July 3, 2015 8:27 World Scientific Handbook of Futures Markets. . . 9.75in x 6.

5in b1892-ch04 page 79

CHAPTER 4

COMMODITY FUTURES PRICES: SOME

EVIDENCE ON FORECAST POWER, PREMIUMS,
AND THE THEORY OF STORAGE
The World Scientific Handbook of Futures Markets Downloaded from www.worldscientific.com

Eugene F. Fama and Kenneth R. French

Abstract: We examine two models of commodity futures prices. The

by LA TROBE UNIVERSITY on 05/21/16. For personal use only.

theory of storage explains the difference between contemporaneous futures

and spot prices (the basis) in terms of interest changes, warehousing costs,
and convenience yields. We find evidence of variation in the basis in response
to both interest rates and seasonals in convenience yields. The second
model splits a futures price into an expected premium and a forecast of
the maturity spot price. We find evidence of forecast power for 10 of 21
commodities and time-varying expected premiums for five commodities.
Keywords: Theory of Storage, Normal Backwardation, Forecast Power,
Commodity Prices.

Introduction

There are two popular views of commodity futures prices. The theory of
storage of Kaldor (1939), Working (1948), Brennan (1958), and Telser (1958)
explains the difference between contemporaneous spot and futures prices in
terms of interest forgone in storing a commodity, warehousing costs, and a
convenience yield on inventory. The alternative view splits a futures price
into an expected risk premium and a forecast of a future spot price. See,
for example, Cootner (1960), Dusak (1973), Breeden (1980), and Hazuka
(1984).
The theory of storage is not controversial. In contrast, there is little
agreement on whether futures prices contain expected premiums or have
power to forecast spot prices. We use both models to study the behavior
of futures prices for 21 commodities. We find that more powerful statistical
tests make the response of futures prices to storage-cost variables easier
to detect than evidence that futures prices contain premiums or power to
forecast spot prices.

79
 July 3, 2015 8:27 World Scientific Handbook of Futures Markets. . . 9.75in x 6.5in b1892-ch04 page 80

80 E. F. Fama and K. R. French

We examine two models of commodity futures prices. The theory of

storage explains the difference between contemporaneous futures and spot
prices (the basis) in terms of interest changes, warehousing costs, and
convenience yields. We find evidence of variation in the basis in response
to both interest rates and seasonals in convenience yields. The second model
splits a futures price into an expected premium and a forecast of the maturity
spot price. We find evidence of forecast power for 10 of 21 commodities and
time-varying expected premiums for five commodities.
The World Scientific Handbook of Futures Markets Downloaded from www.worldscientific.com

The Basis: Evidence on the Theory of Storage

by LA TROBE UNIVERSITY on 05/21/16. For personal use only.

The theory of storage

Let F (t, T ) be the futures price at time t for delivery of a commodity at
T. Let S(t) be the spot price at t. The theory of storage predicts that the
return from purchasing the commodity at t and selling it for delivery at
T, F (t, T ) − S(t), equals the interest forgone, S(t)R(t, I), plus the marginal
storage cost, W (t, T ), less the marginal convenience yield from an additional
unit of inventory, C(t, T ) :
F (t, T ) − S(t) = S(t)R(t, T ) + W (t, T ) − C(t, T ). (4.1)
Equivalently,
[F (t, T ) − S(t)]/S(t) = R(t, T ) + [W (t, T ) − C(t, T )]/S(t). (4.2)
We call F (t, I) − S(t), or [F (t, T ) − S(t)]/S(t), the basis.
The marginal convenience yield, C(t, T ), arises because inventory can
have productive value. For example, there may be a convenience yield from
holding inventories of some commodities (such as wheat) because they are
inputs to the production of other commodities (such as flour). Or there
may be a convenience yield from holding inventories to meet unexpected
demand.
The theory of storage predicts a negative relation between convenience
yields and inventories. Brennan (1958) and Telser (1958) provide detailed
studies of the relations between convenience yields and inventories for
several agricultural commodities. Since good inventory data are not available
for many of the commodities studied here, we take a cruder approach.
Seasonals in production or demand can generate seasonals in inventories.
Under the theory of storage, inventory seasonals generate seasonals in the
marginal convenience yield and in the basis. We test for seasonals in the
basis.
 July 3, 2015 8:27 World Scientific Handbook of Futures Markets. . . 9.75in x 6.5in b1892-ch04 page 81

Commodity Futures Prices 81

Another implication of (4.2) is that, controlling for variation in the

marginal storage cost and the marginal convenience yield, the T − t period
basis for any stored commodity should vary one-for-one with the T −t period
interest rate. We provide (apparently the first) systematic tests of this well-
known implication of the theory of storage.

Data
The World Scientific Handbook of Futures Markets Downloaded from www.worldscientific.com

We construct monthly observations on the basis [F (t, T ) − S(t)]/S(t) and

the interest rate R(t, T ) for 1-, 3-, 6-, and 12-month maturities (T − t). The
interest rates are beginning-of-month yields on Treasury bills calculated from
by LA TROBE UNIVERSITY on 05/21/16. For personal use only.

the quotes in various issues of Salomon Brothers’ Analytical Record of Yields

and Yield Spreads. The sample period for interest rates is January 1967–May
1984.
Measuring the basis presents two problems. The first is that most futures
contracts do not have a specific maturity. Instead, there is a delivery period
of 3–4 weeks at the beginning of the maturity month. We assume that
contracts mature on the first trading day of the delivery month. This means,
for example, that the April 1, 1980, futures price for the May 1980 wheat
contract is used as a 1-month futures price.
The second complication is that good spot-price data are not available
for most commodities. We use futures prices on maturing contracts to
measure spot prices. For example, the spot price for wheat on March
1 is the futures price for the contract that matures in March. Since
futures contracts do not mature each month, this solution limits sample
sizes. The number of observations on the basis is always less than the
number of months in the sample period. On the other hand, using
maturing futures prices to measure spot prices ensures that spot and
futures prices are for the same commodity and are sampled at the same
time.
Table 4.1 summarizes the structure of the commodity-price data. Each
row shows the sample period for a commodity and the standard months in
which contracts mature. Contracts for these standard months usually begin
trading between eight and 12 months before maturity, and they are traded
every year. Copper, gold, and silver have supplemental contracts that fill in
the months between the standard contracts. These supplemental contracts
usually begin trading three or four months before maturity. They are useful
because they augment the spot-price series.
 July 3, 2015
82

8:27
E. F. Fama and K. R. French
The World Scientific Handbook of Futures Markets Downloaded from www.worldscientific.com

Table 4.1. Layout of commodity futures data.

World Scientific Handbook of Futures Markets. . .

by LA TROBE UNIVERSITY on 05/21/16. For personal use only.

Standard delivery months

Sample
Commodity Exchange∗ period Jan. Feb. Mar. Apr. May June July Aug. Sept. Oct. Nov. Dec.

