managing uncertainty and flexibility in the modern …hq284pw2370/... · reshaping the industry. in...

MANAGING UNCERTAINTY AND FLEXIBILITY IN THE MODERN

ENERGY SECTOR: QUANTITATIVE MODELING OF TECHNICAL

RISK, ECONOMIC VALUE, AND STRATEGIC COMPETITION

A DISSERTATION

SUBMITTED TO THE DEPARTMENT OF

MANAGEMENT SCIENCE AND ENGINEERING

AND THE COMMITTEE ON GRADUATE STUDIES

OF STANFORD UNIVERSITY

IN PARTIAL FULFILLMENT OF THE REQUIREMENTS

FOR THE DEGREE OF

DOCTOR OF PHILOSOPHY

KARIM FARHAT

JULY 2016

http://creativecommons.org/licenses/by-nc/3.0/us/

This dissertation is online at: http://purl.stanford.edu/hq284pw2370

© 2016 by Karim Farhat. All Rights Reserved.

Re-distributed by Stanford University under license with the author.

This work is licensed under a Creative Commons Attribution-Noncommercial 3.0 United States License.

ii



http://purl.stanford.edu/hq284pw2370

I certify that I have read this dissertation and that, in my opinion, it is fully adequatein scope and quality as a dissertation for the degree of Doctor of Philosophy.

John Weyant, Primary Adviser


Kathleen Eisenhardt


Stefan Reichelstein

Approved for the Stanford University Committee on Graduate Studies.

Patricia J. Gumport, Vice Provost for Graduate Education

This signature page was generated electronically upon submission of this dissertation in electronic format. An original signed hard copy of the signature page is on file inUniversity Archives.

iii

iv

Abstract

The need to mitigate climate change promises an increasingly different, uncertain, and flexible

energy landscape. In a climate-constrained world, uncertainty and flexibility complicate the

appraisal of new investments in clean energy. They make it more challenging for decision-

makers to quantify the technical risk, economic value, or strategic competitiveness of their

prospective energy initiatives, for orthodox evaluation techniques like worst-case-scenario and

net-present-value become insufficient. Consequently, in order to help investors ride the clean

energy wave, one urgent priority is to clarify and quantify uncertainty and flexibility in

modern energy systems and industries. This dissertation aims to develop assessment models

that achieve this exact goal.

The dissertation takes on three decision-centric research endeavors. The first study sheds light

on the technical uncertainty related to the leakage of anthropogenic carbon dioxide from

geologic storage reservoirs. Specifically, a conceptual methodological framework is developed

to help storage-site managers bridge risk assessment and corrective measures through clear

and collaborative contingency planning. First, a quantitative risk assessment matrix is

presented, highlighting the concept of risk profiles. As the main focus of this study, a

contingency planning matrix is then developed based on the risk assessment matrix, and its

tier structure is discussed. Lastly, the contingency planning matrix is used to guide the design

of a model contingency plan, which covers multiple sections on preparing for leakage risks

and responding to leakage incidents.

The second study switches from technical uncertainty to economic flexibility, investigating

the value of flexible hydrogen-based polygeneration energy systems (PES). PES are multi-

input multi-output industrial facilities. This study models a representative PES that uses coal

as a primary fuel and produces electricity and fertilizers as end-products. A series of economic

v

propositions allows deriving multiple metrics that quantify the levelized cost of hydrogen, the

profitability of PES under both static and flexible operation modes, as well as the real-option

values associated with diversification and flexibility. These metrics are subsequently applied

to evaluate Hydrogen Energy California (HECA), a PES project under development in

California. The results show that the profitability of a static HECA increases in the share of

hydrogen converted to fertilizer rather than electricity. However, when configured as a

flexible PES, HECA almost breaks even. Ultimately, diversification and flexibility prove

valuable for HECA when polygeneration is compared to static monogeneration of electricity,

but these two real options have no value when comparing polygeneration to static

monogeneration of fertilizers.

Finally, aiming to examine uncertainty in strategic competition within an energy industry, the

third study proposes a decision analytic modeling of Porter’s five forces framework, hereby

referred to as DAFF. This work is divided into two parts. The first part addresses the

conceptual foundations of DAFF. After explaining how decision analysis tools can enhance

the operationalization of the five forces theory, this part provides a detailed description of the

various elements in a DAFF model. Subsequently, a series of DAFF models are developed to

fulfill the two main objectives of competitive strategy: positioning in the industry, and

reshaping the industry. In the second part, DAFF is implemented to help inform a major firm’s

competitive strategy in the near-future U.S. residential solar PV industry. A DAFF Bayesian

Network is designed to evaluate competition in the overall industry. The results reveal

moderate competitive powers, with expected earnings before tax (EBT) of 4.05 billion $/year.

Also, due to significant yet asymmetrical competitive interdependence, witnessing a single

competitive force at its strongest or weakest extreme seems sufficient to vary the industry

EBT between 1.86 and 5.77 billion $/year. Analyzing four positioning decisions by the solar

firm expands the Bayesian Network into a Decision Diagram with 32 possible positioning

tracks. Each track yields a unique EBT value ranging between 0.51 and 3.98 billion $/year.

The results show that the highest expected earnings are realized upon: locating in urban areas,

managing customers directly without relying on dealers, and offering loan and lease services

to solar customers.

vi

Acknowledgements

My PhD journey at Stanford University has been one of my best life experiences, thanks to an

amazing group of advisers, instructors, friends, and colleagues.

First and foremost, I would like to express a deep gratitude to my advising team for their

continuous guidance, support, and encouragement over the years. I begin by thanking John

Weyant, my primary adviser, not only for believing in me and being my champion, but also

for nurturing my passion in interdisciplinary energy research. John gave me the freedom to

explore, the intuition to rationalize, and the resources to tackle new and challenging energy

modeling problems. Every time I met with John, I left feeling more confident and enthusiastic

about research, and life. This dissertation would not have been possible without John’s

masterful supervision and advice, and I’m extremely lucky to have had the honor of being one

of his students. I have also had the great fortune of working with Sally Benson, who was my

first research adviser for the Master’s degree in Energy Resources Engineering at Stanford.

Sally taught the engineer in me to appreciate geological sciences, and her deep expertise in

carbon capture and storage was the stimulus for and the facilitator of my continuous education

and interest in this field. Furthermore, Sally introduced me to the extensive web of energy

research and scholars at Stanford; she never hesitated to provide access and opportunities that

expanded my professional horizons far beyond what I could have hoped for on my own.

I want to thank Stefan Reichelstein at the Graduate School of Business, whom I have had the

immense pleasure of learning from and working with. His mentorship and our collaboration

have been essential to my education on energy finance and economic modeling, and his clear

and crisp approach to research has helped me improve the ways I conduct, analyze, and

communicate my scholastic activities. I am also extremely grateful to Kathy Eisenhardt for

introducing me to, teaching me about, and supervising my research on business strategy.

Besides our intellectually stimulating conversations, Kathy always challenged and encouraged

vii

me to give my very best, to walk the extra mile; my research on competitive strategy modeling

would not have been half as good without her guidance.

Outside of my core advising team, I want to thank a number of Stanford faculty members who

equipped me with the knowledge and motivation to undertake this research. I am grateful to

Ronald Howard, Ross Shachter, and Burke Robinson for teaching me everything I know about

decision analysis and for accepting me as a regular guest in their research group meetings.

Terry Root and the late Stephen Schneider have also played an important role in shaping the

energy researcher I am today. I will never forget the 14 hours I spent in a room full of world

leaders negotiating the fate of the Copenhagen Accord in 2009. Terry and Steve were the

reason I was able to gain this life-changing experience, which evolved into a life-long

commitment to advance and promote research on climate change and clean energy.

I would also like to acknowledge my fellow graduate students in the MS&E department.

Specifically, I thank Melanie Craxton, Lauren Culver, Ben Leibowicz, James Merrick, and

Matthew Smith for always being there to brainstorm an idea, help with a complicated

programming code, review a working paper, or grab a beer after a long week. Also, I thank

Mostafa Afkhamizadeh and Marshall Kuypers for the engaging conversations and team

projects we got to share; my research and academic work would not have been as rewarding or

enjoyable without them.

Outside Stanford, I am deeply grateful for meeting and working with Cas Groothuis while he

was the Opportunity Manager Future Energy Technologies at Royal Dutch Shell. Cas

recommended me to join Stanford, and my internship with him in the Netherlands constituted

my first exposure to the world of energy business. Attesting to his superb leadership, he

immediately realized and nurtured my interest in “connecting the dots” and the “big picture,”

and he gave me access to a wide energy network that later helped advance my research. I am

very thankful for having such an amazing first “boss.”

Of course, my family have been very supportive throughout this long process, sending me

their best wishes and prayers, as well as pictures of my cute little nephews, all the way from

Beirut, Lebanon. They are deserving of my gratitude for their love, understanding, and

viii

encouragement. Also, I shall never forget the kindness of my grandmother, who supported me

while in college.

Last, but definitely not least, I want to thank my dear friends Teddy White, Alexander

Greenberg, Bryson Tombridge, David Klein, and Scott McNally, who ensured that my life

beyond graduate school remained enjoyable and fulfilling. These gentlemen were always there

for me when I needed an encouraging word, a refreshing break, or a friendly nudge to help me

stay on task – some of them even endured listening to me defending this very dissertation. For

that, I will always be grateful.

ix

Table of Contents

Abstract .................................................................................................................................. iv

Acknowledgements ................................................................................................................ vi

Table of Contents ................................................................................................................... ix

List of Tables ........................................................................................................................xiii

List of Figures ....................................................................................................................... xv

Chapter 1: Introduction ............................................................................ 1

1 Motivation: A Changing, Uncertain, and Flexible Energy Landscape .............................. 1

2 Scope of Work ................................................................................................................... 6

2.1 Translating Risk Assessment to Contingency Planning for CO2 Geologic

Storage: A Methodological Framework .................................................................. 10

2.2 Economic Value of Flexible Hydrogen-Based Polygeneration Energy

Systems.................................................................................................................... 11

2.3 Decision Analytic Modeling of the Five Forces in Competitive Strategy:

Application in the U.S. Residential Solar PV Industry ........................................... 13

3 Dissertation Organization ................................................................................................ 16

References ............................................................................................................................. 19

Chapter 2: Translating Risk Assessment to Contingency Planning

for CO2 Geologic Storage: A Methodological Framework .................. 29

1 Introduction ..................................................................................................................... 29

2 Risk Management: Assessment, Mitigation, and Contingency Planning ........................ 31

3 Updating the Risk Assessment Matrix ............................................................................ 33

3.1 Functional Subsystems for Risk Identification........................................................ 35

3.2 Bayesian Event Tree for Risk Analysis ................................................................... 36

3.3 Tolerance Levels for Risk Evaluation ..................................................................... 46

x

3.4 Combined Representation of Risk Assessment Elements ....................................... 47

4 Translating the Risk Assessment Matrix to a Contingency Planning Matrix .................. 50

4.1 Transforming Matrix Dimensions ........................................................................... 51

4.2 Transforming Matrix Boundaries ............................................................................ 52

4.3 Classifying Risk into Tiers ...................................................................................... 53

5 A Model Contingency Plan ............................................................................................. 57

5.1 Tiers of Risk-Preparedness and Incident-Response ................................................ 58

6 Conclusions ..................................................................................................................... 69

6.1 Future Work ............................................................................................................ 71

References ............................................................................................................................. 73

Appendix A: Drawbacks of Alternative Three-Tier Systems ............................................... 80

Appendix B: Tier-Based Contingency Planning ................................................................... 82

Chapter 3: Economic Value of Flexible Hydrogen-Based

Polygeneration Energy Systems ............................................................. 84

1 Introduction ..................................................................................................................... 84

2 Research Methodology .................................................................................................... 88

2.1 Levelized Cost of Hydrogen .................................................................................... 89

2.2 Technical Configuration of PES .............................................................................. 91

3 Economic Analysis .......................................................................................................... 93

3.1 Scenario 1: Static PES with Fixed Production Rates .............................................. 94

3.2 Scenario 2: Flexible PES with Variable Production Rates ...................................... 97

4 Profitability and Value of Real-Options ........................................................................ 104

5 Additional Modeling Considerations ............................................................................. 107

5.1 Carbon Capture and Storage.................................................................................. 107

5.2 Time-dependency of prices and variable costs ...................................................... 108

6 Case Study: Hydrogen Energy California ..................................................................... 109

6.1 Technical Configuration ........................................................................................ 109

6.2 Economic Analysis ................................................................................................ 111

xi

7 Conclusions ................................................................................................................... 120

7.1 Future Work .......................................................................................................... 121

References ........................................................................................................................... 123

Appendix A: Derivation of Economic Propositions ............................................................ 127

Appendix B: Cost Estimates for HECA .............................................................................. 144

Chapter 4: Decision Analytic Modeling of the Five Forces in

Competitive Strategy ............................................................................. 147

1 Introduction ................................................................................................................... 147

2 Theoretical Background ................................................................................................ 149

2.1 Decision Analysis .................................................................................................. 149

2.2 The Five Forces that Shape Competition .............................................................. 152

2.3 Decision Analytic Approach to the Five Forces .................................................... 154

3 Methodology: Developing the DAFF Models ............................................................... 159

3.1 Modeling the Competitive Forces, Drivers, and Factors ....................................... 160

3.2 Modeling the Economic Implications of the Five Forces ..................................... 169

3.3 Modeling the Firm’s Actions ................................................................................ 172

4 First Objective of Competitive Strategy: Positioning in the Industry ........................... 176

4.1 First Step: Assess the Profitability of the Overall Industry ................................... 177

4.2 Second Step: Position Competitively and Assess the Profitability of Each

Positioning Segment in the Industry ..................................................................... 181

4.3 Third Step: Assess the Profitability of the Firm in Each Positioning

Segment of the Industry ........................................................................................ 184

4.4 Advantages of the DAFF modeling ....................................................................... 187

5 Second Objective: Reshape the Industry ....................................................................... 194

5.1 First Step: Predict Industry Change ....................................................................... 195

5.2 Second Step: Reshape Industry Change ................................................................ 198

6 Best Practices in DAFF Modeling ................................................................................. 202

7 Broader Alignment between Decision Analysis and Porter’s Competitive Strategy .... 204

xii

8 Conclusions ................................................................................................................... 205

8.1 Future Work .......................................................................................................... 208

References ........................................................................................................................... 212

Chapter 5: DAFF Modeling of Competitive Strategy: Positioning

in the Near-Future U.S. Residential Solar PV Industry .................... 216

1 Introduction ................................................................................................................... 216

1.1 The U.S. Residential Solar PV Industry ................................................................ 218

1.2 The Solar Firm ...................................................................................................... 220

2 Methodology: Developing the DAFF Models ............................................................... 220

2.1 First Step: Assess the Overall Industry ................................................................. 221

2.2 Second Step: Assess Each Positioning Segment in the Industry ........................... 241

3 Results: Outputs from the DAFF Models ...................................................................... 247

3.1 First Step: Assess the Overall Industry ................................................................. 247

3.2 Second Step: Assess Each Positioning Segment in the Industry ........................... 262

4 Conclusions ................................................................................................................... 268

4.1 Future Work .......................................................................................................... 272

References ........................................................................................................................... 274

Appendix A: Degree Characterization for Competitive Uncertainties ................................ 278

Appendix B: Influence of Positioning Tracks on Competitive Forces ................................ 284

xiii

List of Tables

Table 2.1: Examples of Origin FEPs and their corresponding Indicators ................................ 36

Table 2.2: Selection criteria for the three-tier system .............................................................. 53

Table 2.3: Example corrective measures for incident-response ............................................... 63

Table 2.4: Example corrective measures matrix (CMM) for incident-response ...................... 64

Table 2.B1: Tier-based response strategies for contingency planning ..................................... 82

Table 2.B2: Tier-based human and equipment resources for contingency planning ............... 83

Table 3.1: HECA technical parameters .................................................................................. 110

Table 3.2: HECA auxiliary loads ........................................................................................... 111

Table 3.3: Levelized costs of capacity of HECA ................................................................... 112

Table 3.4: Levelized time-averaged fixed operating costs of HECA ..................................... 113

Table 3.5: Levelized time-averaged variable costs of HECA ................................................ 114

Table 3.6: Time-averaged prices of HECA end-products ...................................................... 115

Table 3.7: Economic valuation of HECA .............................................................................. 115

Table 3.B1: System prices of HECA per unit capacity .......................................................... 144

Table 3.B2: System prices of HECA’s subsystems per unit capacity .................................... 144

Table 3.B3: Yearly fixed-operating costs of HECA as fraction of capacity costs ................. 145

Table 3.B4: Yearly fixed-operating costs of HECA Subsystems per unit capacity ............... 145

Table 3.B5: Prices of input commodities and services for HECA ......................................... 146

Table 3.B6: Yearly-averaged variable costs Cost of HECA per unit of production .............. 146

Table 3.B7: Prices of HECA end-products ............................................................................ 146

Table 5.1: DAFF uncertainties influenced by Regional Focus .............................................. 242

Table 5.2: DAFF uncertainties influenced by Downstream Integration ................................ 243

Table 5.3: DAFF uncertainties influenced by Customer Financing ....................................... 244

xiv

Table 5.4: DAFF uncertainties influenced by Panel Manufacturing ...................................... 245

Table 5.A1: Definition of competitive force and driver uncertainties in the U.S.

residential solar industry ..................................................................................... 278

Table 5.A2: Definition of factor uncertainties in the U.S. residential solar PV industry ....... 283

Table 5.B1: Probability of {high} power for each competitive force under each

positioning track ................................................................................................. 284

xv

List of Figures

Figure 2.1: Elements of risk management for CO2 leakage from geologic reservoirs .............. 32

Figure 2.2: Functional subsystems for risk identification ......................................................... 35

Figure 2.3: Example Bayesian event tree (BET) for risk analysis of CO2 leakage ................... 38

Figure 2.4: Sketch of CO2 leakage through caprock high-permeability zone ........................... 40

Figure 2.5: Example probability distributions of CO2 leakage scenarios .................................. 41

Figure 2.6: Example value models of CO2 leakage scenarios ................................................... 45

Figure 2.7: Example risk assessment matrix (RAM) for CO2 leakage. ..................................... 48

Figure 2.8: Translating the risk assessment matrix to a contingency planning matrix ............. 50

Figure 2.9: Tier system tradeoff between resource proximity and resource diversity .............. 55

Figure 2.10: Tier allocation procedure for risk-preparedness and incident-response ............... 60

Figure 2.11: Example decision-making hierarchy of the operating party for

contingency planning ............................................................................................. 66

Figure 2.12: Example notification protocol of the operating party for incident-response ........ 67

Figure 2.13: Example communication scheme for contingency planning ................................ 68

Figure 2.14: Collaborative approach to securing resources for Tiers 2 and 3 ........................... 68

Figure 2.A1: Drawbacks of alternative tier-system approaches for contingency planning ....... 80

Figure 3.1: Simplified process flow sheet of the used PES ....................................................... 92

Figure 3.2: Schematic representation of static and flexible PES ............................................... 94

Figure 3.3: Yearly wholesale prices of electricity in HECA’s region ..................................... 114

Figure 3.4: Profit-margin, value of diversification, and value of flexibility for HECA .......... 117

Figure 3.5: Value of polygeneration for flexible HECA under optimal operations ................ 118

Figure 3.6: Sensitivity analysis on the profitability of flexible HECA ................................... 119

Figure 4.1: Representation of the nodes in a decision diagram ............................................... 152

Figure 4.2: Porter’s five forces framework ............................................................................. 153

xvi

Figure 4.3: Identifying the underlying drivers for the bargaining power of Buyers ............... 161

Figure 4.4: Highlighting the shared underlying drivers for the bargaining power of

Buyers .................................................................................................................. 161

Figure 4.5: Relevance arrows connecting the underlying drivers for the bargaining

power of Buyers ................................................................................................... 162

Figure 4.6: Detailed Network of the five forces, their underlying drivers, and

industry-specific factors ....................................................................................... 165

Figure 4.7: Simple Network of the five forces, their underlying drivers, and

industry-specific factors ....................................................................................... 166

Figure 4.8: Illustrative example of uncertainty assessment in DAFF ..................................... 168

Figure 4.9: DAFF economic sub-model .................................................................................. 172

Figure 4.10: DAFF Bayesian Network for the first objective, first step: assessing the

profitability of the overall industry ...................................................................... 178

Figure 4.11: DAFF Decision Diagram for the first objective, second step: positioning

competitively and assessing the profitability of each positioning segment in

the industry .......................................................................................................... 183

Figure 4.12: DAFF Decision Diagram for the first objective, third step: assessing the

profitability of the firm in each positioning segment of the industry .................. 185

Figure 4.13: Conceptual DAFF Decision Diagram for the first positioning objective ........... 195

Figure 4.14: Dynamic DAFF model for the second objective, first step: predicting

industry change .................................................................................................... 196

Figure 4.15: Guidance on adding temporal relevance arrows between competitive

uncertainties ......................................................................................................... 198

Figure 4.16: Dynamic DAFF model for the second objective, second step: reshaping

industry change .................................................................................................... 199

Figure 4.17: Future opportunities to streamline the DAFF modeling ..................................... 210

Figure 5.1: DAFF modeling of the force of Substitutes and its relevant drivers .................... 222

Figure 5.2: Example probabilistic analysis of a driver: Cost reduction for customer by

industry product ................................................................................................... 224

Figure 5.3: Example probabilistic analysis of a driver: Substitute bill savings ...................... 224

Figure 5.4: Example probabilistic analysis of a driver: Price-performance tradeoff

relative to this industry product ........................................................................... 225

Figure 5.5: Example probabilistic analysis of the power of Substitutes ................................. 225

xvii

Figure 5.6: DAFF modeling of all competitive forces and drivers in the U.S. residential

solar PV industry ................................................................................................. 228

Figure 5.7: DAFF modeling of Technology, Regulation, and Growth factors in the

U.S. residential solar PV industry ........................................................................ 233

Figure 5.8: DAFF modeling of the economic parameters in the U.S. residential solar

PV industry .......................................................................................................... 237

Figure 5.9: A sketch of the complete DAFF Bayesian Network for SunEnergy .................... 241

Figure 5.10: Example of decision influence on conditional probability assignment .............. 245

Figure 5.11: A sketch of the complete DAFF Decision Diagram for SunEnergy ................... 246

Figure 5.12: Competitive landscape in the U.S. residential solar PV industry through

2016 ..................................................................................................................... 248

Figure 5.13: Economic performance of the U.S. residential solar PV industry through

2016 ..................................................................................................................... 252

Figure 5.14: Interdependence between the competitive forces in the U.S. residential


Figure 5.15: Effect of the competitive forces on economics of the U.S. residential


Figure 5.16: Profitability of the various positioning tracks in the U.S. residential solar

PV industry .......................................................................................................... 264

Figure 5.17: The influence of positioning on the competitive forces in the U.S.

residential solar PV industry ................................................................................ 266

1

Chapter 1

Introduction

1 Motivation: A Changing, Uncertain, and Flexible Energy

Landscape

“We are convinced that changing the way that we produce and use energy is

essential to America's economic future – that it will create millions of new jobs,

power new industry, keep us competitive, and spark new innovation. And we are

convinced that changing the way we use energy is essential to America's national

security, because it will reduce our dependence on foreign oil, and help us deal with

some of the dangers posed by climate change … There is no time to waste. America

has made our choice. We have charted our course, we have made our commitments,

and we will do what we say.”

— Barack Hussein Obama, COP15, 2009

These words were some of President Barack Obama’s remarks at the United Nations

Conference of Parties in Copenhagen, Denmark, in 2009 [1]. As leaders from around the

world gathered to address the global challenge of climate change, their message was clear:

climate change is real, and so are the extensive changes in the energy system needed to

mitigate its impacts [2].

Today, fossil fuels meet 87% of the global energy demand [3] and are used to generate 67% of

the global electricity supply [4]. As elaborated in the fifth assessment report by Working

CHAPTER 1 — Motivation: A Changing, Uncertain, and Flexible Energy Landscape 2

Group III of the Intergovernmental Panel on Climate Change, mitigating the impacts of

climate change requires a substantial reduction in carbon dioxide (CO2) emissions below their

present levels [5], which in turn necessitates drastic changes in current energy generation,

delivery, and consumption [6]. Those changes are already underway, as manifested by the

progressive evolution of energy technologies, markets, and policies over the past few years.

Around the world, research and development (R&D) efforts continue to advance new

technologies that can either clean up or substitute fossil fuels. Carbon dioxide capture and

storage (CCS) is one important technology that can reduce carbon emissions from large power

plants and industrial facilities [7, 8]. Although CO2 capture has been in operation since the

1970s [9], the need for CCS as a climate-change mitigation option has accelerated the

emergence of new and improved capture technologies in recent years [10, 11], covering post-

combustion, pre-combustion, and oxy-combustion processes. In fact, one study documents

that, from a pool of 1000 international patents related to CO2 capture, more than 600 patents

were issued after year 2000, including about 250 patents between 2010 and 2012 [12]. On the

storage side, several countries have initiated or expanded their characterization of feasible CO2

storage capacity, both onshore and offshore, all while testing new techniques and equipment to

monitor, verify, and potentially correct CO2 injection operations [13, 14, 15]. The

advancement in emissions-free renewable energy technologies has been at least as impressive.

Bell Labs produced the first practical solar cell as early as 1954 [16]. However, it was only

this decade that thin-film and crystalline-silicon solar systems have been truly commercialized

[17, 18], which is evident by the tens of solar world-record efficiencies and new technologies

documented since 2005 [19]. Wind energy technologies have enjoyed similar improvements,

both onshore and offshore. Between 2002 and 2011, more sophisticated power electronics,

controls, and gearboxes helped double the capacity factor and more than triple the generation

capacity of individual horizontal-axis wind turbines [20, 21, 22].

The progressive change in energy technologies has been accompanied by a progressive

transformation of energy markets. Between 2005 and 2014, developed OECD economies

shifted their electricity generation to cleaner fuel mix; the share of coal in power generation

decreased from 36% to 31% while that of natural gas increased from 19% to 24% [23]. Also

important to fossil-fuel decarbonization, the deployment of large-scale CCS has doubled over


the past five years; 15 projects are operational today, spread across North America, Europe,

and the Middle East [14]. At the same time, both developed and developing countries have

witnessed tremendous market growth for solar and wind energy. Between 2005 and 2013, the

solar installations skyrocketed from about 3 to 140 gigawatts (GW) [24, 25], and wind

installations increased from 58 to 319 GW [24, 26].

Both market and technological changes have been, at least partially, driven and incentivized

by policy changes. Perhaps the most notable example of such changes is the unprecedented

global climate agreement reached in Paris in 2015. After 20 annual summits and numerous

climate meetings under the United Nations’ umbrella, the world nations have committed to a

collaborative effort that caps “the increase in the global average temperature to well below 2

°C above pre-industrial levels” [27]. Building up to that treaty was a set of national policy

initiatives that prepare each country to meet its future emissions’ goals [28]. Focusing

specifically on the U.S., two examples stand out. Earlier in 2015, the U.S. Environmental

Protection Agency (EPA) finalized two regulations that aim to reduce carbon pollution from

the power sector – both new and existing power plants – to 32% below 2005 levels [29, 30].

On the renewables side, the U.S. Congress has enacted a 30% investment tax credit (ITC) for

solar installations since 2006 [31] and a $0.023 per kilowatt-hour (kWh) production tax credit

(PTC) for wind installations since 1993 [32, 33]. Along other forms of governmental subsidies

[34, 35], ITC and PTC have helped accelerate the deployment of wind and solar energy across

the country. Indeed, beyond national measures, state and multi-national policies have also

contributed to shaping, and therefore changing, the energy landscape in the U.S. [36, 37, 38].

Change creates uncertainty, however. Whenever energy technologies, markets, and policies

evolve, the future energy landscape becomes harder to plan or even predict. For instance, as

advanced capture technologies facilitate the large-scale deployment of CCS, it becomes more

urgent to clarify and manage the technical uncertainty of CO2 leakage from storage reservoirs

[39, 40, 41, 42]. Also on the technical side, it is uncertain whether and how the electric grid

can manage the increasing deployment of intermittent renewable energy, especially as wind

and solar technologies become more robust and mainstream [43, 44, 45].


Uncertainty also arises in the evolution of energy markets. Starting with fossil fuels, the role

of natural gas as a “bridge fuel” that eases the transition into a cleaner economy is still

uncertain [46, 47]. In the United States, new hydraulic fracturing techniques have unlocked an

abundance of supply, which has in turn contributed to lower prices [48]. Nonetheless, recent

environmental concerns [49, 50, 51] and liquefied-natural-gas export arrangements [52] make

any long-term forecasts of the domestic natural gas market a rather challenging task.

Furthermore, despite continuous progress, the large-scale nature of most CCS projects render

them vulnerable to economic uncertainties associated with upfront and operational costs,

financing mechanisms, and long-term liability [40, 53]. The development of renewable energy

markets is also far from certain. One example to highlight is strategic competition in the solar

industries. As relatively young players, solar firms have been trying to navigate through the

complicated maze of the power sector. They continue to adjust their business models while

gaining further insights into the uncertain, and still underdeveloped, competitive relationship

with their customers and suppliers, their established substitutes and new rivals, as well as their

regulators and financers [54, 55, 56]. Indeed, one can appreciate the extent of uncertainty in

the solar markets by tracking the stock-price fluctuations of top solar firms, which are

indicative of frequent business successes and failures [57, 58].

While changing energy policies contribute to technological and market uncertainty, they

themselves may also be uncertain. Policies aiming to curb greenhouse gas emissions continue

to be a contentious subject among political parties and energy businesses. Attesting to that

fact, the aforementioned U.S. EPA’s plan to regulate emissions from existing power plants

was challenged in court by several states and utilities; literally just days before composing this

Chapter, the U.S. Supreme Court made an unexpected decision to put the EPA’s plan on hold

[59, 60]. Similarly uncertain are the prospects of enforcing some form of carbon pricing.

Many still argue over the potential benefits of a cap-and-trade regime versus a carbon-tax

regime [61, 62, 63]. Meanwhile, legislations around the world have attempted to adopt either

regime; some succeeded [64], some failed [65], and some were repealed [66]. The evolution of

renewables is also shaped by significant policy uncertainties [67]. As wind and solar

technologies mature and become more competitive, it is not clear whether and what

governmental subsidies remain effective [68]; subsidy mandates often have a short lifetime,


which creates a recurrent need to renew them [31, 69]. Along the same lines, it is not obvious

whether future grid regulations will respond more or less favorably to the penetration of

renewables [70]; judging by the opposing stands on solar grid-charges by the public utilities

commissions (PUC) in Nevada and California [71, 72], both scenarios seem plausible.

Now someone might ask: what exactly is problematic with change and uncertainty? The fact

is, as the changing climate instigates an uncertain energy landscape, new investments in clean

energy infrastructure become more challenging. Fundamentally, uncertainty makes it difficult

for decision-makers to think and act clearly [73]. In our context, technological, market, and

policy uncertainties obstruct the investors’ propensity to assess their energy investments

deterministically. They complicate the decision-makers’ efforts to quantify the technical risk,

economic value, or strategic competitiveness of their prospective energy projects, because

orthodox evaluation tools like worst-case-scenario and net-present-value become insufficient

[74, 75, 76]. As a result, new energy initiatives may be overpriced or underestimated; new

energy facilities may face cost overruns, delays, or even cancellations; and new energy

businesses may suffer curtailment. Unfortunately, examples of these setbacks are already

making news, be it in CCS [77, 78, 79] or solar energy [80, 81].

One way to manage and hedge against uncertainty is flexibility. In a climate-constrained

world, flexible energy investments might be better posed to deal with technical, market, and

policy uncertainties. Here, it is helpful to distinguish between two types of flexibility:

operational and strategic. Operational flexibility refers to built-in technical capabilities that

allow an energy system to switch between different modes of operation within a specific

timeframe. The interest in developing flexible electricity systems has surged in recent years

[82]. These systems promise to manage not only the uncertain availability and variability of

the increasing renewable energy supply but also the uncertain variation in future energy

demand and prices [83]. One promising technology in that regard is flexible CO2 capture.

Flexible capture allows grid-operators or utilities to synchronize the power output from fossil-

fuel plants with intermittent renewables, and it may contribute to enhancing the economics of

CCS deployment [84, 85, 86, 87]. Battery storage has emerged as another promising example

of flexible energy systems, for it can both reduce the curtailment of excess renewable supply

[88, 89] and improve the grid ancillary services [90]. In fact, recent efforts have also examined

CHAPTER 1 — Scope of Work 6

the prospects of grid flexibility by integrating fossil and renewable energy generation. While

some systems combine the operations of distinct single-input/sing-output facilities [91], other

systems operate a single facility with multiple inputs/outputs. Capable of hedging against

unexpected input-price shocks and exploiting sudden output-price peaks, the latter systems are

uniquely labeled “polygeneration,” and their flexibility benefits respond to uncertainties

beyond power generation to include chemicals and fertilizers synthesis [92, 93, 94].

Equally important, strategic flexibility – also called managerial flexibility – refers to the

ability to modify a managerial course of action over an energy investment within a specific

timeframe [95]. In this case, even if the energy project is operationally inflexible, investment

decisions may preserve the option to expand, retrofit, or otherwise modify the project in

response to future risks or opportunities. One example of strategic flexibility is investing in

capture-ready [96, 97, 98]. Bohm et al. [99] classifies a fossil-fueled plant as capture-ready if

“at some point in the future it can be retrofitted for carbon capture and sequestration and still

be economical to operate.” Another prominent example of strategic flexibility is related to the

emergence of strategic competition in solar markets over the past couple of years. Despite

apparent trends of consolidation and vertical integration – aiming to reduce cost and deter new

entrants [56, 100] – major solar firms have considered expanding and diversifying their

business into energy storage, through acquisition [101], direct investment [102], or partnership

agreements [103]. As the power grid evolves, establishing a strong foothold in energy storage

is a strategic flexibility option that gives solar firms the ability, but not the obligation, to

update their future product offerings whenever needed.

2 Scope of Work

I can probably write a few more pages, and cite a dozen more references, to describe how

climate change is making our energy landscape increasingly different, uncertain, and flexible.

I won’t. Hopefully, by now, I have convinced you that this argument holds some truth.

Consequently, in our progressively modernizing energy sector, it is inevitable that uncertainty

and flexibility impact the value of clean energy investments. Because uncertainty and

flexibility increase the complexity of investment decisions, a proper accounting of both


attributes becomes both necessary and beneficial. It facilitates clearer and more accurate

analyses of technical, economic, or strategic challenges; it yields more comprehensive results;

and it conveys more insightful and robust recommendations. Therefore, to help investors ride

the clean energy wave, one of the most urgent priorities – for academic scholars and

business experts alike – is to clarify and quantify uncertainty and flexibility in modern

energy systems and industries. This dissertation aims to develop assessment models that

achieve this exact goal.

To fulfill this mission, this dissertation takes on three major research endeavors, which

augment and build on existing efforts in their respective fields. Motivating the selection of

these specific endeavors is their focus on the “decision-maker” in the energy sector.

Fundamentally, the chosen topics and developed models are intended to be significant to and

usable by executives and managers overseeing energy businesses and facilities.

Now, to formally introduce the three research undertakings in this dissertation, let me explain

what topics are discussed and how their quantitative models are developed. The first study

sheds light on the technical uncertainties related to the safe and reliable deployment of CCS.

Interested in connecting the various elements of risk management for CO2 geologic storage, I

propose a methodological framework that translates quantitative risk assessment to

contingency planning for CO2 leakage from geologic storage reservoirs. The second study

switches from technical to economic assessment, investigating the economic value of flexible

hydrogen-based polygeneration energy systems. Beyond specifying the conditions for

breaking even, this study highlights the value of flexibility and diversification enabled by

polygeneration. Finally, the third study examines the competitive uncertainties and their

relation to positioning strategies within an energy industry. Based on Michael Porter’s five

forces framework (FF) [104], I develop decision analytic models that analyze the uncertain

competitive landscape and economic performance of an overall industry, of several

positioning segments within that industry, and of specific firms within those segments. After

explaining their theoretical foundations, the models are used to assess near-future competitive

strategies in the U.S. residential solar photovoltaic (PV) industry.


Fortunately, these topics have met my aspiration to pursue a consistent yet diversified research

agenda. While all three endeavors attempt to answer the grand question of clarifying and

quantifying uncertainty and flexibility in the energy sector, they differ in their theme,

application, and scope. Thematically, I strive to balance between fossil and renewable

resources, for both will continue to meet our energy needs over the coming few decades. CCS

is a carbon-fuel technology; solar is a non-carbon-fuel technology; and polygeneration is a

hybrid technology that can accommodate both fossil and renewable fuels. The three studies

also differ in their application areas. The CCS study addresses the technical risk associated

with uncertain CO2-leakage events. The polygeneration study addresses the economic value

associated with flexible production operations. As for the solar study, it primarily addresses

strategic positioning, and secondarily alludes to strategic flexibility, in uncertain competitive

landscapes. Lastly, the diversity in scope is rather easy to tell. While the analysis of CCS risk

or polygeneration economics spans an individual energy facility or project, the analysis of

uncertain strategic competition – along with its solar case study – covers a whole energy

industry, multiple segments within that industry, and potentially multiple firms within those

segments.

The diversity and practicality of the aforementioned research topics are mirrored in the

methods and tools used to model and examine them. As my grandmother used to tell me,

there’s more than one way to bake a cake. The same applies to modeling uncertainty and

flexibility in the modern energy sector. To start, my approach to modeling uncertainty – for

both CO2 leakage and solar competitive strategy – is rooted in the field of decision analysis

(DA). DA is a quantitative methodology that analyzes any decision-making situation in terms

of three main components: alternatives, information, and preferences. DA pays special

attention to uncertainty, which accounts for information unknown to the decision-maker, and

it adopts a Bayesian reasoning to express and update the decision-maker’s beliefs about

uncertainty [73].

In the CCS study, I use probabilistic risk assessment (PRA) to characterize potential leakage

features, events, and processes (FEP) in a Bayesian events tree (BET); the PRA technique and

the BET tool are both well-documented and broadly used in the context of decision analysis

[73, 75]. Also, borrowing from the oil and gas industry [105, 106], I design a three-tier system


that helps prepare for tolerable CO2 leakage events before they happen and helps respond to

those events when they actually happen.

In the competitive strategy study and its solar case in point, I use Bayesian networks and

decision diagrams to model Porter’s five forces in a specific industry and for a specific firm.

Porter identifies five forces that shape competition in relatively stable industries: bargaining

power of buyers, bargaining power of suppliers, threat of substitutes, threat of new entry, and

rivalry. In a Bayesian network, these forces – along with their drivers; technological, political,

and growth factors; and economic implications – are modeled as interdependent uncertainties.

Subsequently, a decision diagram illustrates how strategic positioning influences the powers

of these competitive forces and therefore the economic performance of specific market

segments and firms. Porter’s FF is one of the most famous competitive frameworks in

business strategy [107, 104, 108], and Bayesian networks and decision diagrams are

probabilistic graphical tools that are widely utilized in DA [109, 110].

In the polygeneration study, I shift from the uncertainty face to the flexibility face of my

research coin. Here, I rely on concepts and methods in managerial accounting to derive a

series of propositions that assess the economic value of a polygeneration facility, under both

static and flexible operation modes. Essential to these modeling efforts are two concepts:

levelized cost and value of real-options. Levelized cost is a break-even metric that calculates

the ratio of lifetime-cost to lifetime-production of a facility [111, 112]. As will become

evident, this metric allows deriving the unit profit-margin and the unit value of flexibility

associated with polygeneration. Equally important, real-options analysis is widely applied to

quantify the impacts of climate uncertainty and the related value of flexibility in energy

investments [113, 114, 115]. Numerous real-option models have been developed over the

years [116, 117, 76], and they help value both operational and strategic flexibility in the

changing energy landscape [96, 118, 119, 120]. This work takes a unique – rather simple –

approach to expressing the real-option value of polygeneration flexibility; nonetheless, it

adheres to the fundamental premise of real-options as buying the right, but not the obligation,

to modify the operation or management of a project in the future, in response to unknown risks

or opportunities [121].


Now comes the time to put all the pieces together. Combining the previously introduced

subject areas and modeling methods, the following sections provide a more formal and

elaborate overview for each of the three research endeavors pursued in this dissertation.

2.1 Translating Risk Assessment to Contingency Planning for CO2 Geologic

Storage: A Methodological Framework

The uncertain occurrences and consequences of CO2 leakage from potential CCS projects

continue to be a major source of public concern [122, 123]. In order to ensure safe and

effective long-term geologic storage of CO2, existing regulations require both assessing

leakage risks and responding to leakage incidents through corrective measures [124].

However, until now, these two pieces of risk management have been usually addressed

separately. This study proposes a conceptual methodological framework that bridges risk

assessment to corrective measures through clear and collaborative contingency planning. We

achieve this goal in three consecutive steps.

First, a probabilistic risk assessment (PRA) approach is adopted to characterize potential

leakage features, events, and processes (FEP) in a Bayesian events tree (BET), resulting in a

risk assessment matrix (RAM). Allowing the visualization of risk identification, analysis, and

evaluation, the proposed RAM depicts a mutually exclusive and collectively exhaustive set of

uncertain leakage scenarios with quantified likelihood, impact, and tolerance levels.

Second, the risk assessment matrix is translated to a contingency planning matrix (CPM).

Notably, the CPM incorporates a tiered contingency system that guides the preparation for

leakage uncertainties and the response to leakage incidents, whenever they actually occur. The

leakage likelihood and impact dimensions of RAM are translated to resource proximity and

variety dimensions in CPM, respectively. To ensure both rapid and thorough contingency

planning, more likely or frequent risks require more proximate resources while more impactful

risks require more various resources. In addition, the minimum and maximum risk tolerance

levels are translated to contingency thresholds, and all foreseeable risk scenarios are

categorized under three contingency tiers: Tier 1, Tier 2, and Tier 3. Here, we highlight how

the upper, lower, and inter-tier contingency boundaries should be collaboratively pre-


negotiated between the operating party and all relevant stakeholders in order to ensure

effective preparedness and response.

Finally, we present a model contingency plan to demonstrate how all newly introduced

concepts integrate together. Specifically, we focus on explaining how the designed

contingency tiers facilitate important aspects of contingency planning, primarily: evaluating

leakage uncertainties and initiating response procedures; designing a corrective measures

matrix (CMM) that assigns specific control and remediation actions to each potential leakage

scenario; mobilizing, deploying, and sustaining necessary human and equipment resources;

and formulating a decision-making hierarchy, a notification protocol, and a communication

scheme to effectively administer the CO2 storage site.

This study was conducted in collaboration with Professor Sally M. Benson from the

Department of Energy Resources Engineering at Stanford University. While I led the

modeling and the analytical work, Professor Benson provided general guidance and helped co-

author a journal paper, which has been accepted for publication [125].

2.2 Economic Value of Flexible Hydrogen-Based Polygeneration Energy

Systems

Polygeneration energy systems (PES) have the potential to provide a flexible, high-efficiency,

and low-emissions alternative for power generation and chemical synthesis from fossil fuels

[126, 127]. This study aims to develop a set of metrics that calculate the economic value of

fossil-fueled PES, which rely on hydrogen as an intermediate product. To achieve this goal,

we first model a representative PES with carbon capture, which uses coal as a primary energy

input and produces electricity, fertilizer, and pure CO2 as end-products. We then derive a

series of propositions that assess the cost competitiveness of the modeled PES under both

static and flexible operation modes, compare the performance of PES to that of

monogeneration (i.e. single-output) energy systems, and quantify the value of real-options

associated with PES’s diverse and flexible operation.


Based on the levelized cost of electricity (𝐿𝐶𝑂𝐸) metric, the levelized cost of hydrogen

(𝐿𝐶𝑂𝐻) metric is first introduced. Then, we formulate the levelized incremental cost (𝐿𝐼𝐶) of

converting hydrogen into distinct market commodities such as electricity and fertilizers. By

subtracting the cost from the end-products’ sales, we derive the profit-margins of PES under

multiple operation modes: static monogeneration of electricity (𝑃𝑀0𝑒) or fertilizer (𝑃𝑀0𝑓),

static polygeneration of electricity and fertilizer (𝑃𝑀1), and flexible polygeneration of

electricity and fertilizer (𝑃𝑀2). Subsequently, the real-option values enabled by PES are

encapsulated in two terms: value of diversification (𝑉𝑂𝐷) quantifies the option to produce

multiple outputs, and value of flexibility (𝑉𝑂𝐷) quantifies the option to adjust the production

rates of outputs over time. For consistency, every derived economic metric is expressed as a

monetary value per unit of produced hydrogen ($/𝑘𝑔ℎ); nonetheless, all metrics can be easily

converted to value per unit of any end-product, such as electricity ($/𝑘𝑊ℎ).

To illustrate the practical significance of these metrics, we apply them to evaluate the

economics of Hydrogen Energy California (HECA), a real PES project currently under

development in California [128, 129]. Under our technical and economic assumptions,

HECA’s levelized cost of hydrogen is estimated at 1.373 $/𝑘𝑔ℎ. The profitability of HECA

as a static PES depends on the exact production portfolio, and it increases in the share of

hydrogen converted to fertilizer rather than electricity; 𝑃𝑀1 varies between −0.992 and 1.934

$ 𝑘𝑔ℎ⁄ , corresponding to 𝑃𝑀0𝑒 and 𝑃𝑀0𝑓, respectively. However, when configured as a

flexible PES, HECA almost breaks even on a pre-tax basis, with 𝑃𝑀2 = −0.0439 $ 𝑘𝑔ℎ⁄ .

Consequently, we show that diversification and flexibility are valuable for HECA when

comparing polygeneration to static monogeneration of electricity (positive 𝑉𝑂𝐷 and 𝑉𝑂𝐹),

but these two real options have no value when comparing polygeneration to static

monogeneration of fertilizers (negative 𝑉𝑂𝐷 and 𝑉𝑂𝐹).

Conducting a sensitivity analysis on these findings shows that HECA’s economic value is

mostly sensitive to the price of fertilizer and to discount rate. A flexible HECA breaks even

upon modest increase in fertilizer prices beyond 3.1% or decrease in discount rate beyond

7.5%; conversely, the prices of CO2 and electricity need to increase by at least 14.5% and

34%, respectively, to achieve the same goal.


This study was completed in collaboration with Professor Stefan Reichelstein from the

Graduate School of Business at Stanford University. While I led most of the theoretical

modeling as well as HECA’s case study, Professor Reichelstein provided general guidance,

oversaw the economic derivations, and helped co-author and publish a journal paper based on

this work [130].

2.3 Decision Analytic Modeling of the Five Forces in Competitive Strategy:

Application in the U.S. Residential Solar PV Industry

Competitive strategy, pioneered by Michael E. Porter since 1979, explains how an

organization facing competition can achieve superior profitability within its industry. Porter

identifies five forces that shape competition in relatively stable industries: Buyers, Suppliers,

Substitutes, New Entrants, and Rivals. He describes what causes each of the forces to be

strong or weak, and he explains that an incumbent firm gains competitive advantage by

positioning where all forces are weakest [104, 108]. Despite its remarkable contributions to

business strategy over the years, Porter’s FF framework has been mostly applied qualitatively

and deterministically [131, 132, 133]. Few systematic methodologies have been developed to

guide the quantification and operationalization of the competitive forces in real life, and no

sufficient attention has been given to the uncertain and interdependent nature of these forces

and their economic implications [134, 135]. To address these issues, we propose a decision

analytic modeling of the five forces, hereby referred to as DAFF. DAFF uses DA techniques

and tools to link a firm’s positioning decisions with uncertain market competition and

economics. The result is a quantitative model that managers can use to evaluate the

profitability of a specific industry, to properly position their business in the industry, and to

predict and shape the future of that industry. To that end, the DAFF approach is not intended

to refute, replace, or modify the theoretical foundations of FF; rather, DAFF aims to maximize

FF’s practical application.

To build the DAFF models, we first extract and document all important terminology related to

the FF strategic framework from literature. Key elements and essential terms are categorized

as decision-related, uncertainty-related, or value-related. The five competitive forces and their


underlying drivers are modelled as uncertainties. Also counted as uncertainties are industry-

specific regulatory, technological, growth, and complementary factors. Both the generalizable

forces and the industry-specific factors impact the cost, price, and quantity, which are treated

as uncertain economic parameters. Along the way, we describe the probabilistic

interdependence among the uncertainties, and we link them together in a single DAFF

Bayesian Network. We also spend some time explaining how to effectively define

profitability, which is modelled as a value metric.

The final element of the DAFF models is the firm’s actions within the analyzed industry,

which can be categorized into three types of decisions: Value Proposition, Value Chain, and

Economic decisions. Value Proposition decisions dictate what product to make, whereas Value

Chain decisions determine how the product is made. Value Proposition and Value Chain

choices influence the uncertain competitive landscape and therefore yield distinct positioning

alternatives for the firm within its industry. In contrast, Economic decisions do not impact the

competitive landscape; they address three aspects related to the firm’s specific profitability:

Production Scale, Pricing, and Tactical Costing. Adding the firm’s decisions to the DAFF

Bayesian Network results in a DAFF Decision Diagram.

This DAFF Decision Diagram can fulfill the two main objectives of Porter’s competitive

strategy: positioning in the industry and reshaping the industry. Positioning is a short-term

objective, and it requires three consecutive analyses of current and very-near-future

profitability: of the overall industry, of distinct positioning segments in the industry, and of a

specific firm or business within each positioning segment of the industry. Building on the

outcomes from these three steps, reshaping the industry becomes the long-term objective, and

it requires two additional analyses of distant-future profitability, via: predicting the evolution

of the industry structure and modifying the evolution of the industry structure. After detailing

how to model the three steps for industry positioning, we provide a conceptual description of

how to model the remaining two steps for industry reshaping.

Along the way, we explain how the DAFF modeling of these strategic objectives augments

previous efforts to enhance the operationalization of the FF framework. Specifically, we

highlight the following DAFF benefits: generalizing the competitive assessments while


customizing their features to evaluate different industries and firms; explicitly accounting for

competitive and economic uncertainty; clearly mapping the relation between the industry’s

competitive forces and the firm’s competitive actions; clearly mapping the relation between

competitive forces and economic performance; tracking, comparing, ranking, and prioritizing

the interdependence among the competitive forces, drivers, and industry factors; outlining the

firm’s scope of control; and exposing and reducing cognitive biases. Atop all that, the most

important feature of DAFF – indeed, one that enables all aforementioned benefits – is

quantification.

In order to demonstrate DAFF’s applicability, we use it to assess the competitive strategy of a

major firm within the U.S. residential solar PV industry. Specifically, we undertake the first

two steps of the strategic positioning objective, aiming to answer two questions: Is the

competitive landscape in the U.S. residential solar PV industry favorable in the near future?

And if so, where should the solar firm position its residential business?

In the first part of the case study, we describe how the DAFF models are customized for the

specific industry and firm of interest. Subsequently, upon developing and evaluating a DAFF

Bayesian Network, we find that, among the five competitive forces, only rivalry is expected to

be relatively strong. The chance of residential solar firms suffering from strong rivalry –

hereby expressed as {high Rivals} prospect – over the next two years is 0.6, compared to

about 0.4 for {high Substitutes}, {high Buyers}, {high New Entrants}, and {high Suppliers}.

This predicted competitive landscape yields an expected earnings before tax (EBT) of 4.05

billion $/year for the whole U.S. residential PV market, with a probability-weighted average

cost, price, and installation capacity estimated at 1.19 dollar-per-watt ($/W), 3.35 $/W, and

1.93 GW, respectively.

The Bayesian Network also tracks the interdependence among the five forces and its

corresponding economic implications. We document robust interdependence among four

forces: Substitutes, Buyers, New Entrants, and Rivals. For example, observing strong Buyers

increases the likelihood of witnessing strong Rivals and Substitutes; if the decision-maker

observes a {high Buyers} prospect, the probability of experiencing {high Substitutes}

increases from 0.39 to 0.7, and the probability of {high Rivals} increases from 0.62 to 0.74. A

CHAPTER 1 — Dissertation Organization 16

careful analysis shows that the interactions between the competitive forces need not be

symmetrical. For instance, observing strong Buyers increases the likelihood of witnessing

strong Substitutes significantly, but observing strong Substitutes barely updates the decision-

maker’s knowledge about Buyers. Ultimately, experiencing a single force at its weakest or

strongest extreme may lift the overall industry EBT up to 5.77 billion $/year or drag it down to

1.86 billion $/year, respectively.

Using a DAFF Decision Diagram, we also analyze the competitive uncertainties in multiple

positioning segments that are of interest to the solar firm. We investigate four positioning

decisions. Regional Focus is a Value Proposition decision that addresses what customers to

serve: rural or urban. In contrast, Downstream Integration, Customer Finance, and Panel

Manufacturing are Value Chain decisions that determine how to operate the business: whether

to serve customers directly or through intermediate dealers; whether to sell the solar system

through direct-purchase agreements or offer product financing in the form of loans or leases;

and whether to insource or outsource the manufacturing of the solar panels. Combining the

various decision alternatives results in 32 feasible positioning tracks. Each track yields a

unique competitive landscape and, correspondingly, a unique EBT profit value ranging

between 0.51 and 3.98 billion $/year. The results show that the highest expected profit is

realized in a competitive setting where incumbents choose to: locate in urban areas, manage

customers directly without relying on dealers, and offer loan and lease services to their

customers.

3 Dissertation Organization

This dissertation continues with four chapters that document the aforementioned research

initiatives in detail. Given the diversity of the addressed topics, Chapters 2, 3, and 4 are

formatted such that each may stand alone as a self-contained study, which can be reviewed

and referenced independently. Chapter 5 is an extensive case study based on the methods and

results of Chapter 4. For clarity, we avoid discussing all conclusions in one Chapter. Instead,

an elaborate Conclusions section is presented at the end of each Chapter to summarize the

main findings and suggest relevant areas of future research.


Chapter 2, titled Translating Risk Assessment to Contingency Planning for CO2 Geologic

Storage: A Methodological Framework, is based on a journal paper co-authored with

Professor Sally M. Benson and accepted for peer-reviewed publication [125]. The paper starts

with a brief overview of the terminology used in managing the uncertainties of CO2 storage,

highlighting some existing literature on risk assessment methodologies and corrective

measures. The next section discusses how to model a risk assessment matrix, and it introduces

the concept of “risk profiles.” As the main focus of this paper, a contingency planning matrix

is then developed based on the risk assessment matrix, and its tier structure is discussed.

Lastly, we leverage the contingency planning matrix to design a model contingency plan,

which covers multiple sections on preparing for leakage risks and responding to leakage

incidents.

Chapter 3, titled Economic Value of Flexible Hydrogen-Based Polygeneration Entergy

Systems, is based on a paper co-authored with Professor Stefan Reichelstein and already

published in Applied Energy [130]. The first section introduces the economic concepts and

technical configuration used in assessing polygeneration energy systems. Next, we conduct a

detailed economic analysis for a representative PES in three scenarios: Scenario 1 evaluates

static operation while Scenarios 2a and 2b evaluate two modes of flexible operation. As the

main focus of this paper, the economic definitions and derived propositions in all three

scenarios are then used to calculate PES’s profit-margin and real-option values of

diversification and flexibility. Finally, we demonstrate the applicability of all derived metrics

by examining the economic competitiveness of Hydrogen Energy California, a polygeneration

project currently under development.

Chapter 4 is titled Decision Analytic Modeling of the Five Forces in Competitive Strategy, and

it explains the theoretical foundations of DAFF – the decision analytic modeling of the five

forces. The first section provides a brief overview of decision analysis and Porter’s five forces,

and then it outlines how DA tools can enhance the operationalization of the FF theory. The

next section describes how to develop a DAFF model, including how to use uncertainties and

decisions to model the competitive forces and industry-specific factors, the economics of an

industry or a firm, and the strategic positioning actions of the firm. As the main focus of this

study, we then explain how a firm can use DAFF to execute the two main objectives of


competitive strategy: positioning in and reshaping the industry. The subsequent two sections

offer further guidance on applying DAFF; in addition to discussing some modeling best

practices, we highlight the ability of DAFF to preserve and characterize many of Porter’s

strategic insights that stem from, but extend beyond, the FF framework.

Finally, Chapter 5 is titled DAFF Modeling of Competitive Strategy: Positioning in the Near-

Future U.S. Residential Solar PV Industry. It showcases the application of DAFF to fulfill a

specific strategic objective, by a specific firm, in a specific competitive energy industry. The

objective is positioning in the near future; the firm is a major solar developer; and the

competitive industry is residential solar PV in the United States. In the first section, we build a

DAFF Bayesian Network to analyze competition in the overall industry. By considering a

series of important positioning decisions for the firm, the Bayesian Network is expanded into

a Decision Diagram that analyzes various market segments. The next section displays the

results from both modeling steps. For the overall industry, the DAFF Bayesian Network

outputs: a probability distribution over the power of each competitive force as well as over the

average cost, price, and annual sales; an expected profit value; and quantitative measurements

of the interdependencies among the forces. For each positioning segment within the industry,

the DAFF Decision Diagram yields: a distinct probability distribution over the power of each

force as well as a distinct expected profit value. Ultimately, we explain how the DAFF outputs

can be streamlined into a clear and actionable list of recommendations regarding the firm’s

positioning strategy in the U.S. residential solar market.

CHAPTER 1 — References 19

References

[1] B. Obama, "Remarks by the President at the Morning Plenary Session of the United Nations

Climate Change Conference," The White House, Office of the Press Secretary, 18 December

2009.

[2] UNFCCC, "Statements made at COP 15 / CMP 5. United Nations Framework Convention on

Climate Change," 2009. [Online]. Available:

http://unfccc.int/meetings/copenhagen_dec_2009/items/5087.php. [Accessed 2016].

[3] BP, "BP Energy Outlook downloads," 2015. [Online]. Available:

http://www.bp.com/en/global/corporate/energy-economics/energy-outlook-2035/energy-outlook-

downloads.html. [Accessed 2016].

[4] IEA, "2015 Key World Energy Statistics," 2015. [Online]. Available:

http://www.iea.org/publications/freepublications/publication/KeyWorld_Statistics_2015.pdf.

[5] O. Edenhofer, R.Pichs-Madruga, Y. Sokona, E. Farahani, S. Kadner, K. Seyboth, A. Adler, I.

Baum, S. Brunner, P. Eickemeier, B. Kriemann, J. Savolainen, S. Schlömer, C. v. Stechow, T.

Zwickel and J. M. (eds.), "Summary for Policymakers," in Climate Change 2014: Mitigation of

Climate Change. Contribution of Working Group III to the Fifth Assessment Report of the

Intergovernmental Panel on Climate Change, ambridge University Press, Cambridge, United

Kingdom and New York, NY, USA, 2014.

[6] L. Clarke, K. Jiang, K. Akimoto, M. Babiker, G. Blanford, K. Fisher-Vanden, J.-C. Hourcade, V.

Krey, E. Kriegler, A. Löschel, D. McCollum, S. Paltsev, S. Rose, P. R. Shukla, M. Tavoni, B. C.

C. v. d. Zwaan and D. v. Vuuren, "Assessing Transformational Pathways," in Climate Change

2014: Mitigation of Climate Change. Contribution of Working Group III to the Fifth Assessment

Report of the Intergovernmental Panel on Climate Change, Cambridge University Press,

Cambridge, United Kingdom and New York, NY, USA, 2014.

[7] P. Folger, "Carbon Capture and Sequestration (CCS): A Primer," Congressional Research

Review, Washington, DC, 2013.

[8] B. Metz, O. Davidson, H. d. Coninck, M. Loos and L. Meyer, "IPCC Special Report on Carbon

Dioxide Capture and Storage," Cambridge University Press, Cambridge & New York, 2006.

[9] GCCSI, "Large Scale CCS Projects," 2016. [Online]. Available:

https://www.globalccsinstitute.com/projects/large-scale-ccs-projects. [Accessed 2016].

[10] S. D. Kenarsari, D. Yang, G. Jiang, S. Zhang, J. Wang, A. G. Russell, Q. Weif and M. Fan,

"Review of recent advances in carbon dioxide separation and capture," RSC Adv., vol. 3, p.

22739–22773, 2013.

[11] E. S. Rubin, H. Mantripragada, A. Marks, P. Versteeg and J. Kitchin, "The outlook for improved

carbon capture technology," Progress in Energy and Combustion Science, vol. 38, p. 630–671,

2012.


[12] B. Li, Y. Duan, D. Luebke and B. Morreale, "Advances in CO2 capture technology: A patent

review," Applied Energy, vol. 102, p. 1439–1447, 2013.

[13] NETL & DOE, "Carbon Storage Atlas - Fifth Edition," National Energy Technology Laboratory

and The U.S. Department of Energy, Office of Fossil Energy, United States, 2015.

[14] GCCSI, "The Global Status of CCS: 2015, Summary Report," Global CCS Institute, Melbourne,

Australia, 2015.

[15] IEAGHG, "Review of Offshore Monitoring for CCS Projects," IEAGHG, UK, 2015.

[16] A. Chodos, "This Month in Physics History, April 25, 1954: Bell Labs Demonstrates the First

Practical Silicon Solar Cell," APS News, vol. 18, no. 4, April 2009.

[17] V. V. Tyagi, N. A. A. Rahim, N. A. Rahim and J. A. Selvaraj, "Progress insolar PV technology:

Research and achievement," Renewable and Sustainable Energy Reviews, vol. 20, p. 443–461,

2013.

[18] Z. Shahan, "Solar Panel Efficiency Has Come A Long Way (Infographic)," Clean Technica, 6

February 2014.

[19] NREL, "Best Research-Cell Efficiencies," National Renewable Energy Laboratory, United States,

2016.

[20] J. Martino, "Advancements in Wind Turbine Technology: Improving Efficiency and Reducing

Cost," RenewableEnergyWorld.com, 2 April 2014.

[21] F. Blaabjerg and K. Ma, "Future on Power Electronics for Wind Turbine Systems," IEEE Journal

of Emerging and Selected Topics in Power Electronics, vol. 1, no. 3, p. 139–152, 2013.

[22] M. Islam, S.Mekhilef and R.Saidur, "Progress and recent trends of wind energy technology,"

Renewable and Sustainable Energy Reviews, vol. 21, p. 456–468, 2013.

[23] IEA, "Energy Statistics of OECD Countries," International Energy Agency, Paris, France, 2015.

[24] EIA, "International energy data and analysis," 2016. [Online]. Available: http://www.eia.gov/.

[Accessed 2016].

[25] JRC, "PV Status Report," Institute for Energy and Transport, Joint Research Centre, European

Commission, Luxembourg, 2014.

[26] GWEC, "Global Wind Statistics 2014," Global Wind Energy Council, Brussels, 2015.

[27] UNFCCC, "Paris Agreement," United Nations Framework Convention on Climate Change, The

Conference of the Parties on its twenty-first session, FCCC/CP/2015/10/Add.1, 2015.

[28] UNFCCC, "Intended Nationally Determined Contributions (INDCs)," 2016. [Online]. Available:

www4.unfccc.int/submissions/INDC/Submission Pages/Submissions.aspx. [Accessed 2016].


[29] EPA, "Carbon Pollution Emission Guidelines for Existing Stationary Sources:Electric Utility

Generating Units," 80 Fed. Reg. 64662, United States, 2015a.

[30] EPA, Standards of Performance for Greenhouse Gas Emissions From New, Modified, and

Reconstructed Stationary Sources: Electric Utility Generating Units, United States: 80 Fed. Reg.

64510, 2015b.

[31] SEIA, "Solar Investment Tax Credit (ITC). Solar Energy Industries Association," 2016a.

[Online]. Available: http://www.seia.org/policy/finance-tax/solar-investment-tax-credit.

[Accessed 2016].

[32] DOE, "Renewable Electricity Production Tax Credit (PTC). U.S. Department of Energy," 2016.

[Online]. Available: http://energy.gov/savings/renewable-electricity-production-tax-credit-ptc.

[Accessed 2016].

[33] E. Lantz, D. Steinberg, M. Mendelsohn, O. Zinaman, T. James, G. Porro, M. Hand, T. Mai, J.

Logan, J. Heeter and L. Bird, "Implications of a PTC Extension on U.S. Wind Deployment,"

National Renewable Energy Laboratory, United States, 2014.

[34] SEIA, "Depreciation of Solar Energy Property in MACRS," 2016b. [Online]. Available:

http://www.seia.org/policy/finance-tax/depreciation-solar-energy-property-macrs. [Accessed

2016].

[35] DOE, "LPO Overview Brochure," U.S. Department of Energy. Loan Programs Office,

Washington, DC, 2015.

[36] CERC, "U.S.-China Clean Energy Research Center," 2016. [Online]. Available: www.us-china-

cerc.org. [Accessed 2016].

[37] NYSDEC, "High-Volume Hydraulic Fracturing in NYS. New York State Department of

Environmental Conservation," 2015. [Online]. Available:

http://www.dec.ny.gov/energy/75370.html. [Accessed 2016].

[38] AB 32, "California Global Warming Solutions Act," Assembly Bill No. 32, California, United

States, 2006.

[39] T. Napp, K. S. Sum, T. Hills and P. S. Fennell, "Attitudes and Barriers to Deployment of CCS

from Industrial Sources in the UK," Grantham Institute for Climate Change. Imperial College

London, London, 2014.

[40] JRC, "2013 Technology Map of the European Strategic Energy Technology Plan," Institute for

Energy and Transport. Joint Research Centre. European Commission, Luxembourg, 2013.

[41] TERI, "India CCS Scoping Study: Final Report," The Global CCS Institute, Project Code

2011BE02, 2013.


[42] GAO, "Climate Change: Federal Actions Will Greatly Affect the Viability of Carbon Capture and

Storage As a Key Mitigation Option," U.S. Government Accountability Office, Washington, DC,

2008.

[43] L. Bird, M. Milligan and D. Lew, "Integrating Variable Renewable Energy: Challenges and

Solutions," National Renewable Energy Laboratory, United States, 2013.

[44] K. Stefferud, "California’s Solar PV Forecasting Challenges and Opportunities," EnerNex:

Electric power research, engineering and consulting, 13 September 2013.

[45] A. Ipakchi and F. Albuyeh, "Grid of the Future," IEEE power & energy magazine, no.

March/April, p. 53–62, 2009.

[46] MIT, "The Future of Natural Gas," Energy Initiative - Massachusetts Institute of Technology,

Boston, 2011a.

[47] X. Zhang, N. P. Myhrvold, Z. Hausfather and K. Caldeira, "Climate benefits of natural gas as a

bridge fuel and potential delay of near-zero energy systems," Applied Energy, p. (in press), 2015.

[48] IHS CERA, "Fueling the Future with Natural Gas: Bringing It Home," IHS, 2013.

[49] R. W. Howarth, R. Santoro and A. Ingraffea, "Methane and the greenhouse-gas footprint of

natural gas from shale formations," Climatic Change, vol. 106, p. 679–690, 2011a.

[50] S. G. Osborn, A. Vengosh, N. R. Warner and R. B. Jackson, "Methane contamination of drinking

water accompanying gas-well drilling and hydraulic fracturing," Proceedings of the National

Academy of Sciences of the United States of America , vol. 108, no. 20, p. 8172–8176, 2011.

[51] R. W. Howarth, A. Ingraffea and T. Engelder, "Natural gas: Should fracking stop?," Nature, p.

271–275, 15 September 2011b.

[52] J. Bordoff, "How Exporting U.S. Liquefied Natural Gas Will Transform the Politics of Global

Energy," The Wall Street Journal, 17 November 2015.

[53] N. Kulichenko and E. Ereira, "Carbon Capture and Storage in Developing Countries: a

Perspective on Barriers to Deployment," The World Bank, Washington, DC, 2011.

[54] Seeking Alpha, "Competitive Threats To Residential Solar Standouts Are Exaggerated," Seeking

Alpha, 5 January 2016.

[55] Seeking Alpha, "Increasing Uncertainty In Rooftop Solar's Long-Term Business Model," Seeking

Alpha, 15 November 2015.

[56] N. Litvak, "U.S. Residential Solar Financing 2015-2020," GTM Research, United States, 2015.

[57] N. Alster, "Why Solar Power Stocks Are Still Earthbound," The New York Times, 6 April 2013.


[58] A. H. Miller, "Solar stocks fluctuating on market uncertainty," Solar Energy News, 9 March

2011.

[59] J. Pyper, "Breaking: Clean Power Plan Stayed by Supreme Court, Obama Handed a Setback,"

Greentech Media, 9 February 2016.

[60] A. Liptak and C. Davenport, "Supreme Court Deals Blow to Obama’s Efforts to Regulate Coal

Emissions," The New York Times, 9 February 2016.

[61] C. Frank, "Pricing Carbon: A Carbon Tax or Cap-And-Trade?," Brookings, 12 August 2014.

[62] D. M. Uhlmann and R. S. Avi-Yonah, "Combating Global Climate Change: Why a Carbon Tax Is

a Better Response to Global Warming Than Cap and Trade," Stanford Environmental Law

Journal, vol. 28, no. 3, 2009.

[63] L. H. Goulder and A. R. Schein, "Carbon Taxes versus Cap and Trade: A Critical Review,"

Climate Change Economics, vol. 4, no. 3, p. 1350010, 2013.

[64] EC ETS, Directive 2003/87/EC of the European Parliament and of the Council of 13 October

2003 establishing a scheme for greenhouse gas emission allowance trading within the Community

and amending Council Directive 96/61/EC, European Union: European Commission, 2003.

[65] H.R. 2454, American Clean Energy and Security Act of 2009, United States: 111th Congress, 1st

Session, 2009.

[66] R. Taylor and R. Hoyle, "Australia Becomes First Developed Nation to Repeal Carbon Tax," The

Wall Street Journal, 17 July 2014.

[67] J. Hoppmann, J. Huenteler and B. Giroda, "Compulsive policy-making—The evolution of the

German feed-in tariff system for solar photovoltaic power," Research Policy, vol. 43, no. 8, p.

1422–1441, 2014.

[68] M. Yozwiak, "How extending the investment tax credit would affect US solar build," Bloomberg

New Energy Finance, United States, 2015.

[69] UCSUSA, "Production Tax Credit for Renewable Energy," [Online]. Available:

http://www.ucsusa.org/clean_energy/smart-energy-solutions/increase-renewables/production-tax-

credit-for.html. [Accessed 2016].

[70] J. Farrell, "The Future of Solar Economics and Policy," Clean Technica, 20 October 2014.

[71] J. Pyper, "Nevada Regulators Eliminate Retail Rate Net Metering for New and Existing Solar

Customers," Greentech Media, 23 December 2015.

[72] CPUC, Decision Adopting Successor to Net Energy Metering Tariff, California: California Public

Utilities Commission. R.14-07-002, 2015.


[73] R. A. Howard and A. E. Abbas, Foundations of Decision Analysis, 1st ed., United States:

Pearson, 2016.

[74] H. Courtney, J. Kirkland and P. Viguerie, "Strategy Under Uncerainty," Harvard Business

Review, 1997.

[75] E. Pate-Cornell, "Uncertainties in risk analysis: Six levels of treatment," Reliability Engineering

and System Safety, vol. 54, pp. 95-111, 1996.

[76] A. K. Dixit and R. S. Pindyck, Investment Under Uncertainty, Princeton: Princeton University

Press, 1994.

[77] T. Macalister, "Spending watchdog to examine scrapping of £1bn carbon capture plan," The

Gaurdian, 31 January 2016.

[78] A. Neslen, "Europe's carbon capture dream beset by delays, fears and doubt," The Guardian, 9

April 2015.

[79] W. Widmer, "Billions over budget. Two years after deadline. What’s gone wrong for the ‘clean

coal’ project that’s supposed to save an industry?," Politico, 26 April 2015.

[80] SolarCity, "Following Nevada PUC's Decision to Punish Rooftop Solar Customers, SolarCity

Forced to Eliminate More than 550 Jobs in Nevada," Press Releases, 6 January 2016.

[81] L. Stoker, "SunEdison to exit ‘uneconomic’ UK market," Solar Power Portal, 7 October 2015.

[82] H. Chandler, "Empowering Variable Renewables: Options for Flexible Electricity Systems,"

International Energy Agency, Paris, 2008.

[83] MIT, "The Future of the Electric Grid (Section 1.2: Challenges and Opportunities),"

Massachusetts Institute of Technology, Boston, 2011b.

[84] M. T. Ho and D. E. Wiley, "Flexible strategies to facilitate carbon capture deployment at

pulverised coal power plants," International Journal of Greenhouse Gas Control, 2016.

[85] S. M. Cohen, A techno-economic plant- and grid-level assessment of flexible CO2 capture,

Austin: University of Texas at Austin, 2012.

[86] H. Chalmers, M. Leach, M. Lucquiaud and J. Gibbins, "Valuing flexible operation of power

plants with CO2 capture," Energy Procedia, vol. 1, no. 1, p. 4289–4296, 2009.

[87] S. Ziaii, S. Cohen, G. T. Rochelle, T. F. Edgar and M. E. Webber, "Dynamic operation of amine

scrubbing in response to electricity demand and pricing," Energy Procedia, vol. 1, p. 4047–4053,

2009.

[88] C. J. Barnhart, M. Dale, A. R. Brandt and S. M. Benson, "The energetic implications of curtailing

versus storing solar- and wind-generated electricity," Energy & Environmental Science, vol. 6, p.

2804–2810, 2013.


[89] P. Denholm and M. Hand, "Grid flexibility and storage required to achieve very high penetration

of variable renewable electricity," Energy Policy, vol. 39, p. 1817–1830, 2011.

[90] P. Denholm, J. Jorgenson, M. Hummon, T. Jenkin, a. D. Palchak, B. Kirby, O. Ma and M.

O’Malley, "The Value of Energy Storage for Grid Applications," National Renewable Energy

Laboratory, United States, 2013.

[91] C. A. Kang, A. R. Brandt and L. J. Durlofsky, "Optimal operation of an integrated energy system

including fossil fuel power generation, CO2 capture and wind," Energy, vol. 36, p. 6806–6820,

2011.

[92] J. Meerman, A. Ramirez, W. Turkenburg and A. Faaij, "Performance of simulated flexible

integrated gasification polygeneration facilities, Part B: Economice valuation," Renewable and

Sustainable Energy Reviews, vol. 16, p. 6083–6102, 2012.

[93] C. Rubio-Maya, J. Uche-Marcuello, A. Martínez-Gracia and A. A. Bayod-Rújula, "Design

optimization of a polygeneration plant fuelled by natural gas and renewable energy sources,"

Applied Energy, vol. 88, p. 449–457, 2011.

[94] L. Hu, J. Hongguang, G. Lin and H. Wei, "Techno-economic evaluation of coal-based

polygeneration systems of synthetic," Energy Conversion and Management, vol. 52, p. 274–283,

2011.

[95] L. Trigeorgis, "Foreword (by Scott P. Mason)," in Real Options: Managerial Flexibility and

Strategy in Resource Allocation, Cambridge, Massachusetts and London, England, MIT Press,

1996.

[96] X. Zhang, X. Wang, J. Chen, X. Xie, K. Wang and Y. Wei, "A novel modeling based real option

approach for CCS investment evaluation under multiple uncertainties," Applied Energy, vol. 113,

p. 1059–1067, 2014.

[97] GCCSI, "Definition of CCS Ready," 3 November 2010. [Online]. Available:

http://www.globalccsinstitute.com/insights/authors/christophershort/2010/11/03/definition-ccs-

ready. [Accessed 2016].

[98] X. Liang, D. Reiner, J. Gibbins and J. Li, "Assessing the value of CO2 capture ready in new-build

pulverised coal-fired power plants in China," International Journal of Greenhouse Gas Control,

vol. 3, no. 6, p. 787–792, 2009.

[99] M. C. Bohm, H. J. Herzog, J. E. Parsons and R. C. Sekar, "Capture-ready coal plants—Options,

technologies and economics," International Journal of Greenhouse Gas Control, vol. 1, no. 1, p.

113–120, 2007.

[100] R. McIntosh and J. Mandel, "Five Reasons U.S. Solar Installers are Vertically Integrating … For

Now," Rocky Mountain Institute, 10 July 2014.


[101] J. S. John, "SunEdison Buys Solar Grid Storage for Battery-Backed PV and Wind Power,"

Greentech Media, 5 March 2015a.

[102] J. S. John, "First Solar Joins $50M Investment in Younicos, Stakes Claim in Energy Storage

Market," Greentech Media, 8 December 2015b.

[103] D. Cardwell, "SolarCity to Use Batteries From Tesla for Energy Storage," The New York Times, 4

December 2013.

[104] M. E. Porter, "The Five Competitive Forces that Shape Strategy," Harvard Business Review,

January 2008.

[105] H. A. Parker, R. T. Teubner and J. C. Sawicki, "Spill Reponse Planning in the Philippines: 3-Tier

Interaction between Government and Industry," 2009. [Online]. Available:

http://www.interspill.org/previous-events/2009/12-May/pdf/1630_parker.pdf.

[106] IPIECA, "Guide to Tiered Preparedness and Response," International Petroleum Industry

Environmental Conservation Association, London, United Kingdom, 2007.

[107] J. Magretta, Understanding Michael Porter: The Essential Guide to Competition and Strategy,

Cambridge: Harvard Business Review Press, 2012.

[108] M. E. Porter, "How Competitive Forces Shape Strategy," Harvard Business Review, March-April

1979.

