SPE 69594

Significance of Project Risking Methods on Portfolio Optimization Models

P.A. Tyler, Schlumberger GeoQuest, J.R. McVean, Merak Projects
Copyright 2001, Society of Petroleum Engineers Inc.
This paper was prepared for presentation at the SPE Latin American and Caribbean Petroleum
Engineering Conference held in Buenos Aires, Argentina, 2528 March 2001.
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
presented, have not been reviewed by the Society of Petroleum Engineers and are subject to
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
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
for commercial purposes 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 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.

Abstract
Efficient frontier theory is gaining acceptance as a part of the
portfolio analysis process in the petroleum industry. One of
the major difficulties companies encounter in creating
corporate efficient frontiers is in representing project level
risks in a corporate consolidation of value. Full stochastic
evaluations require the management of huge quantities of data.
As a result several current solutions use multiple discrete
outcomes to represent any given project, rather then a
complete stochastic distribution. This paper evaluates the
differences resulting from using these two approaches to
project risking. By examining the optimized portfolio results
and the resultant efficient frontier, we are able to draw direct
conclusions about the added value attributable to detailed
Monte Carlo based evaluations. Since price is the source of
most correlation between projects, the investigation was done
with and without price as an input variable. It was discovered
that the differences between the stochastic and discrete
outcome evaluations are not significant when price uncertainty
is neglected, but the inclusion of price leads to significant
differences if the resulting correlation is not accounted for.
Introduction
Petroleum managers are constantly faced with the decision
of how to invest limited amounts of capital in order to
maximize shareholder value or return. This is usually done by
evaluating all the available investments, and then selecting a
subset in which to invest. This subset is often referred to as
the companys portfolio. Selection of this portfolio is critical
to a companys success and therefore it requires significant
consideration.
The goal of the portfolio selection process is to select the
optimal set of projects. However, this is not a simple

process, and selecting the optimal portfolio while staying

within the corporate strategy and constraints can become a
very daunting exercise. The task of matching project selection
to the company strategy and goals is often referred to as
Portfolio Management. There are many elements that must be
taken into account in portfolio management, including.
Maximizing the value of the portfolio.
Living within the capital spending limits.
Meeting the production requirements.
Achieving both short term and long term cash flow
goals.
Matching forecasted net income targets.
Meeting developmental and/or environmental
constraints.
This is even more complex in our industry where many
projects have significant amounts of uncertainty, or variation
in the possible outcomes. In recent years more and more
attention has been placed on how to represent these
uncertainties within the portfolio evaluation. These
uncertainties exist in various forms within the investment
projects of a petroleum company, and can include.
Existence of hydrocarbons (probability of success).
Geological reservoir properties.
Timing and extent of the development program.
Capital and operating costs.
Oil and gas prices.
These uncertainties in the input data required to make an
economic evaluation of E&P projects lead to uncertainties in
the economic results. Acknowledging that these uncertainties
exist in the individual projects, naturally leads to the concept
that uncertainties exist in the overall portfolio. In recent years
a significant amount of effort has been spent in trying to
define this portfolio risk and to compare portfolios by their
risk vs. reward relationship.
In light of the importance of portfolio level risk, this paper
compares two differing approaches to project evaluation under
uncertain conditions, and the effect the differing approaches
have on the overall portfolio risk measurements. The two
approaches compared will be an uncertainty tree approach
where projects are evaluated as distinct outcomes each with a
differing probability of occurrence (referred to here as simple
stochastic), and a full Monte Carlo evaluation approach
where the projects are evaluated across the full range of input
uncertainties.

P.A. TYLER, J.R. MCVEAN

Theory and Definitions

In a 1952 article, Noble laureate Harry Markowitz explained
the theory of portfolio diversification from the perspective of
minimizing risk. His work completely changed the field of
portfolio optimization (Markowitz, 1952). There are three
basic principles that need to be understood prior to any
discussion of efficient frontier theory.
A rational investor will prefer more value to less value,
but will also prefer less risk to more risk.
A portfolio can be more or less than the sum of its
individual projects, depending on how the investments
interact with each other.
The interaction between
investments is as important as the individual investments
themselves.
There is no single optimal portfolio as it is generally
possible to gain more value by accepting more risk.
This last point is essentially an expression of the concept
of efficient portfolios. A portfolio is said to be efficient if it is
the case that no other portfolio has more value while having
the same or less risk, and there is no other portfolio that has
less risk while having the same or more value. In order to
evaluate the efficiency of any given portfolio it is typically
plotted on an efficient frontier graph (Fig. 1). This graph is a
plot of several portfolios, each with its own value and risk
measurements. Points to the right (higher value) and below
(lower risk) are said to be more efficient portfolios than
portfolios that lie to the left or above. The collection of
portfolios that lie on the lower right edge of this graph are
referred to as the efficient frontier. For the given strategy and
goal, there are no other portfolios that generate the same value
for less risk.

