The Optimal Investment Scale and Timing: A Real Option

Approach to Oilfield Development

Marco Antonio Guimares Dias
Petrobras
[email protected]

Katia Rocha
Instituto de Pesquisa Econmica Aplicada
[email protected]

Jos Paulo Teixeira
Departamento de Engenharia Industrial, Pontifcia Universidade Catlica do
Rio de Janeiro
[email protected]


The Optimal Investment Scale and Timing:
A Real Option Approach to Oilfield Development



Abstract
The oil company holds the investment opportunity to develop a delineated oilfield. The investment
plan must be presented until a specific date or the oilfield rights return to the government. The firm
considers a set of mutually exclusive alternatives of scale to exploit the oilfield. Larger scale means
faster exploitation increasing the present value of revenues, but also higher investment cost. Oil
price uncertainty affects all alternatives. In addition to the scale option, the firm has a timing option
and hence this investment opportunity is analog to a finite-lived American call option on the best of
multiple assets with the same underlying oil price stochastic process but with different benefits and
different exercise prices. We examine both geometric Brownian motion and a mean-reversion
process to model oil prices. We obtain the undeveloped oilfield (real option) value and the optimal
investment rules, i.e., the optimal timing and the optimal scale thresholds.

Keywords: Real Options, Oilfield Development Investment, Optimal Scale of Projects, American Option on
Multiple Assets.
1
1. Introduction
The real options approach is an effective method of economic analysis of investments in projects or
non-financial assets ("real assets") under market and technical uncertainties, because it considers
the value of managerial flexibility to react to these changing scenarios. Although relatively recent,
the real options approach has been adopted by an increasing number of modern corporations
1
.
The presence of managerial flexibility in the decision making process under uncertainty provides
important gains in the valuation of the investment opportunity, especially for low net present value
(NPV) projects, which are similar to at-the-money options. These low NPV opportunities are
more difficult for decision-makers analysis about whether to undertake or reject the project
2
.
Regarding strategic sequential investments, the NPV rule turns out to be a difficult application and
can even fail in trying to quantify the hidden options value provided by the investment
3
. Such
investments are precisely those where the real options approach aggregates more economic value.
Finance literature presents several cases of applications of real options to value natural resource
investments, such as Tourinho (1979) and Brennan and Schwartz (1985)
4
. Paddock, Siegel and

1
A large survey reported in Graham & Harvey (2001) on corporate finance practice, with answers from 392 CFOs of
different firms in the USA and Canada, shows that 26.59% of the firms "always or almost always" consider the value of
real options in projects. This number is about two times the number of firms that use value-at-risk (VaR) or Monte
Carlo simulation. See also at www.realoptions.org articles with recent examples of real options applications for
investment decisions in several corporations, such as HP-Compaq, General Motors, Chevron, Texaco, Pfizer Inc., BP-
Amoco, Dell Computer Corp., J.P. Morgan,, Boeing and Schering Plough, among others.
2
Dixit and Pindyck (1994) present a comprehensive explanation of the differences between NPV and real options
approach to value investment opportunities under uncertainties.
3
Trigeorgis (1996) presents some examples such as: optimal timing of an investment, option to expand, abandon or
suspend the project, strategic options, option to modularity, learning options, etc.
4
Dias (2001) gives an overview of real options applications in petroleum. Schwartz and Trigeorgis (2001) present other
applications of real options to natural resources investment opportunities.
2
Smith (1988) present the classical model of real options for exploration and production (E&P) of an
oilfield, exploiting the analogy between the concession value and the (financial) American call
option.
Usually, after the oilfield delineation at the end of the exploratory phase
5
, the E&P firm holds an
investment opportunity to develop the oilfield by incurring the development costs. In this phase,
uncertainty about the economic value of the reserve, related mainly to future oil prices, is the most
relevant since the exploration phases significantly reduce the technical uncertainties. At any time up
to the option expiration - established by the government agency - the E&P firm can commit to an
investment plan for immediate field development, and therefore holds an equivalent American call
option. At the expiration, the firm can return the oilfield to the government if the investment
opportunity to develop the field is not attractive. We consider both the value of the underlying asset
V (or developed reserve value) and the development cost (D) in present values, so NPV = V D.
Hence, we don't consider the options available during the time to build the project
6
.
This paper follows Dixit (1993a) and Dias (1998, 2001) by calculating the optimal development
investment timing of an oilfield through a finite number of mutually exclusive development

5
The exploratory phase is also an investment option, where the holder of the track rights can exercise the option to drill
the wildcat well. In case of success, the holder can delineate the petroleum field by paying the cost of appraisal wells.
The relevant uncertainties in this phase are mainly oil prices and the technical uncertainties related to the existence,
size and quality of the petroleum reserve. In bids for tracks rights or in wildcat drilling, the strategic uncertainty of the
other firms behaviour is also relevant.
6
For applications of the time to build see Majd and Pindyck (1987), Pindyck (1993), and the discussion in Dixit and
Pindyck (1994, chapter 10). Rocha (1996) provides additional discussions and extensions to the time to build models.
3
alternatives of scale. We do not consider here the option to expand capacity by investing first in a
lower scale and then expanding to a higher scale by adding capacity
7
.
We examine three mutually exclusive alternatives, but the method can be generalized for n different
alternatives of scale. These alternatives have different scale in terms of different numbers of wells,
different processing plant capacities, different pipeline diameters, different types of production
units, etc. There is a trade-off because higher installed capacity results in higher present value for
revenues (higher value of V) but requires higher investment (higher D).
Dcamps, Mariotti and Villeneuve (2003) give an important and mathematically rigorous revision
of Dixit (1993a). They worked with perpetual options and allowed the option to switch from a low-
scale to high-scale project.
We work with finite-lived options and do not allow this option to switch. However, their results
provide important insight for our model. Particularly, like us they also found intermediate waiting
regions that will be discussed in the next topic. In addition, the presence of the option to switch
does not change the threshold to invest in a low-scale project, so that the investment rule in high-
scale project is myopic regarding the possibility to switch to higher scale if the market conditions
improve even more.
Another related reference is Capozza & Li (1994). They model a perpetual option to redevelop an
urban land parcel by choosing both the optimal timing (by looking at market demand) and the
intensity (or scale) of investment. In their case there is a continuum of investment intensities
available to the landowner. Although they use perpetual options, an analytical solution is not
available for this case of continuous investment intensities.