[109] R. A. Howard and J. E. Matheson, "Influence Diagrams," Decision Analysis, vol. 2, no. 3, p. 127–

143, 2005.

[110] R. D. Shachter, "Evaluating Influence Diagrams," Operations Research, vol. 34, no. 6, p. 871–

882, 1986.

[111] S. Reichelstein and M. Yorston, "The prospects for cost competitive solar PV power," Energy

Policy, vol. 55, p. 117–127, 2012.

[112] T. Ramsden, D. Steward and J. Zuboy, "Analyzing the Levelized Cost of Centralized and

Distributed Hydrogen Production Using the H2A Production Model, Version 2," National

Renewable Energy Laboratory, Golden, Colorado, USA, 2009.

[113] J. Anda, A. Golub and E. Strukova, "Economics of climate change under uncertainty: Benefits of

flexibility," Energy Policy, vol. 37, p. 1345–1355, 2009.

[114] S. Hallegatte, "Strategies to adapt to an uncertain climate change," Global Environmental

Change, vol. 19, no. 2, p. 240–247, 2009.

[115] W. Blyth, R. Bradley, D. Bunn, C. Clarke, T. Wilson and M. Yang, "Investment risks under

uncertain climate change policy," Energy Policy, vol. 35, p. 5766–5773, 2007.


[116] D. M. Lander and G. E. Pinches, "Challenges to the practical implementation of modeling and

valuing real options," The Quarterly Review of Economics and Finance, vol. 38, no. 3, p. 537–

567, 1998.

[117] L. E. Brandão, J. S. Dyer and W. J. Hahn, "Using Binomial Decision Trees to Solve Real-

Option," Decision Analysis, vol. 2, no. 2, p. 69–88, 2005.

[118] D. Kroniger and R. Madlener, "Hydrogen storage for wind parks: A real options evaluation for an

optimal investment in more flexibility," Applied Energy, vol. 136, p. 931–946, 2014.

[119] T. K. Boomsma, N. Meade and S.-E. Fleten, "Renewable energy investments under different

support schemes: A real options approach," European Journal of Operational Research, vol. 220,

no. 1, p. 225–237, 2012.

[120] Q. Chen, C. Kang, Q. Xia and J. Zhong, "Real option analysis on carbon capture power plants

under flexible operation mechanism," Minneapolis, 2010.

[121] R. d. Neufville, "Real Options: Dealing With Uncertainty in Systems Planning and Design,"

Integrated Assessment, vol. 4, no. 1, p. 26–34, 2003.

[122] TNS Opinion & Social, "Public Awareness and Acceptance of CO2 capture and storage,"

EuroBarometer, European Commission, Brussels, 2011.

[123] P. Upham and T. Roberts, "Public Perceptions of CCS: the results of NearCO2 European Focus

Groups," NearCO2, 2010.

[124] IEA, "Regulatory Frameworks for CCS," 2015. [Online]. Available:

http://www.iea.org/topics/ccs/subtopics/permittingframeworksforccs/. [Accessed May 2015].

[125] K. Farhat and S. M. Benson, "Translating Risk Assessment to Contingency Planning for CO2

Geologic Storage: A Methodological Framework," International Journal of Greenhouse Gas

Control, (accepted).

[126] J. Meerman, A. Ramírez, W. Turkenburg and A. Faaij, "Performance of simulated flexible

integrated gasification polygeneration facilities. Part A: A technical-energetic assessment,"

Renewable and Sustainable Energy Reviews, vol. 15, p. 2563–2587, 2011.

[127] P. Liu, E. N. Pistikopoulos and Z. Li, "A Multi-Objective Optimization Approach to

Polygeneration Energy Systems Design," AIChE Journal: Process Systems Engineering, vol. 56,

no. 5, p. 1218–1234, 2010.

[128] HECA, "The Project," 2010a. [Online]. Available: http://hydrogenenergycalifornia.com/the-

project. [Accessed 2014].

[129] HECA, "Project Fact Sheet," 2010b. [Online]. Available:

http://hydrogenenergycalifornia.com/factsheets. [Accessed 2013].


[130] K. Farhat and S. Reichelstein, "Economic Value of Flexible Hydrogen-Based Polygeneration

Energy Systems," Applied Energy, vol. 164, p. 857–870, 2016.

[131] Accenture Academy, "Explaining Porter’s Five Forces," 2014. [Online]. Available:

https://www.accentureacademy.com/d/course/1000007629. [Accessed 2015].

[132] FME, "Porter's Five Forces - Strategy Skills," 2013. [Online]. Available: http://www.free-

management-ebooks.com/dldebk-pdf/fme-five-forces-framework.pdf. [Accessed 2015].

[133] R. Marks, "Lecture Notes - Industry Analysis," Australian Graduate School of Management,

2003.

[134] M. E. Dobbs, "Guidelines for applying Porter's five forces framework: a set of industry analysis

templates," Competitiveness Review, vol. 24, no. 1, p. 32–45, 2014.

[135] H. Lee, M.-S. Kim and Y. Park, "An analytic network process approach to operationalization of

five forces model," Applied Mathematical Modelling, p. 1783–1795, 2012.

29

Chapter 2

Translating Risk Assessment to

Contingency Planning for CO2 Geologic

Storage: A Methodological Framework

1 Introduction

Fifty five large-scale carbon capture and storage (CCS) projects exist or are planned around

the world today, the majority of which are located in North America, Australia, Europe, and

China [1]. Ensuring reliable long-term storage of carbon dioxide (CO2) in subsurface geologic

formations is important to gain public support and accelerate the deployment of CCS [2, 3, 4].

To that effect, huge research and policy efforts have been devoted to the development of

technologies and regulations that secure safe operations of CO2 storage sites, with a wide

range of requirements and guidelines on risk management [1, 5, 6].

Different regulatory and legislative bodies have adopted different requirements to permit CO2

geologic storage, specifically regarding the need for risk assessment and corrective-measure

plans to address potential CO2 leakage. In the United States, the Federal Environmental

Protection Agency requires a “corrective action plan” and an “emergency and remedial

response plan” with specific timelines, and it provides detailed guidance on how to provide the

requisite information [7, 8]. Nonetheless, the Agency does not request formal documentation

on risk assessment [9]. In Canada, the environmental protection responsibilities are shared

CHAPTER 2 — Introduction 30

between the federal and provincial governments [10]. The federal government adopted the

CSA Z741 standard, which covers “risk assessment” and “risk treatment” [11, 12]. On the

provincial level, Alberta’s legislation does not require “risk assessment” or “corrective action”

plans, but the legislation allows imposing both requirements through regulation [13, 14, 15].

In addition, while British Columbia’s currently proposed regulations mandate a “corrective

measures / contingency plan” and a “description of measures to prevent significant leakage” as

part of the application for a storage permit [16], Saskatchewan’s CO2 storage operations are

managed under existing oil and gas regulations [10]. Similar policy framework exists in

Australia. Through environmental guidelines, the Commonwealth government calls for

“continuous risk assessment as an essential element of the environmental impact assessment”,

but the legislation governing greenhouse gas storage does not require submitting plans for risk

assessment or corrective measures [17, 18]. At the state level, Victoria demands a “risk

management plan” before granting an injection and monitoring license [19] whereas

Queensland does not [20]. In the European Union, the European Commission published a

CCS Directive, which requires both a “risk assessment plan” and a “corrective measures plan”

[21]. In two of the associated four guidance documents, the Commission provides a detailed

description of the requested plans, which includes example templates, proposed areas of

investigation, as well as recommended tools and formats [22, 23]. Finally, China’s legal

framework for managing CO2 storage is still under development, with no existing rules on risk

assessment and corrective measures [24, 25].

With these different approaches to risk management, a clear link between risk assessment and

corrective action for CO2 leakage is often missing. This reality controverts the wide agreement

among industry experts and policy makers on the need to connect the various aspects of risk

management, including the need to design corrective measures based on risk analysis before

permitting operations [8, 23, 26, 27, 28, 29, 30]. The aim of this work is to propose a

methodological framework that bridges risk assessment to corrective measures through clear

and effective contingency planning. This framework achieves two tasks, which summarize the

novelty of this work. First, it expands the formulation of a risk assessment matrix (RAM) to

make it more action- and decision-oriented, which subsequently facilitates its translation to a

contingency planning matrix (CPM). Second, it explains the significance of the various

CHAPTER 2 — Risk Management: Assessment, Mitigation, and Contingency Planning 31

CPM elements, not only for mapping corrective measures to potential leakage scenarios but

also for facilitating critical coordination between the operating party and the regulatory agency

overseeing CO2 storage.

In pursuing both goals, the proposed framework utilizes the extensive body of literature on

risk assessment and corrective measures for CO2 leakage, offering a mean to bridge the

utilization of existing tools instead of proposing new ones. In addition, when demonstrating its

applicability, this framework considers scenarios of CO2 leakage that start and ends in the

subsurface and propagates through geologic pathways only. When applied in real life, and

using similar techniques to the ones presented in this paper, the framework can be expanded to

include scenarios of CO2 leaks that propagate through man-made pathways and reach the

surface.

In the subsequent sections of this paper, we first provide a brief overview of the terminology

used in the risk management of CO2 storage, highlighting some existing literature on risk

assessment methodologies and corrective measures. Next, we discuss how to update the risk

assessment matrix, introducing the concept of risk profiles of CO2 leakage. As the main focus

of this paper, a contingency planning matrix is then developed based on the updated risk

assessment matrix, and its tier structure is discussed. Lastly, we leverage the contingency

planning matrix to design a model contingency plan, covering multiple sections on preparing

for leakage risks and responding to leakage incidents. When discussing specific response

strategies within the plan, we show how different corrective measures can tackle different risk

profiles under different contingency tiers, effectively linking all elements of risk management

for CO2 leakage from geologic storage.

2 Risk Management: Assessment, Mitigation, and Contingency

Planning

Before discussing the details of the proposed framework, it is important to establish a

consistent risk-management terminology that we can refer to throughout this paper. As

depicted in Figure 2.1, managing the risk of CO2 leakage from geologic storage formations

CHAPTER 2 — Risk Management: Assessment, Mitigation, and Contingency Planning 32

includes three essential steps: assessment of risk, mitigation and avoidance of intolerable risk,

and contingency planning for tolerable risk. A robust risk assessment involves the

identification, analysis, and evaluation of potential leakage scenarios. After identifying a

comprehensive set of scenarios, each scenario is analyzed qualitatively or quantitatively to

determine its likelihood of occurrence and its impact on subsurface formations and surface

ecosystems. Leakage risks are then evaluated according to external mandates by the regulatory

agency and internal procedures by the operating party, resulting in a set of risk tolerance

levels. While risks below the minimum tolerance levels can be safely ignored, risks exceeding

the maximum tolerance levels need to be mitigated or avoided altogether through a variety of

preventative measures [22, 26, 27].

Figure 2.1: Elements of risk management for CO2 leakage from geologic reservoirs

On the other hand, leakage scenarios within the tolerance range are managed through

contingency planning, which aims to prepare for leakage risks and respond to leakage

incidents if they occur [28, 31, 32]. Learning from the oil and gas industry, we envision a tier-

Risk Management

Risk Assessment

Mitigation of Intolerable Risk

Contingency Planning for Tolerable Risk

leakage scenario identification

leakage scenario analysis(likelihood, impact)

leakage scenario evaluation(tolerable, intolerable)

Purpose• prepare for leakage risks

• respond to leakage incidents

Elements• thresholds

• response initiation through triggers

• response strategies• corrective measures:

control (stop and contain) & remediate

• human and equipment resources

• administration and coordination schemes

preventative measures

CHAPTER 2 — Updating the Risk Assessment Matrix 33

system approach to contingency planning for CO2 leakage, which integrates five essential

elements: thresholds, response initiation through triggers, response strategies that include

corrective measures, human and equipment resources, and administration and coordination

schemes [23, 33, 34, 35, 36, 37]. While thresholds refer to specific levels of leakage

likelihoods and impacts that bound risk-preparedness, triggers refer to specific irregular

measurements or observations that initiate incident-response. Additionally, corrective

measures cover subsurface and surface activities that aim to both control (stop or contain) the

leakage and remediate its impacts [32, 38]. To that end, thresholds and triggers shape when

corrective measures should be implemented while human and equipment resources and

administration and coordination schemes define what and how corrective measures should be

implemented.

The aforementioned sequential process of risk management shows that the effective design

and deployment of corrective measures for CO2 leakage necessitates a robust contingency

plan, which in turn should be based on the findings of a comprehensive risk assessment. In our

attempt to present a methodological framework that integrates all three elements of risk

management, we focus primarily on contingency planning, which has received comparatively

little attention in literature. Nonetheless, contingency planning is linked to risk assessment and

corrective measures by utilizing the large body of existing literature on both topics [39, 40,

38]. Specifically, we use the features, events, and processes (FEP) methodology for risk

identification [41] and Bayesian event trees (BET) for risk analysis [40, 42, 43]. The

RISQUE method is another valuable resource to assess impacts and elicit informed

probabilities from experts [44, 45]. Subsequently, for corrective measures, we use

representative examples of containment and remediation activities to combat leakage events in

the subsurface [46, 47, 48].

3 Updating the Risk Assessment Matrix

A risk assessment matrix (RAM) is usually represented as a two-dimensional plot of leakage

impact versus leakage likelihood. Our focus on the RAM is motivated by its application in

some existing CCS policies [22, 49] and projects [29, 50] and by its ability to visualize all


three steps of risk assessment. The leakage scenarios depicted in a RAM result from risk

identification; the likelihood and impact of each scenario are the outcome of risk analysis; and

the determination of insignificant, tolerable, and intolerable scenarios emerge from risk

evaluation. When quantified, a RAM represents a leakage scenario as a single risk point, with

the likelihood and the impact calculated based on expected-average or worst-case estimates

[45, 51]. Other RAMs are qualitative, so a leakage scenario is allocated into a single high,

medium, or low risk zone [29, 52].

While helpful for visualizing and comparing risks, current applications of RAM can still be

improved. For example, a realistic leakage scenario may span more than one risk level

depending on several factors, some of which are uncertain. Such factors include the rate of

leakage, the features of the storage reservoir and surrounding geologic formations, and the

vulnerability of surface ecosystems. In addition, it is hard to distinguish the relative

significance of the various risk drivers in current RAM depictions of leakage. For instance, it

cannot be inferred whether a high likelihood of leakage through a fault into a freshwater

aquifer is due to the high probability that a fault exists or due to the high probability that an

aquifer is nearby the fault given that the latter exists. Adopting a quantifiable probabilistic risk

assessment (PRA) approach that combines FEP and BET offers one way to address those

issues, resulting in a more inclusive RAM and therefore facilitating the transition to a CPM.

PRA relies on systems analysis, decision analysis, and Bayesian reasoning to assess a set of

mutually exclusive and collectively exhaustive scenarios of CO2 leakage in the subsurface

[53]. This approach includes four steps. First, for risk identification, the overall subsurface

system is divided into a series of independent functional subsystems. Potential leakage

scenarios are defined as trajectories that combine multiple subsystems, and FEP guides

specifically the categorization of subsystems where a CO2 leakage may start. Second, the

likelihood of each leakage scenario is assessed as a series of conditional probabilities that

change as a function of a measurable criterion, which is typically a relevant geological feature.

Third, multiple value models are developed to help quantify the impact of leakage on the

subsurface and surface ecosystems. The second and third steps cover risk analysis, and

combining the first three steps results in a BET that systematically describes all foreseeable

conjunctions of leakage events. In the fourth step, specific tolerance levels are determined for


both the likelihood and the impact of leakage in order to evaluate what risks should be

mitigated and what risks can be safely ignored. Eventually, the overall outcome of this PRA is

a RAM that depicts a comprehensive set of risk profiles of CO2 leakage with quantified

likelihood, impact, and tolerance levels.

3.1 Functional Subsystems for Risk Identification

For CO2 leakage through geologic pathways, the system is defined as the set of geologic

formations in the subsurface, above and including the storage reservoir. As depicted in Figure

2.2, this system can be divided into three functional subsystems: Origin, Endpoint, and

Pathway.

Figure 2.2: Functional subsystems for risk identification

Origin: refers to the functional subsystem which is designated to contain the CO2 under

normal conditions. Physically, it encompasses the storage reservoir and any selected

containment zones. Origin can be further decomposed into a list of subsystems, namely, a

comprehensive count of the geologic irregularities – features, events, or processes (FEP) –

through which CO2 may leak and the rest of the formation where CO2 remains safely stored;

we categorize the former subsystems as FEP and refer to the latter subsystem as Safe.

Although FEPs are site-specific, an example list is presented in Table 2.1.

Endpoint: refers to the functional subsystem where the leaked CO2 finally reaches. Endpoint

can be further decomposed into a list of subsystems, which we limit to three: FrW, O&G, and

Other. FrW refers to all subsurface aquifers containing freshwater. O&G refers to all geologic

formations containing oil or gas resources. For simplicity, freshwater, oil, and gas are assumed

to be the only human-valuable subsurface assets. Accordingly, Other includes all geologic

FEP 1 FEP 2 FEP 3 … FEP 8

FrW

Safe

O&G Other

Direction

of CO2

Leakage

Endpoint

Pathway

Origin


formations that trap the leaked CO2 yet are not classified as freshwater aquifers or oil and gas

reservoirs.

Table 2.1: Examples of Origin FEPs and their corresponding Indicators

Origin FEP Symbol

{𝑂} Indicator

Symbol

(𝑖)

Caprock high-permeability zone 𝑂1 permeability α

Caprock-absent zone 𝑂2 size of the opening λ

Caprock fracture due to over-pressurization 𝑂3 size of fracture β

Exceeding capillary pressure due to

over-pressurization 𝑂4 capillary pressure δ

Natural fault or fracture 𝑂5 size of fracture β

Induced fault or fracture due to

over-pressurization 𝑂6 size of fracture β

Induced fault or fracture due to CO2

geochemical reactions 𝑂7 size of fracture β

Induced fault or fracture due to seismic activity 𝑂8 Size of fracture β

Pathway: refers to the functional subsystem between the Origin and the Endpoint. Physically,

this area encompasses all subsurface formations through which CO2 migrates after leaving an

Origin formation until reaching an Endpoint formation.

3.2 Bayesian Event Tree for Risk Analysis

The Bayesian Event Tree (BET) is an effective way to track the likelihood and consequences

of the various scenarios of CO2 leakage. A BET models the likelihood of CO2 leakage through

three sequential and uncertain events: Origination, Propagation, and Destination. As we

explain shortly, these events govern the progression of CO2 leakage through the

aforementioned Origin, Pathway, and Endpoint subsystems. Subsequently, for every leakage

prospect, the BET allows modeling the leakage consequences using various value models; we

introduce two: Value of Flow (VF) and Value of Impact (VI). An example BET is shown in

Figure 2.3, followed by a detailed description of its various components and the procedure to

construct it. Important to note, the depicted BET and all related figures are purely

hypothetical; they aim to demonstrate the concepts outlined here and provide a roadmap for

implementing them.


3.2.1. Probability of CO2 Leakage

For a leakage to occur, three uncertain events must take place sequentially. As explained

below, each event is assigned a probability, and the overall likelihood of a leakage scenario is

the product of those probabilities.

Origination {𝑶}(𝒊): refers to the probability of the existence of a specific Origin, which could

be an FEP or Safe. For an FEP, this likelihood may vary with the subsystem’s exact

characteristics. Therefore, {𝑂} for an FEP is expressed a function of an Indicator 𝑖, which is a

selected attribute of the analyzed FEP. Indicator attributes should be chosen to best-match

their FEP subsystems. Table 2.1 presents a suggested attribute for each of the listed FEP. For

instance, {𝑂1}(𝛼) is the probability of existence of a caprock high-permeability zone 𝑂1 and

is a function of permeability 𝛼. In mathematical terms, Origination {𝑂} can be represented as

a step-wise function of Indicator 𝑖, which is discretized over its total feasible range.

Propagation {𝐏|𝑶}(𝒊): refers to the probability of CO2 entering the Pathway from a specific

Origin, given that this Origin actually exists. Propagation is conditioned on Origination, so

{P|𝑂} can be also represented as a step-wise function over the full discretized range of

Indicator 𝑖. By definition, {P|𝑆𝑎𝑓𝑒} must be always zero. Propagation may be considered a

pinch-point, the point in the analysis at which it doesn’t matter – for subsequent analysis –

how the system reached its current state but how it proceeds from that state [54, 55]. In this

case, if Propagation is positive, CO2 may escape from the Origin to the Pathway. Subsequent

analysis focuses on investigating where the CO2 migrates from the Pathway regardless of how

it reached the Pathway.

Destination {𝑫|𝑶, 𝑷}(𝒊): refers to the probability of CO2 entering a specific Endpoint from

the Pathway, given that it already reached the Pathway through an existing Origin. Here

again, because Destination is conditioned on Origination and Propagation, {𝐷|𝑂, 𝑃} can be

also represented as a step-wise function over the full discretized range of Indicator 𝑖. Also, by

definition, {𝐷|𝑠𝑎𝑓𝑒} is zero because CO2 cannot reach an Endpoint unless it escapes through

an FEP first. Destination is primarily dependent on hydrogeological factors that govern the

transport of CO2 in the Pathway, including the injection locations and rates, local hydrology,

38

Figure 2.3: Example Bayesian event tree (BET) for risk analysis of CO2 leakage

CHPATER 2 — Updating the Risk Assessment Matrix 39

and geologic configuration. However, consistent with the pinch-point definition of

Propagation, CO2 transport in the Pathway is unlikely to be dependent on the specific type of

FEPs in the Origin. Therefore, one simplifying assumption is to treat Destination as

independent of the Propagation {𝑃|𝑂} and Origination {𝑂}(𝑖). In this case, {𝐷|𝑂, 𝑃}(𝑖) =

{𝐷}, which means that Destination is a constant function of Indicator 𝑖.

Leakage Likelihood {𝑳}(𝒊): for CO2 leakage to occur, the CO2 must find an Origin FEP,

move through the Origin FEP into the Pathway, and then enter into an Endpoint. Therefore,

intuitively, the Leakage Likelihood is the product of Origination, Propagation, and

Destination, as illustrated in (1).

{𝐿}(𝑖) = {𝑃}(𝑖) ∙ {P|𝑂}(𝑖) ∙ {𝐷|𝑂, 𝑃}(𝑖) = {𝑃, 𝑂, 𝐷}(𝑖) (1)

The formulation in (1) provides three important insights. First, if Origination, Propagation, or

Destination is zero, the Leakage Likelihood is also zero, and assessing the uncertainty of

subsequent event(s) becomes unnecessary. Second, because {𝐿} is a function of 𝑖, the

likelihood of a specific leakage scenario from a particular FEP is a function of the

characteristics of that FEP. Finally, the analyzed leakage scenarios must be mutually

exclusive and collectively exhaustive. This means that exactly one of the BET scenarios must

occur, and the probability of all scenarios must add up to one. In fact, due to conditional

probability assessment, the branch probabilities at each node of the tree should also sum up to

one. To that end, we note that a leakage scenario need not involve a single Origin and a single

Endpoint. Site-specific data may suggest a leakage incident that occurs through multiple FEPs

and reaches multiple Endpoints, in which case this leakage incident should be presented as a

separate scenario. Such approach ensures that all foreseeable leakage scenarios are accounted

for; specifically, it guards against “perfect storms”, where multiple leakage events of very low

likelihood occur all at once and cause a collective impact greater than the sum of their

individual impacts.

To demonstrate the PRA methodology above, we analyze a subset of three CO2 leakage

scenarios in Figure 2.3, corresponding to three distinct Origin-Pathway-Endpoint trajectories.

Specifically, we assume CO2 storage in a deep saline aquifer (Safe), and we examine the risk


of leakage from a high-permeability zone (𝑂1) in the reservoir’s shale caprock to a freshwater

aquifer (FrW), an oil reservoir (O&G), and another sealed geologic formation (Other). Figure

2.4 depicts the relevant functional subsystems: Origins (Safe, 𝑂1), Pathway, and Endpoints

(FrW, O&G, Other). Subsequently, Figures 2.5a–d show hypothetical probability distributions

of the sequential leakage events: Origination and Propagation through the FEP, Destination

into FrW, O&G, and Other, as well as the overall Leakage Likelihood. The numerical data is

hypothetical and is provided for illustrative purposes only. In practice, the probability inputs

would be obtained based on expert opinions and/or statistical information generated from site

characterization and reservoir modeling; examples of such probabilistic data for the

representation of CO2 leakage risk already exists in literature [45, 56, 57, 58, 59, 60].

Figure 2.4: Sketch of CO2 leakage through caprock high-permeability zone

Because 𝑂1 and Safe are the only two possible Origins, the probability distribution in

Origination is split between {𝑂1} and {𝑆𝑎𝑓𝑒}. Illustrating Origination as function of

Indicator, Figure 2.5a is a log-log plot that shows the probability of existence of a high-

permeability zone in the caprock {𝑂1} with a permeability of 𝛼. The modelled range of 𝛼 is

comparable to that reported by Wang and Small [57], and Griffith [58], for caprock high-

permeability zones and fractures. As shown in Figure 2.5a, {𝑂1} is discretized over seven

intervals of 𝛼. Assuming a typical shale-seal permeability of 10-4

millidarcy (mD) [58, 61], the

CO2FEP = O1

Pathway

FrW

O&G

Other

Origination

Propagation

Destination

CO2

injection well water well oil well

Safe


permeability of a leakage-inducing FEP can only exceed this value, so {𝑂1}(𝛼 < 10−4) is set

at zero. On the other extreme, we assume that it is highly unlikely to find a caprock zone with

permeability greater than 10 mD, so {𝑂1}(𝛼 > 10) is set at 10-5

. For 𝛼 between 10-4

and 10

mD, {𝑂1} decreases almost exponentially from 0.1 to 10-4

. Subsequently, in this specific

example, the probability of having no high-permeability zone in the caprock becomes

{𝑆𝑎𝑓𝑒} = 1 − [∑ {𝑂1}(𝛼)𝛼 ] ≈ 0.839. Though hypothetical, this assumed probability

distribution for Origination is informed by, and is therefore consistent with, the findings of

select literature that addresses uncertainty in reservoir characterization and modeling; for

instance, the results by Wang et al. allow inferring a similar probability distribution from

measurements of moderate pressure buildup in the storage zone [57].

Figure 2.5: Example probability distributions of CO2 leakage scenarios

Because 𝑂1 is the only possible leakage FEP, CO2 will either leak through it or remain in the

designated storage aquifer Safe. As explained already, by definition, the CO2 cannot propagate

1.E-8

1.E-7

1.E-6

1.E-5

1.E-4

1.E-3

1.E-2

1.E-1

1.E+0

1.E-5 1.E-4 1.E-3 1.E-2 1.E-1 1.E+0 1.E+1 1.E+2

{D|O

1,P

} =

{D}

α (mD)

FrW

O&G

Other

1.E-8

1.E-7

1.E-6

1.E-5

1.E-4

1.E-3

1.E-2

1.E-1

1.E+0

1.E-5 1.E-4 1.E-3 1.E-2 1.E-1 1.E+0 1.E+1 1.E+2

{L}

= {O

1,P

,D}

α (mD)

O1, FrW

O1, O&G

O1, Other

1.E-8

1.E-7

1.E-6

1.E-5

1.E-4

1.E-3

1.E-2

1.E-1

1.E+0

1.E-5 1.E-4 1.E-3 1.E-2 1.E-1 1.E+0 1.E+1 1.E+2

{O1

}

α (mD)

1.E-8

1.E-7

1.E-6

1.E-5

1.E-4

1.E-3

1.E-2

1.E-1

1.E+0

1.E-5 1.E-4 1.E-3 1.E-2 1.E-1 1.E+0 1.E+1 1.E+2

{P|O

1}

α (mD)(a) (b)

(c) (d)


from Safe into Pathway, so {P|𝑆𝑎𝑓𝑒} is zero. Figure 2.5b is a log-log plot of Propagation

{𝑃|𝑂1} as a function of Indicator 𝛼, given that the high-permeability zone 𝑂1 actually exists.

For CO2 to leak through 𝑂1, the CO2 plume should first reach 𝑂1 then move through 𝑂1 into

Pathway. In this case, we assume that the likelihood of CO2 transport through the FEP is very

small below a specific permeability threshold of 𝛼 = 0.1 mD. In reality, this threshold could

correspond to capillary entry pressure; as 𝛼 increases, capillary entry pressure decreases until

it finally drops below the capillary pressure at the base of the caprock, at which point CO2 can

escape through the high-permeability zone in the caprock. Therefore, once 𝛼 exceeds 0.1 mD,

Propagation increases rapidly. Still, at the highest permeability range of 𝛼 > 10 mD,

{𝑃|𝑂1} is set at 0.7. Here, we make a realistic assumption that even though CO2 can escape

through 𝑂1, there is still a probability of 1 − {𝑃|𝑂1} = 0.3 that the CO2 plume may not reach

𝑂1 in the first place.

Figure 2.5c plots Destination as function of Indicator. Consistent with the definition of

Propagation as a pinch-point, Destination is assumed to be independent of Origination {𝑂1}

and Propagation {𝑃|𝑂1}. Therefore, {𝐷|𝑂1, 𝑃} is equal to {𝐷}, which is constant across the

whole feasible range of 𝛼. Informed by related findings in existing literature [56, 62], this

example assumes that the leaked CO2 from the saline aquifer is least likely to travel all the

way up to the shallow freshwater aquifer, so {𝐷 = 𝐹𝑟𝑊} is set at 0.05. It is much more likely

that the CO2 gets trapped at a deeper geologic formation along the way, probably in a

subsequent sealed formation (Other) or perhaps in a nearby oil reservoir. Accordingly,

{𝐷 = 𝑂𝑡ℎ𝑒𝑟} and {𝐷 = 𝑂&𝐺} are set at 0.7 and 0.25, respectively. Here, we assume that the

leaking CO2 may reach exactly one of the three aforementioned Endpoints, so the values of

{𝐷} for FrW, O&G, and Other must add up to 1.

Multiplying each probability distribution function in Figure 2.5c by those in Figures 2.5a and

2.5b results in three Leakage Likelihood profiles as a function of 𝛼, corresponding to the three

distinct leakage trajectories. Figure 2.5d shows a log-log plot of the three {𝐿}(𝛼) profiles. As

can be noticed, leakage seems to be less likely at both very high and very low permeability.

This observation can be attributed to two conflicting factors: low Propagation but high

Origination at low permeability and low Origination but high Propagation at high


permeability. In fact, the leakage probabilities at high permeability {𝐿}(𝛼 > 10) are consistent

with some literature findings for similar leakage events through high-permeability zones and

fractures in the caprock [45, 56].

3.2.2. Value of CO2 Leakage

Value models are functions that aim to quantify the consequences of potential CO2 leakage

incidents. As explained in existing literature, these consequences may be characterized in

different metrics and might span a wide spectrum of social, environmental, economic, and

public-safety issues [62, 56]. In this study, we suggest three value models to complement the

aforementioned probabilistic assessment.

Value of Flow – 𝑽𝑭(𝒊): quantifies the amount of the leaked CO2 into each Endpoint

subsystem, which can be expressed as a mass flux, mass flowrate, or total mass during a

specific period of time. While site-specific, VF is usually correlated to Indicator. This

correlation can be derived from characterizing or simulating fluid-flow in the analyzed

subsurface, regardless of the leakage likelihoods. Because the flow of leaked CO2 is

measurable, VF offers a direct way to quantify the consequences of leakage.

At a particular Endpoint, higher VF signifies more severe consequences of leakage. However,

a leakage of a specific VF may lead to different consequences in different Endpoints.

Therefore, while useful for characterizing the consequences of CO2 leakage at individual

Endpoints, VF cannot be used for consistent comparison of the consequences of leakage

across multiple Endpoints. Important for risk assessment and contingency planning,

performing such comparison requires translating VF into monetary terms, whose significance

is the same across all leakage scenarios; proper valuation renders a U.S. dollar spent on

controlling leakage into FrW equivalent to a U.S. dollar spent on controlling leakage into

O&G. Relying on concepts and tools in decision analysis and natural resources economics, it

is possible to express the various social, environmental, economic, and safety consequences of

CO2 leakage in one monetary metric [63, 64, 65, 66]. To that end, we propose two monetary

value models that quantify the consequences of CO2 leakage in the subsurface and on the

surface.


Value of Damage in the Subsurface – 𝑽𝑫𝒔𝒖𝒃(𝒊): corresponds to the cost of any leakage-

induced damages to the subsurface resources. While directly dependent on VF and therefore

on Indicator, separate 𝑉𝐷𝑠𝑢𝑏 models can be designed for FrW, O&G, and Other, influenced

by regulatory requirements. For FrW, 𝑉𝐷𝑠𝑢𝑏 might be a function of several parameters,

including water pH, hardness, and salination, as well as the concentration of any trace metals

or oil and gas contaminants carried by the leaked CO2 stream [38, 56, 67]. For O&G, 𝑉𝐷𝑠𝑢𝑏

may be a function of the quantity and quality of recovery from producing and future

reservoirs, both of which may deteriorate with CO2 leakage. Finally, since Other subsystems

are assumed to have no valuable assets, their corresponding 𝑉𝐷𝑠𝑢𝑏 may be limited to a non-

compliance penalty imposed by the regulatory agency.

Value of Damage on the Surface – 𝑽𝑫𝒔𝒖𝒓(𝒊): corresponds to the cost of any leakage-induced

damages to the surface resources, including environmental and ecological systems and human

structures, activities, and health [62]. In other words, this model accounts for the costs of any

surface damages or harms caused by diminishing the utility of the subsurface resources.

Intuitively, the higher the dependence of ecological and human systems on underground

natural resources, the higher their vulnerability to CO2 leakage, and thus the higher the 𝑉𝐷𝑠𝑢𝑟.

Similar to 𝑉𝐷𝑠𝑢𝑏, distinct 𝑉𝐷𝑠𝑢𝑟 models can be designed for FrW, O&G, and Other, and all

𝑉𝐷𝑠𝑢𝑟 models remain dependent on VF and thus on Indicator. Examples of factors that can be

accounted for in designing 𝑉𝐷𝑠𝑢𝑟 models include the size of human population, per-capita

annual income, size of agricultural activities, and number of natural habitats and ecological

species that depend on the freshwater aquifers where CO2 might leak [62, 56].

Leakage Value of Impact – 𝑽𝑰(𝒊): corresponds to the sum of 𝑉𝐷𝑠𝑢𝑏 and 𝑉𝐷𝑠𝑢𝑟, which allows

expressing all consequences of CO2 leakage in one monetary metric. Consistent with the

formulation of both damage values, VI is a function of VF and thus Indicator 𝑖, as illustrated

in (2). Intuitively, a higher leakage rate (VF) leads to higher contamination of subsurface

resources (𝑉𝐷𝑠𝑢𝑏) and therefore higher disutility of these resources on the surface (𝑉𝐷𝑠𝑢𝑟). In

addition, to facilitate their representation in a BET (Figure 2.3), all value models are

discretized over the same ranges of Indicator 𝑖 used to discretize the conditional probabilities

of leakage.


𝑉𝐼(𝑖) = 𝑉𝐷𝑠𝑢𝑏(𝑖) + 𝑉𝐷𝑠𝑢𝑟(𝑖) = 𝑉𝐷𝑠𝑢𝑏(𝑉𝐹) + 𝑉𝐷𝑠𝑢𝑟(𝑉𝐹) (2)

To complete the risk analysis for the earlier example of CO2 leakage from a high-permeability

caprock zone 𝑂1, we design hypothetical value models for leakage into each of the three

Destination subsystems: FrW, O&G, and Other. In reality, leakage rates are very project-

specific. In this example, we first assume that VF is best represented as a mass flux, and its

values for FrW, O&G, and Other are very similar, as illustrated in Figure 2.6a. In addition, we

assume that the leaking CO2 stream remains relatively concentrated around the high-

permeability FEP. In accordance with relevant literature findings, VF equals about 10-3

kg/m2.s when the permeability of 𝑂1 is high (𝛼 > 10). To put this number in context, Benson

and Hepple report a comparable estimated flux over a 1000 m2 surface area from a storage

site, which leaks 0.1% of its stored CO2 per year after receiving 1 million tonnes of CO2 per

year over a period of 50 years [46]. Furthermore, assuming single-phase flow through the high

permeability zone, VF is set to be proportional to Indicator 𝛼. For very low values of 𝛼, VF is

in the order of 10-7

kg/m2.s, which falls within the lower range of CO2 leakage fluxes reported

from natural analogues [46, 68].

Figure 2.6: Example value models of CO2 leakage scenarios

Subsequently, the VF models are translated to VI models, which are similarly very project-

specific and therefore difficult to generalize. In our example, VI is expressed in (arbitrary)

monetary units and is dependent on Indicator 𝛼 (and therefore on VF), as shown in Figure

2.6b. We assume a high VI for FrW; the leaked CO2 alters the aquifer’s pH and contaminates

it with trace metals (high 𝑉𝐷𝑠𝑢𝑏), and the aquifer is the primary source of freshwater for a

1.E-3

1.E-2

1.E-1

1.E+0

1.E+1

1.E+2

1.E+3

1.E+4

1.E-5 1.E-4 1.E-3 1.E-2 1.E-1 1.E+0 1.E+1 1.E+2

VI (

mo

net

ary

valu

e)

α (mD)

O1, FrW

O1, O&G

O1, Other

1.E-10

1.E-9

1.E-8

1.E-7

1.E-6

1.E-5

1.E-4

1.E-3

1.E-2

1.E-5 1.E-4 1.E-3 1.E-2 1.E-1 1.E+0 1.E+1 1.E+2

VF

(kg/

m2.s

)

α (mD)

O1, FrW

O1, O&G

O1, Other

(a) (b)


large county with a predominantly agricultural economy (high 𝑉𝐷𝑠𝑢𝑏). Figure 2.6b shows four

levels of VI for FrW, simulating the costs of four deterioration levels in water quality –

defined in terms of acidity and trace-metal concentration. One real-life (albeit simplified and

perhaps extreme) interpretation of this trend would be as follows: for 10−4 ≤ 𝛼 ≤ 10−3 mD,

water quality worsens but remains suitable for human and natural use; for 10−3 ≤ 𝛼 ≤ 10−2

mD, water becomes unsafe for drinking; for 10−2 ≤ 𝛼 ≤ 1 mD, water becomes unsafe for all

human use; and for 𝛼 ≥ 1 mD, water becomes unsafe for human, cattle, poultry, agriculture,

and wildlife use. On the other hand, we assume a relatively low VI for O&G; the oil reservoir

is depleting, so the leaked CO2 mildly deteriorates the quality and/or quantity of oil recovery

(low 𝑉𝐷𝑠𝑢𝑏), and oil revenues form a small part of the county’s income (low 𝑉𝐷𝑠𝑢𝑟). Figure

2.6b simulates one example trend for the damage costs associated with CO2 leakage into

O&G: for 10−3 ≤ 𝛼 ≤ 0.1 mD, the oil producer handles the increased flux of leaked CO2 by

progressively adjusting its existing oil extraction techniques and schedule; however, for

𝛼 ≥ 0.1 mD, the large flux of leaked CO2 requires a whole new extraction method, resulting

in a significant increase in operating costs. Furthermore, we assume that the regulatory agency

penalizes the operating party for CO2 leakage into the Other zone. The penalty is fixed, so VI

is independent of VF and Indicator 𝛼. Finally, we note that both VF and VI are discretized

over the same seven intervals of 𝛼 used to discretize leakage probabilities.

3.3 Tolerance Levels for Risk Evaluation

After analyzing the likelihood and impact of all foreseeable leakage scenarios, it is important

to evaluate which of the resulting risk scenarios are insignificant, tolerable, or intolerable.

To that end, maximum and minimum tolerance levels can be identified for the Leakage

Likelihood {𝐿} and the Leakage Value of Impact VI, consistent with existing literature on the

ALARP principle [26]. When {𝐿} is lower than its minimum tolerance level, it is evaluated as

insignificant, and the corresponding leakage risks can be safely ignored. However, when {𝐿} is

higher than its maximum tolerance level, it is evaluated as intolerable, and the corresponding

leakage risks must be mitigated. Similar minimum and maximum tolerance levels can be

determined for VI. Because the monetary metric of VI can consistently characterize all

consequences of leakage, the tolerance levels for VI, like those for {𝐿}, are applicable across


all leakage scenarios. When the likelihood and impact of a leakage scenario are between their

maximum and minimum tolerance levels, the leakage risk is deemed tolerable and is managed

through contingency planning.

Defining effective tolerance levels may prove to be challenging, given the relatively limited

experience in operating large-scale CO2 storage projects for long periods of time. Nonetheless,

such boundaries can still be set by relying on existing experience in similar industries,

primarily oil and gas [69]. Following up on the example of CO2 leakage through a caprock

high-permeability zone, we assume that minimum and maximum tolerance levels for {𝐿}

should be set at 10-7

and 0.1, respectively. Similarly, the minimum and maximum tolerance

levels for VI are set at 10-2

and 10+3

, respectively. While re-emphasizing that all numerical

values are purely hypothetical, one example to rationalize these threshold values would be to

set the monetary unit in million U.S. dollars and to assume an international energy firm

managing the CO2 storage project. In such a world, a VI below $0.01 million may be easily

accommodated within the project’s budget. However, learning from past spill incidents in the

oil and gas industry [70, 71, 72], a VI above $1,000 million might compromise not only the

economic feasibility of the project but also the financial stability of the whole firm.

3.4 Combined Representation of Risk Assessment Elements

The three discussed elements of risk assessment can now be jointly represented in a RAM.

Because the likelihood and impact of leakage are discretized over the same ranges of

Indicator, it is possible to plot all leakage scenarios on a two-dimensional RAM, with VI on

one axis and {𝐿} on the other. The result is a set of risk profiles exhibiting two key

characteristics. First, each risk profile identifies a potential leakage trajectory from a specific

Origin, through the Pathway, and into a specific Endpoint. Second, each data point in the risk

profiles corresponds to a leakage scenario in the BET, with a quantifiable Indication 𝑖,

Leakage Likelihood {𝐿} and Leakage Value of Impact VI. Accordingly, the collective risk

profiles summarize the findings of risk identification and risk analysis. Subsequently, for risk

evaluation, the minimum and maximum tolerance levels for VI and {𝐿} can be plotted on the

RAM, marking clear boundaries for insignificant, tolerable, and intolerable risks. All BET


scenarios can and should be plotted on the RAM except those with a zero probability or

impact of leakage. The risks associated with these scenarios are insignificant by design, so

they can be safely excluded.

Carrying on with our hypothetical example of leakage through a caprock high-permeability

zone, the RAM in Figure 2.7 is a log-log plot of VI versus {𝐿}. The three risk profiles,

corresponding to leakage from the high-permeability zone O1 into FrW, O&G, and Other,

span a range of likelihoods and impacts. The exact {𝐿} and 𝑉𝐼 of a leakage trajectory is

dependent on the permeability 𝛼 of O1. In that regard, the plotted risk profiles exclude the

leakage scenarios corresponding to 𝛼 < 10−4 mD, whose probability is assumed zero. In

addition, applying the tolerance levels for {𝐿} and VI shows that the risks associated with CO2

leakage into O&G and Other are tolerable. However, the impact of CO2 leakage into FrW is

intolerable if the permeability of O1 is 𝛼 > 1 mD, so the associated risks must be mitigated

before proceeding with the project. Equivalently, the likelihood of CO2 leakage into FrW is

insignificant if 𝛼 < 0.1 mD, so the associated risks can be safely ignored.

Figure 2.7: Example risk assessment matrix (RAM) for CO2 leakage.

1.E-3

1.E-2

1.E-1

1.E+0

1.E+1

1.E+2

1.E+3

1.E+4

1.E-8 1.E-7 1.E-6 1.E-5 1.E-4 1.E-3 1.E-2 1.E-1 1.E+0

VI (

mo

ne

tary

un

its)

{L}

O1, FrW

O1, O&G

O1, Other

0.1 - 1 mD

0.01 - 0.1 mD

0.1 - 1 mD

0.01 - 0.1 mD

Insi

gnif

ican

t

Insignificant

Intolerable

Into

lerable

ImpactMaximum

Tolerance Level

Impact Minimum

Tolerance Level

Likelihood Minimum Tolerance Level

Likelihood MaximumTolerance Level


This PRA approach to RAM offers multiple advantages. Broadly, the techniques used to

identify and analyze leakage risks are flexible and generalizable, so they can be expanded and

customized. For example, if multiple freshwater aquifers or oil reservoirs are observed in the

vicinity of the storage zone, each can be assessed as a separate Endpoint, resulting in multiple

FrW and O&G subsystems. Equivalently, the conditional probability analysis in the BET can

be adjusted to achieve clarity [64]; we briefly discuss four examples of such potential

adjustments.

First, depending on available data, the operator may find it clearer to further decompose

Propagation {𝑃|𝑂} into two conditional probabilities: the probability of the CO2 plume

encountering an existing FEP {𝑃𝑒|𝑂}, and the probability of the CO2 plume flowing along the

FEP after encountering it {𝑃𝑓|𝑂, 𝑃𝑒} [60]. In this case, {𝑂, 𝑃𝑒 , 𝑃𝑓} replaces {𝑂, 𝑃} in the BET

to covey the same information: the likelihood that an FEP Origin exists and that the CO2

plume reaches it then escapes through it. Conversely, in the absence of sufficient data, the

operator might find it difficult to assign distinct probabilities to {𝑂} and {𝑂|𝑃}. In this case,

the operator may directly evaluate the joint probability distribution {𝑂, 𝑃}, instead.

Second, because it may be hard to definitively know the long-term-future impacts of leakage,

the VF for each Endpoint can be translated to three mutually exclusive and collectively

exhaustive VI models: high, medium or low. Assuming Destination is a pinch-point, each VI

model occurs with probability {𝐼} referred to as Implication, so the overall Leakage Likelihood

would be updated to {𝐿} = {𝑂, 𝑃, 𝐷, 𝐼}. In this case, RAM would depict each Origin-Pathway-

Endpoint leakage trajectory as a group of three risk profiles, corresponding to the three

possible (high, medium, and low) VI models.

A third BET expansion may account for external events, features, and processes (EFEP),

which occur outside the boundaries of the defined system yet may influence the prospects of

CO2 leakage within the system [73]. In this case, the BET probabilities can be conditioned on

the occurrence of the EFEP, as illustrated in (3–6). Finally, the operator may also choose to

refine the ranges of Indicator 𝑖 over which the probability and impact values are discretized;

eventually, such refinement yields a more detailed representation of risk profiles in RAM.

CHPATER 2 — Translating the Risk Assessment Matrix to a Contingency Planning Matrix 50

{𝑂}(𝑖) = {𝑂|𝐸𝐹𝐸𝑃}(𝑖) ∙ {𝐸𝐹𝐸𝑃} + {𝑂|𝑁𝑜 𝐸𝐹𝐸𝑃}(𝑖) ∙ (1 − {𝐸𝐹𝐸𝑃}) (3)

{𝑃|𝑂}(𝑖) = {𝑃|𝑂, 𝐸𝐹𝐸𝑃}(𝑖) ∙ {𝐸𝐹𝐸𝑃} + {𝑃|𝑂, 𝑁𝑜 𝐸𝐹𝐸𝑃}(𝑖) ∙ (1 − {𝐸𝐹𝐸𝑃}) (4)

{𝐷} = {𝐷|𝐸𝐹𝐸𝑃} ∙ {𝐸𝐹𝐸𝑃} + {𝐷|𝑁𝑜 𝐸𝐹𝐸𝑃} ∙ (1 − {𝐸𝐹𝐸𝑃}) (5)

{𝐼} = {𝐼|𝐸𝐹𝐸𝑃} ∙ {𝐸𝐹𝐸𝑃} + {𝐼|𝑁𝑜 𝐸𝐹𝐸𝑃} ∙ (1 − {𝐸𝐹𝐸𝑃}) (6)

Beyond flexibility and customization, the Bayesian nature of the event tree helps elicit

probabilities from experts and keep the RAM up-to-date; as new information becomes

available, relevant conditional probabilities can be adjusted. Also, the representation of risk in

the form of profiles instead of points allows a leakage trajectory to span multiple risk levels.

As we explain next, all these advantages facilitate translating a RAM to a CPM and thus

designing an effective contingency plan.

4 Translating the Risk Assessment Matrix to a Contingency

Planning Matrix

Figure 2.8: Translating the risk assessment matrix to a contingency planning matrix

The updated risk assessment matrix (RAM) can now be translated into a contingency planning

matrix (CPM), which allows preparing for and responding to tolerable leakage risks. This

translative procedure, shown in Figure 2.8, renders the proposed CPM an effective tool to

design and demonstrate four essential elements of contingency planning: required resources,

VI

{L}

VI

{L}

Likelihood Minimum Tolerance

Level

Likelihood MaximumTolerance

Level

ImpactMaximumTolerance

Level

Impact MinimumTolerance

Level

Max

imu

mC

on

tin

gen

cy T

hre

sho

ld

Risk Assessment Matrix (RAM)

Contingency Planning Matrix (CPM)

Tier 3

Tier 2

Tier 1

risk profileIn

sign

ific

ant

Insignificant

Intolerable

Into

lerable risk

profile

Min

imu

m C

on

tin

gen

cy T

hre

sho

ld

Minimum Contingency Threshold

Maximum Contingency Threshold


agreed-upon thresholds, preparedness and response tiers, and administration and collaboration

schemes.

4.1 Transforming Matrix Dimensions

The axes of the CPM should address the two main goals of contingency planning: risk-

preparedness and incident-response. To successfully fulfill both goals, suitable resources must

be available. More likely or more frequent risks require more proximate resources, so the

Leakage Likelihood {𝐿} in risk assessment is best translated to Resource Proximity 𝑅𝑝𝑟𝑜𝑥 in

contingency planning, which accounts for the closeness and accessibility of the required

resources. Equivalently, more impactful risks require a wider variety of resources, so the

Leakage Value of Impact VI in risk assessment is best translated to Resource Variety 𝑅𝑣𝑎𝑟𝑖 in

contingency planning, which accounts for the uniqueness, complexity, and/or specialization of

the required resources. Several metrics can be used to quantify these axes of CPM. For

example, while an inverse-distance metric may quantify 𝑅𝑝𝑟𝑜𝑥, the number of dispatched

incident-response teams may quantify 𝑅𝑣𝑎𝑟𝑖. In this regard, the ranges of 𝑅𝑝𝑟𝑜𝑥 and 𝑅𝑣𝑎𝑟𝑖

need not be continuous or linear; the operator may choose to define and discretize the ranges

of both CPM axes based on the specific conditions and characteristics of the storage project.

For example, the continuous range of tolerable Leakage Likelihood {𝐿} = [10−7, 10−1] in

Figure 2.7 may be translated into five discretized values of 𝑃𝑝𝑟𝑜𝑥 = {< 1 2000⁄ ; 1 2000⁄ −

1 1000⁄ ; 1 1000⁄ − 1 500⁄ ; 1 500⁄ − 1 100⁄ ; > 1 100⁄ } km-1

, corresponding to the inverse

radial distance below which contingency-planning resources should be accessible; larger {L}

is translated into larger 𝑅𝑝𝑟𝑜𝑥 and therefore shorter distance between the resources and the

storage site. Finally, the increase in the overall level of risk – defined as the multiplication of

{𝐿} by VI – is translated to an increase in the overall amount of resources that should be

available to combat leakage. As such, constant risk-level contours are translated into constant

resource-amount contours. Intuitively, contingency planning requires fewer resources for

less likely and/or less impactful risks but more resources for more likely and/or more

impactful risks.


Ultimately, the translation of RAM axes to CPM axes emphasizes that an effective allocation

of resources for contingency planning shall ensure both timely and thorough preparation and

response, irrespective of whether the addressed leakage scenario is high or low in likelihood or

impact. We further clarify this point by considering two opposing risk scenarios. On one hand,

for a high-impact but low-likelihood leakage risk, the proposed guidelines suggest allocating

more various but less proximate resources. In addition to facilitating a thorough response, the

resources’ high variety can compensate for their low proximity and thus facilitate a timely

response. For example, highly various resources may cover unique logistical expenditures that

accelerate the deployment of specific response equipment (e.g. expedited air shipping) or

teams (e.g. high wages) on short-notice [74], or that boost general response operations (e.g.

high rent of temporary accommodation for relocated communities) [75]. Indeed, because of

the low likelihood, it would be inefficient to make these resources available on or close to the

storage site permanently. On the other hand, for a high-likelihood but low-impact leakage risk,

the proposed guidelines suggest allocating more proximate but less various resources. Here,

the resources’ high proximity can preventively compensate for their low variety and thus

facilitate a thorough, as well as timely, response. For example, learning from analogues in the

oil industry [76], very proximate resources may include on-site devices that enable a rapid –

and if necessary, remote – shutdown of operations (e.g. stopping CO2 injection in case of

over-pressurization), which prevents the need for more complex corrective equipment (e.g.

drilling a relief well). However, if intervention is delayed and pressurization escalates, such

complex resources may become necessary.

4.2 Transforming Matrix Boundaries

Both intolerable and insignificant risks are not presented in the contingency planning matrix.

To that end, the tolerance levels in a risk assessment matrix can be considered contingency

thresholds in the contingency planning matrix. The minimum risk tolerance levels for

likelihood {𝐿} and impact (VI) are translated to minimum contingency thresholds, below

which no contingency planning is required. Equivalently, the maximum tolerance levels for

{𝐿} and VI are translated to maximum contingency thresholds, above which contingency

planning is insufficient and risk mitigation is necessary.


4.3 Classifying Risk into Tiers

Because not all tolerable risks are equal in likelihood or impact, distinct tiers of risk-

preparedness and incident-response must be defined to address different tolerable risks with

different requirements for resources proximity and variety. This study adopts a three-tier

system for preparedness and response.

4.3.1. Scope of the Three-Tier System

The three-tier system is borrowed from the oil and gas industry where it has been extensively

implemented [35, 77, 78, 79]. Table 2.2 lists three main criteria to properly assign a tolerable

leakage risk to one of the three tiers: the geographic location of the leakage scenario and any

resulting response operations; the governance structure among all parties involved in leakage

preparedness and response; and the ownership, proximity, and variety of available resources.

These tiers are discussed in more detail when presenting a model contingency plan later

(Section 5).

Table 2.2: Selection criteria for the three-tier system

Criteria Tier 1 Tier 2 Tier 3

Geographic

location operation site

local vicinity of the

operation site

regional vicinity of the

operation site

Governance

structure

operating party and

its contractors,

regulatory agency

operating party and

its contractors,

regulatory agency,

local stakeholders

operating party and

its contractors,

regulatory agency,

local stakeholders,

regional stakeholders

Resources

ownership,

proximity,

and variety

owned by the operating

party and its direct

contractors

least unique, complex,

and

specialized

trade-off: more proximate

but less various resources


party, its direct

contractors,

and local stakeholders

moderately unique,

complex, and specialized


party,

its direct contractors,

local stakeholders,

and regional stakeholders

most unique, complex, and

specialized

trade-off: more various but

less proximate resources


According to these criteria, Tier 1 is under the direct jurisdiction of the party operating the

storage site and its contractors, and it addresses onsite risks that can be handled through

standard and generic resources of relatively low complexity and specialization. Tier 2 expands

the scope of covered risks to include those that might affect local communities around the

operation site, might require the support and intervention of local governmental authorities

and public-safety departments, or might necessitate the deployment of more complex or

specialized resources. Finally, Tier 3 includes the most risky leakage scenarios whose effects

might expand to the regional level, requiring the support and intervention of regional

authorities, or necessitating the deployment of extensive, highly complex, or highly

specialized resources. In this regard, “regional” in this analysis may refer to district-level,

national-level, or multinational-level geographic zones; the exact definition of “regional”

depends on the scale and conditions of the CO2 storage reservoir and is thus project-specific.

4.3.2. Representation of the Three-Tier System

As can be noticed in Figure 2.8, the three contingency planning tiers cover all tolerable risks.

Tier 1 prepares for risks and responds to incidents that require the smallest 𝑅𝑣𝑎𝑟𝑖 while Tier 3

prepares for risks and responds to incidents that require the largest 𝑅𝑣𝑎𝑟𝑖.

The range of 𝑅𝑣𝑎𝑟𝑖 covered by Tier 1 shrinks as 𝑅𝑝𝑟𝑜𝑥 increases. Tier 1 is primarily

administered by the operating party, which, naturally, tends to concentrate its relevant

resources close to the storage site that it directly manages. Because of space and logistical

constrains, the resources available under this tier tend to be relatively limited in their scope

and variety. Accordingly, as shown in Figure 2.9, Tier 1 covers more proximate but less

various resources relative to constant resource-amount contours. In other words, the closer the

required resources need to be located to effectively address a leakage, the less complex and

specialized they should be to remain covered under Tier 1. However, if the resources need to

be more complex and specialized in addition to being nearby, Tier 1 may not be sufficient, in

which case the leakage should be covered under Tier 2 by bringing onboard further resources

from local stakeholders and communities.


Figure 2.9: Tier system tradeoff between resource proximity and resource diversity

The opposite argument holds for Tier 3. The range of 𝑅𝑣𝑎𝑟𝑖 covered by Tier 3 shrinks as 𝑅𝑝𝑟𝑜𝑥

decreases. Since Tier 3 requires the engagement of multiple parties within a relatively large

geographical area, the scope of resources available for this tier is relatively large. Accordingly,

Tier 3 covers more various but less proximate resources relative to constant resource-amount

contours. In other words, the farther the required resources can be located to effectively

address a leakage, the more complex and specialized they should be to remain covered under

Tier 3. However, if the required resources can be less complex and specialized in addition to

being distant, Tier 3 may not be necessary, in which case the leakage should be covered under

Tier 2.

Another implication of the proposed tier system and resulting contingency planning matrix

(CPM) is the ability to address one leakage profile at multiple tiers. Knowing the measure of

the geologic Indicator associated with each leakage scenario allows determining its likelihood

and impact-value, which in turn allow determining the proper proximity, variety, and amount

of resources required for effective preparedness and response. Consequently, as illustrated in

Figure 2.9, a risk profile (corresponding to a leakage trajectory) can be covered under Tier 1

for relatively low likelihood and impact or under Tier 3 for relatively high likelihood and

impact.

VI

{L}


Tier 3

Tier 2

Tier 1

risk profile

lower variety

higher proximity

Min

imu

m C

on

tin

gen

cy T

hre

sho

ld


Max

imu

mC

on

tin

gen

cy T

hre

sho

ldlower proximity

higher variety


The boundaries between the three tiers can be thought of as additional contingency thresholds,

which necessitate shifting from one strategy of risk-preparedness and incident-response to the

other. Thus, besides the maximum and minimum thresholds identified earlier, contingency

planning requires defining two tier thresholds: Tier 1-2 threshold determines the boundary

between Tier 1 and Tier 2, and Tier 2-3 threshold determines the boundary between Tier 2 and

Tier 3. Important to note, however, the intra-boundaries between the three tiers are not set at

or dictated by constant resource-amount contours due to the trade-offs highlighted earlier.

Appendix A provides further explanation on how the categorization of risk according to

constant-risk contours may cause nontrivial pitfalls in contingency planning.

4.3.3. Negotiating Contingency Thresholds

When setting contingency thresholds, the operating party is usually guided by several

considerations, including: external regulations and standards, internal safety culture and

resource capabilities, compliance with the terms and conditions of insurance policies, and

accommodation of the interests of communities affected by the CO2 storage project. To that

end, the exact specification of minimum, maximum, and tier thresholds are usually negotiated

through a collaborative effort between the operating party and multiple stakeholders.

To start, although formal minimum and maximum contingency thresholds may be dictated by

regulations or industrial standards, their implementation is usually shaped by the

administrative procedures and protocols of the legally liable party managing the CO2 storage

site. To that end, the actual enforcement of these thresholds requires clear and effective

communication. The operating party has to demonstrate its ability to reduce intolerable risks

below mandated maximum contingency thresholds through insurance, safety measures, and

system reinforcement. If the operating party fails to demonstrate such capability, the

regulatory agency may not permit the project. Alternatively, the operating party may find it

beneficial to adopt stricter thresholds than those dictated by regulations. For example,

insurance rates may be lower if the project deploys more frequent monitoring or more accurate

measurement tools than what is legally required. In this case, the adopted minimum and

maximum thresholds would depend on the operating party’s safety culture and resource

CHPATER 2 — A Model Contingency Plan 57

capabilities: trading more aggressive risk mitigation for less aggressive contingency planning,

or vice versa. The regulatory agency would still need to be consulted on such trade-offs.

Equivalently, local, national, or international laws influence how the operating party sets the

boundaries between the three contingency tiers. For example, the operating party may be

legally required to notify local authorities about small leakage incidents while giving regional

authorities the right to unilaterally and directly intervene in the case of large leakage incidents

[21]. In this case, the regulatory agency would argue for keeping small leakage incidents

within Tier 1 while necessitating Tier 2 (and potentially Tier 3) for large leakage incidents. In

addition, the operating party may be legally required to engage with several stakeholders on

the CO2 storage project, including local communities and their emergency-response

departments. Those relationships shape the operating party’s own preferences on how to

categorize risks into different tiers. For instance, a small company might operate and manage a

CO2 storage site within a county that has a large pool of publically funded resources for

contingency planning. In this case, to minimize costs, the operating party would be inclined to

allocate fewer risks under Tier 1 and more risks under Tier 2 and Tier 3. Alternatively, the

operating company might already have a large inventory of emergency-response equipment,

guided by its established safety procedures and long history of international operations. In this

case, to maximize operational and decision-making autonomy, the operating party would be

inclined to allocate more risks under Tier 1 and fewer risks under Tier 2 and Tier 3.

5 A Model Contingency Plan

The proposed RAM and CPM are essential building-blocks in the construction of an effective

and comprehensive contingency plan. To demonstrate their role, we present an overview of

the basic elements of a model contingency plan for CO2 leakage through geologic pathways,

illustrated in the exhibit “Outline of a Model Contingency Plan.” Although the model plan

includes four major sections, we devote our attention to the design of the contingency tiers in

the third section, where the various elements of RAM and CPM are mostly relevant.

Accordingly, while this paper proceeds with a detailed discussion of the elements of the


contingency tiers in the proposed model contingency plan, a brief description of the remaining

three major sections of the plan is included in the referenced exhibit.

5.1 Tiers of Risk-Preparedness and Incident-Response

5.1.1. Thresholds

As explained in the derivation of the proposed CPM, a total of six thresholds should be

identified in a contingency plan: two minimum contingency thresholds corresponding to

minimum tolerance levels of risk likelihood and impact; two maximum contingency

thresholds corresponding to maximum tolerance levels of risk likelihood and impact; Tier 1-2

threshold bordering between Tier 1 and Tier 2; and Tier 2-3 threshold bordering between Tier

2 and Tier 3.

5.1.2. Leakage Evaluation

Two types of leakage assessment frameworks are important to establish: one for risk-

preparedness and one for incident-response. Each framework should include a checklist to

characterize the leakage and categorize it under one of the three contingency tiers.

Risk-Preparedness: in order to adequately prepare for risk, each leakage scenario should be

evaluated through a checklist that identifies its: 1) Indicator, Origin, Pathway, and Endpoint;

2) Leakage Likelihood {𝐿}; and 3) consequences quantified through a predicted Leakage Value

of Impact VI. As explained earlier and illustrated in Figure 2.8, {𝐿} and VI in RAM are

translated to 𝑅𝑝𝑟𝑜𝑥 and 𝑅𝑣𝑎𝑟𝑖 in the CPM, respectively. Then, using the CPM, the adequate

tier to prepare for this prospective leakage scenario is specified at the intersection of the

obtained 𝑅𝑝𝑟𝑜𝑥 or 𝑅𝑣𝑎𝑟𝑖 with the corresponding risk profile, as we demonstrate in Figures

2.10a and 2.10b, respectively. Ultimately, the extent of risks covered under each contingency

planning tier influences the amount, type, and location of resources that should be assigned to

that tier.


Incident-Response: an effective response to a detected CO2 leakage requires a careful yet

prompt evaluation of the leakage consequences. Therefore, the response checklist for a

leakage incident should identify its: 1) Indicator, Origin, Pathway, and Endpoint; and 2)

consequences measured in any feasible form, including Value of Flow 𝑉𝐹, Value of Damage

in Subsurface 𝑉𝐷𝑠𝑢𝑏, Value of Damage on Surface 𝑉𝐷𝑠𝑢𝑟, or Leakage Value of Impact VI. If

VI is not the quickest or most practical way to directly measure the leakage consequences,

5/4/2016 New England Energy System Case Study

7

Directory

Introduction

Tiers of Risk-Preparedness and Incident-Response

Documentation

This section defines the purpose, scope, and relevance of the contingency plan [33]. This section should also introduce the party in-charge of designing and

updating the contingency plan document; such party may be the site-operating firm, a specific team within that firm, or a contracted team by that firm.

• ScopeA comprehensive contingency plan should cover risks associated with surface (e.g. well blowouts) and subsurface releases, through both man-made and

geologic pathways. Accordingly, in this analysis of subsurface leakage through geologic pathways only, this section should list: all possible Origins, Endpoints,

and Pathways, the surface location of Origins; and the location of major human structures (e.g. cities) and ecosystems (e.g. vulnerable habitats) at and in the

vicinity of the storage site. The use of terrestrial maps is important.

• Purpose and ObjectiveThis section defines the roles of the staff members and contractors and the procedures they should follow to prepare for risks and respond to incidents of CO2

leakage from [name] geologic storage reservoir through, in this case, geologic pathways. This section should also outline the objectives of the contingency plan,

which may include: ensuring that preparedness and response are consistent with industrial practices and in compliance with regulatory standards; ensuring a full

and effective integration and utilization of industry and government resources when needed; and prioritizing corrective-actions.

• PrioritiesThis section summarizes the main priorities of the contingency plan, which may include: securing human safety during incident-response; minimizing the

impact of leakage on human health the environment; minimizing the damage to equipment and assets used in incident-response; minimizing the likelihood of

leakage through adequate management of resources; or minimizing disruption to CO2 storage activities.

• Legal and Regulatory ComplianceThis section enlists all relevant governmental requirements, regulations, and laws; industrial standards and protocols; as well as international laws and treaties

with which the contingency plan complies. This section may also include a list of the required reporting to external stakeholders if a leakage incident occurs.

• Plan IntegrationThis section enlists other contingency plans and safety protocols by relevant internal and external stakeholders that complement the contingency plan of

concern, including: health, safety, and environment (HS&E) policies, standards, and guidelines enforced at the storage site by the operating party or its

contractors; governmental plans covering emergency response and relief (e.g. city or county emergency response plans); and industrial plans governing private

service centers for emergency response and relief (e.g. Clean Caribbean and Americas [85], Oil Spill Response and East Asia Response Limited [86]).

A directory is the first section any holder of the contingency plan should have access to in order to reach all decision-making stakeholders. The directory lists

the contact information of people that are directly involved in risk-preparedness and incident-response, including personnel from the operating company,

contractors, regulatory agency, local and regional governmental authorities, and community representatives.

• Thresholds

• Leakage Evaluation

• Response Initiation

• Response Strategies

• General Operations

• Specific Operations: Corrective Measures

• Human and Equipment Resources

• Administration and Coordination

The contingency plan should be a living document, updated as the project progresses from planning operation, and periodically thereafter as more is learned

about the performance of the storage site. To that end, the information included in the contingency plan should be reviewed regularly to incorporate any

changes in risk assessment methods, tier-based preparedness and response procedures, resources inventories and allocation, training and maintenance schedules,

or personnel directory. The operating party should maintain a detailed record of previous leakage incidents, response measures, best practices, and lessons

learned, and it should update the incident-response procedures based on the new information gained from those incidents. In addition, the operating party is

responsible for making the contingency plan available to all relevant internal departments and external stakeholders.

Outline of a Model Contingency Plan


other value models 𝑉𝐹, 𝑉𝐷𝑠𝑢𝑏, 𝑉𝐷𝑠𝑢𝑟 may be used as proxy. Then, using a customized form

of (2) and the guidelines presented in Section 3.2.2, the measured consequences can be

translated to VI, which in turn can then be translated to 𝑅𝑣𝑎𝑟𝑖. Subsequently, using the CPM,

the adequate tier to respond to this leakage incident is specified at the intersection of the

obtained 𝑅𝑣𝑎𝑟𝑖 with the corresponding risk profile, as shown in Figure 2.10b. In this case, we

note that the 𝑅𝑝𝑟𝑜𝑥 axis – derived from the likelihood {𝐿} – plays no role in choosing the

contingency tier; it is maintained in Figure 2.10b for the sole purpose of properly plotting the

risk profiles.

Figure 2.10: Tier allocation procedure for risk-preparedness and incident-response

The leakage characterization checklist for incident-response is different from that for risk-

preparedness. First, Leakage Likelihood is not considered in incident-response because the

leak has already occurred. Second, because leakage is detectable in incident-response, its

consequences can be measured rather than predicted, which should result in a more accurate

characterization.

5.1.3. Response Initiation

The transition from risk-preparedness to incident-response commences upon the identification

of a measurement that exceeds a trigger value. A trigger is best described as an observation

that renders VI above its preset minimum tolerance level. Unlike in the case of risk-

preparedness where VI is predicted yet is still uncertain, a trigger observation renders VI above

VI

{L}

Tier 3

Tier 2

Tier 1

risk profile

VI

{L}

Tier 3

Tier 2

Tier 1

risk profile

(a) (b)

This leakage scenario is allocated under Tier 2

This leakage scenario is allocated under Tier 2


its minimum tolerance level with certainty, which in turn means that the minimum

contingency threshold has been exceeded – and a response procedure must be initiated – also

with certainty. To that end, triggers may be manifested as an irregularity in a wide range of

measurable data; examples of triggers include: excessive pressure-buildup in Safe, high level

of pH or trace metals in a FrW, or high concentration of dissolved CO2 in the oil recovered

from an O&G. In addition, the causes of a trigger may be an FEP from within the analyzed

system (e.g. geochemical erosion of the caprock) or an EFEP from outside the analyzed

system (e.g. new hydrocarbons’ drilling activities in the vicinity of the storage site or a large

earthquake in the vicinity of the storage project).

Whether identified through regular or special monitoring or inspection, and regardless of the

instrumentation or methodology used, once a trigger is identified, a response procedure should

be initiated to detect then control or remediate leakage. Because leakage detection through

trigger identification is the first step in incident-response, contingency plans should specify an

extensive list of triggers. To that end, similar to contingency thresholds, response triggers

should be selected by mutual agreement between the operator, regulatory agencies, and other

appropriate stakeholders. The ability to respond quickly and decisively to abnormal events is

one of the primary benefits of pre-negotiating these triggers.

5.1.4. Response Strategies

For each type of leakage event, a pre-planned course of action should be developed,

depending on which tier it falls under. The response strategy should include two types of

actions: general operations and specific operations. A summary Table of both operation types

is presented in Appendix B.

5.1.4.1. General Operations

Because the general operations are applicable to all leakage incidents, they should be designed

and deployed in a way that reflects the priorities identified in the contingency plan. Most

notably, the safety of all personnel responding to leakage incidents should be secured; in this

case, the safety of the responding teams to subsurface leaks through geologic pathways may


be jeopardized due to either leakage causes (e.g. big earthquakes) or leakage consequences

(e.g. contaminated drinking water). In addition, general response operations should focus on

recovering normal CO2 storage activities as soon as possible.

The responding teams should have a clear action-plan on how to mobilize and deploy

resources. This becomes especially important in Tier 2 and Tier 3 where some required

resources are not owned by the operating party but rather managed and deployed by external

stakeholders (e.g. municipal or state public-safety teams). To that end, Tiers 2 and 3 should

fulfill two additional goals through general operations. First, each local and regional

stakeholder should be assigned a clear responsibility zone to prevent conflicting decisions and

delays in execution. Second, because leakage may impact local and regional populations, the

operating party should be ready to provide supplementary support and services beyond its

typical business. Examples of such services may include: security, evacuation,

accommodation, transportation, and medical supervision for relocated communities; catering

services for both the response teams and the affected communities; and financial

compensation for damages or harms caused to nearby businesses or institutes.

5.1.4.2. Specific Operations: Corrective Measures

Specific response operations involve choosing proper corrective measures to control or

remediate leakage. Several corrective measures have been proposed in literature [31, 38, 46,

48, 80], an example list of which is presented in Table 2.3. As shown, we envision four

primary criteria affecting the feasibility and effectiveness of each corrective measure:

objective, target formation, scale of deployment, and cost of deployment. First, it is important

to identify whether the goal of the corrective action is to control (stop or contain) the leakage

or to remediate its impacts. Second, given the specifics of the leakage trajectory and

subsurface configuration, a corrective action may be carried out at the Origin of the leak, on

the transport Pathway, or at the Endpoint (e.g. a freshwater aquifer). Third, some corrective

measures are best applied only in a particular location of the subsurface formation whereas

others could be applied at multiple locations throughout the subsurface. Finally, depending on

the chosen technique, different corrective measures require different monetary investments

63

Table 2.3: Example corrective measures for incident-response [31, 38, 46, 48, 80]

Corrective Actions Objective Target

Formation

Scale of

Deployment Deployment Components Index

Reduce CO2 injection rate to reduce

pressure-buildup control Origin particular location existing wells A

Stop CO2 injection to reduce

pressure-buildup control Origin particular location existing wells B

Partially extract CO2 from the storage

reservoir to reduce pressure-buildup control Origin

particular location;

multiple locations

existing wells; new wells;

pumps C

Extract CO2 at leakage point control Origin particular location new wells; pumps D

Extract water from the storage reservoir to

reduce pressure-buildup control Origin


multiple locations


pumps; water treatment E

Inject water in upper formations of the storage

reservoir as a hydraulic barrier control Pathway multiple locations


pumps F

Inject water to dissolve the leaking CO2 control Pathway particular location existing wells; new wells;

pumps G

Inject sealing material at leakage point or

pathway (e.g. cement, gels, polymers) control

Origin,

Pathway particular location new wells; pumps; sealants H

Extract CO2 from storage reservoir and re-inject

it into another reservoir remediate Origin


multiple locations


pumps I

Extract contaminated freshwater, treat,

then re-inject remediate

Endpoint

(FrW)


multiple locations


pumps; water treatment J

Treat freshwater in the subsurface

(e.g. inject microbes to restore pH) remediate

Endpoint

(FrW) particular location


chemicals or

microorganisms

K

Extract oil or gas, treat, then use remediate Endpoint

(O&G)


multiple locations


pumps; separators L

Inject water to enhance recovery remediate Endpoint

(O&G) particular location


pumps; water treatment M

CHAPTER 2 — A Model Contingency Plan 64

and time commitments to install and operate wells, pumps, separators, or other water-

treatment systems.

After identifying a list of feasible and effective corrective measures, the contingency plan

should include a corrective measures matrix (CMM) that matches each leakage risk profile

(incorporating a group of leakage scenarios of the same Origin-Pathway-Endpoint trajectory)

with the best corrective-action techniques under each contingency tier. Although no extensive

data currently exists on the best corrective measures for each risk profile, a template of the

proposed CMM is presented in Table 2.4; the matrix is filled-in for illustrative purposes only.

As shown, more than one corrective technique can be assigned to each risk profile, and the

methods should be listed in order of priority.

Populating the CMM is dictated not only by the technical effectiveness of the corrective

measures but also by the economic feasibility of deploying them. To that end, the right

portfolio of corrective measures for each risk profile under each tier should balance between

the cost of corrective measures and their benefits, expressed in terms of the “avoided damage”

that would have occurred otherwise due to leakage. In other words, such cost-benefit analysis

compares the VI of a particular leakage scenario to the cost of the corrective measures

controlling or remediating it, and it is a common approach in managing risks that may widely

impact human and natural ecosystems [81, 82]. Ultimately, designing such a CMM before

leakage incidents occur is crucial for ensuring a rapid response to the leakage incidents when

they occur.

Table 2.4: Example corrective measures matrix (CMM) for incident-response

Corrective Measure Tier 1 Tier 2 Tier 3

Risk profile 1 A, E B, E B, E, C

Risk profile 2 A, E, K B, E, K B, E, J

… … … …

Risk profile 9 B B, D B, D, M


5.1.5. Human and Equipment Resources

In addition to distinguishing between resources proximity and variety, which is captured in the

CPM, the contingency plan should make a clear distinction between human and equipment

resources. Appendix B presents a summary Table for categorizing resources accordingly.

Consistent with the type of response strategies defined earlier, human resources should be

classified into general- and specialized-response teams. In addition, a detailed inventory

should be developed and maintained by the operating party for all internal and contracted

incident-response personnel. Such inventory should identify the name, job title, team

membership, and contact information for each internal personnel, as well as the formal

affiliation for contracted service-providers. For incident-response activities under Tier 2 and

Tier 3, the inventory should also identify which personnel should act as liaison with external

local or regional response teams (e.g. police, firefighters, industrial response centers) and

communities (e.g. residents, schools).

Similar to human resources, the contingency plan should include a detailed inventory of

equipment resources – both general and specialized – as well as a detailed timeline for their

mobilization. In the case of Tier 1, all equipment is owned by the operating party or its

contractors. However, for Tier 2 and Tier 3, the inventory should specify the source of

equipment deployed by external local or regional stakeholders.

Finally, integral to securing effective response is human training and equipment maintenance.

