SPE 63201

Valuing and Comparing Oil and Gas Opportunities: A Comparison of Decision Tree
and Simulation Methodologies
B.S. Mudford, Landmark Graphics Corporation

Copyright 2000, Society of Petroleum Engineers Inc.

Consequently, decision analysis in the oil industry has been an
This paper was prepared for presentation at the 2000 SPE Annual Technical Conference and active area of investigation for at least 50 years. A
Exhibition held in Dallas, Texas, 1–4 October 2000.
comprehensive, but not exhaustive, bibliography of the
This paper was selected for presentation by an SPE Program Committee following review of
information contained in an abstract submitted by the author(s). Contents of the paper, as
decision analysis literature relating to the oil industry is
presented, have not been reviewed by the Society of Petroleum Engineers and are subject to contained in Ref. 1. Over the last half century economic
correction by the author(s). The material, as presented, does not necessarily reflect any
position of the Society of Petroleum Engineers, its officers, or members. Papers presented at analysis has evolved from single point determinations to more
SPE meetings are subject to publication review by Editorial Committees of the Society of
Petroleum Engineers. Electronic reproduction, distribution, or storage of any part of this paper
complex, probabilistically driven techniques. The most
for commercial purposes without the written consent of the Society of Petroleum Engineers is common method of economic analysis uses decision trees to
prohibited. Permission to reproduce in print is restricted to an abstract of not more than 300
words; illustrations may not be copied. The abstract must contain conspicuous derive a risk-weighted mean NPV for the project under
acknowledgment of where and by whom the paper was presented. Write Librarian, SPE, P.O.
Box 833836, Richardson, TX 75083-3836, U.S.A., fax 01-972-952-9435.
consideration. Often a four-point decision tree is built for a
single prospect. For this type of decision tree there are three
‘success’ branches, representing outcomes with ‘high’,
Abstract ‘medium’ and ‘low’ recoverable volumes with each branch
Decision trees have been widely used in the petroleum carrying the NPV forecast for that outcome. The fourth, or
industry for both economic evaluation and decision analysis. ‘failure’, branch carries the NPV of the unsuccessful
In recent times Monte Carlo simulation has begun to be used exploration program. The risk-weighted mean NPV is simply
to value oil and gas projects and to analyze decisions. In this calculated by summing the product of probability and NPV on
presentation I will compare and contrast decision tree and each branch. Decision trees are also frequently used to analyze
simulation methodologies for both economic valuation and management decisions by finding the decision with the
decision analysis. As a starting point, I will take a typical Gulf maximum EMV. Clearly, the complexity of a decision tree
of Mexico project from exploration through to development. grows exponentially as the number of factors for which
The results of a decision tree analysis of the project and a uncertainties are to be analyzed increases. Decision trees are
Monte Carlo simulation of the project will be compared. either not used, or specialist practitioners are employed, when
When using decision trees, the expected monetary value complex decisions or problems with large numbers of
(EMV) is usually calculated and used to value the opportunity. uncertain parameters must be analyzed. This is due to the large
Frequently the impact of uncertainties on the EMV is not amount of work involved in building and analyzing these
assessed. In contrast, with the simulation approach the full net types of decision trees.
present value (NPV) distribution is calculated along with the In contrast to decision tree methods, probabilistic methods
risk. We will show that, for the Gulf of Mexico project, a consider a large number of possible scenarios rather than just a
relatively complicated decision tree is needed if the EMV of few. Most companies use a mixture of probabilistic and
the decision tree is to be close to the mean NPV calculated decision tree methods2. In particular, probabilistic methods are
using the simulation. However, even when the EMV and mean frequently used to analyze likely hydrocarbon volumes, but
NPV are close, the additional information about the NPV are infrequently used in the economic analysis, even though it
distribution that is part of the simulation output enables risk to is recognized that there is considerable uncertainty associated
be an explicit part of the decision-making process. Although with many of the engineering and economic components of an
decision tree methods can be extended to include an oil and gas project. Once hydrocarbons are found, there are
assessment of risk, since the form of the NPV distribution is generally one or more possible development scenarios
not known when using decision trees, a true assessment of depending on the volume found and the production rates. By
project risk is not available. assigning probability distributions to the engineering elements
of production rates and declines, investments and expenses,
Introduction distributions of economic indicators such as NPV, ROR and
Risk is associated with every economic decision involved in P/I can be calculated. Conditional logic is used to build all
exploring for, and developing, oil and gas deposits. possible development scenarios into a single probabilistic
2 B.S. MUDFORD SPE 63201

