a real-option approach for portfolio management of upstream hydrocarbon...
TRANSCRIPT
46
A Real-Option Approach for Portfolio Management of Upstream Hydrocarbon Assets
Bacel Maddah, Ali Yassine, Imad G. Melki
Engineering Management Program, American University of Beirut, P.O. Box 11-0236, Riad El Solh, Beirut 1107 2020, Lebanon
Abstract. We develop an optimization-based decision support system (DSS) that assists oil companies in upstream exploration and production decisions for a portfolio of hydrocarbon assets over time. Specifically, the DSS is based on dynamic programming (DP) and utilizes a probabilistic hydrocarbon price (geometric Brownian motion) and updates the probability distribution of asset reserves across the portfolio based on the outcome of ongoing explorations. The DSS can be seen as valuating a series of two-asset real options, with the underlying assets being the hydrocarbon reserve and price. Given possible realizations of price and reserve, the DSS proposes optimal exploration and development scenarios. The challenge in developing these scenarios is high computer storage. However, we mitigate this by establishing structural properties of the production decisions. Numerical analysis indicates that exploration decisions are greatly affected by the correlation of reserves in different hydrocarbon assets and the volatility of the hydrocarbon price. High correlation between assets leads to successive exploration decisions and high volatility shifts decisions to later periods to benefit from further price fluctuations.
Keywords: hydrocarbon (oil and gas), DSS, dynamic programming, real options, exploration and production decisions
1 Introduction
Oil and gas exploration and production is a risky business with high potential
rewards. The hydrocarbon exploration frontier has moved to deep offshore water where
drilling and production have a high cost and require cutting edge technology and
expertise. Additionally, success in these projects is low and decisions taken incur
millions of dollars of investment and have long-term (e.g., 20- to 30-year) ramifications.
In this high risk / high reward environment, the proper management of portfolio
of O&G assets is essential for an oil company to operate profitably and for a
government to effectively exploit its natural resources. Thus, there is a significant
importance in developing optimization models for planning the oil and gas exploration
and production industry which capture all the complex trade-offs. However, the
46
development of such models is riddled with many complications and challenging factor,
which include,
High uncertainty in critical factors, most notably, the amount of hydrocarbon in
the reserve and the future price,
Complex processes of exploration, development and production stages with
extremely high capital and operating costs and, in many countries, a turmoil of
political and security conditions,
Conflicting interests between the oil company typically baring most of the
exploration and production risk and the government trying to maximize its
benefits from depleting natural resources.
The reserve portfolio effect reflecting the dependency among reserve
uncertainties, in the sense that a commercial discoveries (or lack of) in one asset
affect reserve estimates in other assets.
Coping with uncertainties of price and reserve and price is perhaps the major
challenge in O&G development. On the one hand, prior to exploration drilling the
reserve level is highly uncertain. The existence of such reserves is indicated by
geological formations revealed by seismic studies, which does not guarantee
commercial quantities. After exploration drilling, most of the uncertainty in reserve
is resolved. However, uncertainty persists, since as it is well-known, one cannot
figure out the full reserve potential before the last drop of it is exploited. On the
other hand, complex factors of supply and demand (weather, political conflicts,
wars, technology change, etc.) affect the price of hydrocarbon, which leads to high
volatility and continuous fluctuations. Coupled with high costs that must be covered
by abundant revenues, price volatility has an amplified effect.
Therefore, it is important to capture the effect of high uncertainties and the
complex development factors in a decision support system (that assists in
exploration and production decisions) in upstream O&G. In this paper, we develop
such DSS that attempts to capture effectively the relevant factors. Our DSS is based
on stochastic dynamic programming which stipulates appropriate exploration,
development and production decisions for all possible future scenarios of price and
reserve. It has the following distinctive features:
46
Capturing the uncertainty in reserve through appropriate probabilities
distribution, with correlation factors adjustable based on exploration outcome.
This allows learning from sequential drilling.
Capturing the random price fluctuation over time via an appropriate, geometric
Brownian motion model. This price model, common in financial engineering, is
well-known to account for complex forces of supply and demand (e.g.
Lunenberger 1998).
Adopting a reasonable hydrocarbon cost structure with high capital (capex) and
appropriate operational costs (opex) and with realistic factors such as the
deliverability which restricts production in a given period.
Accounting for the conflicting interest of government and oilcompany by
throwing-in the parameters of production sharing contracts in the decision mix.
The work closest to our proposed research is Skaff (1999) who also considers
various portfolio decisions over time. However our proposed model includes more
features such as probabilistic hydrocarbon price and learning from sequential
drilling, among other things. The sequential drilling is a generalization of Bickel and
Smith (2006) from updating binary (dry/wet) reserve expectations to multiple-level
(e.g., dry, low, medium, high) expectations. This generalization is a useful side
contribution of this paper.
Armed with our multi-faceted DP-based DSS, we perform useful analysis. First,
we demonstrate structural properties of the production decisions, the most important of
which is showing that these decisions are “binary”, in the sense that at a given point in
time either produces the full possible reserve, or produces nothing. This allows reducing
the computer storage requirement and handling large-scale industry-size problems for a
portfolio having a handful of assets. Second, we perform a detailed sensitivity analysis
on the effect of price and reserve uncertainty on the exploration and production
decisions. We observe that high correlation between asset reserves has an important
effect on exploration decisions. In addition, high price volatility shifts decisions to later
in order to benefit from price fluctuations. This is in-line with results from financial
engineering indicating that the value of an option is increasing with its maturity. Here
we have real options and postponing decisions extends maturity.
46
The remainder of this paper is organized as follows. Section 2 surveys the related
literature. Section 3 addressed estimation of DSS input parameters. Section 4 presents
the core DSS dynamic programming model. Section 5 presents analysis on the structural
properties of production decisions and computational complexity. Section 6 presents
extensive numerical results and enlightening managerial insights on the effect of price
and reserve uncertainty on the E&P process. Finally, Section 7 concludes the paper and
presents ideas for future research.
2 Literature review
Managing an exploration and production (E&P) portfolio of hydrocarbon
involves many procedures and many decisions that will affect the performance of a
company. The problem is a typical example of companies which have projects with
high return and high risk. The subject was treated by many researchers in different ways
as described below.
Suslick and Schiozer (2004) presented detailed reviews of this literature. They
mention that there are two major techniques used for the evaluation of these projects:
Decision analysis and real options.
Real option evaluation is a technique that utilizes financial options theory and
is usually used for making one time decision (e.g. bid or not, drill or not, lease or not
and to do the valuation of assets). Example of works here includes Lund and Marketing
(1999), Cortazar et al. (2001), and Armstrong et al. (2004). Decision analysis (DA)
utilizes cash flow estimates and optimization theory and typically involves several inter-
related decisions. Our proposed DP model belongs to this category. For this reason, we
focus in the rest of this chapter on reviewing DA models.
Project selection models utilize DA to choose between multiple projects, but
they do not deal with detailed management decisions like exploration and development
and production decisions. Example of works here includes Davidson and Davis (1995),
Motta et al. (2000), Suslick and Furtado (2001), and Keefer (1991).
Mean-Variance models use Markowitz portfolio theory to select the
appropriate combination of E&P projects to reduce the overall variance of the
investments, while meeting a target profit level. Examples of works here include Walls
(2004), and Al-Harthy and Khurana (2008). These papers provide methods to evaluate
46
probability distributions by running a simulation that will calculate the variance of the
profit of the project (intra-dependency), and calculate the covariance between projects
(inter-dependency). Even though the models provide novel tools, but they do not help in
business unit field decisions, in terms of scheduling exploration and production
decisions over time.