In addition to initial training, responding personnel should undergo regular refresher sessions

in order to update their knowledge and skills as the project progresses. In fact, beyond Tier 1,

joint drills with external (local and regional) parties become necessary [34]. Equivalently,

equipment stockpiles used for safety as well as for leakage analysis, monitoring, control, and

remediation should undergo periodical inspection and testing to ensure proper functionality.

5.1.6. Administration and Coordination

Effective administration of the tier-based contingency planning is key to secure proper

implementation, communication, and accountability. Establishing an administrative


framework for each tier involves identifying a clear hierarchy for decision-making, a

notification protocol to report leakage incidents, and a communication-flow scheme to keep all

relevant parties informed and updated. While the decision hierarchy and communication

scheme are needed for both risk-preparedness and incident-response, the notification protocol

is only relevant if a leakage incident occurs.

Figure 2.11: Example decision-making hierarchy of the operating party for contingency

planning

As shown in Figure 2.11, the decision-making hierarchy within the operating party reflects the

priorities of the contingency plan, expressed earlier in Section 5.1.3. For Tier 1, the primary

decision makers are the onsite operators. The decisions should first secure the safety of

operating personnel during risk-preparedness and incident-response, which is under the broad

supervision of the HS&E team and site management. In addition, securing effective incident-

response requires the presence of skilled and knowledgeable reservoir engineers and modelers

who can refine and update preset corrective-action plans when necessary. Coordinating the

logistics of resources’ mobilization and deployment is also critical. Tiers 2 and 3 require the

deployment of further resources as well as dealing with local and regional stakeholders. To

Tier 3

Tier 2

Tier 1

HeadquartersExecutive Committee

President/CEOVice President for HS&E

Regional Operations CenterRegional Chairman

Vice President for HS&EFinance Member

Legal MemberInformation Coordinator

Local Operations CenterSite Operations Director

Site HS&E Team LeadMaterial/Logistics Coordinator

Finance MemberLegal Member

Information Coordinator

Storage SiteSite Manager on-dutyHS&E Team on-duty

Reservoir Engineering TeamReservoir Modeling Team

Field OperatorsMaterial/Logistics Coordinator


that end, additional administrative support and the involvement of more executive decision-

makers become necessary.

Figure 2.12 shows an example notification protocol that may be followed by the operating

party to report leakage to all relevant parties. One aspect to note is that each notification step

should have a clearly defined timeframe. Internal corporate policies might dictate the

timeframe of internal notifications whereas applied regulations and/or agreements might

determine the notification periods for external stakeholders. Developing this notification

procedure and the aforementioned decision-making hierarchy allows all parties to engage in

an organized communication flow, which can be summarized in the example communication

scheme illustrated in Figure 2.13.

Figure 2.12: Example notification protocol of the operating party for incident-response

Equally important is coordination and collaboration among all parties involved in the

contingency plan in order to maximize efficiency and minimize mistakes and costs. We have

already introduced two venues for collaboration in contingency planning: the negotiations

between the operating party and the regulatory agency on setting the contingency thresholds

and response triggers, as well as the joint training exercises among all response teams. An

additional example of collaboration is illustrated in Figure 2.14, showing multiple cooperative

pathways to pool resources for Tier 2 and Tier 3.

Leakage Detection & Evaluation

notify Field

Operator

notify onsite HS&E and Site Manager

exceed minimum contingency threshold?

resolve onsite

No

Yes

larger than Tier 1-2

threshold?

initiate Tier 1 response

No

notify Regulatory

Agency

Yes

initiate Tier 2 response

notify Local Operations

Center

notify local stakeholders

larger than Tier 2-3

threshold?

Initiate Tier 3 response

notify Local Operations

Centers

NoYes

notify Regional Operations

Centers

notify Regulatory

Agency

notify Regulatory

Agency

Notify local stakeholders

notify regional stakeholders

notify Headquarters


Figure 2.13: Example communication scheme for contingency planning

Figure 2.14: Collaborative approach to securing resources for Tiers 2 and 3

Under Tier 1, coordination is relatively easy, for most dispatched experts and equipment for

leakage preparedness and response are managed directly by the operating party or its

contractors. However, under Tiers 2 or 3, the operating party’s internal and contracted teams

need to further coordinate with local and regional partners. In this case, different parties might

ContractorsExternal Stakeholders

Regional CenterRegional Chairman

Vice President for HS&EFinance Member

Legal MemberInformation Coordinator

Local Operations CenterSite Operations Director

Site HS&E Team LeadMaterial/Logistics Coordinator

Finance MemberLegal Member

Information Coordinator

Administrative Supportexamples: human resources;

media; legal services

Operational Supportexamples: preparation and

deployment of resources; staff training and equipment

inspection; monitoring and detection of leakage;

implementation of general and specific response strategies

Local Stakeholdersexamples: governmental

authorities, public safety services (e.g. police and firefighting);

sensitive neighboring communities (e.g. schools, hospitals, prisons); other

neighboring communities (e.g. residences and businesses)

Regional Stakeholdersexamples: governmental

authorities; public safety services (e.g. police and firefighting,

military)

Regulatory Agency

Storage SiteSite Manager on-dutyHS&E Team on-duty

Reservoir Engineering TeamReservoir Modeling Team

Field OperatorsMaterial/Logistics Coordinator

HeadquartersExecutive Committee

President/CEOVice President for HS&E

Operating Party

• Joint Tier 1 resources of multiple operators in industry

• Joint Tier 1 resources of multiple local public-safety teams

• Specialized Tier 2 resource hub, funded and administrated by multiple operators in industry

Collaboration to fulfill

Tier 2

• Joint Tier 2 resources of multiple operators in industry

• Joint Tier 2 resources of multiple regional (national) emergency-response institutes

• Specialized Tier 3 resource hub, funded and administrated by multiple operators in industry

Collaboration to fulfill

Tier 3

CHAPTER 2 — Conclusions 69

have different priorities or be accustomed to different managerial styles, in which case

coordination becomes challenging. To that end, major contingency planning decisions will

require negotiation and compromise, and they are better handled when all affected parties

have participated in the development and approval of the contingency plan.

6 Conclusions

Although existing regulations on CO2 geologic storage require both assessing leakage risk and

controlling or remediating leakage incidents through corrective measures, these two pieces of

risk management are usually addressed separately. This study proposes a methodological

framework for contingency planning, which links risk assessment and corrective measures.

We achieve this goal in three consecutive steps: updating the representation of the risk

assessment matrix (RAM); translating the risk assessment matrix to a contingency planning

matrix (CPM) that incorporates a tier contingency system; and then using the emerging tiers as

the basis for developing a model contingency plan for risk-preparedness and incident-

response, which encompasses corrective measures. While this study focuses on the risks of

CO2 leakage in the subsurface and through geologic pathways, the proposed framework can be

expanded to include other types of risks, including leakage at the surface or through man-

made pathways.

The updated RAM allows visualizing the three major steps of risk assessment: risk

identification, risk analysis, and risk evaluation, resulting in a comprehensive set of leakage

risk profiles with quantified likelihood, impact, and tolerance levels. Upon dividing the overall

storage site into functional subsystems, various leakage Origins, Pathways, and Endpoints are

identified; leakage scenarios of the same Origin, Pathway, and Endpoint trajectory form a risk

profile. Subsequently, the likelihood of each leakage scenario is analyzed as a series of

conditional probabilities for leakage Origination, Propagation, and Destination, all of which

change as a function of a measurable Indicator. Equivalently, multiple value models are

developed to quantify the leakage flow and impact, which covers the damages caused both in

the subsurface and on the surface. Finally, through tolerance levels specified for both


likelihood and impact, it becomes possible to evaluate what risks are too high and should be

mitigated, and what risks are too small can be safely ignored.

The updated RAM can then be translated to a CPM. Fundamentally, preparing for leakage

risks and responding to leakage incidents require a wide range of resources. To that end, the

likelihood and impact dimensions of risk assessment are translated to resource proximity and

variety, respectively. To ensure both quick mobilization and thorough deployment of

corrective and remediating resources, more likely or frequent risks require more proximate

resources while more impactful risks require more unique, specialized, or complex resources.

In addition, the minimum and maximum risk tolerance levels are translated to contingency

thresholds, defining the upper and lower boundaries for preparedness and response.

Subsequently, to facilitate the assignment of the right resources to each leakage scenario, all

foreseeable risks are categorized under three contingency tiers: Tier 1, Tier 2, and Tier 3. By

design, Tier 1 trades more impactful, less likely risks for less impactful, more likely risks

while Tier 3 does the opposite.

The CPM tier system becomes the cornerstone in the development of a contingency plan. The

model contingency plan presented in this study demonstrates how the three tiers set the

primary criteria for: implementing response strategies; designing a corrective measures matrix

(CMM) that assigns specific control and remediation measures to each leakage profile;

obtaining, mobilizing, deploying, and sustaining the human and equipment resources needed

for incident-response; and formulating a decision-making hierarchy, a notification protocol,

and a communication scheme that allow the operating party to effectively administer the CO2

storage site. After addressing these main topics, it becomes easy to develop the remaining

sections of a contingency plan, which include a directory of response personnel, background

information on the project scope and priorities, a list of triggers that may initiate response, and

a record of previous leakage incidents, best practices, and lessons learned.

Ultimately, the proposed methodological framework presents a dynamic and collaborative

approach to risk management for CO2 geologic storage. Revised experts’ opinions, best

practices from previous leakage incidents, and new learnings from site operations, all can be

captured in the Bayesian probabilities of leakage likelihoods as well as the value models of


leakage consequences, resulting in an updated set of risk profiles. This redistribution of risk

would be translated into a redistribution of resources under each contingency tier as well as

the renegotiation of boundaries among tiers. Through transparent communications and

proactive collaboration among all stakeholders, these systematic updates ensure that the right

leakage risk scenario is covered under the right contingency tier using the right resources, and

that the right corrective measure is deployed quickly and decisively if a leakage occurs.

Consequently, both technological and administrative innovations improve the preparedness for

and correction of leakage incidents.

6.1 Future Work

Like all models, the proposed methodological framework may still face a unique set of

challenges and limitations when implemented for real CO2 storage projects. For completion,

we list some of these limitations and challenges, which present future opportunities to expand

and enhance this work. First, we note the subjectivity of probability assignment and

interpretation in the proposed RAM; different experts might observe the exact same field data

and still assign different probabilities for a leakage scenario. Indeed, such challenge is not

unique to RAM but rather common across probabilistic assessment models. One way to

address this challenge would be to ensure a clear and consistent understanding of each

analyzed uncertainty by all consulted experts; our approach strives to achieve this goal by

relying on sequential conditional probabilities in the Bayesian event tree to untangle the

various uncertain attributes of a leakage event. Also important is using uniform and unbiased

protocols for probability elicitation from experts. While such initiatives are already underway

[45, 62], it would be valuable to explore how to adapt existing protocols in others fields for

the specific context of CO2 leakage risks [83, 84]. Another related challenge is the potential

need for extensive input data in order to support detailed probabilistic analyses. An important

question that remains to be addressed in this regard is: how to balance between the RAM

accuracy and simplicity? And how to test and verify that the analysis is detailed enough?

With CPM, one potential challenge may be the need for multiple metrics to quantify the 𝑃𝑝𝑟𝑜𝑥

and 𝑃𝑣𝑎𝑟𝑖 axes. For example, the site operator may find it necessary to categorize the


complexity of resources not only by the number of dispatched expert teams but also by the

size of deployed equipment. Accordingly, one may envision translating one RAM into a

collection of CPMs, each with two unique metrics quantifying its two axes. To that end, it

might be useful to explore a proper procedure to design such collection of CPMs while

preserving the overarching three-tier contingency system they share.

Moving forward, however, an immediate next-step should be testing the application of the

proposed framework through a real case-study that uses real data corresponding to a real CO2

storage project. By constructing the RAMs, CPMs, and CMMs for multiple sequential stages

of the project’s operational timeline, such case-study can demonstrate this framework’s ability

not only to analyze leakage risks and develop contingency plans but also to track and update

them over time.


References

[1] GCCSI, "The Global Status of CCS," Global CCS Institute, Melbourne, 2014.

[2] TNS Opinion & Social, "Public Awareness and Acceptance of CO2 capture and storage,"

EuroBarometer, European Commission, Brussels, 2011.

[3] P. Upham and T. Roberts, "Public Perceptions of CCS: the results of NearCO2 European Focus

Groups," NearCO2, 2010.

[4] I. Wright, P. Ashworth, S. Xin, L. Di, Z. Yizhong, X. Liang, J. Anderson, S. Shackley, K. Itaoka,

S. Wade, J. Asamoah and D. Reiner, "Public Perception of Carbon Dioxide Capture and Storage:

Prioritised Assessment of Issues and Concerns," CO2 Capture Project.

[5] IEA, "Regulatory Frameworks for CCS," 2015. [Online]. Available:

http://www.iea.org/topics/ccs/subtopics/permittingframeworksforccs/. [Accessed May 2015].

[6] S. Bonham and I. Chrysostomidis, "Regulatory Challenges and Key Lessons Learned From Real

World Development of CCS Projects," CO2 Capture Project, 2012.

[7] EPA, "Federal Requirements Under the Underground Injection Control (UIC) Program for Carbon

Dioxide (CO2) Geologic Sequestration (GS) Wells," Environmental Protection Agency, pp.

Federal Register. Vol. 75, No. 237, 2010.

[8] EPA, "Underground Injection Control (UIC) Program Class VI Well Area of Review Evaluation

and Corrective Action Guidance," Environmental Protection Agency. Office of Water, pp. EPA

816-R-13-005, 2013a.

[9] EPA, "Summary of EPA’s Responses to Public Comments Received on the Draft Class VI Well

Testing and Monitoring Guidance," Environmental Protection Agency. Office of Water, pp. EPA

816-S-13-001, 2013b.

[10] K. Gagnon, "Canada Update: Select CCS Regulatory Developments," Natural Resources Canada,

IEA 6th CCS Regulatory Network Meeting. Paris, 2014.

[11] S. McCoy, "CSA Z741-12 and requirements for geologic storage," 6th International IEA CCS

Regulatory Network Meeting, 2014.

[12] M. Leering, "CSA Z741 – Bi-National Standard for Geological Storage of Carbon Dioxide," CSA

Standards, 2012.

[13] Bill 24, "Carbon Capture and Storage Amendment Act," The Legislative Assembly of Alberta.

Third Session, 27th Legislature, 59 Elizabeth II , 2010.

[14] Alberta Energy, "Carbon Capture and Storage - Summary Report of the Regulatory Framework

Assessment," Alberta Energy, 2012.


[15] Alberta, "Alberta Regulation 68/2011. Mines and Minerals Act. Carbon Sequestration Tenure

Regulation," 2011. [Online]. Available: http://www.qp.alberta.ca/. [Accessed May 2015].

[16] British Columbia, "Carbon Capture and Storage Regulatory Policy - Discussion and Comment

Paper," Ministry of Natural Gas Development. Province of British Columbia, 2014.

[17] Office of Parliamentary Counsel, "Offshore Petroleum and Greenhouse Gas Storage Act 2006,"

ComLaw, Canberra, 2006.

[18] Environment Protection and Heritage Council, "Environmental Guidelines for Carbon Dioxide

Capture and Geological Storage," Commonwealth of Australia and each Australian State and

Territory, 2009.

[19] The Parliament of Victoria, "Greenhouse Gas Geological Sequestration Act," The Parliament of

Victoria, 2008.

[20] Queensland Parliamentary Counsel, "Greenhouse Gas Storage Act," Office of the Queensland

Parliamentary Counsel, 2009.

[21] EC, "DIRECTIVE 2009/31/EC OF THE EUROPEAN PARLIAMENT AND OF THE

COUNCILof 23 April 2009on the geological storage of carbon dioxide," European Commission.

Official Journal of the European Union, 2009.

[22] EC, "Guidance Document 1. CO2 Storage Life Cycle Risk Management Framework," European

Communities, 2011a.

[23] EC, "Guidance Document 2. Characterisation of the Storage Complex, CO2 Stream Composition,

Monitoring and Corrective Measures," European Communities, 2011b.

[24] L.-C. Liu, Q. Li, J.-T. Zhang and D. Cao, "Toward a framework of environmental risk

management for CO2 geological storage in china: gaps and suggestions for future regulations,"

Mitigation and Adaptation Strategies for Global Change, pp. 10.1007/s11027-014-9589-9, 2014.

[25] D. Seligsohn, Y. Liu, S. Forbes, Z. Dongjie and L. West, "CCS in China: Toward an

Environmental, Health, and Safety Regulatory Framework," World Resources Institute,

Washington, DC, 2010.

[26] DNV, "Geological Storage of Carbon Dioxide," DNV, DNV-RP-J203, 2012.

[27] ISO, "Risk Management – Principles. ISO 31000:2009," International Organization for

Standardization, 2009.

[28] S. Vajjhala, J. Gode and A. Torvanger, "An International Regulatory Framework for Risk

Governance of Carbon Capture and Storage," Center for International Climate and Environmental

Research, Oslo, Norway, 2007.


[29] ScottishPower CCS Consortium, "UK Carbon Capture and Storage Demonstration Competition:

Corrective Measures Plan - UKCCS - KT - S7.20 - Shell - 001," UK Carbon Capture and Storage

Demonstration Competition, 2011.

[30] S. C. Energy, "QUEST Carbon Capture and Storage - Risk-Based Measurement, Monitoring &

Verification," in MVA/MMV Knowledge Sharing Workshop, Mobile Alabama, 2012.

[31] V. Kuurskraa and M. L. Godec, "Remediation of Leakage from CO2 Storage Reservoirs," IEA

Greenhouse Gas R&D Programme, 2007.

[32] J.-C. Manceau, D. G. Hatzignatiou, L. D. Lary, N. B. Jensen, K. Flornes, T. L. Guenan and A.

Reveillere, "Methodologies and Technologies for Mitigation of Undesired CO2 Migration In the

Subsurface," IEAGHG, UK, 2013.

[33] Eni Australia, "Joseph Bonaparte Gulf Oil Spill Contingency Plan," Eni Australia B.V., 2009.

[34] C. K. Berry and J. R. Bogner, "Model Emergency Response Plan," North Carolina Department of

Labor (NCDOL), Raleigh.

[35] IPIECA, "Guide to Tiered Preparedness and Response," International Petroleum Industry

Environmental Conservation Association, London, United Kingdom, 2007.

[36] NIST, "Contingency Plan Template, Appendix I-3," [Online]. [Accessed 2012].

[37] S. K. Singh and R. Ranjan, "Oil Spill Contingency Plan for Upstream Petroleum Operations in

Andhra Pradesh Offshore Area, Rajahmundry Asset," Oil and Natural Gas Corporation Ltd,

Rajahmundry, India, 2007.

[38] J.-C. Manceaua, D. Hatzignatiou, d. N. J. L. de Larya and A. Réveillèrea, "Mitigation and

remediation technologies and practices in case of undesired migration of CO2 from a geological

storage unit - Current status," International Journal of Greenhouse Gas Control, vol. 22, pp. 272-

290, 2014.

[39] A. Nicol and M. Gerstenberger, "Risk Assessment in CCS," Sanya China, 2011.

[40] J. Condora, D. Unatrakarna, M. Wilsona and K. Asgharia, "A Comparative Analysis of Risk

Assessment Methodologies for the Geologic Storage of Carbon Dioxide," Energy Procedia, vol. 4,

p. 4036–4043, 2011.

[41] Quintessa, "CO2 FEP Database," 2010. [Online]. Available:

http://www.quintessa.org/co2fepdb_v1.1.0/PHP/frames.php. [Accessed 2015].

[42] M. Gerstenberger, A. Nicol, M. Stenhouse, K. Berryman, M. Stirling, T. Webb and W. Smith,

"Modularised logic tree risk assessment method for carbon capture and storage projects," Energy

Procedia, vol. 1, pp. 2495-2502, 2009.


[43] M. Gerstenberger, A. Christophersena, R. Buxtona and A. Nicola, "Bi-directional risk assessment

in carbon capture and storage with Bayesian Networks," International Journal of Greenhouse Gas

Control, vol. 35, pp. 150-159, 2015.

[44] A. Bowden and A. Rigg, "Assessing Risk in CO2 Storage Projects," APPEA Journal, pp. 677-702,

2004.

[45] A. R. Bowden, D. F. Pershkeb and R. Chalaturnyk, "Geosphere risk assessment conducted for the

IEAGHG Weyburn-Midale CO2 Monitoring and Storage Project," International Journal of

Greenhouse Gas Control, vol. 16S, p. S276–S290, 2013.

[46] S. M. Benson and R. Hepple, "Prospects for Early Detection and Options for Remediation of

Leakage from CO2 Sequestration Projects," in Carbon Dioxide Capture for Storage in Deep

Geologic Formations: Results from the CO2 Capture Project, Vol. 2: Geologic Storage of Carbon

Dioxide with Monitoring and Verification, UK, Elsevier Publishing, 2005, pp. 1189-1203.

[47] V. A. Kuuskraa and M. L. Godec, "Remediation of Leakage from CO2 Storage Reservoirs,"

Advanced Resources International, Inc, Arlington, VA, USA, 2007.

[48] S. M. Benson, "Remediation Technologies," Storage in Saline Formations R&D Workshop,

California, 2011.

[49] NETL, "Risk Analysis and Simulation for Geologic Storage of CO2," National Energy Technology

Laboratory, 2009.

[50] K. Hnottavange-Telleena, E. Chabora, R. J. Finley, S. E. Greenberg and S. Marsteller, "Risk

management in a large-scale CO2 geosequestration pilot project, Illinois, USA," Energy Procedia,

vol. 4, p. 4044–4051, 2011.

[51] K. Edlmann, "Risk Assessment in Geologic Storage of CO2," Oxand.

[52] DNV, "CO2QUALSTORE: Guideline for Selection and Qualification of Sites and Projects for

Geological Storage of CO2," DNV, Hovik, Norway, 2009.

[53] E. Pate-Cornell, "Uncertainties in risk analysis: Six Levels of treatment," Reliability Engineering

and System Safety, pp. 95-111, 1996.

[54] B. J. Garrick, "Recent Case Studies and Advancements in Probabilistic Risk Assessment," Risk

Analysis, vol. 4, no. 4, pp. 267-279, 1984.

[55] E. Pate-Cornell, "Chapter 24 - Nuclear Power Plants: TheOriginalProbabilistic Risk Analysis," in

MS&E250A: Engineering Risk Analysis Class Notes, Stanford, Stanford Bookstore, 2013, p. 177.

[56] FutureGen 2.0, "Final Risk Assessment Report for the FutureGen Project Environmental Impact

Statement," U.S. Department of Energy. Contract No. DE-AT26-06NT42921, 2007.


[57] Z. Wang and M. J. Small, "A Bayesian approach to CO2 leakage detection at saline sequestration

sites using pressure measurements," International Journal of Greenhouse Gas Control, vol. 30, pp.

188-196, 2014.

[58] C. A. Griffith, "Physical Characteristics of Caprock Formations used for Geological Storage of

CO2 and the Impact of Uncertainty in Fracture Properties in CO2 Transport through Fractured

Caprocks," Carnegie Mellon University, Pittsburgh, 2012.

[59] S. Jewell and B. Senior, "CO2 Storage Liabilities in the North Sea: An Assessment of Risks and

Financial Consequences," Department of Energy and Climate Change, UK, 2012.

[60] P. D. Jordan, C. M. Oldenburg and J.-P. Nicot, "Estimating the probability of CO2 plumes

encountering faults," Greenhouse Gases Science and Technology, vol. 1, pp. 160-173, 2011.

[61] R. Dunk, "Assessment of Sub Sea Ecosystem Impacts," IEA Greenhouse Gas R&D Programme.

Report No. 2008/8, Gloucestershire, 2008.

[62] A. R. Bowden, D. F. Pershke and R. Chalaturnyk, "Biosphere risk assessment for CO2 storage

projects," International Journal of Greenhouse Gas Control, vol. 16S, pp. S291-S308, 2013.

[63] R. A. Howard, "On Making Life and Death Decisions," Societal Risk Assessment, pp. 89-113,

1980.

[64] R. A. Howard and A. E. Abbas, Foundations of Decision Analysis, Pearson, 2015.

[65] R. S. d. Groot, M. A. Wilson and R. M. Boumans, "A typology for the classification, description

and valuation of ecosystem functions, goods and services," Ecological Economics, vol. 41, pp.

393-408, 2002.

[66] T. Tietenberg and L. Lewis, Environmental & Natural Resource Economics, Pearson, 2015.

[67] IEA, "Potential Impacts on Groundwater Resources of CO2 Geologic Storage," International

Energy Agency Greenhouse Gas R&D Programme, 2011.

[68] K. Damen, A. Faaij and W. Turkenburg, "Health, Safety, and Environmental Risks of

Underground CO2 Storage - Overview of Mechanisms and Current Knowledge," Climatic Change,

vol. 74, p. 289–318, 2006.

[69] R. Melchers, "On the ALARP approach to risk management," Reliability Engineering and System

Safety, vol. 71, pp. 201-208, 2001.

[70] A. Critchlow, "BP faces never ending legal battle for Deepwater disaster," The Telegraph, 17

January 2015.

[71] L. C. S. Jr., M. Smith and P. Ashcroft, "Analysis of Environmental and Economic Damages from

British Petroleum’s Deepwater Horizon Oil Spill," Albany Law Review, vol. 74, no. 1, pp. 563-585,

2011.


[72] A. Chamberlin, "BP lost 55% shareholder value after the Deepwater Horizon incident," Market

Realist, 10 September 2014.

[73] D. Savage, P. R. Mual, S. Benbow and R. C. Walke, "A Generic FEP Database for the Assessment

of Long-Term Performance and Safety of Geologic Storage of CO2," Quintessa, QRS-1060A-1

verion 1.0, 2004.

[74] M. R. Blood and A. Chang, "Coast Guard defends cleanup response to Santa Barbara oil spill," Los

Angeles Daily News, 30 May 2015.

[75] G. J. Wilcox, "Court order seeks to hasten relocation of residents near Porter Ranch gas leakq," Los

Angeles Daily News, 22 December 2015.

[76] R. D. Shell, "Preventing and Responding to Oil Spills in the Alaskan Arctic," Royal Dutch Shell

plc, for Shell Exploration and Production International B.V, The Hague, 2011.

[77] ENI, "Joseph Bonaparte Gulf Oil Spill Contingency Plan," ENI Australia, West Perth, 2009.

[78] H. A. Parker, R. T. Teubner and J. C. Sawicki, "Spill Reponse Planning in the Philippines: 3-Tier

Interaction between Government and Industry," 2009. [Online]. Available:

http://www.interspill.org/previous-events/2009/12-May/pdf/1630_parker.pdf.

[79] UK, "The National Contingency Plan: A Strategic Overview for Responses to Marine Pollution

from Shipping and Offshore Installations," 2014. [Online]. Available:

https://www.gov.uk/government/uploads/system/uploads/attachment_data/file/408385/140829-

NCP-Final.pdf.

[80] S. M. Benson, R. Hepple, J. Apps, C.-F. Tsang and M. Lippman, "Lessons Learned from Natural

and Industrial Analogues for Storage of Carbon Dioixide in Deep Geological Formations,"

Lawrence Berkeley National Laboratory, Berkeley, California, USA, 2002.

[81] IPCC, "Costs, benefits and avoided climate impacts at global and regional levels. IPCC Fourth

Assessment Report: Climate Change 2007.," 2007. [Online]. Available:

https://www.ipcc.ch/publications_and_data/ar4/syr/en/mains5-7.html. [Accessed 2015].

[82] W. D. Nordhaus, "A Review of the Stern Review on the Economics of Climate Change," Journal

of Economie Literature , vol. 45, no. 3, pp. 686-702, 2007.

[83] M. G. Morgan, M. Henrion and M. Small, Uncertainty: A Guide to Dealing with Uncertainty in

Quantitative Risk and Policy Analysis, New York: Cambridge University Press, 1990.

[84] C. S. Spetzler and C.-A. S. S. V. Holstein, "Probability Encoding in Decision Analysis,"

Management Science, vol. 22, no. 3, pp. 340-358, 1975.

[85] CCA, "Clean Caribbean & Americas," [Online]. Available: http://www.cleancaribbean.org/.

[Accessed 2015].


[86] Oil Spill Response , "Services," [Online]. Available: http://www.oilspillresponse.com/services-

landing. [Accessed 2015].

CHAPTER 2 — Appendix A: Drawbacks of Alternative Three-Tier Systems 80

Appendix A: Drawbacks of Alternative Three-Tier Systems

Figure 2.A1: Drawbacks of alternative tier-system approaches for contingency planning

The proposed tier system for contingency planning in Figure 2.9 avoids potential pitfalls in

alternative tier-system designs, as depicted in Figure 2.A1. The resource-variety tier system

in Figure 2.A1a divides the tolerable risk zone into three tiers based only on the impact of

leakage and thus the variety of needed resources; it covers the least impactful leakage

scenarios under Tier 1 and the most impactful scenarios under Tier 3. The problem with this

approach is that it ignores the need for different levels of Resource Proximity to prepare for

and respond to leakage incidents of different likelihoods but of the same impact.

Conversely, the resource-proximity tier system in Figure 2.A1b divides the tolerable risk

zone into three tiers based only on the likelihood of leakage and thus the proximity of needed

VI

{L}


Tier 3

Tier 2

Tier 1

Min

imu

m C

on

tin

gen

cy T

hre

sho

ld


Max

imu

mC

on

tin

gen

cy T

hre

sho

ld

VI

{L}


Tier 1 Tier 2 Tier 3

Min

imu

m C

on

tin

gen

cy T

hre

sho

ld


Max

imu

mC

on

tin

gen

cy T

hre

sho

ld

VI

{L}


Tier 3

Tier 2

Tier 1

Min

imu

m C

on

tin

gen

cy T

hre

sho

ld


Max

imu

mC

on

tin

gen

cy T

hre

sho

ld

VI

{L}


Min

imu

m C

on

tin

gen

cy T

hre

sho

ld


Max

imu

mC

on

tin

gen

cy T

hre

sho

ld

Tier 3

Tier 2

Tier 1

A1a A1b

A1c A1d

CHAPTER 2 — Appendix A: Drawbacks of Alternative Three-Tier Systems 81

resources; it covers the least likely and infrequent leakage scenarios under Tier 1 and the most

likely and frequent scenarios under Tier 3. The problem with this approach is that it ignores

the need for different levels of Resource Variety to prepare for and respond to leakage

incidents of different impact levels but of the same likelihood.

Additionally, the resource-amount tier system divides the tolerable risk zone into three tiers

based on the overall level of risk, or equivalently, based on the overall amount of available

resources; leakage scenarios of the lowest risk levels (lowest likelihood and impact) are

covered under Tier 1 whereas those of the highest risk levels (highest likelihood and impact)

are covered under Tier 3. This resource-amount tier system may be designed in two forms

using continuous or discretized resource-amount contours, as illustrated in Figures 2.A1c and

2.A1d, respectively. The problem with this approach is that it equates leakage scenarios

requiring most proximate but least various resources to those requiring least proximate but

most various resources. The adopted tier system in this study avoids this problem due to the

tradeoff illustrated in Figure 2.9 and explained in Section 4.3.2.

CHAPTER 2 — Appendix B: Tier-Based Contingency Planning 82

Appendix B: Tier-Based Contingency Planning

Table 2.B1: Tier-based response strategies for contingency planning

Element Tier 1 Tier 2 Tier 3

Res

po

nse

Str

ate

gie

s

General operational procedures

Secure human health and safety

Mobilize and deploy resources onsite

Recover normal business operations as soon as practically possible


Mobilize and deploy resources onsite and in the local vicinity of storage site

Activate responsibility zones among local stakeholders

Provide external support against local health, economic, or environmental damages



Mobilize and deploy resources onsite, and in local and regional vicinity of storage site

Activate responsibility zones among local and regional stakeholders

Provide external support against local and regional health, economic, or environmental damages


Specific operational procedures

Apply corrective measures to control (stop or contain) and remediate leakage



CHAPTER 2 — Appendix B: Tier-Based Contingency Planning 83

Table 2.B2: Tier-based human and equipment resources for contingency planning Element Tier 1 Tier 2 Tier 3

Hu

ma

n a

nd

Eq

uip

men

t R

eso

urc

es

Human resources

Categorize into general and specific response teams

Inventory internal and contracted response personnel

Conduct regular training programs, focusing on achieving, testing, and validating suitable competence (not only awareness or knowledge) to perform the designated role



Identify liaison personnel to external local response teams and stakeholders


Conduct joint training sessions with local stakeholders



Identify liaison personnel to external local and regional response teams and stakeholders


Conduct joint training sessions with local and regional stakeholders

Equipment resources

Inventory owned and contracted, general and specialized equipment

Test and maintain equipment regularly to ensure proper operations

Ensure equipment storage is optimal for easy mobilization


Inventory the exact location and mobilization time of general and specific equipment owned or operated by local stakeholders




Inventory the exact location and mobilization time of general and specific equipment owned or operated by local and regional stakeholders



84

Chapter 3

Economic Value of Flexible Hydrogen-

Based Polygeneration Energy Systems

1 Introduction

Fossil fuels meet 87% of today’s global energy demand [1] and are used to generate 68% of

the global electricity supply [2]. Concerns over climate change, growing energy consumption,

and energy security compel fossil-fuel plants to meet increasing regulatory and market

challenges: lower emissions, higher efficiency, and more flexible operations to complement

intermittent renewables and hedge against fluctuations in energy prices. Polygeneration energy

systems (PES) have the potential to meet all these challenges.

While polygeneration generally describes a wide range of multi-input multi-output industrial

processes [3], this study focuses on polygeneration energy systems that use fossil fuels as

inputs and produce hydrogen as an intermediate product [4]. PES offers multiple advantages

over conventional single-output or monogeneration systems. Technically, polygeneration

allows better process- and heat-integration among various production and ancillary units,

which reduces energy losses and thus results in higher energy-conversion efficiency. This

higher efficiency, combined with the utilization of carbon in chemical synthesis, results in

lower carbon dioxide (CO2) emissions [5, 6]. In addition, the production rates of PES can be

either fixed or adjusted over time. We refer to a system with fixed production rates as static or


Acronyms

AGRU acid-gas removal unit

ASU air separation unit

CCS carbon capture and storage

CO2 carbon dioxide

COE cost of energy

HECA Hydrogen Energy California

HSU hydrogen separation unit

MRU mercury removal unit

NPV net present value

PES polygeneration energy system

PRU particulate removal unit

SRU shift-reaction unit

𝑐𝑙 cost of capacity per one unit of output 𝑙 ($/𝑘𝑊ℎ or $/𝑘𝑔𝑙)

𝐶𝐹 capacity factor

𝐶𝑀𝑙𝑘 contribution margin from converting one kilogram of hydrogen to output 𝑙 in

year 𝑘 ($/𝑘𝑔ℎ)

𝐼𝐶𝑀𝐹𝑙𝑘 incremental contribution margin from flexible switching of hydrogen conversion to

output 𝑙 in year 𝑘 ($/𝑘𝑔ℎ)

𝑗𝑙 time-averaged fixed operating cost per one unit of output 𝑙 ($/𝑘𝑊ℎ or $/𝑘𝑔𝑙)

LCOE levelized cost of electricity ($/𝑘𝑊ℎ)

LCOH levelized cost of hydrogen ($/𝑘𝑔ℎ)

LCOP levelized cost of polygeneration ($/𝑘𝑔ℎ)

𝐿𝐼𝐶𝑙 levelized incremental cost of the subsystem producing output 𝑙 ($/𝑘𝑊ℎ or $/𝑘𝑔𝑙)

𝑚 total number of hours in one year (ℎ𝑟)

𝑚𝑙 yearly hours during which production rate of output 𝑙 should be maximized (ℎ𝑟)

𝑁ℎ production capacity of the hydrogen subsystem (𝑘𝑔ℎ/ℎ𝑟)

𝑃𝑙 price of output 𝑙 ($/𝑘𝑊ℎ or $/𝑘𝑔𝑙)

𝑃𝑀 profit-margin of polygeneration system per one kilogram of produced hydrogen ($/𝑘𝑔ℎ)

𝑆𝑎 storage capacity of ammonia (𝑘𝑔𝑎)

𝑆𝐽𝑙𝑘 fixed operating cost per unit-capacity of output 𝑙 in year 𝑘

(($ 𝑦𝑟⁄ )/𝑘𝑊 or ($ 𝑦𝑟⁄ )/(𝑘𝑔𝑙 ℎ𝑟⁄ ))

𝑆𝑃𝑙 system price; cost per unit-capacity of output 𝑙 (in $/𝑘𝑊 or $/(𝑘𝑔𝑙 ℎ𝑟⁄ ))

𝑇 useful lifetime of the facility (𝑦𝑟)

𝑈𝑐 net CO2 production rate per one kilogram of produced hydrogen (𝑘𝑊ℎ/𝑘𝑔ℎ)

𝑉𝑂𝐷 value of diversification ($/𝑘𝑔ℎ)

𝑉𝑂𝐹 value of flexibility ($/𝑘𝑔ℎ)

𝑉𝑂𝑃 value of polygeneration ($/𝑘𝑔ℎ)

𝑤𝑙 time-averaged variable cost per one unit of output 𝑙 ($/𝑘𝑊ℎ or $/𝑘𝑔𝑙)

𝑥𝑘 degradation factor; the percentage of initial capacity that is still functional at year 𝑘

𝑋𝑙 conversion rate of one kilogram of hydrogen to output 𝑙 (in 𝑘𝑊ℎ/𝑘𝑔ℎ or 𝑘𝑔𝑙/𝑘𝑔ℎ)

𝑦𝑙 fraction of hydrogen allocated to the production of fertilizer 𝑙


Subscripts

𝑎 ammonia without storage

𝑎𝑠 ammonia with storage

𝑐 carbon dioxide

𝑒 electricity

𝑓 fertilizer

ℎ hydrogen

𝑢𝑟𝑒𝑎 urea

𝑈𝐴𝑁 urea and ammonium nitrate solution

𝑚𝑖𝑛 minimum

𝑚𝑎𝑥 maximum

Greek Symbols

𝜏 discount rate

𝛾𝑘 discount factor in year 𝑘

𝜆 fraction of hydrogen production capacity allocated to electricity generation

1 − 𝜆 fraction of hydrogen production capacity allocated to fertilizer generation

𝐾 constant-equivalent fraction of hydrogen capacity allocated for fertilizer generation

𝛷𝑙 correction factor to account for time-dependent variable costs during 𝑚𝑙

steady-state polygeneration and a system with variable production rates as flexible or

dispatchable polygeneration [7]. Flexible polygeneration can exploit frequent variations in

commodity prices; while fuel switching and mixing capabilities help attenuate the risks of

fuel-price shocks, production diversification and dispatchability help capture the benefits of

product-price peaks [7, 8]. Furthermore, merchant hydrogen markets are currently

underdeveloped [9, 10, 11], which renders hydrogen market prices an imperfect indicator of

cost and value. By converting hydrogen to valuable commodities, polygeneration offers an

incentive to expand investments in hydrogen infrastructure.

The advantages of polygeneration systems merit a rigorous analysis of their economic

competitiveness within the broader energy landscape. In this study, we develop a set of

generalizable metrics that can be used to valuate fossil-fuel polygeneration energy systems.

These economic metrics achieve three goals. First, they calculate the levelized cost and

profitability of both static and flexible polygeneration, irrespective of the type of used fossil-

fuels or generated end-products. Second, they facilitate a consistent comparison of the

economics of polygeneration relative to that of monogeneration, with special emphasis on

electricity monogeneration alternatives (e.g. natural gas or wind). Finally, they quantify the


value of two real-options enabled by polygeneration: the value of diversifying end-products

and the value of flexibly varying the production rates of end-products over time.

The main motivation for our analysis stems from the fact that different methodologies have

been used to evaluate polygeneration economics, including net present value [7, 8, 12, 13],

profit index [12], payout time [14], cost of energy [15, 16], and others [17, 18, 19]. While each

methodology has its own merits, the lack of methodological consistency prevents accurate

comparison of polygeneration economics under different technical assumptions and

operational settings. The economic metrics we propose offer one way to overcome this

problem. Specifically, we express all metrics in monetary value per unit of hydrogen

produced, for hydrogen is a common intermediate product across polygeneration energy

systems of various process configurations and end-product portfolios.

While some previous studies have used the cost of energy (COE) ($ 𝑘𝑊ℎ⁄ ) to compare

polygeneration to monogeneration, such an approach faces the following challenges. First,

polygeneration may not necessarily generate electricity as an end-product, in which case the

use of COE becomes impractical. Second, it is problematic to calculate the COE as the cost of

polygeneration less the cost of other non-electricity products from equivalent monogeneration

[15, 16]; this method assigns all cost-savings from polygeneration system-integration to the

power unit and therefore might underestimate the actual cost of electricity. Our approach

addresses this issue by converting the cost per unit of hydrogen to a cost per unit of any end-

product, assuming that all hydrogen is converted to that single end-product. This methodology

facilitates an economic comparison between polygeneration and monogeneration systems,

including traditional power plants.

In addition, an assessment of the economic competitiveness of flexible polygeneration systems

should include a quantification of the economic trade-offs associated with operational

flexibility. Greater flexibility typically implies not only higher revenues but also higher cost of

capacity due to larger equipment size [7, 8]. We address this topic by deriving metrics that

capture the economic impacts of flexible polygeneration, illustrating that production

diversification and flexibility need not always result in economic gains [7].

CHAPTER 3 — Research Methodology 88

On the technical side, there is an extensive literature on optimizing the design and operation of

PES by combining several technologies and processes [6, 13], incorporating investment

planning procedures [20], or investigating the trade-offs associated with operational flexibility

[7]. Other studies also performed detailed techno-economic analyses on specific

polygeneration systems under various input- and output-portfolios [5, 8, 14, 18, 19] and

process configurations [16, 17]. Given the breadth of the available technical analysis, our

work presumes that polygeneration is technically feasible and focuses predominantly on

assessing its economic value. To that end, we use a simple yet generalizable PES

configuration that can operate as both a static and a flexible system. Building specifically on

the work by Chen et al. [7], which optimizes PES operations under uniform levels of

flexibility, we impose different flexibility limits on different production units to explore the

effect of real-life operational constraints on PES economics.

In the following sections, we first introduce the economic concepts and technical

configuration used in assessing PES. Next, we present a detailed economic analysis for the

modeled PES in three scenarios; Scenario 1 evaluates static operations while Scenarios 2a and

2b evaluate two modes of flexible operations. As the main focus of this paper, the economic

definitions and derived propositions in all three scenarios are then used to calculate the profit-

margin and real-options of PES. Lastly, we demonstrate the applicability of the derived

metrics by examining the economic competitiveness of Hydrogen Energy California, a

polygeneration project currently under development.

2 Research Methodology

This section describes the economic concepts and technical specifications used in deriving the

valuation metrics for PES. We first introduce the levelized cost of hydrogen concept, which is

the foundational tool for economic assessment. Then, we explain the process configuration,

fuel-inputs, and product-outputs of the adopted fossil-fuel polygeneration system.


2.1 Levelized Cost of Hydrogen

Similar to the concept of levelized cost of electricity (𝐿𝐶𝑂𝐸) as cost per unit of energy

generation ($ 𝑘𝑊ℎ⁄ ), the metric of levelized cost of hydrogen (𝐿𝐶𝑂𝐻) refers to the cost per

unit of hydrogen production ($ 𝑘𝑔ℎ⁄ ) [21, 22]. Consistent with MIT’s The Future of Coal

definition of 𝐿𝐶𝑂𝐸 [23], this study defines 𝐿𝐶𝑂𝐻 as: “the constant dollar hydrogen price that

would be required over the life of a hydrogen plant to cover all operating expenses, payment

of debt and accrued interest on initial project expenses, and the payment of an acceptable

return to investors”. In other words, the 𝐿𝐶𝑂𝐻 is a break-even metric that calculates the ratio

of lifetime cost to lifetime hydrogen production of a facility.

The 𝐿𝐶𝑂𝐻 formulation adopted in this study is similar to the 𝐿𝐶𝑂𝐸 model in Reichelstein and

Yorston [24]. As shown in (1), the 𝐿𝐶𝑂𝐻 ($ 𝑘𝑔ℎ⁄ ) is the sum of three terms: cost of capacity

per unit output 𝑐ℎ, time-averaged fixed operating cost per unit output 𝑗ℎ, and time-averaged

variable cost per unit output 𝑤ℎ [24, 25].1

𝐿𝐶𝑂𝐻 = 𝑐ℎ + 𝑗ℎ + 𝑤ℎ (1)

Assuming constant returns-to-scale, the cost of capacity per one kilogram of hydrogen can be

expressed as:

𝑐ℎ ($ 𝑘𝑔ℎ⁄ ) =𝑆𝑃ℎ

𝑚 ∙ 𝐶𝐹 ∙ ∑ 𝑥𝑖 ∙ 𝛾𝑖𝑇𝑖=1

(2)

𝑆𝑃 ($ (𝑘𝑔ℎ ℎ𝑟⁄ )⁄ ) in (2) denotes the system price of acquiring one unit of capacity to produce

one kilogram of hydrogen per hour. It includes the cost of engineering procurement and

construction, contingency, and land purchase. The initial investment yields a stream of

hydrogen output over 𝑇 years, with 𝑚 ∙ 𝑥𝑖 ∙ 𝐶𝐹 kilograms delivered in year 𝑖. While 𝑚 =

8,760 refers to the total number of hours in a given year, the system degradation factor, 𝑥𝑖,

accounts for potential losses in generation capacity over time and is technology-specific. In

1 The cost of capacity 𝑐ℎ should be scaled by a factor that accounts for corporate income taxes, as

explained by Reichelstein and Yorston [24]. Our study can be expanded to include the effect of taxes

and subsidies.


addition, since the facility may not be online at all times, the practical capacity is only a

fraction of the theoretical capacity. This fraction is represented by the capacity factor, 𝐶𝐹.

Furthermore, since the 𝐿𝐶𝑂𝐻 is a break-even formula, it is essential to specify an appropriate

discount rate. We denote this discount rate by 𝜏 and the corresponding discount factor by

𝛾 = (1 1 + 𝜏⁄ ).2

Fixed operating costs can change on a yearly basis. The time-averaged fixed operating cost per

one kilogram of hydrogen 𝑗ℎ ($ 𝑘𝑔ℎ⁄ ) is shown in (3). 𝑆𝐽𝑖 (($ 𝑦𝑟⁄ ) (𝑘𝑔ℎ ℎ𝑟⁄ )⁄ ) denotes the

fixed operating cost incurred by operating one-kilogram-per-hour capacity of the hydrogen

facility in year 𝑖. Expenditures in this category include labor, administration and overhead,

maintenance, and insurance. While fixed operating costs are assumed to scale proportionally

with the capacity of the facility, they are independent of the actual amount of hydrogen

generated by the facility.

𝑗ℎ =∑ 𝑆𝐽ℎ𝑖 ∙ 𝛾𝑖𝑇

𝑖=1


(3)

Finally, variable costs can vary over time. Costs in this category include fuel consumption,

raw-material inputs, auxiliary loads, and cash-conversion expenses. We use 𝑤ℎ𝑖(𝑡) ($ 𝑘𝑔ℎ⁄ )

to denote the time-dependent variable cost per one kilogram of hydrogen in year 𝑖, and we

derive 𝑤ℎ𝑖 ($ 𝑘𝑔ℎ⁄ ) in (4) as the yearly-averaged variable cost.

𝑤ℎ𝑖 =1

𝑚∫ 𝑤ℎ𝑖(𝑡)𝑑𝑡

𝑚

0

(4)

Over the life cycle of the facility, the time-averaged variable cost per one kilogram of

hydrogen 𝑤ℎ ($ 𝑘𝑔ℎ⁄ ) becomes as expressed in (5).

𝑤ℎ =∑ 𝑤ℎ𝑖 ∙ 𝑚 ∙ 𝐶𝐹 ∙ 𝑥𝑖 ∙ 𝛾𝑖𝑇

𝑖=1


=∑ 𝑤ℎ𝑖 ∙ 𝑥𝑖 ∙ 𝛾𝑖𝑇

𝑖=1

∑ 𝑥𝑖 ∙ 𝛾𝑖𝑇𝑖=1

(5)

2 The discount rate we specify should be interpreted as a real, rather than a nominal, interest rate that

accounts for inflation.


Similar notation is used to characterize the time-dependency of all economic metrics in this

study, including prices. For instance, 𝑃ℎ𝑖(𝑡), 𝑃ℎ𝑖, and 𝑃ℎ refer to the time-dependent price of

hydrogen in year 𝑖, the yearly-averaged price of hydrogen in year 𝑖, and the time-averaged

price of hydrogen, respectively. Referring back to the definition of 𝐿𝐶𝑂𝐻 as a break-even

value, we can obtain the following benchmark result:

A hydrogen production facility is cost-competitive if and only if:

𝑃ℎ > 𝐿𝐶𝑂𝐻 (6)

Cost-competitiveness is defined as the ability of the facility to achieve a positive NPV. Since a

PES produces hydrogen as an intermediate product only, the price of hydrogen 𝑃ℎ must be

substituted with revenue from the hydrogen-enabled end-products, which are commodities

with well-defined market prices. Therefore, the formulation in (6) needs to be expanded to

assess the economic value of polygeneration.

2.2 Technical Configuration of PES

This study analyzes a simple yet generalizable fossil-fuel PES configuration, which can

operate as either a static or a flexible system. Specifically, we consider a PES that uses coal as

fuel, produces hydrogen then ammonia as intermediate products, and produces electricity and

fertilizer (e.g. urea) as final end-products. In the assumed configuration, coal can be also

mixed with biomass or petcoke as fuel inputs.

In a static PES, all units run at steady-state with fixed output flowrates. In a flexible PES,

however, some units can vary their output flowrates over time while other units should run at

steady-state with fixed flowrates in order to maintain acceptable energy- and chemical-

conversation efficiencies [26]. To account for these real-life operational constraints, and to

allow for a generalizable economic assessment, the adopted PES can be conceptually divided

into four subsystems: hydrogen, electricity, ammonia, and fertilizer. While the electricity and

ammonia subsystems can be either flexible or static, the hydrogen and fertilizer subsystems

are always static.


Figure 3.1 presents a simplified depiction of the process flow sheet for the polygeneration

facility. As shown, coal is first fed into a gasifier where it is mixed with an oxygen (O2)

stream from the air separation unit (ASU) to produce hydrogen-rich syngas. In addition to the

gasifier and the ASU, the hydrogen production subsystem includes: syngas clean-up units

such as particulate removal unit (PRU), mercury removal unit (MRU), and acid-gas removal

unit (AGRU); shift-reaction unit (SRU); and hydrogen separation unit (HSU). All these units

should run at steady-state [26], resulting in a fixed output of hydrogen (H2) and carbon dioxide

(CO2).

Figure 3.1: Simplified process flow sheet of the used PES

Hydrogen is either fed into a combined-cycle turbine unit for electricity generation or mixed

with nitrogen (N2) from the ASU to produce ammonia (NH3), the precursor material for

making fertilizers. Both electricity generation [27, 28] and ammonia synthesis [26] units can

operate flexibly, producing variable power and ammonia flows. Ammonia is then mixed with

a fraction of the CO2 stream to produce fertilizer (e.g. urea). The fertilizer synthesis unit must

run at steady-state [26], resulting in a fixed flow of the end-product. The remaining CO2

gasifier

ASU

syngas

clean-

up

HSU

ammonia

synthesis

fertilizer

synthesis

gas

turbine

steam

turbine

ammonia

storage

air

coal

O2

syngas

(CO +

H2)

steam (H2O)

CO2

H2

CO2

CO2

H2

H2

H2

NH3

electricityhydrogen subsystem

fertilizers

subsystem

electricity

subsystem

vented or

used for CCS

constant flowrate

variable flowrate

N2N2

vented N2

fertilizers

(e.g. urea)

NH3

static units

flexible units

SRU ammonia

subsystem

CHAPTER 3 — Economic Analysis 93

stream is either vented or compressed and transported for geologic sequestration. Finally,

since the steady-state production of fertilizer is dependent on the variable production of

ammonia, intermediate ammonia storage is necessary to buffer the variations in the ammonia

output and secure a fixed ammonia input into the fertilizer subsystem.

Focusing specifically on the ammonia and electricity subsystems, their ranges of flexibility are

constrained by system-integration and efficiency-related standards. For example, part of the

generated electricity is used for auxiliary load within the facility [26], and power turbines must

operate above minimum-capacity limits to avoid severe losses in energy efficiency [28]. In

addition, bounding the range of ammonia production is necessary to cap the size ammonia

storage; the needed intermediate storage becomes increasingly larger as the range of variable

production rates becomes wider. We account for these practical flexibility constraints by

imposing a lower bound on the production rates of both electricity and ammonia. The upper-

bound on production rates is imposed by the name-plate capacity of each production unit.

3 Economic Analysis

Based on the preceding technical configuration, we investigate two scenarios, illustrated in

Figure 3.2. Scenario 1 analyzes a static PES, whereas Scenario 2a and Scenario 2b analyze a

flexible PES. In all scenarios, the PES can still be characterized as a combination of the four

operational subsystems introduced above, with the rate of total hydrogen production fixed at

𝑁ℎ (𝑘𝑔ℎ/ℎ𝑟). For simplicity, we set 𝑁ℎ = 1 in Figure 3.2. For each unit of hydrogen

produced, 𝜆 fraction is allocated to electricity production, and the remaining (1 − 𝜆) is

allocated to ammonia production. While 𝜆 is constant in Scenario 1, 𝜆 may vary with time in

Scenarios 2a and 2b. By definition, one kilogram of hydrogen can be converted to either 𝑋𝑒

kilowatt hours of electricity or 𝑋𝑎 kilograms of ammonia. The subsequent reaction of 𝑋𝑎

kilograms of ammonia with CO2 results in 𝑋𝑓 kilograms of fertilizer; thus, 𝑋𝑎 𝑋𝑓⁄ units of

ammonia are needed to produce one unit of fertilizer.

To account for flexibility constraints, we limit the feasible range of 𝜆, such that 𝜆 ∈

[𝜆𝑚𝑖𝑛 , 𝜆𝑚𝑎𝑥]. 𝜆𝑚𝑖𝑛 is dictated by the minimum allowable rate of electricity generation,


which necessitates that 𝜆 > 𝜆𝑚𝑖𝑛 > 0. On the other hand, 𝜆𝑚𝑎𝑥 is dictated by the minimum

allowable rate of fertilizer generation, which necessitates that (1 − 𝜆) > (1 − 𝜆𝑚𝑎𝑥) > 0, or

equivalently, 𝜆 < 𝜆𝑚𝑎𝑥 < 1. Finally, 𝐾 is a constant parameter related to buffering ammonia

production for fertilizer synthesis, to be formally introduced in Scenario 2b.

Figure 3.2: Schematic representation of static and flexible PES

To reflect current market conditions, the analysis in all scenarios assumes that the price of

electricity changes on an hourly basis [29] whereas the price of fertilizer changes on a yearly

basis [30]. The underlying assumption is that electricity is typically sold in competitive

wholesale markets, but fertilizers can be sold through long-term contracts because they can be

stored relatively easily. Also, to simplify the economic modeling, all fixed and variable costs

are assumed to remain constant in a given year, but they may change across years. Section 5.2

discusses the implications of this assumption in more detail.

3.1 Scenario 1: Static PES with Fixed Production Rates

If the polygeneration system is static, all subsystems run at steady-state with constant

production rates. The ammonia stream is directly and completely converted to fertilizer

Hydrogen

Electricity

Ammonia Fertilizers

Ammonia Storage

Hydrogen

Electricity

Scenario 1:

Static PES

Ammonia Fertilizers

Hydrogen

Electricity

Scenario 2a:

Flexible PES

with flexible

fertilizers

subsystemAmmonia Fertilizers

Scenario 2b:

Flexible PES

with static

fertilizers

subsystem


without the need for an intermediate storage. For every kilogram of hydrogen, the PES outputs

𝜆 ∙ 𝑋𝑒 kilowatt hours of electricity, (1 − 𝜆) ∙ 𝑋𝑎 kilograms of ammonia, and (1 − 𝜆) ∙ 𝑋𝑓

kilograms of fertilizer. We refer to 𝜆 ∙ 𝑋𝑒, (1 − 𝜆) ∙ 𝑋𝑎, and (1 − 𝜆) ∙ 𝑋𝑓 as production

coefficients.

The economic value of this static PES must account for the generation of electricity and

fertilizer at each point in time. Therefore, the expression in (6) is modified by substituting the

revenue from direct sales of merchant hydrogen with the net revenue from converting

hydrogen to both end-products.

For each unit of produced electricity, the net revenue is the difference between the time-

averaged price 𝑃𝑒 ($ 𝑘𝑊ℎ⁄ ) and the levelized incremental cost of installing and operating the

electricity subsystem, defined as 𝐿𝐼𝐶𝑒 ($ 𝑘𝑊ℎ⁄ ). 𝐿𝐼𝐶𝑒 is distinguished from 𝐿𝐶𝑂𝐸. 𝐿𝐼𝐶𝑒

captures the cost of the electricity subsystem only (e.g. combined-cycle turbine). In contrast,

𝐿𝐶𝑂𝐸 accounts for the full cost of electricity generation, which includes the cost of hydrogen.

Therefore, if 𝜆 = 1, we obtain 𝐿𝐶𝑂𝐸 = 𝐿𝐶𝑂𝐻 𝑋𝑒⁄ + 𝐿𝐼𝐶𝑒.

Similarly, for each unit of produced fertilizer, the net revenue is the difference between the

time-averaged price 𝑃𝑓 ($ 𝑘𝑔ℎ⁄ ) and the levelized incremental cost of both the ammonia and

fertilizer subsystems, defined as 𝐿𝐼𝐶𝑎 ($ 𝑘𝑔𝑎⁄ ) and 𝐿𝐼𝐶𝑓 ($ 𝑘𝑔𝑓⁄ ), respectively. Referring to

Section 2.1, Definition 1 presents each 𝐿𝐼𝐶 metric as the sum of three levelized cost

components: a cost of capacity 𝑐, a time-averaged fixed operating cost 𝑗, and a time-averaged

variable cost 𝑤.

Definition 1:

𝐿𝐼𝐶𝑒 = 𝑐𝑒 + 𝑗𝑒 + 𝑤𝑒 (7)

𝐿𝐼𝐶𝑎 = 𝑐𝑎 + 𝑗𝑎 + 𝑤𝑎 (8)

𝐿𝐼𝐶𝑓 = 𝑐𝑓 + 𝑗𝑓 + 𝑤𝑓 (9)

As a result, it is now possible to assess the economic feasibility of this static PES by

formulating Proposition 1.


Proposition 1:

A static polygeneration facility is cost-competitive if and only if:

λ ∙ 𝑋𝑒 ∙ (𝑃𝑒 − 𝐿𝐼𝐶𝑒) + (1 − λ) ∙ (𝑋𝑓 ∙ 𝑃𝑓 − 𝑋𝑓 ∙ 𝐿𝐼𝐶𝑓 − 𝑋𝑎 ∙ 𝐿𝐼𝐶𝑎) > 𝐿𝐶𝑂𝐻 (10)

As proven in the derivation of Proposition 1 in Appendix A, the unit revenue for hydrogen in

(6), 𝑃ℎ, is replaced with two net-revenue terms, one for each end-product: (𝑋𝑒 ∙ 𝑃𝑒 −

𝑋𝑒 ∙ 𝐿𝐼𝐶𝑒) corresponding to the net revenue from hydrogen conversion to electricity and

(𝑋𝑓 ∙ 𝑃𝑓 − 𝑋𝑓 ∙ 𝐿𝐼𝐶𝑓 − 𝑋𝑎 ∙ 𝐿𝐼𝐶𝑎) corresponding to the net revenue from hydrogen conversion

to ammonia then fertilizer. The net revenue from each end-product is weighted by the fraction

of hydrogen capacity allocated to it: λ for electricity and (1 − λ) for fertilizer.

Proposition 1 provides several insights. For the static PES to break even, the prices of end-

products must be high enough to compensate not only for their incremental cost but also for

the cost of hydrogen. Furthermore, optimizing the economic value of the static PES requires

maximizing hydrogen allocation to the end-product contributing the highest net revenue.

Ultimately, this incentivizes setting λ = 1 when (𝑋𝑒 ∙ 𝑃𝑒 − 𝑋𝑒 ∙ 𝐿𝐼𝐶𝑒) ≥ (𝑋𝑓 ∙ 𝑃𝑓 − 𝑋𝑓 ∙ 𝐿𝐼𝐶𝑓 −

𝑋𝑎 ∙ 𝐿𝐼𝐶𝑎) and setting λ = 0 otherwise; in both cases, static PES reduces to static

monogeneration of either end-product. Therefore, Proposition 1 shows that the profitability of

a static PES with multiple end-products is bounded by the profitability of the static

monogeneration of its individual end-products.

To elaborate further on the economics of polygeneration, we introduce the levelized cost of

polygeneration, or 𝐿𝐶𝑂𝑃 ($ 𝑘𝑔ℎ⁄ ). Consistent with the earlier definition of 𝐿𝐶𝑂𝐻, we define

𝐿𝐶𝑂𝑃 in (11) as a weighted-average price of polygeneration end-products that would set the

NPV of the PES to exactly zero.

Definition 2:

𝐿𝐶𝑂𝑃 = 𝐿𝐶𝑂𝐻 + λ ∙ 𝑋𝑒 ∙ 𝐿𝐼𝐶𝑒 + (1 − λ) ∙ [𝑋𝑓 ∙ 𝐿𝐼𝐶𝑓 + 𝑋𝑎 ∙ 𝐿𝐼𝐶𝑎] (11)


Proposition 1 can then be re-arranged to incorporate the mathematical form of 𝐿𝐶𝑂𝑃 in

Definition 2.

Proposition 1’:

A static polygeneration facility is cost-competitive if and only if:

λ ∙ 𝑋𝑒 ∙ 𝑃𝑒 + (1 − λ) ∙ 𝑋𝑓 ∙ 𝑃𝑓 > 𝐿𝐶𝑂𝑃 (12)

The 𝐿𝐶𝑂𝑃 formulation in (11) shows that the levelized cost of a static PES can be expressed

as the sum of the levelized cost of its operational subsystems weighted by their respective

production coefficients. While the levelized costs of individual subsystems can be calculated

using multiple units, (e.g. $/𝑘𝑊ℎ for 𝐿𝐼𝐶𝑒), the conversion rates 𝑋𝑒, 𝑋𝑎, and 𝑋𝑓 ensure that

the overall PES cost is expressed as a monetary value per unit of hydrogen. Clearly, this

approach facilitates comparing the cost of different polygeneration systems with different

configurations, all of which produce hydrogen as an intermediate product.

Furthermore, while (6) shows that the value of monogeneration is dictated by only one price,

Proposition 1’ in (12) shows that the value of polygeneration is determined by a sum of end-

product prices weighted by their respective production coefficients. Consequently, for a fixed

operation mode and thus fixed set of production coefficients, multiple combinations of end-

product prices may achieve break-even. In imperfectly competitive markets, the PES firm can

negotiate multiple portfolios of end-products prices with potential buyers. For instance, the

firm may sell electricity at a competitive market price while having pricing power in selling

fertilizers due to constrained regional supply. Alternatively, in perfectly competitive markets

with preset prices, break-even may be achieved by adjusting production coefficients on both

sides of (12) because 𝜆 is a controllable design parameter. In short, a polygeneration facility

can break even via multiple portfolios of end-product prices and production capacities.

3.2 Scenario 2: Flexible PES with Variable Production Rates

For a PES in flexible mode, both electricity and ammonia generation rates can vary on an

hourly basis; decreasing the power output results in increasing the ammonia output, and vice


versa. While a constant hydrogen generation capacity 𝑁ℎ is maintained, the fraction of

hydrogen converted to electricity and ammonia may vary with time. We use the notation 𝜆(𝑡)

in Figure 3.2 to highlight this fact. Still, due to flexibility constraints, the condition that

𝜆𝑚𝑖𝑛 < 𝜆(𝑡) < 𝜆𝑚𝑎𝑥 remains in place. When 𝜆(𝑡) = 𝜆𝑚𝑎𝑥, electricity production is

maximized while fertilizer production is minimized. Conversely, when 𝜆(𝑡) = 𝜆𝑚𝑖𝑛,

electricity production is minimized while fertilizer production is maximized.

The flexible PES is analyzed sequentially in Scenarios 2a and 2b below. Scenario 2a makes

the simplifying assumption that the fertilizer subsystem can run flexibly, so the variable

fertilizer output is perfectly synchronized with the variable ammonia output. In Scenario 2b,

we acknowledge the real-world need for a static fertilizer subsystem. In other words, Scenario

2a presents a hypothetical operational configuration that aims to benchmark the performance

of the more realistic Scenario 2b. Comparing these two scenarios provides insight into the role

of technical constraints in controlling the economic value of flexible polygeneration.

3.2.1. Scenario 2a: Flexible PES with a Flexible Fertilizer Subsystem

As illustrated in Figure 3.2, a variable fertilizer output results from the direct conversion of the

variable ammonia output at each time interval 𝑡, so no intermediate ammonia storage is

needed in this case. Nonetheless, to accommodate the maximum possible flowrates, the

production capacity of the flexible units should be set at 𝜆𝑚𝑎𝑥 ∙ 𝑋𝑒 (𝑘𝑊) for electricity,

(1 − 𝜆𝑚𝑖𝑛) ∙ 𝑋𝑎 (𝑘𝑔𝑎 ℎ𝑟⁄ ) for ammonia, and (1 − 𝜆𝑚𝑖𝑛) ∙ 𝑋𝑓 (𝑘𝑔𝑓 ℎ𝑟⁄ ) for fertilizer.3 Similar

to Scenario 1, the incurred capacity and fixed operating costs are scaled by these constant

production capacities, independent of the variable production rates.

Since both capacity and fixed operating costs are constant in a given year 𝑖, maximizing the

profitability of the flexible PES requires maximizing the contribution margin of hydrogen

3 Our analysis speaks to the cost competitiveness of a flexible PES whose capacity is chosen to

accommodate the maximum possible flowrate of ammonia. In future work, it would be desirable to

explore under what conditions a flexible PES, accommodating a lower flowrate but requiring

correspondingly smaller capacity investments, could be more economical.


conversion at every point in time 𝑡 of that year.4 Definition 3 introduces 𝐶𝑀𝑒 ($ 𝑘𝑔ℎ⁄ ) and

𝐶𝑀𝑓 ($ 𝑘𝑔ℎ⁄ ) as the contribution margins associated with converting one kilogram of

hydrogen to 𝑋𝑒 kilowatt hours of electricity and 𝑋𝑓 kilograms of fertilizer, respectively.

Definition 3:

𝐶𝑀𝑒𝑖(𝑡) = [𝑋𝑒 ∙ 𝑃𝑒𝑖(𝑡) − 𝑋𝑒 ∙ 𝑤𝑒𝑖(𝑡)] (13)

𝐶𝑀𝑓𝑖(𝑡) = [𝑋𝑓 ∙ 𝑃𝑓𝑖(𝑡) − 𝑋𝑓 ∙ 𝑤𝑓𝑖(𝑡) − 𝑋𝑎 ∙ 𝑤𝑎𝑖(𝑡)] (14)

When 𝐶𝑀𝑒𝑖(𝑡) > 𝐶𝑀𝑓𝑖(𝑡), electricity production should be maximized and fertilizer

production should be minimized; the opposite must hold when 𝐶𝑀𝑒𝑖(𝑡) < 𝐶𝑀𝑓𝑖(𝑡).

Accordingly, we divide the yearly hours 𝑚 into 𝑚𝑒 and 𝑚𝑓, introduced in Definition 4. 𝑚𝑒𝑖

(ℎ𝑟) corresponds to the hours in year 𝑖 when 𝐶𝑀𝑒𝑖(𝑡) ≥ 𝐶𝑀𝑓𝑖(𝑡) whereas 𝑚𝑓𝑖 (ℎ𝑟)

corresponds to the hours when 𝐶𝑀𝑒𝑖(𝑡) < 𝐶𝑀𝑓𝑖(𝑡).5 Clearly, 𝑚 = 𝑚𝑓𝑖 + 𝑚𝑒𝑖 for every year 𝑖.

Definition 4:

𝑚𝑒𝑖 = 𝜇({𝑡|0 ≤ 𝑡 ≤ 𝑚, 𝐶𝑀𝑒𝑖(𝑡) ≥ 𝐶𝑀𝑓𝑖(𝑡)}) (15)

𝑚𝑓𝑖 = 𝜇({𝑡|0 ≤ 𝑡 ≤ 𝑚, 𝐶𝑀𝑒𝑖(𝑡) < 𝐶𝑀𝑓𝑖(𝑡)}) (16)

Flexibility enables choosing the highest contribution margin in every time period. Thus, we

define the difference between 𝐶𝑀𝑒𝑖(𝑡) and 𝐶𝑀𝑓𝑖(𝑡) as the incremental contribution margin

of flexibility (𝐼𝐶𝑀𝐹). In a given year 𝑖, 𝐼𝐶𝑀𝐹𝑒𝑖 is the yearly-averaged sum of flexibility-

enabled contribution margin over 𝑚𝑒𝑖 hours, attributed to switching hydrogen allocation from

fertilizer to electricity. Equivalently, 𝐼𝐶𝑀𝐹𝑓𝑖 is the yearly-averaged sum of flexibility-enabled

contribution margin over 𝑚𝑓𝑖 hours, attributed to switching hydrogen allocation from

4 Contribution margin refers to the difference between sales and variable costs. The analysis in

Reichelstein and Sahoo [45] explains a related idea on quantifying the temporal co-variation between

prices and generation capacity of an intermittent power source. 5 𝜇 in (15) and (16) is the Lebesgue Measure over the set of real numbers between 0 and 𝑚 [46].


electricity to fertilizer. 𝐼𝐶𝑀𝐹𝑒𝑖 ($ 𝑘𝑔ℎ⁄ ) and 𝐼𝐶𝑀𝐹𝑓𝑖 ($ 𝑘𝑔ℎ⁄ ) are introduced in Definition 5.

By design, both terms are always positive.

Definition 5:

𝐼𝐶𝑀𝐹𝑒𝑖 =1

𝑚∙ ∫ [𝐶𝑀𝑒𝑖(𝑡) − 𝐶𝑀𝑓𝑖(𝑡)]𝑑𝑡

𝑚𝑒𝑖

(17)

𝐼𝐶𝑀𝐹𝑓𝑖 =1

𝑚∙ ∫[𝐶𝑀𝑓𝑖(𝑡) − 𝐶𝑀𝑒𝑖(𝑡)]𝑑𝑡

𝑚𝑓𝑖

(18)

The formulation of 𝐼𝐶𝑀𝐹 illustrates the beneficial impacts of price volatility on the economics

of flexible PES. Consistent with the earlier assumption that only electricity price 𝑃𝑒𝑖(𝑡)

changes over time, the effect of higher volatility in 𝑃𝑒𝑖(𝑡) is captured in two ways: higher

𝑃𝑒𝑖(𝑡) leads to higher 𝐶𝑀𝑒𝑖(𝑡) during 𝑚𝑒𝑖 hours and therefore higher 𝐼𝐶𝑀𝐹𝑒𝑖, and lower

𝑃𝑒𝑖(𝑡) leads to lower 𝐶𝑀𝑒𝑖(𝑡) during 𝑚𝑓𝑖 hours and therefore higher 𝐼𝐶𝑀𝐹𝑓𝑖. In short, the

higher the price volatility (around the same price average), the higher the incremental

contribution margin of flexibility.

Similar to the time-averaged variable cost in (6), the time-averaged incremental contribution

margins of flexibility 𝐼𝐶𝑀𝐹𝑒 ($ 𝑘𝑔ℎ⁄ ) and 𝐼𝐶𝑀𝐹𝑓 ($ 𝑘𝑔ℎ⁄ ) are derived from 𝐼𝐶𝑀𝐹𝑒𝑖 and

𝐼𝐶𝑀𝐹𝑓𝑖 in (19) and (20), respectively.

𝐼𝐶𝑀𝐹𝑒 =∑ 𝐼𝐶𝑀𝐹𝑒𝑖 ∙ 𝑥𝑖 ∙ 𝛾𝑖𝑇

𝑖=1


(19)

𝐼𝐶𝑀𝐹𝑓𝑖 =∑ 𝐼𝐶𝑀𝐹𝑓𝑖 ∙ 𝑥𝑖 ∙ 𝛾𝑖𝑇

𝑖=1


(20)

As a result, we can now assess the economic feasibility of the flexible PES in this simplified

scenario by formulating two mathematically equivalent statements of Proposition 2a.


Proposition 2a:

A flexible PES is cost-competitive if and only if:

[𝑋𝑓 ∙ 𝑃𝑓 − 𝑋𝑓 ∙ 𝐿𝐼𝐶𝑓 − 𝑋𝑎 ∙ 𝐿𝐼𝐶𝑎]

+λ𝑚𝑎𝑥 ∙ [𝐼𝐶𝑀𝐹𝑒 − 𝑋𝑒 ∙ (𝑐𝑒 + 𝑗𝑒)]

−λ𝑚𝑖𝑛 ∙ [𝐼𝐶𝑀𝐹𝑓 − 𝑋𝑓 ∙ (𝑐𝑓 + 𝑗𝑓) − 𝑋𝑎 ∙ (𝑐𝑎 + 𝑗𝑎)] > 𝐿𝐶𝑂𝐻

(21)

Equivalently, the flexible PES is cost-competitive if and only if:

𝑋𝑒 ∙ [𝑃𝑒 − 𝐿𝐼𝐶𝑒]

+(1 − λ𝑚𝑖𝑛) ∙ [𝐼𝐶𝑀𝐹𝑓 − 𝑋𝑓 ∙ (𝑐𝑓 + 𝑗𝑓) − 𝑋𝑎 ∙ (𝑐𝑎 + 𝑗𝑎)]

−(1 − λ𝑚𝑎𝑥) ∙ [𝐼𝐶𝑀𝐹𝑒 − 𝑋𝑒 ∙ (𝑐𝑒 + 𝑗𝑒)] > 𝐿𝐶𝑂𝐻

(22)

As proven in Appendix A, the left-hand sides in (21) and (22) are identical. The formulation in

(21) benchmarks flexible polygeneration against static fertilizer monogeneration. Specifically,

the net revenue from hydrogen conversion is divided into three terms. As in Scenario 1, the

first term [𝑋𝑓 ∙ 𝑃𝑓 − 𝑋𝑓 ∙ 𝐿𝐼𝐶𝑓 − 𝑋𝑎 ∙ 𝐿𝐼𝐶𝑎] corresponds to the net revenue from the static

monogeneration of fertilizer. The second and third terms correspond to the additional net

revenues from flexibility. λ𝑚𝑎𝑥[𝐼𝐶𝑀𝐹𝑒 − 𝑋𝑒 ∙ (𝑐𝑒 + 𝑗𝑒)] represents the net revenue associated

with flexible switching from fertilizer to electricity; the flexibility-enabled incremental

contribution margin is weighed against the flexibility-required capacity and fixed operating

costs of generating electricity. This gained net revenue is scaled by λ𝑚𝑎𝑥, the maximum

fraction of hydrogen capacity allocated to electricity. On the other hand, electricity generation

cannot drop below a lower limit defined by λ𝑚𝑖𝑛, so the corresponding net revenue associated

with flexible switching from electricity to fertilizer is lost. This net revenue is captured in

λ𝑚𝑖𝑛 ∙ [𝐼𝐶𝑀𝐹𝑓 − 𝑋𝑓 ∙ (𝑐𝑓 + 𝑗𝑓) − 𝑋𝑎 ∙ (𝑐𝑎 + 𝑗𝑎)], which balances the flexibility-enabled

incremental contribution margin against the flexibility-required capacity and fixed operating

costs of generating ammonia then fertilizer.

The formulation in (22) has a similar and symmetric structure to (21), but it benchmarks the

economics of the flexible polygeneration against static electricity monogeneration. In this

case, the gained and lost net-revenue terms associated with flexibility are scaled by (1 −


λ𝑚𝑖𝑛) and (1 − λ𝑚𝑎𝑥), corresponding to the maximum and minimum fractions of hydrogen

that can be converted to fertilizer, respectively.

In both (21) and (22), the gained and lost net-revenue terms associated with flexibility are

positive only when the incremental contribution margin surpasses the capacity and fixed

operating costs. Thus, Proposition 2a shows that adding flexibility to a PES may not result in

superior economic value; the latter depends on the specifications of the investigated facility.

We analyze this dependency in more detail in Section 4. Also, (21) and (22) show that

maximizing the revenue of flexible polygeneration requires exploiting the ability to vary λ(𝑡)

between λ𝑚𝑖𝑛 and λ𝑚𝑎𝑥; setting λ𝑚𝑖𝑛 = λ𝑚𝑎𝑥 = λ reduces Proposition 2a to Proposition 1.

