Strategic Analysis of Technology Selection and …drbrucehartman.net/MGT400/ppts/LNGErkutSonmez.pdfStrategic Analysis of Technology Selection and Capacity Choices in ... a strategic

Strategic Analysis of Technology Selection and Capacity Choices inthe LNG Industry

Erkut SonmezCarroll School of Management, Boston College, 140 Commonwealth Avenue, Chestnut Hill, MA

02467, USA

[email protected]

Sunder Kekre • Alan Scheller-Wolf • Nicola SecomandiDavid A. Tepper School of Business, Carnegie Mellon University, 5000 Forbes Avenue,

Pittsburgh, PA 15213-3890, USA

{sk0a, awolf, ns7}@andrew.cmu.eduNovember 2010

Abstract

Energy plays a fundamental role in both manufacturing and services, and natural gas is quicklybecoming a key energy source worldwide. Facilitating this emergence is the expanding networkof ocean-going vessels that enable the matching of natural gas supply and demand on a globalscale through its transportation in the form of liquefied natural gas (LNG) for eventual regasi-fication at its destination. Until very recently only one type of technology has been availablefor transporting and regasifying LNG: Conventional LNG vessels and land based LNG regasi-fication. But it is now possible to transport and regasify LNG onboard special LNG vessels.Companies such as Excelerate Energy and Hoegh LNG are currently developing LNG supplychains based on this new technology. Motivated by these developments, we engaged executivesat Excelerate Energy to investigate strategic technology selection and capacity choices regardingthe incumbent and emerging technologies. Our analysis brings to light managerial principlesdelineating when to deploy and how to configure each technology option. It also provides in-sights into the effects of using different models of LNG throughput to support these choices inpractice. These findings informed strategic business development decisions at Excelerate En-ergy, challenge conventional wisdom on the role to be played by the emerging technology andprovide answers to open questions faced by companies engaged in the commercial deployment ofthe emerging technology. Beyond this specific application, our insights have broader potentialrelevance for supporting the strategic evaluation of new technologies and the choice of capacityin general.

1. Introduction

Energy is fundamental to any manufacturing and service activity, and natural gas is rapidly acquir-

ing a prominent role as a source of energy worldwide (Geman 2005, Chapter 10). But, due to local

imbalances, matching the supply to the demand for natural gas requires its transportation from

locations with excess supply to locations with excess demand. Over short distances, natural gas

transportation is done by pipelines; over longer distances, natural gas is transported in the form of

liquefied natural gas (LNG) by ocean-going vessels (Tusiani and Shearer 2007). This LNG industry

is currently developing on a global scale (EIA 2003, Jensen 2003).

LNG must be regasified before it can be consumed as natural gas. Until very recently, there

1

existed only one type of LNG regasification technology. In this incumbent technology (onshore

terminal-based regasification), LNG is regasified at a land based terminal, which receives it from

conventional LNG vessels. In contrast, new regasification technology (onboard regasification) allows

special LNG vessels to regasify LNG onboard without requiring a costly onshore terminal. This

new technology is relatively cheap and fast to build, but features slower unloading of the vessels

compared to the incumbent technology. It is currently being commercially deployed by companies

such as Excelerate Energy and Hoegh LNG.

Companies investing in the development of new LNG supply chains (Jensen 2003) now face the

challenge of selecting between the incumbent and emerging LNG regasification technologies: These

technologies can be deployed using different configurations (architecture and capacity level choices)

of the underlying LNG transportation and regasification processes, and these configurations are

characterized by different operational and financial performance, which in turn affect the technology

and configuration choice. Motivated by this challenge, our objective in this paper is to conduct

a strategic analysis of the technology selection and capacity choice comparing the incumbent and

emerging regasification technologies. To do so, we develop and apply to data an integrated analytic

model and engaged executives at Excelerate Energy to support our strategic analysis.

Our integrated model chooses the resource levels that maximize the net present value (NPV)

generated by LNG network based on each technology option at a given LNG throughput require-

ment. We utilize both stochastic (closed queueing network and simulation) and deterministic

models to calculate the throughput rate that can be sustained by a given resource level. The use of

different throughput models allows us to compute a range of possible throughput values that can

be maintained by the given resource level in practice and examine the impact of modeling approach

on our findings. We value the throughput financially using an NPV model.

The application of our integrated analytic model provides insights that are relevant at the

managerial and modeling levels. At the managerial level, we characterize when each technology

is preferred over the other, depending on two critical factors: The throughput and the lead time

difference (LTD) in revenue generation. Moreover, we measure the merit of ship-to-ship LNG trans-

shipment, a configuration of the emerging onboard technology that companies such as Excelerate

Energy and Hoegh LNG are exploring as a way to improve the profitability of this technology. Our

findings (1) informed strategic business development decisions at Excelerate Energy, (2) challenge

the conventional wisdom on the role to be played by the emerging technology, and (3) provide

answers to open questions faced by companies currently engaged in the commercial deployment of

the new onboard technology.

2

At the modeling level, we provide insights on the role to be played by stochastic OM models in

practice when supporting strategic technology selection and capacity choices. To do so, we compare

the results obtained by employing stochastic and deterministic throughput models. Compared to

the stochastic one, we find that deterministic throughput models are (1) adequate to support

technology selection choices in extreme cases (high and low LTD), but unlikely to be so otherwise,

(2) adequate to examine the net benefit of transshipment in the deployment of emerging onboard

technology, and (3) inadequate to support capacity sizing choices with high throughput requirement,

especially for the emerging onboard technology and in some cases even with the incumbent onshore

technology. The errors in capacity sizing due to overlooking the stochastic variability in the relevant

processing times may lead to up to 13% loss ($12 billion) in the NPV generated and 17.29%

throughput shortfall in our specific application. Thus, our findings reveal that strategic technology

selection and capacity sizing decisions may be non-trivial: The application of deterministic models

may suggest different decisions compared to those found by stochastic models.

While our focus is on a specific segment of the LNG industry, our models has potential appli-

cations for a broader class of technology selection problems. It may be used to evaluate other tech-

nology innovations in the LNG industry, such as floating LNG production (Chazan 2009, Tusiani

and Shearer 2007, Ch. 5) rather than regasification. It may also be used to compare technologies

in other industries; for example in settings where one type is cheaper and requires a shorter time to

install, but can sustain a lower production rate; while the other type is more expensive and requires

a longer time to install, but offers a higher throughput. Companies often face such tradeoffs when

developing new technologies, both in manufacturing and service industries. One example occurs in

emerging markets: A company can typically start manufacturing using cheaper and labor-intensive

systems producing at a lower rate, or can enter the market with a more expensive automated system

that sustains a higher production rate. Such companies face technology decisions as we consider

here.