7
However, PUC and Petrobras (through Pravap-14) developed one research project on alternatives with option to
expand production through optional wells. See Dias (2001) for the basic starting ideas.
4
Brodie & Detemples (1997) analysis of American option on multiple assets seems to be related
with our work because the n underlying assets V(k), k = 1, 2n, are mutually exclusive. In their
paper, each asset follows a different but correlated geometric Brownian motion. However, they
used the same exercise price (analogous to investment in real options) for all assets, which is not so
pertinent for optimal scale choice. Here we use different exercise prices (different investments) for
each alternative, and the stochastic process (for oil price P) is the same for all mutually exclusive
underlying asset V(P, k) alternatives. Here the difference between alternatives is placed in a single
deterministic term in the equation of V (see the next section), in addition to the investment D(k).
So, here we model optimal scale with the trade-off between investment D(k) and present value of
revenues net of operational costs and taxes V(P, k).
The article of Geltner, Riddiough and Stojanovic (1996) is similar to Broadie & Depemple (1997)
because they also assume the same exercise price for the alternatives. However, they studied the
case of two alternatives of land use choice in a perpetual option framework.
Pindyck (1988) is a classical real options paper on capacity choice, but he considers that the
capacity can be incrementally expanded. In our paper we consider that if we install a certain
capacity or scale of production, we continue with the same scale until the oil reserves exhaustion.
This is true in a large range of real oilfield development projects, chiefly because the oil production
reaches a peak in the beginning of its productive life and declines over time due to depletion of
reserves. For other industries, Pindyck's case can be more common.
Childs, Ott and Triantis (1998) examine mutually exclusive projects in a real options framework.
They consider two interrelated projects, each one with two phases (pilot and implementation).
However, they do not address the problem of optimal scale like here and do not consider the option
to delay the investment. They consider technical uncertainty and full information revelation with
5
the pilot phase, which affects the choice of alternatives. They concentrate on comparing parallel
versus sequential pilot projects and on determining their optimal sequence.
This paper is organized as follows. Section 2 presents at the conceptual level the investment model
under oil price uncertainty with finite time to expiration, and discusses the intermediate waiting
regions feature not addressed in the Dixit paper. Section 3 presents the equations for both geometric
Brownian motion and mean-reversion models, as well as the stochastic differential equations and
their boundary conditions. Section 4 presents numerical examples for both stochastic processes,
highlighting the effect of volatility on the optimal timing to develop the oilfield and the optimal
scale of output, and discuss the possible issue of discontinuity of the threshold curve for mean-
reversion. The last section presents the concluding remarks.
2. The Model on Selection of Mutually Exclusive Alternatives to Develop an Oilfield
Feasibility studies by oil companies typically analyze a set of a few discrete mutually exclusive
alternatives to develop an oilfield
8
. Each alternative has a different number of development wells,
processing plant capacities, pipeline diameters, etc. In our framework we consider that the only
source of uncertainty is the oil price P(t), which changes stochastically over time.
Let there be n = 3 alternatives to develop an oilfield. Denote alternative k by A
k
, k = 0, 1, 2, 3, with
alternative A
0
being not to develop. Each alternative k has investment D(k) and benefit V(k). The
net present value (NPV) of the investment opportunity to develop the oilfield at time t by choosing
the optimal alternative k is defined in Eq.(1), where V is the economic value of the developed
oilfield.

8
This practical issue was the first reason for Petrobras to develop a research project with PUC-Rio on this theme. The
second reason - more theoretical - was some intriguing results related to the existence of intermediate waiting regions
not addressed in Dixit (1993a) that Dias found by using approximated models with Excel spreadsheets. These waiting
regions were confirmed in this research project.
6
NPV(P(t), k) = V(P(t), k) D(k) (1)
We consider the fiscal regime of concessions used in about the half the worlds countries. For these
countries, the linear relation between oil prices and the NPV associated with the exercise of the
option to develop the oilfield is at least a very good approximation. One linear model
9
is the
proportional model (V is proportional to P), also called the "business model", expressed as:
V(P(t), k) = P(t) . q(k) . B (2)
where:
P(t) = oil price per barrel at time t ($/bbl);
q(k) = economic quality of the developed reserve by adopting development alternative k;
B = the reserve volume (or number of barrels in the ground) in millions of barrels (million bbl).
Equation (2) allows that the value of the developed reserve (V) can be conveniently given as
proportional to the oil price (P). By using It's Lemma, it is easy to show that V follows the same
stochastic process as P
10
. Note that since V is the present value, all discounting effects are
embedded in the quality factor q.
The proportionality factor q is named the economic quality of the developed reserve because as q
increases, so does the specific value of this developed reserve
11
(for the same oil price and volume).
This quality factor depends on many variables, such as reservoir rock quality (permo-porosity
properties), hydrocarbon quality, country taxes, operational costs, discount rate, infrastructure

9
Another linear model is V(k) = [P(t) . q'(k) . B] C(B), where q' is the economic quality of the reserve for this model
and C(B) is a kind of fixed operational cost in present value. In this case, V and P follow different stochastic processes.
Proof and details at http://www.puc-rio.br/marco.ind/payoff_model.html
10
For the proof see: http://www.puc-rio.br/marco.ind/payoff_model.html
11
Dias (1998) introduced the concept of economic quality of the reserve, suggesting the application of this concept to
model the selection of mutually exclusive alternatives under oil prices uncertainty.
7
proximity, and development capital in place (scale). Only the latter issue is exploited in this paper
because the choice of development scale is one of our control variables in this optimization under
uncertainty problem. In a more general setting (e.g., non-linear payoff models), the economic
quality of the developed reserve for alternative k is defined by:
P
k) V(P,

B
1
k) q(P,

= (3)
We will assume that the value of developed reserve V(P) is linear with P, so that the quality
parameter will be constant (independent of P), differing only across the alternatives of scale.
The motivation for the name "business model" is drawn from the reserves transactions market. For
example, in the United States the average price paid for one barrel of developed reserve is 33% of
the wellhead oil price, which Gruy et al. (1982) named the one-third rule of thumb
12
.
In order to understand the model better, let us examine the case without the option to wait, that is,
the case at expiration (the legal constraint set by the government agency).
Figure 1 shows the NPV(P) linear chart for three different development alternatives.