Agricultural
products
Cocoa CSCE 3/66–7/84





Coffee CSCE 9/72–7/84





Corn CBT 3/66–7/84





Cotton CTN 3/67–7/84





Oats CBT 5/66–7/84





Orange juice CTN 2/67–7/84






Soybeans CBT 3/66–7/84







Soy meal CBT 5/66–7/84








Soy oil CBT 5/66–7/84








Wheat CBT 5/66–7/84





Wood products
Lumber CME 1/70–12/82






Plywood CBT 1/70–9/83

9.75in x 6.5in
b1892-ch04
page 82
 July 3, 2015
8:27
The World Scientific Handbook of Futures Markets Downloaded from www.worldscientific.com

World Scientific Handbook of Futures Markets. . .

by LA TROBE UNIVERSITY on 05/21/16. For personal use only.

Animal
products
Broilers† CBT 8/68–6/81








Eggs† CME 5/66–12/80











Cattle CME 1/72–7/84








Hogs CME 3/66–7/84







Pork bellies CME 5/66–7/84





Metals‡
Copper Comex 3/66–7/84






Gold Comex 2/75–7/84






Platinum NYM 1/68–7/84




Silver Comex 1/67–7/84






∗
CBT = Chicago Board of Trade; CME = Chicago Mercantile Exchange; Comex = Commodity Exchange; CSCE =

Commodity Futures Prices

Coffee, Sugar, and Cocoa Exchange; CTN = New York Cotton Exchange; NYM = New York Mercantile Exchange.
†
The standard delivery months for broilers and eggs change toward the end of the sample period. The delivery months
prevailing over most of the period are indicated here.
‡
Supplemental contracts are traded on copper, gold, and silver. These contracts fill in the months between the standard

9.75in x 6.5in
contracts.

b1892-ch04
83

page 83
 July 3, 2015 8:27 World Scientific Handbook of Futures Markets. . . 9.75in x 6.5in b1892-ch04 page 84

84 E. F. Fama and K. R. French

Standard deviations of the basis

The second column of Table 4.2 shows standard deviations of the 6-month
basis for the 21 commodities. The 6-month maturity is chosen since it is
available for all commodities but cotton. The 3-month basis is used for
cotton. Basis standard deviations differ systematically across commodity
groups. The precious metals have the lowest standard deviations — 2.0%
for gold, 1.5% for silver, and 4.2% for platinum. The standard deviations
The World Scientific Handbook of Futures Markets Downloaded from www.worldscientific.com

for the agricultural products range from 4.6% for corn to 9.7% for oats. The
animal products have the largest basis standard deviations. The standard
deviation for cattle is 5.6%, and the standard deviations for the other four
animal products range from 10.1% for broilers to 22.2% for eggs.
by LA TROBE UNIVERSITY on 05/21/16. For personal use only.

The differences in the basis standard deviations for commodity subgroups

are consistent with the theory of storage. One source of variation in the basis
is seasonals in supply and demand. For example, spot prices for agricultural
commodities usually increase between harvests and fall across harvests.
Because of this pattern in the spot price, the basis is usually positive when
the futures contract matures in the current crop year and negative when
the futures contract matures early in the next crop year. Storage costs are
important in determining the magnitude of the seasonal variation in spot
prices. Higher storage costs imply larger expected spot-price changes to
induce storage between harvests. Thus, seasonal variation in the basis should
be an increasing function of storage costs.
Demand and supply shocks also generate variation in the basis. The effect
of shocks on the basis depends to a large extent on the way inventories adjust
to transmit the price effects of shocks through time. For example, suppose
there is a spell of propitious weather before a harvest that raises expected
future supplies and lowers expected future prices. The expected decline in
the spot price is partly offset by the inventory response it generates. The gap
between current and expected future prices is narrowed as more inventory
is sold immediately. Higher inventory levels allow larger inventory responses
to demand and supply shocks and thus lower variation in expected price
changes. Since storage costs deter storage, the effect of demand and supply
shocks on the variability of the basis should be an increasing function of
storage costs. See French (1986).
The analysis predicts high basis standard deviations for seasonal, high-
storage-cost commodities. Metal storage costs (Table 4.2) are low relative
to value, and the metals are not subject to seasonals in supply or demand.
Thus the low basis standard deviations for the metals are consistent with
the theory of storage. It is also consistent with the theory of storage that
 July 3, 2015
8:27
The World Scientific Handbook of Futures Markets Downloaded from www.worldscientific.com

Table 4.2. Regressions of the 6-month basis on the 6-month interest rate and monthly seasonal dummies:
P12

World Scientific Handbook of Futures Markets. . .

F (t,T )−S(t)
= m=1 αm dm + βR(t, T ) + e(t, T ).
by LA TROBE UNIVERSITY on 05/21/16. For personal use only.

S(t)

Storage Handling
Commodity Obs. SD β s(β) F df R12 R22 (%) (%)

Agricultural products
Cocoa 35 8.1 1.16 1.44 0.00 1 0.00 0.03 0.16 0.35
Coffee 30 9.6 0.29 1.57 1.72 4 0.06 0.03 0.12 0.26
Corn 35 4.6 0.86 0.52 0.01 1 0.05 0.07 1.41 1.73
Cotton 36 4.9 0.84 1.46 1.14 2 −0.02 −0.02 0.32 0.13
Oats 34 9.7 1.06 1.27 6.55 1 0.16 0.01 2.65 3.26
Orange juice 102 9.2 1.39 1.21 3.32 5 0.14 0.04 0.30 0.32
Soybeans 105 7.8 1.88 0.71 5.72 5 0.30 0.14 0.64 0.78
Soy meal 70 7.2 2.03 0.84 0.20 5 0.16 0.21 — —
Soy oil 74 8.9 1.73 1.28 0.79 5 0.06 0.07 0.27 0.30

Commodity Futures Prices

Wheat 35 6.8 1.05 0.86 9.03 1 0.24 0.05 1.39 1.71
Wood products
Lumber 86 13.6 2.41 2.21 1.86 5 0.12 0.07 1.96 3.82
Plywood 82 7.4 1.23 1.17 0.71 5 0.04 0.06 — —

9.75in x 6.5in
(Continued)

b1892-ch04
85

page 85
 July 3, 2015
86

8:27
E. F. Fama and K. R. French
The World Scientific Handbook of Futures Markets Downloaded from www.worldscientific.com

Table 4.2. (Continued )

Storage Handling

World Scientific Handbook of Futures Markets. . .

Commodity Obs. SD β s(β) F df R12 R22 (%) (%)
by LA TROBE UNIVERSITY on 05/21/16. For personal use only.

Animal products
Broilers 64 10.1 1.39 1.65 5.43 11 0.44 0.00 — —
Cattle 70 5.6 −0.06 0.57 4.48 5 0.19 −0.01 — —
Eggs 80 22.2 −4.32 3.34 4.96 11 0.38 0.04 — —
Hogs 102 10.9 2.21 1.36 1.79 9 0.14 0.08 — —
Pork bellies 34 14.3 2.71 1.66 5.86 1 0.19 0.07 0.98 2.54
Metals
Copper 89 6.5 1.39 0.85 1.05 5 0.14 0.13 0.12 0.49
Gold 57 2.0 1.07 0.13 0.29 6 0.81 0.83 0.01 0.03
Platinum 66 4.2 1.18 0.63 0.28 3 0.15 0.18 0.01 0.01
Silver 101 1.5 0.77 0.16 0.31 5 0.58 0.60 0.03 0.06

Note: Obs. is the number of observations. SD is the standard deviation of the 6-month basis. df is the numerator
degrees of freedom for the F-statistic test of the hypothesis that all the seasonal dummies in a regression are equal.
R22 is the coefficient of determination in the simple regression of the basis on the interest rate, and R12 is for the
regression that includes the seasonal dummies. Storage is the monthly warehousing cost per dollar of the June 1984
spot price. Handling is the total cost of loading and unloading the commodity at the warehouse per dollar of the
June 1984 spot price. Storage and handling charges are from futures exchanges, dealers, elevators, and warehouses.