The remainder of this paper is organized as follows: We review the related literature in §2. We

discuss the LNG networks based on the technology options we compare in §3. Section 4 presents

our models. We present our analysis and the insights it generates on the issues we investigate in

§5. We conclude in §6 by discussing further research avenues.

2. Related Literature

Energy has long been an active area of research in both operations management and operations re-

search. Durrer and Slater (1977) review the operations research literature that deals with petroleum

3

and natural gas production. More recently, Smith and McCardle (1998) consider the problem of

valuing oil properties as real options (Dixit and Pindyck 1994, Trigeorgis 1996), and Smith and

McCardle (1999) discuss lessons learned in evaluating oil and gas investments in practice. Hahn

and Dyer (2008) value an oil and gas switching option that arises in the production of these com-

modities. Secomandi (2009b) studies the optimal management of commodity storage assets as real

options and discusses an application to natural gas storage, a topic also explored by Carmona and

Ludkovski (2007), Chen and Forsyth (2007), Boogert and de Jong (2008), and Thompson et al.

(2009). Lai et al. (2009) benchmark practice-based natural gas storage valuation heuristics. Seco-

mandi (2009a) investigates the pricing of natural gas pipeline capacity from various perspectives,

including the real option approach. Enders et al. (2010) study the interaction between technology

and extraction scaling real options in natural gas production. Our work adds to this literature by

considering a novel technology selection problem in the LNG industry.

Closer to the industrial domain that we study, Kaplan et al. (1972), Koenigsberg and Lam

(1976), and Koenigsberg and Meyers (1980) model the shipping stage of an LNG supply chain. In

this paper we use the model of Koenigsberg and Lam (1976) to evaluate the throughput of some

configurations of the technologies that we study, but we also develop original models to evaluate

alternative configurations of the emerging technology. Lai et al. (2010) develop a real option model

to value downstream LNG storage when LNG is regasified using our incumbent LNG regasification

technology. In contrast, we focus on the comparison of this incumbent and the emerging LNG

regasification technologies. Abadie and Chamorro (2009) use Monte Carlo simulation to value

natural gas investments, including an LNG plant, and Ozelkana et al. (2009) use a deterministic

optimization model to analyze the design of LNG terminals. Rodrıguez (2008) develops a real

option model to value delivery flexibility in long-term LNG contracts. None of these authors study

the technology selection problem that we analyze.

Our analysis brings to light managerial insights into the drivers of this technology selection

problem, providing guidance for executives making such technology decisions. Thus, our work is also

related to the operations management literature concerned with establishing principles for guiding

managerial decisions (Fisher 2007, Graves and Jordan 1995). Within this literature, researchers

study technology selection from different perspectives. Krishnan and Bhattacharya (2002) analyze

the relation between product design flexibility and technology selection. Fuchs and Kirchain (2009)

study the impact of production location on technology choice. Van Mieghem (2003) reviews several

papers that deal with capacity management, focusing on the selection between dedicated and flexible

technologies by using stochastic capacity portfolio investment models. In contrast, we study the

4

impact of process configuration and operational and financial performance on technology selection,

by using an integrated evaluation framework.

3. LNG and Regasification Technologies

(a) Option OS: Incumbent onshore terminal based re-gasification.

(b) Option OB: Emerging onboard regasification with-out transshipment.

Figure 1: LNG networks based on the OS and OB options.

LNG is natural gas that has been converted temporarily to liquid form for efficient storage

and economical transportation over long distances. The journey of LNG begins when natural gas,

extracted from underground reservoirs, is sent to a liquefaction facility through a pipeline. At the

liquefaction plant, the natural gas is cooled to minus 260 degrees Fahrenheit transforming it into

LNG. LNG takes 600 times less space than natural gas, thereby making it feasible to transport

it over long distances. LNG vessels load LNG at the liquefaction facility and transport it to

regasification terminals at remote demand locations. At these import terminals, LNG is warmed

back to natural gas. It is then pumped into pipelines feeding the target market.

In this paper, we study the following three regasification technology and architecture options

currently available for LNG supply chains: Incumbent onshore-terminal technology based system

(option OS), emerging onboard technology based system without transshipment (option OB), and

onboard technology based system with transshipment (option OBT). These are described next.

We first contrast the LNG chains based on options OS and OB in Figure 1. In these systems,

ships load LNG at the loading port, transit to the unloading facility, unload their cargos, and

transit back to the loading port. In option OS (Figure 1(a)), conventional LNG carriers (LNGCs)

unload their LNG cargo to the storage tanks of the land-based terminal. LNG in the tanks is then

regasified by the regasification unit and pumped into the local natural gas pipeline. Option OB

(Figure 1(b)), in contrast, has special LNG vessels (LNG regasification vessel-LNGRV) that are

used to regasify LNG onboard at an offshore deepwater port location. In this system, when an

5

Figure 2: Transshipment based configuration of the emerging onboard technology (Option OBT).

LNGRV arrives at an offshore deepwater port, it connects to a submerged unloading buoy. The

LNG is then vaporized onboard the LNGRV and subsequently delivered to shore through a subsea

pipeline.

Figure 2 displays the third option for the LNG chain that features transshipment based ar-

chitecture with the emerging onboard technology (option OBT). In this architecture two types of

ships are used: LNGCs and LNGRVs. In this system, both types of ships keep sailing toward

each other, LNGC from loading port and LNGRV from deepwater port, until the two meet. On

meeting, LNGC transfers its cargo onto LNGRV. After cargo transfer is completed, LNGRV sails

back to deepwater to regasify the cargo while LNGCs sails back to its loading port. The process

then repeats for the next cycle. This type a transshipment network may bring savings in capital

investment by allowing more expensive LNGRVs to dedicate more time for regasification rather

than transportation which can be conducted with cheaper LNGCs.

Typically, LNG supply chains are built as point-to-point networks, where there is one liquefac-

tion plant as the source of supply and one regasification facility at the demand location with LNG

vessels dedicated only to this chain. The vast majority of the global LNG trade is based on this

type static supply chains with long-term project lifetime contracts and constant LNG throughput

requirement rates, due to intensive capital investment required for the facilities and vessels (Tusiani

and Shearer 2007, p. 200). Thus, in this paper we focus on this type of static chains for the three

technology and architecture options we study.

6

4. Models

In this section, we explain the models developed to conduct our strategic technology selection and

capacity configuration choices. We first describe the models developed to calculate the throughput

at a given resource level for each of the LNG network options. Then, we explain the computation

of the net present value (NPV) generated at a given throughput requirement. Finally, we explain

the optimization model employed to select the best resource level that maximizes NPV for a given

throughput requirement.

4.1 Throughput Models

We now describe the stochastic and deterministic models developed to calculate the throughput at a

given resource level for each of the technology/architecture options. The use of both stochastic and

deterministic models in our analysis allows us to examine a range of possible throughput values that

can be maintained by given resource level and configuration. It also helps us to assess the impact

of modeling uncertainty in processing times (stochastic vs deterministic) on strategic technology

and capacity choice decisions.