12
This one-third rule of thumb was used also in Paddock, Siegel and Smith (1988) to perform a numerical example.
However, they model the option value as a function of F(V), not F(P) as here. In addition, they did not devote any
discussion on the properties of the proportionality factor q as here. In our paper, this quality factor has a key role
because it captures the differences of alternatives of scale on the benefit side in the simplest way possible.
8

Figure 1 - NPV Functions and the Optimal Exercise at Expiration
Note that the slope of NPV
k
(P) is one key difference between the alternatives and it is given by
tg(
k
) = q
k
B, so that the economic quality of the developed reserve is related to the slope of
NPV
k
(P). The other difference between the alternatives is the investment, given by the Y-intercept
of the NPV lines, that is, NPV
k
(P = 0) = D
k
.
At expiration (t = T), the traditional NPV rule holds, that is, to exercise the alternative with higher
NPV. This is presented in Figure 1 above by the envelope NPV function (thicker lines), and will be
one boundary condition of our model.
Figure 1 also shows the oil prices at which we would be indifferent between two alternatives. P
0, 1
is
the indifference point between exercising alternative 1 (A
1
) and not investing (alternative 0). P
1, 2
is
the indifference point between exercising alternatives A
1
and A
2
(equal NPVs for these
alternatives). P
2, 3
is the indifference point between alternatives A
2
and A
3
.
9
Before expiration (t < T), due to oil price uncertainty, the "wait and see" policy can be optimal even
if all alternatives have positive NPVs because the option to delay the exercise can be more valuable
than the immediate exercise of any alternative. Before the presentation of the differential equation
for the option value, it is useful at this point to discuss conceptually some results in order to
understand better the boundary conditions for this model in the next section.
Dixit (1993a), working with perpetual options, stated that if the (upper) option curve smooth pastes
the NPV of alternative A
2
at P*
2
, the optimal rule shall be "wait if P < P*
2
, and exercise the higher
NPV alternative in the opposite case". However, Dcamps, Mariotti and Villeneuve (2003) disagree
with this conclusion for perpetual options. We also found at least for the finite-lived options case
that the existence of intermediate waiting regions is possible and even common. Figure 2
illustrates this issue, showing an example where intermediate waiting regions appear between the
exercise regions of the alternatives A
1
, A
2
and A
3
.

Figure 2 - Optimal Decision Map: Investing and Waiting Regions
10
The waiting regions are shaded in Figure 2. In this example, the time to expiration is two years. At
expiration (t = T = 2 years), the NPV rule holds and the reader can compare the indifference points
P
0, 1
, P
1, 2
, P
2, 3
, from Figure 2 with the ones shown in Figure 1. In Figure 2 at t = 0, the decision rule
is
13
: wait if P (0, 22); invest in alternative A
2
if P [22, 27.6]; wait again if P (27.6, 31); and
invest in alternative A
3
if P [31,). So, the decision rule is not to invest in the higher NPV
alternative if P P*
2
(the threshold to invest in A
2
), as in Dixit (1993a). There is an intermediate
waiting region that in this example at t = 0 is in the range P (27.6, 31). The existence of
intermediate waiting regions for a given t < T will depend on the parameters, in particular the
volatility - higher volatility can make it optimal for us to wait for the higher scale alternative A
3
, the
intermediate waiting and investing regions disappearing.
By observing the asymptotic behavior of this intermediate waiting region in Figure 2, we can
conjecture that the existence of this region is possible in the perpetual option case as well.
Dcamps, Mariotti and Villeneuve (2003) worked independently but with perpetual options and
proved mathematically that the waiting region around the indifference point between NPV
2
and
NPV
3
always occurs (non-empty region) if it is sometimes optimal to invest in alternative 2. They
showed that if "investment in the smaller scale project is sometimes optimal, ... the optimal
investment region is dichotomous.". In their proposition 3.2, they show that the indifference point
where NPV
2
= NPV
3
(i.e., P
2, 3
) does not belong to the exercise region. They proved that
proposition with the help of the concept of local time for the continuous semimartingale P applied
at this indifference point. This is true even without the option to switch analyzed in that paper.
Probably this conclusion from Dcamps et al (2003) is even more general, being also true for finite-
lived options except at the expiration, as our numerical simulations indicate for geometric Brownian

13
In case of indifference between wait or invest, we assume invest.
11
motion. However, for the mean-reversion case this can be not true due the discontinuity of the
thresholds curve for the high-scale cases (see our section 4.2).
The existence of this intermediate waiting region led to an apparently surprising result. In Figure 2,
at t = 0, if the oil price is US$ 30/bbl, the wait and see policy is optimal, but as the price drops to
US$ 27/bbl, it is optimal to invest in alternative A
2
. The intuition in this case is as follows: if the
price P is US$ 30/bbl, the hopefulness to invest optimally in the larger scale alternative A
3
is
sufficiently high (because P is close to the threshold P*
3
) to offset the value of exercising the "deep-
in-the-money" alternative A
2
. However, if the price drops to US$ 27/bbl, the probability for the
price hitting the threshold P*
3
before expiration is not high enough to justify the delay in the option
exercise of alternative A
2
.
Note also in Figure 2 that as time passes, e.g. at t = 1.5 years, a region appears of immediate
exercise of alternative A
1
. Together with this new exercise region appears a new intermediate
waiting region. We found that when a new optimal investment region appears, it always divides the
previous waiting region into two waiting sub-regions, so the exercise and waiting regions appear
together as time goes by. If the oil price follows a geometric Brownian motion, the emergence of
new exercise and new waiting regions together is a general result as time approaches expiration for
non-dominated scale alternatives
14
. This occurs because the thresholds curves are continuous (no
jumps) and at expiration there are three exercise regions (see Figure 1), which comprise the limit
(boundary) condition at expiration (t = T). We will see that for the mean-reversion model for oil
prices, due to the possibility of discontinuity in the threshold curves at expiration, some
intermediate waiting regions can never occur.

14
In our model, a higher development cost alternative j is non-dominated if it has higher economic quality q
j
than all
the lower development cost alternatives.
12
Let us define P*
k
and P**
k
, k < k
max
, as the lower and upper oil price levels (so that P*
k
< P**
k
) that
define the optimal exercise region for alternative A
k
during the period this region exists. For the
higher scale alternative k
max
, there is always P*
kmax
, but never a finite value of P**
kmax
. For the
other alternatives, P*
k
and P**
k
can exist only for time t very near expiration. At expiration t = T,
P*
k
collapses to P
k 1, k
and P**
k
collapses to P
k , k + 1
.
The existence of waiting regions reflects the real option value of the additional opportunity to invest
in a higher scale alternative. Figure 3 below illustrates this point by showing the option value F for t
= 0 and the payoffs (NPVs) for immediate exercise in each alternative.