Dynamic models optimize results by running through all possible states of the
portfolio and valuate the outcomes of various decisions dynamically over time till a
certain boundary state (e.g. decommissioning) that has a predetermined value. Haugen
(1993) uses a different approach then the one used in our model where the E&P
Company is bounded by a contract with downstream company to deliver a certain
amount of gas. Our model is different because it is linked to real world market where
the price is fluctuating and it accounts for learning and production decisions. Aydin
(2008) treats the development of an oil field with a dynamic optimization model. The
model uses Bayes formula to calculate probabilities, and thus will require a large
amount of probability estimation. Most feature in Aydin’s model are captured in our
research with additional downstream considerations of the production decision. Bickel
and Smith (2006) treat the problem of dependency in the outcome of exploration of a
portfolio of reserves. They developed a model that updates probabilities by calculating
joint probabilities for all prospects. These calculations are based on minimizing the
Kullback-Leibler distance relative to a worse-case independent distribution of
outcomes. We adopt a similar Kullback-Leibler based approach for probability in our
model, and we generalize it from binary outcomes (wet/dry) to accommodate multiple
reserve outcomes (e.g. Dry, Medium, and High).
Mixed integer non-linear programs (MINLP) is a method used to find optimal
results of variables relative to a constrained non-linear equation, and variables may be
real or integer. Examples of works here include Frair and Devine (1975), Van den
Heever et al. (2000), Goel and Grossman (2004), Goel et al. (2006). These programs
develop a detailed optimization of the development phase which is inconvenient for an
early stage of the development of an area. During this stage, there is no precise
information about the underground formation, and in consequence, we can’t build
detailed deterministic optimization model of the development phase. In our model, we
reduce the number of decisions in the development phase and we account for the
46
possible risk that may arise over the lifetime of the project (e.g. price variability,
dependent random reserve outcomes).
3 Input data estimation
In this section, we will detail the inputs to our model. We show that the
required parameters for our model are measurable and can be estimated with a
reasonable effort.
In Section 3.1, we introduce the contract parameters. In Section 3.2, we
estimate the volatility and growth of gas price. In Section 3.3, we detail the costs
included in the E&P business and explain how they are included in the model. In
Section 3.4, we explain how the reserve probability distribution is currently estimated.
3.1 Contract Parameters
Some of the data inputs that are required for our model can be directly taken
from hydrocarbon contracts or laws. These are typically production sharing agreements
(PSAs) between the government and foreign oil companies. In such agreements, the oil
company bears all the risk of exploration production and shares the revenue of
commercial discoveries with the government. For more details see Younes
(2011).These contract parameters are the royalty ratio, the tax ratio, the exploration
license duration, and the production license duration.
3.2 Estimating the Volatility and Growth of the Gas Price
Modeling the fluctuation of the gas price over time is achieved by collecting
data and estimating the volatility and expected growth rate of gas price. These estimates
are then used to develop a binomial lattice price model.
The Black-Sholes equation is used to calculate the implied volatility of natural
gas price. Using two futures options prices above, we solve for the implied volatility
and the implied current futures price. The Black-Sholes equation is as follows
(Luenberger, 1998).
46
⁄ ⁄
√,
√ ,
where
N(x) is the standard normal cumulative density function,
is the current future price, is unknown,
is the strike price, for K = $5and K = $5.5(Two equations),
r is the risk-free rate, r = 0.08% (Source: www.ustreas.gov on June 2010),
T is the maturity time, T = 3 months = 0.25 year,
is the volatility of gas price; is unknown.
In June 2010, the price of futures option with delivery September 2010 and
strike prices K = $5 and K = $5.5 is C = $0.41 and C = $0.248.
The unknowns in the system of two Black-Sholes equations are and . This
gives an implied current futures price of $4.82 and an implied volatility of 50.38% for
Natural Gas.
3.3 Cost Parameter Estimation
In our model, we are assuming that cost parameters are fixed over time. We
were motivated by Goldsmith et al. (2001) in adapting this cost structure. Goldsmith et
al. describe the method used to evaluate the costs of different deep offshore systems
used relative to capital expenditure, operating expenditure, risk expenditure and loss in
production during operation expenditure. Their objective was to decide upon the
optimal system to be installed in a certain field.
Our model has five cost components for each of the assets: exploration cost,
development cost, fixed production cost, unit production cost and abandonment cost,
which are described in the remainder of this section.
The exploration is the expected cost of drilling exploration wells to know if
there is gas in the asset or not. Development cost includes the cost of drilling production
wells, laying network pipes to connect the wells to the gas system, and buying and
installing the subsea system facilities (For deep offshore). Abandonment cost includes
the cost of dismantling the structure and the cost of meeting the environmental
regulation settled by the government. Fixed production cost includes salaries of
46
operating staff, maintenance cost and blowoff risk cost. Unit production cost includes
unit treatment cost and the transportation cost.
3.4 Estimating reserve Volume Probability Distribution
The volume of gas that may be discovered in an asset is uncertain. The gas
volume in a compartment can be estimated using the following equation (Hyne 2001).
43560 ∅
Where:
V= Volume of the oil in acre-feet. An acre-foot of volume can hold 43560 scf of
gas
Ф= The porosity of the reservoir (Decimal)
Sg= Gas saturation (gas volume relative to porosity, because gas is also mixed
with water)
R= Technical recovery factor
Bg= Gas formation volume factor (scf/bbl); Bg=1 for a gas reserve
The gas saturation and the technical recovery factors are uncertain before
exploration. By assessing three-point probability distribution for these factors one can
estimate the probability distribution of reserve levels. The reserve probability
distribution is simulated into discrete reserve outcome in the dynamic model.
4 Model and Assumptions
The dynamic model will require probability many inputs to simulate the
uncertainty in the E&P project. As shown in Figure 1, estimates of the volatility of the
Natural Gas price and its expected growth are needed for the price outcome binomial
lattice and they are estimated in Section 3.2. The estimation of contract parameters and
cost parameters is explained in Section 3.1 and Section 3.3; they will be needed for the
dynamic model. Reserve probability distribution estimation is explained in Section 3.4.
In Section 4.1, we will introduce the price outcome binomial lattice sub-model.
In Section 4.2, we introduce Kullback Leibler sub-model, and estimate joint
probabilities out of reserve probability distributions and joint probability distributions.
These two sub-models will feed into the dynamic model to estimate the value of the
project and make good E&P decisions. In Section 4.3, we introduce the dynamic model.
4.1
mo
oil
a fa
Wh
Fig
hyd
The
Figure 1: F
Price Ou
A bino
tion which i
and gas (e.g
actor of u or
here
∆ is th
gure 2 illustra
The pr
drocarbon pr
en,
Flow chart o
utcome Bino
omial lattice
is the standar
g. Luenberge
go down by
√∆
√∆
he timestep d
ates the bino
robabilities q
rice will be d
, and
f inputs and
omial Lattic
is a discrete
rd model for
er, 1998). In
y a factor of d
duration; e.g
omial lattice
qu and qd of h
determined b
1 ′
1
46
outputs of s
ce Sub-Mod
e time approx
r the price of
every period
d
g., ∆ 1
model.
having eithe
by setting the
,
sub-models t
del
ximation of
f widely trad
d (year), the
.
er an increas
e current val
1 ′
to the dynam
geometric B
ded commod
price will ei
e or decreas
lue futures c
0
mic model
Brownian
dities such as
ither go up b
e of the
contracts to 0
s
by
0,
46
Figure 2: Binomial lattice model of gas price in the model
Here r' is the rate of increase of different hydrocarbon futures price, which is
different than the risk-free rate because of the storage cost of hydrocarbon (Luenberger
1998, p. 462). r' is calculated from current gas price and gas futures price.
