SPE 124180

Practical Portfolio Simulation: Determining the Precision of a Probability

Distribution of a Large Asset Portfolio When the Underlying Project
Economics are Captured by a Small Number of Discrete Scenarios
B.J.A. Willigers, Palantir Economic Solutions Ltd

Copyright 2009, Society of Petroleum Engineers

This paper was prepared for presentation at the 2009 SPE Annual Technical Conference and Exhibition held in New Orleans, Louisiana, USA, 4–7 October 2009.

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 have not been
reviewed by the Society of Petroleum Engineers and are subject to correction by the author(s). The material does not necessarily reflect any position of the Society of Petroleum Engineers, its
officers, or members. Electronic reproduction, distribution, or storage of any part of this paper without the written consent of the Society of Petroleum Engineers is 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 acknowledgment of SPE copyright.

Abstract
One of the great successes of modern financial risk modelling is the application of computing technology to simulate complex
continuous probability distributions associated with the value metrics of real-life assets. The Monte Carlo simulation technique
is the best known example and is widely applied by economists in the E&P industry. However a Monte Carlo simulation in
which project economics are aggregated into a large asset portfolio is rarely undertaken. The main reason is that present
computing systems cannot handle the vast amounts of data generated in a Monte Carlo simulation of a large asset portfolio.
A pragmatic solution for this issue might be to approximate the continuous distribution of feasible project outcomes using a
small number of probability-weighted discrete scenarios. In a corporate portfolio simulation, a project sample would be drawn
from these discrete distributions as opposed to the original continuous distributions. If either a global assumption needs
revision or a single asset requires recalculating, the economics of a relatively small number of scenarios would be computed.
Thus, there is no need to store or recalculate a large number of outcomes (typically more then 2,000) for each project, as would
be required in a conventional Monte Carlo simulation.
A prerequisite for this approach is that the frequently skewed and complex continuous probability distributions of each of the
assets can precisely be described by a small number of scenarios. This study documents the level of precision that can be
achieved using Swanson’s rule and variations thereof. Our analyses suggest that if the P50 is at least 33% greater than the P10,
the simulated portfolio mean and standard deviation are within 5% and 15% of the actual values, respectively. The
approximation of the lower end of the distribution, i.e., the downside risk of a portfolio, is within 4% even for an extremely
asymmetric lognormal distribution.
The proposed portfolio simulation methodology addresses the largely unmet need of corporate managers to improve their
understanding of key risk and value drivers and their impact on the performance of a corporate asset portfolio.

1. Introduction
In a 2006 survey by Booz, Allen and Hamilton, leaders of 20 Oil & Gas companies indicated a high demand to improve risk
management processes and that the industry is dissatisfied with current procedures (McKenna et al., 2006). McKinsey
interviewed over 1,000 corporate directors, 76% of whom stated that they want to spend more time on strategy and risk
management (Roberts, 2005). Only 11% of the directors indicated that they have a satisfactory understanding of the risks their
companies currently bear, whereas 50% stated that they lack the data to track corporate risk exposure over time.
This lack of insight into the effect of unforeseen events on corporate performance can be catastrophic for those companies
and can be partially attributed to the techniques used to determine the economic value of the company’s assets.
The Boston Consulting Group reported that many energy companies ignore uncertainty and risk and that project decisions
are based on deterministic evaluations (Balagopal and Gilliland, 2005). Clearly, if the economic impact of uncertainty is not
assessed at a project level, this information and the relevant insights cannot be communicated to corporate decision makers.
Although the Oil and Gas industry has started to realize the value of enhanced financial modelling (Bickel and Bratvold,
2007), most efforts to improve decision-making processes have focussed on project management. Risking procedures such as
decision tree modelling or Monte Carlo simulation are routinely applied in project management but are rarely applied at a
corporate portfolio level. As a consequence, corporate decision making tends to be suboptimal.
Even though computing technology has been the cornerstone of portfolio management for several decades, present
computing systems cannot handle the vast amounts of data generated in a Monte Carlo simulation of a large number of assets.
2 SPE 124180

In this study, we discuss a pragmatic approach to developing a risked portfolio of E&P assets. The precision of this
approach is assessed through experiments.

