Commodity Covariance Contracting

Peter Carr Anthony Corso

Banc of America Securities Sithe Energies, Inc.
9 West 57th street 335 Madison Avenue
New York, NY 10019 New York, NY 10017
(212) 583-8529 (617) 351-0390
pcarr@bofasecurities.com tcorso@sithe.com

Personal File Reference: Ccc2.tex

Current Version: October 12, 2000

We thank Keith Lewis, Dilip Madan, and Nedia Miller for comments. Any errors are our own.
I Introduction

A variance swap is a contract which pays the realized variance of the returns of a specified underlying asset

over a specified period of time. Several authors (see eg. Nueberger[12], Dupire[7], and Carr and Madan[4])
have shown that under certain conditions, the payoffs to a continuously monitored variance swap can be
synthesized by combining continuous trading in the underlying with static positions in standard options
maturing with the swap. Carr and Madan[4] have also shown that the covariance of returns between two

currencies can be replicated in this manner.

While the development of a replication strategy for variance and covariance swaps represents a signif-

icant theoretical advance, there are at least three problems with the proposed replication strategy. First,
the strategy replicates perfectly only if there are no jumps in the underlying. Second, the strategy assumes
that the variance swap is continuously monitored, even though all variance swaps are discretely monitored
in practice. Third, the strategy requires continuous trading in the underlying which is problematic in the

presence of transactions costs and market closings.

One purpose of this paper is to propose a solution which addresses all three drawbacks. By changing
the definition of variance in the swap from the realized variance of returns to the realized variance of
price changes, we show that the new payoff can be perfectly replicated in the presence of jumps, discrete
monitoring, and discrete trading opportunities. We further show that a contract paying the realized

covariance of price changes can also be synthesized in this setting.

We illustrate our results in the context of commodity options as the markets for these structures have
many of the features which we require. In particular, to synthesize covariance swaps, we use spread options,
which represent one of the few options written on two assets and listed on an organized exchange. Papers on
the valuation of spread options include Broadie and Detemple[2], Grabbe [8], Heenk, Kemna, and Vorst[10],

Pearson[13], Ravindran[14], and Shimko[16]. All of these papers assume that the covariance between
the two commmodities in the spread is constant. In contrast, this paper assumes that the covariance
between the two commodities is random, and furthermore that the stochastic process governing covariance

1
is unknown. Rather than price spread options in terms of a fixed covariance, we turn the problem around
and show how the covariance between the price changes in two commodity futures can be traded, given
the ability to trade dynamically in the futures and to take static positions in spread options and in options

written on each component of the spread.

The outline of this paper is as follows. The next section reviews the theory of static replication using

options. The following section shows how this theory can be combined with the standard theory of dynamic
replication using futures to create contracts written on realized variance. This theory is also used in the
penultimate section to synthesize contracts written on realized covariance. The final section summarizes
and suggests some avenues for future research.

II Review of Static Hedging Using Options

This section reviews the theory of static replication using options first developed in Ross[15] and Breeden

and Litzenberger[1]. Consider a single period setting in which investments are made at time t0 with all
payoffs received at time tn . In contrast to the standard intertemporal model, we assume that there are
no trading opportunities other than at times t0 and tn . We assume there exists a futures market in a
commodity for delivery at some date T ≥ tn . We also assume that markets exist for European-style

futures options1 of all strikes. While the assumption of a continuum of strikes is far from standard, it is
essentially the analog of the standard assumption of continuous trading. Just as the latter assumption is
frequently made as a reasonable approximation to an environment where investors can trade frequently,
our assumption is a reasonable approximation when there are a large but finite number of option strikes.

It is widely recognized that this market structure allows investors to create any smooth function f (Fn )

of the terminal futures price Fn by taking a static position at time 0 in options2 . When this theory is used to
generate desired volatility exposures, this paper will show that it is only the second derivative of the payoff
1
Note that listed futures options are generally American-style. However, by setting T = tn , the underlying futures will
converge to the spot at tn and so the assumption is that there exists European-style spot options in this special case.
2
This observation was first noted in Breeden and Litzenberger[1] and established formally in Green and Jarrow[9] and
Nachman[11].