3.2.2. Scenario 2b: Flexible PES with a Static Fertilizer Subsystem

We now analyze a flexible PES with a static fertilizer subsystem. Compared to Scenario 2a,

this operational mode induces two technical updates. First, intermediate ammonia storage is

needed to buffer the variable ammonia output (1 − 𝜆(𝑡)) ∙ 𝑋𝑎 (𝑘𝑔𝑎 ℎ𝑟⁄ ) and secure a fixed

ammonia input 𝐾 ∙ 𝑋𝑎 (𝑘𝑔𝑎 ℎ𝑟⁄ ) into the fertilizer processing unit. Intuitively, ammonia is

withdrawn from storage to compensate for shortage when (1 − 𝜆(𝑡)) < 𝐾, and it is added to

storage to save excess when (1 − 𝜆(𝑡)) > 𝐾. The storage capacity is denoted by Sa (𝑘𝑔𝑎),

and we assume optimized storage operations. Specifically, the site can always accommodate

the addition or withdrawal of ammonia at any rate. Also, all ammonia produced in a given

year is converted to fertilizer, resulting in no annual net-storage. These two conditions

maintain the underlying assumption that 𝑚𝑒𝑖 and 𝑚𝑓𝑖 are the same in every year 𝑖 throughout

the facility’s lifetime.

Second, because the fertilizer output is fixed, no excess capacity is needed to accommodate

variable production. Therefore, the capacity of the fertilizer subsystem is reduced from

(1 − 𝜆𝑚𝑖𝑛) ∙ 𝑋𝑓 to 𝐾 ∙ 𝑋𝑓 (𝑘𝑔𝑓 ℎ𝑟⁄ ). A yearly mass balance on ammonia production allows

defining 𝐾 in terms of 𝜆, as depicted in (23). As explained before, ammonia production is

minimized during 𝑚𝑒 hours but maximized during 𝑚𝑓 hours. Equating the yearly variable


output from the ammonia subsystem to the yearly fixed input into the fertilizer subsystem

results in 𝐾 = [𝑚𝑒 ∙ (1 − 𝜆𝑚𝑎𝑥) + 𝑚𝑓 ∙ (1 − 𝜆𝑚𝑖𝑛)] 𝑚⁄ .

[𝑚𝑒 ∙ (1 − 𝜆𝑚𝑎𝑥) + 𝑚𝑓 ∙ (1 − 𝜆𝑚𝑖𝑛)] ∙ 𝑋𝑓 = 𝑚 ∙ 𝐾 ∙ 𝑋𝑓 (23)

With these modifications, we can assess the economic feasibility of flexible polygeneration in

this scenario by deriving two equivalent statements of the new Proposition 2b.

Proposition 2b:


[𝑋𝑓 ∙ 𝑃𝑓 − 𝑋𝑓 ∙ 𝐿𝐼𝐶𝑓 − 𝑋𝑎 ∙ 𝐿𝐼𝐶𝑎𝑠]


−λ𝑚𝑖𝑛 ∙ [𝐼𝐶𝑀𝐹𝑓 − 𝑋𝑓 ∙ (𝑐𝑓 + 𝑗𝑓) − 𝑋𝑎 ∙ (𝑐𝑎𝑠 + 𝑗𝑎𝑠)]

+[(1 − λ𝑚𝑖𝑛 − 𝐾) ∙ 𝑋𝑓 ∙ (𝑐𝑓 + 𝑗𝑓)] > 𝐿𝐶𝑂𝐻

(24)

Equivalently, the flexible PES is cost-competitive if and only if:

𝑋𝑒 ∙ [𝑃𝑒 − 𝐿𝐼𝐶𝑒]

+(1 − λ𝑚𝑖𝑛) ∙ [𝐼𝐶𝑀𝐹𝑓 − 𝑋𝑓 ∙ (𝑐𝑓 + 𝑗𝑓) − 𝑋𝑎 ∙ (𝑐𝑎𝑠 + 𝑗𝑎𝑠)]

−(1 − λ𝑚𝑎𝑥) ∙ [𝐼𝐶𝑀𝐹𝑒 − 𝑋𝑒 ∙ (𝑐𝑒 + 𝑗𝑒)]

+[(1 − λ𝑚𝑖𝑛 − 𝐾) ∙ 𝑋𝑓 ∙ (𝑐𝑓 + 𝑗𝑓)] > 𝐿𝐶𝑂𝐻

(25)

A detailed derivation of (24) and (25) is presented in Appendix A. The formulations in (24)

and (25) are identical to those in (21) and (22) for Scenario 2a, except for two differences.

First, to account for storage, the levelized-cost terms of ammonia 𝑐𝑎, 𝑗𝑎, and 𝑤𝑎 are updated to

𝑐𝑎𝑠, 𝑗𝑎𝑠, and 𝑤𝑎𝑠, respectively; other metrics incorporating these terms are also updated

accordingly. Second, reducing the fertilizer capacity from (1 − λ𝑚𝑖𝑛) in Scenario 2a to 𝐾 in

Scenario 2b results in a net-revenue gain of [(1 − λ𝑚𝑖𝑛 − 𝐾) ∙ 𝑋𝑓 ∙ (𝑐𝑓 + 𝑗𝑓)], which accounts

for savings in capacity and fixed operating costs. All other economic expressions introduced

in Scenario 2a remain valid here. Notably, the flexible PES still captures the full economic

benefits of flexibility even though the fertilizer subsystem is static. Because storage allows all

CHAPTER 3 — Profitability and Value of Real-Options 104

generated ammonia to be eventually converted to fertilizer, flexible ammonia generation is

sufficient to sustain the economic benefits of flexible fertilizer generation.

1 − λ𝑚𝑖𝑛 − 𝐾 = 1 − λ𝑚𝑖𝑛 − 1 +𝑚𝑓 ∙ λ𝑚𝑖𝑛

𝑚+

𝑚𝑒 ∙ λ𝑚𝑎𝑥

𝑚=

𝑚𝑒 ∙ (λ𝑚𝑎𝑥 − λ𝑚𝑖𝑛)

𝑚 (26)

Equally important, (1 − λ𝑚𝑖𝑛 − 𝐾) is directly proportional to 𝑚𝑒, as shown in (26). A larger

𝑚𝑒 means that the flexible PES spends more time maximizing electricity generation on the

expense of fertilizer generation. In this case, a smaller static fertilizer subsystem with

intermediate ammonia storage (Scenario 2b) may achieve better economics than a larger

flexible fertilizer subsystem with no storage (Scenario 2a), even if the latter is technically

feasible.

4 Profitability and Value of Real-Options

Our findings in Scenarios 1 and 2 show that different operation modes result in different

economic values for PES. Compared to a static single-output plant, a polygeneration plant

offers the option of diversifying the static output (Scenario 1) as well as the option of

substituting part of the static output capacity with flexible capacity (Scenario 2). To quantify

the value of these real-options, we first need to characterize the profitability of PES. The

overall net present value (NPV) associated with investing in the capacity to deliver one

kilogram of hydrogen per hour (𝑁ℎ = 1) is given in (27). 𝑃𝑀 is the profit-margin, which

denotes the difference between the net revenue for one kilogram of hydrogen and its levelized

cost. We use 𝑃𝑀 as a profitability metric to assess PES under different operation modes.

𝑁𝑃𝑉 ($) = 𝑃𝑀 ∙ 𝑚 ∙ 𝐶𝐹 ∙ ∑ 𝛾𝑖 ∙ 𝑥𝑖𝑇𝑖=1 (27)

𝑃𝑀0 measures the unit profit-margin of a static single-output plant. 𝑃𝑀0𝑓 ($/𝑘𝑔ℎ) refers to

the profit-margin of a static plant that converts all hydrogen to ammonia and then fertilizer

(𝜆 = 0). Similarly, 𝑃𝑀0𝑒 ($/𝑘𝑔ℎ) refers to the profit-margin of a static power plant that


converts all hydrogen to electricity (𝜆 = 1). 𝑃𝑀0𝑓 and 𝑃𝑀0𝑒 are formally defined in (28) and

(29), respectively.

𝑃𝑀0𝑓 = 𝑋𝑓 ∙ 𝑃𝑓 − 𝑋𝑓 ∙ 𝐿𝐼𝐶𝑓 − 𝑋𝑎 ∙ 𝐿𝐼𝐶𝑎 − 𝐿𝐶𝑂𝐻 (28)

𝑃𝑀0𝑒 = 𝑋𝑒 ∙ 𝑃𝑒 − 𝑋𝑒 ∙ 𝐿𝐼𝐶𝑒 − 𝐿𝐶𝑂𝐻 (29)

Let 𝑃𝑀1 ($/𝑘𝑔ℎ) denote the unit profit-margin of the static PES in Scenario 1, which can be

directly deduced from Proposition 1. 𝑃𝑀1 is derived in (30) by re-arranging the terms in (10),

𝑃𝑀1 = λ ∙ 𝑋𝑒 ∙ (𝑃𝑒 − 𝐿𝐼𝐶𝑒) + (1 − λ) ∙ (𝑋𝑓 ∙ 𝑃𝑓 − 𝑋𝑓 ∙ 𝐿𝐼𝐶𝑓 − 𝑋𝑎 ∙ 𝐿𝐼𝐶𝑎) − 𝐿𝐶𝑂𝐻 (30)

Similarly, 𝑃𝑀2 ($/𝑘𝑔ℎ) refers to the unit profit-margin of the flexible PES in Scenario 2,

which is obtained directly from Proposition 2b. 𝑃𝑀2 can be expressed in two forms, 𝑃𝑀2𝑓

($/𝑘𝑔ℎ) and 𝑃𝑀2𝑒 ($/𝑘𝑔ℎ), presented in (31) and (32), respectively. 𝑃𝑀2𝑓 is derived from

(24) where the economic value of a flexible PES is benchmarked against that of a static

fertilizer plant, and 𝑃𝑀2𝑒 is derived from (25) where the economic value of a flexible PES is

benchmarked against that of a static power plant. Importantly, we note that 𝑃𝑀2𝑒 = 𝑃𝑀2𝑓.

𝑃𝑀2𝑓 = [𝑋𝑓 ∙ 𝑃𝑓 − 𝑋𝑓 ∙ 𝐿𝐼𝐶𝑓 − 𝑋𝑎 ∙ 𝐿𝐼𝐶𝑎𝑠]



+[(1 − λ𝑚𝑖𝑛 − 𝐾) ∙ 𝑋𝑓 ∙ (𝑐𝑓 + 𝑗𝑓)] − 𝐿𝐶𝑂𝐻

(31)

𝑃𝑀2𝑒 = 𝑋𝑒 ∙ [𝑃𝑒 − 𝐿𝐼𝐶𝑒]

+(1 − λ𝑚𝑖𝑛) ∙ [𝐼𝐶𝑀𝐹𝑓 − 𝑋𝑓 ∙ (𝑐𝑓 + 𝑗𝑓) − 𝑋𝑎 ∙ (𝑐𝑎𝑠 + 𝑗𝑎𝑠)]

−(1 − λ𝑚𝑎𝑥) ∙ [𝐼𝐶𝑀𝐹𝑒 − 𝑋𝑒 ∙ (𝑐𝑒 + 𝑗𝑒)]

+[(1 − λ𝑚𝑖𝑛 − 𝐾) ∙ 𝑋𝑓 ∙ (𝑐𝑓 + 𝑗𝑓)] − 𝐿𝐶𝑂𝐻

(32)

These profitability metrics can be directly used to quantify the value of real-options enabled

by polygeneration. We first define 𝑉𝑂𝐷 as the value of diversification from a single output to

a portfolio of multiple outputs. 𝑉𝑂𝐷𝑓 ($/𝑘𝑔ℎ) is the difference between the profit-margin of

static polygeneration 𝑃𝑀1 and that of static monogeneration of fertilizer 𝑃𝑀0𝑓; similarly,


𝑉𝑂𝐷𝑒 ($/𝑘𝑔ℎ) is the difference between 𝑃𝑀1 and 𝑃𝑀0𝑒. Clearly, both 𝑉𝑂𝐷𝑓 and 𝑉𝑂𝐷𝑒 are

dependent on 𝜆, as shown in (33) and (34), respectively.

𝑉𝑂𝐷𝑓(𝜆) = 𝑃𝑀1(𝜆) − 𝑃𝑀0𝑓 (33)

𝑉𝑂𝐷𝑒(𝜆) = 𝑃𝑀1(𝜆) − 𝑃𝑀0𝑒 (34)

We then define the value of flexibility 𝑉𝑂𝐹 ($/𝑘𝑔ℎ) associated with varying power and

ammonia production rates with time. 𝑉𝑂𝐹 is the difference between the profit-margin of

flexible polygeneration 𝑃𝑀2 and that of static polygeneration 𝑃𝑀1, and it is dependent on 𝜆,

as highlighted in (35). This formulation shows that flexible polygeneration is more profitable

than static polygeneration only if 𝑉𝑂𝐹(𝜆) > 0. There might exist some value of 𝜆 for which a

static PES could outperform a flexible PES, in which case 𝑉𝑂𝐹(𝜆) is negative.

𝑉𝑂𝐹(𝜆) = 𝑃𝑀2𝑓 − 𝑃𝑀1(𝜆) = 𝑃𝑀2𝑒 − 𝑃𝑀1(𝜆) (35)

Overall, we define the value of polygeneration 𝑉𝑂𝑃 ($/𝑘𝑔ℎ) as the sum of the real-option

values associated with both diversification and flexibility. As shown in (36) and (37), one

𝑉𝑂𝑃 metric is needed for each end-product; 𝑉𝑂𝑃𝑓 ($/𝑘𝑔ℎ) compares the profit-margin of a

flexible PES to that of a static fertilizer plant, and 𝑉𝑂𝑃𝑒 ($/𝑘𝑔ℎ) compares the profit-margin

of a flexible PES to that of a static power plant.

𝑉𝑂𝑃𝑓 = 𝑉𝑂𝐹(𝜆) + 𝑉𝑂𝐷𝑓(𝜆) = 𝑃𝑀2𝑓 − 𝑃𝑀0𝑓 (36)

𝑉𝑂𝑃𝑒 = 𝑉𝑂𝐹(𝜆) + 𝑉𝑂𝐷𝑒(𝜆) = 𝑃𝑀2𝑒 − 𝑃𝑀0𝑒 (37)

When 𝑉𝑂𝑃𝑓 > 0, flexible polygeneration is more profitable than static fertilizer

monogeneration. Similarly, when 𝑉𝑂𝑃𝑒 > 0, flexible polygeneration is more profitable than

static power monogeneration. However, we showed in (33) and (34) that static polygeneration

is less profitable than the static monogeneration of at least one end-product. Accordingly,

when both 𝑉𝑂𝑃𝑓 and 𝑉𝑂𝑃𝑒 are positive, flexible polygeneration is more profitable than both

static monogeneration and static polygeneration. In that regard, while 𝑉𝑂𝐹(𝜆) identifies the

condition for a flexible PES to be more profitable than a specific static PES with a specific 𝜆,

CHAPTER 3 — Additional Modeling Considerations 107

𝑉𝑂𝑃 identifies the condition for a flexible PES to be more profitable than any static PES with

any 𝜆.

5 Additional Modeling Considerations

5.1 Carbon Capture and Storage

The proposed economic assessment of PES can be robustly expanded to account for technical

supplements such as carbon capture and storage (CCS). The net CO2 output, defined as the

gross CO2 production less the CO2 used for fertilizer synthesis, may be compressed then

transported in pipelines to be either geologically sequestered or used for enhanced oil recovery

[23]. Since CO2 is produced by a steady-state process (gasification) and partially utilized by

another steady-state process (fertilizer synthesis), its net output is a fixed flow, regardless of

whether the PES is static of flexible.

CCS is treated as a separate subsystem of production capacity 𝑈𝑐 ∙ 𝑁ℎ. (𝑘𝑔𝑐 ℎ𝑟⁄ ), where 𝑈𝑐

(𝑘𝑔𝑐 𝑘𝑔ℎ⁄ ) denotes the net CO2 output rate per one kilogram of produced hydrogen. As

before, 𝐿𝐼𝐶𝑐 refers to the sum of 𝑐𝑐, 𝑗𝑐, and 𝑤𝑐, which correspond to the cost of capacity, time-

averaged fixed operating cost, and time-averaged variable cost of CCS per unit of CO2

($ 𝑘𝑔𝑐⁄ ), respectively. Also, if sold for enhanced oil recovery, the CO2 output generates

revenue proportional to its price 𝑃𝑐 ($ 𝑘𝑔𝑐⁄ ). As such, Propositions 1 and 2b can be revised to

incorporate CCS in a static and a flexible PES, as shown in (38) and (39), respectively. All

other economic metrics and propositions can be updated accordingly.

Proposition 1’’:

A static PES with CCS is cost-competitive if and only if:

λ ∙ 𝑋𝑒 ∙ (𝑃𝑒 − 𝐿𝐼𝐶𝑒) + (1 − λ) ∙ (𝑋𝑓 ∙ 𝑃𝑓 − 𝑋𝑓 ∙ 𝐿𝐼𝐶𝑓 − 𝑋𝑎 ∙ 𝐿𝐼𝐶𝑎)

+𝑈𝑐 ∙ (𝑃𝑐 − 𝐿𝐼𝐶𝑐) > 𝐿𝐶𝑂𝐻 (38)

CHAPTER 3 — Additional Modeling Considerations 108

Proposition 2b’:

A flexible PES with CCS is cost-competitive if and only if:




+[(1 − λ𝑚𝑖𝑛 − 𝐾) ∙ 𝑋𝑓 ∙ (𝑐𝑓 + 𝑗𝑓)] + 𝑈𝑐 ∙ (𝑃𝑐 − 𝐿𝐼𝐶𝑐) > 𝐿𝐶𝑂𝐻

(39)

5.2 Time-dependency of prices and variable costs

So far, we have assumed that, except for the price of electricity, all prices and variable costs

are fixed within a given year. If this assumption is not met, the aforementioned analysis will

still generate the exact same results in Scenario 1 and Scenario 2a, but slightly modified

results in Scenario 2b where fertilizer production is fixed. In this particular case, correction

terms should be added to the formulations of Proposition 2b to account for the different

averaging of the fertilizer contribution margin 𝐶𝑀𝑓𝑖(𝑡) over different time periods.

Specifically, (40) and (41) introduce the two correction terms 𝛷𝑓𝑖 ($/𝑘𝑔ℎ) and 𝛷𝑒𝑖 ($/𝑘𝑔ℎ)

in year 𝑖.

𝛷𝑓𝑖 =1

𝑚[𝑚𝑓𝑖

𝑚∫ [𝐶𝑀𝑓𝑖(𝑡)]𝑑(𝑡)

𝑚

− ∫ [𝐶𝑀𝑓𝑖(𝑡)]𝑑(𝑡)

𝑚𝑓𝑖

] =1

𝑚[𝑚𝑓𝑖𝐶𝑀𝑓𝑖 − ∫ [𝐶𝑀𝑓𝑖(𝑡)]𝑑(𝑡)

𝑚𝑓𝑖

] (40)

𝛷𝑒𝑖 =1

𝑚[𝑚𝑒𝑖

𝑚∫ [𝐶𝑀𝑓𝑖(𝑡)]𝑑(𝑡)

𝑚

− ∫ [𝐶𝑀𝑓𝑖(𝑡)]𝑑(𝑡)

𝑚𝑒𝑖

] =1

𝑚[𝑚𝑒𝑖𝐶𝑀𝑓𝑖 − ∫ [𝐶𝑀𝑓𝑖(𝑡)]𝑑(𝑡)

𝑚𝑒𝑖

] (41)

If all prices and variable costs are allowed to vary with time, Propositions 2b can be easily

revised to incorporate 𝛷𝑓 and 𝛷𝑒, as illustrated in (42).

CHAPTER 3 — Case Study: Hydrogen Energy California 109

Proposition 2b’’:





+[(1 − λ𝑚𝑖𝑛 − 𝐾) ∙ 𝑋𝑓 ∙ (𝑐𝑓 + 𝑗𝑓)]

−[𝜆𝑚𝑎𝑥 ∙ 𝛷𝑒 + 𝜆𝑚𝑖𝑛 ∙ 𝛷𝑓] > 𝐿𝐶𝑂𝐻

(42)

6 Case Study: Hydrogen Energy California

To demonstrate the usefulness of our model analysis in the previous sections, we now assess

the economic performance of Hydrogen Energy California (HECA), a polygeneration facility

currently under development in California.

6.1 Technical Configuration

Consistent with the technical configuration presented in Section 2.2, HECA uses a gasification

technology to convert coal and petcoke into clean-burning hydrogen. As an intermediate

product, hydrogen is then converted to electricity and ammonia, which is further processed

into urea and UAN – a solution of urea and ammonium nitrate [31, 32]. The operational

configuration of HECA allows flexible generation of electricity and ammonia, but it requires

static generation of hydrogen, urea, and UAN; the facility also includes a CO2 compression

unit, which can be treated as a separate static CCS subsystem.

The first task is to quantify all technical parameters needed for the economic evaluation. A list

of HECA’s technical parameters and their values is provided in Table 3.1. With a capacity

factor of 𝐶𝐹 = 0.835 and an expected operational lifetime of 25 years, the facility consumes

coal and petcoke at rates equal to roughly 4,209 𝑡𝑜𝑛𝑛𝑒/𝑑𝑎𝑦 and 1,053 𝑡𝑜𝑛𝑛𝑒/𝑑𝑎𝑦,

respectively [33]. The syngas generated from the gasification of coal and petcoke undergoes

shift-reaction to convert most of the carbon monoxide into carbon dioxide, 90% of which is


captured. A fraction of the captured CO2, corresponding to 𝑈𝑐 = 12.1 𝑘𝑔𝑐 𝑘𝑔ℎ⁄ , is

compressed and sold to nearby oil fields for enhanced oil recovery.

Table 3.1: HECA technical parameters

Parameter Value Unit Reference

𝑇 25 𝑦𝑟 [Assumed]

𝐶𝐹 0.835 [unitless] [34]

𝑚𝑒 5,840 ℎ𝑟 [34]

𝑚𝑓 2,920 ℎ𝑟 [34]

𝐶𝑜𝑎𝑙 𝐼𝑛𝑝𝑢𝑡 4,209 𝑡𝑜𝑛𝑛𝑒/𝑑𝑎𝑦 [33]

𝑃𝑒𝑡𝑐𝑜𝑘𝑒 𝐼𝑛𝑝𝑢𝑡 1,052 𝑡𝑜𝑛𝑛𝑒/𝑑𝑎𝑦 [33]

𝑁ℎ 28,748 𝑘𝑔ℎ/ℎ𝑟 [Calculated]

𝜆𝑚𝑖𝑛 0.521 [unitless] [Calculated]

𝜆𝑚𝑎𝑥 0.717 [unitless] [Calculated]

𝑋𝑒 19.66 𝑘𝑊ℎ 𝑘𝑔ℎ⁄ [Calculated]

𝑋𝑎 5.63 𝑘𝑔𝑎 𝑘𝑔ℎ⁄ [Calculated]

𝑋𝑢𝑟𝑒𝑎 9.93 𝑘𝑔𝑢𝑟𝑒𝑎 𝑘𝑔ℎ⁄ [Calculated]

𝑋𝑈𝐴𝑁 13.72 𝑘𝑔𝑈𝐴𝑁 𝑘𝑔ℎ⁄ [Calculated]

𝑈𝑐 12.10 𝑘𝑔𝑐 𝑘𝑔ℎ⁄ [Calculated]

𝑦𝑢𝑟𝑒𝑎 0.532 [unitless] [Calculated]

𝑦𝑈𝐴𝑁 0.468 [unitless] [Calculated]

𝑆𝑎 9,474,036 𝑘𝑔𝑎 [35]

Produced at a fixed flowrate of 𝑁ℎ = 28,748 𝑘𝑔ℎ/ℎ𝑟, hydrogen is converted to electricity

and ammonia at rates equal to 𝑋𝑒 = 19.66 𝑘𝑊ℎ 𝑘𝑔ℎ⁄ and 𝑋𝑎 = 5.63 𝑘𝑔𝑎 𝑘𝑔ℎ⁄ , respectively.

On a daily basis, the facility operates under two modes: “electricity peak” mode from 7 a.m. to

11 p.m., followed by “electricity off-peak” mode for the rest of the time. During “peak” hours

of electricity demand, the plant runs at maximum power and minimum ammonia generation

capacities, corresponding to 𝜆𝑚𝑎𝑥 = 0.717. Alternatively, during “off-peak” hours, the plant

runs at minimum power and maximum ammonia generation capacities, corresponding to

𝜆𝑚𝑖𝑛 = 0.521 [26]. Summing over one year, this results in 𝑚𝑒 = 5,840 ℎ𝑟 and 𝑚𝑓 = 2,920

ℎ𝑟. Importantly, 𝑚𝑒 and 𝑚𝑓 are exogenously imposed in this case instead of being

endogenously optimized through (15) and (16). This regime will have significant impacts on

the economic value of the facility, as we explain in the next section. Table 3.2 outlines the


auxiliary load requirements for the system under each operation mode [34]. As shown, both

operation modes consume 247248 𝑀𝑊 of the gross power output, which is less than

𝜆𝑚𝑖𝑛 ∙ 𝑋𝑒 ∙ 𝑁ℎ = 295 𝑀𝑊. In reality, HECA continues to generate a positive net power output

to the grid even under the “off-peak” mode [34].

Table 3.2: HECA auxiliary loads

System/Unit Peak

Mode

Off-Peak

Mode Unit Reference

Hydrogen subsystem 170 164 MW [34]

Electricity subsystem 12 12 MW [34]

Ammonia subsystem 10 17 MW [Estimated]

Urea + UAN subsystem 15 15 MW [Estimated]

CO2 subsystem 40 40 MW [34]

Total 247 248 MW [34]

The produced ammonia reacts with a fraction of the captured CO2 to synthesize urea, part of

which is then further processed with more ammonia to produce UAN. Because hydrogen is

effectively processed into two fertilizer end-products, every unit of hydrogen allocated to

fertilizer synthesis is split into two fractions: 𝑦𝑢𝑟𝑒𝑎 = 0.532 for urea and 𝑦𝑈𝐴𝑁 = 0.468 for

UAN. We then define 𝑋𝑢𝑟𝑒𝑎 = 9.93 𝑘𝑔𝑢𝑟𝑒𝑎 𝑘𝑔ℎ⁄ and 𝑋𝑈𝐴𝑁 = 13.72 𝑘𝑔𝑈𝐴𝑁 𝑘𝑔ℎ⁄ as the

conversion rates of hydrogen to urea and UAN, respectively. Finally, to buffer the variable

ammonia output and secure a constant input for urea and UAN synthesis, ammonia storage is

needed. The storage capacity is 𝑆𝑎 = 9,474,036 𝑘𝑔𝑎, equivalent to 7 days of full loading at a

rate of 𝐾 ∙ 𝑋𝑎 ∙ 𝑁ℎ = 56,393 𝑘𝑔𝑎 ℎ𝑟⁄ [35].

6.2 Economic Analysis

6.2.1. Cost and Revenue

The cost figures for the hydrogen, electricity, and CCS subsystems are based on a study by the

International Energy Agency that analyzes the economics of coal gasification for co-

production of electricity and hydrogen [36]. The cost of CCS in this case covers CO2

compression only, assuming other parties are responsible for CO2 transportation and


sequestration. In addition, the costs associated with ammonia production, ammonia storage,

urea production, and UAN production, are based on studies by Bartels [37], Leigh [38] and

Morgan [39], Lennon [40], and Damas [41], respectively. Furthermore, because the urea and

UAN units are static, we combine them into one “fertilizer” subsystem. This approach allows

us to directly use the economic metrics derived in Sections 3 and 4. For convenience, the costs

of this joint fertilizer subsystem are expressed per unit of produced urea, so the definition of

𝑋𝑓 and 𝐿𝐼𝐶𝑓 should be updated according to (43) and (44), respectively. All monetary figures

are adjusted to 2012 U.S. dollars, assuming a 1.33 conversion factor from Euro to U.S. dollar

when needed. Finally, as mentioned in Section 2.1, taxes are not accounted for in this analysis.

𝑋𝑓 = 𝑦𝑢𝑟𝑒𝑎 ∙ 𝑋𝑢𝑟𝑒𝑎 (43)

𝐿𝐼𝐶𝑓 ($/𝑘𝑔𝑢𝑟𝑒𝑎) =𝑦𝑢𝑟𝑒𝑎 ∙ 𝑋𝑢𝑟𝑒𝑎 ∙ 𝐿𝐼𝐶𝑢𝑟𝑒𝑎 + 𝑦𝑈𝐴𝑁 ∙ 𝑋𝑈𝐴𝑁 ∙ 𝐿𝐼𝐶𝑈𝐴𝑁

𝑋𝑓 (44)

The levelized capacity, fixed operating, and variable costs are presented in Tables 3.3, 3.4, and

3.5, respectively; a more detailed breakdown of the cost figures is provided in Appendix B.

We assume a constant annual discount rate of τ = 0.07 and no degradation in productivity

over the years (𝑥 = 1) for all cost figures.

Table 3.3: Levelized costs of capacity of HECA

Cost Value Unit

𝑐ℎ 0.5267 $/𝑘𝑔ℎ

𝑐𝑒 0.0123 $/𝑘𝑊ℎ

𝑐𝑎 0.0453 $/𝑘𝑔𝑎

𝑐𝑎𝑠 0.0463 $/𝑘𝑔𝑎

𝑐𝑓 0.0520 $/𝑘𝑔𝑢𝑟𝑒𝑎

𝑐𝑐 0.0016 $/𝑘𝑔𝑐

Table 3.3 lists the levelized costs of capacity for the five major subsystems of HECA. Since

the size of HECA is comparable to that of the facilities analyzed in the referenced literature,

linear scaling factors are used to calculate the capacity cost of each subsystem. Also, we recall

that 𝑐𝑎𝑠 and 𝑐𝑎 correspond to the capacity costs of the ammonia subsystem with and without

intermediate storage, respectively.


The yearly fixed operating costs are calculated as a constant fraction of the overall capacity

cost (refer to Appendix B), and they remain unchanged every year throughout the lifetime of

the project. Accordingly, the levelized time-averaged fixed operating costs of HECA’s

subsystems are listed in Table 3.4.

Table 3.4: Levelized time-averaged fixed operating costs of HECA

Cost Value Unit

𝑗ℎ 0.2071 $/𝑘𝑔ℎ

𝑗𝑒 0.0086 $/𝑘𝑊ℎ

𝑗𝑎 0.0320 $/𝑘𝑔𝑎

𝑗𝑎𝑠 0.0326 $/𝑘𝑔𝑎

𝑗𝑓 0.0367 $/𝑘𝑔𝑢𝑟𝑒𝑎

𝑗𝑐 0.0006 $/𝑘𝑔𝑐

The variable cost for the hydrogen subsystem incorporates the costs of coal and petcoke as

fuel, Selexol™, flux, catalysts, other chemicals, waste-water treatment, and the unit’s

auxiliary load. On the other hand, the auxiliary loads are assumed to be the only variable costs

for all other subsystems. The prices of all physical commodities are fixed with time, assuming

they are purchased through long-term contracts (refer to Appendix B). However, we assume

that HECA’s net power output is sold in the wholesale market, and the cost of auxiliary power

equals the price of sold power. Hence, the yearly costs of the auxiliary loads in Table 3.2 are

obtained by summing up the hourly costs, which are calculated using variable electricity

prices. To simulate a real-life performance, we use the 2012 wholesale one-day-ahead

electricity prices from the SP26 pricing hub, which covers the Southern California region

where HECA plans to operate [29]. The yearly price data, plotted in Figure 3.3, is assumed to

be replicated every year throughout the facility’s lifetime. Under these assumptions, the time-

averaged variable cost equals the yearly-averaged variable cost for each subsystem, and those

costs are presented in Table 3.5. Finally, important to note, the variable cost of ammonia

storage is assumed to be negligible, resulting in 𝑤𝑎𝑠 = 𝑤𝑎.


Figure 3.3: Yearly wholesale prices of electricity in HECA’s region [29]

Table 3.5: Levelized time-averaged variable costs of HECA

Cost Value Unit

𝑤ℎ 0.640 $/𝑘𝑔ℎ

𝑤𝑒 0.00119 $/𝑘𝑊ℎ

𝑤𝑎 0.00644 $/𝑘𝑔𝑎

𝑤𝑎𝑠 0.00644 $/𝑘𝑔𝑎

𝑤𝑓 0.00836 $/𝑘𝑔𝑢𝑟𝑒𝑎

𝑤𝑐 0.00339 $/𝑘𝑔𝑐

The last important set of economic data is the prices of end-products, which account for

HECA’s revenues. The revenues from both fertilizers, urea and UAN, are combined in 𝑃𝑓.

This “price of fertilizers” term, expressed per unit of produced urea, is derived in (45) using a

similar formulation to that of 𝐿𝐼𝐶𝑓 in (43). In addition to fertilizers, Table 3.6 shows the time-

averaged prices of electricity and CO2 sales. More detailed price figures are provided in

Appendix B.

𝑃𝑓 ($/𝑘𝑔𝑢𝑟𝑒𝑎) =𝑦𝑢𝑟𝑒𝑎 ∙ 𝑋𝑢𝑟𝑒𝑎 ∙ 𝑃𝑢𝑟𝑒𝑎 + 𝑦𝑈𝐴𝑁 ∙ 𝑋𝑈𝐴𝑁 ∙ 𝑃𝑈𝐴𝑁

𝑋𝑓 (45)


Table 3.6: Time-averaged prices of HECA end-products

Price Value Unit

𝑃𝑓 0.768 $/𝑘𝑔𝑢𝑟𝑒𝑎

𝑃𝑒 0.0295 $/𝑘𝑊ℎ

𝑃𝑐 0.025 $/𝑘𝑔𝑐

6.2.2. Economic Value

The aforementioned data allows us to calculate the derived metrics in Sections 3 and 4 and

therefore to assess the economic value of HECA under several operation modes. The results

are presented in Table 3.7.

Table 3.7: Economic valuation of HECA

Economic Metric Value Unit

𝐿𝐶𝑂𝐻 1.373 $/𝑘𝑔ℎ

𝐿𝐶𝑂𝐸 0.0953 $/𝑘𝑊ℎ

𝛷𝑒 −0.000354 $/𝑘𝑔ℎ

𝛷𝑓 −0.0105 $/𝑘𝑔ℎ

𝐼𝐶𝑀𝐹𝑓 1.196 $/𝑘𝑔ℎ

𝐼𝐶𝑀𝐹𝑒 −2.223 $/𝑘𝑔ℎ

𝑃𝑀0𝑓 1.934 $/𝑘𝑔ℎ

𝑃𝑀0𝑒 −0.992 $/𝑘𝑔ℎ

𝑃𝑀1 (𝑓𝑜𝑟 𝜆 = 𝜆𝑚𝑎𝑥) −0.163 $/𝑘𝑔ℎ

𝑃𝑀2𝑓 = 𝑃𝑀2𝑒 −0.0439 $/𝑘𝑔ℎ

𝑉𝑂𝐷𝑓 (𝑓𝑜𝑟 𝜆 = 𝜆𝑚𝑎𝑥) −2.097 $/𝑘𝑔ℎ

𝑉𝑂𝐷𝑒 (𝑓𝑜𝑟 𝜆 = 𝜆𝑚𝑎𝑥) 0.829 $/𝑘𝑔ℎ

𝑉𝑂𝐹 (𝑓𝑜𝑟 𝜆 = 𝜆𝑚𝑎𝑥) 0.119 $/𝑘𝑔ℎ

𝑉𝑂𝑃𝑓 −1.978 $/𝑘𝑔ℎ

𝑉𝑂𝑃𝑒 0.948 $/𝑘𝑔ℎ

The levelized cost of hydrogen production is estimated at 𝐿𝐶𝑂𝐻 = 1.373 $/𝑘𝑔ℎ. This cost

can be combined with the cost of the electricity and CCS subsystems to calculate an 𝐿𝐶𝑂𝐸 for


HECA, as illustrated in (46). Notably, the obtained 𝐿𝐶𝑂𝐸 = 0.0953 $/𝑘𝑊ℎ is comparable to

that of coal power plants with CO2 capture, currently estimated at about 0.089– 0.139 $/𝑘𝑊ℎ

[42, 43, 44].

𝐿𝐶𝑂𝐸 = 𝐿𝐶𝑂𝐻 𝑋𝑒⁄ + 𝐿𝐼𝐶𝑒 + 𝑈𝑐 ∙ 𝐿𝐼𝐶𝑐 𝑋𝑒⁄ (46)

To calculate HECA’s unit profit-margin, we use the profitability metrics from Section 4, with

a few updates. Starting with the static mode of operation, 𝑃𝑀0𝑓, 𝑃𝑀0𝑒, and 𝑃𝑀1 are updated

in accordance with (38) in Section 5.1 to account for the CCS subsystem. With 𝑃𝑀0𝑓 = 1.934

$/𝑘𝑔ℎ, HECA is obviously profitable if run as a static fertilizer-only plant. However, the

facility would not break even if run as a static power-only plant, with 𝑃𝑀0𝑒 = −0.992 $/𝑘𝑔ℎ.

The profit-margin of the static polygeneration mode 𝑃𝑀1 is between 𝑃𝑀0𝑓 and 𝑃𝑀0𝑒; the

exact value of 𝑃𝑀1 changes with the hydrogen allocation fraction 𝜆, as illustrated in Figure

3.4. Under assumed prices and costs, a static HECA breaks-even around 𝜆 = 0.66.

Confirming our argument in Section 4, the value of diversification is not always positive. In

this case, diversifying away from power monogeneration increases profitability, evident by the

positive 𝑉𝑂𝐷𝑒. Conversely, diversifying away from fertilizer monogeneration severely

reduces profitability, evident by the negative 𝑉𝑂𝐷𝑓. Ultimately, increasing electricity

generation reduces both profitability and the associated values of diversification, as illustrated

in Figure 3.4.

Shifting to the flexible polygeneration mode, 𝑃𝑀2𝑒 and 𝑃𝑀2𝑓 are updated per Sections 5.1

and 5.2 to account for the CCS subsystem and the correction factors for the time-dependent

variable costs, respectively. 𝛷𝑒 and 𝛷𝑓 in Table 3.7 correct for the fact that the variable costs

change on an hourly basis due to HECA subsystems’ need for auxiliary power. In addition,

although 𝑚𝑒 and 𝑚𝑓 are exogenously imposed rather than endogenously optimized through

(15) and (16), the incremental flexibility contribution margins 𝐼𝐶𝑀𝐹𝑒 and 𝐼𝐶𝑀𝐹𝑓 are

calculated by following their definitions in (19) and (20). Referring to Table 3.7, 𝐼𝐶𝑀𝐹𝑓 is

clearly positive because the contribution margin from fertilizers exceeds that from electricity

during 𝑚𝑓 hours. However, 𝐼𝐶𝑀𝐹𝑒 is negative, contrary to our assertion in Section 3.2 that it

should also be positive. Caused by the exogeneity of 𝑚𝑒 and 𝑚𝑓, this result essentially means


that urea and UAN generate higher revenue than electricity even during 𝑚𝑒 hours when

electricity prices are highest. Therefore, flexible power generation may seem like a poor line

of business.

Figure 3.4: Profit-margin, value of diversification, and value of flexibility for HECA

However, flexibility enables a two-way substitution, so a flexible electricity capacity requires

an equivalent flexible fertilizer capacity. For HECA, while a flexible power capacity may not

be beneficial because electricity prices are relatively low, flexible fertilizer capacity is indeed

beneficial for the exact same reason. This is better understood by looking at 𝑃𝑀2 and the

corresponding 𝑉𝑂𝐹. The profit-margin of a flexible HECA is 𝑃𝑀2𝑒 = 𝑃𝑀2𝑓 = −0.0439

$/𝑘𝑔ℎ, so the facility almost breaks-even. As shown in Figure 3.4, a flexible HECA can be

less or more profitable than a static HECA, depending on the exact value of 𝜆 for the latter.

For a small 𝜆, the static HECA is dominated by fertilizer generation, so adding flexibility

leads to switching from high-price fertilizers to low-price electricity during the exogenously

imposed 𝑚𝑒. In this case, flexibility is not useful, and 𝑃𝑀2 is lower than 𝑃𝑀1, evident by the

negative 𝑉𝑂𝐹. Conversely, when 𝜆 is large, the static mode of HECA is dominated by

electricity generation, so adding flexibility leads to switching from low-price electricity to

-3

-2

-1

0

1

2

3

0 0.2 0.4 0.6 0.8 1

Ec

on

om

ic v

alu

e (

$/k

gh

)

λ

PR1 VODe

PR2 VODf

VOF


high-price fertilizers during 𝑚𝑓. In this case, flexibility is useful, and 𝑃𝑀2 is higher than 𝑃𝑀1,

evident by the positive 𝑉𝑂𝐹.

Ultimately, 𝑉𝑂𝐹 converges to 𝑉𝑂𝑃 at either extreme value of 𝜆. When 𝜆 = 0, 𝑉𝑂𝐹 is at its

minimum value and equal to 𝑉𝑂𝑃𝑓. Conversely, when 𝜆 = 1, 𝑉𝑂𝐹 is at its maximum value

and equal to 𝑉𝑂𝑃𝑒. We conclude that HECA benefits from flexible polygeneration if the

company’s other feasible alternative is investing in a static power-only plant, but it does not

benefit from flexible polygeneration if the other feasible alternative is investing in a static

fertilizer-only plant.

Figure 3.5: Value of polygeneration for flexible HECA under optimal operations

For completeness, we briefly analyze HECA’s performance under a hypothetical optimal

operational schedule, where 𝑚𝑒 and 𝑚𝑓 are obtained endogenously. Under assumed prices and

costs, we find that 𝑚𝑒 = 0 and 𝑚𝑓 = 𝑚, suggesting – as expected – that the facility should

run as a static fertilizer-only plant. Increasing electricity prices, nonetheless, leads to a

different conclusion, as illustrated in Figure 3.5. First, we proportionally increase all prices of

electricity depicted in Figure 3.3, which increases the average price 𝑃𝑒 while preserving the

relative volatility. As electricity prices increase, 𝑚𝑒 increases, signifying the economic

0

500

1000

1500

2000

2500

3000

3500

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

400 450 500 550 600

me

(h

r)

VO

P (

$/k

gh

)

Relative change in Pe (%)

VOPf

VOPe

me


favorability of installing flexible capacity and switching to electricity generation. When 𝑃𝑒 is

550−574% times its current value, both 𝑉𝑂𝑃𝑒 and 𝑉𝑂𝑃𝑓 are positive; in this case, flexible

polygeneration becomes the most profitable alternative for HECA, better than all static

polygeneration or monogeneration alternatives.

6.2.3. Sensitivity Analysis

Figure 3.6: Sensitivity analysis on the profitability of flexible HECA

Since HECA is a first-of-a-kind facility, it seems particularly important to check the

sensitivity of our results. Specifically, we analyze the sensitivity of HECA’s profitability to

the following variables: price of fertilizers, price of electricity, price of CO2, and discount rate.

Figure 3.6 shows that the profit-margin of a flexible HECA 𝑃𝑀2 is highly sensitive to both the

fertilizers price 𝑃𝑓 and the discount rate 𝜏. In fact, the facility can break even upon modest

increase in 𝑃𝑓 beyond 3.1% or upon modest decrease in 𝜏 beyond 7.5%. Conversely, HECA’s

profit-margin seems to be less sensitive to changes in CO2 price 𝑃𝑐 and least sensitive to

changes in electricity prices, characterized by 𝑃𝑒. To achieve break-even, 𝑃𝑐 would need to

increase by more than 14.5%, whereas 𝑃𝑒 would need to increase by more 34%. The

-2

-1.6

-1.2

-0.8

-0.4

0

0.4

0.8

1.2

1.6

2

-100% -75% -50% -25% 0% 25% 50% 75% 100%

PM

2 (

$/k

gh

)

Relative change in input (%)

Pe

Pf

Pc

τ


aforementioned prices and discount rate affect the unit profit-margin of a static HECA 𝑃𝑀1 in

a very similar manner.

7 Conclusions

The levelized cost of electricity is an important economic concept that can be expanded to

assess the economic value of hydrogen-based polygeneration energy systems (PES). In this

study, we derive a set of metrics that quantify the cost, profitability, and value of real-options

associated with fossil-fuel PES. Because a PES can be divided into a distinct set of operational

subsystems, we first define the levelized cost of hydrogen (𝐿𝐶𝑂𝐻) and the levelized

incremental cost (𝐿𝐼𝐶) of converting hydrogen to market commodities such as electricity and

fertilizers. All cost figures can be combined into one term, the levelized cost of polygeneration

(𝐿𝐶𝑂𝑃), expressed as a monetary value per unit of produced hydrogen ($ 𝑘𝑔ℎ⁄ ). Given that

polygeneration systems share hydrogen as an intermediate product, this approach allows a

systematic comparison of polygeneration costs under multiple technical configurations and

operation modes.

By adding end-products’ sales, we derive the optimal unit profit-margin of PES under two

operation modes: static production of electricity and fertilizer (𝑃𝑀1), and flexible production

of electricity and fertilizer (𝑃𝑀2). We then compare both metrics to the profit-margin of static

monogeneration of electricity or fertilizer (𝑃𝑀0). The difference between 𝑃𝑀1 and 𝑃𝑀0 is

coined as the value of diversification (𝑉𝑂𝐷), and it captures the economic trade-offs

associated with allocating hydrogen to multiple end-product units. Similarly, the difference

between 𝑃𝑀2 and 𝑃𝑀1 is coined as the value of flexibility (𝑉𝑂𝐹), and it captures the

economic trade-offs associated with varying hydrogen allocation to each end-product unit over

time. 𝑉𝑂𝐹 and 𝑉𝑂𝐷 can be combined into one term, referred to as the value of polygeneration

(𝑉𝑂𝑃). Also, we demonstrate how to update these metrics to assess PES with carbon capture

and storage (CCS).

Through a series of derived economic propositions, we show that static polygeneration is more

profitable than static monogeneration if 𝑉𝑂𝐷 is positive. Similarly, flexible polygeneration is


more profitable than static polygeneration if 𝑉𝑂𝐹 is positive, and flexible polygeneration is

more profitable than static monognerration if 𝑉𝑂𝑃 is positive. Notably, however, 𝑉𝑂𝐷, 𝑉𝑂𝐹,

and 𝑉𝑂𝑃 need not always be positive because of the aforementioned economic trade-offs. As

such, no specific operation mode is unconditionally superior; the relative competitiveness of

static monogeneration, static polygeneration, and flexible polygeneration is highly dependent

on the assumed commodity prices and investment costs.

Applying the aforementioned economic metrics to a real polygeneration project, Hydrogen

Energy California, reveals their practical significance. Given a set of technical and financial

assumptions, HECA proves to be profitable as a static fertilizer-only plant with 𝑃𝑀0𝑓 = 1.934

$ 𝑘𝑔ℎ⁄ . However, with 𝑃𝑀0𝑒 = −0.992 $ 𝑘𝑔ℎ⁄ , HECA fails to break even as a static

electricity-only plant although its cost at 𝐿𝐶𝑂𝐸 = 0.0953 $ 𝑘𝑊ℎ⁄ is comparable to coal

power plants with carbon capture. As a static PES, HECA’s profit-margin 𝑃𝑀1 is between

𝑃𝑀0𝑓 and 𝑃𝑀0𝑒, with the exact value dependent on the exact splitting of produced hydrogen

between the two end-products. As a flexible PES, HECA almost succeeds to break even, with

𝑃𝑀2 = −0.0439 $ 𝑘𝑔ℎ⁄ . In this case study, the flexible polygeneration is unequivocally

superior to all other operation modes only if electricity prices increase 5.5-5.74 folds under an

endogenously optimized operational schedule.

7.1 Future Work

Moving forward, several opportunities still exist to expand this work. One potential area of

research involves analyzing the economic value of polygeneration systems powered by

renewable energy. While producing the same end-products (e.g. electricity and fertilizer),

renewable polygeneration might differ from fossil-fuel polygeneration in two major ways,

namely, hydrogen production and the need for CCS. Multiple technologies are available to

produce hydrogen in renewable polygeneration, including biomass gasification and water

electrolysis powered by solar PV and wind turbines. Interestingly, in this case, a separate

source of carbon would be needed to synthesize chemicals, and renewable PES may result in

negative emissions if combined with CCS. In addition, hydrogen can be thought of as a form

of energy-storage if it were used to regenerate electricity.


Furthermore, it would be rather important to examine the effect of taxes, including tax

subsidies, on polygeneration. Such endeavor requires a more careful analysis of investment

tax credits, effective corporate income tax rates, and accelerated depreciation rates that could

be applicable to both fossil-fuel and renewable polygeneration. Finally, this work assumes

deterministic commodity prices of inputs and outputs as well as known energy policies. A

more realistic approach is to consider uncertain market prices then optimize the operational

schedule of PES as prices vary with time. A similar approach can be followed to incorporate

uncertain environmental regulations in the form of future carbon pricing.


References

[1] BP, "BP Statistical Review of World Energy," June 2014. [Online]. Available:

http://www.bp.com/en/global/corporate/about-bp/energy-economics/statistical-review-of-world-

energy.html.

[2] IEA, "2014 Key World Energy Statistics," 2014. [Online]. Available:

http://www.iea.org/publications/freepublications/publication/KeyWorld2014.pdf.

[3] L. M. Serra, M.-A. Lozano, J. Ramos, A. V. Ensinas and S. A. Nebra, "Polygeneration and

efficient use of natural resources," Energy, vol. 34, pp. 575-586, 2009.

[4] Nexant, "Polygeneration from Coal: Integrated Power, Chemicals, and Liquid Fuels," Nexant

Chem Systems, White Plains, New York, 2008.

[5] J. Meerman, A. Ramírez, W. Turkenburg and A. Faaij, "Performance of simulated flexible

integrated gasification polygeneration facilities. Part A: A technical-energetic assessment,"

Renewable and Sustainable Energy Reviews, vol. 15, pp. 2563-2587, 2011.

[6] P. Liu, E. N. Pistikopoulos and Z. Li, "A Multi-Objective Optimization Approach to

Polygeneration Energy Systems Design," AIChE Journal: Process Systems Engineering, pp. Vol.

56, No. 5, 2010.

[7] Y. Chen, T. A. Adams II and a. P. I. Barton, "Optimal Design and Operation of Flexible Energy

Polygeneration Systems," Ind. Eng. Chem. Res., vol. 50, pp. 4553-4566, 2011.

[8] J. Meerman, A. Ramirez, W. Turkenburg and A. Faaij, "Performance of simulated flexible

integrated gasification polygeneration facilities, Part B: Economice valuation," Renewable and

Sustainable Energy Reviews, vol. 16, pp. 6083-6102, 2012.

[9] H2Stations, "Hydrogen Filling Stations Worldwide," 2015. [Online]. Available:

http://www.netinform.net/H2/H2Stations/H2Stations.aspx?Continent=AF&StationID=-1.

[10] D. R. Baker, "Hydrogen-fueled cars face uncertain market in California," 2014. [Online].

Available: http://www.sfgate.com/news/article/Hydrogen-fueled-cars-face-uncertain-market-in-

5519890.php. [Accessed 2015].

[11] G. Bromaghim, K. Gibeault, J. Serfass, P. Serfass and E. Wagner, "Hydrogen and Fuel Cells: The

U.S. Market Report," National Hydrogen Association, Washington DC, USA, 2010.

[12] F. Calise, A. Cipollina, M. D. d’Accadia and A. Piacentino, "A novel renewable polygeneration

system for a small Mediterranean volcanic island for the combined production of energy and

water: Dynamic simulation and economic assessment," Applied Energy, vol. 135, pp. 675-693,

2014.


[13] C. Rubio-Maya, J. Uche-Marcuello, A. Martínez-Gracia and A. A. Bayod-Rújula, "Design

optimization of a polygeneration plant fuelled by natural gas and renewable energy sources,"

Applied Energy, vol. 88, pp. 449-457, 2011.

[14] L. A. Pellegrini, G. Soave, S. Gamba and S. Langè, "Economic analysis of a combined energy–

methanol production plant," Applied Energy, vol. 88, pp. 4891-4897, 2011.

[15] S. Li, L. Gao, X. Zhang, H. Lin and H. Jin, "Evaluation of cost reduction potential for a coal based

polygeneration system with CO2 capture," Energy, vol. 45, pp. 101-106, 2012.

[16] L. Hu, J. Hongguang, G. Lin and H. Wei, "Techno-economic evaluation of coal-based

polygeneration systems of synthetic," Energy Conversion and Management, vol. 52, pp. 274-283,

2011.

[17] A. Narvaez, D. Chadwick and L. Kershenbaum, "Small-medium scale polygeneration systems:

Methanol and power production," Applied Energy, vol. 113, pp. 1109-1117, 2014.

[18] K. S. Ng, N. Zhang and J. Sadhukhan, "Techno-economic analysis of polygeneration systems with

carbon capture and storage and CO2 reuse," Chemical Engineering Journal, vol. 219, pp. 96-108,

2013.

[19] C.-C. Cormos, "Assessment of flexible energy vectors poly-generation based on coal and

biomass/solid wastes co-gasification with carbon capture," Internationall journal of hydrogen

energy, pp. 7855-7866, 2013.

[20] P. Liu, D. I. Gerogiorgis and E. N. Pistikopoulos, "Modeling and optimization of polygeneration

energy systems," Catalysis Today, vol. 127, p. 347–359, 2007.

[21] T. Ramsden, M. Ruth, V. Diakov, M. Laffen and T. Timbario, "Hydrogen Pathways: Updated

Cost, Well-to-Wheels Energy Use, and Emissions for the Current Technology Status of Ten

Hydrogen Production, Delivery, and Distribution Scenarios," National Renewable Energy

Laboratory, Golden, Colorado, USA, 2013.

[22] T. Ramsden, D. Steward and J. Zuboy, "Analyzing the Levelized Cost of Centralized and

Distributed Hydrogen Production Using the H2A Production Model, Version 2," National

Renewable Energy Laboratory, Golden, Colorado, USA, 2009.

[23] MIT, "The Future of Coal," Massachusetts Institute of Technology, Boston, 2007.

[24] S. Reichelstein and M. Yorston, "The prospects for cost competitive solar PV power," Energy

Policy, vol. 55, pp. 117-127, 2012.

[25] S. Reichelstein and A. Rohlfing-Bastian, "Levelized Product Cost: Concept and Decision

Relevance," The Accounting Review, vol. 90, no. 4, 2015.

[26] M. Lerdal, "Hydrogen Energy California, LowCarbon Solutions for California," Energy Seminar,

Stanford University, Stanford, California, 2012.


[27] GE Energy, "Heavy duty gas turbine products," 2009. [Online]. Available: http://www.ge-

energy.com/content/multimedia/_files/downloads/GEH12985H.pdf. [Accessed 2014].

[28] J. Wick, "Advanced Gas Turbine Technology GT26," Alstom, Jornada Tecnológica in Madrid,

Madrid, 2006.

[29] CAISO, "Open Access Same-time Information System (OASIS)," 2012. [Online]. Available:

http://oasis.caiso.com.

[30] IndexMundi, "Urea Monthly Price - US Dollars per Metric Ton," 2015. [Online]. Available:

http://www.indexmundi.com/commodities/?commodity=urea&months=120. [Accessed 2015].

[31] HECA, "The Project," 2010a. [Online]. Available: http://hydrogenenergycalifornia.com/the-

project. [Accessed 2014].

[32] HECA, "Project Fact Sheet," 2010b. [Online]. Available:

http://hydrogenenergycalifornia.com/factsheets. [Accessed 2013].

[33] URS, "Responses to CEC Workshop Requests: Nos. A1 through A32. Amended Application for

Certifi cation HYDROGEN ENERGY CALIFORNIA (08-AFC-8A). Kern County, California,"

California Energy Commission, Sacramento, California, 2012.

[34] CEC, "Hydrogen Energy California Project. Preliminary Staff Assessment, Draft Environmental

Impact Statement," U.S. Department of Energy and California Energy Commission, Sacramento,

California, 2013.

[35] URS, "Responses to Sierra Club Data Requests Set Three: Nos. 132 through 146. Amended

Application for Certifi cation for HYDROGEN ENERGY CALIFORNIA (08-AFC-8A) Kern

County, California," California Energy Commission, 2013.

[36] IEA GHG, "Co-Production of Hydrogen and Electricity by Coal Gasification with CO2 Capture -

Updated Economic Analysis," International Energy Agency Greenhouse Gas R&D Programme,

2008.

[37] J. R. Bartels, "A feasibility study of implementing an Ammonia Economy," Iowa State University,

Ames, Iowa, 2008.

[38] B. Leighty, "Energy Storage with Anhydrous Ammonia: Comparison with other Energy Storage,"

in Ammonia: The Key to US Energy Independence, Minneapolis, 2008.

[39] E. R. Morgan, "Techno-Economic Feasibility Study of Ammonia Plants Powered by Offshore

Wind," University of Massachusetts - Amherst, 2013.

[40] D. Lennon, "Refurbishing Used Plants. Relocating Nitrogenous Fertilizer Plants," Capital Plant

International.


[41] C. Damas, "Terra Nitrogen Or CVR Partners: Fertilizer Production Capability," 21 October 2011.

[Online]. Available: http://seekingalpha.com/article/301165-terra-nitrogen-or-cvr-partners-

fertilizer-production-capability.

[42] E. Rubin, G. Booras, J. Davison, C. Ekstrom, M. Matuszewski, S. McCoy and C. Short, "Toward a

Common Method for Cost Estimation for CO2 Capture and Storage at Fossil Fuel Power Plants,"

Global CCS Institute, 2013.

[43] S. Borenstein, "The Private and Public Economics of Renewable Electricity Generation," Journal

of Economic Perspectives, pp. 67-92, 2012.

[44] C. Abellera and C. Short, "The costs of CCS and Other Low-Carbon Technologies," Global CCS

Institute, 2011.

[45] S. Reichelstein and A. Sahoo, "Time of Day Pricing and the Levelized Cost of Intermittent Power

Generation," Energy Economics, vol. 48, pp. 97-108, 2015.

[46] J. E. Marsden, "Elementary Classical Analysis," W. H. Freeman and Co., San Francisco, 1974.

CHAPTER 3 — Appendix A: Derivation of Economic Propositions 127

Appendix A: Derivation of Economic Propositions

Scenario 1: Static PES with fixed production rates

Derivation of Definition 1 and Proposition 1

Below is a step-by-step derivation of Definition 1 and Proposition 1, presented in (7–9) and

(10), respectively. The derivation is based on net present value since the LCOP, as previously

introduced, is a levelized cost figure that equals a weighted-average of end-product prices

such that the NPV of the polygeneration facility is exactly zero.

𝑁𝑃𝑉 ($) = ∑ 𝛾𝑖 ∙ 𝐶𝐹𝐿𝑖

𝑇

𝑖=1

− 𝐶𝑎𝑝𝐸𝑥 (A1)

𝑁𝑃𝑉: net present value ($)

𝛾𝑖: discount factor in year 𝑖

𝐶𝐹𝐿𝑖: cash flow in year 𝑖 ($/𝑦𝑟)

𝐶𝑎𝑝𝐸𝑥: cost of capacity of polygeneration system ($)

𝐶𝑎𝑝𝐸𝑥 = [𝐶𝐴𝑃ℎ + 𝐶𝐴𝑃𝑒 + 𝐶𝐴𝑃𝑎 + 𝐶𝐴𝑃𝑓] (A2)

𝐶𝐴𝑃ℎ: cost of capacity of hydrogen production subsystem ($)

𝐶𝐴𝑃𝑒: cost of capacity of electricity production subsystem ($)

𝐶𝐴𝑃𝑎: cost of capacity of ammonia production subsystem ($)

𝐶𝐴𝑃𝑓: cost of capacity of fertilizers production subsystem ($)

Then, each cost of capacity can be decomposed as follows:

𝐶𝐴𝑃ℎ = 𝑆𝑃ℎ ∙ 𝑁ℎ (A3)

𝐶𝐴𝑃𝑒 = λ ∙ 𝑋𝑒 ∙ 𝑁ℎ ∙ 𝑆𝑃𝑒 (A4)

𝐶𝐴𝑃𝑎 = (1 − λ) ∙ 𝑋𝑎 ∙ 𝑁ℎ ∙ 𝑆𝑃𝑎 (A5)

𝐶𝐴𝑃𝑓 = (1 − λ) ∙ 𝑋𝑓 ∙ 𝑁ℎ ∙ 𝑆𝑃𝑓 (A6)


𝑆𝑃ℎ: system price of hydrogen production per unit capacity ($/(𝑘𝑔ℎ ℎ𝑟⁄ ))

𝑆𝑃𝑒: system price of electricity production per unit capacity ($/𝑘𝑊)

𝑆𝑃𝑓: system price of ammonia production per unit capacity ($/(𝑘𝑔𝑎 ℎ𝑟⁄ ))

𝑆𝑃𝑓: system price of fertilizers production per unit capacity ($/(𝑘𝑔𝑓 ℎ𝑟⁄ ))

And 𝑁ℎ , 𝑋𝑒, 𝑋𝑎, 𝑋𝑓, and λ are as defined before.

Then, by substituting (A3–A6) into (A2):

𝐶𝑎𝑝𝐸𝑥 = 𝑁ℎ ∙ [𝑆𝑃ℎ + λ ∙ 𝑋𝑒 ∙ 𝑆𝑃𝑒 + (1 − λ) ∙ 𝑋𝑎 ∙ 𝑆𝑃𝑎 + (1 − λ) ∙ 𝑋𝑓 ∙ 𝑆𝑃𝑓] (A7)

Now we define the annual cash flow in year 𝑖 as:

𝐶𝐹𝐿𝑖 = −𝐽𝑖

+𝐶𝐹 ∙ 𝑥𝑖 ∙ 𝑁ℎ ∙ ∫ [λ ∙ 𝑋𝑒 ∙ 𝑃𝑒𝑖(𝑡) + (1 − λ) ∙ 𝑋𝑓 ∙ 𝑃𝑓𝑖 − 𝑤ℎ𝑖 − λ ∙ 𝑋𝑒 ∙ 𝑤𝑒𝑖

−(1 − λ) ∙ 𝑋𝑎 ∙ 𝑤𝑎𝑖 − (1 − λ) ∙ 𝑋𝑓 ∙ 𝑤𝑓𝑖] 𝑑𝑡

𝑚

0

(A8)

As before, 𝑚 = 8760 is the total number of hours per year. 𝐶𝐹 and 𝑥𝑖 are the capacity factor

and the system degradation factor, respectively, as defined in the LCOH earlier. Also, as

explained before and illustrated in (A8), the selling price of fertilizers 𝑃𝑓𝑖 is fixed in year 𝑖,

while electricity price 𝑃𝑒𝑖(𝑡) varies on hourly basis. Thus:

∫ [λ ∙ 𝑋𝑒 ∙ 𝑃𝑒𝑖(𝑡) + (1 − λ) ∙ 𝑋𝑓 ∙ 𝑃𝑓𝑖]𝑑𝑡

𝑚

0

= 𝑚 ∙ [λ ∙ 𝑋𝑒 ∙ 𝑃𝑒𝑖 + (1 − λ) ∙ 𝑋𝑓 ∙ 𝑃𝑓𝑖] (A9)

𝑃𝑒𝑖: yearly-averaged price of electricity in year 𝑖 ($/𝑘𝑊ℎ)

𝑃𝑓𝑖: yearly-averaged price of fertilizers in year 𝑖 ($/𝑘𝑔𝑓)

Beside the revenue generated from selling the end-products, 𝑤ℎ𝑖, 𝑤𝑒𝑖, 𝑤𝑎𝑖, and 𝑤𝑓𝑖 refer to the

variable costs of producing one unit of hydrogen, electricity, ammonia, and fertilizers,

respectively. Since all variable costs are assumed constant in a given year, summing over the

all hours of year 𝑖 results:


∫ [𝑤ℎ𝑖 + λ ∙ 𝑋𝑒 ∙ 𝑤𝑒𝑖 + (1 − λ) ∙ 𝑋𝑎 ∙ 𝑤𝑎𝑖 + (1 − λ) ∙ 𝑋𝑓 ∙ 𝑤𝑓𝑖]𝑑𝑡

𝑚

0

= 𝑚 ∙ [𝑤ℎ𝑖 + λ ∙ 𝑋𝑒 ∙ 𝑤𝑒𝑖 + (1 − λ) ∙ 𝑋𝑎 ∙ 𝑤𝑎𝑖 + (1 − λ) ∙ 𝑋𝑓 ∙ 𝑤𝑓𝑖]

(A10)

𝑤ℎ𝑖: yearly-averaged variable cost of hydrogen production per unit output in year 𝑖 (in $/𝑘𝑔ℎ)

𝑤𝑒𝑖: yearly-averaged variable cost of electricity production per unit output in year 𝑖 ($/𝑘𝑊ℎ)

𝑤𝑎𝑖: yearly-averaged variable cost of ammonia production per unit output in year 𝑖 (in $/𝑘𝑔𝑎)

𝑤𝑓𝑖: yearly-averaged variable cost of fertilizers production per unit output in year 𝑖 (in $/𝑘𝑔𝑓)

Similarly, we define the fixed-operating cost in year 𝑖 as:

𝐽𝑖 = 𝐽ℎ𝑖 + 𝐽𝑒𝑖 + 𝐽𝑎𝑖 + 𝐽𝑓𝑖

= 𝑁ℎ ∙ [𝑆𝐽ℎ𝑖 + λ ∙ 𝑋𝑒 ∙ 𝑆𝐽𝑒𝑖 + (1 − λ) ∙ 𝑋𝑎 ∙ 𝑆𝐽𝑎𝑖 + (1 − λ) ∙ 𝑋𝑓 ∙ 𝑆𝐽𝑓𝑖]

(A11)

𝐽𝑖: total annual fixed-operating cost of polygeneration facility in year 𝑖 ($/𝑦𝑟)

𝐽ℎ𝑖: annual fixed-operating cost of hydrogen production subsystem in year 𝑖 ($/𝑦𝑟)

𝐽𝑒𝑖: annual fixed-operating cost of electricity production subsystem in year 𝑖 ($/𝑦𝑟)

𝐽𝑎𝑖: annual fixed-operating cost of ammonia production subsystem in year 𝑖 ($/𝑦𝑟)

𝐽𝑓𝑖: annual fixed-operating cost of fertilizers production subsystem in year 𝑖 ($/𝑦𝑟)

And such that:

𝑆𝐽ℎ𝑖: annual fixed-operating cost of hydrogen production per unit capacity in year 𝑖

(($ 𝑦𝑟⁄ )/(𝑘𝑔ℎ ℎ𝑟⁄ ))

𝑆𝐽𝑒𝑖: annual fixed-operating cost of electricity production per unit capacity in year 𝑖

(($ 𝑦𝑟⁄ )/𝑘𝑊)

𝑆𝐽𝑎𝑖: annual fixed-operating cost of ammonia production per unit capacity in year 𝑖

(($ 𝑦𝑟⁄ )/(𝑘𝑔𝑎 ℎ𝑟⁄ ))

𝑆𝐽𝑓𝑖: annual fixed-operating cost of fertilizers production per unit capacity in year 𝑖

(($ 𝑦𝑟⁄ )/(𝑘𝑔𝑓 ℎ𝑟⁄ ))

Substituting (A9), (A10), and (A11) into (A8):


𝐶𝐹𝐿𝑖 = 𝑚 ∙ 𝐶𝐹 ∙ 𝑥𝑖 ∙ 𝑁ℎ ∙ [ λ ∙ (𝑋𝑒 ∙ 𝑃𝑒𝑖 − 𝑋𝑒 ∙ 𝑤𝑒𝑖) − 𝑤ℎ𝑖

+(1 − λ) ∙ (𝑋𝑓 ∙ 𝑃𝑓𝑖 − 𝑋𝑓 ∙ 𝑤𝑓𝑖 − 𝑋𝑎 ∙ 𝑤𝑎𝑖)]

−𝑁ℎ ∙ [𝑆𝐽ℎ𝑖 + λ ∙ 𝑋𝑒 ∙ 𝑆𝐽𝑒𝑖 + (1 − λ) ∙ 𝑋𝑎 ∙ 𝑆𝐽𝑎𝑖 + (1 − λ) ∙ 𝑋𝑓 ∙ 𝑆𝐽𝑓𝑖] (A12)

Then, the NPV becomes:

𝑁𝑃𝑉 ($) = ∑ 𝛾𝑖 ∙ 𝑚 ∙ 𝐶𝐹 ∙ 𝑥𝑖 ∙ 𝑁ℎ ∙ [ λ ∙ (𝑋𝑒 ∙ 𝑃𝑒𝑖 − 𝑋𝑒 ∙ 𝑤𝑒𝑖) − 𝑤ℎ𝑖


𝑇

𝑖=1

− ∑ 𝛾𝑖 ∙ 𝑁ℎ ∙ {𝑆𝐽ℎ𝑖 + λ ∙ 𝑋𝑒 ∙ 𝑆𝐽𝑒𝑖 + (1 − λ) ∙ [𝑋𝑓 ∙ 𝑆𝐽𝑓𝑖 + 𝑋𝑎 ∙ 𝑆𝐽𝑎𝑖]}

𝑇

𝑖=1

−𝑁ℎ ∙ {𝑆𝑃ℎ + λ ∙ 𝑋𝑒 ∙ 𝑆𝑃𝑒 + (1 − λ) ∙ [𝑋𝑓 ∙ 𝑆𝑃𝑓 + 𝑋𝑎 ∙ 𝑆𝑃𝑎]}

(A13)

PES is economically competitive if and only if NPV is positive, thus:

∑ 𝛾𝑖 ∙ 𝑚 ∙ 𝐶𝐹 ∙ 𝑥𝑖 ∙ 𝑁ℎ ∙ [ λ ∙ (𝑋𝑒 ∙ 𝑃𝑒𝑖 − 𝑋𝑒 ∙ 𝑤𝑒𝑖) − 𝑤ℎ𝑖


𝑇

𝑖=1

− ∑ 𝛾𝑖 ∙ 𝑁ℎ ∙ {𝑆𝐽ℎ𝑖 + λ ∙ 𝑋𝑒 ∙ 𝑆𝐽𝑒𝑖 + (1 − λ) ∙ [𝑋𝑓 ∙ 𝑆𝐽𝑓𝑖 + 𝑋𝑎 ∙ 𝑆𝐽𝑎𝑖]}

𝑇

𝑖=1

−𝑁ℎ ∙ {𝑆𝑃ℎ + λ ∙ 𝑋𝑒 ∙ 𝑆𝑃𝑒 + (1 − λ) ∙ [𝑋𝑓 ∙ 𝑆𝑃𝑓 + 𝑋𝑎 ∙ 𝑆𝑃𝑎]} > 0

(A14)

Upon dividing (A14) by 𝑚 ∙ 𝐶𝐹 ∙ 𝑁ℎ ∑ 𝛾𝑖 ∙ 𝑥𝑖𝑇𝑖=1 , we get:

∑ 𝛾𝑖 ∙ 𝑥𝑖 ∙ [ λ ∙ 𝑋𝑒 ∙ (𝑃𝑒𝑖 − 𝑤𝑒𝑖) + (1 − λ) ∙ (𝑋𝑓 ∙ 𝑃𝑓𝑖 − 𝑋𝑓 ∙ 𝑤𝑓𝑖 − 𝑋𝑎 ∙ 𝑤𝑎𝑖) − 𝑤ℎ𝑖]𝑇𝑖=1

∑ 𝛾𝑖 . 𝑥𝑖𝑇𝑖=1

−∑ 𝛾𝑖 ∙ {𝑆𝐽ℎ𝑖 + λ ∙ 𝑋𝑒 ∙ 𝑆𝐽𝑒𝑖 + (1 − λ) ∙ [𝑋𝑓 ∙ 𝑆𝐽𝑓𝑖 + 𝑋𝑎 ∙ 𝑆𝐽𝑎𝑖]}𝑇

𝑖=1

𝑚 ∙ 𝐶𝐹 ∙ ∑ 𝛾𝑖 ∙ 𝑥𝑖𝑇𝑖=1

−{𝑆𝑃ℎ + λ ∙ 𝑋𝑒 ∙ 𝑆𝑃𝑒 + (1 − λ) ∙ [𝑋𝑓 ∙ 𝑆𝑃𝑓 + 𝑋𝑎 ∙ 𝑆𝑃𝑎]}


> 0

(A15)

[ λ ∙ 𝑋𝑒 ∙ (𝑃𝑒 − 𝑤𝑒) + (1 − λ) ∙ (𝑋𝑓 ∙ 𝑃𝑓 − 𝑋𝑓 ∙ 𝑤𝑓 − 𝑋𝑎 ∙ 𝑤𝑎) − 𝑤ℎ]

−{𝑗ℎ + λ ∙ 𝑋𝑒 ∙ 𝑗𝑒 + (1 − λ) ∙ [𝑋𝑓 ∙ 𝑗𝑓 + 𝑋𝑎 ∙ 𝑗𝑎]}

−{𝑐ℎ + λ ∙ 𝑋𝑒 ∙ 𝑐𝑒 + (1 − λ) ∙ [𝑋𝑓 ∙ 𝑐𝑓 + 𝑋𝑎 ∙ 𝑐𝑎]} > 0

(A16)

𝛌 ∙ (𝑿𝒆 ∙ 𝑷𝒆 − 𝑿𝒆 ∙ 𝑳𝑰𝑪𝒆) +(𝟏 − 𝛌) ∙ (𝑿𝒇 ∙ 𝑷𝒇 − 𝑿𝒇 ∙ 𝑳𝑰𝑪𝒇 − 𝑿𝒂 ∙ 𝑳𝑰𝑪𝒂) > 𝑳𝑪𝑶𝑯

(A17)


Scenario 2a: Flexible PES with a flexible fertilizers subsystem

Derivation of Proposition 2a in (21)

Below is a step-by-step derivation of the formulation of Proposition 2a stated in (21). We

recall from (A1) and (A2) that:

𝑁𝑃𝑉 ($) = ∑ 𝛾𝑖 . 𝐶𝐹𝐿𝑖

𝑇

𝑖=1


𝐶𝑎𝑝𝐸𝑥 = [𝐶𝐴𝑃ℎ + 𝐶𝐴𝑃𝑒 + 𝐶𝐴𝑃𝑎 + 𝐶𝐴𝑃𝑓] (A2)

While 𝐶𝐴𝑃ℎ remains as defined in (A3), the definition of other capacity cost factors are

updated in (A18-A20) to account for the new expanded capacity of all three flexible units:

electricity, ammonia, and fertilizers.

𝐶𝐴𝑃𝑒 = λ𝑚𝑎𝑥 ∙ 𝑋𝑒 ∙ 𝑁ℎ ∙ 𝑆𝑃𝑒 (A18)

𝐶𝐴𝑃𝑎 = (1 − λ𝑚𝑖𝑛) ∙ 𝑋𝑎 ∙ 𝑁ℎ ∙ 𝑆𝑃𝑎 (A19)

𝐶𝐴𝑃𝑓 = (1 − λ𝑚𝑖𝑛) ∙ 𝑋𝑓 ∙ 𝑁ℎ ∙ 𝑆𝑃𝑓 (A20)

Then, by substituting (A3) and (A18-A20) into (A2):

𝐶𝑎𝑝𝐸𝑥 = 𝑁ℎ ∙ {𝑆𝑃ℎ + λ𝑚𝑎𝑥 ∙ 𝑋𝑒 ∙ 𝑆𝑃𝑒 + (1 − λ𝑚𝑖𝑛) ∙ [𝑋𝑎 ∙ 𝑆𝑃𝑎 + 𝑋𝑓 ∙ 𝑆𝑃𝑓]} (A21)

The annual cash flow in year 𝑖 is defined as:

𝐶𝐹𝐿𝑖 = −𝐽𝑖

+𝐶𝐹 ∙ 𝑥𝑖 ∙ 𝑁ℎ ∙ ∫ 𝑚𝑎𝑥⏟𝜆(𝑡)

𝑚

0

[(𝜆(𝑡)) ∙ (𝑋𝑒 ∙ 𝑃𝑒𝑖(𝑡) − 𝑋𝑒 ∙ 𝑤𝑒𝑖) − 𝑤ℎ

+(1 − 𝜆(𝑡)) ∙ (𝑋𝑓 ∙ 𝑃𝑓𝑖 − 𝑋𝑓 ∙ 𝑤𝑓𝑖 − 𝑋𝑎 ∙ 𝑤𝑎𝑖)] 𝑑𝑡

(A22)

Similar to the costs of capacity, the fixed-operating costs in year 𝑖 are updated, such that:

𝐽𝑖 = 𝐽ℎ𝑖 + 𝐽𝑒𝑖 + 𝐽𝑎𝑖 + 𝐽𝑓𝑖

= 𝑁ℎ ∙ [𝑆𝐽ℎ𝑖 + λ𝑚𝑎𝑥 ∙ 𝑋𝑒 ∙ 𝑆𝐽𝑒𝑖 + (1 − λ𝑚𝑖𝑛) ∙ (𝑋𝑎 ∙ 𝑆𝐽𝑎𝑖 + 𝑋𝑓 ∙ 𝑆𝐽𝑓𝑖)] (A23)

The maximization problem in (A22) means that at every time period 𝑡, the flexible PES

operator should allocate hydrogen to electricity and fertilizers production in a way that

maximizes the overall contribution margin.


