Content uploaded by Stephen X. Zhang
Author content
All content in this area was uploaded by Stephen X. Zhang on Oct 02, 2018
Content may be subject to copyright.
Elsevier Editorial System(tm) for Decision Support Systems
Manuscript Draft
Manuscript Number:
Title: An Evolutionary Real Options Framework for Design and Management of Large-Scale
Projects and Systems with Multiple Real Options
Article Type: Full Length Article
Keywords: Uncertainty, Multiple Real Options, Evolutionary Algorithms, Decision Support, Dynamic
Strategic Planning
Corresponding Author: Mr Stephen Zhang,
Corresponding Author's Institution:
First Author: Stephen Zhang
Order of Authors: Stephen Zhang; Vladan Babovic
Stephen ZHANG is a doctoral student at the Department of Engineering and Technology
Management at National University of Singapore and Singapore Delft Water Alliance (SDWA).
He had prior work experiences in semiconductor and management consulting industries. His
research aims to understand how flexibilities can be evaluated and designed in large-scale
projects and systems to adapt with risks. And he had published papers on that topic in peer
reviewed international journals and conferences.
Vladan BABOVIC is an associate professor at National University of Singapore. Prior to joining
NUS, he was Head of Emerging Technologies at Danish Hydraulic Institute and Senior Research
Scientist at WL | Delft Hydraulics. Dr. Babovic presently serves as the director of Singapore Delft
Water Alliance (SDWA) and the Editor-in-Chief for Journal of Hydroinformatics. His research
interests include hydroinformatics, artificial intelligence, machine learning, data mining and
knowledge discovery, data assimilation, real options, and knowledge management. He has
authored and co-authored more than 40 book chapters and papers in peer reviewed international
journals.
Biographical Note of Stephen Zhang And Vladan Babovic
An Evolutionary Real Options Framework for Design and Management of
Large-Scale Projects and Systems with Multiple Real Options
STEPHEN ZHANG* AND VLADAN BABOVIC
Singapore Delft Water Alliance & Department of Civil Engineering,
National University of Singapore,
E1-08-25, Engineering Drive, S117576, Singapore
*Corresponding author.
Tel: 65-65168304,
Fax: 65-67780617,
Email: stephen.zhang@nus.edu.sg
* Manuscript of An Evolutionary Real Option Framework
Click here to view linked References
2
ABSTRACT
This paper addresses the issue of decision support for designing large-scale complex projects and
systems in face of uncertainties, drawing on and integrating real option valuation, genetic algorithms,
Monte Carlo simulations and several other techniques. The proposed evolutionary real option
framework searches for the optimized portfolio of real options and makes adaptive plans to address
uncertainties as future unfolds. Exemplified by a case of designing Maritime Domain Protection
system, the evolutionary framework compares favorably with the traditional Net Present Value
method, and delivers considerable improvements over the prevailing real option methods.
Keywords: Uncertainty, Multiple Real Options, Evolutionary Algorithms, Decision Support,
Dynamic Strategic Planning
3
1 INTRODUCTION
Managers appear to suffer from an “illusion of control” [19] and tend to adopt a single scenario
for the future, come up with a peak design, and compute a single performance measure for a project.
The importance of interdependent options embedded in projects is not always fully appreciated by
managers [2]. Nevertheless, embedding multiple real options and exercising them based on how
future unfolds is a very important means to deal with uncertainties in the design and management of
large-scale projects and systems [29]. Decision making that involves paths of uncertainties together
with planning and exercising of multiple options is complex, and due to the complexity, organizations
often fail in practice to follow a well-structured, accountable and reproducible decision-making
process for assessing and selecting a strategy for project implementation [18]. Consequently a new
systemic modeling philosophy and approach, such as dynamic strategic planning, is required [12].
For example, as illustrated in an example of designing a Maritime Domain Protection (MDP)
system [9], the degree of terrorism in the next decade or longer is uncertain and renders the MDP
system requirements uncertain as well. Therefore, it is desirable to design a flexible system that is
able to adapt according to how terrorism could possibly evolve. This poses a complex practical
challenge to the design and valuation process, because the numbers of pathways along which
uncertainty may unfold and system can adapt are both very large. Indeed, more often than not, a
portfolio of real options are present in a project or a system simultaneously, and the size of the
portfolio and the levels of uncertainty generate a complex structure of project pay-offs where each
option may alter the boundary conditions of other options [22].
The value of a combination of real options is not the combined value of each option in isolation
[30]. To valuate a portfolio of real options, Anand [1] analyzed the determinants to explain the
portfolio effects, and Baldwin and Clark [4] calculated the option value of modules in modular
architecture theoretically. However, it remains unclear how to quantitatively assess a portfolio of real
options when the number of interacting real options becomes large and the design space becomes non-
4
linear and discontinuous. No conventional real option methodologies are able to holistically valuate
and select multiple interdependent real options and their exercising rules in complex projects and
systems.