1 .
.1 12.36%
where
: Gas future price for 1 year
: Current gas price.
Table 1: Binomial lattice parameters Natural Gas (r’ = 12.36%, σ = 50%)
u 1.650
d 0.604
qu 0.494
qd 0.506
4.2 The Kullback-Leibler (KL) Probability Update Sub-Model
The exploration of an asset reveals information about the underground
geological layers that will affect the probability distribution of the reserve’s volume of
neighboring assets. Probability distributions of unexplored assets are then updated
relative to the discovered volume in the explored asset.
Figure 3 shows the geological layers in eastern Mediterranean and how the gas
reserves are found on the same geological layers. Actually, during the exploration of
,
,
,
Tam
pro
of t
dril
Sm
rese
the
For
are
rese
mar, success
obability was
the region. T
Figure 3: D
In our
lling is perfo
mith (2006). T
erve in portf
portfolio.
r multiple ou
: The mar
, : The jo
The in
ervoir engin
∑
s probability
s 50%. This
This learning
Dalit, Tamar
multiple ge
DSS model
ormed throug
This sub-mo
folio and som
utcomes, the
rginal probab
oint probabi
nputs needed
eers. These
, ∈ ∀
was 35%. A
is due to the
g is accessibl
and Leviath
eological lay
, probability
gh a sub-mo
odel requires
me of the joi
potential inp
bility that as
lity that asse
d to be estima
inputs shoul
,
46
As for the ex
e learning tha
le only by dr
han prospects
ers of the thr
y update due
del adapted
the margina
nt probabilit
puts needed
sset i has rese
ets i and j ha
ated are that
ld be consist
∀
ploration of
at was made
rilling explor
s and potent
ree sites (Co
to the outco
from the rec
al probability
ty distributio
for the prob
erve outcom
ave reserve le
t can be prov
tent with bas
∀ , ∀
f Leviathan, t
e on the geol
ration wells.
tial reserves
ook 2011).
ome of explo
cent work by
y distribution
on for pairs o
bability upda
me k.
evels k and l
vided by geo
sic probabilit
(1
the success
ogical layers
.
shown on
oration
y Bickel and
n of assets’
of assets in
ate sub-mode
l.
ologists and
ty theory, i.e
)
s
d
el
e.,
46
∑ ,, ∈ ∀ ∀ , ∀ (2)
where,
S the set of pair joint probability which are inputs of the sub-model
S the set of pair joint probability which are not inputs of the sub-model
The probability update model utilizes the marginal and pair-wise joint
probability to determine joint probability distribution for all assets in the portfolio. The
all-asset joint probability is used within the DSS model to update reserve distribution
based on the outcome of sequential exploration. The all-asset distribution is determined
via optimization on a worse case basis by minimizing the “distance” to an independent
distribution which ignored correlation between assets, subject to constraints imposed by
the input values of pair-wise probabilities. Specifically, the independent distribution is
given by the following joint mass function
, , … , ∏ , 1, 2, … ,
where is the number of volume outcome of asset i. In addition, the
Kullback-Leibler (K-L) metric is used to measure the distance between the independent
and the all-probability joint distribution, with joint probability mass function ,
which is given by
, ∑ (3)
The all-asset joint distribution is then determined as the solution to the
following mathematical program.
min , (4)
Subject to:
∑ 1 (5)
∑ / (6)
46
∑ ⁄ , , , , , ∈ (7)
Solving the above problem directly is demanding mathematically. By
constructing its dual problem which is an exponential objective function formulation
and can be solved as a linear programming, the solution will be easier. In fact, using
Excel solver, we can solve problem having up to 9 assets with 3 reserve outcomes each.
The dual problem is as follows. For each constraint, we assign a dual variable
as shown below.
1 →
/
→
⁄ ,
, , → , ,
Then, the dual problem is as follows,
max
∑ ∗ , ∑ ∑ ⁄
∑ ∑ ∑ ∑ , , , ,, , ∈⁄
(8)
Subject to:
∗ , ,1 ∑ ∑ ⁄
∑ ∑ ∑ ∑ , , , ,, , ∈⁄
(9)
By optimizing the values in vector , we will have ∗ ,
4.3 The Dynamic Model
As shown in Figure 4, the dynamic model will treat E&P decisions. An asset
will start unexplored then exploration decisions will be taken. For a time, the asset will
be explored and after the exploration phase, we will know about the volume of gas in
reserve and we will decide whether to develop it or not. If asset’s development decision
is taken, then for a time, it will be developed. Once asset finishes the development
46
phase, it will be a productive asset and production decisions will be taken relative to gas
price level. At the end of the production license or when asset becomes unprofitable,
asset will be decommissioned.
Figure 4: The different upstream phases of the investment of a hydrocarbon asset
The formulation of the dynamic model is composed of an “Exploration” model
and a “Development” sub-model that feeds its results into the exploration model. The
exploration model will go through all possible combinations of exploration decisions.
Once an asset is explored, it will use another model (Development sub-model) to
calculate the profit from developing the asset and producing from it and then
abandoning it. In Section 4.3.1, we introduce the inputs and variables of the dynamic
model. In Section 4.3.2, we introduce and explain the development sub-model DP
formulation. In Section 4.3.3, we introduce and explain the exploration model DP
formulation. In Section 4.3.4, we calculate and show the complexity of the model.
4.3.1 Nomenclature of the Dynamic Programming Formulation
Inputs
: Number of production decisions
, : Current price of the extracted gas
: The volatility of the gas price; used to calculate binomial lattice parameters
: Risk free interest rate
: Royalty Ratio; contract parameter
: Tax Ratio; contract parameter
46
: Discount factor on revenues taking into consideration royalty and tax
: Number of assets
: Index of the asset varying between 1 and n
, : Current state of asset i
: Exploration cost of asset i
: Development cost of asset i
: Fix production cost of asset i
: Unit production cost of asset i
: Abandonment cost of asset i
: Deliverability ratio of asset i
: Exploration duration of asset i ( 1)
: Development duration of asset i (z 1)
: Exploration license duration of asset i
: Production license duration of asset i (Includes development time)
: Number of reserve outcomes of asset i
j: Index of the reserve outcome varying between 1 and
: The volume of the reserve outcome j of asset i
: The independent probability of the reserve outcome j of asset i
: Joint probability distribution of all assets in the portfolio
: Multiplicative factor of price going up of the binomial lattice
: Multiplicative factor of price going down of the binomial lattice
: The probability of going up of the binomial lattice
: The probability of going down of the binomial lattice
, : Deliverability of the asset i at time step t
∆ , : Increment of production level of asset i at time t.
Dynamic State Variables
t: Index of the timestep
k: Price level; an integer variable
46
: Avectorofs ,
0 ifasset isunexplored1 ifasset isexplored2 ifasset isdeveloped3 ifasset isdecommissionned
∀
: Avectorof , : Discovered reserve index of asset i at timestep t; an integer
variable
,0ifassetreserveisunknown
ifdiscoveredassetreserveindexisthe volume
: Avectorof , ; the remaining timesteps (lag time) till asset i exploration or
development decision is completed at timestep t; an integer variable
: Avectorof , ; The remaining volume in asset i at timestep t; a real valued variable
: Aninformationvectorof , ; it stores the timestep when an asset’s exploration
ended
Decision Variables
: Avectorof , ; Exploration binary decision 0,1 of asset i at timestep t for i=1 to
n; a binary decision.
