Professional Documents
Culture Documents
Ramzi MAALOUF
PARTICULARITIES of the COMMODITIES MARKET
Rsum
Ce rapport de stage prsente une vue globale du march des matires premires. Limportance
de ce march na pas cess daugmenter ces dernires annes. En effet, lanne 2008 tait une
anne trs intressante, durant laquelle le march a t tmoin dune envole des prix du
ptrole, de lor, du cuivre, et des produits agricoles.
La premire partie du document introduit le march et met laccent sur la courbe forward.
Ensuite, elle se concentre sur la convenience yield , une notion particulire au monde des
commos. La deuxime partie labore le sujet des options et volatilits early expiry , qui est
aussi un concept spcifique au monde des commos. La troisime partie dcrit une mthode
utilise pour extraire la corrlation implicite entre deux sous-jacents partir dune option
basket. La dernire partie souligne limportance des options binaires dans le march des
commos et limpact qua le smile des volatilits sur les prix de ces options.
Jury
Guillaume FOUCHERES
Natixis
Commodity Derivatives
Lieu du Stage
Natixis
47, Quai d'Austerlitz
75013 Paris
France
BICH Philippe
Universit Paris 1
Table of Contents
1
St
( ST K ) ( fTT K ) (Q e 0 . ( ST K )
r .dt
(Q e 0 . ( fTT K )
So, theoretically, both call options are worth the same. This can be generalized to all types of
European payoffs. Thus, given the premiums are identical (almost identical in practice),
people tend to choose the most liquid asset as underlying. Consequently, in equity markets,
stocks are preferred. In commodity markets however, futures are preferred for the following
reasons:
1- they avoid all risks associated with spot trading, namely transportation, storage, and
counterparty risk
2- their prices are readily available on exchanges, unlike spot prices which can only be
known by contacting a number of dealers
3- futures and futures options are generally traded on the same exchange, facilitating
hedging and speculation and reducing arbitrage
If we denote :
FtT
S t . e c (T t )
B(t , T )
The other inequality does not hold because of the reasons mentioned above, namely points 1 and 5.
Therfore, to get the equality, we introduce a new parameter b, which represents the benefit from
holding the physical commodity rather than its corresponding futures or forward contract.
FtT
S t . e ( c b ) .(T t )
B(t , T )
S t . e y .(T t )
B(t , T )
The convenience yield y, combined both the storage cost and the physical benefit of holding
the commodity.
Note that for simplicity, the convenience yield is taken to be constant. A more general case
will follow.
ST S t e
y .(T t )
rs .ds
t
1
V 2 s .ds V s .dWs
2t
t
Since the futures price is a martingale under Q, the risk neutral probability, we have :
rs .ds
. ( e t
ft
( { f } ( {ST } St e
Q
t
T
T
Q
t
y .(T t )
Q
t
1
V 2 s .ds V s .dW s
2
Assuming the stochastic interest rate is independent of the brownian motion filtration, we get :
rs .ds
rs .ds 1 V 2 s .ds V s .dWs
Q 2t
t
Q t
y .(T t )
t
St e
. ( e
. (t e
St e
.(t e
ft
y .(T t )
Q
t
On the other hand, since the forward price is a martingale under QT, the forward neutral probability,
we have :
FtT (QtT {FTT } (QtT {ST }
.ST
(Qt e t
y .(T t )
1
QT
Q
. ( t St . e
(t {ST }
B(t , T )
B(t , T )
Ft
1
V 2 s . ds V s .dW s
2
1 V 2 s .ds V s .dWs
y .(T t )
1
St . e
2
t
. St . e y .(T t ) . (Qt e t
B(t , T )
B (t , T )
rs .ds
1
Q t
So, comparing futures to forwards reduces to comparing (t e
.
to
B(t , T )
Using Jensen' s inequality and the fact that f(x) 1/x is a convex function of x, we have :
rs .ds
(Qt e t
1
1
( T
t
T
rs .ds
rs .ds
Q
e t
(t e t
Q
t
1
B (t , T )
rs . ds
t
FtT .