Stochastic Models. We model the systems corresponding to options OS and OB as closed queue-

ing networks (CQNs), following Koenigsberg and Lam (1976), Koenigsberg and Meyers (1980), and

Wang (2008). Figure 3 represents the process flow in the corresponding CQNs. We model the load-

ing and unloading processes as first-come-first-serve (FCFS) exponential queues, and the transit

processes as ample-server (AS) stations with service time distributions having rational Laplace

transforms. Under these assumptions, each CQN has a closed product-form stationary distribution

(Baskett et al. 1975).

Let I be the total number of blocks (four blocks in Figure 3). We denote the number of ships

in block i as si. The state of the shipping system is the array s := (si, i = 1, . . . , I), and satisfies∑Ii=1 si = S, where S is the total number of ships. Let λi and µi be the mean arrival and service

rate of block i, respectively. Denote π(s) as the steady state probability that the system is in state

s. Following Baskett et al. (1975), π(s) = Γ∏Ii=1 γi(λi, µi, si), where Γ is a normalizing constant

chosen to make these probabilities sum to 1 and γi(·) is computed as follows:

γi(λi, µi, si) :=

{(λiµi )

si , If block i is FCFS,1si!

(λiµi )si , If block i is AS.

(1)

In an OB system, there can be multiple unloading buoys/subsea-pipelines at the unloading port

to enable unloading multiple vessels at the same time. In this case, station 1 (the unloading port in

7

Figure 3: Process flow for options OS and OB.

Figure 3) has multiple servers (buoys/subsea-pipelines), and γ1(λ1, µ1, s1) becomes(λ1µ1

)s1∏s1a=1 y(a)

, where

y(a) is the rate of service at station 1 when there are a vessels at this station relative to the service

rate when there is only one vessel at this station, a = 1. If there are B servers at station 1, then

y(a) :=

{a, 1 ≤ a ≤ B,B, a > B.

Let S denote all the possible states of the system. Also denote by S ′ the set of states in which

at least one ship is in station 3 (loading port), i.e., S ′ := {s ∈ S : s3 > 0}. Then the throughput

rate is:

X = cµ3∑s∈S′

π(s), (2)

where c is the cargo size of a ship.

The only difference between the OS and OB systems is the service rate of the unloading block,

µ1; due to onboard regasification, an LNGRV unloads its cargo at a slower rate than an LNGC.

We calculate the throughput of the onboard technology based system with transshipment (op-

tion OBT) by utilizing a Monte Carlo simulation model. Figure 4 displays the flow chart of our

simulation model. An entity representing an LNGRV or an entity representing an LNGC flow into a

match block immediately after they leave the unloading deepwater port and the liquefaction plant,

respectively. When an entity arrives at the match block, it is placed in one of two associated queues,

one for each vessel type. Entities remain in their respective queues until a match occurs. We record

8

Figure 4: Flow chart of the simulation model for option OBT.

this waiting time in the match block queue to obtain the distance traveled by the matching vessel

before the match occurs.

Once a match exists, one entity from each queue is released. After the vessels leave the match

block, they flow into a batch block to form a single entity representing the paired vessels that will

transship. Batched entities are delayed in the transit-to-meet block for the remaining time required

to meet, which is equal to half the difference between the one way transit time and the previously

recorded time waited in the match block. Then, the batched entity is delayed in the transshipment

block for the time required by the ship-to-ship LNG transfer. When this transfer is completed,

the batched entity is separated into its component entities in the separate block. Upon leaving the

separate block, the entities representing the LNGRV and the LNGC are delayed in their respective

transit blocks for the time required for sailing from the location where transshipment is performed

to the deepwater port and the liquefaction plant, respectively.

We use the ARENA simulation software to calculate the throughput, selecting the simulation

run times and number of replications such that the throughput rate becomes insensitive to the

simulation length and the half-width of a 95% confidence interval is at most 0.5% of the mean.

Deterministic Models. We calculate throughput rate of the deterministic OS and OB networks

as follows. Let ci represents the capacity of station i in the OS and OB based networks in Figure

3, i = 1, . . . , I. For FCFS stations (i = 1, 3), c1 = Bµ1 and c3 = µ3; and for AS stations (i = 2, 4),

9

Figure 5: Process flow for deterministic OBT system.

ci = Sµi. The bottleneck capacity of the LNG network is then K := min(ci, i = 1, . . . , I). Let

D denote demand rate of ships in the system: D := S/∑I

i=1 1/µi. The throughput rate of the

deterministic OS and OB networks is the minimum of bottleneck capacity and demand rate:

Xd = min(K,D). (3)

We represent the process flow of a deterministic onboard system with transshipment (option

OBT) in Figure 5. One can think of this system as two conjoined loops that are coupled via

the transshipment block. Let S1 and S2 be number of LNGRVs and LNGCs in loops 1 and 2,

respectively; and τ be the travel time between unloading and the loading ports. In this network,

since uncertainty in the processing times is eliminated and the ships keep sailing until they meet,

transhipment point will be fixed at one of these locations: Between unloading and loading ports,

unloading port, and loading port. If the transshipment point is between unloading and loading

ports, then loops 1 and 2 will be perfectly coordinated: A type of ship never waits for the other

type to conduct transshipment. In this case demand rate of the network is simply total available

capacity divided by the total processing times in the network. If the transshipment point is at

the unloading port, then LNGCs will wait until LNGRVs finish regasification to conduct LNG

transfer. In this case, the demand rate of the network is the demand rate of bottleneck loop 1:

S11µ1

+ 1µ3

. Similarly, if the transshipment point is at the loading port, the demand rate of the network

is the demand rate of bottleneck loop 2: S21µ5

+ 1µ3

. Demand rate of the ships in the network is

10

then DOBT := min( S1+S2

2(τ+ 1µ3

)+ 1µ1

+ 1µ5

, S11µ1

+ 1µ3

, S21µ5

+ 1µ3

). Let KOBT denotes bottleneck capacity of the

network. The throughput rate of the deterministic OBT network is then the minimum of bottleneck

capacity and demand rate:

XOBTd = min(KOBT , DOBT ). (4)

4.2 Valuation Model

We now explain how the computation of NPV generated at a given throughput requirement. This

is the present value of the revenue stream generated minus the operational and capital investment

costs incurred during the project lifetime . Assuming that the capital investment costs are incurred

at time zero, we discount the cash flows over the project lifetime using a constant annual risk-free

rate; that is, we use a risk neutral valuation approach (Smith 2005, Luenberger 1998, Ch. 13).

