Professional Documents
Culture Documents
34
I
August 2003
I
n this article, we aim to describe an efficient option method for
the valuation of an exploration and production project. Consider
the example of a project manager who has the right to start drilling
on a given oil reserve in any one of five designated years, called
decision years. Two years after the drilling starts, crude oil and natural
gas will be produced for 12 consecutive return years, all of which, it is
assumed, the company will sell on the spot market.
This right resembles an American-style option (Bermudan), and if we
are prepared to make the assumptions required for the application of
derivative techniques, we may value it accordingly. However, using tradi-
tional approaches and somewhat realistic oil and gas price models, such
an option can be cumbersome to calculate. For example, the construc-
tion of the binomial tree and the calculation along it can be tedious and
time-consuming. We aim to show that a recently proposed least-squares
Monte Carlo approach (Longstaff and Schwartz (2001)) can solve this
problem very efficiently. It is easier to implement and allows more
general stochastic processes for state variables (oil and gas prices) than
with traditional approaches.
Valuing investment projects of this sort represents a classic example
of real option valuation. The main purpose of this article is to introduce
a new calculation technique for a familiar industry problem. Hence we
will not consider in detail some common but important issues asso-
ciated with real option valuation.
That said, we still want to bring attention to these issues and to the
limitations of our approach. One problem area concerns the discounting
of future cashflows. For investment commodities such as gold, we can
evaluate futures output at current spot prices without discounting. For
consumption commodities such as oil and gas, future output is not quite
the same as current output. Hence, we need to consider the convenience
yield and cost-of-carry respectively the benefit and cost of carrying the
commodity to determine the proper discount rate.
In actual calculation, instead of changing the discount rate, we continue
to use the risk-free interest rate and instead adjust down the drift term of
the spot price process by the convenience yield. If the commodity has a
futures market, we can use its futures prices to determine the average
convenience yield and, therefore, the proper drift (Brennan and Schwartz
(1985)). However, when the commodity does not have a futures market or
one that extends long enough more than 18 years in our example then
we have to determine the drift differently. In this case, the price process
needs to have a risk-adjusted drift that is, the original drift minus a risk
premium determined from an equilibrium model. Similarly, if output has
an unhedgeable risk for example, volume uncertainty we must deduct
another risk premium from the drift (Cox, Ingersoll and Ross (1985);
Schwartz and Trigeorgis, chapter 1, (2001)).
The deduction of risk premiums is an acknowledgement that under
these conditions, the full value of starting an investment project on a flex-
ible date instead of a pre-determined date, is not hedgeable. Therefore the
certainty-equivalent value of it is lower.
In the rest of the article we explain the basics of the least-squares
approach and report results of its application for our example.
The least-squares approach to calculating
an American-style (Bermudan) option
In each decision year, the project manager would compare the expected
value from oil and gas production if drilling were to start immediately
(the exercise value) with the expected value if drilling were to start later
(the continuation value). He starts drilling if the exercise value is larger.
For a given time and price path, the manager calculates the exercise value
from the net present value of cashflows, and the project value is the
discounted exercise value averaged over many simulated price paths.
The innovative part of the least-squares approach is in the calculation
of continuation value. We may estimate the continuation value by project-
ing cashflows realised from continuation on to the space spanned by
current prices we can do this through a simple regression. Detailed steps
for this process are explained in a simple numerical example in Longstaff
and Schwartz (2001, pages 115120). We provide a simple schematic
description in the equation below. But to fully understand the process,
one should work through their example, which can be easily reproduced
using a spreadsheet or by hand.
As shown above, we assume there are n price-paths. On path 1, the
current period price is s
1,t
, and the continuation value (expected project
value) next period is v
1,t +1
. Suppose next period, t +1, is the final period
to exercise, so one has to exercise. The continuation value, v
1,t +1
, is simply
the exercise value (discounted cashflow), given s
1,t +1
. Regressing
{v
1,t +1
,...,v
n,t +1
} on {s
1,t
,..., s
n,t
} yields f (s) =E(v| s), a conditional expecta-
tion function approximating the continuation value given price s.
1
Using
this function, we can compare the continuation value v
1,t
= f (s
1,t
) with the
exercise value at s
1,t
to decide whether to exercise on path 1 at time t.
For the next step, knowing {v
1,t
,...,v
n, t
} allows us to repeat the above
procedure at t 1 and so on, back to time zero. So at the end we will
know the exercise decision and value at each point in time along all price
paths. By discounting the values at the risk-free interest rate and aver-
aging them over all paths, we will obtain the expected project value at
time zero.
Hence, the key point here is that to apply this approach to evaluate
an exploration & production project, one only needs to define first the
price-generating process and second the exercise value along a price
path. We go on to define these two components and present the results
from a case study.