This is not a surprising result, because under deterministic interest rates, the two probability
measures Q and QT are equivalent.
Since in the commodities markets, the interest rates are not the main source of uncertainty,
they are considered to be deterministic. Thus, futures and forwards are used interchangeably.
This is what I will be assuming for the rest of the document.
Consequently, in practice, we work backwards from the forward curve to deduce the implied
convenience yield. Some forward curves may be complex enough requiring a non-constant
convenience yield to explain their shape.
The graph shows the price per barrel on the y-axis and the maturity of the futures contract on the x-axis
ys .ds
St .e t
B (t , T )
FtT
Once the convenience yield is deduced from the available market quotes, it can be used to
determine the prices of futures that are not quoted on the exchanges.
The most common shapes of forward curves in the commodities markets are the next
subsections.
1.5.1 Contago
If we assume for simplicity that the interest rates and convenience yield are constants,
Contago is the case where (r y) > 0.
Usually, it is the case where the current world inventories for the commodity in question are
relaxed, and the common feeling in the market is the fear of a possible supply crunch in the
future, stimulating the buying of long dated futures. We therefore see the long dated end of
the curve higher than the short dated one.
The graph shows the price per barrel on the y-axis and the maturity of the futures contract on the x-axis
1.5.2 Backwardation
The forward curve is in backwardation when (r-y) < 0.
Usually, this is the case where the current world inventories are tight, and consumers prefer to
buying now what they will be using in the near future, fearing that inventories will keep on
decreasing. We therefore see the short dated end of the curve higher than the long dated one.
Below is an example of backwardation:
The graph shows the price per barrel on the y-axis and the maturity of the futures contract on the x-axis
It is worth noting that although it occurs very rarely, a forward curve may change from
backwardation to contago (or vice versa). This was the case for the Brent and WTI crude oil
futures during May and June 2008. Indeed, consumers feared that oil fields and reserves will
soon be depleted because of the massive demand of emerging countries like China and India.
This pushed the forward curves from contago to backwardation.
C0
B ( 0 , T ) { f 0T . N ( d 1) K N ( d 2 ) }
with
ln(
d1
f 0T
1 2
) V basket
T
K
2
V T
d2
d1 V
Shape of forward
Curve
Contango
Backwardation
Call premium
ATM means
K S0 | f 0 first Nearby f 0T
So, the option is actually ITM
ATM means
K S 0 | f 0 first Nearby ! f 0T
So, the option is actually OTM
Expensive
Cheap
Table 1-1
156
150
145
139
134
128
123
117
112
106
101
95
89.
84
78.
73
67.
62
56.
51
45.
40
0.0%
The graph shows the volatility smile of WTI crude oil options with 2 years to maturity. On the x-axis, we
have the strikes. On the y-axis, we have the implied volatilities.
If we consider several maturities, we obtain what is called a volatility surface. In order to have
consistent smiles, we should normalize the strikes by dividing them by the futures price. We
obtain what is called a moneyness ratio.
An example of a volatility surface is shown below:
10
50.00%
40.00%
30.00%
Implied Vol
20.00%
735
553
370
337
309
279
251
Maturity
219
188
160
10.00%
0.00%
45%
60%
75%
127
90%
Moneyness
105%
120%
97
135%
The graph shows the volatility surface of Brent Crude Oil. The maturity is in days.