2. Probabilistic economic analysis

2.1 Deterministic versus probabilistic analysis

Over the past decade, the upstream Oil and Gas industry has gradually accepted the application of probabilistic economic
modelling techniques in place of traditional deterministic project analysis (Bickel and Bratvold, 2007). Probabilistic modelling
avoids single-point estimates for key model variables. Instead, it enters a probability-weighted spectrum of potential outcomes
into an economic cash-flow model. Probabilistic modelling provides insight into the range of feasible outcomes and creates a
much improved understanding on the expected economic performance of a project. The probabilistic results can be translated
in statements like: “There is a 60% chance this project will meet our financial target” or “We are 45% certain that this
investment will yield a positive return on our investment”.

2.2 Discrete versus continuous probability distributions

Uncertainties follow either a discrete distribution or a continuous probability distribution. The former involves a finite number
of states, each of which may be allocated a certain probability. A classic example in Oil and Gas exploration is the uncertainty
whether an exploration well will demonstrate the presence of hydrocarbons. In this example, there are two possible outcomes,
each having a certain probability. Many uncertainties have an indefinite number of outcomes, e.g., the size of an oil prospect
or the future gas spot price. Such uncertainties can be described only by continuous probability distributions.

2.3 Analysing by approximating

2.3.1 Monte Carlo simulation
Modelling a cash flow that depends on uncertainties that follow a continuous distribution is a key challenge faced by
economists, practitioners, and academics. The valuation models applied to real-life assets tend to be too complex to be solved
1
by closed-form solutions. A rare exception is the exact closed-form solution by Black and Scholes (1973) to value American-
type options.
An alternative approach is to approximate the “true” solution using simulation techniques such as Monte Carlo simulation.
The calculation flow in a Monte Carlo model differs from that used in a deterministic model, in that the former consists of
many iterations, each of which considers a new random sample taken from probability distributions that describe each of the
simulated uncertainties. The value metrics will converge to the “true” solution after a large number of iterations.
2.3.2 Swanson’s rule
An effective Monte Carlo simulation requires a large number of iterations to obtain an adequate approximation of the “real”
solution of the analyzed problem. The vast amounts of data this generates cannot be stored or processed using present
computing systems.
A possible solution is to reduce the amount of required data by approximating a continuous distribution using a discrete
probability function. The discrete scenarios are generally referred to as a “P-percentiles.” The P notation refers to the
probability that an uncertain variable has a value less than the specified value, e.g., the uncertainty has a 90% probability of
being higher than the P90 number.
Swanson’s rule has been widely applied in the E&P industry to assign a probability weighting to P10, P50, and P90
percentiles (Megill, 1984; Rose and Thompson, 1992). Swanson initially proposed it in 1972 in an internal Exxon
memorandum (Hurst et al., 2000). Swanson argued that (1) although geoscientists and managers find it difficult to quantify
extreme outcomes, they are reasonably confident in defining P10 and P90 outcomes, and (2) most relevant probability
distributions in E&P modelling are lognormal and that consequently the projects’ expected values and medians are different.
Following Swanson’s rule, the P10, P50, and P90 scenarios are probability weighted at 30%, 40%, and 30%, respectively.
Swanson’s rule preserves the mean and the variance of both normal distributions (see appendix 2 for proof.) and modestly
skewed lognormal distributions.
2.3.3 Extended Pearson-Tukey method
The precision of Swanson’s rule declines when the standard deviation of a lognormal distribution increases. The consequence
of this loss in precision depends on the relative contribution of the upper-end tails of a lognormal distribution to its mean and
standard deviation. A simple method to improve the precision of Swanson’s rule is to sample more extreme values of the
probability distribution. The Extended Pearson-Tukey method, proposed by Keefer and Bodily (1983), provides a probability
weighting for the P5, P50, and P95 of 19%, 62%, and 19%, respectively (see appendix 1 for proof; see also Smith, 1993) This
2
probability weighting is valid for lognormal distributions as well as normal distributions .