: Avectorof , ; Development binary decision 0,1 of asset i at timestep t for i=1
to n; a binary decision.
: Avectorof , ; Abandonment binary decision 0,1 of asset i at timestep t for i=1
to n; a binary decision.
: Avectorofx , ; Production decision of asset i at timestep t for i=1 to n; a real
decision.
4.3.2 The Development Sub-Model
The development model will treat the development and production decision
added the fact that the asset may be decommissioned at any time. In the development
model there will be one asset to be studied and not a combination of assets. So, the asset
index is known and it is i and the reserve index is known and it is , . The model starts
at time t and price level k is assigned from the exploration model. Equation 10 is the
objective function. It is the maximization of the profit over the development decision,
abandonment decision and production decision. In the equation, profit is calculated by
46
adding revenues from selling gas in this timestep, then subtracting the development
cost, the abandonment cost and the production cost, then adding the maximum expected
future profit over the price of gas using the same formulation.
, , , 0 3,
, , , , I
, , ,
, , , , I
, , , , I
(10)
Where
γ 1 ∝ 1 β (11)
, , (12)
0 01 0 (13)
, , 1 1 2& 00
(14)
0 1 1& 0& 10
(15)
0 3 0 1 0
1 & 2
1& 10
0 1
(16)
0 ∆ 2∆ . . 2& 0& 10
(17)
∆ (18)
46
2 13 1 (19)
1 10, 1
(20)
0 , 1
2& 1 1& 1 (21)
Equation 11 calculates the discounting factor on revenues due to royalty and tax.
Equation 12 calculates the gas price at time step t and price level k.
Equation 13 determines whether fixed production cost will be subtracted or not in this.
timestep or not (To be subtracted only if asset is producing).
Equation 14 determines whether development cost will be subtracted or not in this
timestep (To be subtracted only if asset is being developed).
In Equation 15, development decision may be taken if asset is explored and no
abandonment decision is being taken.
In Equation 16, Abandonment decision cannot be taken if asset is decommissioned or it
is being developed or it is producing. Abandonment decision must be taken if either,
production license ends and asset is developed, or there isn’t enough time to develop the
asset before the exploration license ends, or discovered volume in the asset is 0.
Otherwise, abandonment decision may be taken or not.
If asset is productive, the production decisions will take discrete values between 0 and
the deliverability of the asset , relative to the number of production decisions with
a fixed interval between decisions ∆ (See Equation 18).
In Equation 17, production decision may be taken discretely if asset is developed and no
abandonment decision is being taken.
Equation 19 is the state transformation equation of asset state. In this equation, if
development decision is taken, the asset state will be “Developed” or “Being
Developed”, and it will be “Abandoned” if abandonment decision is taken. Otherwise, it
will take current value.
46
Equation 20 is the state transformation equation of the remaining time to develop the
asset. The remaining time till the end of development period will take value if
development decision is taken, and it will be of the duration of the development minus
one timestep, otherwise the remaining time will be reduced by one timestep. At any
timestep, the value of this state should not get below 0.
Equation 21 is the state transformation equation of the remaining volume in reserve.
The remaining volume will take the value of the reserve outcome in the next timestep, if
the asset is developed. It will be reduced to zero if the asset is abandoned. Otherwise,
the produced volume will be deducted from the remaining volume.
4.3.3 The Exploration Model
As shown in Equation 22, the objective function of the exploration is to maximize the
profit over exploration decisions of multiple assets. Profit is calculated by subtracting
exploration cost, adding the discounted expected profit of exploring the remaining
unexplored assets in the future by executing the exploration model, and then adding the
expected discounted profit of assets whose exploration has just finished by executing
the development sub-model.
, , , , 0 , 0& , 0∀ ,
,
∈
∑ , , , , ,
∑ ∏ ∈
, , , ,
∑ , , ,
, , , ,
∑ , , , , I ,
∈
(22)
, , , , I ,
, , , , , , , I , , 1& , 1 , 1& 10
(23)
1, 1 ; , 0
Where
is the set of possible combinations of exploration decisions.
46
, is the set of indices of assets which exploration ends in the next
timestep
, , , is the set of possible reserve outcome combinations
, , , , ,1 , 1 , 1& , 00
(24)
,0 1 , 00
∀ (25)
,1… , 1& 1 , 1& , 1
,∀ (26)
,1 , 1& 1 , 1& , 1
,∀ (27)
′ ,, , 0& 1 ,
0∀ & ∈ (28)
′ ,
,
, & , 1 , 1& , 0 & 1 ,
0
∀ & ∈ (29)
,
1 , 13 1& , 0
,
∀ (30)
,
1 , 1
0, , 1∀ (31)
, , , is the expected profit of the development of asset i whose discovered
reserve index is w , . Development sub-model will be executed once asset is explored.
Equation 24 determines whether exploration will be deducted in this timestep or not.
As shown in Equation 25, exploration decision can be taken only if asset is unexplored.
46
As shown in Equation 26, reserve index will take value at the end of the
exploration phase. Otherwise, its value remains the same.
Equation 27 shows that if exploration ends in the next timestep, , will take
the value of t+1, otherwise its value will remain the same. , vector is meant to
preserve information about exploration outcome timestep.
During the calculations of probabilities, the volume index vector used for the
calculations of probabilities will be the one that was available at exploration decision
timestep. So, if an outcome of an asset happened after the decision ( z 1 I , ), it
will not be taken into consideration. If multiple outcomes are happening at the same
timestep, probabilities will be calculated independently. Reserve index needed for the
calculation of probabilities will be modified in Equations 28 and 29.
Equation 30 is the state transformation equation of asset state vector. In t+1,
asset i will be either explored or being explored if exploration decision is taken, and it
will be decommissioned if there aren’t enough time to explore the asset in the next
timestep. Otherwise, the state of asset i remains the same.
Equation 31 is the state transformation equation of the remaining exploration
time vector. The remaining time till the end of exploration period will take value if
decision is taken, and it will be the current remaining duration for exploration
subtracted by one, otherwise the remaining time will be subtracted by one. At any state,
this value shouldn’t get below 0.
4.3.4 The Complexity of the model
A state is represented as follows
, , , , ,
The number of states will be calculated for each timestep. The number of price
levels in timestep t is t+1. The number of states of an asset i in a certain timestep t can
be accounted for approximately by counting all possible combinations. These
combinations are divided into different types according to the state of the asset. The
number of states in each type will be calculated,
, : The number of states where the asset is unexplored in timestep t
, : The number of states where the asset is abandoned in timestep t
46
, : The number of states where the asset is being explored in timestep t
, : The number of states where the asset is explored in timestep t
, : The number of states where the asset is being developed in timestep t
, : The number of states where the asset is developed and can produce in timestep t
: Numberofreserveoutcomeslargerthan0inthecalculationofthenumberofstates
As a result, the upper bound number of exploration model states is
1 , , ′ , ,
The upper bound number of development sub-model states is
1 ′′ , , ,
The number of states where asset is unexplored
, 0, , 0, , 0, , 0, , 0
,10
The number of states where asset is being explored
, 1, , 1, , 1… 1, , 0, , 0
,, 1, 0 1
0
The number of states where asset is explored
For exploration model
, 1, , 1… , , 0, , 0, , …
′ ,1, 1 1
0
For development model
, 1, , 1… , , 0, , 0
′′ , 1
0
46
The number of states where asset is being developed
, 2, , 1… , , 1… 1, , 0
,1∗
11
11
0
* It is assumed that each asset has a dry outcome
The number of states where asset is developed and can produce
, 2, , 1… , , 1… , , ⋯ , , …
,1∗
11
1
0
The number of states where asset is decommissioned
, 3, , 0, , 0, , 0, , 0
,1 1 10
Table 2: Number of states of the decoupled model compared with general model
2, 2, 3, 2
Number of
Identical
Assets
Number of
States of
the
Exploration
Model
Number of
States of the
Development
Model
Model
Number
of States
2 4 5 2.80E+02 4.90E+02 7.70E+02
5 4 5 7.67E+04 1.23E+03 7.79E+04
5 4 25 7.67E+04 6.34E+04 1.40E+05
5 8 50 3.39E+07 4.61E+05 3.44E+07
5 10 60 1.95E+08 7.84E+05 1.96E+08
5 15 90 4.17E+09 2.57E+06 4.18E+09
5 16 100 6.73E+09 3.51E+06 6.73E+09
46
5 Structural Properties of the Production Decision in the Development Model
This section represents structural properties of the value function model. In
Section 5.1, we prove that the value function of the development sub-model is convex
in xt. In Section 5.2, we prove that the production decisions are binary, either to produce
at maximum level or not to produce. In Section 5.3, we prove that production decision
is not to be taken when marginal profit is negative. In Section 5.4, we show the
computational enhancements by implementing the theorems in the model. In
Section 5.5, we show the speed of execution of the model on excel (visual basic).