model, which can take the place of multiple decision tree consist of subsea completions with tiebacks to existing
cases. Such a model can be used to investigate the inherent infrastructure. If ‘medium’-sized volumes are found then the
risks and provide probabilities of achieving a positive NPV, or development will go forward with a small platform. Finally, if
of adding a certain volume of reserves. ‘large’ volumes are found a large platform will be used for the
Decision tree and simulation methodologies are sometimes development. A major part of the economic analysis of this
presented as being mutually exclusive tools that analyze prospect will involve quantitatively defining what is meant by
different aspects of the evaluation problem3. In particular, it a successful exploration well, and by ‘small’, ‘medium’ and
has been stated that decision tree tools are ideal for analyzing ‘large’ volumes. Monte Carlo simulation techniques will be
management decisions, whereas simulation focuses on used to answer these questions.
analyzing uncertainty for a predetermined project scenario. In
fact, as has been pointed out by other authors4,5,6 Monte Carlo Geologic Probability of Success and Recoverable Volumes.
simulation methods can be used to analyze the economic The geologic probability of success for this prospect has been
impact of different decision options. set at 60%. For the purposes of this paper, the total geologic
The aim of the current paper is to compare the decision probability of success has been used, rather than the individual
tree and simulation approaches to analyzing a relatively risk factors that apply to components of the petroleum system.
straightforward exploration and development project for a The recoverable volume distribution for this prospect has been
typical Gulf of Mexico, shelf gas field. The paper contains determined by combining together probability distributions for
five main sections. In section two the stochastic model used to area, net pay and recovery factor. The cumulative distributions
evaluate the prospect is described. In section three the method for each of the volume components, together with the
used to determine the economic threshold in a stochastic cumulative distribution for the recoverable gas volume are
evaluation is described. Next, in section four, the decision tree show in Fig. 1. For the particular prospect modelled in this
model for the gas project is outlined and the method used to paper it has been determined that maximum volume of gas
determine the economic threshold in a decision tree model is above the spill point is 750 Bcf. Hence the volume distribution
described and applied. In section five the methods used to has been truncated at this value. In addition, there is a
analyze management decisions within stochastic models are dependency between area and net pay for this prospect, which
described and applied to determine the optimum development means that if the area is large then the average net pay will
scenarios for the project of interest. Finally, in section six, the tend to be high. The impact of the relationship between area
analyses of model sensitivities and drivers in decision tree and and net pay is shown in Fig. 2.
stochastic models are compared.
Well Recoverability, Production Rates and Costs. One of
Gulf of Mexico Gas Field Development the major differences between decision tree and full
In a probabilistic model it is convenient to divide the model simulation models is that a decision tree model effectively
into separate geoscience, engineering and economic sections. adds together a relatively small number of discrete cases,
Many existing evaluation processes draw a sharp distinction while the simulation model is a single model incorporating the
between the treatment of reservoir parameters, and full range of variability. This means that the simulation model
engineering and economic parameters. It is common to model must be fully scalable across the full range of recoverable
recoverable volumes probabilistically, and other parameters volumes so that appropriate costs are incurred when both
deterministically or with decision trees2. Often, just a few small and large volumes are found. One way to achieve
cases from the probabilistic volume distribution are selected to scalability is to determine the required number of wells based
use in an economic model. The stochastic procedure presented on the expected recoverability per well. For the area of interest
in this paper carries all the uncertainty from the probabilistic the predicted well recoverability could range anywhere
volume model into a fully probabilistic engineering and between 5 Bcf and 60 Bcf with a most likely value of 20 Bcf.
economics model. This has the advantage of retaining all the In addition, higher recoverabilities are expected when the net
information from the geoscience model in the economics. This pay is high. Hence, there is a dependency between the
approach allows a more robust economic evaluation to be recoverability per well and the net pay, which is shown in Fig.
carried out than is traditionally the case. Another advantage of 3. Field level curtailments of 75 MMcf/day, 175 MMcf/day
modelling in this way is that a multidisciplinary team and 250 MMcf/day were used for the subsea, small platform
approach is necessary to populate the model, which in turn and large platform developments respectively.
produces much better communication between the engineers Another major uncertainty for this prospect is the
and geoscientists involved in evaluating prospects. production profile. The fraction of the ultimate recoverable
The subject of the analysis to be presented in this paper is produced on initial plateau was modelled with a triangular
a single zone, gas prospect on the Gulf of Mexico shelf, which distribution in which the minimum and maximum values were
is close to existing infrastructure. An exploration well will be 5% and 50% respectively and the most likely value was 20%.
drilled to evaluate the structure. If the exploration well is The initial production rate of the wells was also modelled
successful then the prospect will be developed. There are three using a triangular distribution with minimum and maximum
possible development scenarios. If the gas volumes found by values of 5 MMcf/day and 50 MMcf/day respectively and a
the exploration well are ‘small’ then the development will most likely value of 20 MMcf/day. The initial rate is also
 VALUING AND COMPARING OIL AND GAS OPPORTUNITIES:
SPE 63201 A COMPARISON OF DECISION TREE AND SIMULATION METHODOLOGIES 3