To deal with that issue, this paper proposes an integrated framework to quantitatively assess and
optimize complex projects with multiple real options through utilizing Evolutionary Algorithms
(EAs). EAs are generic multi-objective population-based metaheuristic optimization techniques,
inspired by biological evolution [34]. One of such techniques, Genetic Algorithms, are identified to be
suitable to solve very large, path dependent and non-convex problems [3; 16]. GAs can solve complex
problems which are difficult to study analytically or by conventional real options techniques. The
conventional assumptions of variables being independent and identically distributed (i.i.d.) are
relaxed: essentially, one just needs to formulate a GAs problem specifying the objectives and an
environment for their evaluation.
Dias [13] and Lazo et al. [21] have proposed using genetic algorithms to find the exercise regions
regarding to price and time for real options in oil field developments, and recently Hassan et al. [17]
have used it in aircraft design. Following the same direction, this paper develops an evolutionary real
option framework to use GAs to composite and valuate a portfolio of real options and formulates an
overall dynamic strategy plan.
This paper illustrates the framework through the example of designing a Maritime Domain
Protection system, described by [9], which detailed the application of real options to MDP system and
compared the real option design with reference to the classical system engineering design [25]. The
classical system engineering approach that developed the MDP peak design is very generic and
widely adopted. The approach sets an objective hierarchy and based on that subsystems are built in a
modular architecture.
This study inherits the same set of architectural principles and technical modules from MDP study
[25], while applies the evolutionary real option framework as the overarching decision support for
system valuation and selection. In this fashion, the paper examines the added-value of the
5
evolutionary real option framework -specifically from its effectiveness and efficiency in both finding
the suitable pieces of real options and formulating the dynamic adaptive plan to increase the project
value.
2 EVOLUTIONARY REAL OPTION FRAMEWORK
The proposed evolutionary real option framework aims to provide a means to incorporate and
evaluate flexibilities in the design and management of systems and projects. Such flexibilities are
specifically known as “real options” by an analogy with financial options.
Formally, in the sense of both financial and real options, the definition of an option is a right but
not an obligation to take certain actions some time1. Options can be an opportunity to invest (a call
option) or a chance to exit a bad situation (a put option). To act on the option is to “exercise” the
option. The option has a time premium or holding value, because it can derive values from certain
realizations of uncertainties.
Options, practically, can be purely contractual in monetary terms, which are the basis of financial
options, or physical entities embedded in a system or a project, known as real options. Financial
options are traded in markets, while real options are specific to individual project and system. A
classical example of a “real option” is the sizing of the foundation of a parking garage so that
additional floors can be added at a later date if a large demand materializes [33].
The valuation and design of flexibility can use the methods of real option analysis [31], derived
from modern financial option theory. Whereas, the complexity and uniqueness of real options in many
large-scale complex systems [7] render the formulas suitable for financial options inapplicable. New
frameworks need to be developed to reflect the complexity and uniqueness of the ways real options
are embedded and valuated in projects and systems [12; 14; 26].
The real option system design framework recognizes and addresses uncertainties by incorporating
flexibility into the design, operation, and management of system, and that approach changes
1Only the basic real option concepts that are directly used in this paper are introduced here. Please see an article
covering an overview of the topic, e.g. [31].
6
fundamentally the processes of system design - the way managers valuate system, the kinds of
elements embedded into system, and how risk management is practices.
Figure 1. Peak design process, prevailing real option design, and evolutionary real option design process
The traditional peak design practice, the cut real option best practice, and the proposed
evolutionary real option design process are illustrated and compared above in figure 1. Once the
system and its objectives are defined, the traditional system engineering proceeds to optimize its
design based on a set of fixed specifications.
Current real option best practices (circled by a light blue dotted lines in figure 1), even though
vary from each other in terminologies and details, consist of a few essential steps [10; 23; 24; 28; 32].
This study distills the following steps from those state-of-art frameworks. The system design process
starts by identifications of key risk drivers, possible project states, and suitable options in those states.
Then expert opinions or available data are used to estimate thresholds for exercising the real options.
Finally the systems with and without those real options are valuated, with the key risk drivers taken
7
into account, by Monte-Carlo simulations in some cases to handle the non-linearity and discontinuity
[6; 12].
The proposed evolutionary real option framework (circled by a darker blue dotted lines in figure
1) encompasses the current real option best practice, and extends it by deploying evolutionary
algorithms to select the portfolio of options and under which conditions they are exercised.