In order to calculate the revenue, we use New York Mercantile Exchange (NYMEX) natural gas

futures prices. Since we value the revenue stream using futures prices, a risk neutral valuation

approach is appropriate. Moreover, since the futures prices capture the current market view of

future supply and demand conditions, this approach implicitly takes into account uncertainty in

future LNG demand. In addition, we also assume that any regasified LNG can be sold on the

natural gas spot market at the prevailing market price at the time of regasification, i.e. the amount

of natural gas that is vaporized and pumped into the local natural gas pipeline system does not

affect the natural gas price. Given the size of the U.S. natural gas market, this is a reasonable

assumption.

4.3 Optimization of Resource Levels

We now describe the model that calculates the optimal resource levels maximizing the NPV at

a given target throughput requirement for the LNG network options we study. We employ the

following notation to formulate our model in addition to previously introduced ones.

• N : nonnegative integer valued resource level array: N := (S1, S2, B). For option OS, NOS :=

(0, S2, 0), for option OB, NOB := (S1, 0, B), and for option OBT, NOBT := (S1, S2, B);

• C1(N): present value of investment and shipping costs when resource level N is deployed;

• X(N): mean throughput rate that can be obtained when resource level N is deployed;

• XT : LNG throughput requirement rate;

11

• R(XT ): present value of the revenue generated at XT throughput requirement rate;

• C2(XT ): present value of liquefaction and regasification costs at XT throughput requirement

rate;

• V (XT , N): NPV generated at XT throughput requirement rate when resource level N is

deployed: V (XT , N) = R(XT )− (C1(N) + C2(XT ))

The optimization model is

maxN

V (XT , N)

s.t. X(N) ≥ XT .

The objective function captures the NPV generated by the given LNG network when throughput

requirement rate is XT and resource level N is deployed. The constraint ensures that at resource

level N , the system can always sustain XT mean throughput rate.

5. Analysis

We apply our models to conduct a field study using financial and operational data. Some of the

parameter values used in our study were determined in concert with the managers of Excelerate

Energy. Others are based on the existing LNG literature. Table 1 reports the relevant units of

measurement and conversion factors.

5.1 Numerical Values for the Relevant Parameters

We consider an integrated LNG chain with a 25 year lifetime, the length of a typical LNG project

(Flower 1998). Our LNG chain has one liquefaction facility and one regasification facility. With

the incumbent technology, we assume that the regasification terminal is located at Lake Charles,

Louisiana, which indeed hosts an onshore LNG terminal operated by Trunkline LNG. We also

assume that the offshore facility is located nearby; for example, the Gulf Gateway offshore deepwater

port operated by Excelerate Energy is located 100 miles off the Louisiana cost. We assume that

the liquefaction plant is located in Egypt, one of the major LNG exporters (Smith et al. 2004).

The distance between Egypt and Lake Charles is approximately 7,000 NMs.

We use the following parameters in our study.

Shipping: We consider a homogeneous ship cargo size of 3 bcf, which is common in the LNG

industry (Flower 1998). We assume a shipping speed of 19 knots (Cho et al. 2005, Flower 1998, p.

12

Table 1: Units of measurement and conversion factorsbcf Billion Cubic Feetcm Cubic Meterbcf/d Billion Cubic Feet per DayMMTPA Million Tons per AnnumMMBTU Million British Thermal UnitsNM Nautical Mile1 Knot = 1 NM per Hour1 bcf = 1,100,000 MMBTU1 MMTPA = 0.128 bcf/d1 cm = 0.0000215 bcf

100). With this assumption, a one-way trip between the regasification facility and the liquefaction

plant takes approximately 15 days, on average.

Liquefaction Plant: Following Wang (2008), we consider the service time at the liquefaction

plant (loading port) to be exponentially distributed with mean 1 day. The service time is the time

required by a vessel for entering the loading port, loading 3 bcf of LNG, completing the required

paperwork, and leaving the port.

Onshore Terminal: We assume that the regasification capacity of the onshore terminal is 2

bcf/d, which is consistent with the capacity of some of the onshore terminals in the U.S., including

Lake Charles. We set the service time at the onshore terminal (entering the port, unloading 3

bcf of LNG into the storage tanks, completing the required paperwork, and leaving the port) as

exponentially distributed with mean 1 day (Koenigsberg and Lam 1976). Following Lane (2008),

we let the LNGC capital cost be $250M (M denotes million).

The capital cost of an onshore terminal varies considerably depending on factors such as storage

and vaporization capacity, cost of real estate, geological structure, local labor and construction

costs, and marine environment (Tusiani and Shearer 2007). Thus, varying cost figures are reported

in the literature. For instance, Smith et al. (2004) state that a 1 bcf/d regasification terminal costs

$0.5B (B denotes billion), and EIA (2003) states that the cost of a terminal can range from $0.1B

to $2B depending on its regasification capacity. Following Lane (2008) in our base case, we let the

onshore terminal cost be $1.5B. We also conducted an analysis including the cost of the onshore

terminal as a function of its regasification capacity, consistent with these cost figures. We explain

how our conclusions change when we use this cost function instead of a $1.5B fixed cost for all

throughput levels in §5.2.

Tusiani and Shearer (2007) report that the construction time for an LNG terminal does not

generally vary with the size of the facility. Rather, it is determined by the construction schedule

13

Figure 6: NYMEX natural gas futures prices.

for the storage tanks, the most time-consuming and expensive components of a terminal, and it

may take between 2 and 5 years. As our base case, we assume that it takes 5 years to construct the

onshore terminal. In §5.2 we explain how our conclusions would change with lower construction

time of the terminal.

Deepwater Port: We assume that the LNG regasification rate of an LNGRV is 0.5 bcf/d (Energy

Bridge Fact Sheets 2008). We set the service time at the deepwater port (mooring, connecting with

submerged buoy, vaporizing 3 bcf of LNG, and leaving the port) as exponentially distributed with

mean 7 days (Lane 2008). We let the capital cost of an LNGRV be $275M (Lane 2008). We assume

that each buoy/subsea-pipeline structure (each server) at the deepwater port costs $70M, and that

it takes 1 year to construct the deepwater port (Gulf Gateway Fact Sheets 2008), independent of

the number of buoys in the deepwater port. The LNG transshipment service time is taken to be 2

days on average (Lane 2008), and is assumed to be exponentially distributed with this mean.

Operational Cost: This cost has three components: Liquefaction, shipping, and regasification.

Following Wang (2008), we assume that the liquefaction plant operating cost is $8M per MMTPA.

According to Lane (2008), the shipping cost is $47.851M per ship per year (this includes fuel and

crew costs). Finally, we take the regasification variable cost as $0.0285 per MMBTU with a 1.69%

fuel loss (Wang 2008).