predicted to have a strong positive relationship to the thickness estimated mean NPV versus threshold volume is shown. The
of net pay encountered, which is shown in Fig 4. The initial mean NPV for each threshold volume is estimated by
rate is defined as the annual averaged rate for the first year of assuming that all iterations with volumes below the threshold
production, or the initial field production rate. Hence, the large will have an NPV equal to the mean failure NPV.
range of initial rates shown in Fig. 4 is partially due to the fact The plot in Fig. 6 indicates that there is a range of
that in some cases the production is coming from more than threshold volumes over which the mean NPV is relatively
one well bore. The two trends seen in Fig 4 are caused by the insensitive to the precise threshold value. However, if the
presence of both subsea and platform development cases. The threshold is less than about 40 Bcf, the mean NPV begins to
upper limit of 200 MMcf/day is a facility limit. drop rapidly. This is very useful information that could be
There was also considerable uncertainty in the capital used to focus some of the geological and geophysical work on
expenditures. Each capital item in the model was assumed to the prospect. In the case presented in this paper, sufficient data
be triangularly distributed with a minimum 10% below the needs to be gathered to minimize likelihood of developing
most likely, and a maximum 50% above the most likely. It with a recoverable volume of less than 40 Bcf. The best
was also assumed that each capital expenditure was estimate of the threshold is around 60 Bcf, but any value
independent of all others. The range for each capital item may between 40 Bcf and 85 Bcf would not have much impact on
seem extreme, but translates to a P10 to P90 range of 2% below, the overall mean NPV. It should also be noted that, based on
to 33% above the most likely which in many cases is quite the limited amount of information that can be learned from a
realistic. Much less uncertainty was expected in the operating single well, it may not be possible to determine the likely
expenses and accordingly these were modelled with ± 5% recoverable volume to any better accuracy than 60 ± 20 Bcf.
uncertainty about the most likely value. A threshold volume of 60 Bcf has been used for this
evaluation. When the stochastic model is rerun with an
Economic Parameters. The focus of this paper is on economic cutoff of 60 Bcf the success case, or commercial,
comparing decision tree and simulation methods for analyzing recoverable volume distribution has P10, P50 and P90 values of
project economics and development options for the project. 68 Bcf, 131 Bcf and 385 Bcf respectively. The mean
Hence, price risk has not been included in this model and flat recoverable gas volume is 184 Bcf. The cumulative
oil and gas prices in real terms have been used. The price of recoverable volume distribution is shown in Fig. 7.
West Texas Intermediate oil is set at $18/bbl, and the Henry Once the economic threshold has been applied and the
Hub price for gas is assumed to be $2.30/MMBTU. There are stochastic model rerun, the economics for the prospect are
field-level offsets in the model of $1.95/bbl and $0.15/Mcf. In available. In particular, the mean NPV is MM$8.4, and the
addition, a quality adjustment of 1.035 MMBTU/Mcf has been P10, P50 and P90 NPV values are -MM$7, -MM$6 and MM$39
applied to the gas from the field. All prices are in real dollars. respectively. The probability of commercial success (i.e., a
volume above the economic threshold) is 19%, and probability
Economic Threshold of positive NPV is 17%. The cumulative NPV distribution is
The determination of the minimum economic volume, or shown in Fig. 8.
economic threshold, is carried out somewhat differently in a
stochastic model than is the case for a deterministic or Decision Tree Model
decision tree model. Initially, the stochastic model is run The four scenarios that are combined together for the decision
without any economic threshold so that any discovered tree comprise the failure, or dry hole, case and three success
volumes are developed. A cross plot of NPV versus cases. The three success cases correspond to two-well, four-
recoverable volume in the absence of an economic threshold is well and six-well developments for volumes equal to the P10,
shown in Fig. 5. Clearly, a significant number of iterations in P50 and P90 respectively of the recoverable volume
which development occurs are uneconomic. Iterations in distribution. The first order of business for the decision tree is
which geologic failure occurred are those with NPV’s of to determine the economic threshold. The economic threshold
around -MM$7 and no recoverable volume. in the decision tree method is usually determined by finding
If a threshold volume is now invoked any volumes less the volume for which the NPV of the P10, or minimal
than the threshold will cause the exploration well to be development, case is equal to zero. More correctly, for point
plugged and abandoned and no further investment will occur. forward economics, the volume should be that volume which
We can see from Fig. 5 that any iterations with volumes less results in a loss no greater than the dry hole cost. The result of
than a threshold will cause the NPV for that iteration to this determination is shown in Fig. 9 where the after tax NPV
change from the NPV for the development of that volume to for the minimum development is plotted against recoverable
the NPV for a failure case. In many cases the NPV for the gas volume. The threshold recoverable volume for the
failure case will be larger (less negative) than the NPV for the decision tree model is about 46 Bcf.
development case. The final step in choosing the threshold Once the threshold volume has been determined the
volume is to recognize that the optimum threshold volume overall probability of commercial success can be determined
should be that threshold which leads to the largest mean NPV. as the product of the geologic probability of success and the
The result of such a process is shown in Fig. 6, where a plot of probability of finding a volume above the economic threshold.
4 B.S. MUDFORD SPE 63201