1
A mathematical formulation that yields an exact solution to an investment problem
2
The precision of Swanson rule is not affected by the standard deviation is the underlying uncertainty follows a normal
distribution
SPE 124180 3

3. Methods

3.1 Project and portfolio simulation

The development of the portfolio consists of two distinct phases: an initial project valuation and a subsequent aggregation of
the project value into a portfolio assessment. The geological success rate is the probability that an exploration well will
indicate the presence of hydrocarbons. Commercial uncertainty refers to that regarding the quantity of hydrocarbon associated
with an asset, given that hydrocarbon is known to be present (Rose, 2001). Geological risk is modelled as a Bernoulli
distribution that yields either a successful or a non-successful outcome. The simulation of commercial risk is achieved by one
of three approaches: (1) Monte Carlo simulation, (2) Swanson’s rule, or (3) Extended Pearson-Tukey method (Figure 1).
The results generated in the Monte Carlo simulation of commercial risk are used to assess the precision of the results
generated by the other valuation methodologies.
In the portfolio simulation, random samples taken from each of the feasible project outcomes are consolidated into a
portfolio value. The results of the simulation can be translated into a probability distribution that describes the spectrum of
attainable portfolio values. In principle this problem could be solved through the development of a decision tree. However
when simulating 30 projects with four possible outcome each a tree with 810,000 (=304) end nodes would have been required
and trees of this size are difficult or impossible to model.
Swanson’s rule Extended Pearson-Tukey

30% - P10 19% - P5

Exploration Exploration
Success Success
40% - P50 62% - P50

30% - P90 19% - P95

Exploration Exploration
Failure Failure

Monte Carlo simulation

100%
Exploration
Cumulative probability

80%
Success
60%

40%

20%

0%
0 500 1000 1500
Value

Exploration Failure
Figure 1. The project valuation methods simulate (1) the exploration risk as a Bernoulli distribution that yields either a successful or a
non-successful outcome and (2) commercial uncertainty as either discrete with three possible outcomes (Swanson’s rule and
Extended Pearson-Tukey method) or a continuous lognormal distribution (Monte Carlo simulation).

In the hybrid Monte Carlo – Swanson’s rule method, the 10 largest assets representing 70% of the total expected portfolio
value are modelled with the full Monte Carlo method, and the 20 remaining assets are modelled using Swanson’s rule.
3.1.1 Sum of moments
In the “sum of moments” method, each of the first three moments (mean, variance, and skewing) of the individual assets are
respectively summed across the assets to obtain the three moments of the overall project portfolio (Howard, 1971; Stuart and
Ord, 1987). On the basis of the first three moments, a lognormal portfolio distribution is fitted using a simple optimization
routine.
4 SPE 124180

3.1.2 Simulation method

All simulations were performed in a Microsoft Excel© environment and were run for 20,000 iterations. In order to ensure
optimal accuracy for the simulated results, the portfolio distributions generated by the different methodologies are based on the
same set of random numbers.
3.1.3 Valuation methodologies analyzed
In summary, we analyzed five valuation methodologies:
a) Swanson’s rule
b) Extended Pearson-Tukey method
c) Monte Carlo simulation
d) Hybrid Monte Carlo – Swanson’s rule method
e) Sum of moments

3.2 Project and portfolio description

A portfolio of 30 exploration assets was investigated. Two geological risk scenarios were simulated. In the first assessment,
the geological chance of success was set at 50% and was independent from the results of the other exploration targets in the
project portfolio. In the second portfolio assessment, it was assumed that the success rate of the second exploration target is
conditional on the outcome of the first exploration campaign and that the probability of success of all subsequent targets is
dependent on the outcome of the first and second targets (Figure 2). The ranking of the exploration targets, i.e., which target to
drill first, is based on the expected asset values.

First exploration Second exploration Remaining exploration targets

target target
Success 73% Success 85%

Success 50% Failure 15%

Failure 27% Success 50%

Failure 50%

Success 27% Success 50%

Failure 50% Failure 50%

Failure 73% Success 15%

Failure 85%
Figure 2. Conditional probabilities of geological success. The chance of success of the second exploration target is dependent on the
outcome of the first exploration target. The chance of success of any subsequent exploration target is dependent on the outcome of
the previous exploration campaigns.