Revenue: We use NYMEX natural gas futures prices as of 8/8/2008 (Figure 6) for calculating the

relevant revenue figures. For each trading day, NYMEX futures prices are available for maturities

14

Figure 7: NPV and costs difference between the OB and OS systems.

of 148 months in the future. To estimate the futures prices for the months beyond the last available

maturity, we replicate the prices of the last 12 available months. We set the annual risk-free interest

rate as 1.7%, the three-month U.S. Treasury rate as of 8/8/2008.

5.2 Technology Selection: Comparing OS and OB

In this subsection we analyze the conditions that favor selection of each of the regasification technol-

ogy options. To evaluate the impact of modeling approach, we compare the technologies employing

both the stochastic and deterministic models of throughput detailed in §4.1. In our comparison,

LNG networks using each technology option are configured with the optimal resource levels for

a given target throughput requirement as discussed in section §4.3. We then vary throughput

requirements up to and including 2 bcf/d.

We first report our findings obtained employing the stochastic throughput model. The dashed

line in Figure 7 shows the difference between the present values of the total costs of the OB and

OS options under the parameters reported in §5.1. This difference is obtained by subtracting the

total capital and operating costs of the OS system from those of the OB system1. The dashed cost

difference line shows that for “low” throughput levels - less than 0.5 bcf/d - the OB system’s cost

is lower than that of the OS system. This arises due to the lower capital investment required to

1The jittery pattern of the cost difference line is caused by the integer-valued fleet size difference between the OBand OS systems. The magnitude of each peak corresponds to the capital and operating cost of an additional vesselrequired by the OB system compared to the OS system to sustain the throughput interval in which the peak occurs.This fleet size difference also creates the jittery pattern of the NPV difference line in Figure 7.

15

Figure 8: Technology comparison: The LTD effect.

build the offshore deepwater port compared to the capital intensive land based terminal for OS

option. However to sustain higher throughput levels, the OB system needs several unloading buoys

and more vessels than the OS system, due to its lower unloading rate. The capital investment for

multiple unloading buoys and the capital and operating costs of the extra vessels diminish the cost

advantage of the OB system as throughput rises, and soon the total cost of the OB system becomes

significantly larger than that of the OS system for “high” throughput levels - more than 0.5 bcf/d.

The NPV differences show more interesting trends.

The solid line in Figure 7 displays the difference between the NPVs generated by the two

systems; we obtain this difference by subtracting the NPV of the OS system from the NPV of

the OB option. In this case, we find that for all throughput levels, the OB system generates

significantly more NPV than the OS system, although the cost of the OB system is much higher

for high throughput levels. This result arises due to the shorter time required to build an onboard

regasification facility compared to an onshore terminal (recall that we assume it takes one year

to complete the deepwater port and five years to construct the onshore terminal). Thus, the OB

system starts generating revenue four years earlier than the OS option. As a result, as shown

in Figure 7, the OB system is more profitable even its total cost is greater than that of the OS

system for high throughput requirements. This is due to the lead time difference (LTD) in revenue

generation.

Clearly, the NPV difference displayed in Figure 7 is specific to the parameters reported §5.1.

16

In practice, LTD and the NPV advantage it generates for OB system can vary with operational

parameters and market conditions, such as the permit approval process, facility construction time,

availability of the vessels, LNG supply, natural gas futures prices, interest rates, etc. For instance,

due to idiosyncrasies in the LNG industry, building LNGRVs, which are used in the OB option, may

take far more than the year we assumed. This will decrease the LTD. Moreover, due to economic

downturns, the natural gas prices or interest rates can decrease to levels lower than assumed in our

computations, which will reduce the NPV advantage for the OB option arising from LTD. In all of

these cases, the slope of the NPV difference curve would be lower than that displayed in Figure 7,

rotating the curve clockwise as illustrated in Figure 8.

(a) High LTD. (b) Medium LTD. (c) No LTD.

Figure 9: Technology selection.

To further illustrate the impact of LTD on technology selection, we analyze three scenarios:

High, medium and no LTD. In each of these scenarios, we vary the revenue start time of the OB

system (high LTD - year 1, medium LTD - year 3, and no LTD - year 5), while fixing the completion

time for the onshore terminal at year 5. All the other parameter values remain as reported in §5.1.

Figure 9 illustrates the technology choice for each of these LTD cases. Under the high LTD scenario

(Figure 9(a)) the OB option will be the choice at all throughput requirements, due to the factors

explained earlier. Under medium LTD, three regions appear. At either extreme of the throughput

requirement, low (less than 0.5 bcf/d) and high (more than 1.75 bcf/d), OB and OS options

are preferred, respectively. However, at the remaining throughput interval, neither technology

dominates the other. In this no-dominance region, the preferred technology option oscillates as

throughput changes. This is due to the total capital and operating costs associated with under-

utilized shipping capacity (under-utilization arises due to the integer fleet-size assumption) in each

of the technology options. Finally, when there is no LTD (Figure 9(c)), again three regions result.

For a low throughput requirement (less than 0.5 bcf/d), OB is the preferred option and at higher

17

throughput levels (more than 0.6 bcf/d), OS is better from an NPV standpoint. The no-dominance

region still appears, but it is much smaller compared to the medium LTD case.

(a) High LTD. (b) Medium LTD. (c) No LTD.

Figure 10: Technology selection with deterministic throughput model.

We replicated our analysis on technology selection employing the deterministic throughput

models to assess the impact of the modeling approach on the strategic technology adoption choice.

Figure 10 displays the technology selection as a function of the throughput requirement under the

three LTD scenarios analyzed when the deterministic throughput model is utilized. Compared to

the stochastic analysis presented in Figure 9, we again observe dominance and no-dominance regions

arising with different LTD levels. However, the boundaries of these regions (such as no-dominance)

may show differences.

Both the stochastic and deterministic analysis revealed the no-dominance region where the

preferred technology option oscillates as throughput requirement changes, due to under-utilized

shipping capacity arising with integral fleet sizes. In practice, companies can adjust the specifica-

tions of their vessels (e.g. engine size, cargo capacity, speed, etc.) to achieve high utilization levels

for vessels. However, in general, they have access to only a limited set of specification options. This

means full-utilization may not always be achievable. Moreover, the integrality constraint cannot be

ignored. Thus, the no-dominance region poses a formidable challenge for technology selection. In

this region, further analysis is required in which other factors such as available vessel specification

options have to be evaluated. Moreover, technology choice in this no-dominance region is poten-

tially sensitive to the throughput model used. The cost of wrong technology choice in this region

can be as high as $1.21B (capital and operating cost of a vessel) as illustrated in Figure 8.