9.75in x 6.5in
These charges are reported only for commodities that have standard storage arrangements. The absence of such
arrangements implies high storage costs. The 6-month maturity is not available for cotton. The 3-month maturity
is used.

b1892-ch04
page 86
 July 3, 2015 8:27 World Scientific Handbook of Futures Markets. . . 9.75in x 6.5in b1892-ch04 page 87

Commodity Futures Prices 87

the highest basis standard deviations are observed for some of the wood and
animal products (lumber, broilers, eggs, hogs, and pork bellies), where bulk
and perishability make storage expensive.

Regression tests
To obtain more direct tests of the theory of storage, we regress the basis
against the nominal interest rate and monthly seasonal dummies:
The World Scientific Handbook of Futures Markets Downloaded from www.worldscientific.com

12
F (t, T ) − S(t)
= αm dm + βR(t, T ) + e(t, T ) (4.3)
S(t) m=1
where dm equals 1.0 if the futures contract matures in month m and 0.0
by LA TROBE UNIVERSITY on 05/21/16. For personal use only.

otherwise. The hypothesis of the storage equation (4.2) is that the slope
β should be 1.0 for any commodity continuously stored; that is, the basis
should vary one for one with the nominal interest rate. The seasonal dummies
in (4.3) are a crude way to capture variation in the marginal convenience
yield in (4.2), which is due to seasonals in production or demand.
1. Interest-rate relations. Estimates of the slopes in regression (4.3) are in
Table 4.2. The metals produce the strongest evidence of variation in the basis
that tracks interest rates, and gold produces the strongest evidence among
the metals. The interest-rate coefficient in the 6-month gold regression is
1.07. The estimates for one, three, and 12 months to maturity (not reported)
range from 0.99 to 1.06. Table 4.2 shows coefficients of determination (R2 )
for simple regressions of the basis on the nominal interest rate as well as for
the regressions that include seasonal dummies. Nominal interest rates alone
explain 83% of the 6-month basis variance for gold. The nominal interest
rate explains 60% of the variance of the 6-month basis for silver. Explanatory
power is lower for platinum and lower again for copper, but estimated interest
rate coefficients are close to 1.0. The metals regressions for one, three, and
12 months to maturity (not shown) are similar. Metals prices are consistent
with the hypothesis that the basis tracks nominal interest rates.
The regressions for the agricultural and wood products are also consistent
with the hypothesis that the basis varies one for one with the nominal interest
rate. All the interest-rate coefficients are positive, many are close to 1.0, and
only two are more than 1.0 standard error from 1.0. However, the standard
errors of the interest-rate coefficients for the agricultural and wood products
are all greater than 0.5. This lack of precision means that the regression
slopes cannot provide convincing evidence of one-for-one variation in the
basis in response to nominal interest rates. The interest-rate coefficients for
the animal-product regressions are even less precise. The standard errors of
 July 3, 2015 8:27 World Scientific Handbook of Futures Markets. . . 9.75in x 6.5in b1892-ch04 page 88

88 E. F. Fama and K. R. French

the coefficients are typically greater than 1.0, and the estimates are consistent
with a wide range of values for the true slopes, including 0.0 as well as 1.0.
Restated in terms of (4.2), the regressions indicate that variation in the
interest rate is a large fraction of basis variation for gold and silver. For
other commodities, there is suggestive evidence of basis variation in response
to the nominal interest rate, but variation in the [W (t, T ) − C(t, T )]/S(t)
component of the basis leads to imprecise estimates of the relation between
the basis and the interest rate. For agricultural, wood, and animal products,
The World Scientific Handbook of Futures Markets Downloaded from www.worldscientific.com

basis variation must be explained primarily in terms of economic conditions

that generate variation in marginal storage costs, W (t, T ), and marginal
convenience yields, C(t, T ), rather than in terms of the role of the interest
by LA TROBE UNIVERSITY on 05/21/16. For personal use only.

rate in the storage process.

2. Seasonals in the basis. The seasonal dummies in (4.3) are evidence about
seasonal variation in the basis. The F -statistics testing the hypothesis that
all seasonal coefficients in a regression are equal never indicate reliable
seasonals in the basis for any metal. This is not surprising since there is
no presumption of seasonals in the demand or supply of metals.
As expected, there are reliable seasonals in the basis for many of the
seasonally produced agricultural commodities, including corn, oats, orange
juice, soybeans, and wheat. (Although the 6-month basis for corn in
Table 4.2 does not show seasonals, there are reliable seasonals in the 3-month
basis.) On the other hand, it is a bit unexpected that five agricultural
commodities — cocoa, coffee, cotton, soy meal, and soy oil — produce no
reliable evidence of seasonals in the 6-month basis. The absence of seasonals
for soy meal and soy oil is interesting given the strong seasonals in the basis
for soybeans. Apparently, the production process for meal and oil reduces the
effect of seasonals in the price of soybeans.
The animal products produce the strongest evidence of seasonals in
the basis. The coefficients of determination in the seasonal regressions for
broilers, cattle, eggs, and pork bellies are at least 0.19. The seasonals in
the 6-month basis for hogs are weaker, but the coefficients of determination
in the 1- and 3-month seasonal regressions for hogs (not shown) are 0.47
and 0.72, respectively. Since the nominal interest rate explains only a small
fraction of the basis variation for the animal products, much of their basis
variation can be attributed to seasonals.
Since the animal products are subject to seasonals in production and
sometimes in demand (see Bessant 1982), and since bulk and perishability
imply storage costs that are high relative to value, strong seasonals in the
basis confirm the predictions of the theory of storage. On the other hand,
 July 3, 2015 8:27 World Scientific Handbook of Futures Markets. . . 9.75in x 6.5in b1892-ch04 page 89

Commodity Futures Prices 89

lumber and plywood also have high storage costs and seasonals in demand
due to seasonals in building activity, but the regressions for lumber and
plywood produce no reliable evidence of seasonals. One possibility is that
the production of wood products is more easily adapted to seasonals in
demand than the production of animal products. The details of supply and
demand conditions for different commodities and their implications for the
behavior of the basis is interesting material for future research.
The World Scientific Handbook of Futures Markets Downloaded from www.worldscientific.com

The Basis: Forecast Power and Premiums

The theory-of-storage view of futures prices in Equation (4.2) is not

by LA TROBE UNIVERSITY on 05/21/16. For personal use only.