Figure 3 - Option Value at t = 0 and Exercise Payoffs
Note that like in the Figure 2, Figure 3 shows that for t = 0 the threshold P*
1
does not exist (the
option value F is higher than NPV
1
P), there is an exercise region for alternative A
2
(between P
2
*
and P
2
**), and there is an intermediate waiting region between P
2
** and P
3
*. Dcamps, Mariotti
and Villeneuve (2003) showed for perpetual options that this region always exists (if sometimes it
13
is optimal to exercise alternative 2, as occurs in the figure). Numerically, the error can be small if
exercising the option (higher NPV as suggested by Dixit in 1993) rather than waiting in this
intermediate optimal waiting region. So, even not being strictly correct, the error from using Dixit's
insight can be small in practice. However, there can be a large range to the optimal waiting interval
(between P
2
** and P
3
*).
Another insight from Figure 3 is that both the value-matching and smooth-pasting properties exist
at the optimal investment threshold not only for P
k
* but for the upper threshold P
k
** as well, if they
exist. For example, at P
2
** in Figure 3 the option curve F has the same slope as the straight payoff
line NPV
2
. This is not surprising because the case of waiting with P > P
2
** and exercising when P
drops to P = P
2
**, is like an option to abandon the opportunity to invest in the higher scale
alternative A
3
, by taking the irreversible decision to invest in A
2
. The smooth-pasting and value-
matching properties at optimal thresholds occur in practically every real options problem see
Dixit & Pindyck (1994) and Dixit (1993b).
In addition to be optimal "wait and see" policy in this intermediate waiting region, the option curve
value in this region can also indicate something about an optimal project scale. Imagine that the
feasibility study team is willing to study another scale alternative but is looking for an immediate
optimal investment. At t = 0, if P belongs to this intermediate waiting region, what is the optimal
scale project for an immediate investment exercise? It must have a quality q* equal to the derivative
of this option value, that is, q* must be equal to the derivative of the option value F(P, t) with
respect to P. This conclusion is derived from the smooth-pasting property at the optimal option
exercise. This project does not exist in the figure above, but the optimal scale in this case is an
intermediate scale between alternatives 2 and 3 with a payoff curve slope (given by the economic
quality q*) so that q
2
< q* < q
3
. So, there is an interesting practical link between the concept of
quality q and the smooth-pasting property for the optimal scale option exercise.
14
The solution shown in the Figures 2 and 3 must be found numerically, for example by using a
binomial approach and working backwards. In this setting we compare the option value of waiting
with the payoff from exercising the higher NPV alternative for every time t and every oil-price
value in the binomial tree. We will work with the partial differential equation approach for the
option value as function of the stochastic oil price and with the appropriate boundary conditions.
The discussion above facilitates understanding of the boundary conditions.
3. Geometric Brownian Motion, Mean-Reversion and the Differential Equations
3.1) Geometric Brownian Motion
By knowing the current oil price and the stochastic differential equation that governs its future
evolution we can determine both the optimal decision rule and the option value of the undeveloped
oilfield (the option to develop). The real option value is conditional on the optimal decision rule.
The optimal decision rule considers two control variables, the optimal timing to develop the oilfield
and the optimal production scale.
In addition to oil prices (P), the other state variable is the time (t), because we have here a finite-
lived real option, with t [0, T]. After the expiration time T, in case of non-development, the
oilfield returns to the government agency and is worth nothing.
Let F(P,t) be the real option value to develop and let the oil price (P) be given by the following
stochastic differential equation, known as geometric Brownian motion:
dz dt
P
dP
+ = (4)
where, is the drift of the process, is the volatility parameter and dz is the Wiener increment
defined as: N(0,1) t d = dz , with N(0,1) being the standard normal distribution.
15
Lemma: Let F(P, t) be the real option to develop the oilfield by choosing timing and scale from a
set of n investment alternatives, k = 0, 1, n, and NPV
k
(P) the net present value by exercising the
alternative of scale k (being NPV(k = 0) = 0). The following inequality holds:
F(P, t) NPV
k
(P) k {0, 1, n}, t [0, T] (5)
This lemma is derived by construction, given that the oil firm has the freedom to choose the
development timing (constrained by the legal limit T) and the scale from a set of alternatives found
in the feasibility study. Because the firm can exercise the higher NPV alternative, the option value
cannot be lower than the NPV of any alternative. This lemma will help in the boundary conditions
to define the existence of investment thresholds (see below).
Assuming that the oil market is sufficiently complete and that there are no arbitrage opportunities in
equilibrium, by using Ito's Lemma and contingent claims approach it can be shown that
15
the option
value follows the following partial differential equation (PDE), where r is the risk-free interest rate
and is the convenience yield of the commodity:
F r F F P ) r ( F P
2
1
t P PP
2 2
= + + (6)
where the subscripts (P, PP, and t) denote partial derivatives. Let the scale alternatives be ordered,
i.e., D
1
< D
2
<< D
n
. Eq. (6) is subject to the following boundary conditions Eq.(7 12):
F(0, t) = 0 (7)
F(P, T) = Max[NPV
k
(P)], for all k = 0, 1, n (8)
F(P
k
*(t), t) = NPV
k
(P
k
*, t), for t < T and for all k which exist P
k
*(t) (9)

15
See Black and Scholes (1973) or Merton (1973) for a financial option pricing methodology and Dixit and Pindyck
(1994) for a real options approach using contingent claims. The partial differential equation is the same as that of
Black-Scholes-Merton, the differences are in the boundary conditions.
16
| | | |
B .
k
q
P
) t , P ( NPV

P
) t ), t ( P ( F
*
k k
*
k
=

, for t < T and for all k which exist P

k
*(t) (10)
F(P
k
**(t), t) = NPV
k
(P
k
**, t), for t < T and for all k < n which exist P
k
**(t)
(11)
| | | |
B .
k
q
P
) t , P ( NPV