Moreover, system designers can use GAs or other search techniques to discover new options and
exercising rules, if they can help in increasing system value. Some real options may be deleted had
they been deemed not to be useful and exercised in any scenarios in the modelling (the cost of owning
that option outweights the benefit). Thereby, the real option representation is redefined, and of course
system designers, policy makers, and managers have to verify the feasibility and validity of the new
real option representations in the design and business environment. Such a refinement process is
iterative between the system designer and the quantitative evolutionary real option valuation. In this
manner, the cognition tasks of system designers can be better concentrated on the creation of design
alternatives and options, leaving the optimization and valuation of systems and real options to the
rigorous evolutionary real option framework.
The value of real options is derived based on the established approach of comparing the value of
systems with and without incorporated flexibilities [8; 11]. In this fashion, different design approaches
can be compared.
3 AN EXAMPLE APPLICATON: MDP SYSTEM
Terrorism threats in the Straits of Malacca, one of the world’s most important shipping lanes,
require an appropriate Maritime Domain Protection (MDP) system. The architecting and design of
this system, especially to prevent a Weapons of Mass Destruction (WMD) attack, “comprises so many
unknown variables that traditional cost-benefit analysis is rendered nearly impossible” [27]. Terrorism
8
attacks occur irregularly with little predictability. Often, the frequency of occurrences and the impact
or consequences of an event are defined to constitute risk as:
risk = probability (statistical frequency) × consequences (monetary
terms). (1)
The probabilities and the consequences of occurrences are both highly uncertain and
unpredictable in the coming years. Yet a terrorism system must be conceived and designed today in
order to prevent acts of terrorism for a decade or more. A potential solution is to incorporate
flexibility into system design for the reason that the flexibility embedded can be exercised in future as
uncertainty unfolds and updated information arrives.
This study uses the same technical and design configurations and evaluation method as employed
by NPS study [25], which carried out an extensive traditional system engineering process to develop a
MDP system to prevent and defeat terrorism in the Straits of Malacca.
The NPS study selected a modular architecture for MDP system, which consists of five different
subsystems (sensors, C3I, force, land and sea inspection), each of which contains two or three
alternative designs. Various combinations of the alternative subsystems result in various overarching
MDP system. 109 such combinations are considered feasible system configurations. The
performances and costs were determined for each of these system configurations under a forecasted
degree of maritime terrorism. Configurations with higher performances (resulting in lower risk of a
successful attack) and lower costs are considered as a better design. The best of them is referred to as
the peak design. The real option design approach, on the contrast, does not believe the level of
terrorism is certain but identifies a flexible system and subsequent strategies which adapt to varying
degrees of uncertainty.
The optimization problem can be stated as follows:
Objective:
Maximize System Performance = Risk Saved = (Attack Damage without the
protection system – Attack Damage with the system in place)
1, 2, 3
9
Subject to:
- Minimize total system cost1, 2, 3
- Alternatives design technical availability1, 2, 3
Decisions:
- The system in initial stage
1, 2, 3
- Subsequent adaptive plans to change the system
2, 3
- The threshold levels of terrorism to activate subsequent plans
3
Given Input:
Performance and cost of each alternative designs
1, 2, 3
Initial forecasted terrorism prevention requirement
1, 2, 3
Subsequent new information on the terrorism prevention requirement
2, 3
1applied in traditional system engineering design
2applied in prevailing real option design
3applied in proposed evolutionary real option design
3.1 Applying the evolutionary real option framework
This section illustrates the steps of evolutionary real option framework through the Maritime
Domain Protection system study. The key risk driver in MDP system is the uncertain degree of
terrorism risk. As the risk fluctuates, the requirements to the MDP system change accordingly. Higher
maritime risk requires a better performing protection system and justifies a higher associated system
cost, and vice versa. In this setting, Monte Carlo simulation is utilized to generate various fluctuating
paths which terrorism risk may take in the coming decade by a modified lognormal stochastic process.
Because real options lack the relatively real-time trading markets for financial options, they can
only be designed and exercised over a longer period of time. The study assumes that the flexibilities
in the system can be exercised by two stages in the third and sixth year in the ten-year lifespan of the
MDP system.
10
Figure 2. A hand picked real option trinomial scenario tree (“risks for exercising the config”. in every
stage are normalized, and “config.” is one of the 109 feasible system configurations, represented by an
indexs)
A trinomial scenario tree (figure 2) is used to represent the decisions on exercising real options in
three decision stages. The three decision stages in the trinomial path-dependent tree result in nine
branches of paths. In total, there are 13 decision nodes in which system may exercise real options;
there are 1 decision node in stage one (year 2007), 3 in stage two (year 2010), and 9 in stage three
(year 2013). At each decision node, system managers have the options to reconfigure the system
depending on the latest update about terrorism risk at that time. The option exercising thresholds are
initially determined manually by expert opinions. In this case, the MDP system is set to exercise a real
option to reconfigure at a stage when the following conditions are met: if the risk in the decision node
is lower than 0.58 times or higher than 1.72 times of its previous level (equivalent to the 25 percentile
& the 75 percentile in the profile of terrorism risk statistically).