controversial. There is another view that is the subject of long and continuing
controversy. The difference between the futures price and the current spot
price can be expressed as the sum of an expected premium and an expected
change in the spot price:
F (t, T ) − S(t) = Et [P (t, T )] + Et [S(T ) − S(t)], (4.4)
where the expected premium is defined as the bias of the futures price as a
forecast of the future spot price,
Et [P (t, T )] = F (t, T ) − Et [S(T )]. (4.5)
Equation (4.4) and the theory of storage in (4.2) are alternative but
not competing views of the basis. Variation in the expected premium or
the expected change in the spot price in (4.4) translates into variation in
the interest rate, the marginal storage cost, or the marginal convenience
yield in (4.2). For example, the basis for agricultural commodities is often
negative before a harvest when the futures price is for delivery after the
harvest. Under the theory of storage, the basis is negative because inventories
are low and the convenience yield is larger than interest and storage costs.
In terms of (4.4), the explanation for negative values of the basis is that
the spot price is expected to fall when a harvest will substantially increase
inventories. Likewise, positive values of the basis when both the futures and
the spot prices are for the period between harvests can be explained in terms
of storage costs that outweigh marginal convenience yields when inventories
are high, but they are equally well explained in terms of an expected increase
in the spot price necessary to induce storage between harvests.
Despite research that extends from Keynes (1930), Hardy (1940), Work-
ing (1948, 1949), Telser (1958, 1967), and Cootner (1960, 1967) to Dusak
(1973), Bodie and Rosansky (1980), Carter, et al. (1983), and Hazuka (1984),
there is little agreement on whether the expected premium in (4.4) is
 July 3, 2015 8:27 World Scientific Handbook of Futures Markets. . . 9.75in x 6.5in b1892-ch04 page 90

90 E. F. Fama and K. R. French

non-zero or on whether futures prices have power to forecast future spot

prices. We test for time-varying expected premiums and price forecasts in
futures prices with the regression approach in Fama (1984a, 1984b). Consider
the regressions of the change in the spot price and the premium on the
basis:
S(T ) − S(t) = a1 + b1 [F (t, T ) − S(t)] + u(t, T ), (4.6)
F (t, T ) − S(T ) = a2 + b2 [F (t, T ) − S(t)] + z(t, T ). (4.7)
The World Scientific Handbook of Futures Markets Downloaded from www.worldscientific.com

Evidence that b1 is positive means the basis observed at t contains

information about the change in the spot price from t to T . Equivalently,
the futures price has power to forecast the future spot price. Evidence that
by LA TROBE UNIVERSITY on 05/21/16. For personal use only.

b2 is positive means the basis observed at t contains information about the

premium to be realized at T . Predictable variation in realized premiums is
evidence of time-varying expected premiums.

What can we expect?

Regressions (4.6) and (4.7) are subject to an adding-up constraint. The sum
of the premium, F (t, T )−S(T ), and the change in the spot price, S(T )−S(t),
is the basis, F (t, T ) − S(t). Thus, the intercepts in (4.6) and (4.7) must sum
to 0.0; each period’s residuals must sum to 0.0; and, most important, the
slope coefficients must sum to 1.0. In other words, the regressions always
allocate all variation in the basis to the expected premium, the expected
change in the spot price, or some mix of the two.
The point is worth emphasizing. As law-abiding financial economists, we
presume that market forecasts of future spot prices are rational. Moreover,
all the spot and futures prices have been checked twice. Nevertheless, even
basis variance due to measurement errors and irrational forecasts of spot
prices are allocated by the regressions (4.6) and (4.7). It is easy to show that
an irrational forecast of the spot price in the futures price shows up as a
time-varying expected premium (a positive value of b2 ), while measurement
error in the spot price shows up as forecast power (a positive value of b1 ).
Although the regressions allocate all basis variation to expected pre-
miums, expected spot-price changes, or some combination of the two, the
allocation can be statistically unreliable. Since estimates of b1 and b2 in
(4.6) and (4.7) are typically between 0.0 and 1.0, the regressions can fail to
identify the source of variation in the basis — the regressions produce slope
coefficients that sum to 1.0 but are not reliably different from 0.0 — when
basis variation is low relative to variation in realized premiums and changes
in spot prices. We can get a good idea about where to place our bets in
 July 3, 2015 8:27 World Scientific Handbook of Futures Markets. . . 9.75in x 6.5in b1892-ch04 page 91

Commodity Futures Prices 91

Table 4.3. Standard deviations of the 2-month basis, the change in the spot
price, and the premium.

Basis Change Premium

Commodity F (t, t + 2) − S(t) S(t + 2) − S(t) F (t, t + 2) − S(t + 2)

Agricultural products
Cocoa 4.0 14.6 15.2
Coffee 5.2 15.0 14.4
Corn 2.8 9.9 10.6
The World Scientific Handbook of Futures Markets Downloaded from www.worldscientific.com

Cotton 2.4 9.0 9.0

Oats 4.4 11.7 10.5
Orange juice 4.9 13.1 13.4
Soybeans 2.8 12.2 12.0
Soy meal 4.1 13.4 13.4
by LA TROBE UNIVERSITY on 05/21/16. For personal use only.

Soy oil 4.7 13.5 13.7

Wheat 3.3 14.5 14.7
Wood products
Lumber 6.9 11.5 12.1
Plywood 3.2 9.8 10.5
Animal products
Broilers 5.8 11.1 8.7
Cattle 3.3 11.0 10.4
Eggs 13.2 16.3 12.6
Hogs 7.1 12.9 12.1
Pork bellies 1.9 16.9 16.4
Metals
Copper 2.5 12.1 12.4
Gold 0.6 13.1 13.2
Platinum∗ 2.2 16.2 16.1
Silver 0.5 18.5 18.7
∗
2-month maturity is not available; 3-month maturity is used.

the regressions by examining variances of the basis relative to variances of

premiums and changes in spot prices.
Table 4.3 shows standard deviations of the basis, F (t, t + 2) −
S(t), the change, S(t + 2) − S(t), and the premium, F (t, t + 2) − S(t + 2), for
each commodity. The 2-month maturity is chosen because it is available for
all commodities but platinum. In Table 4.3 (and in the regressions below)
all prices are measured in natural logs.
The standard deviations of spot-price changes and premiums are large,
and they do not differ much across commodity groups. For example, the
standard deviations of spot-price changes range from 12.1 to 18.5% for
the metals, from 9.0 to 15.0% for the agricultural products, and from 11.0
to 16.9% for the animal products. It seems that futures markets exist for
 July 3, 2015 8:27 World Scientific Handbook of Futures Markets. . . 9.75in x 6.5in b1892-ch04 page 92

92 E. F. Fama and K. R. French

commodities subject to similar high levels of uncertainty about future spot

prices.
In contrast, basis variability differs systematically across commodity
groups. As in Table 4.2, the basis standard deviations in Table 4.3 are
smallest for the metals, larger for the agricultural and wood products, and
largest for some of the animal products. The standard deviation of the 2-
month basis for gold is 0.6% versus 7.1% for hogs. For commodities such
as the metals, where basis variation is low relative to the variation of
The World Scientific Handbook of Futures Markets Downloaded from www.worldscientific.com

premiums and spot-price changes, it is unlikely that regressions (4.6) and

(4.7) can reliably assign basis variation to expected premiums or expected
spot-price changes. The regressions have a better chance with commodities
by LA TROBE UNIVERSITY on 05/21/16. For personal use only.

like the animal products, where basis variation is substantial.