Fig. 1An illustration of the efficient frontier graph. Portfolio A is

dominated by all portfolios in quadrant II (portfolios B and C) and
dominates all portfolios in quadrant IV. Portfolio D is neither
dominated nor dominates portfolio E because E is in Ds first
quadrant. Portfolios B, C, D, and E are all efficient because they
have no portfolios in their second quadrant.

SPE 69594

This paper will focus on a semi-standard deviation

definition of portfolio risk in order to simplify the analysis.
Semi-standard deviation is a popular measure of downside
uncertainty or risk, and is commonly used by many who are
employing portfolio management in the petroleum industry.
The use of simple stochastic data at the project level leads
to smooth risk profiles at the portfolio level. As the discrete
outcomes are combined in larger and larger numbers of
projects, the discreteness of the outcomes is lost. A portfolio
that has ten projects, each with four discrete outcomes,
contains slightly more then one million possible outcomes. To
illustrate this, the NPV distribution for one of the mid range
portfolios (LV 1) used in this paper, is shown with both the
full Monte Carlo based data (Fig. 2) and the simple stochastic
data
(Fig. 3).

Fig. 2NPV distribution of the portfolio LV 1 from the full Monte

Carlo based optimization.

Fig. 3NPV distribution of the portfolio LV 1 from the simple

stochastic based optimization.

Project Generation
In order to evaluate the differences that the project risk
evaluation method has on portfolio risk, we needed to start
with a typical group of projects. In our case we used thirty
projects based on real evaluations from differing locations

SPE 69594

SIGNIFICANCE OF PROJECT RISKING METHODS ON PORTFOLIO OPTIMIZATION MODELS

around the world. The projects included typical producing

assets, acquisitions and a large proportion of exploration
projects (Table 1). The projects do not however represent
any single company or country, and the values have been
modified slightly to ensure anonymity.
Each project was fully evaluated using a probabilistic
reservoir simulator, and the resultant distributions were used
as the basis for all evaluations. Deterministic development
profiles were created for the most likely successful outcomes.
For the exploration projects; failure, low and high
deterministic evaluations were also created, and the
probability of each outcome was derived from the simulator
results.
The combination of these creates a simple
stochastic evaluation of the projects. These can be
represented by multiple branch uncertainty trees (Fig. 4), and
could have included as many outcomes as required to describe
the development options.

the capital development costs were scaled around the

individual outcomes using a historically validated function.
This leads to a far more robust and complex evaluation of the
individual projects.
Typical results from four of the
exploration projects evaluated are shown in Fig. 5.
Incorporation of Price Uncertainties
One of the issues with portfolio risk evaluations is that if
you need to include price uncertainty, it usually needs to be
fully correlated across all the projects, unlike most technical or
reservoir uncertainty. This can be achieved in Monte Carlo
simulations by applying a single price distribution to the entire
set of projects. This global application of a price uncertainty
allows fully correlated prices across multiple projects. This is
easily accomplished in many of todays Monte Carlo
simulating systems, but adds a considerable complexity to
uncertainty tree analyses.
In order to illustrate the importance of this correlation, a
second set of project data was generated in which the Monte
Carlo evaluated projects used a fully correlated price
distribution, while the simple stochastic evaluations used a
typical 3-branch price uncertainty (Fig. 6).

Fig. 4An example of the typical simple stochastic evaluation of

an exploration project used in the risked portfolio optimization.
The probabilities are derived from a detailed analysis of the
inherent reservoir uncertainties, and the independent scenarios
came from the development engineering group.
Fig. 6An example of the typical simple stochastic evaluation
including the uncertainty associated with price.
This is a
schematic representation, where price uncertainty effects all end
nodes.

Fig. 5Four examples of project risk profiles that are markedly

different from normal distributions. These risk profile shapes are
typical of exploration projects.

In the case of the full Monte Carlo evaluations, the same

specific development assumptions were made as in the simple
stochastic evaluations. However, the entire distribution of
outcomes from the reservoir simulator was incorporated, and

It is possible to account for price-induced inter-project

correlation using a modified simple stochastic approach.
Including the 3-branch uncertainty node, to roughly model
price uncertainty, increases the number of possible outcomes
from four to twelve. When combining projects into portfolios,
perfect price correlation can be achieved by insuring that only
project outcomes of similar price assumptions are combined.
In other words, the high price outcomes of each project can
only be combined with high price outcomes of other projects.
This technique, while requiring more project outcome
evaluations and considerable care in the portfolio generation
stage, may prove to be sufficient to capture the broad affects
of inter-project correlation, provided price is the only
significant contributor to this correlation. For the purposes of
this paper, this approach was not attempted but further study
of this topic is clearly warranted.

