the Chaff

premiums can blur the line between old-school managed futures and modern
active management, which may not be very attractive to the classic investor in
commodity futures markets.

References
Acharya, V.V., L.A. Lochstoer, and T. Ramadorai. 2013. Limits to Arbitrage and Hedg-
ing: Evidence from Commodity Markets. Journal of Financial Economics 109(2):
441–465.
Ankrim, E.M., and C.R. Hensel. 1993. Commodities in Asset Allocation: A Real-Asset
Alternative to Real Estate. Financial Analysts Journal 49(3): 20–29.
Bakshi, G., X. Gao, and A.G. Rossi. 2019. Understanding the Sources of Risk Underly-
ing the Cross Section of Commodity Returns. Management Science 65(2): 619–641.
Basak, S., and A. Pavlova. 2016. A Model of Financialization of Commodities. Journal of
Finance 71(4): 1511–1556.
Bessembinder, H. 2018. The “Roll Yield” Myth. Financial Analysts Journal 74(2): 41–53.
Bessembinder, H., A. Carrion, L. Tuttle, and K. Venkataraman. 2016. Liquidity, Resil-
iency, and Market Quality around Predictable Trades: Theory and Evidence. Journal
of Financial Economics 121(1): 142–166.
Bhardwaj, G., G. B. Gorton, and G. Rouwenhorst. 2015. Facts and Fantasies about
Commodity Futures Ten Years Later. Working Paper 21243, National Bureau of Eco-
nomic Research.
Bhardwaj, G., G.B. Gorton, and K.G. Rouwenhorst. 2014. Fooling Some of the People
all of the Time: The Inefficient Performance and Persistence of Commodity Trading
Advisors. Review of Financial Studies 27(11): 3099–3132.
Bjornson, B., and C.A. Carter. 1997. New Evidence on Agricultural Commodity Return
Performance under Time-Varying Risk. American Journal of Agricultural Economics 79
(3): 918–930.
Black, F. 1976. The Pricing of Commodity Contracts. Journal of Financial Economics 3
(1–2): 167–179.
Bodie, Z., and V.I. Rosansky. 1980. Risk and Return in Commodity Futures. Financial
Analysts Journal 36(3): 27–39.
Boons, M., and M.P. Prado. 2019. Basis-Momentum. Journal of Finance 74(1): 239–279.
Carter, C.A., G.C. Rausser, and A. Schmitz. 1983. Efficient Asset Portfolios and the
Theory of Normal Backwardation. Journal of Political Economy 91(2): 319–331.
Chambers, D.R., and J.S. Zdanowicz. 2014. The Limitations of Diversification Return.
Journal of Portfolio Management 40(4): 65–76.
Cootner, P.H. 1960. Returns to Speculators: Telser vs Keynes. Journal of Political Econ-
omy 68(4): 396–404.
Dugan, I.J. 2013. Pension Funds Cut Back on Commodity Indexes. Wall Street Journal
February 5, https://www.wsj.com/articles/SB1000142412788732476100457828617
2166045096 (accessed November 16, 2018).
Edwards, F.R., and C. Ma. 1988. Commodity Pool Performance: Is the Information
Contained in Pool Prospectuses Useful? Journal of Futures Markets 8(5): 589–616.
Elton, E.J., M.J. Gruber, and J.C. Rentzler. 1987. Professionally Managed, Publicly
Traded Commodity Funds. Journal of Business 60(2): 175–199.
. 1989. New Public Offerings, Information, and Investor Rationality: The Case of
Publicly Offered Commodity Funds. Journal of Business. 62(1): 1–14.
Erb, C.B., and C.R. Harvey. 2006. The Strategic and Tactical Value of Commodity
Futures. Financial Analysts Journal 62(2): 69–97.
Fama, E.F., and K.R. French. 1987. Commodity Futures Prices: Some Evidence on Fore-
cast Power, Premiums, and the Theory of Storage. Journal of Business 60(1): 55–73.
Garcia, P., S.H. Irwin, and A. Smith. 2015. Futures Market Failure? American Journal of
Agricultural Economics 96(1): 40–64.