Upon successful completion of the exploration phase, the asset is exposed to commercial risk, for which the value
distributions are assumed to be lognormal. The effect of the asymmetry of the project value probability distributions was
investigated by defining two asset portfolios. One portfolio consisted of projects with highly skewed value probability
distributions, whereas assets in a second portfolio were modestly skewed. All projects in the skewed portfolio yield P10-P50
and P50-P90 ratios of 6; those in the symmetric portfolio yield ratios of 1.47. 3 Two cases of relative value distribution of
assets across an appraisal portfolio were assessed. In the first case, all assets have a value of 1000; in the second case, the
portfolio consists of a small number of high-value assets and a large number of relatively low-value assets (Figure 3). The
relative value distribution in the latter scenario follows Lavalette’s rule (Lavalette, 1996).

3
For example, P50=100 and P50/P90=10 implies P10=1000 and P90=10.
SPE 124180 5

7000
6000 P10
P50
5000
P90
4000
Value

Expected value
3000
2000
1000
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
Asset number

Figure 3. An appraisal portfolio that consists of 30 assets with a wide range of expected values. The six largest assets constitute 53%
of the expected portfolio value. The probability distribution of the individual assets is highly skewed.

3.3 Portfolio scenarios

A detailed assessment of the different portfolio-risking techniques was made by investigating five portfolio scenarios (Table
1). The analyzed portfolios differ in (1) the value distribution of the assets across the portfolio, (2) the degree of asymmetry of
the probability distribution of the projects, (3) the exposure to geological risk, and (4) the degree of correlation of geological
risk between the different assets.

Portfolio scenario Skewed portfolioa Skewing of projectsb Geological riskc

Geological risk
dependenciesc
1 no modest no no
2 yes modest no no
3 yes strong no no
4 yes strong yes no
5 yes strong yes yes
Table 1. Five portfolio scenarios. aIn the skewed portfolio, the portfolio value distribution follows Lavalette’s rule
(Figure 3). bModestly skewed projects yield P10-P50 and P50-P90 value ratios of 1.47, whereas strongly-skewed
projects yield ratios of 6. cGeological risk equals 50%, given that no geological dependencies exist across the
assets. The conditional chances of success applied in scenario 5 are listed in Figure 2.

4. Results

4.1 Asset valuation

The cumulative probability distributions of two projects were developed using Monte Carlo simulation, Swanson’s rule, and
Extended Pearson-Tukey method (Figure 4). The projects differ only in their P10-P50 ratio, and both projects have a median
value of 1000 and are exposed to a geological chance of failure of 50%. The mean and standard deviation of the project with a
P10-P50 ratio of 1.47 can be precisely reproduced by the three methodologies. 4 The application of Swanson’s rule to the
project with a high asymmetry, i.e., a P10-P50 ratio of 6, results in mean and standard deviation that are significantly lower
than those yielded by the Monte Carlo simulation. Although Extended Pearson-Tukey method yields a standard deviation that
is lower than that of the Monte Carlo simulation, the means yielded by the two methods are similar (Table 2). The results also
suggest that Swanson’s rule can be successfully combined with a discrete fourth scenario.

4
The P50-P10 and P90-P50 ratios are assumed to be equal.
6 SPE 124180

100%
Cumulative probability

80%

60%

40%

20%

0%
0 5000 10000 15000
Value
100%
Cumulative probability

Series1 Series2 Series3

80%
60%
40%
20%
0%
0 1000 2000 3000
Value
Monte Carlo Swanson's rule Extended Pearson‐Tukey
Figure 4. Cumulative probability distributions of two projects. The upper and lower panels correspond to projects with a P10-P50 ratio
of 6 and 1.47, respectively.

P10-P50 ratio Value metric Monte Carlo Swanson's rule Extended Pearson-Tukey method
6 Mean 1268 1068 1249
Standard deviation 3813 2014 2823
1.47 Mean 524 520 520
Standard deviation 577 561 565

Table 2. Value metrics yielded by two distinct projects evaluated by different valuation techniques.