P
) t , (t) P ( F
* *
k k
* *
k
=

, for t < T and for all k which exist P

k
**(t)
(12)
Eq.(7) is usual in option pricing and says that P = 0 is an absorbing barrier so that the option value
is worthless for P = 0. Eq.(8) is the option expiration condition, when the option value is either to
commit to the investment for the higher positive NPV given by the alternative of development k 1
or return the oilfield (k = 0) earning a zero option value otherwise. Eq.(9) and Eq.(10) are the value-
matching and smooth-pasting conditions for the lower threshold P
k
* if this threshold exists for a
given t < T. These conditions set the continuity of the option value and its derivative at the optimal
price P
k
*. Recall our discussion in the previous section showing that depending on the problem
parameters, the threshold to exercise high-scale alternatives may not exist for a given t, e.g., in
Figures 2 and 3, at t = 0, P
1
* does not exist. The value matching condition is equivalent to the
lemma for the equality case, so that the equality case in the lemma sets the existence or not of this
lower threshold. Eq.(11) and Eq.(12) are also value-matching and smooth-pasting conditions, but
for the upper threshold P
k
** if this threshold exists for a given t < T. Note that for k = n, this
threshold does not exist, but P
k
** can exist if it the associated P
k
* exists for k < n. Again, see the
previous section for the intuition behind this point. The four boundary conditions (Eqs. 8-12)
assume by definition that P*
k
< P**
k
when they exist (see the previous section).
This option-pricing problem described by Eq (6) and its boundary conditions is a free-boundary
problem of optimal stopping time, usual in option pricing theory. The option was solved
17
numerically by applying the finite difference methods in the explicit form
16
with an optimization
procedure as shown in Appendix A.
Note that for a number n of alternatives for oilfield development, there are at most (2n 1)
threshold curves for the oil price to calculate. Each region between these threshold curves P
k
*(t)
and P
k
**(t) for the time interval where these regions exist, corresponds to one optimal alternative k
for development during the options lifetime.
Figure 4 shows another example of the investment and waiting regions, considering that the oil
price follows a geometric Brownian motion for development rights that expire in 2 years and for n
= 3 alternatives (A
1
, A
2
, or A
3
) of development scale in increasing order.

Figure 4 - Investment and Waiting Regions: Geometric Brownian Motion
When compared with Figure 2, the example in Figure 4 shows for t = 0 a case where we either
exercise the higher scale alternative A
3
(if P 33.5 $/bbl) or "wait and see" otherwise. There are no
intermediate exercise or waiting regions due to the higher volatility used in this example (25% p.a.).

16
For an application of the finite difference methods to option pricing, see Brennan and Schwartz (1978). More details
about the methodology can be found in Ames (1977) or Smith (1971).
18
Since we are working numerically with three alternatives, there are at most five optimal threshold
curves to determine, performing three areas of option exercise. Again, this maximum number of
threshold curves occurs only near of the expiration.
Figure 5 shows the option value at the current (t = 0) and at the expiration time (t = 2 years), for the
numerical parameters used in Figure 4 and presented in Section 4 below. Note again that at
expiration, we have three different areas for exercise, each one representing a certain development
alternative for the oilfield. The option at current time is only exercised for alternative A
3
with oil
prices equal to or above US$33.50/bbl, since the intermediate waiting and investment regions do
not exist for t = 0.

Figure 5 - Option Value of the Undeveloped Oilfield: Geometric Brownian Motion
19
3.2) Mean Reversion Model
Consider the mean reversion hypothesis, frequently used for commodities, where oil prices (P)
evolve as the following stochastic process, known as inhomogeneous geometric Brownian motion,
or as Battacharyas (1978) mean-reverting process
17
:
Pdz dt ) P P ( dP + = (13)
where, is the reversion speed of the process, is the volatility parameter, P is the long-run
equilibrium mean and dz is the Wiener increment defined as before: N(0,1) , t d dz = .
Following the same contingent claims approach described in the previous section, it can be shown
that
18
the option value F(P,t) follows the partial differential equation shown below, where r is the
risk-free interest rate and is the risk-adjusted discount rate for the underlying (oil price) variable:
rF F PF
P
) P P (
r F P
2
1
t P PP
2 2
= +
(

|
|
.
|

\
|
+

(14)
Again, the subscripts (P, PP, and t) denote partial derivatives. Eq. (14) is subject to the same six
boundary conditions (Eqs.7-12) described in the previous sub-section.
Note that for mean-reverting processes, the convenience yield of the commodity is not constant, it
is a function of the oil price:
(P) =
|
|
.
|

\
|

P
) P P (
(15)
This non-constant convenience yield is a usual characteristic of mean reversion processes. The
parameter is endogenous in our model, and from a market point of view, is used in the sense

17
This equation corresponds to Eq.(4.9) in Kloeden & Platen (1992, p.119). See also an interesting discussion of this
process and comparison with other mean-reverting processes in Robel (2001).
18
See Dixit and Pindyck (1994) chapters 5 and 12 for similar examples of geometric mean reverting process.
20
described by Schwartz (1997, p.2): In practice, the convenience yield is the adjustment needed in
the drift of the spot price process to properly price existing futures prices. Note also that for the
mean-reversion model, the risk-adjusted discount rate appears, even assuming complete markets.
This feature does not occur with geometric Brownian motion because the convenience yield is
constant in that model. Here the convenience yield is a function of the risk-adjusted discount rate
for oil price risk.
See also Dias & Rocha (1999) for a discussion of the value of the convenience yield in mean-
reverting models highlighting the possibility of this parameters becoming negative for low oil
prices. Eq.(15) shows that (P) < 0 is possible for low values of P. In this case, there is the
possibility of discontinuity in the threshold curves at expiration because from the literature of
American options we know that the earlier option exercise is never optimal when the convenience
yield is negative or zero. We will address this question in Topic 4.2.
Figure 6 shows the option value at the current (t = 0) and at the expiration time, using volatility of
25% p.a. and for the numerical parameters for the mean reversion process presented in Section 4
below. Note again that at expiration we have three different exercise areas, each representing a
certain development alternative for the oilfield. The option at current time is exercised for the
alternative A
2
if the oil price is between US$ 22.90 and 28.30/bbl, there is a thin intermediate
waiting region between US$ 28.30 and 29.90/bbl, and for P US$ 29.90/bbl, it is optimal to
exercise scale alternative A
3
.