2
which governs such exposures. Consequently, we will always choose f so that its value and slope vanish at
the initial futures price F0 (i.e. f (F0 ) = f ′ (F0 ) = 0). In this case, the results of Carr and Madan[3] imply
that any twice differentiable payoff can be spanned by the following position in out-of-the-money options:
Z F0 − Z ∞
f (Fn ) = f ′′ (K)(K − FT )+ dK + f ′′ (K)(FT − K)+ dK. (1)
0 F0 +

In words, to create a twice differentiable payoff f (·) with value and slope vanishing at F0 , buy f ′′ (K)dK
puts at all strikes less than F0 and buy f ′′ (K)dK calls at all strikes greater than F0 .

In the absence of arbitrage, a decomposition similar to (1) must prevail among the initial values.
Specifically, if we let V0f , P0 (K), and C0 (K) denote the initial prices of the payoff f (·), the put, and the
call respectively, then the no arbitrage condition requires that:
Z F0 − Z ∞
V0f = ′′
f (K)P0 (K)dK + f ′′ (K)C0 (K)dK. (2)
0 F0 +

Thus, the value of an arbitary payoff can be obtained from the option prices. Given the spectrum of
European options on the spread of two futures prices S = F1 − F2 , one can also create any smooth function
of this spread g(S). If we assume that the value and slope vanish at the initial spread S0 ∈ ℜ, i.e.

g(S0 ) = g ′(S0 ) = 0, then the analogous expression for the value of this payoff is:
Z S0 − Z ∞
V0g = f ′′
(K)P0s (K)dK + f ′′ (K)C0s (K)dK, (3)
−∞ S0 +

where P0s (K) and C0s (K) are the initial prices of European put and call spread options struck at K. Note
that no assumptions were made regarding the stochastic processes governing the futures prices.

III Creating a Contract Paying the Variance of a Commodity

Consider a finite set of discrete times {t0 , t1 , . . . , tn } at which one can trade futures contracts. For simplicity,

we will take these times to be closing times of trading days, but this restriction is not necessary. Let Fi
denote the price traded at on day i, for i = 0, 1, . . . , n. By day n, a standard estimator of the realized
annualized variance of price changes will be:
n
NX
Var(△F ) ≡ (Fi − Fi−1 )2 ,
n i=1

3
where N is the number of trading days in a year. The purpose of this section is to demonstrate a strategy
whose terminal payoff matches the above estimator of variance.

By Taylor’s series, we note that:

Fi2 = Fi−1
2
+ 2Fi−1 (Fi − Fi−1 ) + (Fi − Fi−1 )2 , i = 1, . . . , n.

Re-arranging and summing implies:

n n n
(Fi2 2
(Fi − Fi−1 )2 ,
X X X
− Fi−1 ) − 2Fi−1 (Fi − Fi−1 ) = i = 1, . . . , n.
i=1 i=1 i=1

N
The first sum on the left telescopes to Fn2 − F02 . Thus, multiplying both sides by n
implies:
n
N 2 N
(Fn − F02 ) −
X
Var(△F ) = 2 Fi−1 (Fi − Fi−1 ). (4)
n i=1 n

The first term on the right hand side can be regarded as a function φ(·) of Fn , where:

N
φ(F ) ≡ (F − F0 )2 . (5)
n

The first derivative is given by:

N
φ′ (F ) = 2(F − F0 ). (6)
n
Thus, the value and slope both vanish at F = F0 . It follows from (1) that the payoff φ(Fn ) can be replicated

using options. The number of options held at each strike is proportional to the second derivative of φ,
which is simply:
N
φ′′ (F ) = 2. (7)
n
Substituting (1) into (4) results in:
n
"Z #
N F0 − Z ∞ N
+ +
X
Var(△F ) = 2 (K − Fn ) dK + (Fn − K) dK − 2 Fi−1 (Fi − Fi−1 ). (8)
n 0 F0 + i=1 n

The initial cost of creating the first term on the right hand side of (8) is:

N Z F0 − NZ∞
V0 = 2 P0 (K, T )dK + 2 C0 (K, T )dK. (9)
n 0 n F0 +

4
If we assume that interest rates are constant at r, then the second term on the right hand side of (8) can
be regarded as the cumulative marking-to-market proceeds arising from holding −e−r(tn −ti ) 2 Nn Fi−1 futures
contracts from time ti−1 to time ti . Since futures positions are costless, the theoretically fair price to charge

for this variance contract is V0 given in (9).