Maturity in
days
97
127
160
188
219
251
279
309
337
370
553
735
45%
60%
49.59%
48.10%
47.69%
46.15%
44.93%
44.02%
41.75%
41.91%
39.04%
38.84%
39.17%
37.49%
47.52%
45.63%
45.00%
43.83%
42.90%
42.12%
40.12%
39.92%
37.85%
37.50%
38.07%
36.80%
75%
Moneyness
90%
44.15%
39.31%
43.69%
39.07%
42.40%
38.03%
41.53%
37.35%
40.68%
36.67%
39.99%
36.16%
38.04%
34.92%
37.58%
34.47%
36.18%
33.72%
35.84%
33.47%
35.94%
32.80%
34.53%
31.44%
Table 2-1
105%
120%
135%
34.92%
34.79%
34.31%
33.84%
33.35%
32.95%
32.48%
32.07%
31.81%
31.59%
30.32%
29.05%
33.35%
33.17%
32.68%
32.26%
31.81%
31.35%
31.38%
30.96%
30.85%
30.66%
28.86%
27.56%
34.18%
33.60%
32.76%
32.19%
31.64%
31.05%
31.17%
30.69%
30.56%
30.35%
28.25%
26.89%
11
Tnfuture
Tnoption
Note that the standard options are listed on the same exchange as their corresponding futures
contract, and they obviously mature a few days earlier than the futures contract maturity
option < T future ). The reason for that is to avoid physical delivery of the commodity if
n
( Tn
Future 1
Future 2
Futures
Maturity
Options maturity
(in days)
T1future
T future
2
Moneyness
80%
90%
ATM
110%
120%
T1option = 6
49.12%
43.80%
37.84%
34.71%
36.32%
T2option = 35
45.59%
41.38%
37.56%
35.28%
35.09%
= 66
44.47%
40.62%
37.09%
34.93%
34.29%
T3future
T future
T4option = 97
42.86%
39.31%
36.10%
34.09%
33.35%
T5option = 127
42.45%
39.07%
35.97%
33.98%
33.17%
Future 6
T5future
T future
T6option = 160
41.09%
38.03%
35.34%
33.56%
32.68%
Future 7
T7future
40.23%
37.35%
34.83%
33.14%
32.26%
Future 3
Future 4
Future 5
option
3
option
7
= 188
Table 2-2
12
expiring much earlier than their underlying futures contract, hence the name early expiry
options.
As an example, the WTI crude oil options with maturities beyond 1 year are only listed for the
maturity months of June and December. So, if the current month is July 2008, and a consumer
wants to hedge against an increase in price of oil in September 2009, he/she will not find the
appropriate call option in the exchange. He/she must buy an Over the Counter early maturity
option. In order to price that option, we need to know the early expiry implied volatility.
So, the problem at hand is the following:
Given a standard volatility matrix (in moneyness format) of future contracts on a certain
commodity, we need to generate an array of early volatility matrices (one early volatility
matrix per future).
Now the question is:
Given the standard volatility matrix shown in the table above, we need to find the volatility of
other words, the problem is the one of generating, for each future, the following early
volatility matrix:
Early volatility matrix of future 5
Future
Futures
Maturity
Options maturity
(in days)
T1option = 6
Future 5
T5future
option
2
option
3
Moneyness
80%
90%
ATM
110%
120%
= 35
= 66
T4option = 97
42.45%
39.07%
35.97%
33.98%
33.17%
option
5
= 127
Table 2-3
13
2.3 Methodology
To tackle this problem, we take two main assumptions:
Assumption 1: For a given commodity, ATM instantaneous volatility is stationary
By stationary, we mean that the instantaneous volatility does not depend on time but rather on
the distance to maturity.
Let T2 ! T1
If we denote :
the instantaneous implied volatility of option with maturity T1
V 1 (t )
at time t, t [0, T1 ]
the instantaneous implied volatility of option with maturity T2
V 2 (t )
at time t, t [0, T2 ]
then assumption 1 gives :
E{V 2 (T2 T1 t )}
E{V 1 (t )}
2
If we denote :
6 Tn
Ti , Tj
the average ATM volatility on the period [Ti, Tj] of the option whose standard
maturity is Tn
(6 TT 22 T 1, T 1 ) 2 .T 1
2
V 2 (t ) dt }
E{
T 2 T 1
T1
E{ V 1 (t ) dt}
2
T2
E{V 2 (t )} dt
2
T 2 T1
T1
E{V
2
1
(t )} dt
(6 T0,1T 1 ) 2 .T 1
T2
T 2 T 1, T 1
T1
0, T 1
In words, this means that seen from today, the volatility during the last month of the life of an
option is the same as the volatility of an option that expires in one month. Obviously, this
interpretation applies to all durations other than one month.