Substituting of 𝐶𝑀𝑒𝑖 and 𝐶𝑀𝑓𝑖 in (13) and (14), respectively, the integral can be rewritten as:

∫ 𝑚𝑎𝑥⏟𝜆(𝑡)

𝑚

0

[(𝜆(𝑡)) ∙ (𝑋𝑒 ∙ 𝑃𝑒𝑖(𝑡) − 𝑋𝑒 ∙ 𝑤𝑒𝑖)

+(1 − 𝜆(𝑡)) ∙ (𝑋𝑓 ∙ 𝑃𝑓𝑖 − 𝑋𝑓 ∙ 𝑤𝑓𝑖 − 𝑋𝑎 ∙ 𝑤𝑎𝑖) − 𝑤ℎ

] 𝑑𝑡

= ∫ 𝑚𝑎𝑥⏟𝜆(𝑡)

𝑚

0

[(𝜆(𝑡)) ∙ 𝐶𝑀𝑒𝑖(𝑡) + (1 − 𝜆(𝑡)) ∙ 𝐶𝑀𝑓𝑖]𝑑𝑡 − 𝑚 ∙ 𝑤ℎ

(A24)

The integral aims to maximize the contribution margin over the whole year. Intuitively, this

integral can be decomposed to two parts, over 𝑚�̃� and 𝑚�̃�:


𝑚

0

[(𝜆(𝑡)) ∙ 𝐶𝑀𝑒𝑖(𝑡) + (1 − 𝜆(𝑡)) ∙ 𝐶𝑀𝑓𝑖]

= ∫ 𝑚𝑎𝑥⏟𝜆(𝑡)𝑚𝑒𝑖̃

[(𝜆(𝑡)) ∙ 𝐶𝑀𝑒𝑖(𝑡) + (1 − 𝜆(𝑡)) ∙ 𝐶𝑀𝑓𝑖]𝑑𝑡

+ ∫ 𝑚𝑎𝑥⏟𝜆(𝑡)𝑚𝑓𝑖̃


(A25)

Now resolving the maximization problem in each integral is straightforward. During 𝑚�̃�

hours, the contribution margin from electricity generation is higher than that from fertilizers

generation, by definition. Thus, the solution to the first maximization problem over 𝑚�̃� is to

maximize electricity production by setting 𝜆(𝑡) = 𝜆𝑚𝑎𝑥. Similarly, the solution to the second

maximization problem over 𝑚�̃� is to maximize fertilizers production by setting 𝜆(𝑡) = 𝜆𝑚𝑖𝑛.

Thus:


𝑚

0


= ∫ [𝜆𝑚𝑎𝑥 ∙ 𝐶𝑀𝑒𝑖(𝑡) + (1 − 𝜆𝑚𝑎𝑥) ∙ 𝐶𝑀𝑓𝑖]

𝑚𝑒𝑖̃

𝑑𝑡

+ ∫ [𝜆𝑚𝑖𝑛 ∙ 𝐶𝑀𝑒𝑖(𝑡) + (1 − 𝜆𝑚𝑖𝑛) ∙ 𝐶𝑀𝑓𝑖]

𝑚𝑓𝑖̃

𝑑𝑡

(A26)


Upon re-arranging (A26):


𝑚

0


= ∫ [𝜆𝑚𝑎𝑥 ∙ (𝐶𝑀𝑒𝑖(𝑡) − 𝐶𝑀𝑓𝑖)]

𝑚𝑒𝑖̃

𝑑𝑡 + ∫ 𝐶𝑀𝑓𝑖

𝑚𝑒𝑖̃

𝑑𝑡

− ∫ [𝜆𝑚𝑖𝑛 ∙ (𝐶𝑀𝑓𝑖(𝑡) − 𝐶𝑀𝑒𝑖)]

𝑚𝑓𝑖̃


𝑚𝑓𝑖̃

𝑑𝑡

= 𝜆𝑚𝑎𝑥 ∙ ∫(𝐶𝑀𝑒𝑖(𝑡) − 𝐶𝑀𝑓𝑖)

𝑚𝑒𝑖̃

𝑑𝑡 − 𝜆𝑚𝑖𝑛 ∙ ∫(𝐶𝑀𝑓𝑖(𝑡) − 𝐶𝑀𝑒𝑖)

𝑚𝑓𝑖̃


𝑚

0

𝑑𝑡

(A27)

Then, using the definitions of 𝐼𝐶𝑀𝐹𝑒𝑖 and 𝐼𝐶𝑀𝐹𝑓𝑖 in (17) and (18), respectively:


𝑚

0

[(𝜆(𝑡)) ∙ 𝐶𝑀𝑒(𝑡) + (1 − 𝜆(𝑡)) ∙ 𝐶𝑀𝑓]𝑑𝑡

= 𝑚 ∙ 𝜆𝑚𝑎𝑥 ∙ 𝐼𝐶𝑀𝐹𝑒𝑖 − 𝑚 ∙ 𝜆𝑚𝑖𝑛 ∙ 𝐼𝐶𝑀𝐹𝑓𝑖

+𝑚 ∙ [𝑋𝑓 . 𝑃𝑓𝑖 − 𝑋𝑓 . 𝑤𝑓𝑖 − 𝑋𝑎 . 𝑤𝑎𝑖]

(A28)

Substituting the result of (A23). (A24), and (A28) into (A22), we get:

𝐶𝐹𝐿𝑖 = −𝑁ℎ ∙ [𝑆𝐽ℎ𝑖 + λ𝑚𝑎𝑥 ∙ 𝑋𝑒 ∙ 𝑆𝐽𝑒𝑖 + (1 − λ𝑚𝑖𝑛) ∙ (𝑋𝑎 ∙ 𝑆𝐽𝑎𝑖 + 𝑋𝑓 ∙ 𝑆𝐽𝑓𝑖)]

+𝐶𝐹 ∙ 𝑥𝑖 ∙ 𝑁ℎ ∙ 𝑚 ∙ {𝜆𝑚𝑎𝑥 ∙ 𝐼𝐶𝑀𝐹𝑒𝑖 − 𝜆𝑚𝑖𝑛 ∙ 𝐼𝐶𝑀𝐹𝑓𝑖

+[𝑋𝑓 . 𝑃𝑓𝑖 − 𝑋𝑓 . 𝑤𝑓𝑖 − 𝑋𝑎 . 𝑤𝑎𝑖] − 𝑤ℎ𝑖}

(A29)

Then the NPV becomes:

𝑁𝑃𝑉 ($) = − ∑ 𝛾𝑖 ∙ 𝑁ℎ ∙ [𝑆𝐽ℎ𝑖 + λ𝑚𝑎𝑥 ∙ 𝑋𝑒 ∙ 𝑆𝐽𝑒𝑖

+(1 − λ𝑚𝑖𝑛) ∙ (𝑋𝑎 ∙ 𝑆𝐽𝑎𝑖 + 𝑋𝑓 ∙ 𝑆𝐽𝑓𝑖)]

𝑇

𝑖=1

+ ∑ 𝛾𝑖 ∙ 𝐶𝐹 ∙ 𝑥𝑖 ∙ 𝑁ℎ ∙ 𝑚 ∙ {𝜆𝑚𝑎𝑥 ∙ 𝐼𝐶𝑀𝐹𝑒𝑖 − 𝜆𝑚𝑖𝑛 ∙ 𝐼𝐶𝑀𝐹𝑓𝑖


𝑇

𝑖=1

−𝑁ℎ ∙ {𝑆𝑃ℎ + λ𝑚𝑎𝑥 ∙ 𝑋𝑒 ∙ 𝑆𝑃𝑒 + (1 − λ𝑚𝑖𝑛)[𝑋𝑎 ∙ 𝑆𝑃𝑎 + 𝑋𝑓 ∙ 𝑆𝑃𝑓]}

(A30)



− ∑ 𝛾𝑖 ∙ 𝑁ℎ ∙ [𝑆𝐽ℎ𝑖 + λ𝑚𝑎𝑥 ∙ 𝑋𝑒 ∙ 𝑆𝐽𝑒𝑖

+(1 − λ𝑚𝑖𝑛) ∙ (𝑋𝑎 ∙ 𝑆𝐽𝑎𝑖 + 𝑋𝑓 ∙ 𝑆𝐽𝑓𝑖)]

𝑇

𝑖=1



𝑇

𝑖=1

−𝑁ℎ ∙ {𝑆𝑃ℎ + λ𝑚𝑎𝑥 ∙ 𝑋𝑒 ∙ 𝑆𝑃𝑒 + (1 − λ𝑚𝑖𝑛)[𝑋𝑎 ∙ 𝑆𝑃𝑎 + 𝑋𝑓 ∙ 𝑆𝑃𝑓]} > 0

(A31)

Upon dividing (A31) by 𝑚 ∙ 𝐶𝐹 ∙ 𝑁ℎ ∙ ∑ 𝛾𝑖 ∙ 𝑥𝑖𝑇𝑖=1 , we get:

−∑ 𝛾𝑖 ∙ [𝑆𝐽ℎ𝑖 + λ𝑚𝑎𝑥 ∙ 𝑋𝑒 ∙ 𝑆𝐽𝑒𝑖 + (1 − λ𝑚𝑖𝑛) ∙ (𝑋𝑎 ∙ 𝑆𝐽𝑎𝑖 + 𝑋𝑓 ∙ 𝑆𝐽𝑓𝑖)]𝑇

𝑖=1


+

∑ 𝛾𝑖 ∙ 𝑥𝑖 ∙ {𝜆𝑚𝑎𝑥 ∙ 𝐼𝐶𝑀𝐹𝑒𝑖 − 𝜆𝑚𝑖𝑛 ∙ 𝐼𝐶𝑀𝐹𝑓𝑖

+[𝑋𝑓 . 𝑃𝑓𝑖 − 𝑋𝑓 . 𝑤𝑓𝑖 − 𝑋𝑎 . 𝑤𝑎𝑖] − 𝑤ℎ𝑖}𝑇

𝑖=1

∑ 𝛾𝑖 ∙ 𝑥𝑖𝑇𝑖=1

−{𝑆𝑃ℎ + λ𝑚𝑎𝑥 ∙ 𝑋𝑒 ∙ 𝑆𝑃𝑒 + (1 − λ𝑚𝑖𝑛)[𝑋𝑎 ∙ 𝑆𝑃𝑎 + 𝑋𝑓 ∙ 𝑆𝑃𝑓]}


> 0

(A32)

−𝑗ℎ − λ𝑚𝑎𝑥 ∙ 𝑋𝑒 ∙ 𝑗𝑒 − (1 − λ𝑚𝑖𝑛) ∙ (𝑋𝑎 ∙ 𝑗𝑎 + 𝑋𝑓 ∙ 𝑗𝑓)

+𝜆𝑚𝑎𝑥 ∙ 𝐼𝐶𝑀𝐹𝑒 − 𝜆𝑚𝑖𝑛 ∙ 𝐼𝐶𝑀𝐹𝑓 + [𝑋𝑓 . 𝑃𝑓 − 𝑋𝑓 . 𝑤𝑓 − 𝑋𝑎 . 𝑤𝑎] − 𝑤ℎ

−𝑐ℎ − λ𝑚𝑎𝑥 ∙ 𝑋𝑒 ∙ 𝑐𝑒 − (1 − λ𝑚𝑖𝑛)(𝑋𝑎 ∙ 𝑐𝑎 + 𝑋𝑓 ∙ 𝑐𝑓) > 0

(A33)

𝜆𝑚𝑎𝑥 ∙ (𝐼𝐶𝑀𝐹𝑒 − 𝑋𝑒 ∙ 𝑐𝑒 − 𝑋𝑒 ∙ 𝑗𝑒)

−𝜆𝑚𝑖𝑛 ∙ (𝐼𝐶𝑀𝐹𝑓 − 𝑋𝑎 ∙ 𝑗𝑎 − 𝑋𝑓 ∙ 𝑗𝑓 − 𝑋𝑎 ∙ 𝑐𝑎 − 𝑋𝑓 ∙ 𝑐𝑓)

+[𝑋𝑓 . 𝑃𝑓 − 𝑋𝑓 . 𝑤𝑓 − 𝑋𝑎 . 𝑤𝑎 − 𝑋𝑎 ∙ 𝑗𝑎 − 𝑋𝑓 ∙ 𝑗𝑓 − 𝑋𝑎 ∙ 𝑐𝑎 − 𝑋𝑓 ∙ 𝑐𝑓] > 𝑐ℎ + 𝑗ℎ + 𝑤ℎ

(A34)

(𝑿𝒇. 𝑷𝒇 − 𝑿𝒇. 𝑳𝑰𝑪𝒇 − 𝑿𝒂. 𝑳𝑰𝑪𝒂)

+𝝀𝒎𝒂𝒙 ∙ (𝑰𝑪𝑴𝑭𝒆 − 𝑿𝒆 ∙ (𝒄𝒆 + 𝒋𝒆))

−𝝀𝒎𝒊𝒏 ∙ (𝑰𝑪𝑴𝑭𝒇 − 𝑿𝒇 ∙ (𝒄𝒇 + 𝒋𝒇) − 𝑿𝒂 ∙ (𝒄𝒂 + 𝒋𝒂)) > 𝑳𝑪𝑶𝑯

(A35)


Derivation of Proposition 2a in (22)

The derivation of the formulation of Proposition 2a in (22) follows a very similar path to that

in (21). Following the same steps from (A18) through (A26), we re-arrange (A27) as follows:


𝑚

0


= ∫ [(1 − 𝜆𝑚𝑎𝑥) ∙ (𝐶𝑀𝑓𝑖(𝑡) − 𝐶𝑀𝑒𝑖)]

𝑚𝑒𝑖̃

𝑑𝑡 + ∫ 𝐶𝑀𝑒𝑖

𝑚𝑒𝑖̃

𝑑𝑡

+ ∫ [(1 − 𝜆𝑚𝑖𝑛) ∙ (𝐶𝑀𝑓𝑖(𝑡) − 𝐶𝑀𝑒𝑖)]

𝑚𝑓𝑖̃


𝑚𝑓𝑖̃

𝑑𝑡

= (1 − 𝜆𝑚𝑎𝑥) ∙ ∫(𝐶𝑀𝑓𝑖(𝑡) − 𝐶𝑀𝑒𝑖)

𝑚𝑒𝑖̃

𝑑𝑡

+(1 − 𝜆𝑚𝑖𝑛) ∙ ∫ (𝐶𝑀𝑓𝑖(𝑡) − 𝐶𝑀𝑒𝑖)

𝑚𝑓𝑖̃


𝑚

0

𝑑𝑡

(A36)

Then, using the definitions of 𝐼𝐶𝑀𝐹𝑒𝑖 and 𝐼𝐶𝑀𝐹𝑓𝑖 in (17) and (18), respectively:


𝑚

0

[(𝜆(𝑡)) ∙ 𝐶𝑀𝑒(𝑡) + (1 − 𝜆(𝑡)) ∙ 𝐶𝑀𝑓]𝑑𝑡

= −𝑚 ∙ (1 − 𝜆𝑚𝑎𝑥) ∙ 𝐼𝐶𝑀𝐹𝑒𝑖 + 𝑚 ∙ (1 − 𝜆𝑚𝑖𝑛) ∙ 𝐼𝐶𝑀𝐹𝑓𝑖 + 𝑚 ∙ [𝑋𝑒 . 𝑃𝑒𝑖 − 𝑋𝑒 . 𝑤𝑒𝑖]

(A37)

Substituting the result of (A23) and (A37) in (A22), we get:

𝐶𝐹𝐿𝑖 = −𝑁ℎ ∙ [𝑆𝐽ℎ𝑖 + λ𝑚𝑎𝑥 ∙ 𝑋𝑒 ∙ 𝑆𝐽𝑒𝑖 + (1 − λ𝑚𝑖𝑛) ∙ (𝑋𝑎 ∙ 𝑆𝐽𝑎𝑖 + 𝑋𝑓 ∙ 𝑆𝐽𝑓𝑖) ]

+𝐶𝐹 ∙ 𝑥𝑖 ∙ 𝑁ℎ ∙ 𝑚 ∙ {(1 − 𝜆𝑚𝑖𝑛) ∙ 𝐼𝐶𝑀𝐹𝑓𝑖 − (1 − 𝜆𝑚𝑎𝑥) ∙ 𝐼𝐶𝑀𝐹𝑒𝑖

+[𝑋𝑒 . 𝑃𝑒𝑖 − 𝑋𝑒 . 𝑤𝑒𝑖] − 𝑤ℎ𝑖}

(A38)

Then the NPV becomes:


𝑁𝑃𝑉 ($) = − ∑ 𝛾𝑖 ∙ 𝑁ℎ ∙ [𝑆𝐽ℎ𝑖 + λ𝑚𝑎𝑥 ∙ 𝑋𝑒 ∙ 𝑆𝐽𝑒𝑖

+(1 − λ𝑚𝑖𝑛) ∙ (𝑋𝑎 ∙ 𝑆𝐽𝑎𝑖 + 𝑋𝑓 ∙ 𝑆𝐽𝑓𝑖) ]

𝑇

𝑖=1

+ ∑ 𝛾𝑖 ∙ 𝐶𝐹 ∙ 𝑥𝑖 ∙ 𝑁ℎ ∙ 𝑚 ∙ {(1 − 𝜆𝑚𝑖𝑛) ∙ 𝐼𝐶𝑀𝐹𝑓𝑖 − (1 − 𝜆𝑚𝑎𝑥) ∙ 𝐼𝐶𝑀𝐹𝑒𝑖


𝑇

𝑖=1

−𝑁ℎ ∙ [𝑆𝑃ℎ + λ𝑚𝑎𝑥 ∙ 𝑋𝑒 ∙ 𝑆𝑃𝑒 + (1 − λ𝑚𝑖𝑛)(𝑋𝑎 ∙ 𝑆𝑃𝑎 + 𝑋𝑓 ∙ 𝑆𝑃𝑓)]

(A39)


− ∑ 𝛾𝑖 ∙ 𝑁ℎ ∙ [𝑆𝐽ℎ𝑖 + λ𝑚𝑎𝑥 ∙ 𝑋𝑒 ∙ 𝑆𝐽𝑒𝑖 + (1 − λ𝑚𝑖𝑛) ∙ (𝑋𝑎 ∙ 𝑆𝐽𝑎𝑖 + 𝑋𝑓 ∙ 𝑆𝐽𝑓𝑖) ]

𝑇

𝑖=1

+ ∑ 𝛾𝑖 ∙ 𝐶𝐹 ∙ 𝑥𝑖 ∙ 𝑁ℎ ∙ 𝑚 ∙ {(1 − 𝜆𝑚𝑖𝑛) ∙ 𝐼𝐶𝑀𝐹𝑓𝑖 − (1 − 𝜆𝑚𝑎𝑥) ∙ 𝐼𝐶𝑀𝐹𝑒𝑖


𝑇

𝑖=1

−𝑁ℎ ∙ [𝑆𝑃ℎ + λ𝑚𝑎𝑥 ∙ 𝑋𝑒 ∙ 𝑆𝑃𝑒 + (1 − λ𝑚𝑖𝑛)(𝑋𝑎 ∙ 𝑆𝑃𝑎 + 𝑋𝑓 ∙ 𝑆𝑃𝑓)] > 0

(A40)


−∑ 𝛾𝑖 ∙ [𝑆𝐽ℎ𝑖 + λ𝑚𝑎𝑥 ∙ 𝑋𝑒 ∙ 𝑆𝐽𝑒𝑖 + (1 − λ𝑚𝑖𝑛) ∙ (𝑋𝑎 ∙ 𝑆𝐽𝑎𝑖 + 𝑋𝑓 ∙ 𝑆𝐽𝑓𝑖)]𝑇

𝑖=1


+

∑ 𝛾𝑖 ∙ 𝑥𝑖 ∙ {(1 − 𝜆𝑚𝑖𝑛) ∙ 𝐼𝐶𝑀𝐹𝑓𝑖 − (1 − 𝜆𝑚𝑎𝑥) ∙ 𝐼𝐶𝑀𝐹𝑒𝑖

+[𝑋𝑒 . 𝑃𝑒𝑖 − 𝑋𝑒 . 𝑤𝑒𝑖] − 𝑤ℎ𝑖}𝑇

𝑖=1


−[𝑆𝑃ℎ + λ𝑚𝑎𝑥 ∙ 𝑋𝑒 ∙ 𝑆𝑃𝑒 + (1 − λ𝑚𝑖𝑛)(𝑋𝑎 ∙ 𝑆𝑃𝑎 + 𝑋𝑓 ∙ 𝑆𝑃𝑓)]


> 0

(A41)

−𝑗ℎ − λ𝑚𝑎𝑥 ∙ 𝑋𝑒 ∙ 𝑗𝑒 − (1 − λ𝑚𝑖𝑛) ∙ (𝑋𝑎 ∙ 𝑗𝑎 + 𝑋𝑓 ∙ 𝑗𝑓)

+(1 − 𝜆𝑚𝑖𝑛) ∙ 𝐼𝐶𝑀𝐹𝑓 − (1 − 𝜆𝑚𝑎𝑥) ∙ 𝐼𝐶𝑀𝐹𝑒 + [𝑋𝑒 . 𝑃𝑒 − 𝑋𝑒 . 𝑤𝑒] − 𝑤ℎ

−𝑐ℎ − λ𝑚𝑎𝑥 ∙ 𝑋𝑒 ∙ 𝑐𝑒 − (1 − λ𝑚𝑖𝑛)(𝑋𝑎 ∙ 𝑐𝑎 + 𝑋𝑓 ∙ 𝑐𝑓) > 0

(A42)

(1 − 𝜆𝑚𝑖𝑛) ∙ (𝐼𝐶𝑀𝐹𝑓 − 𝑋𝑎 ∙ 𝑗𝑎 − 𝑋𝑓 ∙ 𝑗𝑓 − 𝑋𝑎 ∙ 𝑐𝑎 − 𝑋𝑓 ∙ 𝑐𝑓)

−(1 − 𝜆𝑚𝑎𝑥) ∙ 𝐼𝐶𝑀𝐹𝑒 − λ𝑚𝑎𝑥 ∙ 𝑋𝑒 ∙ 𝑗𝑒 − λ𝑚𝑎𝑥 ∙ 𝑋𝑒 ∙ 𝑐𝑒 + 𝑋𝑒 ∙ 𝑗𝑒 + 𝑋𝑒 ∙ 𝑐𝑒

+[𝑋𝑒 . 𝑃𝑒 − 𝑋𝑒 . 𝑤𝑒] − 𝑋𝑒 ∙ 𝑗𝑒 − 𝑋𝑒 ∙ 𝑐𝑒 > 𝑐ℎ + 𝑤ℎ + 𝑗ℎ

(A43)

[𝑿𝒆. 𝑷𝒆 − 𝑿𝒆. 𝑳𝑰𝑪𝒆]

+(𝟏 − 𝝀𝒎𝒊𝒏) ∙ [𝑰𝑪𝑴𝑭𝒇 − 𝑿𝒂 ∙ (𝒄𝒂 + 𝒋𝒂) − 𝑿𝒇 ∙ (𝒄𝒇 + 𝒋𝒇)]

−(𝟏 − 𝝀𝒎𝒂𝒙) ∙ [𝑰𝑪𝑴𝑭𝒆 − 𝑿𝒆 ∙ 𝒋𝒆 − 𝑿𝒆 ∙ 𝒄𝒆] > 𝑳𝑪𝑶𝑯

(A44)


Scenario 2b: Flexible PES with a static fertilizers subsystem

Derivation of Proposition 2b in (24)

Below is a step-by-step derivation of the formulation of Proposition 2b stated in (24). We

recall from (A1) that:

𝑁𝑃𝑉 ($) = ∑ 𝛾𝑖 . 𝐶𝐹𝐿𝑖

𝑇

𝑖=1


The cost of capacity in this case should account for the updated capacity of the fertilizers

subsystems as well as the intermediate storage of ammonia.

𝐶𝑎𝑝𝐸𝑥 = [𝐶𝐴𝑃ℎ + 𝐶𝐴𝑃𝑒 + 𝐶𝐴𝑃𝑎𝑠 + 𝐶𝐴𝑃𝑓] (A45)

𝐶𝐴𝑃𝑎𝑠: cost of capacity of ammonia generation and intermediate storage ($)

While 𝐶𝐴𝑃ℎ and 𝐶𝐴𝑃𝑒 are as defined in (A3) and (A18), respectively, the definition of 𝐶𝐴𝑃𝑓 is

updated in (A46), and 𝐶𝐴𝑃𝑎𝑠 is introduced in (A47) to represent the new capacity cost of

ammonia generation and storage.

𝐶𝐴𝑃𝑓 = 𝐾 ∙ 𝑋𝑓 ∙ 𝑁ℎ ∙ 𝑆𝑃𝑓 (A46)

𝐶𝐴𝑃𝑎𝑠 = (1 − λ𝑚𝑖𝑛) ∙ 𝑋𝑎 ∙ 𝑁ℎ ∙ 𝑆𝑃𝑎𝑠 (A47)

𝑆𝑃𝑎𝑠: system price of ammonia generation and intermediate storage per unit of generation

capacity ($/(𝑘𝑔𝑎 ℎ𝑟⁄ ))

Then, by substituting (A3), (A18), (A46) and (A47) into (A45):

𝐶𝑎𝑝𝐸𝑥 = 𝑁ℎ ∙ [𝑆𝑃ℎ + λ𝑚𝑎𝑥 ∙ 𝑋𝑒 ∙ 𝑆𝑃𝑒 + (1 − λ𝑚𝑖𝑛) ∙ 𝑋𝑎 ∙ 𝑆𝑃𝑎𝑠 + 𝐾 ∙ 𝑋𝑓 ∙ 𝑆𝑃𝑓] (A48)

Similar to the costs of capacity, the fixed-operating costs in year 𝑖 are updated, such that:

𝐽𝑖 = 𝐽ℎ𝑖 + 𝐽𝑒𝑖 + 𝐽𝑎𝑠𝑖 + 𝐽𝑓𝑖

= 𝑁ℎ ∙ [𝑆𝐽ℎ𝑖 + λ𝑚𝑎𝑥 ∙ 𝑋𝑒 ∙ 𝑆𝐽𝑒𝑖 + (1 − λ𝑚𝑖𝑛) ∙ 𝑋𝑎 ∙ 𝑆𝐽𝑎𝑠𝑖 + 𝐾 ∙ 𝑋𝑓 ∙ 𝑆𝐽𝑓𝑖 ] (A49)

The annual cash flow in year 𝑖 is defined as:


𝐶𝐹𝐿𝑖 = −𝐽𝑖 + 𝐶𝐹 ∙ 𝑥𝑖 ∙ 𝑁ℎ ∙ ∫ 𝑚𝑎𝑥⏟𝜆(𝑡)

𝑚

0


+(𝐾) ∙ (𝑋𝑓 ∙ 𝑃𝑓𝑖 − 𝑋𝑓 ∙ 𝑤𝑓𝑖 − 𝑋𝑎 ∙ 𝑤𝑎𝑠𝑖)] 𝑑𝑡 (A50)

Maximizing the cash flow from the flexible PES requires following the same operational

policy described in the simplified Scenario 2a. Thus, using the definitions of 𝐶𝑀𝑒𝑖 and 𝐶𝑀𝑓𝑖

in (13) and (14), the maximization problem be expressed as:


𝑚

0


+(𝐾) ∙ (𝑋𝑓 ∙ 𝑃𝑓𝑖 − 𝑋𝑓 ∙ 𝑤𝑓𝑖 − 𝑋𝑎 ∙ 𝑤𝑎𝑠𝑖)] 𝑑𝑡

= ∫ λ𝑚𝑎𝑥 ∙ (𝐶𝑀𝑒𝑖(𝑡))𝑑𝑡

𝑚𝑒𝑖̃

+ ∫ λ𝑚𝑖𝑛 ∙ (𝐶𝑀𝑒𝑖(𝑡))𝑑𝑡

𝑚𝑓𝑖̃

+ ∫ [𝐾 ∙ (𝐶𝑀𝑓𝑖(𝑡)) − 𝑤ℎ] 𝑑𝑡

𝑚

0

(A51)

Upon re-arranging (A51):


𝑚

0



= ∫ λ𝑚𝑎𝑥 ∙ (𝐶𝑀𝑒𝑖(𝑡) − 𝐶𝑀𝑓𝑖(𝑡)) 𝑑𝑡

𝑚𝑒𝑖̃

+ ∫ λ𝑚𝑎𝑥 ∙ (𝐶𝑀𝑓𝑖(𝑡)) 𝑑𝑡

𝑚𝑒𝑖̃

− ∫ λ𝑚𝑖𝑛 ∙ (𝐶𝑀𝑓𝑖(𝑡) − 𝐶𝑀𝑒𝑖(𝑡)) 𝑑𝑡

𝑚𝑓𝑖̃

+ ∫ λ𝑚𝑖𝑛 ∙ (𝐶𝑀𝑓𝑖(𝑡)) 𝑑𝑡

𝑚𝑓𝑖̃

+ ∫ [𝐾 ∙ (𝐶𝑀𝑓𝑖(𝑡)) − 𝑤ℎ] 𝑑𝑡

𝑚

0

(A52)

Then, using the definitions of 𝐼𝐶𝑀𝐹𝑒𝑖 and 𝐼𝐶𝑀𝐹𝑓𝑖 in (17) and (18), respectively, and

expanding 𝐾:


𝑚

0



= 𝑚 ∙ (λ𝑚𝑎𝑥 ∙ 𝐼𝐶𝑀𝐹𝑒𝑖 − λ𝑚𝑖𝑛 ∙ 𝐼𝐶𝑀𝐹𝑓𝑖) + ∫ λ𝑚𝑎𝑥 ∙ (𝐶𝑀𝑓𝑖(𝑡)) 𝑑𝑡

𝑚𝑒𝑖̃

+ ∫ λ𝑚𝑖𝑛 ∙ (𝐶𝑀𝑓𝑖(𝑡)) 𝑑𝑡

𝑚𝑓𝑖̃

+ (𝑚 − 𝑚𝑒𝑖 ∙ 𝜆𝑚𝑎𝑥 − 𝑚𝑓𝑖 ∙ 𝜆𝑚𝑖𝑛) ∙ 𝐶𝑀𝑓𝑖 − 𝑚 ∙ 𝑤ℎ

(A53)


But we know that 𝐶𝑀𝑓𝑖(𝑡) = 𝐶𝑀𝑓𝑖 because we assumed constant fertilizers price and variable

costs in a given year 𝑖. Under these conditions, (A53) can be re-written as:


𝑚

0



= 𝑚 ∙ (λ𝑚𝑎𝑥 ∙ 𝐼𝐶𝑀𝐹𝑒𝑖 − λ𝑚𝑖𝑛 ∙ 𝐼𝐶𝑀𝐹𝑓𝑖) + (𝑚𝑒𝑖 ∙ 𝜆𝑚𝑎𝑥 + 𝑚𝑓𝑖 ∙ 𝜆𝑚𝑖𝑛) ∙ 𝐶𝑀𝑓𝑖

+(𝑚 − 𝑚𝑒𝑖 ∙ 𝜆𝑚𝑎𝑥 − 𝑚𝑓𝑖 ∙ 𝜆𝑚𝑖𝑛) ∙ 𝐶𝑀𝑓𝑖 − 𝑚 ∙ 𝑤ℎ

(A54)

Which then simplifies to:


𝑚

0



= 𝑚 ∙ (λ𝑚𝑎𝑥 ∙ 𝐼𝐶𝑀𝐹𝑒𝑖 − λ𝑚𝑖𝑛 ∙ 𝐼𝐶𝑀𝐹𝑓𝑖) + 𝑚 ∙ (𝑋𝑓𝑃𝑓𝑖 − 𝑋𝑓𝑤𝑓𝑖 − 𝑋𝑎𝑤𝑎𝑠𝑖) − 𝑚 ∙ 𝑤ℎ

(A55)

Substituting the result of (A55) and (A49) into (A50), we get:

𝐶𝐹𝐿𝑖 = −𝑁ℎ ∙ [𝑆𝐽ℎ𝑖 + λ𝑚𝑎𝑥 ∙ 𝑋𝑒 ∙ 𝑆𝐽𝑒𝑖 + (1 − λ𝑚𝑖𝑛) ∙ 𝑋𝑎 ∙ 𝑆𝐽𝑎𝑠𝑖 + 𝐾 ∙ 𝑋𝑓 ∙ 𝑆𝐽𝑓𝑖 ]

+𝐶𝐹 ∙ 𝑥𝑖 ∙ 𝑁ℎ ∙ 𝑚 ∙ {𝜆𝑚𝑎𝑥 ∙ 𝐼𝐶𝑀𝐹𝑒𝑖 − 𝜆𝑚𝑖𝑛 ∙ 𝐼𝐶𝑀𝐹𝑓𝑖

+[𝑋𝑓 . 𝑃𝑓𝑖 − 𝑋𝑓 . 𝑤𝑓𝑖 − 𝑋𝑎 . 𝑤𝑎𝑠𝑖] − 𝑤ℎ𝑖}

(A56)

Then, substituting (A56) and (A48) into (A1), the NPV becomes:

𝑁𝑃𝑉 ($) =

− ∑ 𝛾𝑖 ∙ 𝑁ℎ ∙ [𝑆𝐽ℎ𝑖 + λ𝑚𝑎𝑥 ∙ 𝑋𝑒 ∙ 𝑆𝐽𝑒𝑖 + (1 − λ𝑚𝑖𝑛) ∙ 𝑋𝑎 ∙ 𝑆𝐽𝑎𝑠𝑖 + 𝐾 ∙ 𝑋𝑓 ∙ 𝑆𝐽𝑓𝑖]

𝑇

𝑖=1



𝑇

𝑖=1

−𝑁ℎ ∙ [𝑆𝑃ℎ + λ𝑚𝑎𝑥 ∙ 𝑋𝑒 ∙ 𝑆𝑃𝑒 + (1 − λ𝑚𝑖𝑛) ∙ 𝑋𝑎 ∙ 𝑆𝑃𝑎𝑠 + 𝐾 ∙ 𝑋𝑓 ∙ 𝑆𝑃𝑓]

(A57)



− ∑ 𝛾𝑖 ∙ 𝑁ℎ ∙ [𝑆𝐽ℎ𝑖 + λ𝑚𝑎𝑥 ∙ 𝑋𝑒 ∙ 𝑆𝐽𝑒𝑖 + (1 − λ𝑚𝑖𝑛) ∙ 𝑋𝑎 ∙ 𝑆𝐽𝑎𝑠𝑖 + 𝐾 ∙ 𝑋𝑓 ∙ 𝑆𝐽𝑓𝑖 ]

𝑇

𝑖=1



𝑇

𝑖=1

−𝑁ℎ ∙ [𝑆𝑃ℎ + λ𝑚𝑎𝑥 ∙ 𝑋𝑒 ∙ 𝑆𝑃𝑒 + (1 − λ𝑚𝑖𝑛) ∙ 𝑋𝑎 ∙ 𝑆𝑃𝑎𝑠 + 𝐾 ∙ 𝑋𝑓 ∙ 𝑆𝑃𝑓] > 0

(A58)


−∑ 𝛾𝑖 ∙ [𝑆𝐽ℎ𝑖 + λ𝑚𝑎𝑥 ∙ 𝑋𝑒 ∙ 𝑆𝐽𝑒𝑖 + (1 − λ𝑚𝑖𝑛) ∙ 𝑋𝑎 ∙ 𝑆𝐽𝑎𝑠𝑖 + 𝐾 ∙ 𝑋𝑓 ∙ 𝑆𝐽𝑓𝑖]𝑇

𝑖=1


+

∑ 𝛾𝑖 ∙ 𝑥𝑖 ∙ {𝜆𝑚𝑎𝑥 ∙ 𝐼𝐶𝑀𝐹𝑒𝑖 − 𝜆𝑚𝑖𝑛 ∙ 𝐼𝐶𝑀𝐹𝑓𝑖

+[𝑋𝑓 . 𝑃𝑓𝑖 − 𝑋𝑓 . 𝑤𝑓𝑖 − 𝑋𝑎 . 𝑤𝑎𝑠𝑖] − 𝑤ℎ𝑖}𝑇

𝑖=1


−[𝑆𝑃ℎ + λ𝑚𝑎𝑥 ∙ 𝑋𝑒 ∙ 𝑆𝑃𝑒 + (1 − λ𝑚𝑖𝑛) ∙ 𝑋𝑎 ∙ 𝑆𝑃𝑎𝑠 + 𝐾 ∙ 𝑋𝑓 ∙ 𝑆𝑃𝑓]


> 0

(A59)

−𝑗ℎ − λ𝑚𝑎𝑥 ∙ 𝑋𝑒 ∙ 𝑗𝑒 − (1 − λ𝑚𝑖𝑛) ∙ 𝑋𝑎 ∙ 𝑗𝑎𝑠 − 𝐾 ∙ 𝑋𝑓 ∙ 𝑗𝑓

+𝜆𝑚𝑎𝑥 ∙ 𝐼𝐶𝑀𝐹𝑒 − 𝜆𝑚𝑖𝑛 ∙ 𝐼𝐶𝑀𝐹𝑓 + [𝑋𝑓 . 𝑃𝑓 − 𝑋𝑓 . 𝑤𝑓 − 𝑋𝑎 . 𝑤𝑎𝑠] − 𝑤ℎ

−𝑐ℎ − λ𝑚𝑎𝑥 ∙ 𝑋𝑒 ∙ 𝑐𝑒 − (1 − λ𝑚𝑖𝑛)𝑋𝑎 ∙ 𝑐𝑎𝑠 − 𝐾 ∙ 𝑋𝑓 ∙ 𝑐𝑓 > 0

(A60)

−𝑗ℎ − λ𝑚𝑎𝑥 ∙ 𝑋𝑒 ∙ 𝑗𝑒 − (1 − λ𝑚𝑖𝑛) ∙ 𝑋𝑎 ∙ 𝑗𝑎𝑠 − (1 − λ𝑚𝑖𝑛) ∙ 𝑋𝑓 ∙ 𝑗𝑓

+𝜆𝑚𝑎𝑥 ∙ 𝐼𝐶𝑀𝐹𝑒 − 𝜆𝑚𝑖𝑛 ∙ 𝐼𝐶𝑀𝐹𝑓 + [𝑋𝑓 . 𝑃𝑓 − 𝑋𝑓 . 𝑤𝑓 − 𝑋𝑎 . 𝑤𝑎𝑠] − 𝑤ℎ

−𝑐ℎ − λ𝑚𝑎𝑥 ∙ 𝑋𝑒 ∙ 𝑐𝑒 − (1 − λ𝑚𝑖𝑛)𝑋𝑎 ∙ 𝑐𝑎𝑠 − (1 − λ𝑚𝑖𝑛) ∙ 𝑋𝑓 ∙ 𝑐𝑓

+(1 − λ𝑚𝑖𝑛 − 𝐾) ∙ 𝑋𝑓 ∙ (𝑐𝑓 + 𝑋𝑓) > 0

(A61)

[𝑿𝒇. 𝑷𝒇 − 𝑿𝒇. 𝑳𝑰𝑪𝒇 − 𝑿𝒂. 𝑳𝑰𝑪𝒂𝒔]

+𝝀𝒎𝒂𝒙 ∙ [𝑰𝑪𝑴𝑭𝒆 − 𝑿𝒆 ∙ (𝒄𝒆 + 𝒋𝒆)]

−𝝀𝒎𝒊𝒏 ∙ [𝑰𝑪𝑴𝑭𝒇 − 𝑿𝒇 ∙ (𝒄𝒇 + 𝒋𝒇) − 𝑿𝒂 ∙ (𝒄𝒂𝒔 + 𝒋𝒂𝒔)]

+(𝟏 − 𝛌𝒎𝒊𝒏 − 𝑲) ∙ [𝑿𝒇 ∙ (𝒄𝒇 + 𝒋𝒇)] > 𝑳𝑪𝑶𝑯

(A62)


Derivation of Proposition 2b in (25)

The derivation of the formulation of Proposition 2b in (25) follows a very similar path to that

in (24). Following the same steps from (A45) through (A51), we re-arrange (A52) as follows:


[(𝜆(𝑡)) ∙ (𝑋𝑒 ∙ 𝑃𝑒𝑖(𝑡) − 𝑋𝑒 ∙ 𝑤𝑒𝑖)

+(𝐾) ∙ (𝑋𝑓 ∙ 𝑃𝑓𝑖 − 𝑋𝑓 ∙ 𝑤𝑓𝑖 − 𝑋𝑎 ∙ 𝑤𝑎𝑠𝑖) − 𝑤ℎ

] 𝑑𝑡

𝑚

0

= ∫ (𝜆𝑚𝑎𝑥 − 1) ∙ (𝐶𝑀𝑒𝑖(𝑡))𝑑𝑡

𝑚𝑒𝑖̃

+ ∫(𝐶𝑀𝑒𝑖(𝑡))𝑑𝑡

𝑚𝑒𝑖̃

+ ∫ (𝜆𝑚𝑎𝑥 − 1) ∙ (𝐶𝑀𝑓𝑖(𝑡) − 𝐶𝑀𝑓𝑖(𝑡)) 𝑑𝑡

𝑚𝑒𝑖̃

+ ∫ (λ𝑚𝑖𝑛 − 1) ∙ (𝐶𝑀𝑒𝑖(𝑡))𝑑𝑡

𝑚𝑓𝑖̃

+ ∫ (𝐶𝑀𝑒𝑖(𝑡))𝑑𝑡

𝑚𝑓𝑖̃

+ ∫ (𝜆𝑚𝑖𝑛 − 1) ∙ (𝐶𝑀𝑓𝑖(𝑡) − 𝐶𝑀𝑓𝑖(𝑡)) 𝑑𝑡

𝑚𝑓𝑖̃

+ ∫ [𝐾 ∙ (𝐶𝑀𝑓𝑖(𝑡)) − 𝑤ℎ] 𝑑𝑡

𝑚

0

(A63)

The expression in (A63) can be re-arranged, such that:


[(𝜆(𝑡)) ∙ (𝑋𝑒 ∙ 𝑃𝑒𝑖(𝑡) − 𝑋𝑒 ∙ 𝑤𝑒𝑖)


] 𝑑𝑡

𝑚

0

= ∫ (𝜆𝑚𝑎𝑥 − 1) ∙ (𝐶𝑀𝑒𝑖(𝑡) − 𝐶𝑀𝑓𝑖(𝑡)) 𝑑𝑡

𝑚𝑒𝑖̃

+ ∫ (𝐶𝑀𝑒𝑖(𝑡))𝑑𝑡

𝑚𝑒𝑖̃

+ ∫ (λ𝑚𝑖𝑛 − 1) ∙ (𝐶𝑀𝑒𝑖(𝑡) − 𝐶𝑀𝑓𝑖(𝑡)) 𝑑𝑡

𝑚𝑓𝑖̃

+ ∫(𝐶𝑀𝑒𝑖(𝑡))𝑑𝑡

𝑚𝑓𝑖̃

+ ∫ (𝜆𝑚𝑎𝑥 − 1) ∙ (𝐶𝑀𝑓𝑖(𝑡)) 𝑑𝑡

𝑚𝑒𝑖̃

+ ∫ (𝜆𝑚𝑖𝑛 − 1) ∙ (𝐶𝑀𝑓𝑖(𝑡)) 𝑑𝑡

𝑚𝑓𝑖̃

+ ∫ [𝐾 ∙ (𝐶𝑀𝑓𝑖(𝑡)) − 𝑤ℎ] 𝑑𝑡

𝑚

0

(A64)


Then, using the definitions of 𝐼𝐶𝑀𝐹𝑒𝑖 and 𝐼𝐶𝑀𝐹𝑓𝑖 in (17) and (18), respectively, and

expanding 𝐾:


[(𝜆(𝑡)) ∙ (𝑋𝑒 ∙ 𝑃𝑒𝑖(𝑡) − 𝑋𝑒 ∙ 𝑤𝑒𝑖)


] 𝑑𝑡

𝑚

0

= −𝑚 ∙ (1 − 𝜆𝑚𝑎𝑥) ∙ 𝐼𝐶𝑀𝐹𝑒𝑖 + ∫ (𝐶𝑀𝑒𝑖(𝑡))𝑑𝑡

𝑚𝑒𝑖̃

− ∫(1 − 𝜆𝑚𝑎𝑥) ∙ (𝐶𝑀𝑓𝑖(𝑡)) 𝑑𝑡

𝑚𝑒𝑖̃

+𝑚 ∙ (1 − λ𝑚𝑖𝑛) ∙ 𝐼𝐶𝑀𝐹𝑓𝑖 + ∫(𝐶𝑀𝑒𝑖(𝑡))𝑑𝑡

𝑚𝑓𝑖̃

− ∫(1 − 𝜆𝑚𝑖𝑛) ∙ (𝐶𝑀𝑓𝑖(𝑡)) 𝑑𝑡

𝑚𝑓𝑖̃


(A65)

Re-arranging (A65), we get:


[(𝜆(𝑡)) ∙ (𝑋𝑒 ∙ 𝑃𝑒𝑖(𝑡) − 𝑋𝑒 ∙ 𝑤𝑒𝑖)


] 𝑑𝑡

𝑚

0

= 𝑚 ∙ (1 − λ𝑚𝑖𝑛) ∙ 𝐼𝐶𝑀𝐹𝑓𝑖 − 𝑚 ∙ (1 − 𝜆𝑚𝑎𝑥) ∙ 𝐼𝐶𝑀𝐹𝑒𝑖 + ∫ (𝐶𝑀𝑒𝑖(𝑡))𝑑𝑡

𝑚

0

− ∫ (𝐶𝑀𝑓𝑖(𝑡)) 𝑑𝑡

𝑚

0

+ ∫ 𝜆𝑚𝑖𝑛 ∙ (𝐶𝑀𝑓𝑖(𝑡)) 𝑑𝑡

𝑚𝑓𝑖̃

+ ∫ 𝜆𝑚𝑎𝑥 ∙ (𝐶𝑀𝑓𝑖(𝑡)) 𝑑𝑡

𝑚𝑒𝑖̃


(A66)

But we know that 𝐶𝑀𝑓𝑖(𝑡) = 𝐶𝑀𝑓𝑖 because we assumed constant fertilizers price and variable

costs in a given year 𝑖. Under these conditions, (A66) reduces to:




𝑚

0

= 𝑚 ∙ (1 − λ𝑚𝑖𝑛) ∙ 𝐼𝐶𝑀𝐹𝑓𝑖 − 𝑚 ∙ (1 − λ𝑚𝑎𝑥) ∙ 𝐼𝐶𝑀𝐹𝑒𝑖 + 𝑚 ∙ 𝐶𝑀𝑒𝑖 − 𝑚 ∙ 𝑤ℎ

(A67)

Substituting the results of (A66) and (A49) into (A50), we get:

𝐶𝐹𝐿𝑖 = −𝑁ℎ ∙ [𝑆𝐽ℎ𝑖 + λ𝑚𝑎𝑥 ∙ 𝑋𝑒 ∙ 𝑆𝐽𝑒𝑖 + (1 − λ𝑚𝑖𝑛) ∙ 𝑋𝑎 ∙ 𝑆𝐽𝑎𝑠𝑖 + 𝐾 ∙ 𝑋𝑓 ∙ 𝑆𝐽𝑓𝑖 ]

+𝐶𝐹 ∙ 𝑥𝑖 ∙ 𝑁ℎ ∙ 𝑚 ∙ {(1 − λ𝑚𝑖𝑛) ∙ 𝐼𝐶𝑀𝐹𝑓𝑖 − (1 − λ𝑚𝑎𝑥) ∙ 𝐼𝐶𝑀𝐹𝑒𝑖


(A68)


Then, substituting (A68) and (A48) into (A1), the NPV becomes:

𝑁𝑃𝑉 ($) =


𝑇

𝑖=1

+ ∑ 𝛾𝑖 ∙ 𝐶𝐹 ∙ 𝑥𝑖 ∙ 𝑁ℎ ∙ 𝑚 ∙ {(1 − λ𝑚𝑖𝑛) ∙ 𝐼𝐶𝑀𝐹𝑓𝑖 − (1 − λ𝑚𝑎𝑥) ∙ 𝐼𝐶𝑀𝐹𝑒𝑖


𝑇

𝑖=1

−𝑁ℎ ∙ [𝑆𝑃ℎ + λ𝑚𝑎𝑥 ∙ 𝑋𝑒 ∙ 𝑆𝑃𝑒 + (1 − λ𝑚𝑖𝑛) ∙ 𝑋𝑎 ∙ 𝑆𝑃𝑎𝑠 + 𝐾 ∙ 𝑋𝑓 ∙ 𝑆𝑃𝑓]

(A69)



𝑇

𝑖=1

+ ∑ 𝛾𝑖 ∙ 𝐶𝐹 ∙ 𝑥𝑖 ∙ 𝑁ℎ ∙ 𝑚 ∙ {(1 − λ𝑚𝑖𝑛) ∙ 𝐼𝐶𝑀𝐹𝑓𝑖 − (1 − λ𝑚𝑎𝑥) ∙ 𝐼𝐶𝑀𝐹𝑒𝑖


𝑇

𝑖=1

−𝑁ℎ ∙ [𝑆𝑃ℎ + λ𝑚𝑎𝑥 ∙ 𝑋𝑒 ∙ 𝑆𝑃𝑒 + (1 − λ𝑚𝑖𝑛) ∙ 𝑋𝑎 ∙ 𝑆𝑃𝑎𝑠 + 𝐾 ∙ 𝑋𝑓 ∙ 𝑆𝑃𝑓] > 0

(A70)


−∑ 𝛾𝑖 ∙ [𝑆𝐽ℎ𝑖 + λ𝑚𝑎𝑥 ∙ 𝑋𝑒 ∙ 𝑆𝐽𝑒𝑖 + (1 − λ𝑚𝑖𝑛) ∙ 𝑋𝑎 ∙ 𝑆𝐽𝑎𝑠𝑖 + 𝐾 ∙ 𝑋𝑓 ∙ 𝑆𝐽𝑓𝑖]𝑇

𝑖=1


+

∑ 𝛾𝑖 ∙ 𝑥𝑖 ∙ {(1 − λ𝑚𝑖𝑛) ∙ 𝐼𝐶𝑀𝐹𝑓𝑖 − (1 − λ𝑚𝑎𝑥) ∙ 𝐼𝐶𝑀𝐹𝑒𝑖

+[𝑋𝑒 . 𝑃𝑒𝑖 − 𝑋𝑒 . 𝑤𝑒𝑖] − 𝑤ℎ𝑖}𝑇

𝑖=1


−[𝑆𝑃ℎ + λ𝑚𝑎𝑥 ∙ 𝑋𝑒 ∙ 𝑆𝑃𝑒 + (1 − λ𝑚𝑖𝑛) ∙ 𝑋𝑎 ∙ 𝑆𝑃𝑎𝑠 + 𝐾 ∙ 𝑋𝑓 ∙ 𝑆𝑃𝑓]


> 0

(A71)

−𝑗ℎ − λ𝑚𝑎𝑥 ∙ 𝑋𝑒 ∙ 𝑗𝑒 − (1 − λ𝑚𝑖𝑛) ∙ 𝑋𝑎 ∙ 𝑗𝑎𝑠 − 𝐾 ∙ 𝑋𝑓 ∙ 𝑗𝑓

+(1 − λ𝑚𝑖𝑛) ∙ 𝐼𝐶𝑀𝐹𝑓 − (1 − λ𝑚𝑎𝑥) ∙ 𝐼𝐶𝑀𝐹𝑒 + [𝑋𝑒 . 𝑃𝑒 − 𝑋𝑒 . 𝑤𝑒] − 𝑤ℎ

−𝑐ℎ − λ𝑚𝑎𝑥 ∙ 𝑋𝑒 ∙ 𝑐𝑒 − (1 − λ𝑚𝑖𝑛)𝑋𝑎 ∙ 𝑐𝑎𝑠 − 𝐾 ∙ 𝑋𝑓 ∙ 𝑐𝑓 > 0

(A72)

[𝑿𝒆. 𝑷𝒆 − 𝑿𝒆. 𝑳𝑰𝑪𝒆]

+(𝟏 − 𝝀𝒎𝒊𝒏) ∙ [𝑰𝑪𝑴𝑭𝒇 − 𝑿𝒇 ∙ (𝒄𝒇 + 𝒋𝒇) − 𝑿𝒂 ∙ (𝒄𝒂𝒔 + 𝒋𝒂𝒔)]

−(𝟏 − 𝝀𝒎𝒂𝒙) ∙ [𝑰𝑪𝑴𝑭𝒆 − 𝑿𝒆 ∙ (𝒄𝒆 + 𝒋𝒆)]

+(𝟏 − 𝛌𝒎𝒊𝒏 − 𝑲) ∙ [𝑿𝒇 ∙ (𝒄𝒇 + 𝒋𝒇)] > 𝑳𝑪𝑶𝑯

(A73)

CHAPTER 3 — Appendix B: Cost Estimates for HECA 144

Appendix B: Cost Estimates for HECA

Table 3.B1: System prices of HECA per unit capacity

Subsystem Process Cost Unit Reference

Hydrogen

Subsystem

Coal & petcoke handling and

storage 22,083 $/(𝑡𝑜𝑛𝑛𝑒/𝑑𝑎𝑦) [36]

Gasification 11,528 $/(𝑘𝑔ℎ/ℎ) [36]

Air separation 6,663 $/(𝑘𝑔ℎ/ℎ) [36]

Syngas treatment 3,027 $/(𝑘𝑔ℎ/ℎ) [36]

Acid gas removal 5,732 $/(𝑘𝑔ℎ/ℎ) [36]

Sulfur recovery and trail gas

treatment 1,859 $/(𝑘𝑔ℎ/ℎ) [36]

Hydrogen separation 927 $/(𝑘𝑔ℎ/ℎ) [36]

Utilities and offsites 11,116 $/(𝑘𝑔ℎ/ℎ) [36]

Electricity

Subsystem Power production 1,045 $/(𝑘𝑊 𝑔𝑟𝑜𝑠𝑠) [36]

Ammonia

Subsystem

Ammonia production 3,863 $/(𝑘𝑔𝑎/ℎ) [37]

Ammonia storage 0.686 $/(𝑘𝑔𝑎) [38, 39]

Fertilizers

Subsystem

Urea production 1,643 $/(𝑘𝑔𝑢𝑟𝑒𝑎/ℎ) [40, 41]

UAN Production 1,802 $/(𝑘𝑔𝑈𝐴𝑁/ℎ) [40, 41]

CO2

Subsystem CO2 compression and drying 135 $/(𝑘𝑔𝑐/ℎ) [36]

Aggregating these figures for each subsystem, we then get:

Table 3.B2: System prices of HECA’s subsystems per unit capacity

Subsystem Cost Value Unit Reference

Hydrogen Subsystem 𝑆𝑃ℎ 44,894 $/(𝑘𝑔ℎ/ℎ) [36]

Electricity Subsystem 𝑆𝑃𝑒 1,045 $/(𝑘𝑊) [36]

Ammonia Subsystem

(without storage) 𝑆𝑃𝑎 3,863 $/(𝑘𝑔𝑎/ℎ) [37]

Ammonia Subsystem

(with storage) 𝑆𝑃𝑎𝑠 3,947 $/(𝑘𝑔𝑎/ℎ) [37-39]

Fertilizers Subsystem 𝑆𝑃𝑓 4,435 $/(𝑘𝑔𝑢𝑟𝑒𝑎/ℎ) [40, 41]

CO2 Subsystem 𝑆𝑃𝑐 135 $/(𝑘𝑔𝑐/ℎ) [40, 41]


Table 3.B3: Yearly fixed-operating costs of HECA as fraction of capacity costs

Subsystem Process Cost

(% of capacity cost) Reference

Hydrogen

Subsystem

Coal & petcoke handling and

storage 3% [36]

Gasification 4.8% [36]

Air separation 3% [36]

Syngas treatment 4.8% [36]

Acid gas removal 3% [36]

Sulfur recovery and trail gas

treatment 3% [36]

Hydrogen separation 3% [36]

Utilities and offsites 2.1% [36]

Power

Subsystem Power production 6.1% [36]

Ammonia

Subsystem

Ammonia production 6.1% [estimated]

Ammonia storage 6.1% [estimated]

Fertilizers

Subsystem

Urea production 6.1% [estimated]

UAN Production 6.1% [estimated]

CO2

Subsystem CO2 compression and drying 3% [36]

Again, by aggregating these figures for each subsystem, we get:

Table 3.B4: Yearly fixed-operating costs of HECA Subsystems per unit capacity

Subsystem Cost Value Unit Reference

Hydrogen Subsystem 𝑆𝐽ℎ 1,514 ($ 𝑦𝑟⁄ ) (𝑘𝑔ℎ/ℎ)⁄ [36]

Electricity Subsystem 𝑆𝐽𝑒 63 ($ 𝑦𝑟⁄ )/(𝑘𝑊) [36]

Ammonia Subsystem

(without storage) 𝑆𝐽𝑎 234 ($ 𝑦𝑟⁄ )/(𝑘𝑔𝑎/ℎ) [estimated]

Ammonia Subsystem

(with storage) 𝑆𝐽𝑎𝑠 239 ($ 𝑦𝑟⁄ )/(𝑘𝑔𝑎/ℎ) [estimated]

Fertilizers Subsystem 𝑆𝐽𝑓 268 ($ 𝑦𝑟⁄ )/(𝑘𝑔𝑢𝑟𝑒𝑎/ℎ) [estimated]

CO2 Subsystem 𝑆𝐽𝑐 4 ($ 𝑦𝑟⁄ )/(𝑘𝑔𝑐/ℎ) [36]


Table 3.B5: Prices of input commodities and services for HECA

Input Cost Unit Reference

Coal (sub-bituminous) 60 $/𝑡𝑜𝑛𝑛𝑒 (personal communication)

Petcoke 50 $/𝑡𝑜𝑛𝑛𝑒 (personal communication)

Electricity (average) 29.5 $/MWh [29]

Selexol™ 0.00376 $/𝑘𝑔ℎ [36]

Flux 0.00432 $/𝑘𝑔ℎ [36]

Catalysts 0.00769 $/𝑘𝑔ℎ [36]

Chemicals 0.00601 $/𝑘𝑔ℎ [36]

Waste-water treatment 0.00928 $/𝑘𝑔ℎ [36]

Using this data, as well as the auxiliary loads in Table 3.2 of this study, we can find the

variable costs of HECA:

Table 3.B6: Yearly-averaged variable costs Cost of HECA per unit of production

Subsystem Input Cost Unit

Hydrogen Subsystem

Coal & petcoke 0.443 $/𝑘𝑔ℎ

Flux 0.00338 $/𝑘𝑔ℎ

SelexolTM

0.00294 $/𝑘𝑔ℎ

Catalyst 0.00602 $/𝑘𝑔ℎ

Other chemicals 0.00471 $/𝑘𝑔ℎ

Waste-water treatment 0.00727 $/𝑘𝑔ℎ

Auxiliary power 0.173 $/𝑘𝑔ℎ

Power Subsystem Auxiliary power 0.00119 $/𝑘𝑊ℎ

Ammonia Subsystem Auxiliary power 0.00644 $/𝑘𝑔𝑎

Fertilizers Subsystem Auxiliary power 0.00836 $/𝑘𝑔𝑢𝑟𝑒𝑎

CO2 Subsystem Auxiliary power 0.00331 $/𝑘𝑔𝑐

Table 3.B7: Prices of HECA end-products

End-Product Price Unit Reference

Urea 385 $/𝑡𝑜𝑛𝑛𝑒 (personal communication)

UAN 315 $/𝑡𝑜𝑛𝑛𝑒 (personal communication)

Electricity

(yearly-averaged) 29.5 $/𝑀𝑊ℎ [29]

CO2 25 $/𝑡𝑜𝑛𝑛𝑒 (personal communication)

147

Chapter 4

Decision Analytic Modeling of the Five

Forces in Competitive Strategy

1 Introduction

Despite apparent differences among industries, all firms compete for profit. Competitive

strategy, pioneered by Michael E. Porter since 1979, explains how an organization facing

competition can achieve superior profitability within its industry. Porter identifies five forces

that shape competition in relatively stable industries. Relying on Industrial Organization

theory and using examples from representative industries, he describes what causes each of

the five competitive forces to be strong or weak, and he explains that an incumbent business

gains competitive advantage by positioning where all five forces are weakest [1, 2]. Over the

years, Porter’s five forces framework (FF) has generated valuable strategic insights, has

inspired strategic victories for firms and businesses, and has occupied a permanent position in

strategic management classes and business schools’ curricula. However, despite its remarkable

contribution to competitive strategy, the framework has been mostly applied qualitatively and

deterministically [3, 4, 5]. Few systematic methodologies have been developed to guide the

quantification and operationalization of the competitive forces in real life, and no sufficient

attention has been given to the uncertain and interdependent nature of these forces and their

economic implications [6, 7].


To address these issues, this work proposes decision analysis (DA) as a method to model and

apply the five forces strategic framework in a specific industry and by a specific firm. As a

normative application of decision theory, DA describes how decision-makers, facing many

alternatives and uncertainties, should make an irrevocable allocation of resources to maximize

their utility. In the world of competitive strategy, we propose a decision analytic modeling of

the five forces, hereby referred to as DAFF. This modeling approach describes how an

executive or a manager, facing multiple positioning alternatives and uncertain competitive

powers, should commit to an irrevocable allocation of resources in order to maximize

profitability. Put differently, DAFF uses DA techniques and tools to link the firms’

positioning alternatives with uncertain market competition and economics, in accordance with

the pillars of Industrial Organization. The result is a thorough and quantitative model that

managers can use to evaluate the profitability of a specific industry, to properly position their

business in the industry, and to predict and shape the future of that industry.

Both FF and DA have been used to help firms improve their strategic performance in their

respective industries [8, 9]. By applying the FF theory using the DA method, this work adds

value to both fields of competitive strategy and decision analysis. On one hand, DAFF allows

strategists to quantify the five competitive forces, assess the significance of uncertainty in the

forces and their underlying drivers, capture and exploit interdependencies and interactions

among the forces, link the forces to measurable economic indicators, and specify the decisions

that firms or business units can actually make to influence competition or deter its threat. On

the other hand, the same model allows decision analysts to account for all competitive forces

and their drivers, understand the causal relationships in a competitive industry, and ensure that

the modeling of a firm’s decision problem is consistent with the mature literature on

competitive strategy and competitive advantage.

To explain DAFF and develop its models, the rest of this Chapter is organized as follows.

Section 2 starts by offering a brief formal introduction to both decision analysis and the five

forces strategic framework. Then, as the main motivation behind this work, we highlight key

considerations which have sometimes been neglected or oversighted when implementing the

five forces framework, and we provide a brief overview of previous endeavors aiming to

enhance the application of FF. Center to this study, we describe how DA can augment

CHAPTER 4 — Theoretical Background 149

previous work, accounting for all essential attributes of FF and further improving its

operationalization and implementation in real life. Section 3 provides a detailed description of

how to develop a DAFF model, including how to use DA’s building blocks of uncertainty,

decision, and value to model the five competitive forces and industry-specific factors, the

economics of an industry or a firm, and the firm’s competitive actions. Then, we explain how

a firm can use DAFF to operationalize the two main objectives of competitive strategy:

positioning in and reshaping the industry. Section 4 provides a three-step procedure on how

to model the industry-positioning objective, highlighting the benefits of the decision analytic

approach in each step. Subsequently, Section 5 lays out a conceptual roadmap to modeling the

industry-reshaping objective by augmenting the DAFF model from Section 4. Sections 6 and 7

provide additional guidance on DAFF. While Section 6 discusses additional modeling best

practices, Section 7 highlights the ability of DA to uphold and characterize many of Porter’s

strategic insights that stem from, yet extend beyond, the FF framework. Finally, Section 8

concludes this Chapter by summarizing the main modeling aspects and advantages of DAFF

and then offering some ideas for future work.

2 Theoretical Background

2.1 Decision Analysis

Decision analysis (DA) is a logical procedure that balances the various attributes of a decision

problem. Sometimes, decisions are difficult because they require making trade-offs among

several considerations, some of which are hard to observe, measure, or quantify. Other reasons

that contribute to the complexity of decisions include uncertainty, the lack of clear

information, the lack of sufficient alternatives, and conflicting preferences. Focusing on

strategic decisions, DA comprises of techniques and tools that allow a decision-maker to

achieve clarity of action in her decision, and even more fundamentally, to achieve clarity of

thought. In that regard, a key distinction in DA is the difference between the quality of a

decision and the quality of its outcome; DA emphasizes that the quality of a decision is

determined when making a decision, hence before the outcome is revealed. Put differently, the


decision should never be judged retrospectively based on its outcome, for a good decision may

still result in a bad outcome due to unknown information at the time of making the decision.

DA enables making good strategic decisions [10].

An essential DA foundation is the decision basis, which decomposes every decision situation

into three fundamental elements: alternatives, information, and preferences [10]. The

alternatives describe the options and choices available to the decision-maker at the time of

making a specific decision; this element addresses what the decision-maker can do. For

example, in the automobile industry, an auto manufacturer who wants to launch a new

environmentally friendly car may need to decide between two feasible alternatives: an

[electric car] or a [hybrid car].

Information describes the knowledge that the decision-maker has about the various aspects of

the decision situation. This element addresses what the decision-maker knows, and it

includes both certain information that the decision-maker knows for sure as well as uncertain

information that the decision-maker lacks completely or is unsure about. DA pays special

attention to uncertainty, which is explicitly accounted for in the decision problem. To help

the decision-maker think about uncertainty, DA models uncertainty via two means. First, the

decision-maker is invited to identify all possible realizations of each uncertainty, referred to as

degrees. Then, the decision-maker assigns a probability to each degree, depending on her

beliefs about the likelihood of it actually happening. The degrees should be mutually

exclusive and collectively exhaustive, so only one degree could be realized, and the

probability on all degrees should add up to one. To continue the car example, suppose that the

automaker is unsure about its consumers’ driving habits, which is an important consideration

in deciding what type of car to produce. The consumer’s driving habits become an uncertainty

in this case, and the automaker characterizes this uncertainty using only two degrees: {≤ 100

miles short trips; > 100 miles long trips}. Then, based on market research and other available

yet imperfect information, the automaker assigns a 0.7 probability that consumers will use the

car for short trips and a probability of 0.3 that consumers will drive the car on long trips.

The final element in the decision basis is preferences, which address what the decision-

maker wants. Preferences describe what the decision-maker mostly cares about and thus is


trying to optimize. They are usually presented as monetary values that the decision-maker

assigns to each potential combination of alternatives and uncertainties, called prospects. In

our example, the automaker is mostly concerned about the net present value (NPV) for the

new line of cars it intends to launch, so it assigns an NPV figure for each of the following

prospects: {electric car, short trips}, {hybrid car, long trips}, {hybrid car, short trips},

{electric car, long trips}. The decision-maker then chooses the alternative that achieves the

highest probability-weighted-average value.

On top of the three elements of the decision basis, three additional elements should be

available in every decision problem to ensure good decision quality. The first is a proper

framework, which refers to the context within which the decision problem should be

evaluated. Choosing the framework too broadly or too narrowly results in getting the right

answer to the wrong problem, which is rarely useful. In our car example, the automaker is not

deciding on whether or not to produce a new car (this framework is too broad) or what type of

battery the car should have (this framework is too narrow). The proper framework in this case

already assumes that an environmentally friendly car will be produced, and it postpones the

decision on the battery type till later. The second additional element for good decision quality

is proper logic, which dictates the relationship between the decisions, uncertainties, and

values. Much of the analytical calculations and algorithms needed to solve a decision problem

are governed by Bayesian probability logic. Finally, a good decision requires a clear

commitment to action; there is no point in spending time and resources analyzing a decision

unless the decision-maker is committed to implementing the outcome solution to the decision

problem [10].

Among the various tools that DA uses to model a decision problem, we focus on two tools

called Bayesian networks and decision diagrams [11]. Both tools can model multiple

decisions, multiple uncertainties, and a single value, connected as a network of nodes. While

the decision is generally presented in rectangle-like node, the uncertainty is presented in an

oval-like node, and the value function is presented as a diamond-like node. Sometimes, certain

information is also explicitly modeled to ease the analysis. Certain information is described as

“deterministic” and is presented in oval-like nodes with double-lined borders to distinguish it

from regular single-line ovals presenting uncertainties. A schematic of the various shapes


representing decisions, uncertainties, deterministic information, and values is presented in

Figure 4.1. To reiterate, each decision node incorporates multiple alternatives; each

uncertainty node incorporates multiple degrees; and the value node assigns a value to each

prospect. The process of solving a decision diagram to find the optimal alternative under

uncertainty is mature and well-documented in literature [10, 11, 12, 13], and several

commercial software packages are available for that purpose [14, 15].

Figure 4.1: Representation of the nodes in a decision diagram

2.2 The Five Forces that Shape Competition

The Five Competitive Forces that Shape Strategy by Michael Porter is one of the most well-

known papers in business strategy. In this updated version of his work on competitive

strategy, Porter explains that competition for profits in stable industries extends beyond direct

rivals to include four other players in the marketplace: buyers, suppliers, potential new

entrants, and substitutes. Therefore, despite apparent differences in industries’ structures and

functions, the competition within each industry is shaped by five competitive forces, depicted

in Figure 4.2: the bargaining power of buyers, the bargaining power of suppliers, the threat of

new entrants, the threat of substitutes, and rivalry among incumbents. For brevity, we refer to

these forces throughout the rest of this study as Buyers, Suppliers, New Entrants, Substitutes,

and Rivals, respectively. The strength of each of the five forces is shaped by a set of

generalizable market drivers and another set of industry-specific factors, both of which have

been continuously discussed, refined, and updated by Porter over the years [1, 2, 8, 16]. In that

regard, Porter’s recent work clearly identifies four factors: industry growth rate, technology

and innovation, government regulations and policies, and complementary products and

services. Again, for brevity, we refer to these factors as Growth, Technology, Regulation, and

Complements, respectively. While the five forces shape the underlying structure of any

industry, the four factors are visible attributes of a specific industry.

Uncertainty ValueDecisionDeterministic Information


Figure 4.2: Porter’s five forces framework

Porter’s five forces framework (FF) advises the current incumbents to reduce the power of all

forces in order to increase their profitability. This means that firms should try to limit the

bargaining power of buyers and suppliers, decrease the threat of new entrants and substitutes,

and reduce the intensity of destructive rivalry with direct competitors. However, unlike the

forces, stronger factors may increase or decrease profitability. Depending on the unique

aspects of each industry, each factor influence the five forces in a way that can either increase

or decrease their strength and thus profitability [2]. The role of the internet in the automobile

industry is a good example [17]. As one of the most prominent technological innovations in

the past decade, the internet has impacted automakers in two opposing ways. On one hand,

direct online purchases have lowered the Buyers power by weakening the role of dealers as

intermediate channels [18]. On the other hand, car valuation sites such as Kelly Blue Book

have increased the Buyers power by allowing end-customers to effortlessly compare and shop

for cars at multiple venues [19]. Another important thing to note is that a factor may affect

each force either directly or indirectly through one of the force’s drivers. For example, in the

solar industry, regulation may directly decrease the threat of Substitutes by mandating the

deployment of a specific solar production capacity within a specific region and timeline [20].

Alternatively, regulation may indirectly decrease the bargaining power of Buyers by allowing

BuyersHow much bargaining power

do buyers have?

SuppliersHow much bargaining power

do supplier bear?

New EntrantsWhat is the threat of new entrants into the market?

SubstitutesWhat is the threat of substitute

products or services?

RivalsWhat rivalry exists among

present incumbents?


solar firms to impose a high early-termination penalty on their solar-panel leases, which

effectively increases the switching costs for customers [21].

By analyzing the power of the competitive five forces, their underlying drivers, and the

industry-specific factors, a firm can assess the level and distribution of profitability in the

industry. In turn, such assessment can guide and inform the design of effective strategies that

improve the firm’s performance relative to the industry average. In that regard, the FF

framework facilitates the fulfillment of three strategic objectives by the firm: positioning

where competition is weakest, uncovering new favorable positions by predicting and

exploiting future changes in the industry structure, and becoming an industry leader by

reshaping future changes in the industry structure. Successful firms usually strive to fulfill one

or more of these objectives [2].

2.3 Decision Analytic Approach to the Five Forces

Despite their importance, multiple key considerations have sometimes been downplayed,

oversighted, or neglected in the daily conduct of FF competitive analyses within businesses,

corporations, and consulting practices. Primarily, they include: standardization of the analysis;

role of uncertainty; relation between the competitive forces and economic performance;

relation between the industry’s competitive forces and the firm’s strategic actions;

interdependence among the forces, drivers, and factors; and quantification. DAFF offers one

way to effectively account for and uphold all these considerations in the FF applications,

building on and augmenting a diverse set of previous research endeavors that have aimed to

enhance FF’s operationalization.

Standardization of the analysis: attempting to enhance the clarity and standardization of the

competitive forces’ analyses, Dobbs proposed the design of generalizable templates for FF

practitioners [6]. As will become evident, the proposed DAFF approach expands this effort by

developing generalizable models that represent and incorporate not only the five competitive

forces but also the other building blocks of the FF framework, including: the industry-specific

factors, the economic parameters that characterize the performance of an overall industry or of

a specific firm, and the set of competitive actions and activities taken by the firm.


Role of uncertainty: the vast majority of FF applications discuss the power of the competitive

forces as certain happening facts, assuming that industry players have perfect clairvoyance on

competition. Such an approach is problematic for two main reasons. First, it does not account

for the reality that, due to asymmetric market information, different players may have different

and uncertain beliefs about the strength of the competitive drivers or factors in their present

industry. More importantly, it does not allow predicting the evolution of the competitive

forces and factors in the future. As Magretta nicely explains in her book [8]: “Porter examines

… companies, after the fact, and asks, what explains their success?” This is a clear recognition

that the FF framework has been mostly used to retrospectively assess a company’s

performance in the past. The main question then becomes: how can we use this framework to

proactively assess a company’s performance in the future?

DA resolves this issue by modeling the competitive forces, drivers, and factors – both current

and future – as uncertainties. Accounting for uncertainty is not a new endeavor in the field of

business strategy. For example, cautioning against the dangers of ignoring or misrepresenting

uncertainty, Courtney et al. identifies multiple levels of uncertainty in strategy and proposes

multiple ways to address them [22]. Even earlier work by Ghemawat recognizes the need for

modeling uncertainty in strategic decisions, primarily because these decisions may involve

irrevocable investments and therefore may require an irreversible allocation of nontrivial

resources [23]. In their work on “co-opetition”, Brandenburger and Nalebuff also characterize

uncertainty as “fog”, which, depending on the specific business setting, may be either

beneficial or harmful [24]. In fact, Porter himself notes the significance of uncertainty

assessment in strategy [25]; in the updated publication The Five Competitive Forces that

Shape Strategy, Porter uses the words “likely” and “less likely” multiple times throughout the

manuscript, which validates the inability to perfectly characterize current or future competitive

landscapes [2]. By modeling unknown competitive information as uncertainties, DA helps

decision-makers avoid the false logic that Magretta warns against: “I can’t predict the future.

Strategy requires a prediction of the future. Therefore, I cannot commit to strategy” [8].

Strategy does not require predicting the future. Strategy requires taking an insightful action

about the future after considering all available information, which is exactly what DA allows


doing. To that end, DA and Porter’s FF framework strongly support a common conclusion:

not knowing the future is not a good excuse to dodge strategy.

Relation between the competitive forces and economic performance: we address two

considerations here. First, as clearly explained by Porter himself [2, 25] and further reflected

upon in competitive strategy literature [26, 27], the five forces framework depicts the

economic performance of the firm as a function of competition in the industry. FF literature

explains that weaker competitive forces result in lower costs and/or higher prices, which in

turn result in higher profitability. However, it is unclear what exactly defines a force as

“strong” or “weak”, and to what extent each force can (and does) affect economic parameters

such as price or cost. To resolve this ambiguity, DA models the parameters characterizing the

economic performance (e.g. price and cost) as uncertainties, similar to the five forces.

Consistent with FF theory, the competitive forces’ uncertainties interact directly with the

economic parameters’ uncertainties, capturing all available – albeit imperfect – market

information based on the decision-maker’s personal or corporate intelligence. In addition, DA

modeling allows distinguishing between the effect of the forces on the economics of the

overall industry and the effect of the forces on the economics of a specific firm.

The second consideration to address is FF’s emphasis on maximizing profitability [2].

Because DA models use one value metric to articulate preferences and rank prospects,

profitability should be clearly defined as a single quantitative and measurable metric. In that

regard, profitability may be expressed as “unit-profit-margin” (PM) or “return on invested

capital” (ROIC) as advised by Porter [8], net present value, earnings before tax, or otherwise.

Relation between the industry’s competitive forces and the firm’s strategic actions: when it

comes to strategic decisions, we first highlight that FF and DA are normative not descriptive;

both approaches advise what a firm’s optimal strategy should be rather than describing or

justifying what its strategy actually is. With that in mind, FF identifies three competitive

objectives that each firm should strive to fulfil: positioning in the industry, predicting and

exploiting future industry change, and reshaping the industry [2]. However, pursuing each of

these broad objectives requires a series of specific firm actions – and thus decisions – that are

hard to capture in generalizable – and intentionally simple – strategic frameworks. To that


end, DA augments FF by helping the firm develop a clear recipe of the specific choices it can

make and the specific conditions it can control in order to achieve superior performance. DA

models allow processing as many decisions as necessary, with a specific set of alternatives for

each decision. The DA models then link all these decisions to one another and to the

competitive forces, thus establishing a clear roadmap for the firm to fulfill one or more of the

aforementioned competitive objectives. Indeed, as the experimental work by Song et al.

shows, analyzing this link between the firm’s competitive decisions and the industry’s

competitive forces is essential to understand and predict what, why, and how specific

strategies can prevail in specific industries [28].