4.2 Portfolio valuation

4.2.1 Portfolio scenarios
The results of the portfolio analysis are shown in Figure 5 and Table 3. The results of portfolio analyses 1 and 2 confirm that
all five methodologies yield indistinguishable portfolio value distributions when the probability distributions of the individual
projects are modestly skewed. In portfolio analyses 3, 4, and 5, in which project value distribution is highly skewed,
Swanson’s rule fails to provide precise value estimates. Deviations are particularly pronounced for the P10 estimates. The P10
and expected value yielded by Swanson’s rule in scenario 5 are 30% and 17% lower than by Monte Carlo simulation,
respectively. However, Swanson’s rule does provide a precise estimate of the lower segment of the distribution.
The “sum of moment” method shows a significant deviation from the Monte Carlo simulation in portfolio analyses 3 and 5.
This method shows particularly poor precision at the lower end of the distribution in portfolio analysis 5, in which the P90
estimate is over four times as large as the Monte Carlo result. However, the method does provide a very precise estimate on
the portfolio value distribution in scenario 2.
The precision of Extended Pearson-Tukey method and the hybrid Monte Carlo – Swanson’s rule methodology is high in all
five portfolio analyses. Extended Pearson-Tukey method yields a very precise estimate for the portfolio mean even when the
SPE 124180 7

project value distributions are highly skewed. Although the hybrid Monte Carlo – Swanson’s rule methodology yields the most
precise value distribution in scenario 5, the level of precision is comparable to Extended Pearson-Tukey method.

120%

Cumulative probability
100%
80%
60%
Portfolio analysis 1 40%
20%
0%
25000 27000 29000 31000 33000 35000 37000 39000
Value
100%
Cumulative probability

80% Monte Carlo Swanson's rule

60% adjusted Swanson's rule hybrid MC‐Swanson's rule
Portfolio analysis 2 40%
sum of moments
20%
0%
4500 5000 5500 6000 6500 7000 7500 8000
Value
100%
Cumulative probability

80% Monte Carlo Swanson's rule

60%
adjusted Swanson's rule hybrid MC‐Swanson's rule
Portfolio analysis 3 40%
sum of moments
20%
0%
0 10000 20000 30000 40000
Value
100%
Cumulative probability

80% Monte Carlo Swanson's rule

60% adjusted Swanson's rule hybrid MC‐Swanson's rule
Portfolio analysis 4 40%
sum of moments
20%
0%
0 5000 10000 15000 20000 25000
Value
100%
Cumulative probability

80% Monte Carlo Swanson's rule

60% adjusted Swanson's rule hybrid MC‐Swanson's rule
Portfolio analysis 5 40%
sum of moments
20%
0%
0 5000 10000 15000 20000 25000

Value

Monte Carlo Swanson's rule
Extended Pearson‐Tukey Hybrid MC‐Swanson's  rule
Sum of moments

Figure 5. Cumulative probability plots that show the results of the portfolio analysis.
8 SPE 124180

Extended Pearson- Hybrid MC-

Value metric Monte Carlo Swanson's rule Tukey method Swanson's rule Sum of Moments
Portfolio scenario 1
P10 34416 34180 34573 34247 34381
P50 31371 31347 31383 31374 31350
P90 29155 29152 29187 29134 29176
Average 31400 31382 31446 31388 31400
Portfolio scenario 2
P10 7167 7072 7173 7165 7153
P50 6254 6257 6240 6245 6256
P90 5642 5623 5666 5636 5637
Average 6277 6259 6284 6273 6277
Portfolio scenario 3
P10 31654 20302 25344 30259 33993
P50 13705 13154 14487 12944 13695
P90 8248 8500 8529 8186 6745
Average 15935 13448 15203 15158 15935
Portfolio scenario 4
P10 19989 13066 16489 17928 21322
P50 6332 6397 6559 5933 5946
P90 2732 2936 2802 2813 2198
Average 8013 6870 7619 7547 8013

Portfolio scenario 5
P10 22238 16271 20169 21341 23826
P50 6003 6064 6282 5886 5483
P90 459 493 425 479 1746
Average 8134 6810 7663 7743 8134

Table 3. A summary of the results of the portfolio analysis. Numbers in red are discussed in the text.