A practical challenge is that 13 decision nodes each with 109 feasible configurations make a very
large design space (in the scale of 10 to the power of 26). In addition, the exercising rules for each
option could be optimized as decision variables. The exercising rules in this case depend on the
degree of terrorism, which is continuous. If the degree of terrorism is discretized by a step of 0.01
11
(normalized), the design space, with options and exercising rules both as decision variables, is further
expanded to the scale of 10 to the power of 58.
This study uses Genetic Algorithms (GAs) as the engine to optimize the set of system
configurations in 13 decision nodes (“Config” in figure 2) and their exercising rules under
uncertainties based on multiple objectives. GAs identify 13 optimal configurations altogether (i.e. a
“roadmap” consisting of an initial configuration and subsequent adaptation plan) that optimize the
system value over the ten-year lifespan under various scenarios of uncertainties. The various scenarios
of the fluctuating paths of the terrorism risk are represented by five thousands Monte-Carlo
simulations. The decision tree (figure 2) shows both the configurations and the thresholds to exercise
at a decision node. The flexibility to reconfigure the MDP system by switching each individual
subsystem is understood as a portfolio of (switch) real options. Each switch is an exercise of real
options and associates with some costs. Genetic algorithms are capable of optimization from multiple
objectives – which in this case are Net Present Value (NPV) related. NPV is the difference between
the project performance and project cost, and NPV is dependent on the following decision variables.
ji
jijiji configthresholdfNPV
,
,,, ),( (2)
ji
threshold ,is the exercising threshold to get into branch j at stage i
qp
config ,is the configuration used in branch j at stage i
),,( 1,,, ijijiji configconfigthresholdf is the NPV in branch j at stage i from its precursor in stage i-1
The optimization of NPV is carried out upon the evolutions of terrorism risks, which in this case
follows a modified version of lognormal distribution in stages (figure 3a). The standard deviation of
risk grows over the stages, and this study assumes the standard deviation of risk increase 35% in a
stage over that of previous stage. Figure 3b shows how genetic algorithms divide the risk profile in a
stage and select three different configurations as well their exercising thresholds of exercise
simultaneously. By optimizing the selection of real options and exercising thresholds integrally in
12
every stage, an adaptive roadmap of the flexible system consisting of 13 system configurations at 13
decision nodes is formed, based on maximizing the overall system value.
(a) (b)
Figure 3a. A schematization of the evolution of uncertainties along time.
Figure 3b. A schematization on the identification of configurations to switch to and the definition of
exercising thresholds.
For example, the configuration ji
config ,giving the best performance in a state may not be chosen
because it may not embed suitable real options to cope with the uncertainties and may not deliver the
best overall performance over time, since the ji
config ,affects the NPV in later stages i+n (n=1, 2, …).
Optimization needs to consider the impact of a configuration over the project value not only in that
state, but also over future paths in the decision tree. In dong this, the optimization process becomes
more realistic in how the project adapts to the path-dependent uncertainty.
In addition, due to the nature of switch type real options, managers simultaneously exercise the
rights but not obligations to start an alternative module and to abandon the old module. Determining
the exercising rules for such switch options need to consider the performances and costs of the two
options and option interactions integrally. Because the number of real options is large, the interactions
13
among real options are non-linear, the design space is complex, and many future scenarios exist, it is
unrealistic to expect human cognition to identify the most optimal set of real options.
4 RESULTS AND DISCUSSIONS
The approaches of decision support in designing and managing the system are listed in table 1. A
“fixed peak design” is chosen from a space of 109 alternative design configurations, and there is no
exercising rule, because the design will not change according to circumstances. “Hand picked Real
Option” approach utilizes human cognition alone to determine the exercising rules and select the
suitable options from 109 alternative configurations at 13 decision nodes. In the “EA. selecting Real
Options without exercising rules” approach, exercising rules are still predetermined by experts, but
suitable options are picked and optimized by evolutionary algorithms. In the “EA. selecting Real
Options together with exercising rules” approach, both of real options and their exercising rules are
selected by EAs. As the number of decision variable grows, the computational complexity of selecting
real options and their exercising rules grows exponentially.
Table 1. A summary of design approaches
To evaluate and compare the results from these approaches, the present study follows the
established valuation method [8; 11], which takes the difference between the NPV of the flexible
14
system and that of a traditional peak design as the added value of real options. Likewise, Kulatilaka
and Trigeorgis [20] modeled specifically the value of the flexibility to switch as the difference
between the value of the flexible project and the value of the rigid project. Table 2 summarizes the
NPV of the MDP system from different design approaches and valuation methods.