For the decision tree model the probability of commercial success-side pathways, it is possible to use only three points
success is 22% and the P10, P50 and P90 volumes for the from each of the four distributions. The decision tree approach
commercial distribution are 51 Bcf, 106 Bcf and 352 Bcf breaks the problem into small parts, which can involve a lot of
respectively (c.f., Fig 7). The four-point decision tree for this manual work and is potentially error-prone. In contrast the
evaluation can now be built and is shown in Fig. 10. The mean simulation approach is to input the uncertainties into the
NPV for the four point decision tree is MM$8.7. In building constituent parts of the model and let the computer calculate
this decision tree we have used values for the initial gas the impact of that uncertainty. The user then needs to interpret
production rate and the costs equal to the most likely values the results.
supplied by the engineers, as is common in practice. The threshold volume in a decision tree model is
The difference between the results of the decision tree and deterministic number and, under the assumptions of the
stochastic economic computations are shown in Fig. 8 where model, any volumes greater than the threshold volume will
the cumulative probability distributions of after tax NPV are generate a positive NPV. On the other hand, in a stochastic
shown for both methods of analysis. A more detailed view of evaluation the threshold volume is a distribution. The fact that
the successful part of the cumulative probability distribution is the threshold volume is a distribution in a stochastic model
shown in Fig 8b. The mean, after tax NPV’s computed using gives rise to a fundamental difference between stochastic and
the two methods is similar, as are the P10 and P50 values. There the decision tree evaluations. Namely, that it is possible in the
is considerable difference between the P90 results of MM$39 stochastic model to have a volume above the economic
and MM$31 for the stochastic and decision tree models threshold that generates a negative NPV, much as can happen
respectively. Other major differences between the decision in the real world. It is generally sub-optimal to choose a
tree NPV distribution and the full NPV distribution will sufficiently high threshold volume that there is almost no
clearly occur between the modelled points in the decision tree chance of any development scenario with negative NPV, since
model. In particular, for outcomes between the P10 and P50 too many opportunities for investment will be prematurely
success case NPV’s the decision tree model is optimistic by as discarded. Hence, stochastic models have a probability of
much as MM$30. On the other hand, for outcomes between positive NPV in addition to the commercial success
the P50 and P90 success case NPV’s the decision tree model is probability. If the threshold volume has been chosen
pessimistic by as much as MM$40. Additionally, the decision appropriately these two probabilities will be close, but the
tree model cannot convey the true upside and downside commercial probability of success will always be higher than
volatility, since that is not part of the decision tree input data the probability of positive NPV.
or calculation. For the case analyzed in this paper, the failure
cost is -MM$5.6, which is larger than 63% of the NPV Optimum Development Scenarios
distribution calculated in the stochastic model. Four-point In the project being analyzed the optimum development
models, by their nature, can only be correct at four places in scenarios need to be determined. As was stated above, there
the NPV distribution, and will not provide the true downside are three potential development scenarios: a subsea
and upside to the project. development for ‘small’ volumes; a small platform
In order to bring the decision tree and Monte Carlo development for ‘medium’ volumes; and a large platform
simulation results into closer alignment with respect to the development for ‘large’ volumes. The values of ‘small’,
inherent volatility in the project economics, we would need to ‘medium’ and ‘large’ volumes can be determined by finding
capture the impact of much of the uncertainty present in the the volumes at which the development should be switched
Monte Carlo simulation model. In the stochastic model there from subsea completions to a small platform, and from a small
are three significant areas of uncertainty that are not captured platform to a large platform.
in the decision tree model. Firstly, in the stochastic model the For the stochastic model we can determine the critical
costs have significant uncertainty and their distribution is volumes at which development scenarios should be switched
strongly skewed with the mode, or most likely, falling at about by running three separate cases. The first case is one where all
the 16’th percentile. Secondly, the initial production rates in volumes are developed using subsea completions. For the
the stochastic model have large variability and are dependent second case all volumes are developed using a small platform.
upon the net pay. Thus, higher production rates will tend to Finally, for the third case all volumes are developed using the
occur when the net pay falls on the high side of its large platform. For the prospect being analyzed in this paper
distribution. The third area where uncertainty is introduced in the small platform will contain 10 well slots and the large
the stochastic model is the number of wells. The outcomes platform will contain 20 well slots. The results of all three
from the stochastic model are such that it is possible for any runs are shown in Fig. 11. It is apparent that, for this project,
commercial subsea development to require as many as six the small platform case is largely irrelevant. Over a wide range
wells. Whereas in the decision tree model the recoverability of volumes it is sufficient to develop using subsea
per well is fixed at about 28 Bcf/well for the P10 and P50 cases. completions. Once the subsea completions become sub-
In order to build a decision tree to investigate the four areas of optimal then the large platform development is at least as
variability outlined above (volumes, costs, rates and well economic as the small platform development and has the
recoverability), a decision tree with 81 (34) different success- added advantage of handling far more upside volume
side pathways would need to be modelled. Even with 81 potential.
 VALUING AND COMPARING OIL AND GAS OPPORTUNITIES:
SPE 63201 A COMPARISON OF DECISION TREE AND SIMULATION METHODOLOGIES 5