1, 1, 1
2, 2, 1
, , 1
Current period Next period
price project value
t t
t t
n t n t
s v
s v
s v
+
+
+
Lukens Energy Groups Hugh Li sets out an option method for valuing exploration
and production projects, using a practical example
Valuing exploration and
production projects
1
The function f(s) is assumed to be a linear combination of some basis functions such as s
2
,
exp(s/2) and so on. Our program used simple polynomial functions of s as bases. Calibrating over
from known American option values such as vanilla calls and puts, these basis functions perform as
well or better than other more elaborate choices proposed in Longstaff and Schwartz (2000)
http://www.risk.net/data/eprm/pdf/august2003/technical.pdf
Cutting edge: Exploration & production
36
I
August 2003
The price generating process and exercise value
In a given return year t, some volumes of crude oil and natural gas are
produced. Denote respectively the volumes as v
c
(t) and v
g
(t), and the spot
market prices as s
c
(t) and s
g
(t). We assume the two prices to be mean-
reverting and correlated, as in
,
(1)
.
Note that here both the means and volatilities may be time-varying. With
these prices, we can generate revenue s
c
(t)v
c
(t) +s
g
(t)v
g
(t), which is what
we actually need. Note that this simple price process implies a certain
term structure of volatility. If we derive implied volatilities based on an
option formula assuming a different price process notably Black-
Scholes the volatilities will be different. Nevertheless, a big advantage
of the least-squares approach is that is does not require a specific price
process, so we are free to choose others.
Suppose the drilling starts in year t
*
. After taking into account the
royalty rate (denoted by a); field operation cost for crude, b; tax rate, d;
and discounting, , the project net present value (NPV) of a given sample
path in price space is
.
The exercise value is the expected NPV. Ignoring volumetric uncertainty,
the exercise value is
.
(2)
Given (1), from Schwartz (1997),
,
where
,
and
.
*
*
2 2
2 ( ) 11 12
( ( )) (1 )
2
c
t t
t c
c
Var x t e
+
=
* *
*
2 2
( ) ( ) 11 12
*
( ( )) ln ( ) ( )(1 )
2
c c
t t t t
t c c c
c
E x t s t e e
+
= +
* *
*
( ( )) ( ( )) / 2
( ( ))
t c t c
E x t Var x t
t c
E s t e
+
=
*
* * *
*
13
*
2
( 1)
( ) (( ( ( )) ( ) ( ( )) ( )) (1 )
( ) )(1 )
t
t t c c t g g
t t
t
c
E v t E s t v t E s t v t a
v t b d e
+
= +
= +
*
*
13
( 1)
*
2
( ) (( ( ) ( ) ( ) ( )) (1 ) ( ) )(1 )
t
t
c c g g c
t t
v t s t v t s t v t a v t b d e
+
= +
= +
2
2
1
( ) / ( ) ( ( ) ln ( )) ( ) ( )
g g g g g j j
j
ds t s t t s t dt t dW t
=
= +
2
1
1
( ) / ( ) ( ( ) ln ( )) ( ) ( )
c c c c c j j
j
ds t s t t s t dt t dW t
=
= +
g
(t) =
g
, and
ij
(t) =
ij
do not vary with time. When they do vary with
time, we need to extend the results in Schwartz (1997). The derivation is
similar, but the specifics more complex. The results are listed below.
An extension: exercise value with varying price and volatility
In equation 1, let us suppose (t) and
ij
(t) are step functions (t) =u
k
,
ij
(t) =
ij k
, t[t
k1
, t
k
), k =1,...,T; i =1,2; t
0
=0, t
T
=T. E
t
*
(x(t)) and Var
t
*
(x(t))
in equation 2 have the following different expressions. Because the
expressions for crude oil and gas are symmetrical, for exposition purposes
we suppress the subscripts c and g.
Suppose t
*
[t
I1 -1
, t
I1
) for a certain I
1
, then for t [t
*
, t
I1
)
,
.
For t [t
I2-1
, t
I2
) with I
2
1 I
1
,
1 , 1 ,
1 2
* * 1 * 1
*
1
1 ,
2
1 * * 2 *
2 2
2 2
1
2 ( ) 1 1 2 ( ) 2 ( )
1
2
2
2 ( ) 1 2 ( ) 2 ( )
( ( )) ( ( 1) ( ( ))
2 2
( )) .
2
j I j i
I i i
j I
I
I
t t j j t t t t
t
i I
t t j t t t t
Var x t e e e
e e e
= =
= +
=
= +
1 , 1
*
* 1
* 1
1 , 2
* 1 *
1
1 , 2
1 *
* 2 *
2
2
2
( )) 1 ( )
*
2
2
1
1 ( ) ( )
1
2
2
( ) 1 ( ) ( )
( ( )) ln( ( )) (( )( 1)
2
( )( )
2
( )( )) ,
2
j I
I
j i
i i
j I
I
t t j t t
t I
I
j t t t t
i
i I
t t j t t t t
I
E x t s t e e
e e
e e e
=
= +
=
= +
+ +
1 ,
1
*
*
2
2
1 2 ( )
( ( )) (1 )
2
j I
j t t
t
Var x t e
=
1 ,
1
* *
* 1
2
2
1 ( ) ( )
*
( ( )) ln( ( )) ( )(1 )
2
j I
j t t t t
t I
E x t s t e e
=
= +
= +