Visually,
14
.
Same average ATM volatility
for the specified duration (T1)
T1
T1
T2
T1
T2
T3
Assumption 2: For a given futures contract, the standard volatility smile is replicated in its
early volatility smiles. The reason for that is that traders believe it is the best estimate for the
early volatility smiles. They assume market participants will keep the same skewed
probability distribution for the underlying, independent of the options maturity.
Therefore, the methodology to generate an array of early volatility matrices is the following:
For future n, we first start by computing the ATM early volatilities based on assumption 1.
Then, we compute early ITM and OTM volatilities based on the previously computed early
ATM volatilities and assumption 2.
2
(127
0 , 66 ) u 66
2
(127
66 ,127 ) u 61
(1)
15
2
(127
0 , 66 ) u 66
2
( 61
0 , 61 ) u 61
Moreover, from the standard volatility matrix, we can directly read the following volatilities:
127
0,127 35.97%
66
0,66 37.09%
35
0,35 37.56%
35
Thus, by linear interpolat ion between th e couples (66 , 66
0,66 ) and (35 , 0,35 )
we find : 61
0,61 37.16%
So , 127
0,66
2
61 2
( 127
0,127 ) u127 ( 0,61 ) u 61
34.83%
66
66
Early Volatility
(Unknown)
127
We now calculate the 66days early 80% volatility (the second volatility in question),
denoted by 80%
early
80%
80%
early
Standard
By assumption 2, we have:
(2)
ATM
ATM
early
Standard
16
The right hand side of the equation can directly be read from the standard volatility matrix.
ATM , on the other hand, is calculated in the previous step
early
We therefore have:
42.45
80% 34.83u
%
early
35.97
41.10%
In the same fashion, the rest of the early volatility matrix is computed. The result is shown in
the table below.
Futures
Maturity
Options maturity
(in days)
T1option = 6
Future 5
T5future
option
2
option
3
Moneyness
80%
90%
ATM
110%
120%
41.84%
38.51%
35.45%
33.49%
32.70%
= 35
41.54%
38.24%
35.21%
33.26%
32.47%
= 66
41.10%
37.83%
34.83%
32.90%
32.12%
T4option = 97
41.83%
38.50%
35.45%
33.48%
32.69%
42.45%
39.07%
35.97%
33.98%
33.17%
option
5
= 127
Table 2-4
3 Options on Baskets
3.1 Pricing options on baskets
Options on a basket of underlying assets are very common in commodity markets. We can
imagine a consumer who uses both natural gas (Nat Gas) and crude oil (WTI) for the
manufacturing of his goods. This consumer would be interested in buying a basket call option
to hedge an increase in price rather than buying 2 separate call options, since it is cheaper to
do so. Certainly, this reduction in cost results from a weaker protection, since the basket call
option is only profitable when the average price exceeds the strike.
Mathematically, if we denote
Ft1,T
Ft 2,T
1
t
2
t
17
Note that if the spot price is not readily available or accessible in the market, it is replaced
by the first nearby.
That is, we replace Sti by Ft i , first nearby in all what follows. This is the case for a lot of commodities,
like crude oil and natural gas.
Let us define the process :
A European call option on an equally weighted basket of underlying assets 1 and 2 has the
following payoff at maturity T :
Yt
1 Ft1,T Ft 2,T
2
S0
2 S01
Payoff
(YT K )
Callt (Yt ) d
1
2
1,T
Callt Ft 1
S
F 2 ,T
Callt t 2
S0
18
d1
E Q [Y ] 1 2
ln 0 T V basket
T
K% 2
d2
V basket T
d1 V basket T
So, the two unknowns in the above equations are E0Q [YT ] and V basket .