The critical volume at which the development should be dollar amounts as is traditionally done for decision tree
switched from subsea completions to the large platform models. Clearly, any element to which NPV is highly
development can be found in the following manner. For each correlated should be a strong driver. Fig. 13 shows a tornado
scenario we compute a running average of the NPV plot for the full data set on which we have shown the
distribution ranked according to recoverable volume. The correlation of after tax NPV with recoverable gas volume,
running averages are shown in Fig. 11, where it can be seen well recoverability, initial gas production rate and capital
that the critical volume is around 440 Bcf. costs. The correlation statistic plotted for each model element
in Fig. 13 is the standard correlation coefficient, which is
Sensitivities and Drivers defined as the ratio of the covariance between the model
A key component of the economic analysis of a prospect element and after tax NPV and the product of the standard
involves the investigation of sensitivities and the deviations of the model element and after tax NPV. The
determination of the economic drivers. It is in this area that tornado plot in Fig. 13 shows that recoverable volume is the
another important distinction emerges between decision tree main driver, which should be no surprise. At first glance it
and simulation methods. The customary way to show project also appears to imply that NPV is strongly positively
sensitivities and drivers in a decision tree evaluation is to correlated with capital cost. However, this is merely a
investigate the impact of variability in each of the main reflection of the strong effect of recoverable volumes, since
elements of a project (i.e., volume, well recoverability, initial finding greater recoverable volumes causes more capital to be
production rate and costs) on the overall NPV. Each element is spent to develop those volumes.
varied across some range while holding all other elements Of greater interest is what the drivers look like for P10, P50,
constant. The resulting variation of project NPV due to mean and P90 volumes. The correlations for all four volumes
variability in each of the elements is then shown on a tornado are shown in Fig. 14 where we have taken a subset of the
plot. A tornado plot for the current project is shown in Fig. 12. sampled points around the P10, P50, mean and P90 recoverable
To produce this plot we have assumed that the P50 case is the volumes and recomputed the correlations between NPV and
base case. The variability in the individual elements has been the model elements. It is now clear that the after tax NPV has
taken to be the same as was used in the uncertainty inputs in a strong, negative correlation to the capital, as is to be
the stochastic model, and is as follows. The recoverable gas expected. In addition, NPV has a strong, positive correlation
volumes have then been varied by -50% and +100% about the to well recoverability and initial production rate. It should be
P50 case. The well recoverability has been varied by ± 50% noted that correlations below 50% are usually not indicative of
about the P50 case. The initial, per-well production rate ranges a relationship, since at least half the variability must be
from –75% to +150% of the P50 case. Finally, the capital explained by random factors at this correlation level. Clearly,
expenditures range from –10% to +50% of the P50 case. by this standard, the impact of recoverable volumes has been
There are three major flaws with this approach. The first removed from the analysis. In addition, for recoverable
problem is that the method outlined above neglects any volumes at the P90 level, the after tax NPV is not strongly
dependencies between project elements. For example if the correlated to any of the possible drivers investigated here. If
volumes are high then it might be expected that the well large volumes are found, the project profitability is generally
recoverability, the initial production rates and the capital costs assured, irrespective of the costs or rates.
would all be linked to the volumes and move together. Thus While correlations between model elements do show
looking at the impact of variations on a one-off basis may not project sensitivities, they give no feeling for the monetary
give a good view of the true project sensitivities. The second impact of variations in model elements. In standard
flaw is that there may be too much dependency within a single deterministic, or decision tree analyses spider diagrams are
element. For example, it is probably unrealistic to expect all frequently used to show such details. Spider diagrams can also
capital costs to simultaneously fall by 10%, or rise by 30%. be constructed for a fully stochastic model in the following
The third major flaw is that the range of variability chosen for manner. The average after tax NPV is computed for the top
each element is often somewhat arbitrary, and there is an and bottom 10% of iterations for each model element. If there
implicit assumption that all variations within the range are is a strong relationship between a model element and the after
equally likely. tax NPV, then there will be a large range between the average
The dependencies and drivers in a stochastic model can be after tax NPV of the top and bottom 10% of iterations. A
analyzed in a single pass by looking at the results from the spider diagram can now be constructed in which the range of
base run. This is due to the fact that all variables of interest variability of model elements is fully consistent both between
have been set up as distributions, which should include the full model elements and with the initial variabilities entered into
range of variability expected for each variable. In addition, the model.
any dependencies between elements of the model should Spider diagrams have been constructed for subsets of the
already be present in the sampled data for the base model. full stochastic run around the P10, P50, mean and P90 volumes,
We begin by analyzing the full data set. A simple way to and these are shown in Figs. 15 to 18. For scenarios around
begin is to generate a tornado plot of correlations between the P10 volume (Fig. 15) the capital is the major driver. The
model elements and the after tax NPV, rather than absolute uncertainties applied to the capital expenditures lead to an
6 B.S. MUDFORD SPE 63201