Interdependence among the forces, drivers, and factors: while the FF literature provides a

careful explanation of each competitive force and its possible drivers, the interaction among

forces is rarely emphasized [2, 8, 29]. The illustration of the FF framework in Figure 4.2

shows two-way connections between rivalry and the other four forces. However, it is not clear

why these four forces do not directly interact, and how the different forces’ drivers relate to

one other. Most significantly, a careful analysis of FF literature shows that some drivers are in

fact shared among multiple forces, yet those drivers are sometimes introduced separately for

each force, and the implications of their interdependence are not fully explored. Recently,

some research efforts have relied on Analytical Hierarchy Process (AHP) and Analytical

Network Process (ANP) models to investigate this competitive interaction. However, such

endeavors remain limited in scope, focusing either on a narrow list of competitive drivers or

on specific industries [7, 30, 31]. As a result, the majority of FF applications today fail to

capture the interactions not only between the five forces and the firm’s actions but also among

the five forces, their underlying drivers, and the industry-specific factors.

Augmenting the AHP/ANP research endeavors, DA modeling allows capturing multiple forms

of competitive interaction. By modeling the forces, drivers, and factors as a network of

uncertainties, DA naturally asserts some relevance – or lack thereof – between the various

elements of the competitive landscape. Relevance is an important concept in DA that stems

from ascribing Bayesian conditional probability to the degrees of the various uncertainties.

Similarly, DA allows decisions to directly influence uncertainties by modifying the

probability distribution over their degrees [10]. Relevance and influence allow not only


tracking but also quantifying the interactions among the uncertain competitive forces, drivers,

factors, and economic parameters, as well as between them and the firm’s strategic decisions.

We explain both concepts in more detail in the following sections.

Quantification: although Porter emphasizes the importance of quantifying the FF analyses, the

framework is usually applied qualitatively. With the aforementioned AHP/ANP studies being

a notable exception [7, 30, 31], most FF applications today tend to use fuzzy and simplistic

heuristics for assessing the strength of the forces [3, 4, 5]. Quantifying the FF framework

requires non-trivial effort to gather, analyze, and interpret market data. The first challenge lies

in quantifying uncertain or imperfect information related to the competitive forces, their

underlying drivers, and industry factors. The task becomes trickier when such quantification

has to capture the competitive interdependencies. Another challenge relates to quantifying the

impact of the forces on the profitability of the industry or that of a specific firm. Along those

lines, quantification should also cover how a firm’s action may change the state of an

uncertain force, driver, factor, or economic parameter. DA facilitates quantification in many

distinct ways. Primarily, DA assigns a numerical probability value to each uncertainty degree,

defines a quantitative metric – profitability in this case (e.g. ROIC) – to rank prospects, and

assigns a numerical value to each prospect. Also, to achieve clarity, DA encourages

quantifying the very definitions of the uncertainty degrees and decision alternatives. Referring

to the car example earlier, we see that the “consumer’s driving habits” uncertainty has two

layers of quantification: the degrees’ definitions {≤ 100 miles short trips; > 100 miles long

trips} and their probabilities (0.7 and 0.3) are both expressed numerically.

Broadly, DAFF helps map clear, quantitative, and economically sound connections between

the three main pieces of the competitive strategy puzzle: the firm’s strategic decisions, the

uncertain competitive forces and factors, and the overall as well as firm-specific profitability

in a given industry. The genius of Porter’s work stems from its bridging between economics

and business. Porter aimed to connect the firm’s competitive strategy to its financial

performance. Our work aims to preserve this connection, quantify it, and provide a clear

procedure on how to implement it. Put differently, Porter explained what this connection is;

we aim to augment his work by describing how this connection can be practically drawn.

Porter advised on where a business should be relative to its industry; we aim to explain how

CHAPTER 4 — Methodology: Developing the DAFF Models 159

the business can get there. Ultimately, our goal is to help managers achieve competitive

advantage by providing them with practical tools to model and analyze their decision-making

process for positioning in the industry, for predicting future industry change, or for reshaping

the industry to their advantage.

Here, we realize that the FF theory itself has faced some criticisms since its conception,

including its relatively “abstract”, “static”, and “highly analytical” nature; its “overstress of

macro-analysis at the industry level”; its “oversimplification of the industry value chains”; its

“failure to link directly to possible management actions”; its “overreliance on microeconomic

theory and economic jargon”; and its inability to capture the effects of the “digitalization,

globalization, and deregulation” in today’s economy [29, 32]. This work aims to show how

FF, when modeled using DA tools, can become easily analyzable and operational; can

generate useful insights at the level of both the industry and the individual firms; can capture

the complexities of the industry value chain; and can provide a clear recipe of managerial

strategic actions. In addition, our work illustrates how FF’s deep foundation in Industrial

Organization and microeconomic theory is, in fact, an advantage and a necessity for its

success. Put differently, the DAFF approach is not intended to refute, replace, or modify of the

FF theory. The DAFF models proposed in this study aim to enhance the practical

implementation of the FF framework rather than challenging the theory behind it.

With that in mind, we devote the next section to model the FF competitive framework using

decision analytic tools and techniques.

3 Methodology: Developing the DAFF Models

To build the DAFF models, we first need to extract and document all important terminology

related to the five forces strategic framework from literature. This task is accomplished by

relying primarily on two recent references: Michael Porter’s The Five Competitive Forces

That Shape Strategy [2] and Joan Magretta’s Understanding Michael Porter: Guide to

Competition and Strategy [8]. To the best of our ability, the FF scripts in both references were

carefully reviewed, translated into DA terms, and subsequently transformed into DAFF


models. Accordingly, this section shows how all key elements and essential terms of the FF

framework are categorized as decision-related, uncertainty-related, or value-related. The

five competitive forces, their underlying drivers, the industry-specific factors, and the

economic parameters are all modeled as uncertainty nodes; the firm’s competitive decisions

(or actions) are modeled as decision nodes; and the profitability of the overall industry or of a

specific firm is modeled as a value node.

3.1 Modeling the Competitive Forces, Drivers, and Factors

We start with the most prominent component of the DAFF model: the five forces. An

uncertainty node is used to model the power (strength) of each competitive force: Buyers,

Suppliers, Substitutes, New Entrants, and Rivals. Subsequently, we identify the drivers that

shape the power of each force. Similar to the forces, each driver is represented as an

uncertainty node. Our interpretation of the FF literature reveals asymmetrical sets of drivers

shaping each competitive force. Specifically, we identify 22 drivers for Buyers, 12 drivers for

Suppliers, 3 drivers for Substitutes, 45 drivers for New Entrants, and 21 drivers for Rivals. For

illustration, the 22 driver uncertainties that are relevant to the power of Buyers are shown in

Figure 4.3. Next, we realize that many of these drivers are shared among the forces. In fact,

we identify a total of 13 common drivers, each relevant to two or three of the competitive

forces. For example, as shown in Figure 4.4, Buyers share 10 drivers with other forces: one

with Substitutes, one with Rivals, six with New Entrants, and two with both Rivals and New

Entrants.

After modeling the forces and their underlying drivers as uncertainty nodes, these nodes are

connected using directed arrows to form a Bayesian network [33]. In DA, an arrow

connecting two uncertainties is called a relevance arrow, and it implies a probabilistic

dependency (relationship) between the uncertainties. The existence of an arrow shows that two

uncertainties may or may not be relevant, while the lack of an arrow indicates that the two

uncertainties are definitely irrelevant [10]. Based on our interpretation of any discussed

relevance in the FF literature, we model a set of relevance arrows among the forces and their

drivers. For example, Figure 4.5 shows the relevance arrows for the force of Buyers. We


Figure 4.3: Identifying the underlying drivers for the bargaining power of Buyers

Figure 4.4: Highlighting the shared underlying drivers for the bargaining power of Buyers

Relative dependence of the buyer on industry

product

Buyer’s switching cost between this

industry’s products

Product differentiation

among incumbents

Bargaining Power of Buyers

price sensitivity of

the buyer

Product cost as

fraction of the

buyer’s budget

concentration of buyers relative to

incumbents

Buyer’s threat to integrate backward

Ability to influence

buyers downstream

High fixed costs

Buyer’s need to trim

purchase cost of the

product

Profits earned by the buyer

Amount of cash available to the

buyer

Industry product pays for

itself

product leads to performance

improvement for buyer

product reduces costs for buyer

Number of

buyers

Volume of

purchase per buyer

Buyer’s switching cost

from this industry’s

products to substitutes

Willingness of price

discounting by

incumbents

Buyer’s ability to

alter product


retain trained

employees


alter production

system


product




among incumbents



the buyer

Product cost as

fraction of the

buyer’s budget


incumbents



buyers downstream

High fixed costs



product



buyer


itself




Number of

buyers

Volume of

purchase per buyer





discounting by

incumbents


alter product


retain trained

employees


alter production

system

Substitutes

Buyers

New Entrants

Rivals


Figure 4.5: Relevance arrows connecting the underlying drivers for the bargaining power

of Buyers

make a simplifying assumption – as DA analysts usually do – that undiscussed relevance

implies no relevance. To that end, the relevance arrows we show in this work represent a

foundational, yet not necessarily final, set of relevance relations among the forces and their

drivers. When using the proposed DAFF models in practice to assess a specific industry, some

relevance arrows may be added or subtracted; those adjusted arrows would indicate potential

relevance relations that may or may not be important in the considered industry.

Relevance arrows are usually assigned in the direction that makes it easiest for the decision-

maker to assess the uncertainties. As a matter of terminology, the uncertainty where the arrow

originates is a parent while the one where the arrow ends is a child. Exploiting this “family

tree” analogy, commonly used in DA [34], we refer to the parent of a parent uncertainty as a

grandparent uncertainty, to the parent of a grandparent uncertainty as a great-grandparent

uncertainty, and so on. Accordingly, a child node may have many ancestral nodes.

The probability distribution in the child uncertainty node is conditioned on that in the parent

uncertainty node. Probability computations are most intuitive when the parent node is a


product




among incumbents



the buyer

Product cost as

fraction of the

buyer’s budget


incumbents



buyers downstream

High fixed costs



product



buyer


itself




Number of

buyers

Volume of

purchase per buyer





discounting by

incumbents


alter product


retain trained

employees


alter production

system

Substitutes

Buyers

New Entrants

Rivals


contributing cause to, or an underlying driver of, the child node. Adhering to this logic, we

model the five force uncertainties as children, and their underlying driver – mostly causal –

uncertainties as ancestors. The relevance arrows usually point from the drivers towards the

forces, as we demonstrate in Figure 4.5. A careful interpretation of the FF literature suggests

that not all drivers affect the forces equally, for some drivers are causal parents of other

drivers. Therefore, driver uncertainties are generally grouped in levels: driver uncertainty

nodes at the parent level have arrows pointing directly into the five forces; grandparent driver

nodes have arrows pointing into parent driver nodes; great-grandparent driver nodes have

arrows pointing into grandparent driver nodes; and so on.

As a side note, we briefly discuss what happens if the direction of a relevance arrow is flipped.

The causal reasoning is lost, so the probability assessment becomes harder. Nonetheless, by

DA rules, such assessment is still possible and should be equally robust. Referring to Figure

4.5, it might be hard – rather odd – for a manager to analyze the Product differentiation among

incumbents driver based on her knowledge (i.e. by conditioning it) on the Price sensitivity of

the buyer driver. That said, the manager may still infer something about the uniqueness of a

product by learning about the sensitivity of the buyer to its price; luxury-sports cars is a good

example: learning that a buyer is extremely price insensitive may increase the decision-

maker’s belief that the product is highly unique. Inference and arrow-flipping are mature

notions in DA [10], and they both stem from the Bayesian nature of relevance. Accordingly,

relevance is broader than and superior to causality because it allows two-way information flow

between uncertainties. Causality is nice to have to facilitate probability elicitation, but it is not

necessary for DAFF modeling.

In addition to the five forces and the 88 relevant drivers, the proposed DAFF adds one

uncertainty node for each of the four industry-specific factors: Growth, Regulations,

Technology, and Complements. A decision analytic interpretation of the FF literature reveals

that the factor uncertainties may share relevance arrows either with the forces directly or with

their underlying drivers. For example, Growth may be directly relevant to Suppliers power, as

well as to three drivers: Commitment of incumbents to retain and fight over market share,

Basis of competition, and Expected retaliation.


As a result of identifying and connecting the uncertainty nodes of the forces, drivers, and

factors, we generate two Bayesian networks, presented in Figures 4.6 and 4.7. We label the

diagram in Figure 4.6 Detailed Network because it maps the complete set of competitive

forces, drivers, and factors that appear in the two examined references of FF literature [2, 8].

On the other hand, the DAFF model in Figure 4.7 is labelled Simple Network because it

depicts only two ancestral levels of drivers for each force (parents and grandparents), thus

limiting the number of uncertainties that need to be assessed.

Here, we make two key modeling notes. First, the common drivers (shared among two or

more forces) in both the Detailed and Simple networks are distinguished in a separate level

above all other uncertainty nodes. This representation aims to highlight the role these common

nodes play in shaping the interactions among the forces. Second, because the goal of DAFF is

to achieve clarity, and due to the discrepancy in the number of and relevance among the

drivers and factors shaping each force, the decision-maker may indeed choose to expand or

reduce the number of analyzed competitive uncertainties. If the decision-maker is not able to

assess a specific driver or factor because it is too complex or vague, new parent or children

nodes can be added to facilitate the assessment of that uncertainty through relevance (and

conditional probability elicitation). Magretta emphasizes that “a good five forces analysis

allows [the decision-maker] to see through the complexity of competition” [8]; this is exactly

what these Bayesian networks aim to accomplish. In contrast, if the decision-maker deems

some drivers or factors unimportant or unnecessary for properly analyzing the competitive

landscape, those uncertainties can be removed from the Bayesian network. The latter option

should be exercised with great care, however, for all uncertainties and their relevance

relationships in Figures 4.6 and 4.7 are directly derived from the FF literature, whose findings

– according to Porter – are generalizable across industries.

After identifying and connecting the competitive uncertainties in a Bayesian network, we

transition to describe how to characterize the degrees of each uncertainty. To start, each driver

or factor uncertainty node may incorporate multiple degrees, so long as they are mutually

exclusive and collectively exhaustive (i.e. their probabilities add up to one). We make a

decision to customize the characterization of the degrees for both the drivers and the factors,

165

Figure 4.6: Detailed Network of the five forces, their underlying drivers, and industry-specific factors

Threat of new entrants

Barriers to entry

Expected retaliation

Size-independent advantages

for incumbents

Capital needed by new entrant

Network effects

Supply-side economies of scale for incumbent

Unequal access to

distribution channels for new entrant

Government regulations and

policies

Previous responses

by incumbents

Extent of Resources

available for incumbent

Capital availability

for new entrant

Relative Dependence of the buyer on industry

product



Bargaining Power of Suppliers

Concentration of suppliers relative to

incumbents

Fragmentation of industry

Relative dependence of suppliers

on this industry profits

Incumbent switching

costs between suppliers

Supplier switching

costs between

incumbents


among suppliers

Availability of

substitutes for what

the suppliers provideSuppliers

threat to integrate foreword

Threat of substitutes

Price-performance trade-off to this industry

product

Commitment of

incumbents to retain and

fight over market share

Industry growth

rate


among incumbents

Efficiency of expansion of Incumbents production

capacity



the buyer

Product cost as

fraction of the

buyer’s budget


incumbents



buyers downstream

Rivalry

Intensity of competition

Basis of competition

Ability to enforce

practices desirable for the industry

Extent of exit

barriers

High commitment to business

Technology and

innovation

Complementary products and

services

High fixed costs

High fixed costs and

low variable

costs

Perishable product



product


discounting by

incumbents

Extent of market

segments

Ability to meet the needs of multiple customer segments

Inability to read

incumbents’ signals

Lack of familiarity

with incumbents


Amount of cash

available for the buyer


itself

Product leads to

performance improvement

for buyer

Product reduces costs for

buyer

Number of buyers

Buyer’s trust in

incumbents

Well-established

brand

Proprietary technology

Prime location

Preferential access to resources

Cumulative in-house industry

experienceLimitation

to wholesale or retail channels

Reserved space on shelves

Control over

wholesale or retail channels

Efficiency of capital markets Fixed

facilities

Need to build

inventory

R&D spending

Marketing spending

Ability to command

better terms with

suppliers

Spread of fixed

costs over large

volumes

Excess cash

Borrowing power

Available production

capacity

Price advantage

for incumbents

Cost or quality

advantage for

incumbents

Need to bypass

incumbent existing

advantages by new entrant

Need to invest large

financial resources

by new entrant

Need for large-scale production

by new entrant

Consumer adoption

rate of product by

new entrant

Importance of non-

profit goals

Presence of an

industry leader

Importance of prestige in industry

Importance of job-

creation in industry

Buyer’s ability to retrain

trained employees

Buyer’s ability to alter product

specifications

Buyer’s ability to alter

production system

Incumbent’s investment in services by current suppliers

Incumbent’s production

location near current

suppliers

Number of

industries the

suppliers serve

Profits extracted

by suppliers

from other industries

Availability of

Substitutes for this

industry’s product

Volume of

purchase per

buyer

Buyer’s switching cost from this

industry’s products to substitutes

Substitutes

Buyers

New Entrants

Rivals

Suppliers Factors

166

Figure 4.7: Simple Network of the five forces, their underlying drivers, and industry-specific factors


Barriers to entry



for incumbents Capital

needed by new entrant

Network effects


Unequal access to


Government regulations and

policies

Previous responses

by incumbents

Extent of Resources available

for incumbent


for new entrant


product





incumbents




Incumbent switching


Supplier switching

costs between

incumbents


among suppliers

Availability of






product

Commitment of



Industry growth

rate


among incumbents


capacity



the buyer

Product cost as

fraction of the

buyer’s budget


incumbents



buyers downstream

Rivalry



Ability to enforce


Extent of exit

barriers


Technology and

innovation

Complementary products and

services

High fixed costs

Perishable product



product


discounting by

incumbents

Extent of market

segments

Ability to meet the needs of multiple

customer segments

Inability to read


Lack of familiarity

with incumbents

Number of buyers

Importance of non-

profit goalsIncumbent’s investment in services by current suppliers



suppliers

Number of

industries the

suppliers serve

Profits extracted

by suppliers


Availability of



Volume of

purchase per

buyer



Substitutes

Buyers

New Entrants

Rivals

Suppliers Factors


per the specific context of the competitive analysis. Because we encourage the degrees’

quantification, it is hard to uniformly define the degrees for all drivers and factors in all

industries, especially that drivers and factors are heavily shaped by the specific industry of

interest. For example, let us attempt to characterize the degrees of the driver Product cost as

function of buyer’s budget in two industries: the auto industry and the cellphone industry.

Clearly, the cost of a car is very different than that of a cellphone. Thus, assuming a simple

pay-upfront purchase, we may want to express “budget” as “monthly income” and define two

degrees {< 5% monthly income; ≥ 5% monthly income} for this driver in the cellphone

industry while expressing “budget” as “yearly income” and defining two degrees {< 10%

yearly income; ≥ 10% yearly income} for this driver in the auto industry. Similar rationale

applies when customizing the characterization of the degrees for the factor uncertainties;

Regulation overseeing grid-connectivity in the residential solar industry, for instance, is very

different that Regulation overseeing direct sales in the auto industry.

Finally, we explain how to characterize the degrees of the five forces. Each force uncertainty

is modeled with two degrees: {high; low}. The direct reason behind this characterization is

that Porter’s scripts (as well as Industrial Organization) provide no better option. Both FF

literature and mainstream FF applications always assess the forces’ power as “high” or “low”,

“strong” or “weak”. Such an approach is understandable, for it is hard to come up with a more

specific or quantitative way to characterize all forces in all industries; practically, managers

can intuitively comprehend what a “high” or “low” power of the force means. Ultimately,

because the purpose of our work is to operationalize the FF theory rather than adjust or expand

it, we choose to standardize the definition of the degrees for all five force uncertainties. This

modeling approach preserves DA’s ability to quantify the competitive analysis and to track the

competitive interdependencies. By linking each of the forces’ uncertainties to the economic

parameters’ uncertainties, the decision-maker can still quantify the implication of each force

on the profitability of the overall industry or of the firm. In addition, standardizing the forces’

degrees does not impact our ability to model their relevance and thus to track their

interdependence. We address these points in more detail in the next sections.

To demonstrate all newly introduced concepts in this section, we refer back to our earlier

example of launching a new environmentally friendly car. For simplicity, let us focus on only


three competitive uncertainties, illustrated in Figure 4.8: the bargaining power of Buyers and

its two parental drivers Price sensitivity of the buyer and Buyer’s switching cost between this

industry’s products. The analysis proceeds as follows. First, the automobile manager defines

two quantifiable, mutually exclusive and collectively exhaustive degrees for the Price

sensitivity of the buyer: {price elasticity of demand = [-3, -2]; price elasticity of demand = [-2,

-1]}. Similarly, she defines two degrees for the Buyer’s switching cost between this industry’s

products: {resale depreciation ≤ $2000; resale depreciation > $2000}. She also assigns two

degrees to the power of Buyers: {high; low}.

Figure 4.8: Illustrative example of uncertainty assessment in DAFF

Following the direction of the relevance arrows, the manager then starts her uncertainty

assessment from the most ancient ancestral level. Therefore, the manager first assesses the

probability of the car buyer’s price sensitivity. Based on her market intelligence, she assigns a

probability of 0.6 to the degree {price elasticity of demand = [-3,-2]} and a probability of 0.4

to {price elasticity of demand = [-2,-1]}. Similarly, the manager assesses the probability of the

buyer’s switching cost, and she assigs a 0.8 and 0.2 probabilities to the {resale depreciation ≤

$2000} and {resale depreciation > $2000} degrees, respectively. Finally, the manager assesses

the probability of the power of Buyers, conditioned on the two parent drivers. For example,

the manager asks: given that the Price sensitivity of the buyer is {price elasticity of demand =

Force

Drivers




Price sensitivity of

the buyer

Price elasticity of demand = [-3, -2]

Price elasticity of demand = [-2, -1]

0.6 0.4

Resale depreciation

≤ $2000

Resale depreciation

> $2000

0.8 0.2

Price elasticity of demand = [-3, -2] Price elasticity of demand = [-2, -1]

Resale depreciation

≤ $2000

Resale depreciation

> $2000

Resale depreciation

≤ $2000

Resale depreciation

> $2000

high low high low high low high low

0.9 0.1 0.6 0.4 0.5 0.5 0.2 0.8

high = 0.68low = 0.32


[-3, -2]} and that the Buyer’s switching cost between this industry’s products is {resale

depreciation ≤ $2000}, what is the probability that the power of Buyers is {high}? Based on

her market intelligence, the manager assigns a 0.9 probability to that prospect. After assigning

similar conditional probabilities to all eight prospects, the DAFF model computes the overall

probability of {high} or {low} bargaining power of Buyers. In this case, Figure 4.8 shows that

the probability of {high} is 0.68 and that of {low} is 0.32.

3.2 Modeling the Economic Implications of the Five Forces

FF emphasizes that maximizing profitability is the main goal of competitive strategy. The first

point that we should therefore clarify is: what defines profitability? Porter’s literature cites two

metrics to evaluate profitability: return on invested capital (ROIC) [2, 8] and unit profit

margin (UPM) [8]. ROIC is defined in (1) – as Porter recommends – as the ratio of annual

earnings before tax (EBT) over invested capital (CAPEX) [35]. As we know from managerial

accounting, EBT can be further decomposed into revenue (REV) less variable and fixed

operating expenses, denoted by VOPEX and FOPEX, respectively. Porter asserts that ROIC is

an effective metric to assess the industry’s profitability because it accounts for invested

capital. If ROIC is higher than the weighted-average cost of capital (WACC), the industry is

generating value. On the other hand, UPM is originally defined by Magretta as price less unit

cost [8]. To add clarity, we adopt a more elaborate definition in (2), where UPM is expressed

as a ratio of annual earnings before tax over annual revenue [36].

The use of either metric comes with its own set of challenges. To start, Porter cautions about

the use of UPM in assessing the profitability of the overall industry because it does not

account for CAPEX. Equally important, within a single firm, CAPEX may be shared among

multiple business units that operate in distinct industries; in this case, the use of ROIC to

assess a business unit’s relative profitability becomes challenging. In fact, multiple business

units might even share VOPEX or FOPEX, which further complicates the computation of both

UPM and ROIC in their respective industries. For instance, because of different product

specifications, geographic locations, customers, and substitutes, residential-scale and utility-

scale solar energy systems are distinct enough to be classified as separate industries. Yet, a

solar firm may operate in both industries [37]. Such a firm may own a solar-panel


manufacturing facility that caters to both its utility and residential businesses (shared

CAPEX), and its advertising endeavors may be designed to raise brand awareness, which

would also benefit both businesses (shared FOPEX). As a result, selecting a proper

profitability metric is dependent on the decision-maker’s specific objectives for the

competitive analysis. We discuss this topic in detail in the coming sections.

𝑅𝑂𝐼𝐶 =𝐸𝐵𝑇

𝐶𝐴𝑃𝐸𝑋=𝑅𝐸𝑉 − 𝑉𝑂𝑃𝐸𝑋 − 𝐹𝑂𝑃𝐸𝑋

𝐶𝐴𝑃𝐸𝑋=𝑄 ∙ (𝑃 − 𝐶𝑣 − 𝐶𝑓)

𝐶𝐴𝑃𝐸𝑋 (1)

𝑈𝑃𝑀 =𝐸𝐵𝑇

𝑅𝐸𝑉=𝑅𝐸𝑉 − 𝑉𝑂𝑃𝐸𝑋 − 𝐹𝑂𝑃𝐸𝑋

𝑅𝐸𝑉=𝑄 ∙ (𝑃 − 𝐶𝑣 − 𝐶𝑓)

𝑄 ∙ 𝑃 (2)

The formulations in (1) and (2) highlight two considerations related to linking the

aforementioned profitability metrics to the five competitive forces. First, the cited profitability

metrics incorporate three types of costs: CAPEX, VOPEX, and FOPEX. We know from the

FF literature that the five forces affect cost, but there is no clear reference as to how each force

affects each type of cost. Second, it is important to realize that ROIC and UPM share the same

basic economic parameters of: price, cost, and quantity. In fact, we demonstrate in (1) and (2)

how both profitability metrics can be reformulated in terms of price P, quantity Q, and two

per-unit costs Cv, and Cf, associated with VOPEX and FOPEX, respectively. VOPEX and

FOPEX can be (and usually are) reported per unit of goods or services because they are highly

dependent on sales, so together they reflect the ongoing cost of running the business.

The second consideration focuses on the need to evaluate quantity. While the FF literature

explicitly explains how stronger forces reduce price and increase cost, it provides little, if

any, guidance on the relationship between the forces and quantity. We address this issue now

by modeling a clear mapping between the competitive forces and all three broad categories of

economic parameters: Price, Cost, and Quantity. The impact of the forces on specific types of

economic parameters (e.g. sales quantity versus production quantity, or CAPEX versus

OPEX) is addressed in detail in the next section, when modeling the specific objectives of

competitive strategy.


Price, Cost, and Quantity are modeled in DAFF as uncertainty nodes. As explained by Porter

[2] and further clarified by Magretta [8], all five forces affect Cost while only four forces

affect Price: Buyers, Substitutes, New Entrants, and Rivals. Accordingly, nine relevance

arrows are added, five extending from the forces nodes into the Cost node and four extending

from the forces nodes into the Price node. Now, because FF provides no exact guidance on the

relevance between the competitive forces and Quantity, we undertake this modeling task by

referring back to the foundational pillars of FF that govern competition: microeconomic

theory and Industrial Organization. The economic models in both fields highlight four

relationships that are important to our task. In any given industry, demand-curves describe a

clear relationship between Price and Quantity, and cost-curves describe a clear relationship

between Cost and Quantity. Both relationships can be modeled as decision analytic relevance.

In addition, industry growth changes the shape of the demand-curve and thus the optimal

quantity output at market-equilibrium. As such, it is reasonable to model potential relevance

between the Growth factor and Quantity. Moreover, because substitutes lie outside the

industry’s formal boundaries, their desirability does not only affect the price or cost of the

product or service but also the quantity purchased by consumers; multiple mature economic

notions such as “elasticity of substitution” and “indifference curves” clearly capture this

relevance between Substitutes and Quantity [38]. As a result, we propose four relevance

arrows extending into Quantity from: Price, Cost, Growth, and Substitutes.

Subsequently, the profitability metric is modeled as a value node, and it is connected to the

Price, Cost, and Quantity nodes using a new type of arrows that we call functional arrows.

While relevance arrows convey potential probabilistic dependence, functional arrows convey

discrete mathematical relationships between the DA nodes; they always extend into the value

node. Figure 4.9 combines all the economic modeling steps discussed thus far and presents an

economic sub-model that can be integrated into a complete DAFF model.

After modeling Price, Cost, and Quantity and linking them to both the profitability value as

well as the force and factor uncertainties, we clarify how to characterize the degrees of each of

these economic uncertainties. Similar to drivers and factors, economic uncertainty nodes can

incorporate multiple degrees, which are industry-specific. Here again, Industrial Organization

provides necessary guidance on two tasks: defining the degrees of the Price, Cost, and


Quantity and assigning probabilities to those degrees. The widely documented Industrial

Organization models are simple yet effective means to estimate the range of Price, Cost, or

Quantity values in the industry. The difference between a purely monopolistic and a perfectly

competitive optimal price (or quantity) output is one good example of such range [38]. Along

the same lines, these models allow computing the exact optimal economic outputs under

specific competitive scenarios (equivalent to specific prospects of the five forces), which then

helps the decision-maker determine the likelihoods of the preset degrees of specific economic

uncertainties under these scenarios. For example, entry-deterrence models allow computing

the “limit-output”, defined as the quantity that a monopolist needs to produce in order to deter

entry [38]. Upon knowing the limit-output, a decision-maker can assign a more accurate

probability distribution over the Quantity degrees, given that both Rivals and New Entrants are

weak; in this case, the Quantity degree that is closest to the limit-output should be assigned the

highest conditional probability.

Figure 4.9: DAFF economic sub-model

3.3 Modeling the Firm’s Actions

The final element that needs to be modeled in DAFF is the firm’s list of competitive actions

within the analyzed industry. These actions can be categorized into three types of decisions:

Value Proposition decisions, Value Chain decisions, and Economic decisions.

Substitutes BuyersNew

EntrantsRivals Suppliers

Price CostQuantity

Profitability

Substitutes

Buyers

New EntrantsRivals

Suppliers Factors

Economic parameters

Industry growth rate


Value Proposition and Value Chain decisions are the two ingredients in Porter’s recipe for

strategic positioning [8]. Value Proposition decisions relate to the manager’s choice of what

product to make: what needs to meet in the industry, and what customers to serve. Then, Value

Chain decisions involve a series of managerial choices on how the product should be made:

how to design, produce, test, distribute, sell, and service. In other words, Value Proposition

decisions require the decision-making manager to look externally, focusing on the overall

industry; in contrast, Value Chain decisions require the manager to look internally, focusing

on the firm’s own activities and operations. In this regard, DAFF is a useful tool to bridge

between analyzing the five forces and making proper positioning decisions. The powers of the

five forces are different in different segments within the same industry. Upon choosing a

series of distinct Value Proposition and Value Chain alternatives, the decision-maker positions

her firm in a specific segment within the industry, hoping that the selected position yields

weak competitive forces and thus superior profitability for the firm.

Unlike Value Proposition and Value Chain decisions, Economic decisions are not strategic.

Economic decisions focus on three important choices: Production Scale, Pricing, and Tactical

Costing. Clearly, these decisions relate specifically to the firm’s profitability, and they shape

the firm-specific price, quantity, and cost. No matter what Value Proposition or Value Chain

decisions a firm makes to position in the industry, it should always optimize three aspects of

its business in order to maximize performance: how many units to produce, how to price these

units, and how to cost these units. The firm may choose to set the price at, above, or below the

optimal market-equilibrium price; similarly, it may underproduce or overproduce relative to

its expected demand. On the cost front, we make an important note. Big part of the firm’s

costs is already dictated by Value Chain decisions because, as Porter explains, value-chain

“activities [are] the elemental units of cost behavior” [25]. However, full costs are determined

not only by Value Chain strategic decisions that reflect positioning within the industry but

also by more detailed and mundane tactical decisions that reflect operative efficiency within

the firm. It is those latter decisions that we account for in Tactical Costing.

All three types of the firm’s actions are modeled in DAFF as decision nodes, and the

decision-maker may consider any number of alternatives for each decision. Decision nodes are

linked to the competitive and economic uncertainty nodes using a specific type of arrows


called influence arrows. By design, an influence arrow always extends from the decision

node to the uncertainty node [10]. Combining the firm’s decision nodes with the economic

sub-model in Figure 4.9 and the Bayesian network in Figure 4.6 (or 4.7) results in a complete

DAFF model, referred to as a decision diagram.

Value Proposition and Value Chain decision nodes can directly influence the five forces, their

underlying drivers, the four industry factors, as well as the economic parameters. Referring

back to our earlier car example, the auto manufacturer has to make a Value Proposition

decision regarding the type of car to produce: [electric] or [hybrid]. This decision may

influence the Buyer’s switching cost from this industry’s products to substitutes driver because

it may be easier to resell a hybrid car than an electric car [39, 40]. Notably, the influence of

this decision propagates all the way to the Substitutes and Buyers, for the aforementioned

driver is shared between these two forces. Additionally, this decision may directly influence

Cost because of the inherent technical differences between the two types of vehicles,

regardless of market competition [41, 42, 43].

While the competitive and economic uncertainties in DAFF are generalizable across

industries, Value Proposition and Value Chain decisions are industry- and firm-specific.

Accordingly, influence arrows are customized for each DAFF decision diagram corresponding

to a specific competitive analysis in a specific industry. Focusing on the competitive

uncertainties, one useful modeling practice to simplify the decision diagram is to add

influence arrows from Value Proposition and Value Chain decisions to the most ancient

ancestral level of drivers. In this case, the decisions update the probabilistic information in

these driver nodes only, but the Bayesian structure of the uncertainty network guarantees that

their influence propagates through the subsequent descendant nodes until reaching the five

competitive forces.

Finally, because Economic decisions relate to the firm’s specific profitability, they may only

influence the economic parameters. Specifically, Tactical Costing influences Cost while

Production Scale and Pricing influence Quantity, where they contribute to determining the

firm’s sales; the specifics of this modeling will become clearer when we discuss the objectives

of competitive strategy in the next section. For now, we note that, unlike Value Proposition


and Value Chain decisions, Economic decisions are modeled uniformly in all DAFF decision

diagrams across all industries.

This brings us to the end of this section. So far, we have introduced the various DA elements

and tools and used them to characterize and connect the different components of the FF

framework. The result is a complete DAFF model. In the next section, we move to explain

how DAFF models can fulfill each of the objectives of the FF competitive strategy. Along the

way, we highlight how modeling these strategic objectives resolves multiple deficiencies in

the operationalization and practical application of the – otherwise insightful – FF framework.

Here, we recall that the FF literature discusses three competitive-strategy objectives for every

firm: properly position in the industry, predict and exploit future industry change, and reshape

future industry change [2]. Based on our DA methodology, however, we propose

consolidating the goals of FF into two strategic objectives only, each of which would be

realized through sequential steps of analyzing the industry structure and the firm’s actions.

Specifically, the two DA-oriented competitive-strategy objectives are: position in the

industry and reshape the industry.

Positioning in the industry is a short-term objective, and it requires three consecutive

analyses of current and very-near-future profitability: of the overall industry, of distinct

positioning segments in the industry, and of a specific firm or business upon positioning in

each segment of the industry. Building on the outcomes from these steps, reshaping the

industry becomes the long-term objective, and it requires two additional analyses of distant-

future profitability: via predicting the evolution of the industry structure and via modifying the

evolution of the industry structure. These two strategic objectives and the steps required to

fulfill them are the focus of the following two sections. Because positioning in the industry is

the central concern of this study, we provide a detailed account on how to model its three steps

using the DAFF components introduced so far. Then, we describe, conceptually, how to

expand the DAFF approach to model the remaining two steps of reshaping the industry.

CHAPTER 4 — First Objective of Competitive Strategy: Positioning in the Industry 176

4 First Objective of Competitive Strategy: Positioning in the

Industry

Michael Porter argues that “strategy explains how an organization, faced with competition,

achieves superior performance” [8]. Translating this into decision analytic terms, strategy

explains how a decision-maker, faced with uncertain competition, can make choices that

achieve superior profitability, consistent with his unique set of preferences.

To achieve superior profitability, the business must first properly position in its industry.

Effective positioning requires rigorous analysis of the competitive landscape, which can be

achieved in three steps: assessing the profitability of the overall industry; identifying a series

of strategic Value Proposition and Value Chain alternatives that define unique positioning

segments in the industry then assessing the profitability of each of those segments; and finally

assessing the profitability of the business in all potential positioning segments. When thinking

through all three steps, the manager has to decide on one important aspect before delving into

the modeling details: the timeframe of the competitive analysis.

We leave it to the decision-maker to specify the exact timeframe for the positioning objective,

based on his specific goals from the competitive analysis and the specific industry of his

business. Nonetheless, we emphasize that any chosen timeframe must meet two criteria:

consistent across all three steps, and short-term in the context of the analyzed industry. The

consistency of the timeframe across the three positioning steps is necessary because their

analyses are sequential, one building on the top of the next. Equally important, the short-term

criteria is based on the premise that the decision-maker aims to locate his business in the most

preferred industry segment, very soon, before any structural changes affect the profitability of

that segment. Here, the decision-maker does not aim to change the behavior of other

industry players just yet; he only aims to understand their behavior and locate where it

is least threatening.


4.1 First Step: Assess the Profitability of the Overall Industry

Before making any decisions, including what to produce, how to produce, and how much to

produce and charge, the firm must have a clear understanding of the industry’s overall

structure and its average profitability. To that end, we first need to develop a DAFF model

that assesses the industry’s competitive landscape and its economic implications. The

complete DAFF model, including a detailed economic sub-model, for this first step of the first

competitive objective is presented in Figure 4.10. We call it the DAFF Bayesian Network.

To start, either the Simple or Detailed network, presented respectively in Figures 4.6 and 4.7,

can be used to represent the uncertain powers of the five forces and their underlying drivers.

We use the Simple Network here purely for convenience. Also, because the effects of the

factors on the powers of the forces are very industry-specific, all but one factor uncertainties

are removed from the displayed Bayesian Network; the Growth factor is maintained because it

has generalizable economic impacts, as will become evident. Furthermore, per Porter’s

recommendation, we adopt ROIC as the value metric to express the industry’s average

profitability.

The remaining task is to design a more elaborate DAFF economic sub-model that links the

competitive uncertainties to the ROIC value metric through a well-defined set of economic

parameters. These overall-industry parameters are labeled with an asterisk (*), and they

account for the fact that ROIC incorporates an average price P and three measures of cost:

CAPEX, Cv, and Cf. As explained earlier, Cv and Cf refer, respectively, to VOPEX and

FOPEX per unit of produced goods or services. Consistent with our approach in Section 3, the

price and the three cost parameters are modeled as uncertainties and are connected to the force

uncertainties using relevance arrows. As a reminder, the stronger the forces are, the more

likely the price is low and the costs are high [8]. Focusing on costs, because the FF literature

does not detail how the different forces affect the different types of cost, we model potential

relevance between all the forces and all the cost parameters. For example, in the automobile

industry, technological advances or strong customer preferences may necessitate additional

safety features in passenger cars. Higher safety measures may entail not only more expensive

parts in the car (higher Cv) but also more labor hours (higher Cf) or more complicated

178

Figure 4.10: DAFF Bayesian Network for the first objective, first step: assessing the profitability of the overall industry


Barriers to entry





Network effects


Unequal access to


Previous responses

by incumbents


for incumbent


for new entrant


product





incumbents




Incumbent switching


Supplier switching

costs between

incumbents


among suppliers

Availability of






product

Commitment of



Industry growth

rate


among incumbents


capacity



the buyer

Product cost as

fraction of the

buyer’s budget


incumbents



buyers downstream

Rivalry



Ability to enforce


Extent of exit

barriers


High fixed costs

Perishable product



product


discounting by

incumbents

Extent of market

segments


customer segments

Inability to read


Lack of familiarity

with incumbents

Number of buyers

Importance of non-




suppliers

Number of

industries the

suppliers serve

Profits extracted

by suppliers


Availability of



Volume of

purchase per

buyer



Substitutes

Buyers

New Entrants

Rivals

Suppliers Factors

Economic parameters


machinery (higher CAPEX) to assemble and install these parts. For every possible grouping of

a Cv degree and a Cf degree, we can calculate their exact sum as per-unit average operating

cost (CAO). CAO is modeled as a deterministic node because it is known with certainty for

every realized combination of Cv and Cf.

Subsequently, we transition to model quantity in the specific context of the overall industry.

Quantity here is defined as total annual sales volume (Q), and it is relevant to four

uncertainties: price P, average operating cost CAO, Growth factor, and Substitutes force.

Because the total sales volume is reached through some sort of market-equilibrium between

supply and demand, we assert that only P and CAO are relevant to Q, implying no relevance

between the CAPEX and the sales volume. To support this assertion, we first clarify that our

managerial-accounting definition of average operating cost CAO is equivalent to the

microeconomic-theory definition of average variable cost (CAV) 1

; effectively, both measures

refer to expenses that can change with the scale of business operations in the short-term

(which is the timeframe of this positioning objective). We know that market-equilibrium is set

where marginal revenue equals marginal cost. CAV, and thus CAO, is considered a good

approximation of marginal cost in the short-term. In addition, P equals marginal revenue under

the simplifying assumption of a linear demand-curve [38]. Accordingly, the equilibrium

between marginal revenue and marginal cost can be approximated as an equilibrium between

P and CAO, both of which then dictate Q.

The relevance between P and Q is shaped by the industry’s demand-curve, so higher P

increases the likelihood of smaller Q (assuming the product is not Giffen goods).

Subsequently, for a specific pair of P and CAO degrees, the closer CAO is to P, the more likely

the industry is competitive, and therefore the more likely Q is large; of course, we assume here

that the industry rivals always price above cost (P > CAO). One straightforward illustration of

this relevance is monopoly behavior: assuming all else equal, the sales volume in a monopoly

setting is always less than that in a perfectly competitive market, resulting in a deadweight

1 Our choice to use a managerial-accounting metric stems from our interest in making the analysis as simple and

accessible to business managers as possible. While relying on Industrial Organization to model the economic

implications of FF, we shall not forget that, eventually, managers are the primary decision-makers and the target

audience in this crusade to operationalize the FF framework. To that end, we strive to bridge between strategy,

managerial accounting, and Industrial Organization to achieve robust yet intuitive economic modeling.


loss. Finally, consistent with the explanation in Section 3, we emphasize that the relevance of

Growth and Substitutes to Q is highly industry-specific and can be further informed and

shaped by testing a few Industrial Organization models. However, as a broad generalization, Q

is more likely to be large when Growth is fast and Substitutes are weak [38].

We continue this discussion of the industry’s economic sub-model by focusing on the time

horizons. As already established, assessing the overall industry profitability for positioning is

a short-term analysis; this means that the probability distribution over the degrees of each

uncertain economic parameter should reflect the decision-maker’s beliefs about the expected

value of that parameter in the very near future. But if probability assignments over the

economic degrees are guided by the analysis’s (short-term) timeframe, what guides the

definition of the degrees themselves? In other words, how do we characterize the range of

feasible degrees for each economic parameter in the first place? To answer this question, we

refer to the FF literature. Porter suggests that the ROIC, and therefore its economic attributes,

should be characterized over a full business cycle. In other words, when modeling each

economic uncertainty, the numerical-value range of its degrees should be comparable to the

numerical-value range observed in the industry during a full business cycle. An example here

might be helpful. Let us assume that the full business cycle in a mining industry is one decade

[2], and that a current incumbent wants to use the DAFF model to analyze the profitability of

the whole mining industry next year. When modeling the CAPEX uncertainty, the incumbent

first defines three CAPEX degrees whose numerical values cover the full range of observed

CAPEX over the last decade (full-business-cycle time horizon). Then, he assigns a probability

to each CAPEX degree, under all possible scenarios of the forces’ powers, based on his beliefs

about the value of CAPEX next year (near-future time horizon).

Ultimately, the resulting DAFF Bayesian Network in Figure 4.10 generates two primary

outcomes. First, it computes a probability distribution for the power of each of the five

competitive forces, thus providing a snapshot of the industry’s overarching competitive

landscape. Second, it yields a probabilistic-weighted-average value of ROIC, referred to as

expected ROIC. This expected value indicates the average profitability of the whole industry

within the short-term timeframe set by the decision-maker. One can then quickly gain insight

about the extent of the industry profitability by comparing the resulting expected ROIC to its


historical values. Intuitively, stronger competitive forces will result in lower ROIC; the DAFF

model quantifies this trend after accounting for all the interactions among the competitive

forces and factors. Building on these results, the DAFF Bayesian Network can also test the

sensitivity of the expected ROIC to the power of one or more of the five forces. The outcome

would be an ROIC range corresponding to the power range of the investigated force(s).

Additionally, the resulting DAFF model can compute how observing one force at either

extreme (strong or weak) may update the probability distribution of other forces, thus

quantifying the extent of interdependency and interaction among these forces.

4.2 Second Step: Position Competitively and Assess the Profitability of Each

Positioning Segment in the Industry

After assessing the competitive forces and their economic implications for the whole industry,

a manager should make a series of strategic choices that position her business in the most

profitable segment of the industry. As noted by Song et al. [28], the prospects of any

positioning decision are highly dependent on the uncertain competitive landscape, so a clear

mapping between the management’s feasible positioning alternatives and the industry’s forces

is essential for guiding the firm’s competitive strategy. In our DAFF approach, strategic

positioning is characterized by two of the three types of actions undertaken by firms: Value

Proposition and Value Chain decisions.

As explained earlier, Value Proposition and Value Chain decisions allow the firm to focus and

tailor its activities and to target a specific segment in the industry, thus narrowing down the

group of competitive players that the firm interacts with. In other words, positioning decisions

influence the power of the competitive forces not by changing the behavior of other industry

players but by changing the identity of those players. Every combination of Value Proposition

and Value Chain alternatives results in a unique positioning track, which in turn defines a

unique segment of the industry where the firm can locate and operate. For example, a

decision-maker in the auto industry may want to analyze the influence of three Value

Proposition and/or Value Chain decisions on the competitive landscape, and two alternatives


are available for each decision. In this case, the automaker has a set of eight (i.e. 23) possible

positioning tracks to choose from, each of which defines a unique industry segment.

Because positioning decisions are very circumstance-, industry-, and firm-specific, Figure

4.11 shows how to add generic Value Proposition and Value Chain decisions to the DAFF

Bayesian Network introduced in Figure 4.10, resulting in a DAFF Decision Diagram.

Needless to say, the non-generalizable nature of positioning means that the influence relation

between a specific Value Proposition or Value Chain decision and a competitive uncertainty

may change from one industry to the other. Let us consider the influence relation between

Product design decision and Preferential access to resources uncertainty as an example. In

China, where silicon is abundant, the decision by a solar-panel-manufacturing firm to produce

[photovoltaic solar systems] instead of [concentrated solar systems] might influence (increase)

the likelihood of it gaining easy access to raw materials. On the other hand, the decision by a

US automaker to produce an [electric car] instead of a [hybrid car] will not influence the

likelihood of it gaining wide access to skilled labor.

In addition, the DAFF Decision Diagram in Figure 4.11 highlights two important modeling

aspects. First, the economic parameters are still industry-based not firm-based; more

specifically, the degrees in Cost, Price, and Quantity uncertainties, as well as the profitability

measures in the ROIC value node, still represent average industry values. Second, we recall

from Section 3.3 that a positioning decision node can influence any uncertainty node,

including the forces, drivers, industry-specific factors, and economic parameters. In fact, one

positioning decision might influence more than one uncertainty, and one uncertainty might be

influenced by more than one positioning decision. Regardless of what and how many

uncertainties the decision-maker draws influence arrows into, the Bayesian nature of the

decision diagram ensures that the influence of each positioning decision will propagate all the

way to the value node. As a result, the DAFF Decision Diagram in Figure 4.11 computes a

unique expected ROIC value for each unique positioning track (combination of positioning

alternatives) and therefore for each unique industry segment. Following up on the earlier

example, analyzing three Value Chain decisions, each with two alternatives, outputs eight

positioning tracks and therefore eight expected ROIC values. Intuitively, the firm should

choose to position in the industry segment with the highest ROIC.

183

Figure 4.11: DAFF Decision Diagram for the first objective, second step: positioning competitively and assessing the profitability of each

positioning segment in the industry


Barriers to entry





Network effects


Unequal access to


Previous responses

by incumbents


for incumbent


for new entrant


product





incumbents

Industry concentration



Incumbent switching


Supplier switching

costs between

incumbents


among suppliers

Availability of






product

Commitment of



Industry growth

rate


among incumbents


capacity



the buyer

Product cost as

fraction of the

buyer’s budget


incumbents



buyers downstream

Rivalry



Ability to enforce


Extent of exit

barriers


High fixed costs

Perishable product



product


discounting by

incumbents

Extent of market

segments


customer segments

Inability to read


Lack of familiarity

with incumbents

Number of buyers

Importance of non-




suppliers

Number of

industries the

suppliers serve

Profits extracted

by suppliers


Availability of



Volume of

purchase per

buyer



Substitutes

Buyers

New Entrants

Rivals

Suppliers Factors

𝑃 𝐶𝑣

𝑅𝑂𝐼𝐶

𝐶𝑓

𝐶𝐴𝑃𝐸𝑋 𝐶

𝑄

Economic parameters

Value Chain 1 Value Chain 2Value

Proposition 1

Firm’s strategic decisions


4.3 Third Step: Assess the Profitability of the Firm in Each Positioning

Segment of the Industry

The last step in modeling the positioning objective requires calculating the firm-specific

profitability in the industry segment where it chooses to position. To achieve this goal, we

only need to update the economic sub-model of the DAFF Decision Diagram in order to

account for the firm’s economic performance, as shown in Figure 4.12. Here, we add the three

Economic decisions we introduced earlier: Tactical Costing, Pricing, and Production Scale.

Based on the available resources and/or implemented management system, a firm may

implement a series of tactical measures that affect its cost structure without affecting (and

regardless of) its strategic positioning in the industry. Those tactical measures are captured in

a set of Tactical Costing alternatives that increase or decrease the cost for the firm by a

specific percentage points relative to its industry-segment average. For example, an electric

vehicle manufacturer who has decided to produce a luxury sports car (strategic positioning

decision) still needs to make a decision on how to schedule the car assembly (tactical

decision). If the company deploys state-of-the-art labor-scheduling protocols and procedures,

it may endure below-average utility expenses; the opposite is true if the company deploys

outdated labor-scheduling protocols. Either way, the company’s labor-scheduling practices

affect its cost without affecting its strategic positioning, so it is accounted for in Tactical

Costing. To that end, the cost structure for the firm is dictated by both the average cost in the

positioning industry segment and the tactical operational practices in the firm, which shift the

cost above or below the industry’s average.

To properly model the firm’s costs in our DAFF Decision Diagram, we add firm-specific

economic parameters Cv0, Cf

0, and CAPEX0 corresponding, as before, to capital expenditure,

unit-based fixed operating cost, and unit-based variable operating cost, respectively. CAPEX0

for a specific firm is determined by the Tactical Costing of that firm as well as by CAPEX of

the industry segment where the firm positions. Accordingly, CAPEX0 is modeled as a

deterministic node that is known with certainty for a given pair of CAPEX degree and

185

Figure 4.12: DAFF Decision Diagram for the first objective, third step: assessing the profitability of the firm in each positioning segment of

the industry


Barriers to entry





Network effects


Unequal access to


Previous responses

by incumbents


for incumbent


for new entrant


product





incumbents

Industry concentration



Incumbent switching


Supplier switching

costs between

incumbents


among suppliers

Availability of






product

Commitment of



Industry growth

rate


among incumbents


capacity



the buyer

Product cost as

fraction of the

buyer’s budget


incumbents



buyers downstream

Rivalry



Ability to enforce


Extent of exit

barriers


High fixed costs

Perishable product



product


discounting by

incumbents

Extent of market

segments


customer segments

Inability to read


Lack of familiarity

with incumbents

Number of buyers

Importance of non-




suppliers

Number of

industries the

suppliers serve

Profits extracted

by suppliers


Availability of



Volume of

purchase per

buyer



Substitutes

Buyers

New Entrants

Rivals

Suppliers Factors

𝑃 𝐶𝑣 𝐶𝑓

𝐶𝐴𝑃𝐸𝑋 𝐶

𝑄

Economic parameters

Value Chain 1 Value Chain 2Value

Proposition 1

Firm’s decisions

𝐶𝐴𝑃𝐸𝑋0

Tactical Costing

Pricing

Production Scale

𝑀 0 0

𝑄0

𝑅𝑂𝐼𝐶


Tactical Costing alternative. A relevance arrow and an influence arrow extend into CAPEX0

from CAPEX and Tactical Costing, respectively. We follow a similar procedure to model Cv0

and Cf0.

Subsequently, the firm has the ability to choose how to price its product, regardless of what

the optimal market-equilibrium is. Here, we introduce a new economic parameter, the firm’s

Market Share (MS0), defined as a fraction of the overall annual sales volume in the firm’s

industry segment (Q ). Relying on Industrial Organization theory, we model four economic

attributes affecting MS0: the firm’s Pricing, industry average Price P

*, Industry concentration

driver, and the power of New Entrants force. Industry concentration refers to the number and

relative size of current incumbents in the industry. Reasonably, learning about current

incumbents and future new-entrants informs the decision-maker’s beliefs about her firm’s

likelihood to gain, sustain, or lose market share in the near-future [38]. Also, for any given

setting of incumbents and new entrants, one can argue that as Pricing increases beyond P*,

customers will increasingly refrain from purchasing the firm’s product, leading to a smaller

MS0. To account for all these effects, the DAFF Decision Diagram in Figure 4.12 presents the

firm’s MS0

as an uncertainty node with four arrows directed into it: one influence arrow from

the Pricing decision node and three relevance arrows from the New Entrants, Industry

concentration, and P* uncertainty nodes.

The rest of the economic modeling is just math. The product of MS0 and Q

* results in D

0,

defined as the expected demand for the firm’s product (or service). Then, the firm’s annual

sales volume (Q0) become the minimum of {D

0, Production Scale}. If the firm overproduces

beyond its expected demand (Production Scale > D0), its sales will be constrained by demand

(Q0 = D

0); equivalently, if the firm underproduces below its expected demand (Production

Scale < D0), its sales will be constrained by supply (Q

0 = Production Scale). As shown in

Figure 4.12, both D0 and Q

0 are modeled as deterministic nodes because they are fully

specified upon knowing their parents. Finally, recalling the two definitions in (1) and (2), we

can compute the expected profitability for the firm in each positioning industry segment either

in the form or ROIC0 or PM

0. Figure 4.12 models ROIC

0 as the value node with functional


arrows extending into it from the firm’s Pricing, expected annual sales volume Q0, and all

three types of costs Cv0, Cf

0, and CAPEX0.

Before ending this section, we underscore that this proposed modeling of the firm’s

profitability is an extension to – not a formal part of – the FF framework or theory. While

necessary for operationalizing the positioning objective in competitive strategy, the discussed

economic sub-model for the firm’s profitability represents our own assessment – and

assumptions – on a how a manager values a specific line of business. Unlike all previous

modeling endeavors so far, this economic sub-model is not a direct translation of the FF

literature. To that end, we shall leave it to the decision-maker to use, modify, or even replace

this sub-model in order to evaluate her business’s specific profitability.

4.4 Advantages of the DAFF modeling

We now bring all newly introduced concepts together in order to highlight the benefits of

using the DAFF approach to model this first objective of competitive strategy.

Comprehensive representation of competitive strategy: Broadly, Porter’s literature tends to

discuss FF and value-chain as two distinct strategic frameworks, aiming, respectively, to

evaluate the whole industry then to develop a competitive advantage for the firm within the

industry [8]. Evidently, DAFF combines the modeling of both frameworks, which further

enhances their operationalization. The presented DAFF Decision Diagram in Figure 4.11

illustrates that FF and value-chain are highly interconnected, and when modeled together, they

offer a clear procedure for firms to properly position in their respective industries. For

example, value-chain choices on product design and distribution channels can easily affect

whether a firm “enters, stays in, or exits” a specific industry, which Porter emphasizes as

important positioning decisions. Also, the pool of possible Value Proposition and Value Chain

alternatives allows the firm to be both creative and flexible in choosing its positioning

strategies, which in turn widens the scope of the “cost-leadership”, “differentiation”, and

“focus” strategies highlighted by Porter [28].


Balance between generality and customization: An important advantage of the DAFF

Bayesian Network in Figure 4.10 is balancing between the need to generalize and the ability

to customize the probabilistic assessment of the competitive uncertainties. By adding or

eliminating drivers at the various ancestral levels, the model allows the decision-maker to

emphasize (or deemphasize) specific attributes in her industry. Yet, despite what drivers the

decision-maker chooses to analyze, she will always end up with a clear probability distribution

for each of the five force uncertainties. For example, a manger in the residential solar industry

may not be able to properly assess the power of Substitutes because the Price-performance

tradeoff to this industry product driver seems too vague. In this case, the manager can add a

few parents to that driver in order to better inform her assessment. Such parent nodes may

include Bill savings from substitutes, Installation and maintenance time for substitute, or

Climate impact of substitute. Then, the manager investigates: what is the probability that the

Price-performance tradeoff to this industry product is favorable, given that Bill savings from

substitutes is {25% reduction in the monthly utility bill}, Installation and maintenance time

for substitute is {15 minutes per month}, and Climate impact of substitute is {50% reduction

in the household carbon emissions}?

The Decision Diagrams in Figures 4.11 and 4.12 further emphasize DAFF’s ability to balance

between generality and customization. To properly position, every firm in any industry should

examine a series of Value Proposition and Value Chain decisions and should link those

decisions to the competitive and economic uncertainties. Equivalently, to compute its

profitability, every firm in any industry should factor in its Tactical Costing, Production

Scale, and Pricing decisions. While DAFF explains how to model each of these decision

categories, it leaves it to the manager to define every particular decision and its alternatives,

based on the specifics of the analyzed business and industry. Because of that, the DAFF

Decision Diagrams can model clear competitive actions for a wide range of decision problems

that are material to a wide range of decision-makers.

In addition, through customization, the DAFF modeling endorses Porter’s recommendation to

the firm to be “unique” instead of “best” in its industry [8]. This is achieved in two ways.

First, DAFF forces the firm to choose only one of many feasible positioning tracks, which

encourages specialization. Second, the DAFF inputs are subjective in nature; even if we


assume a single DAFF model with fixed configuration, different decision-makers may have

different market information and business intuition, which results in different design of

uncertainty degrees or different probability assignments over those degrees. Inevitably, the

subjective input will generate subjective outputs and recommendations, unique to the

decision-maker’s outlook on the industry. Here, we note that as DAFF facilitates the modeling

of “being unique”, it also obstructs the modeling of “being best.” Choosing the “best”

positioning alternative cannot be properly modeled by DAFF, for “best” is too vague to pass

the DA clarity-test and to define a proper value metric for ranking the decision-maker’s

preferences.

Quantification: In her book, Magretta quotes Porter that “strategy requires clear, analytic

thinking” [8]. As evident by now, the DAFF models allow quantifying all the necessary

components for a competitive analysis, including: the description of feasible alternatives for

every competitive decision; the description of possible realizations (i.e. degrees) for every

competitive driver, factor, force, or economic uncertainty; the likelihood (i.e. probability) of

each degree; and the value of profitability. In turn, this multi-layered quantification enables

other DAFF benefits, which we subsequently discuss.

Comparison, ranking, and prioritization: Customization and quantification facilitate consistent

comparison of the forces, their drivers, and industry-specific factors by examining their impact

on profitability. More powerful forces lead to lower prices, higher costs, and therefore lower

ROIC. To that end, the range of power for every force can be translated to a range of ROIC,

and the forces can be compared and ranked based on their respective ranges of expected

profitability. Extremely strong (probability of {high} = 1) Buyers may not lead to the same

ROIC as extremely strong Substitutes, for instance. In general, a wider ROIC range signifies a

more impactful force.

Because drivers and factors are probabilistically relevant to the forces, changing any of them

can also result in a distinct ROIC range. However, depending on the exact conditional

probability distributions, drivers or factors at the same ancestral level – even those sharing the

same child nodes – may have different impacts on profitability. To that end, the relative


significance of drivers and factors may vary from one industry to the other even though their

structural configuration – in the DAFF models – remains consistent across industries.

In general, this form of sensitivity analysis that investigates the effect of the various

uncertainties on the value is common in DA and is widely applied in the form of tornado

diagrams [10]. Indeed, this ranking and comparison exercise also resolves one of the perceived

deficiencies in applying the FF framework, raised by Grundy [29]. As a practical implication

of this sensitivity analysis, the manager may choose to prioritize the most impactful drivers or

factors and focus all subsequent analyses on them. Such prioritization may entail eliminating

the least-impactful drivers or factors, seeking more information on or control over the most

impactful drivers or factors, or both.

Similar logic applies to assessing the impact of the firm’s positioning decisions on

profitability. Because every positioning track is associated with a distinct ROIC, positioning

tracks can be ranked from most to least profitable. The manager may then decide to spend

additional resources further analyzing the most profitable tracks while discarding the least

profitable ones.

Mapping interdependence and interaction: The FF literature stresses that a good industry

analysis should investigate “how shifts in one competitive force [might] trigger reactions in

others” [2]. DAFF offers one method to conduct such an investigation both rigorously and

quantitatively. The presented DAFF Bayesian Network in Figure 4.10 tracks the interactions

among the forces via two means: shared drivers and relevance paths – the latter referring to

a sequence of relevance arrows. In both cases, the uncertainty nodes communicate because of

their Bayesian probability distributions. When new information becomes available regarding a

specific uncertainty, the decision-maker may adjust the probability distribution over the

degrees of that uncertainty, which may simultaneously update the probability distributions of

that uncertainty’s parents and/or children.

As mentioned before, one example of a shared driver node is the Buyer’s switching cost from

this industry’s products to substitutes, which is a common ancestor to both Buyers and

Substitutes. Similarly, one example of a relevance path is the sequence of relevance arrows


from Basis of competition to Barriers to entry. To simulate this relevance, we start by

assuming that the decision-maker receives new information that reveals price-based

competition between the incumbents. Given this information, the decision-maker may infer

that {low} Production differentiation among incumbents is more likely (causing incumbents

to compete on price). This in turn triggers the following sequential probability updating:

{low} Buyer’s switching cost between this industry’s products is more likely; {high}

Consumer adoption rate of product by new entrant is more likely; and therefore {low}

Barriers to entry are more likely. Of course, we simplistically use {high} and {low} degrees

for all uncertainties in this example for illustrative purposes only. Practically, the exact

relevance path(s) from one uncertainty to another in a Bayesian network can be tracked via

established DA techniques such as Bayes-Ball [44].

Furthermore, the DAFF models are beneficial not only for highlighting relevance but also for

capturing irrelevance. As explained earlier, the lack of a relevance arrow between two

uncertainties signifies that – in the absence of additional information – the two uncertainties

are not relevant. When such an assumption does not hold in a specific industry, more

relevance arrows can be added. For example, the Detailed Network in Figure 4.6 shows no

relevance between Buyers trust in incumbents and Well-established brands because we didn’t

detect any assertion regarding such relevance in the reviewed FF scripts. Nonetheless, it is

easy to imagine multiple industries (e.g. automobiles, electronics, and sports merchandise)

where such relevance is possible and thus should be captured.

Finally, the DAFF modeling enables – rather imposes – a consistency check on the

interdependencies among the force, driver, and factor uncertainties by exposing, then ruling

out, illogical or conflicting evaluations of those uncertainties. For example, the DAFF model

allows checking whether it is possible for the probability of {high} Buyers power to be less

than 0.6, given that the probabilities of {high} Substitutes and {high} New Entrants are both

greater than 0.8.

Outlining the firm’s scope of control: Though substantial, the effect of positioning on the five

forces’ powers is still limited because decisions can influence some, but not all, of the forces’

ancestral drivers and factors. This DAFF feature is consistent with Porter’s caution against


“competitive convergence” where competing firms eventually run out of ways to

differentiate themselves [2]. To that end, DAFF modeling captures both the firm’s ability to

control being unique and its inability to control competitive convergence, and how both

phenomena may co-exist in the industry. On one hand, “best vs. unique” can be controlled by

the firm through positioning: being unique, as opposed to being best, is driven by the firm’s

need to choose one positioning track, which determines the identity and type of players it

interacts with. On the other hand, “competitive convergence” cannot be controlled by the

firm’s positioning decisions: competitive convergence stems from the firm’s inability to

change the short-term activities and decisions of the players it interacts with. Tesla Motors is a

good example here [45, 46]. By positioning in luxury-sports electric vehicles, Tesla was able

to limit its competitors to high-end car manufacturers like Porsche and Cadillac while

avoiding direct interaction with mass-market manufacturers such as Ford and Toyota.

Nonetheless, Tesla was not able to change its competitors’ decisions on whether to produce

electric vehicles. As such, Tesla succeeded in being unique at the beginning with Roadster and

then Model S. However, eventually, competitive convergence emerged, with Porsche rolling

out Panamera S E-Hybrid and Cadillac rolling out the ELR plug-in hybrid, both with similar

features and price-range to Tesla’s Model S [47].

Along the same lines, we recall that the FF framework asserts a disproportional relation

between the powers of the five forces and average profitability; firms should position in

industry segments that maximize the latter by minimizing the former. Nonetheless, while

specific positioning can push the forces to a relatively weaker level, it does not guarantee that

all forces will reach their absolute weakest level because it cannot simultaneously influence

all their underlying drivers and factors. Here, DAFF modeling can manifest how optimal

positioning may not reduce the power of all forces simultaneously; instead, it might be

achieved at some level of trade-off or balance among the forces’ powers. Let us consider

SolarCity for instance [48]. As the largest player in the residential solar industry, SolarCity

adopts a direct-to-customer retail model whereby it internalizes all system-installation services

for its customers [49]. This positioning strategy may decrease Buyers power by canceling the

need for intermediary channels, but it may also increase Rivals power by limiting the majority

of business to urban areas where the number of competitors, and thus the intensity of


competition, is high [50, 51]. A positioning track is judged by its expected profitability, so

direct-to-customer retailing and the resulting five-force balance may indeed be optimal, but

only if it maximizes the ROIC.

Valuating broader control or additional information: By robustly quantifying and mapping all

the competitive forces, drivers, and factors, their interactions, as well as their influence by the

firm’s positioning decisions, the proposed DAFF models in Figures 4.10 and 4.11 allow

measuring two key metrics: value of control (VOC) and value of information (VOI). VOC is

a DA concept that quantifies the cost of controlling an uncertainty; effectively, it assumes that

the decision-maker can somehow purchase the ability to adjust the outcome of that uncertainty

to her benefit. For example, to compute the value of fully controlling Buyers power, the

decision-maker sets the power of this force to its absolute maximum (probability of {high} =

1) or minimum (probability of {low} = 1) then calculates VOC as the difference between the

ROIC values with and without control. The resulting VOC plays an important role in

benchmarking how much a manager should spend on controlling a competitive force; a

decision-maker should never pay more than the VOC of a specific uncertainty to control the

outcome of that uncertainty. Equivalently, VOI is a similar DA metric that quantifies the cost

of resolving an uncertainty; in this case, the decision-maker gets to learn what the outcome of

the uncertainty is rather than dictating what it should be [10].

Exposing and reducing cognitive biases: A decision-maker usually thinks of few drivers,

factors, or forces when assessing competition. Naturally, this go-to set of competitive

attributes is highly dependent on the decision-maker’s prior experiences and knowledge, and it

may not be sufficient to understand and analyze the industry of her current business. To that

end, designing degrees and assigning probabilities for every uncertainty node in the DAFF

Bayesian Network urges the decision-maker to think critically about a wide range of

competitive attributes, which helps expose and reduce any cognitive biases that may magnify

the role of some attributes while attenuating (even ignoring) the role of others.

Moreover, at the end of the assessment, the decision-maker may dislike or reject the output

probability distribution for a particular force or the output ROIC value for a particular industry

segment. Because the outputs are calculated based on all the inputs into the model, the

CHAPTER 4 — Second Objective: Reshape the Industry 194

apparent discrepancy between the expected and actual results must be due to either a cognitive

bias by the decision-maker (that needs to be exposed) or an erroneous belief-input into one or

more uncertainty nodes (that needs to be fixed). It is easy to track back the information in the

DAFF models until reaching the real cause of this discrepancy. At that point, either the

decision-maker will realize her cognitive bias and become convinced with the present output,

or he will adjust her input. Ultimately, this iterative process of rationalizing the output with

the input achieves consistency and clarity of thought when assessing the competitive

landscape in the industry, no matter how complex or detailed the DAFF model might be.

5 Second Objective: Reshape the Industry

Fundamentally, this second objective of competitive strategy requires tracking the changes in

the five forces’ powers and their economic implications in the long run. The managerial

decision-maker may be interested either in passively predicting or in actively modifying the

competitive structure in the various positioning segments of her industry. In the next two

sections, we provide a conceptual overview on how to model these two tasks explicitly using

the DAFF approach, and we highlight the intuitive transition from one task to the other. As

will become evident, predicting the competitive evolution of the industry is a necessary first

step to reshaping it, and the decision-maker can achieve both goals by continuously

understanding, anticipating, and influencing the behavior of its buyers, suppliers, rivals, new

entrants, and substitutes.

Because reshaping the industry requires tracking competition over time, the DAFF modeling

of this second objective can be perceived as a dynamic extension of the DAFF modeling of

the first objective. We recall that one main distinction between the two competitive strategy

objectives is timeframe. As the decision-maker aims to properly position in the industry in the

short-term, she aspires to reshape the industry in the long-term. In essence, then, the long-term

timeframe is nothing more than a sequence of short-term periods, and the decision-maker can

dynamically assess the distant future of competition by modeling how the competitive forces

transition from one short-term period to the next.


To facilitate the development of the dynamic DAFF models, we condense the representation

of the original positioning DAFF models in Section 4. The DAFF Decision Diagram in Figure

4.11 can be reduced to a three-node conceptual model, presented in Figure 4.13. The decision

node D in the conceptual model encapsulates all Value Proposition and Value Chain

decisions, and the uncertainty node F incorporates all force, driver, factor, and economic

uncertainties. The value node V remains as defined previously, reflecting the metric used to

assess profitability in the various positioning segments of the industry. Here, we clarify our

explicit intention to base the conceptual model on the decision diagram in the second step

(Figure 4.11), rather than the third step (Figure 4.12), of the first objective. Because the

manager’s strategic goal is to track the evolution of the competitive forces in every positioning

segment of the industry, she ought to focus on industry-level profitability; once that

profitability is properly evaluated, it becomes relatively easy to deduce the firm’s profitability

by applying the economic sub-model and following the procedure we outlined in Section 4.3.

Figure 4.13: Conceptual DAFF Decision Diagram for the first positioning objective

5.1 First Step: Predict Industry Change

The dynamic DAFF model for predicting industry change is presented in Figure 4.14. The

model assumes that the decision-maker wants to assess the change in the competitive

landscape over three sequential periods of time: (t), (t + 1), and (t + 2). Period (t), on its

own, represents the short-term timeframe for effective positioning. Positioning decisions D(t)

influence the competitive landscape F(t) exactly as we explained in Section 4. However, the

new dynamic nature of the DAFF model is manifested in the relevance arrows that extend

F

D

V

Forces, drivers,

factors, economic

parameters

Value Proposition

Value Chain

Profitability


from F(t) to F(t + 1) and then from F(t + 1) to F(t + 2). Those arrows signify that each of

the five force uncertainties at (t + 2) is relevant to its earlier state at (t + 1), which in turn is

relevant to its state at (t). The same is true for each factor, driver, and economic uncertainty.

Effectively, this means that learning about the competitive and economic uncertainties in the

near future helps inform their assessment in the distant future. Finally, the competitive

landscape at (t + 2) dictates the expected profitability of each positioning segment in the

industry at (t + 2), per our earlier explanation in Section 4.

Figure 4.14: Dynamic DAFF model for the second objective, first step: predicting industry

change

The evolution of the competitive uncertainties over time is the main distinctive feature of this

dynamic DAFF model. In that regard, we make three important modeling notes. First, we

leave it to the decision-maker to determine the scale of each time period and the number of

time periods over which she wishes to assess industry change in the long-term, for both

features are very industry-specific. For example, the mining manager in our earlier example

may wish to set (t) equal to one year while (t + 1) and (t + 2) equal to three years each; such

timeframe would allow the manager to predict industry change over the coming seven years.

Second, to keep the dynamic DAFF model simple, relevance arrows from (t) to (t + 1) and

from (t + 1) to (t + 2) should be added very selectively; we call both types of arrows

Short-term

Long-term

Long-term


temporal. In details, if an uncertainty node has parents, it shall not originate or receive a

temporal relevance arrow. Conversely, if an uncertainty node has no parents, it shall originate

or receive only one temporal relevance arrow. Figure 4.15 shows an example from the force of

Buyers. The Price sensitivity of the buyer driver has six ancestral nodes with no parents.

Assessing the probabilities of these nodes based on prior knowledge may become harder the

further the decision-maker looks into the future. However, the near-future state of these

uncertainties at (t) may inform the decision-maker’s assessment of their distant-future state at

(t+1). Accordingly, a temporal relevance arrow is added for each of the six ancestral nodes,

extending from its state at (t) to its state at (t + 1). Subsequently, the probabilistic assessment

of each of the remaining four drivers – including the Price sensitivity of the buyer itself – at

(t + 1) is informed by its parents at the same time period. Because the Price sensitivity of the

buyer driver has four parent nodes that facilitate its analysis by the decision-maker, we advise

that no temporal relevance arrow is needed between its states at (t) and (t + 1). However, an

analyst may still find it useful to add that arrow to achieve clarity. In fact, the analyst may

even decide to add a temporal relevance arrow from a parent at (t) to a child at (t + 1). We

distinguish both types of arrows as dotted lines in Figure 4.15. While adding either arrow is

not technically wrong, it might increase the complexity of the competitive analysis. If we

assume two degrees per uncertainty in Figure 4.15, then adding two additional temporal

relevance arrows to the Price sensitivity of the buyer at (t + 1) increases the number of

required probability assignments for that node from 16 (24) to 64 (2

6), which is a difficult task

for managers to handle.

The final note is related to positioning. Similar to Figure 4.11, Figure 4.14 shows that the

firm’s positioning decisions influence the competitive landscape in the short-term. However,

in this first step of the second objective, the decision-maker aims not only to understand the

short-term behavior but also to anticipate the long-term behavior of the industry players in

each positioning segment. Subsequently, the expected profitability for each segment is

evaluated at the end of the last period of the selected timeframe; for example, in Figure 4.14,

the expected profitability value corresponds to the end of period (t + 2). In short, while the

DAFF model in Figure 4.11 provides managers with a tool to identify the most profitable

positioning segment in their industry in the short-term, the DAFF model in Figure 4.14 allows


managers to predict the most profitable positioning segment in their industry in the long-term.

Both DAFF models highlight where the competitive behavior is least threatening, but they

provide no guidance on how to modify that behavior. Modeling the manager’s efforts to

modify her competitors’ behavior, and consequently to reshape her industry, is what we

discuss in the next section.

Figure 4.15: Guidance on adding temporal relevance arrows between competitive

uncertainties

5.2 Second Step: Reshape Industry Change

Reshaping the industry requires influencing the competitive behavior. We now explain how

DAFF can model the firm’s attempts to change the behavior of its industry players over time

in order to gain competitive advantage. The dynamic DAFF model for reshaping industry

change is presented in Figure 4.16. Compared to the earlier model in Figure 4.14 for

predicting industry change, the DAFF model in Figure 4.16 adds three major elements, each of

which enables and informs the firm’s strategic interactions with the other competitive players.

First, a decision node D is added at (t + 1) and (t + 2), which allows the decision-maker to

re-assess positioning over time. Second, a new influence arrow is added from D(t) to F(t + 1)

and similarly from D(t + 1) to F(t + 2). While the arrow from D(t) to F(t) influences the


product

Product differentiatio

n among incumbents


the buyer

Product cost as fraction of the buyer’s

budget



product

product leads to

performance improvement for buyer


Amount of cash

available to the buyer

Industry product pays

for itself

product reduces costs for

buyer

?

?