Let :
M 1 E0Q [YT ]
1 F01,T F02,T
(
2 ) Y0
2 S01
S0
Now that we found the first unknown, let us try to find the second, V basket :
To do so, we have to pass by an intermediate step:
Let :
M 2 E0Q [YT2 ]
1
4
S 01 .S 02
( S 01 ) 2
( S 02 ) 2
But since the futures price is an exponential martingale under Q, and assuming V is deterministic,
we have :
Q
E0 { ( Ft1,T ) 2 }
1,T 2
0
2
To
o
1 T 2
V 1 s .ds V 1s .d W s
2
Q
0
0
. E0 e
F . e
1,T 2
0
E0 { ( Ft 2,T ) 2 }
2 Vo1s .d Wo s
Q 0
E0 e
V 2 s .ds
0
F . e
1,T 2
0
V 1 s .ds
2
2
To
o
1 T 2
V 2 s . ds V 2 s .d W s
2
Q
0
. E0 e 0
F
2 ,T
0
F . e
2 ,T 2
0
V 2 s .ds
0
2 Vo 2 s .d Wo s
Q 0
E0 e
F . e
2 ,T 2
0
V 2 s .ds
2
19
1,T
E0 { Ft . Ft
2 ,T
1,T
0
2 ,T
0
F .F
.e
1
V 1 2 s .ds
2
1
V 1 2 s . ds
2
.e
F01,T . F02,T .e
1,T
0
2 ,T
0
F .F
1
V 1 2 s .ds
2
T o o
V 1 V 2 .ds
.e 0
(Vo1s
Q 0
.E0 e
.e
1
V 1 2 s .ds
2
.e
V 2 s ).d W s
o
1 o
V 1 s V 2 s .ds
2
1,T
0
2 ,T
0
F .F
U V 1 V 2 .ds
.e 0
M 2 E0Q [YT2 ]
2
2
F 1,T
1
F 2 ,T
F 1,T . F 2,T
{ ( 0 1 ) 2 . e V 1 T ( 0 2 ) 2 . eV 2 T 2. 0 1 02 . e U V 1 V 2 T
S0
4
S0
S0 .S0
(1)
Now that we found M2, we have to find a relation between M2 and V basket .
For any exponential martingale, and for Yt in particular, we have:
1
YT
Y0 .e
2
V basket
2
.T
V basket .W T V basket
M
1
. ln 22
T
M1
M2
Y0 . e V basket .T . E 0 {e
2
2.V basket .W T
} M 1 . e V basket .T
2
( 2)
Now that we found all the parameters we need, B&S formula can readily be applied.
20
2
T
1 F01,T F02,T V basket
1 2 e
Then, from equation (2) above, we find M 2 M .e
2 S
S0
0
After determining M 2 , we have everything we need to determine the implied correlation
2
1
V2
basket
2
2
F 2, T
1 F 1,T
M 2 [ ( 0 1 ) 2 . e V 1 T ( 0 2 ) 2 . eV 2 T ]
4
S0
S0
ln{
}
1, T
2,T
1 F0 . F0
2 S 01 .S 02
V1 V 2 T
In most cases, the implied correlation turns out to be different than the historical correlation,
just like implied and historical volatilities are different. Implied correlation is interpreted as
the correlation the market is using to price basket options.
From equations (1) and (2), we can see that M2 is an increasing function of U , and V basket is
an increasing function of M2. We also know that the call value is an increasing function of
volatility. This shows that the price of a call basket is monotonically increasing with U .
So, if the market is pricing a high implied correlation, it means there is a common sentiment
that both underlying assets are going to increase together. This creates a high demand on call
baskets for hedging purposes, pushing their price higher.
Once this implied correlation is determined, it can either be used to price other basket options,
or it can be used to build a correlation matrix, which is in turn used to price basket options on
more than 2 underlying assets.
4 Digital options
4.1 Popularity of digital options
A European digital (or binary) option with maturity T and strike K on an underlying futures
contract FTt, is the option whose payoff at T is 1^F T t K `.
T
Digital or binary options are very common in the commodities market. They are used in the
pricing of popular payoffs, like European barrier options (knock in and knock out options).
The reason for that popularity is that the option structure disregards the highly improbable
events with very large payoffs, reducing the premium for the client.