On the other hand, in regions where one technology option dominates the other, the preferred

option will remain the same even when optimal specifications that can sustain full-utilization of

vessels can be deployed. In other words, technology selection in these dominance regions is rela-

18

Figure 11: NPV difference with linear fleet size.

tively easy compared to the no-dominance region. To support this, we replicated our analysis on

technology selection by relaxing the integrality constraint: We compute the throughput with the

deterministic throughput model as in equation 3, while the number of vessels, S, can take fractional

values. Linearization of the fleet size allows full utilization of the assigned shipping capacity at a

given throughput requirement. This mimics the best case scenario in which companies can adjust

vessel specifications such that all vessels are full-utilized. Figure 11 presents the NPV difference

between the OB and OS technology options when this linear deterministic throughput model is

employed. Compared to our analysis with integrality constraint, we observe that the preferred

technology options in dominance regions remains same.

Our analysis in this section reveals the conditions specifying when each technology should be

adopted as a function of throughput (operational performance) and LTD (financial performance)

as summarized in Figure 12. We find that:

• If the throughput requirement is low, an OB system is always more profitable than an OS

system due to the OB system’s lower capital investment cost.

• As the throughput requirement increases, the technology adoption choice depends on LTD.

Although the OB system’s total cost is greater than that of the OS system for high throughput

requirement levels, the extra NPV that may be obtained by the OB system due to faster

revenue generation may make the it more profitable.

19

Figure 12: Technology adoption choices (the graph is drawn by studying various LTD levels andthen connecting the boundary points of no-dominance regions).

The former finding is consistent with the literature (Jensen 2003, Smith et al. 2004). The latter

finding contrasts with those obtained by Jensen (2003) and Smith et al. (2004): These authors state

that the emerging onboard technology is well-suited for seasonal and occasional usage, that is, the

low throughput case; they also report that the incumbent onshore technology is more profitable than

the emerging onboard technology in the high throughput case. In contrast, our models demonstrate

that the onboard technology can also be preferred to sustain high throughput, provided the NPV

advantage arising with LTD is high.

At first glance, it is surprising that the slower onboard technology, which features a longer time

for vessel unloading, outperforms onshore technology even at high throughput levels. The reason

for this apparently counterintuitive result is that there is possibility of configuring the capacity of

the onboard technology using multiple unloading buoys, a factor extant literature and practitioners

have seemingly ignored. Optimization of resource level by selecting the number of unloading buoys

overcomes the disadvantage of slower unloading rates of the onboard technology. This, together

with NPV advantage arising due to LTD, leads to the onboard technology to outperform the onshore

technology at higher throughput requirements. Thus, our analysis challenges the LNG industry to

think differently about the emerging onboard technology: Our insights should help in promoting

the adoption of this emerging technology.

In §5.1, we explained that the capital cost of an onshore terminal varies considerably depending

20

Figure 13: Comparing OB and OBT options with stochastic throughput models.

on factors such as storage and vaporization capacity, cost of real estate, geological structure, local

labor and construction costs, and marine environment. Therefore, we also analyzed the sensitivity

of our findings on technology choice to the capital cost of the onshore terminal by modeling it as a

function of the terminal’s regasifaction capacity. We now assume that capital cost of the onshore

terminal increases $0.5B for every 0.5 bcf/d increase in throughput requirement. These numbers

are consistent with the cost figures reported in the literature as explained in §5.1. Compared to

our analysis with a fixed terminal cost, we observe that the OS system may be more profitable

than the OB system for some throughput intervals in the low throughput region. However, for

high throughput requirements, our findings on technology choice remain robust: Technology choice

depends on LTD as when fixed terminal cost is used.

5.3 The Benefit of LNG Transshipment with the Emerging Technology: Com-paring options OB and OBT

In this section we compare options OB and OBT to study the merit of ship-to-ship LNG trans-

shipment, a configuration aspect that companies such as Excelerate Energy and Hoegh LNG are

exploring to improve the profitability of the OB option. Transshipment allows a firm to configure a

fleet of ships as a mix of the cheaper LNGCs and more expensive LNGRVs. Such a configuration can

potentially reduce capital investment costs at the expense of introducing an additional processing

step in the LNG network: Ship-to-ship LNG transfer. It also partially decouples the transportation

from storage and regasification processes, which leads to higher utilization of the capital intensive

21

Figure 14: Comparing OB and OBT options with deterministic throughput models.

assets of regasification, namely LNGRVs. In other words, while incurring additional processing

time in the network, transshipment enables existing LNGRVs to dedicate more time for regasifi-

cation, rather than transportation. We examine these tradeoffs by computing the net benefit of

transshipment in terms of improved profitability of an onboard technology based system. We also

evaluate how such an advanced configuration impacts on technology selection insights presented

earlier in §5.2.

Figure 13 displays the difference between the NPV of the systems without and with transship-

ment. We obtain this difference by subtracting the NPV of the system with transshipment from the

NPV of the system without transshipment. This figure quantifies the tradeoff between the capital

investment savings obtained by replacing the expensive LNGRVs with the cheaper LNGCs and

the value of lost throughput due to additional transshipment activity. We observe that for most

throughput levels, the OB option generates significantly more NPV than the OBT option. This

arises since an OBT system typically requires more vessels than a OB system to sustain a given

throughput requirement, due to the additional time required for the ship-to-ship LNG transfer and

the synchronization of the ships. The capital and operating costs of these extra ships far exceed

the savings resulting from using the cheaper LNGCs in the transshipment network. Only at some

throughput intervals less than 0.5 bcf/d, does transshipment pay off in terms of NPV, but as seen

in Figure 13, even in these cases the benefit is marginal.

We also study the merit of transshipment utilizing deterministic models of throughput for OB

22

Figure 15: Optimal resource levels calculated by the deterministic and stochastic models - OB.

and OBT systems. Figure 14 displays the difference between the NPV of the systems without and

with transshipment when deterministic models are employed. We observe that in the deterministic

analysis, the OB option also generates significantly more NPV than the OBT option in most

throughput requirement levels.

Our analysis, both with stochastic and deterministic models of throughput, points out that

LNG supply chains based on OB regasification technology should be developed, when possible,

only using dedicated LNGRVs, rather than both vessel types. The use of transshipment should

only be considered as a way to circumvent capacity restrictions if the availability of LNGRVs is an

issue in the market. This finding provides an answer to the process architecture and fleet structure

choices faced by LNG companies planning to use or already using emerging onboard technology

(Bryngelson 2007). Our analysis also reveals that the insights on technology selection as driven

by throughput and LTD presented in §5.2 are not affected by the potential deployment of the

transshipment architecture.

5.4 Capacity Sizing

In this subsection we investigate the optimal resource levels obtained using stochastic and determin-

istic throughput models for each of the technology options compared. Obviously, optimal resource

levels assigned by utilizing stochastic throughput models, which account for uncertainty in the

processing network, are higher compared to the deterministic models. What is less obvious is the

23

magnitude of the difference in resource levels and its effect on throughput and NPV. We quantify

these effects.