21

Figure 6 - Option Value of the Undeveloped Oilfield: Mean Reversion Model
4. Numerical Simulations and the Effect of Volatility
4.1) Geometric Brownian Motion
Consider three oilfield development alternatives, A
1
, A
2
and A
3
. The investment opportunity to
develop the oilfield expires in T = 2 years and the manager has to decide which is the optimal
timing for development as well the optimal production scale that maximize the investment option
value.
Let the following be the parameters for the base case
19
(MM = million): q
1
= 0.08, q
2
= 0.16, q
3
=
0.22, D
1
= US$ 400 MM, D
2
= US$ 1000 MM, D
3
= US$ 1700 MM, B = 400 MM bbl, r = 8% p.a.,
= 8% p.a., = 25% p.a.
20
and P
0
= US$ 20/bbl. With these parameters, the NPVs for immediate
exercise are NPV
1
= US$ 240 MM; NPV
2
= US$ 280 MM; and NPV
3
= US$ 60 MM.

19
Some values were estimated using available data about oil prices or using available related literature.
20
Based on the volatility estimation of Dias and Rocha (1999).
22
Table 1 shows the managerial flexibility value for having a certain number of development
alternatives. Note that the option value increases with the number of available alternatives,
reflecting the option value of additional managerial freedom in choosing the investment scale.
However, the rate of additional value is decreasing with the number of alternatives, so that in
practice after 3 alternatives the additional value may not compensate the time spent by the
feasibility study team in detailing additional scale alternatives.
Table 1: Managerial Flexibility Value .
Development Alternatives : Production Scale Option Value (US$ MM)
k = 2 : Medium Scale 310.98
k = 1 or 2 : Small or Medium Scale 322.65
k = 1 or 2 or 3: Small, Medium or Large Scale 323.33

Note also that these values are higher than the highest NPV alternative (NPV
2
= US$ 280 MM).
Figure 7 compares the oil price threshold curves with a change in the volatility parameters (15%
p.a. and 25% p.a.). A lower volatility means lower probability of changes in oil price evolution. In
this situation the option is exercised sooner (note how the exercise area for the development
alternatives becomes higher) because the option of waiting become less valuable. At t = 0, the
intermediate waiting and exercise regions appear only in the low volatility case. These regions
appear even for very high volatility values, but only at times near or very near expiration.
23

Figure 7 - Geometric Brownian Motion: The Effect of Volatility on the Thresholds
Table 2 shows the real option values at t = 0 and the optimal investment rule for different
volatilities (including an intermediate volatility not shown in Figure 7) and different current oil
prices.
Table 2: Option to Develop Values (US$ MM) and Optimal Action for Geometric Brownian Motion
Current Oil Price (US$/bbl)
Volatility (% p.a.) 15 25 30
15 85.89 wait 600 exercise A
2
942.21 wait
20 102.55 wait 600 exercise A
2
948.65 wait
25 122.29 wait 605.21 wait 958.72 wait

4.2) Mean-Reversion Process
We use the base case parameters from the previous section (except ) plus the specific parameters
of the mean reverting process: = 12% p.a., = 0.3466
21
, P = US$20/bbl
22
.

21
This reversion speed implies a half-life of about 2 years. Bradley (1998, p.59) finds a half-life of 1.39 years.
24
Before showing the threshold chart, it is important to calculate the break-even prices (for which
NPV
k
= 0), and the oil price that makes the convenience yield equal to zero, due to the possibility of
discontinuity in the threshold curves at expiration, as explained below.
The break-even prices for the three alternatives are: P
be1
= US$ 12.50/bbl; P
be2
= US$ 15.635/bbl;
and P
be3
= US$ 19.32/bbl, calculated with the equation P
be(k)
= D
k
/(q
k
B).
By using the Eq.(15) and the base-case parameters, it is easy to find that the oil price that makes the
convenience yield equal to zero is P = US$ 14.86/bbl. So, for oil prices equal to or lower than US$
14.86/bbl, it is never optimal to exercise the option before expiration. However, exactly at
expiration, if the oil price is in the range 12.50 P 14.86, even with 0, it is optimal to
exercise alternative A
1
because we have a non-negative NPV
1
and so exercising the option
increases the shareholder value if P > 12.5 in this numerical example. This means that there is a
discontinuity in the threshold curve at least for alternative A
1
: the threshold to invest in this
alternative at expiration is US$ 12.5/bbl, but at an infinitesimal time before the threshold for this
alternative jumps to a much higher value in order to avoid exercise with negative, zero or too low
values for the oil convenience yield (P).
In the presence of other scale alternatives, this individual behavior of alternative A
1
can make both
the intermediate exercise and waiting regions vanish even very shortly before expiration. This will
occur in our example for alternative A
1
.
Figure 8 compares the threshold curves for oil price and volatility effect (25% p.a. e 45% p.a.).
Note that the region for investment in alternative A
1
no longer exists except at expiration, even for a

22
Baker et al (1998, p.129) estimate the long run oil price as $18.86/bbl (in 1995 dollars) and used (pp.138-140)
US$20/bbl as the initial long run level in their model. This value is also adopted in Bradley (1998, pp.59-61) and
shown in Cortazar & Schwartz (1996, Figure 4). Econometric tests including more recent data point to a long-run level
higher than US$ 20/bbl.
25
time very near expiration. This is caused by the discontinuity of the threshold curve for this
alternative at expiration, as explained in the previous paragraphs. For alternative A
1
, the oil prices
region for optimal exercise at expiration implies either a negative convenience yield or values very
near zero. So, for alternative A
1
alone, a very small time before expiration requires a threshold
much higher than the upper oil price value that makes this alternative optimal at expiration.
However, at expiration, it is optimal to exercise alternative A
1
for a range of prices of 12.50 P <
18.75, shown as a vertical line at t = T.