overall variability in total capital ranging from –40% to +40% the case that uncertainties are included in the model on an
of the mean capital expenditure for recoverable volumes arbitrary basis that cannot be rigorously related to the
around the P10 volume. This variability in capital drives an uncertainty in the basic model elements. In addition, it is very
after tax NPV variation from MM$10 to –MM$15. Similar difficult to include dependencies between model elements
conclusions also hold for scenarios with recoverable volumes when analyzing sensitivities and drivers in a decision tree
close to the P50 (Fig, 16) and mean (Fig. 17) recoverable model. In contrast, dependencies are built into a stochastic
volumes. Again, capital is a strong negative driver, and well model from the beginning.
recoverability and initial production rate are strong positive
drivers. For the P90 recoverable volume case (Fig. 18) the Nomenclature
impact of model elements on the project after tax NPV is not P =probability of success or exceedance percentile
as strong as when smaller volumes are found.
Subscripts
Conclusions 10 = 10’th percentile
In this paper a relatively straightforward Gulf of Mexico, shelf 50 = 50’th percentile
gas development project has been analyzed using both 90 = 90’th percentile
decision tree and stochastic methods. There are numerous
differences in modelling methodologies and outputs between Acknowledgements
decision tree and stochastic evaluation techniques. Decision David Santley of Anadarko Petroleum Corporation is thanked
tree models are constructed by assigning probabilities to a few for his suggestion of the method used in this paper to generate
(four cases, for four-point economics) isolated scenarios. The a stochastic version of the spider diagram. In addition, I thank
project economics are then evaluated by combining together Ned Butler of Landmark Graphics for many useful discussions
the scenarios with their assigned probability weighting to around stochastic economics. This paper benefited from the
produce an EMV. Stochastic modelling begins by assigning reviews by Ian Beck, Bill Diggons and John Davidson, all of
probability distributions to input parameters, together with Landmark Graphics. Any errors that remain are my
dependencies between inputs. A simulation is then run to responsibility. The Monte Carlo simulations reported in this
determine output economic distributions such as after tax paper were performed with TERAS software, courtesy of
NPV. The mean of the NPV distribution from a stochastic Landmark Graphics.
model corresponds to the EMV from a decision tree model.
The main similarities and differences between the decision References
tree and stochastic methods are set out below 1. Harbaugh, J.W., Davis, J.C., and Wendebourg, J.: Computing
1. For the project analyzed in this paper the mean after tax Risk for Oil Prospects: Principles and Programs, Elsevier
NPV’s computed using the two methods are similar. However Science Ltd., Oxford, UK (1995) 452.
the degree of difference between the mean NPV determined 2. Simpson, G.S., Lamb, F.E., Finch, J.H. and Dinnie, N.C. “The
Application of Probabilistic and Qualitative Methods to Asset
using the two methods is project specific, and depends on the Management Decision Making”, paper SPE 59455 presented at
complexity of the project. Four-point models, by their nature, the 2000 Asia Pacific Conference on Integrated Modelling for
can only be correct at four places in the NPV distribution, and Asset Management, Yokohama, Japan, Apr. 25-26.
will not provide the true downside and upside to the project. 3. Gailli, A.G., Armstrong, M. and Jehl, B., “Comparison of Three
2. For a stochastic model, as in the real world, commercial Methods for Evaluating Oil Projects” JPT (October 1999) 44.
success (i.e., finding a volume larger than the economic 4. Gutleber, D.S., Heiberger, E.M. and Morris, T.D. “Simulation
threshold volume) does not guarantee positive NPV. Analysis for Integrated Evaluation of Technical and
3. Management decisions can be analyzed as easily using Commercial Risk” JPT (December 1995) 1062.
stochastic models, as can be done using decision tree models. 5. Murtha, J.A. “Monte Carlo Simulation: Its Status and Future”
JPT (April 1997) 361.
The advantage of stochastic models is that the impact of 6. Moore, L.R. and Mudford, B.S. “Probabilistic Play Analysis
engineering and development uncertainty is an integral part of From Geoscience to Economics: An Example From the Gulf of
the decision analysis. Mexico”, paper SPE 52955 presented at the 1999 SPE
4. Project sensitivities and drivers can be analyzed in an Hydrocarbon Economics and Evaluation Symposium, Dallas,
internally consistent manner using stochastic models. This is TX, Mar. 20-23.
in sharp contrast to decision tree models where it is frequently
 VALUING AND COMPARING OIL AND GAS OPPORTUNITIES:
SPE 63201 A COMPARISON OF DECISION TREE AND SIMULATION METHODOLOGIES 7