Regression results

Estimates of the change regression (4.6) and the premium regression (4.7)
are given in Table 4.4. Because they are the focal point of the evidence on
forecast power and time-varying expected premiums, the slopes b1 and b2
for both (4.6) and (4.7) are reported, even though they must sum to 1.0.
To limit the size of the table, only results for two, six, and 10 months to
maturity are shown. These maturities tend to have the largest samples, and
they are spaced fairly evenly among the possible maturities from one to 12
months.
We first categorize the regressions for different commodities according
to whether futures prices show time-varying expected premiums, power to
forecast spot prices, both, or neither. Then we relate the results to differences
in conditions of production and storage.
Type SF — strong forecast power. Futures prices for broilers, eggs, hogs,
and oats have reliable forecast power at every maturity (including those not
shown in Table 4.4), and they show no reliable evidence of time-varying
expected premiums. The slopes in the change regressions for these com-
modities are all more than 2.6 standard errors from 0.0, and most are more
than 4.0 standard errors from 0.0. Moreover, the forecast power of futures
prices is non-trivial. For example, the coefficients of determination (R12 ) in
the change regressions for broilers are 0.40 and 0.42; for oats they range from
0.20 to 0.35.
Type GF — good forecast power but not for all maturities. The change
regressions for cattle, pork bellies, soybeans, and soy meal indicate reliable
forecast power in futures prices for at least one maturity and suggestive
 July 3, 2015
8:27
Table 4.4. Regressions of the spot price change and the premium on the basis: S(T )−S(t) = a1 +b1 [F (t, T )−S(t)]+u(t, T ),
F (t, t) − S(T ) = a2 + b2 [F (t, T ) − S(t)] − u(t, T ).
The World Scientific Handbook of Futures Markets Downloaded from www.worldscientific.com

Two Months Six Months Ten Months

Commodity Max. Obs. b1 b2 t(b1 ) t(b2 ) R2

1 R2
2 Obs. b1 b2 t(b1 ) t(b2 ) R2
1 R2
2 Obs. b1 b2 t(b1 ) t(b2 ) R2
1 R2
2

Agricultural

World Scientific Handbook of Futures Markets. . .

products
by LA TROBE UNIVERSITY on 05/21/16. For personal use only.

Cocoa 221 56 −0.03 1.03 −0.07 2.09 0.00 0.07 36 −0.08 1.08 −0.16 2.26 0.00 0.13 54 0.24 0.76 0.42 1.32 0.01 0.08
Coffee 137 38 0.86 0.14 1.91 0.30 0.09 0.00 — — — — — — — 37 0.46 0.54 0.65 0.75 0.03 0.04
Corn 221 56 −0.40 1.40 −0.84 2.93 0.01 0.13 36 −0.59 1.59 −0.81 2.20 0.02 0.12 54 0.48 0.52 0.91 0.98 0.03 0.03
Cotton 207 53 0.55 0.45 1.06 0.87 0.02 0.01 — — — — — — — 50 1.10 −0.10 2.27 −0.21 0.16 0.00
Oats 217 54 1.18 −0.18 3.63 −0.55 0.20 0.01 34 1.05 −0.05 4.19 −0.19 0.35 0.00 42 1.02 −0.02 3.28 −0.07 0.29 0.00
Orange juice 207 101 0.28 0.72 1.07 2.69 0.01 0.07 100 0.57 0.43 1.79 1.36 0.06 0.04 97 1.00 −0.00 2.85 −0.00 0.20 0.00
Soybeans 221 110 0.80 0.20 1.95 0.49 0.03 0.00 108 0.71 0.29 2.36 0.95 0.09 0.02 106 0.63 0.37 1.71 1.01 0.07 0.03
Soy meal 217 108 0.44 0.56 1.26 1.59 0.02 0.03 68 0.50 0.50 1.14 1.15 0.03 0.03 93 0.65 0.35 2.85 1.56 0.14 0.05
Soy oil 217 112 0.40 0.60 1.41 2.11 0.02 0.04 75 −0.02 1.02 −0.07 2.75 0.00 0.16 105 0.01 0.99 0.03 2.11 0.00 0.14
Wheat 219 55 0.18 0.82 0.29 1.36 0.00 0.03 35 −0.65 1.65 −1.33 3.39 0.05 0.25 52 −0.78 1.78 −1.62 3.69 0.07 0.27
Wood
products
Plywood 163 81 −0.00 1.00 −0.02 3.42 0.00 0.13 79 0.53 0.47 1.46 1.29 0.06 0.05 69 1.27 −0.27 3.81 −0.80 0.34 0.02
Lumber 173 86 0.35 0.65 1.97 3.71 0.04 0.14 84 0.28 0.72 1.35 3.55 0.05 0.27 52 0.16 0.84 0.42 2.21 0.01 0.24
Animal
products
Broilers 152 108 1.22 −0.22 7.68 −1.40 0.40 0.02 64 0.93 0.07 5.39 0.40 0.42 0.00 — — — — — — —
Cattle 147 51 1.12 −0.12 2.51 −0.27 0.11 0.00 67 0.84 0.16 1.54 0.30 0.07 0.00 — — — — — — —
Eggs 173 145 0.80 0.20 8.58 2.15 0.42 0.04 80 0.97 0.03 6.24 0.18 0.53 0.00 — — — — — — —

Commodity Futures Prices

Hogs 217 117 0.72 0.28 4.59 1.81 0.16 0.03 103 0.66 0.34 2.67 1.38 0.12 0.04 78 0.80 0.20 2.76 0.70 0.22 0.02
Pork bellies 219 37 2.77 −1.77 1.95 −1.25 0.10 0.04 33 1.39 −0.39 5.12 −1.43 0.45 0.06 36 1.12 −0.12 5.08 −0.53 0.53 0.01
Metals
Copper 223 157 −0.03 1.03 −0.08 2.57 0.00 0.05 92 0.64 0.36 1.54 0.88 0.05 0.02 98 0.66 0.34 1.36 0.70 0.06 0.02
Gold 115 107 −2.20 3.20 −0.80 1.17 0.01 0.02 55 −1.74 2.87 −0.68 1.04 0.02 0.05 52 −2.83 3.83 −0.89 1.20 0.07 0.12
Platinum 199 — — — — — — — 65 0.73 0.27 0.82 0.30 0.02 0.00 — — — — — — —

9.75in x 6.5in
Silver 211 174 −8.56 9.56 −2.45 2.73 0.06 0.07 100 −7.82 8.82 −2.50 2.82 0.13 0.16 99 −6.12 7.12 −2.03 2.36 0.14 0.17

Note: R12 and R22 are the coefficients of determination for the change and premium regressions, respectively. Since the change
and premium regressions have the same explanatory variable and the two residuals sum to .0, their slope coefficients have
the same standard error. The t-statistics, t(b1 ) and t(b2 ), are based on standard errors adjusted for the autocorrelation of
the regression residuals induced by the overlap of the observations on S(T ) − S(t) and F (t, T ) − S(T ). (See Hansen and

b1892-ch04
Hodrick, 1980.) Obs. is the number of observations in a regression, and Max. is the number of months in the sample period.