Figure 8 - Investment and Waiting Regions for the Mean-Reversion Model
This issue of threshold curve discontinuity never occurs with geometric Brownian motion because
the convenience yield is constant (and positive) in that model. So, intermediate exercise and waiting
regions always occur at least for times very near expiration in that model, but the same is not true in
the mean-reversion model. Note also in Figure 8 that for the base-case volatility of 25% p.a., at t =
0, there is a thin intermediate waiting region between US$ 28.30 and 29.90/bbl and a region for
exercising alternative A
2
between US$ 22.90 and 28.30/bbl. Compare these regions from Figure 8
with the option chart in Figure 6.
26
The existence of these intermediate waiting regions between the optimal exercise of alternatives 2
and 3 (for = 25% at t = 0 and for = 45% at t > 1.2 in Figure 8, and in the other simulations we
performed) indicates that the conclusion is more general valid for other stochastic processes that
these waiting regions always occur if it is sometimes optimal to invest in alternative 2 (intermediate
scale alternative). This is a strong conjecture demonstrated only for the geometric Brownian motion
and perpetual options by Dcamps et al. (2003).
The real option to develop, considering all three scales options, is worth US$ 313.86 MM if the
current oil price is US$ 20/bbl. Table 3 shows the option values at t = 0 and the optimal investment
rule for different volatilities and different oil prices.
Table 3: Option Value (US$ MM) and Optimal Action for Mean-Reverting Process
Current Oil Price (US$/bbl)
Volatility (% p.a.) 15 25 30
15 126.21 wait 600 exercise A
2
940 exercise A
3

20 140.92 wait 600 exercise A
2
940 exercise A
3

25 158.45 wait 600 exercise A
2
940 exercise A
3


Comparing Tables 2 and 3, we can see the differences in the option values due to the hypothesis
about the oil price evolution. Options values for mean-reversion are higher for low initial prices
(lower than P ) and slightly lower for high initial oil prices.
Note in Table 3 that the option under the mean-reverting hypothesis is exercised immediately when
the current oil price is above the long-run equilibrium mean (US$20/bbl). This is due to the
expectation of the mean-reverting process, leading to a small probability for oil prices to be too far
from the mean
23
, decreasing the value of waiting for a higher scale alternative that has a threshold

23
The mean-reverting process is a stationary process with a bounded variance. The reversion speed is the force that
pulls back the oil price towards the long-run equilibrium mean. As far is the oil price from the equilibrium as higher is
the reversion force, like a spring force.
27
very far from the long-run equilibrium price to which the prices are attracted. Immediate exercise,
in this case, maximizes the option value.
However, for oil prices lower than the long-run equilibrium mean, the option is not exercised even
for the lower volatility case. In this case, the expectation that prices will revert to the mean
increases the option to wait. These results vary according to the parameters of the model (speed of
reversion, current price, long-run mean, volatility, and time to expiration) and should be taken with
caution. For example, at t = 0, Figure 8 shows that for the very high volatility case of 45% p.a.,
waiting is the optimal policy even for prices of US$ 25 or 30/bbl.
5. Concluding Remarks
This paper analyses the investment opportunity of developing an oilfield. It calculates the
investment option value and also the optimal developing timing and production scale, considering
oil price uncertainties and applying the real options approach.
The investment opportunity in the oilfield is analogous to an American call option on several assets
but with the same underlying oil price stochastic process and with different exercise prices and
different payoffs (different oil price functions) from the different scale alternatives. The payoff
(NPV) of exercising a specific alternative k is the value of the developed reserve using this scale
alternative minus the development cost of the alternative. The economic quality of the reserve
parameter distinguishes the scale alternatives on the benefit side.
The presence of managerial flexibilities, i.e., the options to choose when to develop the oilfield and
to set the optimal production scale, increases the investment option value.
We determine the oil price threshold curves for the exercising options. Instead of one threshold
curve as in standard (financial or real) American call options, we have regions for option exercise
28
and waiting regions. The existence of intermediate waiting regions has not been addressed before in
the real options literature and can occur with both stochastic processes that we analyzed.
According to oil price volatility and the stochastic process for the oil price, the area for exercising
the option can degenerate and the option to wait can dominate immediate development. An increase
(decrease) in the volatility parameter decreases (increases) the area for exercising the option for
both stochastic processes: geometric Brownian motion and mean reversion. This is due to the fact
that the option to wait for better conditions to commit to the investment (waiting for a higher scale)
is higher (lower) in such cases.
Finally, the mean-reverting process - frequently used in commodities modeling - presents different
results than geometric Brownian motion. Specific production scales can never be optimal before
expiration. This is caused by discontinuity of the threshold curves at expiration if the break-even oil
prices for these alternatives imply a negative or zero convenience yield. For geometric Brownian
motion, this issue never occurs because the convenience yield is constant (independent of the oil
price) and commonly assumed to be positive.