a) b) c) d)
100

80
C u m u la tive P ro b a b ility (% )

60

40

20

0
0 1 2 3 4 0 200 400 200 500 800 0 200 400
A re a (M acre ) N e t P a y (ft) R f (M cf/a cre ft) V o lu m e (B cf)

Fig. 1 --- Cumulative probability distributions for: a) area; b) net

pay; c) recovery factor; and d) recoverable gas volume.

50 0
80

40 0
W ell R ecoverability (B cf/w ell)

60
N et P ay (ft)

30 0

40
20 0

10 0 20

0
0
0 10 00 20 00 30 00 40 00 50 00
A rea (acre) 0 200 400 600
N et P ay (ft)
Fig. 2 --- Net pay versus area, where each point is the result from
a single iteration in the Monte Carlo simulation. Fig. 3 --- Well recoverability versus net pay, where each point is
the result from a single iteration in the Monte Carlo simulation.
8 B.S. MUDFORD SPE 63201

200 100

C um ulative P robability (% )
160 80
Initial R ate (M M cf/day)

120 60

80 40

40 20

0 0
0 200 400 600 0 200 400 600 800
N et P ay (ft) R ecoverable G as (B cf)

Fig. 4 --- Average first year production rate versus net pay, where Fig. 7 --- Cumulative probability distribution for technically
each point is the result from a single iteration in the Monte Carlo recoverable gas volumes above the threshold volume.
simulation.

100 a) b)
100 25
75
S to ch a stic
A fter T ax N P V (M M $)

Cumulative Probability (%)

Cumulative Probability (%) 80 20
50 4 -P o in t

60 15
25

0 40 10

-2 5 20 5

-5 0 0 0
0 40 80 120 160 200 0 100 200 0 100 200
R ecoverable G as (B cf) A fter T ax N P V (M M $) A fter T ax N P V (M M $)

Fig. 5 --- After tax NPV versus recoverable gas with no economic Fig. 8 --- Cumulative probability distributions of after tax NPV for
threshold, where each point is the result from a single iteration in stochastic and decision tree models. a) Full cumulative
the Monte Carlo simulation. distribution and b) the commercially successful part of the
distribution

8 4
M e a n A fte r T a x N P V (M M $ )

6
0
A fter T ax N P V (M M $)