Table 2. The project values based on design approaches
The remainder of the results is organized as follows. Section 4.1 discusses the single peak design,
valuated based on one scenario (a single forecast) and on many scenarios (through Monte Carlo
simulations). Section 4.2 compares the results from hand-picked real option selections with those of
peak design, both valuated through Monte Carlo simulations. These two results form the baseline to
compare the results from the proposed evolutionary real option framework, which are discussed in
section 4.3 and 4.4. Section 4.5 examines the value of flexible project from a Value-At-Risk graph
and section 4.6 studies the value of real options as a function of the level of uncertainty. Finally, a
synthesized discussion of the results and managerial implications are given in Section 4.7 and 4.8
respectively.
15
4.1 The project value based on a fix design in deterministic & Monte Carlo
simulations
Traditional peak design process valuates the system based on a forecasted level of terrorism risk,
and gives a deterministic project value of $574M. Utilizing Monte Carlo simulations of the key
uncertainty driver reveals that the system value may fluctuate a lot in reality. A lower realization of
terrorism risk than the initial forecasted level will lower the value of the MDP system, while a higher
realization of terrorism risk will increase the value of the MDP system, provided that the system can
handle the increased requirements.
The impact of the uncertainty driver is not symmetrical in the case, as it would not be in most of
the complex systems. In the case of MDP system, a higher realization of terrorism risk has a larger
impact on project value than a lower realization has; consequently valuation with Monte Carlo
simulations gives a higher average NPV than the deterministic valuation does to the same peak
design. Results from Monte Carlo simulations also point out some alarming findings concerning risk
management. The minimum NPV and the 5 percentile NPV of the peak design are deeply negative,
which are hidden in the deterministic valuation method.
4.2 The project value based on hand-picked real option approach
In the common practice of real options to project and system design, experts pick the real options,
set their exercise regions, and valuate those options [15; 32; 33]. We follow the same approach to get
the project value to compare with the peak design approach and the evolutionary real option design
later. In the hand-picked real option approach, the project states, options are first established via a
decision tree. The trinomial decision tree was branched according to the evolution of the degree of
terrorism in all the decision nodes. The top 25 percentile of the degree of terrorism forms a decision
point of option; the lower 25 percentile forms a decision point of option, and the middle 50% forms
another a decision point of option (as in figure 2).
16
On one hand, configurations that have lower costs and correspondingly lower levels of protection
are exercised when the realization of terrorism is in the lower 25% branch. This elevates the minimum
project value and the 5 percentile project value compared to the single peak design (-$2,288M to -
$1267M and -$1,780M to -$1083M). On the other hand, more capable yet more expensive
configurations are utilized to capture a larger project value once the degree of terrorism reaches the
upper 25 percentiles (as measured by the top 5 percentile increasing from $16,160M to -$34,708M
and the maximum project value increasing from $141,874M to -$579,905M). The average NPV of the
Monte Carlo simulation increases significantly from $3,295M to $10,043M. These increments in NPV
through expert based hand-picked design approach are the values of real options demonstrated by the
real option literature.
4.3 The project value based on the “EA. selecting Real Options without
exercising rules” approach
One step further, the issue of compositing the right pieces of real options for the 13 decision
nodes can be tackled by genetic algorithms. GAs are used to search for suitable real options in all the
13 decision nodes to make a portfolio of real options that maximize two objectives: the average and
the 5 percentile of project value. The objectives are customizable -the average and the 5 percentile of
project value are chosen here to cover both the value and risk management of the project. The same
predetermined exercise regions (the 25, 50, and 75 percentile of risk profile) as those in “hand-picked
real option design approach” are used.
Figure 4 below shows such an optimized set of configurations selected by Genetic algorithms
with predetermined exercising regions. Genetic algorithms are able to composite a portfolio of real
options to improve the average NPV to $10,277M and the 5 percentile project value to -$298M (table
2). The maximum and 95 percentile project values remain relatively the same.
17
Figure 4. GAs selecting real options in a trinomial scenario tree (asterisks denotes decision variables
optimized by GAs – in this case configurations)
4.4 The project value based on the “EA. selecting Real Options together with
exercising rules” approach
Even more, genetic algorithms can be deployed to composite a portfolio of real options as well as
their exercising thresholds simultaneously. Instead of the fixed expert estimated option exercising
thresholds at the 25, 50 and 75 percentile of the risk profile, the evolutionary real option framework
combines various real options with various exercising rules to attempt for higher project values.