Acknowledgements: We would like to thank Petrobras for the financial support of this PUC-
Petrobras project, through the systemic research project named Pravap-14. We thank Thomas
Mariotti for the thorough discussions. Any error is ours.
29
Bibliographical References
1. Ames, W.F. (1977). Numerical Methods for Partial Differential Equations. Academic Press.
New York.
2. Baker, M.P.; Mayfield, E.S; Parsons, J.E. (1998): Alternative Models of Uncertain Commodity
Prices for Use with Modern Asset Pricing. The Energy Journal 19(1):115-148.
3. Batthacharya, S. (1978). Project Valuation with Mean-Reverting Cash Flow Streams. Journal
of Finance 33:1317-1331.
4. Black, F., and M. Scholes. (1973). The Pricing of Options and Corporate Liabilities. Journal of
Political Economy 81:637-659.
5. Bradley, P.G. (1998). On the Use of Modern Asset Pricing Theory for Comparing Alternative
Royalty Systems for Petroleum Development Projects. The Energy Journal 19(1):47-81.
6. Brennan, M.; Schwartz, E. (1978): Finite Difference Methods and Jump Processes Arising in
the Price of Contingent Claims: a Synthesis. Journal of Financial and Quantitative Analysis 13:
461-474
7. Brennan, M.; Schwartz, E. (1985). Evaluating Natural Resources Investments. Journal of
Business 58:135-157.
8. Brodie, M.; J. Detemple (1997): The Valuation of American Options on Multiple Assets.
Mathematical Finance 7(3):241-286
9. Capozza, D.R.; Y. Li (1994): The Intensity and Timing of Investment: The Case of Land.
American Economic Review 84(4):889-904
10.Childs, P.D.; S.H. Ott; A.J. Triantis (1998): Capital Budgeting for Interrelated Projects: A Real
Options Approach. Journal of Financial and Quantitative Analysis 33(3):305-335
30
11.Cortazar, G.; Schwartz, E. (1996): Implementing a Real Options Model for Valuing an
Undeveloped Oilfield. Working Paper, Catholic University of Chile, 16 pp.
12.Dcamps, J-P.; T. Mariotti; S. Villeneuve (2003): "Irreversible Investment in Alternative
Projects". Working Paper, Universit de Toulouse and London School of Economics and
Political Science, June.
13.Dias, M.A.G. (1998): "Investment under Uncertainty in Exploration & Production - Update and
Overview of the M.Sc. Thesis". Workshop on Real Options, Stavanger, May. See the discussion
on economic quality of a reserve at http://www.puc-rio.br/marco.ind/quality.html
14.Dias, M.A.G. (2001): Valuation of Exploration & Production Assets: An Overview of Real
Options Models. Forthcoming in Journal of Petroleum Science and Engineering.
15.Dias, M.A.G.; Rocha, K. (1999). Petroleum Concessions with Extendible Options: Investment
Timing and Value using Mean Reversion with Jumps to Model Oil Prices. Working Paper,
Petrobras and IPEA, at: http://www.puc-rio.br/marco.ind/extend.html
16.Dixit, A. (1993a). Choosing Among Alternatives Discrete Investment Projects Under
Uncertainty Economic Letters 41:265-268.
17.Dixit, A.K. (1993b): The Art of Smooth Pasting. Fundamentals of Pure and Applied
Economics. Harwood Academic Publishers.
18.Dixit, A.; Pindyck, R. (1994). Investment under Uncertainty. Princeton University Press.
19.Geltner, D.; Riddiough, T.; Stojanovic, S. (1996). Insights on the Effect of Land Use Choice:
The Perpetual Option on the Best of Two Underlying Assets. Journal of Urban Economics
39:20-50
20. Graham, J.R.; C.R. Harvey (2001): "The Theory and Practice of Corporate Finance: Evidence
from the Field". Journal of Financial Economics 60:187-243
31
21.Gruy, H.J.; Grab, F.A.; Wood, J.W. (1982). Determining the Value of Oil and Gas in the
Ground. Word Oil, CXCIV:105-108.
22.Kloeden, P.E.; E. Platen (1992): Numerical Solution of Stochastic Differential Equations.
Springer-Verlag Berlin Heidelberg.
23.Majd, S.; Pindyck, R. (1987). Time to Build, Option Value, and Investment Decisions. Journal
of Financial Economics 18:7-27.
24.Merton, R.(1973). Theory of rational option pricing. Bell Journal of Economics and
Management Science 4 (4):141-183.
25.Paddock, J.; Siegel, D.; Smith, J. (1988).Option Valuation of Claims on Real Assets: The Case
of Offshore Petroleum Leases. Quarterly Journal of Economics 103:479-508.
26.Pindyck, R.S. (1988): Irreversible Investment, Capacity Choice, and Value of the Firm.
American Economic Review 78(5):969-985.
27.Pindyck, R.S. (1993): Investments of Uncertain Cost. Journal of Financial Economics 34:53-
76.
28.Robel, G.F. (2001): "Real Options and Mean-Reverting Prices". 5
th
Annual International
Conference on Real Options, UCLA, Los Angeles, July. Available at www.realoptions.org.
29.Rocha, K.M.C. (1996): "Decises de Investimentos Sequenciais em Regime de Incerteza"
(Sequential Investment Decisions in the Presence of Uncertainty). M.Sc. Dissertation, Dept. of
Industrial Engineering, Catholic University of Rio de Janeiro.
30.Smith, G.D. (1971). Numerical Solution of Partial Differential Equations. Oxford
Mathematical Handbooks. Oxford University Press.
31.Schwartz, E.S. (1997): Valuing Long Term Commodity Assets Working Paper, 7-97, UCLA,
23 pp.
32
32.Schwartz, E.; Trigeorgis, L. (2001). Real Options and Investment Under Uncertainty. Classical
Readings and Recent Contributions. MIT Press.
33. Tourinho, O. (1979). The Option Value of Reserves of Natural Resources. Doctoral
Dissertation, University of California at Berkeley.
34.Trigeorgis, L. (1996) Real Options: Managerial Flexibility and Strategy in Resource Allocation;
MIT Press.

33
Appendix A: Finite Difference Method in Explicit Form for Numerically Solving the Partial
Diferential Equation (PDE)
The finite difference method transforms the partial differential equation, Eq. (6) in the text, and its
respective boundary conditions into a difference equation that can be solved numerically.
By using a specific discretization mesh (time step, and oil price step), the explicit form converges to
the exact solution of Eq. (6) and it is easier and faster (especially for a low number of state
variables) than the implicit forms or the Monte Carlo simulations techniques associated with
optimization procedures.
Regarding the free-boundary problems defined in Eq. (6), the explicit form and the discretization
mesh can easily handle the optimization algorithm used to solve those optimal stopping time
problems, like the backward induction style of a stochastic dynamic programming approach.
Implicit forms, however, have to deal with a simultaneous system of equations together with the
optimization procedure.
Numerical solutions for partial differential equations can be found in Ames (1977) or Smith (1971).
Dixit and Pindyck (1994), Chapter 10, applies the same procedure (explicit form together with an
optimization algorithm) to solve an option-pricing problem about sequential investment.
Let F(P,t) at point (P,t) be represented by F
i,j
, where P = iP for i (0,m) and t = jt for j (0,n)
Assume the following partial derivative approximations:
F
PP
[F
i+1,j+1
- 2F
i,j+1
+ F
i-1,j+1
] / (P)
2
; F
P
[F
i+1,j+1
- F
i-1,j+1
] / 2P; F
t
[F
i,j+1
F
i,j
] / t (A1)
We use the central difference approximation to variable price (P), and forward difference
approximation to variable time (t). Applying these expressions in Eq.(6), we have the following
difference-equation:
1 j , i
0
i 1 j , 1 i i 1 j , 1 i i j , i
F p F p F p F
+ +

+ +
+
+ + = (A2)
34
t
1
r
2
i ) r (
i
2
1
p
2 2
P
i

+
=
+

t
1
r
2
i ) r (
i
2
1
p
2 2
P
i

t
1
r
t
1
i
p
2 2
P
0
i

+
+
= (A3)
We can apply the same procedure to the boundary conditions of Eq. (6). It can be shown
24
that the
solution to equation (A2) converges to the solution of Eq.(6) if all the p
i
coefficients in Eq.(A3)
are non-negative numbers. Therefore, we have to choose a discretization time-step, and a price-step
in order to guarantee that condition.

24
See the theorem in Ames (1977), page 65.
35