5.1 Value Function of the Development Sub-Model is Convex in xt
Lemma1: , , , , isconvexin
Proof.SeeMelki(2012)1.∎
5.2 Binary Decisions
Theorem 1: Production decisions are binary that is, ∗ ,
Proof.SeeMelki(2012).∎
5.3 Production Decision is not to be Taken When Marginal Profit is Negative
Theorem 2: if at time t and price level k, the marginal profit is negative, ,
∀ , then ∗
Proof.SeeMelki(2012).∎
5.4 Computational Enhancement Results
Theorems 1 and 2 have a great effect on the model. Before, we had a very large
number of decisions and resulting states leading to high storage requirement. These
theorems significantly reduce storage requirement. Table 3 shows the percent decrease
in the number of states relative to taking three discrete timesteps
1 I. G. Melki, “Optimizing Exploration and Production Decisions in Upstream Hydrocarbon Portfolio Management”, American University of Beirut (AUB), 2012.
46
Table 3: The percent decrease in the number of states relative to taking 3 discrete
production decisions
Deliverability\Tp 5 10 15 20 30 0.0005 79.08% 91.76% 93.63% 94.74% 95.80% 0.005 70.41% 83.88% 88.27% 91.23% 93.93% 0.01 62.24% 80.36% 86.34% 90.08% 93.37% 0.05 57.14% 78.55% 85.44% 89.58% 93.15% 0.1 57.14% 78.15% 85.22% 89.47% 93.13%
5.5 The Speed of Execution of the Model
We also tested the duration of the execution of the model on a PC having the
following characteristics.
Processor: Intel (R) Core (TM) i7 CPU Q720 @ 1.60GHz
Installed Memory (RAM): 6.00GB
System type: 64-bit operating system
The duration of the execution of the model is affected by the number of
timesteps and the number of assets. We fixed the duration of the exploration period to
be 5 years and the duration of the development and production period to be 30 years
with timestep duration of 1 year. We increase the number of assets from 1 to 6. The
resulting execution durations are shown in Error! Reference source not found. and
Figure 5.
0
100
200
300
400
500
600
0 2 4 6
Execution Duration (in
seconds)
Number of Assets
46
Figure 5: The execution duration function of the number of assets
As it can be seen in Figure 5, the model runs within few minutes for up to five
assets. For a large number of assets, storage requirement become excessive despite our
efforts to reduce storage in Theorems 1 and 2.
6 Numerical results
In this section, we present numerical results and analyze the behavior of the
exploration, development and production decisions. In section 6.1, we analyze a case of
five assets having a five year exploration period and 30years production period. We
observe exploration results and analyze them. In section 6.2, we observe and analyze the
results of sensitivity on the volatility of the gas price, the interest rate and the
correlation between assets.
6.1 Illustrative Example
A company had just won the bid for developing an offshore block. The contract
with the government licenses the company to explore the block for 5 years and if
hydrocarbon is discovered, the company has the right to develop and produce from asset
for an additional 30years (Inspired from Tamar contract). In return, the government will
get 10% royalty and 40% taxes on hydrocarbon revenues (These values are within
range, see Younes 2011). The company is now looking to exploit the hydrocarbon
reserves in this region.
The company did some geological studies in the area and found out that there
are 5 potential spots where gas reserves can be found, and these assets have correlated
reserve outcomes (Figure 6).
F
6.1.
pro
asse
Tab
acc
Exp
dril
igure 6: Cas
.1 Inputs
Each a
oduction wel
et. The expe
ble 4, and th
cording to ex
ploration), th
lling a produ
se Study Ass
asset will req
ls in order to
ected number
e expected n
xperts that w
he cost of dr
uction well w
sets within th
quire a numb
o produce. T
r of explorat
number of de
we met at LIP
rilling an exp
will be $100M
46
he licensed b
ber of explor
The number o
tion wells fo
evelopment w
PE (Lebanon
ploration we
M. We assum
block and co
ration wells
of these well
or each of the
wells is as sh
n Internation
ll will be $1
me an additi
orrelation bet
and a numbe
ls is relative
e assets is as
hown in Tab
al Petroleum
50M, and th
ional cost of
tween assets
er of
to the size o
s shown in
ble 5. And,
m
he cost of
f $300M will
s
of
l
46
be added to development cost to account for the common facilities to be installed and
for laying pipes to connect the installed subsea system to the existing gas network.
Table 4: The number of exploration wells for each of the assets and the corresponding
exploration cost
Number of
Exploration Wells Exploration Cost
Asset A 1 150M
Asset B 3 450M
Asset C 1 150M
Asset D 1 150M
Asset E 4 600M
Table 5: The number of development wells for each of the assets and the corresponding
development cost
Number of
Development WellsDevelopment Cost
Asset A 4 700M
Asset B 26 2,900M
Asset C 2 500M
Asset D 8 1,100M
Asset E 44 4,600M
In our analysis, we assumed an expected abandonment cost of $25M per drilled
well. Fixed production cost is proportional to the number of production wells. The unit
production cost includes the transportation cost and the treatment cost of gas and it is
assumed to be $0.0005/scf. Inputs to the model are summarized in Table 6, Table 7 and
Table 8.
46
Table 6: Summary of exploration, development, abandonment, fixed production cost
and unit production cost of each of the assets
Asset Name Ce($) Cd($) CA($) Cfp($) Cup($/scf)
Asset A 150M 700M 125M 5M 0.0005
Asset B 450M 2,900M 725M 40M 0.0005
Asset C 150M 500M 75M 5M 0.0005
Asset D 150M 1,100M 225M 10M 0.0005
Asset E 600M 4,600M 1,175M 65M 0.0005
Table 7: General parameters to be used in the estimation of the exploitation of the block
Royalty Ratio, α 10%
Tax Ratio, β 40%
Yearly Risk Free Interest Rate, r 0.30%*
Current Gas Price, p0,0 ($/cfg) 0.004781**
Deliverability 6%***
Volatility of Gas Price, σ 50%****
Gas Price Growth, r’ 12.36%****
u 1.649*****
d 0.607*****
qu 0.496*****
qd 0.504*****
*Treasury bill interest rate for 1year on June 2010 **The price of gas on June 2010 ***Deliverability value used for the development of Tamar gas reserve, Cook
2011 ****Refer to Section 3.2 *****Refer to Section 4.1 for calculation
46
Table 8: Reserve probability distribution of each of the assets
Relative Asset
Name
Discrete Reserve
Outcome Name
Volume in
Reserve (scf)
Marginal
Probability
Expected Reserve
given Wet (scf)
Asset A
A-Low - 0.5 ∗ 0.2 1 ∗ 0.10.2 0.1
0.67T
A-Medium 0.50T p1(2)=0.20
A-High 1.00T p1(3)=0.10
Asset B
B-Low -
4.00T B-Medium 3.00T p2(2)=0.20
B-High 6.00T p2(3)=0.10
Asset C
C-Low -
0.33T C-Medium 0.25T p3(2)=0.20
C-High 0.50T p3(3)=0.10
Asset D
D-Low -
1.33T D-Medium 1.00T p4(2)=0.20
D-High 2.00T p4(3)=0.10
Asset E
E-Low -
6.67T E-Medium 5.00T p5(2)=0.20
E-High 10.00T p5(3)=0.10
The area has not been explored yet, and it has potential gas reserves. Reserve
probability distributions shown in Table 8 are inspired from the reserves already
discovered in Eastern Mediterranean. Compared to Tamar field (8.40T scf of gas) and
Dalit (0.5T scf of gas), these assets are within range.