P.A. TYLER, J.R. MCVEAN

SPE 69594

Portfolio Optimization Process

In order to evaluate the effect that these two approaches had
on the portfolio risk measurement, an efficient frontier was
developed for each of the four project evaluations.
Full Monte Carlo evaluations without price
uncertainty.
Simple stochastic evaluations without price
uncertainty.
Full Monte Carlo evaluations with fully correlated
price uncertainty.
Simple stochastic evaluations with price uncertainty
without correlation.
This was accomplished by defining a strategy that
consisted of the following requirements.
A minimum annual produced oil volume and gas
volume was required from year 2003 to 2010.
Annual exploration capital and production capital
expenditure was limited for all years
Cumulative reserve replacement over the first five
years and over the first ten years were specified.
Only portfolios that met all of the above goals were
included in the efficient frontier graph. In this case, the
projects could either be included or not. The model was not
permitted to vary the percent working interest in any of the
projects. This was representative of the majority of projects
we were working with, and is often how some companies
evaluate portfolio selection.
The optimizer objective value was set to maximize NPV at
a 10% discount rate, or to minimize the semi-standard
deviation of NPV at a 10% discount rate, our risk measure.
By varying the minimum required NPV for a given
optimization, and optimizing on minimum possible risk,
specific portfolios from the set of the most efficient portfolios
found were noted. Randomly generated portfolios were used
to supplement the efficient portfolios, in order to populate the
efficient frontier graph. In total over fifteen optimizations
were run for each graph.
The exact same optimization procedure was performed for
all sets of input data (Figs. 7 to 10), and the efficient frontiers
were compared for both project selection and overall risk
measures.

evaluations were selected. They are labeled on Fig. 7, and

their characteristics are listed in Table 1. The same portfolios
(groups of projects) were then identified in the simple
stochastic efficient frontier graph (Fig. 8). As can be seen
most of the portfolios remained efficient or did not shift
markedly from the efficient frontier in the simple stochastic
optimizations. This indicates the differences in the evaluation
detail did not significantly effect the portfolio selection or
portfolio risk comparisons, when price uncertainty was
ignored.

Results
Using the full Monte Carlo evaluations, the resultant portfolio
NPVs ranged from MM$1,510 to MM$3,070, while the
associated risk ranged from MM$225 to MM$660 (Fig. 7).
While using the simple stochastic evaluations, the resultant
portfolio NPVs ranged from MM$1,510 to MM$3,070, and
the associated risk ranged from MM$200 to MM$490 (Fig. 8).
The identical NPV values are no surprise as the expected
values of the simple stochastic evaluations matched the mean
values (expected) from the full Monte Carlo evaluations
(Table 2).
In order to compare the optimized project selection
between the two project risking methods, fourteen of the most
efficient portfolios found based on the full Monte Carlo

Fig. 8The efficient frontier graph based on the simple stochastic

project evaluations, without price uncertainty. Note that most of
the portfolios identified as efficient in Fig. 7 are efficient or close
to efficient here.

Fig. 7The optimized efficient frontier graph based on the full

Monte Carlo based project evaluations without price uncertainty.
The highlighted efficient portfolios were chosen for comparison
with the simple stochastic based efficient frontier.

The same procedure was carried out using the projects

evaluated with price uncertainty. Using the full Monte Carlo
evaluations with correlated price uncertainty, the resultant
portfolio NPVs ranged from MM$1,565 to MM$3,040, while
the associated risk ranged from MM$610 to MM$1,000 (Fig.
9).
Using the simple stochastic evaluations with non
correlated price uncertainty, the resultant portfolio NPVs
ranged from MM$1,560 to MM$3,040, and the associated risk
ranged from MM$310 to MM$590 (Fig. 10).

SPE 69594

SIGNIFICANCE OF PROJECT RISKING METHODS ON PORTFOLIO OPTIMIZATION MODELS

The difference in the risk measurement is a significant

factor in this case. Including correlated price uncertainty in the
full Monte Carlo based project evaluations leads to a
significant increase in the range of possible values for
resulting portfolios. This increase in the width of the portfolio
risk profiles manifests itself as larger semi-standard
deviations. In the case of the simple stochastic approach with
price uncertainty, the lack of price correlation means that this
affect is not observed.
In order to compare the optimized project selection and the
effect of price correlation, 14 of the most efficient portfolios
found based on the full Monte Carlo with correlated price
evaluations were selected. They are labeled on Fig. 9. The
same portfolios (groups of projects) were then identified in the
simple stochastic efficient frontier graph with non correlated
price uncertainty (Fig. 10). In this case only four of the
fourteen portfolios are still efficient using the simple
stochastic project data with non correlated prices. The
portfolios are close to the efficient frontier, indicating that the
reduced evaluation detail still has some value.