An example of such a payoff is shown below:
21
Payoff
25
50
75
100
125
150
175
Percentage
Moneyness
We can see that the investor is protected against the rise in the price of the underlying up to
the level of 175%, after which a rebate of 25% of the nominal is paid. Since this event is
considered as highly unlikely, giving away this payoff is not much of a problem for the
investor, and will reduce the overall cost of the structured product.
On the other hand, if the price drops, the client is not exposed to the downside loss unless the
drop is bigger than 25%. In the latter case, the investor experiences a 100% participation loss,
which is capped at 50% of the nominal.
The above payoff is easily structured using vanilla and digital calls and puts.
In fact, if we denote
N
C(K)
P(K)
DC(K)
DP(K)
22
DC 0 E 0Q e 0 .1^F T t K `
T
DC 0
FTT t K F0T .e
1
V 2T V .WT
2
B(0, T ). QT FTT t K
FT 1
ln 0 V 2T
K 2
K 1
t K V .WT t ln T V 2T z t
V. T
F0 2
d2
where z ~ N(0,1)
So, DC 0
Now, we will study the first derivative of d2 with respect to volatility, which will directly give
us the behavior of the price since N(x) is an increasing function of x.
wd 2
wV
FT
ln 0
K
2
V . T
1
T
2
23
wd
t 0 2 0
wV
So, the price of the digital call is always decreasing with volatility. (displays a -ve Vega)
0.8
0.6
0.4
0.2
145%
139%
133%
127%
121%
115%
109%
97%
103%
91%
85%
79%
73%
67%
61%
55%
49%
43%
37%
31%
25%
19%
7%
13%
1%
Volatilities
FT
ln 0
K
wd 2
0 wV ! 0
wd 2 0
wV
FT
ln 0
K
when V 2
T
otherwise
So, the price of the digital call is increasing with volatility. (+ve Vega) up to a certain point,
where more volatility will actually hinder the calls value.
24
0.1
0.08
0.06
0.04
0.02
193%
185%
177%
169%
161%
153%
145%
137%
129%
121%
113%
97%
105%
-0.02
89%
81%
73%
65%
57%
49%
41%
33%
25%
9%
17%
1%
Volatilities
K1 K
K2
Spot Price
25
In practice, this replicating portfolio rarely has the same value as the formula for the price of a
digital call derived in the previous section. This difference is clearly seen when the volatility
smile is not flat.
Intuitively, when the volatility skew is upward sloping, the replicating portfolio consists of
buying a large number of calls with strike K1 and selling the same number of calls with strike
K2. Since C(K1) is priced with a volatility V1 a bit smaller than V2 ( which is used to price
C(K2) ), the replicating portfolio is worth less than the case where V1 = V2 (when the
volatility smile is flat). The opposite can be said when the volatility is downward sloping.
So, given the same parameters (Maturity, Strike, Volatility, and interest rates), and the
different volatility smiles shown below,
Volatilities
Smile 3
Smile 2
Smile 1
Strikes
wC S , K , V K , T , r
C S , K H , V K H , T , r
H o0
2.H
wK
wC S , K , V , T , r
dV K
wC S , K , V , T , r
.
dK
wK
wV
K
V V K
VV
Slope of volatility
lim
B ( 0 ,T ). N ( d 2 )
Vega of
Vanilla Call
First, we observe that when the smile is flat (B&S world), the slope is zero, and the equation
reduces to the formula obtained in the previous section.
Second, since the Vega of a vanilla call option is always positive, the price is only affected by
the sign of the slope parameter. In the case it is positive, the price decreases and vice versa.
26
smile
5 References
1- Commodities and Commodity Derivatives: Modeling and Pricing for Agriculturals,
Metals, and Energy by Helyette ,Prof. Geman
2- Options, Futures and Other Derivatives, by John C. Hull
3- Commodity trade definitions and pricing, ECDA Analytics, Natixis
4- Interest Rate Models Theory and Practice, by Damiano Brigo Fabio Mercurio
27