Assets’ outcomes are either correlated or have a low correlation. Correlated assets have
joint probabilities as shown in
31
Table 9. Assets that have low correlation between each other have joint
probabilities as shown in Table 10.
Table 9: Correlated Assets’ Pair-Joint
Probabilities
Low Medium High
Low
Medium 0.12 0.04
High 0.04 0.03
Table 10: Low Correlation Probabilities
6.1.2 Results
The company will optimize exploration and production decisions using our
DSS model. The DSS model was executed using the inputs in Section 6.1.1 and with
timestep duration of 1year.
The DSS model chose not to explore any asset at t = 0 and t = 1. This behavior
is justified by the fact that gas price is expected to increase (i.e., r’=12.36%) and
postponing decisions will likely increase profit. At t = 2, there are three price level
scenarios to be studied, price increases twice, no price change, and price decreases
twice.
At timestep 2 and for a price level k=2 (i.e., price increased twice), exploration
decisions are as shown in Figure 7. In this case, there is no learning between outcomes
of assets. At this level of the gas price, and knowing the trend of gas price fluctuation, it
is better to wait and get knowledge of the price level. All assets finish their exploration
phase at timestep t=5 when the exploration license ends. As a result, assets B and E are
explored at timestep 3 and assets A, C and D are explored at timestep 4.
Low Medium High
Low
Medium 0.08 0.03
High 0.03 0.02
32
Figure 7: Optimal sequential exploration decisions if project price level is k = 2 at t = 2
If, at timestep t=2, price level is k=0 (no price change), exploration decisions
are as shown in Figure 8.
Figure 8: Optimal sequential exploration decisions if project price level is k = 0 at t = 2
Assets A and C are to be explored first at timestep t=2. At timestep t=3, there
are 2 scenarios whether reserve outcome of asset A is Low or whether it was Medium or
33
High. For a Medium or High reserve outcome of asset A, asset B is explored and the
company benefits from learning of asset A because assets A and B are correlated. For a
Low reserve outcome of asset A, exploration of asset B is not to be happening; knowing
that asset A is dry. At timestep t=3, Asset E is explored regardless of outcomes, it is an
asset that has low correlation with the other assets and it needs two years to be explored.
As for asset D, it will be explored at timestep t=4 and it has low correlation with the
other assets and it needs one year to be explored.
For timestep t=2 and price level k=-2 (decreased twice), exploration decisions’
behavior is the same as of price level k=0 (no price change).
6.2 Sensitivity Analysis on the Model
In this section, we present numerical results analyzing the effect of
environmental parameters especially those involving uncertainty on the profit and
exploration decisions. Uncertainty in this model is represented by the volatility of the
commodity price and the probability distribution of the reserve outcomes. We focus on
analyzing the expected profit and decision making behavior by doing sensitivity
analysis on the volatility of the gas price and the correlation between assets reserve
represented by a parameter which is the value of / , 1
(i.e., the conditional probability of assets j have medium (or high) reserve level given
that asset i has the same reserve level). Accordingly, we calculate the value of joint
probabilities , 1; The other pair joint probabilities are
not inputs to the KL sub-model, they are to be estimated according to the available
information.
We are considering typical general inputs of the model to do the sensitivity
analysis as shown in Table 11.
Table 11: Summary of the base inputs to the sensitivity analysis
Royalty Ratio, α 10%
Tax Ratio, β 40%
Yearly Risk Free Interest Rate, r 0.30%
34
Current Gas Price, p0,0 ($/cfg) 0.004781
Volatility of Gas Price, σ 50%
Gas Price Growth, r’ 12.36%
We are also considering identical assets having the characteristics shown in Table 12.
Table 12: Asset inputs used in the sensitivity analysis
Exploration
Cost
Unit Production
Cost
Exploration
Duration
Exploration License
Duration
150M 0.0005 1 2
We are considering that initial probability distributions are fixed as shown in
Table 13 and we are not doing any sensitivity on them.
Table 13: Reserve outcome inputs used in the sensitivity analysis
Reserve Outcome Name Volume (CFG) Independent Probability
R-L -
R-M 1000B 0.25
R-H 2000B 0.25
Considering a 2 asset independent case ( 0.25), the profit from exploiting
these assets is $2304M.
6.2.1 Sensitivity on Profit
Base inputs used in this analysis represent the current situation of investment in
the E&P of gas reserves. A sensitivity analysis on the base case was done by changing
key model parameters, namely the risk-free rate r, the price growth r’, volatility of gas
price σ, and the correlation parameter ρ. The parameters will be modified discretely
between a minimum value and a maximum value mentioned in Table 14. The vertical
axis of Figure 9 show the percent change in value relative to the base case.
35
Table 14: The range of parameters
Parameter Min. Value Max
Value
r 0% 50.30%
r' 0% 200%
σ 12.50% 250%
ρ 0% 100%
Figure 9: Percent increase/decrease relative to base case profit function the
increase/decrease in the value of each of the parameter (interest rate r, volatility of the
gas price σ, conditional probabilities’ value ρ, gas price growth r’)
‐100%
‐80%
‐60%
‐40%
‐20%
0%
20%
40%
60%
80%
100%
‐100.00% 0.00% 100.00% 200.00% 300.00%
Percent increase/decrease relative to base case
profit
Increase/Decrease in the value of the parameter (r, σ, ρ, r')
r
σ
ρ
r'
36
As shown in Figure 9, in terms of sensitivity magnitude, the expected profit is
highly sensitive to price growth rate r’, followed by the interest rate r, the volatility σ,
and finally the correlation ρ.
6.2.2 Sensitivity on Exploration Decisions for Fixed Gas Price and Fixed Interest
Rate
In this section, the price of gas is fixed ( 0 ’ 0%). For independent
assets, exploration decisions are as shown in Figure 10.
Figure 10: Optimal exploration decisions for a project with two identical independent
assets (r’=0%, 0, r=0.3%)
Both assets are explored at t=0 because of the time value of money. In this
case, postponing exploration and production decisions will generate less profit.
If the two assets’ outcomes are perfectly correlated (ρ=1), then exploration
decisions will be as shown in Figure 11. So, the first asset is explored first, and the
second asset is explored in the next timestep when the first asset’s exploration ends with
a medium or high reserve outcome (Note that if the first asset’s exploration outcome is
dry, the second asset will not be explored).