4.2.2 Project asymmetry and valuation methodology

The impact of asymmetry of the project value probability distribution on the precision of the portfolio valuation methodology
was further assessed using the basic assumptions used in portfolio scenario 5, except that portfolios were developed with
underlying assets that have a wide range of P10-P50 ratios. The results shown in Figure 6 were normalized to the Monte Carlo
simulation results, and values smaller than one indicate that the corresponding methodology yields a smaller value than does
the Monte Carlo simulation.
The results illustrate that for a P10-P50 ratio that is less than 5, Swanson’s rule approximates the project’s mean within
10% and standard deviation within 35%. The approximation of a portfolio whose underlying assets have the same P10-P50
ratio yields a mean within 10% and a standard deviation within 25%. For a P10-P50 of 3, the mean is within 5% and the
standard deviation is within 15%. The loss of precision in the assessment of a portfolio with asymmetric distributed underlying
assets is dominated by the relatively poor approximation of the upper part of the portfolio distribution. The approximation of
the lower part of lognormal distributions is very precise even for highly skewed distributions. The approximation of the
portfolio P90 is within 4% for a P10-P50 ratio of 5.
For a highly skewed project or a portfolio of such projects, Swanson’s rule yields quite low estimates for the expected
value, P10 value, and standard deviation. However, significant improvements are achievable by the application of Extended
Pearson-Tukey method. When value metrics have lognormal distribution and a P10-P50 ratio less than 5, a project’s mean can
be estimated within 2% and standard deviation within 20%. If the probability distributions can be described by a normal
distribution, the precision of Swanson’s rule does not decrease when standard deviation is increased.
Given a P10-P50 ratio of 5, the hybrid Monte Carlo - Swanson’s rule method approximates the mean, standard deviation,
and P10 within 5% and the P90 within 7%.
SPE 124180 9

The expected value and the standard deviation yielded by the sum of moments method are by definition equal to the mean
of the Monte Carlo simulation. 5 The P10 estimate is within 5% for a P10-P50 ratio of 5, but the P90 value is overestimated by
350%.

1.05
1
0.95
Normalised mean

Normalised mean
0.9
0.85
0.8
0.75
0.7
0.65
0.6
0 2 4 6 8 10 12
1.2 P10‐P50 ratio
Norm standard deviation 1 Series1 Series2 Series3 Series4
Normalised standard deviation
0.8

0.6

0.4

0.2

0
0 2 4 6 8 10 12
1.1 P10‐P50 ratio
1 Series1 Series2 Series3 Series4
Normalised P10
Normalised P10

0.9

0.8

0.7

0.6
0 2 4 6 8 10 12
P10‐P50 ratio
4.1
3.6
Swanson's rule Adjusted Swanson's rule
Normalised P90

Normalised P90 3.1
Hybrid MC‐Swanson's rule Sum of moments
2.6
2.1
1.6
1.1
0.6
0 2 4 6 8 10 12

P10‐P50 ratio

Swanson's rule Extended Pearson‐Tukey
Hybrid MC‐Swanson's  rule Sum of moments

Figure 6. Portfolio value metrics, normalized to the Monte Carlo simulation results, as a function of project value spread. The portfolio
value spread is expressed as the ratio between the P10 and P50 of the individual projects, the latter applying to all projects within the
portfolio.

5
The mean and standard deviation of the Monte Carlo simulation results are used to define the sum of moments probability
distribution.
10 SPE 124180

5. Discussion

5.1 The precision of Swanson’s rule

The analysis executed in this study confirms the finding of Hurst et al. (2000) that Swanson’s rule provides a precise estimate
of a modestly skewed lognormal distribution. It was also found that Swanson’s rule progressively underestimates values in the
upper part of the probability distribution, including their mean and standard deviation. The lower end of the distribution,
however, is accurately approximated even for highly skewed lognormal distributions.
The precision of Swanson’s rule is not reduced with increasing standard deviation if the probability distributions can be
described by a normal distribution.
The hybrid Monte Carlo - Swanson’s rule method yields the best approximation of a full Monte Carlo simulation. It
involves applying the Monte Carlo method to the high-value assets and Swanson’s rule to the low-value assets. This method
could be used as a compromise between Swanson’s rule and a conventional Monte Carlo simulation, especially if the overall
portfolio value is highly concentrated in a limited number of projects that follow complex probability distributions that cannot
be reasonably approximated with a discrete probability distribution. The application of the hybrid Monte Carlo - Swanson’s
rule method would reduce the requirement on data processing and storage during a portfolio simulation.