25
Applied Economic Perspectives and Policy

Gorton, G. B., F. Hayashi, and K. G. Rouwenhorst. 2007. The Fundamentals of Com-

modity Futures Returns. Working Paper 13249, National Bureau of Economic
Research.
Gorton, G.B., F. Hayashi, and K.G. Rouwenhorst. 2013. The Fundamentals of Com-
modity Futures Returns. Review of Finance 17(1): 35–105.
Gorton, G., and K.G. Rouwenhorst. 2006. Facts and Fantasies about Commodity
Futures. Financial Analysts Journal 62(2): 47–68.
Greer, R.J. 1978. Conservative Commodities: A Key Inflation Hedge. Journal of Portfolio
Management 4(4): 26–29.
. 2000. The Nature of Commodity Index Returns. Journal of Alternative Investments
3(1): 45–52.
Hamilton, J.D., and J.C. Wu. 2015. Effects of Index-Fund Investing on Commodity
Futures Prices. International Economic Review 56(1): 187–205.
Irwin, S.H. 1994. Further Evidence on the Usefulness of CTA Performance Information
in Public Commodity Pool Prospectuses and a Proposal for Reform. In Advances in
Futures and Options Research, ed. D.M. Chance and R.R. Trippi, 251–265.
Greenwich, CT: JAI Press.
Kolb, R.W. 1992. Is Normal Backwardation Normal? Journal of Futures Markets 12(1):
75–90.
Labys, W.C., and H.C. Thomas. 1975. Speculation, Hedging and Commodity Price
Behavior: An International Comparison. Applied Economics 7(4): 287–301.
Main, S., S.H. Irwin, D.R. Sanders, and A. Smith. 2018. Financialization and the
Returns to Commodity Investments. Journal of Commodity Markets 10: 22–28.
Moran, N.M., S.H. Irwin, and P. Garcia. 2020. Who Wins and Who Loses? Trader
Returns and Risk Premiums in Agricultural Futures Markets. Applied Economic Per-
spectives and Policy This issue.
Moskowitz, T.J., Y.H. Ooi, and L.H. Pedersen. 2012. Time Series Momentum. Journal of
Financial Economics 104(2): 228–250.
Mou, Yiqun 2011. Limits to Arbitrage and Commodity Index Investment: Front-
Running the Goldman Roll. Working Paper, Columbia School of Business.
Nussbaum, A., and L.-A. Javier. 2016. Buy, Hold & Lose: How a Commodities Roll Under-
cuts Investors. Bloomberg.com, September 21, http://www.bloomberg.com/news/
articles/2016-09-21/buy-hold-lose-how-commodities-roll-is-undercutting-investors
(accessed November 16, 2018).
Pindyck, R.S. 2001. The Dynamics of Commodity Spot and Futures Markets: A Primer.
Energy Journal 22(3): 1–29.
Samuelson, R. J. 2015. Behind the Commodities Bust. Washington Post. August 26, https://
www.washingtonpost.com/opinions/behind-the-commodities-bust/2015/08/26/
a56b35d4-4c07-11e5-902f-39e9219e574b_story.html?utm_term=.4f035bc43824 (accessed
November 16, 2018).
Sanders, D.R., and S.H. Irwin. 2012. A Reappraisal of Investing in Commodity Futures
Markets. Applied Economic Perspectives and Policy 34(3): 515–530.
Shumsky, T. 2014. Commodity ETFs Get a Performance Boost This Year. Wall Street Journal,
July 16, http://blogs.wsj.com/totalreturn/2014/07/16/commodity-etfs-get-a-
performance-boost-this-year/ (accessed November 16, 2018).
Summers, L.H. 1986. Does the Stock Market Rationally Reflect Fundamental Values?
Journal of Finance 41(3): 591–601.
Szymankowska, M., F. de Roon, T. Nijman, and R.V.D. Goorbergh. 2014. An Anatomy
of Commodity Futures Risk Premia. Journal of Finance 69(1): 453–482.
Telser, L.G. 1958. Futures Trading and the Storage of Cotton and Wheat. Journal of
Political Economy 66(3): 233–255.
. 1967. The Supply of Speculative Services in Wheat, Corn, and Soybeans. Food
Research Institute Studies 7(Suppl): 131–176.
Tomek, W.G. 1997. Commodity Futures Prices as Forecasts. Review of Agricultural Eco-
nomics 19(1): 23–44.

26
 Separating the Wheat from the Chaff

Willenbrock, S. 2011. Diversification Return, Portfolio Rebalancing, and the Commod-

ity Return Puzzle. Financial Analysts Journal 67(4): 42–49.
Williams, J.C., and B.D. Wright. 1991. Storage and Commodity Markets. Cambridge, UK:
Cambridge University Press.