37
Figure 11: Optimal exploration decisions for a project with two identical assets (ρ=1,
r’=0%, ~0, r=0.3%)
As shown in Figure 12, a one way sensitivity analysis is done. At a certain
positive correlation, learning will start profiting the project. Unlike what is presented in
Figure 9, profit in this case increased by 35% due to the increase in the value of .
Figure 12: Percentage increase in profit relative to two independent asset profit function
of ρ (r’=0%, ~0, r=0.3%)
6.2.3 Sensitivity on Exploration Decisions for Volatile Gas Price and Interest Rate
In this section, we are analyzing the effect of having two uncertainties in the
model. We analyze how decisions are taken relative to the volatility of gas price and the
correlation between reserve outcomes of different assets.
0%
5%
10%
15%
20%
25%
30%
35%
0.2 0.4 0.6 0.8 1
Percentage
increase in
profit
relative to two independent
assets
ρ
38
For σ = 50%, the exploration of both assets ended when their exploration
license ended. So, exploration decisions were “shifted to the end” as shown in Figure
13, in order to benefit from the potentially high price levels in the production phase.
Figure 13: Optimal exploration decisions for a project with two identical
Figure 14 analyzes the trade-off between volatility σ and learning effects on the
exploration decisions timing prospective. Exploration decisions for both assets are being
taken at the beginning for low volatility and low correlation. As the value of ρ increases,
exploration decisions will become subsequent and learning will occur. For high
volatility, exploration decisions will be shifted to the end of exploration license. As
aforementioned, this occurs in order to benefit from a potential prices increase under
high volatility.
Figure 14: Optimal exploration decision behavior of two assets in two way sensitivity
on σ and ρ (r’=0%, r=0.3%, 2 assets)
By increasing the number of similar assets to three, similar results will be
reflected, but learning effect becomes more important as shown in Figure 15. Learning
σ\ρ 0.25 0.3 0.40 0.50 0.60 0.70 0.80 0.90 0.99
0.10%
2.00%
5.00%
10.00%
15.00%
20.00%
30.00%
50.00%
80.00%
100.00%
140.00%
200.00%
All Explorations at t=0
Learning
Exploration Decisions at t=1
39
happens (as reflected by sequential exploration decisions) over a wider range of
volatility σ, and correlation ρ.
Figure 15: Optimal exploration decision in two way sensitivity on σ and ρ
Next, we investigate the effect of the risk-free rate. It is interesting to study the
effect of an increase in the interest rate (0.3%), as it may be the case in practice.
6.2.4 Sensitivity on Exploration Decisions for a Volatile Gas Price and Different
Values of Interest Rate
Considering the base case with fixed price, we increased the value of the interest rate to
5%. On the other hand, and as shown in Figure 16, exploration decisions behaved
almost the same as in Figure 14 where interest rate is 0.3%.
Figure 16: Optimal exploration decisions of two assets in two way sensitivity on σ and ρ
σ\ρ 25% 30% 40% 50% 60% 70% 80% 90% 99%
0.10%
2.00%
5.00%
10.00%
15.00%
20.00%
30.00%
50.00%
80.00%
100.00%
140.00%
200.00%
Learning
Exploration Decisions at t=1
All Explorations at t=0
σ\ρ 0.25 0.3 0.40 0.50 0.60 0.70 0.80 0.90 0.99
0.10%
2.00%
5.00%
10.00%
15.00%
20.00%
30.00%
50.00%
80.00%
100.00%
140.00%
200.00%
Learning
Exploration Decisions at t=1
All Explorations at t=0
40
Figure 17: Optimal exploration decision behavior of two assets in two way sensitivity
on σ and ρ (r’ = 0%, r = 15%, 2 assets)
At high interest rates, exploration costs have a higher weight in the present
value of the expected profit. Therefore, exploration/production decisions are shifted to
the future to benefit from high potential prices that could offset the high present value of
the exploration cost. As shown in Figure 17, exploration decisions are taken at timestep
t = 1 for lower value of volatility than that shown in Figure 14 (i.e., interest rate is
0.3%). As the interest rate increases to 25%, as shown in Figure 18, projects with low
volatility and low correlation are becoming unprofitable, and for higher volatility the
model is choosing to wait for a probable increase in price to take exploration decisions.
Exploration decisions’ behavior is shown in Figure 19.
Figure 18: Optimal exploration decision behavior of two assets in two way sensitivity
on σ and ρ (r’ = 0%, r = 25%, 2 assets)
σ\ρ 0.25 0.3 0.40 0.50 0.60 0.70 0.80 0.90 0.99
0.10%
2.00%
5.00%
10.00%
15.00%
20.00%
30.00%
50.00%
80.00%
100.00%
140.00%
200.00%
Exploration Decisions at t=1
Exploration
Decisions at t=0 Learning
σ\ρ 0.25 0.3 0.40 0.50 0.60 0.70 0.80 0.90 0.99
0.10%
2.00%
5.00%
10.00%
15.00%
20.00%
30.00%
50.00%
80.00%
100.00%
140.00%
200.00%
No Exploration Decisions, Assets
are unprofitable
Exploration Decisions at t=1
Learning makes the whole project
profitable
41
Figure 19: Optimal exploration decisions for a project with two identical independent
assets (r’ = 0%, 10%, r = 25%)
6.2.5 Decision Making for High Price Growth Rate
In this subsection, we observe the results of sensitivity analysis where the
expected price growth is high (r’ = 12.36%). In this case, for a two asset portfolio, when
the interest rate is 0.3%, all exploration decisions are at t=1 (i.e., the second timestep) as
shown in Figure 20. When interest rate increases to 25% (Figure 21), then for low
volatility values and high correlation between asset, learning is happening; otherwise,
exploration decisions are taken at t = 1.
Figure 20: Optimal exploration decision behavior of assets in two way sensitivity on σ
and ρ (r’=12.36%, r=0.3%, 2 assets)
σ\ρ 0.25 0.30 0.40 0.50 0.60 0.70 0.80 0.90 0.99
12.00%
15.00%
20.00%
30.00%
50.00%
80.00%
100.00%
140.00%
200.00%
Exploration Decisions at t=1
42
Figure 21: Optimal exploration decision behavior of assets in two way sensitivity on σ
and ρ (r’=12.36%, r=25%, 2 assets)
Figure 22: Optimal exploration decision behavior of assets in two way sensitivity on σ
and ρ (r’=12.36%, r=0.3%, 3 assets)
Figure 23: Optimal exploration decision behavior of assets in two way sensitivity on σ
and ρ (r’=12.36%, r=25%, 3 assets)
Learning becomes more important when the number assets increases, this is
demonstrated between Figure 20 and Figure 22, and between Figure 21 and Figure 23.
σ\ρ 0.25 0.30 0.40 0.50 0.60 0.70 0.80 0.90 0.99
12.00%
15.00%
20.00%
30.00%
50.00%
80.00%
100.00%
140.00%
200.00%
Exploration Decisions at t=1
Learning
σ\ρ 25% 30% 40% 50% 60% 70% 80% 90% 99%
12.00%
15.00%
20.00%
30.00%
50.00%
80.00%
100.00%
140.00%
200.00%
LearningExploration Decisions at t=1
σ\ρ 25% 30% 40% 50% 60% 70% 80% 90% 99%
12.00%
15.00%
20.00%
30.00%
50.00%
80.00%
100.00%
140.00%
200.00%
Learning
Exploration Decision at t=1
43
7 Conclusions
In this research, we develop a dynamic model that handles multiple decisions
over time. The model is capable of simulating hydrocarbon price fluctuation and
learning from exploration drilling. We experimented on the model and observed that
profit in E&P projects are sensitive to the expected price growth and volatility of the gas
price. On the other hand, E&P projects are negatively sensitive to the interest rate. We
observed while doing sensitivity analysis on the model that exploration decision taking
is affected the expected gas price growth, the volatility of the gas price, the interest rate
and the correlation between assets. For a fixed price (or low expected price growth and
low volatility) and high interest rate, the decisions (i.e., exploration and production
decisions) will be shifted to earlier timesteps. As for having a high expected price
growth or high volatility of the price, decisions will be shifted to later timesteps. Even
with high interest, it is more profitable to wait for later timesteps to benefit from a
probable price growth. Finally, for a high correlation between assets, exploration
decisions will be successive to benefit from learning.