As a result, arrays of solutions are found -each with a set of real options and their corresponding
exercising conditions. For instance, figure 5 shows one of such solutions. In the second stage (year
2010), GAs use the configuration number as long as risk is within 4.88 times (i.e. the entire 98
percentile of the risk profile from the bottom) of its previous level in the first stage (year 2007). And
GAs will exercise a high performing configuration (index number 84) for the 98-99 percentiles and
for the 99-100 percentiles of very high risk scenarios. The exercising thresholds in stage 2 are highly
asymmetrical, and so are those in stage 3.
18
Figure 5. GAs selecting real options and their exercising thresholds in a trinomial scenario tree (asterisks
denotes decision variables optimized by GAs – in this case the configurations and the degree of risk to
exercise the configurations)
Such an option exercising plan is found to benefit the project value more. The average value of
the project increases further to $11,329M (table 2). The exercising regions of real options in this plan
are exclusively focused on the high terrorism risk side (to switch at the 98 and 99 percentiles or to
switch only when risk escalates by about 5 folds in other words). This approach clearly enhances the
ability of the system to capture more values from the higher degrees of terrorism risk, as shown by the
considerable improvement of the 95 percentile of NPV ($47,395 from $34,610 when GAs are only
deployed to optimize the option but not the exercising rules).
The system configurations in the adjacent branches of the decision trees could be the same
sometimes (as found in both the hand-picked solution in figure 2 and GA-based solution in figure 4
and 5). This implies an area to further improve the decision tree representation. For instance, project
managers can combine some branches or split them to form a new tree of decision nodes. To frame
and reflect deeper, a different decision tree represents a different decision structure of how the project
manager plans to act to new information about the rise and fall of the degree of terrorism. Hence,
19
optimizing the decision tree, - which options are worthwhile to have and at what conditions to
exercise them, is a fundamental way to improve the design and management of projects.
4.5 Real option values in detail by Value-At-Risk Graph
Table 1 has given a few key value indicators such as the average, maximum, and minimum NPV,
but in any case they are only parts of a picture.
The Value-At-Risk (VAR) graph of both the peak design and the evolutionary real option design
(GAs selecting both real options and exercising rules) is shown in figure 6. The fixed design has 52%
probability of ending up with a negative net present value, while the flexible design has only 38%.
Also, the maximum possible loss for the fixed design is -2,288 million USD versus -1,625 million
USD for the flexible design. Although the graph is truncated on its right, it can be seen the maximum
NPV obtainable with a flexible system is much larger - for instance the flexible design far
outperforms the fixed design by around $13M at 95% cumulative percentile.
0%
25%
50%
75%
100%
-4,000 3,000 10,000 17,000 24,000 31,000 38,000 45,000
NPV
Cumulative Percent
Peak Design
GA optimized Real
Option Design
52%
38%
95%
Figure 6. VaR graph of the NPVs of peak design and evolutionary real option design
20
Only in a narrow range sandwiched between 60% to 85% in the VAR graph, where the curve of
the fixed design is to the right of the curve of the flexible design, the fixed design outperforms the
flexible one. This is the results of some scenarios when future turns out similarly to what forecast
tells. In that narrow range, peak design is outperforming, because real option designs carry extra costs
from embedding flexibilities. Overall, the VAR clearly indicates the better risk management of the
real option design compared to a peak design.
4.6 Sensitivity of Real Option value to the level of the uncertainty
In addition, the sensitivity of the value of real options to the level of uncertainty around the degree
of terrorism was studied. Figure 7 shows the value of real options with respect to the levels of
uncertainties. The value of real options is the difference of the project value of the evolutionary real
option design and the project value of the fixed design.
0
2000
4000
6000
8000
10000
12000
14000
0 0.1 0.2 0.3 0.4 0.5
Standard Deviation
Mean NPV of Embeding RO
Figure 7. VaR graph of peak design NPV and evolutionary real option design NPV
As the uncertainty around the degree of terrorism becomes larger, the demand to the MDP system
is more unpredictable, the likelihood to reconfigure the system with real options increase, and as a
result embedding real options in the project becomes more valuable. The level of uncertainty used to
21
generate the results in this paper is higher than that in the paper of MDP case study [9] (standard
deviation of risk over a stage: 35% vs. 25%), and correspondingly the values of real options vary, as
shown in figure 7. In principle, when the level of uncertainty associated is higher, real options have a
higher leverage in improving the systems and projects.