5.2 The precision of extended Pearson-Tukey method

In a detailed study Smith (1993) demonstrated the precision that can be achieved by application of the extended Pearson-
Tukey method. The distributions of many uncertainties are much more skewed than the lognormal distribution with a P10-P50
ratio of 1.11 that was analysed by Smith (1993). The examples of lognormal distributions discussed by Rose (2001) yield P10-
P50 ratios in excess of 2.5 and the 2008 swings in oil prices point to a much higher asymmetry than assumed by Smith (1993).
On the basis of the results reported in this study we conclude that the Extended Pearson-Tukey method yields more precise
results than Swanson’s rule especially for highly skewed lognormal distribution.
The Extended Pearson-Tukey method has the drawback that more extreme outcomes are used in the calculation and it has
been reported in several studies that individuals find it difficult to accurately visualize and quantify extreme project outcomes
(Goodwin and Wright, 2004; Hora, 2007).

5.3 Sum of moments

The potential to apply the sum of moments method to evaluate a real-life corporate portfolio is limited. The method can only
be used when no dependencies exist between the assets within the portfolio (Howard, 1971; Stuart and Ord, 1987), which
applies to few E&P portfolios.
The shortcomings of the sum of moments technique have been illustrated in this study by applying the approach to a
portfolio that consists of assets whose chance of geological success is conditional on the outcome of other assets within the
portfolio. Figure 6 shows that sum of moments fails to precisely describe the lower end of the cumulative probability
distribution.

5.4 Probability distributions and the central limit theorem

The lognormal distribution is a frequently used probability distribution in E&P economic analysis. Many input variables used
in economic models are the result of multiplying a series of independent random variables, and the central limit theorem states
that such a product has a lognormal probability distribution.
A lognormal distribution with a high standard deviation will yield high extreme values, which have a pronounced effect on
the mean of the distribution. Several authors (e.g., Rose, 2001) commented that those extreme values are frequently
geologically unrealistic and that windfall taxes often prevent companies from capitalizing on such scenarios. As a
consequence, probability distributions of real-life assets are unlikely to be as highly skewed as some of the distributions
simulated in this study, although they may be more complex.
The central limit theorem also states that summing a large number of lognormal distributions will result in a normal
distribution. However, in many real-life portfolios, the contribution of a small number of assets will dominate the overall
portfolio value. If these projects follow a lognormal distribution, the portfolio will not have a normal distribution.

5.5 Precision of Monte Carlo simulation and other valuation techniques

This study implicitly assumed that Monte Carlo simulation yields “true” solutions for the different experiments. In reality,
Monte Carlo simulation generates an approximation of the true solution. In the context of valuation methodologies, the terms
“precise” and “precision” refer to the discrepancy between the results from the specific method under consideration and the
corresponding results of the Monte Carlo simulation.

6. Conclusions
A large unmet need exists for techniques and tools that can generate insight into the probability-weighted range of portfolio
performance. The processing and storage of large volumes of data that are generated in a portfolio simulation is a challenge in
modern corporate financial management. The approach proposed in this study vastly reduces the amount of data to be
SPE 124180 11

managed by replacing the continuous probability distributions for individual assets with a small number of probability-
weighted discrete scenarios.
The modelled project scenarios include one failure scenario of zero value and three positive scenarios. The latter three
scenarios are probability weighted according Swanson’s rule or a variation thereof. We have demonstrated that the proposed
methodology generates precise results in the vast majority of cases. The portfolio mean can generally be estimated within 10%
and the standard deviation within 25%. If the underlying assets have a P10-P50 ratio of 3, the portfolio simulation yields a
precision of 5% for the mean and 15% for the standard deviation. The approximation of the lower end of the distribution, i.e.,
the downside risk of a portfolio, is within 4% even for an extremely asymmetric lognormal distribution. For situations where
the precision of the approximation is insufficient, we suggested several approaches to improve it. The simulation of the P5,
P50, and P95 in Extended Pearson-Tukey method significantly improves the precision of the approximation. A hybrid
approach, in which the few high-value assets are modelled in a Monte Carlo simulation while the remaining assets are
modelled using Swanson’s rule, greatly enhances the precision of the simulation and significantly reduces the computing
resources required.