Fig. 9The optimized efficient frontier graph based on the full

Monte Carlo based project evaluations, including fully correlated
price uncertainty.
The highlighted efficient portfolios were
chosen for comparison with the simple stochastic based efficient
frontier.

The simple stochastic approach to project evaluation

inevitably leads to the loss of inter-project correlations.
However, in many cases there is little cause for inter-project
correlation, with the notable exception of price. As we have
seen, the loss of this correlation can significantly affect the
understanding of the level of risk involved with the resulting
portfolios, although the selection of optimal portfolios may not
be affected as significantly

Fig. 10The efficient frontier graph based on the simple

stochastic project evaluations, including non correlated price
uncertainty. Note location of the Portfolios, from Fig. 9, which
were efficient in the full Monte Carlo efficient frontier with
correlated price uncertainty.

Conclusions
The method of describing risk at the individual project level, is
a factor in the portfolio optimization and analysis process.
Using full Monte Carlo based evaluations does lead to
improved risk measurement and more accurate portfolio
analysis.
Using a relatively simple uncertainty tree approach to
evaluate the projects will lead to similar portfolio optimization
results when using risk measures as the main selection criteria,
with significantly less effort in the creation of the project
evaluations.
The incorporation of price uncertainty leads to large
differences in the two approaches due to the effects of the
correlation. The authors feel more investigation is warranted
in the effects of incorporating correlated prices in the simple
stochastic evaluations.
This paper firmly demonstrates that companies can start
down the road of risked based portfolio optimization and
analysis without having to put together detailed Monte Carlo
based evaluations for every project. A mix of the approaches
can be used to perform risk based portfolio optimization and
analysis, allowing companies to use the detailed data where
available, without having to force everyone into long and
detailed analysis of every project.
References
1. Markowitz, H.: Portfolio Selection, The Journal of Finance,
Vol. VII, No. 1, (1952) 77-91.
2. McVean, J.R..: The Significance of Risk Definition on Portfolio
Selection, SPE 62966, 2000 SPE annual technical conference
and exhibition in Dallas, Texas, 1-4 Oct, 2000.

P.A. TYLER, J.R. MCVEAN

Appendix AProject Characteristics

Appendix BPortfolio Properties

TABLE 1 Expected Value Project Characteristics

Name
EP 1
EP 2
EP 3
EP 4
EP 5
EP 6
EP 7
EP 8
EP 9
EP 10
EP 11
EP 12
EP 13
EP 14
EP 15
EP 16
EP 17
EP 18
EP 19
EP 20
EP 21
EP 22
EP 23
EP 24
EP 25
EP 26
EP 27
EP 28
EP 29
EP 30

AT Cash
NPV 10%
(MM$)
127
102
200
251
367
68
90
77
125
128
80
85
30
92
51
47
8
7
5
6
106
38
110
148
184
50
118
187
283
23

Capital
Undisc.
(MM$)
142
62
133
119
345
75
137
28
88
204
71
121
86
95
31
96
21
20
9
5
452
182
352
42
387
73
143
82
69
51

Oil Prod.

Type

(MMSTB)

56
50
182
212
47
57
88
48
44
111
80
85
38
66
21
47
11
10
7
6
182
79
138
28
237
46
103
32
39
39

SPE 69594

TABLE 2 Portfolio Properties, w/o price uncertainty

Probability
of Success
(%)

Acq.
Dev.
Dev.
Dev.
Acq.
Del.
Expl.
Expl.
Expl.
Expl.
Del.
Expl.
Expl.
Dev.
Dev.
Expl.
Expl.
Expl.
Del.
Del.
Expl.
Expl.
Expl.
Expl.
Dev.
Expl.
Del.
Expl.
Del.
Expl.

90
72
35
41
17
94
30
27
40
30
20
95
90
30
60
23
43
80
90
15
90
25

Name

Max V
HV - 1
HV 2
HV 3
HV 4
AV 1
AV 2
AV 3
AV 4
LV 1
LV 2
LV 3
LV 4
Min R

AT Cash
NPV 10%
(MM$)
3,070
2,860
2,796
2,690
2,613
2,552
2,420
2,352
2,240
2,186
2,034
1,889
1,771
1,510

Semi-Standard
deviation based
on Full
Monte Carlo
Inputs
(MM$)
660
639
611
565
540
523
460
430
403
380
362
337
311
225

Number
of
Projects
25
22
23
23
25
21
24
24
23
22
20
21
19
19

Semi-Standard
deviation based
on Simple
Stochastic
Inputs
(MM$)
490
477
485
442
378
377
350
315
310
285
276
255
225
200