Amusingly, the dynamic strategy in futures can be interpreted as an attempt to hedge the payoff φ(Fn )

made at tn , conducted under the false assumption of zero volatility. Given this ridiculous assumption,
φ
the value function is Vi−1 (Fi−1 , ti−1 ) = e−r(tn −ti−1 ) Nn (Fi−1
2
− F02 ) for i = 1, . . . , n. Recognizing that the
marking-to-market proceeds are realized one trading day after the position is put on, the zero vol hedger
φ
∂Vi−1 (Fi−1 ,ti−1 )
holds er(ti −ti−1 ) ∂F
= e−r(tn −ti ) Nn 2Fi−1
2
futures contracts from time ti−1 to time ti . This is exactly

the dynamic strategy needed to create the last term in (8). Since realized volatility will in fact be positive,
an error arises, and the magnitude of this error is given by the left side of (8).

IV Creating a Contract Paying the Covariance of Two Futures

Prices

Recall that:
Si ≡ F1,i − F2,i , i = 0, 1, . . . , n (10)

denotes the spread on day ti , where F1,i and F2,i denote the contemporaneous futures prices of the
two components of the spread. This section shows how to create a contract paying Cov(△F1 , △F2) ≡
n
N
(F1,i − F1,i−1 )(F2,i − F2,i−1 ) at time tn by combining static positions in options with dynamic trading
P
n
i=1
in the underlying futures.

We begin by recalling the well-known3 result that:

Var(△S) = Var(△(F1 − F2 )) = Var(△F1 ) − 2Cov(△F1 , △F2 ) + Var(△F2 ).

3
See Zerolis[17] for an intuition derived from the law of cosines.

5
Re-arranging this expression gives:
1 1 1
Cov(△F1 , △F2 ) = − Var(△S) + Var△(F1 ) + Var(△F2 ). (11)
2 2 2
Thus, one can synthesize a covariance swap by selling half a variance swap on the spread and buying half

a variance swap on each of the spread components. As these variance swaps are unlikely to be explicitly
available, they can be synthesized. Substituting (8) in each term on the right hand side of (11) implies:
n
"Z #
N S0 − Z ∞ N
(K − Sn )+ dK + (Sn − K)+ dK +
X
Cov(△F1 , △F2 ) = − Si−1 (Si − Si−1 ) (12)
n 0 S0 + i=1 n
 
(1) n
N
Z F0 − Z ∞ N (1) (1) (1)
(K − Fn(1) )+ dK + (Fn(1) − K)+ dK  −
X
+ Fi−1 (Fi − Fi−1 )
n 0
(1)
F0 + i=1 n
 
(2) n
N Z F0 − Z ∞ N (2) (2) (2)
(K − Fn(2) )+ dK + (Fn(2) − K)+ dK  −
X
+  Fi−1 (Fi − Fi−1 ).
n 0
(2)
F0 + i=1 n
From (10), the second term on the right hand side can be created by dynamic trading in futures on the
spread components:
n n
X N X N (1) (2) (1) (2)
Si−1 (Si − Si−1 ) = Si−1 [(Fi − Fi ) − (Fi−1 − Fi−1 )]
i=1 n i=1 n
n n
X N (1) (1) X N (2) (2)
= Si−1 (Fi − Fi−1 ) − Si−1 (Fi − Fi−1 ). (13)
i=1 n i=1 n
Substituting (10) and (13) in (12) and simplifying implies:
" #
N Z S0 − Z ∞
Cov(△F1 , △F2 ) = − (K − Sn )+ dK + (Sn − K)+ dK (14)
n 0 S0 +
 
(1) n
N
Z F0 − Z ∞ N (2) (1) (1)
(K − Fn(1) )+ dK + (Fn(1) − K)+ dK  −
X
+ Fi−1 (Fi − Fi−1 )
n 0
(1)
F0 + i=1 n
 