(t) (t+1)

temporal

relevance

arrows


product

Product differentiatio

n among incumbents


the buyer

Product cost as fraction of the buyer’s

budget



product

product leads to

performance improvement for buyer


Amount of cash

available to the buyer

Industry product pays

for itself

product reduces costs for

buyer


identity of the firm’s competitors, the arrow from D(t) to F(t + 1) influences the behavior of

those competitors. The same argument applies to the influence arrows from D(t + 1) to

F(t + 1) and F(t + 2), respectively. Third, a new type of arrows is added from the

competitive uncertainties F at one time period into the positioning decisions D at the next time

period. This is illustrated in two arrows from F(t) to D(t + 1) and then from F(t + 1) to

D(t + 2) in Figure 4.16. In decision analysis, an arrow extending from an uncertainty into a

decision is called an information arrow; it means that the decision-maker will know exactly

how that uncertainty will be resolved before making the decision.

Figure 4.16: Dynamic DAFF model for the second objective, second step: reshaping

industry change

The firm’s original positioning inevitably triggers a series of reactions from other industry

players. Those reactions induce the firm to respond, which in turn induces other players to

respond back. This continuous strategic interaction between the firm and other industry

players is captured effectively in the sequence of influence and information arrows illustrated

in Figure 4.16. When a firm selects a positioning track at (t), it not only chooses a set of

players to interact with at (t) (short-term) but also induces a reaction from those players and

causes a change in their behavior at time (t + 1) (long-term). When (t + 1) begins, the firm

Short-term

Long-term

Long-term


learns how the competitive landscape turned out in the previous time period (t), and it uses

this new information to re-assess its positioning. Based on the change in the competitive

behavior, the new positioning allows the firm to update the set of players it chooses to interact

with at (t + 1) (which is the short-term now), and it induces a reaction from those players and

causes a change in their behavior at time (t + 2) (which is the long-term now). This cycle

goes on as long as the decision-maker desires, repeating with every new time period in the

decision-maker’s timeframe. Eventually, the DAFF model produces an expected profitability

for every positioning policy, defined as the sequence of positioning tracks that the firm adopts

over all time periods. A positioning policy may reveal that a firm position in one industry

segment throughout all time periods, or it may reveal positioning in different segments during

different periods. As before, the firm should adopt the positioning policy that results in the

highest expected value of profitability.

Let us consider one of our earlier examples from the automobile industry to clarify this

modeling procedure. Going back to 2006, suppose that an auto manufacturer, we call Electra,

is launching a new car line and is thinking of three positioning decisions: car technology

[hybrid, electric], car model [luxury, economy], and customer sales [direct, dealer]. Electra is

interested in assessing its ability to reshape the competitive landscape over the coming nine

years, assuming that three years is a reasonable estimation of a short-term period.

Accordingly, we model three time periods (t), (t + 1), (t + 2), each spanning three years. If

Electra positions in [electric, luxury, direct] in 2007, it will compete with a unique set of

players and thus be exposed to a unique competitive landscape up until 2009. For example,

less direct rivals may exist in the [electric, luxury, direct] segment in comparison with the

[hybrid, economy, dealer] segment. Then, Electra has to analyze how its positioning in 2007

will change the behavior of the players it chooses to compete with up until 2012. For instance,

six years may be sufficient for a big manufacturer like General Motors to enter the [electric,

luxury, direct] segment and start competing directly with Electra. As 2010 commences,

Electra gains more information about the competitive landscape in each positioning segment

up until 2009. This new information helps Electra decide whether to adopt a new positioning

track, such as switching to the economy car model in order to enter the mass market and

increase sales. If Electra updates its positioning to [electric, economy, direct] in 2010, it needs


to consider how this new positioning may change the identity of the players it competes with

up until 2012, and how those players may react to its new positioning up until 2015.

Eventually, the DAFF model outputs the expected profitability for every possible positioning

policy in 2015, which advises Electra on the exact sequence of positioning tracks it should

pursue every three years till the end of 2015. A positioning policy example is presented in (3)

for illustrative purposes only. Also, for full disclosure, we confess that our hypothetical story

on Electra is undeniably inspired by that of Tesla Motors [52].

[electric, luxury, direct]𝑡 where (𝑡) = 2007 − 2009

[electric, economy, direct]𝑡+1 where (𝑡) = 2010 − 2012

[electric, economy, direct]𝑡+2 where (𝑡) = 2013 − 2015

(3)

After explaining the details of the dynamic DAFF model, it is important that we take a step

back to discuss how this approach upholds FF’s theories in competitive strategy. To start, the

notion that positioning influences the five forces by changing the behavior of other industry

players is consistent with Industrial Organization theory and supported by Porter. Porter

recognizes early in his work that “there are feedback effects of firm conduct (strategy) on

market structure” [53]. Equally important, the DAFF dynamic model can demonstrate how

and why a business succeeds or fails over time, which is a central question to tackle when

designing a dynamic theory of strategy [25]. For a business to continue to succeed, its

positioning at every period should not only influence future market conditions but also be

informed by and consistent with historic market conditions. The dynamic DAFF models

clearly capture both requirements through the influence and information arrows, respectively.

Here, we emphasize the role of information arrows in accommodating sudden or rapid

industry changes. When an industry faces new or quickly evolving challenges, information

arrows preserve and transfer that intellegence from one period of the competitive analysis to

the next. A firm’s positioning strategy that does not account for the most recent competitive

information may be outdated, resulting in long-term failure despite short-term superior

performance. Ultimately, the dynamic DAFF modeling succeeds in fulfilling a key issue that a

dynamic theory of strategy must address: tracking the sequential interdependence between the

performance of the firm and that of the industry it operates in over time [25]. In the DAFF

CHAPTER 4 — Best Practices in DAFF Modeling 202

model of Figure 4.16, we see how the industry’s current structure informs the firm’s future

decisions, and how the firm’s current decisions influence the industry’s future structure.

6 Best Practices in DAFF Modeling

So far, we have discussed the various aspects and advantages of applying the DA method to

model and operationalize the FF framework and to fulfill its corresponding objectives in

competitive strategy. Reasonably, however, the introduced DAFF models may become too

complex or confusing in the absence of clear guidance on their development and use. This

section aims to highlight five major best-practices that business analysts and managers should

keep in mind when building and assessing the proposed DAFF Bayesian networks and

decision diagrams.

First, any DAFF model should start with and include all five competitive forces. The decision-

making manager can then proceed to add the driver and factor uncertainties that would help

her assess the five force uncertainties. As mentioned before, the decision-maker need not

analyze every driver mapped in the Detailed Network of Figure 4.6. Sometimes, the decision-

maker may have deep and perfect knowledge of specific market drivers, or she may deem

some drivers immaterial to competition in her industry; in both cases, these driver

uncertainties can be excluded from the DAFF model. On the other hand, if the decision-maker

is not able to assess specific drivers or factors due to their complexity or vagueness, new

parent uncertainties can be added to facilitate their assessment.

Second, the number of arrows extending into an uncertainty – either from a decision or from a

parent uncertainty – should be limited. Mathematically, when a decision node with N

alternatives extends an influence arrow into an uncertainty, the number of probability

assignments required at that uncertainty increases N folds; the same is true for a relevance

arrow extending from a parent uncertainty with N degrees. Therefore, it is preferable to cap

the number of arrows pointing into an uncertainty at four, but surely no more than five. For an

uncertainty node with five parents, each with two degrees, the number of probability

assignments required is 32 (25), which is already a challenging task for managers to handle.

CHAPTER 4 — Best Practices in DAFF Modeling 203

Third, also related to uncertainty modeling, probability elicitation from experts and decision-

makers should be handled efficiently. While an iterative analysis of the uncertainties in the

DAFF model may help expose cognitive biases, we caution that this iterative process should

not be abused. As obvious as it may sound, we stress that every probabilistic input should be a

true and honest representation of the decision-maker’s information, which in turn should be

shaped, informed, and validated by objective and factual data. If the probabilistic input into

the model is manipulated to yield a desirable output by the decision-maker, the model will

almost certainly fail to offer any new insights, defeating the purpose of the DAFF modeling

exercise in the first place. To ensure effective modeling of uncertainty, several protocols for

performing probability assessment have been proposed in literature [54]; among others, the

Stanford/SRI protocol [55] is widely used by DA practitioners.

The fourth modeling note is related to choosing a proper measure of profitability. We reiterate

that Porter recommends the use of ROIC because it comprehensively accounts for revenue and

cost figures, including CAPEX. The additional benefit of ROIC is that it can be computed

from the firms’ balance sheets and income statements – documents that are readily available

and that managers are familiar with. From a different perspective, NPV is a preferred and

widely used metric for assessing value in DA. Unlike the ROIC ratio, NPV is an absolute

measure, which facilitates clear ranking of decision prospects. If two prospects have equal

NPV, we can reasonably conclude that they have equivalent scale and profitability. The same

argument does not apply to ROIC, for two prospects of different scales and/or different

combination of EBT and CAPEX may still have the same ROIC ratio. Of course, the use of

NPV has its own challenges, especially in the dynamic assessment of competitive landscapes.

The NPV calculation assumes that the future discounted cash flows are known throughout the

business’s lifetime. However, as demonstrated in the DAFF dynamic model, the future

economics of the business might change with the uncertain competitive landscape. Ultimately,

we understand that different managers may be inclined to use different profitability metrics

based on their firms’ established ways of doing and evaluating business. Whatever

profitability metric the firm chooses to use in its DAFF modeling, it must always meet three

criteria: account for the effect of the competitive forces on the basic economic parameters; be

quantitative and computable; and be intuitive to the decision-maker.

CHAPTER 4 — Broader Alignment between Decision Analysis and Porter’s Competitive

Strategy 204

Finally, we briefly discuss how managers should think about their positioning alternatives. To

be unique, different firms within the same industry may – and perhaps should – choose to

model different positioning decisions, or different alternatives for the same positioning

decisions. Working towards that goal, we assert that, consistent with strategy literature, the

manager in a specific firm should select the set of feasible positioning alternatives based on

her firm’s own competences and available resources [26], and in accordance with her firm’s

cultural and behavioral norms [56]. Indeed, because different decision-makers may have

different information and beliefs (guiding their probability assessment) about the industry and

how their actions change the competitive landscape, firms with the same set of feasible

positioning alternatives may end up pursuing different positioning strategies.

7 Broader Alignment between Decision Analysis and Porter’s

Competitive Strategy

As we progressed in modeling the five forces framework using decision analytic tools, we

uncovered additional areas of alignment between decision analysis and Porter’s competitive

strategy, which we briefly overview in this section.

To start, a notable similarity exist between Porter’s display and description of tailored value

chains [8] and DA’s display and description of strategy tables [10, 57]. Both concepts

suggest that the set of strategic activities defining the firm’s competitive behavior cannot be

chosen randomly; those activities must be guided by an overarching theme or trend. In

Porter’s world, the firm’s activities along its value chain must serve the firm’s value

proposition and therefore be different than those of its rivals. In the DA world, the firm’s

activities must make sense together, which excludes a nontrivial pool of positioning

combinations that are technically correct yet practically unintuitive for the decision-maker.

Along the same lines, both DA and Porter highlight the importance of fit. Porter notes that fit

among the value-chain activities deters imitation, which allows the firm to sustain its

competitive advantage. Magretta explains: “as fit lowers the probability of successful

imitation, it raises the penalty of failure precisely because the activities are interconnected. A


small shortfall in one can produce ripple effects elsewhere.” DA can effectively capture this

“fit” and its corresponding “ripple effects” in its Bayesian networks and decision diagrams.

One way to demonstrate such capability is to examine how two positioning decisions

influence the same uncertainty. Our story on Electra provides a good example. Let us suppose

that switching from [dealer-sales] to [direct-sales] of an electric economy car reduces the

probability of strong Rivals from 0.6 to 0.4. Equivalently, switching from an [economy] to a

[luxury] model of an electric dealer-sold car yields the same outcome. However, if both the

[luxury] and [direct-sales] alternatives are combined in one positioning track, the probability

of strong Rivals drops from 0.6 to 0.1. In this case, the fit between the [luxury] and [direct-

sales] alternatives reinforces their collective effect, resulting in a less competitive landscape.

The probabilistic interdependence between the multiple elements of a DA decision diagram

can also manifest the “ripple effect”. Changing a positioning decision may change the

probability distributions of the influenced driver uncertainties, which in turn may change the

probabilities of their relevant uncertainties. As a result, the stronger the dependencies between

the nodes of a DA model, the harder it is to copy the firm’s network of positioning activities.

By facilitating the design of tailored positioning alternatives, enforcing trade-offs when

choosing positioning alternatives, and ensuring fit among the chosen positioning alternatives,

DAFF allows the firm to fulfill Porter’s ultimate recipe for establishing and sustaining

competitive advantage. While tailoring and trading-off positioning decisions “prevent existing

rivals from copying [the firm’s] good strategy, either by straddling or repositioning”, fit helps

sustain competitive advantage “against new entrants, even the most determined of them” [8].

8 Conclusions

This study explains the foundations of DAFF, a decision analytic modeling approach to

Michael Porter’s five forces framework in competitive strategy. We start by briefly

introducing the concepts of decision analysis (DA) and the five forces framework (FF), and

we identify key areas where DA can augment previous endeavors to further the

operationalization and implementation of FF by real firms and in real industries. As the main

focus of this research, we then provide a detailed description of the various DAFF


components, and we develop a series of DAFF models to fulfill the two main objectives of

competitive strategy: positioning in the industry, and reshaping the industry.

The various elements of the FF strategic framework are first translated into DA terms and

categorized as uncertainty-related, value-related, or decision-related. The five competitive

forces – Substitutes, Buyers, Suppliers, New Entrants, and Rivals – and their underlying

drivers are modelled as uncertainties. Also counted as uncertainties are industry-specific

factors related to governmental regulations (Regulation), technological advances

(Technology), market growth rate (Growth), and complementary products and services

(Complements). Both the generalizable forces and drivers as well as the industry-specific

factors impact Cost, Price, and Quantity, which are treated as uncertain economic parameters.

After modeling all these uncertainties, we explain how to effectively define profitability and

model it as a value metric. The final element of DAFF is the firm’s set of competitive actions

within the analyzed industry, which can be categorized into three types of decisions: Value

Proposition, Value Chain, and Economic decisions. Putting these elements together results in

two types of DAFF models: the DAFF Bayesian Network evaluates the competitive

performance of the whole industry whereas the DAFF Decision Diagram evaluates the

performance of specific industry segments and/or firms.

Subsequently, we show how the proposed DAFF models allow fulfilling the two main

objectives of Porter’s competitive strategy: positioning in and reshaping the industry.

Positioning is a short-term objective, and it entails three consecutive analyses focusing on

current and very-near-future profitability: of the overall industry, of distinct positioning

segments in the industry, and of a specific firm or business within each positioning segment.

Building on the outcomes from these three steps, reshaping the industry becomes the long-

term objective, and it requires two additional analyses focusing on distant-future profitability:

predicting the evolution of the industry structure then modifying the evolution of the industry

structure. After detailing how to model the three steps for industry positioning, we provide a

conceptual description of how to model the remaining two steps for industry reshaping.

Practically, the DAFF modeling of industry reshaping proves to be a dynamic extension of the

DAFF modeling for industry positioning.


Along the way, we highlight several benefits for using DAFF to conduct FF-based competitive

analyses, primarily the ability to: generalize the assessment model while customizing its

features to evaluate different industries and firms; explicitly account for competitive and

economic uncertainty; clearly map the relation between the industry’s competitive forces and

the firm’s competitive actions; clearly map the relation between competitive forces and

economic performance; compare, rank, prioritize, and track the interdependence among the

competitive forces, drivers, and industry factors; outline the firm’s scope of control; and

expose and reduce cognitive biases. Atop all that, perhaps the most important feature of DAFF

– indeed, one that enables all aforementioned benefits – is quantification. DAFF allow

quantifying multiple aspects of the competitive analyses, including: the definition of all

possible realizations for each force, driver, factor, or economic uncertainty; the likelihood of

each realization; the firm- or industry-specific profitability value; as well as the maximum

value that the firm should pay to reduce uncertainty through additional information or broader

control.

Eventually, this work shows that DAFF upholds and incorporates Porter’s “four principle

issues [that] emerge … as one contemplates a theory of strategy”: approach to theory building,

chain of causality, time horizon, and testing [25]. First, DAFF balances between what Porter

refers to as the “framework” and the “model” approaches to building a strategy. On one hand,

the generalizable structure of the decision diagrams allows them to both “capture much of the

complexity of actual competition” and “help the analyst to better think through the problem by

understanding the firm and its environment and defining and selecting among the strategic

alternatives available, no matter what the industry and starting position [are]” [25]. On the

other hand, DAFF relies on “mathematical models of limited complexity” to quantify the

economic impact of the competitive forces, realizing the need to make some simplifying and

manager-friendly assumptions and adjustments along the way. The second principle issue, the

chain of causality, is captured in the “flexible ancestral” structure of the Bayesian networks.

As already established, DAFF allows the decision-maker to analyze competitive drivers at

multiple ancestral levels in order to achieve clarity of thought, and those driver uncertainties

are usually connected in a causal direction to facilitate probability assessment. Third, the time

horizon aspect of the strategy theory is clearly accounted for in modeling the short-term and


long-term competitive objectives, where we demonstrated how a firm can comprehend,

anticipate, and respond to the behavior of other industry players over multiple periods of time.

Finally, on the issue of testing, Porter states that “empirical testing is vital for both

frameworks and models.” The DAFF approach is hard to test empirically, for DAFF models

aim to assess the competitive landscape in the near or distant future, not in the past. However,

both DA and Porter converge on recognizing the important role of case-studies as a testing

tool. Evident in both competitive strategy literature [2, 8] and DA consulting practices [58,

59], “in-depth case studies [can] identify significant variables, explore the relationships among

them, and cope with industry and firm specificity in strategy choices” [25].

8.1 Future Work

The ultimate objective of this research is to present and promote DAFF models as generic

robust tools that managers and executives can easily and intuitively use to evaluate

competitive strategies for their line of business. However, like any modeling approach, DAFF

has its own challenges and shortcomings. Perhaps the most apparent challenge is related to the

potential need for extensive data input. If the number of uncertainties in the DAFF model is

too big, the decision-maker might find their probabilistic analysis too time-consuming.

Similarly, a congested network of relevance arrows linking these uncertainties might force the

decision-maker to consider very hard trade-offs when assigning conditional probabilities, thus

compounding the difficulty of the competitive analysis. Fundamentally, this potential

computational burden exposes DAFF’s most significant vulnerability: seeing the “trees” while

losing the “forest.” Originally, we asserted that the detailed nature of the DAFF models help

the decision-maker gain clarity regarding the competitive landscape in her industry. Ironically,

however, this same need to dig deep into each DAFF element (tree) might also obstruct the

decision-maker’s ability to see the big picture (forest), which is essential in strategy.

Consequently, even if managers understand the DAFF approach and appreciate its benefits,

they might find the DAFF models too complex to apply on daily basis in real life.

To resolve this dilemma, and to encourage DAFF’s wide adoption and utilization by

businesses and firms, future work should focus on exploring ways to streamline the DAFF

Bayesian networks and decision diagrams. We offer a glimpse of how to propel and expand


the DAFF research in that direction. The premise is rather straightforward: the large number of

driver uncertainties might render a DAFF model too big or complex; yet, these drivers are

necessary to properly assess the competitive five forces; so, how about we eliminate all the

drivers after assessing the forces, and while preserving the competitive intelligence derived

from them? In theory, this proposition is feasible and can be achieved by reversing the

direction of the relevance arrows between the competitive forces and drivers; if applied

correctly, it can simplify the DAFF analyses drastically.

As sketched in Figure 4.17, the proposed endeavor expands the scope of the formal execution

process of DAFF modeling, asserting it is best conducted in three sequential stages and as a

collaboration between the decision-making manager and an expert analyst. In Stage 1, the

manager identifies the positioning concerns that should be addressed in the competitive

analysis, and the analyst translates those concerns into a clear set of positioning alternatives.

The outcome of Stage 1 is a simple yet comprehensive DAFF representation that lists the to-

be-analyzed firm’s positioning decisions and industry’s five force uncertainties. In Stage 2, the

analyst works with the decision-maker to add and examine the driver uncertainties needed to

help assess the five force uncertainties. The analyst then develops a complete DAFF Bayesian

Network or Decision Diagram, connecting the positioning decisions to the drivers and

connecting the drivers to the forces. Stage 2 is all what we have expounded in this work.

Instead of stopping at Stage 2, we envision a Stage 3 where the analyst works on condensing

and simplifying the complete DAFF model from Stage 2 while preserving all its competitive

intelligence. To do this, the analyst reverses the direction of the relevance arrows between the

force and driver uncertainties, a technique commonly referred to as “tree flipping” in DA

[10]. Upon applying this technique, the five forces become the most ancestral uncertainties,

and their probabilistic distributions become unconditioned on the drives. At this point, the

drivers are no longer needed for assessing the competitive forces or the economic performance

in the industry, and they can be removed from the DAFF model altogether. Although

reversing relevance arrows is an established practice in DA [10, 13, 44], future work should

focus on developing robust tree-flipping algorithms that specifically simplify the DAFF

models; ideally, such algorithms would automate the transformation of the complete DAFF

model in Stage 2 to the simplified DAFF model in Stage 3.


Figure 4.17: Future opportunities to streamline the DAFF modeling

While eliminating the drivers, tree flipping preserves all the probabilistic intelligence in the

complete DAFF model. Eventually, as shown in Figure 4.17, Stage 3 simplifies the DAFF

model representation drastically; the firm’s positioning decisions become directly linked to the

industry’s force uncertainties, and the force uncertainties become directly linked to one

another. Equally important, Stage 3 helps preserve the “forest”; the decision-making manager

remains focused on the big-picture related to competition and positioning, while the expert

analyst leads the work on fully examining then eliminating the detailed drivers in the analysis.

Positioning Decision A

Positioning Decision B

Force 1

Positioning Decision C

Force 2

Force 3

Force 4

Force 5

Stage 1

Add drivers to connect the positioning decisions to the

competitive forces

Flip the tree to remove all drivers



Force 1


Force 2

Force 3

Force 4

Force 5

Factor I Factor IIStage 2



Force 1


Force 2

Force 3

Force 4

Force 5

Stage 3

Forest

Trees

Forest

Co

mp

leted research

Futu

re research


Overall, the streamlined representation in Stage 3 helps managers better comprehend the

DAFF model and grasp its findings, which in turn incentivizes DAFF’s adoption as a generic

assessment tool in competitive strategy.


References

[1] M. E. Porter, "How Competitive Forces Shape Strategy," Harvard Business Review, March-April

1979.


January 2008.

[3] Accenture Academy, "Explaining Porter’s Five Forces," 2014. [Online]. Available:

https://www.accentureacademy.com/d/course/1000007629/?tabId=1&moduleId=507. [Accessed

2015].

[4] FME, "Porter's Five Forces - Strategy Skills," 2013. [Online]. Available: http://www.free-

management-ebooks.com/dldebk-pdf/fme-five-forces-framework.pdf. [Accessed 2015].

[5] R. Marks, "Lecture Notes - Industry Analysis," Australian Graduate School of Management, 2003.

[6] M. E. Dobbs, "Guidelines for applying Porter's five forces framework: a set of industry analysis

templates," Competitiveness Review, vol. 24, no. 1, p. 32–45, 2014.

[7] H. Lee, M.-S. Kim and Y. Park, "An analytic network process approach to operationalization of

five forces model," Applied Mathematical Modelling, pp. 1783-1795, 2012.

[8] J. Magretta, Understanding Michael Porter: The Essential Guide to Competition and Strategy,

Cambridge: Harvard Business Review Press, 2012.

[9] SDG, "Strategic Decision Group International LLC," 2015. [Online]. Available:

http://www.sdg.com/about-sdg/. [Accessed 2015].

[10] R. A. Howard and A. E. Abbas, Foundations of Decision Analysis, 1st ed., United States: Pearson,

2016.

[11] R. A. Howard and J. E. Matheson, "Influence Diagrams, INFORMS," Decision Analysis, pp. Vol.

2, No. 3, pp. 127–143, 2005.

[12] R. D. Shachter, "Evaluating Influence Diagrams," Operations Research, pp. Vol. 34, No. 6, pp.

871-882, 1986.

[13] R. D. Shachter and D. Bhattacharjya, "Solving influence diagrams: Exact algorithm," Wiley

Encyclopedia of Operations Research and Management Science, 2010.

[14] Decisions Systems Laboratory, "GeNIe and SMILE," 2013. [Online]. Available:

https://dslpitt.org/genie/. [Accessed 2013].

[15] Lumina, "Influence Diagrams," 2015. [Online]. Available:

http://www.lumina.com/technology/influence-diagrams/. [Accessed 2015].


[16] M. E. Porter, Competitive Strategy Techniques for Analyzing Industries and Competitors, Free

Press, 1998.

[17] M. E. Porter, "Strategy and the Internet," Harvard Business Review, 2001.

[18] The Boston Globe, "Automaker Tesla looks to bypass car dealers," 2013. [Online]. Available:

https://www.bostonglobe.com/business/2013/11/20/tesla-battles-auto-dealers-direct-sales-

consumers/3f1xBFN21xH8QqQc3jijTP/story.html. [Accessed 2015].

[19] KBB, "Kelly Blue Book," 2015. [Online]. Available: http://www.kbb.com/. [Accessed 2015].

[20] CPUC, "California Renewables Portfolio Standard (RPS)," 2015. [Online]. Available:

http://www.cpuc.ca.gov/PUC/energy/Renewables/. [Accessed 2015].

[21] E. Halper, "Rules prevent solar panels in many states with abundant sunlight," Los Angeles Times,

2010.

[22] H. Courtney, J. Kirkland and P. Viguerie, "Strategy Under Uncerainty," Harvard Business Review,

1997.

[23] P. Ghemawat, Commitment: The Dynamic of Strategy, New York: Free Press, 1991.

[24] A. M. Brandenburger and B. J. Nalebuff, Co-opetition, New York: Doubleday, 2011.

[25] M. E. Porter, "Towards a Dynamic Theory of Strategy," Strategic Management Journal, pp. 95-

117, 1991.

[26] Y. E. Spanos and S. Lioukas, "An Examination into the Causal Logic of Rent Generation:

Constrasting Porter's Competitive Strategy Framework and the Resource-Based Perspective,"

Strategic Management Journal, vol. 22, pp. 907-934, 2001.

[27] N. J. Foss, "Research in Strategy, Economics, and Porter," Journal of Management Studies, vol.

33, pp. 1-24, 1996.

[28] M. Song, R. J. Calanton and C. A. D. Benedetto, "Competitive Forces and Strategic Choice

Decisions: An Experimental Investigation in the United States and Japan," Strategic Management

Journal, vol. 23, pp. 969-978, 2002.

[29] T. Grundy, "Rethinking and reinventing Michael Porter's five forces model," Strategic Change, pp.

213-229, 2006.

[30] K.-J. Wu, M.-L. Tseng and A. S. Chiu, "Using the Analytical Network Process in Porter’s Five

Forces Analysis – Case Study in Philippines," Procedia - Social and Behavioral Sciences, pp. 1-9,

2012.

[31] J. Franek and A. Kresta, "Competitive strategy decision making based on the five forces analysis

with AHP/ANP approach," VŠB-Technical University of Ostrava, Ostrava, 2013.


[32] L. Downes, "Beyond Porter," Context Magazine, 1997.

[33] R. D. Shachter, "Model Building with Belief Networks and Influence Diagrams," 2007. [Online].

Available: http://www.usc.edu/dept/create/assets/001/50850.pdf. [Accessed 2015].

[34] R. D. Shachter, "An ordered examination of influence diagrams," Networks, vol. 20, no. 5, pp. 535-

563, 1990.

[35] Investopedia, "Return On Invested Capital - ROIC," 2015. [Online]. Available:

http://www.investopedia.com/terms/r/returnoninvestmentcapital.asp. [Accessed 2015].

[36] Investopedia, "Profit Margin," 2015. [Online]. Available:

http://www.investopedia.com/terms/p/profitmargin.asp. [Accessed 2015].

[37] SunPower, 2015. [Online]. Available: http://us.sunpower.com/.

[38] J. Church and R. Ware, Industrial Organization: A Strategic Approach, United States: The

McGraw-Hill Companies, 2000.

[39] B. Halvorson, "Will Low Resale Values Spoil The Cost Benefits Of Electric-Car Ownership?,"

2013. [Online]. Available: http://www.thecarconnection.com/news/1089368_will-low-resale-

values-spoil-the-cost-benefits-of-electric-car-ownership. [Accessed 2015].

[40] C. Rogers, "Resale Prices Tumble on Electric Cars," 2015. [Online]. Available:

http://www.wsj.com/articles/resale-prices-tumble-on-electric-cars-1424977378. [Accessed 2013].

[41] RAND, "Vehicle Production and Lifecycle Cost," [Online]. Available:

https://www.rand.org/content/dam/rand/pubs/monograph_reports/MR1578/MR1578.ch4.pdf.

[Accessed 2015].

[42] K. Aguirre, L. Eisenhardt, C. Lim, B. Nelson, A. Norring, P. Slowik and N. Tu, "Lifecycle

Analysis Comparison of a Battery Electric Vehicle and a Conventional Gasoline Vehicle," 2012.

[Online]. Available:

http://www.environment.ucla.edu/media/files/BatteryElectricVehicleLCA2012-rh-ptd.pdf.

[Accessed 2015].

[43] E. Dodge, "The Case for Electric Vehicles, Part 2: EV Costs," 2014. [Online]. Available:

http://breakingenergy.com/2014/10/02/the-case-for-electric-vehicles-part-2-ev-costs/. [Accessed

2015].

[44] R. D. Shachter, "Bayes-Ball: The Rational Pastime (for Determining Irrelevance and Requisite

Information in Belief Networks and Influence Diagrams)," Morgan Kaufmann Publishers Inc. San

Francisco, 1998.

[45] Tesla Motors, "About Tesla," 2016. [Online]. Available: https://www.teslamotors.com/about.

[46] The Dialogue, "Tesla Cars Evolution," 2016. [Online]. Available: http://www.the-

dialogue.com/en/en21-tesla-cars-evolution/.


[47] UCSUSA, "Electric Vehicle Timeline," 2015. [Online]. Available:

http://www.ucsusa.org/clean_vehicles/smart-transportation-solutions/advanced-vehicle-

technologies/electric-cars/electric-vehicle-timeline.html#.Vi6kuLerQdU. [Accessed 2015].

[48] SolarCity, "SolarCity delivers Better Energy," 2016. [Online]. Available:

http://www.solarcity.com/company/about.

[49] GTM, "Here Are the Top 5 Residential Solar Installers of 2014," Greentech Media, 2015. [Online].

Available: http://www.greentechmedia.com/articles/read/Here-Are-the-Top-Five-Residential-

Solar-Installers-of-2014. [Accessed 2015].

[50] SEIA, "Solar Industry Data," 2015. [Online]. Available: http://www.seia.org/research-

resources/solar-industry-data. [Accessed 2015].

[51] J. Burr, T. Dutzik, J. Schneider and R. Sargent, "Shining Cities - At the Forefront of America’s

Solar Energy Revolution," Environment America Research & Policy Center , 2014.

[52] E. Musk, "The Secret Tesla Motors Master Plan (just between you and me)," Tesla Motors, United

States, 2006.

[53] M. E. Porter, "The Contributions of Industrial Organization to Strategic Management," The

Academy of Management Review, pp. 609-620, 1981.

[54] M. G. Morgan, M. Henrion and M. Small, "Performing Probability Assessment," in Uncertainty: A

Guide to Dealing with Uncertainty in Quantitative Risk and Policy Analysis, New York,

Cambridge University Press, 1990.

[55] C. S. Spetzler and C.-A. S. S. V. Holstein, "Probability Encoding in Decision Analysis,"

Management Science, vol. 22, no. 3, pp. 340-358, 1975.

[56] M. A. Hitt, M. T. Dacin, B. B. Tyler and D. Park, "Understanding the Differences in Korean and

U.S. Executives' Strategic Orientations," Strategic Management Journal, vol. 18, no. 2, pp. 159-

167, 1997.

[57] R. A. Howard, "Decision Analysis: Practice and Promise," Management Science, vol. 34, no. 6, pp.

679-695, 1988.

[58] L. Neal and C. Spetzler, "An Organization-Wide Approach to Good Decision Making," Harvard

Business Review, Cambridge, 2015.

[59] C. S. Spetzler, "Chevron Overcomes the Biggest Bias of All," Strategic Decision Group, Palo Alto,

2011.

216

Chapter 5

DAFF Modeling of Competitive Strategy:

Positioning in the Near-Future U.S.

Residential Solar PV Industry

1 Introduction

Decision analytic modeling of the five forces strategic framework (DAFF) offers a robust and

quantitative approach to competitive analysis. After introducing the foundations of DAFF in

Chapter 4, this Chapter aims to demonstrate the applicability of DAFF by using it to evaluate

the competitive strategy of a major solar firm in the United States. Specifically, we undertake

a case study that aims to answer two questions: Is the competitive landscape in the U.S.

residential solar photovoltaic (PV) industry favorable between 2014 and 2016? And if so,

where should the solar firm position its residential business?

To answer these two questions, the authors of this study joined the solar firm’s strategic

planning team in the summer of 2014 and worked on developing two models: a DAFF

Bayesian Network and a DAFF Decision Diagram. Consistent with our explanation in Chapter

4, both DAFF models help the firm fulfill the first objective of competitive strategy, namely,

choose an appropriate positioning track within the target industry. In collaboration with the

company’s residential and strategic planning teams, this study completes the first two steps of

the positioning objective. First, we assess the competitive landscape and profitability of the

overall U.S. residential solar PV industry. Then, we identify a series of unique positioning


segments in the industry, and we analyze the profitability of each of those segments. Although

the work does not proceed to the third step of assessing the firm’s specific profitability in

every positioning segment, the first two steps prove sufficient to deliver valuable insights into,

and therefore to make clear recommendations on, the optimal positioning strategies. Here, we

note that in order to freely discuss and publically share the DAFF models’ inputs and outputs,

it was necessary that we refrain from disclosing the real identity of the solar firm in this case

study. Therefore, we shall conveniently refer to the firm as “SunEnergy” throughout this

Chapter.

The subsequent sections of this case-study competitive analysis proceeds as follows. First, we

provide a brief overview of the residential solar PV industry in the United States and some

background information on the examined solar firm. In Section 2, we explain how to develop

the DAFF models for SunEnergy’s positioning. In the first step aiming to analyze competition

in the overall industry, we build a DAFF Bayesian Network and describe its various

components, including: the five forces, their underlying drivers, the industry-specific factors,

and the economic parameters. Then, in the second step aiming to analyze various positioning

segments within the industry, we describe a series of Value Proposition and Value Chain

decisions that are important to SunEnergy. Adding those decisions to the DAFF Bayesian

Network yields a DAFF Decision Diagram. In Section 3, we explain the results from both

modeling steps. The first step allows assessing not only the uncertain power of each of the five

competitive forces but also the probabilistic interdependence among these forces. The DAFF

Bayesian Network also outputs probability distributions for the average cost, price, and annual

sales, as well as an expected average profit value for all industry incumbents. In the second

step, the DAFF Decision Diagram yields an expected profit value for each possible

positioning track, and we demonstrate the influence of the various positioning alternatives on

the power of each competitive force. We end this section by distilling the outputs from the

DAFF models into a clear and actionable list of recommendations regarding SunEnergy’s

positioning strategy in the U.S. residential solar PV market. Finally, Section 4 concludes this

case study by offering a summary of the main findings, reflecting on the advantages and

limitations of the DAFF modeling endeavors, and discussing how this analysis can be further

expanded in the future.


1.1 The U.S. Residential Solar PV Industry

The United States has witnessed a tremendous growth in the deployment of solar photovoltaic

(PV) energy systems over the past few years [1]. Between 2010 and 2014, the yearly solar

installations increased from 850 to about 6000 megawatts (MW) [2], and the total solar

workforce expanded from 93000 to roughly 174000 employees [3]. To put these figures in

context, we note that the capacity of an average nuclear power plant is on the order 1000 MW,

and its operation requires 400–700 full-time employees [4]. In addition, as reported by

Greentech Media, solar energy was the second-largest source of new electricity generation

capacity in 2013, exceeded only by natural gas [5]. Thus, beyond its vital role as a renewable

energy resource that can mitigate climate change [6], solar PV is progressively earning a

competitive and permanent place in the U.S. energy mix.

As the market for solar energy continues to develop and mature, the evolution of business

practices has carved out three distinct and relatively stable industries for solar PV:

residential, commercial and institutional (C&I), and utility. In accordance with Michael

Porter’s guidance on how to define industry boundaries [7], we recognize distinct product

offerings and competitive structures in the three solar industries, despite their shared

geographical scope. The major difference in product offerings is related to scale. Dictated by

the habits and needs of the electricity consumer, the size of the solar system varies by orders

of magnitude from one industry to the other: residential systems are usually 0.001–0.01 MW;

C&I systems are on the order of 0.01–1 MW; and utility systems are typically 1–100 MW [8].

Also related to system size is system maintenance. While utility solar systems are usually

equipped with sophisticated monitoring and remote-control capabilities to accurately predict

and adjust their real-time power output, C&I and residential systems may have simpler

capabilities, or even lack them altogether [9].

In terms of competitive structure, the most notable differences between the three industries are

those involving buyers and substitutes. The industries’ very labeling highlights their distinct

buyers: while the residential solar industry caters to households and homeowners, the C&I

solar industry targets for-profit and not-for-profit organizations, and the utility solar industry

focuses on serving electric utilities. Obviously, these solar customers differ drastically in their


organizational hierarchies, decision-making processes, financial capabilities, and needs.

Specifically, the different needs are illustrated not only in the aforementioned scale and

specifications of the solar system but also in what can feasibly substitute it. A homeowner

aiming to reduce his electricity bill may substitute the rooftop solar panels with a direct-load-

control or demand-response subscription [10, 11]. On the other hand, a commercial customer

aiming to offset its carbon emissions or to enhance its “green” branding may choose to build

or invest in a wind farm, an option that is not feasible for households [12]. Finally, in order to

reduce the carbon intensity of its power generation fleet, a utility may choose to retrofit some

of its existing fossil-fuel plants with carbon capture and storage systems instead of developing

new solar farms [13]; such alternative is available to neither residential nor C&I customers.

Although different solar regulations may be imposed in different states, these regulations share

the same underlying objectives such as managing access to the grid, providing financial

subsidies, or designing roadmaps to accelerate large-scale deployment [14]. This common

regulatory foundation, in addition to interstate free trade, makes it reasonable to assume that

the whole U.S. is the proper geographical scope for all three solar industries.

For this case study, we focus solely on the residential solar PV industry in the United States.

In 2014, about 20% of the 6000 MW installed solar capacity was residential in nature,

corresponding to a market size of roughly $5.6 billion [15]. With an average annual growth

rate around 50% between 2011 and 2014, industry analysts predict that “the outlook for the

U.S. residential solar market is extremely bright” [15, 16]. Different firms adopt different

business models in the industry, with activities that include the production, sales, installation,

financing, and servicing of solar panels. In 2014, despite the relatively large number of solar

installers, only 10 firms were responsible for installing 60% of the added capacity, two of

which installed more than 45% of that capacity. In addition, about two-thirds of the residential

systems today are categorized as “third-party ownership” (TPO), which means that they are

owned (e.g. leased) by the solar firm instead of the actual end-customer. In fact, 90% of the

TPO systems are owned by seven solar firms only, many of which are major installers. This

business reality, enforced by a series of recent mergers and acquisitions, seems to be

consolidating the residential solar industry around a few incumbents [15], one of which is the

focus of this case study: SunEnergy.


1.2 The Solar Firm

Through 2014 – the time of conducting this analysis – SunEnergy was an established solar

company with diverse business operations that include the design, production, installation,

operation, monitoring, maintenance, and financing of solar energy projects. The company

housed a large number of employees and installed numerous solar systems in multiple

locations across the U.S.

SunEnergy succeeded in becoming an established player in the utility and C&I solar

industries, but its residential business was still in relatively early stages. At the same time, the

company was aware of the wide consensus among experts’ predictions regarding the steady

growth of residential solar this decade [16, 17, 18]. Given its history in the solar business, and

motivated by the prospects of furthering its growth, SunEnergy became increasingly interested

in expanding its activities in the U.S. residential solar industry.

2 Methodology: Developing the DAFF Models

Before delving into any modeling details, we first need to define a proper timeframe for the

competitive analysis and therefore for the DAFF models. In consultation with SunEnergy’s

strategic planning and residential solar business teams, the timeframe was set to cover the

upcoming two years, extending from the fall of 2014 through the end of 2016. Consistent with

our notes in Chapter 4, this near-future timeframe allows SunEnergy to position where its

competitors are least threating, but it is not sufficient, nor intended, to exploit or modify the

strategic behavior of those competitors.

Also important to DAFF modeling is designating a specific decision-maker. Multiple parties

may contribute to information gathering and data analysis, but one decision-maker shall have

the final say on the inputs into the DAFF models. In this study, multiple members of

SunEnergy’s strategic planning and residential business teams collaborated to supply and

analyze market information, but the Director of Strategic Planning was the designated

decision-maker.


With that in mind, we rely on GeNIe to perform all the DAFF modeling in this case study.

Developed by the Decision Systems Laboratory at the University of Pittsburgh, GeNIe is an

open-access software package that is widely used to design and analyze decision-theoretic

models [19].

2.1 First Step: Assess the Overall Industry

In this step, we develop a DAFF Bayesian Network to analyze competition and its impact on

the economic performance of the overall residential solar PV industry in the U.S. As discussed

in Chapter 4, this endeavor requires a detailed characterization of: the five competitive forces

and their underlying drives; the regulatory, technological, and growth-rate factors; and the

economic parameters that determine the industry’s profitability.

2.1.1. Competitive Forces and Drivers

For every competitive force, the DAFF modeling requires: identifying a set of driver

uncertainties that are relevant to the force uncertainty, defining the possible realizations (i.e.

degrees) of the force and each driver uncertainty, assigning probabilities over these degrees,

and then computing the output probability distribution over the power of the force.

We offer a detailed account of these requirements for the force of Substitutes, whose network

of driver uncertainties is illustrated in Figure 5.1. To start, we define substitutes as “any

product or service that residential customers can use to reduce their electricity bill.” In that

regard, potential substitutes may include smart meters, smart thermostats, more efficient home

appliances, or direct-load-control and demand-response subscriptions. As shows in Figure 5.1,

the power of Substitutes is conditioned on three parent drivers: Price-performance tradeoff

relative to this industry product, Regulatory and technical feasibility, and consumer switching

cost from this industry product to substitutes. The power of Substitutes is more likely to be

high when potential solar substitutes offer better price-performance tradeoff, are backed by

more favorable regulations, are more technically feasible, or can be easily adopted by a

disgruntled solar customer.


Figure 5.1: DAFF modeling of the force of Substitutes and its relevant drivers

To achieve clarity, we condition the probabilistic analysis of the Price-performance tradeoff

relative to this industry product uncertainty on three criteria that are deemed important by

SunEnergy: Substitute performance, Substitute upfront cost, and Substitute bill savings; each

of these criteria is modeled as a parent driver node to the Price-performance tradeoff relative

to this industry product node. Subsequently, to further inform the analysis of the Substitute

performance driver, we condition it on three additional drivers: Substitute installation and

maintenance, Substitute control and operation, and Substitute climate impact; reasonably, a

substitute is more likely to outperform a solar system if it is faster to install and maintain,

easier to automate, or more capable of reducing greenhouse gas emissions. Similarly, already

illustrating one form of interaction between two competitive forces, we assert that the solar

Product perceived cost as fraction of the customer budget (Buyers driver) may help inform the


Price-performance tradeoff relative to

this industry product

Product perceived

cost as fraction of

the customer

budget

Regulatory and technical

feasibility

Customer switching cost from this

industry product to substitutes

Substitute upfront

cost

Substitute bill savings

Substitute performance

Substitute climate impact

Substitute control & operation

Substitute installation & maintenance

Cost reduction

for customer

by industry product

Substitutes

Buyers


analysis of the Substitute upfront cost, and the Cost reduction for customer by industry

product (another Buyers driver) may help inform the analysis of the Substitute bill savings.

Finally, because home energy technologies and regulations continue to evolve, the analysis of

the Regulatory and technical feasibility driver is conditioned on uncertain regulatory and

technological factors, as we discuss in the next Section.

After introducing and connecting the force and driver uncertainties, we proceed to define and

characterize their degrees. As explained in Chapter 4, each uncertainty is characterized by two

or more mutually exclusive and collectively exhaustive (MECE) degrees, representing the

possible resolutions of that uncertainty in the future. In our work with SunEnergy’s teams, we

also make an effort to quantify the definitions of the driver degrees, to the extent possible.

Each degree is then assigned a probability, indicative of the decision-maker’s beliefs about its

future realization. In this case, SunEnergy’s probability assignments reflect the company’s

collective intelligence on the competitive landscape over the next two years.

Referring back to the uncertainty network for Substitutes in Figure 5.1, we demonstrate the

degree definitions and probability assignments for all uncertainty nodes with red borders.

Figures 5.2–5.5 are snapshots of the assessment tables for these uncertainties, obtained

directly from the GeNIe DAFF model. We start with the most ancestral driver in Figure 5.2:

Cost reduction for customer by industry product. Here, the two quantifiable MECE degrees

benchmark cost reductions from the solar system against the homeowner’s electricity bill; they

are defined as: {<15% reduction in power bill} and {>15% reduction in power bill}. Because

this uncertainty has no parents, each of its degrees is assigned an unconditional probability

value, and the two probability values sum up to one.

Similarly, two MECE degrees are defined for the Substitute bill savings driver in Figure 5.3:

{less bill savings} and {more bill savings} (compared to a solar system). However, this

uncertainty is conditioned on one parent – the aforementioned Cost reduction for customer by

industry product driver – and so is the probability distribution over its two degrees.

Accordingly, for every prospect of the parent uncertainty, the two degrees are assigned distinct

probability values that add up to one. For instance, given that a solar system achieves

224

Figure 5.2: Example probabilistic analysis of a driver: Cost reduction for customer by industry product

Figure 5.3: Example probabilistic analysis of a driver: Substitute bill savings

225

Figure 5.4: Example probabilistic analysis of a driver: Price-performance tradeoff relative to this industry product

Figure 5.5: Example probabilistic analysis of the power of Substitutes


{<15% reduction in power bill}, the probability of customers enjoying {less bill savings}

from substitutes is 0.25 whereas the probability of {more bill savings} is 0.75; conversely,

given that a solar system achieves {>15% reduction in power bill}, these probabilities change

to 0.8 and 0.2, respectively. Accounting for all these potential scenarios, the DAFF model then

computes an overall probability distribution for Substitute bill savings, outputting a 0.61

probability for {less bill savings} and a 0.38 probability for {more bill savings}.

Three degrees are also modelled for the Price-performance tradeoff relative to this industry

product driver in Figure 5.4: {substitute superior to solar}, {substitute equivalent to solar},

and {substitute inferior to solar}. In this case, each degree is assigned eight distinct

conditional probability values, corresponding to eight distinct parental prospects; for each

prospect, the probability values of the three degrees sum up to one. For example, given that

the substitutes yield {less bill savings}, {high upfront costs}, and {inferior performance}

compared to the solar system, the probability of {substitute superior to solar} is 0.01 whereas

the probability of {substitute inferior to solar} is 0.98. By balancing the evaluations of all

prospects, the DAFF model outputs an overall probability distribution for the three degrees of

the Price-performance tradeoff relative to this industry product driver: 0.29 for {substitute

superior to solar}, 0.25 for {substitute equivalent to solar}, and 0.46 for {substitute inferior to

solar}. This means that, per SunEnergy’s intelligence, solar panels are likely to offer better

price-performance tradeoff to customers than other home energy equipment or services.

Analyzing the two degrees {high} and {low} of the power of Substitutes in Figure 5.5 follows

the same logic. Each degree is assigned 12 distinct conditional probabilities corresponding to

12 distinct combinations of parental prospects; for each parental prospect, the probability

values of the two degrees add up to one.

Before ending this discussion on modeling the force of Substitutes and its drivers, we make an

important note. Compared to the original DAFF modeling of Substitutes in Chapter 4, Figure

5.1 introduces three changes. First, we expand the analysis of Price-performance tradeoff

relative to this industry product driver by conditioning it on six new ancestral drivers. Second,

we add the parent driver Regulatory and technical feasibility. Third, we eliminate the original

driver Availability of substitutes for this industry’s products. Based on available market data,


we know, with certainty, that substitutes do exist for residential solar panels, including gadgets

such as smart thermostats [20] and services such as demand-response programs [21, 22].

SunEnergy’s belief about the persistence of the substitutes’ availability and variety in the near

future cancels the need for analyzing this driver uncertainty. As will become evident, all these

DAFF customizations are necessary not only to fit the specific context of the investigated

residential solar industry in the U.S. but also to help SunEnergy’s decision-makers understand,

analyze, and communicate relevant information and findings.

The DAFF modeling of the other four competitive forces and their underlying drivers

proceeds in a very similar fashion. The resulting Bayesian network connecting the various

competitive force and driver uncertainties is presented in Figure 5.6, and a full list of their

degrees is presented in Appendix A. At this point, we provide a brief overview of the DAFF

modeling for the remaining four forces: Buyers, New Entrants, Rivals, and Suppliers.

When analyzing the power of Buyers, the nature of the residential solar business necessitates

that we distinguish between two types of buyers: end-customers and dealers. Similar to their

role in the auto industry, dealers act as intermediaries that help deliver the solar system to the

final end-customer – the household in this case. However, unlike auto manufacturers in most

states, residential solar developers are allowed to directly sell and service end-customers

without the need to engage dealers in the transaction.

To capture this competitive configuration, we condition the uncertain power of Buyers on two

uncertain parent drivers: Power of end-customers and Power of dealers. Subsequently, the

analysis of these two drivers is further informed by conditioning them on three levels of

additional ancestral drivers. The Power of end-customers is conditioned on the Customer price

sensitivity and the Customer switching cost away from this industry product, which in turn are

further shaped by investigating the customers’ need for the solar system and their ability to

afford it. Equivalently, the Power of dealers is conditioned on the Dealer ability to influence

customers downstream, Dealer volume of sales per incumbent, and Dealer threat to integrate

backward, which in turn are further shaped by examining the dealers’ brands, business

models, and geographical scope of reach. Intuitively, dealers with strong brands, preferential

access to populated residential areas, or national footprint are more likely to have

228

Figure 5.6: DAFF modeling of all competitive forces and drivers in the U.S. residential solar PV industry

Threat of new

entrants

Barriers to entry


Previous responses

by incumbents

Relative reliance of customer on

industry product


Customer switching cost

among industry products


incumbents

Fragmentation of the industry



Incumbent switching


Supplier switching

cost between

incumbents


among suppliers

Availability of substitutes for

what the supplier provides

Supplier threat to integrate foreword


Price-performance tradeoff relative to

this industry product

Commitment of incumbent to

retain and fight over market

share


among incumbents

Bargaining Power of

Buyers

Costumer price

sensitivity

Product perceived

cost as fraction of

the customer

budget

Dealer ability to influence

customers downstream

Dealer threat to integrate backward

Customer switching cost away from the

incumbent product

Rivalry

Intensity of competitionBasis of

competition

Ability to enforce

practices desirable for whole industry

Extent of exit

barriers


Customer need to trim immediate

cost of industry product


discounting by incumbents

Inability to read other


Importance of non-profit

goals

Incumbent joint

investments with current

supplier

Number of industries supplier

serve

Profits extracted by

suppliers from other

industries


feasibility

Dealer partnership

with incumbent

Customer switching cost from this

industry product to substitutes

Substitutes

Buyers

New Entrants

Rivals

Suppliers

Substitute upfront

cost


Substitute performance



Substitute installation & maintenance

Power of end-customers

Power of dealers

Dealer volume of sales per

incumbent

Dealer reach

Dealer brand

Performance improvement for customer by industry

product

Amount of cash

available for

customer

Cost reduction

for customer

by industry product

Size dependent

disadvantages for new entrant

Size independent

disadvantages for new entrant

Unequal access to

distribution channel by

new entrant

Customer adoption

rate of product by new entrant

Extent of resources available

for incumbents

Incumbent economies

of scaleIncumbent

prime location

Incumbent cumulative

in-house experience

Incumbent established

brand

Incumbent IP

Control over

distribution channel by incumbent

Limitation of distribution

channels

Size and availability of capital

needed by new

entrant

Incumbent assets

Incumbent R&D

spending

Incumbent customer acquisition cost

Efficiency of capital markets

Incumbent network effects

Customer trust in

incumbent

Excess cash

Borrowing power


capacity

Market segmentation


customer segments

High fixed costs

and low variable

costs

Presence of an

industry leader

Incumbent inventory


higher bargaining power. Ultimately, the power of Buyers is more likely to be high when the

power of either the end-customers or the dealers is high.

The ancestral drivers for the Power of end-customers are consistent with the original DAFF

modeling of the Buyers force in Chapter 4. In contrast, the Power of dealers ancestral drivers

emerged during our discussions with SunEnergy’s teams; business experts believe that these

uncertainties play an important role in shaping competition within the residential solar

industry and therefore must be accounted for.

Moving on, Figure 5.6 shows that the analysis of the power of New Entrants is informed by

several drivers. These drivers are structured over three ancestral levels, starting with two

parents: Barriers to entry and Expected retaliation. Reasonably, the power of New Entrants is

more likely to be low when new firms in the industry either struggle to overcome steep

barriers to entry or expect fierce retaliation by incumbents upon entry. To facilitate its

probabilistic assessment by the decision-maker, Barriers to Entry is conditioned on five

drivers. The Size dependent disadvantages for new entrant driver accounts for the economies

of scale achieved by current solar incumbents whereas the Size independent disadvantages for

new entrant driver accounts for the incumbents’ brands, intellectual property (IP), human

capital, and geographical spread. Indeed, current residential solar firms seem to be proactive in

building strong brands and IP, with examples ranging from SolarCity’s famous “green trucks”

to SunPower’s, SolarCity’s, and Panasonic’s continuous fight to claim the “most efficient

rooftop solar panel” title [23, 24]. Equivalently, the Unequal access to distribution channel by

new entrant driver examines the extent to which current incumbents have access and control

over the distribution channels of solar systems. Then, the fourth driver, Customer adoption

rate of product by new entrant is shaped by the networking effects in the industry as well as

by the reliance of the customer on the solar system; new entrants are more likely to enjoy high

adoption rates of their product when the networking effects among existing incumbents are

weak, when the solar savings are high, and when the switching costs for the customer are low.

Finally, the Size and availability of capital needed by new entrant driver is informed by key

financial attributes such as the efficiency of the available capital markets, as well as the

customer acquisition cost, R&D spending, assets, and inventories of the incumbents.


Similarly, Expected retaliation is conditioned on four drivers: Extent of resources available for

incumbents, Previous responses by incumbents, Willingness of price discounting by

incumbents, and Commitment of incumbent to retain and fight over market share. The Extent

of resources available for incumbents is further clarified by accounting for the incumbent’s

excess cash, production capacity, and borrowing power. With that in mind, a new entrant is

more likely to experience fierce retaliation if the solar incumbents have deep pockets and/or

are capable and willing to engage in price wars, fight over market share, pursue hostile

acquisitions, or match their competitors’ offers.

While this DAFF modeling of the New Entrants preserves the overarching Bayesian structure

presented in Chapter 4, it eliminates and rearranges multiple driver uncertainties – and the

relevance arrows linking them – in order to ease the probabilistic analysis. For example, the

decision-maker felt more comfortable evaluating how the Unequal access to distribution

channels directly affects the strength of Barriers to Entry, without considering the Need to

bypass incumbent existing advantages by new entrant; as a result, the latter driver does not

appear in Figure 5.6.

As for the power of Rivals, Figure 5.6 shows that Rivalry is conditioned on three direct parent

drivers: Basis of competition, Intensity of competition, and Inability to read other incumbents’

signals. Those drivers are, in turn, conditioned on two ancestral levels of causal drivers. For

example, the assessment of the Basis of competition driver is better informed by conditioning

it on four additional drivers that convey the following story: rivals are more likely to lead

price-based competition if the solar market is homogenous and segmentation is limited (e.g.

all solar customers are homeowners with bad credit score); if the differentiation in solar

offerings is short-term or nonexistent (e.g. all solar systems come with free installation or in

one color); if the customers’ switching costs are low (e.g. customers face little legal

complications or penalties); and if the competitors’ willingness to discount price is high (e.g.

companies make frequent special offers around the year). Equivalently, the Intensity of

competition driver is shaped by another four drivers that examine the rivals’ inclination to:

collaborate, exit the industry, defend current revenue from the industry, and maintain current

market-share in the industry. Ultimately, the power of Rivals is more likely to be high if

competition among the current incumbents is price-based, intense, and not transparent.


In comparison to the original DAFF modeling presented in Chapter 4, we, here again, either

eliminate or rearrange relevance among some of the Rivals drivers in order to simplify their

probabilistic evaluation. For instance, the decision-maker was able to evaluate the Inability to

read other incumbents’ signals driver directly, without needing to condition it on the Lack of

familiarity with incumbents; the latter driver uncertainty is therefore excluded from Figure 5.6.

Finally, to evaluate the power of Suppliers, we first limit our definition of solar suppliers to

“component providers” (e.g. solar modules, microinverters, etc.) and “customer-lead

generators.” With that in mind, we condition the force of Suppliers on five direct parent

drivers. Supplier switching cost between incumbents and Incumbent switching cost between

suppliers drivers gauge the relative ability of either an incumbent or a supplier to credibly

threaten the other party with ending their dealings; for instance, if solar firms can easily

replace their microinverter suppliers, the latter are more likely to have low bargaining power.

In that regard, analyzing the Incumbent switching cost between suppliers can be further

informed by examining three additional ancestral drivers: Product differentiation among

suppliers, Availability of substitutes for what the supplier provides, and Incumbent joint

investments with current supplier. The third driver of Suppliers’ power is the Relative

dependence of suppliers on this industry profits, which in turn can be better evaluated by

looking into the Profits extracted by suppliers from other industries as well as the Number of

industries the supplier serve. The last two drivers that help assess Suppliers power are:

Supplier threat to integrate forward and Concentration of suppliers relative to incumbents.

Eventually, Suppliers are more likely to have high bargaining power in the residential solar

industry if they: can easily replace their customer incumbents, are diversified enough and only

generate a minor portion of their revenue or profit from this industry, have enough resources

to credibly threaten entering into this industry, or are highly concentrated.

Compared to the original DAFF modeling of Suppliers in chapter 4, only one driver -

Incumbent’s production location near current suppliers – is excluded from the Bayesian

network in Figure 5.6. This driver is eliminated because it is known with certainty: current

market data confirms the existence of a multitude of solar suppliers all around the globe, many

of which can be easily accessed by the residential solar firms in the U.S. [25, 26].


Broadly, while clarifying and simplifying the competitive analysis, all aforementioned

adjustments in the DAFF modeling of the competitive uncertainties remain consistent with,

and representative of, FF theory. We recall that the original goal of analyzing the competitive

drivers is to better inform the analysis of the competitive forces. Unavoidably, different

drivers may be more or less important in different industries. The competitive drivers in

Figure 5.6 are the ones that SunEnergy deem significant in the residential solar PV industry.

2.1.2. Technological, Regulatory, and Growth Factors

The DAFF modeling of the factor uncertainties is very industry-specific. Residential solar,

like all major energy businesses in the U.S., is heavily shaped by technological advances,

governmental regulations, and growth opportunities. In this case study, we model an extensive

list of Technology, Regulation, and Growth factors that, according to SunEnergy, are likely to

influence the industry’s competitive landscape in the near future. The DAFF modeling of

these factor uncertainties in a Bayesian network is presented in Figure 5.7, and the definitions

of their degrees are listed in Appendix A. As with the competitive force and driver

uncertainties, we aim to quantify the degrees of the factor uncertainties, to the extent possible.

Starting with the Technology factors, we do not envision substantial breakthroughs in solar

technologies within the timeframe of this competitive analysis. However, we recognize the

potential proliferation of two important technologies that may be relevant to residential solar:

electric vehicles [27] and storage batteries [28, 29]. Because electric vehicles are mostly

charged at home [30, 31], their adoption may change the profile of grid-electricity usage in the

household, which we describe in terms of “total demand” and “peak demand”. The rise of

electric vehicles is also relevant to the spread of battery storage systems since, fundamentally,

both consumer products rely on the advancement in battery technologies [32]. While battery

storage systems do not change the household’s total demand for electricity from the grid, they

can help shift this demand from one time period to the other, therefore reducing the home’s

peak demand. Ultimately, the change in the home’s total and peak demand due to electric

vehicles and battery storage might change the size of the home solar system, which is

otherwise heavily dictated by enacted regulations. In that regard, electric vehicles and storage

233

Figure 5.7: DAFF modeling of Technology, Regulation, and Growth factors in the U.S. residential solar PV industry

Substitutes

Buyers

New Entrants

R1: Power market structure

R2: magnitude of utility rates

R3: hourly variation in utility rates

R5: Solar system control

R4: Solar system

connectivity charges

R8: Solar rate

R7: Solar system cap

R9: Solar net-negative

compensation structure

R6: Solar territorial

cap

T1: Proliferation of electric vehicles

T2: Proliferation of storage batteries

T3: Change in home total

demand

R10: Application

of ITC & Depreciation

T4: Change in home peak

demand

T5: Change in size of home solar system

T6: Optimal size of home solar

system

R11: Exploitation

of FMV

G1: Industry growth rate




feasibility


Cost reduction for customer by industry

product

Performance improvement for customer by industry

product

Product perceived cost as fraction of the customer

budget

Relative reliance of

customer on industry product

Customer need to trim immediate cost of industry

product

Commitment of incumbent to retain and


Excess cash

Rivals

Suppliers

Factors


batteries can be viewed not only as relevant technologies but also as major complements to

residential solar, and by analyzing their prospects, we account for two of Porter’s four factors.

We model the impact of both complementary technologies in a series of six Technology

uncertainty nodes, labelled T1 through T6 in Figure 5.7.

In addition, we investigate 11 Regulation uncertainties that cover both the mechanical and the

financial aspects of the residential solar business; those factors are labelled R1 through R11 in

Figure 5.7. First, we consider the overall structure of electricity markets in the U.S. (R1),

focusing on the extent of deregulation that may occur in the next two years [33, 34, 35]. Given

our knowledge about deregulation, we then assess the magnitude (R2) and hourly variations

(R3) of electricity rates that homeowners pay their utilities. By triggering certain behavioral

changes in the way people consume power, those rates might impact the home’s total and/or

peak demand. Other Regulation factors assess whether a utility would charge specific fees to

connect the residential solar systems to the grid (R4), would demand direct and full control of

the residential solar systems under its jurisdiction (R5), or would impose an upper cap either

on the total capacity of solar installations in a specific area (R6) or on the capacity of a single

solar system (R7). We also evaluate the uncertainty around the rate (R8) and payment

schedule (R9) for any excess solar energy that the household produces and sends back to the

grid. As shown in Figure 5.7, the regulated solar rates and capacity caps influence not only the

optimal size of the residential solar system but also how it is likely change in the presence (or

absence) of the aforementioned complementary technologies.

The last two Regulation factors – R10 and R11 – focus on the financing of the residential solar

systems. Currently, solar installations across the country are eligible for a 30% “investment

tax credit” (ITC), granted to the legal owner of the solar system [36]. If the project developer

(e.g. SunEnergy) owns and operates the system for the homeowner, it can realize additional

savings by applying an accelerated depreciation schedule to the system’s net present value

(NPV) [37]. While both forms of subsidies remain active and available till the end of 2016,

R10 captures the uncertainty around how and to what extent they will be utilized. In R11, we

investigate whether the “fair market value” (FMV) rules are exploited by industry incumbents.

A solar company that owns and operates a residential solar system has to report that system’s

FMV in its tax forms [38], which in turn determines the amount of tax subsidies it receives.


Recently, the Treasury Department has detected a controversial practice whereby some solar

firms may be inflating the FMV in order to increase their tax subsidies [39]. The last factor

uncertainty we model is the industry growth rate (G1), which, as depicted in Figure 5.7, might

be relevant to the proliferation of complementary technologies.

Now, after identifying and connecting the various factor uncertainties, it is crucial that we

analyze the relevance relations between these factors and the competitive forces and drivers.

In Figure 5.7, we connect the factors to multiple competitive drivers using relevance arrows.

Those arrows reflect SunEnergy’s beliefs on how the aforementioned Technology, Regulation,

and Growth factors shape competition in the residential solar industry. Obviously, SunEnergy

seems to be more concerned with industry factors that affect the downstream side of business.

By influencing the attractiveness of the solar system relative to its substitutes, the modeled

Regulation and Technology uncertainties may be relevant to two competitive forces:

Substitutes and Buyers. In turn, the industry Growth may be relevant to three forces: Buyers,

New Entrants, and Rivals. While Growth may be dependent on the end-customers’ needs and

budgets for the solar panels, it may change the urgency of market-share fights by, as well as

the availability of investment cash for, both current rivals and future new entrants.

Notably, Figure 5.7 shows no relevance arrows from or into the FMV factor. The reasoning is

that FMV does not impact the competitive landscape, but it directly affects the economics of

residential solar projects and therefore the profitability of the overall industry. In fact, multiple

other factors also have a direct impact on the industry economics. These attributes of the

competitive analysis are expounded in the next section, where we model both the economic

parameters for the residential solar industry as well as the relevance relations that these

parameters exhibit with the competitive forces, drivers, and factors.

2.1.3. Economic Parameters

The economic parameters are the last necessary component for developing a complete DAFF

Bayesian Network and therefore for assessing competition in the overall industry. As

explained in Chapter 4, these economic parameters account for the industry’s average cost,


price, quantity, and ultimately, profitability. The DAFF modeling of the economic parameters

is presented in Figure 5.8.

To start, SunEnergy chooses the “cost of goods sold” (COGS) as the metric to measure the

average cost in the U.S. residential solar PV industry. By SunEnergy’s accounting, COGS

cover both the fixed operating expenses (FOPEX) and variable operating expenses (VOPEX)

of the business, spanning the following major categories: cost of the solar kit (e.g. solar

modules and balance of system); installation fees (e.g. labor and permitting); sales and

marketing (e.g. customer acquisition cost); and general, admin and R&D. Nonetheless,

because many of the U.S. solar firms operate either internationally [40, 41] or in multiple solar

industries (residential, C&I, and utility) [9], it is hard to distinguish the capital investments

associated with their residential solar business in the U.S. only; for instance, investments in

panel manufacturing facilities or administrative offices are shared across multiple corporate

business units. Consequently, to simplify the analysis, the capital costs are not accounted for

in this case study.

Based on public market information [42] and private corporate data, the COGS of residential

solar in the U.S. is modeled in Figure 5.8 as a Real Cost uncertainty with three possible

degrees: {2 $/W}, {3.5 $/W}, and {5 $/W}. Here, we make four important notes. Firstly, this

economic parameter captures the real cost of installing and operating one unit capacity of the

solar system, before accounting for any governmental subsidies. Secondly, we assert that the

selected numerical range – between 2 and 5 $/W – is informed by, and therefore is comparable

to, the cost figures observed in the industry over the past 3–5 years. From Chapter 4, we recall

that the industry economics should be evaluated over a full business cycle; in this case, we

estimate that a full business cycle in the U.S. residential solar PV industry is around 3–5 years.

Thirdly, the three numerical degrees denote an approximate discretized probability distribution

over a linear cost range between 2 and 5 $/W. For instance, if the decision-maker thinks that

the competitive circumstances render the cost around 4 $/W, he may assign a 0, 0.65, and 0.35

probability to the {2 $/W}, {3.5 $/W} and {5 $/W} degrees, respectively. This discretization

of an otherwise continuous random variable is robustly applied in decision analysis to

facilitate uncertainty assessment [43], and it proves beneficial here for computing the industry

237

Figure 5.8: DAFF modeling of the economic parameters in the U.S. residential solar PV industry

Substitutes

Buyers

New Entrants

Rivals

Suppliers

Factors

Real CostEffective

CostEffective

Price

Threat of new

entrants



Bargaining Power of

BuyersRivalry

System UnitsTotal

Quantity

G1: Industry growth rate

T5: Change in size of

home solar system

T6: Optimal size of home solar system

R10: Application

of ITC & Depreciation

R11: Exploitation

of FMV

R4: Solar system

connectivity charges

EBT Economic Parameters


economics. Lastly, consistent with our DAFF modeling of Porter’s work, cost is influenced by

all five competitive forces, so a relevance arrow is added from each force uncertainty into the

Real Cost uncertainty.

Subsequently, for a given realization of Real Cost, we can model the effective cost that a solar

developer like SunEnergy actually incurs by factoring in any governmental subsidies (ITC and

depreciation), exploitations of FMV, and/or utility-imposed connectivity charges to the grid.

To that end, an Effective Cost uncertainty is modelled in Figure 5.8 with four parents: Real

Cost, R4: Solar system connectivity charges, R10: Application of ITC and Depreciation, and

R11: Exploitation of FMV. In addition, we model four numerical degrees for Effective Cost:

{0 $/W}, {2 $/W}, {4 $/W}, and {$6/W}. As before, those degrees denote a discretized

probability distribution over an effective-cost range between 0 and 6 $/W, which is

comparable with market data over the past 3–5 years.

In modeling the Effective Cost, we make a simplifying yet important assumption that

applicable governmental subsidies are always captured by the industry incumbent, even if the

end-customer is the legal owner of the solar system. One way to vindicate this assumption is

to think of a situation where the end-customer transacts subsidies freely and instantaneously

with the solar firm in exchange for a lower price. For example, if the homeowner buys a solar

system for $X and receives a subsidy of $Y, his effective price will be $(X–Y). Alternatively,

he can agree to transfer that $Y subsidy to the firm, and the firm will in return agree to reduce

the system price to $(X–Y). Once again, the homeowner’s effective price becomes $(X–Y).

Evidently, this modeling approach raises the notion of Effective Price, which we model as an

uncertainty in Figure 5.8. Formally, we define Effective Price as “the end-customer’s net

payment, and the industry incumbent’s net revenue, over the entire lifetime of the solar system

and after clearing all transactions of applicable governmental subsidies.”

Consistent with our DAFF discussions in Chapter 4, the Effective Price is conditioned on four

of the five competitive forces: Substitutes, Buyers, New Entrants, and Rivals. Furthermore,

since both costs and prices are ultimately decided by the firm, the Effective Price is also

conditioned on the Effective Cost. Three primary arguments justify this relevance. First, it is

reasonable to assume that the solar developer shall never want to set the Effective Price below


the Effective Cost, in which case it would suffer negative returns. Second, firms with

drastically different costs are unlikely to set the same price for their product; intuitively, a

low-cost competitor may be inclined to adopt a low-price strategy in order to attract more

customers. Third, this relevance relation captures the aforementioned effect of governmental

policies and subsidies on price through cost; as shown in Figure 5.8, R4, R10, and R11 are

grandparent nodes to Effective Price. Eventually, public [25, 42, 44, 45] and private data over

one business cycle suggests that the Effective Price may range from 1 to 7 $/W, which we

model as an approximate discretized probability distribution with four degrees: {1 $/W}, {3

$/W}, {5 $/W}, and {7 $/W}.

Beyond cost and price, we investigate potential changes in the sales volume of residential

solar systems over the next two years. Consistent with our explanation in Chapter 4, the

number of sold systems is more likely to be high when the threat of substitutes is low, the

industry growth is fast, the system price is low, and the margins are tight (price is set close to

the cost). As such, the unit sales are modeled as a System Units uncertainty in Figure 5.8,

conditioned on four parents: Effective Cost, Effective Price, Substitutes, and G1: Industry

growth rate. Public [46] and corporate projections estimate the annual installations to range

between 90000 and 540000 units. To capture this range, we define four System Units degrees:

{0}, {90000}, {300000}, and {540000}. Notably, the {0} degree accounts exclusively for all

prospects where the realized cost exceeds the realized price; in this case, we know with

certainty that solar developers will not sell their rooftop solar panels for negative returns.

The economic modeling of the average annual unit sales, along with the previous

technological modeling of the average unit size, informs our analysis of the average annual

capacity sales throughout the timeframe of this competitive analysis. As a result, Figure 5.8

shows a Total Quantity uncertainty with three parents: System Units, T5: Change in size of

home solar system, and T6: Optimal size of home solar system. Upon estimating the average

peak demand of various types of households in various geographic locations, SunEnergy

believes that the added annual capacity of residential solar in the U.S. over the next two years

might be anywhere between 0.1 and 5 gigawatts (GW). Therefore, we model the Total

Quantity uncertainty with six degrees: {0 GW}, {0.1 GW}, {1 GW}, {2 GW}, {3 GW}, {4

GW}, an {5 GW}. Here, again, the {0 GW} degree is needed to properly handle the lack of


solar installations due to unfavorable economics; whenever System Units is {0}, the Total

Quantity must be {0 GW}.

To conclude the DAFF modeling of the economic parameters, we use “earnings before tax”

(EBT) as the proper value metric to measure the overall industry profitability. Figure 5.8

shows the EBT value metric with three functional arrows extending into it from the Effective

Cost, Effective Price, and Total Quantity uncertainties. As formulated in (1), EBT calculates

the average annual net income of all incumbents in the industry, after accounting for the

governmental subsidies but before accounting for federal and state income taxes [47]. In other

words, the output EBT represents the expected average of annual profits from the residential

solar PV industry in the next two years. The emphasis on using an absolute-profit metric

instead of a relative-profitability ratio stems from SunEnergy’s interest in assessing not only

the relative favorability but also the absolute scale of the industry’s economics; a small

market is less likely to be a priority business opportunity for SunEnergy, even if the

profitability ratio is high.

𝐸𝐵𝑇 ($ 𝑦𝑒𝑎𝑟⁄ ) = 𝑇𝑜𝑡𝑎𝑙 𝑄𝑢𝑎𝑛𝑡𝑖𝑡𝑦 ∙ (𝐸𝑓𝑓𝑒𝑐𝑡𝑖𝑣𝑒 𝑃𝑟𝑖𝑐𝑒 − 𝐸𝑓𝑓𝑒𝑐𝑡𝑖𝑣𝑒 𝐶𝑜𝑠𝑡) (1)

Similar to those of forces, drivers, and factors, the assigned probabilities over the degrees of

all economic uncertainties are reflective of the SunEnergy’s beliefs regarding the industry’s

near future. Ultimately, combining the DAFF elements in Figures 5.6, 5.7, and 5.8 results in a

complete DAFF Bayesian Network that assesses the profitability of the overall residential

solar PV industry in the U.S. A simplified sketch of the complete DAFF Bayesian Network is

shown in Figure 5.9. Here, for clarity, the representation of competitive drivers and factors is

condensed in the form of uncertainty “clouds”, each displaying the total number of uncertainty

nodes it incorporates and accounts for.

Now, after modeling the competitive forces and their economic implications for the whole

industry, SunEnergy shall make a series of strategic choices that position the company in the

most profitable segment of the industry. To achieve this goal, we transition to the second step

of this competitive analysis where we model several positioning decisions.


Figure 5.9: A sketch of the complete DAFF Bayesian Network for SunEnergy

2.2 Second Step: Assess Each Positioning Segment in the Industry

An effective strategy for SunEnergy requires a clear mapping between its feasible positioning

alternatives and the industry’s competitive forces. As explained in Chapter 4, strategic

positioning encompasses two types of decisions that can be undertaken by firms: Value

Proposition and Value Chain. While Value Proposition decisions define what product to

make, Value Chain decisions define how the product is made. In this case study, we model

four positioning decisions that SunEnergy deems important for clear market segmentation: one

Value Proposition decision and three Value Chain decisions. By modeling these decisions and

their influence on the aforementioned competitive forces, drivers, factors, and economic

parameters, we complete the construction of a DAFF Decision Diagram for the residential

solar PV industry in the U.S.

To start, the Value Proposition decision addresses the firm’s ability – rather need – to choose

the regions where it sells and services solar systems. One key distinction to make here is

between rural and urban regions. Compared to rural areas, urban centers are characterized by

Regulation (11)

Drivers ofSubstitutes (9)

Drivers of Buyers (20)

Drivers of New Entrants (30)

Drivers of Rivals (16)

Drivers of Suppliers (11)

Substitutes

Growth (1) Technology (6)

BuyersNew


Real Cost

Effective Cost

Effective Price

System Units

Total Quantity


higher population but smaller houses [48, 49], which translates into more but smaller solar

systems. The solar preferences of urban and rural customers may also differ for a variety of

reasons, including: weather, accessibility to the electric grid, and the nature of daily activities

within the household. Consequently, our DAFF modeling incorporates Regional Focus as a

Value Proposition decision with two alternatives: [Urban, Rural]. This decision influences the

probability distribution over multiple driver, factor, and economic-parameter uncertainties,

which we list in Table 5.1.

Table 5.1: DAFF uncertainties influenced by Regional Focus

Classification Uncertainty

Substitutes Customer switching cost from this industry product to substitutes

Regulatory and technical feasibility

Buyers

Customer switching cost among industry products

Dealer brand

Dealer reach

Product differentiation among incumbents

Relative reliance of customer on industry product

New Entrants


Incumbent established brands

Incumbent network effects

Limitation of distribution channels

Previous responses by incumbents

Rivals Ability to meet the needs of multiple customer segments

Suppliers

Availability of substitutes for what suppliers provide

Concentration of suppliers relative to incumbents

Supplier switching cost among incumbents

Factors G1: Industry growth rate

Economic

parameters Total Quantity

After choosing what region to serve, SunEnergy cares to examine where to operate along the

industry’s value chain. To address this point, we assess the feasible extent of vertical

integration, especially downstream. The solar firm may choose to manage its sales either

directly with the end-customer or through an intermediary dealer. In that regard, the dealer’s

responsibility may span a wide range of marketing, engineering, and financial services,


including (but not limited to): lead generation or closing, system installation or maintenance,

and financial securitization [47]. In the absence of such intermediary, the solar firm maintains

a direct and exclusive relationship with its residential solar customers on all fronts.