93

page 93
 July 3, 2015 8:27 World Scientific Handbook of Futures Markets. . . 9.75in x 6.5in b1892-ch04 page 94

94 E. F. Fama and K. R. French

evidence of forecast power for other maturities. They show no reliable

evidence of time-varying expected premiums.
Types SP and GP — expected premiums. The premium regressions for two
commodities, soy oil and lumber, produce reliable evidence of time-varying
expected premiums at every maturity, while three commodities — cocoa,
corn and wheat — seem to have time-varying expected premiums at some
but not all maturities. The evidence of time-varying expected premiums for
The World Scientific Handbook of Futures Markets Downloaded from www.worldscientific.com

these commodities is weaker than the evidence of forecast power for the
SF and GF commodities. For example, t-statistics above 5.0 are common
in the change regressions for the SF commodities, but the largest t-statistic
in the premium regressions for the SP commodities is 3.71. Similarly, the
by LA TROBE UNIVERSITY on 05/21/16. For personal use only.

coefficients of determination are often above 0.40 in the change regressions for
the SF commodities, but they never exceed 0.27 in the premium regressions
for the SP commodities.
Type F&P — forecast power and expected premiums. There are two com-
modities, orange juice and plywood, for which futures prices seem to show
both forecast power and time-varying expected premiums. However, the
evidence for forecast power occurs at long maturities, whereas the evidence
for expected premiums is observed for shorter maturities.
Type W — weak. The regressions for coffee, copper, and cotton produce
suggestive evidence that futures prices contain both time-varying expected
premiums and power to forecast future spot prices; that is, for many
maturities (including those not shown), the slope coefficients b1 and b2 for the
change and premium regressions are both well above 0.0 and below 1.0. Some
of the regression slopes for these commodities are more than two standard
errors from zero, but most are not reliably different from 0.0. In short,
the regressions fail to identify any commodities for which the basis shows
reliable simultaneous variation in expected premiums and forecasts of spot
prices.
For gold and platinum, basis variability is so low relative to the variability
of realized premiums and changes in spot prices that regression coefficients
equal to 1.0 would usually be less than one standard error from zero. For
these commodities, the regressions cannot reliably identify situations in
which all basis variation reflects either time-varying expected premiums or
forecasts of spot-price changes. This is in contrast to the theory-of-storage
regressions in Table 4.2, where the basis is the dependent variable and the
low basis variances of the precious metals allow the most reliable inferences
that the basis tracks nominal interest rates.
 July 3, 2015 8:27 World Scientific Handbook of Futures Markets. . . 9.75in x 6.5in b1892-ch04 page 95

Commodity Futures Prices 95

The silver regressions in Table 4.4 are puzzling. The regression slopes are
more than 2.0 standard errors from zero, but the coefficients seem bizarre.
For example, the estimated slope in the 2-month change regression is −8.56;
taken literally, a 1.0% increase in the basis implies an 8.6% drop in the
expected price change. We are examining these and other metal results in
more detail.
The World Scientific Handbook of Futures Markets Downloaded from www.worldscientific.com

Interpretation of the regressions

Table 4.5 summarizes the regressions, both those in Table 4.4 and the theory-
of-storage regressions in Table 4.2. Commodities are allocated to the columns
by LA TROBE UNIVERSITY on 05/21/16. For personal use only.

of Table 4.5 depending on whether their futures prices show forecast power
or time-varying expected premiums in (4.6) and (4.7). The numbers after the
commodity names are the reverse order of their basis standard deviations for
the 2-month maturity in Table 4.3. An asterisk means that the commodity’s
basis shows seasonals in Table 4.2.
1. Basis variability, forecast power, and premiums. As expected, there is
a relation between basis variability and evidence that futures prices have
time-varying expected premiums or power to forecast future spot prices. Of
the four commodities that show strong forecast power (SF) in the estimates
of (4.6), three — broilers, eggs, and hogs — rank in the top four in basis
variability. Two commodities, lumber and soy oil, show evidence of time-
varying expected premiums (SP) for all maturities; lumber ranks third in
basis variability, and soy oil is seventh. At the other end of the spectrum,
copper, cotton, gold, and platinum have relatively low basis variation and
unreliable results in the tests for forecast power and premiums.
Since the slopes in the regressions of S(T ) − S(t) and F (t, T ) − S(T ) on
F (t, T ) − S(t) sum to 1.0 and are typically between 0.0 and 1.0, it is almost
a matter of arithmetic that regression coefficients statistically far from 0.0
in (4.6) and (4.7) occur when basis variation is high relative to the variation
of the changes and premiums to be explained. The interesting question is
why basis variation is high for some commodities and low for others.
2. Storage costs and forecast power. As discussed earlier, the theory of storage
predicts that storage costs are important in explaining differences in the
variability of the expected spot-price change in the basis. For agricultural and
animal products, which are subject to seasonals in production or demand,
the amount of predictable seasonal variation in the spot price should be an
increasing function of storage costs. Likewise, stored stocks act to smooth
predictable adjustments in the spot price in response to demand and supply
 July 3, 2015 8:27 World Scientific Handbook of Futures Markets. . . 9.75in x 6.5in b1892-ch04 page 96

96 E. F. Fama and K. R. French

Table 4.5. Regression scoreboard.

Expected
Forecast power premiums Both Neither (weak)

Futures prices show

Broilers* (SF) 4 Lumber (SP) 3 Orange juice* Coffee (W) 5
(F&P) 6
Eggs* (SF) 1 Soy oil (SP) 7 Plywood Copper (W) 16
(F&P) 13
The World Scientific Handbook of Futures Markets Downloaded from www.worldscientific.com

Hogs* (SF) 2 Cotton (W) 17

Oats* (SF) 8 Cocoa (GP) 10 Gold (W) 21
Corn* (GP) 15 Platinum (W) 18
Cattle* (GF) 11 Wheat* (GP) 12 Silver (W) 20
by LA TROBE UNIVERSITY on 05/21/16. For personal use only.

Pork bellies* (GF) 19

Soybeans* (GF) 14
Soy meal (GF) 9

Note: An asterisk means the commodity’s basis shows reliable evidence of

seasonals in the estimates of (4.3). The numbers after the commodities are
the (reverse) order of their basis standard deviations for the 2-month maturity
in Table 4.3. For example, the 1 after eggs indicates that its basis has the
highest standard deviation. The letters in parentheses after the commodity names
categorize their regression results in the estimates of (4.5) and (4.6) in Table 4.4.
The categories are as follows. SF = strong forecast power: statistically reliable
power to forecast changes in spot prices across all maturities; no evidence of
time-varying expected premiums. GF = good forecast power: statistically reliable
power to forecast changes in spot prices for most but not all maturities; no
evidence of time-varying expected premiums. SP = strong expected premiums:
statistically reliable evidence of time-varying expected premiums across all
maturities. GP = expected premiums: statistically reliable evidence of time-
varying expected premiums for most but not all maturities. F&P = forecast
power and expected premiums: statistically reliable forecast power for some
maturities and statistically reliable time-varying expected premiums for others.
W = weak: regression evidence is unreliable or extreme.