4.7 Synthesized results
The results of the test case confirm the findings from others [13; 32; 33] that embedding real
options design by human cognition alone can be hugely beneficial alone (the mean of NPV of hand
picked real option design is 3 times higher than that of fixed peak design). Furthermore, the
evolutionary framework aided by GAs extends the prevailing real option frameworks and this
extension is found to be valuable (the mean of NPV from evolutionary real option framework is 13%
higher than that from conventional hand picked real option approach). The more sophisticated
framework is able to valuate multiple interdependent real options and their exercising rules and
provides an integrated approach to system planning, design, management and decision making. Such
valuation enables the incorporation of multiple real options and exercising rules, which fundamentally
increase the capability of the project to adapt to future circumstances. Therefore the project value is
increased beyond the prevailing real option frameworks and much beyond the traditional disciplinary
engineering peak design approach.
4.8 Managerial and other implications
As the design approaches become more flexible -from the peak design with a single project value,
to the hand-picked real option design, and ultimately to the evolutionary real option design, managers
and designers of projects appear to have less and less controls. On the contrary, losing such controls
(in the sense of controlling a project statically) is rather an illusion [19]. In an uncertain environment,
a more meaningful way to control a project is through incorporating various real options to enable the
system to adapt to the many paths future could take. As the number of future paths increases and the
interactions among options become more intricate, the appreciation and adoption of more advanced
22
techniques in more sophisticated decision support frameworks becomes more very crucial. For this
reason, the evolutionary real option framework is developed for designing and managing “evolvable
design” that would be difficult for human to conceive and valuate.
The computational time used by genetic algorithms to reach a solution is 3 minutes in the case
study by a standard stand alone Dell personal computer. The framework could be partly incorporated
into a computer program as an iterative decision support tool to enable system designers to create
more evolvable design. That is not to say the paper suggests substituting the decision making with the
framework, instead the paper proposes to use the framework to provide quantitative knowledge about
the “design for evolvability” [5] to support the dynamic decision planning, so that experts can focus
more on the more creative process of generating design alternatives to harness more from flexibility.
The framework is generic to be applicable to a wide range of large-scale systems and projects.
Most large-scale systems and projects have options to switch its subsystems or project modules to
cope with uncertainty. To a certain extent, realism is that absolute peak designs do not exist, and part
of the job responsibilities of project managers is to exercise judgments over the project according to
realizations of future. Therefore, the issue becomes whether to explicitly bring the “design for
evolvability” upfront to integrate it into the design process. This study represents such an effort to
analyze and design systems and projects with real options in a holistic framework.
5 CONCLUSIONS
The evolutionary real option framework valuates many flexibilities in a project holistically and
delivers a flexible roadmap for deploying real options. The evolutionary framework is proved to
further increase the project value from the prevailing real option approaches and provide better
decision support in the presence of multiple real options.
There could be a lot of potential advances in optimizing the usages of multiple real options along
many uncertain paths, drawing upon and integrating evolutionary algorithms, Monte Carlo
23
Simulations, scenario trees, and other techniques. This study makes the initial attempt of creating an
evolutionary real option framework to aid human cognitions and enable practitioners – managers,
investors, and system designers to valuate and plan flexibilities into projects and systems and exercise
them dynamically.
24
ACKNOWLEDGEMENTS
The authors thank Defense Science and Technology Agency of Singapore, who funded the project
and gave invaluable insights. The authors gratefully acknowledge the support & contributions of this
project from Singapore-Delft Water Alliance (SDWA) -- R-264-001-001-272. For more information,
please visit http://www.sdwa.nus.edu.sg. Finally, authors gratefully acknowledge support by
Academic Research Fund under grant "Data Assimilation and Data-Driven Knowledge Discovery" R-
264-000-199-133/112.
25
REFERENCES
[1] J. Anand, R. Oriani, and R.S. Vassolo, Managing a Portfolio of Real Options. in: J.J. Reuer, and
T.W. Tong, (Eds.), Advances in Strategic Management, Emerald Group Publishing Limited,
West Yorkshire, England, (2007), pp. 275-303.
[2] B. Asrilhant, R.G. Dyson, and M. Meadows, On the strategic project management process in the
UK upstream oil and gas sector, Omega 35 (2007) (89-103).
[3] V. Babovic, Emergence, Evolution, Intelligence: Hydroinformatics, Balkema, Rotterdam, (1996).
[4] C.Y. Baldwin, and K.B. Clark, "Design rules: The power of modularity", The MIT Press,
Cambridge, MA, (2000).
[5] P.D. Borches, and G.M. Bonnema, On The Origin Of Evolvable Systems, Seventh International
Symposium on Tools and Methods of Competitive Engineering, Kusadasi, Turkey, (2008).
[6] P.P. Boyle, Options: A Monte Carlo approach, Journal of Financial Economics 4 (1977) (323-
338).
[7] M.A. Brach, Real Options in Practice, Wiley, Hoboken, New Jersey, (2003).
[8] R.A. Brealey, and S.C. Myers, Principles of Corporate Finance, McGraw-Hill, Boston, (1991).