(2) n
N
Z F0 − Z ∞ N (1) (2) (2)
(K − Fn(2) )+ dK + (Fn(2) − K)+ dK  −
X
+  Fi−1 (Fi − Fi−1 ).
n 0
(2)
F0 + i=1 n

Since futures positions are costless, the fair price to charge for the covariance swap is the cost of creating
the static options position:
N Z S0 − s NZ∞ s
V0 = − P0 (K, T )dK − C (K, T )dK
n 0 n S0 + 0
Z (1)
N F0 − (1) N ∞
Z
(1)
+ P0 (K, T )dK + (1)
C0 (K, T )dK
n 0 n F0 +
(2)
N
Z F0 − N
Z ∞
(2) (2)
+ P0 (K, T )dK + C0 (K, T )dK. (15)
n 0 n (2)
F0 +

6
The investor must also trade futures on a daily basis, holding −e−r(tn −ti ) F2,i−1 units of the first futures
contract and −e−r(tn −ti ) F1,i−1 units of the second from time ti−1 to time ti .

Summary and Suggestions for Future Research

We showed that by combining static positions in options with dynamic trading in futures, investors can
synthesize contracts paying the realized variance of a commodity or paying the realized covariance between

two commodities. Importantly, these contracts were created without assuming anything about the under-
lying price. It would be interesting to extend our results to other payoffs besides variance and covariance.
Indeed, Carr, Lewis, and Madan[6] characterize the entire set of continuously paid cash flows which can be
spanned in our structure. It would also be interesting to consider nonlinear functions of realized variance

or covariance, such as options on these moments. Finally, it would be interesting to develop contracts on
other statistics of the sample path such as the standard deviation, the Sharpe ratio, skewness, correlation,
etc. In the interests of brevity, such inquiries are best left for future research.

References

[1] Breeden, D. and R. Litzenberger, 1978, “Prices of State Contingent Claims Implicit in Option Prices,”
Journal of Business, 51, 621-651.

[2] Broadie, M. and J. Detemple, 1994, “The Valuation of American Options on Multiple Assets”,
Columbia University working paper.

[3] Carr P. and D. Madan, 1997, “Optimal Positioning in Derivative Securities”, forthcoming in Review
of Derivatives Research.

[4] Carr P. and D. Madan, 1999, “Currency Covariance Contracting”, Risk Magazine, Feb., 47–51..

[5] Carr P. and D. Madan, 1998, “Towards a Theory of Volatility Trading”, in Volatility, R. Jarrow, ed.,
Risk Publications, 417–427.

7
 [6] Carr P., K. Lewis, and D. Madan, 2000, “On the Nature of Options”, Banc of America Securities
working paper.

[7] Dupire B., 1993, “Model Art”, Risk. Sept. 1993, pgs. 118 and 120.

[8] Grabbe, O., 1995, “Copper-Bottom Pricing”, Risk Magazine, 8, 5, May., 63-6.

[9] Green, R.C. and R.A. Jarrow, 1987, “Spanning and Completeness in markets with Contingent Claims,”
Journal of Economic Theory, 41, 202-210.

[10] Heenk, B.A., A.G.Z. Kemna, and A.C.F. Vorst, 1993, “Asian Options on Oil Spreads”, The Review

of Futures Markets, 9, 3, 510–531.

[11] Nachman, D., 1988, “Spanning and Completeness with Options,” Review of Financial Studies, 3, 31,
311-328.

[12] Neuberger, A. 1990, “Volatility Trading”, London Business School working paper.

[13] Pearson, N., 1995, “An Efficient Approach for Pricing Spread Options”, Journal of Derivatives, Fall,
76–91.

[14] Ravindran, K., 1993, “Low Fat Spreads”, Risk Magazine, 6, 10, Oct., 66-7.

[15] Ross, S., 1976, “Options and Efficiency”, Quarterly Journal of Economics, 90 Feb. 75–89.

[16] Shimko, D., 1994, “Options on Futures Spreads: Hedging, Speculation, and Valuation”, Journal of
Futures Markets, 183–213.

[17] Zerolis, J., 1996, “Triangulating Risk”, Risk, 9, 12, 36-8.