Consequently, we add Downstream Integration as a Value Chain decision in our DAFF

model, with two alternatives: [Dealer, No Dealer]. Inevitably, this decision influences multiple

competitive drivers that shape not only the bargaining power of Buyers but also the power of

both incumbent Rivals and future New Entrants. The probabilistic influence of Downstream

Integration on the competitive landscape is documented in Table 5.2. In general, while the

reliance on dealers may crowd the industry with an additional group of influential players and

therefore escalate competition, it may also allow more flexibility in the solar firm’s business

model; such flexibility results in a wider reach to underserved customers and therefore yields a

higher industry growth rate.

Table 5.2: DAFF uncertainties influenced by Downstream Integration


Buyers

Dealer partnership with incumbent

Dealer threat to integrate backward

Dealer volume of sales per incumbent

New Entrants

Incumbent assets


Incumbent economies of scale


Incumbent inventory

Size-dependent disadvantages for new entrants

Rivals

Ability to meet the needs of multiple customer segments

Extent of exit barriers

Fragmentation of the industry

High fixed costs and low variable costs

Factors G1: Industry growth rate

R11: Exploitation of FMV

Another significant aspect of the solar firm’s business model is the type of financing services

available to its customers. If the homeowner wants to own his rooftop solar panels, he may

purchase the system either directly (i.e. full-upfront payment) or through a loan.

Alternatively, the homeowner may lease the solar system for either a full-upfront payment or


monthly payments (with no upfront fee). Leasing maintains the firm as the legal owner of the

solar system while providing the end-customer with the solar service only. Depending on its

financial structure and capabilities, a solar firm may choose to provide any of these financing

services, either directly or through a financing partner (i.e. a dealer). To capture the

competitive tradeoffs associated with these various options, we model Customer Financing as

a Value Chain decision with four alternatives: [Direct Purchase, Loan, Full-upfront Lease, No-

upfront Lease]. The DAFF uncertainties influenced by this decision are presented in Table 5.3.

Table 5.3: DAFF uncertainties influenced by Customer Financing


Substitutes Customer switching cost from this industry product to substitutes

Buyers

Amount of cash available for customer

Customer switching cost among industry products

Cost reduction for customer by industry product

Dealer ability to influence customers downstream

Dealer partnership with incumbent

Dealer threat to integrate backwards

Product differentiation among incumbents

Product perceived cost as fraction of the customer budget

New Entrants

Buyer trust in incumbent

Efficiency of capital markets


Rivals Ability to meet the needs of multiple customer segments

Factors R10: ITC and depreciation

R11: Exploitation of FMV

Importantly, the combination of both Customer Financing and Downstream Integration

heavily influences the exploitation of fair market value. If the solar firm sells its systems either

through direct purchase or a loan, exploiting the FMV is infeasible, regardless of the firm’s

extent of vertical integration; in these two cases, the system price on the company’s income

statement has to match the one on the customer’s receipt. However, under a lease agreement,

exploiting the FMV becomes a feasible (albeit controversial) possibility if the solar firm is

vertically integrated. In this case, the firm acts as a solar developer, installer, and financer, and

it can buy/sell the system from/to itself; in other words, it can transact the system internally.


Lastly, focusing on the opposite side of the Value Chain, industry incumbents have the ability

to choose how to source the components of their solar system. One decision of particular

interest to SunEnergy is whether to manufacture the solar-panel components (e.g. cells,

modules, or inverters) in-house or purchase them from a third-party supplier. Solar modules

still contribute a significant portion of the overall system cost [42], so carefully managing their

supply is necessary for an effective competitive strategy. Consequently, we add Panel

Manufacturing as a Value Chain decision in our DAFF model, and we define its two

alternatives as: [Insource, Outsource]. Table 5.4 lists the Supplier and other competitive

uncertainties that are probabilistically influenced by Panel Manufacturing.

Table 5.4: DAFF uncertainties influenced by Panel Manufacturing


Buyers Product differentiation among incumbents

New Entrants

Incumbent assets

Incumbent IP

Incumbent R&D spending


Rivals Extent of exit barriers

High fixed costs and low variable costs

Suppliers

Availability of substitutes for what suppliers provide


Incumbent joint investments with current suppliers

Supplier threat to integrate forward

Figure 5.10: Example of decision influence on conditional probability assignment

In terms of probability assignment, we note that conditioning the probability distribution on a

decision alternative is similar to conditioning it on an uncertainty degree. To illustrate this


concept, Figure 5.10 shows the two degrees for the Incumbent economies of scale driver,

conditioned on the two alternatives of Downstream Integration. For every alternative, the two

degrees are assigned distinct probability values. Here again, the numerical values of

probability modifications by positioning decisions are reflective of SunEnergy’s beliefs and

market intelligence.

Figure 5.11: A sketch of the complete DAFF Decision Diagram for SunEnergy

Adding the aforementioned four decisions to our Bayesian Network yields a complete

Decision Diagram, a simplified version of which is depicted in Figure 5.11. For clarity, only

influence arrows extending from the decisions to the uncertainties are displayed in Figure 5.11

– relevance and information arrows are not. Every combination of the Value Proposition and

Value Chain alternatives results in a unique positioning track, which in turn defines a unique

segment of residential solar where SunEnergy can locate and operate. In that regard, we recall

from Chapter 4 that, by design, the alternatives of a particular decision are mutually exclusive;

Regulation (11)

Drivers ofSubstitutes (9)

Drivers of Buyers (20)

Drivers of New Entrants (30)

Drivers of Rivals (16)

Drivers of Suppliers (11)

Substitutes

Growth (1) Technology (6)

BuyersNew


Real Cost

Effective Cost

Effective Price

System Units

Total Quantity

Regional FocusDownstream Integration

Customer Financing

Panel Manufacturing

Value Proposition Value Chain Value Chain Value Chain

CHAPTER 5 — Results: Outputs from the DAFF Models 247

in other words, a solar firm chooses one and only one alternative for each decision. Because

the positioning alternatives are mutually exclusive, so are their combinations in the form of

positioning tracks. Furthermore, although positioning decisions affect the probability

distribution of only specific uncertainty nodes, their influence propagates through the decision

diagram until ultimately impacting the industry’s profitability. As a result, the DAFF Decision

Diagram model yields a unique competitive landscape, and therefore a unique expected EBT

value, for every positioning track.

3 Results: Outputs from the DAFF Models

After completing the DAFF modeling of the five forces, their economic implications, and the

multiple positioning alternatives available for SunEnergy in the residential solar PV industry,

we now transition to presenting the results of this competitive analysis. We follow the same

order in the previous section, first reflecting on the performance of the overall industry, and

then evaluating the attractiveness of the various positioning segments. The outputs from the

DAFF models highlight several insights that help inform SunEnergy’s competitive strategy.

3.1 First Step: Assess the Overall Industry

3.1.1. Competitive Landscape

The DAFF Bayesian Network allows quantifying, visualizing, and reflecting on the powers of

the five competitive forces, which account for both the intelligence in and interaction among

the extensive set of analyzed drivers and factors. As direct outputs of this DAFF model, the

probability distributions for the power of each competitive force and its parent drivers are

plotted in Figure 5.12. The results show that, according to SunEnergy’s beliefs and

knowledge, the residential solar PV industry is likely to witness limited bargaining powers for

both buyers and suppliers, as well as limited competitive threats by both substitutes and new

entrants, through the end of 2016. Nonetheless, the industry is likely to experience stronger

rivalry among current incumbents, especially as they try to expand their business with the

growing market.


Figure 5.12: Competitive landscape in the U.S. residential solar PV industry through 2016

As manifested in Figure 5.12, SunEnergy believes that the threat of Substitutes through 2016

will be {high} with probability of 0.39 and {low} with probability of 0.61. This limited threat

emerges from balancing the effects of three market drivers. Although it is not clear whether

solar substitutes will enjoy distinctively more favorable regulatory or technological

ecosystems (probability of {more favorable} is 0.52), it is very unlikely that they will offer a

superior price-performance tradeoff to solar panels (probability of {superior} is 0.28).

Additionally, it is unlikely that customers will have the luxury to feely give up or replace their

panels with another home-energy-saving gadget or service, due to strict contract agreements

and/or high penalties imposed by the solar firms (probability of {no legal complications or

high penalty} is 0.35).

1

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Supplier threat to integrate forward


Relative dependence of suppliers on this industry profits

Incumbent switching costs between suppliers

Supplier switching cost between incumbents


Inability to read other incumbents’ signals



Rivalry


Barriers to entry

Threat of New Entrants

Power of dealers

Power of end-customers


Customer switching cost from this industry product to substitutes

Regulatory and technical feasibility

Price-performance tradeoff relative to this industry product

Threat of Substitutes

Probability

high

high

high

high

high

low

low

low

low

low

high low

high low

superior equivalent inferior

more favorable

legal complications or high penalty(opposite)

less favorable

highlow

highlow

price non-price

high low

able to read other incumbents' signals(opposite)

<10% increase in production cost >10% ...

<10% increase in production cost>10% ...

solar is top profit source(opposite)

>10 suppliers per incumbent<10 ...

<1 supplier forward-integration per year>1 ...


Similarly, SunEnergy believes that the power of Buyers through 2016 will be {high} with

probability of 0.38 and {low} with probability of 0.62. Two drivers contribute to this result.

First, end-customers will likely maintain limited bargaining powers (probability of {low}

Power of end-customers is 0.56) due to a host of reasons, including: the customers’ original

perception of achieving attractive utility-bill savings (>15%) and of affording the solar

purchase through loans and leases, and the customers’ subsequent constraints to switch away

from the solar system due to the aforementioned strict contracts or high penalties. Second,

dealers will likely have even weaker bargaining powers (probability of {low} Power of

dealers is 0.68), for two main reasons. The majority of dealers tend to be small and local

vendors – unable to influence a wide range of customers or credibly threat to take over the

incumbents’ business. At the same time, solar incumbents seem to be increasingly interested

in either securing exclusive partnerships with dealers or acquiring the dealers’ business and

doing it themselves.

The threat of New Entrants is projected to be {high} with probability of 0.41 and {low} with

probability of 0.59. This relatively limited threat is a result of slightly challenging barriers to

entry (probability of {high} Barriers to entry is 0.54) but relatively fierce retaliation expected

from incumbents (probability of {high} Expected retaliation is 0.63). The analysis shows that,

because the solar energy market is still relatively recent and growing, it is less likely that the

incumbent’ economies of scale or size of capital will play a significant role in deterring entry.

However, the proactive attempts by the incumbents to control distribution channels (e.g. build

partnerships with dealers) and exploit networking effects (e.g. win the customers’ trust and

advocacy) will likely compel new entrants to find novel ways for promoting and delivering

their products. Equivalently, SunEnergy believes that the retaliatory behavior of incumbents

will likely be driven by the inclination to either acquire new entrants or engage them in

discount wars.

Unlike the other four competitive forces, Rivalry over the next two years is expected to be

{high} with probability of 0.62 and {low} with probability of 0.38. This result accounts for

three main considerations. To start, the industry’s visibility will likely help incumbents read

their rivals’ signals and guess their moves (probability of {able to read other incumbents’

signals} is 0.75). Although this visibility may incentivize collaboration, its benefits are


drastically diminished by the other two considerations. First, competition will mostly likely be

price-based, due to both the absence of significantly differentiated products by solar firms

(e.g. efficiency ranges, colors, and dimensions of solar panels) and – congruently – the

absence of significantly differentiated needs by solar customers (e.g. purpose and time of use,

maintenance and control services, and physical appearance of the solar system); in other

words, the homogeneity of both the solar products and the solar needs leave incumbents with

price as the main dimension to compete on (probability of {price} Basis of competition is

0.65). In addition, as incumbents focus on growing and gaining market-share, without secure

and diversified revenue streams from other non-solar industries, the intensity of their

competition will likely escalate (probability of {high} Intensity of competition is 0.63).

Finally, the competitive analysis reveals that the bargaining power of Suppliers will be {high}

with probability of 0.39 and {low} with probability of 0.61. This outcome balances the effect

of five market drivers. On one hand, because the residential solar systems and services are

pretty standardized, the incremental costs that suppliers may face upon switching between

incumbents are likely to be minor (probability of {<10% increase in production cost} is 0.7);

this increases the suppliers’ leverage. On the other hand, four drivers seem to disfavor

suppliers. To start, because of standardization, the incumbents’ switching costs between

suppliers are equally likely to be minor (probability of {<10% increase in production cost} is

0.61). This reality is further amplified by the low concentration of suppliers relative to

incumbents (probability of {>10 suppliers per incumbent} is 0.71). Add to that, the majority

of suppliers will probably continue to rely on the solar industry as their major source of profit

(probability of {solar is top profit source} is 0.8), and they are unlikely to integrate forward

and start competing directly with incumbents (probability of {<1 supplier forward-integration

per year} is 0.75).

Given this overview of the competitive landscape in the U.S. residential solar PV industry

over the next two years, we pause to make an important statement. After going through this

modeling exercise, we assert that the aforementioned insights and conclusions about

competition could have been obtained by conducting a detailed qualitative analysis of the

five forces instead of building a DAFF model. However, if such an assertion is true, then one

must wonder: what are the additional benefits from using the newly proposed DAFF approach


instead of the conventional qualitative-analysis approach? Simply put, while a qualitative

analysis of the five forces can generate robust insights regarding the overarching competitive

trends in the industry, its benefits end there. Conversely, the benefits of DAFF extend beyond

highlighting the competitive trends, to include: establishing a quantitative relation between

competition and the economic performance in the industry; identifying and quantifying

competitive interdependence and its economic implications; and quantifying, comparing, and

ranking various positioning strategies for the firm in the industry. We proceed to demonstrate

each of these DAFF advantages in the following sections.

3.1.2. Economic Performance

Before delving into the economics, one consideration to keep in mind is the subjective

significance of the results in Figure 5.12. As we note in Chapter 4, DAFF standardizes the

characterization of the five force uncertainties, via the two degrees {high} and {low}, in order

to simplify and generalize their assessment across industries and decision-makers. The

tradeoff associated with this modeling approach is that the significance of the forces’ powers

becomes subjective; the probability distribution over {high} and {low} is interpreted

differently by different decision-makers. To give a hypothetical example in the context of this

case study, SunEnergy’s Director of Strategic Planning may interpret a 0.6 probability for

{high} Buyers power as a “mildly challenging” competitive prospect, but the Director of

Business Development of a rival solar firm may interpret the same information as a “severely

challenging” competitive prospect. To resolve this ambiguity, we need to translate the

subjectively interpretable competitive forces into objectively explicable and quantifiable

metrics, whose numerical values are perceived consistently by all players. Surely enough, this

goal is fulfilled in the DAFF model through the economic parameters and their precise

relations with the competitive forces.

We summarize the expected economic performance of the U.S. residential solar industry over

the next two years in Figures 5.13a–c. These figures plot discretized probability distributions

for the Effective Cost, Effective Price, and Total Quantity of installed solar systems, after

accounting for all relevant driver and factor uncertainties. In Figure 5.13a, SunEnergy’s

intelligence about competition and governmental regulations allows deducing that the


Effective Cost will very likely be lower than 4 $/W, with an expected (i.e. probability-

weighted average) value of 1.19 $/W through 2016. Equivalently, Figure 5.13b shows that the

Effective Price of the solar system will likely be in the 3–5 $/W range, with an expected value

of 3.35 $/W. These cost and price prospects, as well as SunEnergy’s intelligence on the power

of Substitutes, the industry Growth rate, and the average size of the solar system, yield

attractive projections for annual sales. In Figure 5.13c, the DAFF output shows a high level of

confidence that the Total Quantity – representing the yearly installation capacity – will be

between 1 and 3 GW, with an expected value of 1.93 GW. With this information, we can now

complete the first objective of this competitive analysis by calculating the expected profit for

the whole industry: given the probabilistic outcomes for all three economic parameters, the

DAFF model computes the expected EBT of the U.S. residential solar PV industry

through 2016 at 4.05 billion $/year.

Figure 5.13: Economic performance of the U.S. residential solar PV industry through 2016

3.1.3. Competitive Interdependence

Beyond quantifying the impact of competition on the industry economics, the DAFF Bayesian

Network provides crucial insights regarding the strategic interdependence and interaction

among the competitive forces. On way to comprehend this DAFF characteristic is to think of

the Bayesian network as a living model that can, and should, be continuously updated; the

decision-maker can always modify the probability assignments for uncertainties in order to

capture newly attained market information. Because of probabilistic relevance, observing new

information about a specific uncertainty may allow inferring new information about relevant

0

0.1

0.2

0.3

0.4

0.5

0.6

0 2 4 6

Pro

bab

ility

Effective Cost ($/W)

0

0.1

0.2

0.3

0.4

0.5

0 0.1 1 2 3 4 5

Pro

bab

ility

Total Quantity (GW/year)

0

0.1

0.2

0.3

0.4

0.5

0.6

1 3 5 7

Pro

bab

ility

Effective Price ($/W)

(a) (b) (c)


uncertainties [43]. Both observations and inferences result in updated probability distributions

over relevant competitive uncertainties.

To demonstrate both notions of information-update in and interdependence among the

competitive forces, we describe a thought experiment whereby we examine the sensitivity of

the obtained results to extreme competitive scenarios. In a nutshell, the experiment goes as

follows: if the decision-maker observes, with certainty, that the power of a force is at its

highest or lowest extreme, how does this new information about the power of one force update

his beliefs about the power of other forces? Put differently, if the decision-maker gets perfect

clairvoyance on one force, what would he infer about the other forces?

To conduct this experiment, we test two scenarios for each force: all the driver uncertainties of

a given force are resolved to render that force’s power either strongest (probability of {high} =

1) or weakest (probability of {low} = 1). Then, for each scenario, we examine how

maximizing (or minimizing) the power of the observed force updates the power of the other

four forces. The results are summarized in Figures 5.14a–e.

In the tornado plot of Figure 5.14a, we analyze how our beliefs about the power of Substitutes

changes upon observing each of the other four forces (and their drivers) at their extreme

values. The vertical axis shows the competitive forces we observe: Buyers, New Entrants,

Rivals, or Suppliers. The power of each force is observed as either {high} or {low}. The

horizontal axis then indicates how the probability of {high power of Substitutes} changes

upon observing each of the other four forces. Given the decision-maker’s base-case

information about competition, the probability of {high power of Substitutes} is 0.39, as

documented in Figure 5.12. Now, observing {high power of Buyers} updates our beliefs about

Substitutes, such that the likelihood of {high power of Substitutes} increases from 0.39 to 0.7.

Conversely, if the decision-maker observes {low power of Buyers}, the probability of {high

power of Substitutes} decreases from 0.39 to 0.25. In plain English, observing strong Buyers

increases the likelihood of witnessing strong Substitutes, and observing weak Buyers

decreases the likelihood of witnessing strong Substitutes. Overall, these results signify a

strong dependence of the threat of Substitutes on the bargaining power of Buyers.


Figure 5.14: Interdependence between the competitive forces in the U.S. residential solar

PV industry

0.2 0.3 0.4 0.5 0.6 0.7 0.8Probability of {high power of Substitutes}

Substitutes

Observed as "Low" Observed as "High"

0.2 0.3 0.4 0.5 0.6 0.7 0.8

Probability of {high power of Buyers}

Buyers


0.2 0.3 0.4 0.5 0.6 0.7 0.8

Probability of {high power of New Entrants}

New Entrants


0.2 0.3 0.4 0.5 0.6 0.7 0.8

Probability of {high power of Rivals}

Rivals


Buyers

New Entrants

Rivals

Suppliers

Substitutes

New Entrants

Rivals

Suppliers

Substitutes

Buyers

Rivals

Suppliers

Substitutes

Buyers

New Entrants

Suppliers

0.2 0.3 0.4 0.5 0.6 0.7 0.8

Probability of {high power of Suppliers}

Suppliers


Substitutes

Buyers

New Entrants

Rivals

(a) (b)

(c)

(e)

(d)

Observed as {low} Observed as {high} Observed as {low} Observed as {high}

Observed as {low} Observed as {high} Observed as {low} Observed as {high}

Observed as {low} Observed as {high}


To understand the reasons behind this interdependence, we refer back to the Bayesian network

in Figures 5.6 and 5.7. Evidently, multiple Substitutes drivers have three Buyers drivers as

direct parents: Cost reduction for customer by industry product, Product perceived cost as

fraction of the customer budget, and Customer switching cost from this industry product to

substitutes. Observing the Buyers drivers changes the probability of their degrees to either 0 or

1. Subsequently, the updated probability distributions for the parent Buyers drivers yield

updated probability distributions for the children Substitutes drivers. A second form of

probabilistic relevance also contributes to the interaction between the two forces: several

Buyers and Substitutes drivers have common ancestors, especially the Regulation

uncertainties. Observing the children Buyers drivers allows the decision-maker to infer new

probability distributions for the parent Regulation factors, which then yields new probability

distributions for the children Substitutes drivers. For example, if the Director of Strategic

Planning at SunEnergy gets to know, undoubtedly, that the residential solar system will

achieve {< 15% reduction in power bill} and will {not improve power purchase and

utilization} in the next two years, he may infer that the optimal size of the home solar system

will more likely be {less than 50% of home peak demand}, which in turn allows deducing

that substitutes will more likely achieve {higher emissions reduction}. As discussed in

Chapter 4, Bayes-ball is a good technique to track these interdependence relations [50].

Continuing with Figure 5.14a, we notice that the power of Substitutes is also dependent on the

threat of New Entrants; the probability of {high power of Substitutes} increases to 0.49 upon

observing {low power of New Entrants} and decreases to 0.33 upon observing {high power of

New Entrants}. Interestingly, in this case, the forces move in opposite directions: observing

weak New Entrants increases the likelihood of witnessing strong Substitutes; the opposite

is also true. This relation is rather intuitive: if an industry has a large number of substitutes,

those substitutes may credibly discourage new players from entering the market. In our case

study, this interaction can be tracked in the DAFF network of Figure 5.6. Observing the power

of New Entrants entails observing the Relative reliance of customer on industry product

driver, which in turn allows inferring new information about its parent Cost reduction for

customer by industry product driver. Changing the probability distribution of the latter driver

leads to updating the probability distribution of its child Substitute bill savings, which is a


Substitutes driver. In simpler words, the story goes as follows. Firms prefer to enter markets

where customers rapidly adopt the product because it is of real value to them. Accordingly,

one indication of low threat of entry is the observation that residential customers are not

heavily reliant on the solar system. Such observation allows inferring that solar panels enable

no significant reductions in the customers’ utility bills. In turn, the poor bill savings from solar

systems raise the relative value of savings from solar substitutes, which eventually contributes

to increasing the overall threat of substitutes.

Another important insight from Figure 5.14a is that, for a particular force, the

interdependence relations with other forces need not be in the same direction or of the

same magnitude. In terms of direction, we just explained that the likelihood of witnessing

strong Substitutes in the industry increases upon observing strong Buyers but decreases upon

observing strong New Entrants. Equally noteworthy, the magnitude of change in the

probability of {high power of Substitutes} is larger upon observing {high power of Buyers}

than it is upon observing {low power of Buyers}. This result is primarily a modeling feature

rather than a strategic intuition. Observing {high power of Buyers} changes the probability of

{high power of Substitutes} by 0.62 (from 0.38 to 1), whereas observing {low power of

Buyers} changes the probability by 0.38 (from 0.38 to 0). As new information updates the

probabilities throughout the Bayesian network, larger changes in the probability of the

observed force (Buyers in our example) cause larger changes in the probability of the tested

force (Substitutes in our example). In fact, because the Bayesian network has to always

balance the numerous probabilistic relations among various uncertainties, observing one force

at either extreme is unlikely to cause another force to reach either extreme. In other words, if

the probability of a {high} force at base-case is less than 0.5, it becomes progressively harder

for new competitive observations to further weaken that force; similarly, if the probability of a

{high} force at base-case is greater than 0.5, it becomes progressively harder for new

competitive observations to further strengthen that force.

The same logic applies to all the remaining interdependence relations depicted in Figure

5.14b–e; we proceed to discuss the most prominent of these relations. Moving on to Figure

5.14b, the tornado diagram shows the sensitivity of the power of Buyers to observing the four

other forces: Substitutes, New Entrants, Rivals, and Suppliers. Starting with the dependence of


Buyers on Substitutes, we see that, observing strong Substitutes increases the likelihood of

witnessing strong Buyers; the opposite is also true. Compared to the results in Figure 5.14a,

we see that the forces still move in the same direction, but the dependence of Buyers on

Substitutes is less significant than the dependence of Substitutes on Buyers. This outcome

highlights an important takeaway: the interdependence between two competitive forces

need not be symmetrical.

As explained before, the interdependence between the forces is governed by the probabilistic

relevance between the uncertainties, which in turn is dictated by the assignment of conditional

probabilities. The conditional probabilities between two uncertainties need not be the same in

both the forward and reverse directions of assessment. A simple example to clarify this point

is the relevance between “lung disease” and “smoking”; here, we can make two assessments:

the probability that a person {has a lung disease given that he smokes}, and the probability

that a person {smokes given that he has a lung disease}. Obviously, these two conditional

probabilities having different meanings and may have different values. The asymmetry in the

interdependence between Buyers and Substitutes can be explained similarly. In our DAFF

logic, the customer advantages (e.g. cost reductions) from the solar system are modelled as

drivers for Buyers. Observing Buyers means observing the exact customer advantages from

the solar system, which then may update the decision-maker’s information about the relative

appeal of potential substitutes. Equivalently, observing Substitutes means observing the exact

customer advantages from solar substitutes, which then may update the decision-maker’s

information about the relative appeal of the solar system. Nonetheless, in this case study,

SunEnergy believes that the performance of substitutes does not help infer much about the

performance of residential solar. Also, the force of Buyers is shaped by other drivers beyond

the customer savings from the solar system, including the customer switching cost and the

power of dealers. Both attributes dilute the sensitivity of Buyers to Substitutes, resulting in the

weak dependence we observe in Figure 5.14b.

Another noticeable result in Figure 5.14b is the interdependence between Buyers and Rivals:

observing strong Rivals increases the likelihood of witnessing strong Buyers; the opposite

is also true. This interdependence is primarily attributed to three shared ancestral drivers,

illustrated in Figure 5.6: Product differentiation among incumbents, Customer switching cost


among industry products, and Willingness of price discounting by incumbents. Included in the

observation of {high power of Rivals} are the observations that {no product differentiation}

exists among the solar offerings in the residential industry, that customers suffer {no legal

complications and low financial penalty} upon switching between solar providers, and that

solar firms are {willing to share > 15% of profits with customers}. The lack of product

differentiation leaves the end-customer with price as the only dimension to bargain over.

Accordingly, the lack of product differentiation effectively increases the customer’s price

sensitivity, which, along with the customer’s ability to freely switch between solar vendors,

increases the bargaining power of Buyers. Equivalently, the observation that the solar

incumbents are willing to share a substantial portion of the system value with the Buyers (e.g.

high dealer commission) allows inferring that the power of Buyers will more likely be {high}.

Interestingly, the same shared drivers are responsible for the comparable dependence of Rivals

on Buyers, depicted in Figure 5.14d: observing strong Buyers increases the likelihood of

witnessing strong Rivals; the opposite is also true. After explaining how these drivers affect

the bargaining power of Buyers, let us now do the reverse analysis and explain how they affect

Rivalry among incumbents. Simply put, all three shared drivers are related to the economic

performance of the solar system. Upon observing {high power of Buyers}, both {no product

differentiation} and {no legal complications and low financial penalty} induce the

homeowners to prioritize “economics” as the most important consideration when shopping for

solar panels. Likewise, observing the willingness of solar firms to {share > 15% of profits

with customers} reflects a business culture that is strongly biased towards managing

economics instead of, for example, delivering a fast service or gaining customer loyalty. When

both the customers and the firms focus on the solar system’s economics, price-based

competition becomes inevitable, and therefore {high} Rivalry becomes more likely.

Rivals also exhibits symmetrical interdependence with New Entrants, evident in the tornado

plots of both Figures 5.14d and 5.14c. The probabilistic dependence between the two forces is

similar in magnitude, but the two forces move in opposite directions. In Figure 5.14d,

observing strong New Entrants decreases the likelihood of witnessing strong Rivals.

Equivalently, in Figure 5.14c, observing strong Rivals decreases the likelihood of

witnessing strong New Entrants. The opposites hold true in both cases. Here again, the


strategic interaction between the two forces balances the effects of their common ancestral

drivers: Commitment of incumbent to retain and fight over market share, Willingness of price

discounting by incumbent, and Customer switching cost among industry products. In a

competitive landscape where rivalry is high, we observe that solar firms {focus on relative

growth} and express clear willingness to {share > 15% of profits with customers}. Both

behaviors indicate a high likelihood that industry incumbents will seek to retaliate from new

entrants, therefore deterring the threat of new entry. However, the ramifications of both

behaviors are counteracted by the third driver, which shifts the forces of Rivals and New

Entrants in the same direction. High rivalry may imply {no legal complications and low

financial penalty} when customers switch among solar providers, which then may accelerate

the adoption rate of the solar offerings by new entrants and therefore increase the overall

threat of entry.

Figure 5.14c depicts two additional results concerning New Entrants that are worth discussing.

Specifically, observing strong Buyers or Suppliers increases the likelihood of witnessing

strong New Entrants. This result stems from the ability of either Buyers or Suppliers to

extend their activities into the solar industry; observing {high power of Buyers} signifies a

real threat of backward integration by dealers, and observing {high power of Suppliers}

signifies a real threat of forward integration by suppliers. In both cases, the danger of new

entry becomes more imminent.

The last strategic insight in Figure 5.14 concerns the limited interaction between Suppliers

and the other competitive forces. As evident in Figures 5.14a, 5.14b, and 5.14d, observing

the power of Suppliers (and its drivers) does not update the decision-maker’s beliefs about the

power of Substitutes, Buyers, or Rivals, respectively, in this case study. Equivalently, Figure

5.14e shows that the power of Suppliers is not sensitive to the change in the power of

Substitutes, Buyers, or New Entrants. The minor dependence of Suppliers on Rivals can be

attributed to the single common ancestral driver between both forces: Fragmentation of

industry. Observing {high Rivalry} accounts for the existence of oligopolistic competition

among established solar firms. The presence of multiple, similarly sized firms signifies

relatively low population (i.e. high concentration) of suppliers relative to incumbents, which in

turn implies higher bargaining power for Suppliers.


3.1.4. Economic Interdependence

Equally important to examining the interactions among the competitive forces is testing the

sensitivity of the industry’s economic performance to their change. To that end, Figures

5.15a–d plot the range of the Effective Cost, Effective Price, Total Quantity, and EBT

associated with observing each force between its two extremes. Consistent with Porter’s

teachings, and consequently with DAFF’s underlying logic, Figures 5.15a and 5.15b show that

stronger forces yield higher costs and lower prices. In fact, observing a force at its highest

extreme may yield a COGS as high as 1.83 $/W and a price as low as 3.09 $/W. On the other

hand, observing a force at its lowest extreme may yield a cost as low as 1.00 $/W and a price

as high as 3.66 $/W.

Notably, the power of Buyers has the highest impact on Effective Cost, which we attribute to

two main reasons. First, referencing our discussions on Figure 5.14, we know that observing

strong Buyers increases the likelihood of witnessing strong Substitutes (Figure 5.14a), strong

New Entrants (Figure 5.14c), and strong Rivals (Figure 5.14d); in other words, observing

powerful Buyers implies a competitive reality in which three other forces are likely to be

powerful (Substitutes, New Entrants, and Rivals). Inevitably, such combination of four strong

competitive forces increases the likelihood of witnessing high production, installation, and

servicing costs. The second reason has to do with the direct probabilistic relevance between

the Buyers’ driver uncertainties and the Regulation uncertainties, especially those affecting

solar financing. Observing strong Buyers entails knowing, with certainty, that the solar

Product perceived cost as fraction of the customer budget is {> 20% of household annual

income}. Given this information, the decision-maker may reasonably infer that the R10: ITC

and Depreciation subsidies are not fully utilized, which raises the expected Effective Cost of

the solar system.

Beyond price and cost, the total sales of the solar system are also sensitive to the force of

Buyers, as illustrated in Figure 5.15c. That said, Total Quantity is even more sensitive to the

resolution of two other forces: Substitutes and New Entrants. Per our DAFF modeling,

observing {high power of Substitutes} increases the likelihood of witnessing limited solar

sales. However, observing {high power of New Entrants} induces a probabilistic inference in


the opposite direction; it allows inferring a more likely {rapid Growth rate} and a less likely

{high power of Substitutes}, both of which then imply higher chances of witnessing vast solar

sales.

Figure 5.15: Effect of the competitive forces on economics of the U.S. residential solar PV

industry

Subsequently, upon balancing the relevance relations among the numerous uncertainties, the

DAFF Bayesian Network allows us to evaluate how sensitive the solar industry’s EBT is to

observing the extremes of each competitive force. Remarkably, Figure 5.15d shows that in a

competitive setting where the power of Buyers is {high}, profits can be as low as 1.82 billion

$/year; conversely, in a competitive setting where the power of Buyers is {low}, profits can be

1 1.25 1.5 1.75 2

Effective Cost ($/W)

3 3.25 3.5 3.75 4

Effective Price ($/W)

1 1.25 1.5 1.75 2 2.25 2.5

Total Quantity (GW)

1.5 2.5 3.5 4.5 5.5 6.5

EBT (billion $/year)

Suppliers

Rivals

New Entrants

Buyers

Substitutes

(a) (b)

(c) (d)

observed as {high}

observed as {Low}


as high as 5.77 $/year. Beside this wide range of expected EBT, the DAFF model offers a key

insight that may seem counterintuitive: unlike the other four competitive forces, observing

{high power of New Entrants} yields a higher expected EBT and therefore signifies a more

favorable competitive landscape. To comprehend this result, we first reemphasize the need to

distinguish between “relevance” and “causality” in probabilistic assessment. We assert that

higher threat of entry does not cause more favorable competitive landscape in the residential

solar industry. Rather, observing a high threat of entry implies that the competitive landscape

is likely attractive; naturally, the more profitable the industry, the more firms are interested in

joining it.

Overall, Figure 5.15 shows that Substitutes and Buyers are the top two competitive influencers

of economic performance in the U.S. residential solar PV industry, followed by Rivals and

New Entrants. While observing the competitive forces is a beneficial thought-experiment to

understand their impacts on the overall industry, reaching perfect clairvoyance on competition

is almost impossible. Therefore, when designing its competitive strategy, SunEnergy has to

not only gather information about the competitive powers in its environment but also control

these powers through proper positioning. The effect of positioning decisions on SunEnergy’s

success is what we discuss in the next section.

3.2 Second Step: Assess Each Positioning Segment in the Industry

This case study examines a set of 32 feasible positioning tracks for SunEnergy in the U.S.

residential solar PV industry. Each positioning track yields a unique competitive landscape

and, accordingly, a unique profit (EBT) value. Figure 5.16 ranks the expected EBT for all

positioning tracks. For convenience, we abbreviate the labeling of the Customer Financing

alternatives such that “purchase”, “loan”, “leaseF” and “leaseN” correspond to [Direct

Purchase], [Loan], [Full-upfront Lease], and [No-upfront Lease], respectively.

We start with a series of modeling notes in order to guarantee the precise and proper

interpretation of the numerical findings in Figure 5.16. First, we remind that the EBT value of

a positioning segment corresponds not to an individual solar firm (e.g. SunEnergy) but to all

solar firms positioning in that segment. Second, because positioning tracks are mutually


exclusive, their EBT values are not additive; rather, each positioning track characterizes a

prospective reality about the residential solar industry and therefore should be assessed

independently. Third, while our DAFF value node accounts for the operational costs and

revenues in a specific positioning segment, it does not account for any capital investment that

might be needed upfront to start a business in that segment. To that end, the EBT values

presume that firms are already operational in the positioning segment. To demonstrate these

points, let us interpret one positioning track as an example: the EBT value for the [Urban,

Dealer, Loan, Insource] positioning track represents the “expected earnings before tax for the

whole residential solar market in the U.S., where all solar firms locate in urban areas, rely on

dealers to acquire and manage customers, sell their solar systems through loans, and

manufacture their own solar components.”

Given our specific market assumptions, information, and beliefs, Figure 5.16 shows that

different positioning tracks may render the expected EBT of the residential solar industry as

high as 3.98 billion or as low as 0.51 billion $/year. The highest profits are realized in a

competitive setting where incumbents choose to: locate in urban areas, cut dealers out,

and offer leasing services to their customers. On the other hand, the lowest profits result

when firms decide to: run the solar business in rural areas, use dealers, and sell through direct-

purchase agreements. Notably, both the top-two and the bottom-two positioning tracks score

very comparable EBT values despite their different Panel Manufacturing alternatives.

Replicated multiple times throughout the numerical outputs of Figure 5.16, this observation

signifies that, relative to Regional Focus, Downstream Integration, and Customer Financing,

Panel Manufacturing is the least influential positioning decision in shaping the competitive

landscape and the industry’s profitability.

Another important insight is related to the collective influence of positioning decisions. While

the [Urban, No-upfront Lease, No Dealer, Insource] positioning track is likely to yield the

highest profit, this result is not sufficient to conclude that [Urban], [No-upfront Lease], [No

Dealer], and [Insource] are, independently, the best positioning alternatives for their respective

decisions. It is only the combination these alternatives that collectively influences the

competitive forces and factors in the most favorable way to solar firms. For instance, when


combined with [Rural, Loan, and Outsource], [Dealer] results in slightly higher expected EBT

than [No Dealer].

Figure 5.16: Profitability of the various positioning tracks in the U.S. residential solar PV

industry

This characteristic of positioning alternatives is a good manifestation of “strategic fit”, an

important notion introduced by Porter and discussed earlier in Chapter 4. When positioning

activities fit together, not only they become harder to imitate by competitors, but also their

combined benefit becomes larger than the sum of their individual benefits. A representative

0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0

Rural, Purchase, Dealer, Outsource

Rural, Purchase, Dealer, Insource

Rural, Purchase, NoDealer, Outsource

Rural, Purchase, NoDealer, Insource

Rural, Loan, NoDealer, Outsource

Rural, Loan, Dealer, Outsource

Rural, Loan, NoDealer, Insource

Rural, Loan, Dealer, Insource

Rural, LeaseF, Dealer, Outsource

Rural, LeaseF, Dealer, Insource

Rural, LeaseF, NoDealer, Outsource

Rural, LeaseF, NoDealer, Insource

Rural, LeaseN, Dealer, Outsource

Rural, LeaseN, Dealer, Insource

Rural, LeaseN, NoDealer, Outsource

Rural, LeaseN, NoDealer, Insource

Urban, Purchase, Dealer, Outsource

Urban, Purchase, Dealer, Insource

Urban, Purchase, NoDealer, Outsource

Urban, Loan, Dealer, Outsource

Urban, Purchase, NoDealer, Insource

Urban, Loan, Dealer, Insource

Urban, LeaseF, Dealer, Outsource

Urban, LeaseF, Dealer, Insource

Urban, Loan, NoDealer, Outsource

Urban, Loan, NoDealer, Insource

Urban, LeaseN, Dealer, Outsource

Urban, LeaseN, Dealer, Insource

Urban, LeaseF, NoDealer, Outsource

Urban, LeaseF, NoDealer, Insource

Urban, LeaseN, NoDealer, Outsource

Urban, LeaseN, NoDealer, Insource

EBT (billion $/year)

base-case

individual benefit

combined benefit

example of "strategic fit"


example of fit in this case study is highlighted in Figure 5.16. We start with a base-case

positioning track of [Urban, Direct Purchase, Dealer, Insource], which achieves an expected

EBT of $1.81 billion $/year. If the solar firms cut the dealers out and choose to position in

[Urban, Direct Purchase, No Dealer, Insource] instead, the expected EBT increases to 2.29

billion $/year (gains of 0.48 billion). Otherwise, if solar firms shift from a direct-purchase

sales model to full-upfront lease contracts and position in [Urban, Full-upfront Lease, Dealer,

Insource], the expected EBT increases to 2.81 billion $/year (gains of 1 billion). Now, if solar

firms choose to both cut dealers out and offer full-upfront leases, thus positioning in [Urban,

Full-upfront Lease, No Dealer, Insource], the expected EBT increases all the way to 3.62

billion $/year; the combined gains of 1.81 billion here are larger than the sum of the two

individual gains of 0.48 and 1 billion, which can be attributed to the strategic fit between the

[Dealer] and [Full-upfront Lease] positioning alternatives.

Subsequently, to better understand how positioning influences the competitive forces, Figure

5.17 plots the range of change in the power of each competitive force under the various

feasible positioning tracks in the industry. The principal message from Figure 5.17 is

consistent with that conveyed in Figure 5.16: positioning can significantly influence the

competitive landscape faced by the solar firm. For example, we see that some positioning

tracks reduce the likelihood of {high threat of Substitutes} all the way to 0.28 while other

positioning tracks raise the likelihood of this prospect to 0.54. Buyers seem to be the most

prone to control by the four considered decisions; different positioning strategies may set the

probability of {high bargaining power of Buyers} anywhere between 0.18 and 0.71.

Conversely, Suppliers seem to be the least sensitive to positioning; the 32 positioning tracks

result in a slim change in the probability of {high bargaining power of Suppliers}, ranging

between 0.39 and 0.41. Appendix B tabulates the power of each competitive force for each

positioning track, in more detail.

Interestingly, comparing the results in Figures 5.14 and 5.17 shows that the four competitive

forces mostly sensitive to observing new information are also the ones mostly sensitive to

positioning: Substitutes, Buyers, New Entrants, and Rivals. The opposite is true for Suppliers.

While this outcome is not necessarily generalizable, it does reflect our intentional attempt in

this case study to focus on controlling the forces that are strongly interdependent. These forces


play a major role in shaping the economics of the residential solar industry, so influencing

them further SunEnergy’s ability to position where the industry performance is superior.

Figure 5.17: The influence of positioning on the competitive forces in the U.S. residential

solar PV industry

Also in Figure 5.17, we highlight the positioning alternatives that cause four of the

competitive forces to be strongest or weakest: Substitutes, Buyers, New Entrants, and Rivals.

This mapping provides additional insight into the role of individual positioning decisions, and

the results reaffirm our earlier conclusions in Figure 5.16. To start, managing customers

directly and cutting dealers out seem to be the optimal Downstream Integration alternative;

[No Dealer] both reduces competition (Figure 5.17) and increases profitability (Figure 5.16).

Equivalently, allowing customers to finance their solar system through loans or no-upfront

leases seem to be the optimal Customer Financing alternatives; [Loan] and [No-upfront

Lease] minimize competitive threats and enhance profitability. On the other hand, the

influence of the Regional Focus positioning decision is rather unique. Here, positioning in

[Urban] areas seem to expose solar firms to higher competitive forces (Figure 5.17) even

though it eventually results in higher expected EBT (Figure 5.16); the opposite is true for

positioning in [Rural]. One explanation for this trend is that urban regions offer a bigger solar

market with more potential customers. In this case, although stronger competitive forces may

0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

Substitutes

Buyers

New Entrants

Rivals

Suppliers

Probability of {High} power of force

Urban

Dealer

Purchase

Rural

NoDealer

LeaseNLoan


erode some of the firms’ profit margins, the total sales are high enough to overcompensate for

this effect and still uphold [Urban] as an attractive positioning alternative.

Building on the preceding results, we now attempt to answer the central question of this case

study: where should SunEnergy position in the U.S. residential solar PV industry over the next

two years? In line with Porter’s teachings and our DAFF modeling, SunEnergy should

position where it achieves the highest expected EBT. If SunEnergy anticipates gaining the

same market-share and enduring comparable (or no) upfront capital costs in all market

segments, then it should position in the highest-EBT track of Figure 5.16: [Urban, No-upfront

Lease, No Dealer, Insource]. If, for whatever reason, implementing this positioning strategy is

not possible, SunEnergy should pick the second-highest-EBT track [Urban, No-upfront Lease,

No Dealer, Outsource], then the third [Urban, Full-upfront Lease, No Dealer, Insource], and so

forth. Following the same rationale, SunEnergy should avoid positioning in the lowest-EBT

track [Rural, Direct Purchase, Dealer, Outsource]. Nonetheless, if SunEnergy believes that its

prospective market-share or upfront capital investments in the residential business are likely to

differ with its positioning, then the best positioning strategy cannot be directly deduced from

our results, and further analysis is needed; as explained in Chapter 4, it is precisely here that

the third step of the first objective in competitive strategy becomes essential. Still, even

without the third-step analysis, our results from the first and second steps provide SunEnergy

with clearer intuition on the optimal competitive strategy in the U.S. residential solar PV

industry over the next two years. By now, SunEnergy knows how to better prioritize its

positioning plans: focus on the top tracks and avoid the bottom tracks in Figure 5.16. Along

the same lines, SunEnergy also knows that it is beneficial to: locate in urban areas with higher

sales despite stronger competition; manage customers directly without relying on dealers –

except in some rural setting; offer leases and loans to finance its customers’ solar systems

while avoiding direct-purchase agreements; and combine lease-offerings with direct-customer-

management in one business model to generate a strategic fit that is hard to imitate. These

concrete positioning recommendations would have been hard to materialize without DAFF.


4 Conclusions

Is the competitive landscape in the U.S. residential solar PV industry favorable in the near

future? And if so, where should SunEnergy position? The purpose of this case study is to

answer these two key questions, and our attempt to do so centers on developing and evaluating

decision analytic five forces (DAFF) models. After presenting a brief overview of the solar

photovoltaic industries in the United States as well as of SunEnergy’s business, we delve into

constructing the various elements of the DAFF models. First, we describe how to design the

five competitive forces and their underlying drivers as uncertainties. Also accounted for as

uncertainties are several important regulatory, technological, and growth factors that shape the

residential solar business. Both the generalizable forces and the industry-specific factors

impact the cost, price, and sales of solar systems, all of which are treated as uncertain

economic parameters. Here, we spend some time explaining how to account for governmental

subsidies and tax regulations in our economic modeling, and we justify the use of earnings

before tax (EBT) as the value metric. Along the way, we describe the probabilistic relevance

among the competitive, regulatory, technological, growth, and economic uncertainties, so that

we can eventually connect them in a single DAFF Bayesian Network. This Network captures a

lot of SunEnergy’s knowledge and beliefs about the overall residential solar industry in the

U.S. and its potential advancement in the next two years.

On its own, the DAFF Bayesian Network provides important results on the competitive

performance of the overall industry. Among the five competitive forces, only rivalry is

expected to be relatively strong; the chance of residential solar firms suffering {high power of

Rivals} over the next two years is 0.62, compared to about 0.4 for {high} powers of

Substitutes, Buyers, New Entrants, and Suppliers. This predicted competitive landscape yields

an expected EBT of 4.05 billion $/year for the whole U.S. residential solar market, with a

probability-weighted average cost, price, and installation capacity estimated at 1.19 $/W, 3.35

$/W, and 1.93 GW, respectively.

The DAFF Bayesian Network also highlights the interdependence among the five forces. Most

notably, we document robust interdependence among Substitutes, Buyers, New Entrants, and

Rivals. A careful tracking of the competitive DAFF logic shows that the interdependence


relations associated with a given force need not have the same magnitude or direction, and the

interdependence between two competitive forces need not be symmetrical. In this case study,

observing strong Buyers increases the likelihood of witnessing strong Substitutes significantly,

but observing strong Substitutes increases the likelihood of witnessing strong Buyers only

mildly. Conversely, the interdependence between Buyers and Rivals is more symmetrical,

where observing strong Buyers signifies strong Rivals, and vice versa. In addition, both

Substitutes and Rivals show inversely proportional dependence on New Entrants; observing

strong New Entrants decreases the likelihood of witnessing strong Rivals or strong Substitutes.

One important DAFF element that plays a major role in shaping these interactions among the

forces is the shared parent drivers. When the decision-maker gains new market intelligence,

these shared drivers facilitate the probabilistic updating of his beliefs about one or multiple

competitive forces through conditional reasoning and inference.

Along the same lines, Substitutes and Buyers prove to be the top two competitive influencers

of economic performance in the U.S. residential solar PV industry. The industry’s profitability

seem to be mostly sensitive to the power of Buyers, for changing the latter updates the

decision-maker’s beliefs about three other forces: Substitutes, New Entrants, and Rivals.

Accordingly, EBT can be as small as 1.82 billion $/year in a competitive setting where the

power of Buyers is {high} and as large as 5.77 billion $/year in a competitive setting where

the power of Buyers is {low}. Solar incumbents can capitalize on this insight to reduce

competition in their industry, regardless of how they choose to position. Effectively,

incumbents can work collaboratively to deter the threat of Substitutes (e.g. lobby collectively

for more favorable regulations), or they can work individually to secure advantageous

relations with their Buyers (e.g. negotiate exclusive partnerships with dealers).

After assessing the competitive performance of the overall industry, we analyze the

competitive performance in multiple positioning segments that are of interest to SunEnergy.

Four positioning decisions are investigated, each influencing multiple force and factor

uncertainties. Regional Focus is a Value Proposition decision that addresses what customers

to serve: [Rural] or [Urban]. Then, Downstream Integration, Customer Finance, and Panel

Manufacturing are Value Chain decisions that determine how to operate the business: whether

to serve customers directly [No Dealer] or through intermediate dealers [Dealer]; whether to


sell the solar system through [Direct Purchase] or offer product financing in the form of

[Loan], [Full-upfront lease], or [No-upfront lease]; and whether to [Insource] or [Outsource]

the manufacturing of the solar panels. Adding the four decisions to the Bayesian Network

results in a complete DAFF Decision Diagram, and combining the various decision

alternatives results in 32 possible positioning tracks.

Each positioning track is characterized by a unique competitive landscape and,

correspondingly, a unique EBT profit value. The highest EBT of 3.98 billion $/year is realized

by positioning in [Urban, No Dealer, No-upfront Lease, Insource] while the lowest EBT of

0.51 billion $/year is obtained upon positioning in [Rural, Dealer, Direct Purchase, Insource].

Overall, the outputs from the Decision Diagram show that Customer Financing and

Downstream Integration are influential in shaping the residential solar industry. While loans

and leases consistently achieve lower competitive forces and higher EBT, direct-purchase

agreements prove to be robustly inferior. Similarly, cutting dealers out and managing

customers directly prove to be a favorable positioning strategy in most cases, especially in

urban areas. In contrast, the influence of Panel Manufacturing seems to be diluted by other

positioning decisions, resulting in no clear impact on the five forces or EBT. Regional Focus

plays a unique role in market segmentation; locating in urban areas exposes the solar firm to

higher competitive forces but also larger customer pool. Consequently, while urban

positioning is likely to yield lower prices and higher costs per solar system, its relatively high

sales are likely to compensate for its relatively low margins and ultimately produce superior

EBT compared to rural positioning.

As should be clear by now, the DAFF models track and quantify strategic interactions not only

among the industry’s competitive forces but also among the firm’s positioning decisions and

between the forces and the decisions. These interactions are neither universal across industries

nor generalizable in a given industry, however; rather, they are primarily shaped by the firm’s

subjective information and beliefs. Even if we assume a standardized DAFF model for the

U.S. residential solar PV industry within the next two years, different decision-makers from

different solar firms may have different market intelligence and business intuition, which

results in different design of uncertainty degrees or different probability assignments over

those degrees. In fact, beyond the subjective assessment of uncertainty, different corporate


cultures and managerial experiences may incentivize decision-makers to model different

positioning decisions or alternatives. For instance, a decision-maker with an economics

background may want to focus on financing decisions whereas a decision-maker with an

engineering background may want to focus on product-design decisions. Inevitably, the

subjective input into the DAFF models results in subjective outputs and recommendations,

unique to the decision-maker’s outlook on the industry. In a way, we assert this DAFF

characteristic as an advantage, for it preserves and promotes Michael Porter’s teachings that

firms should compete to be “unique” not “best.” In this case study, the DAFF models ensures

that SunEnergy adopts a unique competitive strategy that matches its unique information,

preferences, and resources in the residential solar industry.

Although DAFF may provide different results and recommendations to different decision-

makers, it always provides the same clarity to all decision-makers. Regardless of whether the

competitive forces turn out to be strong or weak, and irrespective of what positioning tracks

yield the highest EBT, DAFF ensures that the decision-maker achieves “clarity of thought”

when assessing competition in his business. In fact, when we asked SunEnergy’s Director of

Strategic Planning about the benefits he gained from this modeling exercise, clarity was on the

top of his list. As he insightfully explained, DAFF incorporates and enforces clarity in every

step of modeling the five forces and the positioning decisions. From a structural standpoint,

the model requires the decision-maker to think about an extensive list of drivers and factors

that shape the competitive landscape. As emphasized in Chapter 4, examining this list of

uncertainties reduces cognitive biases that may magnify the role of some competitive

attributes while attenuating (or even ignoring) the role of others. One level deeper, clarity is

also ensured via the need to design mutually exclusive and collectively exhaustive degrees, as

well as the need to analyze prospects that involve multiple tradeoffs among those degrees. On

multiple occasions, our probability elicitations from SunEnergy’s decision-maker have led to

“aha!” moments with regards to the industry’s activities and the competitive players’ behavior.

A similar level of clarity is achieved when thinking about positioning. Rather than cramming

all positioning decisions in one vague or seemingly indivisible plan, DAFF allows the

decision-maker to model a distinct decision for virtually every step along the business’s value


chain. Additionally, examining the various combinations of positioning alternatives helps

widen the decision-maker’s perspective about the range of strategic choices at his disposal.

4.1 Future Work

Despite its many advantages, the proposed DAFF model in this case study can still be

improved on multiple fronts. To start, the current profit-value metric covers all solar firms in

every positioning segment. Only if we assume that SunEnergy’s market-share and upfront

capital costs are comparable across all positioning segments will the current model allow an

accurate ranking of the company’s own profit in those segments. Because such assumptions

may be impractical, future work may extend the DAFF modeling in two ways. First, it would

be rather useful to undertake the third step of this competitive analysis, which allows adding

SunEnergy’s market-share as an uncertainty and updating the decision diagram’s value node

into SunEnergy-specific EBT. Second, for every positioning track, SunEnergy’s residential

solar team can provide refined estimates of any upfront capital investments that the company

may need in order to initiate or expand its residential business.

Beyond these economic updates, SunEnergy may gain deeper insight into competitive

positioning either by adding new positioning decisions or alternatives or by further

disintegrating current positioning decisions along the value chain. In that regard, for some

positioning decisions, it may be more realistic to introduce new alternatives as the

combination of multiple existing alternatives. For example, in addition to positioning in either

[Urban] or [Rural] areas, Regional Focus would examine positioning in [Urban and Rural]

combined. Moreover, it would be beneficial to add another alternative to Customer Financing

in order to account for Power Purchase Agreements (PPA) in the industry. PPA is similar to

leasing, for both options maintain the system ownership under the solar firm. However, while

the lease requires the customer to pay a fixed monthly fee for the solar energy, PPA requires

the customer to pay per unit of solar energy consumed. Lastly, it might be helpful to

decompose Downstream Integration into multiple decision nodes that distinguish between

different types of dealers: financial dealers (e.g. asset securitization), operation dealers (e.g.

maintenance) and customer-relations dealers (e.g. sales generation). True to the premise of


DAFF, all these modeling extensions would further clarify positioning and therefore would

further inform SunEnergy’s competitive strategy.


References

[1] S. Kann, M. Shiao, C. Honeyman, N. Litvak, J. Jones, L. Cooper, T. Kimbis, J. Baca, S. Rumery

and A. Holm, "U.S. Solar Market Insight: Q1 2015 Executive Summary," Greentech Media and

Solar Energy Industries Association, United States, 2015.

[2] SEIA-a, "Solar Industry Data," 2015. [Online]. Available: http://www.seia.org/research-

resources/solar-industry-data. [Accessed 2015].

[3] The Solar Foundation, "National Solar Jobs Census," 2015. [Online]. Available:

http://www.thesolarfoundation.org/solar-jobs-census/national/. [Accessed 2015].

[4] NEI, "Fact Sheets: Nuclear Power Plants Benefit State and Local Economies," 2015. [Online].

Available: http://www.nei.org/Master-Document-Folder/Backgrounders/Fact-Sheets/Nuclear-

Power-Plants-Contribute-Significantly-to-S. [Accessed 2015].

[5] M. Munsell, "US Solar Market Grew 41%, Had Record Year in 2013," Greentech Media, 07

March 2014.

[6] O. Edenhofer, R. Pichs-Madruga, Y. Sokona, K. Seyboth, P. Matschoss, S. Kadner, T. Zwickel, P.

Eickemeier and G. Hansen, "Summary for Policymakers," in IPCC Special Report on Renewable

Energy Sources and Climate Change Mitigation, Cambridge, United Kingdom and New York,

NY, USA, Cambridge University Press, 2011.


January 2008.

[8] A. Goodrich, T. James and M. Woodhouse, "Residential, Commercial, and Utility-Scale

Photovoltaic (PV) System Prices in the United States: Current Drivers and Cost-Reduction

Opportunities," National Renewable Energy Laboratory. Contract No. DE-AC36-08GO28308,

Golden, Colorado, 2012.

[9] C. Brehaut, "Global PV Monitoring 2014-2018: Technologies, Markets and Leading Players,"

GTM Research, Greentech Media, United States, 2014.

[10] V. Bugnion, "Clearly Energy: Residential Demand-Response Programs," 2014. [Online].

Available: https://www.clearlyenergy.com/residential-demand-response-programs. [Accessed

2015].

[11] Navigant Research, "Residential Demand Response. Direct Load Control and Dynamic Pricing

Programs, DR Markets, and DR Management Systems for Residential Customers: Global Market

Analysis and Forecasts," 2014. [Online]. Available:

https://www.navigantresearch.com/research/residential-demand-response. [Accessed 2015].

[12] J. Niccolai, "Google invests in another wind farm for its data centers," Computerworld, 21 April

2011.


[13] S. Specker, J. Phillips and D. Dillon, "The Potential Growing Role of Post-Combustion CO2

Capture Retrofits in Early Commercial Applications of CCS to Coal-Fired Power Plants," MIT

Coal Retrofit Symposium, 2009.

[14] K. Farhat, S. Comello, S. Reichelstein, F. Mormann and D. Reicher, "Appendix D: State-Level

Incentives," in The Federal Investment Tax Credit for Solar Energy: Assessing and Addressing the

Impact of the 2017 Step-Down, Stanford, Stanford University, 2015.

[15] N. Litvak, "U.S. Residential Solar Financing 2015-2020," GTM Research, United States, 2015.

[16] M. Yozwiak, "How extending the investment tax credit would affect US solar build," Bloomberg

New Energy Finance, 2015.

[17] S. Kann, M. Shiao, C. Honeyman, N. Litvak, J. Jones, C. Smith, L. Cooper, T. Kimbis, J. Baca, S.

Rumery and A. Holm, "Solar Market Insight Report 2015 Q2: Market Outlook," 2015. [Online].

Available: http://www.seia.org/research-resources/solar-market-insight-report-2015-q2. [Accessed

2015].

[18] K. Aanesen, S. Heck and D. Pinner, "Solar power: Darkest before dawn," McKinsey & Company,

2012.

[19] Decision Systems Laboratory, "GeNIe & SMILE," 2013. [Online]. Available:

https://dslpitt.org/genie/. [Accessed 2015].

[20] Nest, "Saving energy starts with your thermostat," 2015. [Online]. Available:

https://nest.com/thermostat/real-savings/. [Accessed 2016].

[21] EcoFactor, "Optimized Demand Response," 2015. [Online]. Available:

http://www.ecofactor.com/services/#drcloud. [Accessed 2016].

[22] Navigant, "Residential Demand Response Revenue is Expected to Reach $2.3 Billion Annually by

2023," Navigant Research, 24 November 2014.

[23] Ranking the Brands, "SolarCity," 2015. [Online]. Available:

http://www.rankingthebrands.com/Brand-detail.aspx?brandID=3097.

[24] E. Wesoff, "‘World’s Most Efficient Rooftop Solar Panel’ Revisited," Greentech Media, 13

October 2015.

[25] D. Feldman, R. Margolis and D. Boff, "Q1/Q2 ‘14 Solar Industry Update," SunShot - U.S.

Department of Energy, 2014.

[26] ENF, "Solar Panel Manufacturers," 2014. [Online]. Available:

http://www.enfsolar.com/directory/panel. [Accessed 2014].

[27] E. Atkin, "We Are On The Verge Of An Electric Car Battery Breakthrough," ClimateProgress, 31

August 2014.


[28] S. Lacey, "Storage Is the New Solar: Will Batteries and PV Create an Unstoppable Hybrid

Force?," Greentech Media, 15 June 2014.

[29] D. Savenije and E. Howland, "10 predictions for the electric sector in 2014," Utility Dive, 6

January 2014.

[30] J. Smart and S. Schey, "Battery Electric Vehicle Driving and Charging Behavior Observed Early in

The EV Project," Advanced Vehicle Testing Activity - Idaho National Laboratory, 2012.

[31] D. R. Baker, "Most electric vehicle drivers charge them at home," SFGate, 21 November 2013.

[32] D. Howell, "Tesla Battery Factory Might Power Up SolarCity, Apple," Investors.com, 21 02 2014.

[33] W. Warwick, "A Primer on Electric Utilities, Deregulation, and Restructuring of U.S. Electricity

Markets," Pacific Northwest National Laboratory, Richland, Washington, 2002.

[34] EIA, "Status of Electricity Restructuring by State," U.S. Energy Information Administration, 2010.

[Online]. Available: http://www.eia.gov/electricity/policies/restructuring/restructure_elect.html.

[Accessed 2016].

[35] W. Pentland, "After Decades Of Doubt, Deregulation Delivers Lower Electricity Prices," Forbes,

13 October 2013.

[36] SEIA-b, "Solar Investment Tax Credit (ITC)," 2015. [Online]. Available:

http://www.seia.org/policy/finance-tax/solar-investment-tax-credit. [Accessed 2015].

[37] SEIA-c, "Depreciation of Solar Energy Property in MACRS," 2015. [Online]. Available:

http://www.seia.org/policy/finance-tax/depreciation-solar-energy-property-macrs. [Accessed

2015].

[38] Investopedia-a, "Fair Market Value," 2016. [Online]. Available:

http://www.investopedia.com/terms/f/fairmarketvalue.asp. [Accessed 2016].

[39] H. K. Trabish, "Why Treasury Is Investigating SolarCity and Solar Third-Party Funds," Greentech

Media, 19 December 2013.

[40] A. H. Miller, "SunPower expanding internationally," CleanEnergyAuthority.com, 21 December

2012.

[41] RePower, "Solar Universe Enters International Market and Debuts First Franchise in Puerto Rico,"

2013. [Online]. Available: https://repower.solaruniverse.com/press-releases/solar-universe-

announces-local-franchise-offices-in-puerto-rico. [Accessed 2015].

[42] S. Kann, M. Shiao, S. Mehta, C. Honeyman, N. Litvak, J. Jones, J. Baca, S. Rumery and A. Holm,

"U.S. Solar Market Insight Report - Q1, 2014," Greentech Media and Solar Energy Industries

Association, United States, 2014.


[43] R. A. Howard and A. E. Abbas, Foundations of Decision Analysis, 1st ed., United States: Pearson,

2016.

[44] Z. Shahan, "What Is The Current Cost Of Solar Panels?," CleanTechnica, 4 February 2014.

[45] M. Munsell, "Solar PV Pricing Continues to Fall During a Record-Breaking 2014," Greentech

Media, 13 March 2015.

[46] N. Litvak, "U.S. Residential Solar Financing: 2014-2018," GTM Research, California, 2014.

[47] Investopedia-b, "Earnings Before Tax - EBT," 2016. [Online]. Available:

http://www.investopedia.com/terms/e/ebt.asp. [Accessed 2016].

[48] R. Dietz, "The Geography of Home Size and Occupancy," National Association of Home Builders,

2 December 2011.

[49] United States Census Bureau, "Census Urban and Rural Classification and Urban Area Criteria,"

2010. [Online]. Available: https://www.census.gov/geo/reference/ua/urban-rural-2010.html.

[Accessed 2015].

[50] R. D. Shachter, "Bayes-Ball: The Rational Pastime (for Determining Irrelevance and Requisite

Information in Belief Networks and Influence Diagrams)," Morgan Kaufmann Publishers Inc. San

Francisco, 1998.

CHAPTER 5 — Appendix A: Degree Characterization for Competitive Uncertainties 278

Appendix A: Degree Characterization for Competitive Uncertainties

Table 5.A1: Definition of competitive force and driver uncertainties in the U.S. residential

solar industry

Uncertainty Degrees

Substitutes

Power of Substitutes high low

Price-performance tradeoff

relative to this industry’s

product

substitute superior to

solar

substitute equivalent

to solar

substitute inferior to

solar


feasibility

substitute more favorable than

solar

substitute less favorable than

solar

Substitute performance superior inferior

Substitute upfront cost higher upfront cost lower upfront cost

Substitute bill savings less bill savings more bill savings

Substitute installation &

maintenance faster or easier slower and harder

Substitute control &

operation

more

automated

with higher

consumer

control

less automated

with higher

consumer

control

more

automated

with lower

consumer

control

less automated

with lower

consumer

control

Substitute climate impact more emissions’ reduction less emissions’ reduction

Buyers

Bargaining Power of Buyers high low

Power of end-customers high low

Power of dealers high low

Costumer price sensitivity elasticity of demand =

[−3.5 , −2.5] elasticity of demand =

[−2.5, −1.5]


away from the incumbent

product

legal complications or high

financial penalty

no legal complications and low

financial penalty

Dealer ability to influence

customers downstream

heavily influence

customer choice

mildly influence

customer choice

not influence

customer choice


Dealer volume of sales per

incumbent

dealer involved in < 10% of

incumbent sales

dealer involved in > 10% of

incumbent sales

Dealer threat to integrate

backward

backward integration of less

than one dealer per year (on

average)

backward integration of more

than one dealer per year (on

average)

Customer need to trim

immediate cost of industry

product

customer can afford upfront cost customer cannot afford upfront

cost

Dealer partnership with

incumbent exclusive partnership non-exclusive partnership

Dealer reach local national

Dealer brand impactful not impactful

Product perceived cost as

fraction of the customer

budget

< 20% of annual household

income

> 20% of annual household

income

Cost reduction for customer

by industry product < 15% reduction in power bill > 15% reduction in power bill

Amount of cash available for

customer

less than ten-times the system

price

more than 10-times the system

price

Performance improvement

for customer by industry

product

improves power purchase or

utilization

Does not improve power

purchase and utilization

New Entrants

Threat of new entrants high low

Barriers to entry high low

Expected retaliation high low

Size independent

disadvantages for new

entrant

cost disadvantage or need large

scale

no cost disadvantage and no

need for large scale

Size dependent

disadvantages for new

entrant

sustainable size independent

disadvantage

no sustainable size independent

disadvantage

Unequal access to

distribution channel by new

entrant

use existing retail channels need to invest in new retail

channels

Customer adoption rate of

product by new entrant < 70% market-share annual

growth

> 70% market-share annual

growth

Size and availability of

capital needed by new

entrant

need working capital is < $100

million

need working capital is > $100

million


Previous responses by

incumbents acquisition or match offering

no acquisition and no match

offering

Extent of resources available

for incumbents

time before cash runs out is < 1

year

time before cash runs out is > 1

year

Incumbent economies of

scale

economies of scale are fully

realized

economies of scale are in-

progress

Incumbent established brand one or more well-established

brands no well-established brands

Incumbent IP new technology retention rate is

< 3 years

new technology retention rate is

> 3 years

Incumbent cumulative in-

house experience

employee retention rate per year

is < 90% employee retention rate per year

is > 90%

Incumbent prime location proximity to innovation

resources

no proximity to innovation

resources

Limitation of distribution

channels saturated dealer channels unsaturated dealer channels

Control over distribution

channel by incumbent

exclusive dealer-incumbent

agreements

non-exclusive dealer-incumbent

agreements

Customer trust in incumbent positive reputation for

incumbent in specific segments

no positive reputation for

incumbent in any segment

Incumbent network effects weak network effects strong network effects

Efficiency of capital markets weighted average cost of capital

(WACC) is < 13%

weighted average cost of capital

(WACC) is > 13%

Incumbent customer

acquisition cost > 0.5 $/𝑊 < 0.5 $/𝑊

Incumbent R&D spending > 3% of annual revenue, and

less than annual profits

< 3% annual revenue, or greater

than annual profits

Incumbent assets available funds are > $1 billion

per year

available funds are < $1 billion

per year

Incumbent inventory < 100 MW in total capacity > 100 MW in total capacity


capacity

production capacity meets

market demand

production capacity does not

meet market demand

Borrowing power high asset growth and

performance

low asset growth and

performance

Excess cash cash-in-hand is > 10% of total

assets

cash-in-hand is < 10% of total

assets

Rivals

Rivalry high low

Basis of competition price non-price


Intensity of competition high low

Inability to read other


incumbent able to read other

incumbents’ market signals

incumbent not able to read other

incumbents’ market signals

Market segmentation less than 2 segments and

dimensions

more than 2 segments and

dimensions

Extent of exit barriers

exist cost prevent incumbent’s

investment in the top 30% of

the business

exist cost does not prevent

incumbent’s investment in the

top 30% of the business

High commitment to

business < 50% of the incumbent

revenue is from this industry

> 50% of the incumbent

revenue is from this industry

Ability to enforce practices

desirable for whole industry

industry leaders engage in

collaborations

industry leaders do not engage

in collaborations

High fixed costs and low

variable costs

ratio of fixed cost to variable

cost is < 1

ratio of fixed cost to variable

cost is > 1

Ability to meet the needs of

multiple customer segments

less than 2 segments or

dimensions

more than 2 segments or

dimensions

Importance of non-profit

goals

care for branding or social

impact

does not care for branding or

social impact

Presence of an industry

leader

at least one incumbent with

≥ 30% market share

no incumbent with ≥ 30%

market share

Suppliers

Bargaining Power of

Suppliers high low

Supplier threat to integrate

foreword

backward integration of less

than one supplier per year (on

average)

backward integration of more

than one supplier per year (on

average)

Supplier switching cost

between incumbents

< 10% increase in production

cost for supplier

> 10% increase in production

cost for supplier

Incumbent switching costs

between suppliers

< 10% increase in production

cost for incumbent

> 10% increase in production

cost for incumbent

Relative dependence of

suppliers on this industry

profits

this industry is the top source of

profit for supplier

this industry is not the top

source of profit for supplier

Concentration of suppliers

relative to incumbents < 10 suppliers per incumbent > 10 suppliers per incumbent

Availability of substitutes for

what the supplier provides

< 2 current and < 1 future

substitute per year

< 2 current or > 1 future

substitute per year

Incumbent joint investments

with current supplier

incumbent partner with or invest

in the supplier’s business

incumbent does not partner with

and does not invest in the

supplier’s business


among suppliers

< 50% (by monetary value) of

the system is standardized

> 50% (by monetary value) of

the system is standardized


Profits extracted by

suppliers from other

industries < 50% of supplier profits > 50% of supplier profits

Number of industries

supplier serve < 3 industries ≥ 3 industries

Substitutes & Buyers (shared driver)


from this industry product to

substitutes


financial penalty


financial penalty

Buyers & New Entrants (shared driver)

Relative reliance of customer

on industry product

ratio of solar value

to lifetime utility

cost for customer is

< 15%


to lifetime utility


between 15% and

25%


to lifetime utility


> 15%

Buyers, New Entrants, & Rivals (shared driver)


among industry products


financial penalty


financial penalty


discounting by incumbents willing to share < 15% of

profits with customers

willing to share > 15% of

profits with customers

Buyers & Rivals (shared driver)


among incumbents

no product

differentiation

short-term product

differentiation

long-term product

differentiation

New Entrants & Rivals (shared driver)

Commitment of incumbent to

retain and fight over market

share

focus on absolute growth focus on relative growth

Rivals & Suppliers (shared driver)

Fragmentation of the

industry monopoly leaders & followers oligopoly


Table 5.A2: Definition of factor uncertainties in the U.S. residential solar PV industry

Uncertainty Degrees

Technology (and Complements)

T1: Proliferation of electric

vehicles

electric vehicle in the

household

no electric vehicle in the

household

T2: Proliferation of storage

batteries

battery storage system in the

household

no battery storage system in the

household

T3: Change in home total

demand

smaller total

demand same total demand larger total demand

T4: Change in home peak

demand

smaller peak

demand same peak demand larger peak demand

T5: change in size of home

solar system 35% smaller same 35% bigger

T6: Optimal size of home

solar system

< 50% of

home peak

demand

Between 50%

and 90% of

home peak

demand

Between 90%

and 110% of

home peak

demand

> 110% of

home peak

demand

Regulation

R1: Power market structure regulated market deregulated market

R2: magnitude of utility rates high rates low rates

R3: hourly variation in

utility rates high variations low variations

R4: Solar system

connectivity charges connectivity charges no connectivity charges

R5: Solar system control controlled by utility not controlled by utility

R6: Solar territorial cap territorial supply cap no territorial supply cap

R7: Solar system cap below home peak

demand

at home peak

demand

above home peak

demand

R8: Solar rate equal to retail rate below retail rate but

above LCOE equal to LCOE

R9: Solar net-negative

compensation structure carry-over credit monthly payments

R10: Application of ITC &

Depreciation

30% ITC and

accelerated

depreciation

30% ITC only none

R11: Exploitation of FMV applicable not applicable

Growth

G1: Industry growth rate fast slow

CHAPTER 5 — Appendix B: Influence of Positioning Tracks on Competitive Forces 284

Appendix B: Influence of Positioning Tracks on Competitive Forces

Table 5.B1: Probability of {high} power for each competitive force under each positioning

track

Probability of {high} power of…

Positioning Track Substitutes Buyers New

Entrants Rivals Suppliers

Rural, Purchase, Dealer, Outsource 0.388 0.623 0.548 0.644 0.417

Rural, Purchase, Dealer, Insource 0.388 0.621 0.501 0.661 0.409

Rural, Purchase, NoDealer, Outsource 0.388 0.230 0.359 0.682 0.398

Rural, Purchase, NoDealer, Insource 0.388 0.227 0.314 0.696 0.396

Rural, Loan, NoDealer, Outsource 0.277 0.188 0.349 0.607 0.398

Rural, Loan, Dealer, Outsource 0.277 0.518 0.516 0.566 0.417

Rural, Loan, NoDealer, Insource 0.277 0.185 0.304 0.615 0.396

Rural, Loan, Dealer, Insource 0.277 0.515 0.468 0.578 0.409

Rural, LeaseF, Dealer, Outsource 0.344 0.537 0.514 0.600 0.417

Rural, LeaseF, Dealer, Insource 0.344 0.534 0.466 0.611 0.409

Rural, LeaseF, NoDealer, Outsource 0.344 0.213 0.349 0.637 0.398

Rural, LeaseF, NoDealer, Insource 0.344 0.209 0.305 0.644 0.396

Rural, LeaseN, Dealer, Outsource 0.287 0.412 0.502 0.532 0.417

Rural, LeaseN, Dealer, Insource 0.287 0.409 0.454 0.552 0.409

Rural, LeaseN, NoDealer, Outsource 0.287 0.185 0.344 0.576 0.398

Rural, LeaseN, NoDealer, Insource 0.287 0.183 0.300 0.592 0.396

Urban, Purchase, Dealer, Outsource 0.539 0.705 0.536 0.696 0.404

Urban, Purchase, Dealer, Insource 0.539 0.704 0.487 0.720 0.398

Urban, Purchase, NoDealer, Outsource 0.539 0.249 0.331 0.715 0.385

Urban, Loan, Dealer, Outsource 0.396 0.590 0.499 0.627 0.404

Urban, Purchase, NoDealer, Insource 0.539 0.247 0.290 0.737 0.385

Urban, Loan, Dealer, Insource 0.396 0.588 0.450 0.645 0.398

Urban, LeaseF, Dealer, Outsource 0.488 0.615 0.498 0.661 0.404

Urban, LeaseF, Dealer, Insource 0.488 0.612 0.449 0.678 0.398

Urban, Loan, NoDealer, Outsource 0.396 0.201 0.329 0.641 0.385

Urban, Loan, NoDealer, Insource 0.396 0.198 0.284 0.656 0.385

Urban, LeaseN, Dealer, Outsource 0.400 0.482 0.482 0.586 0.404

Urban, LeaseN, Dealer, Insource 0.400 0.481 0.433 0.616 0.398

Urban, LeaseF, NoDealer, Outsource 0.488 0.233 0.325 0.677 0.385

Urban, LeaseF, NoDealer, Insource 0.488 0.230 0.282 0.692 0.385

Urban, LeaseN, NoDealer, Outsource 0.400 0.194 0.323 0.604 0.385

Urban, LeaseN, NoDealer, Insource 0.400 0.193 0.281 0.631 0.385

managing uncertainty and flexibility in the modern …hq284pw2370/... · reshaping the industry. in...

Documents