shocks. Since storage costs that are high relative to value deter storage, the
theory predicts that variation in expected spot-price changes in response to
shocks is also an increasing function of storage costs.
These predictions can explain broad features of the estimates of the
change regression (4.6). The regressions for eight commodities indicate that
the basis F (t, T ) − S(t) has reliable information about the future change
in the spot price S(T ) − S(t) for most maturities (T − t). Five of these
commodities are animal products (broilers, cattle, eggs, hogs, and pork
bellies), whose bulk and perishability imply high storage costs relative
to value. Storage costs (Table 4.2) are also high for the remaining three
commodities (oats, soybeans, and soy meal), whose futures prices show
 July 3, 2015 8:27 World Scientific Handbook of Futures Markets. . . 9.75in x 6.5in b1892-ch04 page 97

Commodity Futures Prices 97

consistent forecast power. Forecast power is not found in futures prices for
gold and platinum, whose storage costs are low relative to value and basis
variances are low relative to variances of spot-price changes.1
3. Seasonals and forecast power. Our analysis of storage costs and forecast
power predicts that seasonal variation in the basis generates forecast power
in the change regression (4.6). Table 4.5 suggests that this prediction is
generally correct. The basis has reliable power to forecast changes in the
spot price for eight of the ten commodities that have reliable basis seasonals.
The World Scientific Handbook of Futures Markets Downloaded from www.worldscientific.com

Of the ten commodities with reliable evidence of forecast power, eight have
reliable basis seasonals.
by LA TROBE UNIVERSITY on 05/21/16. For personal use only.

Univariate tests for expected premiums

The estimates of (4.7) find evidence of expected premiums for only 7 of 21

commodities. The regressions, however, are designed to detect variation in
expected premiums. Failure to identify time-varying expected premiums does
not imply that expected premiums are zero. To examine the issue further,
we have computed average values of the premium, F (t, T ) − S(T ), for each
maturity of each commodity. This approach has no power (i) because the
variances of realized premiums are so large (see Table 4.3) and (ii) because
futures contracts for a given maturity (T − t) are available for only a fraction
of the sample months (compare the maximum and actual observations in
Table 4.4).
Following Bodie and Rosansky (1980), we increase the power of univariate
tests for expected premiums by (i) combining contracts for a commodity to
ensure that an observation for the commodity is available every month and
(ii) combining commodities into portfolios. Monthly returns are computed
for each commodity using the shortest futures contract with at least 1 month
to maturity on the first trading day of each month. A simple return is defined
as the change in the futures price over the month, divided by the price of
the contract at the beginning of the month. Contracts chosen generally have
maturities of 1, 2, or 3 months.
The average simple returns for individual commodities (Table 4.6) suggest
that futures prices show normal backwardation — the expected return from

1
Hazuka (1984) estimates (4.6) for 1-month spot-price changes and for a shorter list
of commodities. His conclusions about the relation between storage costs and forecast
power are similar to ours. He does not recognize that the change regression (4.6) has a
complement, the premium regression (4.7).
 July 3, 2015 8:27 World Scientific Handbook of Futures Markets. . . 9.75in x 6.5in b1892-ch04 page 98

98 E. F. Fama and K. R. French

Table 4.6. Average monthly simple and continuously compounded returns for
portfolios and individual commodities.

Continuously
Simple compounded

Obs. M SD t(M ) M SD t(M )

All commodities 222 0.54 4.3 1.87 0.45 4.2 1.57

Agricultural products 222 0.83 5.4 2.29 0.69 5.2 1.97
The World Scientific Handbook of Futures Markets Downloaded from www.worldscientific.com

Cocoa 220 1.59 10.1 2.33 1.10 9.8 1.67

Coffee 140 1.84 10.2 2.13 1.35 9.6 1.67
Corn 220 0.12 6.6 0.27 −0.09 6.5 −0.21
Cotton 208 0.32 7.1 0.66 0.09 6.8 0.18
Oats 218 0.10 7.6 0.19 −0.19 7.6 −0.36
by LA TROBE UNIVERSITY on 05/21/16. For personal use only.

Orange juice 209 1.37 11.2 1.76 0.83 10.0 1.20

Soybeans 220 0.66 9.7 1.01 0.21 9.4 0.34
Soy meal 213 0.86 10.9 1.15 0.31 10.3 0.44
Soy oil 218 1.91 11.4 2.47 1.31 10.6 1.83
Wheat 218 0.06 9.1 0.10 −0.30 8.4 −0.54
Wood products 177 −0.23 7.5 −0.42 −0.51 7.5 −0.91
Lumber 177 −0.53 7.9 −0.89 −0.86 8.1 −1.40
Plywood 165 0.44 8.0 0.71 0.13 7.8 0.22
Animal products 220 0.00 6.3 0.00 −0.20 6.3 −0.46
Broilers 151 0.55 7.4 0.92 0.31 6.9 0.56
Cattle 151 0.24 7.0 0.42 −0.01 7.1 −0.01
Eggs 175 −2.01 10.4 −2.56 −2.60 10.7 −3.20
Hogs 220 1.11 8.0 2.06 0.79 8.0 1.47
Pork bellies 218 0.26 10.4 0.37 −0.28 10.5 −0.39
Metals 222 0.57 8.3 1.02 0.23 8.1 0.43
Copper 222 0.35 8.2 0.63 0.01 8.2 0.02
Gold 114 0.32 9.0 0.38 −0.06 8.7 −0.08
Platinum 119 0.17 9.4 0.26 −0.26 9.4 −0.40
Silver 209 0.95 12.3 1.11 0.25 11.8 0.31

Note: M is the average return. SD is the standard deviation of the return. t(M ) is
the t-statistic for the average return.

a long futures position is positive. The average simple returns for 19 of

the 21 commodities are positive, and the average returns for cocoa, coffee,
orange juice, soy oil, and hogs are larger than 1.0% per month. However, the
evidence for normal backwardation is weaker than these averages suggest.
Standard deviations of monthly returns for individual commodities are often
greater than 10.0%. As a consequence, the average simple returns for only
five commodities produce t-statistics greater than 2.0. With continuous
compounding, only the average return for eggs is more than 2.0 standard
errors from 0.0 — and that return is negative.
 July 3, 2015 8:27 World Scientific Handbook of Futures Markets. . . 9.75in x 6.5in b1892-ch04 page 99

Commodity Futures Prices 99

If normal backwardation is the normal case, combining commodities into

portfolios does not smear information about expected premiums, and the
power of tests for expected premiums is improved. We average the simple
monthly returns on individual commodities to get equally weighted portfolio
returns. A portfolio of all 21 commodities and portfolios that include natural
subgroups are examined.
The average simple return on the portfolio of all commodities is 0.54%
per month. Its t-statistic is 1.87. Thus, on the average, the monthly changes
The World Scientific Handbook of Futures Markets Downloaded from www.worldscientific.com

in futures prices for commodities show marginally reliable normal backwar-

dation that is also non-trivial in magnitude. This conclusion is tempered
by the continuously compounded returns. With continuous compounding,
by LA TROBE UNIVERSITY on 05/21/16. For personal use only.

the average portfolio return falls from 0.54 to 0.45% per month, and its
t-statistic falls to 1.57. The non-trivial differences between average simple
and continuously compounded returns are easily explained. Even the return
on the portfolio of all commodities has substantial variability. The standard
deviations of its simple and continuously compounded returns are 4.3% and
4.2% per month.2 Among subgroup portfolios, the highly diversified portfolio
of agricultural products produces t-statistics around 2.0 in the simple and
continuously compounded returns, but the t-statistics for the average returns
for the remaining portfolios are 1.02 or less.
In short, large average premiums sometimes produce marginal evidence of
non-zero expected premiums when the futures contracts for commodities are
combined into portfolios. Even for portfolios, inference is sensitive to whether
we use simple or continuously compounded returns, and the evidence is never
strong. These results provide a good perspective on the problems of inference
posed by the variability of futures prices — and on the persistence of the
debate about the existence of expected premiums.