(a) Target and achieved throughput levels (b) NPV shortfall

Figure 16: Impact of modeling approach on throughput and NPV - OB.

Figure 15 illustrates the optimal OB system configuration (number of buoys and vessels) com-

puted by employing both deterministic and stochastic throughput models at various throughput

requirements. This figure shows that the resource levels prescribed by using stochastic throughput

model are never lower than those obtained with the deterministic throughput model, a consequence

of the congestion that the stochastic model accounts for. While the fleet sizes are often similar,

the stochastic model typically suggests a higher number of unloading buoys. In order to increase

throughput, the optimization model based on the stochastic model of throughput first chooses to

install an additional buoy, since the capital and operating cost of an additional vessel is far higher

than the cost of an unloading buoy. When adding an extra buoy can no longer boost the through-

put, this model demands to add an extra ship to meet the additional throughput requirement. This

increases the throughput dramatically, so the target throughput can be met with fewer buoys (note

that in Figure 15 when the number of buoys decreases, this always coincides with an increase in

fleet size).

Figure 16(a) displays the throughput levels achieved when optimal resource levels computed

using the deterministic throughput model are evaluated with the stochastic throughput model.

This yields the gap between the targeted and achieved throughput levels, and is seen to sometimes

be as large as 17.29%. It indicates the potential throughput shortage relative to the targeted

level when deterministic models are used. Figure 16(b) quantifies the NPV shortfall when optimal

resource levels suggested by deterministic model are evaluated with the stochastic model. The NPV

shortfall is seen to be as large as $12 billion, 13% of the total NPV generated2. These numbers

2Due to the integrality assumption on fleet size, NPVs generated by OB and OS systems are non-monotone

24

Figure 17: Optimal fleet size calculated by the deterministic and stochastic models - OS.

reflect the potential huge errors resulting from deterministic analysis.

(a) Target and achieved throughput levels (b) NPV Shortfall

Figure 18: Impact of modeling approach on throughput and NPV - OS.

Figures 17 and 18 are analogous to Figures 15 and 16 and relate to the onshore system. Although

the throughput and NPV shortfalls due to employing the deterministic throughput model are still

present and can be as large as 7.86% and $5.5 billion (8.4% of the total NPV), they appear to be

smaller than the OB option due to the shorter vessel unloading times.

The stochastic models of throughput employed in our analysis assume exponentially distributed

processing times. The exponential and deterministic processing times should be interpreted as two

functions of the throughput requirement. This leads to the negative valued peaks in NPV shortfall graphs 16(b) and18(b). However, the throughput requirements where NPV shortfall is negative can never be the target for both theOB and OS options since more NPV can be generated with sustaining less throughput.

25

extreme cases: Koenigsberg and Lam (1976) and Kaplan et al. (1972) report that the coefficient of

variation (CV) for actual processing times will probably be in the range 0.15-0.25. To account for

this, we also compute the throughput and NPV when CV of processing times are 0.15 and 0.25.

In these CV cases, processing time distributions are assumed to be normal and negative processing

times are reset to zero. We use simulation to calculate the throughput.

First, we evaluate the optimal resource levels computed with the deterministic throughput

models using our simulation model where CV of processing times are 0.15 and 0.25. We find that,

for the OB system, throughput and NPV shortfalls are seen to be as large as 6.31% and $4.95 billion

(5.3% of total NPV) when the CV is 0.15; and 8.9% and $6.47 billion (6.96% of total NPV) when

the CV is 0.25. For the OS system, throughput and NPV shortfalls are seen to be as large as 3.89%

and $3.62 billion (3.44% of total NPV) when the CV is 0.15; and 4.47% and $4.25 billion (4.1%

of total NPV) when the CV is 0.25. These numbers reflect that throughput and NPV shortfalls

resulting from the deterministic analysis are significant even when the CV of the processing times

are small.

We also compare exponential(CV = 1) and simulation (CV = 0.15 and CV = 0.25) models

to see the cost of over provision. We find that, for both OB and OS systems, exponential model

suggests at most one more vessel compared to the simulation models with CV values 0.15 and 0.25.

The capital cost of an extra vessel is $275M for OB and $250M for OS system. These numbers

reflect that the cost of over provision on the uncertainty in processing times is much lower than the

cost arises when it is ignored.

To sum up, our analysis reveals that deterministic models of throughput are inadequate for

sizing resource levels for both the technology options, OB and OS. They lead to substantially lower

investments (resource levels) and missed opportunities (lost NPV) although they may yield the

technology choice right.

6. Conclusions

In this paper, we study LNG regasification technology selection and capacity choices motivated by

the recent developments in this industry. Our analysis brings to light the conditions under which a

specific regasification technology and its configuration is appropriate for adoption. We also analyze

the effects of using different models of LNG throughput to support these choices in practice. Some

of our insights attribute a different role to the emerging technology than currently envisioned; others

offer new perspectives on pressing issues encountered by companies that are currently deploying

this technology on a commercial scale. The application of our models provided novel guidelines

26

to executives at Excelerate Energy for their strategic-level planning for capital investments and

operating decisions.

Our work could be extended in several directions. In this paper, we focus on technology innova-

tions in the regasification and transportation of LNG. Increased global LNG demand has also led to

several technology innovations in the upstream portion of LNG supply chains; for example, floating

offshore liquefaction facilities (FOLFs). Companies that are seeking alternatives to conventional

onshore natural gas liquefaction plants have expressed growing interest in FOLFs (Chazan 2009,

Tusiani and Shearer 2007, Ch. 5). These facilities can offer greater flexibility and lower cost and

capacity installation time compared to onshore liquefaction plants (Loo 2009). One could adapt

our models to study the selection of technology for natural gas liquefaction.

We analyze the profit of an integrated LNG chain. However, LNG chains may include multiple

parties that manage different stages of the chain, such as LNG producers, shippers, and merchants,

who may have conflicting objectives. Our models could be extended to include the perspectives of

different parties within a game-theoretic framework. These models could be used to analyze the

impact of ownership and contract terms on the architecture and technology choice within an LNG

supply chain.

Due to competition between LNG import markets, some LNG importers have recently exper-

imented with diverting their cargoes to the markets with the greatest profit margins, such as the

U.K., Spain, and Japan. Our models could be extended to assess the value of delivery flexibility,

that is, the ability to deliver a cargo to the most attractive market, and to support the development

of practical vessel routing policies.

Acknowledgments

The authors thank Captain Mark K. Lane, Vice President-Operations at Excelerate Energy, for his

help and support in determining the operational parameters in our numerical study.

References

Abadie, L. M., J. M. Chamorro. 2009. Monte Carlo valuation of natural gas investments. Review of FinancialEconomics 18 10–22.