F a ilu re C o st
4

-4
2

-8 T h res h o ld V o lu m e
0

-2 -1 2
0 40 80 120 160 200 40 44 48 52 56 60
T hreshold V olum e (B cf) T hres hold V olum e (B c f)

Fig. 6 --- Estimated mean after tax NPV versus threshold volume. Fig. 9 --- Decision tree after tax NPV versus recoverable volume
for a minimal development, showing the threshold volume.
 VALUING AND COMPARING OIL AND GAS OPPORTUNITIES:
SPE 63201 A COMPARISON OF DECISION TREE AND SIMULATION METHODOLOGIES 9

R ecovera ble V olum e

In itial P rod u ction R a te

W ell P rod ucibility

C apita l E xp end itu re

Fig. 10 --- Four-point decision tree for the project being analyzed
in this paper.
0 0.2 0.4 0.6 0.8 1
400 A fte r T a x N P V C o rrelation
B o u nd ary b e tw e e n s ub se a
a n d p la tfo rm d e v e lo pm e n ts Fig. 13 --- Stochastic tornado plot showing the correlation of after
300 tax NPV to recoverable volumes, well recoverability, initial
A fter T ax N P V (M M $)

production rates and capital expenditures.

200 P 1 0 V o lum e
P 5 0 V o lum e
M ea n V olum e
100 P 9 0 V o lum e
C apita l E xpen ditu re

Subsea
0 S m a ll P la tfo rm
L a rg e P la tfo rm
W ell P rodu cibility

-1 0 0
Initial P ro du ctio n R ate
0 200 400 600 800
R ecoverable G as (B cf)

Fig. 11 --- After tax NPV versus recoverable volume for subsea R ecoverable V o lum e
completion, small platform and large platform development
scenarios. The NPV is computed by taking a running average of
the NPV distribution ranked according to recoverable volume.
-1 -0.5 0 0.5 1
R ecovera ble V olum e A fter T ax N P V C orrelation

Fig. 14 --- Stochastic tornado plots for subsets of the data around
the P10, P50, mean and P90 success case volumes. The correlation
Initial P rod uction R ate of after tax NPV to recoverable volumes, well recoverability, initial
production rates and capital expenditures is plotted.

C apital E xpend itu re

W ell P roducibility B ase C ase

P 50 A T N P V

-40 0 40 80 120
A fter T ax N P V (M M $)

Fig. 12 --- Decision tree tornado plot showing the impact on after
tax NPV of variability in recoverable volumes, well recoverability,
initial production rates and capital expenditures.
10 B.S. MUDFORD SPE 63201

20 95 R e cove rab le G a s
R eco vera ble G a s
V ariation about M ean A T N P V (M M $)

In itia l P rod uctio n

V ariation about M ean A T N P V (M M $)

In itia l P ro du ctio n
C a pita l
C ap ital
W e ll P rod u cib ility
W ell P rodu cibility 90
10

85
0

80

-1 0
75

-2 0 70
-0 .8 -0 .4 0 0 .4 0 .8 -0 .4 -0 .2 0 0.2 0.4
V ariation A bout M ean P aram eter V alue V ariation A bout M ean P aram eter V alue
Fig. 15 --- Spider diagram for recoverable volumes around the P10
Fig. 17 --- Spider diagram for recoverable volumes around the
success case volume, showing the impact on mean after tax NPV
mean success case volume, showing the impact on mean after
of variations in recoverable volumes, well recoverability, initial
tax NPV of variations in recoverable volumes, well recoverability,
production rates and capital expenditures is plotted.
initial production rates and capital expenditures is plotted.

65 20 4
V ariation about M ean A T N P V (M M $)

V ariation about M ean A T N P V (M M $)

60 20 0

55 19 6
R e cove rab le G a s
In itia l P rod uctio n
50 C a pita l 19 2
W e ll P rod u cib ility
R e co ve ra b le G a s
45 18 8 In itia l P ro d u ctio n
C a p ita l
W e ll P ro d u c ib ility

40 18 4
-0 .4 -0 .2 0 0.2 0.4 0.6 0.8 -0.4 -0.2 0 0.2 0.4
V ariation A bout M ean P aram eter V alue V ariation A bout M ean P aram eter V alue
Fig. 16 --- Spider diagram for recoverable volumes around the P50 Fig. 18 --- Spider diagram for recoverable volumes around the P90
success case volume, showing the impact on mean after tax NPV success case volume, showing the impact on mean after tax NPV
of variations in recoverable volumes, well recoverability, initial of variations in recoverable volumes, well recoverability, initial
production rates and capital expenditures is plotted. production rates and capital expenditures is plotted.