Summary

Two views of commodity futures prices are common. The theory of storage
summarized in Equation (4.2) explains the difference between a futures
price and the contemporaneous spot price (the basis) in terms of interest

2
Fama and Schwert (1979) report that the standard deviation of monthly rates of change
in the U.S. Consumer Price Index (CPI) is about 0.25%. The standard deviations of the
monthly rates of change of the nine major components of the CPI never exceed 0.62%.
These numbers are trivial relative to the standard deviations in Table 4.6. It seems safe
to conclude that general inflation is a negligible component of the short-term variation in
futures prices.
 July 3, 2015 8:27 World Scientific Handbook of Futures Markets. . . 9.75in x 6.5in b1892-ch04 page 100

100 E. F. Fama and K. R. French

forgone in storing a commodity, warehousing costs, and a convenience yield

from inventory. The alternative view of Equation (4.4) splits a futures price
into an expected premium and a forecast of the maturity spot price.
Although the two models are alternative perspectives on the same eco-
nomic phenomena, developing evidence on them presents different statistical
problems. Evidence on forecast power and expected premiums in futures
prices is extracted from realized spot-price changes and premiums, in the
manner, for example, of regressions (4.6) and (4.7). A typical characteristic of
The World Scientific Handbook of Futures Markets Downloaded from www.worldscientific.com

commodities traded in futures markets is highly uncertain future spot prices.

For many traded commodities, basis variances that are large in absolute
terms are small relative to the variances of realized premiums and spot-price
by LA TROBE UNIVERSITY on 05/21/16. For personal use only.

changes. As a consequence, although regressions (4.6) and (4.7) allocate all

basis variation to expected premiums and expected changes in spot prices,
for many commodities, the allocation is not statistically reliable, and non-
zero variances of the two expected values in the basis cannot be separately
identified.
Likewise, the large variances of realized premiums mean that average
premiums that often seem economically large are usually insufficient to infer
that expected premiums are non-zero, especially in the data for individual
commodities. When commodities are combined into portfolios, statistical
power is increased and marginal evidence of normal backwardation is
obtained. But the evidence is not strong enough to resolve the long-standing
controversy about the existence of non-zero expected premiums.
In contrast, regressions in which the basis is the dependent variable are
used to test whether the basis varies with interest rates, warehousing costs,
and convenience yields in the manner predicted by the storage-cost model
of (4.2). Since variation in the basis is not buried in extraneous noise —
even with a crude approach like the regression (4.3) of the basis on the
nominal interest rate and seasonal dummies — the tracks of the storage-cost
variables in the basis are identified more easily than variation in the basis
due to expected premiums and forecasts of future spot prices.
The results for the precious metals and the animal products illustrate the
different statistical problems. The low basis variances of the precious metals
allow precise estimation of the interest-rate response predicted in (4.2), but
they preclude a reliable split of the basis between the expected premium and
the expected spot-price change in (4.4). At the other extreme, the high basis
variances of the animal products preclude reliable estimates of the interest-
rate coefficient in (4.2), but they allow us to infer that futures prices have
power to forecast future spot prices.
 July 3, 2015 8:27 World Scientific Handbook of Futures Markets. . . 9.75in x 6.5in b1892-ch04 page 101

Commodity Futures Prices 101

Acknowledgments

This research is supported by the National Science Foundation (Fama) and

the Chicago Board of Trade (French). The comments of Wayne Ferson, John
Long, G. William Schwert, and workshop participants at the University of
Chicago, the University of California, Los Angeles (where Fama spends
winter quarters), the University of Rochester, and Yale University are
gratefully acknowledged.
The World Scientific Handbook of Futures Markets Downloaded from www.worldscientific.com

References
by LA TROBE UNIVERSITY on 05/21/16. For personal use only.

Bessant, L (1982). Commodity Trading Manual. Chicago, IL: Chicago Board of Trade.
Bodie, Z and VI Rosansky (1980). Risk and return in commodity futures. Financial
Analysis Journal, 36(May–June), 27–39.
Breeden, DT (1980). Consumption risks in futures markets. Journal of Finance, 35(May),
503–520.
Brennan, MJ (1958). The supply of storage. American Economic Review, 48(March),
50–72.
Carter, CA, GC Rausser, and A Schmitz (1983). Efficient asset portfolios and the theory
of normal backwardation. Journal of Political Economy, 91(April), 319–331.
Cootner, PH (1960). Returns to speculators: Telser vs. Keynes. Journal of Political
Economy, 68(August), 396–404.
Cootner, PH (1967). Speculation and hedging. Food Research Institute Studies, 7(Suppl.),
65–106.
Dusak, K (1973). Futures trading and investor returns: An investigation of commodity
market risk premiums. Journal of Political Economy, 81(November–December),
1387–1406.
Fama, EF (1984a). Forward and spot exchange rates. Journal of Monetary Economics,
14(November), 319–338.
Fama, EF (1984b). The information in the term structure. Journal of Financial Economics,
13(December), 509–528.
Fama, EF and GW Schwert (1979). Inflation, interest, and relative prices. Journal of
Business, 52(April), 183–209.
French, KR (1986). Detecting spot price forecasts in futures prices. Journal of Business,
59(Pt. 2, April), S39–S54.
Hansen, LP and RJ Hodrick (1980). Forward exchange rates as optimal predictors of
future spot rates: An econometric analysis. Journal of Political Economy, 88(October),
829–853.
Hardy, C (1940). Risk and Risk Bearing. Chicago, IL: University of Chicago Press.
Hazuka, TB (1984). Consumption betas and backwardation in commodity markets.
Journal of Finance, 39(July), 647–655.
Kaldor, N (1939). Speculation and economic stability. Review of Economic Studies,
7(October), 1–27.
Keynes, JM (1930). Treatise on Money. London, UK: Macmillan.
Salomon Brothers (various issues). Analytical Record of Yields and Yield Spreads.
New York, NY: Salomon Brothers.
 July 3, 2015 8:27 World Scientific Handbook of Futures Markets. . . 9.75in x 6.5in b1892-ch04 page 102

102 E. F. Fama and K. R. French

Telser, LG (1958). Futures trading and the storage of cotton and wheat. Journal of Political
Economy, 66(June), 233–255.
Telser, LG (1967). The supply of speculative services in wheat, corn and soybeans. Food
Research Institute Studies, 7(Suppl.), 131–176.
Working, H (1948). Theory of the inverse carrying charge in futures markets. Journal of
Farm Economics, 30(February), 1–28.
Working, H (1949). The theory of the price of storage. American Economic Review,
39(December), 1254–1262.
The World Scientific Handbook of Futures Markets Downloaded from www.worldscientific.com
by LA TROBE UNIVERSITY on 05/21/16. For personal use only.