Baskett, F., K. M. Chandy, R. R. Muntz, F. G. Palacios. 1975. Open, closed, and mixed networks of queueswith different classes of customers. Journal of the Association for Computing Machinery 22 248–260.

Boogert, A., C. de Jong. 2008. Gas storage valuation using a Monte Carlo method. Journal of Derivatives15 127–147.

Bryngelson, R. 2007. Northeast Gateway Energy Bridge deepwater port project status. Available athttp://www.northeastgas.org/pdf/r bryngelson1107.pdf.

27

Carmona, R., M. Ludkovski. 2007. Valuation of energy storage: An optimal switching approach. WorkingPaper, Princeton University, Princeton, NJ, USA.

Chazan, G. 2009. Shell plans to build floating gas plant. Wall Street Journal,http://online.wsj.com/article/SB10001424052748704882404574461321132480810.html.

Chen, Z., P. A. Forsyth. 2007. A semi-Lagrangian approach for natural gas storage valuation and optimaloperation. SIAM Journal on Scientific Computing 30 339–368.

Cho, J. H., H. Kotzot, F. de la Vega, C. Durr. 2005. Large LNG carrier poses economic advantages, technicalchallenges. LNG Observer 2. October.

Dixit, A. K., R. S. Pindyck. 1994. Investment under Uncertainty . Princeton University Press, Princeton,NJ, USA.

Durrer, E.J., G.E. Slater. 1977. Optimization of petroleum and natural gas production - A survey. Manage-ment Science 24(1) 35–43.

EIA. 2003. Energy Information Administration, The global liquefied natural gas market: Status and outlook.U.S. Energy Information Administration, U.S. Department of Energy, Washington, DC, USA.

Enders, P., A. Scheller-Wolf, N. Secomandi. 2010. Interaction between technology and extraction scalingreal options in natural gas production. IIE Transactions Forthcoming.

Energy Bridge Fact Sheets. 2008. Available at http://www.excelerateenergy.com/downloads/Excelerate energybridge.pdf.

Fisher, M. 2007. Strengthening the empirical base of Operations Management. Manufacturing and ServiceOperations Management 9 368–382.

Flower, A. R. 1998. LNG project feasibility. In G. B. Greenwald (ed.), Liquefied Natural Gas: Developingand Financing International Projects. Kluwer Law International, London, UK.

Fuchs, E., R. Kirchain. 2009. Design for location? The impact of manufacturing offshore on technologycompetitiveness in the optoelectronics industry. Working Paper, Carnegie Mellon University, Collegeof Engineering.

Geman, H. 2005. Commodities and Commodity Derivatives: Modeling and Pricing for Agriculturals, Metalsand Energy . John Wiley & Sons, Chichester, UK.

Graves, S. C., W. C. Jordan. 1995. Principles on the benefits of manufacturing process flexibility. ManagementScience 41 577–594.

Gulf Gateway Fact Sheets. 2008. Available at http://www.excelerateenergy.com/downloads/Excelerate gulfgate08.pdf.

Hahn, Warren J., James S. Dyer. 2008. Discrete time modeling of mean-reverting stochastic processes forreal option valuation. European Journal of Operational Research 184 534–548.

Jensen, J. T. 2003. The LNG revolution. The Energy Journal 24 1–45.

Kaplan, M., R. C. Wentworth, R. J. Hischier. 1972. Simulation and optimization of LNG shipping sys-tems. ASME Transactions of the Petroleum Mechanical Engineering and Pressure Vessels and PipingConference, New Orleans, September 17-21, 1-11.

Koenigsberg, E., R. C. Lam. 1976. Cyclic queue models of fleet operations. Operations Research 24 516–529.

Koenigsberg, E., D. A. Meyers. 1980. An interacting cyclic queue model of fleet operations. The Logisticsand Transportation Review 16 59–71.

Krishnan, V., S. Bhattacharya. 2002. Technology selection and commitment in new product development:The role of uncertainty and design flexibility. Management Science 48 313–327.

Lai, G., F. Margot, N. Secomandi. 2009. An approximate dynamic programming approach to benchmarkpractice-based heuristics for natural gas storage valuation. Operations Research Forthcoming.

Lai, G., M. X. Wang, S. Kekre, A. Scheller-Wolf, N. Secomandi. 2010. Valuation of the real option to storeliquefied natural gas at a regasification terminal. Working Paper 2006-E99, Tepper School of Business,Carnegie Mellon University, Pittsburgh, PA, USA.

Lane, M. K. 2008. Vice President - Operations at Excelerate Energy. Personal Communication.

28

Loo, F. 2009. Preview-LNG industry seeks answers to untried floating LNG. Available athttp://www.reuters.com/ article/marketsNews/idUSSP50927120090727?sp=true.

Luenberger, D. G. 1998. Investment Science. Oxford University Press, New York, NY, USA.

Ozelkana, E. C., A. D’Ambrosio, S. G. Teng. 2009. Optimizing liquefied natural gas terminal design foreffective supply-chain operations. International Journal of Production Economics 111 529–542.

Rodrıguez, R. Y. 2008. Real option valuation of free destination in long-term liquefied natural gas supplies.Energy Economics 30 1909–1932.

Secomandi, N. 2009a. On the pricing of natural gas pipeline capacity. Manufacturing & Service OperationsManagement Forthcoming.

Secomandi, N. 2009b. Optimal commodity trading with a capacitated storage asset. Management ScienceForthcoming.

Smith, E. N., J. W. McFarland, J. M. Trapani, S. E. Michaelides, J. R. Moroney, C. W. Nelson. 2004.Liquefied natural gas imports and their impact on the state, regional, and national economies. Entergy-Tulane Energy Institute.

Smith, J. E. 2005. Alternative approaches for solving real-options problems. Decision Analysis 2 89–102.

Smith, J. E., K. F. McCardle. 1999. Options in the real world: Lessons learned in evaluating oil and gasinvestments. Operations Research 47 1–15.

Smith, J.E., K.F. McCardle. 1998. Valuing oil properties: Integrating option pricing and decision analysisapproaches. Operations Research 46(2) 198–217.

Thompson, M., M. Davison, H. Rasmussen. 2009. Natural gas storage valuation and optimization: A realoptions application. Naval Research Logistics 56 226–238.

Trigeorgis, L. 1996. Real Options: Managerial Flexibility and Strategy in Resource Allocation. The MITPress, Cambridge, MA, USA.

Tusiani, M. D., G. Shearer. 2007. LNG: A Nontechnical Guide. PennWell Corporation, Tulsa, OK, USA.

Van Mieghem, J. A. 2003. Capacity management, investment, and hedging: Review and recent developments.Manufacturing and Service Operations Management 5 269–302.

Wang, M. X. 2008. Supply chain management and economic valuation of real options in the natural gas andliquefied natural gas industry. Ph.D. thesis, Carnegie Mellon University.

29