References
[1] M. Al-Harthy and A. Khurana, "Portfolio Optimization: Part 1—The Modeling Approach of Inter-and Intra-dependencies", Energy Sources, Part B: Economics, Planning, and Policy, vol. 3, no. 4, pp. 315-323, 2008.
[2] M. Al-Harthy and A. Khurana, "Portfolio Optimization: Part 2—An Application of Oil Projects with Inter-and Intra-dependencies", Energy Sources, Part B: Economics, Planning, and Policy, vol. 3, no. 4, pp. 324-330, 2008.
[3] M. Armstrong, A. Galli, W. Bailey and B. Couët, "Incorporating technical uncertainty in real option valuation of oil projects", Journal of Petroleum Science and Engineering, vol. 44, no. 1-2, pp. 67-82.
[4] C.O. Aydın, "Sequential investment planning for complex oil development projects", Massachusetts Institute of Technology, 2008.
44
[5] W. Aylor Jr, "Measuring the impact of 3D seismic on business performance", J.Pet.Technol., vol. 51, no. 6, pp. 52-56, 1999.
[6] C.Y. Baldwin, "Optimal sequential investment when capital is not readily reversible", The Journal of Finance, vol. 37, no. 3, pp. 763-782, 1982.
[7] P.I. Barton and A. Selot, “Short-term supply chain management in upstream natural gas systems”, Massachusetts Institute of Technology, 2009.
[8] J.E. Bickel and J.E. Smith, "Optimal sequential exploration: A binary learning model", Decision Analysis, vol. 3, no. 1, pp. 16-32, 2006.
[9] R. Cook, S. Cunningham, "Eastern Mediterranean", Noble Energy, 2011.
[10] G. Cortazar, E.S. Schwartz and J. Casassus, "Optimal exploration investments under price and geological‐technical uncertainty: a real options model", R&D Management, vol. 31, no. 2, pp. 181-189, 2000.
[11] P.I. Davidsen, J.D. Sterman and G.P. Richardson, "A petroleum life cycle model for the United States with endogenous technology, exploration, recovery, and demand", System Dynamics Review, vol. 6, no. 1, pp. 66-93, 1990.
[12] L. Davidson and J. Davis, "Simple, effective models for evaluating portfolios of exploration and production projects", pp. 325-332, 1995.
[13] E. Durrer and G. Slater, "Optimization of petroleum and natural gas production-A survey", Management Science, vol. 24, no. 1, pp. 35-43, 1977.
[14] Energy Information Administration (EIA) and US Energy Information Administration, "Annual Energy Review, 2008", 2009.
[15] J. Etherington, T. Pollen and L. Zuccolo, “Comparison of selected reserves and resource classifications and associated definitions”, Society of Petroleum Engineers, 2005.
[16] L. Frair and M. Devine, "Economic optimization of offshore petroleum development", Management Science, vol. 21, no. 12, pp. 1370-1379, 1975.
[17] V. Goel and I.E. Grossmann, "A stochastic programming approach to planning of offshore gas field developments under uncertainty in reserves", Comput.Chem.Eng., vol. 28, no. 8, pp. 1409-1429, 2003.
45
[18] V. Goel, I.E. Grossmann, A.S. El-Bakry and E.L. Mulkay, "A novel branch and bound algorithm for optimal development of gas fields under uncertainty in reserves", Comput.Chem.Eng., vol. 30, no. 6-7, pp. 1076-1092, 2006.
[19] K.K. Haugen, "A Stochastic Dynamic Programming model for scheduling of offshore petroleum fields with resource uncertainty* 1", Eur.J.Oper.Res., vol. 88, no. 1, pp. 88-100, 1996.
[20] N.J. Hyne, '"Nontechnical guide to petroleum geology, exploration, drilling, and production", 2001.
[21] D.L. Keefer, '"Resource allocation models with risk aversion and probabilistic dependence: offshore oil and gas bidding," Management science, vol. 37, no. 4, pp. 377-395, 1991.
[22] R. Goldsmith, R. Eriksen, M. Childs, B. Saucier and F. J. Deegan, “Lifecycle Cost of Deepwater Production Systems”, Offshore Technology Conference, 2001.
[23] Klett TR, "United States geological survey's reserve-growth models and their implementation," Natural Resources Research, vol. 14, no. 3, pp. 249-264, 2005.
[24] J. Laherrere, "The Evolution of the World‘s Hydrocarbon Reserves" Retrieved, vol. 7, no. 15, pp. 2004.
[25] D.G. Luenberger, "Investment science", 1998.
[26] M.W. Lund and N.G. Marketing, "Real Options in Offshore Oil Field Development Projects".
[27] R. Motta, G. Caloba, L. Almeida, A. Moreira, M. Nogueira, L. Cardoso and L. Berlink, "Investment and Risk Analysis Applied to the Petroleum Industry", 2000.
[28] Pemex Exploración y Producción (PEP), “Hydrocarbon Reserves of Mexico”, 2007.
[29] M.A. Skaf, "Portfolio management in an upstream oil and gas organization," Interfaces, pp. 84-104, 1999.
[30] S. Smidt, "Decisions under Uncertainty: Drilling Decisions by Oil and Gas Operators" Adm.Sci.Q., vol. 6, no. 1, pp. 112-114, .
[31] J.P. Stern, D. Fridley and A. Flower, "Natural gas in Asia: the challenges of growth in China, India, Japan and Korea", 2008.
46
[32] S.B. Suslick and R. Furtado, "Quantifying the value of technological, environmental and financial gain in decision models for offshore oil exploration", Journal of Petroleum Science and Engineering, vol. 32, no. 2-4, pp. 115-125, 2001.
[33] S.B. Suslick, D. Schiozer and M.R. Rodriguez, "Uncertainty and Risk Analysis in Petroleum Exploration and Production".
[34] S. Suslick and D. Schiozer, "Risk analysis applied to petroleum exploration and production: an overview", Journal of Petroleum Science and Engineering, vol. 44, no. 1-2, pp. 1-9, 2004.
[35] S. Tanaka, Y. Okada, Y. Ishikawa, “Offshore drilling and production equipment”, Encyclopedia of Life Support Systems (EOLSS), Eolss Publishers, Oxford ,UK, [http://www.eolss.net], 2005.
[36] S.A. van den Heever, I.E. Grossmann, S. Vasantharajan and K. Edwards, "Integrating complex economic objectives with the design and planning of offshore oilfield infrastructures", Comput.Chem.Eng., vol. 24, no. 2-7, pp. 1049-1055, 2000.
[37] M.R. Walls, "Combining decision analysis and portfolio management to improve project selection in the exploration and production firm", Journal of Petroleum Science and Engineering, vol. 44, no. 1-2, pp. 55-65, 2004.
[38] US Treasury, www.ustreas.gov
[39] Wall Street Journal, http://online.wsj.com
[40] N. Younes, “On structuring offshore hydrocarbon production sharing contracts Lebanon's case”, American University of Beirut (AUB).