[9] J. Buurman, S. Zhang, and V. Babovic, Reducing Risk Through Real Options In Systems Design:
The Case Of Architecting A Maritime Domain Protection System Risk Analysis 29 (2009)
(366-379).
[10] M. Cardin, W. Nuttall, R. de Neufville, and J. Dahlgren, Extracting Value from Uncertainty: A
Methodology for Engineering Systems Design, 17th. Annual International Symposium of the
International Council on Systems Engineering (INCOSE), San Diego, (2007).
[11] T. Copeland, and V. Antikarov, Real Options: A Practitioner's Guide, Texere Publishing, New
York (2001).
[12] R. de Neufville, Dynamic strategic planning for technology policy, International Journal of
Technology Management 19 (2000) (225-245).
26
[13] M.A.G. Dias, Investment in Information for Oil Field Development Using Evolutionary
Approach with Monte Carlo Simulation, 5th Annual International Conference on Real
Options - Theory Meets Practice, UCLA, (2001).
[14] T.W. Faulkner, Applying 'options thinking' to R&D valuation, Research Technology
Management 39 (1996) (50-56).
[15] D.N. Ford, D.M. Lander, and J.J. Voyer, A real options approach to valuing strategic flexibility
in uncertain construction projects, Construction Management and Economics 20 (2002) (343-
351).
[16] D. Goldberg, Genetic algorithms in search, optimization and machine learning, Addison-Wesley
Professional, Boston, MA, USA, (1989).
[17] R. Hassan, R. De Neufville, and D. McKinnon, Value-at-risk analysis for Real Options in
complex engineered systems, Proceedings of IEEE International Conference on Systems, Man
and Cybernetics, Waikoloa, HI, (2005), pp. 3697-3704.
[18] C. Hilhorst, P. Ribbers, E. van Heck, and M. Smits, Using Dempster-Shafer theory and real
options theory to assess competing strategies for implementing IT infrastructures: A case
study, Decision Support Systems 46 (2008) (344-355).
[19] D. Kahneman, and D. Lovallo, Timid choices and bold forecasts: A cognitive perspective on risk
taking, Management Science 39 (1993) (17-31).
[20] N. Kulatilaka, and L. Trigeorgis, The general flexibility to switch: Real options revisited,
International Journal of Finance 6 (1994) (778-798).
[21] J. Lazo, M. Pacheco, M. Vellasco, and M. Dias, Real Option Decision Rules For Oil Field
Development Under Market Uncertainty Using Genetic Algorithms And Monte Carlo
Simulation, 7th Annual Real Options International Conference, DC Washington, (2003).
[22] R.G. McGrath, A real options logic for initiating technology positioning investments, Academy
of Management Review 22 (1997) (974-996).
27
[23] K.D. Miller, and H.G. Waller, Scenarios, real options and integrated risk management, Long
Range Planning 36 (2003) (93-107).
[24] J. Mun, Real options analysis : tools and techniques for valuing strategic investments and
decisions, John Wiley & Sons, Hoboken, N.J., (2006).
[25] NPS, Maritime Domain Protection in the Straits of Malacca, Naval Postgraduate School,
Monterry, California, (2005).
[26] E. Pennings, and O. Lint, The option value of advanced R & D, European Journal of Operational
Research 103 (1997) (83-94).
[27] C. Raymond, (Ed.), Maritime Terrorism, A Risk Assessment: The Australian Example, World
Scientific; Institute of Defence and Strategic Studies, Singapore, (2005).
[28] M. Schneider, M. Tejeda, G. Dondi, F. Herzog, S. Keel, and H. Geering, Making real options
work for practitioners: A generic model for valuing R&D projects, R and D Management 38
(2008) (85-106).
[29] A. Triantis, and A. Borison, Real options: State of the practice, Journal of Applied Corporate
Finance 14 (2001) (8-24).
[30] L. Trigeorgis, The nature of option interactions and the valuation of investments with multiple
real options, Journal of Financial and Quantitative Analysis 28 (1993) (1-20).
[31] L. Trigeorgis, Real options: Managerial flexibility and strategy in resource allocation, The MIT
Press, Cambridge, Mass., (1996).
[32] T. Wang, and R. de Neufville, Real options "in" projects, 9th Annual International Conference on
Real Options, Paris, (2005).
[33] T. Zhao, and C.L. Tseng, Valuing flexibility in infrastructure expansion, Journal of Infrastructure
Systems 9 (2003) (89-97).
[34] E. Zitzler, and L. Thiele, Multiobjective evolutionary algorithms: a comparative case study and
the strength Pareto approach, IEEE Transactions on Evolutionary Computation 3 (1999) (257-
271).