Appendix: Mathematical Development of the Cost-of-Carry

Model and Commodity Futures Returns
The “cost-of-carry” model (e.g. Tomek 1997; Pindyck 2001) is a well-
developed theoretical model of pricing storable commodities. The cost-of-carry
model can be used to formally analyze the drivers of returns to long-only com-
modity investments. The model is the same as that found in Bessembinder et al.
(2016) except we assume a constant storage cost in order to simplify the exposi-
tion. To be consistent with the graphical analysis in the article, let Pt represent
the spot price at date t, Ft, T represent the futures price at time t for delivery on
date T, and C is the per period cost-of-carry which includes foregone interest,
physical storage costs, and the convenience yield associated with having stocks
on hand. The cost-of-carry C normally is dominated by interest and physical
storage costs, in which case the futures market is in a normal carry or contango
(C > 0). Other times, the convenience of having stocks on-hand dominates, in
which case the futures market is inverted (C < 0). In all cases, the cost-of-
carrying inventory is revealed by the term structure of the futures market.
It should be noted that there is one important exception to the result that the
cost-of-carrying inventory is revealed by the futures term structure. Garcia,
Irwin, and Smith (2015) show that the futures term structure provides a
downward-biased measure of the cost-of-carrying inventory when the mar-
ket price of physical storage exceeds the maximum storage rate allowed by
the delivery terms of the futures market. This situation occurred frequently
over 2005–2010 for grain futures markets, and Garcia, Irwin, and Smith show
how this explains the much-discussed episodes of nonconvergence that pla-
gued the markets during this time period.
The no-arbitrage cost-of-carry relationship can be expressed in mathemati-
cal terms as follows,

Ft,T = Pt eCðT − tÞ : ðA1Þ

The return on the spot commodity net of storage costs is,


Pt + 1
U t + 1 = ln : ðA2Þ
P t eC

Bessembinder et al. (2016) call Ut + 1 the ex post premium. It has two compo-
nents: (i) the ex ante risk premium (π), which is the return that holders of the
commodity expect to earn as compensation for risk; and (ii) the ex post price
shock (εt + 1), which includes unforeseen supply and demand shocks. To
make this clear, we write Ut + 1 = π + εt + 1.
Market forces imply that ex post price shocks εt + 1 should average zero, and
therefore the ex post premium is determined, on average, by the risk premium.
For example, if traders expect demand for the commodity to increase in the
future, then they will hold some of the commodity off the market to store in
anticipation of higher future prices. This action will cause current prices to

27
Applied Economic Perspectives and Policy

rise and eliminates any excess returns from storage. Finally, note that strictly
speaking this result applies to the mean of the exponential rather than the
level of Ut + 1. In a rational expectations equilibrium, the expected price next
period equals the current price plus the carrying  cost and the ex ante risk pre-
mium, i.e. E(Pt + 1) = Pteπ + C, which implies E eUt + 1 = eπ and, from Jensen’s
inequality, E(Ut + 1) < π .
Equations (A1) and (A2) can be used to express the continuously com-
pounded returns to holding spot and futures positions,


Pt + 1
ln = π + εt + 1 + C ðA3Þ
Pt


Ft + 1,T
ln = π + εt + 1 ðA4Þ
Ft,T

where, as above, Ut + 1 = π + εt + 1. The gross return to the holder of the cash or

spot commodity in (A3) is the sum of the ex post premium Ut + 1 and the
market-implied cost-of-carrying the inventory C. From (A4), the return to a
long futures position is the ex post premium Ut + 1. Note, the difference in
returns to the spot (A3) and the futures (A4) is the cost of carry C. The spot
market prices will appreciate by the cost of carry, whereas that cost C is incor-
porated into the futures price Ft + 1, T.
Equations (A3) and (A4) yield three important predictions regarding
returns in storable commodity markets. First, if there is a risk premium
(π 6¼ 0), it appears in both the spot and futures returns. Different risk pre-
miums in spot and futures prices would be a violation of no-arbitrage condi-
tions. Second, if there is no risk premium (π = 0), then cash prices will change
by exactly the carrying costs (C) and futures returns are determined entirely
by ex post supply and demand price shocks. Carrying costs are incorporated
in the period t futures price Ft, T so they are not a component of the return from
period t to t + 1. Third, long futures returns in (A4) are not directly deter-
mined by the carry term structure of the futures markets or the level of C. In
other words, the long futures return is not a function of whether the futures
market is in contango (C > 0) or backwardation (C < 0).
Finally, we show how to derive the popular decomposition of long-only
commodity futures returns into the sum of spot and “roll” returns.
Equations (A3) and (A4) can be easily combined to demonstrate this common
decomposition. That is, solving (A3) for π = ln[Pt + 1/Pt] − εt + 1 − C and
substituting into (A4) we obtain the relationship that futures returns equal
spot returns minus the carry:



Ft + 1,T Pt + 1
ln = ln −C ðA5Þ
Ft,T Pt

In (A5), the carry C adds to the futures returns when the market is inverted
(negative C) and detracts from returns when the market is in contango (posi-
tive C). Note that C is defined exactly the same as the roll return in the stan-
dard decomposition (e.g. Moskowitz, Ooi, and Pedersen 2012), except the
sign is the opposite in our case. Therefore, by simply reversing the sign on C
in (A5) we obtain the identity commonly used to decompose the return to a
long futures position as Futures Return = Spot Return + Roll Return.

28