7. References
Balagopal, B. and Gilliland, G. 2005. Integrating value and risk in portfolio strategy for energy companies.
http://www.bcg.com/impact_expertise/publications/files/Integrating_Value_Risk_Energy_Companies_Sep2005.pdf
Bickel, J. E. and Bratvold, R. B. 2007. Decision making in the Oil and Gas industry: From blissful ignorance to uncertainty induced
confusion. SPE 109610.
Black, F. and Scholes, M. 1973. The pricing of options and corporate liabilities. Journal of Political Economy 81 637-654.
Goodwind, P. and Wright, G. 2004. Decision analysis for management judgement, Third edition, John Wiley & Sons, Ltd.
Hora, S. C. 2007. Eliciting probabilities from experts. In Advances in decision analysis, edited by Edwards, W., Miles Jr., R. F. and
vonWinterfeldt, D., Cambridge University press, 129-153.
Howard, R. A. 1971. Proximal decision analysis. Management Science 17 (9) Reprinted in Howard, R. A. and Matheson, J. E. (Eds.) (1984)
Readings on the principles and applications of decision analysis, 2, Strategic decisions group, Menlo Park, CA.
Hurst, A., Brown, G. C. and Swanson, R. I. 2000. Swanson’s 30-40-30 rule. AAPG Bulletin 84 (12) 1883-1891.
Keefer, D.L. and Bodily, S.E. 1983 3-point approximation for continuous random variables. Management Science 29 595-609
Lavalette D. 1996. Facteur d’impact: impartialité ou impuissance? Internal Report, INSERM U350, Institut Curie - Recherche, Bât. 112,
Centre Universitaire, 91405 Orsay, France, http://www.curie.u-psud.fr/U350/
McKenna, M. G., Wilczynski, H. and VanderSchee D. 2006. Capital project execution in the Oil and Gas industry.
http://www.boozallen.com/media/file/Capital_Project_Execution.pdf
Megill, R. E. 1984. An introduction to risk analysis, Second edition: Tulsa, Pennwell Publishing.
Roberts, T. 2005. Appetite for destruction: Ensuring sound risk management and governance.
http://www.accaglobal.com/pubs/publicinterest/activities/library/governance/presentations/2705952/241105_TimRoberts.pdf
Rose, P. R. 2001. Risk analysis and management of petroleum exploration ventures. AAPG methods in exploration series 12.
Rose, P. R. and Thompson, R. S. 1992. Economics and risk assessment. In Morton-Thompson, D. and Woods, A. H., eds., Development
geology reference manual. AAPG Methods in exploration series 10 21-56.
Smith J.E. 1993 Moment methods for decision analysis. Management Science 39 340-358
Stuart, A. and Ord, J. K. 1987. Kendall’s advanced theory of statistics, Vol. 1, Distribution theory. Oxford Publications, New York.

Appendix 1: Theoretical justification of Extended Pearson-Tukey method

Swanson’s rule can be justified by equation (1). (See Hurst et al. [2000] for details.) This formula yields a = 0.304 with a
choice of of 40 (as in Swanson’s rule) and (0.1) =1.282.

(1)

= number of percentage points difference between the minimum / maximum and the median
= probability
= Standard deviation of log(X)
= Probability P(Z> )
Assuming = 0.19 and (0.05) = (0.95) = 1.62, Extended Pearson-Tukey method can be developed in which the
P5, P50, and P95 are weighted at 19%, 62%, and 19%, respectively. The theoretical validity of Extended Pearson-Tukey
method equals that of the original formulation, given that it is based on the same mathematical formulation.

Appendix 2: Theoretical justification of the application of Swanson’s rule to normal distributions

The variance of a population is given by
∑
∑
(2)
= population mean
= the probability of occurrence of
Equation (2) can be rewritten as
(3)
12 SPE 124180

For normal distributions, and 0, hence (3) can be rewritten as

(4)
For standard normal distributions, where 1 and 1.282, equation (4) can be solved to
0.3048. This result suggests a probability weighting of 30%, 40%, and 30% for the P10, P50, and P90, respectively, and
demonstrates that Swanson’s rule also applies to normal distributions.
Substituting for in equation (4), where 1.62, yields 0.19, which is the same
result developed in Appendix 1 for Extended Pearson-Tukey method.
The approximation of a normal distribution using Swanson’s rule will preserve both the mean and the standard deviation of
the original continuous distribution.